aspects of the theory of clitics

Aspects of the Theory of Clitics

O X F O R D S T U D I E S I N T H E O R E T I C A L L I N G U I S T I C S

general editors: David Adger, Queen Mary College London; Hagit Borer, University ofSouthern California

advisory editors: Stephen Anderson, Yale University; Gennaro Chierchia, University ofMilan; Rose-Marie Dechaine, University of British Columbia; Elan Dresher, University of Toronto;James Higginbotham, University of Southern California; Pat Keating, University of California, LosAngeles; Ruth Kempson, King’s College, University of London; James McCloskey, University ofCalifornia, Santa Cruz; Gillian Ramchand, University of Tromsø; Maria-Luisa Zubizarreta,University of Southern California

published

1 The Syntax of Silenceby Jason Merchant

2 Questions and Answers in Embedded Contextsby Utpal Lahiri

3 Phonetics, Phonology, and Cognitionedited by Jacques Durand and Bernard Laks

4 At the Syntax-Pragmatics Interfaceby Lutz Marten

5 The Unaccusativity Puzzleedited by Artemis Alexiadou, Elena Anagnostopoulou, and Martin Everaert

6 Beyond Morphologyby Peter Ackema and Ad Neeleman

7 The Logic of Conventional Implicaturesby Christopher Potts

8 Paradigms of Phonological Theoryedited by Laura Downing, T. Alan Hall, and Renate Raffelsiefen

9 The Verbal Complex in Romanceby Paola Monachesi

10 The Syntax of Aspectedited by Nomi Erteschik-Shir and Tova Rapopart

11 Aspects of the Theory of Cliticsby Stephen Anderson

i n p r e pa r at i o n

Phi Syntax: A Theory of Agreementby Susana Bejar

Aspect and Reference Timeby Olga Borik

Prosodic Morphologyby Laura Downing

Stratal Optimality Theoryby Ricardo Bermúdez Otero

Tense, Mood, and Aspectedited by Alessandra Giorgi, James Higginbotham, and Fabio Pianesi

The Ecology of English Noun-Noun Compoundingby Ray Jackendoff

A Natural History of Infixationby Alan Chi Lun Yu

The Oxford Handbook of Linguistic Interfacesedited by Gillian Ramchand and Charles Reiss

[ published in association with the series]

Aspects of the Theoryof Clitics

S T E P H E N R . A N D E R S O N

1

3Great 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 YorkAuckland Cape Town Dar es Salaam Hong Kong KarachiKuala Lumpur Madrid Melbourne Mexico City NairobiNew Delhi Shanghai Taipei Toronto

With offices inArgentina Austria Brazil Chile Czech Republic France GreeceGuatemala Hungary Italy Japan Poland Portugal SingaporeSouth Korea Switzerland Thailand Turkey Ukraine Vietnam

Oxford is a registered trade mark of Oxford University Pressin the UK and in certain other countries

Published in the United Statesby Oxford University Press Inc., New York

© Stephen Anderson 2005

The moral rights of the author have been assertedDatabase 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 appropriatereprographics rights organization. Enquiries concerning reproductionoutside 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 coverand you must impose this same condition on any acquirer

British Library Cataloguing in Publication DataData available

Library of Congress Cataloguing in Publication DataData available

Typeset by SPI Publisher Services, Pondicherry, IndiaPrinted in Great Britain on acid-free paper by Biddles Ltd., King’s Lynn

ISBN 0–19–927990–x 978–0–19–927990–6ISBN 0–19–927991–8 978–0–19–927991–3 (Pbk.)

10 9 8 7 6 5 4 3 2 1

Contents

General Preface vii

Acknowledgements viii

1 Introduction 1

2 What is a Clitic? 9

2.1 Words, Clitics, and Affixes 9

2.2 Case I: An Introduction to Kwakw’ala Clitics 14

2.3 Dimensions of Clitic-hood 22

2.4 Case II: How “Simple” are English Auxiliary Clitics? 24

2.5 Special Clitics 31

2.6 Clitics and Affixes 33

3 The Phonology of Cliticization 37

3.1 Prosodic Structure 37

3.2 Dimensions of Phonological Cliticization 44

3.3 Prosodic Structure and Syntactic Structure 55

3.4 Phonological Clitics and Cliticization in English 64

4 Special Clitics and their Grammar 75

4.1 Phenomenology 76

4.2 Special Clitics as the Morphology of Phrases 82

4.3 Some Examples 89

5 Theories of Special Clitics 107

5.1 The Nature of the Problem 108

5.2 Syntactic Theories of Clitic Placement 116

6 An Optimal Theory of Clitic Positioning 127

6.1 Special Clitics as Phrasal Morphology 127

6.2 Second Position: Anchors and Domains 142

6.3 “Endoclitics” 152

6.4 Tagalog Second-Position Clitics 165

vi Contents

7 Verb Second as Alignment 177

7.1 Verb Movement and Second Position 178

7.2 Icelandic 183

7.3 Kashmiri 187

7.4 Breton 193

7.5 Surmiran Rumantsch 204

7.6 Conclusions 224

8 Pronominal Clitics 227

8.1 Pronouns and Agreement 228

8.2 Clitics, Agreement, and Doubling 239

8.3 Clitic Climbing 245

8.4 Subject Clitics 249

9 Clause Structure and the Grammar of Incorporation 257

9.1 Introduction: Two Approaches to Noun Incorporation 258

9.2 Fleshing Out the Syntactic and Lexical Accounts 264

9.3 Noun Incorporation: Syntax or Lexicon? 275

9.4 Denominal Verb Formation in West Greenlandic 281

9.5 Conclusion 286

References 289

Index of Subjects 303

Index of Names 311

Index of Languages 315

General Preface

The theoretical focus of this series is on the interfaces between subcomponentsof the human grammatical system and the closely related area of the interfacesbetween the different subdisciplines of linguistics. The notion of SinterfaceŠhas become central in grammatical theory (for instance, in ChomskyŠs recentMinimalist Program) and in linguistic practice: work on the interfaces betweensyntax and semantics, syntax and morphology, phonology and phonetics etc.has led to a deeper understanding of particular linguistic phenomena and ofthe architecture of the linguistic component of the mind/brain.

The series covers interfaces between core components of grammar, includ-ing syntax/morphology, syntax/semantics, syntax/phonology, syntax/ pragmat-ics, morphology/phonology, phonology/phonetics, phonetics/speech process-ing, semantics/pragmatics, intonation/discourse structure as well as issues inthe way that the systems of grammar involving these interface areas are ac-quired and deployed in use (including language acquisition, language dysfunc-tion, and language processing). It demonstrates, we hope, that proper under-standings of particular linguistic phenomena, languages, language groups, orinter-language variations all require reference to interfaces.

The series is open to work by linguists of all theoretical persuasions andschools of thought. A main requirement is that authors should write so as tounderstood by colleagues in related subfields of linguistics and by scholars incognate disciplines.

Clitics appear to live at the interfaces between morphology, syntax andprosody. In our eleventh volume, Stephen Anderson addresses the theory ofcliticization, and argues that the properties of simple clitics are best capturedin terms of independently motivated prosodic structures, couched within anOT framework. He then shows that special clitics can be understood as beingthe phrasal analogues of morphological affixation, and that an OT account ofthese can be naturally extended to V2 constructions and pronominal clitics,with important implications for clause structure and the plausibility of headmovement.

David AdgerHagit Borer

Acknowledgements

The work reported here has benefited from the generous support over manyyears of the National Science Foundation, to whom I have been promising thismonograph for all too long. First and foremost, therefore, I would like toacknowledge the support of awards SBR 95--14682 and BCS 98--76456 from theNSF to Yale University. I am extremely grateful for the confidence the NSF, andespecially Paul Chapin as director of the Linguistics Program, has shown in mywork.

A language that figures prominently in the present book is Kwakw’ala,on which my research over several years in the 1970s was also supported bythe NSF, under award BNS 78--15395 to UCLA, as well as by a grant from theWenner-Gren Foundation for Anthropological Research. My study of Kwakw’alawas made possible by the assistance of Sam Henderson, Daisy Moon, and JimHenderson 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 mate-rials of Franz Boas in the library of the American Philosophical Society, whichI 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 sup-ported by NSF award BCS 98--76456 and by grants from the Yale UniversitySocial Science Research Fund, and continues under NSF award BCS 04--18410to Yale University. I am grateful to Petra Uffer for many hours of patient in-struction in Surmiran, as well as to Ursus Baltermia and my other friends andconsultants in Salouf and Savognin.

It is a cliché of prefaces such as this one, that too many people have pro-vided me with help, advice, suggestions, and corrections over the years in pur-suing this research for it to be possible to enumerate them all without causingthe reader’s eyes to glaze over—poor thanks, indeed. I would, however, liketo thank the students who have participated in my seminars on clitics at Yaleover a number of years, including especially Lizanne Kaiser and others whosework appears in Kaiser (1997). I would also like to acknowledge the help ofJulie Legate, who helped to keep me a bit more honest over the course of apresentation of much of this material in the Fall of 2003 at Yale, and who hasprovided numerous comments and suggestions on the manuscript. Additionalcomments and assistance on portions of the manuscript from Geert Booij,

Acknowledgements ix

Luigi Burzio, Andrew Carstairs-McCarthy, Sandra Chung, Jerry Sadock, andIda Toivonen have been extremely valuable. Obviously, none of these peo-ple 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 generoussupport of a sabbatical leave from Yale. During that time, I have been a VisitingScholar in the Department of Linguistics, School of Classics and Linguistics, atthe University of Canterbury in Christchurch, New Zealand. I am extremelygrateful to the University and the Department for their hospitality and for thegenerosity with which they have made their facilities available to me, providingme with a delightful atmosphere in which to carry out this work. Particularlypleasant have been frequent conversations with Andrew Carstairs-McCarthyboth about matters treated here and about more general issues in morphologyand phonology.

I must also single out Paul Smolensky. Paul introduced me to the meritsof an Optimality Theoretic approach to clitic phenomena in the course of aconversation years ago when I was seeking exactly the opposite: namely, a goodargument for ignoring OT in favor of the rule-based analysis I had offered inearlier 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 withmonographs, specialized collections, articles, notes, and other research deal-ing with the properties of clitics—indeed, entire books (such as Nevis, Joseph,Wanner, and Zwicky 1994) are devoted to nothing but lists of references rel-evant to the study of clitics. And yet a search of several standard online dic-tionaries, including Webster’s Unabridged, turns up no mention of the word.The Oxford English Dictionary first notes it in 1946, and defines it thusly: “Anenclitic or proclitic.” Not a lot of immediate help there, but at least we havea link to a much older tradition, for the notions of enclitic and proclitic arethoroughly grounded in grammatical description dating back at least to theseventeenth century.

The OED’s entry is supported by a quotation from Nida (1949: vii): “Manylanguages have elements that (1) combine phonologically with words withwhich they do not form morphological constructions, and (2) do not con-stitute derivational or inflectional formatives . . . The term ‘clitics’ may . . . beemployed in this general meaning.” Etymologically, clitic is from Greek klinein‘to lean,’ and refers to this tendency of clitics to depend on otherwise unrelatedmaterial.

Our understanding thus far is essentially negative: a clitic is somethingthat is not integrated into the sentence in the way “normal” words are, and/ornot integrated into words in the way affixes are. But despite its negative char-acter, this is the basis not just of one but of two nearly independent researchtraditions.

One of these, the older, is associated in linguistics primarily with tradi-tional grammarians and Indo-Europeanists. This is essentially a phonologicalunderstanding, on which a (pro- or en-)clitic is a stressless “little” word thatlacks independent accent, and that (as a result) depends prosodically on anadjacent word. The OED defines enclitic, for instance (following Liddell andScott’s 1843 Greek-English Lexicon) as a word that “ ‘leans its accent on the pre-ceding 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 ac-cent to be laid on the last syllable of the word which they follow. Hence appliedto the analogous Latin particles -que, -ve, -ne, etc., and in mod[ern] use (withextension of sense) to those unemphatic words in other lang[uage]s that aretreated 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 andproclitics. Such an interpretation is not at all controverted by the point ofthat paper, which was the further claim that clitics (defined in this way) inthe ancient Indo-European languages—and presumably also in Proto Indo-European—may huddle together in some particular location (in that case, in“second position,” after the first word of the clause). For example, in theHomeric 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 thesentence.1

(1.1) polees=te=min eresanto hippees phoreeinmany-and-it prayed riders carryAnd many riders prayed to carry it (Iliad 4.143)

For Wackernagel, the occurrence of these words in this position followed di-rectly 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 phonologicalone. Clitics are little words that lack accent of their own.

Modern syntacticians, on the other hand, have their own distinct ideasabout the notion of “clitics.” For them, the study of clitics is essentially thestudy of a particular class of pronominal elements (and others that behave likethem), most commonly the Romance (conjunct) pronouns. This conception isparticularly associated with a rich tradition of syntactic research originating inthe work of Richard Kayne. In Kayne (1975) a number of distinctive propertiesare identified with the relevant subset of pronominals, including not only theirdistinctive position immediately preceding the verb (as opposed to other nom-inal expressions), but also their inability to be modified or conjoined. Theseproperties go together with certain phonological and semantic idiosyncrasies,and appear to define a category of grammatical elements. For Kayne and thosewho have continued this line of research, the description of clitics is primarilya 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 anythingabout the specific phonological or syntactic consequences of such a relation, but only to highlightthe items as clitics.

Introduction 3

That notion is extremely widespread. For example, when we consult theindex of a recent reference grammar of Catalan (Wheeler, Yates, and Dols 1999)for “clitics,” we find “see pronominal clitics.” This is not because Catalan hasno other clitic elements of interest (in Chapter 3, we will see a class of unac-cented prepositions in this language that have properties quite worth exam-ining), but rather because syntacticians are particularly interested in elementssuch as those underlined in example (1.2).

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

For researchers of this persuasion, what is remarkable about the under-lined elements in (1.2) is not their lack of accent (a phonological property), butrather the fact that (in Catalan, as in most Romance languages) they all pileup immediately before the finite verb of the clause—a position in which cor-responding non-clitic elements would not appear. This sense of “clitic” is thusa (morpho)syntactic one.

My goal in this book is to explore both of these senses of clitic, their in-terrelations, and the implications of their analysis for the theory of grammarmore generally. The general perspective will be one that grows out of the the-ory of A-Morphous Morphology (Anderson 1992), where clitics are regarded asthe phrasal analog of (word-level) morphology. In more general terms, theo-retically satisfying accounts and explanations will be sought within a view ofgrammar based primarily on systems of constraints (a version of OptimalityTheory) rather than on rules. Much of the analysis on the morphosyntacticside is prefigured by that of Anderson (2000c), which is both extended hereand complemented by more attention to the phonological side of the problem.

The basic architecture of the book is determined by the distinction be-tween the two logically independent senses of the notion of “clitic” introducedabove. I begin in Chapter 2 by endeavoring to separate these by distinguishingsimple clitics—referring to a purely phonological dimension of clitic-hood—from special clitics in more or less the sense of Zwicky (1977). I consider twosystems that might be taken to illustrate simple clitics, whose behavior we canlater compare and contrast with that of morphosyntactically unusual specialclitics. The first of these is a set of clitic elements from the Wakashan languageKwakw’ala, whose properties (while initially remarkable in some respects) arerelatively straightforward.

The second case considered here yields a rather more complex story, how-ever. It has become fairly common to cite English contracted auxiliary formssuch as the ’s in Nixon’s the one as near canonical instances of simple clitics.


Syntacticians, on the other hand, have developed a rich literature on these el-ements: if they are really so simple, it is remarkable that so many people havereceived Ph.D.s for dissertations devoted (entirely or in large part) to a dis-cussion of intricate syntactic conditions on their occurrence. I explore thesematters with the goal of seeing what a phonologically based analysis must ac-count 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 phonol-ogy of the relevant elements does not, of course, actually provide in itself a the-ory of that dimension. In fact, the phonology of cliticization is closely boundup with the nature and status of the much broader theory of prosodic structurein language, and we can only understand phonological clitics by placing theirbehavior within this larger context. In Chapter 3, accordingly, I outline a viewof prosodic categories and their relations within the framework of OptimalityTheory, 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 Eng-lish reduced auxiliaries do indeed fall within a phonological account, and thusthat 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, Ithen move on in Chapter 4 to the morphosyntax of special clitics. Starting fromthe 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 andnot syntax furnishes the appropriate context for understanding them. Specialclitics, that is, constitute the morphology of phrases—a suggestion that hasoften been made and which I try here to substantiate.

Some morphology consists not of affixes but of other sorts of change inshape (“Non-concatenative” morphology). There is a case to be made that thekind of function filled by clitics is sometimes realized by changes in shape otherthan the addition of segmentable material to a phrase, and instances of such“Non-concatenative” clitics are proposed and analyzed. And just as word-levelmorphology can be divided between inflection and derivation, there seem tobe two corresponding classes of clitics: some that represent grammatical ma-terial, such as pronominal arguments; and some that represent more semanticcontent, such as discourse markers, various adverbials, etc. I discuss the extentto which this distinction among clitics is substantively parallel to that found inmorphology.

Among the various types of special clitics, by far the most interesting (inthe 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 areso important for the development of an adequate theory of special clitic posi-tioning. Proceeding from a fuller account of the Kwakw’ala clitics consideredearlier, I explore second-position clitics in a variety of languages. I consider arange of possible theories that have been proposed for the positioning of spe-cial clitics, and conclude that neither the syntax nor the phonology suffices todescribe the grammar of special clitics—a result that is of course entirely inaccord with my earlier proposal that the nature of these elements is essentiallymorphological.

I next proceed in Chapter 6 to elaborate a more precise account of thephrasal morphology that is responsible for the appearance of special clitics,within an Optimality Theoretic framework. A limited set of constraint typessuffice to describe these (and other possible special clitics) concisely and in-sightfully, as has also been argued in a series of papers by Géraldine Legendre inrecent years as well as in Anderson 2000c. The OT-based account of “Clitics asPhrasal Affixes” is compared with a variety of other theories: those postulatinga purely syntactic account of cliticization, and those locating the unusual prop-erties of special clitics in their phonology or in other properties of the interfacebetween syntax and other parts of grammar. The system of second-positionclitics in Tagalog provides a complex and nuanced example of a number of theproperties discussed to this point.

In Chapter 7, I turn from the analysis of clitics per se to an area of syntaxthat has not generally been taken within the generative tradition to be closelyrelated. Here I pose the question of whether or not it was correct for JakobWackernagel, whose classic discussion of second-position clitics in the earliestIndo-European languages has been so often cited (if less often read), to suggesta connection between these clitics and another set of second-position phenom-ena, the verb-second regularities of Modern German and a number of otherlanguages. I conclude that while Wackernagel’s own notion of the explanatoryconnection was undoubtedly incorrect, there is indeed a deep link, and themorphosyntactic apparatus used to describe second-position clitics can pro-vide us with an account of verb second in German, Icelandic, Breton, and otherlanguages as well. In the process, we learn some things about the way Optimal-ity Theoretic mechanisms play out in the domain of syntax, where they havebeen 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 classof clitics, those traditionally analyzed as pronominals. I begin by examiningthe nature of (predicate-argument) agreement, and comparing it with well-known phenomena arising in the analysis of special clitics. This requires an


elaboration of the analysis of the Morphosyntactic Representations of cate-gories, in order to account for phenomena such as (the presence versus ab-sence versus optionality of) clitic doubling, clitic climbing, and the like. Whilethe bulk of the literature devoted to pronominal clitics focuses on object cli-tics, some languages (including several spoken in northern Italy and in nearbyareas of Switzerland) also have special clitics referring to subjects. I exploretheir properties, including those of Surmiran and a range of “northern Italiandialects.” I discuss the significance of the morphological approach to specialclitics for the syntax of functional categories and the proposal that all suchcategories constitute syntactically autonomous heads (each with its own pro-jection) in syntactic representation.

The account of clause structure that results from viewing clitics in the wayadvocated here has a variety of other wider consequences. One of these involvesthe plausibility of an approach to incorporation phenomena based on syntacticmovement, as developed for instance by Sadock, Baker, and others. If syntacticmovement is not involved in incorporation constructions, as I argue, then thefoundation for the whole apparatus of Head Movement collapses, and withit much of the plausibility of the elaborately articulated functional structureassumed by syntacticians since the late 1980s. This is a result which is quite inharmony with the picture of clause structure developed in the present work,which assumes that much functional content is present in the form of featureson a limited number of structural categories, rather than as a set of functionalheads in their own right. In Chapter 9, I explore these matters, and attempt todraw some general morals.

From various remarks above it will perhaps be clear that the view of syn-tactic structure represented in this book does not line up completely withany of the established positions that have dominated discussion over the pastdecade or so. Apart from occasional turns of phrase, I do not adopt the spe-cific assumptions that have emerged within the Minimalist Program for syntax,though I think that most of what I say could be transposed into that frameworkwith little substantive alteration. I assume an overall picture involving base-generated structures and operations of displacement relating these to surfaceform, subject to conditions including those of Binding Theory and the like in away reminiscent of much work in the framework of Government and Bindingor Principles and Parameters.

I do not, however, follow the common development of that framework toinclude an increasing proliferation of structure based on functional categories.Much of this functional content is associated here with an elaboration of thefeature structure of categories drawn from a somewhat more limited set, withconsiderably 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 ar-ticulation, than those to which students of generative grammar are typicallyintroduced today, and involve much less movement within functional struc-ture. In part, these differences reflect the fact that for my purposes, much ofthis functional structure is simply not relevant to the points at issue, and thuscan usefully be abbreviated. It should also be clear, however, that there are realdifferences of principle involved.

I assume (and attempt to argue for) the relevance of a constraint-basedsystem of an Optimality Theoretic sort in the development of morphosyntacticstructure. This is most prominent in relation to the principles that introducespecial clitic elements into sentences, but also plays a role elsewhere (e.g., inproducing “verb-second” patterns). This view is of a piece with the OptimalityTheoretic interpretation of principles of prosodic structure which forms thebasis of the account of phonological clitics offered in Chapter 3.

This is probably the longest work published to date based on assump-tions from Optimality Theory but which does not include a single constrainttableau. My principal excuse for that is the fact that had I included tableauxto 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 ofthese displays for confirming the correctness of complex constraint rankings,I am not convinced of their more general perspicuity. Most of the constraintinteractions proposed here are quite straightforward, and I hope they can beappreciated on the basis of the discussion. In any event, I leave the constructionof formal tableaux as an exercise for the reader.

I cannot hope to argue for all of the details of my (morpho)syntactic as-sumptions in this book, but I hope that I have made a case for some of the morenoticeable deviations from current practice, and that in other instances I haveat least made clear what my assumptions are so that others can see whether thedifferences matter to the real arguments.

2

What is a Clitic?

The modern study of clitics within generative grammar can reasonably be seenas beginning with the survey provided by Zwicky (1977). A number of ear-lier papers had dealt with the special properties of these elements, especiallyin terms of their syntax; and after the fashion of the time, Zwicky’s paper wasnever formally published, but only distributed in mimeographed form. Thedistinctions proposed there, however, and the observations on which they werebased, established a research agenda that has dominated discussion of cliticsevery since.

2.1 Words, Clitics, and Affixes

Zwicky’s classic paper starts from the question of how we distinguish wordsfrom affixes, enumerating a number of properties in terms of which thesediffer. Like many such lists, this one contains some properties that might beregarded as definitional, mixed in with others that are simply common symp-toms of an element’s status.

Ordering: Affix order within the word is quite rigid, while word orderwithin phrases can vary. The degree of variation differs from language to lan-guage, of course, but virtually all languages allow some alternative word orderscorresponding to the same essential meaning. In contrast, any variation wefind in the order of word-internal affixes is virtually always correlated with adifference in sense.

Internal sandhi: Different phonological adjustments may apply withinversus across words. The distinction of (word-)internal sandhi from externalsandhi goes back (as the etymology of the word sandhi suggests) to the ancientSanskrit grammarians. It corresponds in large part to the difference betweenLexical and Post-lexical phonology in more modern terms (cf. Kaisse and Shaw1985).

Binding: This is Zwicky’s term for the fact that some morphological ele-ments can appear alone, while others only occur in combination. The latter(“Bound Morphemes”) are affixes, while words are free.


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 supportthe claim that an element in Madurese which occurs between two identifiableaffixes in a form whose base can be independently established must itself be anaffix.

Rule immunity: Syntactic rules, such as deletion under identity, do notaffect affixes, since these are proper parts of words. Seen as a criterion fordistinguishing words from affixes, this is clearly based on strong Lexicalist as-sumptions 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, wemight be content, but of course that is not true, and Zwicky identifies threecommon types of “mixed” case. These, of course, are three varieties of thingswe might call clitics. On this account, then, a clitic is a linguistic element whichthe tests just given do not classify unambiguously as being either a word or anaffix. 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 vari-ants of free forms (similar, that is, in sound and meaning to some non-cliticword); and which display “special” syntax. This might mean simply “different”syntax from the corresponding free form, as French Je la vois ‘I see her’ versusJe vois la femme de mes rêves ‘I see the woman of my dreams.’ In this case (thesyntactician’s prototypical instance of a clitic), the pronoun precedes the finiteverb, while a corresponding non-pronoun is completely excluded from thatposition, and instead appears post-verbally (where the pronoun, in its turn,is excluded). More dramatically, the syntax of special clitics may involve theirappearing in a position which is not in general accessible per se to rules of thesyntax at all—e.g., second position, interpreted as immediately following thefirst full phonological word of a phrase or clause.

Simple clitics are unaccented variants of free morphemes, which may bephonologically reduced and subordinated to a neighboring word. In termsof their syntax, though, they appear in the same position as one that can beoccupied by the corresponding free word.

Finally, bound words are forms that are always unaccented and phonolog-ically subordinated to a neighboring word. They are often syntactically associ-ated with a whole phrase, while being phonologically associated with a singleword in (or adjacent to) it.

What is a Clitic? 11

Examining this classification, we can see that it has some slightly peculiarproperties. For example, the difference between special and simple clitics, onthe one hand, and bound words, on the other, is based on whether or not arelated non-clitic (free) form exists. But what reason do we have to believe thatthis is really significant? That is, assuming that a cluster of properties (not justa 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 dependingon whether or not the form in question alternates with another (synonymousbut free) form of the same lexical item.

Consider the English pronouns which can appear either reduced (Ifhe comes near me, I’ll hit’m [hiRm

"]) or unreduced (she wanted to meet him

[mijthim]). The former is identified as a clitic by Zwicky’s criteria, in virtueof the latter’s existence. The assumption made at the outset of Zwicky’s pa-per, motivating some of his decisions about terminology and categories, is thatclitic and non-clitic in cases like that of English object pronouns are related byphonological rule. When we look more closely, though, this turns out to befairly implausible, at least in the general case.

Zwicky and Pullum’s (1983) paper on reduced forms of negation in Eng-lish, for instance (which I will discuss in more detail below), observes that theclitic forms of English auxiliaries like will and would suggest processes that areotherwise quite unprecedented in English phonology. If these clitics are linkedto corresponding non-clitics, then this is almost certainly a fact about the rel-evant (phonologically complex) lexical entries, not a matter of phonologicallypredictable variants of simple forms. But why should the structural charac-ter of a linguistic form as a “clitic” depend crucially on the contingent matterof 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 lan-guage Mixtec, and makes this point in greater detail. She shows that in thatlanguage there are phonological rules deleting glottal stops, coalescing sequen-ces of identical vowels, and deleting word-initial syllables, in rapid speech.These processes have the effect of reducing disyllabic forms to monosyllablesunder appropriate conditions, and some forms show an alternation that isdriven by these rules. We can see that the variation between disyllabic andmonosyllabic forms is a consequence of these rules, because in slower speechthe lost syllables are “restored.”

But there are other forms for which the ‘short’ form does not correspondto any longer slow-speech form, and in some other cases, the alternation is notsomething that can be derived from the (independently motivated) phonology.


For instance, the third person masculine object pronoun is /càà/ in its fullform, but /–re/ when reduced. The full form of the first person familiar pro-noun 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 anobject is the (phonologically predictable) shape /-ru/. These diverse relationsshow 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 oftenlexical alternates. For instance, English would has two lexical shapes: the cliticform /–d/ and the non-clitic /wUd/. The Mixtec third-person masculine objecthas the forms /càà/ and /–re/, etc. In each of these cases, the forms in thelexical entry must be accompanied by some statement of their distribution.But in that case, the difference between clitic forms and bound words losesmuch of its theoretical interest: a bound word is just an unaccented lexicalitem that does not happen to have any non-clitic lexical alternant—surely nota fundamental difference of status.

What about cases where there really is some motivated phonological re-duction that relates the clitic to the non-clitic form? This is the case withmuch (though not all) of the Mixtec lexicon, for example, and perhaps it isalso true for English object pronouns. In that case, we want to say the lex-ical entry provides only one lexical shape, and the phonology has an effectthat yields a reduced form (a clitic). Here it appears that we have somethingwhich is different both from pure non-clitics and from lexically distributedclitics.

But what is it that we have in such cases? Everything depends on what wewant to say the nature of a “clitic” is, and in particular, whether the existenceof 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 surelyquite anomalous to suggest that a phonological rule could convert somethinginto 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 propertyshared by most (though as we will see, not all) affixes; and (b) phonologicalsubordination to another word. In fact, as soon as there came to be a realtheory of prosodic structure within which to articulate the problem (well af-ter the appearance of Zwicky’s paper), it became possible to give a coherentdefinition of the phonological property that makes something a clitic withouttreating “clitic” as a lexical or grammatical category. The necessary framework,which we will discuss in more detail in Chapter 3, is grounded in discussionsof the so-called “prosodic hierarchy,” for which a classic reference point is thepresentation in Selkirk (1984) or Nespor and Vogel (1986).


A first approximation to a characterization of clitics, along lines we willamplify later, is as follows. Let us say that the phonological substance of a wordis organized into a relatively high-level prosodic domain, the “prosodic word.”A constituent of this type is composed of feet, which in turn are composedof syllables, composed of segments, linked to features. Affixes, in contrast tolexical words, consist of segments, or syllables, or even feet; but these are not(generally) organized into a prosodic word. Prosodic words, in turn, are theconstituents of phonological phrases, etc. When phonological material doesnot have enough prosodic structure to be integrated into the prosodic struc-ture of the whole utterance on its own (by virtue of not being organized intoa prosodic word), it must be dependent on some adjacent material that canprovide the necessary bridge between lower- and higher-level prosodic cate-gories. This sort of incorporation into an adjacent word is just the behavior weassociate with clitics (in the phonological sense), which we can thus proposeto 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 integratedinto the prosodic hierarchy of syllable, foot, word, phrase, etc. As a result,where a potential utterance contains some ‘deficient’ material, some princi-ple of adjunction has to operate within the phonology to incorporate it intothe structure if it is to be pronounced (and the overall structure is to be well-formed). This usually happens by adjunction into an adjacent constituent ofthe 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 possiblelevel of structure. I will refer to whatever principle(s) a language may havefor prosodic incorporation of such stray material as its rule(s) of Stray Ad-junction.1 Let us call the adjacent element that serves as the target of thisadjunction the clitic’s host. This operation, which I will treat in more detail inChapter (3), is the phonological form of cliticization.2

By virtue of not projecting structure at the level of the phonologicalword—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. Phono-logical 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 con-straints are the appropriate way to think about Stray Adjunction, or indeed whether Stray Ad-junction is always “adjunction” in formal terms. We simply need a convenient term for the wayin which languages deal with prosodically deficient material.

2 The proposal that integration of otherwise deficient material into prosodic structure is whatis involved in phonological cliticization has also been made and defended by other authors, in-cluding Lahiri, Jongman, and Sereno (1990) and Booij (1996).


domain as adjacent material to which they are adjoined, though whether “Lex-ical” or “Post-lexical” phonology will apply at the boundary between cliticsand their hosts, whether the presence of a clitic can affect the location of stresswithin the host, and other phonological matters will depend on the intrica-cies of the precise structure which results from the adjunction, and also on thedetails of the phonologies of particular languages. But this much gives us anoverall framework within which we can hope to explore the phonological di-mension of cliticization, leaving the morphosyntactic peculiarities that are im-portant 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 Kwakw’ala Clitics

As an example of a set of clitics whose distinctive properties arise as a conse-quence of their phonology, I look first at the determiner and pronominal cli-tics of Kwakw’ala. Although the syntax of these elements is not unusual, theirphonology has consequences that are quite striking at first glance, and indeedthere 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, butmy focus here is on elements whose behavior derives from the fact that they areprosodically deficient in the sense of the preceding section.

The name Kwakw’ala is commonly used to designate a number of distinctbut related dialects spoken by a number of distinguishable tribal groups onthe northeast coast of Vancouver Island, British Columbia, and in areas of theadjacent mainland (of which the Kwaguì are one, but only one). The areain which Kwakw’ala is spoken is identified as “Kwakwaka’wakw” on the mapin Figure 2.1, following current local usage. It is a member of the NorthernWakashan family of languages, related to the other Northern Wakashan lan-guages Oowikyala (“Owikeno”), Heiltsuk (Bella Bella), and Haisla; and moredistantly, to the languages of the Southern Wakashan family (“Nootka,” “Niti-naht,” and Makah). The time depth of the entire Wakashan family is probablysomething like 3000 years, and that of its northern branch perhaps 2000 (Bachand Howe 2002).

Kwakw’ala is better known as “Kwakiutl”3 from the name Franz Boas usedfor 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 actuallyintended to represent the name of the people (the [kwag,uì]), as opposed to the name of theirlanguage, ([kwakw’ala]). Although this writing system fails to represent several major phonolog-ical 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 phonologyof Kwakw’ala inevitably involve a fair number of diacritics and unusual characters, a feature thatspeakers dislike since it seems to make their language look strange and exotic.


Figure 2.1. Some First Nations languages of British Columbia. “Kwakwaka’wakw” =

Kwakw’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 ofBritish Columbia. Copyright 1994.

(Boas 1911, 1947) and many volumes of texts, prepared over a period of manyyears. The description here is based on my own fieldwork during several sum-mers in the 1970s, but I have drawn most of my examples and cited forms fromBoas’s work, to facilitate comparison. I have, however, modified Boas’s tran-scriptions to reflect my own understanding of Kwakw’ala phonological struc-ture. As noted in Anderson (1985b), Boas employed representations that areextremely close to the surface phonetics, in ways that sometimes obscure theirstructural interpretation.

To appreciate the role of the clitics in Kwakw’ala morphosyntax, it is nec-essary to introduce some of the basic features of the language’s syntax (cf. An-derson (1984) and references cited there for further details). The basic clausestructure is quite rigidly VSO, and conforms to the schema in (2.1).

(2.1) V—Sbj(—x. -Obj)(—s-Obj)(—PP*)


The verb always comes first in the clause, although the sentence-initialelement is often a semantically empty auxiliary verb, with the lexical verb fol-lowing. The verb is then immediately followed by the subject if this is overt. Asdiscussed in Anderson (1984), a sentence may contain more than one verb.In that case, the subject may immediately follow either the sentence-initialverb—typically, a semantically empty auxiliary—or the last verb in the se-quence. Regardless of the position of the subject, remaining arguments andadjuncts follow all verbs, in the sequence to be described below. I abstract awayfrom this limited optionality in the position of the subject in the discussionbelow.

The subject is followed in its turn by one or both of two kinds of ob-ject argument phrase. These are called “objects” and “instrumentals” by Boas,though their precise semantics do not always correspond to these descriptiveterms, and they should be regarded as formal categories. “Objects” are markedwith 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 isgiven in (2.2).

(2.2) y@lkw@mas=ida b@gwan@ma=x. -a ’watsi=s-a gwax.ňux.w

cause hurt-dem man-obj-dem dog-inst-dem stickThe 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@gwan@ma=x. -a g@nan@ma=s

struck-dem man-obj-dem child-instrThe man struck the child with it

b. x.w@sPid=ida b@gwan@ma=q

struck-dem man-objThe man struck him

c. x.w@sPida=∅=q=s

struck-he-obj-instHe struck him with it

Pronominal forms also exist for the subject, but these behave somewhat dif-ferently from the non-subject pronominals. Since the third-person form istypically null phonologically, I will ignore the subject forms for the present.

In the examples of (2.2) and (2.3), we have several representative in-stances of simple clitics. Note first that each full nominal expression is pre-ceded by a determiner element: -ida, -x. a, -sa, etc. Although this provides casemarking and deictic information about the nominal that follows, it attaches


phonologically to the preceding word, regardless of that word’s syntactic af-filiation. This prosodic attachment shows us that these elements are clitics,and the way it works shows us that Stray Adjunction in Kwakw’ala operatesrigidly to adjoin prosodically weak material to the left, despite the syntacticallycounter-intuitive nature of the result.

As illustrated in (2.3), any of the overt (third-person) nominal argumentsmay be substituted by a pronominal form. These are also (simple) clitics, andattach to the preceding word. Consistent with the claim that these are ‘simpleclitics,’ they appear in the position of the corresponding full nominal argumentexpression. And like the determiners, they attach to their left.

The core of the clause can be followed by any number of adjunct expres-sions. These take the form of prepositional phrases, although it is worth not-ing that the language has only a very small number of prepositions. Most PPsemploy the same preposition, la4 with the semantics of the adjunct relationsupplied by affixes within the verb.

As an alternative to a full case-marked nominal expression in argumentposition, 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 kwix.ayu-kaux-fut-I strike-you with-dem club-demI’ll strike you with this club

b. noìa-∅-s-is kwix. ayu lax. -is ts’a’yathreaten-he-inst-his club to-his younger brotherHe threatened his younger brother with his club

c. la-’mis-@s ńiqala-ň@-s aňa’n@m gax-@naux-conn-you name-fut-inst wolf to-meAnd 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 referencemust be supplied by an adjunct (such as gax en ‘me’ in (2.4c)). Further, there areno non-clitic pronouns that occupy argument positions. The only full wordswith “pronominal” reference are predicates (e.g., nugwaP em ‘it is I’). These areentirely parallel to the language’s ‘wh-words’, which similarly are predicates andnot nominals (e.g., engwa- ‘(to be) who?’, ’mas-‘(to be) what?’).

To some extent, Adjuncts alternate with argument nominals, as shownin (2.5).

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


(2.5) a. ’nik@-∅-x. -is x.w@nukw

say-he-obj-his childHe said to his child

b. ?’niki-∅ lax. -is x.w@nukw

say-he to-his childHe said to his child

The same argument cannot be represented simultaneously by an adjunctand 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@-∅-qsay-he-objHe says to him

b. *nik@-∅-q lax. -is x.w@nukw

say-he-obj to-his childHe says to his child

The basic set of clitic pronouns is given in Figure 2.2.

Person Subject Object Instrumental1sg -@n(ň) — -@n(ň)

1Incl -@nts — -@nts1Excl -@nux.

w — [email protected]

2nd -@s -uň -us3rd ∅ -q -s

Figure 2.2. Clitic pronouns in Kwakw’ala

We see that argument positions can be filled by any of the following: a(case-marked) full nominal expression; a (phonologically) weak pronominalin the same position as a full nominal, attaching phonologically to its left; orelse by an empty pro, which I assume to occupy argument positions when thereferent of the corresponding argument is specified by an Adjunct. In this, Ifollow the line argued for by Baker (1995) with respect to all argument posi-tions in Mohawk. I will return to this analysis in later chapters to explore itsconsequences in more detail.

Notice that the clitic elements introduced above really do form part ofa phonological word with the preceding word. This is confirmed by thephonology: at the boundary between a clitic and its host, we get word-internalphonology rather than the phonology which otherwise occurs across word


boundaries. One small example of this, cited by Boas (1947), is the rule ofconsonantal epenthesis given in (2.7), a rule which does not apply to /ìs/ se-quences formed across the boundaries between words.

(2.7) a. ∅ →t/ì—s

b. dug.w eìts < dug.

w eł=s ‘it was seen by him’

Even more diagnostic in this regard is the placement of stress. Stress inKwakw’ala appears on the first full vowel (or non-glottal syllabic resonant) ofthe word, where the relevant notion of “word” includes a host and any fol-lowing clitics. Clitics thus display their dependent status by not initiating newstress domains.

We could express the phonological behavior of clitics in Kwakw’ala infor-mally by the rule in (2.8).

(2.8) Stray Adjunction (Kwakw’ala): Adjoin stray material to a PWord on itsleft.

In Chapter 3, I will refine this description somewhat by recasting it withina framework based on constraints and situating it within a broader account ofprosodic categories in general. There is actually a good deal more to say aboutthe Kwakw’ala determiner system, and I will explore it further in Chapter 4.

The clitic nature of determiners in Kwakw’ala is not simply a matter of theway in which they are pronounced. It has significant implications for the rangeof syntactic possibilities in the language. These follow from the fact that theleft edge of a nominal expression in Kwakw’ala is always marked with a cliticdeterminer, and this must attach to its left as a consequence of (2.8). Fromthis, it results that a nominal expression cannot be pronounced by itself, sinceit must always be preceded by a host for its initial clitic. When there is nopreceding constituent, the place of such a host must be supplied by an initialprothetic stem, as in (2.9).

(2.9) *(yi)=x. ux. da g enan em

(∅)-dem child

(it’s) that child

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


We can compare this situation with that in closely related languages whoseclitic systems are different from that of Kwakw’ala. Both Haisla and Heilt-suk (Bella Bella) lack determiner clitics at the left edge of nominal expres-sions, 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 Kwakw’ala. The examplesin (2.10) are taken from Boas’s (1928) collection of bilingual texts, and provideboth Oowiky’ala and Kwakw’ala versions for comparison.

(2.10) a. (O) gi sukwa’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-hisThen shei took hisj,∗i arrow

In Heiltsuk (Bella Bella; cf. Rath (1981)), the determiner system is fairlyintricate and involves a number of component parts (like Kwakw’ala, as weshall see in Chapter 4). As the examples in (2.11) illustrate, however, none ofthese parts is a clitic appearing at the left edge of the nominal.

(2.11) a. p’ála wísm=á=x. i la uxwňiás=a=x. iwork 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 uxwňiás=a=x. iwork diligent-det1-conn man-det2 on roof–det1-det2

The diligent man worked on the roof

c. p’ála ’wála=ya=s p’ác’uá=s wísem=x. i lawork really-det1-conn diligent-conn man-det2 onuxwňiás=a=x. iroof–det1-det2

The really diligent man worked on the roof

Corresponding to this difference in the determiner systems amongWakashan languages, we find a syntactic difference: languages without leftedge clitic determiners also allow a construction with an initial nominalrepresenting a topic, though otherwise their word order is very similar to whatwe find in Kwakw’ala. The examples in (2.12) with initial topics are presented


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

(2.12) a. (gi) ’m@nuk pkw’ala(then) one-person say(Then) one man said . . .

b. h@lxPainoxw láoňde g.@n@m-xde=a=seKiller whale removed wife-former-det1-yourA killer whale took away your (removed from presence) wife

Haisla, like the other Northern Wakashan languages except Kwakw’ala, lacksdeterminer clitics at the left edges of nominals and allows initial topics as il-lustrated in (2.13), drawn from a published text (Lincoln, Rath, and Windsor(1986)).

(2.13) a. kwa’nalasgułd h’x.w h’a’maka mia-gila-su=si

yellow cedar try first fish-make-pass-3agtYellow cedar wood was the first thing he tried to turn into salmon

b. hnńm-’wsm sa’wati hs=qids Gala=yads bgwanm=x. i duarrow-still use by-3inv first-det1 people-det2 andłk’w=isbow-detThe arrow was still used by them, the first people, and the bow

Finally, we find a very similar situation in Ditidaht (“Nitinaht”), a south-ern Wakashan language (cf. Klokeid (1976)). In this language, as the examplesin (2.14) show, a determiner appears after the first word of the nominal andcase is indicated by a separate word (Poyoqw ‘acc’ in these sentences).

(2.14) a. tl’itchitl=ibt=PPa John bowatc Poyoqwshoot-past-decl John deer acc

John shot a deer

b. tl’itchitl=ibt=PPa John bowatc=PPaq Poyoqwshoot-past-decl John deer-det acc

John shot the deer

c. tl’itchitl=ibt=PPa John Pıx.=PPaq bowatc Poyoqwshoot-past-decl John big-det deer acc

John shot the big deer

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


(2.15) [ oyoqw=obt=PPa [ Pıx.=PPaq bowatc ] ] tl’itcitl Johnacc-past-decl big-det deer shoot John

The big deer, John shot (it)

The correlation between determiner systems and the possibility ofsentence-initial topics make it clear that the absence of this latter constructionin Kwakw’ala must be due to the impossibility of providing a host for thedeterminer clitics that appear at the left edges of nominals in this language.Although the distinctive properties of these elements are phonological innature, they have syntactic consequences.

I will return to the clitics of Kwakw’ala in Chapter 4, where additionalelements with the properties of “special” clitics will be discussed. There isconsiderably more to be said about the phonological dimension of cliticiza-tion before we move on to those syntactic matters, however. At this point Iwould like to return to the foundational issues of how to identify clitics, andhow to disentangle the diverse senses in which something can be thought ofmeaningfully as deserving that designation.

2.3 Dimensions of Clitic-hood

Let us return to the typology of clitic elements, as presented originally inZwicky (1977). On that picture, the elements Zwicky called simple clitics aremerely one (or more) of the alternative phonological forms of certain lexicalitems (in particular, the prosodically deficient realization(s) of the item inquestion). “Bound words” are the limiting case of this: lexical items thatonly have prosodically deficient phonological forms. And where a free formundergoes phonological reduction to become a clitic, what that means is thatthe 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 thatit comes under the same phonological regularities as forms that are lexicallydeficient.

On this account, there are three quite distinct categories of element thatdisplay the same prosodic behavior: bound words, which are uniformlydeficient in prosody; (simple) clitics, sensu stricto, the prosodically deficientalternants of certain free forms; and free forms which have undergonephonological reduction. But it is far from clear that this distinction is a mean-ingful one. For instance, we saw above that Kwakw’ala has no full forms withpronominal reference. Only the clitics of Figure 2.2 serve that purpose in thelanguage. Since these words have no lexical non-clitic alternants, and are notphonologically reduced from free forms, they are (in the classification underdiscussion) bound words. In the related language Heiltsuk (Bella Bella), how-ever, first- and second-person forms have both clitic and independent, full-word alternants, as in Figure 2.3 (drawn from Boas (1947:296) and Rath (1981)).


Person Independent form Subject Object Instrumental1st nugwa -nugwa, -@n -@nňa q@s-

1Incl nugw@nts -@nts -@nň@nts q@nts-1Excl nugw@ntkw -@ntkw -@nň@ntkw q@ntkw-2nd qsu -(@)s(u) -uň -us3rd — ∅ -q -s

Figure 2.3. Pronouns in Heiltsuk (Bella Bella)

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

(2.16) a. a@mbaya his núgwa /qs= a@mbayaahead inst me 1sg-inst aheadone ahead of me one ahead of me

b. a@mbaya his qsú /a@mbay =usahead inst you ahead 2-instone ahead of you one ahead of you

In the third person, however, only the clitic form (as in a embaya=s ‘oneahead of him’) is found. The first- and second-person clitic forms in Heiltsukare truly clitics in Zwicky’s classification, since they have full-form variants,but the third-person form is instead a “bound word.” Clearly this distinctiondoes not correspond to the one we want: rather, we want to classify all of theprosodically dependent forms (i.e., all of the pronominals in Kwakw’ala, andboth the proclitic and enclitic forms in Heiltsuk) as clitics, regardless of whetheror not they have non-clitic lexical alternants.

The most appropriate sense of clitic, then (at least in the present phono-logical context) seems to be one that is independent of the existence (or lackthereof) of non-clitic alternate forms of the item in question.

(2.17) Phonological Clitic: A linguistic element whose phonological form isdeficient 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 whoseonly claim to clitic-hood is phonological in the sense of (2.17), though I willusually use phonological clitic instead in this book, to highlight the nature ofthe elements under discussion.

Recall that classical usage (e.g., that of Wackernagel) equated clitic sta-tus with lack of independent accent, and that property follows directly for


elements that satisfy (2.17). Phonologists have assumed at least since the ap-pearance of Liberman and Prince (1977) that stress accent is definitionally aproperty of the metrical structure of Prosodic Words, and obviously an ele-ment that does not have such structure does not, by itself, project stress. Butthat analysis also allows us to accommodate one of the types of clitic thatZwicky (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 suchenclitics 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 seemsto be compromised by the fact that it has an accent.

The answer to this apparent conundrum is that, while it does not consti-tute a Prosodic Word in its own right, a clitic does (indeed, must) get incor-porated into the larger prosodic structure projected by its host. The relevantstress which appears on the clitic is assigned not to the clitic per se, but ratherto 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 Chap-ter 3. Assume for now, however, that an appropriate structure has already beenassigned 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 ad-ditional syllable is incorporated into the word (by whatever principle of StrayAdjunction applies in Modern Greek), as in [ðósemu] ‘give me!’ no new footis constructed on this added material, since feet in Greek must be minimallybisyllabic, and [mu]σ remains a syllable—part of the Prosodic Word (and thuspronounceable), 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 newfoot (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 projectsstress on its own: rather, it is part of a larger Prosodic Word, whose base is theclitics’ host [ðóse].5 There is thus no contradiction between the clitic status ofthe pronominal elements =mu and =to, on the one hand, and their manifes-tation 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 seenone set of “simple clitics,” in the Kwakw’ala pronouns and determiners dis-cussed 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.


we might be tempted to attribute distinctive syntax to a set of clitics, but whereit turns out that the relevant properties follow from their phonological charac-ter alone. What makes the Kwakw’ala elements of section 2.2 clitics is a purelyphonological property, but one with syntactic consequences. The need fora prosodic host entails the restriction that they cannot appear in construc-tions where they would be initial within a phrase, such as in the position of asentence-initial topic. Their syntax follows without additional stipulation fromtheir phonology.

Before proceeding to a more general theoretical framework for the phonol-ogy of cliticization, I look first at a somewhat more notorious example, wherethe “simplicity” of the designation “simple clitic” is rather more seriously inquestion. This is the case of the reduced forms of a number of auxiliary verbsin English. These have often been suggested to be simple clitics, but an enor-mous literature attests to the fact that at least superficially, they have distinctivesyntactic properties. We need to determine whether, as in the Kwakw’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 auxiliaryverbs, including is and has, can appear either in a full form or reduced—inthese 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 re-duced form for the full form of the auxiliary is not entirely free. Under somecircumstances, 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 (atHalloween).

While not all of the English auxiliary verbs have these reduced forms, quitea 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 formscan appear: that is, an unreduced form can always substitute for a reducedone, but not vice versa. The fact that the reduced auxiliaries appear in positionswhere the syntax would put corresponding non-clitic forms suggests that they


Full form Reducedis ’s

are ’ream ’mhas ’s

have ’vehad ’dwill ’ll

would ’d

Figure 2.4. English Reduced Auxiliaries

are simple rather than special clitics (in Zwicky’s terminology). On the otherhand, 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 thepossibility of these elements and also of the limitations on their appearance.

We also want to keep another consideration in mind. Whatever the cor-rect analysis turns out to be, it has to be one that children can plausibly acquirequite easily. A basic fact about this matter is the observation that transcripts ofchildren’s utterances contain essentially no instances of reduced forms wherethese are impossible in adult speech. Furthermore, in experiments designed tobias children toward producing inappropriate instances of the reduced forms,they still do not do so.6 This suggests that the correct account must be quitedeeply embedded in principles of Universal Grammar that plausibly governchildren’s grammars at all stages, rather than based on detailed language-particular stipulations.

Kaisse’s (1985) Analysis A convenient starting point is Kaisse’s (1985) exten-sive discussion of a range of cliticization phenomena in English, even thoughKaisse’s notion of “simple clitics” was evidently somewhat different from theone 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 ischaracteristic of simple clitics that their syntax has nothing special about it. Ifthese items arise through the intervention of specifically syntactic effects, theywould not be simple clitics in my sense.

Although she follows Zwicky (1977) in proceeding from a definition ofclitic that assumes a relation between reduced and full forms (which I sug-gested 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 ofthese claims.


essentially on this. She argues that another distinction is rather more impor-tant: among the items that have both a full and a phonologically reduced form,the difference is sometimes attributable to the phonology alone, and some-times idiosyncratic.

For example, don’t displays the shapes [dõ] and [don] which arise from or-dinary fast-speech reduction processes. Reduction of a final cluster /nt/ to thenasal alone, and further reduction of a final nasal to nasalization of a precedingvowel, are processes that apply rather generally in the post-lexical phonologyof English, especially in rapid speech. The special form [d@] (as in I don’t wantto, pronounced as [ad@"wan@]), though, appears to be a lexically idiosyncraticvariant of don’t, rather than being produced by productive processes. This issuggested 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 noform [w@].

Kaisse argues that all of the reduced auxiliaries and modals in Figure 2.4are lexically specified, and do not arise through the operation of phonologicalrules. Since English does not have productive phonological rules that could de-rive the observed reduced forms of is, are, am, has, have, had, will, would, thesevariants 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 ofthe phonology of cliticization. Given a lexically determined alternation, suchas we find for instance with has versus ’s, we can study the distribution of thereduced form (’s) and ask what the conditions are under which it can be used orinserted. 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 isinserted everywhere, and then the post-lexical phonology does what it likes onthe basis of derived structure. Because of this difference, when we evaluate theconditions on cliticization, we have to be clear in each case that we are actuallydealing with a clitic form, and not just with a phonetic variant derived by thephonology from a full form.

Kaisse’s view of what is going on in the reduced forms was premised onthe notion that the syntax is implicated. Her resolution of the problem was asfollows: under certain circumstances, one syntactic element may be cliticizedto another in the specific sense of being syntactically adjoined to it. Whenthat happens, the clitic variant of the lexical item is chosen, and otherwise, thenon-clitic variant is used. The basic operation involved in cliticization is thus asyntactic one, and the conditions on the appearance of phonological clitics areto be sought in the properties of the syntactic representation.


A representative derivation illustrating this view is given in (2.20).

(2.20) [Jack]NP

[is]aux

[leaving]VP

(output of syntax)

[[Jack] is]NP

[t] [leaving]VP

(cliticization)

[[Jack] is]NP

[t] ["leaving]VP

(sentence prosody)

[[Jack]z]NP

[t] ["leaving]VP

(allomorph selection)

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 isaltered when cliticization occurs. That is, why should we believe that syntacticrestructuring is involved in cliticization? One possible argument for that claimwould be provided if we found cases where the clitic moves as a unit withits host, and where the corresponding uncliticized sequence did not behavethat way. When we construct potential examples, however, that is not whatwe 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 relevantsyntax is completed. To say that, however, is to say that no possible evidencecould 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 auxiliaryplus 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, Zwickyand Pullum (1983) have shown convincingly that won’t in such sentences isactually a unitary inflected verbal form, and not a host+clitic sequence. Assuch there is no question that it is a single word,7 while will not is a sequenceof two separate categories, not a unit and hence not movable. Thus we have noevidence for the claim that cliticization in English has syntactic consequences.

7 Flagg (2003) argues that Zwicky and Pullum’s analysis is “internally inconsistent,” on thegrounds that if has and hasn’t are really just inflectionally distinct forms of the same word, thecliticization of has as ’s ought to imply (counterfactually) the corresponding cliticization of hasn’tas *’sn’t. No such conclusion follows, of course, since as we have seen the existence of a reducedform 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.


Reduced Auxiliaries and Syntactic Gaps Of course, the lack of evidence infavor of an analysis of auxiliary reduction in English as involving syntacticadjunction still does not rule out the possibility that there are syntactic con-ditions on the appearance of the reduced forms; and indeed there is quite arobust literature claiming that such is the case. The best-known conditionof this sort is the requirement that auxiliaries do not appear in reduced formwhen followed by a syntactic gap. In sentences such as those in (2.23), theimpossibility of a reduced form of the auxiliary appears to be correlated withthe presence of an immediately following gap, which may result either fromdeletion 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 con-trary to appearances, English cliticization is actually obligatory adjunction tothe right. The restriction that reduced forms cannot immediately precede agap would then follow from a plausible constraint to the effect that the reducedauxiliaries cannot adjoin to a phonologically null element (the gap).

This position has persisted in the thought of many otherwise right-think-ing linguists, despite its completely counter-intuitive nature from a phonolog-ical point of view. The phonological interaction of the clitic (assimilation, inthis case), after all, is with the item to the left, not the right. This direction ofassociation is supported by the fact that the pronunciation of the clitic variesfrom [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. Thevariation in shape which we can observe is exactly the same as that seen in theshape of the regular plural ending (spelled (e)s but pronounced in the samethree 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 al-ways 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]) replac-ing 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


Ignoring these patent phonological facts while claiming that the reducedauxiliaries “cliticize” (invisibly) to the right is unfortunately typical of the loose-ness with which linguists (especially syntacticians) sometimes invoke thenotion of “cliticization” as a kind of magic wand, with no requirement thatit have any observable correlate. Phonologists, however, have generally beenquite uncomfortable with the notion of an operation of rightward cliticizationaffecting English reduced auxiliaries.

Apart from the obvious motivation provided by the facts in (2.24) forseeing auxiliary reduction as an operation which (at least phonologically)associates the reduced form with material to its left, it has also been pointedout that if there were (syntactic) adjunction to the right, we would expectthe sequence of (reduced) auxiliary plus following word to act like a unit forpurposes of deletion and movement. Examples such as those in (2.25) showthat 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 followingsyntactic gap, it does not appear to favor an analysis on which cliticization isrightward-adjunction in the syntax. In fact, we have no reason to believe thatit 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 simpleclitics in the sense that their only exceptional property is the phonologicalone referred to in (2.17), I must provide an account of how such apparentlysyntactic facts as the presence of a following gap can be relevant to theirappearance. I cannot attempt to do this, however, until I have developed amuch fuller account of prosodic structure, especially in English, and its relationto syntactic structure. This task will be undertaken in Chapter 3 below, wherewe will see that it is ultimately possible to vindicate the claim that Englishreduced auxiliaries are indeed simple clitics.

First, however, there are two other issues that must be addressed withrespect to the overall ontology of clitics and related elements. One of theseis to provide at least a sketch of the nature of the second major class of cliticelements, “special clitics,” and the other is to address the outward and visiblesigns by which clitics can be differentiated from morphological affixes inpractical terms.


2.5 Special Clitics

If “simple” clitics are elements whose only definitional property is phonologi-cal as formulated in (2.17), what should we say about Zwicky’s class of specialclitics? I will return in Chapter 4 to an account of their morphosyntacticproperties. It is clearly necessary to note, though, that with few exceptions (seebelow), these elements display the same prosodic dependency as simple clitics,in addition to whatever sets them apart from the rest of the languagemorphosyntactically. As far as the phonology is concerned, the analysis thatfollows from this is the same as for the class of simple clitics: the clitic stemis lexically characterized as prosodically deficient, while a possible non-cliticalternant (if one exists) is not.

On Zwicky’s (1977) interpretation, special clitics are also related tocorresponding non-clitic forms with full-word phonology and ordinary syntax.Surely, however, there is no logical requirement that a special clitic must havea non-clitic alternant, any more than that the same should hold for simpleclitics. All of the characteristics of the clitic itself are ideally to be accountedfor as ‘local’ properties of that element (assuming of course that we eventuallycome up with an appropriate account of the morphosyntax, as promised). Andindeed, there are many special clitics that do not correspond in any obviousway to non-clitic elements: consider connective particles such as the second-position clitic =te ‘and’ in the Homeric example (1.1) for instance. Indeed,it is questionable whether there is ever a special relationship (beyond near-synonymy) between clitics and non-clitics. In particular, there is probably lessreason than Zwicky assumed to treat “conjunctive” and “disjunctive” forms(e.g., French me versus moi) as synchronically alternant forms of the samelexical 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 mor-phosyntax.

(2.26) Morphosyntactic clitic: a linguistic element whose position withrespect to the other elements of the phrase or clause follows a distinctset of principles, separate from those of the independently motivatedsyntax of free elements in the language.

As opposed to the case of phonological clitics, this definition requires consid-erably more fleshing out. In particular, we need a substantive account of thekinds of positioning principle that may govern the location of morphosyn-tactic clitics, their place within the overall architecture of a grammar, andtheir relation to other areas of structure (specifically, the lexicon, morphology,


and syntax). I will address those matters in Chapters 4 and following; forthe present, I will simply assume that the cases for which such treatment isrequired 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 require-ment of a matched non-clitic form should be dropped from the original defi-nition; and (b) most special clitics are phonological clitics, as well: that is, theydisplay the property in (2.17) as well as that in (2.26).

This observation, in fact, allows us to accommodate another apparentanomaly in Zwicky’s (1977) system: special clitics that are not prosodicallydeficient. Examples include standard Italian loro8 ‘to them’ or Tagalog tayo‘we (dual)’. Whatever is at the heart of the distinctive morphosyntax of specialclitics, it is apparently orthogonal to the prosodic property displayed by phono-logical clitics. The important point to note is that some special clitics, whiledisplaying the relevant special morphosyntax, do not happen to be prosod-ically weak. These can be presumed to have full prosodic structure, despitedisplaying morphosyntactic behavior which is (also) associated with prosodi-cally weak items. While the properties in (2.17) and (2.26) often coincide, andmorphosyntactic clitics are typically also phonological clitics, these notions arelogically separable, and empirically distinct.

On the present account, the special class of bound words disappears as anindependent category. These are simply elements whose lexical representationis always prosodically weak (or deficient). They can be divided into two classes:those with special morphosyntax, and those without. There is no reasonto believe that this distinction is correlated with any other interestingdistinguishing property, however.

In later work (cf. Zwicky and Pullum 1983), Zwicky also abandons theindependent class of “bound words,” though with little discussion. In anyevent, the simpler ontology adopted here has the merit of being grounded inthe theory of grammar, rather than being just a set of descriptive categories.This, of course, is just what Zwicky (1985 and elsewhere) counsels: pre-systematic tests and descriptive labels eventually have to be cashed out interms of theoretical constructs.

8 The classification of this element as a clitic is a chronic problem: see Cardinaletti (1991) andNespor (1994) among other sources for discussion. In brief, loro displays some of the propertiesof other pronominal clitics in Italian, while differing in others. It appears following the verbwith 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 distributionin terms of its atypical prosody, a matter I will not attempt here.


2.6 Clitics and Affixes

We now have a reasonable understanding of what clitics are: they are linguisticelements that display prosodically deficient phonology, anomalous morpho-syntax, or both. The comparisons to this point have been between clitics andordinary, non-clitic words, but another contrast to consider is that betweenclitics 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 needto be able to distinguish them from other word-forming elements.

The locus classicus for these matters is Zwicky and Pullum’s (1983) exam-ination of the ways in which we distinguish, in practice, between clitics andaffixes. They propose a number of tests for separating the two, summarizedin (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 paradigmaticgaps than host+clitic combinations.

c. Affixed words are more likely to have idiosyncratic shapes thanhost+clitic combinations.

d. Affixed words are more likely to have idiosyncratic semantics thanhost+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 contain-ing clitics.

As formulated, these points are merely descriptive observations about differ-ences in the behavior of two pre-systematically understood classes of item.Some linguists9 content themselves with lists of behavioral properties of thissort, considering such a more or less comprehensive diagnostic symptomatol-ogy to constitute an analysis of a phenomenon. A list like (2.27), however, doesnot represent an explanation: rather, it lays out what is to be explained. Letus then see if we can ground these differences in the analytic framework I haveproposed for studying clitics.

Some minimally controversial assumptions about the architecture of agrammar will suffice for that purpose. First, I assume that words are built(including affixation) within the lexical phonology. As a result, affixationprocesses have access to the form and meaning of stems, and can depend on

9 Not Zwicky and Pullum, I hasten to point out.


(and affect) this in idiosyncratic ways. Second, words are combined with oneanother post-lexically, through the syntax. On the assumptions of the Lexical-ist Hypothesis (cf. Anderson (1992)), the syntax does not manipulate or haveaccess to the internal form of words—in the sense of lexical items, which mayor may may not be coextensive with the P(rosodic) Words that concern thephonology. After a syntactic structure has been developed,10 it is then subjectto 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 or-dinary lexical items that happen to be unusual from a phonological point ofview, are combined with other items in the sentence by the ordinary mecha-nisms of the syntax: hence, post-lexically. Special clitics are not the provinceof the ordinary syntax—that is what makes them ‘special’—but I assume thatthey are nonetheless introduced into the structure post-lexically. Specifically, Iwill argue in Chapters 4 and 6 below that they are introduced as phonologicalmodifications of the (post-Spell-Out) shape of phrases, similar to the modifi-cations made in words by the Word Formation Rules of the morphology. Thatprecise mechanism is not important at this point, however—the only impor-tant claim is that special clitics are also introduced post-lexically.

From these architectural assumptions, the facts in (2.27) can be derivedas theorems. Property (2.27a), for example, tells us that while affixes may beassociated only with lexically idiosyncratic classes of bases, clitics attach to anyhost, depending only on its position in the structure or at most, its basic lexicalcategory.11 This follows immediately from the fact that affixation is a lexicalprocess, and has access to individual word-level properties, while clitics appearonly post-lexically, where they have access only to syntactic structure.

Essentially the same result holds for (2.27b,c,d). Elements introduced inthe lexicon (affixes and other modifications performed by Word FormationRules) can behave differently in construction with different lexical items, whileelements introduced post-lexically (clitics) have no opportunity to refuse se-lectively to appear (in the case of gaps) or to make item-specific modificationsof form or content, since they have no access to the internal properties of thespecific lexical elements with which they combine.

Property (2.27e) refers to the fact that while a lexical base is affected bysyntactic 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 thestructure 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 underdiscussion in this section, however.

11 Items such as wanna in English represent lexicalized combinations of specific items, ratherthan a clitic attaching to specific host. See section 3.4 for some discussion.


movement or deletion of a host does not have corresponding consequencesfor clitics dependent on it. The fact that the syntax treats affixed words asunits 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 theaffixes as distinct elements. On the other hand, clitics and their hosts do not ingeneral form constituents, and since these are the currency in which the syntaxtrades, it has (in general) no opportunity to affect clitic–host combinations ina unitary way.

Finally, (2.27f ) also follows from the Lexicalist Hypothesis. Clitics areintroduced post-lexically, after lexical items are spelled out, and at this pointthe 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 couldnot be introduced to a form which “already” hosts a clitic. It is perfectlypossible, however, for new clitics to attach to a prosodic word which alreadyhosts other clitics—no architectural considerations bear on this issue, thoughother considerations (phonological and/or morphosyntactic) may limit theextent of clitic accumulation in a given language.

The tests in (2.27) are thus thoroughly grounded in the structure of nat-ural language, and the evidence they provide is correspondingly quite strong.On this basis, Zwicky and Pullum (1983) argue for an initially quite surpris-ing 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 auxiliaryverbs 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 thefollowing stages, on the basis of fairly straightforward reanalysis at each stage.First, in Old English, negation was expressed by the full word combination nawiht ‘no thing’, reduced to nought, and by Middle English, not. This word waspresumably an adverb at this stage in the language, and appeared in adverbialpositions subject to semantic considerations based on its scope.

At this point, however, its semantics had been reduced to that of a logicaloperator expressing negation. Partly as a consequence of this reduced semanticcontent, it was subject to de-emphasis and consequent phonological reductionin many instances, leading to a simple clitic form [n

"t]—a phonological alter-

nant of full word not. Given its semantics, however, and the fact that (forreasons 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, itcould be reanalyzed as a special clitic, attached at the left edge of a VP bearing afeature [+Neg]—or at the right edge of the immediately preceding Infl. But aspecial clitic at the right edge of Infl could easily be reanalyzed, in its turn, as


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 his-torical scenario, but it seems plausible. While the conclusion that -n’t repre-sents 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 distin-guishing clitics from full words. I will not address these in detail, becausemost of them follow quite directly from the understanding of clitics articulatedabove. One exception is the principle that clitics are generally ordered ratherrigidly, while words often are not. This is a matter that can only be addressedin 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 intraditional 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 isthat 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 mark-ers, pronominals, negation, etc.), which have attracted most of the attentionin the clitics literature. Many things that are effectively dismissed descriptivelyby being designated as ‘particles’ have considerable semantic content. This istrue, for example, of the objects Zwicky calls “discourse markers.” These ob-servations suggest that from a descriptive point of view, there are (at least) twosomewhat different kinds of clitic element: grammatical markers, and seman-tically content-ful ones. I will eventually want to draw an analogy between thisdifference and that between inflection and derivation in morphology, and Ireturn to this matter in Chapters 4 and 6.

At this point, I have developed a general picture of the kinds of clitics thatought to be accounted for by grammatical theory, and some notions of howthat might be done, at least for the phonological dimension of clitic-hood. Inthe following chapter, I turn to a closer examination of the prosodic bases ofphonological cliticization. In the course of that discussion, we will arrive ata resolution of the problem posed by the English auxiliaries discussed abovein section 2.4, showing that their unusual properties can indeed be reduced totheir nature as phonological clitics.

3

The Phonology of Cliticization

Let us proceed on the assumption that the proposal to analyze English reducedauxiliaries as simple (phonological) clitics in Chapter 2 is on target. To take thatidea from a gleam in the eye to a genuine analysis, however, we need a morearticulated theory of how the prosodic structure of utterances is organized, aswell as how that structure is related to syntactic structure. It makes sense toaddress the nature of prosody first, beginning with the set of proposals madeby various authors (especially Selkirk (1981, 1984, 1995), as well as Nespor andVogel (1986)) concerning prosodic organization. Excellent summaries of theseissues, including reference to their application to clitics, are to be found inPeperkamp (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 repre-sentations, whose repudiation has since been the source of much phonologicaltheorizing. One of these concerned the interface between the syntax and thephonology. The claim that (morpho)syntactic organization is relevant to theoperation of the phonology, originating in Chomsky, Halle, and Lukoff (1956),led the phonologists of the 1960s to assume that this connection should be im-plemented by allowing the rules of the phonology to refer directly to syntacticstructure.1 I will return to the appropriate characterization of these interac-tions in section 3.3 below.

First, though, my concern is with another basic assumption of the SPEtheory: the claim that phonological representations are matrices of features,whose only internal organization is given by the rows and columns of a matrixwith no further hierarchical structure. The division of the representation intounits such as syllables, morphemes, words, phrases, etc., was carried out (tothe 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 sim-ply the extension of syntactic mechanisms to smaller domains, “syntax” also includes “morphol-ogy” here.


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

(3.1) [p@$théj$Do]

Kahn (1976) argued that the boundary-element approach to phonologi-cal units larger than the segment (specifically, the syllable) had a number ofdeficiencies that could be remedied by adopting a looser, Autosegmental pic-ture (Goldsmith 1979). Instead of identifying syllables as stretches of segmentsbetween two instances of an appropriate boundary element, Kahn proposedtreating them as structural units in their own right, constituting a structuraltier linked to segmental units in a potentially many-to-many fashion. On thatview, 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 servedto establish the idea that syllables ought to be treated as structural units—andwith it, the possibility that phonological representations have hierarchical in-ternal organization rather than being simply a succession of structurally equiv-alent units (segments and boundaries). And this, in turn, opened the doorsfor the more fundamental revisions that lie at the heart of modern views ofprosodic organization.

If the nature of syllables motivated a hierarchical organization of the stringof segments, why stop there? In his 1975 dissertation, Mark Liberman ar-rived at a vastly superior representation of stress by assuming that the sylla-bles themselves were organized into larger constituents, and that a relation de-fined on this structure was the right way to interpret the phonetically slipperyconcept of “stress.” This work is best known from its presentation in Liber-man and Prince (1977); the result is a picture on which potato is represented asin (3.3).

(3.3) PWordw s

Foots w

σ σ σ

p@ thej Do

The Phonology of Cliticization 39

Actually, the representation in (3.3) is somewhat anachronistic in provid-ing labels such as “PWord” and “Foot” for the constituents above the level ofthe syllable. Liberman and Prince simply grouped constituents together intohigher-order units (recursively), but the provision of a principled inventory ofstructural types for these units would follow shortly.

The Prosodic Hierarchy

Telescoping the history somewhat, work of Selkirk (1981, 1984) and Nesporand Vogel (1986) proposed a range of additional constituent types inter-mediate 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 phrasesinto intonational phrases (IPh),4 and intonational phrases into wholeutterances.

Implicit in these theories, and necessary to make them really significant,is the assumption that the prosodic organization they impose is an exhaustiveone. 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., pro-nounced), phonetic content has to be incorporated into prosodicstructure.

That is, stray phonological material not connected ultimately to the root of theprosodic tree is not interpretable, and therefore disallowed.

Central to this work is the claim that “Foot,” “PWord,” etc., are not simplylabels, but represent substantive categories with properties of their own thatdistinguish them. Among these properties is the assertion that each constituenttype (syllable, foot, word, phrase, etc.) defines a kind of domain. At least forNespor and Vogel, phonological rules are expressed with respect to the domainof their application. Thus, where two feet are joined into a word, foot-levelrules 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 concernus at this point.

3 This constituent is called the Phonological Word by some authors, and the Prosodic Wordby others, without entailing any significant difference. The abbreviation “PW(or)d” employed inthis book may be interpreted either way, as the reader chooses. As a phonological/prosodic con-struct, 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 andthe IPh has been a matter of some controversy in the literature, and it is possible that morethan one type of PPh must be recognized. This matter is not relevant to the concerns of thisbook, however. One specific intermediate constituent type, the Clitic Group, will be specificallydiscussed and rejected below.

�

�

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

�

�

�

�

�


their combination as a unit. As a result, the specific label given to a constituentformed at a given point has consequences.

An example where the correct assignment of prosodic structure is crucialto getting the (non-prosodic) phonology right is provided by Polish, asdiscussed by Booij and Rubach (1987). Polish has a number of rules whichaffect the voicing of obstruents, especially in clusters. One of these devoices africative following another voiceless obstruent, as formulated5 in (3.5).

(3.5) Progressive Devoicing:[+Obst+Cont

]−→ [−Voice] / [−Voice]

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 /listıv/. Itdoes not, however, apply across the boundaries between words within largerconstituents such as the Phonological Phrase.

Another rule requires that clusters of obstruents agree in voice whenformed within the larger unit of the Phonological Phrase. Clusters that woulddisagree are adjusted regressively, as expressed by the rule in (3.6).

(3.6) Voice Assimilation: [+Obst] −→ [αVoice] /[+ObstαVoice

]

As a rule of the PPhrase level, Voicing Assimilation applies across wordboundaries (though not across phrase boundaries), as illustrated by the exam-ples 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 expressedin (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, orwhen the next word begins with a sonorant, as in the examples in (3.9).

(3.9) a. zaba ‘frog (fem. nom. sg.)’; zab [p] (gen. pl.)

b. głazy ‘stones’; głaz [s] lodowcowy ‘glacier stone’

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


When the following word within the PPhrase begins with an obstruent, what-ever effect Final Devoicing might have is of course obliterated by the operationof 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 aparticular prosodic constituent type in order to apply correctly, a situationalso 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 asdescribed above. Specifically, prepositions followed by their complements donot 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) asa regularity of the PWord domain. The generalization is that final devoicingis PWord final. The prepositions in (3.10) are phonological clitics, and as suchdo not constitute phonological words. The prosodic structure is thus as givenin (3.11).

(3.11) [PPhr

[PWd

pod[PWd

owocem]]]vs. [

PPhr[PWd

głaz (→s)][PWd

lodowcovy]]

Both Selkirk (1984) and Nespor and Vogel (1986) go beyond the associa-tion of different domains with distinct phonological behavior, proposing a setof specific conditions that came to be known as the prosodic hierarchy condi-tion. One central aspect of this is the claim that in moving up the tree fromsegments to complete utterances, constituent types are layered in the sense ofincreasing monotonically in scope. For instance, Feet cannot contain PWordsor PPhrases, PWords cannot contain PPhrases, etc.

Somewhat more controversially, Nespor and Vogel (1986) claim that theanalysis is exhaustive and non-recursive at each step. That is, every bit ofphonological content dominated by a constituent of a given type has an analy-sis as part of a constituent of the immediately lower type. Thus, PhonologicalPhrases may not contain syllables that are not part of some Foot, which in turnis 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 directlyas a constituent of a PWord without being part of any Foot.


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, oreven, perhaps, all that plausible, and much subsequent work has rejected it.Nonetheless, it serves as quite a useful starting point for discussion, since itis about as restrictive as one could imagine. Whatever we may find out aboutprosodic structure empirically, it can probably be understood as involvingpotential relaxation of the conditions just discussed.

These matters are of direct significance for an analysis of phonologicalclitics, as indeed should already be evident from the Polish facts cited above.Phonologists dealing with the structure of the prosodic hierarchy have gener-ally thought of cliticization in terms of the formation of prosodic constituents.Clitics differ from non-clitics in being combined with other material into asingle prosodic unit, (e.g., the PWord in the Polish example) rather than beingautonomous units themselves. On this account, much of the phonology ofcliticization comes down to the principles of prosodic organization and theirconsequences.

The Clitic Group

One of the more controversial aspects of the prosodic hierarchy as describedby Nespor and Vogel 1986 (apart from the exhaustivity claim just discussed, towhich I will return below) was a particular specific level of prosodic structurewhich they assumed. Following a suggestion from Bruce Hayes, they posit aspecial kind of constituent composed of a sequence of clitics together with theirhost. This unit, the Clitic Group, was assumed to be intermediate betweenPWords and PhPhrases. A Clitic Group is constructed by combining a non-clitic word with any suitable adjacent clitics. “Suitable” here refers to the notionthat some clitics are lexically specified to attach in only one direction.6

Otherwise, clitics form a Clitic Group with the host with which they sharethe most syntactic structure.

This looks rather innocuous, but in the overall context of the prosodichierarchy, it involves several claims that have been questioned by others in sub-sequent work. First of all, clitics have to be distinguishable from non-clitics atthe point prosodic organization is built. And notice that, given the strong formof the prosodic hierarchy assumed here, this distinction cannot be a prosodicone, because clitics have to be PWords themselves by virtue of the exhaustivitycondition, 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 someother authors. I will suggest in later sections of this chapter that this type of lexical specificationcan probably be eliminated.


phonology, and not simply that of phrases or other constituents. Finally, thistheory 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 differentclitics 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 CliticGroup as a separate prosodic category? In the first instance, they must showthat Clitic Groups have special phonological properties that cannot be assim-ilated to those of other constituent types. In this connection, they cite theexample of stress in Modern Greek, noted above in section 2.3. They ob-serve that when words are combined into a single word in a compound, theresult 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 getextra 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 wehave already seen, when two separate enclitics are added, an additional stressappears as in [ðósemúto] ‘give it to me!’ Nespor and Vogel suggest that this isdue to a rule of stress readjustment whose domain is the Clitic Group.

This argument does not succeed, however, as pointed out by Zec and Inke-las (1992) and Peperkamp (1997). We need to say that stress in compounds, asin simple words, is governed by some set of lexical principles, building metri-cal (foot) structure within a PWord. With respect to these rules, compoundscount as a single domain of the relevant type. The main stress within sucha PWord is associated with a binary (disyllabic) Foot constructed at the rightedge of the word, allowing a single following syllable that is not part of thatFoot. The generalization that results is that at most two unstressed syllablesmay follow the main stress within the PWord.

Suppose that we have metrified a host word (e.g., [ðóse] ‘give!’) in thisway, and we now adjoin some additional material, such as a single clitic (yield-ing, e.g., [ðósemu] ‘give me!’). This form conforms to the stress regularity ofthe Modern Greek PWord, since its main stress foot is followed by only oneadditional syllable. If we add two clitics, however, as in [ðósemúto] ‘give itto me!’ the additional material will cause the generalization “No sequences ofunstressed 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 newFoot is built following the original main stress Foot, since there are now at leasttwo syllables to work with (the minimum for constructing a well-formed Footin the language).


On this analysis, no special property needs to be assigned to the CliticGroup that would distinguish it from the PWord. The behavior of host-plus-clitic 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 domainsof type PWord) with a post-lexical rule that applies when phonological wordswith unstressed sequences arise in the syntax (hence post-lexically). No dis-tinct constituent type need be recognized for this account to go through.

The Modern Greek situation can be contrasted with what happens in someItalian dialects (and Latin), where the basic stress pattern takes the presenceof a clitic into account. For example, Lucanian vínn e‘sell!’; v enníll e‘sell it!’;vinn emíll e‘sell me it!’ display stress that remains consistently penultimate inthe form including all clitics. Here we simply say that the basic stress rule ispost-lexical, and no metrical structure is built lexically. Again the behaviorof host-plus-clitic groups can be described without assuming a distinct con-stituent 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 surelyone that says as little as possible specifically about these items, while gettingtheir properties to fall out from other factors. In that sense, eliminating theClitic Group is a desirable move. The examples Nespor and Vogel provide, aswell as those offered by Hayes (1989) and Vogel (1990), seem to be eliminableif we recognize a difference between lexical and post-lexical phonology. Otherpotential examples of unique properties of the Clitic Group disappear as a re-sult of the fact that within a PWord, the boundaries of included PWords can bevisible 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 categoryare lacking, and the other restrictive claims mentioned above that result frompositing this category within the overall system turn out not to be correct (asargued by Zec and Inkelas 1992), most authors have abandoned it as a part ofthe prosodic hierarchy.7

3.2 Dimensions of Phonological Cliticization

Let us assume that the phonology of cliticization is to be understood withinthe following scenario: a lexically interpreted surface structure is interpretedby a hierarchical structure of prosodic units within which lexical and func-tional 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 ofthe prosodic hierarchy.


part, the syntactic structure, and serves as the input to the post-lexical phonol-ogy.8 A framework for talking about the phonology of cliticization, based onthe combination of Lexical Phonology and Prosodic Phonology, would go asfollows:

(3.12) a. Lexical principles9 determine the phonological shapes of lexicalitems (members of lexical categories). This includes assigning pro-sodic structure to them, up to the level of the PWord, as argued byBooij (1988).

b. Post-lexical principles govern the relation between syntactic (S-)st-ructure and the prosodic organization of the whole utterance. Thisincludes building structure above the level of the PWord, and alsoincorporating material that (for whatever reason) is not alreadypart of a PWord, so as to satisfy Full Interpretation (3.4).

c. Post-lexical phonological adjustments apply to the resultantstructure.

If we think of (phonological) cliticization in these phonological terms, itbecomes a matter of describing the ways in which clitics are incorporated intoprosodic structure along with non-clitic elements, together with the phono-logical 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 non-clitic host?

With respect to the first of these questions, I have already suggested thatthe best theory would be one on which there is no specific property such as“[±Clitic]” unique to clitics, but rather where clitics are distinguished fromnon-clitics in terms of some independently available dimension within thephonology. The proposal of section 2.3 in the previous chapter follows ex-actly this program, within the architecture of (3.12) above. Non-clitic lexicalwords are assigned prosodic structure up to the level of the PWord within theLexical Phonology. (Phonological) clitics, in contrast, are “prosodically defi-cient” in the sense that they do not have PWord structure at the point they are

8 I will return to the nature and content of this relation in section 3.3 below.9 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.


introduced into S-structure. To satisfy Full Interpretation, they must be in-corporated 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 agiven clitic, a matter to which I return in section 3.3. For any given host-plus-clitic combination, there are essentially four formal possibilities as to theprosodic structure that might result from combining them, as Selkirk (1995)points out.11

(3.14) a. PWord Clitic: PPh

PWd PWd

Host Clitic

b. Free Clitic: PPh

PWd

Host Clitic

c. Internal Clitic: PPh

PWd

Host Clitic

d. Affixal Clitic: PPh

PWd

PWd

Host Clitic

The first of these, the PWord clitic, is the case where the clitic is assignedindependent status as a PWord in the phonology. A Free clitic is associated (asa non-PWord) with its host within the PPhrase. An Internal clitic is incorpo-rated into the same PWord as its host, as opposed to an Affixal clitic, which isadjoined to the host to form a recursive PWord structure. While all of thesestructures are logically and formally possible, however, it remains to be shownwhich of them are actually instantiated in natural languages, and how to dis-tinguish among them in the phonology of a particular language to the extentthat more than one possibility is admitted by Universal Grammar.

10 This is far from an original proposal. Previous work that treats the properties of phonologicalclitics in essentially the same way includes Berendsen (1986); Selkirk (1986, 1995); Zec (1988); andInkelas (1989), among others.

11 These possibilities are of course independent of the linear order of the clitic with respect toits host. For the sake of concreteness, I represent the specific case of enclitics here.


The Strict Layer Hypothesis

Two of the formal possibilities in (3.13) above violate the Strict Layer Hypothe-sis, as this was informally described in section 3.1. Free clitics involve structuresin 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 ofwhether such structures are ever motivated thus bears on the plausibility ofthe Strict Layer Hypothesis itself.

Inkelas (1989) argued that this claim, as it appeared originally in worksuch as Selkirk (1984) and Nespor and Vogel (1986), should not be treated asa monolithic notion, but rather broken down into a number of logically in-dependent claims. Some of these are inherent in the nature of the prosodichierarchy itself. That theory of representations claims not only that prosodicconstituent types are drawn from a limited universal inventory, but also that(unlike, say, the constituent types that appear in syntactic representations, suchas 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 couldexpress the basic nature of the prosodic hierarchy as involving two fundamen-tal requirements.

(3.16) Layeredness: No C i dominates a C j where j > i (e.g., no Foot con-tains a PWord)

Headedness (first approximation): Every C i directly dominatessome C i−1 (e.g., every PWord contains a Foot)

The Strict Layering Hypothesis can be expressed as the claim that repre-sentations 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., noPWord contains another PWord; adjunction structures do not exist)

In order to maintain its logical independence from Non-recursivity, theformulation 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 hi-erarchy, and will not be questioned here. The requirements in (3.17), however,


make substantive claims about the range of prosodic structures found in thelanguages of the world, and as such, are subject to empirical confirmation. Ev-idence suggests, in fact, that they are violated in some instances, and this callsfor 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 twopossible outcomes of the post-lexical combination of a phonological clitic andits host: either a PWord is built over the clitic, and the two become part ofthe same PPhrase, or else the clitic is incorporated into an adjacent elementof the next level up (a syllable into a foot, or a foot into a PWord). These arethe “PWord clitics” and “Internal clitics,” respectively, of (3.14a,c). A languagein which Free clitics (3.14b) appeared would violate Exhaustivity, since a sylla-ble (or Foot) would be directly dominated by PPhrase; while a language withAffixal clitics (3.14d) would violate Non-recursivity (and perhaps Exhaustivityas well, if the clitic consisted only of a syllable). If these structures are motivatedin some language(s), the theory must be amended so as not to impose the StrictLayering requirements (3.17) as absolute conditions on prosodic structure.

English Function Words Selkirk (1995) argues that both of these problematicstructures occur in English. She analyzes the phonology of “function words,”typically monosyllabic prepositions, determiners, auxiliary verbs, and personalpronouns. These have full forms that appear, with stress, under some condi-tions: in isolation, when focused (Bettina can speak, but refuses to), or in mostinstances 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. Selkirkanalyzes the strong forms as cases where PWord structure is built over the itemin question, either to support the phonetics of focus or because ends of phrasesare preferably also ends of words. In the remaining cases, the function wordsenter the structure as prosodically deficient items—phonological clitics in thesense of (2.17). They must thus be integrated into prosodic structure in someway to satisfy Full Interpretation.

Consider first the cases of weak monosyllabic function words that are notphrase final. Since these do not bear stress, they cannot be PWords. They can-not even be organized into feet, because the resulting structure would produceadditional secondary stresses. In a phrase such as or for cónferences the two ini-tial syllables are without stress, showing that neither is the head of a foot. Wecan conclude a fortiori that neither alone, nor the two taken together constitutea PWord, and thus the PWord clitic analysis is excluded. But then how are theygrouped together with the lexical word conferences into a PPhrase? They can-not have been incorporated into the PWord [kánf@r@ns@z] as Internal clitics


either, because English excludes sequences of two or more unstressed syllablesat the beginning of a PWord. But a series of non-final phonological clitics cangive rise to much longer unstressed stretches, as in or for a convérsion.

The stress patterns found with English function words in non-phrase-finalposition thus argue that they are attached to the PPhrase but are not part of aPWord. This is confirmed by other phenomena as well. PWord initial voice-less stops are aspirated, even when the syllable they initiate is unstressed, as inCanadian [kh@néjdi@n], but this is not true of PPhrase initial function words,as in can aid (orphans) [kn

"éjd]. It is necessary to conclude that function words

that are non-final in their phrases are phonological clitics of the type identifiedas 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 ap-pears in this form of English after a PWord-final low vowel, when the followingword begins with a vowel. The one class of weak function words that do notreceive stress (and PWord status) when appearing in phrase-final position areweak object pronouns, as in hit’m again! In a sequence of verb plus unstressedobject pronoun, the verb itself retains its status as a PWord, as shown by thefact that intrusive r appears at its right edge under appropriate conditions: saw-r us, withdraw-r it. But the pronoun also behaves as if it were at the right edgeof a PWord, in sequences such as I saw ya-r and asked about it. The struc-ture of verb plus weak object pronoun is thus evidently as in (3.20), violatingNon-recursivity at the level of the PWord.

(3.20) PPh

PWd

PWd

saw us

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


dialects of Serbo-Croatian analyzed by Zec (1993). The properties of tonal ac-cent in these three dialects provide a minimal triplet for phonological cliticstructures. The same unstressed preposition u ‘(in)to’ combines with a follow-ing PWord such as graad ‘city’ or glaavu ‘head’ as an Internal clitic in one di-alect (Eastern Herzegovina), a Free clitic in another (Belgrade), and an Affixalclitic in a third (Šrem, Macva). We will see other examples of these structureslater in this book.

Constraint Systems We might take these examples to suggest that Strict Lay-ering is just a “tendency,” or a matter of “markedness,” but surely that is just acop-out. Or we might take them to show that the principles in (3.17) are justwrong. But throwing the baby out with the bathwater in that way is not sat-isfactory either. Selkirk suggests in each case that what is actually going on issubtler 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 prin-ciples of grammar are not absolute. Rather, they are violable, true to the extentthat some more important principle does not require them to be violated. Thisis a familiar enough notion in life . . . “Thou shalt not steal,” and we do not, atleast not until we need to in order to feed our family. This is the core notion ofOptimality Theory (or “OT”), which forms the basis of several recent analysesof cliticization phenomena (including Selkirk’s).

The basic idea of OT is that the structure of a grammar comprises a systemof ranked but violable constraints rather than a set of ordered rewriting rules.Rules and constraints provide alternative algorithms for answering the ques-tion: given the abstract (input) representation of a form, how do we determinethe overt shape of its realization? The rule-based answer says: start with theinput, and apply a sequence of systematic modifications to it, resulting eventu-ally in the output. This format allows us to express a great deal, but each suchmodification has an all-or-nothing character to it. Either a given rule appliesor it does not. Of course, another rule may apply later and obscure what thisrule has done, but either it applied or it did not.

Within OT, things are quite different. For any given input, the outputcould in principle differ from it in any arbitrary way. Therefore, again in prin-ciple, any well-formed structure within the vocabulary of phonological expres-sions might be the output corresponding to the given input. Taking that as thestarting point, we could assume that the problem is to provide a set of “triage”conditions—the constraints—for sorting out all of these (logically) possibleoutputs and arriving at the most suitable: the optimal one.


In those terms, we can say that the constraints may (indeed, must) beranked. This ranking establishes a balance between two contradictory influ-ences on the output form. On the one hand, the output should look as muchlike the input as possible (Faithfulness), but on the other hand, the outputshould also conform to general principles of Markedness. The relative rankingsof specific Faithfulness and Markedness constraints determine the way theseconflicts are resolved in the choice of the optimal output form.

Some things in the theory of OT as it has evolved in phonology (and morerecently, in syntax) are essential: notably, the description of an input–outputrelation as based on a ranked set of violable constraints evaluated simultane-ously. Other matters are contingent, part of the specific implementation ofthis idea that has been widely explored in phonology. One of these is the claimof universality for constraints, a proposal which sometimes seems intendedprimarily to avoid the question of how specific constraints might be learned.There is clearly a general component to constraint systems, but whether every-thing but the ranking can be ascribed to Universal Grammar is much less ob-vious.

A second, even more controversial claim is the principle that there is onlya single step between input and output, mediated by a single ranked constraintsystem. Some fairly common and well-established situations in phonology,such as opaque interactions among regularities, pose very serious problemsfor this idea, and have led to rather convoluted supplements to the theory thathave not gained wide acceptance. There are also reasons to believe that ba-sic insights of the cyclic organization of phonology and morphology as thisdeveloped within the theory of Lexical Phonology are quite realand should be preserved, although that grammatical architecture is quiteat odds with the one-step conversion process assumed in much of the OTliterature.

In this book, I adopt a modified version of OT, along lines developed inrecent work of Kiparsky (2000, forthcoming); see also Booij (1997). Specifi-cally, I assume a cyclic organization of the Lexical Phonology of words, whereconstraints operate to instantiate the phonology of a given level of the LexicalPhonology, as well as that of the Post-lexical phonology, in such a way thatthere is more than one constraint-satisfaction operation in the phonology as awhole.

The analysis of phonological clitics is almost exclusively a matter of thePost-lexical constraint system, on this view. For particular items, several is-sues must be resolved. First, we must establish whether the item’s prosodicdeficiency is inherent or derived by rule. I assume (with Selkirk, for instance,


among others) that items that do not belong to lexical categories are prosodi-cally deficient, at least to the extent of not having inherent PWord structure.This follows from their not having passed through the parts of the LexicalPhonology that create such structure. Perhaps, however, some other princi-ples (e.g., fast speech reductions) can deprive even lexical material of some ofits prosodic organization.

Second, we must establish the way in which phonologically clitic materialis to be incorporated into the prosodic structure of neighboring material so asto satisfy Full Interpretation (3.4). At what level of prosodic structure (sylla-ble, Foot, PWord or PPhrase) is the item attached, and what is the resultingstructure? 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 requirementsin (3.21), and so presumably the only candidates submitted to the constraintsystem for evaluation conform to these.

(3.21) Layeredness: No category dominates a higher-level category.

Headedness: Every category directly dominates (at least) oneelement no more than one level below it on the hierarchy.

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

(3.22) Exhaustivity(Ci): Every element of category Ci is exhaustively com-posed of elements of category Ci−1.

NonRecursive(Ci): No element of category Ci directly dominatesanother instance of Ci.

A description of the phonology of Stray Adjunction (in the terminologyof Chapter 2) includes a ranking of these constraints (perhaps with respect toothers), as well as a resolution of such other matters as the choice of a host onthe 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 var-ious modes of phonological cliticization, let us review the descriptions of aset of three Italian dialects discussed by Peperkamp (1997). These provide anice contrasting set, differing minimally in the way clitics are incorporatedinto prosodic structure which can be described in terms of varied rankings of


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

(3.23) Neapolitan:Verb Imperative Imperative + ‘it’ Imperative + ‘you’ + ‘it’do fá fáll@ fattíll@tell cónta cóntal@ cóntatíll@comb péttina péttinal@ péttinatíll@

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

(3.24) a. PPh

PWd

PWd

F. . . σ

lex cl

b. PPh

PWd

PWd

F. . . F

lex 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 introducean additional stress. Peperkamp’s discussion suggests that there are aspects offormal suppletion that require the treatment of the two-clitic sequence as asingle unit, which is eligible to be a Foot. Alternatively, we could asume sim-ply that the two monosyllabic units are introduced together, and subsequentlyorganized 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 can-didates that are to be compared, so they play no part in the ranking. It is alsothe case that prosodic structure assigned lexically is generally preserved, so weassume a high-ranking Faithfulness constraint to this effect.

(3.25) Prosodic Faithfulness: Prosodic structure in the input should bepreserved in the output.

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


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

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

The prosodically deficient material (i.e., the clitics) must be incorporatedinto the structure somewhere, and there are not a lot of choices available. In-corporation into a Foot would violate well-formedness conditions on feet, aswell as Faithfulness to existing prosodic structure. Incorporation into the ex-isting PWord would also violate faithfulness. Incorporation at the PPhraselevel would violate Exhaustivity(PPh). The Affixal clitic structures we ac-tually 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 dominatedby 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 withthat employed in another dialect, Lucanian.

(3.28) Lucanian:

a. vínn e‘sell’; v enníll e‘sell it’

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

We see in (3.28a) that the addition of a clitic in this language causes stress toshift rightward.13 Apparently a binary trochaic foot is constructed over the lasttwo syllables of the form, including both stem and any following clitics. Theforms in (3.28b) with two clitics have this foot constructed entirely over cliticmaterial. In this language, Stray Adjunction produces Internal clitics, sacri-ficing Faithfulness to maintain the Strict Layering constraints. The resultingstructure 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 beresolved 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] corre-sponding to unstressed [@].


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, illustratedin (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, sug-gesting 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 assignedeven though two syllables of additional material would support the construc-tion of a new Foot if this material were within the PWord. Apparently, then,Stray Adjunction in Standard Italian produces Free clitics by attachment to thePPhrase, 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 differentrankings of the prosodic constraints, yielding three different structural typesof 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 assumedin each case that the appropriate host for a given clitic is apparent, and that theproblem is to determine how that clitic and its host relate to one another inthe structure resulting from Stray Adjunction. When a clitic appears betweentwo prosodically complete potential hosts, however, it is necessary to establishthe direction in which incorporation will take place. In some cases this is apurely phonological matter, but in others it depends (at least apparently) onthe syntactic structure within which the material in question appears. I turn


now to the issue of how syntactic and phonological representations interact todetermine these matters.

As I noted at the beginning of this chapter, “classical” Generative Phonol-ogy assumed that the relation between syntactic structure and phonology wasa basically simple one: phonological rules referred directly to syntactic form,and could, at least in principle, make use of any and all information they mightfind there. A complication recognized at least as early as Chomsky and Halle(1968), however, was that the structures provided by the syntax might not becompletely appropriate for the phonology. For instance, syntactic organizationtends to involve greater degrees of embedding as sentences get longer, but thedivision of sentences into phrases is comparatively flat. Indeed, the phrasingsuggested by the syntax may not be the same as that suggested by the phonol-ogy. The classic example of this is a sentence like This is the cat that chasedthe rat that ate the cheese . . . , where syntactic phrasing groups nominal headstogether with the following relative clause but the phrase boundaries in pro-nunciation come between the head and the relative clause.

(3.34) Syntax: [CP

This is [DP

the cat [CP

that chased [DP

the rat [CP

that ate

[DP

the cheese [CP

that . . .]]]]]]]

Phonology: [PPh

This is the cat][PPh

that chased the rat][PPh

that

ate the cheese] [PPh

that . . .]

To accommodate such divergences, SPE assumed that a set of Readjust-ment Rules operated before any of the phonological rules proper, to massagethe structure in various ways so as to make it appropriate for phonologicalinterpretation. Though this notion was never pursued in any detail in the lit-erature of the period, the principal lesson to be drawn from the necessity ofsuch readjustments is surely that the organization of an utterance for syntacticpurposes on the one hand, and for phonological ones on the other, is simplynot the same.

The development of prosodic phonology, as sketched in the sections above,brings this point into greater relief. If substantial hierarchical organizationhas to be imposed on a phonological representation for its own purposes,the need for the phonology also to consult another similar but non-identicaland logically independent hierarchical structure (that provided by the syn-tax) is not obvious, and risks substantial redundancy where there is not actualconflict.

In fact, the approach that has emerged (and which I will follow here)assumes that the role of syntactic representations in phonology is limited toproviding information that may be consulted (along with other, purely


phonological factors) in the construction of prosodic structure. The rules ofthe 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 inthe phonology to categories and structures of the syntax is by the principlesthat construct prosodic organization. In the context of an OT approach tophonology, the relevant principles are Alignment constraints, which specifypreferred relations between two distinct analyses of the same material. Thegeneral 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 specifyingthat “For any instance of Cati, its Right/Left edge should be aligned with theRight/Left edge of an instance of Catj.”

Perhaps the simplest instantiation of this system is the description of thefact that lexical words project PWords in the phonology. We can express thepreference for this structure in terms of the ranking of the two constraintsin (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 leftand right edges should be aligned with the left and right edges, respectively,of a PWord. Various other factors, however, especially the need to incorpo-rate phonological clitics into the structure, may lead to circumstances in whichthis alignment is not perfectly satisfied. These are described in terms of otherconstraints that outrank one or the other of those in (3.36).

In terms of higher-level categories, we want prosodic structure to reflectsyntactic structure, in the absence of overriding factors. We can accomplishthis, 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 leastmaximal projections) should project PPhrases in the prosodic structure.

Left to their own devices, these constraints would produce much the samerecursive embedding in the phonology as that of the syntax, but we alreadyknow from examples like (3.34) that a rather flatter structure should be pre-ferred. In fact, however, this will follow (to a first approximation, at least)from the presence in the grammar of the independently necessary constraint


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

(3.38) a. [PPh

the cat] [PPh

that chased the rat]

b. [PPh

the cat [PPh

that chased the rat]]

Now let us see how this apparatus helps us describe Stray Adjunction,the incorporation of prosodically deficient material into existing structure. Inmany languages, Stray Adjunction prefers to follow syntactic structure, in thesense that stray material is incorporated into the neighboring constituent withwhich it bears the closest syntactic relation. Recall the case of weak preposi-tions in Polish, for example, which form a recursive PWord with the followingmaterial that constitutes (part of) their object, as in (3.11).

We need to ensure that a new PWord is not introduced to project non-lexical prosodically deficient material. Assume that some form of a constraint*Struct (“Do not build structure unless necessary”), highly ranked, producesthis effect for function words (where (3.36) cannot apply). Full Interpreta-tion still requires the preposition to be incorporated into prosodic structurein some way. In this case (3.37a) requires the left edge of a PPhrase to coincidewith the left edge of the PP, thus including the preposition in that PPhrase. Thechoices are to incorporate the stray syllable either directly into the PPhrase, orelse into the following PWord, recursively (as an Affixal clitic) or not (as anInternal clitic). The constraint ranking in (3.39) will produce Affixal clitics asrequired for Polish.

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

For the case in which clitic attachment follows the direction of syntacticaffinity, then, we can say that the language prefers to have the edges of syn-tactic phrases coincide with the edges of PPhrases, as mandated by (3.37). Asa result, clitics attach either as proclitics or as enclitics because the other di-rection would be counter-syntactic. For instance, Romance clitics are procliticwhen pre-verbal, and enclitic when post-verbal, because (as I will argue in laterchapters) their closest syntactic affiliation is with the verb.

Consistent adjunction of a phrase-initial item to its left would require usto violate the constraint (3.37a), and in languages such as Kwakw’ala, that isjust what happens. We saw in section 2.2 that phonological clitics in that lan-guage systematically adjoin to material on their left, usually resulting in a vio-lation of the similarity between syntactic and prosodic structure found in other


languages such as Polish. It must therefore be the case that in Kwakw’ala someother constraint outranks one or both of those in (3.37).

Suppose we want to force clitics always to associate to their left, as inKwakw’ala. In the present system, there is no way to say that directly. But ifa clitic attaches in the counter-syntactic direction, this must be because someother constraint is thereby satisfied, a constraint that (in this language) out-ranks (3.37a). What is gained by having the clitic attach to its left? A possibilityis that this avoids having the PPhrase begin with the clitic, rather than withlexical material. On that view, we can describe Kwakw’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 materialout of the left periphery of the PPhrase than it is to have syntactic and prosodicphrasing be isomorphic. The high ranking of a constraint requiring prosodicconstituents to begin with lexical material can probably be connected to thefact that Kwakw’ala is a language that has essentially no prefixes: the only ex-ceptions to this are reduplicative elements whose content can be argued to belexical.

We can relate this situation to that in other languages where similar align-ment 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 nom-inals constitute a set of simple clitics. According to Wheeler, Yates, and Dols(1999: 262), “[w]hen a finite complement clause, beginning with que ‘that’, isthe object of a weak [i.e., atonic] preposition [. . . ], the weak preposition isdropped” as in the examples of (3.41).

(3.41) a. Estem contents (*de) que hagis vingut ‘We are pleased that you havecome’

b. Estem contents *(de) la teva vinguda ‘We are pleased at yourcoming’

c. Els hem d’acostumar (*a) que no hi vagin ‘We must get them usedto not going there’

d. S’ha acostumat *(a) no anar-hi ‘They have got used to not goingthere’

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


prepositions to a constraint requiring the left edge of a clause (CP) to coincidewith the beginning of a prosodic word, assuming that NonRecursive(PWord)would force these phonological clitics to attach directly to the PPhrase as Freeclitics.

If it is in fact more important for CPs to begin at a word boundary than forthe empty prepositions that serve to assign case to be overtly realized, we couldget the facts of deletion by ranking the alignment constraint higher than a pre-sumed constraint Max(Pfnc) requiring semantically empty structural elementsto 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 phonol-ogy, including constraints that enforce Stray Adjunction in this language in away 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 speci-fying the direction of their attachment. For Klavans, this directionality para-meter was to be specified individually as a property of particular clitics, ratherthan being reducible to general properties of the language. The view presentedhere, in contrast, treats direction of attachment as a matter that follows fromthe overall prosodic properties of the language, and which is not available forlexical specification with respect to individual items. In fact, the literature pro-vides very few instances of directional attachment of clitics which are plausiblyitem-specific, and even fewer explicit arguments for the necessity of such de-scription.

Nespor and Vogel (1986: chapter 5) argue such a case for some pronominalclitics in Modern Greek. They show that clitics like mu ‘me’ and to ‘it’ areproclitic to a following verb in examples like

(3.43) o ðáskalos mu= to= ípeart teacher me it saidThe teacher said it to me

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

(3.44) a. o ðáskalòs =mu to= ípeart teacher me it saidMy teacher said it


b. to prósfatò =mu árTroart recent 1sg articlemy recent article

c. ta iposiníðità =mu esTímataart unconscious 1sg feelingsmy unconscious feelings

d. o pio filóðoksòs =mu fílosart most ambitious 1sg friendmy most ambitious friend

They then suggest that in their possessive use, clitics like mu must be specifiedas attaching only to their left, as enclitics.

This does not follow, however, within the present framework. There isa clear difference between the environment of the clitic in (3.43), on the onehand, and in (3.44) on the other. In (3.43), the Verb Phrase “me it said” surelyconstitutes a PPhrase separate from the subject “the teacher.” A clitic suchas mu in this example, introduced as a special clitic associated with the verbípe ‘said’ could only attach to its left by crossing the boundary between thesetwo PPhrases. Within the DPs of (3.44), however, there is no reason to assumesuch a boundary between PPhrases, and thus nothing to prevent the clitic fromattaching to its left as an enclitic.

Suppose we say, then, that in Modern Greek (as described in Nespor andVogel 1986) prosodically deficient elements attach to a preceding PWord as en-clitics where this is consistent with PPhrase boundaries projected from the syn-tax, but as proclitics to a following PWord otherwise. There is then no needto say anything about the direction of attachment of any particular clitic on anitem-specific basis.

The motivation for Nespor and Vogel’s (1986) wish to specify the posses-sive markers as obligatorily enclitic may be to force them into a non-initialposition within the DP, by requiring them to have a appropriate host on theirleft. There is no need to do that, though, if we simply treat them as specialclitics introduced in second position within the DP, within the framework tobe developed below in Chapter 6. From example (3.44d), it appears that “sec-ond position” in this case means “following the first full PWord,” since initialunstressed elements (articles, and the intensifier pio ‘most’) do not serve ashosts. None of this, however, requires us to specify the clitics themselves forthe 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 re-quire an item-specific indication of direction of attachment. He first establishes


that the basic direction of prosodic integration of Dutch clitics is leftward, asenclitics to a preceding host. This is true for a wide range of pronouns, deter-miners, and other weak elements, which he argues are incorporated as Internalclitics into the preceding PWord, adducing a substantial number of phonolog-ical phenomena in support of this analysis. The leftward attachment of cliticsis attributed to general principles of Dutch prosody, and is entirely consistentwith the position espoused here.

Booij then notes that when clitics appear in a position where there is nopossible host on their left (particularly in sentence-initial position), they at-tach 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 shownto be entirely parallel to the structure of weak derivational affixes in the lan-guage. Since the same clitic may attach either leftward (in the general case)or rightward (when necessary), he notes that these elements could not be lex-ically specified for a direction of attachment. Rather, that question must bedetermined on the basis of the way general properties of Dutch prosody affectparticular structures, without regard to the identity of the individual element.

There are two Dutch pronominal clitics, however, for which Booij sug-gests that a lexical specification of direction of attachment is necessary: theweak form ie ‘he’, [i], and its allomorph [di]. These, he argues, can only attachleftward. When no host is available on the left, they cannot be used, unlikeother Dutch clitics.

(3.45) a. Komt-ie? ‘does he come?’ [σ

kOm][σ

ti]b. *Ie komt ‘he comes’

c. Kan-die ‘can he’

d. *Die kan ‘he can’

e. Het gaat wel goed ‘it goes well’ [σ

@t][σ

xat]

This argument is not decisive, however. In fact, [i] and [di] differ fromthe rest of the Dutch clitics in another way: they are the only ones whose weakform contains a full vowel, rather than a schwa. Syllables containing schwaare not eligible to constitute PWords by themselves in Dutch, while syllablescontaining a full vowel can stand by themselves, even when vowel initial like[i]. This suggests that while most of the phonological clitics of Dutch—thosewhose 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 consis-tent with the position taken here. As Booij shows, a proclitic form integrateswith its host as an Affixal clitic, as in the structure (3.46) for de keer ‘the turn’.


(3.46) PWord

PWord

Ftσ σ

d@ ker

If a clitic constituting a complete foot such as [i] or [di] were placed in the sameway within this structure, the resulting initial foot would create a new initialstress within the larger PWord, violating the Dutch stress pattern or (if theoriginal stress on the host were deleted) producing a stress pattern unfaithfulto that of the input host. It seems reasonable to attribute the ill-formedness ofproclitic [i] and [di] to such factors, rather than to a lexical specification fordirection of attachment.

Booij points out (personal communication) that there are prefixed wordsin Dutch, such as overkomen ‘to happen to somebody’ in which the prefix(here, over-) consists of a Foot, and this Foot bears (secondary) stress. Thisconstruction is a lexical item, however, and so the prosodic structure involvedis created by the lexical phonology, not the post-lexical phonology as would bethe case with the integration of a clitic such as ie into prosodic structure. If wesay that a new secondary stress in a Foot preceding and adjoined to the PWordwith main stress cannot be created post-lexically (though the same configura-tion created lexically is preserved by high-ranking Prosodic Faithfulness), wecan ensure that initial ie will be blocked without having to stipulate its direc-tion of attachment as a lexical property of the clitic.

I conclude that the Dutch data do not prevent us from dispensing withKlavans’s parameter of (item-specific) directionality. Quite generally, thechoice of proclitic or enclitic attachment can be made to follow from moregeneral principles of a language’s prosodic organization.

Sometimes the direction of cliticization is partially obscured by post-lexical adjustments of prosodic structure. An example of this is furnished byGallo-Romance (Old French) facts cited by Jacobs 1993. Here a clitic betweenthe subject and the verb is generally enclitic to the preceding word, but procliticif 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 (similarto Kwakw’ala). As a result, there was normally a PWord boundary before theVerb, causing the clitic to associate leftwards so as to produce the usual encliticstructure.

On the other hand, the language also ranked highly the constraint Onsetrequiring syllables to have onsets. In the post-lexical phonology, this outrankedthe PWord–LexWord alignment constraint. When a lexical verb began with a


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

(3.47) a. Jot (= jo+te) vi ‘I saw you’

b. jo t’aim (= te+aim) ‘I love you’

c. . . . qu’il parloient einsi . . . ‘that they were speaking thus’

d. qu j’en recevroie tel cop ‘that I would receive such a blow from it’

e. 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 asin (3.48).

(3.48) Onset�Align(PWord,L,LexWord,L),Prosodic Faithfulness

Along similar lines, Chamicuro (Parker 1999) shows a constraint (or con-straint 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 situ-ation is disturbed, however, when the preceding word ends in a heavy sylla-ble and the article itself is light. Under those circumstances the clitic articleshifts to its left prosodically, becoming an enclitic on the preceding word andthus satisfying the final light syllable requirement on that PWord. Developingthe relevant constraints to express the prosodic requirements on PWords inChamicuro would take us too far afield here, and I do not propose to formu-late a complete analysis that would achieve the observed interaction betweenprosodic and syntactic factors, but the general lines should be clear within thesystem of this book.

3.4 Phonological Clitics and Cliticization in English

Let us now return to the case of contracted auxiliary verbs in English, whichwas raised in section 2.4 of the previous chapter. Recall that I proposed to fol-low Kaisse in treating items like ’s, ’d as prosodically deficient lexical variantsof full word forms (is, has, had, would, etc.). The problem, if we are to main-tain a treatment of them as simple clitics, is to account for the distribution ofthese elements in terms of the phonology alone in the face of apparently syn-tactic restrictions on their occurrence. Within the theory adopted here, it isnot possible to allow the phonology to consult the syntactic representation di-rectly (to discover the presence of a following gap, for instance): the only way


syntactic form can influence the phonology is through its role in constructingthe prosodic representation.

We need to account for the following apparent paradox. On the one hand,the reduced auxiliaries display an obvious phonological dependency on thematerial to their left, and no sensitivity to that on their right. On the otherhand, the presence of a syntactic gap on the right precludes the use of the re-duced form of the auxiliary, and requires the full (non-clitic) form, while leftsyntactic context is irrelevant. Finally, whatever is going on, it has to be some-thing that follows largely from general principles, since children learn how touse the reduced auxiliaries quickly, without making mistakes and without be-ing provided with negative data.

If we reject an account on which cliticization is actually a syntactic op-eration, how could syntactic effects (like the presence of a gap) show up asconditions on cliticization? Let us take a hint from Kaisse’s proposal above insection 2.4, on which the choice of allomorph (i.e., determination of the possi-bility of the clitic form) is made after sentence prosody has been assigned. Thatsuggests that sentence prosody might be what is really relevant. The appear-ance of a syntactic condition on clitic auxiliaries could be just the reflex of theeffects of syntactic form on sentence prosody.

A simple possibility along those lines is suggested by the core cases inwhich 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 thatin each case the potentially clitic auxiliary appears at the end of a phonologi-cal phrase. We might then note that a phrase-final verb must bear the nuclearstress of the phrase. The clitic forms of the auxiliaries, however, are prosodi-cally ineligible to be stressed, due to their lack of phonological content (theyconsist of a single segment, not even a syllable). As a result cliticization, con-strued as choice of a clitic allomorph, will be blocked. The relevance of thegap is that its filler would otherwise follow the item in question within thephrase, and the absence of that material causes the auxiliary to be final in itsphrase.

While promising, this approach is probably insufficient. One problemis that it does not connect these facts with the other thing we know clearlyabout the contracted auxiliary forms: phonologically, they attach to a preced-ing word. Another is that in constructions like (3.50), the degree of stress on


the full auxiliary verb is quite low, but that does not improve the acceptabilityof 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, thoughthat does not mean prosodic structure is irrelevant.

Recall that our problem is to account for the fact that not only are sen-tences like (3.51a) impossible in English, but language learners know this with-out ever having been explicitly told not to use reduced auxiliary forms in suchpositions.

(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 near-twin (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 distinctionof syntactic structure or meaning, this does have one consequence. If each ofthe syntactic phrases in these examples is reflected by a PPhrase in the prosodicstructure, 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 theexception of post-verbal object pronouns) function words in English (includ-ing auxiliaries) are Free clitics, attaching directly to the PPhrase rather thanhaving a PWord build over them. The cases she considered were those in whichthe 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, nota syllable, and their phonological behavior shows that they must be integratedinto prosodic structure in a different way.

Let us assume that (in English, at least) stray segments can only be incor-porated as parts of a syllable, not directly as constituents of a Foot, PWord,PPhrase, or higher level category. But then if Prosodic Faithfulness is rankedhighly enough to require the preservation of existing syllables at the pointwhere the clitic auxiliaries are under consideration, there is only one way toincorporate a stray segment such as one of these forms: by adjunction, as a syl-labic affix to the syllable immediately to its left. The complete prosodic struc-ture 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 claimthat the predicate of a sentence constitutes a phrase.


(3.52) IntPhrasePPhrase PPhrase

PWord PWord

Ft Ftσ

σ σ σ

tIm z 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 incodas, and about relations between nucleus and coda, is violated by syllablescontaining the “syllabic affixes” /z/ and /d/. Final clusters such as that in texts[tEksts] occur only under these circumstances, and it seems reasonable to re-gard the syllabic affixes as adjoined to the basic syllable, rather than incorpo-rated into it.

The adjoined structure is the one assigned to a number of diverse con-structions in English: regular plurals and past tenses, as well as the third-person singular present of regular verbs, all formed in the morphology;possessives such as Rick’s office, formed in the syntax, and sequences of a hostword plus reduced auxiliary clitic. All display the same relation to normalsyllable structure, and all display the same phonological modifications of thesequences that result (cf. (2.24) in section 2.4 above).

Adjunction of the stray segment represented by a clitic auxiliary to the pre-ceding syllable as an affix is thus possible. Given other constraints, includingProsodic Faithfulness and the impossibility of syllable-initial affixes, this is in-deed the only possible way for these elements to satisfy Full Interpretation. Itis also exactly what we expect, given the clear evidence which has already beennoted that the reduced auxiliaries depend phonologically on material to theirleft.

But now notice another consequence of this manner of incorporating theclitic auxiliaries into prosodic structure. When the auxiliary would otherwisebe PPhrase-initial, adjunction to a preceding syllable effectively shifts it out ofthe PPhrase to which it would otherwise belong, since the preceding syllablebelongs to a distinct PPhrase, and a single syllable cannot be shared by morethan one higher-level constituent.

This in turn suggests an approach to the ungrammaticality of (3.51a). Inthis sentence, the final PPhrase which is motivated on syntactic grounds, cor-responding to the syntactic phrase [

VP’s [e]], would have the structure [

PPhz].


But after the restructuring required to incorporate the stray element /z/, thisPPhrase would have no phonological content at all. We can thus suggest thatwhat blocks the use of clitic auxiliaries in such contexts is the fact that theywould lead to violations of (3.53), a constraint which is arguably a universalwell-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 contractedauxiliaries are not possible, it must be established that the prosodic structureof the relevant sentence types contains a PPhrase whose only phonetic contentis a reduced auxiliary, an element that is necessarily aligned out of its originalPPhrase as a result of Stray Adjunction. This, in turn, requires a full account ofthe principles that construct prosodic representation on the basis of syntacticstructure in English. I do not attempt a comprehensive description of this sorthere, but rely instead on the proposals in Selkirk 1995.

To a large extent, prosodic and syntactic structure are similar. In par-ticular, Selkirk argues that the edges of lexical words and of PWords preferablycoincide.

(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 categoriespreferably coincide with the edges of PPhrases.

(3.55) LPhCon: Boundaries of lexical maximal projections should be PPhraseboundaries.

a. Align(LexMax,L,PPhrase,L)

b. Align(LexMax,R,PPhrase,R)

Furthermore, at least the right edge of a PPhrase is preferably aligned withthe right edge of a PWord.

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


These constraints do not by themselves determine the way in which non-lexical items, or “function words,” are incorporated into prosodic structure,except through their interaction with the constraints of the prosodic hierar-chy more generally and the condition of Full Interpretation. The result is thatPPhrase 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 precedethe lexical head of a phrase are incorporated as Free clitics, attached directly tothe PPhrase. Selkirk shows that the correct structures follow from the rankingin (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 ma-terial comprising a syntactic XMax. This PPhrase may also include followingfunction words belonging to the same item, like the at of look at, over whichPWords are constructed as just noted. The lexical material is preceded withinthe PPhrase by Free clitic function words belonging to the same or an includingsyntactic phrase until the right edge of another PPhrase is reached.

Consider now some of the sentence types from which contracted auxil-iaries 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] ofSchopenhauer.

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 justbefore the auxiliary is. This auxiliary, in turn, is immediately followed bya syntactic gap, and it is exactly the claim of the standard account of con-tracted auxiliaries that they are excluded when followed by a syntactic gap.But attention to the prosodic structure shows another property of these struc-tures: in each case, a new PPhrase clearly begins in pronunciation immedi-ately after the gap. We could say, then, that the conditions which exclude con-tracted auxiliaries do not refer directly to the syntactic gap per se, but ratherdepend 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 suchas (3.59).


(3.59) Align([e],R,PPhrase,L): Syntactic gaps are followed by a PPhraseboundary.

This constraint encourages the construction of a PPhrase in each of thesentences 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 withother phrase-final function words. If the contracted form is chosen, however,Stray Adjunction forces it into the preceding PPhrase, producing a violationof (3.53) and rendering the sentence unacceptable.

If this were the whole story, the resulting analysis would be a fairly minorvariation on the traditional picture, which relates the ill-formedness of a con-tracted auxiliary to the presence of a following syntactic gap. That is not thecase, however. My claim is that the factor which excludes the reduced auxil-iaries 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 gapsblock the appearance of reduced auxiliaries, despite the fact that the gaps inquestion do not immediately follow the position of the auxiliary element, asin (3.60).

(3.60) Comparative Subdeletion: Marie’s a better scientist thanPierre is/*’s an [e] engineer.

Subject-Aux Inversion: Marie’s better known than is/*’s herhusband [e].

In these cases, we can note that the material following the auxiliary elementis set off by the left edge of a new PPhrase. In the case of Subject-Aux Inver-sion, this phrasing follows from the general principles we have discussed. It isnot obvious that the same is true for Comparative Subdeletion, but regardless,it is clear that this phrasing is associated with the construction. As a conse-quence, in each of these structures the auxiliary element will find itself alone ina PPhrase beginning at its left edge, such that its phonological attachment tothe preceding word will produce a violation of (3.53). Only the selection of thefull form of the auxiliary (over which a PWord can be built, given its PPhrasefinal 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 dislocationsite 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 ar-gued that phonologically empty syntactic categories are not visible to the processes that constructprosodic structure.


Indeed, contracted auxiliaries are excluded even in structures with no (rel-evant) gap at all, but where violations of the condition (3.53) arise. Bresnan(1978) observed that a following parenthetical expression as in (3.61a) blockscontracted 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 sup-posed cliticization to the right, but in fact the clear phrasal break around par-entheticals suffices to explain the phenomenon. Whether the contracted formis also blocked following the parenthetical depends on prosody: if pronouncedwith a clear break before the following phrase, (3.61b) can only have the fullform of the auxiliary, but if no such break is present, many speakers accept thecontracted form as well. This makes clear the dependence of the contractedauxiliary on prosodic, rather than syntactic conditions. The same is true ofthe 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 ofAuxiliary Reduction in English. Indeed, there is no syntactic condition at allper se that governs the appearance of clitic auxiliaries.16 The role of the syntaxin this area is limited to the effect of syntactic constituency in the constructionof prosodic structure. All we need to say about the auxiliaries is that each ofthese have two forms, either of which can in principle be chosen freely. Ifthe full form is chosen, it will behave either as a final or a non-final functionword, depending on its prosodic position. If the reduced (non-syllabic) formis chosen, it will necessarily undergo reattachment as an affix to a precedingsyllable, resulting in a violation of (3.53) where this results in a phoneticallyempty PPhrase. The reduced forms of English auxiliaries are thus simple orphonological clitics in these sense of this book.

16 Kaisse (1985) argues for additional conditions on the relation between the auxiliary and pre-ceding material, specifically the “Government Condition” which stipulates that auxiliaries maycliticize only onto a constituent that they govern. This would block sentences such as Speakingtonight’s our star reporter to which Kaisse assigns a star, but which I and nearly all speakers withwhom I have discussed the matter find perfectly acceptable. If some such condition neverthelessexists, I would interpret it as governing the formation under appropriate syntactic circumstancesof prosodic constituents at higher levels, such as the IntPhrase, whose presence blocks the incor-poration via Stray Adjunction of the clitic auxiliary into the preceding PPhrase, resulting in itsstranding in a prosodically ill-formed position.


Let us now return to the observation that children acquire the use of thecontracted auxiliary forms in English rapidly and essentially without error. Onthe traditional account, this is quite mysterious. If “Auxiliary Reduction” is aprocess specific to English, depending on a comprehensive analysis of rathersubtle syntactic factors, including as a central component the role of syntacticelements which are not manifested overtly in sentence form, it is hard to seehow the learner could arrive at the correct analysis as directly as seems to bethe case.

My claim is that the phonology of these clitics depends only indirectlyon syntactic form, and only to the extent this is a significant determinant ofprosodic form. Syntactic phrasing is not always faithfully reflected in phono-logical phrasing, but neither is the relation between the two arbitrary and un-constrained. Children are sensitive at a very early age (around eight to tenmonths)17 to the phonological organization of utterances into phrases, andthere is no doubt that they use this structure as a crucial key to discoveringthe syntactic organization of sentences and phrases. In order to do this, theymust assume that syntactic phrases correspond largely to phonological ones.It seems clear that the prosodic structure is available to the learner before thesyntactic form.

On the prosodic analysis, there is no mystery to the rapidity with whichuse of the clitic auxiliaries is acquired. All the child needs to learn is thefact that certain function words have optional variants which are prosodi-cally deficient (i.e., phonological clitics). This information is directly avail-able in surface forms, since children hear both The tiger is in that cage andThe tiger’s in that cage and have no reason not to treat them as optional vari-ants. 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 knowl-edge provides them with the phonological apparatus to account for the pho-netic forms of the clitic auxiliaries. They must also gain access to the prosodicstructures within which these regularities play out, but as noted above, thisinformation is probably in place before a complete syntactic interpretation isarrived at. No language-specific conditions, syntactic or otherwise, govern thecontracted auxiliaries per se, once it is determined that they are phonologicalclitics.

An understanding of the significance of prosodic structure can probablylead to the resolution of another chronic puzzle in English, the analysis of con-tracted forms such as wanna, gonna, hafta, and a few others. These representcombinations 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.


It is widely assumed that the presence of a syntactic gap intervening betweenthe base word and to is responsible for blocking the “Contraction” in cases suchas 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 outwhether 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 numerousauthors to attempt to assimilate this phenomenon in some way to that of thecontracted auxiliaries.

Arguably, however, it is not the presence of a gap per se that blocks theuse of wanna, gonna, hafta in these examples, but rather a fact about prosodicstructure. Notice that in pronunciation, there is a clear PPhrase boundary pre-ceding the to in each of the sentences where the contracted forms are impos-sible. No such boundary is necessary, however, in the cases where contractedforms can be used. Since this is not a book about the syntax of complementa-tion, I will not propose explicit structures from which this intonational differ-ence should be derived, but there is little doubt that structural distinctions arepresent 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 con-tracted forms consists of two syllables, forming a trochaic foot. If they have thisprosodic structure already at the point where they are inserted, it is clear whytheir use will be blocked when a PPhrase boundary should intervene betweenthe syntactic positions of want, going, have, etc., and that of to. The prosodichierarchy prevents a single foot from belonging simultaneously to two distinctconstituents at a higher level, such as the PPhrase. As in the case of the reducedauxiliaries, no reference to a condition on syntactic structure is necessary hereat all, but only an appreciation of the way such structure may influence theformation of prosodic representations.

Prosodic structure, then, plays an important part in governing the well-formedness of sentences. For present purposes, the most significant aspect ofthis component of linguistic form is the fact that phonological (or “simple”)clitics, material without the prosodic structure of normal lexical items, must


be integrated into the prosodic hierarchy through Stray Adjunction and thephonological adjustments that may arise as a consequence. The study of theseeffects across a variety of languages forms the content of one of the two distinctdimensions 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 sec-tion 2.1. These were defined as unaccented bound forms that are variants offree 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 fromthe corresponding free form, as in the French sentences in (4.1). Assumingthe underlined expressions in these sentences are intended to refer to the sameindividual, when that reference is made by a full DP, this phrase follows thefinite verb, while the corresponding argument appears pre-verbally when it isa clitic pronoun.

(4.1) a. Je vois l’homme dont je t’ai parléI see the man of whom I to you have spokenI see the man I told you about

b. Je le voisI him seeI see him

Alternatively, it might be something more unusual such as obligatory ap-pearance in second position, as in the Ngiyambaa examples in (4.2).

(4.2) a. Nunhi=lu=na Nadhigave-3erg-3abs me-oblHe gave it to me

b. Ninu: Na:nhi Nadhuyou-obl saw I-nomI saw you

In Ngiyambaa (Donaldson 1980), bound pronouns are all located togetherin a fixed sequence in second position within the clause, although the orderof other words in the clause (including non-bound pronouns) is free. Theplacement of the pronouns in (4.2a) is thus the only one available, while their


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

4.1 Phenomenology

A useful initial step toward understanding the morphosyntactic dimension ofclitic structure would be a survey of the phenomena, in the form of a catalog ofthe varieties of special distribution which clitics may show. After some initialclarification of terminology, I turn to such a survey, as a prelude to the questionof 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 shouldprobably be factored out of Zwicky’s definition. For one thing, to the extentspecial clitics are unaccented, this can be treated as following from the fact thatthey may be phonological clitics as well, independently of their distinctive dis-tribution. I propose, then, that rather than treating simple clitics and specialclitics as two distinct and non-overlapping categories, we regard the phono-logical and the morphosyntactic dimensions of clitic behavior as separable. Infact, most special clitics are also phonological clitics, but the two properties arelogically quite distinct and neither necessarily implies the other.

Furthermore, not all special clitics are unaccented, as we have seen. Forinstance, in Lucanian (as discussed in section 3.2), stress may appear ona pronominal special clitic because in this language stress is always pen-ultimate within the PWord, and clitic pronominals are incorporated intothis constituent—where they may receive stress.

It is by no means obvious that all special clitics are also phonological cli-tics, though. Recall that the basis of phonological clitic behavior is prosodicdeficiency (failure to constitute a PWord, in particular). Where a clitic consistsof isolated segments, or a single light syllable (in a language whose minimalPWord consists of two syllables or two moras), such a diagnosis is obvious. Insome instances, though, a special clitic may contain enough material to con-stitute 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 havemore to say in Chapter 6) and Italian loro.

Even in such a case, however, it is still difficult to say with certainty that theelement 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 secondposition within the clause, as illustrated by saku ‘2sg.sbj;1sg.obj’ in (4.3).

(4.3) Wada =saku ini qtayipret 2sg→1sg neg seeYou 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 unitsget stressed quite regularly. Furthermore, according to Holmer, the word pre-ceding one of them undergoes ‘end of word’ phonology. This certainly suggeststhat the clitic constitutes a new word, but in fact it is consistent either witha two-word PPhrase (e.g., [

PPh[PWd

wada][PWd

saku]]) or with a recursive

PWord structure (e.g., [PWd

[PWd

wada]saku]). To distinguish between these

possibilities (and thus to establish the prosodic status of the clitic), we needevidence from distinctive sandhi at the boundary between two PWords.

An example of the appropriate sort is apparently furnished by Warlpiri. Inthis language, as is well known, auxiliary elements and pronominal clitics clus-ter together in second position within the clause. In addition to the auxiliarybases (ka and lpa as well as a hypothetical phonologically null base), comple-mentizers can also appear in this position. Given the contrast between thegenerally free word order of Warlpiri sentences and the rigidity of ordering ofclitics, auxiliaries and complementizers, all of these elements are undoubtedlyto be analyzed as special clitics.

While the auxiliary bases are monosyllabic, and thus prosodically deficientwith respect to the minimally bisyllabic Warlpiri PWord, the complementizers(with the exception of yi ‘for, since’) are bisyllabic, and thus at least candidatesfor independent PWord status. This status is supported by a fact noted byLegate (forthcoming): vowel harmony, a PWord-internal process assimilatingthe backness of high vowels in successive syllables, does not apply betweenthe sentence-initial stem and a complementizer. An example is given in (4.4),where the vowels that are underlined should harmonize in backness if theywere part of the same PWord.

(4.4) Nganayi kuja=lpa=lu liwanja-paju-rnu, . . .whatchamacallit decl.comp-pst.imperf-3pl.sbj fish-call-pstThat thing they called fish . . .

Further analysis of the precise interaction between prosodic structure andvowel harmony in Warlpiri is still necessary before such examples can be se-curely interpreted as instances of prosodically autonomous special clitics. I will


assume, though, that it is possible for a special clitic to constitute a PWord in itsown right, and thus that special clitics are not necessarily phonological cliticsas 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 “boundwords” which are just like special clitics, except that they are not variants ofany free form. That suggests that a clitic may or may not alternate with a freeform, but this is a lexical fact about particular elements rather than a definingproperty of a linguistically significant class of items.

In French, for instance, we have (clitic or ‘conjunct’) me, te, etc. but non-clitic (or ‘disjunct’) moi, toi. The latter appear in ordinary DP positions; theformer as clitics attached to the finite verb of the clause. On Zwicky’s defini-tion, it is the existence of the disjunct pronouns that authorizes us to call theconjunct forms special clitics, since we can treat the latter as weak variants ofthe former. I suggest, though, that it is a mistake to treat corresponding mem-bers of these two sets as alternate forms of the same lexical item. It is true thatthey are generally in complementary distribution, but this is due to the generalimpossibility in French of clitics that “double” or have the same reference as afull DP expression.1 This is true not only for disjunct pronouns, but for anyother 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 theroom’

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 thathappen to have the same reference, rather than one single item. Their mutualexclusivity within a single sentence results from the broader regularity that, forreasons 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 cliticpronouns to double full DP expressions, including disjunct pronouns. This does not affect thepoint being made here.

2 If the full DPs in these examples are set off intonationally, as right-dislocated elements, theyare perfectly compatible with a coreferring clitic, but this is not directly relevant to the impossi-bility of clitic doubling within the clause in (standard) French.


A more complete account of this (which I will offer in Chapter 8) is based onprinciples of anaphora, not syntactic movement, but need not occupy us atthis point.

A Typology of Special Clitics

What remains of Zwicky’s defining properties of special clitics, then, is theircharacteristic special positioning. A special clitic is thus a “little word” whosesyntax is not assimilable to that of full words that might seem to be syntacticallyparallel. To develop a theory of special clitics, we need to survey the range ofsuch “special syntax” that we might find. In work that has served as the classicfoundation 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 tosome domain with which it is (syntactically and semantically) associated. Inpractice, this yields three general sorts of clitic. First, there are sentence clitics,located with respect to an entire clause. This includes auxiliary elements, manysorts of pronominals, particles marking discourse or illocutionary force, andmany other sorts of clitics taking scope over the whole sentence. A second typeis specifically associated with nominal expressions (NP/DP clitics such as casemarkers, determiners, or possessives as in some Balkan and Uralic languages).Finally, we have clitics which can be associated with phrases of any type, asmarkers of emphasis, constituent negation, interrogation, or other similar op-erators. The syntax of any particular special clitic is thus characterized in partby the domain within which it is located (CP/IP, DP/NP, XP).

Relative to its domain, Klavans proposed two further parameters that de-termine where the clitic is located. The first of these, which I will call Anchor-ing instead of Klavans’s term Dominance, characterizes a clitic as oriented withrespect either to the first or to the last element within the domain. The secondpositional parameter then specifies whether the clitic is placed before or afterthe anchoring element.

These parameters accommodate four kinds of clitics in terms of their po-sition within a domain: initial and second position (before/after the first el-ement); pre-final and final position (before/after the last element). Three ofthese types are abundantly attested in natural languages: initial clitics, such asthe Kwakw’ala determiners we saw in Chapter 2, second-position clitics whichwill be discussed in much more detail below, and final-position clitics such asthe English possessive marker ’s.

From considerations of symmetry within the system, we would also ex-pect pre-final or penultimate position clitics, but in fact it is hard to find anyreal examples of this type. Klavans proposes (Kugu) Nganhcara (Smith and


Johnson 2000) pronominals as a possible instance, but there are two problemswith that analysis. First, in some sentences, the clitics in question are actuallyfinal, rather than pre-final. Secondly, in those sentences with pre-final clitics,they always immediately precede the verb (which is itself in final position). Itis therefore difficult to say that the positional regularity here is “before the lastelement of the clause” rather than “before the verb.” It is not unreasonable toask, in fact, whether there are any genuine examples of specifically pre-finalclitics, a question which I will note again below.

Are these in fact the only possibilities for clitic positioning? Sometimesit appears that rather than being anchored to the first or last element withintheir domain, clitics are anchored by the head element within that domain.Romance pronominals, for example, are clearly anchored by the finite verb oftheir clause (if there is one; alternatively, by the infinitive or other main verbin some types of non-finite clauses). The anchoring verb is clearly the head ofits VP (and when bearing Tense, Agreement, etc., of IP as well). For a non-pronoun example, Finnish –kin ‘also’ (cf. Nevis 1986) is a sentential clitic thatattaches to the right of the first tensed verb of the clause (as head of IP).

(4.6) Kalle on=kin ostanut autonKarl is-also bought carKarl also bought a car

Kaisse (1985), Nevis (1986), Zwicky (1987), and Anderson (1992), among oth-ers, have thus suggested the availability of a third value for the Anchoring pa-rameter, one which would allow for orientation with respect to the head of therelevant phrasal domain.

There is another way to look at these clitics, however. Instead of seeingthem as placed within the phrasal domain, and anchored by its head, we couldregard them as placed within a smaller domain: that of the head constituentitself (V or perhaps I).3 On this picture, the Domain parameter can be set toany category that is either a maximal phrasal projection or a lexical categorywhich projects such a phrase. There would then be no need to allow a thirdvalue for the Anchoring parameter.

In some languages, multiple possibilities for the positioning of a givenclitic may be instantiated. In most of the Romance languages, for example,pronominal clitics generally appear initially within the domain of the finiteverb, 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 domainof V is thus not to be confused with treating it as an inflectional prefix or suffix on the verb. Thisdifference will be significant in some instances to be discussed later.


in Portuguese, sometimes with finite verbs) they may instead appear finallywithin this verbal domain. Similar facts obtain in some South Slavic languages.To the extent the conditions for initial versus final appearance can be formu-lated exactly, this complication does not impugn the overall typology of cliticpositioning, however.

With respect to second-position clitics, an important point that will oc-cupy us at some length below is the fact that “second position” can have morethan one meaning. It often means “following the first syntactic daughter of theconstituent defining the domain,” but in some languages it may instead mean“following the first PWord, PPhrase, etc. within the phonological realizationof the domain.” The best-known example of this type is the clitic system ofsome forms of Serbo-Croatian, as described by Browne (1974), which will bethe subject of closer attention in Chapter 5. This possibility corresponds to thefact that (at least in descriptive terms) the anchoring element may be inter-preted either as a syntactic or as a prosodic object.

All of the possible placements for clitics considered so far assume that cli-tics are placed at the periphery of some anchoring element. Some exampleshave been proposed, however, in which so called “endoclitics” (in the termi-nology of Zwicky 1977) appear internal to a word. Putative examples includePashto (Tegey 1977; van der Leeuw 1995a; Roberts 1997) á-de-xist e‘you werebuying it’; Udi (Harris 2002) a-ne-q’-sa ‘she takes’ (root /aq’/); and Portuguesedar-no-lo-á ‘he will give it to us’. These cases will be examined more closely inChapter 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 direc-tion of their phonological attachment (proclitic or enclitic). One might expecta clitic to take as its host the element with respect to which it is anchored, orperhaps the element with which it is most closely associated syntactically, butin fact clitics may attach phonologically to an element which does not evenform a part of the syntactic domain in which the clitic is positioned. We havealready seen this with respect to Kwakw’ala determiners and pronouns (a casealso cited by Klavans). A closely similar example is provided by Yagua (Payneand Payne 1990), as in (4.7). In this language, Object DPs are immediatelypreceded by a coreferring clitic pronominal which attaches to the word thatprecedes it, regardless of the syntactic relation (if any) which that word maybear to the object.

(4.7) sa-púúchiy Pauro rooriy-v11mu=níí Anita3sg.sbj-lead/carry Paul house-inside-3sg.obj AnitaPaul leads/carries Anita inside a/the house

Another example is provided by Northern Vogul, as in (4.8) from Nevis(1990).


(4.8) Xum jot=ke åleGem naurem Xaniman with-if I live child clingsIf I live with a man, the child belongs to me

Here the clitic ke ‘if ’ is located immediately before the verb, but attaches to itsleft.

Similarly, in Kugu Nganhcara (Smith and Johnson 2000), pronominal cli-tics precede the verb but attach to the preceding word, as in (4.9).

(4.9) ngaya ku’an hingkurum ka’im=ngkurum kala-ng1sg.nom dog 2sg.abl neg-2sg.abl take-1sgI didn’t take your dog

These and a number of other examples of the same sort are discussed andanalyzed by Cysouw (forthcoming). According to Klavans, they motivate anadditional parameter of individual clitics, their phonological liaison. In sec-tion 3.3 above, however, I argued that the direction of phonological attachmentis 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 respectto its prosodic structure. I will therefore assume that we can dispense withany such parameter as an item-specific property within the typology of specialclitics.

In summary, an exhaustive typology of special clitic positions is given bythe parameters in (4.10).

(4.10) A clitic is located

a. within the Domain of some syntactic constituent (X0 or Xmax forsome value of X);

b. by reference to the first versus last daughter constituent of thatdomain (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 adequatetaxonomy of what we find in the languages of the world, but it does not go veryfar beyond simple observation. That is, it does not really tell us how specialclitics are to be accounted for in an explicit grammar. It does not even tellus directly where to look for such an account: in the syntax (as syntacticiansassume virtually without discussion), in the phonology (where we have already


found all that is apparently necessary to describe simple clitics), or perhapssomewhere else.

Certainly the most popular picture is the syntactic one, according to whichspecial clitics are introduced into sentences in the same way as other (lexical)items, and then displaced from where we might expect to find syntacticallycorresponding words to their surface position. On that view, the parametersof the preceding section become parameters that characterize a set of syntacticrules.

A priori, though, that is not a very attractive position. For one thing, as Ihave already noted and will discuss below in more detail, second-position cli-tics may come literally after the first word in some languages; and this (phono-logical) notion is not in general available to the syntax (which deals in phrasalconstituents). For another, this kind of “syntax” is quite incompatible with theway syntactic movement is otherwise characterized.

The notion of a set of special syntactic rules of the sort implied by this ty-pology is clearly a relic of the more general conception of syntax as groundedin a collection of construction-specific rules (“Passive,” “Dative Movement,”“Subject-to-Object Raising,” etc.). Most modern theories of syntax4 deploresuch rules and attempt to derive the properties of individual constructionsfrom more general properties of structure and displacement operations. Thismight, of course, turn out to be impossible, but our first impression surelyought to be that if a syntactic theory of special clitics must take this form, weshould consider the possibility that some other part of the grammar might bebetter suited to the purpose.

The alternative which I will pursue in this book is the proposal that cliticsare not syntactic objects at all, but morphological ones: overt morphologicalmarkers of the morphosyntactic properties of phrases. Clitics on this view arephrasal affixes. Later chapters will be devoted to filling out and justifying thisview, but to establish its a priori attractiveness, let us compare the typologyof special clitics developed above with the typology of morphological markersrealizing the morphosyntactic properties of words: primarily (but not exclu-sively) the typology of affixes.

Most affixes are either prefixes or suffixes of course, but there are alsoinfixes. 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 “ConstructionGrammar” (Goldberg 1995). Various other syntactic theories also admit at least a limited amountof parochial, construction-specific structure in syntax, without going so far as to suggest thatsuch formations constitute the core of grammatical knowledge. I take it for granted that anythingthat can be reduced to general principles ought to be, and that rules which directly stipulate theproperties of a construction are an admission of failure in this endeavor—perhaps necessary, butsurely not desirable.


find morphological markers infixed not just anywhere in a word, but only ina 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 elsethey precede the last element of such a type in the word.

This of course looks rather like what we saw with special clitics, and es-sentially 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 ofinfix are located after the first or before the last.

Among affixes, there is an analog to the issue of the domain within whichclitics are placed. Recall that some special clitics are located within the domainof the head of a phrase (e.g., Romance conjunct pronominals). The same canalso be true for affixes. It is suggested in Anderson (1992) that words some-times have internal structure, as with Icelandic [[kalla]st] ‘to be called s.t.’ orGeorgian [mo[k’lav]] ‘(will) kill’. In such a case the affix may be located as aprefix (or suffix) to the internal head of the word, as in [[köll-um]st] ‘we arecalled 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-finalposition are comparatively quite rare. In this they are rather like penultimateclitics, which may exist, but as a distinctly unusual type. This similarity isof course only a tendency (assuming penultimate clitics exist at all), but it isinteresting 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 matterof ordering. Within a word, the order of morphological markers is virtually al-ways strictly determined and invariable, holding meaning constant. Of course,the passive of a causative will typically differ in the order of markers from thecausative of a passive: this is the basis of the “Mirror Principle” proposed byBaker (1985). But for given content, no known language allows the order ofaffixes to vary freely. In contrast, practically all languages allow for some free-dom in the ordering of constituents or even individual words while preservingsemantic content.

When we look at clitics, we find that their order is largely fixed, like that ofmorphological affixes, and essentially never free, like that of syntactic units.The contrast can be stark within an individual language. In Warlpiri, forexample, word order is extremely free, but the one constant is the fact thata set of special clitics (including auxiliary elements and pronominals) mustcome in second position, and must conform to a strict internal template. Inthis respect, special clitics clearly behave much more like morphology thanlike syntax.


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

(4.11) a. Só o Pedro o viuonly art Pedro him= sawOnly Pedro saw him

b. *Só o Pedro viu-o

c. Viu-o só o Pedrosaw=him only art PedroOnly Pedro saw him

d. *O viu só o Pedro

Other alternations in order have already been mentioned above in otherRomance and Balkan languages, where clitics often precede the finite verb butfollow non-finite and imperative forms.

The alternation in position which we find in these cases is nothing likefree word order or other “Scrambling” phenomena, however. Rather, the or-der which is strictly required under one set of circumstances is replaced byanother under a complementary set of conditions. In the Portuguese case,pronominals appear in enclitic position when the finite verb is initial within itsclause (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 thesame affix may show up either as a prefix or as a suffix, depending on specificfactors. 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 aconsonant or the low vowel [A], but as prefixes when the verb stem begins witha 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 isnotable 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 onsome other morphological category which is copresent. For instance, in Huaveas described by Noyer (1994), a theme vowel appears as a suffix to the stem indetransitivized forms but otherwise as a prefix. Both clitics and word-internalaffixes only display alternations in order under clearly specifiable conditions,


as opposed to the much greater freedom of word (or constituent) order foundin many languages.

I have suggested that post-initial infixes are analogous to second-positionclitics, both occurring after the first element of some specified type within thedomain whose properties they specify. In general, the anchoring element forword-internal infixes is prosodically defined: a consonant, syllable nucleus,syllable, foot, etc. In at least one case, however the parallel with clitics iseven more striking. Nevis and Joseph (1992) argue that a reflexive/passive affixin Balto-Slavic (Lithuanian –s(i)) occurs precisely in a word-internal secondposition, as illustrated in (4.12).

(4.12) No prefix: matymasi-s ‘seeing each other’; kalbúo-si ‘I converse’

One prefix: pa-si-matymas ‘a date, meeting’; be-si-kalbant ‘whileconversing’

Two prefixes: su-si-pa-žìnti ‘to become acquainted with’; pri-si-pa-žìnti ‘to confess, avow’

Nevis and Joseph call –s(i) a “Wackernagel affix,” and establish that it ispart of the same PWord with the stem and other affixes. I suggest that thisis indeed a case of “suffix to the first element in the word.” Assuming thatthe prefixes after which –s(i) appears are structure-building, like the Georgianand Icelandic affixes discussed above, the element that counts as “first” is astructural one. This account is consistent with dialectal variation which theymention in whether the reflexive marker, when it comes after the unprefixedstem, comes before or after inflectional suffixes. The similarity between thisaffix and second-position clitics is thus unusually striking.

Just as not all morphological markers are concatenative affixes, there arealso clitics that are processual (or “non-concatenative”). I will discuss belowthe analysis of the “definitive accent” in some Polynesian languages. In Ander-son 1992, I also cited in this connection the marking of (some) accusative ob-jects in Welsh by soft mutation (a non-concatenative morphological process)at the left edge of the nominal phrase. Another interesting example is furnishedby some languages of the Algonquian family.

Algonquian languages such as Potawatomi display person-marking pre-fixes at the left edge of the verbal phrase. Though these have often beenregarded as inflection markers on the verb (and were so treated in Anderson1992), Hockett (1948) showed that they should actually be analyzed as phrase-initial clitics, a view also adopted by Halle and Marantz (1993). In theirpositioning, the Potawatomi personal pronoun clitics are interestingly simi-lar to another element in Algonquian morphosyntax which appears in several


languages of the family: a replacive vowel change affecting the leftmost vowelin the same verbal projection as that to which the personal proclitics are at-tached. This “initial change” (to adopt the terminology of Bloomfield 1946 andelsewhere) functions as a sort of complementizer in marking certain subordi-nate constructions, as in (4.13). Vowels affected by initial change (or related toothers that are) are underlined.

(4.13) Potawatomi: a. Pe ki mpotprt past diedHe died

b. ka mpotpast (< ki) diedthe one who died

Menomini: a. ne=pa:pam-nato:nE:hokme=going about-he seeks meHe goes about seeking me

b. (emeq) pE:pam-esiat(yonder) going about-he goesIt is over yonder that he goes about

Fox: a. e:shki kano:n-ehkafirst (<ashki) speak-3sg to 2sg

The one who first spoke to you

b. ke:no:n-ehkaspeak-3sg to 2sg

The one who spoke to you

Non-segmental apophonic relationships such as the initial change illus-trated in (4.13) can serve functions which might in other languages be filledby a phrasal clitic: here, that of a complementizer. If we think of clitics asthe phrase-level analog of word-level morphology, it is natural to expect suchphenomena, on a par with comparable markers in the form of words. Interest-ingly, both sorts of clitic here—the initial change and the person markers—arelocated at the left edge of the same projection of V (VP or V).

Reduplication is a type of formal relationship between forms that occurs inmany languages as a marker of morphological categories at the word level. Anexample where this same formal device marks a category at the phrase level isfurnished by Chamorro (Chung 2003). In this language, a predicate phrase XPof any one of several types is marked for Progressive aspect by reduplication


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

(4.14) a. Ni hagu mismu ti un-li’i’ amanu un-hanánaguieven you self not agr-see where? agr-go.to.progEven you yourself do not see where you are going

b. Ti atrasásaonot agr.late.progHe’s never late

c. Á’aguaguat =ha’ na patgun =gui’ esta pa’gunaughty.prog emph linker child he until nowHe’s still a naughty child

d. Ma’estrótro-nña =ha’ si Juan si Mr Perez?teacher.prog-agr emph art Juan art Mr PerexIs Mr Perez still Juan’s teacher?

e. Ginin= i= gima’-yúyu’us si nana-ñafrom the house-God.prog art mother-agrHer 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 morphol-ogy 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 uswith a further analog at the phrase level to a morphological type which is wellestablished in the analysis of words.

It looks, then, as if there are a number of substantive parallels between themorphological markers we find in the structure of words, on the one hand, andspecial 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 ofthe onset, but not the coda: e.g., tristi ‘sad’ gives (for some speakers) triristi ‘(still) sad’ and (forother speakers) tritristi; planta ‘set (the table)’ gives plalanta ‘set (prog)’ for some speakers andplaplanta 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 thusappear as proclitics attached to the following PWord.


same direction are also worth noting. First, both special clitics and affixes aregenerally deficient in prosodic structure, though in both cases there are a fewinstances with structure at least up to the level of the foot. In this, both distin-guish themselves from ordinary lexical items, which typically project a PWord(in most languages). Second, it seems useful to recognize a distinction be-tween ‘inflectional’ (grammatical, functional) markers and ‘derivational’ onesboth for clitics and for word-level morphology. Somewhat remarkably, thisdistinction (which in both cases can be grounded in the architecture of gram-mar) correlates with a difference in ordering, with inflectional elements com-ing “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 oftheir possible positions of occurrence. It was precisely the unusual placementof these elements, indeed, that led to the positing of a class of special clitics inthe first place. These factors suggest the plausibility of treating special clitics assomething other than syntactic objects manipulated by syntactic rule, and inparticular, 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 bytreating special clitics not as lexical items inserted and moved around withinthe syntax, but rather as phonological material inserted (like affixes) into thephonological content of a phrase. Special clitics express the morphosyntacticproperties of a phrase (the content of its functional categories, more or less),or else indicate (in the “derivational” case) modifications to the semantics, dis-course properties, and the like, of the phrase. It remains, of course, to constructan 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 concreteexamples of special clitics in a bit more detail. I will consider three cases: the ’sthat marks nominal possession in English, a kind of “poster child” of specialclitics; a potential example of a non-affixal, or processual determinerelement in some languages of Polynesia; and finally, the special clitic systemof Kwakw’ala, expanding on the description of simple clitics in that language inChapter 2.

Case I: The English Possessive

Virtually every discussion of clitic phenomena involves at least a passing refer-ence to the possessive marker ’s in English. Some scholars simply take it as self-evident that ’s is attached as a clitic to the right edge of a nominal expression,


and it is easy to see how immediately appealing this analysis is. As opposed tolanguages that mark possession by inflecting the head of the nominal in an ap-propriate way (usually called “genitive case”), the possessive marker in Englishattaches freely to heads and non-heads alike, provided only that their host belocated 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 inwallpaper is appalling.

e. Even that attractive young man who is trying to flirt with you’s tastein wallpaper is appalling.

In its fondness for this “group genitive,” modern English is somewhat unusualamong the languages of the world.

Assuming that the right account of clitics is one that treats them as phrasalmorphology, it seems straightforward to describe the English possessive interms of a feature—let us call it [Poss]—which is assigned to DPs. The distri-bution of this feature is also straightforward: it is assigned (as what Anderson1985a describes as a “configurational” property) to a DP that appears as the de-terminer of another DP. We can then say that a rule of phrasal morphology in-troduces a phonological marker /z/ at the right edge of a string of phonologicalmaterial whose syntactic analysis is [DP,Poss]. The resulting string undergoesappropriate phonology, so that /z/ appears phonetically as [s], [z], or [1z] de-pending 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 ofthe process by which the marker is introduced into the representation, thoughit is worth highlighting the assumption that phrasal categories (like DP) havean internal feature structure that can include elements like [Poss] which arenot simply markers of the type and bar level of the phrase.

If this were all there were to the English genitive, we could pass quicklyon to other matters, but a fairly abundant literature (including, but not at alllimited to, Allen 1997; Carstairs 1987; Stemberger 1981; Stump 2001; Zwicky1987) 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 thebasic phrase with an appended /z/. That is the case, notably, for DPs consistingsimply 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


that have a distinct possessive form, and since this is unpredictable (supple-tive), it is presumably listed in their lexical entries.

Let us assume that pronouns originate not as the heads of NPs, but ratheras determiners without NP complements, heading DPs—an analysis originat-ing in Postal (1966) which has returned to vogue in recent years. Then theexceptional possessives are exactly those DPs with the feature [Poss] whichconsist of nothing but a determiner. Assume further that a principle of Englishgrammar requires the feature [Poss] to be inherited by D from the DP to whichit 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 featurecomplex [D,Poss] can be interpreted by an item from the lexicon when this isa pronoun—and only then—and I assume that some principle of disjunctionblocks 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) determinerdo not generally take the possessive /z/ either: *These’s illustrations are morecompetently drawn than those’s, *Of the books I lent you, two’s/some’s/many’scovers were soiled when you brought them back, etc.7 This suggests that a DPconsisting 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, therule introducing the possessive clitic does not apply to it, and the only way thefeature [Poss] can be realized in this case is if a lexically listed form exists toserve that function. Non-pronoun determiners have defective paradigms inthe sense that they have no listed possessive forms.

Note also that the possessive pronouns have special forms when they serveas determiners of a DP containing a null NP complement: I want mine, yours,his, etc. These are presumably lexically listed and associated with the specificenvironment [

DPD [

NP∅]].

The fact that [Poss] is realized sometimes by the form of an individualword (for pronouns) might be seen as evidence against the view that it is aphrasal affix, but in my view this is actually interesting evidence in favor ofthis position. It is significant that the same feature shows up sometimes as adeterminant of the morphology of a word and otherwise as a marker clearlyappended to the right edge of a phrase. This argues that although a givenproperty may be realized in more than one way, that realization is still a matterof 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 thenotion that since one also serves as a pronoun, it might have a listed genitive like other pronouns.


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 itas an inflectional element that is confined to an element at the right edge of aDP. On this view, ’s is a word-level affix, capable of occurring on any word ofthe language (or at least any word that can end a DP), realizing an “edge fea-ture.” Aside from the conceptual difficulties posed by that notion, note that ’scan 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 personwho cooked mine’s hairs are floating in my soup.

c. Fred’s mother’s walk-in closet alone is bigger than mine’s entireapartment.

If [Poss] were regarded as a feature assigned to the word that occurs at theright edge of a DP, further apparatus would be needed to ensure the correctforms in these cases.

There is a further set of environments in which the possessive ’s is notphonologically realized. When ’s is added to a noun which is already markedfor 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 bedue to a general loss of possessive marking with plurals, since plural formationsother than /z/ are perfectly compatible with the marker of the possessive: thethree children’s/women’s/deer’s/mice’s etc., feet.

Beyond the combination of the /z/ plural and the possessive, there is somedisagreement about the exact facts here. Zwicky (1987) says that /z/ is lost afterany word that ends in an inflectional /z/. For him, this is true not only whenthe possessive is added directly to a regular plural (the two boys’ books) but alsowhen 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 toa single /z/. An instance of /z/ is not, however, lost following a stem that endsphonetically in the same way as one of the allomorphs of /z/, as shown in (4.18).


(4.18) 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 pho-netic degemination. In fact, however, there is a fairly direct way to formulatethe necessary simplification of sequences of /z/s. Recall the proposal I made insection 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 toa 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 isthus as in (4.19).

(4.19) σ

σ

d O g z

As a result, suffixal /z/ is structurally distinct from stem-final /z/, and thereis no difficulty in formulating a regularity that reduces two identical syllabicaffixes to a single one.

This is still not the end of the story, however. Zwicky (1987: 140, n. 6) notesthat “POSS is occasionally suppressed in speech (as it regularly is in writing, ac-cording 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 segmentof these names is treated by some speakers (or writers) as a syllabic affix, ratherthan as part of the stem. Those who say (or write) Jeeves’ rather than Jeeves’s,that is, represent the name lexically as [

σ[σjijv]z]. As a result, the addition

of a further affixal /z/ invokes the reduction process just discussed. Speakerswho say Jeeves’s, on the other hand, have a lexical representation consisting ofa single unaffixed syllable [

σjijvz].

Carstairs-McCarthy (1995) rejects Zwicky’s claim that the multiple affixa-tions 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 twoboys’s caps is just as bad for him as it is for Zwicky (and virtually all otherspeakers of English). It is only when /z/ is added directly to a regular pluralnoun that affix reduction must take place for Carstairs-McCarthy: elsewhere,it is at most optional.


A proposal for describing the grammar of a speaker for whom the multiplyaffixed forms in (4.17) are acceptable might go as follows. We must recognizethat while the /z/ affixes are added in a structurally adjoined position phono-logically, they must eventually be incorporated into simple syllables by thetime they are produced phonetically. Assume, therefore, a post-lexical processof Syllable-affix Incorporation. If this process is allowed to apply cyclically,it might already have applied on an earlier cycle when another affix is intro-duced later, with the result that the earlier affix no longer occupies an adjoinedposition, 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 al-ways identifiable as such for him, and always undergo reduction when morethan one is present on the same word form.

For Carstairs-McCarthy (and many other speakers), however, the facts aremore complex. A possessive suffix added directly to a plural noun will alwaysundergo reduction, because there is no intervening cycle on which the pluralcould have undergone Syllable-affix Incorporation so as to lose its identity asan 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 mayhave incorporated the affix of an included plural noun so that the reductionof multiple affixes is no longer applicable. There is obviously some work to bedone to flesh out this suggestion fully, but it seems to show some promise fordealing 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 complicatingfactors which have been cited in the literature can in general be treated as theresult of (a) the transfer of the [Poss] feature to a pronoun D head when this isthe only content of the DP; and (b) phonological processes reducing multiplesyllabic 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 consti-tutes an example of a non-affixal clitic: a process which serves as a phrasalmarker comparable to other clitics. Closer examination shows the facts to bemore 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, *thequeens of England’s, *the men I mentioned’s. I agree to some extent with this judgment, attributedby Zwicky to Kruisinga (1932) and partially invoked by Carstairs-McCarthy (1995) as well, but Icannot see how to make it follow from anything else in the present analysis.


(or perhaps a cognate of it found in another language, Rotuman) as a proces-sual special clitic. The definitive accent is described carefully by Churchward(1953); its semantics and usage are analyzed by Aitchison (2001) within theframework of HPSG. Other relevant literature will be noted below.

In Tongan, the normal location of stress is on (the syllable containing) thepenultimate mora. Definiteness9 is marked by a stress shift to the final mora ofthe entire DP. The examples in (4.20) show that the stress shift occurs only onthe last word of the DP.

(4.20) a. kuo maumau ’a [e salioté ]perf broke abs art cart-defThe cart is broken

b. kuo maumau ’a [e saliote ’a Feletí ]perf broke abs art cart gen Fred-defFred’s cart is broken

c. kuo maumau ’a [e saliote ’a Feleti mo Sioné ]perf broke abs art cart gen Fred and John-defFred 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-defI will go to the house in which they have assembled

e. na’a ke ’alu ki [he fakataha lahí ] aneafipast you go to art meeting big-def yesterdayDid you go to the big meeting yesterday?

The phrasal nature of this process is quite clear: it occurs at the right edgeof a DP without regard to the syntactic role of its “host.” Furthermore, thedefinitive accent is clearly analogous to other elements in Tongan that are un-controversially clitics: the demonstratives in the first column of Figure 4.1.

The demonstrative clitics ni ‘this’ and na ‘that’ are also introduced at theright edges of DPs which they determine. Their (phonological) clitic status isshown by the fact that (a) being monomoraic, they do not meet the language’sminimal PWord constraint, and thus must be stray syllables; and (b) they areevidently incorporated into the preceding PWord as Internal clitics, since they

9 This dimension is distinct from that marked by articles appearing initial within the TonganDP. 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 asdefinite, while normal noun phrases are indefinite.”


Clitic Pronoun Locative Adverb— e he pehe

this here like thisthat there like that

ni eni heni pehenithis this here like thisna ena hena pehenathat that there like that

Figure 4.1. Tongan demonstratives

cause the accent on that word to shift one mora to the right (from penult toultima): he falé ni ‘this house’, not *he fále ni.

Semantically as well as formally, it is tempting to see the definitive accentas 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 ofan original demonstrative *a that would fill that gap. Apparently, this elementcame to be assimilated to the quality of the preceding (word-final) vowel, andthe 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 aprocessual shift of the accent from penultimate to final mora.

These facts would not constitute an instance of a non-affixal special cliticif it were possible to maintain something like the “edge inflection” analysisexamined (and rejected) above for the English possessive marker. And in factsuch 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 involvedis deeply embedded in the phonology of the language.

The definitive accent involves a shift of stress which treats underlying longvowels as sequences of two identical moras. Poser assumes that a lexical rulereduces such sequences to a single long vowel at some point. This rule is arguedto be lexical since it requires access to properties of specific lexical items, andnot simply to phonological form. If the definitive accent rule precedes thislexical rule, it must itself be lexical.

There are several problems with this line of reasoning, though. First, thereis no reason to assume a rule collapsing sequences of identical (short) vowelsinto a unitary long vowel, if long vowels are represented simply as two segmen-tal positions linked to the same vowel melody, the standard assumption today(though not as generally at the time Poser wrote). The putative lexical rule ofvowel coalescence, then, is no rule at all: it is simply the phonetic interpreta-tion assigned to a sequence of vowel moras linked to the same melody whenthese are syllabified together.


Some of Poser’s arguments for the lexical nature of processes necessarilyfollowing 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 obser-vation that sequences of identical vowel segments do not coalesce—that is, inthe terms of the present discussion, are not syllabified together—when they areformed in certain ways. Specifically, Poser notes that vowel sequences formedby reduplication and by the prefixation of faka- do not undergo coalescence,as illustrated in (4.21).

(4.21) a. ongoongo ‘news’; ongoongoa ‘famous’

b. piko ‘crooked’; angaangapiko ‘somewhat crooked’

c. ava ‘open’; fakaava ‘to open’; fakaava’i ‘to open completely’

d. 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 penulti-mate position,10 extra stresses appear in several classes of words. Among theseare words with prefixes of two or more moras (such as faka- ), compounds,and fully reduplicated words. I assume that these extra stresses represent thefact that the components of such words are independently footed, with a bi-moraic trochee formed at the right edge of each part. When, e.g., fakaava ‘toopen’ is formed, then, the final vowel of the prefix faka- constitutes the weakmora in such a foot, and the stem initial /a/ which follows is the strong morain another foot. The requirements of the prosodic hierarchy prevent moras be-longing to distinct feet from forming a single syllable, and these two as are thusnot interpreted as a unitary long vowel. Even when main stress on the wholeword shifts one mora to the right in fakaava’i ‘to open completely’ the two ad-jacent as cannot be syllabified together, due to the pre-existing foot structure.

Similar reasoning in the other cases noted by Poser results in their elimina-tion as evidence for a lexical process of vowel coalescence, and a fortiori for thelexical nature of the definitive accent shift. As a result, there is nothing to pre-vent 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 havebeen assuming: by a shift in the location of the stress from the penultimatemora 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.


have been raised in the subsequent literature which appear to undermine thisaccount of the facts.

Recall that historically, the definitive accent appears to originate in an ear-lier clitic demonstrative *a which shifts stress one mora to the right, assim-ilates to the final vowel, and then disappears. But what if this element hasnot, in fact, disappeared in the modern language? We might then assumethat 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 thestress onto a final syllable, which now contains two moras. Assuming the newmora 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 aphrasal special clitic for the definitive accent, but since it has a segmentableform (an empty vocalic mora), it would no longer serve as an example ofa non-affixal clitic.

Distinguishing the two analyses at issue here would seem to be a simplematter: either the final vowel is short, in which case we have a non-affixalclitic, or else it is long, in which case we have a straightforward phrasal affixnot much different from the English possessive. Churchward’s description isfairly explicit in favoring the first interpretation, but Taumoefolau (2003), anative speaker, argues that the final vowel on which the definitive accent falls inTongan is geminated, or long. Measurements by Anderson and Otsuka (2002)do find some lengthening in these vowels. Furthermore, the related languagePukapukan has a very similar marker for definiteness, and Salisbury (2002)reports that the stressed final vowels are lengthened in this language. Theseaccounts would seem to favor the “empty mora affix” version of the Tongandefinitive accent over Churchward’s.

The most careful set of phonetic measurements available, though, seemsto suggest that neither account is quite right: the final stressed vowels in wordswith the definitive accent are shown by Condax (1989) to be longer than other(stressed) short vowels, but shorter than other (stressed) long vowels. Further-more, there are other differences between words with final definitive accentand words with final long vowels. One of these is the fact that the penulti-mate syllable (which would have primary stress in the absence of definitiveaccent) displays a secondary stress, yielding a stress clash (or sequence of twoconsecutive stressed syllables) that does not occur otherwise in the languageand whose exact phonetic properties are therefore difficult to project indepen-dently. In addition, there are pitch contours associated with definitive accentedwords that are similarly isolated within Tongan phonology.

Condax concludes that the definitive accent consists neither in a simpleshift of stress (as implied by much of the literature, based on Churchward’s


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 whichserves to mark definiteness. In the absence of a full analysis of the facts asCondax reports them, it is hazardous to venture a categorial interpretation.Nonetheless, it seems plausible to suggest that the definitive accent in Tongan ismarked by a process that applies at the right edge of phrase-final words, ratherthan by the addition of specifiable segmental material (an affix). As such, itdoes 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 discussionof the phonetic differences between phrases with and without the accent, in-cluding comparisons with the Tongan facts cited above, Salisbury (2002: 42)concludes that “the definitive accent is a grammatical morpheme which is notrealized 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 thepitch contour, but the resultant length is not phonemically equivalent to thatof a long vowel.”

And there is more. Tongan is not the only language that Churchwardworked on and in which he found phenomena of the sort we have just beendiscussing. In his Tongan Grammar (Churchward 1953: 268) he suggests that“[f]undamentally the distinction between the ordinary form of a Tongan wordand its definitive form corresponds to the distinction between the two ‘phases’of Rotuman words” and refers us to his grammar of that language11 (Church-ward 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, byweakening it, makes it indefinite, while Tongan takes the original form as in-definite (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 wordis comparatively definite, the weaker form comparatively indefinite.” Recentphonological analyses of the difference between the phases in Rotuman (e.g.,McCarthy 2000) make this analogy even more compelling by analyzing the in-complete phase as resulting from an accent shift applied to the complete phaseform.

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


The differences between the “complete” and “incomplete phase” formsin Rotuman have been the object of much discussion in the phonologicalliterature in connection with the appearance of metathesis, perhaps morpho-logically conditioned, in some of the relevant forms. There are several dis-tinct phonological changes involved in different sets of forms, but the choiceof one or the other is completely determined by the shape of the stem in-volved. 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 com-plete phase, and (b) a significant distinction between the two appears onlyat the right edge of a full DP. Non-final words within the DP all appear inthe incomplete phase, as do words preceding a specific set of suffixes and cli-tics. Examples can be found in the works cited above; it is uncontroversialthat the phase distinction appears only on the final word of thephrase, and so this is an excellent candidate for the status of a specialclitic.

Hale and Kissock (1998) propose, however, that there is no independentmarker of “complete phase” (or of “incomplete phase”) at all. On their ac-count, one set of affixes and clitics conditions the appearance of incompletephase forms, while another set conditions the appearance of complete phaseforms. The latter set includes, somewhat suspiciously, a marker –∅ ‘definiteplural/locative’ which contributes an abstract mora that never appears in sur-face forms.13 The distinction between the two phases, according to Hale andKissock, is nothing more than a complex but phonologically predictable con-comitant of the difference between the two sets of affix. McCarthy (2000) fol-lows essentially the same analysis, though differing with Hale and Kissock onother points.

However appealing a purely phonological analysis of the Rotuman phasedistinction may be, Kurisu (2002; see also Kurisu 2001) shows that it does notcompletely account for the facts. At least some affixes and clitics require aphase form of a preceding word which is not predictable from its phonologi-cal shape, and this includes one category with no formal marker (the definiteplural/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 specificformal 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 factthat a complete phase form with no overt marker serves as definite in contrast with a similarlyunmarked indefinite form in the incomplete phase. For Hale and Kissock this corresponds to theabsence versus the presence of this –∅.


Kurisu concludes that the alternation is triggered (at least in part) by mor-phological factors. But since its realization is at the right edge of a phrasalconstituent, this must be phrasal morphology: a special clitic. And if thisreasoning is correct, the resulting special clitic appears to provide us with yetanother non-affixal instance.

Case III: Kwakw’ala Special Clitics

A final and somewhat more elaborate example involving an entire system ofspecial clitics is furnished by Kwakw’ala. Much, but not all, of the complexityof this system is located in the distinctions made by its determiners, whoseanalysis was begun in Chapter 2 and to which I now return.

The pronominal simple clitics already mentioned distinguish only gram-matical function and person, but Kwakw’ala has a much richer set of distinc-tions that are carried by its determiners and pronouns. In addition to the basicproperties previously illustrated, the weak pronominals also display deictic dis-tinctions. In principle, at least, the form of every pronominal indicates whetherits referent is near the speaker, near the addressee, or distant from the speechsituation, as well as whether it is visible or not. Some dialects recorded by Boasalso indicate whether the referent still exists at the time of the speech event ornot, but these distinctions were not present in forms of the language which Iwas able to record in the late 1970s. As an example of the kind of distinctionthis system makes possible, the form –quP in (4.22) makes explicit the fact thatthing forgotten is presently near the addressee, but not visible to the speaker.

(4.22) la-x.@nt=@n ň’@lsPola=quPaux-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 showingthese distinctions is given in Figure 4.2.

Not only pronouns show this range of distinctions; full nominals do aswell. All of the deictic distinctions made by the pronouns in Figure 4.2 also

1st.vis 1st.inv 2nd.vis 2nd.inv 3rd.vis 3rd.invSubj: -k -gaP -ux. -uP -iq -iPObj: -q@k -x. gaP -qw -quP -q -qiInst: -s@k -sgaP -sux. -suP -s -si

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


appear in full nominal phrases. In these, however, the complete formal expres-sion 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 elementsis given in Figure 4.3. Notice that the two separate elements in a full determinercombine to give the full deictic category of the nominal expression.

(4.23) a. la-ň=@n qaq@s@l-is-ň lax.=gadaaux-fut-I go along-beach-fut to-dem(1st)aňi’nagwis=@kcountry-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.)?

c. ka’yadzoxiňa=x. -ux. wis=aq’drive away-obj-dem(2nd) boy-dem(2nd.inv)Drive away this boy (near 2nd person, invisible)!

d. la’laPi dux. ‘waň@la [email protected]=i=x.a

aux(evid) saw [email protected](3rd.inv)-obj(3rd)

gukw=ihouse-dem(3rd.inv)Then, it is said, Laqw@layug.

wa (name) saw the house(neither L nor house visible)

Complicating matters still further, there is yet another dimension of in-formation which is carried by determiners: that of possessor person. Where anominal is possessed, the determiner indicates whether the possessor is first,second, or third person. For third-person possessors, a further distinction ismade 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) -xwa, -x.ux. (da) -x. (a)Inst: -sga(da) -sa, -sux(da) -s(a)

Post-nominal: -k -ga -ix. -aq’ -∅ -a/i1st.vis 1st.inv 2nd.vis 2nd.inv 3rd.vis 3rd.inv

Figure 4.3. Kwakw’ala DP demonstrative clitics (unpossessed)


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

(4.24) a. Pix’m=is wałd@m=osgood-your word-dem(2nd poss of 3rd.vis)Your word is good

b. g.w@nał gax. -@n lax.=@n gagima-xd=a@n

pay to-me to (for)-my loan-past-dem(1st poss of 3rd.inv)l-oňto-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.@[email protected] wife-have-instr-dem(2nd.vis)k’idała=x. sprincess-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 thenominal. We could treat it as we have to this point, as an item appearing ina structural determiner position which is initial within the nominal. Alterna-tively, we could describe it as a left-edge special clitic, but the choice between

Demonstrative ofPossessor 1ST.VIS 1ST.INV 2ND.VIS 2ND.INV 3RD.VIS 3RD.INV

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 -is3RD, 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 -is3RD, POSS=SBJ, POST-N -k -gaP -q, -ix. -q’, -aq’ ∅ -a

Figure 4.4. Kwakw’ala DP demonstrative clitics (possessed)


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 theD position.

But what of the second, “post-nominal” component of these complex de-terminers? In the sentences we have seen to this point, we might describe thatas appearing either (a) at the right edge of the nominal expression, or (b) asfollowing the head noun, since all of the nominals in those sentences consistedexactly of a single word, the head.

When we consider nominals containing more than one word, though, itbecomes clear that neither of these accounts of the “post-nominal determiners”is adequate. The sentences in (4.25) illustrate this.

(4.25) a. m@x’id=ida ’walas=i b@gwan@m=x.a g@nan@mhit-dem(3rd) big-dem(3rd.inv) man-obj(dem) childThe big man hit the child

b. (vs.) m@x’id=ida b@gwan@m=a=x.a g@nan@mhit-dem(3rd) man-dem(3rd.inv)-obj(dem) childThe 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) houseShe goes home to her own house

d. g. ilagaP la=xwa k’is=ix. ał’@m wałdemacome to-dem(2nd) not-dem(2nd.vis) new wordCome to this not new speech

In fact, rather than being “right-edge” or “post-head” clitics, these el-ements are actually located in second position within the nominal, follow-ing the first independent word—regardless of that word’s function. Exampleslike (4.25d) show that it is indeed the first word of the DP (and not, for ex-ample, the first syntactic constituent of the DP, which in this case would be thetwo-word AdjP k’is ał em ‘not new’) that serves as the anchor for the seconddeterminer 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 basisin the syntax of the DP. If we assume that the ‘left-edge’ Determiner elementoccupies the D position, the other element is a second-position clitic withinthe complement of D. Note that these structural consequences are the sameregardless of whether we regard the nominals as basically DPs or as NPs, asshown in (4.26).


(4.26) a. DP

NP

D1 Word D2 X

b. NP

N

D1 Word D2 X

Special clitics of this type pose particularly serious challenges to standardtheories of clitic placement. The notion of ‘second position’ that they are basedon requires reference to a ‘first position’ that is filled by the first PWord ofthe domain (or in some languages another prosodic constituent such as thePPhrase: see the discussion of Chamorro in Chapter 5 below). These arephonological notions, and as such are generally assumed not to be access-ible to syntactic processes. In the following chapter I will discuss a range ofapproaches that have been taken to such second-position phenomena withinvarious frameworks, and offer reasons to believe that neither the syntax northe phonology can provide the relevant apparatus.

Kwakw’ala also displays other clitic elements, each with its own interestfor the overall analysis of the language, but these do not present new sorts ofproblem for a general theory. Some of them are arguably simple clitics. Forexample, a Complementizer such as (i)x. ‘that’ in example (4.27) appears at theleft edge of the complement clause and like other simple clitics in Kwakw’ala,attaches to a host on its left. In fact, this host will always be a word that is partof the matrix clause, as in (4.27).

(4.27) dux. ’waň@la=q=ix.=s w@nq@la=ida xw@p’asaw-it-comp-3sbj deep-dem holeHe saw that the hole was deep

Other grammatical elements in Kwakw’ala are special clitics, such as theSubject pronominals which appear in second position within the CP. Withinmain clauses, these will come after the clause-initial verb, but in subordinateclauses ‘second position’ will be immediately after the Complementizer andthus before the verb. Both possibilities are illustrated by the sentence in ex-ample (4.28).

(4.28) qa=suPP g.wag.

wixsPala la=qw q@=ida ’n@mukw=iif-you talk about on-it to-dem one-dem(3rd.inv)b@gwan@m . . .man . . .la-ň=@@n kwix.Pid@-ň=uňňaux-fut-I club-fut-youIf you talk about this to (any) one man. . . I’ll club you


The domain here is different from what we have already seen in DP specialclitics, but the parameters affecting these elements’ positioning are the same.

In this chapter, I have surveyed a range of special clitic phenomena whichappear to represent what we find in the languages of the world. Although Ihave argued that close parallels with word-level morphology suggest that it isappropriate to view special clitics as the morphology of phrases, and to de-scribe them by essentially morphological mechanisms, I have not yet shownthat alternative, more traditional treatments within the syntax are not avail-able. In Chapter 5, I address that issue, as the foundation for providing a moreexplicit account of what such a “morphological” theory of special clitics mightlook like in Chapter 6.

5

Theories of Special Clitics

Up to this point we have seen a representative range of phenomena that fallunder the rubric “special clitic,” and we have also seen some interesting analo-gies between these elements and the formal markers that appear within wordsas a consequence of their morphological properties. However interesting suchan analogy may be, though, it does not by itself constitute a theory of specialclitics. And indeed, most of the existing literature assumes that a real theoryof these phenomena should be located within the domain of syntax—or, for afew authors, the phonology. The task of this chapter is to flesh out the sugges-tion that both of these views are mistaken, and that neither the constructs andmechanisms of the syntax nor those of the phonology are fully appropriate tothe description of special clitics. Instead, the possibility raised in Chapter 4 thatthese constitute the morphology of phrases is more than a suggestive metaphor,and the formal devices of a theory of word structure should be recruited forthis purpose as well.

I begin by showing something of the nature of the problem, in the form ofan apparent conflict between the fundamental terms of the generalizations wesaw in Chapter 4 and those of the core components of a grammar as standardlyconstrued, the syntax and the phonology. These issues come into focus inconnection with second-position clitics,1 in particular, and the reasons for thiswill be developed. I will then review a careful survey by Boškovic of the rangeof 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 Boškovic, however, on the appropriate response to this and,in Chapter 6, I will develop the outlines of a morphological view, based onmechanisms of Optimality Theory.

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


5.1 The Nature of the Problem

If we want to explore the question of what kind of theory is best suited todescribe the distribution of special clitics, it is clear that only a limitedsubset of possible clitic types is likely to be relevant. Of the four possiblevarieties of special clitics discussed in Chapter 4, only the second-positiontype (or the penultimate type, if any such examples exist) will be of proba-tive value. That is because essentially any sort of theory should have nodifficulty locating elements in initial or final position within a domain. Second(or penultimate) position, on the other hand, is somewhat more theoreticallychallenging.

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 prosodicconstituent (PWord, PPhrase, possibly even IntPhrase), whether or not thiscorresponds to a syntactically unitary (and recognizable) constituent. Thiswill prove to be a central issue in developing an adequate theory of specialclitics.

It is apparent that Wackernagel’s (1892) idea of “his” position was “afterthe first word,” and not “after the first (syntactic) constituent.” The exampleshe cites quite generally involve clitics that appear after exactly one clause-initialword, rather than after a phrasal constituent. Although he himself undoubt-edly interpreted these in terms of (what we would see today as) a prosodic unitrather 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 providesome other examples) allow rather free word order. If we treat this flexibilityas resulting from a syntactic operation of “Scrambling,” it follows that theselanguages must allow phrases to be broken up into smaller constituents ratherfreely. In consequence, it is next to impossible to find evidence against theclaim that the single word preceding a second-position clitic in Wackernagel’sexamples 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 constituentsinto their minimal “scramble-able” units must have been obligatory, even inthe absence of reordering. Nonetheless, the evidential value of Wackernagel’sexamples for the point at issue is much less than it might be, in light of otheraspects 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 notappear to have presented much variation in word order of the sort describedby “Scrambling.” Hittite has generally rather fixed word order, and there is noevidence in this language for the existence of scrambling processes that wouldpresuppose a break-up of phrasal constituents into their component words.Nonetheless, we find that Hittite clitics generally follow the first full word ofthe clause, even when this involves locating them inside phrases of the sortthat 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-nithis=quot=refl baby-acc take conn=dir house-locanda it . . .in go . . .Take this baby and go into the house . . . (KUB XXIV 7 IV 45–6)

b. nepisas=as=sta DIM-unni assus estaheaven-gen=he=ptc Weather-God-dat dear-nom he-wasHe was dear to the Weather-God of Heaven (StBoT 18.2)

c. Lupakkin=ma=kan Uzalman=na INA Amka paraaLupakkis=quot=ptc Uzalaas=and into Amka forthnaistahe-sentHe sent L. and U. forth to Amka (JCS 10, p. 94)

The material preceding Hittite second-position clitics is not in general consis-tent with the claim that it should consist of exactly the first syntactic daughterconstituent of the clause.

Another system in which “second-position” placement of clitics is definedby an initial word rather than an initial phrase is West Greenlandic. Sadock(2003: 61) says that “[c]ertain clitics are positioned syntactically before a syn-tactic phrase and are suffixed to the first word of that phrase. This property ofclitics provides a test for wordhood” (my emphasis). Among the relevant cl-itics are the conjunctional elements =lu ‘and’, =luunniit ‘or’, and =li ‘but’,each of which is “positioned syntactically between phrases [and] will show upas a suffix to the first word of the phrase it precedes” (ibid). An example inwhich this entails placement within an initial phrasal constituent is providedby (5.2) from the text in Sadock (2003: 71).

2 The Hittite material available to us today was still undeciphered when Wackernagel wrote in1892, of course. His interests were in the early history of Indo-European, and the Indo-Europeanaffiliation of Hittite was not even suggested until nearly a quarter of a century later. It is hardlysurprising that he does not cite this evidence, although it provides interesting confirmation of hisproposals.


(5.2) Inuil=li tamaani nunaqartut iluanniperson.dir.pl-but here.loc land.have.ing.dir.pl midst.loc.3p.spinngoriartorpoq 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, fromLynge 1978) is the underlined instance of =lu ‘and’ in the second conjunct ofthe sentence in (5.3).

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

In the modern syntactic literature, the claim that “second position” mightbe defined in some instances by a prosodic unit (e.g., a word) that did notconstitute the first syntactic daughter of the clause first appeared in work byWayles Browne (1974, 1975).3 Browne argued that in Serbo-Croatian, either thefirst full word or the first full constituent of the clause can precede the clitic(s),as in (5.4).

(5.4) a. i. Moja ce mladja sestra doci u utorakmy fut younger sister come on TuesdayMy younger sister will come on Tuesday

ii. Moja mladja sestra ce doci u utorakmy younger sister fut come on TuesdayMy younger sister will come on Tuesday

b. i. Lav je Tolstoi veliki ruski pisacLeo is Tolstoi great Russian writerLeo Tolstoy is a great Russian writer

ii. Lav Tolstoi je veliki ruski pisacLeo Tolstoi is great Russian writerLeo Tolstoy is a great Russian writer

3 See also the extensive reviews in Schutze (1994) and Boskovic (2001), as well as literature citedin those sources.


Some native speaker linguists have questioned the acceptability of the datacited by Browne, but it seems that (as so often in such discussions) what isat stake is more a difference among dialects than a categorical distinction. Ingeneral, and simplifying somewhat, dialects spoken in the western part of theSerbo-Croatian area (grossly, “Croatian” dialects) admit clitics after the firstword, while dialects more to the east (grossly, “Serbian”) only allow them afterthe first full syntactic constituent. Thus, the (i) examples in (5.4) are much lessacceptable than the (ii) examples for these eastern speakers, who prefer cliticsto be located after a full syntactic phrase (and not, for instance, after the firstword of a two-word proper name).

This sort of variation is quite interesting, of course, and it is importantto be able to characterize it formally. It does not, however, directly impugnthe claim that some languages allow clitics to be placed after an anchor whoseunitary status is prosodic in nature rather than syntactic. There is no doubtthat systems exist where the initial anchor is defined in syntactic terms: whatis at stake is whether the prosodic anchor type occurs as well. If this possibil-ity is instantiated in even a single dialect of a single language, it must still beaccommodated within the terms of Universal Grammar, thus posing a prob-lem for theories that assume a uniformly syntactic solution to the problem ofsecond-position clitic placement.

DP Second-Position Clitics

The examples of second-position clitics we have seen thus far all involve cli-tics whose domain is the clause, but of course some languages have second-position clitics within other domains. For example, in section 4.3 I presentedthe set of second-determiner elements in Kwakw’ala, clitics that are placed im-mediately after the initial word of the DP. Somewhat better-known cases ofsecond-position elements within nominals, though, are the postposed deter-miners found in several Balkan languages, including Albanian, Bulgarian,Macedonian, and Romanian. I illustrate this type from Bulgarian, a languagein which definite articles appear immediately after the first word of the nomi-nal 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, as-sociated particularly with Halpern (1992a, 1992b, 1995) is Phonological in char-acter, and based on a proposed operation of Prosodic Inversion. This view


assumes that the clitic is “basically” positioned at the left edge of its domain,by syntactically trivial mechanisms. Subsequently, inversion of the clitic with afollowing word occurs somewhere in the mapping between syntactic structureand Phonetic Form so as to locate the clitic in second position.

Such a post-syntactic manipulation of word order is assumed to be forcedin order to satisfy prosodic requirements of the clitic. If this element is sub-categorized to require that it attach to a PWord on its left within the phrasethat constitutes its domain, that will only be possible (given its syntacticallyphrase-initial position) if it moves so as to assure that there is (at least) a singlephonological word preceding it within that domain.

The notion that grammatical elements can be subcategorized for a spe-cific prosodic environment in which they may (or must) appear is a widelyheld conception, developed most explicitly perhaps by Inkelas (1989). It wouldsurely be preferable, however, if we could derive such restrictions from moregeneral principles. I have already suggested in Chapter 3 that similar item-specific stipulations about the direction of attachment can (and should) bereplaced by a full account of general principles of Stray Adjunction within alanguage. To the extent possible, it would be desirable to maintain a similarposition with respect to the broader notion of “Prosodic Subcategorization.”In that case, however, the analysis of second-position phenomena in terms ofProsodic Inversion would not be possible. Rather than identifying particularitems with respect to the prosodic affiliation they should exhibit, I assume thatthese facts should follow from an identification of the item’s prosodic category(syllable, Foot, PWord, etc.) together with the principles of Stray Adjunctionthat govern the language as a whole.

Regardless of that consideration, though, the Prosodic Inversion analysiscannot accommodate the Kwakw’ala second determiners. Assuming that theseclitics are initially generated in the left periphery of the DP, they would alreadybe preceded by material that could support left attachment, without requir-ing any inversion. We can tell that the material preceding the left edge of thenominal is an appropriate host for clitics, because in fact the phrase-initial de-terminer 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 second-position clitic placement below. In any event, most linguists have not adoptedthat position, but have rather preferred to see second-position clitics as locatedwithin the syntax, by syntactic mechanisms. Nearly all of the discussion of thismatter has concerned clitics located within a clausal domain (such as those ofSerbo-Croatian), but before passing on to a more detailed examination of syn-tactic theories of such clitics, I will consider briefly the way such a theory plays


out in the particular case of DP-internal clitics, like the Kwakw’ala determinersor the Balkan determiners.

On the syntactic view, the syntax generates the clitic in the head D positionwithin a DP. A subsequent operation of displacement must then raise preciselythe first word of the embedded NP to SpecDP, as illustrated (for Bulgarian)in (5.6).

(5.6) DP

SpecDP D′

D NP

mnogoi =to [ei] interesni knigi

The syntactic account is motivated by a theory-internal assumption thatthe syntax must be the locus of description for such facts, so it is perhapsappropriate that it raises some general theory-internal problems even withinBalkan 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 in-volves a single word rather than a complete phrase. As such, it must be thekind of displacement known as “Head Movement,” rather than normal phrasalmovement. But the target of the displacement in this case is not a head posi-tion, but phrasal: SpecDP. To the extent Head Movement is assumed to haveproperties distinct from those of ordinary phrasal movement, this presents aconceptual 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 pre-sumed prosodic requirements), and not by those of the word that moves. Butwithin at least one version of the sort of theory that is at issue here, move-ment is only supposed to be driven by the needs of the item that moves, ratherthan by the resulting configuration. The movement in (5.6) would present aconceptual problem for any such account.

Additional problems for this analysis arise in attempting to extend it tothe case of the Kwakw’ala second determiners. First, it is unclear where theseshould be generated within the DP. There is already a clitic in phrase-initial po-sition (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


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 realizingthe content of D. In an analysis whose ontology of linguistic elements is basedresolutely on the traditional morpheme, this is difficult to resolve, although theanalogy suggested in Chapter 4 between clitics and morphology allows us to re-gard it simply as one more instance of “multiple exponence” (Anderson 2001).

Furthermore, just as in the phonological account, there is no motiva-tion (syntactic or prosodic) for the necessary displacement of the initial non-determiner word of the DP to the left of the second determiner element. Thisword need not move for its own purposes; and even if we allow for “altru-istic movement” there is no prosodic requirement that would go unmet if itremained in its basic position, as noted above in connection with the ProsodicInversion account. I conclude that an analysis of these second-position cliticsas resulting from movement within either the syntax or the phonology cannotbe maintained.

Other Prosodically defined Clitic Types

Kwakw’ala, Hittite, West Greenlandic, and (perhaps some forms of) Serbo-Croatian provide us with examples in which the anchoring element relativeto which a clitic appears in second position is a prosodic unit, the PWord, andnot a syntactic one. In principle, at least, if the PWord can constitute suchan anchor, it should be possible for some language to determine the notion ofsecond position by reference to some other prosodic constituent type.4 Exactlysuch a system is described and analyzed in detail by Chung (2003).

Chamorro has a set of weak pronominals that can represent direct objectsand certain intransitive subjects.5 These elements are prosodically deficient,and must attach phonologically to a preceding PWord. Clearly, then, they havethe properties of phonological clitics, as Chung shows. Furthermore, they canoccur in positions where corresponding non-clitic DPs cannot, as illustratedby the pair of sentences in (5.7).

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

4 Indeed, Halpern (1992b, 1995) entertains this possibility in connection with the operation ofProsodic Inversion, though he does not develop it in detail.

5 Transitive subjects and the subjects of irrealis verbs and adjectives are cross-referenced byagreement material in Chamorro, and it is a general principle of the language that pronominalsindexed by agreement must be phonologically null. This complementarity between agreementand the possibility of a weak pronominal is interestingly consistent with the position I will developin Chapter 8, according to which pronominal special clitics are themselves a form of agreementmaterial.


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 pro-

clitic attached to the following word or outside the domain within which theclitic appears, such a sentence would suggest that these elements are located insecond position, anchored by the first PWord of their domain. This is consis-tent with many other examples, such as (5.8).

(5.8) Fahani =yu’ =fan gä’-hu ga’lagubuy.for me please pet-agr dogPlease buy me a pet dog

However, in many other examples such as (5.9) it appears the clitics followan 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 alsoThey forget that we are American citizens too

b. [Más yä-hu na taotao] =haomost wh[obj].like-agr linker person youYou’re the person I like most

c. [Ginin San Roque na songsung] =yu’from San Roque linker village II’m from San Roque village

Furthermore, in other examples such as (5.10) the material preceding theclitic is more than a single PWord but not a syntactically complete phrase,either.

(5.10) Taotao San Roque =yu’ na songsungperson.linker San Roque I linker villageI’m a person of San Roque village

Chung shows in detail that the material preceding the clitic need not con-stitute 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’anthe most old she linker sibling-agr femaleShe’s my oldest sister


b. Malak ädyu =yu’ i gima’ ni gininagr.go.to that I the house Comp imperf

s-um-ásaga =yu’ stabaagr-stay.prog I used.toI went to that house where I used to live

How are we to reconcile these apparently conflicting facts about the place-ment of Chamorro pronominal clitics? Chung argues that the correct (andexceptionless) generalization is the formulation in (5.12).

(5.12) Pronominal clitics are located immediately after the first PPhrasewithin the IntPhrase with which they are associated.

In support of this, she develops a set of specific principles that relate syntac-tic structure to prosodic structure in Chamorro, showing that the PPhrasesthey project are precisely the units that serve as initial anchors for the clitics.Indeed, where these principles provide more than one option for phrasing, acorresponding option exists for the placement of a clitic.

The details underlying this analysis are not immediately relevant here, andthe reader is referred to Chung (2003) for much additional discussion. Whatmatters to us is the conclusion that in this language, the prosodic constituentPPhrase serves as the anchor for a set of second-position clitics, showing thatthe PWord is not the only prosodic type that can serve this purpose.6 Theseprosodic units are constructed on the basis of syntactic structure, but do notcorrespond 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 thepositioning of special clitics that allocates this task entirely to the syntax. Themost extensive survey of the adequacy of syntactic theories is that of Boškovic(2000, 2001), a discussion that concentrates on Serbo-Croatian and related lan-guages of the Balkans while addressing the relevant issues in a rather generalfashion. I turn now to a review of Boškovic’s arguments.

Assuming that the available options for a theory of second-position cliticsreduce to a balance between syntactic and phonological mechanisms, Boškovicdistinguishes four general classes of theory:

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


Strong Syntax: Clitic positioning is fully determined by the syntax.

Strong Phonology: Phonology is fully responsible for placing clitics in secondposition. The syntax generates clitics in normal argument positions. They arethen moved into second position as necessary in the phonology.

Weak Syntax: Most movement of clitics is syntactic. However, Prosodic Inver-sion may intervene in the phonology to satisfy the requirement of a clitic foran appropriate host.

Weak Phonology: Movement of clitics takes place in the syntax, and involvesa considerable amount of freedom of positioning. The role of phonology ispassive, filtering out certain syntactically well-formed sentences that violatephonological requirements of the clitics.

Accounts that involve movement of syntactically specified items withinthe phonology are undesirable a priori, according to Boškovic, and most an-alysts who have discussed the matter agree. We have already seen some rea-sons to believe such phonological movement is an unsatisfactory answer to theproblem of second-position clitics, in the discussion of Kwakw’ala DP cliticsin section 5.1. With reference specifically to Serbo-Croatian, the language thathas been at the center of discussion in this regard, Boškovic concludes thatthere is no persuasive evidence in favor of phonological movement (includingProsodic Inversion), and some evidence against it.

One such argument comes from the fact that clitic placement seems to besensitive to syntactic information of a sort that ought not to be available in thephonology. Placement of clitics in clitic-climbing constructions depends onthe syntactic/semantic relation between the matrix verb and a complement, assuggested by the examples in (5.13).

(5.13) a. Milan kaže da =ga vidiMilan says that him seesMilan says that he sees him

b. *Milan =ga kaže da vidi

c. Milan želi da =ga vidiMilan wants that him seesMilan 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 reasonto believe it is based on properties that are plausibly present in Phonetic Form.


A related fact arguing against Prosodic Inversion as the source of post-initial positioning of clitics is that there are some circumstances in Serbo-Croatian 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 movementmust be invoked. This is illustrated in (5.14).

(5.14) a. Prema Mileni =su Milan i Jovan išlitoward Milena aux Milan and Jovan walkedMilan and Jovan walked toward Milena

b. *Prema =su Mileni Milan i Jovan išlitoward aux Milena Milan and Jovan walkedMilan and Jovan walked toward Milena

Unlike most prepositions in Serbo-Croatian, prema bears stress and is pre-sumably a full prosodic word, so if the clitics are generated in initial positionand then undergo Prosodic Inversion around a following word, that shouldlead to placement after the preposition in this case.

Boškovic relates the positioning of the clitic in (5.14) to another propertyof the phrase prema Mileni ‘toward Milena’ in Serbo-Croatian: while word or-der in the language is generally quite free, such a phrase must remain as a unitand cannot be separated through Scrambling. I will return later to the impor-tance of this observation, but for present purposes its significance is due to itsinteraction with the putative rule inverting a clitic with a following word so asto satisfy prosodic requirements. Apparently the generalization is: “ProsodicInversion” can locate a clitic after an initial word if and only if that word con-stitutes a syntactic unit that can move independently of what follows it. Sinceprema 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 forthe placement of the clitic, it cannot be something that is purely phonologicalin character, and so we are not really dealing with “Prosodic” inversion in thegeneral case of second-position clitics.

On the premise (introduced in previous chapters) that access to syntacticstructure in the phonology can in general be reduced to access to the prosodicstructure induced by syntactic phrasing, I conclude (with Boškovic) that thereis no advantage to be gained from having movement that takes place in thephonology. Apparently, that excludes both ‘Strong Phonology’ and ‘Weak Syn-tax’ positions within Boškovic’s typology, and leaves us with accounts in whichclitics are placed entirely by syntactic mechanisms. Focusing on second-pos-ition clitics, two major purely syntactic theories (and one minor one) are to befound in the literature.


One of these assumes that clitics are initially placed in phrase-initial po-sition. Subsequent syntactic operations then have the consequence of movingor adjoining exactly one constituent to the left of the clitics. Alternatively, wemight assume that clitics are placed (either initially or as the result of move-ment) in a head position quite high within the structure. The Spec positionassociated 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 andProgovac 1999), can be seen as a variant of these. On this picture, it is as-sumed that the clitics are attached to the verb, which then moves (crucially, bya copying operation rather than simple movement) into second position. Thecopy of the verb, while in second position, is not however pronounced, thoughthe copies of the clitics are. Correspondingly, the copy of the verb in its originalis pronounced, while the instances of the clitics in this position are not. Thisapproach, frankly, strikes me as simply bizarre, especially given that there isno evidence for the claim that verb second is operative in every language withsecond-position clitics. Boškovic (2001), nonetheless, finds advantages for thisview within the range of theories based exclusively on syntactic mechanisms.It is not necessary to deal directly with these matters, however, since this viewof second-position clitics has the same general consequences as the other two,as these are explored by Boškovic.

Any theory describing the placement of clitics in second position by suchpurely syntactic mechanisms makes several predictions, some of which are ex-plored by Boškovic:

• 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 thanone constituent to precede it.

Thus, there ought to be some consistent syntactically motivated positionthat clitics occupy. On one of the variants just described, this will be the posi-tion of an adjoined phrase on the left periphery, preceded by another adjoinedphrase. On another, it will be a head position, perhaps head of CP, TP, or someother functional projection, and preceded by the specifier of that category. Thekey to all of these analyses is the presence of a syntactically consistent targetfor clitic placement. All have in common the (implicit or explicit) claim that“second position” is only an accidental, epiphenomenal consequence of thepresence of clitics in this structurally defined location.

Syntactic accounts have generally attempted to argue that second-positionclitics do indeed appear in some such uniform position in the structure of the


clause. Phonology-based accounts make a similar claim. Halpern (1992b, 1995)argues that second-position clitics are structurally initial in their clause (andeither come to be preceded by another constituent, or else undergo phonolog-ical, but not syntactic movement). Schütze (1994), within a syntactic account,argues that Serbo-Croatian second-position clitics appear in C. Others haveadvocated other functional heads, such as T.

Boškovic argues that there is no consistent structural position which is oc-cupied by Serbo-Croatian second-position clitics. He establishes this by show-ing that the assumption of such a position leads to a contradiction. First, he ob-serves 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 PetraJovan aux forgotten completely PetarJovan forgot Petar completely

b. Jovan=je nesumnjivo istukao PetraJovan aux undoubtedly beaten PetarJovan undoubtedly beat Petar

c. *Jovan=je istukao nesumnjivo PetraJovan aux beaten undoubtedly Petar

*Jovan beat Petar undoubtedly

From this he concludes that sentence adverbs are located higher in the struc-ture of the clause than VP adverbs; and that a participle can move from belowthe location of VP adverbs to a position between the two, but no higher.

Consider what this evidence tells us about the position of clitics, assum-ing these occupy some specific structural position. The clitics in the examplesof (5.15) are to the left of (and thus, ex hypothesi, higher than) participles andboth kinds of adverb. This shows that the designated position for clitics mustbe located above that of sentence adverbs (or VP-Adverbs, a fortiori) and par-ticiples. But now compare the previous examples with (5.16).

(5.16) Odgovorio=je pravilno Marijianswered aux correctly Marijaonly (He) answered Marija correctlynot (He) was correct to answer Marija

In terms of the location of the clitic je, nothing has changed, in at least onesense: like the clitics in the other Serbo-Croatian examples above, this one is insecond position. The adverb pravilno ‘correctly’ is potentially interpretable as


modifying either the VP or the entire sentence, but since in (5.16) the partici-ple odgovorio ‘answered’ precedes it, we predict (correctly) that it must here beinterpreted only as a VP adverbial. The participle also precedes the clitic, andthus, on the assumptions about how word order is related to syntactic structurebeing made here, it occupies a higher position in the tree than that in whichclitics occur. But this contradicts our earlier conclusion that the highest posi-tion for participles is lower in the tree than that for clitics, at least if we assumethat the position occupied by clitics is structurally uniform.7

Boškovic concludes from these and a variety of other facts that the syntaxby itself cannot accomplish all that it has to. The notion of “second position”does not have a natural reconstruction in purely syntactic terms, because itdoes not correspond to any structurally unitary position. Nonetheless, exactly“second position” seems to be the notion that is relevant for placing clitics inSerbo-Croatian. Something other than a theory of the “Strong Syntax” sortmust therefore be required.

Within the spectrum of theories Boškovic considers, the remaining possi-bility is a “Weak Phonology” view, and it is such an analysis that he presentsand defends. In support of the claim that the phonology is involved, he arguesthat the very domain within which clitics are placed in this language requiresreference to the phonology. That is because various constructions that induceunusual phrasal organization also result in unusual clitic placement, as illus-trated in (5.17).

(5.17) a. Sa Petrom Petrovicem srela =se samo Milenawith Peter Petrovic met refl only MilenaWith Peter Petrovic, only Milena met

b. Znaci da, kao što rekoh, oni =ce sutra docimeans that as said they aux tomorrow arriveIt means that, as I said, they will arrive tomorrow

c. Ja, tvoja mama, obecala =sam =ti sladoledI your mother promised aux 2sg ice creamI, your mother, promised you an ice cream

Initial topics, inserted parentheticals and appositives all induce IntPhraseboundaries at their edges, and under these circumstances we find that the no-tion of “second position” also shifts. Aside from providing further argumentsagainst a purely syntactic view of clitic placement (for which the presence of

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


these intonational elements ought not to matter), these facts suggest that cli-tics are located in second position within a domain that does not necessarilyhave a syntactically unitary status. They are also, of course, strongly reminis-cent of the English constructions considered in Chapter 3, where special phras-ing (with contrastive stress, parentheticals, gapping, etc.) affected the possibleoccurrence of clitic auxiliaries in ways that do not make sense in terms of thesyntax alone.

Boškovic concludes that the real generalization is that in (5.18).

(5.18) Serbo-Croatian clitics occur in the second position of their into-national phrase.

He proposes to achieve that effect by allowing the syntax to manipulate theorder of words (including clitics) fairly freely, and then filtering the result interms of stipulated requirements governing the position of clitics in PhoneticForm. The statement in (5.18), roughly, is thus imposed as an output con-straint, and any ordering of elements that includes clitics not conforming to itis blocked as ill-formed.

Boškovic argues that the syntax must play a central role in this, because apurely prosodic definition of the initial anchor preceding the clitics does notsuffice. In particular, while word order in Serbo-Croatian is fairly free, thereare some constructions that cannot be separated, and must move as syntacticunits. When one of these occurs initially, the clitics must follow the entirestructure, 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 otišliyour mother and Petar aux leftYour mother and Petar left

ii. *Tvoja =su majka i Petar otišli

b. i. Roditelji uspešnih studenata =su =se razišliparents successful students aux self dispersedParents of successful students dispersed

ii. *Roditelji =su =se uspešnih studenata razišli

Interestingly, speakers differ somewhat as to the separability of these con-structions, and it appears to be the case that if and only if a given speaker allowsone or another of them to be split up, that speaker also allows clitics to occurafter the first PWord of the structure, and not (only) after the entire syntacticunit.

A similar fact can be cited with respect to sentences in which a two-partproper name occurs, such as Browne’s example (5.4b). In such DPs, inflection


may occur either on only the last part of the name, or independently on bothparts. The distinction is not evident in (5.4b), where the nominative has noovert inflectional marker, but can be seen in other cases. Crucially, if and onlyif both parts of the name show independent inflection, they can be formallyseparated—and it is exactly in this case that clitics can appear after the firstpart of the name, as well.

(5.20) a. i. Lava Tolstoja citamLeo.acc Tolstoy.acc read.1sgLeo Tolstoy, I read

ii. Lav Tolstoja citamLeo Tolstoy.acc read.1sgLeo Tolstoy, I read

b. i. Lava =sam Tolstoja citalaLeo.acc aux Tolstoy.acc read.ppartLeo Tolstoy, I read

ii. *Lav =sam Tolstoja citalaLeo aux Tolstoy.acc read.ppartLeo Tolstoy, I read

These facts appear to establish the conclusion that the syntax must play adetermining role in characterizing the initial anchor for second-position cliticsin 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 thatthis anchor is also a syntactically autonomous unit. Combining these consid-erations with the evidence for the insufficiency of syntactic mechanisms aloneand for a role played by the phonology, Boškovic concludes that the “WeakPhonology” position is strongly supported for this language.

Whatever the correct analysis of Serbo-Croatian may be, we have alreadyseen evidence that an account along these lines (with syntax controlling theorder of elements, including clitics, and phonology playing a filtering role toenforce a second-position constraint) cannot be maintained in general. In theHittite examples in (5.1), for example, there is no warrant for the assumptionthat the anchors preceding the clitics are syntactic units that can be permutedinto this position, since Hittite in general does not allow the kind of reorder-ing characteristic of Serbo-Croatian. Similarly, as we have already seen in sec-tion 5.1 above, Chung (2003) argues explicitly that the anchor preceding cliticsin 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 ofspecific constructions in Serbo-Croatian is not what Boskovic claims.


with any syntactically unitary constituent. We must conclude for the gen-eral case of second-position clitic placement that while the “Weak Phonology”position is an improvement over accounts based either on exclusively syntacticmechanisms or on Prosodic Inversion, it is still not adequate to cover the fullrange of second-position phenomena.

Another class of arguments against the treatment of special clitics as syn-tactic elements is due to Géraldine Legendre. She shows that special cliticelements in Balkan languages are syntactically “inert.” For instance, in Mace-donian (Legendre 1998) full verbs and auxiliaries invert with the subject inquestions, 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-defDid Ana give the vase to you?

b. Go= ima li Ivan svrseno?it have-3 q Ivan solvedHas Ivan solved it?

c. Ke= ti= ja= dade li Penka knigata?will 3sg-dat it gave-3 q Penka book-defWill Penka give you the book?

d. Kade bi= ja= stavil ti knigata?where would it put you book-defWhere would you put the book?

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

(5.22) a. Izpraznet =ke =e stanotvacated fut be-3 apartment-defThe apartment will be vacated

b. Procel =sum knigataread have-1 book-defI have read the book

Clitics also do not induce strong crossover effects, as shown by the possi-bility of examples like (5.23).

(5.23) Covetkoti

kogoi

što goi= vidov [e]

iman-def whom that him saw-1 tthe man whom I saw


Similar arguments (and others) can be given from other Balkan languages(Legendre 1999, 2000a, 2000b, 2000c) to the effect that clitics do not behave inthe way we would expect if they were heads (or phrases) in the syntax. Thisconfirms that it is probably incorrect to think of clitics as simply a special classof lexical items, whose location within the sentence is specified by the mecha-nisms of the syntax of words and phrases.

Accounts that place special clitics in their surface position by purely syn-tactic means make other predictions that can be tested, as well. If the syntaxis doing this work, then in cases where more than one special clitic appears ina sequence, the relative order within the clitic sequence ought to reflect someaspect(s) of syntactic structure, such as relative scope of corresponding func-tional categories. Real examples do not in general conform to this expectation,however. For instance, in French the preverbal clitics appear in the order givenin (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 place-ment within the generative literature, this ordering does not reflect syntac-tic structure, and must be imposed by some other means (for Perlmutter, atemplate acting as a constraint on Surface Structure). In some languages,the sequence of clitics may show coherence in terms of morphological cate-gories (e.g., first person before second person before third person in Ngiyam-baa (Donaldson 1980), regardless of grammatical relations) while largely dis-regarding syntactic roles.

Hittite also shows a rigid template for the sequence of second-positionclitics, 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 pointof view of the syntax.

Yir-Yoront (Alpher 1991) moves even further from a syntactically basedprinciple of clitic ordering. Clusters of pronominal clitics in this language havea first-person enclitic first (regardless of the grammatical relation to which thiscorresponds), followed by second- and third-person markers ordered accord-ing to a phonological principle, with “lighter” clitics preceding “heavier” ones.


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

(5.26) a. dadik´rvn˙a ıd u nu carkirama 4.40.1

b. mitrás cid hí s.ma juhuran. ó dev´an 10.12.5

Schachter (1973) offers a comparable argument from Tagalog. In this lan-guage, monosyllabic clitics always precede polysyllabic ones, a distinctionwhich cannot be related to any syntactically relevant dimension. I will discussthe Tagalog facts in more detail below in section 6.4.

A wide variety of considerations thus force us to recognize a role for phon-ological (specifically, prosodic) considerations, along with morphological ones(such as person and number, as well as case) in the regularities of special cliticplacement. Since we have also seen that purely syntactic accounts of thesematters fail in a variety of ways, it is clear that we must look for somealternative—indeed, for a theory of a sort that does not fall entirely withinany of the categories contemplated by Boškovic. In Chapter 6, I will developthe basic principles of such a theory.

6

An Optimal Theory ofClitic Positioning

If the syntax is not the appropriate place to look for the principles of cliticplacement, what alternatives are there? Clitics are apparently not just membersof a special class of lexical items, elements whose location within the sentenceis specified by the syntax of words and phrases. But if the syntax is not the ap-propriate place to look for the principles of clitic placement, what alternativesare there?

In this chapter, I return to the suggestion of Chapter 4 that there are sub-stantial similarities between special clitics and word-level morphology. Fillingout that proposal, I propose an account of the positioning of these elementswithin the framework of Optimality Theory that brings out the connectionsbetween the two domains. I consider some special cases, particularly thosethat seem to argue for the existence of “endoclitics,” and then illustrate theworkings 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 thatclitic ordering does not in general follow from syntactic principles. Rather itseems to be a surface, or PF, phenomenon, although a purely phonologicalaccount of the ways in which special clitics diverge in their positioning fromthe predictions of the syntax, based on Prosodic Inversion, does not succeedeither. A phonological account would in any event be of no help in avoidingthe consequences of arguments suggesting that special clitics are syntacticallyinert.

The phonology can thus be excluded as providing the mechanisms neces-sary to locate clitics. But clitics do look a lot like morphology, as I suggestedin 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


its domain results from a rule of modification operating on the phonologi-cal realization of that domain. Of course, most of these rules perform simpleaffixation of fixed phonological material, but in this respect they are entirelyparallel to the rules of word-internal morphology, most of which are affixalas well.

Let us then accept the argument that clitics and word-level morphologyare parallel phenomena, differing primarily in the scale of the unit (words ver-sus 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 modifi-cation of the phonological form, such as mutation, stress shift, reduplication,etc.)1 specifies what change should be made in the overall phonological repre-sentation, but it does not by itself tell us where that change should be made. Forany particular clitic or word-level affix, what is at issue is the location within acomplete form at which a particular change takes place, regardless of whetherthat change consists in the introduction of new phonological material or ina modification of material already present. An appropriate formalization ofthe rules of clitic placement remains to be settled on, but whatever mecha-nism is involved, it must involve a kind of non-syntactic formal modification,something distinct from the syntax of independent minimal signs, and a spec-ification 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 in-troduced within some domain. Let us confine our attention for now to whatwe will refer to as “inflectional” clitics, such as pronominals representing ar-guments of the lexical verb within a sentential domain, or the Tense/Aspectproperties associated with such a domain, or determiner and/or possessor in-formation associated with a nominal (DP) domain. Grossly, we might say thatclitics are introduced by rules parallel to those that introduce affixal materialinto words, reflecting the grammatical properties (e.g., case, number, agree-ment, etc.) of an inflected word.

1 In what follows here, I will often omit the fuller qualifications necessary to provide for non-concatenative modifications when talking about “clitics” and “affixes.” This is purely a matterof convenience and brevity of expression, and should not be interpreted as limiting the range ofexpression 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 gram-matical properties—features—of the node dominating some appropriatedomain, such as CP, IP, or DP, we can ask how these formal reflections of func-tional content can be made to appear in the correct relative positions in thesurface form. If affixes and clitics are introduced by individual rules of modifi-cation, an obvious possibility is to specify the locus of affixation directly as partof the rule that introduces the material in question, and then to relate the linearsequence of clitics to order of application of these rules. This is the proposal Ipursued in Anderson (1992, 1993).

There are, however, some formal and mechanical problems in implement-ing this account. One of these is the fact that sometimes the shape of a cliticdepends on that of another clitic, where the most natural description wouldhave the conditioning clitic introduced later in the descriptive sequence thanthe clitic whose shape it influences. For instance, in Serbo-Croatian, the fem-ine singular accusative clitic =je is obligatorily replaced by the shape =ju whenanother clitic with the shape =je occurs after it, as shown in (6.1).

(6.1) a. Oni =su =je/*ju zaboravilithey aux her forgottenThey forgot her

b. On =ju/*je =je zaboraviohe her aux forgottenHe forgot her

If the clitics are introduced one by one, in an order reflecting their left-to-right sequence, the conditioning environment for this change will not beavailable at the point at which the clitic undergoing it is to be added. Since thechange 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 betreated as an aspect of the rule that adds fem sg acc=je/ju). This appears to bean 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 Boškovic (2001:103) observes). That intuition is not expressed, however, if the second =je isnot present to trigger the change, which must then be conditioned by the setof morphosyntactic properties that will subsequently introduce the auxiliary=je.

A similar issue arises with respect to clitic allomorphy in Italian. Here theclitics mi, ti, si are replaced by me, te, se immediately before ne, lo, la, li, le, andfem.dat. le is replaced by glie (Monachesi 1995: 43ff), as illustrated in (6.2).


(6.2) a. Martina te/*ti= lo= spediràMartina 2sg.dat 3msg.acc will.sendMartina will send it to you

b. Martina me/*mi= ne spedirà moltiMartina 1sg.dat part will.send manyMartina will send me many of them

c. Glie-/*Le= le= ho date3fsg.dat 3pf.acc I.have givenI have given them to her(/him)

This may well have had a phonological basis at some point in the history ofthe language, but it is now a matter of idiosyncratic allomorphy that oughtto be expressed as a contingency in the choice of inserted material. Such anaccount is not available, however, if clitics to the right of a given element in thesequence are not introduced until after that element.

These problems would disappear if the clitics were introduced in right-to-left order, instead of left-to-right. But Italian also illustrates an allomorphicdependency in the opposite direction. In the literary language, the locativeclitic ci= can also be realized as vi=. However, if the locative clitic is precededby an object clitic, it can only have the form ci=, as shown in (6.3).

(6.3) Mi/Ti/Vi= ci/*vi= porta1sg.acc/2sg.acc/2pl.acc loc takesHe takes me/you(sg)/you(pl) there

Such examples suggest that introduction of clitics one at a time, in a se-quence intended to determine their order in the surface form, will run intologistical difficulties. Instead, it seems that several clitics forming a unitarycluster need to be introduced simultaneously in a way that would potentiallyallow each to be sensitive to the properties of the others.

A rather different set of problems for this approach to clitic positioningresults from the fact that the linear order of a sequence of clitics may remainthe same regardless of whether the sequence as a whole precedes or followssome anchor point. Thus, in Italian, clitics precede the finite verb, but followimperatives. What is important to note is that the sequence is the same in bothcases, despite the fact that the cluster of clitics appears on opposite sides of theanchoring verb.

(6.4) a. Me= lo= dice1sg.dat 3sg.acc you.tellYou tell me it


b. Dim=me=lotell.imp-1sg.dat-3sg.accTell me it!

In Macedonian, parallel to Italian, clitics appear in the same sequencebefore finite verbs and after non-finite verbs as exemplified in (6.5).

(6.5) a. Ne =bi =mi =go dalneg cond 1sg.dat 3sg.acc givesHe wouldn’t give it to me

b. Dajte =mi =gogive.imper 1sg.dat 3sg.accGive it to me!

c. nosejki =mi =go. . .bring.Pres.Part 1sg.dat 3sg.accbringing it to me. . .

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

(6.6) a. Ivanco =mi =go pokazaIvancho 1sg.dat 3sg.acc showedIvancho showed it to me

b. Pokaza =mi =go Ivancoshowed 1sg.dat 3sg.acc IvanchoIvancho showed it to me

Why does this pose a problem for a theory that introduces special cliticsone by one by rule, and derives their linear sequence from the order of applica-tion of these rules? The problem is that when the same set of clitics can appeareither before or after their anchoring element, we would expect the sequencesin the two cases to be mirror images, not the same. If the same set of rulesintroduces these clitics, differing only in the before/after parameter, the cliticclosest to the anchor in each case ought to be the same, yielding reverse ordersin the two cases. This situation, which would be expected at least to be theunmarked option, is virtually never found,2 and this requires an explanation.

This fact is noted by Kayne (1991), who accounts for the fact that cliticsappear 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 maynot even be relevant, since it is not obvious that the second pronoun (moi) in these imperativeforms is a special clitic comparable to the pre-verbal form (me).


anchoring verb by not placing them individually with respect to that verb atall. Rather, Kayne attaches the clitics to an empty functional category and thenmoves that category to a position either before or after the verb, dependingon properties such as finiteness. In at least some cases, however, this abstractanalysis will be difficult to maintain. Legendre (1998, 2000a) develops argu-ments for Bulgarian and Macedonian similar to those provided by Boškovicfor Serbo-Croatian, showing that there is no uniform category to which theclitics could be attached in implementing this analysis.

So where do we stand with respect to the placement of second-positionclitics? I have argued that we want a theory that can accommodate both cliticsand affixes as the phrasal and word-level instantiations of very similar mecha-nisms, because the positional possibilities are so similar in the two cases. Inthe case of clitics, we need to be able to define “second position” with referenceeither to a prosodic unit (a PWord, PPhrase, etc.) or to a syntactic unit (aphrasal constituent) at the beginning of the relevant domain. Ascribing thisplacement entirely to syntactic movement is apparently excluded on severalgrounds, as we have seen. In earlier work (Anderson 1992, 1993), I explored thealternative of introducing clitics by rules of modification of the phonologicalform of the phrase, where the relative ordering of clitics is specified as relativeordering of the clitic introduction rules. But this approach also has significantproblems, and so it is necessary to seek an alternative.

A Constraint-Based Theory

An alternative view is provided by Optimality Theory (“OT”: Prince and Smol-ensky 1993).3 To see this, let us note that a single “rule” of the sort consideredin the previous subsection actually does two things. First, it introduces somechange in the shape of the form (typically, the addition of a chunk of phono-logical material). In addition, the formulation of the rule specifies just wherethat change is to take place. The primary function of the rule is to provide anexponent for some aspect of linguistic content, and secondarily to determineexactly how a complex form is to be structured. Separating the two yields apotential improvement in the clarity with which each can be expressed.

Optimality Theory can provide us with a way to describe the placement ofspecial clitics if we invoke a constraint system to describe the way a complexinput should be realized overtly. That input consists of a basic form as supplied

3 As in Chapter 3, I assume that the overall workings of constraint-based theories like OT arefamiliar, and do not attempt to provide a general introduction here. The reader whose needsare thereby left unmet can consult general works such as Kager (1999), or more specifically forapplications to morphosyntax, the papers in Barbosa, Fox, Hagstrom, McGinnis, and Pesetsky(1998) and Legendre, Vikner, and Grimshaw (2001).


in a lexically interpreted syntactic representation, together with an indicationof one or more changes (phrasal affixes to be introduced or non-segmentalalterations to be made) that should affect the form to reflect associated mor-phosyntactic properties. I assume that this input is constructed in one of twoways.

First is the case of special clitics representing “grammatical” material, suchas pronominals identifying an argument, tense and aspect markers, determin-ers, etc. Here I assume that the input to rules of phrasal affixation is a pairconsisting of the syntactic representation and its phonological shape as thishas been developed by other aspects of the grammar (particularly lexical in-terpretation). The structural description of such a rule identifies features inthe complete morphosyntactic representation of the syntactic phrase that isthe clitics’ domain. These will include much of what is often represented as thecontent of separate functional categories, here presumed to be represented aspart of a complex symbol. For discussion of these matters at the word level, seethe discussion of morphosyntactic representations in Anderson (1992); furtherelaboration of these notions will be provided in Chapter 8. The effect of therule is then to specify additional phonological content that is to be associatedwith the phonological representation of the phrase. The result (possibly in-volving several such components of added material) is then submitted to theconstraint system, which locates the modifications within the form, and ef-fects any necessary phonological adjustments that may be necessary in lightof the additional material. Such adjustments will include the effects of “StrayAdjunction,” for instance, as well as other changes such as assimilations, etc.

The second case involves special clitics that do not reflect grammatical orfunctional material, but which rather have semantic and/or pragmatic contentof their own. We will see some examples in the discussion of Tagalog below insection 6.4; such elements are often (and unilluminatingly) called “particles”in grammatical descriptions. They constitute the special clitic analog of deriva-tional morphology, while the grammatical clitics mentioned above correspondmore or less to inflectional morphology. Again, the discussion of Tagalog be-low will suggest that this parallel is not entirely without force.

Special clitic rules of this “derivational” sort take as their input a syntacticstructure 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, justas in the “inflectional” case. On the other hand, a correlated change is alsoeffected in one or both of the syntactic and semantic representations. A specialclitic may convert a sentential structure to a nominalized form, for example(see Kaiser 1998 for some examples), by affecting its syntactic categorization.


Somewhat more commonly, perhaps, the associated change is semantic, suchas the introduction of a negation operator, adverbial material, etc.

The output of one or more operations of phrasal affixation then creates arepresentation consisting of several components, and this is then submitted tothe constraint system to determine its phonological realization. It is this con-straint system which is responsible for determining the placement, as opposedto the presence, of special clitics, as well as their integration into an overallphonological shape.

For present purposes, the immediate advantage of this approach as op-posed to the rule-based view comes from the fact that in OT, several of theaffixes in a word or the clitics in a phrase can be treated as simultaneouslyintroduced, and thus essentially copresent. Descriptive order can reflect the si-multaneously evaluated relative ranking of element-specific constraints, ratherthan a sequence by which the elements are introduced. Several clitics may allneed 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 simul-taneously and with reference to one another. We will see other advantages ofthe constraint-based view as we proceed to develop it.

The possibility of simultaneous evaluation of constraints on the position-ing of multiple clitics need not mean that all of the clitics in an utterance, orall of the affixes in a word, are always introduced in a single step. This com-plete parallelism is, it is true, a common assumption in the OT literature, butit is one that is logically separable from other aspects of a constraint-based ap-proach. Indeed, Kiparsky (2000, forthcoming) has developed a constraint-based version of Lexical Phonology in which serial, cyclically organized deriva-tions play a role, with each stage of phonological adjustment expressed by asystem of constraints. I follow essentially the same approach, and will note be-low where the standard strictly parallel view is not maintained. Nonetheless, inmany cases, I assume that multiple adjustments are made to the same structurein a single step. This is true, in particular, where a number of clitics cluster inways that interact with one another.

Word-Level Affixation Let us begin to develop the required theory by dis-cussing the word-level case. How does a constraint-based view describe theplacement of affixes? Relative ordering results from the fact that a number ofaffixes are all constrained to be located in the same position. These constraintsare “soft” or violable, however, and the demands of some affixes will outweighthose of others.

Suppose a number of affixes are all constrained to be prefixes. That meansthat for each of them, there is a constraint to the effect that it should appear atthe left edge of the word. Since these constraints are strictly ranked, however,


one of them will outrank the others; and of the remaining constraints, therewill again be one that is next highest ranked, etc. In the optimal realization ofa given form, the affix corresponding to the constraint with the highest rankwill actually be an initial prefix; the next highest ranked will occupy a positionwhich violates minimally its requirement of being initial (i.e., as the secondprefix), etc. Exactly similar considerations apply, mutatis mutandis, for collec-tions of suffixes.

To produce the result that prefixes or suffixes appear in a particular orderwith respect to one another (and to the left edge of the word), we invoke mem-bers of a family of constraints EdgeMost(e, L/R, D) (where e is some linguisticelement such as an affix or clitic, and L/R refers to the left or right edge of thedomain D within which e appears). These “EdgeMost” constraints are clearlya subset of what are more commonly referred to in the OT literature as “align-ment” constraints. Thus, the interpretation of EdgeMost(Afi, L, D) is the sameas a constraint saying that the left edge of the affix Afi should be aligned withthe left edge of some including category D, such as the PWord or the LexicalWord.

I use a somewhat simpler formulation here simply for convenience, andindeed I will often omit specification of the Domain when this is obvious. Aconstraint-requiring element e to appear at the left edge (of D) will generallybe expressed below as LeftMost(e(, D)); each such constraint says that the ele-ment e should appear as close to the left edge (of D) as possible. An analogousconstraint family RightMost(e(, D)) says that (the right edge of) e should ap-pear at the right edge of D.

A given affix Afi is characterized as a prefix or as a suffix by whether Left-Most(Afi,D) dominates RightMost(Afi,D) or vice versa. The descriptive orderof Afi and Afj is determined by the dominance relation that obtains betweentheir corresponding LeftMost (or RightMost) constraints relative to D.

Now consider the way infixes are treated in OT, a point on which a sub-stantial literature is in essential agreement. A post-initial infix is generallydescribed by treating it as a type of prefix, and thus subject to a high-rankingLeftMost constraint. The location of this affix in initial position, however, isprevented by some other higher-ranking constraint C that keeps it from beingabsolutely initial. The optimal form resolves this conflict by locating the af-fix 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 thuspositions the element as a (post-initial) infix.

The nature of the higher-ranking constraint C that forces a would-beprefix to be located as an infix is clearly crucial to this account. In some word-level cases, C may be a phonological requirement on permissible syllable struc-tures. As an example, the Sundanese plural infix –ar–/–al– immediately follows


a word-initial consonant. This is arguably due to high-ranking constraints thatrequire syllables to have onsets while deprecating codas. When –ar–/–al– isadded to niis ‘cools self ’, the affix is constrained to be LeftMost; but if it wereactually to occur at the left edge of the word, the result would be *arniis, whoseinitial syllable (a) has no onset, and (b) has a coda. Locating the infix after thefirst consonant avoids violations both of Onset and of *Coda, at the cost of vi-olating LeftMost(ar/l) minimally. Given the ranking of the syllable-structureconstraints 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 consonantavoids both types of violations equally while violating LeftMost minimally. Inother instances, however, it is possible to be more specific. Consider the case ofChamorro, where some stems begin with an initial cluster. When infixed, forexample, tristi becomes trumisti, trinisti. The fully prefixed forms *umtristi,*intristi clearly violate both Onset and *Coda. If we were to position the infixafter the first consonant, yielding *tumristi, *tinristi, we would have repairedthe violations of Onset, but not those of *Coda. The actual forms avoid vio-lating *Coda at the cost of a greater degree of violation of LeftMost, suggestingthat it is the coda constraint rather than the requirement of onsets that is re-sponsible for the infixed position of “LeftMost” VC affixes in this language.

Both in Chamorro and in Sundanese, the conditions forcing infixationmay plausibly be seen as purely phonological. Most of the examples of infix-ation that have been treated in the recent literature have a similar basis in syl-lable structure or related phenomena, but in other cases, it is hard to see howthe placement of an infix has any effect in improving the syllable structure ofthe resulting form.

A number of Austro-Asiatic languages, for instance, make extensive use ofinfixes that are placed immediately after the first consonant of a word. Thus inKamhmuP4 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 resultingfrom 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 informationon this language is available in Svantesson (1983). Similar facts can be attested from a variety ofrelated languages, such as Atayal, Semai, Temiar, and Khasi.


c. krlam ‘pole carried on the shoulders’; cf. klam ‘carry suspended froma 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 wouldbe of adding rn to such a form. Nonetheless, there is no obvious sense in whichinfixation is preferable to plain prefixation in the forms of (6.7).

To achieve this, we might assume another constraint family NonIni-tial(e,D), each of whose members expresses the requirement that the element eshould not be initial within the domain D. To characterize an affix Afi as “sec-ond position” on this view, we would then say that NonInitial(Afi, D) dom-inates LeftMost(Afi, D). In the optimal form, then, the affix will appear in aposition 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 whetherthere 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 Chap-ter 3. I observed there that some instances of Stray Adjunction seem to operatein a way that suggests a preference within a given language for the left (or po-tentially, the right) edges of certain constituent types to coincide with lexical,rather than grammatical material. We might invoke the same general principlehere, at the word level, if we supposed that in KamhmuP the left edge of thegrammatical word is preferentially aligned with lexical rather than grammati-cal 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 theleft edge of the word, the optimal form will be one that violates left alignmentof rn by a minimal amount—a single segment.

This account has a disadvantage, however, from the point of view of the-ories of morphology such as that of Anderson (1992) or Stump (2001), sinceit presumes that the component parts of a morphologically complex word arevisible in the grammar. This can be avoided, though, if we replace (6.8) with afaithfulness requirement. The desired effect can be achieved if we rank highly(in the grammar of KamhmuP and similar languages) a constraint (6.9), to theeffect that the left edge of the input word must correspond to the left edge ofthe output word.


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

Such a constraint is presumably motivated by the fact that lexical iden-tification in many languages is highly sensitive to the beginnings of words.Requiring that the left edge be conserved under affixation can be seen as sup-porting this preference. And if we rank (6.9) above LeftMost(rn, Word), thisaffix will once again be placed after exactly a single segment at the beginningof the word in the optimal form.

There may be instances of infixation which must be achieved through aconstraint such as NonInitial(e, D), but we would like to avoid that wherepossible. In the case of word-level affixation, that can generally be achievedeither by invoking phonological factors or by appeal to (6.9).

Phrase-Level Affixation (Clitics) The extension of this apparatus to at leastsome classes of special clitics is obvious. Phrase-initial and phrase-final cliticsare directly analogous to prefixes and suffixes, respectively. The descriptionof their placement involves constraints from the same family, with clitic ele-ments specified as being attracted to the left or right edge of an appropriate do-main. The placement of second-position clitics, however, requires additionaldiscussion.

Recall that second-position clitics were claimed in section 4.2 to be thephrase-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 mor-phological affixes on the basis of morphosyntactic properties of a word,5 a ruleof special clitic formation specifies material that should be inserted into thephonological form corresponding to the phrasal domain within which the cl-itics are motivated. The location within this domain where the specified ma-terial appears is governed primarily by a constraint of the LeftMost family,with some other constraint C intervening to prevent them from occurring inabsolute phrase-initial position.

When we ask what that other constraint might be, the answer is not asimmediately obvious as in the word-level case. There are no known ex-amples, for instance, in which the placement of clitics in second position can

5 Derivational morphology is introduced similarly, as affixal material associated with the op-eration of a Word Formation Rule. The positioning of this material follows from a system ofconstraints just as in the inflectional case. Since many core cases of special clitics are similar ininteresting ways to productive inflectional morphology, I will discuss the general phenomenonin those terms, for concreteness’ sake, but the same mechanisms apply for more “derivational”material, both word level and phrase level.


be motivated by considerations of syllable structure. Nonetheless, there aresome 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 thesystem of second-position clitics in Warlpiri.

In this language, a cluster of clitics appears typically in second positionwithin the clause, composed of an auxiliary base (possibly ∅) followed by oneor more pronominal clitics. Word order in Warlpiri is generally quite free; thefixed positioning of the clitic cluster is striking in contrast, and suggests thatmechanisms 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 posi-tion or initially. He proposes a principle governing the placement of the cliticcluster as follows: when the base of the auxiliary is monosyllabic (or ∅), theauxiliary follows the first constituent of the sentence. When the base of theauxiliary is bisyllabic, however, the auxiliary can appear either initially or insecond position. This difference is illustrated in (6.10).

(6.10) a. Wawiri =ka=rna purra-mikangaroo PresImpf-1sgSbj cook-NPastI am cooking the kangaroo

b. Kapi=rna wawiri purra-miFut-1sgSbj kangaroo cook-NPastI will cook the kangaroo

The pronominal clitics appear immediately following the base, wherever thatis. The role of phonological form in this principle suggests strongly that thephonology should play a part in any account of it.

For the case of null or monosyllabic bases, we can note that the “minimalword” in Warlpiri, as in many languages, is bisyllabic. A monosyllabic ele-ment presumably cannot be independently footed, and thus must be treatedas prosodically deficient. I assume that all of the pronominal clitic elements inWarlpiri are non-lexical, and thus not analyzed as PWords, even if they containmore than a single syllable (e.g., -pala ‘non-1st dual’). Let us assume furtherthat stray adjunction in Warlpiri always operates so as to incorporate prosodi-cally 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 monosyllabicauxiliary base will have to be preceded by some other material. If we say thatauxiliary bases are subject to (highly ranked) LeftMost(cli), the furthest to the


left that they can go is after the first constituent,6 if they are to be prosodicallywell-formed. That accounts for their appearance in second position.

This account is supported by a remark Simpson (1991) makes, to the effectthat even short bases can be initial if the clause is closely linked phonologicallywith a preceding one. No example is provided in that source, but Jane Simpson(personal communication) has kindly provided one from the electronic files ofKen Hale’s original texts, which is reproduced here as (6.11). The monosyllabicauxiliary base which appears in clause-initial position is underlined.

(6.11) Jinarn-kiji-ni-ji. Ngulanya ka-rnalu ngarri-nitrip-throw-npast-euph. that pres-1pl.ex.sbj call-npastjinarn-kiji-ni-ji, kaji-lpa-npa watiya-rla-rlangutrip-throw-npast-euph, if-past-2sg.sbj log-loc-for.examplewanti-yarla. ka jinarn-kiji-ni.fall-irr, pres trip-throw-npast“Jinarn-kijirni” ‘to cause to fall’. We call it “jinarn-kijirni” if you fallover on, say, a piece of wood, it trips one up. (Hale 1966 notes [0125])

A bisyllabic base, unlike shorter ones, contains enough material to consti-tute a foot (and thus a PWord) on its own. Suppose we say that these baseshave two variants: one whose foot is structured as a PWord, and one where itis not a PWord. If we choose the prosodically autonomous form, then Left-Most(cli) will locate it in initial position. If we choose the deficient alternant,LeftMost(cli) cannot be fully satisfied, since that would yield a prosodicallyill-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, unlesspreceding 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 beenwidely cited in the literature), Legate (forthcoming) shows that it is basedon a mis-analysis. The bisyllabic elements such as kapi in (6.10) which Haletreated as Auxiliaries are not members of that class, but rather of the set ofcomplementizers. These latter appear optionally either in first or in secondposition, and their placement is without regard to number of syllables: thecomplementizer yi ‘for, since’ is monosyllabic, but shows the same (limited)freedom of placement as bisyllabic members of the class. Furthermore, evenwhen occurring in second position, the complementizers show behaviortypical 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 addressedbelow in section 6.2.


as observed earlier in section 4.1. Whatever the basis of this optionality, it mustbe grammatical rather than phonological.

While there may be languages for which a phonological account of thenon-occurrence of clitics in phrase-initial position is possible, no well sup-ported analyses of this kind have been presented. And in fact it is extremelyunlikely that such a prosodically based account will be adequate in general.That is because some special clitics that must be placed post-initially are notprosodically deficient. Tagalog, for instance, has a huge system of clitics, mostof which are prosodically autonomous and bear their own stress. There seemsno 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 syntacticphrase that constitutes the clitic’s domain to correspond to the left edge of theprosodic phrase that realizes it. On the theory being developed here, the cliticis not itself a syntactic constituent of its phrasal domain, but rather a phono-logical marker of some feature(s) associated with the domain. As a result, ifthe clitic itself appears in phrase-initial position, it will produce a violation ofthis 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 inboth directions in this language, aligning syntactic and prosodic boundarieswhere possible. The difference between ne, šti, and other clitics can be ascribedto the fact that while all are subject to LeftMost(cli), the LeftMost require-ments of ne and šti dominate the alignment constraint (which we might treatas 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 mo-tivated by faithful alignment, we can still achieve the required positioning bytreating the clitics in question as subject to a descriptive requirement that, al-though anchored by the left edge of their phrase, they must nevertheless notoccur in absolute initial position. A constraint NonInitial(cli), that is, domi-nates the otherwise highly ranked LeftMost(cli). In such a case, we can say thatthe placement of the clitics in second position represents a basic descriptivegeneralization, one not reducible to the effects of some other requirement(s).

The constraint family NonInitial(e), where we need to invoke it for thecase of clitic positioning, is well established in other areas. In particular, in theanalysis of stress and accent, NonInitial(e) and the mirror-image constraint


type NonFinal(e) play much the same role in constraint-based analyses thatExtrametricality (Hayes 1982) had in earlier, rule-based descriptions.

Second-position clitics and post-initial infixes are relatively common inthe languages of the world, but we have already noted that “penultimate po-sition” is at best rarely instantiated in the case of pre-final infixes and in thecorresponding clitic type. Whatever the role of NonInitial(e) and LeftEdge-Faith(P) in the description of morphological and phrasal affixes, their mirrorimages Non-Final(e) and RightEdgeFaith(P) are much less prominent, if in-deed they exist at all.7 It is possible to suggest as a motivation for this asym-metry that the identification (and stability) of left edges is important in itselffor morphosyntactic parsing, while no corresponding significance is attachedto right edges. On this picture, the apparent symmetry of the parameters forclitic position that arises on views such as those of Klavans (1985) and Ander-son (1992) is less pervasive than it initially appears—a conclusion which seemsto be warranted on empirical grounds.

6.2 Second Position: Anchors and Domains

At this point, we have the beginnings of a more “morphological” theory ofspecial clitics. Such a theory is based on mechanisms that modify the basicphonological realization of a syntactic phrase by introducing new segmentalmaterial (the simplest kind of “clitic”), or by making other modifications ofthe sort discussed in section 4.2, as a reflection of some of the morphosyntacticcontent of the phrase. The location of the phonological modifications thusintroduced is determined in a way entirely parallel to the way morphologicalaffixation (whether concatenative or not) is positioned: through a series ofconstraints, specifying preferred alignment for the elements in question in aninteraction with other influences. There are still many aspects of this theorythat remain to be made precise, however. In this section, I address two ofthese: the specification of the “first” element with respect to which a clitic mayappear in second position, and the specification of the domain within whichthe placement of a given clitic may be constrained.

Anchors for Clitics

A question that was finessed in the first approximation to a constraint-basedtheory of clitic placement above is that of how we ought to define “secondposition.” That is, what can appear in the first position within the same do-main? 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, atleast with respect to prosodically defined domains.


second-position clitic? In the case of infixation within the word, the infix eitherappears in a position determined by prosodic constraints on syllable structurewithin the word, or following some prosodically defined unit, such as after thefirst segment, syllable, or foot. We certainly do not want clitic positioning tobe determined in the same way, since the result would in general be to locatethe clitic somewhere within the initial word, and it is virtually never the casethat clitics appear in a position where they are properly included within a pre-existing PWord.8

To accommodate this observation, let us assume that in the derivation ofthe form of phrases, there is an overall constraint to the effect that materialcannot be inserted inside a PWord. Perhaps new material can be adjoined toa PWord at an edge, but an existing word can not be interrupted. Let us callthis constraint Integrity(Word). In virtually all languages, this constraint isundominated.

Now consider the case of a clitic cli subject to LeftMost(cli), where someother constraint prevents cli from appearing in absolute phrase-initial position.As a result of high-ranking Integrity(Word), there will always be at least onephonological word between the second position occupied by cli and the leftedge of the phrase, since the only way there could be less would be by violat-ing Integrity(Word); or else by violating the other constraint NonInitial(cli),9

which (ex hypothesi) dominates LeftMost(cli). But given LeftMost(cli), theonly way there could be more than one word before the clitic would be if someother higher-ranking constraint required it.

This apparatus is all we need to describe the specific case in which secondposition is interpreted as “following the first word of the phrase,” as in Hittite(and perhaps others of the ancient Indo-European languages, as Wackernagel1892 assumed) or West Greenlandic. For the somewhat commoner cases inwhich second position means “after the first syntactic phrasal daughter of thedomain-defining phrase,” however, something else is clearly necessary to forceplacement 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 elementsthat are not themselves members of that phrase.

Note that this will allow clitics to appear within the largest containingphrase 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 litera-ture will be addressed below.

9 For convenience, I represent whatever constraint forces second position by NonInitial(cli),although (as mentioned above) this is only one of the possibilities that might force such place-ment. This choice is not intended as a claim about the actual constraint that might be involved.


since they reflect properties of that phrase. It does, however, mean that theycan not appear within another phrase which is itself contained within thatdomain.

Since syntactic phrases come in many varieties, it seems plausible that In-tegrity(XP) is actually a family of constraints, such that in some languagessome phrasal types might be more “permeable” than others. In his carefulstudy of Serbo-Croatian word order, for instance, Boškovic (2001), followingmuch earlier literature, notes that while subparts of most phrasal types can beseparated, there are a few that cannot. As noted above in section 5.2, these in-clude certain prepositional phrases (at least those with the preposition prema);two-part proper names with overt case marking only on one member; andnominals including a governed genitive expression. Phrases of these types mustalways remain together as units, despite the general freedom of word order inthe language.

Boškovic takes these facts to be an argument for a theory that derives cliticpositioning via syntactic movement. He proposes that the generalization con-cerning first position is that exactly one syntactically autonomous constituentcan appear before the second-position clitics. That may be a multi-word con-stituent, but for many speakers of “eastern” or “Croatian” dialects, it can alsobe a single word—if and only if that word is capable of “scrambling” in a wayindependent of other words with which it forms a constituent. Syntacticallyautonomous movement thus correlates with the possibility of occupying firstposition; and Boškovic argues that this correlation goes unexpressed in a the-ory like the present one, where the set of constituents that must group togetherin first position is described by constraints of the Integrity family, rather thanby syntactic movement.

This objection does not hold, however. It is true that the set of constituenttypes which cannot be interrupted by clitics is described, on the view beingdeveloped here, by some collection of constraints of the type Integrity(P) forspecified phrasal types P. But in fact, exactly the same constraints can be takento be responsible for the facts Boškovic attributes to the syntax. That is, wecan assume that scrambling is (from the point of view of the syntax, thoughnot, presumably, with regard to semantics and discourse structure) a mat-ter of free re-ordering. Some constituents, however, cannot be broken upin the process—and it is precisely because these are subject to highly rankedIntegrity(P) that they are impermeable, both to clitic placement and to inter-ruption as a consequence of scrambling. Whatever generalizations there maybe about the set of constituent types that must remain intact under scrambling,exactly the same descriptive apparatus that accounts for this will simultane-ously account for their behavior with respect to clitic placement. There is thus


no lost generalization, and no concomitant argument against the present lineof analysis.

I assume, then, that XP in Integrity(XP) is parametrically variable.Integrity(Word) and Integrity(XP), obviously, are instances of the samefamily: constraints to the effect that material cannot be entirely containedwithin a domain of a certain sort unless it represents a member or elementof that domain.

Now suppose that the integrity constraints (including Integrity(P) forvarious values of P, like Integrity(Word)) are generally undominated. In thatcase, the earliest that “second position” can come is after the first phrasaldaughter of a type P in the containing phrase, and so that is exactly wheresecond-position clitics will be found. Where Integrity(XP) is lower rankedthan various relevant alignment constraints (but Integrity(Word) is still highlyranked), “second position” will mean “after the first word.” Variation in the in-terpretation of “second position,” accordingly, is a matter of variable rankingof 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 thephonological representation of a syntactic unit on the basis of properties ofthat unit (phrase, clause). Alignment constraints influence the positioning ofthe clitic within the form by preferring realizations at one edge or the other(LeftMost(e), RightMost(e)), representing the Anchoring parameter of earlierdescriptive theories. When the clitic is not positioned at the actual peripheryof the form (in particular, when it is located in second position) this is dueto some higher-ranking constraint: perhaps a requirement that the left edgeof the output form correspond to the left edge of the input; or an alignmentconstraint specifying that the left edge correspond to lexical material; or simplythe requirement that the clitic be NonInitial. These represent the interactionof the Anchoring and Precedence descriptive parameters, while constraints ofthe 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 Do-main within which special clitics appear. In most of the cases considered thusfar, 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 correspond-ing prosodic constituent. In fact, however, domain specifications are not al-ways that simple and direct, and we will have to allow for the possibility thatconstraints such as those introduced above have additional parameters speci-fying the domain of their applicability. Thus, a clitic is specified not simply asLeftMost and/or NonInitial, but LeftMost within some domain D, etc.


As I have already suggested, this provides us with a way to limit the de-scriptive apparatus somewhat. In addition to being anchored by the Initial orFinal element of a domain such as the clause, clitics sometimes appear to beanchored by the Head of that domain (in particular, the finite verb). That pos-sibility was not included in Klavans’s (1985) theory of the parameterization ofclitic 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 needto 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 dis-cussed, 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 Bulgarianand 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 casesby saying that properties of the larger phrase are inherited (via something likethe ‘Head Feature Convention’ of Gazdar, Klein, Pullum, and Sag 1985), andthe rules introducing clitics apply within a domain that is circumscribed tothat head verb. They are then subject to alignment constraints such as Left-Most(cli,V) which apply within that domain, ranked in such a way as to yieldthe observed order.

This line of analysis immediately yields a solution to the problem of how topreserve clitic order regardless of whether the actual clitic sequence precedes orfollows its anchoring verb. Ceteris paribus, a clitic constrained by high-rankingLeftMost(cli,V) will appear to the left of the lexical verb within the syntacticdomain V, and the order of several clitics within this domain will follow fromthe relative ranking of their corresponding LeftMost constraints. Suppose wesay now that an even higher-ranking constraint requires that infinitives (andsuch other non-finite forms as are relevant) must be aligned with the left edgeof the same domain. LeftMost(Verb[−Finite],V), that is, outranks all of the spe-cific 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 sin-gle additional constraint on the non-finite form: its LeftMost constraint dom-inates those of the clitics, while the LeftMost constraint applicable to finiteverbs is itself outranked by those applying to the clitics. But regardless of theposition of the verb itself within the head V domain, the relative ordering of itsassociated clitics remains the same, determined in all cases by the same rankingof clitic-specific alignment constraints.

In Bulgarian, there is an additional wrinkle. Even when linked to a fi-nite verb, clitics within the V domain follow the verb rather than preceding it


when the V itself is sentence initial. To accommodate the facts of V-initial sen-tences 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 insentence-initial position. This particular pattern, in which enclisis versus pro-clisis 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) theclitics 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 notedin section 6.1 above for the rule-based morphological approach to cliticpositioning.

In the descriptive theories of Klavans (1985) and Anderson (1992), a singledomain is relevant to both of the parameters of clitic placement (Anchoringand Precedence), but on the constraint-based view, since the constraints areseparate their relevant domains might be as well, as we have just seen for Bul-garian. A somewhat different example illustrating this theoretical possibility ispresented by Richardson (1997) for Czech. Richardson first shows that Czechin general has a collection of second-position clitics similar to those of Serbo-Croatian, as illustrated in (6.12).

(6.12) a. Vcera =jsem =se =mu konecne omluvilyesterday past-1sg refl 3sg-dat finally apologizedYesterday I finally apologized to him

b. Cervené tulipány =se objednalred tulips refl orderedHe did order red tulips

The second of these sentences suggests that what constitutes first position inCzech is a phrase, not a single PWord, and we assume that Integrity(XP) isranked above the LeftMost constraints relevant to the clitics.

In embedded sentences, clitics generally come immediately after the com-plementizer, as in (6.13).

(6.13) Rekl, že =mi =ho mužete ukázatsaid-ppl that 1sg-dat 3sg-acc can showHe said that you can show him to me


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

(6.14) a. Helena ríkala, [CPže [IP=se Petr odstehoval]]Helen said that refl Peter movedHelen said that Peter had moved

b. Helena ríkala, [CPže Petr [IP=se odstehoval]]Helen said that Peter refl movedHelen said of Peter that he had moved

The first of these is as we expect. But in the second, the clitic appears tobe in third position. Richardson argues that in this example, the subject Petris not in its normal position within the clause, but rather has been moved outand adjoined to IP as a sort of topic, as shown by the bracketing. What we seethen is that the clitics are really constrained differently in two domains: theyare drawn to the left edge of IP, but constrained not to be initial in CP. Theconstraint 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 placedcorrectly in second position when they occur there. In dependent clauses, how-ever, there is material in CP that is not part of IP. As long as that consists of onlya single element (the complementizer), the result still looks like simple second-position placement within CP. It is only when more than one element occursin CP outside of IP that we can tell the difference, and here we find that theposition of the clitic is bounded by the left edge of IP no matter how far thatmay be from the left edge of CP.

Another analysis in which LeftMost and NonInitial have distinct domainsis offered for European Portuguese by Barbosa (1996). In this language, cli-tics are attracted to the Verb, but they appear either preceding or following itdepending on various factors:

(6.16) a. Ninguém/Alguém o= viuNo one/Someone 3sg sawNo one/Someone saw him

b. O João não a= viuthe John not her sawJohn didn’t see her


c. O João nunca a= vêthe John immediately her seesJohn sees her immediately

d. Viu =o o Joãosaw him the JohnJohn saw him

e. Esses livros, dei =os á Mariathose books 1sg-gave them to MaryThose books, I gave (them) to Mary

f. A Maria viu =othe Mary saw himMary saw him

Barbosa argues that the clitics are attracted to the left edge of a verbal pro-jection (which she identifies as IP, a specification that I would replace with thatof the finite V), but that they are prohibited from appearing initially in an Into-national Phrase. This is obvious in the case of V-initial sentences, and plausiblewhen something is clearly dislocated. She also argues that referential subjectsare always dislocated, thus establishing a new IntPhrase boundary. This couldbe 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, anda syntactic one relevant to the other.

Galves and Sandalo (2004) argue that while Barbosa’s analysis is basicallycorrect for Classical Portuguese, the modern language is subtly different. Theyshow that in the language as written in Portugal between the fifteenth and theeighteenth century, proclisis and enclisis are both found in sentences with def-inite subjects like (6.16f ), and that the variation is correlated with other factors(contrastive focus, constituent length) that are plausibly related to the issueof whether the initial subject constitutes a separate phrase prosodically. Theyargue that a change has occurred in modern European Portuguese, however.Definite (and pronoun) subjects now uniformly trigger enclisis under circum-stances that cannot be based on prosodic differences, and so the domain of therelevant NonInitial constraint must have shifted from a prosodic category to agrammatical one some time around the end of the eighteenth century. It wouldtake us too far afield to examine in detail the syntactic assumptions on whichtheir reanalysis is based; I simply note their proposal as potential evidence thatsuch a historical change may be possible.


Another instance in which syntactic and prosodic domains interact is pro-vided by the facts noted by Boškovic to which I referred in section 5.2 above.Recall that in Serbo-Croatian, clitics do not appear where we expect themwhen unusual intonation intervenes, as in the examples of (5.17) repeated hereas (6.18).

(6.18) a. Sa Petrom Petrovicem srela =se samo Milenawith Peter Petrovic met refl only MilenaWith Peter Petrovic, only Milena met

b. Znaci da, kao što rekoh, oni =ce sutra docimeans that as said they aux tomorrow arriveIt means that, as I said, they will arrive tomorrow

c. Ja, tvoja mama, obecala =sam =ti sladoledI your mother promised aux 2sg ice creamI, your mother, promised you an ice cream

In each of these cases, we would expect the clitics to come after the initialphrase, but they do not. As Boškovic notes, the basic regularity (5.18) is thatthe clitics appear in second position within an intonational phrase. We mightsuggest that in Serbo-Croatian, the clitics are attracted to the left edge of the(syntactic) clause, but prevented from occurring initially in the Intonationalphrase, 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 inthe sentences with unusual intonation, however, since it would appear thattheir optimal location under this ranking would be at the end of the initialIntPhrase, rather than in second position within a later one. We can note with anumber 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 followedby an intonation break would be worse than having them come later in thesentence (after the parenthetical, appositive, etc.). This approach is promising,but there must be more to the matter than this, since Serbo-Croatian cliticssometimes do occur in IntPhrase final position in sentences like (6.21).

(6.21) Moja sestra, koja =je u Sarajevo, sjeca =vas =semy sister who 3sg-pres in Sarajevo remembers 2pl refl

My sister, who is in Sarajevo, remembers you


Location of the clitics in this example after the first constituent of either thefirst or the last IntPhrase will violate (6.20). Since the position at the end ofMoja sestra is closer to the left edge of the clause, this would appear (counter-factually) 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 cliticinsertion. What we need to say to implement Boškovic’s insight about theseexamples is that where a clause is divided into multiple IntPhrases, the cliticsare introduced within the IntPhrase that contains the head of the clause (themain verb). It is then constrained to occur as close as possible to the left edgeof that phrase, but not initially. On this account, all of the relevant domainparameters refer to prosodically defined units.

As a final example, let us consider some interesting facts about Bulgarianclitics pointed out by Legendre (2000a). Like those in Portuguese, these cliticsare introduced within a projection of V, identified by Legendre as V ′. As inEuropean Portuguese, most of the clitics are pre-verbal except when the verb issentence initial, as shown in (6.22).

(6.22) a. Az =sum =mu =go dalI aux-1sg him it givenI have given it to him

b. Dal =sum =mu =gogiven aux-1sg him itI have given it to him

When multiple wh-words appear at the left edge of the clause, the cliticsmay be located in a position within the clause later than second, as illustratedin (6.23).

(6.23) Koj kakvo mu= e= dal?who what him aux-3sg givenWho 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 thatwhile some of the Bulgarian clitics behave in this way, others are allowed toappear 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 toan additional constraint requiring its left edge to be aligned with the right edgeof a prosodic word. Sentences like (6.24) illustrate these possibilities.

(6.24) Šte= go= viždaš =lifut him see-2sg QWill you see him?


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

6.3 “Endoclitics”

One property of special clitics which has been widely accepted as general atleast since the appearance of Zwicky’s (1977) survey is the claim that these donot appear properly included within a word. That is, while there are both pro-clitics and enclitics, there are no “mesoclitics” or “endoclitics.” Some exam-ples have been discussed in the literature, however, which appear to controvertthis. With the apparatus developed to this point, I can address the three mostprominent cases of this sort: Pashto clitics and verbs with prefixal stress, Euro-pean Portuguese pronominal clitics with certain specific verbal forms, and theclitics of Udi.

Pashto

The second-position clitics of Pashto were first discussed in this connectionby Tegey (1977), who noted cases in which they appear to interrupt a mono-morphemic word. Kaisse (1981, 1985) argued that the relevant cases should beanalyzed in a somewhat less dramatic way, but that the clitics still should beseen as interrupting lexical units. Subsequent discussion in terms of Optimal-ity Theoretic analyses can be found in van der Leeuw (1995a, 1997) and Roberts(1997, 2000). The present discussion largely (but not entirely) follows that ofRoberts (1997). Most of the examples below are drawn from that work, whichcites them in turn from Tegey (1977).

Pashto has a set of clitics that appear in second position, in a sense tobe made clear below.10 These include elements of several sorts, as shownin (6.25).

(6.25) Pronominal modalme 1sg ba will, might, must, should, mayde 2sg de should, had better, letye 3sg, 3pl adverbialam 1pl, 2pl xo indeed, really, of coursemo 1pl, 2pl no then

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


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

(6.26) a. xušal =me zyaati ne wah-iKhosal 1sg anymore neg hit-3sgPresKhosal doesn’t hit me anymore

b. zyaati =me ne wah-ianymore 1sg neg hit-3sgPresHe doesn’t hit me anymore

c. ne =me wah-ineg 1sg hit-3sgPresHe doesn’t hit me

d. wah-i =mehit-3sgPres 1sg

He hits me

e. z@ GwAó@m ce tor =me wuguriI want comp Tor 1sg seeI want Tor to see me

Sentence (6.26d) is particularly interesting, in light of the fairly rigid verb-finalstructure of sentences in Pashto.

When the initial constituent of the sentence is a multi-word nominal or apostpositional phrase, the clitics follow the entire phrase rather than its initialword, as shown in (6.27).

(6.27) a. aGa šel kal@na danga aw khAysta peGla =me n@nthat 20 year tall and pretty girl 1sg todaybyA w@lidaagain sawI saw that twenty-year-old tall and pretty girl again today

b. xušal aw patang =ba =ye d@r ta rAwóiKhosal and Patang will 3sg you to bringKhosal and Patang will bring it to you

c. laylA na =de A-xist@Layla from 2sg buyYou 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@ rAwalidonkey should neg bringHe should not bring the donkey


From these facts, I conclude that the clitics in (6.25) are each oriented to-ward the left edge of IP, but this alignment is outranked by the requirementthat they not be initial within that domain, and also by the requirement thatDPs and PPs not be interrupted. The resulting constraint ranking is givenin (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 thegeneral 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 inthe clitic’s appearing far from the left edge of the clause, as in (6.30).

(6.30) ra ta te rA škAw@ =deme 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 defi-cient 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 withthat PWord. The fact that second-position clitics do not interrupt this se-quence suggests that Integrity(PPhrase) appears with the other highly rankedIntegrity constraints in (6.29).

The role of prosodic structure in these matters is highlighted by the mini-mal pair of sentences in (6.31).

(6.31) a. rA sara wí =deme with be letLet it be with me

b. rA sará =de wime with let beLet it be with me

When the normally unstressed PP rA sara ‘with me’ is assigned contrastive orfocal stress, the clitic follows it rather than the copula wi. This is presum-ably because the presence of stress promotes the PP to the status of a separatePPhrase 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 withoutinterrupting a PPhrase.

So what has any of this to do with the question of “endoclitics”? The an-swer emerges when we consider aspectual distinctions in Pashto verbs. The


difference between imperfective and perfective forms is often indicated by ashift of the location of stress. In verbs containing a prefix, or consisting of anoun, adjective, or adverb plus a “light” verb, the imperfective generally hasstress on the second element, but shifts this to the first element in the per-fective. Thus, úel-wAhé ‘(I) was pushing it,’ but úél-wAhe ‘I pushed it’; pore-w@st@ ‘(I) was carrying it across,’ but póre-west@ ‘I carried it across.’ In somemonomorphemic verbs, a prefix is recruited in the perfect to bear this shiftedstress: 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@ =mepfx push 1sg

I was pushing it

ii. úél =me wAh@pfx 1sg pushI pushed it

b. i. pore- west@ =meacross carry 1sg

I was carrying it across

ii. póre =me west@across 1sg carryI carried it across

c. i. pezand@ =derecognize 2sg

You were recognizing (him)

ii w@ =de pezand@pfx 2sg recognizeYou 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 “endocli-tics,” but it seems at least equally reasonable to see each part of these construc-tions as a distinct unit—let us call it a ‘lexical word’ – in which case the cliticsdo not interrupt lexical words.

There is another (small) class of verbs showing an alternation in the loca-tion of stress within a monomorphemic stem, where either possibility occursin imperfective forms. Crucially, the location of stress in these cases is notrelevant to the location of clitics, as shown in (6.33).


(6.33) a. p@rebd@ =mebeat 1sg

I was beating (him)

b. p@rebd@ =mebeat 1sg

I was beating (him)

c. *p@=me rebd@

These cases confirm that clitic placement cannot in general interrupt a sin-gle lexical word. This suggests that Integrity(LexWord) is among the high-ranking 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 dependingon the location of the stress as in (6.34).

(6.34) a. i. axist@l@ =mebuy 1sg

I was buying (them)

ii. á =me xist@l@? 1sg buyI was buying (them)

b. i. aGust@ =mewear 1sg

I was wearing (it)

ii. á =me Gust@? 1sg wearI was wearing (it)

If verbs such as those in (6.34) are genuinely single lexical words, the (ii) sen-tences here involve real endoclitics, and this is the crux of the issue posed byPashto 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 behav-ior of other complex verbs such as those in (6.32). Tegey (1977) argues thatthis is incorrect: while some of these verbs actually represent the continuationof forms with a distinct prefix a-, others (such as ax- ‘buy’) are monomor-phemic stems that never contained such a prefix. Kaisse (1981, 1985), however,argues (correctly, in my view) that this historical difference is irrelevant, and


that there is no reason not to assume that many verbs with initial a- that wasnot historically a distinct prefix have been reanalyzed so that synchronicallytheir structure is like that of the complex verbs in (6.32). In that case, there areno instances of real “endoclitics” to consider in Pashto.

The analysis suggested here, then, is as follows. The prosodic structureof Pashto involves assigning PWord status to any lexical element that bears astress. PPhrases are constructed on the basis of PWords, including precedingunstressed material as free clitics within the PPhrase and ending at the rightedge of the head PWord. The second-position clitics in (6.25) are then all sub-ject to LeftMost(cli, IP), itself outranked by NonInitial(cli, IP).11 The place inwhich the clitics actually lodge is further constrained not to interrupt a lexicalword, a (syntactic) DP, or a (prosodic) PPhrase. If we assume that prosod-ically deficient postpositions are incorporated as affixal clitics into a preced-ing PWord, we can dispense with the constraint Integrity(PP) appealed toin (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 withinthe domain of V. They are then subject to LeftMost(cli,V), but these constraintsare outranked by NonInitial(cli,IntPhr). The result is proclisis to the finiteverb unless this is initial within an IntPhrase (or some other constituent, asGalves and Sandalo 2004 argue for Modern Portuguese), in which case enclisisresults. This is indeed the normal state of affairs, but under specific circum-stances there are additional complications. In particular, when the verb is aform in the synthetic future or conditional, the clitic appears to be located in-ternal to the form, between the stem and the ending. This apparent case ofendocliticization has been discussed in Optimality Theoretic terms by van derLeeuw (1995b, 1997), and the prosodic organization of the relevant forms isclearly established by Vigário (2003). I follow those discussions in part, thoughthe analysis below diverges in some ways from them.

The forms involved show the endings in Figure 6.1, added to a stem whichis usually the same as the infinitive, except in a small number of (rather fre-quent) verbs.

11 In place of NonInitial(cli, IP), much of the literature on Pashto in OT terms appeals to analignment constraint requiring the clitics to be aligned such that their left edge coincides with theright edge of a PPhrase. The difference between that description and the one assumed here willnot be explored further, because it does not bear on the question of possible “endoclitics” withwhich I am concerned.


Future Conditional1sg -éi -ía2sg -ás -ías3sg -á -ía1pl -émos -íamos2pl -éis -íeis3pl -ão -íam

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

When there is no clitic present, the resultant form has a single stress, lo-cated on the ending: daríamos ‘(we) would give,’ perceberás ‘(you sg.) will un-derstand,’ dirá ‘(s/he) will tell,’ mostrará ‘(s/he) will show.’ When enclitics areadded, 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.’ Notealso that forms with enclitics following the entire verbal form are quite impos-sible: *dariamos-te, *perceberas-me, *dira-lhe, *mostrara-no-los. There are twoquestions to answer, then: (a) when the clitic cannot be proclitic, why does itappear before the ending and not after the entire verb form; and (b) why isthere an extra stress on the stem precisely when a clitic intervenes between itand the ending?

The history of the forms in question provides some tantalizing hints aboutthe answers. As opposed to all other tense forms in the paradigm of the Por-tuguese verb, the inflected future and conditional result from the relatively re-cent fusion of a separate auxiliary verb (a form of haver ‘to have’) with theinfinitive form of the lexical verb which it formerly governed. This is similar todevelopments in some other Romance languages (e.g., French); it is also quiteconsistent with a comparison of the endings for the future and conditional inFigure 6.1 to the present and past forms of haver provided in Figure 6.2.

Present Past1sg hei havia2sg hás havias3sg há havia1pl havemos havíamos2pl haveis havíeis3pl hão haviam

Figure 6.2. European Portuguese haver ‘to have’


If we could regard the future and conditional “endings” and the infinitivebase 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 theauxiliary verb, and the resulting form would show two stresses because thereare two distinct words, each of which bears stress. Such an analysis involvingformation of the future and conditional in the syntax, rather than the mor-phology, is argued for by several authors, including Vigário (2003). From asyntactic point of view, however, it faces a number of problems, discussed inpart by Crysmann (2000). In addition, it fails to account (except by stipula-tion) for why two stresses should appear when an “endoclitic” is present, butnot otherwise. This last fact suggests that prosodic structure is relevant.

An alternative is available, however, that allows us to have our cake andeat it too. This is the proposal that while the future and conditional forms arenot internally complex from the point of view of the syntax, they display inter-nal structure with respect to the morphology (and the prosodic phonology).In Anderson (1992), I argued that most forms composed by word formationrules on the basis of other forms do not have an internal structure of the sortstandardly assumed, with morpheme divisions represented structurally. Theonly internal structure, once a morphologically complex form has been con-structed is that motivated by the phonology, a position shared with some otherviews of morphology such as that of Stump (2001). While this is the case formost morphologically complex forms, however, there are others for which it ismuch less attractive. Compounds are an obvious example, since they appear inmany languages to display distinct morphological and phonological domainsfor each of their elements. I suggested in Anderson (1992) that some non-compound formations in some languages, which I labeled “composites,” alsodisplay some internal structure of a non-phonological sort. Icelandic “middlevoice” verbs such as kallast ‘be called X, have the name X,’ which derive his-torically from the fusion of a reflexive pronoun with the verb, have a structuresuch as that in (6.35).

(6.35) VV

kalla st

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


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

(6.36) VLexWord

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 thatof 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 syntac-tic verbal domain which the verb in (6.36) interprets. If the verb is in medialposition within an IntPhrase, of course, the clitic is adjoined at the left as aproclitic, and nothing of further interest occurs. If the verb is initial withinIntPhrase, however, the clitic cannot be realized in LeftMost position withinthe verbal domain. It must, however, be positioned as close to the left edgeof this domain as possible, consistent with other, higher-ranking constraints.Suppose that in Portuguese lexical words (though not PWords) are subject tohigh-ranked Integrity. The clitic will then be positioned at the right edge ofthe lexical word, but preceding the conditional ending.

The ranking of Integrity(LexWd), then, in conjunction with what we al-ready know about European Portuguese clitics, accounts for the positioningof the clitics in the future and conditional forms. The additional stress resultsfrom a prosodic reorganization which affects them when the stem of the struc-ture in (6.36) acquires an enclitic. Enclitics in Portuguese are structured asaffixal clitics together with their hosts, as shown in detail by Vigário (2003) onthe basis of a variety of phonological processes. The verbal stem thus becomesthe head of an independent PWord, resulting in the emergence of stress on thestem. This, in turn, requires the desinence to be restructured as a PWord in itsown 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 com-pounds, as well as these forms with internal clitics. It may or may not be iden-tifiable with the PPhrase, a matter which I will not address here.


Enclitics associated with future and conditional forms in European Por-tuguese, then, are not really “endoclitics.” They do not interrupt lexical words,and their position within prosodic structure is not properly included withina PWord, but only within a higher prosodic type associated with internallycomplex units, compounds and other composites. Their positioning followsfrom the constraints applicable in the language, as a relatively straightforwardvariation on the notion of second position, here relative to the category V.

Udi

Without doubt, the most extensively argued (and most successful) case for gen-uine endoclitics is that presented by Harris (2000, 2002) for the North-EastCaucasian language Udi. The facts here are intricate, and it is quite impossiblefor me to review them comprehensively here (especially since that has alreadybeen done quite thoroughly in Harris (2000), to which the reader is referredfor 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 theanalysis of endoclitics.

Udi has a set of person-marking clitics that show agreement with the sub-ject of a clause.12 There are four sets of these, shown in Figure 6.3. The sin-gle 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 ofother verbs, and the “General” set elsewhere. The choice of one of the variantsof a given form is entirely based on phonological factors.

Under some circumstances, these appear as simple enclitics. Certain spe-cific categories of Tense and Aspect, for example, require the person-marking

General Inversion Possession Question1sg -zu, -z -za -bez, -bes2sg -nu, -n, -ru, -lu -va -vi3sg -ne, -le, -re -t’u -t’a -a1pl -yan -ya -beš2pl -nan, -ran, -lan -va, -va.n -e.f3pl -q’un -q’o -q’o

Figure 6.3. Person Markers in Udi

12 The behavior of another clitic -qa- “Subjunctive” falls together with that of the personmarkers, and will not be discussed separately here.


clitics to follow them in their position after the verb. Also, when some con-stituent of the clause bears focus, the person markers immediately follow it.This includes negative particles and questioned constituents, which are oblig-atorily focused, as well as any other focused element. Finally, when the clauseinvolves predication by means of a phonetically null copula, the person markerfollows the associated predicate nominal. We can note that in all of these casesthe clitics are constrained by high-ranking RightMost(cli) within the domainin which they appear, and nothing else prevents them from appearing as finalenclitics.

The interesting issues arise when no element in the clause takes focus inone of these ways. In that case, the person markers appear together with theverb itself, of which there are two principal varieties to be considered. SomeUdi verbs are simple, monomorphemic stems (at least from a synchronic pointof view), but many others are internally complex. These consist of an initialelement which may be practically anything: nouns, adjective, locative pre-verbs, historically simplex verb stems, borrowed verbs, infinitives (to formcausatives), or opaque elements that do not occur independently. This ini-tial element (which may itself be internally complex) is followed by one of anumber of what Harris refers to as “light” verbs. Some of these occur indepen-dently with a lexical sense which may or may not be relevant to the meaning ofcomplexes that they head, while others occur only in this use. When the verb isof this type, the person marker is placed immediately before its final light verb,as in the examples of (6.38).

(6.38) a. . . . pascaG-un Gar-muG-on lašk’o=q’un-b-esaking-gen boy-pl-erg wedding-3pl-do-pres

The king’s sons marry [the girls they had rescued]

b. nana-n äyel-ax ak’-es=ne-d-e k’ucanmother-erg child-dat see-inf-3sg-caus-aorII puppy.absThe mother showed a puppy to the child

c. áyel kala=ne-bak-echild.abs big-3sg-become-aorIIThe child grew (up)

Let us assume that when the person-marking clitics are associated with theverb (perhaps because the verb takes default focus in the absence of another fo-cused constituent), they are introduced into a domain consisting of exactly thelexical 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 fu-ture and conditional in Portuguese. Harris provides a wide range of arguments


that these complex verbs constitute lexical and syntactic units, but these do notcontradict the claim that from a morphological point of view, they are com-posed of more than one distinct unit. And it is precisely between these piecesthat the clitics are located.

From the cases in which the person markers are associated with some-thing other than the verb, we know that their positioning is governed byRightMost(cli,D). If that were all there were to the story, we would expectthat when associated with the verb, they would follow it, but instead they im-mediately precede its final element. In the discussion of KamhmuP above insection 6.1, I appealed to a constraint LeftEdgeFaith(Word) (6.9) to ensure thatthe infix -rn- in that language follows exactly the initial segment of the word inwhich it appears. What we need in Udi, apparently, is a similar constraint onthe 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 corre-spond to the element at the right edge in the input.

If RightEdgeFaith(Verb) dominates RightMost(cli,Verb), the clitics willbe located in an infixed position. What we observe is that they precede thefinal structural component of a complex verb, regardless of the size of thatelement. This suggests that an appropriate constraint from the Integrity fam-ily also dominates RightMost(cli,Verb). For convenience, I will assume thatthe components of a complex verb are all lexical words, although this specificchoice is not important: there is no doubt that the components of a complexverb are structural units of some kind, and the correct constraint will requirethat these units not be interrupted.

Now, however, we come to the central examples. When the verb is notinternally complex, but consists of only a single unit, we find the clitics locatedbefore the final consonant of that unit, as in the example of (6.40).13

(6.40) a. pascaG-un Gar-en gölö be.=ne-G-sa met’a-laxoking-gen boy-erg much look1-3sg-look2-pres this.gen-onThe prince looks at this for a long time (root be.G ‘look’)

b. kaGuz-ax a=z-q’-eletter-dat receive1-1sg-receive2-aorIII received the letter (root aq’ ‘receive’)

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


c. q’acaG-G-on bez tänginax baš=q’un-q’-ethief-pl-erg my money.dat steal1-3pl-steal2-aorIIThieves stole my money (root bašq’ ‘steal’)

d. ba=ne-k-sa sa pašcaG-k’ena adamarbe1-3sg-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 re-ally are clitics, and (b) there is no motivation for seeing the material on eitherside of these clitics as a structurally autonomous unit.

We do not really have to add more to the analysis above in order to accom-modate these forms, however. True, the constraints already introduced for Udiwould be expected to place a clitic, when introduced in the domain of a verbthat is not structurally decomposable, at its left edge (the rightmost positionthat preserves the right edge of the verb and does not interrupt a structuralunit). But we can remedy this incorrect prediction by invoking a condition offaithfulness at the left edge of the verb as well. Constraint (6.41) is simply themirror image of (6.39).

(6.41) LeftEdgeFaith(Verb): The element at the left edge (of the verb) inthe output form should correspond to the element at the left edgein the input.

Assuming that LeftEdgeFaith(Verb) dominates Integrity(LexWd), a singlestructural unit will be interrupted precisely when necessary to preserve Input-Output Faithfulness at the edges of the verb. In this case, the clitic will ne-cessarily be positioned within the verb, as far to the right as possible withoutbeing at the right edge: that is, before the final consonant of the root. Theoverall 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 inlanguage—either in the limited sense of clitics introduced between elementsof a composite form, as in Pashto and European Portuguese, or in the evenmore 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 correctlywith verb stems consisting of a single consonant, which are treated in Harris (2002: 127ff) alongwith some other cases as exceptions to the generalizations above. I do not deal with these excep-tional forms here, but they do not appear to present problems for the proposed analysis.


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

6.4 Tagalog Second-Position Clitics

To illustrate the application and force of the ideas explored in this chapter, Iprovide here an account of the system of second-position clitics in Tagalog, alanguage with a particularly interesting and intricate system of clitics.

Tagalog (Schachter and Otanes 1972; Schachter 1973) displays clitic ele-ments with two distinct sorts of function, all following the first element ofthe sentence. One set of these, consisting of pronominal elements representingarguments of the clause, is presented in Figure 6.4.

Within the set of pronominal clitics, there are two subsets, depending onthe grammatical role of the element they represent. Since the analysis of gram-matical relations in Tagalog clause structure is not at issue here, and nothing inparticular follows from the terminology, I simply follow Schachter in referringto these as the “Topic” set and the “Complement” set.

A second set of clitics consists of a set of “particles” filling a variety offunctions, as given in Figure 6.5.

All of the elements in Figures 6.4 and 6.5 constitute special clitics in thesense of Chapter 2, because they are all subject to a particular regularity intheir placement: all come in second position, following the first element of

“Topic” “Complement”1sg ako ko2sg ka mo3sg siya niyadual kata nita1pl (excl) kami namin1pl (incl) tayo natin2pl kayo ninyo3pl sila nila2sg(T)+1sg(C): ka+ko→kita

Figure 6.4. Tagalog Pronominal Clitics


ba (interrogative) na ‘already’kasi ‘because’ naman ‘instead’kaya (speculation) nga ‘really’daw (reported speech) pa ‘still’din ‘too’ pala (surprise)ho (politeness) po (politeness)lamang ‘only’ sana (optative)man ‘even’ tuloy ‘as a result’muna ‘for a while’ yata (uncertainty)

Figure 6.5. Tagalog Particle Clitics

the sentence. When more than one clitic is present, other regularities governtheir sequence relative to one another, and these will be the object of discussionbelow.

Descriptive Regularities

The overall framework for the description of clitics in the present book at-tributes their complex surface properties to the interaction of (individuallyquite simple) principles from a number of areas of grammar: at least phonol-ogy, morphology, and syntax. The analysis of the Tagalog clitics illustratesvirtually the full range of considerations that can interact in determining theplacement of clitics.

The property of appearing in second position characterizes particularclitics in Tagalog, and not the entire language. It is not the case, that is, thatall phonologically weak elements appear in this position (as Wackernagel as-sumed in his 1892 analysis of the early Indo-European languages), since Tagalogalso has a set of particles a, e, ha, o (Schachter and Otanes 1972: 461–3) whichappear 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 sub-ject of some subsequent discussion. The basic ordering regularities are thosestated in (6.43).

(6.43) a. Monosyllabic members of the set of pronoun clitics in (6.4) alwaysprecede particles (the elements listed in Figure 6.5).

b. The particles in Figure 6.5 have an internal ordering among them-selves.

c. Disyllabic pronoun clitics follow all particles.


Within each of these three subsets of clitics, there is a fixed relative order-ing. In the case of the two subsets of pronominals, this is presumably stipu-lated by the ranking of corresponding LeftMost constraints, since it does notappear 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 sev-eral subclasses, and I presume their relative order can be made to follow from acombination of semantic scope and language-particular stipulation, much likethe relative order of derivational elements in morphology, though I will notattempt to develop the relevant principles here.

There is some agreement in the descriptive literature, such as Bloomfield1917 and Schachter and Otanes 1972, that the clitic elements in figures 6.4 and 6.5generally follow the first stress-bearing word of the sentence, though there areas usual some complications in this picture. The examples that are most di-rectly consistent with this simple picture involve clitics that appear to inter-rupt otherwise unitary constituents so as to come after a single sentence-initialword, as in (6.44).

(6.44) Ganu =ka =na =ba kakinis?how you already int cleverHow clever are you now? (Bloomfield 1917: 143)

On the other hand, there are various constructions consisting of more thanone phonological word that do not get broken up (Schachter and Otanes 1972:187ff): proper names, numerical expressions, times of day, ages, amounts ofmoney, etc., as in (6.45).

(6.45) Bukas ng gabi nang alas otso =siya aalistomorrow night at eight o’clock he leavesIt’s tomorrow night at eight that he’s leaving (Schachter and Otanes1972: 188)

Schachter and Otanes (1972) describe these facts by treating certain con-struction types, ones they call “obligatory non-pre-enclitics,” as uninterrupt-able. In the present theory this is expressed by an appropriately parameterizedhigh-ranking Integrity constraint. I have nothing particularly illuminating tosay here about why some constructions have this property while others do not.Where a sentence-initial constituent can be interrupted, Integrity(Word) incombination with NonInitial(cli,D) nonetheless prevents clitics from appear-ing 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 do-main that excludes certain preposed, focused constituents. I will discuss this


point in more detail below; for now let us simply assume that the domainwithin which some element appears in “second position” can be identified withthe one within which it is introduced, and that this follows (in some way) fromits grammatical properties. These complications do not affect the major pointsto be made below: what matters is that “second position” has the same basicsense 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 purelysyntactic mechanisms. First, there is the usual difficulty of placing clitics “afterthe first word,” a notion that does not have a natural reconstruction in termsof those parts of grammar that normally deal in terms of phrases as basic ele-ments. This is especially true if the relevant notion of “word” is a phonologicalone, which would result in difficulties for a principle of “Phonology-free Syn-tax” (Pullum and Zwicky 1988). A different (and potentially more serious)puzzle is how to derive the regularities of order among clitics. If these are asstated in (6.43) above, it is not clear that they can be described syntactically atall.

This matter is the subject of Schachter (1973): the problem is that the or-der appears to depend on the phonological property of how many syllables agiven clitic has; and syntactic rules, even of the sort that were accepted backin 1973, ought not to have access to the specifics of the phonological realiza-tion of elements they affect. But how can this be avoided? Note in particularthe portmanteau element kita. This represents the combination of a 2sg.Topicand a 1sg.Complement, and replaces the expected sequence ka+ko. Each ofthe latter elements individually is monosyllabic, and thus should precede anyparticles (by 6.43a), but when they come together, the combination is replacedby a disyllabic portmanteau which must follow particles (by 6.43c). Even a setof rules that place the pronouns (as syntactic elements) one by one, and thattreats the length regularities as accidental, will fail in this case.

Schachter (1973) evaluated several brute force solutions to these difficul-ties utilizing reordering transformations, and showed that they either fail al-together, or at least fail to capture the generalization about length. His ownproposal, following Perlmutter’s (1971) analysis of French and Spanish clitics,is a surface constraint: a template like (6.46), stipulating explicitly that mono-syllabic pronouns precede particles, and that these in turn precede longer pro-nouns. This, he suggests, could serve either to filter the output of a scramblingrule or else to linearize elements at surface structure.

(6.46) Pro1σ < Particles < Pro2σ


Such a description would allow us to state the regularity about syllable struc-ture, but note that it does not explain either of the relevant facts: (a) monosyl-labic pronouns come first; and (b) otherwise, pronouns come after particles.Subsequent work (e.g., Kroeger 1993) similarly leaves this pattern as a matterof stipulation.

In Anderson (1992), I noted that an explanation of the fact that at leastmost of the pronouns follow particles might be derived from the analysis ofclitics as sentential morphology. Within words, derivational morphologycomes ‘inside of ’ inflectional morphology. If we ignore the monosyllabic pro-nouns for the moment, and think of clitics as suffixes to the first word orphrase, then we could say the ordering of particles before pronouns reflects thesame regularity: particles have various semantic and pragmatic content, andare thus “derivational,” as opposed to pronouns, which are (on this analysis)agreement markers, and thus “inflectional.” It would be natural for pronomi-nal 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-internalcase, this follows (as argued in Anderson 1992) from the place of lexical inser-tion in a grammar, given the nature of (productive) inflection. What needsto be shown is that in the case of clitics, the corresponding result also followsfrom the architecture of grammar.

Adopting the overall terms of Chomsky (1995), the introduction of seman-tically contentful particle clitics can be seen as a generalized instance of the op-eration “Merge.” In the development of complex syntactic structures, the usualstep 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 clausalagreement- like material has to be realized phonologically, the particle cliticswill all be in place at the point where the structure exists to support theintroduction 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 withclitics, it remains syntactically inert and is not visible for syntactic purposes in the way lexicalwords are.


Obviously, this involves serialism in the overall derivation of a sentence—a notion which is not fundamentally at odds with the invocation of principlesfrom OT to describe what happens at a given stage of the derivation, though itis at odds with the common (but incorrect) assumption that constraint systemsmust always involve a single monolithic collection of constraints applied inparallel. As noted above, much of the non-monolithic architecture of LexicalPhonology is in fact perfectly consistent with the basic insights of a constraint-based approach such as OT.

But what about the syllable-length regularity? It is clearly an embarrass-ment for this account to have some “inflectional” clitics coming inside of the“derivational” ones. In Anderson (1992), I proposed to treat this by saying thatthe monosyllabic clitics are introduced by “head-inflection” rules, as opposedto all the others, which are “word inflection.” This, however, still misses thefact that the difference depends on prosodic size. In a step backward fromSchachter’s analysis, it does not even state this fact, let alone explain it.

Suppose that the relation between particles and disyllabic pronouns doesindeed follow from an appropriate generalization of the “derivation precedesinflection” theorem. That is, all of the derivational clitics are already in place,attached to their host, at the point the pronominal ones are introduced. Thedifference between principles (6.43a) and (6.43c) must then follow from someother aspect of the grammar of clitics. When we then ask where a differencebetween the behavior of disyllabic and monosyllabic clitics might lie, the an-swer is clear: the difference itself is phonological, so the obvious place to look isin the phonology. Let us then suppose that monosyllabic and disyllabic cliticsdiffer in the way they undergo Stray Adjunction. This difference could theninteract with the system of constraints determining their behavior as second-position elements so as to yield the observed placements.

The minimal foot in Tagalog is bimoraic, and disyllabic clitics are prosod-ically rich enough to be structured as feet. Monosyllabic pronominals, on theother hand, are monomoraic, and thus cannot constitute feet on their own:they are simply stray syllables. The difference is represented in (6.47). Whilediffering in their prosodic characterization, both are phonological clitics in thesense of Chapter 2, since they are not PWords, and Stray Adjunction must thusincorporate both types into an adjacent PWord.

(6.47) Disyllabic clitic: Ftσ σ

si ya

Monosyllabic clitic: σ

ko

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


Chapter 3, that a stray foot can (and must) be incorporated into the prosodicword to its left, while a stray syllable is adjoined to a foot on its left. The pro-nouns are true second-position clitics: that is, they are subject to high-rankingLeftMost and NonInitial constraints while not being allowed to violate avariety of Integrity constraints. In particular, lexical items are Integral in thissense, in addition to various uninterruptable phrasal types as noted above.

Stray Adjunction thus presents the two cases in (6.48). The differencebetween incorporation (in the Foot case, 6.48a) and adjunction (in the caseof stray syllables, 6.48b) in these rules follows from the well-known fact thatthe internal constituency of feet is typically rather narrowly constrained, whilephonological words are somewhat more loosely structured and may display anorganization that is flatter and internally more homogeneous in comparisonwith 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 asentence such as (6.49), with a monosyllabic clitic, a particle, and a disyllabicclitic, in that order.

(6.49) nakikita =ka =na =niyasees you (sg.) already heHe sees you (sg) now

In the construction of this phrase, we begin with a sentence-initial hostconsisting of a single word nakikita. Within the syntactic system, this comes tobe followed by a particle =na as a consequence of the application of “deriva-tional” cliticization. The structure in (6.50a) thus becomes that of (6.50b) atthe output of the syntax.

(6.50) a. PWordFt Ft

σ σ σ σ

na ki ki ta

LexWord

b. PWordFt

Ft Ftσ σ σ σ σ

na ki ki ta =na

LexWord


The pronominal clitics are then introduced into this structure. When thepronoun is disyllabic, it must be parsed as a foot, and thus has to follow theentire lexical word plus particle sequence. The latter is (as a whole) a PWord(perhaps derived by previous instances of Stray Adjunction incorporating par-ticle clitics into the PWord defined by the host). The clitic must follow thisentire sequence because (as shown in 6.48a above) the only place where a footcan undergo Stray Adjunction is at the right edge of a PWord.

Where the pronominal clitic is monosyllabic, on the other hand, it is onlya syllable, not a foot. Potentially it can (and thus, to avoid unnecessary viola-tions of LeftMost, must) be placed further to the left within the word. Theseclitics must be placed at the right edge of a foot, so that they can undergo StrayAdjunction as in (6.48b). They must also not violate higher-ranking Integrityconstraints, including the integrity of lexical words, but they must otherwisebe as close as possible to the left edge without being initial. As a result of theinteraction of these factors, they will lodge at the right edge of the leftmost footthat 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. Thesetwo types are both present in (6.51).

(6.51) PrWdFt

FtFt 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 in-nermost foot within the already complex foot [

f[f

ki ta]—=na]. Depending

on the details of stress contours in such a configuration, this analysis mightwell be replaced with one where the two adjoined syllables are restructured toestablish a new foot of their own, as in (6.52).

(6.52) [PWord

[Ft

na ki][Ft

ki ta ][Ft

ka na][Ft

ni ya]]

I do not attempt to develop the mechanical details of this refinement of theproposal here, but they do not appear to present conceptual problems.

The phonology of Stray Adjunction, then, taken together with the con-straints that define second position, can actually explain why monosyllabicclitics behave differently than disyllabic ones. It is important to note that this


result depends crucially on the constraint-based approach adopted here. Thatis, on the present picture, instead of specifying a unique position in hierarchi-cal (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 therelevant domain as possible, subject to the higher-ranking requirement thatthey not be absolutely initial. The definition of “second position” that resultsis a consequence of the interaction of a number of other effects (especially thedetails of the language’s phonology and of the relevant Integrity constraints).Since those factors play out differently in the cases of monsyllabic and of di-syllabic clitics, these elements appear in different positions, in ways that couldnot 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 Fig-ure 6.5 and the pronouns of Figure 6.4 alike—are to be characterized collec-tively as “second-position” elements in the sense that for each clitic cli there isa constraint NonInitial(cli, D) that outranks an otherwise highly ranked con-straint LeftMost(cli, D). We noted in section 6.1, however, that there mightbe languages in which the “NonInitial” effect results not from an independentconstraint, but rather from phonological requirements alone. Could this bethe case for the Tagalog clitics?

We might be able to examine a possible difference between the effects ofNonInitial(clj) and a potentially phonological effect if we could find a circum-stance where the domain D in which some clitic is subject to NonInitial(cli, D)phonologically includes preceding material that is part of a distinct syntacticdomain D′. In that case, we might expect that the clitic would appear ini-tially within the domain D; while if NonInitial(cli, D) in fact outranks Left-Most(cli, D), the clitic should continue to be post-initial within D.

It is always difficult to be absolutely certain that the phonological adja-cency necessary to establish this kind of situation is in fact met, but there is oneconstruction in Tagalog that appears to provide us with evidence of the type weseek. A range of inversion constructions (described by Schachter and Otanes1972: 485ff) involve the movement of a topical constituent to an initial position,presumably Spec(CP). When this happens, there may be a pause between thetopic phrase and the rest of the sentence (which would, I assume, block strayincorporation 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 (presumablyin C). It would appear that in this case the preceding phrase would providean appropriate prosodic host for an IP-initial clitic, if the latter element werenot subject to NonInitial(cli, IP), parallel to the facilitating role of a preceding


connected sentence in Warlpiri examples like (6.11). In such sentences in Taga-log, however, a pronominal clitic continues to appear in second position withinIP, as in (6.53) below.

(6.53) Ang sulat ay tinanggap =ko kahaponthe letter ay received 1sg yesterdayI received the letter yesterday (Schachter and Otanes 1972: 486)

This is at least weak evidence that the position of pronominal clitics is in factgoverned by high-ranking grammatical constraints of the type NonInitial(cli,IP). Furthermore, there is no evidence of the sort suggested for Warlpiri whichwould 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 pandanggodance Q they fandango

bukas ng gabi?]]tomorrow night

Will they dance a fandango tomorrow night?

b. [CP

bukas =ba ng gabi ’y [IP

sasayaw =silatomorrow 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” asexpected, both coming in second position. In sentence (6.54b), however, wherethe phrase bukas ng gabi “tomorrow night” appears in fronted topic position,the two clitics are separated. Sila appears as expected in second position withinthe nucleus of the clause, but ba appears within the topic phrase.

We cannot account for the facts in (6.54) by saying that particles, unlikepronouns, are not subject to NonInitial(cli,IP) and are only constrained to benon-initial within IP by phonological factors. If that were the case, we wouldexpect the particle to appear at the left edge of IP—which it cannot, as shownby the ungrammaticality of (6.55).

(6.55) *[CP

bukas ng gabi ’y [IP

=ba sasayaw =silatomorrow night ay Q dance they

ng pandanggo?]]fandango

(Will they dance a fandango tomorrow night?)


The fact that the particle appears embedded within the phrase bukas nggabi (following its first word) in (6.54b) shows that it must in fact be locatedwith 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 willcoincide, and so the particles and the pronouns will form a single sequencelocated in the same “second position.” Where there is a preposed topic phrase,however, the difference between the two domains is revealed and the cliticsbecome separated.

One group of particles behaves like ba in (6.54) above, coming inside thetopic phrase. Other particles, however, behave like the pronominal clitics inalways appearing within IP, such as lamang in the sentences in (6.56).

(6.56) a. [CP

[IP

sasayaw =lamang =sila ng pandanggodance just they fandango

bukas ng gabi ]]tomorrow night

They’ll just dance a fandango tomorrow nightb. [

CPbukas ng gabi ay [

IPsasayaw =lamang =sila

tomorrow night ay dance just they

ng pandanggo]]fandango

They’ll just dance a fandango tomorrow nightc. *[

CPbukas =lamang ng gabi ay [

IPsasayaw =sila

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 ofwhich is subject to high-ranking NonInitial(cli,CP) while the other is subjectto NonInitial(cli,IP). This difference is a property of the particles as grammat-ical elements, and does not follow simply as a consequence of their phonolog-ical form. According to Schachter and Otanes (1972: 429ff), the particles thatbehave like ba in (6.54) include both monosyllabic elements (ba, man) anddisyllabic ones (kasi, kaya). Similarly, the set that are excluded from sentence-initial topic phrases includes not only disyllabic particles like lamang, munabut also monosyllables (na, pa).

A third subset, including both monosyllabic and disyllabic items, can ap-pear in either position, depending (apparently) on considerations of semantic


scope.16 If this speculation is indeed correct, it points the way to the essence ofthe differences we find here. Assuming that particles are introduced within adomain that constitutes their scope, as we might wish if the operation Mergeis to be semantically compositional, we could say that the relevant LeftMostand NonInitial constraints are relativized to the same domain, whichever itis in a given sentence. The domain of introduction of pronominal elements ispresumably uniformly IP, so their placement in second position within IP isconsistent with this account.

It seems, then, that there is no evidence supporting the notion that thenon-initial position of any of the Tagalog clitics follows from phonologicalconsiderations alone. Instead, each clitic’s location is governed by the interac-tion of appropriate LeftMost and NonInitial constraints, relative to the gram-matical domain (either IP or CP, depending on the clitic) within which it isintroduced. Legendre’s (2000a) suggestion that Tagalog clitics are placed insecond position within a phonological constituent (specifically, a phonologicalphrase) rather than a grammatical one is thus not confirmed.

Tagalog thus presents a particularly interesting and intricate illustrationof 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 do-main (IP or CP);

iii. Hierarchical ranking of alignment constraints for specificclitics;

iv. Integrity constraints, which characterize uninterruptible sub-parts of a domain and thus define what counts as occupying“first” position;

b. An architecture of grammar in which “derivational” clitics resultfrom an operation of Merge, while “inflectional” ones realize thefunctional categorial morphology of phrases and clauses; and

c. Phonological effects, in particular the details of Stray Adjunctionphenomena.

This complex picture has a natural place within a theory that draws on thebasic ideas of Optimality Theory, such as the research program developed inthe present work.

16 See Condoravdi and Kiparsky (1998) for a discussion of the interaction of scope with align-ment 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 surelyWackernagel (1892). Most of this paper is dedicated to the special clitics ofthe early Indo-European languages, and on the basis of the preceding chap-ters, we now have a theory of the grammatical mechanisms behind this varietyof special clitic positioning (among others). But Wackernagel’s paper raisesanother point, one for which it is rather less often cited. This is the questionof whether the notion of “second position” wherever it arises in grammar hasa uniform basis.

In particular, reference to “second position” comes up in an area thatseems quite remote from the analysis of special clitics: the placement of the fi-nite verb in second position in German (and most other Germanic languages,apart from English). Wackernagel claimed explicitly that “verb second” con-structions and second-position clitics were connected, and the goal of thepresent chapter is to explore the possibility that they do indeed have acommon basis.

Wackernagel’s own argument was simple and straightforward. Recall that,for him, “clitics” were precisely and by definition “unstressed words.” Heshowed that these consistently appeared after the first word of the sentencein the earliest known Indo-European texts, a conclusion further confirmed bymaterial from the Anatolian languages that later became available. It seems,then, that unstressed words in proto-Indo-European and its immediate de-scendants 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 sec-ondary development (as in Greek). From this it should follow that, as unac-cented words, finite verbs should behave as clitics—and thus that they shouldalso be attracted to second position. As a result, he suggested that the locationof the verb in second position in modern German main clauses might well bea reflex of its earlier link to the broader class of clitics.

Elegant as this explanation would be if it could be sustained, it has seri-ous problems. For one thing, the verb-second construction has been argued


(Kiparsky 1995) to be a comparatively recent innovation in Germanic, andnot a property inherited from earlier Indo-European origins.1 And what-ever the historical facts may be, this explanation does not really bear on thesynchronic state of affairs in modern German, since (a) the language doesnot have second-position clitics, and (b) its finite verbs are not unstressed.Therefore, Wackernagel’s proposed synthesis cannot be literally correct as ananalysis 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 bothof these areas of grammar? Is there in fact a common basis that would sug-gest that they should be unified in theoretical terms? Previous chapters havedeveloped a theory of special clitics, within which “second position” is recon-structed in a specific way. If clitics are introduced into the phonological formsof phrases by a generalization of essentially morphological mechanisms, andfind their position within the surface form of sentences through a system ofhierarchically ranked violable constraints, as proposed here, is there perhapssome relation between second-position clitics and verb second, as Wackernagelproposed?

I begin by considering how the constraint-based account of special cliticssuggested in this book relates to the standard analysis of verb second as involv-ing movement, as instantiated in modern standard German. I then go on toextend the analysis to other Indo-European languages in which verb second isfound, including Icelandic, Kashmiri, Breton, and Surmiran Rumantsch beforedrawing some general conclusions.

7.1 Verb Movement and Second Position

A major theme in recent syntactic discussion is the notion that movement isforced by “morphological” considerations. Theories invoking “morphologi-cally based movement” are not generally very closely related to any theory ofmorphology, however. The image is that of a scavenger hunt, where word for-mation has already done its work, the grammar has scattered inflectionally rel-evant features here and there within sentence structure, and it is a constituent’stask 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 morecentrally morphological concerns than the need to check features.

1 It should be noted, though, that others (Eythorsson 1996) have argued that verb second isactually somewhat older than Kiparsky suggests.

Verb Second as Alignment 179

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

(7.1) Verb Second: The verb which indicates the Tense, Mood and Agre-ement properties of a clause appears immediately after the firstconstituent of the clause.

Another way to put this is: the formal markers of a clause’s relational prop-erties appear as morphology on a verb which appears immediately after thethe first unit of the clause’s syntactic analysis. That is not very different fromthe descriptive regularity governing second-position clitics: these, too, appearimmediately after the first (non-“permeable”) consituent of the clause. Andboth finite main verbs and (many) sentential clitics realize a clause’s inflec-tional features.

From this perspective, a verb-second language such as Icelandic differsfrom a second-position clitic language like Tagalog in that the clause’s gram-matical 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 ismade by Legendre (2001), who argues along lines very similar to those devel-oped below that a single set of constraints on the location of the functional con-tent of the finite clause governs the location both of the verb (in verb-secondconstructions) and of clitics such as those expressing tense and agreement.

Taking the analogy seriously, how can we relate the properties of verb-second structures to the regularity of second-position clitic placement? In thepresent framework, this is straightforward. We describe second-position cliticsas occurring as far to the left within their domain as possible without being atthe absolute left edge. We can characterize the position of the verb in a verb-second structure in parallel fashion by saying that the locus of realization ofTense, 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 whatdoes 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 featuresof Tense, Aspect, Agreement, etc. When those features are expressed on the verb, the result isverb second; when they are expressed by special clitics, these appear in second position. Thedifference between that formulation and the present one is not essential to the immediate issueof establishing that such a parallelism exists.


are not. When constraints conflict, their ranking determines which of them ismore important to observe. We thus want the ranked constraints in (7.2) toserve as a mechanism to choose the correct structure from among alternativesprovided by the syntax.

In the syntactic literature of recent years, a derivation of verb-secondclauses as sketched in the Phrase Marker of Figure 7.1 has come to be widelyaccepted. Let us make the common assumption that German clauses are ba-sically 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 theVP into Spec(IP)). Movement of the verb to I in Figure 7.1 simply representsthe fact that this verb is the one that bears finite inflection—such movementis not needed on a view where functional content propagates as features ratherthan 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 aspossible.

Movement of the AdvP gestern ‘yesterday’ to initial (Spec(CP)) position inFigure 7.1 presumably represents an instance of topicalization, and is motivatedby the information structure of the utterance. But the crucial question in thisanalysis has always been the reason for movement of the inflected Verb fromI to C. Does such movement take place because of some property of the Cposition itself ? Proposals have been made to force this movement by positing

CP(=S)

Spec C′

C IP(=S)

Spec I′

I′

VP I

I VP

AdvP V NP AdvP NP V V

Gestern habe ich t eine Zeitung gelesen t tyesterday have I a newspaper read

‘Yesterday I read a newspaper’

Figure 7.1. Movement in the German verb-second construction


some feature-checking relation between the positions Spec(CP) and the verb inC, or by requiring C to be lexically filled, or in other ways; but these all amountto camouflaged language-particular stipulations of the requirement “Move I toC.” The movement remains, accordingly, without independent motivation.

The proposal here is that this movement is not motivated by the need tocheck some abstract feature or the like, but rather by the fact of second positionitself. Movement to this position occurs when (a) it is syntactically possible,violating no constraints of the syntactic computational system; and (b) thestructures that result violate relevant Alignment constraints such as (7.2) to alesser degree than structures in which such movement has not taken place, sothat the finite verb would otherwise be farther from the left edge of the sentencethan in the preferred structure.

The theory of verb second espoused here does not per se deny that verbsecond involves movement of the Verb into a C position, though it does notrequire that specific landing site for movement, either. It simply proposes thatthe reason for any such movement is to get the verb to a position as far to theleft 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 ofthe proposal concerns the motivation for the movement, and not its mechan-ics. Its essence is the notion that “second position” is the real generalizationabout the position of the finite verb and not an epiphenomenal consequenceof movement carried out for some other purpose.

The similarity of this analysis of verb second to the treatment of second-position clitics developed in Chapter 6 is apparent. In the case of specialclitics in second position, I argued that what unifies them is not occupancyof a consistent hierarchical position in phrase structure, as syntactic move-ment accounts predict (or assume), but rather the fact that they are in secondposition—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-positionclitics result from essentially similar mechanisms, however, another questionarises. If verb second results from a “morphological” imperative, akin to thatgoverning clitic placement, and not from a purely syntactic requirement, whydo 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, “verbsecond” treats as its anchor the first syntactic daughter within the domain inwhich the verb appears, rather than the first prosodic constituent of some type(such as a PWord). There are nowhere near as many languages with productiveverb second to consider as those that instantiate second-position clitics, butit is still striking that a position defined in prosodic terms is apparently notavailable for verbs while it is for clitics.


The answer to this apparent conundrum is that the syntax cannot ingeneral access a position defined in this way. In this respect, the mechanismsthat implement the placement of special clitics and of the verb are fundamen-tally different, although the two may interact with similar constraint systems.Special clitics, on the one hand, are placed by “affixation” rules that modify thephonological shape of the form. It is thus possible for these rules to introducematerial anywhere in the structure, subject only to the Integrity constraintsthat prevent insertion into units of some particular (prosodically or syntacti-cally defined) sort. Verb second, on the other hand, is the result of syntacticmovement, and the only structures that the syntax provides for comparativeevaluation by the constraint system are ones that instantiate well-formed syn-tactic operations. Since these depend exclusively on the syntactic analysis ofa structure, and not on its prosody, there is no way for a prosodically definedposition to play a role directly in syntactic movement.

Recall the way in which a constraint-based system works. On such a the-ory, the computational system involves (1) a subsystem (usually referred to as“gen” in the OT literature) which produces a set of formally possible struc-tures; and (2) a set of constraints on output form that evaluate these candidatestructures with respect to a hierarchically organized set of requirements so asto choose the optimal one.

It seems reasonable to assume that gen incorporates universal notions ofsyntactic well-formedness, in that only structures conforming to fundamentalprinciples such as those of X-theory, and involving syntactically permissibledisplacement relations, are presented for comparative evaluation by the con-straint 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 sentence-initial PWord, where that may be internal to a larger containing phrase, willthen be disallowed by the general nature of movement. Candidate structuresof this sort are not found in the output of gen, and thus no language preferringthem could exist.

On the other hand, no such limitation applies to the candidates gen pro-vides for evaluation in the case of special clitics. Since these are (on the presenttheory) instances of phrasal morphology, the input to the system that placesthem includes both a prosodic and a syntactic analysis, just as word-levelmorphology operates on the basis of both phonological and morphosyntacticrepresentations of the word. As a result, clitic placement can “see” a positiondefined prosodically, while syntactic movement cannot.


Now notice that this provides us with an account of the “asymmetric” na-ture of verb second in German. As is well known, verb second does not takeplace in German subordinate clauses just in case they have an overt comple-mentizer, as illustrated in (7.3).

(7.3) a. Hans behauptet, daß er davon nichts gewußt habeHans maintained that he thereof nothing known haveHans maintained that he knew nothing about it

b. Hans behauptet, er habe nichts davon gewußtHans maintained he have nothing thereof knownHans maintained he knew nothing about it

Suppose that the Verb Movement giving rise to verb second in mainclauses is in fact a movement from I to an otherwise empty C, and that thisC position is in fact the only possible target of syntactic movement of the verbin German. It then follows that we should fail to find verb second exactly wherethe relevant movement is blocked by general conditions—and of course whenthe C position is filled by an overt complementizer like German daß, this willbe the case. That is, where C is already filled, gen does not provide a candi-date structure in which the finite verb is displaced from clause-final I position,because there is no well-formed movement available that would reduce theviolation of LeftMost(Vfin,S). This is essentially the same explanation that isstandardly offered for the asymmetry within other syntactic accounts, and itextends quite naturally to the present view.

The proposed analysis for verb-second constructions is thus based on thecomparative evaluation of well-formed outputs of the underlying syntacticcomputational system with respect to a constraint system including (7.2). Inthe remainder of this chapter, I will examine the applicability of this accountto verb-second phenomena in a range of other languages. These are all drawnfrom (distinct subfamilies within) Indo-European, but they appear to be in-dependent of one another, and not simply diverse modern reflexes of a singleoriginal type.

7.2 Icelandic

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


second symmetrically, in both main and subordinate clauses. Consider thesentences in (7.4).

(7.4) a. Jón harmar að þessa bók skuli ég hafa lesiðJohn regrets that this book shall I have readJohn regrets that I have read this book

b. Ég veit að það hefur enginn lesið bókinaI know that there has no one read the bookI know that no one has read the book

The finite verbs are italicized in these examples, while the initial elementswhich they follow within the sentential domain are underlined. These in-stances 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 stúdentarI thought that kissed had her many studentsI thought that many students had kissed her

b. *Ég hélt að hana kysst hefðu margir stúdentarI thought that her kissed had many studentsI thought that many students had kissed her

Sentence (7.5a) illustrates “Stylistic Fronting,” a construction with well-established and distinctive characteristics (Maling 1980; Jónsson 1991) quiteseparate from those of Topicalization. It involves the movement to a positionbefore the inflected verb of a single word (participle, negative ekki, or certainadverbs). Stylistic Fronting displays two notable characteristics. First of theseis the fact that a single word, rather than a phrase is moved. Second, Stylis-tic Fronting is constrained by a “Subject gap condition”: it is possible exactlywhen the subject position is not overtly occupied. This is the case in imper-sonal sentences, which may receive an expletive subject in their surface form ifnothing else intervenes, or as a consequence of extraction, postposing, or anyother operation rendering the subject position unfilled. These various possi-bilities, and the consequences of ignoring the Subject gap condition in similarstructures, are illustrated in (7.6).

(7.6) a. i. Honum mætti standa á sama, hvað sagt værito him would be the same what said would beum hannabout himIt would be all the same to him, what was said about him


ii. *Honum mætti standa á sama, hvað sagt hefði Hjördísto him would be the same what said had Hjördísum hannabout himIt 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 komahe is the only that not is likely to comeHe is the only one that is not likely to come

ii. *Hún spurði, hvort líklegur væri hann til að komashe asked whether likely would be he to comeShe asked whether he would be likely to come

c. i. Þetta er bærinn, þar sem fæddir eru margir frægirthis is the town where born are many famousÍslendingarIcelandersThis is the town where many famous Icelanders were born

ii. *Þetta er bréfið, sem ekki skrifaði Helgithis is the letter that not wrote HelgiThis is the letter that Helgi didn’t write

Let us assume, as seems uncontroversial, that Icelandic has basic SVOorder, and that complete clauses have the structure of (7.7).

(7.7) [(XP) (X) [(Subject) [V′

(Verb) . . .]]]

The “X” and the “XP” in (7.7) might be “C” and “Spec(CP),” but nothinghinges on that categorial analysis. All we require is that two clause-initial pos-itions are available, one of which can contain a simple head and the other ofwhich can contain a phrasal category.

It might be objected that in the sentences of (7.4), the subordinate clausesare already introduced by a complementizer (að). As a result, it is impossiblethat they should have the structure of (7.7), since the “X” position there is al-ready filled by að. Precisely this aspect of verb second in Icelandic has led someauthors to propose that in languages of the “symmetric V-2” type, the categoryCP is recursive, making available a new instance of “X” in (7.7). It is not ne-cessary to make that move however. Instead, we can regard að as a marker ofsubordinate clauses, rather than a complementizer. Specifically, we can treatað as an initial (special) clitic within subordinate clauses, introduced at the leftedge of the structure in (7.7) when this appears in embedded positions, and notoccupying the structural position “X” within that structure.


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 mainverb, where they appear as its inflection. Rank NonInitial(Vfin,S)above LeftMost(Vfin,S) (and both high), thus preferring a structurein which the finite Verb appears immediately following the first overtsyntactic daughter of the clause.

There are several ways to satisfy the conditions in (7.8). First, it might be thatnothing moves from the core of the clause into the available XP or X posi-tions. In this case everything remains in its “natural” position, and the verb issecond by virtue of following exactly the subject DP. Second, it might be thecase that some phrase moves into the available XP position (presumably as atopic or for some similar information–structurally motivated reason), and theverb itself moves to the available X position. In this case, the verb is in sec-ond position by virtue of following exactly the topic phrase. Note that thesetwo situations are distinct. Sentences with basic SVO constituent order mayrepresent prominently “topical” subjects, but on the other hand, they may beinformationally neutral. When something other than the subject appears insentence-initial position, however, it must be interpreted as topical.

As a third possibility, it may be that no phrase moves to the topic XP pos-ition, but the subject position is not overtly filled by lexical material. Thismight be the case in impersonal sentences, or where the verb comes to precedethe subject through an operation of postposing (if this a possibility, as assumedin much of the traditional, descriptively oriented literature). In these cases, anexpletive is inserted in the pre-verbal X position, as illustrated in (7.9).

(7.9) Það eru margir frægir Íslendingar fæddir í þessum bæexpl are many famous Icelanders born in this townThere were many famous Icelanders born in this town

Finally, in the case where no phrase moves to topic XP, and there is alsoa gap in the subject position, Stylistic Fronting can move a participle or otherelement 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 itappears in a consistent configurational position, but rather the fact that ineach instance, the verb is second within its clause. This is exactly the claimof an analysis on which verb second results from a mechanism similar to thatgoverning second-position clitics: compare these facts with the lack of a con-sistent structural position occupied by second-position clitics in a variety oflanguages cited by authors such as Boškovic (1995) for Serbo-Croatian, and thebroader discussion of this matter in Chapter 5.


This account also provides an immediate way to derive the Subject gapcondition, 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 nolonger 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, sinceif it did so (even disregarding the problem posed by a head moving to a phrasalposition), it would then be initial rather than second. Stylistic Fronting canonly take place where it leads to a preferred structure, and thus is only possiblewhen the subject position is unfilled.

The analysis of verb second as parallel to the positioning of second-position clitics thus seems motivated for Icelandic, a core instance of a lan-guage displaying this property. I now move on to other languages which havebeen less commonly cited in this regard.

7.3 Kashmiri

The literature on the verb-second construction has concentrated largely onGermanic (for the very good reason that nearly all of the languages of thisfamily display it in some form). As I will discuss in subsequent sections ofthis chapter, verb second is also found in early stages of Romance and in atleast one modern Romance language (Rumantsch), and related word-order ef-fects are also found in Breton within Celtic. At least one language in the Indicfamily also shows a form of verb second that we can compare with the Ger-manic phenomena, however: Kashmiri.3

Following Hock (1982, 1991), Bhatt (1999) suggests a scenario for the devel-opment 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 de-veloped a large number of verbs that are normally conjugated with the aid ofan auxiliary element. Middle Indic acch- ‘be’ underwent reduction in Kashmirito a clitic form with base ch- , resulting in its being placed in second positionalong with other clitics. As a later development, the positioning of non-cliticfinite verbs was assimilated to this, resulting in verb second. Much of this

3 Hendriksen (1986, 1990) reports that in at least two “Himachali” languages, Kot.gar.hi andKoci, he finds word-order patterns quite similar to those of Kashmiri. Verb second is reportedas 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 sub-ordinate clauses. It is quite difficult to determine the details of these patterns on the basis ofHendriksen’s descriptive materials, however, and I will not discuss these languages further. Theyare spoken in a region of north India near but not adjacent to that in which Kashmiri is the dom-inant language. Like Kashmiri, they are sometimes said to be part of a “Dardic” subfamily ofIndo-Iranian, though modern scholarship treats all of these languages as Indo-Aryan.


account remains speculative, but the resemblance to Wackernagel’s (1892) ex-planation of verb second is not hard to see.

The basic verb-second construction in Kashmiri main clauses is illustratedin (7.10).4

(7.10) a. me dits mohn-as yi k@mi:z ba:gas manzI.erg gave Mohan-dat this shirt garden inI gave Mohan this shirt in the garden

b. yi k@mi:z dits me mohn-as ba:gas manzthis shirt gave I.erg Mohan-dat garden inI gave Mohan this shirt in the garden

c. mohn-as dits me yi k@mi:z ba:gas manzMohan-dat gave I.erg this shirt garden inI gave Mohan this shirt in the garden

d. ba:gas manz dits me monh-as yi k@mi:zgarden in gave I.erg Mohan-dat this shirtI gave Mohan this shirt in the garden

As can be seen here, the finite verb immediately follows the first constituentof the main clause. Surprisingly, this is also true for imperatives and yes/noquestions, as shown in (7.11).

(7.11) a. ko:r-i di-t1 kita:bgirl-dat give-pol.imper book(Please) give the book to the girl!

b. (k’a:) ts1 gatsch-kh-a: paga:h gar1?(qpart) you go-2sg-q tomorrow homeWill you go home tomorrow?

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

(7.12) a. ma:st.ar-an ch-u l@d. k-as kita:b div-a:nteacher aux-m.sg boy-dat book give-nonperfThe teacher gives a book to the boy

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


b. tem-sund ch-u asi maka:n bad. 1 pasandhe-of aux-m.sg us.dat house very likeWe like his house very much

Essentially the same facts obtain in subordinate clauses introduced by thecomplementizer ki ‘that’ as in (7.13).

(7.13) a. me ch-u pata: ki yi k@mi:z dits ra:m-anI aux-m.sg know comp this shirt gave Ram-ergmohn-asMohan-datI know that Ram gave this shirt to Mohan

b. me ch-u ba:sa:n ki su ch-u diliI aux-m.sg believe comp he aux-m.sg Delhi.locro:za:nlive.nonperfI believe that he lives in Delhi

On the basis of these facts, we could suggest an analysis (similar to that ofIcelandic in section 7.2) along the following lines. From the position of the lex-ical verb in sentences like (7.12), it appears that Kashmiri has a fundamentallySOV organization. Let us assume that this basic clause structure is embeddedin a layer of functional structure similar to that proposed for Icelandic, withan initial Specifier position that can be filled by a phrasal unit, and an imme-diately following head position. We can then say that the initial topic phrasesin (7.10) occupy the Specifier position in this layer of structure, and the finiteverb is displaced from its underlying final position to occupy the correspond-ing head. The requirement that the initial topic position be filled when thefinite verb is fronted is apparently so important that “[w]hen nothing in a fi-nite matrix clause is topicalized, not even the subject which generally acts as thediscourse topic by default, dummy yi is inserted” (Bhatt 1999: 97). Unfortu-nately, neither Bhatt nor Wali and Koul (who suggest the same thing) providesclear examples, although presentational sentences such as (7.14) may representthis phenomenon.

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


Since we find the same word order in the embedded clauses of (7.12), it cannotbe the case that the functional layer in which topics and the finite verb appear isthat 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 introducedinitially in complement clauses, like að in Icelandic.

The observed displacement of the finite verb can then be attributed tothe operation of high-ranking LeftMost(Vfin) within the including functionaldomain, with the obligatory presence of an initial topic phrase due to evenhigher ranked NonInitial(Vfin) within the same domain. With the exceptionthat its basic order within the VP is SOV and not SVO, Kashmiri then looksmuch like Icelandic with regard to the basis for its verb-second construction.

A complication to this picture appears to be presented by the structure ofwh-questions. In these sentences, the wh-phrase appears immediately beforethe finite verb as in (7.15).

(7.15) a. ramesh-an k’a: d’utnay tse ra:thRamesh-erg what gave you.dat yesterdayWhat did Ramesh give you yesterday?

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

c. ? k’a: d’utnay ramesh-an tse ra:thwhat gave Ramesh-erg you.dat yesterdayWhat did Ramesh give you yesterday?

If (7.15c) were the standard form of wh-questions, we could account for thissimply by saying that wh-phrases move to the same position as topicalizedelements. Unfortunately, this cannot be the case: while this order is marginallypossible, the order of the other sentences in (7.15), with a topic phrase preced-ing 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 andKoul (1997) and Bhatt (1999).

(7.16) yi kus o:s bar-asexpl who was door-atWho was at the door?

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


immediately precede the finite verb. Furthermore, the presence of a pre-verbalwh-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 structuresuch as Figure 7.2.

In multiple-wh questions, more than one wh-phrase can adjoin to the verbin this fashion. Alternatively, one or more (but not all) of the wh-phrases canremain in situ, as in (7.17b).

(7.17) a. rameshan k@m’is k’a: d’utRamesh.erg who.dat what gaveWhat did Ramesh give to whom?

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

Assuming that the complex V created by adjoining one or more wh-phrasesto the finite verb is the constituent whose position is constrained by NonIni-tial(Vfin), the data from wh-questions fall together with those from declarativesunder this analysis.

There is still one further set of facts to consider, however. In Kashmirirelative clauses, the finite verb appears in final rather than second position, asin the examples of (7.18).

XP

X′

V VP/IP

DP DP

V

DP Adv PP DP V

ra:j- an k’a: kor [e] ra:th pan1ni gari [e] [e]Raj-erg what did yesterday his-at houseWhat did Raj do at his home yesterday?

Figure 7.2. Movement in Kashmiri wh-Questions


(7.18) a. su l@d. k yus ra:th yath kamras manz bat1dem boy rel yesterday dem room in foodkh’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. . .

c. yet’i swa bat1 ran-a:n ch-a. . .where she food cook-nonperf aux-fsg. . .where she cooks food. . .

Relative clauses can occur as modifiers immediately following their asso-ciated 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:vrel servant you yesterday met cor servant leftvun’just nowThe servant who met you yesterday has left just now

Apparently, relative clauses (of both sorts) are based on a full clause struc-ture. Instead of being embedded within the functional structure that supportstopicalization and verb second, however, as in the other Kashmiri examplesabove, relative clauses are apparently embedded within a distinct structure inwhich 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 byan abstract operator corresponding to the anaphoric relation between the rel-ative phrase and a gap in the relative clause; in any event it is apparently nota position to which the verb can move in the way it moves to the head of thetopicalization structure in other sentence types.

We can tell from these facts that the functional structure in non-relativeconstructions is not an instance of IP, such that the verb must be displaced toits head in order to inherit finite inflection. Relative clauses show us that theposition in which finite inflection is assigned is clause-final, and thus that thereis no structural difference between VP and IP.

Since the head position in a relative construction is not one to whichthe verb can move, and since there is no alternative position to which dis-placement of the verb would be well-formed in the syntax, the only candidate


structures which can be evaluated with respect to the alignment constraintsLeftMost(Vfin) and NonInitial(Vfin) are those in which the finite verb is finalwithin 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 subordi-nate clauses introduced by the complementizer daß, which also pre-empts theonly site to which syntactically well-formed displacement of the verb fromfinal position could occur in such clauses. German and Kashmiri are thus both“asymmetric” verb-second languages, in the sense that verb second obtains insome structures and verb final in others, although the dimension that definesthe asymmetry is distinct in the two languages. In both cases, the motivationfor the movement that takes place to yield verb second where possible can beseen as essentially the same set of Alignment constraints as those that are oper-ative in second-position clitic constructions in other languages. As mentionedat the beginning of this section, the connection with second-position cliticsmay have historical roots in Kashmiri, though there is no longer a synchroniclink between the two in the grammar either of that language or of those in theGermanic 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 otherCeltic languages; while on the other hand, most sentences in fact look like whatwe expect to find in a verb-second language. The analysis that I will proposeextends the account of section 7.1 to Breton. The results are quite similar tothose of Hannahs and Tallerman (2000), and also to Legendre (2001), a papersharing most of the assumptions and goals of the present chapter.

A number of years ago, I made some proposals about Breton clause struc-ture, beginning with those in Anderson and Chung (1977). A later revisionof that analysis appeared as Anderson (1981), and served as the basis of someremarks on Breton Agreement in Anderson (1982). A substantial literature onBreton syntax has developed in the meantime, including Borsley (1990); Bors-ley, 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) criti-cal of one or another proposal in my earlier analyses. In some respects I thinkthese criticisms are well taken, though in others I am less sure of that. In anyevent, only a limited range of those issues will be relevant here.


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-finiteverbal form, etc. In all but one of the sentences of (7.20) below, the finite formin question is the perfect auxiliary, disregarding some specific issues raised bythe structure of this element (to be mentioned below).

(7.20) a. Yann en deus debret e voued er weturYann he-has eaten his food in the carYann has eaten his food in the car

b. E voued en deus debret Yann er weturhis food he-has eaten Yann in the carYann has eaten his food in the car

c. Skuizh int da c’hortozTired they-are to waitThey’re tired of waiting

d. Debret en deus Yann e voued er wetureaten he-has Yann his food in the carYann has eaten his food in the car

Remarkably for a “VSO” language, on the other hand, the verb itself cannotbe first, as illustrated by the ungrammaticality of (7.21).

(7.21) *a/e d/t/zebr Yann e voued er weturprt eats Yann his food in the carYann eats his food in the car

The element in initial position typically has some particular informationvalue, such as that of a topic. When nothing in the sentence is appropriatefor isolation in this way, there are two alternative constructions. One of theseinvolves fronting the lexical verb, in its verbal noun form, with an inflectedform of ober ‘do’ occupying the position of the verb. The other involves anempty expletive bezañ, homophonous with the verb ‘to be’ but syntacticallyinert in this usage. These two possibilities are illustrated in (7.22).

(7.22) a. Debriñ a raio Yannig krampouezh hiziveat (VN) prt will-do Yannig crêpes todayYannig’ll eat crêpes today

b. Bez’ e vo pesketaer Yannig a-hed e vuhez[BE] prt-he-will-be fisherman Yannig all his lifeYannig will be a fisherman all his life


c. Bez’ e tebr Mona he boued er gegin[BE] prt-eats Mona her food in-the kitchenMona eats her food in the kitchen

d. Bez’ eo bet labourer-douar Yannig daou vloaz[BE] is been farm worker Yannig two yearsYannig’s been a farm worker for two years

Breton, like English, has three verbs do: a lexical main verb (as in do thedishes), the pro-form of do so, and an empty auxiliary (as in I don’t like thatas much as you do). Compound forms of one or another of these do verbs canalso appear (with their finite part in immediate post-initial position) with apreposed lexical verb, as in (7.23).

(7.23) Debriñ en deus graet Yann e voued er wetureat he-has done Yann his food in the carYann has eaten his food in the car

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

(7.24) a. Debriñ krampouezh a raio Yannig hiziveat crêpes prt-will-eat Yannig todayYannig will eat crêpes today

b. Debriñ e voued en deus graet Yann er wetureat his food he-has done Yann in the carYann has eaten his food in the car

Forms in which the object accompanies the verb appear to be instances oftopicalization of a VP, and Anderson and Chung (1977) suggested that the con-struction 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 Ander-son (1981), however, I argued that what is actually involved is a nominalizationwith or without its complement. On this account, the verbal form is a “verbalnoun” and the complement is an accompanying genitive. Arguments purport-ing to demonstrate the verbal nature of these phrases have concentrated on thefacts of Welsh, and although the corresponding structures in that related lan-guage may well be headed by verbs, the arguments of Anderson (1981) to theeffect that they are nominals have not been refuted (in my opinion), and I willassume that structure here.


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

(7.25) a. Kollet am eus ma hentlost I-have my roadI have lost my way

b. *kollet ma hent am euslost my road I-haveI have lost my way

c. *ma hent kollet am eusmy road lost I-haveI have lost my way

d. Ma hent am eus kolletmy road I-have lostI have lost my way

Although fronted participles may not take an object complement in thatposition, they may be accompanied by certain (common, short) adverbs, asin (7.26).

(7.26) Debret mat en deus d’ he goaneaten well he-has for his supperHe has eaten well for supper

There is one further possibility: the first part (ne) of the Breton two-part negation can also fill this pre-verbal position. In negative sentences suchas (7.27), nothing else can be topicalized.

(7.27) a. N’ en deus ket lennet Tom al levrneg he-has neg read Tom the bookTom has not read the book

b. *ne lennet en deus ket Tom al levrneg read he has neg Tom the bookTom has not read the book

c. *lennet n’ en deus ket Tom al levrread neg he has neg Tom the bookTom has not read the book

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


Breton sentences, then, generally have some element preceding the finiteverb. 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 in-flected finite verb in Breton appears at the left edge of some constitutent—callthis IP—and is preceded by exactly one element, which may be either a phraseor a single word (including the first part of negation), but not both at the sametime. The structural possibilities seem to be those offered in (7.28).

(7.28) [CP

(XP) (X) [IP

Vfin . . .]]

The XP here can be filled with any maximal projection, including the ver-bal noun together with an optional complement. The X can be filled by ahead such as a participle or an adjective, or by the first element of the nega-tion. What matters is that the finite verb appears at the very left edge of IP, butsecond within CP. Superficially at least, this is very similar to the account ofIcelandic offered in section 7.2, or other verb-second languages.

How are we to describe these facts? Hannahs and Tallerman (2000) showthat previous purely syntactic accounts involve either unmotivated differencesin 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 ofthe various sentence types, on standard assumptions, involve (at least) move-ment of the verb. But we need to ask why the verb moves, as well as what elsemoves (or fails to move) and why.

There are really two things to account for. First, we need to know howthe 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 displacementinvolved. Second, we need to know what principles govern (and limit) theappearance of pre-verbal material. This whole complex of facts and issuesis typical for verb-second languages, although Breton shows some uniquefeatures of its own.

Breton As a Verb-Initial Language

Breton, like other Celtic languages, is typically said to be “basically” a VSOlanguage, a claim that the reader may find puzzling on the basis of the evidencesupplied thus far. The evidence that is usually cited for this is the order whichwe 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 gozhI say to-you prt was the horse-dem too oldI tell you that horse is too old


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 wedifferentiate the order in main clauses from that in embedded clauses? Assum-ing the structure in (7.28), what we need to do is require that within CP, exactlyone of the positions C and SpecCP is filled. This effect could be obtained bysaying that the finite verb in Breton should be as far to the left as possiblewithin IP (the “Verb-initial” constraint), and also within CP, but should notbe absolutely initial within the CP. The relevant constraints and their rankingwould 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 theVfin within CP than what is necessary to ensure that it is not initial withinthat constituent, and the high ranking of LeftMost(Vfin, IP) ensures that theSpec(IP) position is not filled.

The position to which the verb moves is structurally uniform---I, initialwithin IP. The generalization about verb second is not a structurally uniformone, however: it requires that the finite verb not be initial within CP, but thisis not due to a requirement that some particular pre-verbal position must befilled. Rather, exactly one pre-verbal position within CP must be filled, with-out 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 discus-sion, though. If embedded clauses are CPs, we must ensure that verbsecond is not enforced there. Legendre (2001) suggests that the relevant differ-ence does not depend on main versus subordinate status but rather on initialversus non-initial position within the entire sentence. The domain of Non-Initial might then be a prosodic one, consistent with Legendre’s proposal thatNonInitial (or “Tobler-Musaffia”) effects are always relative to a prosodic ratherthan a grammatical domain.

At a minimum, the category involved cannot be one including the entiresentence, since the second of two coordinated clauses often shows verb secondas in (7.31).

(7.31) Me a breno ar sae, hogen c’hwi a wisko anezhiI prt will buy the dress but you prt will wear itI’ll buy the dress, but you’ll wear it

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


An alternative account of the (limited) asymmetry of verb-second effectsin Breton is to say that complements typically are simply IPs, not CPs, and thatthere is thus no C or Spec(CP) position available to be filled. Complementswith overt complementizers (apart from the pre-verbal particles, which arethemselves 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 (orpermit) further topicalization.

Another difficulty for the analysis of Hannahs and Tallerman (2000) arisesfrom the facts of negative sentences, and this is less easily eliminated. I havealready noted that in negative sentences, nothing need precede the negativeparticle ne ; and in fact a preposed participle or other single head cannot appearin that position, as shown in (7.32).

(7.32) a. N’en deus ket gwellet Yann e vignonezhasn’t seen Yann his girlfriendYann hasn’t seen his girlfriend

b. Gwellet en deus Yann e vignonezseen has Yann his girlfriendYann has seen his girlfriend

c. *Gwellet n’ en deus ket Yann e vignonezseen hasn’t Yann his girlfriendYann hasn’t seen his girlfriend

This is consistent with the consensus in the literature that the negativeparticle ne, unlike the other verbal particles a, e, occupies C or some otherstructural position and thus allows the finite verb to be non-initial within CP,while precluding the movement of other elements to C. But preposed topicalmaterial is found with negatives when that material is phrasal, as in (7.33).

(7.33) Gwelout e vignonez ne reas ket Yannseeing his girlfriend didn’t YannYann didn’t see his girlfriend

Hannahs and Tallerman propose to account for this by saying that anotherconstraint, one requiring that ne appear in C, outranks all others, thus forcingne to appear in this position even when there is a phrase in topic position. Thisdoes not really solve the problem, though. If ne is in the C position, the con-straints 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 Left-Most(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


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 inBreton. Verbs show inflectional agreement only when the subject is not overtlypresent, either in VP-internal subject position or in the position of topic. Whenthe subject DP is overt within the CP of (7.28), the verb appears in a uniformnon-agreeing form homophonous with the third-person singular. The factsare illustrated in (7.34). The mechanisms of this “dis-agreement” effect are notimmediately relevant here, since what matters is its structural characterization.

(7.34) a. Me a lenn al levrI prt read the bookI read the book

b. Al levr a lennanthe book prt I-readI read the book

c. Ar wazed a lenn al levrthe men prt read the bookThe men read the book

d. Al levr a lenn ar wazedthe book prt read the menThe men read the book

e. Al levr a lennontthe book prt 3pl-readThey read the book

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

(7.35) a. Ne lennan ket al levrI-don’t-read the bookI don’t read the book

b. Me ne lennan ket al levrI I-don’t-read the bookI don’t read the book

c. Ar wazed ne lennont ket al levrthe men they-don’t-read the bookThe men don’t read the book

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


but rather in some position external to the structure of (7.28). If that is thecase, though, the presence of this phrase would not produce violations of theverb-second regularity within (7.28).

If initial subjects in negative sentences are outside of the clause, where arethey? A number of authors have pointed out that the generalization of An-derson and Chung (1977) that only a single “topicalization” is possible in agiven sentence is incorrect. Initial topic-like phrases are found even when someother element must be present in C or SpecCP, as in left-dislocated structuressuch as (7.36).

(7.36) a. Ha me, se en deus bet merket ac’hanon e-barzand I that has been marked of-me inma vuhezmy lifeAnd me, that marked me for the rest of my life

b. Marhadizion amonenn ha uieu, deu e oentbutter-and-eggs merchants two prt 3pl-wereButter and eggs merchants, they were two

It seems plausible to suggest that left-dislocated constituents do not oc-cupy the position of Spec(CP), but instead are left-adjoined to the CP, in aposition where they are not relevant to agreement. I suggest that pre-verbalsubjects in negative sentences such as (7.35) are actually left-dislocated, andthus 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 theconstraints of (7.30).

Two further complications for the analysis of verb-second regularities inBreton have their origins in the incompletely assimilated effects of historicalchange. First is the behavior of the unusual verb emañ. This serves as a copulain locatives and related expressions; it also combines with a non-finite verbalform 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 tibe-loc my brother in the houseMy 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 usedin non-situational contexts when preceded by a DP.


(7.38) Va breur a zo en timy brother prt be in houseMy brother is in the house

From a historical perspective, the anomalous behavior of emañ is straight-forward. Its paradigm originates, essentially, as a combination of hemañ ‘here’plus appropriate forms of the verb ‘to be.’ Presumably, the included elementhemañ originally had the effect of satisfying the requirement that the verbappear in second position. When an overt topic appeared, there was no longera place for the ‘here’ element within the constraints of verb second. The copulawould then be immediately preceded by the nominal topic phrase (presumablyin 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 verbin modern Breton, and not as a phrasal collocation; and that means that there isno obvious way to implement the generalization above by regarding a subpartof emañ as occupying the C position. This is the kind of puzzle that restruc-turing in historical change often presents for the goal of a perfectly consistentsynchronic grammar. There does not seem to be any more elegant solution inthis case than simply to stipulate that emañ itself must appear leftmost withinCP, by means of a specific Alignment constraint (7.39) that outranks the generalrequirements in (7.30), and in particular the constraint NonInitial(Vfin,CP).Given the unique status of this verb, a more principled solution does not seemto 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 throughsuffixes, but rather by complex internal modification. Finite forms (which bearno 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 dis-plays explicit agreement even with an overt subject, as opposed to the behaviorof other verbs as illustrated in (7.34).

(7.40) a. Lennet en deus Yann al levrread has (m) Yann the bookYann has read the book

b. Lennet he deus Mari al levrread has (f) Marie the bookMarie has read the book


c. Me am eus lennet al levrI have (1sg) read the bookI have read the book

Present Future1sg am eus am bo2sg az peus az po3sg,m en deus en devo3sg, f he deus he devo1pl hon eus hor bo2pl hoc’h eus ho po3pl o deus 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 mean-ing “have” were originally composed of an inflected preposition plus a non-agreeing form of the existential verb. As a result, the position corresponding tosubject person marking was that of a prepositional object. This, presumably,stood in a different structural relation to the rest of the clause from the agree-ment material associated with other verbs. Again, however, it is difficult to seehow to implement this historically derived insight in the description of mod-ern Breton, since the forms are synchronically those of a verb with an unusualinternal form of agreement, and there is no other justification for viewing it asa phrasal collocation.

Breton VPs and the Structure of IP

The basic structure of the clause in Breton is thus plausibly viewed as thatin (7.28). But what of the internal structure of the IP in (7.28)? This presum-ably has a head position (I) at its left edge, to which the verb moves in order torealize the features of Tense and Agreement. An associated preceding SpecIPposition could never be filled, by virtue of the high ranking of the constraintLeftMost(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 Fillmoreand McCawley), on which the base position of the subject is within VP, thecomplement 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 comesfrom the fact that non-finite clauses in the language, when they contain anovert subject, display SVO order as in (7.41).


(7.41) a. Kavet am eus ur bluenn vat din da skrivañ gwelloc’hfound I have a good pen for-1sg to write betterI have found a good pen so that I can write better

b. Pedin a reas Lennaig ar vugale dezho da zebrininvite prt did Lennaig the children for-3pl to eatkrampouezhcrêpesLennaig invited the children to eat crêpes

c. Da Yann da welout e oant holl azeFor Yann to see prt were all thereAs far as Yann could see, they were all there

If we take this to be the basic structure, the occurrence of VSO will followfrom Verb Movement to initial I, while movement of the reverse sort has noclear motivation. This establishes the basic structure of clauses in the languageas that of (7.42).

(7.42) [CP

(XP) (X) [IP

I [VP

DPSbj [V′ V (DPObj)]]]]

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 thisposition. It cannot be preceded there by other material, since that would pro-duce (uncompensated) violations of LeftMost(Vfin,IP). The verb-second prop-erties of sentences as complete CPs then follow from the remaining constraintsof (7.30) with the exception of sentences with the verb emañ, where (7.39) takesprecedence and requires that element to appear initially.

7.5 Surmiran Rumantsch

Among the modern Romance languages, the clearest example of a verb-secondsystem is found in Rumantsch, one of Switzerland’s (four) official languages.6

Whether this forms part of a larger Rhaeto-Romance family together with Friu-lian and Dolomitic Ladin is disputed (see Haiman and Benincà 1992; Liver1999), but Rumantsch is clearly distinct within Romance from the various lan-guages referred to as “Italian,” as well as from French, Franco-Provençal, andother surrounding members of the family. Opinions differ as to the originof its verb-second structure, even within the scope of a single work: one ofthe 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.


borrowed from German, while the other argues that it originated as an inde-pendent common northern Italian medieval pattern. Regardless of its origins,however, verb second is a pervasive and thoroughly assimilated aspect of thesyntactic structure of Rumantsch, and thus important to examine in compari-son with the other languages discussed in this chapter.

Most Rumantsch speakers live in the canton of Graubünden, and nearlyall are at least bilingual (mostly in German, though Italian is important insome parts of the canton). In the 1990 census, about 66,000 people indicatedRumantsch as the language of which they had the best command or which theymost used, and of these 41,000 lived in Graubünden.

I focus here on Surmiran, one of five different literary standards that cor-respond roughly to linguistically distinguishable dialect areas. The amount ofvariation among these languages is significant, but not extreme: most speak-ers of one can understand the others fairly well. Surselvan is probably thebest known and best described, and Surmiran is thus a minority languageeven within the minority Rumantsch community, with about 3,200 speakers(Simeon 1996).

Like other forms of Rumantsch, Surmiran is strongly verb second. The ex-istence of verb-second constructions in the medieval Romance languages haslong been recognized (Foulet 1928), and the analysis of these facts in medievalFrench (Adams 1987; Roberts 1993b; Vance 1997), Spanish (Fontana 1993) andItalian (Benincá 1994) has played a part in discussion of broader theoreticalissues. Verb second in the “Rhaeto-Romance” languages of Northern Italy(Benincá 1985, 1994, 1995; Poletto 2000) has also figured in the syntactic litera-ture, 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 purposeshere, some of these are subject clitics, post-verbal clitic elements which un-der quite specific conditions appear in agreement with the subject. Theseare not simply phonologically reduced forms of full pronoun subjects, sincethey do not replace but rather may double the subject nominal. They arealso not simply verbal morphology: verbs agree with their subjects quite in-dependently of the presence of the clitics, and phonological and other con-siderations indicate that the two sets of markers are quite distinct. These twoaspects of Surmiran syntax are closely interconnected, in that the subject cliticsonly occur when the full subject is inverted with the verb. Subject clitics andthe grammatical mechanisms that implement verb second thus form part of asingle complex.

A final factor in this mix is the impersonal subject element ins. At firstglance this appears simply to be a pronominal, equivalent to German man or


French on. The interaction of ins with verb second, however, suggests that ithas 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 clausein Surmiran is SVO, as illustrated in (7.43).7

(7.43) Ursus discorra stupent rumantschUrsus speaks.3sg excellently RumantschUrsus speaks Rumantsch very well

In both main and subordinate clauses, the major constituents appearconsistently in this order and are not subject to significant scrambling or othervariation, with one exception. In main clauses, a single topical constituent(typically an argument of the verb, or an adverbial) may appear in initialposition. When a topical non-subject thus appears initially, the verb and thesubject invert, leaving the verb in second position, as illustrated in (7.44). Theremainder of the clause preserves the same order as in uninverted sentences.

(7.44) a. Rumantsch discorra Ursus stupentRumantsch speaks.3sg Ursus excellentlyUrsus speaks Rumantsch very well

b. Stupent discorra Ursus rumantschexcellently speaks.3sg Ursus RumantschUrsus speaks Rumantsch very well

Thus far everything is straightforward, but an additional possibility arisesin 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 stupentRumantsch speaks.3sg-3sg.m Ursus excellentlyUrsus speaks Rumantsch very well

7 Surmiran examples are presented in the standard orthography, based on German and Italianconventions. The digraphs gl and gn represent palatal laterals [L] and nasals [ñ] as in Italian. Tschrepresents 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. Thevoiced correspondent of tg is written as g before front vowels. Voiced and voiceless coronal frica-tives 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 onthe penult if the last vowel is schwa.


b. *Ursus discorra=’l stupent rumantschUrsus speaks.3sg-3sg.m excellently Rumantsch(Ursus speaks Rumantsch very well)

Sentences like (7.45a) where a clitic appears together with an overt sub-ject have an interpretation with mild emphasis on the doubled subject. Thepresence of the clitic, however, licenses the deletion of the subject (subjectto referential recoverability in discourse, naturally) as in (7.46), in which caseemphasis is no longer implied.

(7.46) Rumantsch discorra=’l stupentRumantsch speaks.3sg-3sg.m excellentlyHe speaks Rumantsch very well

Surmiran is not a pro-drop language in general, and subjects cannot beomitted in the absence of such a clitic, regardless of whether the conditions forinversion are present. These points are illustrated by the ungrammaticality ofsentences such as those in (7.47).

(7.47) a. *Discorra rumantsch stupentspeaks.3sg Rumantsch excellently(He speaks Rumantsch very well)

b. *Rumantsch discorra stupentRumantsch speaks.3sg excellently(He speaks Rumantsch very well)

These facts are not limited to sentences with third-person nominalsubjects --- the same pattern shows up in other persons as well, as in (7.48).

(7.48) a. Ia/*∅ discor mal rumantsch(I) speak.1sg badly RumantschI speak Rumantsch badly

b. Rumantsch discor ia/*∅ malRumantsch speak.1sg (I) badlyI speak Rumantsch badly

c. Rumantsch discorr=a malRumantsch speak.1sg-1sg badlyI speak Rumantsch badly

d. Rumantsch discorr=a ia malRumantsch speak.1sg-1sg I badlyI speak Rumantsch badly


The full set of post-verbal subject clitics is given in Figure 7.4. These areovert in all cases except the second-person plural. I will discuss their analysisfurther in Chapter 8, where I will show that they cannot be treated simplyas part of the morphology of the verb. Summarizing the points above, thesesubject clitics are introduced optionally (and post-lexically) in constructionsthat call for subject--verb inversion.8 When present, subject clitics sanction theotherwise ungrammatical replacement of overt subjects by phonetically nullpronouns. Such omission of the subject is not obligatory, however, and theclitics may be doubled by an overt nominal or pronominal subject phrase.

Person/Number(/Gender) Subject clitic1sg =a2sg =t3sg m =’l3sg f =’la3sg impersonal =(i)gl1pl =s(a)9

2pl ∅3pl m/f =igl

Figure 7.4. Surmiran subject clitics

The Inversion Construction

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

(7.49) a. La steiva ò Ursus nattagea bagnthe living room has.3sg Ursus cleaned wellUrsus cleaned the living room well

8 Thus far, we have only seen inversion in the presence of a sentence-initial non-subject topicconstituent. Other constructions will be discussed below which also call for inversion, and wherethe 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 onthe penult. This includes (nous) ıschan “(we) are,” and is general in tenses other than the presentindicative where the 1pl form has penultimate stress, such as the conditional (nous) cantessan“we would sing.”

10 Examples of inversion below may have either a post-verbal clitic, a full nominal subject fol-lowing the verb or both, with this focused or emphatic interpretation. Except where necessary,I will not differentiate these cases.


b. Tar igl gi da Rummy vala igl joker adegnain the game of rummy is.worth.3sg the joker always25 puncts25 pointsIn the game of rummy, the joker is always worth 25 points

c. Giond ier a spass ò Ursus scuntro Ladinagoing yesterday for a walk has.3sg Ursus met LadinaWhile 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 nattageaYesterday the living room has.3sg Ursus cleaned

Under the conditions of inversion, the (main or auxiliary) finite verbprecedes the subject, accompanied by any clitic elements dependent on it asin (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 othernegatives. This particle is quite often omitted in spoken Surmiran, as ne isin spoken French. Apart from the position of the finite verb, other elementsof the clause remain in the same relative position they occupy in uninvertedsentences.

(7.51) a. Cleramaintg n’=ò=’l Ursus betg savia cheglObviously neg-has.3sg-3sg.m Ursus not known thatObviously Ursus didn’t know that

b. Ier seira n’=ans=ò Maria betg telefonoYesterday evening neg-1pl-has.3sg Maria not phonedYesterday 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 withthe verb do not constitute additional constituents in the relevant sense.

In compound tenses, the verbal past participle alone can appear at thebeginning of the sentence. Although the participle represents the main lexicalverb of its clause, and arguably heads a verbal phrase, complements of thisverb cannot accompany it in initial position. They remain where they wouldbe in an uninverted clause as shown in (7.52), although as the marginal statusof (7.52e) illustrates, frequent adverbs may to some extent appear inverted withthe participle.


(7.52) a. Maglea va ia en traclo cun caschieleaten have.1sg I a sandwich with cheeseI ate a cheese sandwich

b. *Maglea en traclo cun caschiel va iaeaten a sandwich with cheese have.1sg I

c. La notg passada ò Gion durmia malamaintglast night has.3sg John slept badlyLast night John slept badly

d. Durmia ò Gion malamaintg la notg passadaslept has.3sg John badly last nightJohn slept badly last night

e. ??Durmia malamaintg ò Gion la notg passadaslept badly has.3sg John last night

This construction is reminiscent of the Icelandic “Stylistic Fronting” con-struction, which we saw above in section 7.2. A variety of other differencesbetween Surmiran and Icelandic, however, make it difficult to establish a clearparallel.

It is also possible to emphasize the lexical main verb of the sentence, inwhich 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 infinitivescannot be accompanied by complements, as (7.53b) shows.

(7.53) a. Cantar canta=’l Ursus ena canzungto.sing sings.3sg-3sg.m Ursus a songUrsus is singing a song

b. *cantar ena canzung canta=’l Ursusto.sing a song sings.3sg-3sg.m Ursus

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

(7.54) *screiver fatsch ia en codeschto-write do-1sg I a book


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

(7.55) a. Cantar cantava=’l Ursus bagnto.sing sang.3sg.imperf-3sg Ursus wellUrsus was singing well

b. */?Cantar ò=’l canto Ursus bagnto.sing has3sg-3sg.m sung Ursus wellUrsus sang well

On the other hand, Breton shares with Surmiran the possibility of having averbal participle in sentence-initial position, and also the restriction that sucha 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 paral-lels among verb-second languages from three distinct families within Indo-European must remain a topic for future research.

I analyze these basic aspects of Surmiran syntax as follows. The basic orga-nization of the core clause in the language is SVO. In simple declarative clauseswith no topic, there is no distinction between IP and VP, and the structure isas 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 withinwhich the VP is embedded. The Specifier position within this phrase is filledby the topic. The finite verb then moves to the I position; this might be eitherbecause it must be located in the head of IP in order to inherit the Tense andAgreement features of the clause, or in order to remain in second positionwithin IP. On either analysis, it is this movement of the verb to I within anincluding IP that constitutes “Subject--Verb Inversion” in Surmiran. On thisaccount the structure of sentence (7.44a) above is as shown in Figure 7.5.

In terms of this structure, we can now characterize the circumstancesunder which subject clitics can appear. Specifically, when the verb in I C-commands the subject phrase in Spec(VP), a special clitic agreeing with theproperties of the subject may be introduced following the verb. When such aclitic appears on the verb (and only then), a null pronominal (pro) is licensedin the corresponding subject position. I leave additional details concerning theoperation of this proposal to Chapter 8, but the overall outline should be clear.


IP

I′

DP VP

I V′

V′

DP V DP Adv

Rumantsch discorra 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 Surmi-ran. Languages with verb second can differ in the class of clauses in whichthese effects can appear: in German, verb second is found only in main clauses(and subordinate daß clauses with no overt complementizer), while most sub-ordinate clauses have the finite verb in final position. Icelandic, in contrast,shows verb-second effects in both subordinate and main clauses. It is worthasking 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 con-struction requires the presence of a topic element, and Topicalization is in gen-eral a root phenomenon. It is thus pragmatically unusual to find the conditionsfor inversion in a subordinate clause, but examples such as (7.57) where initialtopics are well-formed in embedded clauses do show inversion.

(7.57) a. Cartez tg’igl settember turnan=sbelieve.2pl that-art September return.sbjnctve.1pl-1plainten chel hotelin this hotelDo you think in September we’ll come back to this hotel?

b. Ia pains tgi dultschems vegiaI think.1sg that sweets have.sbjnctve.3sgCorinna gugentCorinna gladlyI think Corinna likes sweets


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

(7.58) a. Tge ò=’la (Ladina) cumpro?what has.3sg-3sg.f Ladina boughtWhat did Ladina/she buy?

b. Cura ò=’la (Ladina) cumpro en auto?when has.3sg-3sg.f Ladina a carWhen did Ladina/she buy a car?

c. Igl auto da tgi ò=’la (Ladina) cumpro?the car of whom has.3sg-3sg.f Ladina boughtWhose car did Ladina/she buy?

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

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

When a non-subject wh-phrase is fronted, inversion occurs along with thepossibility of a subject clitic, as in (7.58). We can tell that these effects areassociated with the configuration that results from the movement, and notwith wh-movement itself, because a wh-phrase fronted out of a complementclause produces them not in the clause from which the movement takes placebut rather in the matrix clause, as illustrated in (7.60). Wh-movement causesthe wh-phrase to precede the matrix subject, where it is no longer within thecomplement 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?boughtWhat do you think that Ladina bought?


Sentences such as (7.60) show that the complementizer tgi “that” does notitself trigger inversion in a subordinate clause. In the analyses of Icelandic (sec-tion 7.2) and Kashmiri (section 7.3) above, I suggested that the elements aðand ki /yi which appear initially in complement clauses in these languages arenot complementizers in the sense of occupying the C position in clause struc-ture, but rather initial special clitics introduced as a reflection of a CP’s subor-dinate position. I propose to treat Surmiran tgi= in the same way.

Multi-word expressions can also introduce subordinate clauses, withoutprovoking inversion, as in (7.61).

(7.61) Siva tg’els on en unfant, stat el pi savenssince that-they have.3pl a child is.3sg he more oftena tgesaat homeSince they have a child, he is home more often

Rather than treating siva tgi “since” in (7.61) as a complex complementizeroccupying a C position, I suggest that this construction involves the indepen-dently occurring preposition siva “after, following” governing a clause markedwith special clitic tgi= like other embedded clauses.

Example (7.61) also demonstrates that an initial subordinate clause can it-self occupy the structural Spec(IP) position, triggering movement of the verbto 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 mightbe because the wh-phrase itself occupies the same Spec(IP) position as top-ics, and thus provides an I position outside of the VP to which the verb mustmove, as in declaratives. This is probably not the best analysis, however, as weshall see.

Turning to relative clauses, we find that these do not in general show in-version regardless of whether the relativized element is subject or non-subject,as the examples in (7.62) illustrate.

(7.62) a. Igl codesch tgi è sen meisa sas=t aveirthe book which is on the table can.2sg-2sg haveThe book which is on the table you can have

b. Igl velo tgi Ursus ò cumpro n’=è betg novthe bike which Ursus has bought neg-is.3sg not newThe bike which Ursus bought is not new


c. Igl gioven agl qual ia va scretgThe youngster to.the which I have.1sg writtenè sto igl mies scolaris.3sg been the my studentThe youngster to whom I wrote was my student

d. La matta dalla qualla te ast survagniathe girl from.the which you have.2sg receiveden canaster mareida proximamaintga basket marries.3sg in the near futureThe girl from whom you got a basket is getting married soon

Relative pronoun expressions thus appear to be unlike wh-question phrases innot occupying a structural position that would result in movement of the verb.

Another construction in which we do find inversion suggests a basis forthis structural difference between relative pronouns and question words. Inyes/no questions such as (7.63), inversion takes place with no phrase of any sortpreceding the verb in its displaced position.

(7.63) a. È igl viadi sto tger?is.3sg the trip been expensiveWas the trip expensive?

b. Ast er te gost da neir?have.2sg also you desire to comeDo you want to come too?

c. Lain=sa nous eir cugl tren?want.1pl-1pl we go with.the trainDo we want to take the train?

d. At=ò=gl plaschia an Sicilia?2sg-has.3sg-3impers pleased in SicilyDid you like Sicily?

The possibility of subject clitics in this construction (as with other instancesof inversion) suggests that the position of the verb here, as in declaratives withinitial topics and content questions with initial non-subject wh-phrases, is onefrom which it C-commands the original subject position.

Inversion in yes/no questions cannot be associated with a requirement thatthe 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


a consequence of additional functional structure introduced to host an initialphrase, since no such phrase occurs. Something else must be present in thesestructures to result in movement of the verb.

Assume that questions involve an interrogative operator “Q” appearingin I, the head of the clause. In content questions, the questioned wh-phrasemust appear as the Specifier of I[+Q]; as a result, where the wh-phrase is anon-subject, such sentences have the core VP structure of the clause embeddedwithin a higher functional IP projection like that of topicalized sentences, withthe verb attracted to I[+Q] and the wh-phrase in Specifier position. If the wh-phrase is the subject, however, no additional structure is motivated beyondthe 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. Itmust thus appear on the head of an IP within which the VP is embedded, andthis position again attracts the verb. Inversion (or Verb Raising) thus occursboth in yes-no questions and in content questions, except when the questionedphrase is the subject.

Relative clauses, in contrast, involve a different operator which is unrelatedto I and thus does not attract the verb. The relativized phrase appears as thespecifier of this operator, and in this position binds a gap inside the VP towhich it is related. Relative clauses do not contain topics, and there is no othermotivation for Inversion in their structure.

“Inversion,” then, consists in the movement of the verb to an I positionoutside of its basic VP, triggered by the IP structure necessary to support aninitial 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, thedisplaced verb precedes and C-commands the VP-internal subject position,the condition which sanctions the presence of a subject clitic. Inversion andthe 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 beconsidered: a set of complications that arise in connection with a class of im-personal sentences involving the element ins ‘one.’ This is a modern reflex ofLatin unus, similar in its interpretation to French on and German man. It in-dicates an impersonal subject, in sentences like (7.64). Not surprisingly, suchsentences show third-person singular agreement.

(7.64) Ins pò eir quant spert tg’ins votins can-3sg go as fast that-ins wants.3sgsen las autostradas svizrason the freeways SwissYou can go as fast as you want on the Swiss freeways11

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


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

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

b. *Mintgign digls guids ò la sia modaeach of.the guides has.3sg the his wayda trattar cun insof 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 indefinitepronoun. Just like any subject, it precedes the verb, and is otherwise sentenceinitial. But surprisingly, when a topic phrase precedes ins as in (7.66), the verbdoes not invert.

(7.66) a. Dalla derivanza digls rets ins so tant scuof.the origin of.the Rhaeti ins know.3sg so-much asnavotnothingOf the origins of the Rhaeti12 we know almost nothing.

b. D’anviern ins pò eir sur tot igls pass cun autoIn winter ins can.3sg go over all the passes with carIn 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 uins travels.3sg more good with-the train orcugl auto sch’ins fò viadis pi lungs?with-the car if-ins makes trips more longDoes one travel better by train or by car when making longer trips?

b. Tge meis digl onn ins dovra pneuswhat month of.the year ins needs.3sg tiresd’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 does not undergoinversion with a following finite verb.

12 Early indigenous people of the Rumantsch area.


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

(7.68) a. Ainten chell’ustareia ins na magl=igl betgin this-inn ins neg= eat.3sg-3impers notschi bagn, on=igl detgso well have.3pl-3pl saidIn this inn you don’t eat so well, they said

b. Ins pò=gl fimar cò?ins can.3sg-3impers to-smoke hereCan you smoke here?

c. Cun tge tren ins vo=gl igl miglerwith what train ins goes.3sg-3impers the betterper eir da Sargans a Wien?to go from Sargans to ViennaWhich train is better to go from Sargans to Vienna?

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

(7.69) a. Chel ins dei=’l dapertotthat ins says.3sg-3sg.m everywhereThat they say everywhere

b. Ena schi greva lavour ins sto=’lA so hard job ins should.3sg-3sgmbetg far tot suletnot do all aloneSuch 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.


(7.70) a. Ins dei dapertot tgi . . .ins says.3sg everywhere thatThey say everywhere that. . .

b. Ins sto betg far tot sulet ena schi greva lavourins should.3sg not do all alone a so hard jobOne shouldn’t do such a hard job all alone

Although ins seems like an indefinite subject pronoun, its behavior is morelike that of a special clitic. Like the negative element na and the object pronom-inal clitics, it appears in a fixed position in relation to the finite verb and doesnot invert with it. Its presence, however, does satisfy the verb’s need for a sub-ject. This pattern resembles that of impersonal se in Spanish or (perhaps morerelevantly) si in Italian, as in the examples of (7.71).

(7.71) Spanish: En México se trabaja muchoin Mexico se works.3sg muchIn Mexico one works a lot

Italian: Si lavora sempre tropposi works.3sg always too muchOne always works too much

In some other forms of Rumantsch, particularly those spoken in the Enga-dine, we find a much more direct correspondence with the pattern of(7.71) characteristic of other Romance languages. Engadine Rumantsch hasadopted 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 qualchevoutasPassing across the desert 3sg.refl finds.3sg sometimesskeletsskeletonsCrossing the desert, one sometimes finds skeletons

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

(7.73) Passand tres il desert chatta ün qualchevoutaspassing across the desert finds.3sg man sometimesskeletsskeletonsCrossing the desert, one sometimes finds skeletons


The form ün in Puter (like Surmiran ins, a reflex of Latin unus) behaveslike German man in occupying the subject position and inverting with the verbunder the appropriate circumstances. Surselvan also has this construction, forwhich that language uses the pronoun ins, homophonous with the Surmiranform and derived from the same source but which inverts with the verb whenrequired 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 endNa, alla fin ein ins mai.no at-the end is.3sg ins neverWhere 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 usagein Sotses (Casti [Tiefencastel], Alvaschagn).

(7.75) Chel dei ins dapertotThat says.3sg ins everywhereThat they say everywhere

Speakers in the Surses area of Surmeir (on whose speech the present de-scription is based) also recognize sentences with inverted ins as possible,though in the region roughly between Mon and Marmorera, and also in thenormative description of Signorell, Wuethrich-Grisch, and Simeon (1987), un-inverted ins is the rule. Speakers who do not invert ins characterize sentenceswith inversion of this element as imitative of German usage.

The structure with inverted ins calls for no particular comment, since thatis just what we would expect of an impersonal pronoun occupying the subjectposition. What is much more remarkable is the possibility of the uninvertedstructure. Uninverted ins patterns structurally as a clitic, similar to Spanishse, Italian si, or Vallader as, despite origins which would lead us to expect itto be treated as a non-clitic pronoun. How reasonable is it that ins has beenreanalyzed from a full indefinite pronoun to a clitic, and what might have ledto that development?

A glimpse of the relevant history is provided in Grisch (1939), a descriptionof what we might see as an earlier, transitional stage of the language. In thatsource, the phonetic form in which ins appears is actually [@ns]. This wouldbe homophonous with the first-person plural object clitic pronoun ans, whichmight in turn have led to a conflation of the two. Note that in French and


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

(7.76) French: Nous, on fait pas ça iciwe on does.3sg not that hereWe don’t do that here

Italian: Si è contenti quando =ci scrivonosi is.3sg happy.pl when 1pl write.3plWe are happy when they write to us (Burzio 1992)

To the extent ins from unus was pronounced [@ns], it could readily beconfused with the first-person plural object clitic ans. Despite the fact thatins and ans are clearly distinct elements in contemporary Surmiran, the pos-sibility that they might be confused is apparently still alive. This is suggested bythe fact that the normative description of the language contains a remarkable(and quite isolated) warning that “Igl pronom subject ‘ins’ è da disfranztgerbagn digl pronom ‘ans’ tgi serva scu pronom reflexiv e pronom persunal objectdall’amprema persunga plural”13 (Signorell, Wuethrich-Grisch, and Simeon1987: 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 theverb and is itself preceded by the negative particle na if that is present, ins itselfprecedes na, as in (7.77).

(7.77) Da lò ins n’=ans= vei betg cleramaintgfrom there ins neg-1pl sees.3sg not clearlyFrom there one doesn’t see us clearly

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

(7.78) a. El vot ans= tarmetter ena factura dumangHe wants.3sg 1pl to.send a bill tomorrowHe wants to send us a bill tomorrow

b. Mintgatant ins stò(=gl) spitgier en pooften ins must.3sg(-3impers) wait a bitOften you have to wait a bit

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


c. *Mintgatant stò(=gl) ins spitgier en pooften must.3sg(-3impers) ins wait a bitOften you have to wait a bit

In compound tenses such as the perfect, object pronoun clitics like anscan appear either before the (inflected) auxiliary or before the participle, butins always precedes the auxiliary.

For these reasons (among others), ins and ans must be treated as quitedistinct grammatical elements. Nonetheless, it appears that at some point inthe history of the language, they were at least partially confused, and ins cameto be treated as if it were a clitic like ans. This confusion was probably en-couraged by the fact that in Italian, a language with which many Surmiranspeakers are at least somewhat familiar, the indefinite subject interpretationprovided by ins is supplied by a construction with no overt subject and a cliticpronoun si---which can also have first-person plural reference.

As a result, ins has taken on the behavior of a clitic associated with thefinite verb, rather than that of an autonomous nominal expression. This hasthe strange consequence that the subject position in ins-sentences is not overtlyfilled: strange, because Surmiran normally allows the subjects of finite verbs tobe unexpressed only when they are doubled by a subject clitic in the inversionconstruction. 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, thisought to mean that in inversion contexts when ins represents the subject, thereis nothing to invert with the verb, and we have seen that the word order inthese cases remains fixed. We also saw, however, in (7.68), that the subjectclitic =(i)gl, normally a diagnostic of inversion, can appear under just thesecircumstances.

In contemporary Surmiran of the variety in which sentences like thoseof (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 reana-lyzed as a special clitic associated with the finite verb. The subject positionin impersonal sentences, accordingly, came to be filled not by ins itself, butrather by a phonetically null pronoun with impersonal reference: the PROArb

often assumed to represent the subjects of non-finite predicates like those ofComing in second is never fun and It is more fun to come in first. Thespecial clitic ins= is then introduced at the left edge of a finite verb whosesubject 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


with respect to ins. On the other hand, when the verb is displaced to sucha position and thus comes to C-command its original subject, the subjectclitic =(i)gl may be introduced as in any other case where a displaced verbC-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 asthe one we refer to in describing other languages such as German or Icelandicas “verb second” languages?

In Surmiran, unlike the other languages we have surveyed thus far, it ap-pears that the correspondent of verb second involves exactly displacement ofthe verb from its base position (within VP) to a head position in a higher layerof structure. The displacement, however, appears to be motivated only by theproperties of the head itself.

In particular, Alignment constraints such as LeftMost and NonInitial playno obvious role in the inversion construction. This is most apparent in imper-sonal sentences with ins, where we have seen the subject position is phoneti-cally empty, and the verb (together with its clitics, including ins) thus appearsat the left edge of the constituent in which it heads.

Just the sort of analysis that has commonly been proposed for Germaniclanguages with verb second thus turns out to be appropriate for Surmiran.Somewhat ironically, however, the evidence that supports that conclusion moststrongly comes precisely from the fact that Surmiran is not consistently a verb-second language. This is quite distinct from the properties of special construc-tions such as yes/no questions, which must be provided for separately in anycase in other verb-second languages like German. In Surmiran, however, itturns out that under the one set of circumstances in which the pre-verbalsubject position can be phonetically empty in a neutral declarative clause (inins-sentences), the verb shows no disposition to avoid the result that it therebyappears initially.

What gives Surmiran the appearance of a true verb-second language is theinversion construction, especially as this is associated with initial topic phrases.Surmiran (and Rumantsch more generally) may have borrowed the structuralproperties of this construction (topic phrase in Spec(IP) and displacement ofthe verb to I) from Germanic, or it may have developed within the history ofRomance, but this historical matter has no bearing on the resultant syntax.

We come down to the conclusion that the “verb second” property in Sur-miran is not really a requirement that the verb be in second position at all, or


indeed in any structurally uniform place. What gives rise to that impression isthe 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 moti-vated, properties of their heads may provoke a displacement of the finite verbto a position immediately following an initial specifier. In this case, at least,Wackernagel’s (1892) proposed unification has little to recommend it. Verbsecond here does not fall together with the mechanisms describing second-position clitics.

7.6 Conclusions

Second-position clitics and verb second have in common the fact that bothmay reflect constraints requiring the grammatical properties (i.e., the func-tional content) of a phrase or clause to be aligned as closely as possible with theleft edge of the inflectional domain, without being absolutely initial(in IP, or in some languages CP). They differ in that clitics are introduced by aphonological mechanism of affixation, while verbs that inherit the functionalproperties of the clause they head are subject to movement by normal syntacticmechanisms.

On the other hand, all that glitters is not gold: verb-second constructionsin German, Icelandic, Kashmiri, and Breton do appear to involve essentiallythe same Alignment constraints as those responsible in other languages forsecond-position special clitics, but the superficially very similar constructionin Surmiran does not. To the extent it is valid for some core examples, Wack-ernagel’s (1892) suggested unification thus does not generalize to all instancesof verb-second languages.

It is also important to note that while second-position clitics are found ina wide range of languages throughout the world, clear instances of verb secondlanguages are difficult (if not impossible) to find outside of Indo-European.Furthermore, with the possible exception of Kashmiri, languages that displayverb 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 con-nection between the two phenomena. If the unification we set out to exploreat the beginning of this chapter were really pervasive, we would expect thesetypological matters to work out rather differently.

This does not, however, contradict the claim that the theory of second-position clitics can be extended to an account of at least a significant subsetof verb-second phenomena. The two constructions have in common that they


involve requirements for aligning grammatical material as close to the begin-ning of a particular structure as possible, while at the same time not allowingit to appear at the absolute beginning of the structure. The main differencebetween the two types of phenomena is that the clitics are introduced intothe relevant position by a phonological affixation mechanism while the verbsarrive in their position by syntactic displacement. What is especially interestingin this account is the possibility of maintaining a morphological/phonologicalaccount of the clitic phenomena and a syntactic account of the verb phenom-ena, while at the same time, using essentially identical OT constraints to coverboth cases.

8

Pronominal Clitics

The canonical examples of special clitics in the minds of many linguists arepronominals such as those found in association with the main verb in Romancelanguages1 or those found in second position in languages like Warlpiri orTagalog. Pronominal clitics have a number of interesting properties in theirown right. Some of this interest is more or less independent of their specialclitic status, and other matters are important to explore in connection withthe very notion that these elements can be treated linguistically in the sameway as other, non-pronominal special clitics. The purpose of this chapter is todeal with (or at least acknowledge) some of these issues. Given their complex-ity, and the richness of the existing literature dealing with them, this surveywill inevitably have something of the character of a superficial whirlwind tour.Nonetheless, it is important to show that the present framework provides abasis for discussing and analyzing these classic problems.

I have been assuming we can call pronominal clitics the functional mor-phology of phrases, but what does that mean as far as the underlying grammat-ical structure is concerned? Pronominal clitics are also frequently assimilatedin various ways to agreement markers—what connections are there betweenthe two? A key role in understanding these matters is played by Clitic Dou-bling, a construction whose properties help to clarify the relation betweenpronominal clitics and the argument positions they are related to. A furthertopic which has been prominent is the construction known as “Clitic Climb-ing.” While a relatively traditional analysis of this construction fits well with thegeneral framework adopted here, its extension to cases of Long Distance Agree-ment is perhaps less familiar. Finally, while attention to pronominal clitics hasgenerally focused on elements corresponding to objects of various types, some

1 Vast amounts have been written about the Romance pronominal clitics, and it is quiteimpossible for me to cover the complete research history relevant to these elements, althoughmuch of the discussion below is devoted to them. A very useful survey of relevant phenomenaand 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 discussedhere because these would take us too far afield from the main point, but I do not think any ofthese matters pose serious problems for the analysis to be presented.


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

8.1 Pronouns and Agreement

What is the structural relation between a pronominal clitic and an argumentposition which it is interpreted as specifying or filling? In “classical” generativegrammar, this was considered a straightforward instance of movement. Suchclitics were generated as pronouns of a certain sort, occupying an argumentposition just as any other nominal expression might. By virtue of some identi-fiable property (e.g., “[+Clitic]”) they then underwent movement to adjoin tothe Verb—or wherever else they might appear in a particular language, such assecond position.

The movement analysis explained one important property of clitics: inmany cases, they are mutually exclusive with overt nominal expressions ina given argument position, as illustrated in sentences (8.1) from (standard)French.2

(8.1) a. Les enfants l’= ont déjà mangéThe kids it have already eatenThe kids have already eaten it

b. *Les enfants l’= ont déjà mangé le gateauThe 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, atleast 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 consistentwith the simultaneous presence of the full phrase al professor.

(8.2) Le= entregué el libro (al professor)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 toan overt argument expression. This is called clitic doubling, and it obviouslyraises 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 significantpause, indicating a right-dislocated structure. This is irrelevant to the point here, which concernsthe complementarity of clitics and full nominals specifying the same argument within a clause. Itshould also be pointed out that some forms of spoken French differ from the “standard” languagein 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 cliticsare not elements that have been moved to where they appear in the surfaceform from some argument position. Rather, they are the overt reflection ofproperties of that position, construed as part of the functional content of theclause and realized by a principle of phrasal morphology as a modificationof the phonological form of the clause. In fact, the phenomena surroundingpronominal special clitics are rather similar to what we find in verbal agree-ment (as various authors have noted). I will thus develop a view of these cliticsby starting from corresponding facts in that domain.4

Kinds of Agreement

In many (perhaps most) languages, agreement as marked on the verb registerscertain properties of an argument, and does not supplant the overt expressionof the argument. This is true in French, for example, where agreement anda full nominal subject are both present in all finite clauses. In other cases,however, agreement morphology and overt argument expressions are in com-plementary distribution. A particularly clear example is provided by theVenezuelan Carib language Pemon, as illustrated by the examples in (8.3).5

(8.3) a. kamicha ke Antonio-da mure ponte-’poclothes with Antonio-erg child dress-pastAntonio dressed up the child with clothes

b. kamicha ke mure ponte-’po-i-yaclothes with child dress-past-3-ergHe dressed up the child with clothes

c. kamicha ke i-ponte-’po Antonio-daclothes with 3-dress-past Antonio-ergAntonio dressed him up with clothes

d. kamicha ke i-ponte-’po-i-yaclothes with 3-dress-past-3-ergHe dressed him up with clothes

3 This is true even if one adopts a “copy and delete” view of displacement, since the clitic is notliterally 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 theoriesentertained for agreement within Lexical Functional Grammar. See Bresnan (2001: ch. 8) for anextended discussion in that framework.

5 The data below are from Jose Alvarez, unpublished field notes. These facts were reported byAlvarez on the Linguist List, vol. 6, no. 574, in 1995.


An extreme variant of this situation is proposed by Baker (1995) for Mo-hawk (and other “Polysynthetic” languages). Baker argues that the Mohawkverb always includes agreement (and/or an incorporated Noun), but that overtnominal argument expressions appear in the position of adjuncts, rather thanthat of arguments. The presence of agreement is claimed to preclude the pres-ence of a nominal in the argument position, as in Pemon. The difference isthat the presence of agreement itself (ignoring for now the case of incorpo-rated nouns, to which I will return in Chapter 9) is obligatory in a language ofthis 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 morearguments may always be present. While such agreement is compatible withthe simultaneous presence of an overt argument, though, that argument mayalso be absent from the surface form, in which case the argument it refers tois interpreted in the same way a pronoun would be. An example is providedby Georgian, as in (8.4). The verb momc.era ‘he wrote it to me’ agrees simul-taneously with all three of its arguments (although there are only two overtaffixal markers of agreement present in the form). Any one of these may bephonologically 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.aorVano/he wrote a letter/it to me

We need to provide, then, for several distinct possibilities. Sometimesagreement marking simply reflects the properties of the argument (as for in-stance in French subject–verb agreement). Sometimes it alternates with a fullargument (as an option, as in Pemon, or perhaps obligatorily, as in Baker’sanalysis of Mohawk). Finally, its presence sometimes sanctions (while notrequiring) a null argument.

I propose to regard the case of mere “registration”6 of an argument as in-volving an operation that copies some relevant morphosyntactic features (typ-ically, person, number, and possibly gender; see Corbett 2003 for a survey)from the argument to the Morphosyntactic Representation7 of the verb, whererules of the morphology may spell them out. Alternative views on which thesefeatures are generated independently in the Morphosyntactic Representationor its equivalent, and then checked for the required identity at some point

6 The useful terminological distinction between “registration” and “agreement” proper wasintroduced in early work by Perlmutter and Postal on a theory of Relational Grammar. Unfortu-nately, it does not seem to have survived into current discourse, but it is well worth reviving.

7 See Anderson (1992) for this notion.


in the syntax, can be regarded for present purposes as notational variants ofthis picture.

The second type of agreement is supplied by Baker’s (1995) analysis ofMohawk, which I adopt here at least for the purposes of discussion. In thiscase, we have obligatory agreement coupled with the impossibility of full DPsin argument position. Baker argues at length that the argument positions arein fact present, and occupied by phonetically null (but referential) pro elementswhose features are reflected in verbal agreement.

To describe this situation, we need to establish a link between the par-ticular syntactic configuration posited for Mohawk and something in the mor-phology of the language. For Baker, that link is provided by the MorphologicalVisibility Condition (“MVC”). This is a parametric choice in the grammarsof individual languages: in some languages, such as Mohawk, the conditionholds, while in others, like English, it does not. Where the MVC obtains, itrequires that all arguments of a head (such as the verb) must be reflected inthe morphology of that head. Baker assumes that the markers which appear inthis function will then prevent full nominals from receiving Case in argumentposition. As a result, they must appear (if at all) in some external, adjoinedposition. This yields the particular clause structure which Baker argues forin Mohawk.

A key role in this explanation is played by the assumption that agreementelements in Mohawk absorb the structural case which the verb would other-wise assign to its arguments. This must obviously be interpreted as a fact aboutMohawk, at least in part, since agreement is not universally incompatible withthe appearance of full nominals in argument positions. Something extra thusneeds to be said in the grammar of such a language if the desired consequencesare to follow.

Mohawk verbs do, of course, subcategorize argument positions, but onBaker’s analysis these can only be filled by phonetically null pro with some ap-propriate set of features. In this language, these are (roughly) [±me], [±you],[±sg], [±pl], [±masc], [±fem], [±zoic], where dual number can be repre-sented as [-sg, -pl] and neuter gender as [−masc, −fem, −zoic]. Because ofthe presumed Case absorption effect, overt nominals appear in clause-externaladjunct position, where they form a chain with the presumed pro.

An alternative to Baker’s assumption that Mohawk agreement absorbs Caseis simply to say that Mohawk verbs do not assign structural Case at all—orrather, that the only principle of structural-case assignment in the language isthe one licensing nominals in adjoined position. Notice that since agreementis obligatory, any hypothetical structural Case related to the verb would never,in fact, be assigned on Baker’s assumptions.


The exclusion of full DPs from argument positions need not, however, berelated to the assignment of abstract Case at all. I will suggest below that thiscan be derived instead from Binding relations, which would disallow anythingbut pro in positions with which the verb agrees referentially (as is the case inMohawk). On that assumption, the assignment of Case is quite orthogonal tothe issue of what fills argument positions.

The presence of agreement establishes reference, so let us assume (withBaker) that argument positions are also indexed referentially in the verb byagreement. This relation involves a coindexing relation in addition to thecopying (or identity) of features characteristic of agreement in general. Bakerbelieves that what is at work here is the presence in syntactic representationsof functional categories (instances of Agr) which bear indices. These are sub-sequently incorporated (via adjunction) into a single word with the head verb,thus satisfying the MVC. You might notice that the agreement elements thuslook like little clitic pronouns.

There are (at least) two ways in which Baker’s assumptions differ fromthose of the present book. First, like much recent work within the MinimalistProgram (but for different reasons), I consider it important to avoid posit-ing independent functional categories like Agr as syntactic heads. In thatcase, however, there is no obvious constituent to bear the referential burdenof Baker’s Agr elements. Second, in line with the Lexicalist Hypothesis as ar-ticulated in Anderson (1992), the present framework does not allow for thecreation of words within the syntax.

What is necessary is to coindex something in the representation of the verbwith the argument positions to which that verb assigns T-roles. In the presentterms, that “something” clearly has to be the verb’s Morphosyntactic Repre-sentation.

I thus want to say that agreement rules come in two basic flavors, cor-responding to the difference between registration and agreement mentionedabove. Note that Baker needs to say that Agr absorbs Case in some languagesand not in others: the difference to be developed is intended to have the sameconsequences.

The primary function of an agreement rule is to copy some set of featuresfrom an argument to the head, as for instance in English or French subjectagreement. As a result, the Morphosyntactic Representation of the verb (orother predicate) will contain those features, and they can then trigger theintroduction of overt morphological markers in the derivation of the appro-priate word form. This kind of agreement—registration—has no further syn-tactic consequences.


As a second possibility, an agreement rule can also establish a relation ofcoindexation between the argument position in question and the Morphosyn-tactic Representation itself, as is arguably the case in Mohawk. Such a relationis obviously very similar to the one Baker assumes, except that the coindexationis not with a “morpheme” internal to the Verb, but rather with a subset of thefeatural content of the verb’s Morphosyntactic Representation. In Mohawk,then, we could say that agreement is of this second, coindexing sort. Wheresome proi is coindexed with a layer of agreement features, the two must agreein those features.

Now suppose we take seriously the fact that, as a result of agreement, theverb (through its Morphosyntactic Representation) is coindexed with those ofits argument positions with which it agrees. Since the verb is the head of theVP, the “Head Feature Convention” (which I borrow from work in GPSG, Gen-eralized Phrase Structure Grammar: see Gazdar et al. 1985) says that features inits Morphosyntactic Representation are transmitted to that of the VP. In casethe constituent type IP is distinct in a given language from VP, when the verbmoves to I (the head of IP) its features are also transmitted to the Morphosyn-tactic Representation of the IP node.

Note that in many languages, Agreement establishes a relation between theverb and more than one of its arguments. In such a case, the agreement mate-rial corresponding to various arguments must in general be kept distinct. The“Layering Convention” of Anderson (1992) covers just this issue for word-levelmorphology, and extends directly to the case of Morphosyntactic Representa-tions for phrasal categories.

Just as the content of agreement may be transmitted from the verb tosuperordinate phrasal categories which it heads, it may itself inherit other prop-erties from such a phrasal category. Suppose we say that Tense, for example, isa property of IP—part of the Morphosyntactic Representation of that phrasalnode. The Head Feature Convention, by asserting identity between the prop-erties of a phrase and of its head, has the effect of transmitting Tense propertiesto a verb when it occurs in I, the head of IP, and the morphology of the verbmay thus reflect this phrasal property.

In the case of interest to us here, features and/or referential indices as-signed to the verb by Agreement are inherited by the Morphosyntactic Repre-sentation of the clause which it heads, where they constitute (perhaps a por-tion of) the “functional content” of the clause. On this view, “functionalcategories” such as Tense and Agreement are characteristic of a clause, whilenot projecting additional layers of syntactic structure. When realized as inflec-tional morphology on the finite verb, this content characterizes that word aswell by virtue of the identity established by the Head Feature Convention.


Now let us return to the case of genuine agreement: the sort which involvescoindexing as well as feature identity. Here a referential index will be presentby the mechanisms just described at the level of the clause, coindexed with theagreeing argument position. Now suppose an overt nominal expression wereto appear in such an argument position, one with which the verb (and thus theclause node) is coindexed via Agreement. This ought to produce a violation ofthe Binding Conditions, since the nominal would thus be bound (coindexedwith a C-commanding category) within its clause.

This would exclude both R-expressions and lexical pronominals fromappearing in positions with which the verb stands in a relation of agreement(and not merely registration). It is plausible, however, to assume that phono-logically null pronominal elements (pro) are not similarly excluded. In thisway, we could derive the exclusion of overt nominal expressions in argumentpositions 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 havenull pronouns in their lexicon: the only way a motivated argument positioncan be phonetically unfilled is when lexicalization is blocked by virtue of ref-erential agreement. Thus pro is not really a pronoun, but rather just a kindof empty category which (like other empty categories) must be sanctionedby principles of grammar. In this case, the relevant principle is grounded inthe binding relation between referential agreement material and the positionpro occupies.

Based on these considerations, it is possible to propose a typology of therelations between agreement and the arguments it indexes. English or Frenchrepresent one extreme, where agreement is always non-coindexing registra-tion. As a result, argument positions in such languages are never required(or allowed) to be empty in the absence of other specific displacementsor deletions.

In Georgian, or in Italian and the other classic pro-drop languages, theagreed-with position can optionally remain empty. This can be regarded as anoptionality in the operation of Agreement: this always copies the features of theargument, and it may optionally also coindex. Where coindexation appears, weget pro; where it does not, we have a normal pronoun or full lexical nominalexpression.

In contrast, accepting Baker’s extensively argued proposal for clause struc-ture 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 ofphonologically null DPs is somewhat different and more like that of genuine lexical alternantsof full DP expressions.


argument positions are always empty. Mohawk deals with this situation bylicensing the formation of chains which relate an empty (but referential) argu-ment position to the content of a nominal in adjunct position.

In Pemon, a given argument position can be identified either by an overtnominal or by verbal agreement, but not both. We can describe this by say-ing that Pemon agreement is of the referential kind, but Agreement itself isoptional. When it occurs, it triggers the associated morphology and precludesan overt nominal; when it does not, there is no morphology, phonologicallynull pro is not licensed, and the relevant argument position must be filled withan 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 agreementin some languages “identifies” the corresponding position(s). It also comesas close as we are likely to get to formalizing the notion of “strong” agree-ment, in treating this as a parameter of grammatical variation rather than de-riving it from some aspect of the formal nature of agreement itself, such asthe precise combination of features represented, or the extent to which variousforms are distinct from one another—two tempting but ultimately unproduc-tive approaches that have been taken in the literature to grounding the distinc-tion. I suggest that the present proposal provides a more nearly satisfactoryreconstruction.

A Complex Example: Finnish Agreement

A more intricate agreement system than those considered above is found inFinnish, and it is worth exploring how this is to be characterized in the presentframework.

Finnish has a set of markers whose status has been the subject of much dis-cussion, appearing on head nouns to indicate properties of a possessor. On thebasis of their failure to trigger the well-known process of consonant gradationin a previous syllable, as opposed to clear inflectional affixes, some authors(e.g., Nevis 1986) have described these elements as (special) clitics. Despitetheir phonology, Kanerva (1987) shows clearly that they must be treated asword-level affixes, and not as clitics. They thus constitute markers of agree-ment within the nominal phrase. They also show interesting similarities tothe markers of verbal agreement in the language, and both the nominal andthe verbal markers should fall within a theory of agreement relations. Thephonology of these affixes in the broader context of Finnish noun inflectionis analyzed by Kiparsky (2003), and will not concern us here. The account oftheir syntax below follows in large part that of Toivonen (2000), an analysis in


the terms of Lexical Functional Grammar and translated here into the frame-work of the present book.

The facts concerning the nominal agreement system are as follows. Whena nominal has a first- or second-person possessor, the head of the phrase bearsobligatory possessor marking, and an accompanying genitive pronoun may ormay 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” withthe appropriate person.

(8.5) a. Poika myi (minun) marsu-niboy sell.3sg.pst 1sg.gen guinea pig-px1sgThe boy sold my guinea pig

b. (Sinun) kissa-si on sairas2sg.gen cat-px2sg is sickYour(sg) cat is sick

c. Pekka näkee (meidän) ystävä-mmePekka see.3sg.pres 1pl.gen friend-px1plPekka sees our friend

d. (Teidän) kissa-nne on ruma2pl.gen cat-px2pl is uglyYour(pl) cat is ugly

This is straightforward: when the possessor is first or second person, Agree-ment copies its features of person and number onto the head noun, and mayoptionally coindex that noun’s Morphosyntactic Representation with the pos-sessor. If the coindexation option is taken, the presence of that referential indi-cation prevents an overt possessor phrase, and the position of the possessor isfilled by phonologically null pro. If there is no coindexation, pro is not licensedin that position and a full pronoun must appear. This is the normal state ofaffairs described above for “pro-drop” languages.

In the third person, the situation is more complicated. Here we must dis-tinguish three separate cases as far as the applicability of Agreement is con-cerned. 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-nsaboy.pl see.3pl.pres 3pl.gen friend-px3The boysi see theirj,*i friend

Agreement here is obligatory and consists only in copying the person fea-ture(s) onto the head noun. Since this does not involve coindexation, an empty


pro possessor is not licensed and the pronoun must appear. From independentconsiderations of binding, this must always be interpreted as disjoint in refer-ence from the subject of the sentence.

When an overt possessor expression appears that is either (a) non-prono-minal, 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, Agree-ment does not apply at all. No possessive affix appears on the head noun, andthe possessor cannot be replaced by null pro.

The third case, that of sentences like (8.8), is particularly interesting. Whena third-person possessive marker appears on the head noun, and no overt pos-sessor is present, the nominal is interpreted as possessed by the subject ofthe clause. Note that there is no requirement that the possessor in this casebe human.

(8.8) a. Pekka näkee ystävä-nsäPekka see.3sg.pres friend-px3Pekkai sees hisi friend

b. Se heiluttaa häntää-nsäIt wiggle.3sg.pres tail-px3It wiggles its tail

The interpretation of (8.8) shows that a possessor is present, which mustbe a phonologically null anaphoric element controlled by the subject of thesentence. We might plausibly take this to be PRO, the element that also appearsas the subject of certain non-finite complement clauses. When the possessor ina nominal expression is PRO, then, agreement copies its person and numberfeatures onto the head noun along with its referential index. This referentialindex, in turn, is controlled by the subject of the clause in which the nominalappears.

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 doesnot affect the basic point.


as in the first and second person. Rather, each of these possibilities correspondsto 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. HereAgreement is always obligatory, and the morphology of the verb reflects theperson and the number of its subject. When the subject is first or second per-son, it can be phonologically null as shown in (8.9).

(8.9) a. (Minä) ammuin karhun1sg shoot.1sg.pst bearI shot a/the bear

b. (Sinä) näit karhut2sg see.2sg.pst bear.plYou saw the bears

Agreement here is exactly parallel to what we saw above in the nominalcase. With a first-or second-person (possessor or) subject, Agreement copiesthe person and number features onto the head, and may or may not also intro-duce coindexing. When coindexing occurs, the possessor or subject expressionis replaced by phonologically null pro.

In the third person, the verbal and nominal cases diverge, but not en-tirely. First, verbal Agreement is obligatory regardless of the nature of the sub-ject as human or non-human, pronominal or non-pronominal, as illustratedin (8.10).

(8.10) a. Poika/Hän myi kissa-nsaboy/he sell.3sg.pst cat-px3The boyi/Hei sold hisi cat

b. Kissa-ni kuolicat-px1sg die.3sg.pstMy cat died

c. Se oli sairasit was.3sg illIt was ill

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


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

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

This sort of generic interpretation is often found in other languages withnon-finite clauses where there is no controller for the subject position, such asEnglish To know her is to love her. We saw this in the discussion of ins in Sur-miran in section 7.5 of Chapter 7, where I suggested that we should regard suchsentences 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 “imper-sonal” 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 agree-ment system into a closer relation with the nominal pattern. Recall that in thatcase, coindexing agreement was limited to the case of controlled anaphoricPRO as possessor. It seems then that the major difference between the ver-bal and the nominal systems is that in the verbal case, Agreement is alwaysobligatory at least to the extent of copying person and number features. Coin-dexation is also possible if the argument triggering Agreement is first or secondperson; and obligatory if the trigger is PRO (interpreted as Arbitrary in theabsence 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 non-pronominal overt third-person possessor, but otherwise the two systems areentirely similar.

There are further phenomena to be discussed in connection with Finnishagreement, such as the facts of non-finite clauses of various types, the ex-act nature of the control relation relevant to PRO possessors, among others.Nevertheless the present typology of agreement relations seems to provide anappropriate framework for making the distinctions necessary to an adequateanalysis.

8.2 Clitics, Agreement, and Doubling

Now let us return to the analysis of pronominal clitics. I propose to regardclitic pronominals as in fact a form of agreement, differing from verbal agree-ment only in whether the functional content is realized as the morphology


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 Func-tional Grammar and Generalized (or Head-Driven) Phrase Structure Gram-mar based on feature systems, and similar suggestions have been entertainedwithin other frameworks, such as the proposals of Franks and King (2000).

The overt manifestation of agreement material by pronominal special cli-tics can appear in various places, as we have already seen. Clitics may appearwith reference to the beginning of the clause—in second position, often, as inWarlpiri or Tagalog. The relevant agreement representation, a set of featuresassociated with the S (or IP) node, triggers the introduction of phonologicalmaterial (the clitic(s)) within this domain. The linear position in which theclitics are found is determined by Alignment constraints in the ways developedin chapter 6.

Alternatively, as in Romance, we may find the clitics associated with thefinite verb. The functional content of the clause is shared by its Morphosyntac-tic Representation and that of the Verb node which is its head by virtue of theHead Feature Convention, assuming the relevant features have the status of“Head Features” in the language at issue. In Romance languages, those parts ofthis feature structure relating to non-subject positions are realized within therestricted syntactic domain of the Verb constituent, while the subject materialis realized by inflection on the verb itself. In other languages, such as Geor-gian, the features inherited by the verb are all realized as inflectional propertiesof that word.

If pronominal special clitics are closely comparable to verbal agreement,they ought, like agreement, to come in both “weak” and “strong” flavors, cor-responding to the distinction drawn above between registration and agreementper 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, wherethe presence of the clitic merely registers the properties of the correspondingnominal. Consider the Bulgarian examples in (8.12).

(8.12) a. Decata =ja obicat nejachildren.def her.acc love.3pl her.accThe children love her

b. Na Svetozar =mu xrumna edna misâlto Svetozar him.dat dawned.3sg one thoughtA thought occurred to Svetozar

c. Šte =ti kaža az na tebe koj e= predatelwill you.dat tell.1sg to you who is traitorI’ll tell you who’s a traitor


d. Uvažavat =go edin ucitel zaradi vseotdajnostta =murespect.3pl him.acc one teacher for devotion.def hisA teacher is respected for his devotion

e. Kogo kazvaš =sa =go uvolnili?who.acc say.2sg aux.3pl him.acc firedWho do you say they fired?

f. Na nego =mu =e mâcnoto him him.dat is homesickHe is homesick

As summarized from a variety of authors by Franks and King (2000), thesource from which the sentences in (8.12) are taken, doubling occurs in Bulgar-ian 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 othernon-specific contexts, such as sentence (8.13).

(8.13) Târsjat (*=go) nov ucitelseek.3pl him.acc new teacherThey are looking for a new teacher

The description of this state of affairs is not trivial, but neither are thefacts themselves. We say that Agreement with object nominals (which will ofcourse be realized by the introduction of a clitic rather than directly by ver-bal morphology) is optional, unless the argument is non-referential, as in theintensional reading of (8.13) or an idiom chunk. When it occurs, it is option-ally coindexing, except in the case of impersonal sentences with no nominativesubject such as (8.12f ), where it is obligatory. As in the case of agreement real-ized as verbal morphology, where this agreement is referential (or coindexing),it precludes the presence of an overt nominal in the corresponding argumentposition, and thus the clitic gives the impression of substituting for such anexpression. Where the option of coindexation is not taken, simple registrationof an overt argument phrase results.

A well-known case of a similar sort is supplied by standard Castilian Span-ish, in examples such as those of (8.14).

(8.14) a. *(Lo=) veo a élhim.acc see.1sg prep him(I see him)

b. (*lo=) veo a Juanhim.acc see.1sg prep JohnI see John


c. (*la=) veo la mesaher.acc see.1sg the tableI see the table

d. *(Le=) hablo a élhim.dat speak.1sg to himI speak to him

e. (Le=) hablo a Juanhim.dat speak.1sg to JohnI speak to John

In Spanish, doubling is possible (but not obligatory) with full indirect ob-ject nominals, but not with full direct objects. In contrast, when the object is apronoun, doubling is obligatory in both cases. As Miller and Monachesi (2003:87) describe matters, some dialects, particularly Latin-American, double fullindirect object expressions whether pronominal or not, and have a strong pref-erence for doubling when the direct object is animate and specific. As inBulgarian, the facts are in some ways rather intricate, but the descriptive pa-rameters are clear. We say that where clitics can occur, Agreement (at least inthe sense of feature copying) can apply. Where doubling is impossible, we saythat this agreement must also be coindexing; and where doubling occurs, wesay that coindexation does not.

The kind of complications which we see in these cases are not at all peculiarto clitic constructions. Bresnan (2001: 146-7) notes that verbal agreement andclitics may both be sensitive to the same factors, such as pronouns versus fullnominals, the character of a nominal as definite, specific, animate, etc. We sawabove that the nature of possessor agreement in Finnish depends on whetherthe possessor expression is a human pronoun or not. In general, it is clear thatagreement systems (in the broad sense of this notion) can involve the sameparameters, regardless of whether the results are realized by verbal morphologyor 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 theclitic. 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 Surmi-ran, 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 associatedwith the verb, but there is no doubling: that is, any given argument is


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 nousUrsus has.3sg bring.pp the mushrooms to usUrsus brought the mushrooms to us

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

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

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

Non-subject agreement in Surmiran is thus like French, with one differ-ence. In French, non-subject (“disjunct”) pronouns are not generally availableexcept in special contrastive or other strongly stressed contexts, while in Sur-miran, pronouns are freely usable in non-subject positions. Another differencecomes from the fact that French freely allows combinations of multiple clitics,while combinations such as that found in (8.15d) are tightly restricted in Surmi-ran. In particular, the only acceptable combinations consist of a third-persondirect object clitic followed by a non-third-person indirect object. Substitutingclitics 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 GelgiaWho is giving the bouquet to Gelgia?

b. ?*Tgi igl= la datWho it her gives

c. ?*Tgi l’= igl datWho her it gives

Pronominal special clitics are generally used to index arguments of a pred-icate, but in some languages, clitics appear that do not correspond to anyargument. An example is provided by verbs in French (and many other lan-guages) that require the presence of a reflexive clitic without projecting anycorresponding argument, as in (8.17). These “lexical reflexives” are sometimesknown as “pronominal verbs” in the literature.


(8.17) 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 inAnderson (1991), verbs in some languages may show agreement with morearguments than they have. For instance, in most Algonquian languages, theconjugation of a verb indicates clearly whether it is treated as transitive (i.e.,taking an object) or intransitive; and whether the Theme argument (intran-sitive subject or transitive object) is grammatically Animate or Inanimate. Insome instances, however, verbs are conjugated in a way that indicates the pres-ence of an argument that they do not project. The Menomini examples in (8.18)all involve syntactically intransitive verbs whose inflection includes “agree-ment” 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 Inanimateinflection; plural object)

c. mo;hkanam ‘he uncovers it’ or ‘it (a heavenly body) rises’ (Transi-tive Inanimate inflection)

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

(8.19) can pitk@me cik@niJohn pack.ai.indic.3 apple.anim.obv.plJohn packed the apples

The complex inflectional patterns of Georgian include a number ofdeviations between syntactic and morphological argument structure. Someof these are illustrated by the sentences in (8.20), all of which include inflec-tion (underlined) for an extra argument which is not present in the syntacticrepresentation.

(8.20) a. kari uberavs ‘the wind blows’ (inflected for 3sg io)

b. sonat.a dauk. ari (cem-tvis) ‘you played a sonata (for me)’ (verbinflected for 3sg io)

c. mjinavs ‘I’m sleeping’ (3sg do)

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


case, is inherited by the including phrase) is not necessarily isomorphic to itssubcategorization or argument structure. In such cases, the exact collection offeatures that appears in the Morphosyntactic Representations triggering mor-phological Agreement and/or the introduction of special clitics is motivatedonly in part by the syntactic representation. Such a situation has sometimesbeen asserted to be impossible by those working in heavily syntactic theoriesof morphology, such as Distributed Morphology, but in fact it is not uncom-mon in the languages of the world.

Sometimes a clitic results from something other than the argument struc-ture 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 chicaJuan me her ruined the life to that girlJuan ruined that girl’s life on me

These datives do not appear to correspond to any argument positionpresent in the syntactic representation, and are apparently limited to first andsecond person. One possible account would make appeal to a “derivational”rule (along lines hinted at in the discussion of Tagalog in section 6.4) whichmaps the Morphosyntactic Representation of the clause onto a new one withan extra agreement layer corresponding to the ethical dative while also addingthe semantic content corresponding to “on me/us/you” to the meaning of thesentence. The extra Morphosyntactic content then triggers the introductionof 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 seemspromising.

8.3 Clitic Climbing

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

(8.22) Mario lo= vuole leggereMario it wants to.readMario wants to read it

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


associated with a syntactic reanalysis of the embedded structure that is trig-gered by the matrix verbs in question. While their complements are introducedwith the structure of a clause (including some sort of CP layer), the result ofthis Restructuring (as the reanalysis has come to be called, following Rizzi) isthat the matrix and embedded verbs are no longer separated by a clause bound-ary 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 =loMario wants to.read itMario wants to read it

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

(8.24) [VP[V

vuole] [VP[V

leggere] pro3sg.m]]

What we want in this case is for the clitic lo that specifies the phonologi-cally null proi that is the object of leggere to be introduced within the domainof the finite verb vuole rather than in the domain of the embedded verb. Let usassume that Agreement operates as expected, adding content to theMorphosyntactic Representation of [

Vleggere] in agreement with the object

argument proi. This content, by the Head Feature Convention, is inherited bythe VP that this verb heads.

Of course, since Italian has no rule introducing pronominal clitics at theVP level (but only within V), the presence of this material at this level of struc-ture has no direct effect. However, we can then say that features characteristicof an embedded VP are also inherited by an including VP. This requires anotion akin to that of Grimshaw’s (2000) “Extended Projection” to character-ize the scope of the relation involved here, but the intuition seems clear. Suchinheritance will be blocked by the intervening CP structure when Restructuringhas not taken place. Given that the agreement material thus characterizes thematrix VP, it is naturally inherited (again, via the identity required by the HeadFeature Convention) by the verb that projects that phrase—in this instancevuole, where it can trigger the introduction of a pronominal special clitic.

Where more than one clitic is present, either all of them “climb” or nonedoes, as illustrated in (8.25).

(8.25) a. Mario vuole dar =glie =loMario wants to.give him itMario wants to give it to him


b. Mario glie= lo= vuole dareMario him it wants to.giveMario 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 cli-tics in association with a single verb. Recall that in Surmiran, sequences oftwo third-person clitics are not allowed: one or both of the arguments of averb like dar ‘give’ must be represented by a full pronoun in argument pos-ition, since only one can be represented only by a clitic. Clitic Climbing inSurmiran is obligatory with causatives (formed with lascher ‘let’ and far ‘make,do’). According to the normative grammar (Signorell, Wuethrich-Grisch, andSimeon 1987), it is not possible with “modal” matrix verbs like leir ‘want’, butmany speakers accept sentences with a clitic associated either with the matrixor with the embedded infinitive with such verbs, as well as with the past-tenseauxiliary, as illustrated in (8.26).

(8.26) a. Nous lagn la= tarmetter dumangwe want.1pl it.fem to.send tomorrowWe want to send it tomorrow

b. (?) Nous la= lagn tarmetter dumang

c. Te n’=ast betg igl= cumproyou.sg neg-have.2sg not it boughtYou 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 excludedto have both together. Some of the possibilities are laid out in (8.27).

(8.27) a. Ia vi dar el ad ellaI want.1sg to.give it.masc to herI 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


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 thatwould result from applying Agreement to two separate third-person argumentsare ill-formed, regardless of where one might try to realize the agreementmaterial as clitics.

On the analysis proposed here, pronominal special clitics are a kind ofphrase-level agreement phenomenon. If the functional content that triggersthe introduction of such clitics can “climb” in the way we have seen, we wouldexpect that the same might be true of functional content triggering word-levelagreement in the form of verbal inflection. And in fact, when we look for sucha parallel, we find it. Bobaljik and Wurmbrand (2002) show that the Siberianlanguage Itelmen has a restructuring construction comparable to that of Ital-ian, and that when Restructuring applies, agreement material associated withan embedded verb can show up on the matrix verb. This construction, whichBobaljik and Wurmbrand refer to as “Long Distance Agreement (LDA),” isillustrated in (8.28).

(8.28) a. t’-utu-z-in @lcqu-aì-iì1sg.sbj-unable-pres-2sg.obj see-fut-infI am unable to see you

b. na @ntxa-Bum=nin kma jeBna-she forget-1sg.obj-3 me meet-infHe forgot to meet me

While the Itelmen facts appear to provide a rather close analogy to thebetter-known construction in Italian, it must be noted that other instancesof apparent “LDA” do not yield to the same analysis. Polinsky and Potsdam(2001) for example discuss a similar construction in the North-East Caucasianlanguage Tsez, and show in detail that it is not based on Restructuring. Thisconstruction has no clear analog in the grammar of clitics, and I will not dis-cuss it here.

Another construction which should probably be mentioned here is thewell-known phenomenon of Complementizer Agreement found (at least) insome dialects of Dutch and Flemish and in some German dialects as well (Hae-geman 1992; Hoekstra and Smits 1998). The examples in (8.29) from the Lim-burg dialect of Dutch are illustrative.


(8.29) a. veurtot-s tiech te bruk zuu-sbefore-2sg you.sg the bridge see-2sgbefore 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 isinherited by all of the heads in the verb’s “extended projection,” up to the levelof a complementizer head of CP. Morphological agreement can then take placenot only on the verb itself, but also on the complementizer. The propertiesof this construction have been the basis of an extensive literature; much morediscussion is required to incorporate the results of that literature and resolve itsoutstanding issues, but the line of attack suggested here does appear to providea basis for that work.

Obviously, there is much more to be said about Clitic Climbing, and thepresent remarks do not purport to constitute a full theory of the phenomenon.An extensive survey of a broader class of these “Complex Predicate” construc-tions in the principal Romance languages is provided by Abeillé and Godard(2003), and I have only touched the surface of the intricate array of propertiesthey discuss. Nonetheless, it seems clear that these facts do not in themselvespose special problems with regard to the overall framework proposed here forthe grammar of pronominal special clitics. Indeed, that framework appears tooffer new possibilities for unifying the analysis of phenomena that have previ-ously been treated in isolation from one another.

8.4 Subject Clitics

Most of the attention in the literature dealing with pronominal special clitics isdevoted to clitics that correspond to (direct and/or indirect) objects, but theseare not the only pronominals that a theory must account for. In at least somelanguages we also find clitics that correspond to subjects, and these present is-sues of their own. I refer here not simply to the phonologically weak pronounsthat may occupy subject position, as in English, but rather to special clitics in-troduced in association with the verb (or perhaps elsewhere), which may (andin 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 out-side the VP from which it C-commands the VP-internal subject position),special clitics may be introduced post-verbally in agreement with the subject.


These do not necessarily replace, but may rather double the subject nomi-nal. Similar subject clitics are found (under slightly different conditions) inEngadine Rumantsch (Puter and Vallader); their occurrence in Surselvan, thebest-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 currentusage. In some languages of northern Italy (Poletto 1993, 2000) and Franco-Provençal (Roberts 1993a), as well, special clitics associated with subjectposition have been discussed, and we will turn briefly to those facts after asummary of the Surmiran data and some reasons for calling the elements inquestion clitics rather than verbal agreement markers.

The full set of Surmiran subject clitics was given in Figure 7.4 in section 7.5of the preceding chapter. Recall that these include overt markers in all casesexcept the second-person plural. But why should we call these elements (spe-cial) 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 phonologi-cal) properties of their hosts, and they are attached “outside of” clearly affixalmaterial. 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 Romancelanguages (including Surmiran), or do they simply register the person, num-ber, and gender of the subject, like the pre-auxiliary subject clitics of Franco-Provençal Valdôtain (Roberts 1993a)?

The post-verbal elements in Figure 7.4 cannot simply be morphology reg-istering the properties of the subject, like agreement markers. One piece ofevidence in support of this is the fact that they play an essential role in sanc-tioning the omission of a pronominal subject. As we have seen, (cf. exam-ples (7.47) and (7.48) in Chapter 7), subjects cannot in general be omitted inSurmiran, despite the presence of relatively “rich” agreement on the verb. Thisshows 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 notrequire) the subject to be omitted. The agreement that gives rise to the subjectclitics, that is, involves optional coindexing with the subject position; and whenthis occurs, phonologically null pro occupies that position and excludes anovert nominal. At a minimum, then, the operation of Agreement which givesrise to verbal morphology is distinct from the Agreement that gives rise to thesubject clitics.

The referential nature of the clitics is also supported by the fact that theycannot appear in association with non-referential subjects, such as navot


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

(8.30) a. Nign n’am ò(*’l/*’la) anvidono one neg=me has-3sg(=3sg-m/f) invitedNo one invited me

b. Ossa n’am ò(*’l/*’la) nign anvidotoday neg=me has-3sg(=3sg-m/f) no one invitedToday no one invited me

Another difference between the clitic elements in Figure 7.4 and verbalagreement is a phonological one, which shows up in the second-person sin-gular form. The regular second-person singular desinences in all tense formsin Surmiran end in -s (e.g., cantas ‘you (sg) sing’). A very general rule of Sur-miran phonology replaces /s/ by [S] before all consonants except [l]. We wouldtherefore 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 pro-nounced [kánt@st], with [s] and not [S]. In this respect, the sequence of 2sgending followed by =t is comparable to what we find in compound forms like(las) clavs-tgesa [kláfstCèz@] ‘(the) house-keys’, not *[kláfStCèz@]. Evidently, thepre-consonantal s→S process (whether this is expressed as a rule or as someset of constraint(s) that require this replacement) is an aspect of the lexicalphonology which is no longer active at the point post-verbal clitics are intro-duced (or when compounds are formed). These facts suggest that the elementsin Figure 7.4 are added post-lexically, as would be expected of clitics (thoughnot of inflectional agreement markers).

I conclude, then, that the elements in Figure 7.4 are clitics and not ver-bal morphology. They are introduced optionally (and post-lexically) in con-structions that call for subject–verb inversion, and when present, sanction the(otherwise ungrammatical) omission of pronominal subjects. Such omissionis not obligatory, however, and the clitics can be doubled by an overt nominalor pronominal subject phrase, corresponding to the optionality of the coin-dexation aspect of this distinct Agreement process.

The great bulk of work on clitics linked specifically to subjects (as opposedto subject clitics forming part of a much more general system, as in Tagalogor Warlpiri) has focused on the languages of Northern Italy, including boththose considered “dialects” of Italian and others, such as Friulian and Franco-Provençal. Poletto (2000) provides many useful references and discusses a widerange of data from these systems, which differ in many intricate details from


one another. It is quite impossible to cover anything like the whole of thatground here, and in any event the differences between the theoretical assump-tions of most of this literature and those of the present book would requireeven more extensive discussion. All I can realistically hope to accomplish is arapid sketch of the ways in which some of these facts might be expected to fitinto the present framework.

Poletto (2000) distinguishes four types of subject clitic. She argues thateach of these is associated with a different functional category in clause struc-ture, an account which is at odds with the assumptions of this book and thusnot directly available here. Fortunately, that is not a necessary interpretationof 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 markerthat is the same regardless of the person and number of the subject. For in-stance, the Swiss Lombard dialect of Lugano has an initial particle a whichoccurs with all subjects as in (8.31).

(8.31) a. A= vegni micl come.1sg II come

b. A= ta= vegnat ticl 2sg come.2sg you.sgYou (sg) come

c. A= vegn luucl come.3 heHe comes

d. A= vegnumcl come.1plWe come

e. A= vegnufcl come.2plYou (pl) come

f. A= vegn lurcl come.3 theyThey 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 notedfirst by Benincá (1983), invariable “subject clitics” are found in sentences thatconvey new information or in exclamative contexts. Since there is no apparent


morphosyntactic property of a clause corresponding to this characterization, itappears that they should be treated along the lines of the Tagalog “particle cli-tics” discussed in section 6.4. I will therefore assume that they are introducedderivationally at the level of CP, contributing their phonological content to-gether with the semantic/pragmatic interpretive content identified by Benincá.

A second type of “subject clitic” is illustrated by the Friulian dialect ofSan Michele al Tagliamento. As illustrated in (8.32), an initial element in thislanguage appears as i when the subject is first or second person, but as a withthird-person subjects.

(8.32) a. I= mangicl eat.1sgI eat

b. I= ti= mangiscl 2sg eat.2sgYou (sg) eat

c. A= l= mangiacl 3sg.m eat.3sgHe eats

d. I= mangincl eat.1plWe eat

e. I= mangècl eat.2plYou (pl) eat

f. A= mangincl eat.3plThey eat

Poletto calls these “deictic clitics.” For reasons to be discussed below, I willtreat them as special clitics introduced in the domain of the head of CP, andthus as a type of clitic analogous to the phenomenon of Complementizer Agree-ment mentioned above.

The third and fourth types of subject clitic treated by Poletto involve cliticsattached to the verb that reflect specific person and number combinations inthe subject. Some of the languages she considers have such clitics specificallyfor second-person singular subjects, and some for second- and third-personsingular; these she refers to as “person clitics.” Other languages have such cli-tics for all third-person subjects, distinguishing gender and number, and these


are labeled “number clitics.” For reasons that seem primarily theory-internal,these two types are distinguished and assigned to distinct functional categoriesin Poletto’s (2000) analysis. I will treat them together as “personal subject cli-tics” and consider their central property to be their orientation with respect tothe main verb, as opposed to the head of CP. I assume they are introduced asspecial clitics within the domain of the head of IP.

The northern Italian “subject clitics” thus fall into two major groups, whichwe 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 architec-ture of Poletto’s analysis, though not in detail.

Support for this broad division comes from several sources, as Polettoshows. First, we can note that CP clitics and IP clitics can co-occur. In ex-ample (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 orIP clitics.

The possibility of combining CP and IP clitics follows from the proposalthat agreement material is (potentially) inherited by all of the heads withinthe extended projection of the agreeing verb, and can potentially be realizedin any of several ways: as agreement on the main verb; as a clitic associatedwith this verb or (in the case of clitic-climbing constructions) as agreement ora 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 thedominating CP. Each of these realizations is potentially a rule distinct fromthe 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 tohave multiple instances of the same realization driven by the same agreementmaterial. Thus we find (as in Surmiran) that in clitic-climbing structures theclitic may be realized either on the lower verb or on the higher one, but neveron both simultaneously. The possibility of having “subject clitics” of the CPand IP classes co-occur, then, argues that these are two quite distinct types ofrealization 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 im-possibility 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 notco-occur, though this may in fact be an accidental gap, since the number of languages attestingeither is relatively limited.


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

(8.33) a. Ara ch’=a vegnolook that-cl come.1sgLook, I am coming

b. *Ara che a vegno

c. No so s=a vegnonot know.1sg if-cl come.1sgI do not know whether I will come

d. *No so se a vegno.

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

(8.34) a. Ara che el= vienlook that 3sg.m come.3sgLook, he is coming

b. Ara ch’el vien

Since CP clitics like Loreo invariable a are introduced in association withthe complementizer, it is reasonable that they could trigger allomorphy in thatelement. Personal subject clitics like el, on the other hand, are introduced inassociation with the verb, and their combination with a preceding complemen-tizer element is governed by the phonology. In this case, the reduction of suchvowel 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 mark-ers (in the sense of Zanuttini 1997), while pronominal subject clitics generallyfollow them. This follows if such negation appears within the IP, typically atits left edge. A CP clitic will be outside this structure, and thus precede thenegation, while a clitic introduced within IP in association with the verb willfollow it.

Poletto’s (2000) study and the literature on which it is based explore anumber of additional properties of subject clitics in the languages of northernItaly and adjacent regions, and I cannot address them all. Nevertheless, it seemsclear that the basic distinctions to be made in that work are natural ones withinthe present framework, just as I have argued above for other aspects of thebehavior of pronominal special clitics.

9

Clause Structure and the Grammarof Incorporation

The topic of Noun Incorporation may appear to have little to do with the gram-mar of clitics.1 Nevertheless, the choice between syntactic and lexical analysesof this construction and the framework developed in this book for the de-scription of special clitics both bear on a set of central issues in the theory ofgrammar: the nature and scope of “Head Movement” and the role of abstractfunctional categories in clause structure.

I assume that the functional categorial structure of clauses is re-presented by the features assigned to a limited set of fairly traditional syntacticcategories—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 syntacticliterature since the late 1980s, however, presumes that a much wider array ofcategories, functional rather than lexical in their basis, define structural con-stitutent types that stand in a hierarchical relation to one another.

If special clitics are not structurally autonomous elements in syntactic rep-resentation but instead the overt markers corresponding to the (possibly quitecomplex) feature structure of syntactic categories, that suggests that the num-ber of such categories that are fundamentally distinct from one another maybe quite limited. If this formal possibility exists for the overt expression offunctional content, then simply identifying some component of functional or-ganization does not automatically require us to posit new layers of projectingstructure. Instead we might assume, in the absence of evidence to the contrary,that such content corresponds to a feature associated with a level of structurewhose internal complexity can be established independently.

The possibility of analyzing syntactic structure in terms of a vast array offunctional categories, each representing essentially a single property but eachprojecting 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).


cumulate the content of a number of layers of structure. This in turn relieson the generality of the mechanism of Head Movement, the operation whicheventually conflates the presumed functional elements into a more limited de-gree of arboreal complexity in the surface form.

The presumed properties of Head Movement itself, however, are heavilydependent on arguments originally supplied by Baker (1987), founded in turnon the analysis of Incorporation constructions—particularly Noun Incorpora-tion. But if the constructions on which this theory is grounded do not actuallyinvolve syntactic movement at all, then the presumed generality of an oper-ation of Head Movement, and with it the apparatus of functional categorialstructure that depends on it, is much less self-evident. If Noun Incorporationdoes not involve Head Movement, the foundations for that apparatus are lesssecure—or at least different. And if special clitics and related phenomena donot depend on elaborated functional structure, the need for a theory of HeadMovement 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 haveassumed verbs move from the head of VP to the head of IP and/or CP. How-ever, the extension of this kind of displacement to the full range of construc-tions that have been attributed to Head Movement does not follow, and indeeddoubts about this operation have begun to surface among those pursuing theMinimalist Program. The research strategy of the present book converges withthe adoption of a non-syntactic approach to Noun Incorporation in facilitatingthe 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 ofboth syntax and morphology, because it involves the construction of units thatare unquestionably words from material that gives the appearance of havingbeen combined within the syntax. If this impression is indeed correct, thisoperation presents an important prima facie problem for most versions of theLexicalist Hypothesis.

Noun Incorporation is a feature of the grammar of a wide range oflanguages, including (as a semi-random, quite small sample) Mohawk,Chuckchee, Southern Tiwa, Classical Nahuatl, and many others. Grossly, itinvolves 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 anincorporated noun corresponds to the direct object of a transitive verb, as inthe examples in (9.1) below from Southern Paiute (Sapir 1911: 263).

The Grammar of Incorporation 259

(9.1) a. qam;Ú- yaai- nÚm- pÚGa‘jackrabbit- hunt- usitative- remote past

He used to hunt jackrabbits

b. cû´q˙uc˙u qam;Ú- vaxqa- q;a‘one jackrabbit- kill- comp (ss)Having killed one jackrabbit,. . .

In some languages, the underlying object (derived subject) of an unac-cusative Verb can incorporate; in other languages, unergative subjects canincorporate as well. These possibilities are illustrated in the Chuckchee ex-amples in (9.2) below taken from Polinsky (1990).2

(9.2) a. Neyk-@k P@l-@-lg-@-gPihill-loc snow-evid-melt-evid-aor3sgOn the hill, the snow melted

b. @tlon Ninqe-et-@-lPet-gPehe child-intens-evid-come-aor3sgHe 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-kocillo-tetePki panci3sg 3sg.it-knife-cut breadHe cut the bread with a knife

b. Chukchee (Skorik 1948, apud Spencer 1995):g@tg-@lq@t-gPe walw@[email protected] raven.absRaven went to the lake

Noun Incorporation has long been an important subject of discussion inthe theory of grammar. In the early years of the twentieth century, it wasasserted by some to constitute a particularly characteristic feature of NorthAmerican languages. Kroeber (1909), in a somewhat extreme attempt to rebutthat 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 con-struction across a variety of languages.

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


There are two basic views of the nature of Noun Incorporation, and bothhave long histories. One approach, which I will refer to as the “syntactic”analysis, treats Noun Incorporation as a syntactic process by which an ar-gument of the verb (or at least part of that argument) is actually displacedfrom 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 theincorporated stem comes to fill the semantic (T) role of a corresponding unin-corporated noun. Proponents of this view in the generative literature includeSadock (1980, 1986), who argued, from the apparent need to form words inthe syntax by moving a noun into the same word as a verb, that the Lexicalistprohibition against syntactic manipulation of the internal form of words mustbe 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 thisfrom Noun Incorporation. From the morphologist’s point of view, classi-cal Noun Incorporation involves combining two stems, each of which is (inprinciple) a distinct lexical entry, and by definition, this is an instance ofcompounding. Denominal verb formation in languages like those of theEskimo-Aleut or the Wakashan families, in contrast, involves the addition ofone of a large inventory of bound affixes to a stem. Morphologically, this is akind of derivation, not compounding, because the affixes involved never occurindependently as stems; and that is what causes authors like Mithun and Gerdtsto 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 thepossibility that bound morphological material can be syntactically autono-mous. Regardless of definitional issues, the analysis proposed by Sadock forWest Greenlandic certainly involves Head Movement, and so it is relevant tomy concerns here.

A conceptually similar (but technically very different) position is devel-oped by Mark Baker, who has explored it at some length in one book (Baker1987) and made considerable use of it in another (Baker 1995). Baker also main-tains that noun heads move in the syntax to take up their incorporated surfacepositions inside verbs. But Baker’s theory of the kind of movement involvedgoes much farther, and in fact is the basis of the rather more general notionof Head Movement in syntax, since he extends the operation founded on theproperties of Noun Incorporation to many other sorts of putative movementof heads.


The alternative to these syntactic movement analyses is the “lexical”account of Sapir (1911), Mithun (1984), Anderson (1992), and others. On thispicture, the noun-plus-verb combinations we refer to as instances of “NounIncorporation” are built in the lexicon, not in the syntax.

The formation of such [V[N

stem1] stem2] combinations, with [N

stem1]interpreted as a specific argument of [

Vstem2], is surely within the scope of

the type of regularity found in the lexicon. Something quite similar is surelyinvolved in the case of “synthetic” compounds: duck hunting ([

N[N

duck][N[V

hunt]ing]]) is built from [N

duck] and [V

hunt] and has an interpre-

tation in which the noun is taken as specifying one of the arguments of theassociated verb.

Some have argued for a syntactic incorporation analysis of such com-pounds, but the appeal of such a view is limited. Compounds are evidentlystructurally parallel to non-compound members of basic lexical categories,and are typically opaque to syntactic processes. The possibility of buildingall lexical compounds in the syntax was tried out in the early days of genera-tive grammar (Lees 1960), but the context for that proposal was one in whichno serious alternative existed: at the time, there really was no theory of thelexicon. Most authors today, though, agree that the formation of compoundstakes place in the lexicon and not in the syntax.

Compounds like duck hunting involve not only a relation between thenoun and the argument structure of the verb, but one which is quite like whatwe find in Noun Incorporation constructions. In both cases the noun mosttypically corresponds to an argument of the verb that would fill the syntac-tic 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 nounapparently corresponds to the subject of an intransitive verb.

In these examples, we might invoke the Unaccusative hypothesis and saythat the argument in question represents an underlying direct object. Suchan account is less plausible, though, for examples such as crybaby, flashlight,workman, playboy, etc., that seem to involve a noun specifying the agentiveargument of an unergative verb. Such verbs, on at least one understanding,are exactly those that cumulate the agent and theme roles and associate bothwith 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 associ-ated 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 theseverb-argument(/adjunct) relations can be established by a lexical rule, as they


must be if we are to deal with true compounds, the initial motivation for asyntactic account of Noun Incorporation becomes much less persuasive.

As with the syntactic analysis, the lexical one has some apparent primafacie advantages. One argument that seems to point to a lexical analysis isthe fact that the shape of incorporated nominal elements often differs phono-logically from a corresponding free form in unpredictable ways. In some in-stances, such variation is limited to a small number of suppletive forms: e.g.,in Mohawk (Mithun 1984: 876) “-nahskw-‘domestic animal’ appears only in-corporated, while a semantically equivalent stem -tshenv appears only as anindependent N[oun].” In other languages, however, this phenomenon maybe quite pervasive. Thus, in the Australian language Tiwi (Osborne 1974: 48),“[i]ncorporated forms and free forms are generally not cognate, [. . . ] as thefree forms which were cognate with the existing incorporated forms have longsince disappeared, owing to the rapid rate of lexical replacement in Australianlanguages.” Some representative pairs of incorporated and free forms drawnfrom Osborne (1974) are given in (9.4). In some cases, a degree of phono-logical resemblance can be seen (especially in some body part names), but inothers there is no apparent similarity. Also, in some cases a single free formcorresponds to more than one incorporated form, while in others the reverseis true.

(9.4) incorporated free glosswuôati- -wuôatiNa foreheaduNintami- -j@r@Nintamuôa noseaw@ri- -wuôara bellyaw@ri- t@muripa navelmaratiNa t”@raka (live) wallabyNiliwan”t”@(N@) t”@raka (dead) wallabypaN@laônti- -t”araNa buffalo

A language may show semi-systematic relations between free and incor-porated forms that do not reflect phonological regularities found elsewhere inthe language. A particularly interesting instance of this is found in the Mundalanguage 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 regardedas derived from verbal or other roots by the addition of prefixes, suffixes, or in-fixes. 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 formof such a noun can be formed by simply removing such added morphologi-cal material: thus, the combining form of k-@n-oN ‘razor’ is -koN, and that ofg etasi ‘play’ is -g esi (cf. the verb g esi-‘to play’ from which the noun is derived


by infixation). This relation has been massively overgeneralized and inverted,however. A huge number of polysyllabic free-form nouns have incorporatedforms produced by removing non-existent prefixes, suffixes, and infixed -VC-sequences so as to reduce the form to a canonical -CVC shape. For details, seeZide (1976); the point is simply that the elements appearing in incorporationstructures are quite distinct from semantically similar free forms in ways thatdo not follow from the independent phonology of the language, but ratherpoint to lexical processes.

Just as the phonological form of incorporated and non-incorporated ele-ments may require specifically lexical description, the semantics of an incor-porated nominal construction may also differ, again unpredictably, from theinterpretation of a corresponding construction in which the nominal is repre-sented by a full DP in argument position. Some Mohawk examples taken fromBaker (1995) are given in (9.5), where “#” in (9.5b) means that the sentence isanomalous 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-puncShe started her journey (lit.: picked up the road)

b. #tu-s-a-yú-([a]te)-hkw-e’ ne oháhadup-iter-fact-f.sg.subj-srfl-pick.up-punc art roadShe picked up the road (literal reading only)

Further examples from (Huauhtla) Nahuatl (Merlan 1976) are givenin (9.6); essentially every language with extensive Noun Incorporation providesinstances of this sort of non-compositional interpretation of the construction.

(9.6) a. i. ∅-nec-maka-∅-k paPtli3sg-1sg-give-pst-sg medicineHe gave me medicine

ii. ∅-nec-paP-maka-∅-k3sg-1sg-medicine-give-pst-sgHe 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 illustratedare to be expected if Noun Incorporation is simply a syntactic process that


may (or may not) apply to a fully articulated syntactic structure to move someof its lexical material from one position in the tree to another. Such argumentshave not received much attention in the syntactic literature, but this does notmean they are negligible: syntacticians tend to assume that both the phonol-ogy and the lexical semantics will take care of themselves in the “interpretive”components of the grammar, and the difficulties presented by actually work-ing this out often carry less weight than they should. In fact, though, theseaspects of Noun Incorporation structures point away from the syntax in theiranalysis, and toward the part of the grammar which is classically the locus ofitem-specific information, the lexicon.

Sapir (1911) was probably the first to propose that Noun Incorporationconstructions are actually instances of lexical compounding. Since the possi-bility 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 con-sideration that incorporated structures involve a combination of two stems,like compounds, together with phonological and semantic idiosyncrasies ofthe sort adduced above.

9.2 Fleshing Out the Syntactic and Lexical Accounts

Noun Incorporation, then, is a construction type with some specific proper-ties: the combination in a single word of verbal and nominal components,where the nominal generally supplies information about one of a small num-ber of thematic possibilities in the argument frame of the verb; and where thewhole functions as a verb. Its analysis can be approached from either of twoperspectives (syntactic or lexical) each of which seems to have some naturalconceptual affinities with the construction’s basic character. The question tobe addressed is which of these (if either; or perhaps some combination of thetwo) is correct.

It is quite important for syntactic analyses based extensively on HeadMovement operations that syntactic Noun Incorporation be at least possibleand common, even if other sorts of “Noun Incorporation” constructions existas well. For those who maintain that “the syntax neither manipulates nor hasaccess 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 isquite important not to allow syntactic rules to put words together in this way.The choice of analyses for Noun Incorporation constructions thus has a gooddeal 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 whichNoun Incorporation is the most obvious empirical domain) is central to the


articulated-infl account of clause structure common to much current syntax,so any challenge to the syntactic nature of Noun Incorporation undercuts thisview, at least indirectly.

A choice between the two accounts does not seem to follow from the basicdescriptive properties of Noun Incorporation. Both views are able to describethe fact that the noun part supplies content associated with a thematic positionin the argument structure of the verb part. As far as the limitations on whichpositions can be involved, the Lexicalist view accommodates these thematicrestrictions rather straightforwardly: lexical rules often refer to the relationof theme, and compounding in particular does so. If Noun Incorporationis simply a form of noun–verb compounding, this is exactly what we wouldexpect. As we have seen, the incorporated noun is not always a theme: it issometimes a locative or an instrumental, as in (9.3). But this is again quiteparallel to the facts about compounds, and supports the view that there is asingle regularity at work in the two cases.

The syntactic view derives this result in a way that is, at a minimum, ratherless direct: some might find Baker’s (1995: sect. 7.3) discussion of why only thedirect-object position is accessible somewhat tortuous, but for the sake of dis-cussion let us assume that such an account is at least possible. Similarly, I willassume that the phonological and semantic idiosyncrasies referred to abovecan find a home within the syntactic account: idioms, for instance, providea clear precedent for the assignment of non-compositional interpretations tosyntactically complex structures, and it is at least possible that suppletion andother non-phonological variation occurs as a function of syntactically derivedenvironments. If we are to choose between the two theories of Noun Incorpo-ration under consideration, it will probably be on the basis of detailed study,not simply as a consequence of the basic descriptive regularities that define theconstruction.

Mohawk Noun Incorporation

Baker proposes to do just this. His primary evidence that Noun Incorporationis syntactic comes from Mohawk, for which an analysis is developed at lengthin Baker (1995). On this account, a lexical noun can be generated (as the ex-haustive content of a DP) in argument position, and then moved to adjoin tothe governing verb. The question naturally arises of why this movement shouldtake place in Mohawk (but not in English).

I have already introduced some aspects of Baker’s analysis of Mohawkclause structure, which he supports in detail, in section 8.1 above. According tothis account, overt DPs are not licensed in argument positions in Mohawk, butonly appear in adjunct positions, where they serve as appositive expressions


(a view similar to, but not identical with, that of Jelinek 1984). If such a[DP

Noun]were generated in an argument position, it would thus be ill-formed.

In order to get T-marked, such a DP has to be coindexed with an element in theverb (a key subpart of the Morphological Visibility Condition, or MVC, dis-cussed in the preceding chapter). If that element were Agreement, the Agree-ment would (on Baker’s hypothesis) absorb the case-assigning property of theverb, so the overt DP would still be ill-formed. But if the noun moves so as toadjoin to the verb, the MVC is satisfied, the otherwise caseless DP no longerhas phonetic content, and no well-formedness conditions are violated.

The Morphological Visibility Condition is a parametrically determinedcharacteristic of certain languages (called “Polysynthetic” by Baker, in a us-age that deviates somewhat from the traditional sense of this term). We mayask, naturally, which languages have (syntactic) Noun Incorporation of thissort. Baker’s answer is: those that have to. That is, in Mohawk, incorpora-tion 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 Minimalistassumptions, if nothing requires some operation to take place, it is blocked. Asa result, Noun Incorporation is impossible in English, since it is not forced.

Now in fact many languages have constructions that look like NounIncorporation (i.e., cases in which noun plus verb together seem to form asingle word, and where the noun is interpreted as specifying one of the argu-ments of the verb). They differ quite a bit from one another, though. For onething, in most Noun Incorporation languages (but not, e.g., Mohawk), theincorporated noun is always interpreted as indefinite and/or generic. This iscomparable to the interpretation of nouns in (English) lexical compounds, asshown 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 anaphoricreference, since it is necessarily generic (or at least referentially under-determined).

In contrast, in Mohawk, an incorporated noun can refer to something thatis referentially specific or definite, as the short discourse in example (9.8) fromBaker (1995: 288) illustrates.

(9.8) Thet2re’ wa’-ke-nakt-a hnínu-’yesterday fact-1sg.subj-bed-∅-buy-puncI bought a bed yesterday


Í-k-ehr-e’ Uwári 2-ye-núhwe’-ne’∅-is.subj-think-impf Mary fut-f.sg.subj-like-puncI think Mary will like it (the bed)

This is just what would be expected if the incorporated noun representsa DP in an argument position, since such a DP can perfectly well be specificor definite. If the lexical analysis is to remain viable, it must do somethingto accommodate this possibility. Apparently, the verb-plus-noun structurethat constitutes a lexically compound (Noun Incorporating) verb must be ableto take an argument in the thematic position that is (also) specified by theincorporated noun stem, at least to the extent of assigning reference to thatargument.

The Lexical Analysis

The most common view to be found in the literature assumes that when anoun is compounded with a verb, the noun satisfies (or saturates) the corre-sponding argument in the verb’s argument structure, which is also interpretedas generic or indefinite. Assuming that nouns themselves have an externalT-role to discharge (the “R-role,” connected with the noun’s possibility of re-ferring), this generic interpretation corresponds to a certain sort of binding.The noun’s own external T-role (R), along with the corresponding variable inthe interpretation of the verb, are jointly bound by a generic or indefinite op-erator Genx. The relevant compounding operation, like many others, intro-duces this generic operator binding the logical variable corresponding to thenoun’s R-role and the verbal T-role identified with it, thus saturating both andprecluding an external referring expression linked to the same argument.

To accommodate the facts of languages where an incorporated noun ispotentially referential, we need to extend this account. Let us say that in somelanguages, at least, Noun–Verb Compounding is an operation that “unifies”the semantics of the noun with the argument position of the verb, but withoutsaturating the argument itself. Explicit semantic analyses of such an operation,including its application to Noun Incorporation, are provided (along some-what different lines) by Bittner (2001) and by Chung and Ladusaw (2004); I willadopt a somewhat less formal account here, similar in the relevant respects tothese. I will refer to the operation unifying the semantics of the noun with theverb’s argument position as “Restrict,” employing the terminology of Chungand Ladusaw.

On this view, in a language of the relevant type the Noun Incorporat-ing verb fish-catch is a transitive verb meaning “X catches Y, Y a fish.” Thisis essentially what Rosen (1989) has called “Classifier Incorporation,” and whichother writers have proposed for at least some incorporation structures

�

�

“Clitics” — 2005/7/26 — 17:39 — page 268 — #278�

�

�

�

�

�


(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. [V[N

fish][V

catch]] = “X catches Y, Y a fish.”

b.fish[—R

fish

]+

catchcatch

[—

Agent

],[ —

Theme] ⇒

fish-catch

catch[

—Agent

],[

—Theme(,R)

fish

]

This verb is still syntactically transitive; and while its theme argument isspecified as something having the properties of a fish, it is not logically or ref-erentially 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 whichincorporated arguments are always necessarily generic, not in having syntac-tic (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 withlexical operations of compounding, the lexical suffixes of Wakashan andSalish 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 theymodify 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 argumentsof the verb, as illustrated in the examples of (9.10) taken from Bergsland (1997).

(9.10) a. hla-kucha-x hila-ku-xboy-little-abs read-fin-3The little boy is reading

b. hila-kucha-ku-xread-little-fin-3He (the little one) is reading

c. tayagu-xsiida-x saga-ku-xman-poor-abs sleep-fin-3The poor guy is sleeping

d. saga-xsiida-ku-x hamasleep-poor-fin-3 dem(up there, invisible)He is sleeping, the poor one


e. kida-kucha-ku-nghelp-little-fin-1sg/3sgI am helping him (the little one) or I (the little one) am helpinghim

When Aleut -kucha, -xsiida, etc., are added to nouns, they are semanticallyintegrated with the meaning of the base noun by means of Restrict operatingon the noun’s R T-role. When added to verbs, they are semantically integratedin the same way, with Restrict operating on some one of the T-roles in the verb’sargument structure.

Kwakw’ala similarly has lexical operations that result in combiningthe 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 Bach1998).

(9.11) a. la-i hoqw@w@ls-ida q’isq’asde-x.a k’@lx-i q’isinaaux-they go out-dem eat currants-obj raw-dem currantsthose who have been eating the raw currants go out

b. la-i k’@lxk’[email protected]’-ida q’isinaaux-they eat raw-pass-really-also-dem currantsraw currants are also eaten

c. k’@lxk’axa-Pax.a-ida bak’[email protected] t’@mxwalieat raw-also-dem Indians-obj gooseberriesThe 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 reduplicatedsequences are underlined, converts a nominal stem into a verb meaning ‘eatstem’ and it is this derivational process that contributes the argument frame,where the base noun or adjective provides a semantic interpretation to be com-bined (via Restrict) with the theme argument position.

Notice that both in Aleut and in Kwakw’ala, the Head Movement analysisof the apparent “Incorporation” is not available, since the element that wouldhave 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 Movementviolation. At least on Baker’s account, that would seem to entail the conclusionthat some sort of lexical operation is involved rather than Head Movement inthe syntax.

The proposal that Noun Incorporation involves a lexical operation(Restrict) that unifies the semantics (and perhaps other properties, as we willsee below) of a lexical base with those of a thematically characterized position


in the argument structure of another base is quite close in spirit to the analysesoffered by Malouf (1999) and Bittner (2001) for Greenlandic. While Malouf ’sanalysis is presented within the framework of HPSG, and Bittner’s is in termsof Dynamic Semantics, their essential claims are quite parallel (mutatis mutan-dis) to those of the present account.

“Doubling” of Incorporated Nouns

Another typological difference among languages with respect to Noun Incor-poration derives from the fact that in some languages an incorporated nominalcan be doubled by an external phrase whose content also specifies the corre-sponding argument position. There are two subcases to consider, dependingon 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 (Myfather [eight bullheads] fish-bought).

The first of these subcases is straightforward for the Head-Movement ac-count: 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 stillpresent 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 prob-lem. On that view, the doubling nominal is actually an adjunct, related to aseparate (phonetically null) DP in argument position, not only in the presenceof an incorporated noun but for all overt DPs. Figure 9.1 shows how this worksfor the relevant parts of a Mohawk sentence taken from Mithun (1984: 870).Dotted lines here indicate anaphoric links, and the arrow indicates a move-ment relation.

The lexical analysis of constructions in which an incorporated noun stemis doubled by all or part of a nominal phrase in argument position is simi-lar in part to the syntactic account. When an incorporated noun is doubled,the analysis is essentially the same as Baker’s. The clause-internal argumentposition is filled by a phonologically null pro, and the overt DP appears asan adjunct and forms a chain with that pro. The only structural difference isthat the empty category in the argument position is an instance of pro ratherthan of trace, since it does not result from displacement. Semantically, the in-corporated noun unifies with an argument of the verb (by Restrict), which issaturated by its link to the clause-internal argument (pro, forming a chain withthe external adjunct expression).

Notice that the presence of the empty pro in argument position in Mohawkfollows not from anything about Noun Incorporation, but rather from Baker’s


IP

IP DP

DP IP

DP VP

DP V

N N V

8 bullheadsj proi [e]j fishj- bought my fatheri

Tohka niyohserá:ke tsi nahéseveral so-it-year-numbers so it-goessha’té:ku nikú:ti rabahbót wahu-tsy-ahní:nu ki rake’níhaeight of-them bullhead he-fish-bought this my-father[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 clauseas adjuncts. In the absence of evidence for that, we could simply say that theovert nominal phrase is in the expected argument position. Since the incorpo-rated noun did not originate there, there is no syntactic reason why some otherexpression should not fill this position.

In the case of the apparently headless DPs, where it looks as if Head Move-ment has taken place, we can note that the expressions themselves are well-formed 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 bymovement of its head out of the DP to adjoin to a governing verb, but ratherby the selection of pro as head of the NP within the DP, with the semanticsof the nominal being supplied from context (or verb-internally via Restrict, ina case like (9.9b)). On this view the headed and headless cases of doublingexpressions fall together, at least in principle.

With this apparatus in place, we can approach a typology of referentialityand “doubling” within the class of Noun Incorporation languages. As we shall

�

�

“Clitics” — 2005/7/26 — 17:39 — page 272 — #282�

�

�

�

�

�


see, this is in part similar to what we saw about “doubling” with agreement andwith pronominal clitics in Chapter 8, and in part different.

Three major classes of construction must be distinguished. The first ofthese is the case in which the noun is necessarily interpreted as indefinite, non-specific, or generic, and where no doubling expression can be present. Here Iassume that the semantic side of the compounding operation which forms the“noun incorporating” verb identifies the semantic variables corresponding tothe noun’s external R T-role and the appropriate verbal argument. It also intro-duces an appropriate logical operator binding this combined variable, result-ing in a semantics like (9.12). The syntactic subcategorization of the resultingverb is reduced by the elimination of the argument position corresponding tothis variable.

(9.12) Genx (catch[

—Agent

],

[x

Themefish

])

In some languages, the result is that otherwise transitive verb stems sub-categorize as intransitives as a result of incorporation. In others, the mor-phological properties of the derived verb may remain transitive, althoughno additional argument corresponding to the “incorporated” element can bepresent, since it could not be bound to the already saturated argumentposition. I assume that this is an instance of the formal dissociation of mor-phological and syntactic transitivity, as discussed in Anderson (1991).

The second case is that in which the object of a noun-incorporating verbcan have independent (definite and/or non-generic) reference, and where thisargument can be represented by an overt DP which “doubles” the incorpo-rated noun. This type (“Classifier Incorporation”) is similar to the first. Thesemantics of forming a noun-incorporating compound in such a language donot involve the introduction of a logical operator binding (and saturating)the relevant argument, but only the identification of the incorporated noun’sR role with that of the verbal argument. As a result, a corresponding seman-tic variable is (still) available for external specification by an expression in theargument position. Of course, since the semantics of one argument of the verbhave been Restricted by the interpretation of the incorporated noun stem, theexternally specified expression must be consistent with that: thus, the informa-tion provided by the object of (9.9b) must (at minimum) be consistent with theproperty 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 doublingexpression can be present, as in, e.g., Nahuatl or Southern Tiwa. Clearly, thesemantics of the incorporated noun must unify here with a verbal argument


position, but the resulting verb apparently has some property that is incom-patible 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 thatexcludes overt expressions in the argument position corresponding to anincorporated noun in such a language. Baker’s (1995) account of a languagelike Southern Tiwa, where doubling is excluded but non-generic reference inassociation with an incorporated noun is allowed, is that the condition licens-ing adjuncts in association with argument positions only allows them to belicensed 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 corre-spond to pro in argument position, and not to the trace of head movement.

Apparently, overt phrases in argument positions that correspond to an in-corporated noun in such a language have some property that is incompatiblewith the structure of a verb derived by Noun Incorporation. Let us supposethat Noun Incorporation in a language like Mohawk (which permits externaldoubling phrases) not only unifies the semantics of the noun with that of thecorresponding argument within the semantics of the Verb, but also eliminatesthe Noun’s external R T-role (or merges it with the T-role assigned to that argu-ment). In Southern Tiwa, on the other hand, the incorporated noun’s R T-roleis identified (but not merged) with the T-role of the argument. That meansthat if the argument position contains an expression headed by a noun thatassigns its own R T-role, there is a T-criterion violation. Only a null pro, whichhas no R role to assign, can occupy such a position.

Another aspect of the typology of noun-incorporating languages requiresfurther investigation, but the outlines of a solution are reasonably clear. Lan-guages that permit doubling may differ in which of the following schematicsentence 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 incorporatednoun stem allow sentences such as (9.13a), where the doubling expressionis more specific and detailed than the incorporated noun alone. Many(though apparently not all) allow sentences like (9.13b), where the two are es-sentially synonymous. None, however, appear to allow sentences like (9.13c),where the doubling expression is actually less specific and detailed than theincorporated noun.


These facts appear to result from a requirement (semantic or pragmatic,depending on one’s view of where the line between these aspects of grammaris to be drawn) that overt expressions be at least minimally informative withrespect 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 aproper 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. Lan-guages apparently differ on the basis of whether they consider an essentiallyequivalent expression “informative” (perhaps by virtue of the possibility it in-troduces of independent referentiality) or not.

One further point should be noted about the properties of doubling ex-pressions. In many cases, these provide information about the nature of thecorresponding argument itself: in a Mohawk sentence like (9.14), for instance,the overt argument expression and the incorporated noun combine to specifya single referent ‘a polka-dotted dress’.

(9.14) Kanekwarúnyu wa’-k-akya’tawi’tsher-ú:niit.dotted.dist past-I-dress-makeI made a polka-dotted dress

In other cases, however, the external NP may supply information about apossessor of the argument in question, as in the Rembarrnga sentence (9.15)cited in Dixon (1980).

(9.15) tiNP paNa-warnta-na-∅woman 1sg/3pl-track-see-pastI saw the women’s track

Note that although it is a single track that is referred to, the agreement isplural, due to the fact that it is the track of more than one woman, even thoughthe DP tiN? ‘women’ is not overtly marked as plural. The overt argumentexpression, 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 ∅]). Agree-ment here is with a third-person plural, showing that ‘women’ is the headof the object DP, and not simply a modifier. In such cases (often referred toas “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 externalargument expression as a specification of the possessor of an (already partlyspecified) argument, and not as restricting the argument itself.


9.3 Noun Incorporation: Syntax or Lexicon?

By and large, up this point, the syntactic and lexical theories of Noun Incor-poration are “tied” in that each can be said to account for roughly the samerange of phenomena. Baker (1995) considers the lexical account (though notwith 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 NounIncorporation of the lexical variety, in addition to the syntactic type. Thismakes the theory rather close to unprovable, since any fact that appears toargue against the syntactic account can be dealt with by saying that in just sucha case, the incorporation is lexical.

But Baker also discusses some phenomena which he feels argue for thesyntactic account over the lexical one. As the syntactic and lexical views havebeen elaborated above, they converge to a great extent as far as the representa-tions they assume. But there is one difference: for Baker, the empty categorypresent in argument positions in association with a verb that has undergonesyntactic Noun Incorporation is a trace, whereas when the formation is lex-ical it is pro. As a result, any way in which these two possibilities could beteased apart could provide a way of discriminating between the theories. Onthis basis, Baker (1995: 314–29) offers three arguments that at least some casesof syntactic Noun Incorporation exist.

The first of these concerns agreement. He argues that in general, thereis no agreement with the position corresponding to the source of Noun In-corporation. 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 in-corporation ought to leave not a pro, but a trace and Baker makes the unusualassumption that this should not produce agreement. Absence of agreementwith noun-incorporating verbs, as opposed to its presence in association withsimple 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 atNoun Incorporation structures and see whether agreement with the incorpo-rated position is present or not. But things are not so simple as that. Languagestend to be devious and evasive at just the points where the analyst really wantsthem to commit themselves, and Mohawk is no exception.

To evaluate this matter, we must bear in mind some limitations onMohawk Noun Incorporation. In fact, it is almost exclusively inanimates thatare incorporated. This is particularly interesting in light of the fact that inMohawk, agreement with an inanimate object is formally indistinguishable

�

�

“Clitics” — 2005/7/26 — 17:39 — page 276 — #286�

�

�

�

�

�


from no agreement at all. That is, when the object is inanimate the agreementmarkers are the same as in intransitive agreement with the subject alone. Incor-poration of animates is generally disfavored, and regarded as pejorative (pos-sibly implying the treatment of a person as an object).

Why should this restriction obtain? Baker admits to having no explanationfor the avoidance of incorporation of animates. One possibility, though, is thatit is a consequence of the fact that it is precisely with inanimate objects that thespeaker can “fudge” the issue of whether agreement is or is not present, similarto the use of modal constructions in English in those cases where no particularagreement seems right.

(9.16) Neither Fred nor I

{??am/??is/??are coming

will come

}to the party.

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 dativeobjects with those taking accusative objects is only felicitous when the objectNP is ambiguous between the two cases.

(9.17) a. Er findet und hilft Frauenhe finds(+ACC) and helps(+DAT) womenHe 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 toinanimates might be described as follows: where the verb contains an incorpo-rated noun referring to an overt argument expression, that seems like “enough”to a speaker, and so the presence of a separate agreement element (as wouldbe required if the object were animate) seems superfluous. Where the transi-tive object is inanimate, however, the agreement morphology is ambiguous asto whether it contains a marker referring to the object or not, so no surfacediscomfort results.

In any event, in most cases of Noun Incorporation there is no overt in-dication of whether agreement is present with the ‘incorporated’ position ornot, because of the formal similarity between Sbj/InanimateObj markers andintransitive Sbj markers. In addition, as Baker notes, in sentences like (9.18)where there is no external ‘doubling’ expression we can always say that thelexical operation of incorporation has constructed an intransitive verb (by sat-urating the argument), so the absence of agreement would follow on either thelexical or the syntactic view.


(9.18) Tu-t-a-yako-kétoht-e’ ts-e-wir-2háwidup-cis-fact-f.sg.obj-appear-punc iter-f.sg.subj-baby-carry/statShe appeared carrying a baby.

In other sentences, though, such as (9.19), there is clearly an animate nounincorporated. 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.

(9.19) Ra-wir-a-núhwe’-s thík2 (owirá’a)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 argu-ment, since (on his analysis) lack of agreement is forced by the fact that therelevant 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 mustbe being treated as an inanimate. Since (animate) agreement is obligatory insuch cases when incorporation has not taken place, Baker concludes that theargument position in a sentence like (9.19) must be filled with a trace ratherthan with (inanimate) pro.

Baker (1995: 335) rejects the lexical analysis, asserting that “if Mohawks canfreely view babies as inanimate entities, then one would expect that they couldtrigger inanimate agreement even when not incorporated [as in (9.20b) below–SRA], contrary to fact.”

(9.20) a. shako-núhwe’-s (ne owirá’a)m.sg.subj/3obj-like-hab art babyHe likes them (babies)

b. *ra-nuhwe’-s (ne owirá’a)m.sg.subj-like-hab art babyHe likes them (babies)

c. ra-wir-a-núhwe’-sm.sg.subj-baby-∅-like-habHe likes babies

This conclusion does not necessarily follow, however. It might well be the casethat the semantic content of the incorporating stem -wir-‘baby’ differs subtlyfrom that of the independent Noun owirá’a, perhaps in being under-specifiedfor animacy. The notion that incorporating stems are less fully specified thanindependent nominals in some respects is entirely consistent with the tendency


across languages for such stems to have a somewhat more generic interpreta-tion than full lexical nouns.

It seems, however, that agreement is at least optional with animate incor-porated objects, as in (9.21).

(9.21) a. Uwári ye(-ruwa)-kst2-hser-2haw-e’Mary f.sg.subj(/m.sg.obj)-old.person-nom-carry-impf

ne rake-’níhaprt my-fatherMary is holding my father

b. Wa’-ke (-hi)-kst2-hser-áhset-e’fact-1sg.subj(/m.sg.obj)-old.person-nom-hide-puncI hid the old person (the old man)

And in fact Baker (1995: 336) observes in a footnote that “when the dou-bling material makes explicit the gender of the argument in question, the NounIncorporation plus agreement construction is preferred where possible.” Sinceagreement is claimed to be impossible in any instance of (syntactic) incorpo-ration via movement, this would appear to be a strike against the syntacticanalysis.

On the lexical analysis, what is the explanation for the fact that explicitagreement and Noun Incorporation do not comfortably co-occur in Mohawk?The facts are obviously rather complicated. There appears to be a preferencefor avoiding the appearance within a single verb of both overt agreement ma-terial and an incorporated noun referring to the same participant. On the otherhand, when a verb (regardless of its internal composition) takes an object, itagrees with that object. One way to resolve this tension is to treat the agreed-with position in a Noun Incorporation construction as if it were inanimate, inwhich case no overt marker will appear. That provides the basis for the prefer-ence for inanimates as incorporated elements. It also underlies the sense thatincorporation of animates is somehow pejorative: overt agreement with ananimate object can be avoided through the treatment of the DP in question asif it were inanimate, but this is clearly derogatory. We can then say that NounIncorporation constructions do indeed have morphosyntactic agreement, eventhough (with objects are—or are treated as—inanimate) this has no overtphonological consequences. There is thus no motivation for seeing theirobjects as (non-agreeing) traces rather than as pro.

Baker notes further that in some languages which are otherwise syn-tactically similar to Mohawk (Tanoan languages like Southern Tiwa andGunwinjguan languages like Mayali), overt agreement does appear with

�

�

“Clitics” — 2005/7/26 — 17:39 — page 279 — #289�

�

�

�

�

�


positions that are also associated with an incorporated noun. In Ainu, whichhe puts in the class of polysynthetic languages (incorrectly, as shown in Kaiser1997a), some dialects have agreement with the “incorporated” argument pos-ition and some do not. Wherever we find agreement with an incorporatedposition in a “polysynthetic” language, that presents a problem for Baker’sanalysis. On the other hand, the lexical account derives these cases withoutdifficulty, since they represent plain transitive Verbs.

Baker’s other two arguments similarly do not provide clear proof of theneed for a Head Movement analysis of Noun Incorporation. One of thesederives from the fact that in Mohawk the object position associated with anoun-incorporating verb cannot be questioned with a general purpose ques-tion word. For Baker, this follows on the syntactic analysis from the fact thatthe argument position cannot simultaneously contain a noun to be incorpo-rated and a question word, a conclusion that does not follow from the lexicalanalysis.

There are two subcases to this argument. The first is that of who questionslike (9.22).

(9.22) Úhka wa’-{

*kekhe

}ksá-ht-a-ya’k-e’?

who fact-

{*1sS

1sS/FsO

}child-nom-∅-hit-punc

Who (a child) did I slap?

In this case, we see that if agreement is overtly present, the sentence isacceptable; while the absence of agreement leads to ungrammaticality. AsBaker points out, the variant with agreement validates the predictions of thelexical account; but he suggests that the impossibility of the alternative withoutagreement supports the syntactic analysis, since it would only follow (accordingto him) from the incompatibility of agreement and movement traces. But thisdoes not in fact follow: the ungrammaticality of the sentence without agree-ment (treated on the lexical view as involving inanimate agreement) is duesimply to an evident agreement conflict. That is, the question word ‘who’makes it explicit that the object is animate, and this conflicts with the apparentinanimate agreement.

For what- questions the account is a bit different. In sentence (9.23) thereis 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-puncWhat did he meat-eat?


Baker again concludes that the incompatibility of question words withincorporation in such sentences follows from his syntactic analysis, but there isan alternative that is readily available on the lexical analysis. The ungrammati-cality of (9.23) would also follow from the requirement that an overt argumentlinked to an argument position that is already partially specified (via Restrict)has to be informative. Since ‘what’ adds nothing to the semantics of ‘meat-eat’ that would further specify the properties of its object argument, it fails tomeet that requirement. The lexical account accommodates these facts withoutsyntactic movement.

Baker’s final argument is based on “condition C effects.” He notes that insentences like those in (9.24), coreference between the matrix subject and theargument corresponding to an incorporated noun is at best marginal.

(9.24) a. (Sak) ra-tsháni-s tóka 2-ke-ksá-ht-a-ya’k-e’Sak MsS-fear-hab maybe fut-1sS-child-nom-∅-hit-puncHe (Sak)i is afraid that maybe I will slap the childj/??i

b. Sak í-hr-ehr-e’ ks-kst2-hser-akérasSak ∅-m.sg.subj-think-impf n.sg.subj-old.person-nom-stinkSaki thinks the old personj/??i smells bad

c. Sak í-hr-ehr-e’ a-ke-kst2-hser-áhset-e’Sak ∅-m.sg.subj-think-impf opt-1sg.subj-old.person-nom-

hide-puncSaki wants me to hide the old personj/??i

Baker interprets the verb forms in the embedded clauses in these exam-ples as involving an incorporated object and intransitive agreement. He sug-gests that the difficulty of getting coreferential interpretations would follow ifthe position corresponding to the incorporated noun contained a trace, ratherthan pro, and that trace exhibited condition C effects (rather than the condi-tion B effects we would expect of a pro).

Another interpretation is quite possible, however. We might say that theagreement with the positions corresponding to incorporated nouns in (9.24)is actually inanimate, rather than missing (as is indeed overt in the case ofthe embedded intransitive in (9.24b)). In that case, the difficulty of getting acoreferential reading for these sentences would follow from an apparent gen-der conflict between the matrix subject (overtly animate and masculine in allcases) and the lower DP, apparently identified as inanimate. This interpreta-tion is strengthened by Baker’s example given here as (9.25), where appropriatetransitive animate agreement with the “incorporated” position appears, andwhere coreference is perfectly acceptable.


(9.25) (Sak) ra-tsháni-s tóka 2-hi-ksá-ht-a-ya’k-e’Sak m.sg.subj-fear-hab maybe fut-1sg.subj/m.sg.obj-child-

nom-∅-hit-puncHe (Sak)i is afraid that maybe I will slap the childi/j

Given the uncertain status of condition C effects in the syntactic litera-ture, it would not be wise to put too much weight on this argument in eitherdirection, but it certainly appears that even assuming the disjoint referenceprinciples Baker invokes, it does not necessarily follow that the position corre-sponding 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 ratherthan trace.

In summary, the lexical account of Noun Incorporation seems entirelyviable, even for the phenomena Baker treats as most centrally syntactic. Sincethe lexical analysis is clearly motivated in some cases, even within Mohawk,it is incumbent on the advocate of an alternative analysis based on syntacticHead Movement to show that this is the only one available under at least someconditions. There does not appear to be evidence of that sort, however, andso it follows that Mohawk and similar languages do not support the claim thatNoun 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 thetype that all analysts agree are core instances of Noun Incorporation. Accord-ing to Sadock (1980, 1986, 1991, 2003; see also Woodbury and Sadock 1986), thelarge class of suffixes in “Eskimo” languages like West Greenlandic that deriveverbs from noun stems as in (9.26) also involve composition of surface words inthe syntax, a view similar to Baker’s and contrary to the Lexicalist Hypothesis.

(9.26) a. Sapanngamik kusanartumik pisivoqbead-inst beautiful-nom-inst thing-get-indic-3sgHe bought a beautiful bead

b. Kusanartumik sapangarsivoqbeautiful-nom-inst bead-get-indic-3sg

These denominal verbs differ in some ways from basic Noun Incorpora-tion constructions, as remarked above. As Sadock shows, though, they poseessentially the same issues for the morphology–syntax interface. That is, inthese structures as well we appear to have a syntactic relation between part of a


(complex, derived) word and a position in syntactic structure. Sadock (2003)says explicitly that “[s]ome affixes in all classes also count as independentsyntactic elements in syntactic representations. This is a highly characteris-tic feature of W[est] G[reenlandic] grammar that is responsible for the restof its polysynthetic character. A single word can not only have the mean-ing of an entire phrase or sentence in more isolating languages, but also thesyntax thereof.” Sadock’s usage of “polysynthetic” is much closer to thetraditional understanding of this term than Baker’s, but like Baker’s, the rel-ation between a syntactic noun head and the nominal component of acomplex denominal verb, at least at first glance, appears to be best expressed asdisplacement.

Sadock’s (1991) own approach to this construction is actually somewhatdifferent. Instead of moving a noun to adjoin it to a verbal element, he treatsthis construction (among others) as a mismatch between syntactic and mor-phological structures, by which the morphology is allowed to construct wordsout of pieces that are not co-constituents in the syntactic representation ofa sentence. For present purposes, the differences between this view and theHead Movement analysis of Baker (1995) are not significant. Both assert thatthe nominal element in the constructions at issue is to be identified with ma-terial constituting an argument of the verbal element, while the lexical analysisto be proposed here does not make such an identification.

A number of arguments are cited in Sadock (1986) in support of the notionthat parts of these complex words participate in the external syntax of thestructures they head. I will address those points in the present section, andsuggest that a syntactic analysis of the sort assumed by Sadock is not in factrequired. This conclusion is in accord with several other analyses in the lit-erature (e.g., van Geenhoven 1998; Malouf 1999; and Bittner 2001). Sadock’spoints are very significant, and any account of Noun Incorporation (and in-deed the general nature of the relation between syntax and morphology) mustcome to terms with them. I can not take up all of these matters in as muchdetail as they deserve, but I will try to provide enough discussion to sub-stantiate my optimism with respect to the lexical account, as it extends toGreenlandic.

The first of Sadock’s arguments is based on the observation that in-corporating verbs alone, as opposed to all others in the language, can haveexternal 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 beinterpreted as the possessor of the object ‘(many) daughters,’ and not as the subject of a transitiveverb ‘have.’


a nominal phrase consisting only of a possessive (in the ergative case) modify-ing a presumably null head cannot in general appear outside of incorporationstructures, as illustrated in (9.27c).

(9.27) a. Puissip neqitorpungaseal-erg meat-consume-indic.1sgI ate seal meat

b. Kunngip panippassuaqarpoqking-erg daughter-many-have-indic.3sgThere are many princesses

c. *Kunngip takuvunga/takuaraking-erg see-indic.1sg/1sg.3sg(I saw the king’s)

Elsewhere, nominals without an overt head noun do occur, consisting only ofmodifiers such as a quantifier, and it is important to understand why strandedpossessors 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 apossessor, rather than DPs headed by a lexical noun which becomes part of acomplex verb. What needs to be explained is why such DPs can only occur asarguments of noun-incorporating verbs.

“Incorporation” only applies to objects4 in Greenlandic, and, the objectassociated with an “incorporated” noun stem is always and necessarily indef-inite. 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-headedDPs do not appear in positions where they should exhibit agreement, becausethe null head is indefinite, and Greenlandic verbs agree with their objects onlywhen these are definite. This shows us why the transitive form verbal takuarais 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 froma requirement that the null head is only acceptable where some informationis provided to specify some of its semantic properties. Some modifiers (e.g.,quantifiers or adjectives) might be argued to have this effect, but not posses-sives. 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 constitutean exception to this generalization, but as I will suggest below, these should be regarded as “quasi-transitive” and any incorporated argument as corresponding to a kind of object.

�

�

“Clitics” — 2005/7/26 — 17:39 — page 284 — #294�

�

�

�

�

�


such information (via Restrict), and so possessed null-headed nominals arepossible as objects of such verbs.

The distribution of null-headed nominals in Greenlandic remains to beexplored in more detail, but a difference in this regard between possessedexpressions and those with other modifiers does not seem implausible. Therestriction of the possessed phrases to incorporation structures, where theincorporated stem provides some semantic specification of the referent ofthe null head, then follows from a more general informativeness requirementthat cannot be satisfied by null heads alone.

Sadock’s second point, which has become particularly well known in theliterature, is that incorporating verbs alone may impose purely formal, se-mantically unmotivated restrictions on the plurality of an external DP. This isillustrated in example (9.28).

(9.28) a. Ataatsinik qamuteqarpogone-inst.pl sled.pl-have-indic.3sgHe has one sled

b. *Ataatsimik qamuteqarpogone-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 tantemform in Greenlandic, and appears as grammatically plural even when its ref-erent is semantically singular. Even though this noun does not appear in theovert argument expression ataatsinik, this is inflected as a plural in apparentdisregard of the semantically singular nature of the sled. This fact would ofcourse follow if the grammatically plural noun qamut ‘sled’ were present inthe underlying form of the DP, and displaced to the “incorporated” positionafter triggering Agreement.

I suggest that this behavior follows relatively naturally from the accountproposed here, though, without the need to invoke such displacement. Theverb qamuteqarpoq ‘he has sled(s)’ is built on the noun qamut, and thus pre-sumably inherits the lexical idiosyncrasies of that word, including its otherwiseunmotivated number.

(9.29)qamut[

—sled

+Plural

]+

–qar

have[

—Agent

],[ —

Theme] ⇒

qamuteqar–

have[

—Agent

],⎡⎢⎢⎣

—Themesled

+Plural

⎤⎥⎥⎦


In this structure, the derived complex verb inherits the grammaticalidiosyncrasy of its base noun in the form of a grammatical restriction on theobject 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 beassigned the formal feature [+Plural], and the modifier agrees with that.

Sadock’s third point is that incorporating verbs alone may have an addi-tional absolutive nominal associated with them, which follows the verb andis 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 tusaamasoqJørgen priest-become-indic.3sg famous-nomJørgen became a famous priest

b. ????Joorut toquvoq tusaamasoqJørgen die-indic.3sg famous-nomThe 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 argumentposition. Tusaamasoq ‘famous’ in example (9.30a) thus occupies the posi-tion 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 prob-lem with having parts of the same DP separated by the verb: in this examplethere is only a single DP, whose parts are non-contiguous, while in thesuperficially similar (9.30a) the two parts correspond to two separatephrasal constituents. Predicate nominatives are of course non-referential, asargued long ago by Kuno (1970), and in the context of Greenlandic case-assignment 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 natureof incorporation. He notes that Greenlandic displays sporadic instances of“the polysynthetic equivalent of Gapping” in which the first of two conjunctsconsists of an incorporating verb form and the second simply of an absolutiveDP with no suffix (see sentence (5.3) in Chapter 5 above for an example). Whileit might of course be possible to derive the apparent Gapping structures bysyntactic deletion, it seems likely that ellipsis of this sort is a matter of semanticinterpretation rather than of syntactic deletion or reconstruction.

Greenlandic (though not the otherwise similar Yup’ik) also disallows anexternal occurence of the same nominal that has been “incorporated” with adenominal verbal affix. But this simply shows that Greenlandic does not regard


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 con-structions does not involve movement of the nominal head per se, but rathera mechanism that allows parts of complex words to be interpreted as syntacti-cally autonomous elements. On this picture, sentence (9.28a) is not formed bymoving qamut ‘sled’ out of the DP and into the verb, but rather by allowingthe 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 forthe analysis: “Within the noun phrase, the possessor must precede the headnoun and any modifiers must follow it (Fortescue 1984: 117). Under Sadock’sanalysis, examples like [(9.31)], where a nominal modifier precedes an incor-porated nominal, violate the linear precedence constraints fornoun phrases.”

(9.31) kissartu-mik kavvi-sur-puthot-inst coffee-drink-3pl.indicThey drank hot coffee (Fortescue 1984: 83)

Sadock must thus allow the word-order constraints within the DP to berelaxed exactly when the head noun is part of an incorporation structure. Ifwe treat the nominal in (9.31) as having a null head, distinct from the stem thatappears as part of the (lexically formed) denominal verb, no such proviso isnecessary. Of course, an analysis such as Baker’s, on which the nominal headmoves into the verb in the syntax, also avoids this difficulty, though we havealready seen other reasons to doubt the motivation for either of these accountsas opposed to the lexical one.

9.5 Conclusion

So where does this survey of incorporation phenomena leave us with respectto a choice between the two primary views of Noun Incorporation? It appearsthat even Baker agrees that much Noun Incorporation is in fact lexical, notsyntactic, even in the language for which he feels the strongest syntactic casecan be made (Mohawk). And in fact, the limited sets of data for which the syn-tactic account is still said to be necessary can also be accommodated within thelexical account, without invoking extraordinary mechanisms. That means thata purely lexical account of Noun Incorporation, without syntactic movement(or its equivalent in Sadock’s system), is almost certainly possible. But that, inturn, means that the best putative support for an operation of syntactic Head


Movement may be non-existent---a conclusion with extensive consequences formany areas of contemporary syntax. This is especially true for the “split-Infl”analysis of clause structure, with its proliferation of functional heads. Sincethe framework offered in the present book also accommodates the grammarof special clitics without recourse to extensive layers of functional structure,these conclusions dovetail to suggest that clause structure in natural languageis formally simpler (and thereby closer to surface form) than it is often sug-gested to be.

References

Abeillé, Anne and Godard, Danièle (2003), “Les prédicats complexes dans les languesromanes”, 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”. Mas-

ter’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 Seman-tics 16 (New York: Academic Press), 21–75.

—— (1985a), “Inflectional morphology”, in T. Shopen (ed.), Language typology andsyntactic description, vol. III (Cambridge: Cambridge University Press), 150–201.

—— (1985b), Phonology in the twentieth century (Chicago: University of ChicagoPress).

—— (1991), “Syntactically arbitrary inflectional morphology”, Yearbook of Morphology4: 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: AkademieVerlag GmbH), 123–42.

—— (2000c), “Towards an optimal account of second position phenomena”, inDekkers, van der Leeuw, and van de Weijer (2000), 302–33.

—— (2001), “On some issues of morphological exponence”, Yearbook of Morphology2000: 1–18.

—— (2004), “Subject clitics and verb-second in Surmiran Rumantsch”, MIT WorkingPapers in Linguistics 47: 1–22.

—— and Lightfoot, David W. (2002), The language organ: Linguistics as cognitive phys-iology (Cambridge: Cambridge University Press).

290 References

Anderson, Stephen R. and Chung, Sandra (1977), “On grammatical relations and clausestructure in verb-initial languages”, in P. Cole and J. Sadock (eds.), Grammatical re-lations, Syntax and Semantics 8 (New York: Academic Press), 1–25.

Anderson, Victoria and Otsuka, Yuko (2002), “Phonetic correlates of length, stress anddefinite accent in Tongan”. Paper presented at 9th meeting of the Austronesian For-mal Linguistics Association, Cornell University, Ithaca, NY.

Bach, Emmon (1998), “Semantic relations in and outside of complex words”. Paperpresented 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”, Linguis-tic Inquiry 16: 373–415.

—— (1987), Incorporation: A theory of grammatical function changing (Chicago: Uni-versity 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 positionclitics 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 (Ox-ford: 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 diGiovan Battista Pellegrini (Pisa: Pacini), 25–32.

—— (1985), “L’interferenza sintattica: di un aspetto della sintassi ladina considerato diorigine 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 Illi-

nois Studies in Language and Literature, vol. III, nos. 2–4 (Urbana: University ofIllinois).

References 291

—— (1946), “Algonquian”, in Cornelius Osgood (ed.), Linguistic structures ofNative America, Viking Fund Publications in Anthropology, 6 (New York),85–129.

Boas, Franz (1911), “Kwakiutl”, in F. Boas (ed.), Handbook of American Indianlanguages, vol. 1 (Washington D.C.: Smithsonian Institution, Bureau of AmericanEthnology, 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 theAmerican Philosophical Society 37(3): 201–377.

Bobaljik, Jonathan and Wurmbrand, Susan (2002), “Long distance agreement, restruc-turing 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 phonol-

ogy”, 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 Bre-

ton”, in R. D. Borsely and I. Roberts (eds.), The syntax of the Celtic languages (Cam-bridge: Cambridge University Press), 1–52.

Boškovic, Željko (1995), “Participle movement and second position cliticization inSerbo-Croatian”, Lingua 96: 245–66.

—— (2000), “Second position cliticisation: Syntax and/or phonology?”, in Beukemaand den Dikken (2000), 71–119.

—— (2001), On the nature of the syntax-phonology interface (Oxford: Elsevier ScienceLtd.).

Boysson-Bardies, Bénédicte de (1999), How language comes to children (Cambridge,MA: MIT Press).

Bresnan, Joan (1978), Contraction and the transformational cycle (Bloomington: IndianaUniversity 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 SlavicMaterials, vol. 10 (Ann Arbor), 36–52.

292 References

Browne, Wayles (1975), “Serbo-Croatian enclitics for English-speaking learners”, Con-trastive 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 Con-ference 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: Aus-tralasian Medical Publishing Company).

—— (1953), Tongan grammar (London: Oxford University Press).Clark, Ross (1974), “On the origin of the Tongan definitive accent”, Journal of the Poly-

nesian 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 pre-

sented 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 Philo-logical Society, vol. 101 (2) (Oxford: Blackwell), 155–202.

Crysmann, Berthold (2000), “Clitics and coordination in linear structure”, in Gerlachand Grijzenhout (2000), 121–59.

Cysouw, Michael (forthcoming), “Morphology in the wrong place: A survey of pre-posed enclitics”, in W. U. Dressler (ed.), Morphology and its demarcations (Amster-dam: Benjamins).

References 293

Dekkers, Joost, van der Leeuw, Frank, and van de Weijer, Jeroen, eds. (2000), Optimal-ity Theory: Phonology, syntax and acquisition (Oxford: Oxford University Press).

Di Sciullo, Anna-Maria and Williams, Edwin (1987), On the definition of word (Cam-bridge, MA: MIT Press).

Dixon, Robert M. W. (1980), The languages of Australia (Cambridge: Cambridge Uni-versity Press).

Donaldson, Tamsin (1980), Ngiyambaa: The language of the Wangaaybuwan (Cam-bridge: Cambridge University Press).

Eythórsson, Thórhallur (1996), “Functional categories, cliticization, and verb move-ment 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 <http://www.ledonline.it/snippets/>.

Fontana, Josep M. (1993), “Phrase structure and the history of clitics in the history ofSpanish”. 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 phono-

logically driven morphology”, Proceedings of the West Coast Conference on FormalLinguistics 9: 189–203.

Galves, Charlotte and Sandalo, Filomena (2004), “Clitic placement in modern and clas-sical European Portuguese”, MIT Working Papers in Linguistics 47: 115–28. CollectedPapers on Romance Syntax.

Ganzoni, Gian Paul (1977), Grammatica Ladina: Grammatica sistematica dal rumantschd’Engiadina Ota (Samedan: Lia Romauntscha).

—— (1983), Grammatica Ladina: Grammatica sistematica dal rumantsch d’EngiadinaBassa (Samedan: Lia Romantscha).

Gazdar, Gerald, Klein, Ewan, Pullum, Geoffrey, and Sag, Ivan (1985), Generalized phrasestructure grammar (Cambridge, MA: Harvard University Press).

van Geenhoven, Veerle (1998), Semantic incorporation and indefinite descriptions: Se-mantic 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.), Hand-book of morphology (London: Basil Blackwell), 84–100.

Gerlach, Birgit and Grijzenhout, Janet, eds. (2000), Clitics in phonology, morphologyand syntax (Amsterdam: Benjamins).

http://www.ledonline.it/snippets/

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 argumentstructure (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. Ever-

aert, and J. Grimshaw. (eds.), Lexical specification and insertion (Amsterdam: Ben-jamins), 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 studyin West Flemish, Cambridge Studies in Linguistics (Supplementary Volume) (Cam-bridge: 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 Rotu-man”, 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 RoyalSociety of Canada 6: 59–105. Section II, 1888.

Halle, Morris and Marantz, Alec (1993), “Distributed morphology and the pieces ofinflection”, 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. man-uscript, Stanford University, CA.

—— (1992b), “Topics in the Placement and Morphology of Clitics”. Ph.D. thesis, Stan-ford 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 verb-

second 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 Univer-

sity 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: Rei-

del).—— (1990a), “Breton pronominals, binding, and barriers”, in Hendrick (1990b),

121–65.Hendrick, Randall, ed. (1990b), The syntax of modern Celtic languages, Syntax & Se-

mantics 23 (New York: Academic Press).

References 295

Hendriksen, Hans (1986), Himachali studies III: Grammar, Historisk-filosofiske Med-delelser 48 (Copenhagen: Det Kongelige Danske Videnskabernes Selskab).

—— (1990), “Sentence position of the verb in Himachali”, Acta Linguistica Hafniensia22: 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 Ameri-

can 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 Linguis-tics, 189–200.

Holmer, Arthur J. (1996), A parametric grammar of Seediq (Lund: Lund UniversityPress).

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 andold French”, in W. J. Ashby, M. Mithun, G. Perissinotto and E. Raposo (eds.), Lin-guistic 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 Semantics19 (Orlando: Academic Press), 15–42.

Jelinek, Eloise (1984), “Empty categories, case and configurationality”, Natural Lan-guage and Linguistic Theory 3: 39–76.

Jónsson, Jóhannes Gísli (1991), “Stylistic fronting in Icelandic”, Working Papers in Scan-dinavian Syntax 48: 1–43.

Kager, René (1999), Optimality Theory (Cambridge: Cambridge University Press).Kahn, Daniel (1976), “Syllable-based generalizations in English phonology”. Ph.D. the-

sis, MIT, distributed by Indiana University Linguistics Club.Kaiser, Lizanne, ed. (1997), Studies in the morphosyntax of clitics, Y.A.L.E. working pa-

pers, vol. 1 (Dept. of Linguistics, Yale University).Kaiser, Lizanne (1998a), “The interaction of noun incorporation and applicative for-

mation 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 Inquiry22: 647–86.

Kiparsky, Paul (1995), “Indo-European origins of Germanic syntax”, in Battye andRoberts (1995), 140–69.

—— (2000), “Opacity and cyclicity”, The Linguistic Review 17: 351–67.—— (2003), “Finnish noun inflection”, in D. Nelson and S. Manninen (eds.), Gener-

ative 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 Inter-

national Conference on Salishan Languages (Seattle: Department of Anthropology,University of Washington).

Koopman, Hilda and Sportiche, Dominique (1991), “The position of subjects”, Lingua85: 555–88.

Kroeber, Alfred (1909), “Noun incorporation in American languages”, Proceedings ofthe 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 andsyntax, vol. 2, 5th edn. (Groningen: P. Noordhoff).

Kuno, Susumu (1970), “Some properties of non-referential noun phrases”, in R. Jakob-son and S. Kawamoto (eds.), Studies in general and oriental linguistics presented toShiro 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, Uni-versity of California, Santa Cruz.

—— (2002), “Rotuman incomplete phase as morphologically driven word optimiza-tion”. Paper presented at the Linguistics and Phonetics 2002 conference, Meikai Uni-versity, 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”, inM. 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 fromWarlpiri”. Manuscript, University of Delaware.

References 297

Legendre, Géraldine (1998), “Second position clitics in a V2 language: Conflict reso-lution 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-Slavicclitics”, Proceedings of the West Coast Conference on Formal Linguistics XVII, 436–50.

—— (2000a), “Morphological and prosodic alignment of Bulgarian clitics”, in Dekkerset al. (2000), 423–62.

—— (2000b), “Optimal Romanian clitics: A cross-linguistic perspective”, in V. Mota-panyane (ed.), Comparative studies in Romanian syntax, North-Holland LinguisticSeries 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 Legen-dre et al. (2001), 241–77.

Legendre, Géraldine, Sten Vikner, and Jane Grimshaw, eds. (2001), OT syntax (Cam-bridge, 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 ofBaxwbakwalanusiwa, as told by Kitlope elder Gordon Robertson”, Working papers forthe 21st International Conference on Salishan Languages (Seattle: University of Wash-ington).

Linder, Karl Peter (1987), Grammatische Untersuchungen zur Charakteristik des Rätoro-manschen 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-nunaanninaqiterisitsisarfik).

Macaulay, Monica (1987), “Cliticization and the morphosyntax of Mixtec”, Interna-tional Journal of American Linguistics 53: 119–35.

Maling, Joan (1980), “Inversion in embedded clauses in modern Icelandic”, Íslenskt Mál2: 175–93.

Malouf, Robert (1999), “West Greenlandic noun incorporation in a monohierarchicaltheory of grammar”, in G. Webelhuth, A. Kathol, and J.-P. Koenig (eds.), Lexical andconstructional aspects of linguistic explanation (Stanford, CA: CSLI Publications), 47–62.

Mardirussian, Galust (1975), “Noun incorporation in universal grammar”, Proceedingsof the Chicago Linguistic Society 11: 383–89.

McCarthy, John J. (1991), “Synchronic rule inversion”, Proceedings of the Annual Meetingof the Berkeley Linguistics Society 17: 197–207.

—— (1995), “Extensions of faithfulness: Rotuman revisited”. Unpub. manuscript, Uni-versity of Massachusetts; Rutgers Optimality Archive paper ROA-110.

298 References

McCarthy, John J. (2000), “The prosody of phase in Rotuman”, Natural Language andLinguistic Theory 18: 147–97.

Merlan, Francesca (1976), “Noun incorporation and discourse referencein 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 ofLinguistics).

Miller, Philip H. and Monachesi, Paola (2003), “Les pronoms clitiques dans les languesromanes”, 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 Univer-sity.

Moravcsik, Edith (1977), On rules of infixing (Bloomington: Indiana UniversityLinguistics Club).

Mussafia, A. (1898), “Enclisi o proclisi del pronome personale atono qual oggetto”, Ro-mania 27: 145–6.

Nespor, Marina (1994), “The phonology of clitic groups”, in L. Hellan and H. vanRiemsdijk (eds.), Clitic doubling and clitic groups, vol. 5, ESF-Eurotyp Working Pa-pers, 67–90.

—— and Vogel, Irene (1986), Prosodic phonology (Dordrecht: Foris).

Nevis, Joel A. (1986), Finnish particle clitics and general clitic theory. Working Papers inLinguistics 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 bibliogra-phy 1892–1991 (Amsterdam: Benjamins).

—— and Joseph, Brian (1992), “Wackernagel affixes: Evidence from Balto-Slavic”, Year-book of Morphology 1992: 93–111.

Nida, Eugene A. (1949), Morphology: The descriptive analysis of words, 2nd edn. (AnnArbor: University of Michigan Press).

Noyer, Rolf (1994), “Mobile affixes in Huave: Optimality and morphological well-formedness”, 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 AboriginalStudies).

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 andG. K. Pullum (eds.), Handbook of Amazonian languages, vol. 2 (Berlin: Mouton deGruyter), 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. Dzi-

wirek, P. Farrell, and E. Mejfas-Bikandi (eds.), Grammatical relations: A cross-theoretical perspective (Stanford, CA: Center for the Study of Language and Infor-mation), 349–64.

—— and Potsdam, Eric (2001), “Long distance agreement and topic in Tsez”, NaturalLanguage and Linguistic Theory 19: 583–646.

Poser, William, (1985), “Cliticization to NP and lexical phonology”, Proceedings of theWest Coast Conference on Formal Linguistics 4: 262–72.

Postal, Paul (1966), “On so-called pronouns in English”, in F. Dineen (ed.), Report of the17th 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 Col-orado.

Pullum, Geoffrey K. and Zwicky, Arnold M. (1986), “Phonological resolution of syn-tactic feature conflict”, Language 62: 751–73.

—— (1988), “The syntax-phonology interface”, in F. J. Newmeyer (ed.), Linguistics: TheCambridge survey, vol. I (Cambridge: Cambridge University Press), 255–80.

—— (1998), “Licensing of prosodic features by syntactic rules: The key to auxiliaryreduction”. Paper presented at January Annual Meeting of the Linguistic Society ofAmerica.

Rath, John C. (1981), A practical Heiltsuk-English dictionary with a grammatical intro-duction, National Museum of Man Mercury Series, 75 (2 volumes) (Ottawa: NationalMuseums 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.), Recenttransformational 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. vande Weijer (eds.), Phonology in progress: Progress in phonology, HIL Phonology PapersIII (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 Auck-

land.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 (Cam-

bridge, 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 MassachusettsOccasional Papers in Linguistics 18 (Amherst: Dept. of Linguistics, University ofMassachusetts), 439–70.

Sherwood, David Fairchild (1986), Maliseet-Passamaquoddy verb morphology, Cana-dian Ethnology Service, Paper No. 105 (Ottawa: Canadian Museum of Civilization).

Signorell, Faust, Wuethrich-Grisch, Mena, and Simeon, Gion Pol (1987), Normas sur-miranas (Coira: Tgesa editoura cantunala per stampats e meds d’instrucziun).

References 301

Simeon, Gion Pol (1996), Igl muvimaint rumantsch an Surmeir (Savognin: Uniun ru-mantscha da Surmeir).

Simpson, Jane (1991), Warlpiri morpho-syntax: A lexicalist approach, Studies in NaturalLanguage and Linguistic Theory (Dordrecht: Kluwer Academic).

Skorik, Pjotr Ja (1948), Ocerk po sintaksisu cukotskogo jazyka: inkorporacija (Leningrad:Ucpedgiz).

Smith, Ian and Johnson, Steve (2000), “Kugu Nganhcara”, in R. M. W. Dixon andB. Blake (eds.), Handbook of Australian languages, vol. 5 (Oxford: Oxford Univer-sity 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. Row-

land (eds.), Celtic linguistics (Amsterdam: John Benjamin), 151–67.Steriade, Donca (1988), “Greek accent: A case for preserving structure”, Linguistic In-

quiry 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: Cam-

bridge 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 For-mal Linguistics Association.

Tegey, Habibullah (1977), “The Grammar of Clitics: Evidence from Pashto and OtherLanguages”. Ph.D. thesis, University of Illinois.

Tobler, Adolf (1875/1912), “Besprechung von J. Le Coultre, De l’ordre des mots dansChrétien de Troyes”, Vermischte Beiträge zur französischen Grammatik, vol. 5 (Leipzig:Hirzel), 395–414.

Toivonen, Ida (2000), “The morphosyntax of Finnish possessives”, Natural Languageand Linguistic Theory 18: 579–609.

Truckenbrodt, Hubert (1999), “On the relation between syntactic phrases and phono-logical phrases”, Linguistic Inquiry 30: 219–55.

Vance, Barbara (1997), Syntactic change in medieval French: Verb-second and null sub-jects (Dordrecht: Kluwer).

van der Leeuw, Frank (1995a), “Cliticization as alignment to morphological slots”, Pro-ceedings of the Chicago Linguistic Society 31: 168–80 vol. 2: Proceedings of the Parases-sion 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 deGruyter).

Vogel, Irene (1990), “The clitic group in prosodic phonology”, in J. Mascaró and M. Ne-spor (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 comprehensivegrammar (London: Routledge).

Woodbury, Anthony and Sadock, Jerrold (1986), “Affixal verbs in syntax: a reply toGrimshaw and Mester”, Natural Language and Linguistic Theory 4: 229–44.

Zanuttini, Raffaella (1997), Negation and clausal structure: A comparative study of Ro-mance languages (Oxford: Oxford University Press).

Zec, Draga (1987), “Interactions of prosodic and syntactic constituency”. Unpub. man-uscript, Stanford University.

—— (1988), “Sonority Constraints on Prosodic Structure”. Ph.D. thesis, Stanford Uni-versity.

—— (1993), “Rule domains and phonological change”, in S. Hargus and E. Kaisse(eds.), Studies in lexical phonology, Phonetics and Phonology 4 (New York: Acad-emic Press), 365–405.

—— and Inkelas, Sharon (1990), “Prosodically constrained syntax”, in S. Inkelas andD. Zec (eds.), The phonology–syntax connection (Chicago and Stanford: University ofChicago Press and CSLI Publications), 365–78.

—— (1992), “The place of clitics in the prosodic hierarchy”, Proceedings of the WestCoast Conference on Formal Linguistics 10: 505–20.

Zide, Arlene R. K. (1976), “Nominal combining forms in Sora and Gorum”, Oceaniclinguistics special publication 13, Austroasiatic Studies, Part II (Honolulu: Universityof Hawaii Press), 1259–94.

Zwicky, Arnold M. (1977), On clitics (Bloomington: Indiana University LinguisticsClub).

—— (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”, Lan-

guage 59: 502–13.

Index of Subjects

accent 10, 12, 24affixation 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; suffixesagreement 161, 169, 186, 228–54,

248–51, 275–80, 283–4long-distance 227, 248vs. “registration” 230, 235

Alignment constraints:Align(Cati,Edger/l ,Catj ,

Edger/l) 57, 135, 145, 181Align(CP,L,PWord,L) 60Align(GrWord,L,LexWord, L) 137Align(LexMax,L,PPhrase, L) 68Align(LexMax,R,PPrase, R) 68Align(LexWord,L,PWord, L) 57,

68Align(LexWord,R,PWord, R) 57,

68Align(PPhrase,L,LexWord, L) 59Align(PPhrase,R,PWord, R) 68Align(PWord,L,LexWord, L) 64,

68Align(Pword,R,LexWord, R) 68Align(XP,L,PPhrase,L) 57, 59Align(XP,R,PPhrase,R) 57Align([e],R,PPhrase,L) 70see also EdgeMost (Alignment)

constraintsA-Morphous Morphology 3anchors

for clitic placement 79, 80, 85,115, 123, 142, 147

auxiliaries 16, 25, 48, 140–1clitics 24, 65–8, 72contracted 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” insyntax 28–30

clitics

Affixal 46–50

complementizer 105, 254–5

304 Index of Subjects

clitics (Continued)demonstrative 95, 103Free 46–9, 66“inflectional” 128, 169Internal 46, 48, 50Klavans’s typology of 79phonological 23, 31, 37, 41, 58, 76pronominal 11, 14, 16–18, 62, 78,

80–2, 152, 227–8PWord 46, 48Zwicky’s classification of 9–11,

22–6, 31–2, 36, 76, 81see also enclitics; proclitics;

second-position clitics; simpleclitics; special clitics; subjectclitics

∗Coda 136comparative subdeletion 70Complementizer

in Breton 199in German 77, 87, 105, 140, 148,

183, 193, 249, 254, 255in Icelandic 185in Kashimiri 189in Kwakw,ala 105in Surmiran 214

Complementizer Phrase (CP)recursion 185

complete/incomplete phase formsin Rotuman 100–1

compositesnon-compound formation 159in Portuguese 159prosodic structure of 160

compounds 261, 264–7constraints (Optimality Theory)

ranking of 55, 58–60, 64, 69,148–50, 154, 164, 179, 198

correlative constructionsin Kashimiri 192

definiteness 95, 99marked 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 Kwakw,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 Kwakw,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


Features 90–1, 231, 233, 236–7[+neg] (English) 35[D(eterminer)] 91[Poss] 90–1

Final Devoicingin Polish 40

Foot 38–9, 41, 43, 47, 53Full Interpretation (at PF) 39, 45–6,

52–3, 58, 67, 69functional categories

proliferation of 287function words

in English 48–9, 69Gapping 69, 285gaps, syntactic 29, 65, 73

post-auxiliary 29, 70Generalized Phrase Structure

Grammar (GPSG) 240Generate (gen) 182Generative Phonology, Classic 56Genitive

“group genitive” in English 90

Head-Driven Phrase StructureGrammar (HPSG) 95, 240, 270

Headedness constraint 52–3Head Feature Convention 146, 233,

240, 246Head-inflection (rule) 170Head Movement 113, 124, 258, 260,

264, 269, 270–1, 279, 282

impersonal constructionsin Icelandic 184in Rumantsch 216, 218–19

incorporation 261, 263, 267, 275,279–80, 285–6

in Aleut 269classifier 267, 272with doubling 270–3, 278in Kwakw,ala 269noun 257, 258–9, 260–83, 286syllable-affix 94

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


Layeredness 47, 52–3Layering Convention 233Left Dislocation

in Breton 201LeftEdgeFaith constraints

LeftEdgeFaith(CP) 141LeftEdgeFaith(P) 142LeftEdgeFaith(Verb) 163–5LeftEdgeFaith(Word) 138, 163

LeftMost (alignment) constraints:LeftMost(e(, D)) 135, 167, 171–2,

176LeftMost(Afi , D) 137LeftMost(ar/l) 136LeftMost(cl, Vfinite) 149LeftMost(cli) 139, 140–1, 143LeftMost(cli, D) 173LeftMost(cli,IP) 150, 154, 157LeftMost(cli,V) 146, 147, 157LeftMost(e) 145LeftMost(emañ, CP) 202LeftMost(V,CP) 199LeftMost(Verb[−Finite], V) 146LeftMost(Vfin) 190, 193LeftMost(Vfin, CP) 198LeftMost(Vfin, IP) 198, 203–4LeftMost(Vfin, S) 179, 182–3

Lexical Functional Grammar (LFG)236, 240

Lexicalist Hypothesis 34–5, 258, 264Lexical Phonology 9, 45, 51–2, 96,

134Post- 9, 14, 27, 51

LPhCon 68–9LWdCon 68–9

Max(Pfnc) 60Merge (syntactic operation) 169, 176mesoclitics See endocliticsMinimalist Program for syntax 91,

232Morphological 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 ofcategories 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 Kwakw,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


Non-recursive constraints 47–9NonRecursive(Ci) 52NonRecursive(PPh) 58NonRecursive(PW(or)d) 54–5, 60

Numeration 169Onset 63–4, 136Optimality Theory (OT) 50, 57, 132,

134, 157, 170, 176, 182, 225

particlesgrammatical- and discourse

markers 36Phonetic Form (PF) 34Phonological Phrases

(PPh/PPhr(ase) ) 39–41, 47–8,61, 66, 70, 116, 160

boundaries of 69, 73and syntactic phrases 67two-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 160Phonology 9, 12, 25, 56

autosegmental 38lexical 9, 14, 33, 45, 51–2, 96,

134post-lexical 9, 14, 27, 34, 51prosodic 45, 56

phrasing groupssyntactic and prosodic 56, 72

polysynthetic languages 266, 279,282

portmanteauxlexicalized 73

possessivesin English 89–94

possessorsin Finnish 236

in Kwakw,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 inventory47

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 PhonologicalPhrase (PPh/PPhr(ase) )

prosodic structure 45, 57, 66,69, 73

Prosodic Word See PhonologicalWord (PW(or)d)

PWdCon 68

Readjustment Rules in Sound Patternof English 56

Reanalysis See Restructuring


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 Adjunction171

stress 13, 19, 24, 38, 43–4, 62, 77,98

Strict Layering Hypothesis 42,47–50, 52

Struct∗ 58Stylistic Fronting

in Icelandic 184, 186, 187, 210subject clitics 208, 213, 215, 249–55,

249–255four types of 252, 253, 254

Subject gap conditionin Icelandic 184, 187

subordinate clausesin Breton 197in German 183in Icelandic 184–5in Kashimiri 189in Kwakw,ala 105in Pashto 153in Surmiran 206, 212, 214

suffixes 33, 66, 83, 85, 92, 128, 135English plural 92Huave theme vowel 85

Syllable-affix Incorporation 94syllables 13, 38, 49, 66–7Syntactic (S-)structure 46, 55

and prosody 45relevant to phonology 37

SyntaxPhonology-free 168

theta-roles (θ-roles) 232, 269external (R) 267, 269, 273

Tobler-Mussafia effects 147, 198topicalization

in Breton 195, 201in German 180in Icelandic 184, 195in Surmiran 210, 212

topics 19–22, 121, 148, 202, 209in Breton 194, 199–202in Bulgarian 214in Icelandic 186


in Kashimiri 189–90in Kwakw,ala 19–22in Surmiran 206, 211–12, 215–7,

233in Tagalog 165

Unaccusative Hypothesis 261underspecification

of animacy 277verb initial

in Breton 197–8verbs

in Breton 194–5, 197–8,201–3

incorporating 282in Kashimiri 188–91in Udi 162

verb second 177–224

in Breton 193, 199, 201in German 180–1, 183in Icelandic 183–6in Kashimiri 187–193in Surmiran 208, 212, 223–4in West Greenlandic 283

Voice Assimilationin Polish 40

wh-movement 69in Surmiran 213

wh-words/phrasesin Bulgarian 151, 241in Kashimiri 190–1in Kwakw,ala 17in Surmiran 213–16

Word Formation Rules

(morphology) 34

Index of Names

Abeillé, Anne 249Adams, Marianne P. 205Aitchison, Therese M. 95Allen, Cynthia 90Alpher, Barry 125Anderson, 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, 269Baker, Mark 6, 18, 84, 230–4, 258,

260, 263–6, 269–70, 273,275–82, 286

Barbosa, Pilar 85, 148, 149Barshi, Immanuel 274Benincá, Paola 204–5, 252–3Bergsland, Knut 268Bhatt, Rakesh Mohan 187, 189–90Bittner, Maria 267, 270, 282Bloomfield, Leonard 87, 167Boas, Franz 14–16, 19–22, 101Bobaljik, Jonathan 248Bolinger, Dwight 126Booij, Geert 40, 42, 45, 51,

61–3Borsley, Robert D. 193Boškovic, željko 107, 116–23, 126,

129, 132, 144, 150–1, 186Bresnan, Joan 29, 71, 242Browne, Wayles 81, 110–11, 122Burzio, Luigi 245

Carstairs, Andrew 90Carstairs-McCarthy, Andrew 93–4Chomsky, Noam 37, 56, 169Chung, Sandra L. 87, 114–16, 123,

193, 195, 201, 267Churchward, C. Maxwell 95, 97–9Clark, Ross 96Condax, Iovanna D. 98–9

Corbett, Greville 230Crysmann, Berthold 159Cysouw, Michael 82

Di Sciullo, Anna-Maria 268Dixon, Robert M. W. 274Dols, Nicolau 3, 59Donaldson, Tamsin 75, 125

Fillmore, C. J. 203Fontana, Josep M. 205Fortescue, Michael 286Foulet, Lucien 205Franks, Steven 119, 240–1Fukui, Naoki 203Fulmer, Sandra Lee 85

Galves, Charlotte 85, 149, 157Ganzoni, Gian Paul 219Gazdar, Gerald 146, 233Gerdts, Donna B. 260Godard, Danièle 249Goldsmith, John 38Grimshaw, Jane 246Grisch, Mena 218, 220

Haegeman, Liliane 248Haiman, John 204Hale, Kenneth 139–40Hale, Mark 100Halle, Morris 37, 56, 86Halpern, Aaron 111, 120Hannahs, S. J. 146;193,

197–99Harris, Alice C. 81, 161–2, 164Hayes, Bruce 42, 44, 142Hendrick, Randall 193Hock, Hans Heinrich 187Hockett, Charles F. 86Hoekstra, Eric 248

312 Index of Names

Holmer, Arthur J. 76Howe, Darin 14

Inkelas, Sharon 43–4, 47, 112Insler, Stanley 126

Jacobs, Haike 63Jaeggli, Osvaldo 245Jelinek, Eloise 266Johnson, Steve 80, 82Jónsson, Jóhannes Gísli 184Joseph, Brian 1, 85–6

Kahn, Daniel 38Kaiser, Lizanne 133, 279Kaisse, Ellen M. 4, 9, 26–8, 64–5, 80,

152, 156Kanerva, Jonni 235Kayne, Richard 2, 131–2King, Tracy Holloway 240–1Kiparsky, Paul 51, 134, 178, 235Kissock, Madelyn 100Klavans, Judith L. 4, 60, 63, 79, 81–2,

142, 146–7Klein, Ewan 146, 233Klokeid, Terry J. 21Koopman, Hilda 203Koul, Omkar N. 189–90Kroeber, Alfred 259Kroeger, Paul 169Kuno, Susumu 285Kurisu, Kazutaka 100–1

Ladusaw, William A. 267Lees, Robert B. 261Legate, Julie Anne 140Legendre, Géraldine 5, 124–5, 132,

151, 176, 179, 193, 198Liberman, Mark Y. 24, 38, 39Liddell, H. G. 1Lincoln, Neville J. 21Linder, Karl Peter 250Liver, Ricarda 204

Lukoff, Fred 37Lynge, Hans 110

Macaulay, Monica 11McCarthy, John J. 49, 99, 100McCawley, James D. 203Maling, Joan 184Malouf, Robert 270, 282, 286Marantz, Alec 86Mardirussian, Galust 260Merlan, Francesca 259, 263Miller, Phillip H. 240, 242Mithun, Marianne 260–2,

270–1Monachesi, Paola 129, 240, 242Moravcsik, Edith 83Mussafia, A. 147, 198

Nespor, Marina 12, 37, 39, 41–4, 47,60–1

Nevis, Joel A. 1, 80–1, 85–6, 235Nida, Eugene 1Noyer, Rolf 85

Ortiz de Urbina, Jon 193Osborne, C. R. 262Otanes, Fe T. 165–7, 173–5Otsuka, Yuko 98

Parker, Steve 64Payne, Doris L. 81, 274Payne, Thomas E. 81Peperkamp, Sharon 37, 43, 52–3Perlmutter, David 125, 168Poletto, Cecilia 205, 250–5Polinsky, Maria S. 248, 259Poser, William 96, 97Postal, Paul 91, 275Potsdam, Eric 248Prince, Alan 24, 38, 39, 132Progovac, Liljana 119Pullum, Geoffrey K. 11, 28, 32–3, 35,

70–1, 146, 168, 233, 276

Rath, John C. 20–3Richardson, Matthew 147–8

Index of Names 313

Rivero, Maria-Luisa 193Rizzi, Luigi 245–6Roberts, Ian 205, 250Roberts, Taylor 81, 152Rosen, Sara Thomas 267Ross, John Robert 126Rubach, Jerzy 40

Sadock, Jerrold 6, 109, 110, 260,281–2, 284–6

Sag, Ivan 146, 233Salisbury, Mary 98–9Sandalo, Filomena 85, 149, 157Sapir, Edward 258–64Schachter, Paul 126, 165–8, 170,

173–5Schafer, Robin 193Schapensky, Nathalie 193Scott, R. 1Selkirk, Elizabeth 12, 37, 39, 41,

46–51, 66, 68–9Shaw, Patricia 9Sherwood, David Fairchild 244ShÖtze, Carson 120Signorell, Faust 220–1, 247Simeon, Gion Pol 220–1, 247Simpson, Jane 140Skorik, Pjotr Ja. 259Smith, Ian 79, 82Smits, Caroline 248Smolensky, Paul 132Speas, Margaret 203Spencer, Andrew 259, 268

Sportiche, Dominique 203Stemberger, Joseph 90Stephens, Janig 193Stump, Gregory T. 90, 137, 159, 193

Taumoefolau, Melanaite 98Tegey, Habibullah 81, 152, 156Tellerman, Maggie 193, 197–9Tobler, Adolf 147, 198Toivonen, Ida 235

Vance, Barbara 205van der Leeuw, Frank 81, 152, 157van Geenhoven, Veerle 282Vigário, Marina 37, 157, 159–60Vogel, Irene 12, 37, 39, 41–4, 47,

60–1

Wackernagel, Jacob 2, 5, 23, 86, 108,143, 166, 177–8, 188, 224

Wali, Kashi 189–90Wanner, Dieter 1Wheeler, Max 3, 59Williams, Edwin 268Windsor, Evelyn 21Wuethrich-Grisch, Mena 220–1, 247Wurmbrand, Susan 248

Yates, Alan 3, 59

Zec, Draga 43–4, 50

Zide, Arlene R. K. 262–3

Index of Languages

Afar 85Ainu 279Albanian 111, 113Aleut 268–9Algonquian languages 86, 244

See also Fox; Menomini;Potawatomi

Balkan languages 79, 85, 111, 113,124–5

Balto-Slavic languages 86Bella Bella See HeiltsukBreton 5, 178, 193–204, 210–11, 224,

239Bulgarian 111–13, 131–2, 141,

146–7, 151, 240, 241

Catalan 3, 59, 60Chamicuro 64Chamorro 87–8, 105, 114–16, 123,

136Chuckchee 258–9Czech 147–8

Ditidaht 21Dolomitic Ladin 204Dutch 61–3

Limburg dialect 248–9

EnglishEastern Massachusetts dialect 49Middle 35Old 35

Eskimo See West GreenlandicEskimo-Aleut languages 260, 268

Finnish 80, 235–9, 242Flemish 248Fox 87Franco-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, 163Kashimiri 178, 187–93, 214–15, 224Kugu Nganhcara 79, 82Kwakiutl See Kwakw,alaKwakwakawakw See Kwakw,alaKwakw,ala 3, 5, 14–5, 17–25, 58–9,

63, 79, 81, 89, 101–105, 112–14,269

Latin 2, 44, 216, 220early 108

Lithuanian 86

Macedonian 111, 113, 124, 131–2,146

Madurese 10Makah 14Maliseet 244Mayali 278Menomini 87, 244Mixtec 11, 12Mohawk 18, 230–1, 233–5, 258,

262–8, 270–81, 286

Nahuatl (Huauhtla) 259, 263, 272Classical 258

Ngiyambaa 75, 125Nitinaht See DitidahtNootka See Nuu-chah-nulthNuu-chah-nulth 14

Oowiky’ala 14, 20Oweekeno See Oowiky’alaOwikeno See Oowiky’ala

Pashto 81, 152–7, 164Pemon 229–30. 235, 242Polish 40–2, 58–9Portuguese 81, 85, 148, 151, 157–60,

162Claasical 149European 85, 148–9, 151–2,

157–8, 160–1, 164Potawatomi 86–7

Pukapukan 98–9Puter 219–20, 250

Rembarrnga 274Rhaeto-Romance 205Romance languages 3, 58, 80, 84–5,

146–7, 158, 223–4, 240Romanian 111, 113Rotuman 99, 100Rumantsch 204–5, 219, 223, 242,

247, 250Engadine 219, 250Swiss 205See also Puter; Surmiran;

Surselvan; Vallader

Sanskrit 9, 126, 177Vedic 108

Seediq 76Serbo-Croatian 49–50, 81, 110–12,

116–23, 129, 132, 144, 147, 150,186

Belgrade 50Eastern Herzegovina 50eastern (“Serbian”) dialects 111Neo-Štokavian 49–50Srem, Macva 50western (“Croatian”) dialects 111,

144Sora 262Southern Paiute 258–9Southern Tiwa 258, 272–3, 278Spanish 168, 205, 219, 228–9, 241–2,

245Castilian 241Latin American 242

Sundanese 135–6Surmiran 6, 204–24, 242–3, 247–51,

254Surselvan 220, 250

Tagalog 32, 76, 126–7, 133, 141,165–176, 179, 227, 240, 245,253

Index of Languages 317

Tiwi 262Tongan 94–99Tsez 248

Udi 161–4Uralic languages 79

Vallader 219–20, 250Vogul

Northern 81

Wakashan languages 14, 20–1,260

Warlpiri 77, 84, 139–40, 174, 227,240

Welsh 85West Greenlandic 109–10, 143, 174,

260, 270–86

Yagua 81

Yir-Yoront 125

aspects of the theory of clitics

Documents