The Syntax of the Celtic Languages: A Comparative Perspective

This volume brings together ten chapters on the Celtic languages, using the insights of principles-and-parameters theory. The leading researchers in the field examine Welsh, Irish, Breton and Scots Gaelic in a comparative perspective, making reference to recent work on English, French, Arabic, German and other languages. The editors have provided a substantial introduction which seeks to make the volume accessible to theoreticians unfamiliar with the Celtic languages and also to Celtic specialists who are less familiar with the theoretical framework underpinning the work. The syntax of the Celtic languages makes a substantial contribution both to linguistic theory and to our understanding of the Celtic languages. It will appeal to those interested in comparative syntax and to those specializing in the Celtic languages.

The syntax of the Celtic languages

The syntax of the Celtic languages A comparative perspective

edited by ROBERT D. BORSLEY and IAN ROBERTS University of Wales, Bangor

CAMBRIDGE UNIVERSITY PRESS

CAMBRIDGE UNIVERSITY PRESS Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, Sao Paulo Cambridge University Press The Edinburgh Building, Cambridge CB2 2RU, UK Published in the United States of America by Cambridge University Press, New York www. Cambridge. org Information on this title: www.cambridge.org/9780521481601 © Cambridge University Press 1996 This publication is in copyright. Subject to statutory exception and to the provisions of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published 1996 This digitally printed first paperback version 2005 A catalogue recordfor this publication is available from the British Library Library of Congress Cataloguing in Publication data The syntax of the Celtic languages: a comparative perspective / edited by Robert D. Borsley and Ian Roberts. p. cm. Includes bibliographical references and index. ISBN 0 521 48160 0 (hardback) 1. Celtic languages — Syntax. I. Borsley, Robert D. II. Roberts, Ian G. PB1071.S96 1996 491.6-dc20 95-14193 CIP ISBN-13 978-0-521-48160-1 hardback ISBN-10 0-521-48160-0 hardback ISBN-13 978-0-521-02324-5 paperback ISBN-10 0-521-02324-6 paperback

Contents

List of contributors Introduction Robert D. Borsley and Ian Roberts

page viii 1

1 Long head movement in Breton Robert D. Borsley, Maria-Luisa Rivero and Janig Stephens

53

2 Some syntactic effects of suppletion in the Celtic copulas Randall Hendrick

75

3 Fronting constructions in Welsh Maggie Tallerman

97

4 Bod in the present tense and in other tenses Alain Rouveret

125

5 Pronominal enclisis in VSO languages Ian Roberts and Ur Shlonsky

171

6 Aspect, agreement and measure phrases in Scottish Gaelic David Adger

200

7 A minimalist approach to some problems of Irish word order 223 Jonathan David Bobaljik and Andrew Carnie 8 Subjects and subject positions in Irish James McCloskey 9 Negation in Irish and the representation of monotone decreasing quantifiers Paolo Acquaviva 10 On structural invariance and lexical diversity in VSO languages: arguments from Irish noun phrases Nigel Duffield

241

284

314

References

341

Index

357

Contributors

Paolo Acquaviva, University College Dublin David Adger, University of York Jonathan David Bobaljik, Massachusetts Institute of Technology Robert D. Borsley, University of Wales, Bangor Andrew Carnie, University of Calgary Nigel Duffield, McGill University Randall Hendrick, University of North Carolina, Chapel Hill James McCloskey, University of California, Santa Cruz Maria-Luisa Rivero, University of Ottawa Ian Roberts, University of Wales, Bangor Alain Rouveret, Universite de Paris VIII Ur Shlonsky, Universite de Geneve Janig Stephens, Cardiff Institute of Higher Education Maggie Tallerman, University of Durham

Introduction Robert D. Borsley and Ian Roberts

1 Preliminary remarks This book grew out of a conference on Comparative Celtic Syntax held at the University of Wales, Bangor, on 25-7 June 1992.1 Earlier versions of seven of the ten chapters collected here were given at that conference. The idea behind the conference was to bring together researchers working on the syntax of the Celtic languages from a 'principles-and-parameters' perspective (the assumptions behind this perspective are outlined below in section 2.1), and, in particular, to provide a forum where comparative work on Celtic syntax could be presented. The comparative work was intended to be both internal and external to the Celtic family. Hence, one goal of the conference was to encourage those working on Celtic to make comparisons with non-Celtic languages, and to bring relevant phenomena and analyses of Celtic languages to the attention of those working on non-Celtic languages. Although the precise contents differ from the conference, and this volume should not be taken as a conference proceedings, we have compiled this collection with the same general goals in mind. This introduction is intended to provide the background to the chapters that follow, both for those who are unfamiliar with the principles-and-parameters framework and for those who are unfamiliar with the Celtic languages. In this section, we briefly sketch the historical, geographical and social situation of the languages. Section 2 provides background to the principles-and-parameters framework. This section is of most relevance for readers who may be familiar with the languages but who are less familiar with this framework. Sections 3 and 4 then discuss a number of aspects of the syntax of Celtic that are of particular interest from a principles-and-parameters perspective, either for general theoretical reasons or because of the comparative interest of the particular Celtic phenomena in relation to other languages. These sections are of most relevance to readers who are familiar with work on other languages in the principles-and-parameters framework, but who are less familiar with the Celtic languages. Of course, none of these introductory sections is exhaustive: a complete introduction to the syntactic framework is given in Haegeman (1994), and fuller discussions of Celtic syntax are provided in the chapters themselves, as well as in the collections in Hendrick (1990b) and the special issue of Natural Language and Linguistic Theory devoted 1

2

Robert D. Borsley and Ian Roberts

to these languages (NLLT 7.3, 1989). A general survey of the Celtic languages, covering all descriptive areas as well as sociolinguistic and dialectological questions, is MacAulay (1992). The Celtic languages are a subgroup of Indo-European. Certain morphological similarities have led to the suggestion that Celtic and Italic form a common subgroup, although this view is no longer widely held. The Celtic languages are grouped together on the basis of shared innovations with respect to IndoEuropean, such as the absence of a reflex of Indo-European *p in initial and medial positions (cf. Irish athair 'father' vs. Latin pater, English father, etc.), a number of which are not found elsewhere in the family. Within Celtic, the basic division is between Continental and Insular Celtic. Continental Celtic refers to the languages spoken by the Celts in continental Europe between roughly 500BC and 500AD. These languages are only known through Latin and Greek versions of proper names, and a small number of inscriptions from France and northern Italy. The principal Continental Celtic languages were Gaulish, Celtiberian and Lepontic. Since, given the nature of the attested evidence, very little can be known about the syntax of these languages, they are not mentioned in any of the chapters in this collection. Insular Celtic refers to the Celtic languages that are historically rooted in the British Isles (including Breton, which is the result of fifth- and sixth-century migrations from Cornwall). These languages are divided into two groups: Brythonic (or Brittonic or British) and Goidelic (or Gaelic). The principal distinguishing feature of the two groups is the reflex of Indo-European *kw, which in Brythonic gives p and in Goidelic gives k: for example, Welsh pump ('five') corresponds to Irish coic (cf. Latin quinque). For this reason, the two branches are sometimes referred to as T-Celtic' and 'Q-Celtic' respectively (the term 'Q-Celtic' originates from the Latin transliteration of the Ogam symbol for /ku/). There are six Insular Celtic languages: three Goidelic and three Brythonic: the three Goidelic languages are Irish, Manx and Scots Gaelic; the three Brythonic ones are Breton, Cornish and Welsh. Of these six languages, Cornish and Manx are no longer spoken. The remaining four languages are the only ones that still have native speakers, and these are the only ones that are discussed in this volume. We will now briefly sketch the historical, geographical and social situation of each one. For more details, see the respective chapters of MacAulay (1992). Irish, as the official first language of the Republic of Ireland, is the only Celtic language to be a national standard. It is not known when the Celts migrated to Ireland, but they were certainly there in Classical times. The earliest attestations of Insular Celtic are the forms of Irish recorded in the indigenous Ogam script (approximately 300-500). Old Irish (600-900) provides the oldest body of literary texts in Celtic. Until the consolidation of colonial English power and the transplantation of English-speaking migrants to Ireland in the seventeenth century, Irish was the language of essentially the whole island of Ireland. From

Introduction

3

the beginning of the colonial period, Irish declined. In 1851, at the first census which asked about language, only 23 per cent of the population were Irishspeaking. At present, between 30,000 and 70,000 native speakers of Irish remain (estimates vary), scattered in small communities in the extreme west of the country. Scots Gaelic was imported into Scotland from Ireland some time before the fifth century AD. It has always been linguistically very close to Irish, especially to the Ulster dialects, and until the seventeenth century the written language was essentially Irish. It has never been the language of the whole of Scotland: Pictish was spoken in Classical times, although it was eventually replaced by Gaelic. Brythonic languages were also spoken in the south until roughly the eleventh century. A northern dialect of English has been spoken in the south-east, around Edinburgh, for over a thousand years. The last few centuries have seen a steady retreat of Gaelic to the north and west, aided by official policy from the seventeenth century to the twentieth. The number of speakers has diminished from about 300,000 in 1800 to about 80,000 in 1981. The largest proportion of these speakers is in the Western Isles. The local authority in this area has an official bilingual policy, and there is some bilingual education. Breton was, as mentioned above, imported into north-western France from Cornwall around the fifth and sixth centuries. Although Brittany was largely politically independent until it was absorbed into France in 1532, the ruling classes were probably French-speaking much earlier. The earliest connected Breton texts date from the fifteenth century, and the orthography was standardized in the early nineteenth century. Until 1951 Breton was prohibited in schools, and children were punished if they were heard speaking it. The traditionally recognized linguistic frontier separating Breton from French runs across the peninsula from St Brieuc in the north to Vannes in the south. West of this line, Breton is spoken in rural communities, particularly inland. No statistical evidence is available about Breton, since it has no official status within the French Republic. Estimates of the number of speakers vary from 400,000 to 600,000, though all agree that the figure is declining rapidly. Welsh is the linguistic survivor of the Anglo-Saxon invasions of the main island of Britain, which began in the fifth century AD. Because its immediate ancestor was spoken throughout England during the period of Roman administration (43-405), Welsh contains a large number of Latin loan words. The period from the fifth to the seventeenth century was one of steady Anglo-Saxon encroachment westwards. Old Welsh is attested from the eighth century, by which time the language was confined to Wales, having been separated from Cornish in the south in the sixth century. Wales ceased to be an independent country in 1288, and lost all political autonomy with the Act of Union in 1536. However, the form of Modern Literary Welsh wasfixedby the Bible translation of 1588. The Methodist revival of the eighteenth century arguably saved Welsh, spreading both literacy

4

Robert D. Borsley and Ian Roberts

and liturgy in the language. Until the industrial revolution, almost all of Wales was wholly Welsh-speaking. In 1921, 37.1 per cent of the total population was Welsh-speaking; this had declined to 18.9 per cent by 1981 (a total of 503,549 people). Since 1967, the language has had official status throughout Wales. Bilingual schools are found everywhere, and there has been a Welsh television station since the 1980s. Although its long-term future is far from secure, Welsh appears to be the most firmly established and widely spoken Celtic language at present. This concludes our brief overview of the languages that are discussed here. In the next section, we turn to more theoretical questions, providing a sketch of the basic assumptions about syntactic theory that are common to the chapters collected here.

2 Theoretical framework 2.1 Principles and parameters

All the contributions to this volume assume the 'principles-and-parameters' approach as the framework for the study of comparative syntax. Here we very briefly describe what that approach involves. In the present context, we cannot give a full introduction; for a full discussion and illustration of the concept of a parameter of Universal Grammar, see the introduction to Jaeggli and Saflr (1989), and for a general introduction to the principles and parameters approach, see Haegeman (1994). Chomsky was the originator of the notion of Universal Grammar in the context of modern linguistics (see in particular Chomsky 1975, 1980, 1986a, 1988 for a fuller discussion and defence of this idea). Universal Grammar (UG) consists of an invariant core of constitutive principles which are common to all possible human languages. These principles are innately given: that is, they form part of the human genetic endowment. In order to account for the attested variation among the world's languages, these principles are associated with parameters of variation which make it possible for a principle to be realized in different ways in different languages. Since the principles of UG are fairly abstract in nature, a minimal difference in the value of an associated parameter in two different languages may give rise to dramatic surface differences in the wellformed sentences of the languages. In this way, the principles-and-parameters approach is able to account for the differences among languages while maintaining the idea that all languages are cut from the same cloth. An example of the interaction of principles and parameters comes from certain well-known differences between French and English regarding the position of the inflected main verb in finite clauses (see Emonds 1978; Pollock 1989). It was

Introduction

5

originally argued by Emonds (1978) that French has a rule moving finite verbs out of VP, while English does not. The basic form of the observation is as follows: there is a class of elements X that can be plausibly regarded as positioned on the left edge of VP. These elements include VP-adverbs, clausal negation and floated quantifiers. In French, finite main verbs must precede X, while English main verbs always follow X. The relevant paradigms are as follows: (1)

Adverb a. Jean embrasse souvent Marie. *Jean souvent embrasse Marie. b. *John kisses often Mary. John often kisses Mary.

(2)

Negation a. Jean (ne) mange pas du chocolat. *Jean (ne) pas mange du chocolat. b. *John eats not chocolate. John does not eat chocolate.

(3)

Floated quantifiers a. Les enfants mangent tous le chocolat. *Les enfants tous mangent le chocolat. b. T h e children eat all chocolate. The children all eat chocolate.

The evidence clearly shows that finite verbs are in different positions in the two languages. The alternative is to suggest that the X-elements differ between the two languages (this has been suggested by Williams 1994). The usual account of these differences is that French verbs move to I (for Inflection), a node comparable to the Aux node of early transformational work which contains features of tense, agreement, etc. in a tree structure such as the following (see the discussion of clause structure in section 2.2):

6 (4)

Robert D. Borsley and Ian Roberts IP

Les enfants mangent tous

I

J

le chocolat

Here, the V-to-I rule places the finite verb in a position preceding X. The operation of this rule in French thus derives the orders seen in (l)-(3) above, and its non-operation in English derives the English orders seen in those examples. Standard assumptions about subject-verb inversion deriving from the seminal work of den Besten (1983) treat this operation as involving movement of I to C. Given the Head Movement Constraint (see Travis 1984; Baker 1988; and section 2.4 below), V cannot move directly to C, and so inversion of main verbs depends on the prior operation of V-to-I movement to feed it. Thus we find that French main verbs are able to undergo inversion (subject to the independent restriction that the subject be a clitic - see Rizzi and Roberts 1989), while English main verbs are unable to do so: (5)

a. Voit-il le cheval? b. *Sees he the horse?

The contrast in (4) is further evidence that French main verbs move to I while their English counterparts do not.2 Chomsky (1993) proposes that the relevant parameter concerns the value of an abstract morphological feature that licenses verbs, and is associated with I. This feature is called Fs V-feature. In Chomsky's system, such features are generated both on V and on I, and must be cancelled out by a checking operation prior to LF since they have no semantic content and will thus violate the Principle of Full Interpretation, which applies at LF, unless eliminated. The feature varies parametrically as either strong or weak. If it is strong, it is visible to the PF component, and hence must be eliminated prior to the mapping to that level of representation, that is, prior to S-structure. Since feature-checking takes place in a highly local domain, V must move to I in order for feature-checking to take place. Thus where the V-feature is strong, V raises overtly to I. Where the feature is

Introduction

7

weak, the Procrastinate Principle, which delays movement to the covert, postSpell-Out part of the grammar wherever possible, prevents this movement from taking place overtly. In these terms, then, the UG principle is that V-features must be checked, and the parametric variation consists in French I having a strong V-feature and English I a weak V-feature. This outline of the basic ideas of the principles-and-parameters approach to cross-linguistic variation and typology, although extremely sketchy and simplified, is enough for our purposes here. One important facet of this approach is that it makes it possible in principle to isolate structural isomorphisms among unrelated languages: that is, we expect to find interesting similarities between languages with no historical or other connection. This can be attributed to their choosing the same parameter settings in a given domain. In this way, very enlightening and unexpected comparisons become possible. The present volume offers two such instances: the similarities between Celtic languages and Semitic languages in the nature of pronouns and word order studied by Roberts and Shlonsky, and the iong head movement' construction shared between Breton and various Slavonic and Romance languages studied by Borsley, Rivero and Stephens. Duffield's article also brings out similarities among Irish, Hebrew and Maltese. We now have a notion of how current theory accounts for syntactic differences and similarities between languages. The principles-and-parameters approach was designed to account for synchronic variation, and, as the chapters presented here attest for the case of a relatively little-studied and somewhat 'exotic' language family, it provides the right kind of restricted yet flexible analytic framework for this.

2.2 Clause structure and functional categories

Most of the work contained in this volume takes its fundamental impetus from the recent upsurge of research on clause structure. The current interest in clause structure revolves around a hypothesized system of non-lexical categories which carry essentially 'morphological' information: the functional categories. This work was stimulated initially by Chomsky (1986b), and the approach was further developed by Pollock (1989) and Chomsky (1991, 1993). Work on functional categories inside nominals has also been very important; we will discuss this in section 2.3. The X-bar schema gives the internal structure of syntactic categories.3 (6)

a. XP -> b. X' -+

YP X

X' ZP

8

Robert D. Borsley and Ian Roberts

Originally this schema was taken to apply only to the lexical categories N, V, A and P. It was assumed that the 'clausal' categories S and S' were generated by the following PS-rules: (7)

a. S ' ^ C O M P S b. S - » N P A u x V P

This was clearly an undesirable state of affairs given the general programme, initiated by Chomsky (1981) and Stowell (1981), for the elimination of independent PS-rules from the theory. Accordingly, Chomsky (1986b) proposed that S and S' are projections of the non-lexical categories I(nfl) and C(omp), respectively. As we mentioned above, I corresponds roughly to the Aux node of earlier work, and contains in particular features specifying tense and agreement; this position had originally been argued to be the head of S in Hale, Jeanne and Platero (1977). In the system of Chomsky (1986b), the specifier of I is the subject position, and the complement of I is VP. C is the earlier COMP position; its specifier is the landing site for w/z-movement and (in verb-second languages see below) fronted topics, and its complement is IP. So we have the following clause structure:

Here CP and IP are functional categories, and together make up what is sometimes called the 'functional domain' of the clause. All the chapters in this volume assume some version of the structure in (8), frequently with further elaboration. One important result of Pollock's (1989) work on verb movement in English and French is the 'split-InfT hypothesis. The initial evidence for this comes from the behaviour of French infinitives. First, Pollock observes that French infinitives show the same split between auxiliaries and main verbs as English finite verbs, in that only the auxiliaries etre and avoir can move over the negative pas (see Emonds' (1978) rule of have•/'be raising which applies in English finite clauses): (9)

a. N'etre pas content est une condition pour ecrire. T o be not happy is a condition for writing.' b. *Ne sembler pas content . . . T o seem not happy . . .'

Introduction

9

However, the situation regarding the placement of main-verb infinitives in relation to adverbs is more complex. While infinitives cannot raise over negation, they can precede some adverbs, for instance: (10)

a. A peine parler l'italien apres cinqans d'etude . . . hardly to-speak Italian after five years of study . . . b. Parler a peine l'italien apres cinq ans d'etude . . .

Pollock is led to propose a 'short' movement of main-verb infinitives. Following the Structure Preservation Condition of Chomsky (1986b), the landing-site of this movement must be a head. Pollock capitalizes on the fact that I node of Chomsky (1986b) was a rather unnatural combination of the features of Tense (T) and Agreement (Agr), and proposes that these two kinds of features should each project their own X-bar structure. This gives the two separate functional projections TP and AgrP. The 'short' movement of main-verb infinitives in French is then seen as movement to the lower of these two heads, while the longer movement of tensed main verbs in French is to the higher of these heads. Pollock assumed that TP dominates AgrP, so this gives the following clause structure: (11)

Pollock's 'split-Infl' hypothesis has given rise to a vast amount of work on basic clause structure and functional categories. Almost any property that can be reasonably ascribed to an auxiliary system - Aspect, Modality, Negation, Voice, etc. - has been associated with its own functional category. Thus, just as it is unclear what the inventory of 'possible auxiliary notions' is, it is unclear what the full inventory of functional categories might be. It would be impossible here to outline all the proposals for clause structure and functional categories that have been made in the past few years. However, two elaborations of Pollock's system are worthy of particular note and so we briefly discuss them here. Belletti (1990) argues that AgrP should be taken to dominate TP, apparently the inverse of Pollock's split-Infl structure. Chomsky (1991) elaborates Belletti's proposal further by suggesting that there are separate functional projections for subject and object agreement: there is an Agr-projection above T which is the

10

Robert D. Borsley and Ian Roberts

position of affixes specifying agreement with the subject, or AgrS, and there is also an Agr-projection below TP, the position for agreement with the object, AgrO (see Kayne 1989a). The structure of the clause is thus as in (12): (12)

CP Spec

^

C

\

Spec

AgrS'

Spec

AgiO' AgiO

^ Spec

Most current work assumes this, usually with further functional projections for Negation and, frequently, a further projection between C and AgrS. The above paragraphs have been concerned mainly with verb movement within IP, and consequently have not addressed the analysis of inversion constructions, i.e. constructions in which the verb moves over the subject. The most prominent type of construction where the verb appears to regularly move out of IP is the verb-second phenomenon, found in all Germanic languages (except for contemporary English). The verb-second (or V2) construction features movement of the finite verb coupled with fronting of some XP. The precise nature of XP is immaterial; it may be the subject, a complement or an adverbial element. The following German sentences (from Tomaselli 1989), illustrate the phenomenon: (13)

a. Ich I b. Ich I

las read habe have

schon letztes already last schon letztes already last

(14)

a. Diesen Roman las ich schon letztes Jahr. this book read I already last year b. Diesen Roman habe ich schon letztes Jahr gelesen. this book have I already last year read

(15)

a. Schon letztes already last b. Schon letztes already last

Jahr year Jahr year

Jahr year Jahr year

las read habe have

diesen this diesen this

ich diesen I this ich diesen I this

Roman. book Roman gelesen. book read

Roman. book Roman gelesen. book read

Introduction

11

The topic of V2 is of considerable interest for comparative Celtic syntax; Breton appears to show a similar constraint (though not precisely what we find in Germanic) and the other languages all feature V-fronting operations which may be similar to the Germanic verb-fronting. A very influential analysis of V2 (although certainly not the only one) was first put forward in den Besten (1983). Den Besten proposed that the inflected verb moves to the C-position in matrix declaratives in V2 languages. This operation is associated with the fronting of some XP to SpecC. In this way, the root nature of the phenomenon is explained: embedded complementizers are frequently filled (and at a more abstract level of analysis, they perhaps always are) and so cannot serve as the landing site for the fronted verb. However, there are two further questions here. One concerns the treatment of SVO sentences like (13) in V2 languages - is the subject in SpecAgr' or in SpecC7? For differing points of view, see Zwart (1993) and Vikner and Schwartz (forthcoming). The other concerns the nature of the allegedly 'symmetric' V2 languages Icelandic and Yiddish, in which V2 is possible in all clause types (see Rognvaldsson and Thrainsson 1990 on Icelandic; Santorini 1989 on Yiddish). One analysis of these languages, however, assumes that verbs are not fronted to C but to some lower functional head position. For different versions of this idea, see Cardinaletti and Roberts (1991) and Diesing (1990). Another important development in our conception of clause structure is logically distinct from the elaboration of the functional structure of the clause, but has interacted with it in important ways. This is the VP-internal subject hypothesis, originally proposed in generative grammar by Fillmore (1968), and developed in recent theories by Fukui and Speas (1986), Koopman and Sportiche (1991), Kuroda (1988) and others. The central idea, as Koopman and Sportiche put it, is that one can show that I (to revert to the 'non-split' Infl for ease of presentation) has the properties of a raising predicate (see Ross 1969). These properties are (a) that it assigns no thematic role of its own, and (b) that its surface subject can be anything selected by a structurally lower predicate, including expletive elements (which are non-thematic by definition) and idiom chunks. The following comparison of a classic raising verb, seem, and what is generally taken to be an I-element, the modal will, illustrate the parallels: (16)

a. b. c. d.

John seems to speak Chinese. It seems to rain a lot here. There seems to be a problem. The cat seems to be out of the bag.

(17)

a. b. c. d.

John will speak Chinese. It will rain a lot today. There will be a problem. The cat will be out of the bag.

12

Robert D. Borsley and Ian Roberts

In the (a) sentences, the subject is clearly the agent of speak, not of seem or will. In the (b) and (c) sentences, different expletives are associated with the meteorological and existential main predicates respectively. Finally, the (d) examples show a piece of an idiom appearing as subject, without seem or will taking on any idiomatic interpretation. This parallel syntax of raising verbs and I-elements led to the idea that I is a raising predicate. In these terms, the subject must be thought of as generated inside VP and raised to its surface position (in SVO languages - see section 3). In this section we have given only the barest outline of the extent of recent work on clause structure and functional categories. The analyses to be found in the articles contained in this volume, which we will review in the next section, will refine this outline somewhat.

2.3 The structure of nominals The general tendencies seen in the development of theories of clause structure that we reviewed in the previous subsection are also apparent in recent work on nominals. Above all, functional categories have been shown to play a central role in this domain, too. The most important functional category in nominals is the Determiner Phrase, or DP. This was originally proposed by Abney (1987) and by Fukui and Speas (1986). Abney (1987) motivated the postulation of DP for the analysis of the English POSS-mg gerundives, such as John's building a spaceship. These elements have the external distribution of a nominal phrase, but an internal structure that appears to contain the VP build a spaceship. The nominal-like external distribution is shown by the fact that these gerundives can appear in NP-positions where true clauses cannot appear, for example subject position and object of preposition: (18)

a. *Did [that John built a spaceship] upset you? b. Did [John's building a spaceship] upset you?

(19)

a. *I told you about [that John had built a spaceship], b. I told you about [John's building a spaceship].

The gerundive itself, however, is verb-like in that its object is accusative rather than genitive (20a), it allows raising (20b), exceptional Case-marking (20c), double-object constructions (20d) and verb-particle constructions (20e), and is modified by an adverb rather than an adjective (20f). All these properties are characteristic of verbs as opposed to nouns: (20)

a. John's destroying/*destruction the spaceship b. John's appearing/*appearance to be dead c. John's believing/*belief Bill to be innocent

Introduction

13

d. John's giving/*gift Mary a Fiat e. John's explaining/*explanation the problem away f. John's deliberately building a spaceship. X-bar Theory prevents us from positing a structure like (21), which would otherwise be able to account for the facts in (18)—(20): (21)

Example (21) violates X-bar Theory because NP t is exocentric. A claim of X-bar Theory is that all syntactic categories are endocentric. Instead of (21), Abney argues that we should posit the following structure: (22)

John's

building a spaceship

Where the component to D is NP, we have a standard nominal. D is overtly realized by articles and some quantifiers, as in John's every book, etc. We must also postulate a Spec-head agreement relation between Dj and DP 2 in Spec, D( in order to account for the impossibility of * John's the book alongside the book. Cross-linguistic evidence supports this idea, in that many languages show agreement between possessor and possessed within nominals. Turkish does, and also shows the same agreement in POSS-/«g constructions: (23)

a. on-un el-i he-GEN hand-3sG 'his hand' b. Halil'-in kedi-ye yemek-0 er-me-dig-i H.-GEN

Cat-DAT food-ACC give-NEG-GER-3SG

'H.'s not giving food to the cat' This motivates the association of agreement features with D, and the related suggestion that English D contains abstract agreement in POSS-/«g constructions and perhaps possessives. Following the initial postulation of a functional category inside nominals, there was, as in the case of clauses, a proliferation of proposals. Here again, almost any

14

Robert D. Borsley and Ian Roberts

morphological property associated with nominals has been assigned to its own projection: Number, Gender, Case, Agreement, etc. Space prevents us from reviewing all of these proposals, although some are utilized in the following chapters (see in particular Duffield's contribution). A further parallel with work on clause structure concerns the idea that N may raise to D (or to another functional head inside DP). This idea can account for two kinds of phenomena. First, it has been used (by Delsing 1990; Taraldsen 1990) to account for postnominal articles of the type found in the Scandinavian languages. Thus, a form like hus-et ('house-the') is derived by N-movement adjoining hus to -et in D in a structure like the following: (24)

DP D

NP

-et

hus-

The phenomenon of postnominal articles can then be reduced to the operation of a movement rule. Longobardi (1994) argues that this movement rule applies at LF in those languages where it not does apply overtly, in order to give referential value to the N. Second, the Semitic construct-state construction may feature D-to-N raising. This has been argued for by Ritter (1988) and by Siloni (1991a, 1994) for Semitic, and the analysis carries over naturally to Celtic possessives (see section 3.7 below for illustration). The parallel with the way in which verb movement in clauses gives rise to VSO order is striking, especially given that constructs are found just in languages that are either fully or residually VSO: Celtic and Semitic. These issues are explored in detail by Duffield, and to some extent also by Roberts and Shlonsky.

2.4 Head movement We mentioned above in connection with (5) that English auxiliaries can undergo movement to I and therefore can undergo inversion: that is, movement from I to C. In embedded clauses, complementizers occupy C°, blocking I-to-C movement; this gives the well-known result that subject-aux inversion is blocked in many kinds of embedded clauses (e.g. */ wonder if has John left). I-to-C movement is an instantiation of a general process of head-to-head movement, itself just that variant of the general schema for movement, move-a, where the value of a is X°. Head-to-head movement, or incorporation, was

Introduction

15

explored and elaborated in detail from a theoretical point of view by Baker (1988). We will not go into the technical details of Baker's proposals here, but simply indicate the main formal properties of the operation: 1 Head-to-head movement is local. This means that V cannot move directly to C, 'skipping' I. This restriction is formulated as the Head Movement Constraint (HMC) of Travis (1984). This constraint may be violated in particular languages under particular conditions - see Borsley, Rivero and Stephens' chapter in this volume. 2 Head-to-head movement is cyclic. This is another property head movement shares with other kinds of movement. It is thus possible to reiterate the movement such that it passes through several heads; this in fact happens in subject-aux inversion, where the auxiliary moves from V to I and on to C. 3 Head-to-head movement is structure-preserving. This means that heads can only move to other heads; movement which attaches a head to XP or to X' is ruled out. It follows from structure preservation, combined with the Head Movement Constraint, that when we see a verb occupying the C position, we must conclude that it has passed through I. Thus head movement is local, cyclic and structure-preserving. All these properties are the direct result of the fact that head movement is nothing more than the instantiation of move-a where a is a head.

2.5 Clitics Clitics are 'weak', unstressed items which are dependent on another item which functions as 'host'. The dependency may be phonological, morphological or syntactic, or, in principle, any of these three together. Many different kinds of elements can be clitics: pronouns, auxiliaries, conjunctions, articles, etc. Here we restrict our attention, following a tradition of generative studies initiated by Kayne (1975), to pronominal clitics. We also restrict attention to syntactic clitics, that is, elements which can be shown to be syntactically dependent on their hosts and indeed are often held to be moved into 'special' derived positions by cliticplacement rules. For example, Kayne (1975) argued for a clitic-placement rule moving la into preverbal position from the usual postverbal object position in an example like the following: (25)

a. Jean John b. Jean John

voit sees la her

Marie. Mary voit. sees

16

Robert D. Borsley and Ian Roberts

Clitics are thus usually seen as syntactically dependent elements that appear in 'special' positions. Clitic doubling is an important phenomenon which is relevant for the Celtic languages. There are two principal kinds of doubling: that involving subjects and that involving complements. In Romance, these are easily distinguishable: subject-clitic doubling does not require a special marking of the subject, and is not sensitive to semantic properties of the subject. Subject-clitic doubling is found in most northern Italian dialects; the following example is from Fiorentino (Brandi and Cordin 1989: 113): (26)

La Maria la parla. the M. she speaks 'M. speaks.'

Object-clitic doubling is found in some Latin American dialects of Spanish, particularly in Platense. Here the object must be marked with a, and only nonreferentially quantified objects are allowed (Jaeggli 1982: 14, 45f.): (27)

a. Lo him 'We b. *^A A c. *Lo him

vimos a Guille. we-saw A Guille saw Guille.' quien lo viste? who him you-saw vi a un chico. I-saw A a boy

It is unclear how valid the Romance asymmetry between subjects and objects is cross-linguistically. Clitics and, arguably, clitic doubling are pervasive phenomena in the Celtic languages (see in particular Roberts and Shlonsky's contribution, and the discussion in section 3.8 below). Clitic doubling raises the important question of the relationship between clitics and agreement. We can illustrate the issue using Spanish data. Spanish is a nullsubject language, in that it allows finite clauses with an empty subject which is interpreted as a definite pronoun. Example (27a) illustrates this; here the subject is unambiguously first-person plural, although the relevant pronoun is not present. A standard observation, which goes back to traditional grammar, is that the verbal agreement marking permits the relevant information, here person and number features, to be 'recovered'. Hence, 'dropping' the subject pronoun is allowed. In English on the other hand, a sentence like saw John is ungrammatical because the content of the missing subject pronoun cannot be recovered, owing to the lack of relevant inflectional marking on the verb. These ideas have been formalized in terms of the notion that a phonologically empty pronoun, pro, is allowed (or 'licensed') in Spanish and similar languages. The verbal inflection plays a crucial role in licensing pro by recovering its content (see Rizzi 1986;

Introduction

17

Jaeggli and Saflr 1989). If the agreement inflection licenses pro in subject position in (27a), could we not say that the clitic licenses pro in object position in a comparable example without the overt direct object, such as (28)? (28)

Lo vimos. him we-saw 4 We saw him.'

The clitic lo gives information about the person, number and gender of the direct object, and hence, just like the subject agreement, permits the content of a putative object pro to be recovered. On this view (28) is to (27a) for the object as (27a) is to (29) for the subject: (29)

Nosotros lo vimos a Guille. we him saw A Guille 'We saw Guille.'

In other words, pro can alternate with an overt DP in a subject or object position, as long as there is some morphological marker to permit recovery of its content. At least functionally, then, lo is acting just like an agreement inflection on this kind of analysis. This leads naturally to the suggestion that, at least in languages where clitic doubling is tolerated, it may be correct to regard clitics as a kind of Agr. There are various properties, however, that distinguish clitics from agreement inflections. First, clitics do not condition morphological variation in their hosts, while agreement morphemes may do this (see Zwicky and Pullum 1983). Second, agreement marking is always obligatory, while clitics are optional. This is true in the case of Spanish direct objects just given: lo can be dropped in (29), but the agreement marking on the verb cannot (although, perhaps significantly, lo cannot be dropped where the object is a strong pronoun: *(Lo) vimos a el 'We saw him'). Third, as we have just mentioned, some clitics prohibit doubling, but agreement almost always allows it. The Celtic languages constitute potentially significant counterexamples to the last two claims: certain clitics are just as obligatory as agreement marking in these languages, and subject agreement is in complementary distribution with overt non-pronominal subject DPs - see section 3.8 and the contributions by Adger, Duffield and Roberts and Shlonsky. Afinalissue connected with clitics stems from these last considerations. What is the correct derivation of clitics? We mentioned above that Kayne (1975) proposed a clitic-placement transformation. Although the details of this operation should be revised (it has become clear that French object clitics do not left-adjoin to V but to a functional category), the basic idea that syntactic clitics undergo syntactic movement is attractive for several reasons. First, it accounts for the complementary distribution of clitics and direct object DPs (this argument clearly falls foul of the clitic-doubling phenomenon, but is valid for French).

18

Robert D. Borsley and Ian Roberts

Second, clitic placement obeys standard conditions on movement, such as the Specified Subject Condition (SSC). The SSC prevents movement of certain classes of elements across a subject. French paradigms like the following indicate that clitic movement is subject to the SSC: (30)

a. Jean a laisse Pierre parler a Marie. Jean has let P. speak to M. b. Jean l'a laisse lui parler. J. him-has let to-her talk 'J, let him talk to her.' c. *Jean lui a laisse Pierre parler. J. to him has let P. speak

In (30c) clitic placement would move lui over the subject of the embedded clause, in violation of the SSC. If we assume that clitics are derived by movement, we can account for the impossibility of this example. Third, clitic objects trigger participle agreement while in situ objects do not (in some varieties of French): (31)

a. Jean a peint(*e) la porte J. has painted( + FEM) the door(FEM) b. Jean Fa peint*(e). J. it(FEM) has painted( + FEM) 'John has painted it.'

This can be accounted for if we assume, following Kayne (1989a), that the clitic must transit through SpecAgrOP and hence triggers object agreement. The agreement is thus a regular case of Specifier-head agreement, analogous to standard subject-verb agreement (which takes place at the AgrSP level; see section 2.2). On the other hand, certain considerations speak for a base-generation approach. First, there is the existence of clitic doubling of the type described above. Second, even in languages like French which lack clitic doubling we find clitics that are not readily linked to a grammatical function. One example is the ethical dative, as in the following (the example and the idiomatic translation are from Sportiche 1992:18): (32)

Je t'acheterais un cadeau a Pierre. I you-would-buy a present to P. 'I tell ya, I would buy P. a present.'

It is unclear what a plausible launching-site for clitic placement of te would be in this case. If clitic placement is a movement operation, we can ask what kind of movement operation it is. An influential proposal is that clitics are Ds, like other kinds of pronoun, heading DPs of their own. In that case, clitic movement could be A-movement (as proposed in Kayne 1975; the argument based on SSC given

Introduction

19

above depends on this idea), A'-movement (this is plausible to the extent that clitics move to adjoined positions - see Chomsky 1981) or head movement, that is, D-movement (or Agr-movement - see above and section 3.8). In the last case, however, we need to explain (a) how clitics are subject to the SSC, since this constraint does not apply to head movement; (b) why clitic movement appears to violate the Head Movement Constraint, since, if it moves from object position to a functional head, it 'skips' V; (c) how clitic movement triggers participle agreement. All of these questions can be answered if we treat clitic movement as a combination of A-movement and D-movement - see Sportiche (1988). While such an analysis may work for Romance, it is not clear whether it is appropriate for Celtic - see section 3.8.

3 Some issues in Celtic syntax Having introduced the basic approach and some of the main theoretical ideas assumed by the chapters in this volume, we turn now to the syntax of the Celtic languages. In this section, we focus on areas of syntax with which the following chapters are concerned, and in the process introduce the chapters. In the next section, we consider some further features. We will concentrate on three languages: Welsh, Breton and Irish. Scots Gaelic is in most respects like Irish. Hence, we will largely ignore it. 3.1 VSO clauses The most discussed feature of the Celtic languages is their VSO clause structure, and this is something that a number of the chapters in this volume are concerned with. All the languages have VSO as their normal order in root and subordinate finite clauses of all kinds. The following illustrate VSO root clauses. (33)

(34)

(35)

Gwelai Emrys ddraig. would-see Emrys dragon 'Emrys would see a dragon.' Email Yann war an hent. is Yann on the road 'Yann is on the road.' Bheadh se ann. would-be he there 'He would be there.'

(Welsh)

(Breton)

(Irish)

We will see later that there are important complications in the case of Breton. In early work, the central question was: is the VSO order basic or is it derived through a V-fronting process? The VSO order was assumed to be basic in Awbery

20

Robert D. Borsley and Ian Roberts

(1976) and McCloskey (1979). However, works such as Jones and Thomas (1977), Emonds (1981) and Harlow (1981) argued in favour of a V-fronting analysis and within transformational approaches this analysis is generally accepted. For Harlow (1981), example (33) has something like the following structure: Welsh

(36)

Gwelodd

Emrys

t

ddraig

In more recent work in which a more complex picture of clause structure is assumed (see section 2.2), the main questions have been: where do V and the subject appear in VSO clauses? and how does V-fronting in Celtic VSO clauses compare with V-movement in verb-second languages such as German and certain SVO languages such as French? A priori, there are a number of possible analyses for VSO clauses. One possibility is that V appears in C, as it is generally assumed to do in verb-second languages. This is proposed for Irish in Stowell (1989) and Doherty (1992). This would give the following structure for (33): Welsh

(37)

gwelodd

Emrys

ddraig

However, as noted by Guilfoyle (1990) and Bobaljik and Carnie (this volume), this seems implausible given that the Celtic languages have VSO order not only in main clauses but also in subordinate clauses. The following illustrate: (38)

Dywedodd Megan y gwelai Emrys ddraig. said Megan PRT would-see Emrys dragon 'Megan said Emrys would see a dragon.'

(Welsh)

Introduction (39)

(40)

21

Gouzout a ran eman Yann war an hent. know PRTdo.lsG is Yann on the road 4 I know Yann is on the road.' Shil me go mbeadh se ann. thought I PRT would-be he there 'I thought he would be he there.'

(Breton)

(Irish)

In contrast, the main verb-second languages only have verb-second order in main clauses. (There are two exceptions, Yiddish and Icelandic; see the discussion in section 2.2.) McCloskey (forthcoming) argues on the basis of preposed adverbials that verbs are in I in Irish. He observes that preposed adverbials precede the verb in subordinate clauses. The following illustrates:

(41)

Deiridis an cheadNollaig eile go dtiocfadh se anios. they-used-to-say the first Christmas other PRT would-come he up 'They used to say that next Christmas he would come up.'

On the assumption that such adverbials are adjoined to IP, the verb must be in I. It is possible to maintain that V is in Infl in VSO clauses given the VP-internal subject hypothesis outlined in section 2.2. It is possible to assume that V is in Infl, as it is in French and certain other languages, and that it is followed by the subject because subjects, unlike in SVO languages, remain within VP. It is possible, then, to propose something like the following structure for (33): Welsh

Gwelodd

Emrys

ddraig

This is similar to the analysis that is proposed for Welsh in Rouveret (1990), and similar analyses are proposed for Irish in Guilfoyle (1990) and McCloskey (1991b). Within such an analysis, one can claim that subjects remain in VP because case is assigned under government (Guilfoyle 1990; Koopman and Sportiche 1991) or because the N-features of I are weak so that the subject does not have to move to SpecIP until LF (Chomsky 1993).

22

Robert D. Borsley and Ian Roberts

It seems, however, that this analysis is too simple at least for Welsh and Irish. In both languages, there is evidence that subjects are outside VP. The evidence comes from sentences in which subjects precede elements which it is plausible to assume are outside VP. In Welsh, we have negative sentences like the following: (43)

Welodd Emrys ddim draig. saw Emrys NEG dragon 'Emrys didn't see a dragon.'

(Welsh)

Here, we see that the subject precedes the negative particle ddim. If we assume that ddim is outside VP, the subject must be too since it precedes ddim. In Irish, we have examples like the following highlighted in McCloskey (this volume): (44)

Nior shaotaigh Eoghan ariamh pingin. NEG earned Owen ever penny 'Owen never earned a penny.'

(Irish)

Here, the subject precedes the adverb ariamh 'ever'. If this is outside VP, the subject must be too. The split-Infl clause structure discussed in section 2.2 offers a solution to this problem. As Rouveret (1991) notes, it is possible for V to be in one functional head position and the subject to be in the specifier position of a lower functional head. If we assume just two functional heads instead of Infl, Fl and F2, this would give the following structure for (33): Welsh

F1P

(45)

gwelodd

Emrys

t

t

V,

DP

t

ddraig

We would have a similar structure for (35). What about Breton? Here, we may have a rather different situation. In Breton, a subject follows the counterpart of ddim, ket, as (46) illustrates. (46)

Ne lenn ket ar vugale levriou. PRT read not the children books 'The children do not read books.'

(Breton)

It looks, then, as if subjects may be inside VP in Breton.

Introduction

23

Further arguments that subjects are outside VP in Irish are presented by Bobaljik and Carnie and McCloskey in the present volume. Bobaljik and Carnie argue that objects of Irish VSO clauses are outside VP. If this is right, subjects must also be outside VP. McCloskey argues that subjects cannot be in SpecVP because a DP complement must move into subject position in unaccusatives and perfective passives and such obligatory movement must be to the specifier position of a functional head. This line of argument suggests that subjects may also be outside VP in Breton since Breton also seems to have obligatory movement to subject position in unaccusatives and passives. The obvious question about the structure in (45) is: what exactly are Fl and F2? The obvious answer is that one is T(ense) and the other AgrS. It is not at all obvious, however, which is which. For Welsh, Rouveret (1991) and Hendrick (1991) argue that the higher head is AgrS, but Tallerman (1993a) argues that it is T. For Irish, Duffleld (1990) and McCloskey (this volume) argue that the higher head is T, but Bobaljik and Carnie (this volume) assume that it is AgrS. McCloskey's argument that T is above AgrS in Irish involves the properties of unaccusatives. He points out that Irish has unaccusatives in which the single argument of the verb is a subject, like (47), and unaccusatives in which the single argument is the object of a preposition, like (48). (47)

(48)

Mheadaigh mo shaibhreas. increased ISG wealth 'My wealth increased.' Mheadaigh ar mo shaibhreas. increased on ISG wealth 'My wealth increased.'

However, Irish has no unaccusatives like French // est arrive trois hommes Three men have arrived', in which the single argument is a complement of the verb. He notes that the situation is similar with perfective passives. He argues that the data can be explained on the assumption that Irish does not have expletives. He proposes that this is because TP is above AgrSP in Irish and expletives are associated with SpecTP, a position which is never overtly realized in Irish.

3.2 Breton VAuxSO clauses All the Celtic languages have VSO finite clauses, but Breton also has a distinctive VAuxSO clause type. This is the main focus of Borsley, Rivero and Stephens' chapter. The following are typical examples. (49)

Lennet en deus Yann al levr. read 3SG.M has Yann the book 4 Yann has read the book.'

(Breton)

24

Robert D. Borsley and Ian Roberts

(50)

Lenn a ra Anna al levr. read PRT does Anna the book 'Anna reads the book.'

(Breton)

They argue that it involves long head movement', a process which moves a V directly to C° across certain intervening heads. They point out that this process is found in a variety of languages, including Bulgarian, Czech, Serbo-Croatian and Old Spanish. The following illustrate its operation in Bulgarian and Czech. (51)

(52)

Procel sum knigata. read have. 1 SG book-the 'I have read the book.' Koupil jsem knihy. bought have.lsG books 'I have bought books.'

They argue that long head movement is one way of licensing Tense and that it is a last-resort process that does not apply if Tense is not licensed in some other way. This accounts for the fact that it does not apply in subordinate clauses, negatives and sentences where SpecCP is filled by a topic or a w/j-phrase. In all these clause types the verb follows the auxiliary (and precedes the subject). (53)

(54)

(55)

Lavaret he deus Anna [en deus lennet Tom al levr] said 3SG.F have Anna 3SG.M have read Tom the book 'Anna said Tom had read the book.' N' en deus ket lennet Tom al levr. NEG 3SG.M has NEG read Tom the book Tom has not read the book.' Al levr en deus lennet Tom. the book 3SG.M has read Tom Tom has read the book.'

(Breton)

(Breton)

(Breton)

They argue that the licensing conditions associated with Tense account for the fact that finite verbs and auxiliaries are generally impossible in initial position in root clauses. The copula in (34) is the main exception to this generalization. The following illustrate the general situation: (56)

(57)

*Lenn Anna al levr. read Anna the book 'Anna reads the book.' *En deus lennet Tom al levr. 3SG.M has read Tom the book Tom has read the book.'

(Breton)

(Breton)

Introduction

25

The following show that finite verbs like finite auxiliaries can appear in initial position in subordinate clauses, negatives and sentences where SpecCP is filled. (58)

(59) (60)

Al levr a lenn Anna. (Breton) the book PRT read Anna 'Anna reads the book.' Ne lennket Anna al levr. (Breton) NEG readNEG Anna the book 'Anna didn't read the book.' Gouzout a ran [e lenn Anna al levr] know PRT do.lsG PRT read Anna the book 4 1 know that Anna read the book.'

(Breton)

Here, then, we have some major differences between Breton and the other Celtic languages.

3.3 Clause-initial particles

All the Celtic languages have a variety of clause-initial particles in finite clauses, both root and subordinate. An important question here is whether these particles are complementizers or something else. Welsh has the following particles: (61)

Root affirmative Subordinating Interrogative Root negative Embedded negative 'Direct' relative 'Indirect' relative

fe, mi, y y a ni(d) na(d) a y

Perhaps the most important point to note here is that we have three different root affirmative particles, one of which is identical to the subordinating particle. Fe and mi are characteristic of spoken Welsh. Fe is used in South Wales and mi in North Wales. They are used very widely. Y, on the other hand, is only used with the present and imperfect of bod 'be', and only in Literary Welsh. The following illustrate these particles: (62)

(63)

Fe/mi welais i ddraig. PRT saw.lsG I dragon. 'I saw a dragon.' Fe/mi fydd ef ym Mangor. PRT will-be he in Bangor 'He will be in Bangor.'

(Welsh)

(Welsh)

26

Robert D. Borsley and Ian Roberts

(64)

Y mae ef ym Mangor. (Welsh) PRT is he in Bangor 'He is in Bangor.' A further point to note about Welsh is that it has distinct root and subordinate negative particles. These, however, are mainly a feature of Literary Welsh. In Colloquial Welsh only their mutation effects appear. Turning to Breton, we find the following, quite simple situation. (65)

Subordinating Interrogative Negative 'Direct' relative 'Indirect' relative

e hag-en ne a e

Unlike Welsh, Breton has no root affirmative particles. The negative particle, like its counterparts in Welsh, is a feature of the written language. Irish has the most complex set of particles. Following Chung and McCloskey (1987), we can summarize the Irish situation as follows: (66) Subordinating Interrogative Root negative Embedded negative 'Direct' relative 'Indirect' relative

Non-past go an ni nach a a

Past gur ar nior nar a ar

The most important point to note here is that we have distinct non-past and past particles. Duffield (1990) sees this as evidence that Tense and not Agr is the first functional category below C in Irish. A further point to note is that like Breton, Irish has no root affirmative particles. There has been considerable debate about the status of these particles. For Welsh, Hendrick (1988) and Sadler (1988) assume that they are complementizers, while Harlow (1983) and Rouveret (1990) argue that they are part of a verbal complex. For Breton, Stephens (1982) argues that they are part of a verbal complex, but Hendrick (1988) assumes that they are complementizers, and Stump (1989) argues that negative ne is a complementizer but that a and e are not. For Irish, Chung and McCloskey (1987) argue that they are combinations of a complementizer and an Infl element, and Duffield (1990a) argues that they are realizations of separate C, T, Neg and Agr elements. McCloskey (forthcoming) argues that although they are basically complementizers, they occupy the Infl position on the surface as the result of a lowering process. As (41) shows, preposed adverbials precede preverbal particles. If preposed adverbials are adjoined to IP the particles must be in IP on the surface. If some of the particles

Introduction

27

are not complementizers in some of the languages, we will have certain subordinate clauses that cannot contain a complementizer. This situation arises anyway in Breton since no particle appears in affirmative subordinate clauses containing the 'have' auxiliary. The subordinate clause in (60) illustrates. One matter that we have not commented on is the appearance in all three languages of direct and indirect relative particles. We will return to these particles in section 4.1.

3.4 Clefting A notable feature of the Celtic languages is that they do not have SVO finite clauses, contrary to the statements of Greenberg (1963) about VSO languages. Stump (1984) argues that Breton examples like the following are SVO finite clauses: (67)

Ar vugale a lenn levriou. the children PRT read books The children read books.'

(Breton)

However, Borsley and Stephens (1989) show that the crucial clauses are in fact instances of topicalization or clefting. What this suggests is that the specifier position of the highest functional head below C is for some reason never overtly occupied in the Celtic languages (at least in finite clauses). Clefting is an important process in all the Celtic languages. The following illustrate: (68)

(69) (70)

Gafra weloddy dyn. (Welsh) goat PRT saw the man 'It was a goat that the man saw.' Levriou a lenn ar vugale. (Breton) books PRT read the children 'It's books that the children read.' Capall mor ban chonaic me (Irish) horse big white saw I 'It was a big white horse that I saw.'

Two chapters in the present volume deal (among other things) with cleft sentences in Welsh: Tallerman and Rouveret. Tallerman is concerned with the relation between cleft sentences like (71) and so-called abnormal sentences like (72). (71)

Y dynion a werthodd y ci. the men PRT sold the dog 'It was the men who sold the dog.'

(Welsh)

28

Robert D. Borsley and Ian Roberts

(72)

Y dynion a werthasant y ci. the men PRT sold.3PL the dog The men sold the dog.'

(Welsh)

Abnormal sentences are obsolete but were an important feature of Middle Welsh. In both constructions a variety of constituents can appear in initial position, but when a subject occupies the initial position the verb shows no agreement in the cleft construction, but does show agreement in the abnormal construction. Tallerman argues that the former involves the movement of a constituent into SpecCP while the latter involves a constituent base-generated adjoined to CP. An important fact about cleft sentences is that they can appear as subordinate clauses introduced by the element mai (or in South Wales, taw). We have examples like the following: (73)

Dywedais i mai'r dynion a werthodd y ci. said.lsG I PRT-the men PRT sold the dog 'I said that it was the men that sold the dog.'

Tallerman argues that this element is a complementizer which unusually only takes a CP as its complement. She notes that Welsh also has an interrogative complementizer ai that appears to take a CP complement. (74)

Gofynodd Emrys ai Gwyn a welodd y ddraig. asked Emrys PRT Gwyn PRT saw the dragon 'Emrys asked if it was Gwyn that saw the dragon.'

(Welsh)

Breton has a similar interrogative complementizer, as Borsley, Rivero and Stephens note. (75)

(76)

N' ouzon ket ha Yann en deus lennet al NEG know.lsGNEG PRT Yann 3SG.M have read the 'I don't know whether Yann has read the book.' N' ouzon ket ha lennet en deus Yann al NEG know. ISG NEG PRT read 3SG.M have Yann the 'I don't know whether Yann has read the book.'

levr. (Breton) book levr. (Breton) book

Example (76) illustrates the one situation in which a non-finite verb can be moved to a pre-auxiliary position in a subordinate clause. Rouveret is concerned not only with ordinary cleft sentences but also with sentences in which some constituent appears in initial position followed immediately by a form of bod 'be' with no intervening complementizer. The following illustrate: (77)

Meddyg yw Emrys. doctor is Emrys 'Emrys is a doctor.'

(Welsh)

Introduction (78)

Glasyw'r mor. blue is-the sea The sea is blue.'

29 (Welsh)

Rouveret argues that the initial constituent, whether a subject or a non-subject, is in SpecCP and that the following form of bod is in C, hence the impossibility of a complementizer. In other words, he argues that Welsh has a class of verb-second sentences rather like those of German and other Germanic languages (see section 2.2). Among other pieces of evidence for this proposal he cites the fact that such clauses can appear as subordinate clauses introduced by mai. This suggests rather strongly that they are CPs and hence that the initial constituent is in SpecCP.

3.5 Non-finite clauses

We can look next at non-finite clauses. Here, all the Celtic languages have subjectinitial order, although Welsh also has a class of what look like verb-initial nonfinite clauses. Welsh and Breton are broadly similar in this area, while Irish is rather different. One point that we should note immediately is that non-finite verbs are traditionally referred to as verb-nouns. However, when they correspond to English non-finite verbs they are fairly ordinary non-finite verbs. They take the same proclitics as nouns, as the following illustrate: (79)

Ceisiodd Megan ei tried

(80) (81) (82)

weld (ef).

(Welsh)

Megan 3SG.M see he

'Megan tried to see him.' ei wraig(ef) (Welsh) 3SG.M wife he 'his wife' Gallout a raYanne welout. be-able PRT do Yann 3SG.M see 'Yann can see him.' e di

(Breton) (Breton)

3SG.M house

(83)

(84)

'his house' TaCathal i ndiaidh a mholadh. is Cathal after 3SG.M praise 'Cathal has praised it.' a mhac (Irish) 3SG.M son

'his son'

(Irish)

30

Robert D. Borsley and Ian Roberts

However, they differ from nouns in taking adverbs. In Welsh, we have the following contrasts: (85)

(86)

Dylai Megan ganu *(yn) hyfryd. ought Megan sing PRT pleasant 'Megan ought to sing pleasantly.' y swn (*yn) hyfryd the sound PRT pleasant 'the pleasant sound'

(Welsh)

(Welsh)

In Welsh, adverbs are distinguished from attributive adjectives among other things by the preceding particle yn. This is obligatory in (85) but impossible in (86). Turning to Breton, we have the following data: (87)

(88)

a. EmanYann o tistrujin al lizher prim. is Yann PROG destroy the letter quick 'Yann is destroying the letter quickly.' b. *Eman Yann o tistrujin prim al lizher. is Yann PROG destroy quick the letter a. an distruj euzhus eus Jerusalem the destruction terrible of Jerusalem 'the terrible destruction of Jerusalem' b. *an distruj eus Jerusalem euzhus the destruction of Jerusalem terrible

(Breton)

(Breton)

In Breton, adverbs and attributive adjectives are formally identical, but they differ in their distribution. Adverbs normally follow the complements of the associated verb, but an attributive adjective immediately follows the associated noun and among other things precedes any complements. Finally, Irish is much like Welsh in this area. In Irish, adverbs are distinguished from attributive adjectives by the preceding particle go. This is obligatory in (89) and impossible in (90). (89)

(90)

Tase ag ceol *(go) binn. is he PROG sing PRT pleasant 'He is singing pleasantly.' an fonn (*go) binn the tune PRT pleasant 'the pleasant tune'

(Irish)

(Irish)

Non-finite verbs also differ from nouns in taking 'bare' DP objects. As we will see in section 3.7, nouns have complements marked with a preposition. For further discussion of the Welsh data, see Borsley (1993). All three Celtic languages have superficially subjectless non-finite clauses in both control and raising sentences. The former are illustrated in (91)—(93) and the latter in (94)-(96).

Introduction (91)

(92)

(93)

(94)

(95)

(96)

Ceisiodd Emrys weld y ddraig. tried Emrys see the dragon 'Emrys tried to see the dragon.' C'hoant en deus Yann da vont da Baris desire 3SG.M have Yann to go to Paris 'Yann wants to go to Paris.' Ba mhaithliom an teach a dhiol. I-would-like the house to sell 'I would like to sell the house.' Dechreuodd Megan ddarllen y llyfr. began Megan read the book 'Megan began to read the book.' Annaig a zeuas da vezan klanv. Annaig PRT came to be ill 'Annaig happened to be ill.' D'fheadfadh Ciaran a bheith breoite. can.coND Ciaran be(-FiN) ill 'Ciaran could be ill.'

31 (Welsh)

(Breton)

(Irish)

(Welsh)

(Breton)

(Irish)

An important difference between Irish and the other two languages is illustrated here. In Irish non-finite clauses, the object normally appears before the verb. This is illustrated by (93). Chung and McCloskey (1987) and McCloskey and Sells (1988) suggest that this order is the result of adjunction of the object to the predicate of the small clause. Bobaljik and Carnie (this volume) argue, however, that it is the result of movement to SpecAgrO. Adger (this volume) argues that object-verb order in Scots Gaelic is a result of the same process. Another notable feature of Irish is that it has sentences which appear to involve raising to a prepositional object position. The following is a relevant example: (97)

Thiocfadh le Ciaran teach a cheannach. come.coND with Ciaran house to buy 'Ciaran could buy a house.'

(Irish)

According to McCloskey (1984), Ciaran here is the underlying subject of a nonfinite clause, but the surface object of the preposition le 'with'. Stowell (1989) argues against this analysis, but the analysis is defended by McCloskey (this volume). All three Celtic languages have non-finite clauses with overt subjects. In Welsh and Breton, these are rather like English for-to clauses. The following illustrate: (98)

Disgwyliodd Emrys i Megan fynd i Fangor. expected Emrys to Megan go to Bangor 'Emrys expected Megan to go to Bangor.'

(Welsh)

32

Robert D. Borsley and Ian Roberts

(99)

Mad eo d' ar vugale mont d' an aod. good is to the children go to the beach 'It's good for the children to go to the beach.'

(Breton)

It seems reasonable to assume that the prepositional elements / and a" assign case to the following subjects in these examples. Irish non-finite clauses with overt subjects look rather like so-called exceptional Case-marking clauses, e.g. Bill to be a genius in / consider Bill to be a genius. The following illustrates: (100)

Ba mhaith Horn iad Ciaran a fhostu. I-would-like them Ciaran hire(-FiN) 'I would like them to hire Ciaran.'

(Irish)

These clauses, however, are not restricted in their distribution in the way that exceptional Case-marking clauses are. McCloskey (1985) points out that they can appear as complements of nouns and adjectives, citing the following: (101)

(102)

Bheadh luchair air iad a bheith i lathair. (Irish) would-be joy on.3sG.M thembe(-FiN) present 'He would be delighted for them to be present.' B' eadoiche iad cruinniu aris (Irish) is improbable them assemble again 'It would be improbable that they would assemble again.'

He also notes that the subject of such a clause has accusative Case even when the clause is complement of a preposition that assigns genitive Case. He illustrates with the following, in which i ndiaidh 'after' is a preposition which assigns genitive Case: (103)

i ndiadh an pobal imeacht after the congregation( + ACC) leave(-FiN) 'after the congregation leaves/left'

(Irish)

Negated non-finite clauses take a rather different form in the three languages. Welsh has peidio a, literally 'cease with', after the subject. (104)

Disgwyliodd Emrys i Megan beidio a mynd i Fangor. expected Emrys to Megan cease with go to Bangor 'Emrys expected Megan not to go to Bangor.'

(Welsh)

Breton has the form nompas, calqued on French ne pas. (105)

Mad eo d' ar vugale nompas mont d' an aod. good is to the children NEG go to the beach 'It's good for the children not to go to the beach.'

(Breton)

Introduction

33

Irish has the element gan, which Chung and McCloskey (1987) argue is a complementizer. (106)

Ba mhaith Horn gan iad posadh ro-6g. I-would-like NEG them marry(-FiN) too-young 'I would like them not to marry too young.'

(Irish)

How exactly Celtic non-finite clauses should be analysed is far from clear. In all three languages, however, there is some evidence that the subject is outside VP. In Welsh, as we have seen, the subject of a non-finite clause precedes the negative element beidio. If we assume that negation originates outside VP then the subject must also be outside VP. When we turn to Breton, the fact that the subject of a non-finite clause precedes the negative element nompas suggests that it too is outside VP. It suggests in fact that subjects are in a higher position in non-finite clauses than in finite clauses given that subjects follow negative ket in finite clauses. What, then, of Irish? Chung and McCloskey (1987) argue that the subject of a non-finite clause is within the complement of Infl. However, if one accepts the view of Bobaljik and Carnie (this volume) that preverbal objects in non-finite clauses are in Spec AgrO, then subjects must also be outside VP in Irish. It seems reasonable to assume that / in (98) and d" in (99) assign Case to the following subjects. Borsley (1986) analyses the former as a prepositional complementizer like English for, and the same analysis is proposed for the latter in Hendrick (1988). Bobaljik and Carnie (this volume) suggest that the subjects of Irish non-finite clauses are Case-marked by an empty prepositional complementizer. If the subjects of non-fmite clauses are assigned Case by preceding prepositional complementizers, then it may be that they occupy the specifier position of the functional head immediately below C. If so, they will be in a higher position than in finite clauses. It is possible, however, that / and d" are not in C but in Agr. If so, at least the Welsh subjects might be in the same position in nonfinite clauses as finite clauses. We noted at the outset that Welsh also has a class of what look like verb-initial non-fmite complements. The following illustrate: (107)

(108)

Credaf fod Emrys yn dod. believe. ISG be Emrys in come 'I believe Emrys is coming.' Dywedodd Megan fod Emrys yn hwyr. said Megan be Emrys in late 'Megan said Emrys was late.'

(Welsh)

(Welsh)

These clauses allow just one verb, bod 'be'. Although they appear to be non-fmite, Awbery (1976) pointed out that there is evidence that they are in fact finite. Examples (107) and (108) contain the non-finite form bod where we would expect

34

Robert D. Borsley and Ian Roberts

either a present or an imperfect form. These forms, which only bod has, cannot appear in affirmative declarative subordinate clauses, as the following illustrate: (109) (110)

*Credaf y mae believe. 1 SG PRT is *Dywedodd Megan said Megan

Emrys yn dod. Emrys in come yr oedd Emrys yn hwyr. PRT was Emrys in late

(Welsh) (Welsh)

We do, however, find present and imperfect forms of bod in negative declarative and interrogative subordinate clauses. Thus, we have the following: (111)

(112)

(113)

(114)

Credaf nad ydyw Emrys yn dod. believe. ISG NEG is Emrys in come 'I believe Emrys is not coming.' Dywedodd Megan nad oedd Emrys yn hwyr. said Megan NEG was Emrys in late 'Megan said Emrys was not late.' Yr wyf i 'n gofyn a ydyw Emrys yn dod. PRT am I in ask Q is Emrys in come 'I am asking if Emrys is coming.' Gofynnodd Megan a oedd Emrys yn hwyr. asked Megan Q was Emrys in late 'Megan asked whether Emrys was late.'

(Welsh)

(Welsh)

(Welsh)

(Welsh)

This suggests fairly strongly that the subordinate clauses in (107) and (108) are really finite despite appearances. It appears that certain other Welsh subordinate clauses which appear to be non-finite are really finite. Consider the following: (115)

Dywedodd Megan i Emrys fyndi Gaergybi. said Megan to Emrys go to Holyhead 'Megan said Emrys went to Holyhead.'

(Welsh)

Sadler (1988) points out that we have not (116) but (117) as a related example with a negated subordinate clause. (116)

(117)

*Dywedodd Megan i Emrys said Megan to Emrys 'Megan said Emrys didn't go Dywedodd Megan nad aeth said Megan NEG went

beidio a mynd i Gaergybi. (Welsh) cease with go to Holyhead to Holyhead.' Emrys i Gaergybi. (Welsh) Emrys to Holyhead

She also notes that these subordinate clauses, unlike ordinary non-finite subordinate clauses, can be readily conjoined with an ordinary finite subordinate clause. Thus, there are quite good reasons for thinking that these are really finite clauses.

Introduction

35

3.6 Participial constructions We can turn now to participial constructions. In fact, we have already considered a Breton participial construction exemplified by (49) and (53)—(55). More characteristic of Celtic, however, are participial constructions involving a non-finite verb preceded by what we will call a particle. We can look first at what we will call the progressive construction. Here, all three languages have a non-finite verb preceded by a particle. The following illustrate: (118)

(119)

(120)

Mae Rhiannon yn cysgu. is Rhiannon PRT sleep 'Rhiannon is sleeping.' Emanva breur o vont a-hed an hent. is ISG brother PRT go along the road 'My brother is going along the road.' Ta me ag togail teach ur. are I PRT build house new 'I am building a new house.'

(Welsh)

(Breton)

(Irish)

An important point to note here is that the Irish example in (120) has verb-object order and not the object-verb order characteristic of ordinary non-finite clauses. Perhaps the main question to ask about these constructions is: what are the particles? In all three languages, they look rather like prepositions, but in all three languages there are grounds for thinking that they are not. Welsh yn resembles a preposition meaning 'in', but it differs from the preposition in a number of ways. Most notably, it triggers no mutation whereas the preposition triggers so-called nasal mutation. The following illustrate: (121)

(122)

Mae ef yn darllen y papur. is he PRT read the paper 'He is reading the paper.' Mae ef yn Nolgellau. is he in Dolgellau 'He is in Dolgellau.'

(Welsh)

(Welsh)

In (121), we have the basic form of a verb. In contrast, in (122), we have a mutated form of the name Dolgellau. Breton o is related historically to a preposition ouzh 'from'. It is clear, however, that it is a quite different element. It triggers the so-called mixed mutation whereas the preposition triggers no mutation. The following illustrate: (123)

o

vont go 'going'

PRT

(Breton)

36

Robert D. Borsley and Ian Roberts

(124)

ouzh mogerenn an ti (Breton) from wall the house 'from the wall of the house'

In (123), vont is a mutated form of the verb, the basic form being mont. In (124), mogerenn is a basic form. Irish ag appears to be identical to a preposition meaning 'at', but the written language is misleading here. The particle is in fact pronounced differently from the preposition. It is realized as [a] before consonants, [g] before a back vowel, and [g'] (a voiced palatal stop) before a front vowel. The preposition has just one allomorph, [eg'] in most dialects. It is doubtful, then, whether these particles are prepositions. Note, however, that Rouveret (this volume) assumes that Welsh yn is a preposition. If these particles are not prepositions, what are they? One possibility is that they are realizations of a functional category Asp(ect) (Hendrick 1991). If they are, we might have something like the following structure for (118): (125)

F1P

Asp

VP V

Rhiannon

e

t

yn

t

cysgu

We can turn now to what I will call the perfective construction. Here, again, Welsh and Irish have a non-finite verb preceded by a particle. The following illustrate: (126)

(127)

Mae Rhiannon wedi mynd adref. is Rhiannon after go home 'Rhiannon has gone home.' Ta me i ndiaidh Eoghan a fheiceail. is I after Owen see(-FiN) 'I have just seen Owen.'

(Welsh)

(Irish)

Notice that the Irish example has OV order just like an ordinary non-finite clause. It seems clear that the particles in these examples are prepositions. Elsewhere, the same items introduce ordinary non-finite clauses with an overt subject. The following illustrate:

Introduction (128)

(129)

37

wedi i Megan weld y ddraig after to Megan see the dragon 'after Megan saw the dragon' i ndiaidh an pobal imeacht after the congregation leave(-FiN) 'after the congregation leave/left'

(Welsh)

(Irish)

It looks, then, as if these constructions may involve a preposition taking a fairly ordinary non-finite clause. Unlike Welsh and Irish, Breton has a perfective construction containing a participle. We have already seen examples in (49) and (53)—(55). The following are two further examples: (130)

(131)

Kolleten deus ar martolod e gasketen. lost 3SG.M have the sailor 3SG.M cap 'The sailor has lost his cap.' Nijet eo al labous kuit. flown is the bird away 'The bird has flown away.'

(Breton)

(Breton)

These examples show that the Breton perfective construction can contain a form of 'have' or a form of 'be'. Which one appears seems to be determined by semantic considerations, 'have' being used when an event is referred to and 'be' when a state is referred to. We can look finally at passive constructions in the three languages. The Welsh passive is illustrated by the following: (132)

Cafodd Gwyn ei daro gan Emrys. got Gwyn 3SG.M hit with Emrys 'Gwyn was hit by Emrys.'

(Welsh)

The point to note here is that the non-finite verb is preceded by a clitic agreeing with the subject. Irish also has a passive construction which is rather like the Welsh construction. McCloskey (1983) calls this the progressive passive. The following illustrates: (133)

Ta teach a thogail aige. is house 3SG.M build at.3sG.M 'A house is being built by him.'

(Irish)

As in the Welsh construction, the non-finite verb is preceded by a clitic agreeing with the subject. Unlike Welsh, however, but like Breton, Irish has a passive participle. This is used in what McCloskey (1983) calls the perfective passive, which is illustrated by the following:

38

Robert D. Borsley and Ian Roberts

(134)

Ta an teach togtha acu. is the house built at.3PL They have built a house.'

(Irish)

Breton only has passives with a passive participle. The following illustrates: (135)

Chanchet e oa e vuhez. changed PRTwas 3SG.M life 'His life was changed.'

(Breton)

3.7 Nominals We can turn now to the central features of Celtic nominal phrases. Here, there are various points to note. All the Celtic languages have a definite article but only Breton has an indefinite article. In all the Celtic languages, the definite article can co-occur with a demonstrative element, which appears after the noun. Thus, we have data like the following: (136)

(137)

(138)

y dyn hwn/hwnnw the man this that 'this/that man' al levr -man/-se the book this that 'this/that book' an teach seo / sin the house this that 'this/that house'

(Welsh)

(Breton)

(Irish)

In all the Celtic languages, adjectives generally follow the noun, whereas numerals precede. The following illustrate: (139)

(140)

(141)

(142)

pentrefbach village small 'a small village' levr bihan book small 'a small book' cat mor cat big 'a big cat' dau afal two apples 'two apples'

(Welsh)

(Breton)

(Irish)

(Welsh)

Introduction (143)

(144)

daou levr two book 'two books' da chat two cat 'two cats'

39 (Breton)

(Irish)

All the languages have a few adjectives which appear prenominally, for example hen 'old', gwahanol 'different' in Welsh, gozh 'old' in Breton and priomh 'chief in Irish. In all the Celtic languages, possession is expressed by a bare postnominal DP. The following illustrate: (145)

(146)

(147)

brawd yr athro brother the teacher 'the teacher's brother' levr ar paotr book the man 'the man's book' pictiur Chathail picture Cathal(GEN) 'Cathal's picture'

(Welsh)

(Breton)

(Irish)

Where a noun has both a possessor and a complement, the complement follows the possessor. (148)

(149)

(150)

lun Emrys o Megan picture Emrys of Megan 'Emrys's picture of Megan' karantez Lenaig evit ar beorien love Lenaig for the poor 'Lenaig's love for the poor' pictiur Chathail den chapall picture Cathal(GEN) of-the horse 'Cathal's picture of the horse.'

(Welsh)

(Breton)

(Irish)

As noted in section 3.5, nominal complements are marked by a preposition. Possessors are often seen as subjects of nominal phrases. Thus we have nounsubject-complement order within nominal phrases in much the same way as we have verb-subject-complement order in finite clauses. In other words, the form of nominal phrases echoes that of finite clauses. This is the case not only in the Celtic languages but also in the Semitic languages, where we have examples like the following from Syrian Arabic:

40

Robert D. Borsley and Ian Roberts

(151)

Sourt Salwa la Kamal picture Salwa of Kamal 'Salwa's picture of Kamal'

The relation between Celtic and Semitic nominals, or more precisely that between Irish and Hebrew and Maltese nominals, is the subject of Duffield's chapter. He proposes that nominal phrases are DPs in both Irish and Hebrew, but he proposes that there are two other functional heads between D and N: Agr and Num. He argues that nominal phrases have the same basic structure in all three languages and that the differences between them can be attributed to different derivational processes.

3.8 Agreement, clitics and null arguments We turn now to an area where the Celtic languages have some very distinctive properties, although, as Roberts and Shlonsky (this volume) point out, the Semitic languages are quite similar. In all the Celtic languages, finite verbs show no agreement with non-pronominal DPs. In Welsh, they show agreement with pronouns either overt or empty. The following illustrate: (152)

(153)

(154)

Gwelsan (nhw) ddraig. saw.3PL they dragon They saw a dragon.' Gwelodd y dynion ddraig. saw the men dragon 'The men saw a dragon.' *Gwelsan y dynion ddraig. saw.3PL the men dragon 'The men saw a dragon.'

(Welsh)

(Welsh)

(Welsh)

Here, we see that the third-person plural form of the verb is used with a thirdperson plural pronoun but the unmarked form of the verb with a plural nonpronominal DP. Some dialects of Breton seem to be like Welsh in this area. Others only allow agreement with an empty pronoun. We have data like the following: (155)

(156)

Levriou a lennont. books PRT read.3PL 'They read books.' Levriou a lenn ar vugale. books PRT read the children 'The children read books.'

(Breton)

(Breton)

Introduction (157)

41

*Levriou a lennont ar vugale. books PRT read.3PL the children T h e children read books.'

(Breton)

Speakers seem to vary as to whether they allow a pronoun in an example like (155). In Irish, agreement is only possible with an empty pronoun. The following illustrate: (158)

(159)

(160)

D'imiodar. lefUpL 'They left.' D'imighna fir. left the men T h e men left.' *D'imiodar na fir / siad. left.3pL the men they T h e men/they left.'

(Irish)

(Irish)

(Irish)

Here, we see that the third-person plural form of the verb cannot appear with an overt subject whether a non-pronominal DP or a pronoun. All the Celtic languages have inflected prepositions, which agree with their objects under the same conditions as finite verbs agree with their subjects. The following parallel (152)-(157): (161)

(162)

(163)

(164)

(165)

i 'r dynion to the men 'to the men' iddyn (nhw) to.3PL they 'to them' *iddyn y dynion to.3PL the men 'to the men' daYannig to Yannig 'to Yannig' dezhan

(Welsh)

(Welsh)

(Welsh)

(Breton)

(Breton)

tO.3SG.M

(166)

'to him' *dezhan Yannig tO.3.SG.M

'to Yannig'

(Breton)

42

Robert D. Borsley and Ian Roberts

(167)

ar na fir on the men 'on the men' orthu on.3PL 'on them' *orthu na fir / siad on.3PL the men they 'on the men/them'

(168)

(169)

(Irish)

(Irish)

(Irish)

It seems, then, that we have a single phenomenon here. Notice that the languages either allow or require an empty object with an inflected preposition. Thus, they are not just null-subject languages but null-argument languages, to use McCloskey and Hale's (1984) term. As we saw in section 3.5, both nouns and non-finite verbs in Celtic take proclitics. They appear under the same conditions as agreement morphology. We will just illustrate with nouns. (170)

ei

thy

(hi)

(Welsh)

3SG.F house she

(171)

(172)

(173)

'her house' ty Megan house Megan 'Megan's house' *ei thy Megan 3SG.F house Megan 'Megan's house' e dad

(Welsh)

(Welsh)

(Breton)

3SG.M father

(174)

(175)

(176)

'his father' tad Yannig father Yannig 'Yannig's father' *e dad Yannig 3SG.M father Yannig 'Yannig's father' a mhac

(Breton)

(Breton)

(Irish)

3SG.M son

(177)

'his son' mac Chathail son Cathal(GEN) 'Cathal's son'

(Irish)

Introduction (178)

*a

43

mhac Chathail /

3SG.M son

Cathal(GEN)

se

(Irish)

he

'Cathal's/his son' It seems, then, that clitics and agreement morphology constitute a single phenomenon. Welsh clitics are a major concern of Roberts and Shlonsky (this volume). They show that they are quite different from Romance clitics but quite like Semitic clitics. They argue that Welsh and Semitic clitics are Agr heads whereas Romance clitics are D's. Essentially the same view of Scots Gaelic clitics is advanced in Adger (this volume), and Duffield (this volume) takes a similar view of clitics in Irish, Hebrew and Maltese nominal phrases. 3.9 The copula Another notable feature of the Celtic languages is the complex array of forms that they have corresponding to the English verb be. In Irish, it is traditional to distinguish between a copula and a substantive verb. These are exemplified by (179) and (180), respectively. (179)

(180)

Is dochtuir Sean. COP doctor Sean 'Sean is a doctor.' Ta Sean ar muisce. be Sean drunk 'Sean was drunk.'

(Irish)

(Irish)

They differ in a number of ways. Most obviously, the copula takes a clause-final subject, whereas the substantive verb appears in VSO clauses like other verbs. The copula also takes an accusative subject whereas the substantive verb takes a nominative subject, again like other verbs. The copula takes predicates that express a permanent property, individual-level predicates in the terminology of Kratzer (1989), while the substantive verb takes predicates that express more transient properties, stage-level predicates in Kratzer's terms. The Irish copula is discussed at length by Doherty (1992), who argues that they involve the following structure:

44

Robert D. Borsley and Ian Roberts

(181)

IP

subject COP

predicate

If this is right, Irish has finite clauses in which the specifier position of the first functional head below C is overtly filled, unlike typical finite clauses in the Celtic languages. Rouveret (this volume) argues that the distinction between individual- and stage-level predicates is also relevant to the Welsh copula. He is particularly concerned with the contrasting behaviour of the two third-person singular present-tense forms mae and yw. Mae cannot appear in negative or polar interrogative sentences and cannot have a definite DP as its complement. (182)

a. Mae Megan yn cysgu. (Welsh) is Megan in sleep 'Megan is sleeping.' b. Mae Megan yn ddiog. is Megan in lazy 'Megan is lazy.' c. Mae Megan yn athro is Megan in teacher 'Megan is a teacher.' d. *Ni mae Megan ddim yn cysgu. NEG is Megan NEG in sleep 'Megan isn't sleeping.' e. *A mae Megan yn cysgu? Q is Megan in sleep 'Is Megan sleeping?' f. *Mae Megan yn yr athro. is Megan in the teacher 'Megan is the teacher.'

Yw, on the other hand, cannot appear in simple affirmative declarative clauses. It is limited to negatives, polar interrogatives and the verb-second clauses referred to earlier. We have data like the following: (183)

a. *Yw Megan yn cysgu. is Megan in sleep 'Megan is sleeping.'

(Welsh)

Introduction

45

b. *Yw Megan yn ddiog. is Megan in lazy 'Megan is lazy.' c. *Yw Megan yn athro is Megan in teacher 'Megan is a teacher.' d. Nid yw Megan ddim yn cysgu. NEG is Megan NEG in sleep 'Megan isn't sleeping.' e. A yw Megan yn cysgu? Q is Megan in sleep 'Is Megan sleeping?' f. Yr athro yw Megan, the teacher is Megan 'Megan is the teacher.' Rouveret proposes that the particle yn creates stage-level predicates and that mae incorporates a locative adverbial clitic which is an operator at LF and can bind the spatio-temporal external argument characteristic of such predicates. Neg and Q can also bind this argument. Hence, mae does not occur when they are present. Hendrick argues against Rouveret's approach to the mae/yw distinction, and develops a very different approach drawing on recent proposals of Pesetsky's. These involve a Principle of Telegraph which deletes functional categories subject to a series of partially ordered constraints. He proposes that mae is inserted next to the particle y. This prevents it from appearing in negatives and polar interrogatives. He suggests that it is required to be at the left edge of AgrP in Literary Welsh, but at the left edge of CP in Colloquial Welsh. The latter requirement forces the deletion of y. Hendrick is also concerned with the distribution of another third-person present-tense form of the copula, sydd. This appears when a subject is extracted, as (186) illustrates: (184)

Rhys sydd yn athro Rhys is in teacher 'Rhys is a teacher.'

(Welsh)

Hendrick proposes that sydd licenses a following trace. An important feature of sydd is that it never co-occurs with the particle a, which normally appears when a subject is extracted. Hendrick proposes that both sydd and a are required to be on the left edge of CP, but that the sydd constraint is higher ranked than the a constraint. The result is that a is deleted to allow sydd to be at the left edge of CP. Hendrick also discusses the Breton copula. He argues that whereas the Welsh copula is a substantive element, the Breton copula is generally a grammatical element, a spell-out of tense and agreement features. In this connection, he

46

Robert D. Borsley and Ian Roberts

highlights the very different form of copula + predicate adjective sentences in the two languages. He notes, however, that Breton also has a substantive copula taking PP and progressive complements. Hendrick also discusses the form zo, which, like sydd, appears when a subject is extracted. (185)

Mona a zo brav. Mona PRTis fine 'Mona is fine.'

(Breton)

Unlike sydd, zo co-occurs with the particle a. Hendrick proposes that the particle licenses the following trace and that zo is the result of a low-level morphological rule.

3.10 The realization of negation One further interesting feature of the Celtic languages, with which a number of the chapters are concerned, is the realization of negation. As we have seen, both Welsh and Breton have two negative particles rather like French ne and pas. As we have also seen, subjects precede the Welsh counterpart of pas, ddim, but follow the Breton counterpart, ket. This suggests that subjects are in a higher position in Welsh than in Breton. However, some Welsh dialects are more like Breton. In Pembrokeshire Welsh, indefinite subjects always follow ddim and definite subjects optionally follow (see Awbery 1990). In Standard Welsh, only indefinite subjects of the copula follow ddim (which appears as dim). We have examples like the following: (186)

Does dim defaid yn y cae. NEG.is NEG sheep in the field There are no sheep in the field.'

(Welsh)

This fact is discussed in Rouveret's chapter. Irish does not have the two-part negation of Welsh and Breton, but Irish negation is of considerable interest, as Acquaviva's chapter shows. As Acquaviva discusses, Irish negation never takes the form of a 'negated quantifier' comparable to nobody or nothing. More generally, he points out, Irish lacks what Barwise and Cooper (1981) call monotone decreasing quantifiers, quantifiers like few, which have the property that when one appears in a subject truth is preserved if the predicate is replaced by another predicate denoting a smaller set. He argues that monotone decreasing quantifiers are not syntactic quantifiers but indefinites locally bound by a negative operator at LF. This means that NegP must be projected whenever a monotone decreasing quantifier is present. However, Italian data suggest that monotone decreasing quantifiers are incompatible with an overt Neg°. Acquaviva proposes that Irish lacks monotone decreasing quantifiers because Neg° is always lexically realized in Irish.

Introduction

47

4 Some further issues In section 3, we were concerned with aspects of Celtic syntax about which the chapters in this volume have quite a lot to say. We turn now to two further notable features of Celtic syntax about which the following chapters have little to say.

4.1 A-binding constructions A'-binding constructions such as relative clauses and w/z-questions are a particularly interesting feature of the Celtic languages. The constructions have broadly similar properties in all the Celtic languages although they are certainly not identical. In all the Celtic languages, they involve resumptive pronouns and two types of empty categories, one associated with agreement including clitics and another not associated with agreement, and in all the Celtic languages, they show interesting clause-initial particle alternations. Irish A'-binding constructions are studied in some detail in McCloskey (1979, 1990). McCloskey argues that empty categories associated with agreement are empty resumptive pronouns, while empty categories associated with agreement are traces. He shows that A'-binding involving traces but not A'-binding involving resumptive pronouns is subject to Subjacency. This means that resumptive pronouns are possible in many positions where traces are not possible. He shows that there is one position where a trace is possible but not a resumptive pronoun, the highest subject position below the binding operator. There is a variety of positions in which both traces and resumptive pronouns are possible. McCloskey is particularly concerned with the clause-initial particle alternations that occur in Irish A'-binding constructions. We saw earlier that the present-tense forms of the direct and indirect relative particles have the same form. They are both a. They differ, however, in that the direct particle triggers the mutation process known as lenition, while the indirect particle triggers the mutation process known as eclipsis. For this reason, McCloskey represents them as aL and aN, respectively. With traces, the direct particle appears everywhere between the binding operator and the trace, as schematized in (187) and illustrated in (188). (187) (188)

[ DP DP yaL... [s,aL ...*...]]]] an rud a shil me a duirt tu a dheanfa the thing PRT thought I PRT said you PRT do.coND.2sG 'the thing that I thought you said you would do'

(Irish) (Irish)

With resumptive pronouns, the indirect particle appears adjacent to the binding operator, but the ordinary subordinating particle appears everywhere else between the binding operator and the resumptive pronoun, as follows:

48

Robert D. Borsley and Ian Roberts

(189) (190)

[ D P D P [ s , a N . . . y g o . . . ^ g o ...pro...}}}} an fear ar shil me go dtiocfadh se the man PRT thought I PRT come.coND he 'the man that I thought (he) would come'

(Irish) (Irish)

Alternatively, the indirect relative particle appears everywhere between the binding operator and the resumptive pronoun, as follows: (191) (192)

[ DP DP [SraN... [s,aN... an meid den dan the amount of-the poem 'as much of the poem as

[s,aN.. .pro...]]]] (Irish) ar mheas se a raibh feidhm leis PRT thought he PRT was use with.3sG.M he thought he needed'

McCloskey (1990) argues that the form of the particle in A'-binding constructions is determined by Spec-head agreement. In movement structures, there is an empty operator in the topmost SpecCP and traces in any intermediate SpecCPs. In nonmovement structures, there is an empty operator in the topmost SpecCP and empty operators may also occur in intermediate SpecCPs. He proposes that empty operators and intermediate traces have the same value for the features ibanaphoric and ipronominal as the associated traces and resumptive pronouns. This entails that empty operators and intermediate traces in movement construction are [-anaphoric, -pronominal], while empty operators in resumptive pronoun constructions are [-anaphoric, + pronominal]. The A'-binding constructions of the other Celtic languages have received rather less attention than those of Irish. For Welsh, Sadler (1988) argues that empty categories associated with agreement are empty resumptive pronouns. However, Hendrick (1988) argues that such empty categories are traces and that the associated agreement is required by the Empty Category Principle (ECP). Welsh and Breton A'-binding constructions are quite similar to Irish A'-binding constructions, but there are some important differences. Most notably, we have certain affirmative-negative contrasts in Welsh and Breton, which are not found in Irish. Consider, for example, the following: (193)

(194)

(195)

y dynion a welodd ddraig the men PRT saw dragon 'the men that saw a dragon' y dynion na welsan ddraig the men NEG saw.3PL dragon 'the men that didn't see a dragon' ar vugale a lenne al levriou the children PRT read the books 'the children that read the books'

(Welsh)

(Welsh)

(Breton)

Introduction (196)

49

ar vugale ne lennent ket al levriou the children PRT read.3PL NEG the books 'the children that didn't read the books'

(Breton)

In (193) and (195) we have relativization of a subject in an affirmative relative clause, while in (194) and (196) we have relativization of a subject in a negative relative clause. In (193) and (195) the verb shows no agreement, but in (194) and (195) there is agreement. There is no such contrast in Irish, as the following illustrate: (197)

(198)

na daoine a bhuail an cu (Irish) the men PRT struck the dog 'the men that struck the dog' na daoine nar bhuail an cu (Irish) the men NEG struck the dog 'the men that didn't strike the dog'

For some recent discussion of these contrasts, see Ouhalla (1993). There is much more that could be said about A'-binding constructions. We will note just one point. The particles that appear in relative clauses also appear in clefts and w/*-questions, in which SpecCP is overtly filled. If these particles are complementizers, we must conclude that the Celtic languages have no constraint against having both an overt complementizer and an overt occupant of SpecCP.

4.2 Mutation One of the most distinctive features of the Celtic languages is their so-called mutation systems, systems of morphophonological alternations affecting initial consonants. The following Welsh examples provide a striking illustration: (199)

a. tad 'father' c. dy dad 'your(sG) father'

b. fy nhad 'my father' d. ei thad 'her father'

(Welsh)

Welsh has three sets of mutations, which means that some words have four different initial consonants. Breton has five, and Irish has two. Mutation is typically triggered by some item immediately preceding the affected constituent. A variety of lexical items are triggers. The examples in (199) show that proclitics can trigger mutation. Example (200) shows that some prepositions trigger mutation, (201) shows that some numerals do, and (202) shows that the predicative particle yn does. (In each case the mutated form is in bold and the basic form given in parentheses.)

50

Robert D. Borsley and Ian Roberts

(200)

i Fangor (Bangor) to Bangor 'to Bangor' dau fachgen (bachgen) two boy 'two boys' Mae Gwyn yn feddyg (meddyg) is Gwyn in doctor 'Gwyn is a doctor.'

(201)

(202)

(Welsh)

(Welsh)

(Welsh)

It is possible for a whole class of lexical items to trigger mutation. Thus, in Welsh, all feminine singular nouns trigger mutation on a following adjective. We have contrasts like the following: (203)

a. cath fawr (mawr) cat big 'a big cat' b. ci mawr dog big 'a big dog'

(Welsh)

A further important point to note is that mutation can be triggered by an empty category. Consider, for example, the following: (204)

Welais i ddim draig. (gwelais) saw.lsG I NEG dragon 'I didn't see a dragon.'

(Welsh)

Here, it seems reasonable to assume that the mutation is triggered by an empty counterpart of the negative particle ni referred to in sections 3.3 and 3.10. What exactly is mutation? It has sometimes been suggested that it is a manifestation of Case. For example, Zwicky (1984) suggests that the contrast between the mutated object in (205a) and the unmutated object in (205b) represents a contrast between accusative Case and genitive Case. (205)

a. Gwelodd y dyn gi (ci) saw the man dog 'The man saw a dog.' b. Mae'r dyn wedi gweld ci is-the man after see dog 'The man has seen a dog.'

(Welsh)

(Welsh)

There is, however, a very important difference between mutation and case. As Harlow (1989) points out, a moved w/z-phrase inherits Case from its trace, but a

Introduction

51

moved constituent does not inherit mutation from its trace. Thus, the w/z-phrase in the following is unmutated: (206)

Pwy a welodd y dyn? who PRT saw the man 'Who did the man see?'

(Welsh)

Hence, it seems very unlikely that mutation can be seen as a manifestation of Case. The mutation in (205a) is particularly interesting. Harlow (1989) argues that this mutation is triggered not by a preceding lexical item but by a preceding DP. He suggests that the mutation in the following has the same source: (207)

(208)

Mae yna fuwch yn yr ardd. (buwch) is there cow in the garden There is a cow in the garden.' cyn i Emrys fynd i Aberystwyth (mynd) before to Emrys go to Aberystwyth 'before Emrys went to Aberystwyth'

(Welsh)

(Welsh)

Revising and extending Harlow's proposal, Borsley and Tallerman (1996) argue that all phrases trigger mutation under certain conditions. If either of these proposals is along the right lines, they add to the evidence that mutation is not a manifestation of Case given that Case is normally assigned by heads.

5 Conclusion

We hope that the foregoing remarks will provide an adequate background to the chapters that follow. We believe that these chapters testify both to the intrinsic interest of the Celtic languages and to their relevance for current theorizing.

Notes We would like to thank the British Academy and the Welsh Department of the University of Wales, Bangor, for help in funding the conference on which this collection is based. We would also like to thank Janig Stephens and Cathair O Dochartaigh for help with the Breton and Irish data, respectively. The straightforward implication that a language allows inversion only if it has the French-style orders in (la), (2a), (3a) does not hold. The Mainland Scandinavian languages - Swedish, Danish and Norwegian - are verb-second in root clauses, i.e. root clauses feature inversion of subject and verb when a non-subject comes first, and pattern like English with respect to the Pollock-Emonds tests in embedded clauses (Platzack 1987; Holmberg and Platzack 1991; Vikner 1994). If at least some verb-second clauses

52

Robert D. Borsley and Ian Roberts

involve V-to-I movement, as proposed by Travis (1984) and Zwart (1993), then the generalization can be maintained. 3 We give the schema in the form which derives specifier-head-complement order; this order is the only one relevant for Celtic, and, if Kayne (1995) is correct, may be the only one that is available at all.

1

Long head movement in Breton

Robert D. Borsley, Maria-Luisa Rivero and Janig Stephens

1 Introduction

A notable feature of Breton is sentences in which a non-finite verb of some kind appears in initial position followed by an auxiliary and a subject (or if there is no overt subject whatever complement(s) the non-finite verb requires).1 The following is a typical example: (1)

Lennet en deus Yann al levr. read 3SG.M has Yann the book 4 Yann has read the book.'

Sentences of this kind are not found in the other Celtic languages or in the most intensively studied non-Celtic languages. Not surprisingly, therefore, they have figured quite prominently in discussions of Breton syntax. In this chapter, we will argue that these sentences are the product of long head movement, a process which moves a verb directly to C over certain intervening heads. This process has previously been identified in Bulgarian, Czech, Slovak, Serbo-Croatian, Rumanian, Old Spanish and European Portuguese before the twentieth century (see Rivero 1991, 1994, Lema and Rivero 1989a, 1989b, 1991). We will show that Breton sentences like (1) have all the properties of similar sentences in these languages. It seems, then, that Breton is another language with long head movement. This is of interest for two fairly obvious reasons. Firstly, it means that long head movement is not limited to the Slavonic and Romance families. Secondly, since Breton is a VSO language, it means that long head movement can occur in a VSO language. Breton long head movement is also of interest because it appears to be motivated by a rather different constraint from that which motivates it in the other languages in which it occurs. We will argue, however, following Rivero (1993), that the same constraint is operative in Breton and the other languages. We will also argue, following Rivero (1991) and especially Roberts (1994a) that the Breton data require a distinction between two types of heads, and a version of the ECP sensitive to this distinction. Breton provides two apparent problems for the conception of long head movement developed in earlier work. It is suggested in this work that auxiliaries 53

54

Robert D. Borsley, Maria-Luisa Rivero and Janig Stephens

fall into two mutually exclusive classes: functional auxiliaries, which may license long head movement, and lexical auxiliaries, which may license VP-preposing. Breton appears to have an auxiliary which licenses both. We will argue, however, that appearances are misleading here and that there are in fact two different auxiliaries. It is also argued in earlier work that functional auxiliaries do not impose semantic restrictions on their complements. Breton appears to have auxiliaries which license long head movement and hence must be functional but which impose semantic restrictions on their complements. We leave this problem for further research. The chapter is organized as follows. In section 2, we introduce the Breton sentences and discuss in a preliminary way how they should be analysed. Then, in section 3, we summarize earlier work on long head movement and argue that the process occurs in Breton. In section 4, we consider various analytic questions that arise about Breton long head movement. In section 5, we look at the two apparent problems. Finally, in section 6, we summarize the chapter.

2 Verb + auxiliary sentences It is fairly well established that Breton is a VSO language. It has VSO order in all types of clauses, and SVO clauses appear to be instances of topicalization (see Anderson and Chung 1977; Anderson 1981; and especially Borsley and Stephens 1989). However, an important feature of Breton is sentences which are not VSO but VAuxSO. Example (1) illustrates, and so do the following: (2)

(3)

Lennet e oa al levr gant Yann. read PRT was the book by Yann The book was read by Yann.' Lenn a ra Anna al levr. read PRT does Anna the book 'Anna reads the book.'

In (1), we have a past participle followed by a perfect auxiliary. In (2), we have a passive participle followed by a passive auxiliary. (As in English, the passive participle is identical in form to the past participle but has different syntactic properties.) Finally, in (3), we have the basic non-finite form of the verb, traditionally known as a verb-noun, followed by a counterpart of English auxiliary do. As we have said, such sentences have had considerable attention in the literature on Breton syntax. How, then, should sentences like (l)-(3) be analysed? One proposal, advanced in Anderson and Chung (1977), Anderson (1981) and Schafer (1992), is that they are instances of topicalization. One thing that casts doubt on this idea is that it appears to be a head and not a phrase that is fronted in such sentences. If

Long head movement

55

topicalization is movement to SpecCP, one would not expect a head to undergo it. One might suggest, however, as Schafer appears to do, that these sentences involve so-called remnant topicalization, the topicalization of a phrase from which a constituent has been extracted. It has been argued that this is involved in Dutch sentences like the following from Koster (1987): (4)

Gelezen heeft hij het boek niet. read has he the book not 'He has not read the book.'

Here, we have just the verb gelezen in initial position. Thus, remnant topicalization can look like head movement. Notice, however, that one would expect remnant topicalization to be subject to the same constraints as ordinary topicalization. As was noted in Stephens (1982), Breton verb-fronting is more constrained than topicalization. Firstly, it is clause-bound. Thus, an example like (5), in which the initial verb has been extracted from a subordinate clause, is ungrammatical. (5)

*Desket am eus klevet he deus Anna he c'henteliou. learnt ISG have heard 3SG.F have Anna 3SG.F lessons 'I have heard that Anna has learnt her lessons.'

Topicalization is not clause-bound. We have examples like (6), in which an NP has been extracted from a subordinate clause, and examples like (7), in which a VP has undergone the same process. (6)

(7)

Al levr a lavaras Yann e lennas. the book PRT said Yann PRT read 4 Yann said that he read the book.' O lenn al levr a ouian email Yann. PROG read the book PRT know.ISG is Yann 'I know Yann is reading this book.'

Secondly, Breton verb-fronting is blocked by negation. Thus, (8a) does not have (8b) as an alternative form. (8)

a. N' en

deus ket lennet Tom al levr. has NEG read Tom the book 'Tom has not read the book.' b. * Lennet n' en deus ket Tom al levr. read NEG 3sG.Mhas NEG Tom the book NEG 3SG.M

Topicalization is not blocked by negation. Thus (9), in which an NP has been topicalized, is perfectly acceptable.

56

Robert D. Borsley, Maria-Luisa Rivero and Janig Stephens

(9)

Al levr na lennas ket Yann. the book NEG read NEG Yann 'Yann didn't read the book.'

So is (10), in which a VP has been topicalized. (10)

O lenn al levr n' email ket Yann. PROG read the book NEG is NEG Yann 'Yann isn't reading the book.'

The ungrammaticality of examples like (8b) not only distinguishes Breton verbfronting from Breton topicalization, it also distinguishes it from remnant topicalization in Dutch, which, as (4) shows, is not blocked by negation. It is fairly clear, then, that examples like (l)-(3) are not instances of topicalization.2 Within current theoretical assumptions, a natural proposal is that examples like (l)-(3) involve movement to C. For (1), this will give something like the following analysis: CP

(11)

IP VP

V; NP

VP NP

lennet

en deus

Yann

al levr

This analysis embodies the widely held assumption that subjects of VSO clauses are in a VP-internal position (see among others Koopman and Sportiche 1991; McCloskey 1991b; Rouveret 1990). It is rather like analyses proposed in Hendrick (1990a, 1991) except that Hendrick assumes that the auxiliary en deus originates in a specifier position. We see no advantage in this assumption and at least one disadvantage: it requires a process of specifier-to-head movement, something which is not generally assumed. We would have a more complex structure here if we assumed separate T and AGR categories instead of I (see e.g. Pollock 1989 and Chomsky 1991). We would also have a more complex structure if the auxiliary originated in a lower position and was moved into its surface position. However, we will ignore both these possibilities in subsequent discussion. Examples (2) and (3) appear to be somewhat more complex than (1) given that the particles e and a precede the auxiliaries. In Hendrick (1988), these particles are analysed as complementizers. There is, however, evidence against such an analysis, most notably in Stump (1989). What their underlying status is is far from

Long head movement

57

clear. Superficially, however, they appear to form a constituent with the following finite element. We assume, then, that (2) and (3) have structures like (11) except that they have a complex I constituent. One further point that we should note here is that we seem to have a violation of the Head Movement Constraint, which stipulates that a head can only move to the nearest c-commanding head position. We will return to this matter in section 4. One argument against an analysis like that in (11) is advanced in Borsley (1992). It involves examples like the following: (12)

Lennet en deus ha komprenet en deus Yann al levr. read 3SG.M have and understood 3SG.M have Yann the book 'Yann has read and has understood the book.'

On the assumption that only constituents can be conjoined, such examples suggest that the combination of participle and auxiliary is a constituent, contrary to the analysis in (11). It is not necessary, however, to analyse (12) as involving the co-ordination of participle + auxiliary combinations. Instead, it can be analysed as involving the right node raising of the (verbless) VP Yann al levr. The following examples suggest that this is a plausible analysis: (13)

(14)

Dec'h e welas ha hirie e prenas Yann al levr. yesterday PRT saw and today PRT bought Yann the book 'Yesterday Yann saw and today he bought the book.' Dec'h en devoa gwelet ha hirie en deus prenet yesterday 3SG.M had seen and today 3SG.M have bought Yann al levr. Yann the book 'Yesterday Yann had seen and today he has bought the book.'

In (13) the conjunction is preceded and followed by the sequence adverbial, particle and finite verb, and in (14) it is preceded and followed by the sequence adverbial, auxiliary and participle. Neither of these sequences can be plausibly analysed as a constituent. Thus, these examples must involve the right node raising of a VP. It is natural to assume that (13) does too.

3 Long head movement So far we have said very little that is new. We can now move on to something which we assume is new. This is the central observation of the chapter: that V + Aux order is not a peculiarity of Breton, but is a feature of a number of languages discussed in Rivero (1991, 1994) and Lema and Rivero (1989a, 1989b, 1991), and as in these languages is the result of long head movement (LHM). In particular, as noted at the outset, it is a feature of Bulgarian, Czech, Slovak, Serbo-Croatian,

58

Robert D. Borsley, Maria-Luisa Rivero and Janig Stephens

Rumanian, Old Spanish and European Portuguese before the twentieth century. The following illustrate: (15)

(16)

(17)

(18)

(19)

(20)

(21)

Procel sum knigata. (Bulgarian) read have.lsG book-the 'I have read the book.' Koupiljsem knihy. (Czech) bought have.lsG books 'I have bought books.' Napisal som list. (Slovak) written have. 1 SG letter 'I have written a letter.' Citao sam knjigu. (Serbo-Croatian) read have. ISG book 'I read a book.' Spune mi va? (Rumanian) tell me will.3sG 'Will she/he tell me?' Dar- te he un exemplo. (Old Spanish) give you will. 1 SG an example 'I will give you an example.' Seguir- te- ei por toda a parte. (European follow you will.ISG by all the part Portuguese) 'I will follow you everywhere.'

Here and subsequently, we indicate which language non-Breton examples come from by giving the language in parentheses after the example. In all these examples, we have a non-finite verb followed by an auxiliary. Note, however, that the auxiliary does not immediately follow the verb in the last three examples because a clitic pronoun intervenes. A number of properties of LHM are identified in the papers we have cited. We will show that Breton verb-fronting has all these properties. The first property of LHM that is identified in the earlier papers is that it is restricted to root clauses. The following Bulgarian data illustrate this property: (22)

a. Znam ce sum procel knigata. know. 1 SG that have. 1 SG read book-the 'I know that I have read the book.' b. *Znam ce procel sum knigata.

(Bulgarian)

In (22a) the participle follows the auxiliary in the subordinate clause. Hence, LHM has not applied. In (22b), the participle precedes the auxiliary, indicating that LHM has applied. The result is ungrammaticality. The following data show that Breton verb-fronting is subject to the same restriction:

Long head movement (23)

59

a. Lavaret he deus Anna en deus lennet Tom al levr. said 3SG.F have Anna 3SG.M have read Tom the book 'Anna said Tom had read the book.' b. *Lavaret he deus Anna lennet en deus Tom al levr.

In (23a) the participle in the subordinate clause follows the auxiliary but in (23b) it precedes as a result of verb-fronting and the result is ungrammaticality.3 There is in fact one situation in which verb-fronting can occur in subordinate clauses. We will discuss this in section 4. The second property of LHM that the earlier papers discuss is that it cannot cross a negative particle. It either takes the negative particle with it or it does not occur in negative sentences. The former is the case in Czech, while the latter is the case in Bulgarian, as the following illustrate: (24)

(25)

Ne koupil jsem knihy. NEG bought have.lsG books 'I did not buy books.' a. Ne sum procel knigata. NEG have. ISG read book-the 'I have not read the book.' b. * Procel ne sum knigata. read NEG have.ISG book-the

(Czech)

(Bulgarian)

We saw earlier that Breton verb-fronting is impossible in negative sentences. The relevant examples are repeated here as (26). (26)

a. N' en deus ket lennet Tom al levr. NEG 3SG.M has NEG read Tom the book T o m has not read the book.' b. *Lennet n' en deus ket Tom al levr. read NEG 3SG.M has NEG Tom the book

A further property of long head movement that is highlighted in the earlier papers is that it is incompatible with movement to SpecCP. Again, we can illustrate with Bulgarian data: (27)

a. Kakvo e procel Petur? what has read Peter 'What has Peter read?' b. *Kakvo procel e Petur? what read has Peter

(Bulgarian)

In these examples we have w/j-movement, one instantiation of movement to SpecCP. In (27a), the participle follows the auxiliary. Hence, there is no LHM.

60

Robert D. Borsley, Maria-Luisa Rivero and Janig Stephens

In (27b), the participle precedes the auxiliary. Hence, LHM has applied. This example is ungrammatical. Consider now some Breton data. (28)

a. Al levr en deus lennet Tom. the book 3SG.M has read Tom Tom has read the book.' b. *A1 levr lennet en deus Tom. the book read 3SG.M has Tom

Here, we have topicalization, another instantiation of movement to SpecCP. In (28a), there is no verb-fronting. In (28b), verb-fronting has applied and ungrammaticality results. The final property that the earlier papers highlight is that an auxiliary allows LHM or VP-fronting, but not both. Auxiliaries that allow LHM are termed functional auxiliaries and auxiliaries that allow VP-fronting are termed lexical auxiliaries, and a number of properties of the two types are identified. The following examples show that the Bulgarian and Czech perfect auxiliaries, which allow LHM, do not allow VP-fronting: (29)

(30)

*Procel knigata sum. read book-the PRES.ISG 'I have read the book.' *Koupil knihy jsem. bought books have.lsG 'I have bought books.'

(Bulgarian)

(Czech)

In contrast, the Czech future auxiliary allows VP-fronting but not LHM. The following illustrate: (31)

a. Kupovat knihy budu. buy books will.lsG 'I will buy books.' b. * Kupo vat budu knihy. buy will.lsG books

(Czech)

Returning now to Breton, the following shows the Breton perfect auxiliary does not allow VP-fronting:4 (32)

*Lennetal levr en deus Yann. read the book3sG.M has Yann 'Tom has read the book.'

Long head movement

61

The following shows that the same is true of the Breton passive auxiliary: (33)

*LennetgantYann e oa al levr. read by Yann PRT was the book T h e book was read by Yann.'

Contrasting with the perfect and passive auxiliaries, we have the progressive auxiliary. This allows VP-fronting (as we saw earlier) but not verb-fronting. The following illustrate: (34)

a. O lenn al levr email Yann. PROG read the book is Yann 4 Yann is reading the book.' b. *O lenn email Yann al levr. PROG read is Yann the book

Thus, Breton V-fronting has all the properties that are associated with LHM in earlier work. Therefore, we conclude that it is a further case of LHM.

4 Analytic questions As we noted at the outset, a number of questions arise about LHM in any language in which it occurs. In this section we will look at these questions and consider how they should be answered in Breton. The most obvious question is: why does LHM occur? Superficially, it may appear that the answer to this question is rather different in Breton than in the other LHM languages. We will argue, however, that this is not the case, and propose a unified account based on licensing requirements for Tense. For LHM languages other than Breton, it is often suggested that LHM is one way of avoiding weak, clitic elements such as auxiliaries or pronouns in sentence-initial position. In Breton, it is possible to have a pronominal clitic in sentence-initial position, as the following illustrate: (35)

(36)

E c'halvet en deus Yann. 3SG.M called 3SG.M have Yann 4 Yann has called him.' O gwelet am eus. 3PL seen ISG have 'I have seen them.'

In both of these examples, we have an object clitic in initial position followed by a participle. Thus, weak clitic elements can appear in sentence-initial position in Breton. What is not in general possible in sentence-initial position is a finite verb. Thus, the following is ungrammatical:

62

Robert D. Borsley, Maria-Luisa Rivero and Janig Stephens

(37)

*Lenn Anna al levr. read Anna the book 'Anna reads the book.'

In contrast, we have examples like the following: (38)

(39)

(40)

Al levr a lenn Anna, the book PRT read Anna 'Anna reads the book.' Ne lenn ket Anna al levr. NEG read NEG Anna the book 'Anna didn't read the book.' Gouzout a ran e lenn Anna al levr. know PRT do. ISG PRT read Anna the book 'I know that Anna read the book.'

In (38) the finite verb is preceded by a topicalized NP, in (39) it is preceded by the negative element ne, and in (40) we have a finite verb in initial position in a subordinate clause. All of these examples are acceptable because they do not have a finite verb in sentence-initial position. In general, then, finite verbs cannot appear in sentence-initial position in Breton. There is, however, one important exception to this. This is a form of the copula that takes a PP or a progressive complement. The following illustrate: (41)

(42)

EmanYann war an hent. is Yann on the road 'Yann is on the road.' Eman Anna o lenn al levr. is Anna PROG read the book 'Anna is reading the book.'

There is a further point that we should note here. Although (35) and (36) are perfectly grammatical, clitics are not always acceptable in sentence-initial position. Consider, for example, the following: (43)

(44)

*E c'halvas Yann. 3SG.M called Yann 'Yann called him.' *O gwelis. 3PL saw.ISG

'I saw them.' In these examples the clitics are followed by a finite verb, whereas in (35) and (36) they are followed by participles. What this suggests is that the combination of a

Long head movement

63

clitic and a finite verb counts as a finite verb and is impossible where a finite verb is impossible. In view of the above phenomena, it may appear that we have a rather different restriction in Breton than in the other LHM languages. However, Rivero (1993) argues that all LHM languages including Breton are characterized by a syntactic licensing system for Tense which accounts for the properties of this language type and separates it from other types such as the verb-second type. We will summarize the proposed analysis. The central idea of the analysis is that Tense may be licensed in two ways in LHM languages. On the one hand, it may be licensed by a V which is adjoined to it. In other words, it may be licensed in its checking domain in the sense of Chomsky (1993). On the other hand, it may be licensed by a governing C with certain properties. In other words, it may be licensed in the internal domain of C in Chomsky's (1993) terminology. In Breton, Tense is generally licensed in this way. Only the form of the copula illustrated in (41) and (42) can license Tense in its checking domain. A governing C may license tense under three conditions: (a) it is filled either by negation or by a non-finite verb; (b) its specifier position is filled; or (c) it heads a CP which is L-marked. Within this analysis, LHM languages may differ in how much they utilize the two ways of licensing Tense. As we have noted, Breton tense is generally licensed in the internal domain of C. In Slavonic only the functional auxiliaries traditionally labelled clitics are licensed in this way. Verbs and lexical auxiliaries can license Tense in its checking domain. Hence, they can appear in sentenceinitial position. In addition to the two ways of licensing Tense, one of which derives second position effects for tensed items, all LHM languages including Breton have a long X° movement process for non-tensed verbs and auxiliaries. This is a last-resort operation subject to economy. As such, it applies only if nothing else licenses Tense. Although LHM languages share a common system for licensing Tense, this does not mean that they share the same systems for licensing other functional categories. In Slavonic, licensing systems for pronominal clitics of the functional category D resemble in great measure the system used for Tense. In Breton, however, pronominal clitics adopt the licensing system of the category they attach to. This means that in (35) and (36) the pronominal clitic is licit in sentence-initial position because a non-finite verb is licit in this position. In contrast in (43) and (44) clitics are impossible in sentence-initial position because finite verbs are impossible in this position. Here, then, we have an account of why LHM occurs in Breton which treats it as essentially similar to other LHM languages despite the superficial differences. A second question that we need to ask about LHM is: why is it possible? As we noted earlier, LHM is incompatible with the Head Movement Constraint. Since

64

Robert D. Borsley, Maria-Luisa Rivero and Janig Stephens

Chomsky (1986) the HMC has been viewed as a consequence of the Empty Category Principle and in particular the requirement that traces be antecedentgoverned. Assuming that this approach is basically sound, the question that arises is: how can the ECP allow certain exceptions to the HMC while ensuring that it generally holds? If antecedent-government is subject to Relativized Minimality, as in Rizzi (1990), the right results can be obtained if two types of head are distinguished. This idea is developed in Roberts (1994a). Building on the proposals of Chomsky and Lasnik (1993), Roberts distinguishes between L-related and non-L-related heads. A head is L-related if it contains a feature of some lexical head. Otherwise, it is non-L-related. In the present context, what is important is that I is L-related (or T and Agr are if they are distinguished) while C and Neg are non-L-related. If Relativized Minimality is sensitive to the distinction, the ECP will allow a head to cross over another head as long as the head that it crosses is not of the same type as the head to which it moves. Hence, it will allow a verb to move to C across I since while I is L-related C is not. This approach receives important support from sentences which contain two auxiliaries, one finite and the other non-finite. Here, we have examples like the following: (45)

a. Bet am eus kavet al levr. had have.lsG found the book 'I have found the book.' b. Kavet am eus bet al levr. found have.lsG had the book

In (45a), the non-finite auxiliary appears in initial position, followed by the finite auxiliary and the main verb. In (45b), the main verb is in initial position, followed by the finite auxiliary and the non-finite auxiliary. These are active sentences. We have similar passive sentences. Consider the following: (46)

a. Bet eo lennet al levr gant Yann. been is read the book by Yann T h e book has been read by Yann.' b. Lennet eo bet al levr gant Yann. read is been the book by Yann

Again, we have the non-finite auxiliary in initial position in the (a) example and the main verb in initial position in the (b) example. As Rivero (1991) notes, similar examples occur with the Bulgarian renarrated mood: (47)

a. Bil sum cetjal knigata. had have. 1 SG read book + the 'According to someone, I am reading the book.'

(Bulgarian)

Long head movement

65

b. Cetjal sum bil knigata read have. 1 SG had book + the In examples such as Breton (45b) and (46b) and Bulgarian (47b), two heads and not just one intervene between the verb and its trace in VP. We assume that (45b) has something like the following structure, and that (46b) and Bulgarian (47b) have similar structures: (48)

kavet

al levr

We assume that the untensed verb in C antecedent-governs its trace in such examples because the auxiliaries are all L-related. Hence, there is no violation of the ECP. While these long' LHM patterns can be easily accommodated within an approach which recognizes two types of heads and appeals to Relativized Minimality, they appear problematic for other recent analyses of LHM, such as those proposed by Cavar and Wilder (1992), Dobrovie-Sorin (forthcoming: ch. 1) and Manzini (1992b), which seek to avoid an appeal to Relativized Minimality. These treatments differ from each other in several respects but they share the core idea that the fronted non-tensed verb and the following tensed auxiliary form a complex in C in LHM sentences. In simple LHM sentences, this complex head is able to antecedent-govern both the trace of the tensed auxiliary which is in IP and the trace of the non-tensed item which is in VP, since it contains indices for both traces. However, in examples like (45b), (46b) and (47b), there is a non-tensed auxiliary between the tensed auxiliary and the trace in VP. Hence, even if the tensed auxiliary is in C, the non-tensed verb will not antecedent-govern the trace in VP without an appeal to Relativized Minimality.5 A third question that arises about LHM is: what sort of movement is it? Rizzi and Roberts (1989) identify two main types of head movement: movement triggered by the m(orphological)-selection properties of the host head, and adjunction. There are two reasons for rejecting the first of these options. Firstly, as Roberts (1994a) notes, m-selecting heads are one type of L-related head, and our account of why LHM is possible crucially involves the assumption that C is a non-L-related head. Secondly, there is evidence that the verb and the following

66

Robert D. Borsley, Maria-Luisa Rivero and Janig Stephens

particle do not form a complex word, as they presumably would if we were dealing with m-selection. The evidence comes from examples like the following: (49)

(50)

(51)

Lennet ha komprenet en deus Yann al levr. read and understood 3SG.M have Yann the book 'Yann has read and understood the book.' Lennet ha komprenet e oa al levr gant Yann. read and understood PRT was the book by Yann T h e book was read and understood by Yann.' Lenn ha kompren a ra Yann al levr. read and understood PRT did Yann the book 'Yann read and understood the book.'

In these examples, the particles are preceded by conjoined Vs. It is presumably impossible to have a conjunction inside a complex word. If so, we do not have complex words here, and hence we are not dealing with m-selected movement. Examples like (49)—(51) are also possible in Bulgarian, as the following illustrates: 6 (52)

Vidjal i procel e knigata. seen.SG.M and read.sG.M has book-the 'He has seen and read the book.'

(Bulgarian)

In Bulgarian too, then, there is strong evidence that LHM cannot be m-selected movement. It looks, then, as if we have adjunction here. This will give us the following S-structure for (1). (53)

lennet

en deus

Yann

al levr

One might object to this analysis on the grounds that it does not explain why Breton LHM is clause-bound. Roberts (1991a) argues that what he calls excorporation is possible when a head is adjoined to another head: that is, the adjoined head can be moved to a higher head position. One might, then, suggest that there is no reason why a V that is adjoined to a C should not be moved to a higher

Long head movement

67

position. Interestingly, however, Roberts (1994a) argues on the basis of Romance clitic-climbing data that excorporation is impossible from a non-L-related head. As we have seen, C must preclude further movement of a V that is adjoined to C. The final question that arises about LHM is: why is it restricted as it is? There are a number of sub-questions here. The first is: why is it clause-bound? We have already answered this question. The combination of the ECP and the ban on excorporation from a non-L-related head means that a verb can move no further than the nearest C. The next sub-question is: why is LHM incompatible with negation? We can attribute this to its last-resort nature. This means that LHM cannot apply when Tense is licensed without it. Tense is licensed when the negative particle ne is in C. Hence, LHM cannot apply. The incompatibility of LHM and negation can also be seen as a consequence of the ECP given plausible assumptions about Breton negation. If we assume, building on ideas of Pollock (1989), that ne originates as the head of a NegP which has ket as its specifier, we will have the following structure for the ungrammatical (8b):7 (54)

lennet

en deus

allevr

Here, the negative head rt is moved into C and the verb is adjoined to C. The verb has been moved to one non-L-related head, C, across another, Neg. Hence, we have a violation of the ECP. There will be no violation of the ECP in a simple negative sentence such as (39). Here, we will have the following structure:

68

Robert D. Borsley, Maria-Luisa Rivero and Janig Stephens

(55)

Yann e

al levr

Here, the verb is moved across a non-L-related head to an L-related head, and there is no violation of the ECP. There are two further sub-questions that we must ask: why is LHM impossible in subordinate clauses? And why is it incompatible with topicalization? Both these properties of LHM can be attributed to its last-resort nature. As we noted earlier, C can license Tense within its complement if the associated specifier position is filled and if it heads an L-marked CP. Given the last-resort nature of LHM, it will be impossible in both these situations. Notice now that these proposals suggest that C cannot license Tense within its complement if it heads a CP which is not L-marked even if it is a subordinate clause. Hence, LHM should be possible in the situation. Examples like the following suggest that this is indeed the case: (56)

N' ouzon ket ha lennet en deus Yann al levr. NEG know.lsGNEG Q read 3SG.M have Yann the book 'I don't know whether Yann has read the book.'

Here, we have an interrogative subordinate clause. This is introduced by the particle hag. We assume that hag is a functional head which takes a CP complement, but does not L-mark it. Given this assumption, we expect LHM to be possible within such complements.8

5 Two apparent problems We turn now to two features of Breton which appear to pose problems for some aspects of the conception of LHM developed in earlier work. The first apparent problem involves sentences containing various forms of ober 4 do\ Consider the following examples: (57)

Lenn a ra Yann al levr. read PRT does Yann the book 'Yann reads the book.'

Long head movement (58)

69

Lenn al levr a ra Yann. read the book PRT does Yann

Example (57) is identical to (2). Example (58) looks rather similar, but it is clearly an instance of VP-preposing since both the verb and its object appear in initial position. On the face of it, we have a counterexample here to the claim that an auxiliary allows LHM or VP-fronting but not both. However, as Stephens (1982: ch. 3) has shown, there is evidence that there are two different auxiliaries here. (See also Hewitt 1990.) Before we introduce this evidence, we can note that both LHM and VPpreposing have the expected properties in these examples. The following show that VP-preposing is unbounded, while LHM is clause-bound: (59)

a. Debrin krampouezh ed-du a ouian e rae Yann. eat pancakes buckwheat PRT know.lsG PRT did Yann 'I knew that Yann ate buckwheat pancakes.' b. * Debrin a ouian e rae Yann krampouezh ed-du. eat PRT know.lsG PRT did Yann pancakes buckwheat

The following show that VP-preposing is compatible with negation, while LHM is not. (60)

a. Debrin krampouezh ed-du ne ra ket Yann. eat pancakes buckwheat NEG do NEG Yann 'Yann does not eat buckwheat pancakes.' b. * Debrin ne ra ket Yann krampouezh ed-du. eat NEG do NEG Yann pancakes buckwheat

We turn now to the evidence that there are two different auxiliaries here. The crucial data are the following: (61)

(62)

a. Lennal levr en deus graetYann. read the book 3SG.M have done Yann 'Yann has read the book.' b. *Lenn en deus graet Yann al levr. read 3SG.M have done Yann the book a. Lennal levr a c'hellan ober. read the book PRTmay.lsG do 'I may read the book.' b. *Lenn a c'hellan ober al levr. read PRTmay.lsG do the book

Here, we see that VP-preposing but not LHM is possible with non-finite forms of ober. It may be that some independently motivated constraint blocks LHM in (62b). However, this is most unlikely in (61b). This is very similar to (45b) and (46b). We conclude, then, that there are two different auxiliaries here, one that

70

Robert D. Borsley, Maria-Luisa Rivero and Janig Stephens

has just finite forms and one that has the full range of forms. We will refer to the former as functional ober and to the latter as lexical ober. If this is right, there is no problem here for the idea that an auxiliary allows LHM or VP-fronting but not both. We will say a little more about these auxiliaries. Functional ober appears to be quite like the English do of do-support. That is, it seems that it is inserted in I just in case nothing else can be inserted. In the examples we have considered, there is nothing else that can be inserted because the verb has undergone LHM. Another situation in which there may be nothing else to be inserted in I is in tag questions. Here, we have examples like the following from Wojcik (1976): (63)

Skrivan a ra Yann, ne ra ket? write PRT does Yann NEG does NEG 'Yann writes, doesn't he?'

Obviously, this is very much like the do of do-support. Turning to lexical ober, this is unlike ordinary lexical auxiliaries in that it not only allows VP-fronting but requires it. Thus, the following are ungrammatical. (64)

(65)

*Yann a ra lenn al levr. Yann PRT does read the book 'Yann reads the book.' *Yann en deus great lenn al levr. Yann 3SG.M have done read the book 'Yann has read the book.'

We suggest that lexical ober requires an empty VP as its complement. As noted in Borsley (1990), it seems that one English do is rather similar. Consider the following data: (66) (67) (68)

What John does is annoy Mary. What John may do is annoy Mary. What John seems to do is annoy Mary.

Here, we have pseudo-clefts containing a form of do. The important point about these examples is that related simple sentences are ungrammatical. The following illustrate: (69) (70) (71)

*John does annoy Mary. *John may do annoy Mary. *John seems to do annoy Mary.

It seems, then, that this do also requires an empty VP as its complement. To conclude this discussion, we can propose the following structures for the examples in (57) and (58):

Long head movement

71

(72)

lenn

e

a ra

Yann

e

a ra

e

al

levr

(73)

lenn

al

levr

e

Obviously, these are very different structures. The second apparent problem involves the contrast referred to earlier between functional and lexical auxiliaries. An important claim in earlier work is that only the latter can select a semantically defined subset of VPs. This idea receives considerable support from the Romance and Slavonic languages. In the Romance languages, LHM occurs with the future and conditional auxiliaries, which combine with all VPs, but not with the two perfect auxiliaries corresponding to have and be, which combine with a subset of VPs. In the Slavonic languages, the perfect auxiliary combines with all VPs and allows LHM. In Serbo-Croatian, the future auxiliary also combines with all VPs and allows LHM. In Czech, however, the future auxiliary only combines with imperfective VPs and does not allow LHM. Like the Romance languages, Breton has two perfect auxiliaries endevout and bezan, corresponding to have and be, which combine with a subset of VPs. However, LHM occurs with both of them. This would be no problem if the two auxiliaries combined with syntactically defined subsets of VPs. It seems unlikely, however, that this is the case. VPs headed by the same verb appear sometimes with endevout and sometimes with bezan depending on how exactly they are interpreted. Consider the following: (74)

a. Nijet o deus diwar o neizh. left 3PL have from 3PL nest They have left their nest (but may come back)/

72

(75)

Robert D. Borsley, Maria-Luisa Rivero and Janig Stephens b. Nijet int diwar o neizh. left are from 3PL nest They have left their nest (permanently).' a. Kouezhet en deus en ur zont. fallen 3SG.M have while come 'He has fallen while coming (but probably got up again).' b. Kouezhet eo en ur zont. fallen is while come 'He has fallen while coming (and is probably still there).'

The (a) examples here contain a form of endevout and the (b) examples a form of bezan. As the translations make clear, the VPs in the (a) examples have an action interpretation, while those in the (b) examples have a state interpretation. Thus, it seems that endevout combines with VPs that have an action interpretation, while bezan combines with VPs that have a state interpretation. It looks, then, as if the two Breton perfect auxiliaries have semantic selection properties. If this is right, we must draw one of two conclusions: either functional auxiliaries can have semantic selection properties or lexical auxiliaries can allow LHM. Either way, we will be revising earlier conceptions. However, before it can be concluded that distinctions between endevout and bezan do not reduce to syntax, a detailed syntactic analysis of Breton constructions with these auxiliaries is required (for a recent syntactic approach to the distinction in Romance see Kayne 1993).

6 Conclusions In this chapter, we have looked at Breton sentences in which a non-finite verb of some kind appears in initial position followed by an auxiliary. We have argued that these sentences are the product of LHM, a process previously identified in a number of Slavonic and Romance languages. Superficially, LHM appears to be motivated by a somewhat different constraint in Breton from that which motivates it in the other languages. We have argued, however, following Rivero (1993), that the same constraint is operative in all the languages in which LHM has been identified. In all these languages, LHM is one way that they have of licensing Tense. We have also argued that Breton LHM provides evidence for an approach to LHM which distinguishes two types of head and incorporates a version of the ECP sensitive to the distinction. We have also looked at two problems which Breton seems to pose for the view of LHM developed in earlier work. We have argued that one of them - that Breton seems to have an auxiliary which licenses both LHM and VP-fronting - is illusory because there are in fact two different auxiliaries. The other problem - that Breton has two auxiliaries which license LHM and seem to have semantic selection properties - we leave to further research.

Long head movement

73

Notes 1 Earlier versions of this chapter were presented at the Conference on Comparative Celtic Syntax, University of Wales Bangor, June 1992, at the Plenary of the Eurotyp Project, San Sebastian, Spain, September 1992, and at the Autumn Meeting of the Linguistics Association of Great Britain, University of Surrey, September 1992. A French version of the chapter was presented at the annual meeting of the Centre Regional d' Etudes Bretonnes et Celtiques, Brest, Brittany in December 1992. We are grateful to Luigi Rizzi for helpful comments. 2 Schafer (1992) seeks to account for the contrasts between clause-initial non-finite verbs and clause-initial phrases by assuming that the latter are base-generated. However, while clause-initial phrases are not constrained in the way that clause-initial non-finite verbs are, they are not unconstrained. In particular, they show standard island effects. The following show that an NP cannot be extracted from a w/i-complement or a relative clause: (i) (ii)

*A1 the *A1 the

levr book levr book

e PRT a PRT

sonjen piv a lennas. wondered. ISG who PRT read ouian an den a lennas. know. ISG the man PRT read

It is also impossible to extract a VP from a w/z-complement or a relative clause, as the following show: (iii) (iv)

*Lenn read *Lenn read

al levr e sonjen piv reas. the book PRT wondered.ISG who did al levr a ouian an den a reas. the book PRT know. ISG the man PRT did

In the light of such data, the conclusion that clause-initial phrases are the result of movement seems inescapable. 3 Notice that the participle precedes the subject in the subordinate clause in (23a). We will not consider what sort of process is responsible for this order. 4 The impossibility of examples like (32) is noted in Anderson (1981). He suggests that they are ungrammatical because participles are adjectives and adjectives cannot take direct objects. It seems to us that this is a very unconvincing explanation. Breton participles do take direct objects. It is simply that they are commonly separated from their objects by movement processes. As far as we know, there is no reason to think that they are adjectives in the sorts of example that we are concerned with here. 5 Borsley (1990) develops an analysis within the GPSG framework of Gazdar et al. (1985) which allows examples like (45b) and (46b) but there is no obvious way to accommodate (45a) and (46a) within this analysis. 6 We are grateful to Mila Dimitrova-Vulchanova for providing us with this example. 7 This is essentially the analysis proposed for the broadly similar Welsh negation in Rouveret (1991).

74

Robert D. Borsley, Maria-Luisa Rivero and Janig Stephens

8 Some Breton speakers have a rather different type of interrogative subordinate clause illustrated in the following: (i)

N' ouzon ket hag-en en deus lennet Yann al levr. NEG know. ISG NEG Q 3SG.M have read Yann the book 'I don't know whether Yann has read the book.'

Here, we have a finite auxiliary in initial position. It may be that hag-en is a C that takes an IP complement. Alternatively, it may be that hag is the same element as in (56) and en an expletive pronoun in SpecCP.

Some syntactic effects of suppletion in the Celtic copulas Randall Hendrick

1 Introduction This study addresses broadly the issue of how syntax and morphology interact.1 I approach this topic by examining some of the syntactic consequences of irregular forms that appear in the third-person present tense of the copula in many of the Celtic languages. I suggest in section 2 that the Welsh and Breton copulas are fundamentally different and reproduce a distinction familiar from traditional descriptions of Irish: the Welsh copula is a substantive (i.e. lexical) copula while its Breton counterpart is a functional (or 'grammatical') copula, a spellout of tense and agreement features. I then argue in section 3 that, although both languages exhibit irregular forms in the copula's third-person present tense, they behave quite differently syntactically, and that this fact correlates with the distinction between substantive and functional copulas. Only the (Welsh) substantive copula exhibits suppletion that has syntactic effects; the (Breton) functional copula varies morphophonemically in terms of its spell-out and is syntactically inert. I make use of ideas of Pesetsky (1994) to model variation in the morphosyntactic restrictions of the Welsh substantive copula. Both the substantive and functional copula are, I claim, similar in terms of their selectional properties. Their difference, like other kinds of 'syntactic' variation between languages, is a formal, morphosyntactic one. In essence my claim is that the variation in the copulas reduces to a formal one between lexical formatives and functional formatives.2 In this respect I depart from the tack taken by Rouveret (this volume) which attributes the suppletion in Welsh of the third-person present-tense copula to a contrast between forms of the copula that semantically license stage-level predicates and other forms that lack that ability, as my analysis in section 4 of the Welsh suppletive form mae makes clear. The point is reinforced in section 5: the Welsh suppletive forms of the copula appear in a construction that suggests that, for purely morphosyntactic reasons, there are two positions outside VP that license subjects. This phenomenon poses a challenge for (mixed) semantic accounts such as that favoured by Rouveret that attribute semantic properties (in addition to morphosyntactic properties) to these two positions in which subjects are licensed. I conclude in section 6 by drawing out some of the implications that the syntactic effects of suppletive forms have for our understanding of how 75

76

Randall Hendrick

grammars organize themselves. My point is that syntax should not be insulated from the process of lexical instantiation that matches complex symbols of grammatical features with lexical forms.

2 Two Celtic copulas Traditional grammars commonly distinguish two copulas in Irish because one encounters in that language two morphologically different forms of the copula. This claim has recently been carefully defended in a generative framework by Doherty (1992), where a functional copula (is) is isolated that spells out tense and agreement, and a substantive copula (ta) formed on a verbal stem.3 In Hendrick (1994) I try to show that a similar distinction appears in the Brythonic languages, Welsh and Breton, as well, and explains some of the otherwise idiosyncratic differences between these languages. The variation between the substantive and grammatical copula is reflected in the surface word order. The functional copula, which appears in Breton sentences like (1), requires the predicate adjective (or nominal) to raise so that it appears in initial position to the left of the subject. (1)

(2)

Setu ar verc'h[Cp e oa spletus he levr]. this the woman PRT be.iMPERF profitable her book This is the woman whose book was profitable.' *Setu ar verc'h[ CP e oa he levr spletus].

Welsh on the other hand has a substantive copula. The Welsh substantive copula co-occurs with a predicative particle; the predicate adjective (or nominal) raises to that particle and appears to the right of the subject, as (3) exemplifies. (3)

Y mae Ifanyn saer. PRT be.PRES Ifan PRT carpenter 'Ifan is a carpenter.'

In this section we search for an explanation for this gross distinction in what are in many other respects rather similar languages. 2.1 The Welsh substantive copula In Modern Welsh predicate adjectives and predicate nominals occur in a copular construction that contains, in addition to the copula bod, a predicative particle yn. Thus we encounter structures of the form (PRT) bodNPyn AP/NP, as illustrated by the examples in (3) and (4).4 (4)

Y mae'r ty yn fawr. PRT is-the house PRT big The house is big.'

Suppletion in copulas

77

It is important to recognize that sentences (3) and (4) contrast with sentences containing transitive and intransitive verbs such as (5) where yn is not a predicative particle but still functions as a preverbal particle.5 (5)

Mae ef yn gweithio. be.PRES he PRT work 'He is working.'

The important difference between these various occurrences of yn is the following: the predicative use of the particle triggers the appropriate consonant mutation on the following phrase. Understanding why the predicative yn triggers consonant mutation on the following formative provides an important clue to a structural difference between (3) and (4) on the one hand and (5) on the other. The Welsh soft mutation is summarized in (6). (6)

p- b b->f[v] t-+d d -> dd [6] c-+g g->0 m —> v

Sproat (1982 and 1986) presents an insightful account of soft mutation. Instead of deriving [d] from [t] under lenition, or [6] from [d] under frication, all the variants are derived from an archiphoneme /T/ underspecified for the continuant and voice features. Specification of this archiphoneme takes place in the lexicon. On Sproat's view preflxation and cliticization induce mutation.6 This claim accounts for the fact that the vast majority of prefixes in the language are followed by lenition on the initial element of the following root. For example cam'mis' + deall 'understand' is camddeall 'misunderstand'. Similarly rhag- 'pre' + barn 'judgement' gives rhagfarn 'prejudice'. In virtually all compounds the initial consonant of the second element is also lenited. Gwag 'empty' + Haw 'hand' gives gwaglaw 'empty-handed'. Similarly rheilffordd 'railway' + gorsaf 'station' gives rheilffordd orsaf 'railway station'. From this perspective, the predicate adjective construction entails cliticizing or affixing the predicate adjective to the particle yn. This can be accomplished syntactically by head-raising the predicate adjective to the particle yn. The mutation facts raise the question of why head-raising to the particle should happen with the predicate adjective and not with the other constructions involving yn such as (5). Predicate adjectives have a thematic role to discharge and their subject discharges that theta-role thereby saturating the predicate. On this view the copula is required syntactically but remains semantically inert because it assigns

78

Randall Hendrick

no independent theta-role. The subject of the adjective originates in a small clause at D-structure and is raised to the subject of the copula to satisfy Case Theory. I will assume that NPs can be licensed for the purposes of Case Theory in either the specifier position of AGRP or of TP. Building on ideas developed in Pesetsky (1994), I will also take it for granted that functional categories may be specified to occur on the left edge of a phrasal domain. 7 A version of Pesetsky's notation would state that T is specified LE(T, CP) in Welsh, meaning that it must appear at PF on the left edge of CP. The only way to Case-license the subject of the small clause and satisfy the left-edge requirement on T is to raise the small clause subject to SPEC, TP and raise V to T and then to AGR. 8 Thus the structure of (4) is (7). (7)

[cp[AGRpMae7[TP'r t y ^ t v p ^[ASPP yn fawrJAp'/[^]]]]]]]]-

I assume here that yn is a functional head of the category ASPP, but is crucially not an L-marker.9 As a consequence, the direct raising of y(r) /j>out of AP to subject position in [SPEC, TP] of (7) is not possible because it would cross at least one barrier, ASPP, preventing proper government of its trace. ASPP will inherit barrier status from the AP that it governs but does not L-mark. It is this fact that leads mawr to raise to yn by head-to-head movement, capturing Sproat's (1982) suggestion that yn is a 'prefix' dependent on the adjective. After raising mawr to yn, yn has the necessary lexical material required to L-mark, and ASPP is no longer a barrier. 10 On this account, a different structure must be given to other occurrences of yn. The subject in (5) cannot originate within a small clause embedded under yn, as I claimed for the predicate adjectives. If it were to originate there, we would expect the verb to raise to yn, just as the predicate adjective does, and exhibit consonant mutation. For this reason I posit that there is a structural asymmetry between the D-structure position of subjects in predicate adjectives and in clauses with noncopular verbs.11 In (5) the subject appears outside yn + VP at D-structure. For the moment we can assume that it is a D-structure specifier of the ASPP headed by yn.12 In this way (5) has an S-structure like (8) after mae fronts to clause-initial position to satisfy the left-edge requirement of T by raising to AGR. (8)

icptAGRpMae^T-pefyfvp/^^pp^-ynlvpgweithio]]]]]]].

I am suggesting here that the external theta-role of the VP is assigned compositionally through ASPP and its functional verb mae, much as in Tenny (1987).

Suppletion in copulas

79

2.2 Substantive copulas versus functional copulas Predicate adjectives in Breton look quite different from their Welsh counterparts. We have already seen in examples (l)-(2) that the predicate adjective positions itself between the finite verb and the subject. The following examples reemphasize this fact.13 (9)

(10)

N'eo ket kenbihan-se e di. NEG.isNEG as small-that his house 'His house is not so small as that.' N'eo ket ken krenv an avel hirio. NEG is NEG as strong the wind today T h e wind is not as strong today.'

In Hendrick (1994) I propose that these sentences have the D-structure in (11). (11)

ARGP AGR

TP

AP NP

AP

I conjecture that the copula eo/oa is a spell-out of T + AGR and that the subject of an AP is not in a Case-marked position. What is more, neither T nor AGR are L-markers. Given these assumptions, the subject of the small clauses must raise to [SPEC, TP] for the purposes of Case Theory. As a result, A raises to T in order to L-mark the AP and permit raising of the NP to [SPEC, TP]. The A + T complex raises further to AGR to satisfy a left-edge requirement on Breton T. 14 Crucially, the Breton copula is not a lexical verb, and differs in this respect from the Welsh copula in (4).

3 The syntactic consequences of suppletion in the third-person present form of the copula Let us now consider the paradigm for the Welsh copula bod given in (12). The pattern here is straightforward: the verbal stem appears to be (yd)\vy combined with agreement suffixes and either an affirmative prefix r-, or a negative prefix d-.

80

Randall Hendnck

(12)

Interrogative 1st 2nd 3rd

Singular wyf, ydwyf wyt, ydwyt yw, ydyw, oes

Plural ym, ydym ych, ydych ynt, ydynt

rwyf, rydwyf rwyt, rydwyt mae, oes, sydd

rym, rydym rych, rydych rynt, rydynt

dwyf, dydwyf dwyt, dydwyt dyw, dydyw

dym, dydym dych, dydych dynt, dydynt

Affirmative

1st 2nd 3rd Negative

1st 2nd 3rd

There is a departure from this straightforward pattern in the third-person affirmative. The paradigm for the Welsh bod would lead us to expect rydy in the affirmative third-person singular when in fact we find the suppletive forms mae and sydd. In the interrogative mood ydy appears as expected and similarly for the negative dydy. (The form oes is used with an indefinite subject obligatorily in the negative and interrogative moods, optionally in the affirmative mood.) I would like to pursue in greater detail the appearance of the suppletive form sydd and mae in the affirmative mood. When a subject is extracted in Welsh we typically find the structure NP/ + a + V + f/ as illustrated in (14). In Hendrick (1988) I argue that the complementizer a is forced to appear in structures like (14) in order to license the trace for the purposes of the ECP. In this way the trace of the dislocated subject is not properly governed in (15). Yet bod 'be' curiously departs from this otherwise regular pattern: when a subject is extracted over bod, we find [NP/sydd tt...] as in (17) rather than [NPjaydyti...] as in (16). (13)

(14)

(15)

(16)

Darllenodd John y llyfr. read.PAST John the book 'John read the book.' John a ddarllenodd y llyfr. John PRT read.PAST the book 'John read the book.' *John y /0 darllenodd y llyfr. John PRT/ 0 read.PAST the book 'John read the book.' *Rhys a ydy yn athro. Rhys PRT be.PRES PRT teacher 'Rhys is a teacher.'

Suppletion in copulas (17)

81

Rhys sydd yn athro. Rhys be.PRES PRT teacher 'Rhys is a teacher.'

The formative sydd surfaces only when a subject is extracted, as the ungrammatically of (18) suggests. (18)

*Sydd Rhys yn athro. be.PRES Rhys PRT teacher 'Rhys is a teacher.'

In addition, sydd remains invariant regardless of whether a singular or plural subject is dislocated over it. This fact can be seen by contrasting (17) with (19) (19)

Y dynion sydd wedi darllen y llyfr. the men be.PRES PRT read the book 'The men read the book.'

Summarizing our observations thus far, we have identified three idiosyncratic properties of the suppletive form sydd that seem true of all dialects of Welsh: 1 Sydd appears only with a dislocated subject. 2 Sydd fails to co-occur with the preverbal particle a that typically licenses the traces of dislocated subjects for the ECP. 3 Sydd does not vary for number; it remains invariant. Suppose that there is no privileged level of lexical insertion, and that lexical items can be introduced to a phrase marker after the operation of move-a. To account for the appearance of sydd one could frame a lexical entry for sydd that inserts it under an appropriate C + V + AGR complex when adjacent to a trace. This amounts to stipulating the first property of sydd noted above, that it only cooccurs with dislocated subjects. To encode properties B and C, let us further stipulate that sydd is co-indexed to the trace on its right. (20)

sydd, [C[[V, T] AGR]],

[NP /,]

The structure in (21) exemplifies how sydd will surface after lexical insertion.

82

Randall Hendrick

(21)

C

AGR NP;

sydd,

As a lexically listed suppletive form, sydd will morphologically block out (in the sense of Aronoff 1976) the regular form of the copula, ydy, in the context of insertion specified by the subcategorization restriction in (20). In Hendrick (1988) I have argued that both the complementizer a and the overt rich agreement traditionally termed synthetic agreement serve as proper governors for traces in Welsh and Breton. Where one of these elements must remain in a structure to license a trace, the complementizer a takes precedence over the synthetic agreement. In Hendrick (1994) I argue that a deletion operation affects synthetic agreement and the complementizer a eliminating one or both so long as independent principles of Universal Grammar such as the ECP are satisfied. We might choose to equate this operation with the Principle of Telegraph proposed by Pesetsky (1994).15 Telegraph applies to delete functional categories subject to absolute universal principles (such as the Recoverability of Deletion Principle) and a series of (partially) ordered relative constraints that prefer some morphemes at constituent boundaries.16 These constraints may vary in their ordering between languages and operate on pools of possible structures to select an 'optimal' structure at PF. Where one of these elements must remain in a structure to license an empty category, the a in C takes precedence over synthetic agreement. In Pesetsky's framework we could stipulate that a, synthetic agreement (SAGR) and sydd all require being on the left boundary of a clause but that these claims are prioritized by the hierarchy in (22).17 Structure S/ that violates a more highly ranked constraint defers to a competing structure S, that violates less highly ranked constraints, or no constraints at all; structures {S/, S,} whose leftmost violations are equally ranked order themselves in such a way that S7 defers to S7 if S/ violates a more highly ranked constraint than S,. This procedure is applied recursively from left to right on the table and defines the most preferred, or 'optimal', structure or set of structures. The table in (23) is intended to show graphically that sydd without the a complementizer or synthetic agreement forms the optimal structure in this sense. Violations of a constraint are indicated by *; constraints that make a structure non-optimal are signalled by!

Suppletion in copulas (22)

83

Recoverability Principle < ECP < LE(sydd, CP) < LE(a, CP) < TEL < LE(SAGR, CP)

(23)

RCV

ECP

LE(sydd,

LE(a,

CP)

CP)

*Y dynion, syddSAGR /, wedi darllen y llyfr.

LE(SAGR, CP)

*

*t

*Y dynion, a sydd tt wedi darllen y llyfr.

TEL

* *

Y dynion, sydd tt wedi darllen y llyfr.

By the ranking of constraints in (22) Telegraph will eliminate synthetic agreement in favour of the complementizer a, and will eliminate both of those in favour of sydd, so long as the Recoverability Principle is respected. We have modelled the general properties A, B and C of sydd that hold in Welsh generally. There is, however, considerable variation as to the co-occurrence of sydd with the other complementizer: the negative {ni)djna\ and the affirmative (y)r.n I turn now to account for the variation involving the negative complementizer. The negative particle (ni)d/na is in complementary distribution with the particle a, a fact that is readily accounted for if we class them both as complementizers competing for the same structural position. (24)

a. y

dyn na welodd Wyn

the man PRT see.PAST Wyn

'the man that Wyn didn't see' b. y dyn a welodd Wyn the man PRT see.PAST Wyn

'the man that Wyn saw' c. *y dyn a na/naa welodd Wyn the man PRT PRT see Wyn 'the man that Wyn didn't see' In the negative construction of formal Welsh, just where a cannot appear the synthetic agreement appears instead, precisely as the hierarchy of edge effects in (22) would lead us to expect. That is to say, in the negative a verb agrees in person and number with a preposed subject, as can be seen in (25). (25)

a. y

dynion na

the men

ddaethant

NEG come.PAST.3PL

b. *y dynion na ddaeth the men NEG come.PAST In some dialects the negative complementizer can co-occur with sydd, as in (26), although it may not appear with a. This asymmetry between the negative

84

Randall Hendrick

complementizer and a should be attributed to a difference in how Telegraph applies modulo the Recoverability of Deletion Principle: while a can be deleted, the application of Telegraph to na would delete non-recoverable material. It is interesting that in such dialects sydd does not change its shape to show number agreement as the verbs in (25) do. (26)

Dyma'r bechgyn na sydd yn canu yn y cor. here the boys NEG be.PRES PRT sing in the choir There are the boys who will not be singing in the choir.'

This fact shows that even when a is not present synthetic agreement is blocked; we deduce that there must be another element that is able to satisfy the ECP and that is higher than synthetic agreement in the hierarchy (27). Our decision to recognize sydd as a licenser for the ECP by giving it an index was on the right track. (27)

Recoverability Principle < ECP < LE(na, CP) < LE(sydd, CP) < LE(a, CP) < TEL < LE(SAGR, CP)

(28)

RCV

LE(«a, CP)

LE(sydd,

LE(a,

CP)

CP)

TEL

LE(SAGR, CP)

*

Y dyn na sydd t wedi.

*

*Y dynion na syddSAGR I wedi. *Y dyn sydd / wedi.

ECP

*!

* *

*

Other dialects of Welsh do not allow sydd to appear with the negative complementizer na(d). Just in these cases we find the non-suppletive form of the copula ydy that co-occurs with synthetic agreement and a resumptive pronoun in subject position. 19 (29)

Dyma'r tim nad ydy e (ddim) wedi ennill gem eleni. here-is-the team NEG be.PRES it (NEG) PRT win game this-year 'Here's the team that hasn't won a game this year.'

Empirical considerations like those outlined in Shlonsky (1992) might lead us to posit a constraint that prefers traces to resumptive pronouns. Call this constraint Avoid Resumptive Pronoun (ARP). Dialects that accept (26) might be thought to rank this constraint relatively highly while dialects that prefer (29) rank it lower. More specifically, the dialect accepting (29) would be (32) while the dialect preferring (26) would be (30). These rankings will prefer nad ydy in the former case and na sydd in the latter as the tables below show. (30)

Recoverability Principle < ECP < ARP < LE(na, CP) < LE(sydd, CP) < LE(a, CP) < TEL < LE(SAGR, CP)

85

Suppletion in copulas (31)

RCV

ARP

ECP

LE(na, CP)

LE(sydd, CP)

LE(a, CP)

*y dynion, na syddSAGR /, wedi darllen y llyfr

*!

*

*

*!

* *

*

*!

*y dynion, nad ydy pronoun, wedi darllen y llyfr

(32)

LE(SAGR, CP)

*

y dyn na sydd / wedi

*y dynion, sydd /, wedi darllen y llyfr

TEL

Recoverability Principle < ECP < LE(na, CP) < LE(sydd, CP) < ARP < LE(fl, CP) < TEL < LE(SAGR, CP)

(33)

RCV

ECP

LE(na,

LE(sydd,

CP)

CP)

*y dynion, na sydd /, wedi

*!

*y dynion, na syddSAGR tt wedi

*!

*y dynion, sydd f, wedi

ARP

TEL

LE(SAGR, CP) *

* *

*! *

y dynion,- nad ydy pronoun,- wedi

The Breton paradigm for bezan is as in (34) and offers a valuable contrast to the Welsh paradigm examined above. 20 (34) Indicative 1st 2nd 3rd Habitual 1st 2nd 3rd

Singular on out eo, a zo, eus

Plural omp oc'h int

bezan bezez bez, a zo, eus

bezomp bezit bezont

The third person shows considerable complexity here as well, but it departs somewhat from Welsh. The form eus is tangential to our interests, and I set it aside.21 What is important for our purposes at this juncture is the special form a zo, which is employed just when the subject is extracted, parallel to Welsh sydd. Our conjecture that Breton differs from Welsh in that the former has only a grammatical copula spelling out grammatical features of tense and agreement by phonological rule makes it impossible to extend our treatment of the Welsh sydd to its Breton counterpart. In Welsh we found an irregular non-complex lexical item that morphologically blocks the appearance of a rule-governed complex form. Breton should not have a suppletive form in this sense. We thus expect that,

86

Randall Hendrick

despite the superficial similarity between Welsh sydd and Breton a zo, they should have a very different syntactic distribution. It is crucial that, unlike its Welsh counterpart sydd, the formative a zo is never able to co-occur with the negative complementizer. (35)

(36)

(37)

Mona a zo brav. Mona be.PRES fine 'Mona is fine.' Mona n'eo ket brav. Mona NEG be.PRES NEG fine 'Mona isn't fine. *Mona n'a zo ket brav. Mona NEG PRT be.PRES NEG fine 'Mona isn't fine.'

The unacceptability of (37) is most naturally reduced to the descriptive observation that ne and a compete for the same structural position, presumably C (see Hendrick 1991). This claim in turn suggests that the distribution of Breton a zo is sensitive to the presence of the (affirmative) complementizer a. While the Welsh sydd is apparently triggered by a following trace, the Breton a zo is restricted to co-occur with a trace indirectly; it is limited to appearing with a, which itself only appears with traces in affirmative clauses; hence the restricted distribution of a entails only co-occurring with a trace in affirmatives. Let us posit the language-specific rules in (38) to account for the behaviour of the Breton copula. (38)

Breton: a (e)o = > a z o

The Breton rule (38) is purely morphophonemic in that it only changes the phonological shape of the copula when adjacent to the complementizer a; the Welsh sydd not only changes the shape of the copula but has a syntactic effect as well because it carries an index licensing traces. The Breton preverbal particle a in (37) is in C and is co-indexed with a subject. It, rather than the functional copula, serves to license the trace in subject position for the ECP, as I have argued in Hendrick (1988). The change in the shape of the Breton formative is syntactically inert because it does not alter a syntactically relevant dimension of the form. Welsh, however, behaves differently: the suppletive verb stem serves as an antecedent governor for the ECP and Telegraph eliminates functional categories such as C and AGR in its favour because of its position on the Pesetsky-style optimality hierarchy. Corroborating evidence for the treatment of the Breton copula offered here comes from a copula form that is reserved for locatives.

Suppletion in copulas (39)

Singular 1st emaon 2nd emaout 3rd eman

87

Plural emaomp emaoc'h emaint

These forms should be analysed as a substantive copula. I do not place this form in the paradigm of the grammatical copula bezan; rather, I regard it as a substantive verb formed on the stem ema 'here'. This substantive copula does not appear readily with adjectives but to the extent that it does so marginally, it does not require raising of the adjective to the copula. Thus, Press (1986) cites (40). (40)

?eman va breur fur. is my brother good 'My brother is good.'

My claim that eman is a lexical auxiliary converges with the analysis of Breton auxiliaries offered in Borsley et al. (this volume).

4 The distribution of Welsh mae If we contrast the paradigm of the Welsh copula with its Breton counterpart the presence of the Welsh mae is striking because, unlike sydd, it has no Breton counterpart. Welsh mae is of further interest because it is another suppletive form, and is limited to present affirmative contexts. We obviously need to ask what determines its distribution. Welsh mae has the lexical entry in (41).22 T (AU \\V 1 AGR 1 lfM 1 (41) mae, , , , , , + (y)r v J ' |_|_ [-past]\ [-person]\' Lwy This statement is intended to guarantee that mae appears in third-person present-tense structures. The specification that mae is inserted adjacent to the complementizer (y)r will insure that it surfaces only in affirmative clauses.23 In some dialects of spoken Welsh, mae need not co-occur with (y)r, although in formal Welsh it must. These facts will not follow simply by making the adjacency statement optional since Welsh contains interrogative and negative complementizers that we must prevent mae from co-occurring with. We can, however, account for this variation by appealing to the statement of left-edge requirements of mae. In formal Welsh, mae must be on the left edge of AGRP, as in (42), while in spoken Welsh mae must be on the left edge of CP, as in (43).

(42) (43)

LE(mae, AGRP) LE(mae, CP)

In spoken Welsh the left-edge requirement (43) of mae will be ordered higher

88

Randall Hendrick

than that of (y)r9 making it possible for Telegraph to delete (y)r. This treatment makes the distribution of mae dependent on the adjacent complementizer (much as our analysis of Breton a zo does). We correctly predict that when any complementizer other than the affirmative declarative complementizer is present in Welsh, mae will be suppressed in favour of the regular form ydy. Specifically, the negative complementizer (na(d) in formal Welsh, d- in spoken Welsh), or the interrogative complementizer (a null complementizer in spoken Welsh, a in formal Welsh) will both prevent the appearance of mae. Rouveret (this volume) takes a very different approach to the distribution of mae. In essence, he looks for a syntactic/semantic explanation for why mae is limited to affirmative declarative present-tense sentences. The basis of his explanation is in many respects appealing. Rouveret embeds his analysis within a framework similar to Diesing's (1992), which distinguishes stage-level predicates, which have a spatio-temporal ('Davidsonian') argument in need of binding, from individual-level predicates, which lack such a spatio-temporal argument. Diesing (1992) argues that some subjects are VP-internal while others are external. Subjects of stage-level predicates are VP-internal; subjects of individual-level predicates are external to the VP. The motivation behind this division in the class of subjects is the claim that stage-level predicates have an external 'spatiotemporal' argument that is bound. Individual-level predicates lack such a spatiotemporal argument and thereby permit the subject to appear in an argument position external to the VP. Following Carlson (1977), stage-level predicates are transient, while individual-level predicates are relatively stable temporally. Applying Diesing's general analysis to Welsh, Rouveret suggests that mae is a complex form containing the copula and a clitic able to bind a spatio-temporal argument; as such, mae does not appear with individual-level predicates or in structures that have an extra operator available to bind the spatio-temporal argument. The Welsh ydy appears in structures that have an extra operator such as the negative, interrogative or past-tense operators that all serve to license a spatio-temporal argument. Attractive though it is, I do not accept Rouveret's analysis here. On my view morphosyntactic properties of complementizers are relevant to the distribution of mae, rather than the syntactic/semantic character of the predicate. The literature on stage-level and individual-level predicates distinguishes between transitory attributes such as being sad from more permanent ones such as being tall. Yet no such typology of attributes can be correlated with the distribution of mae and ydy. Both stage- and individual-level predicates co-occur with mae in the present affirmative: (44)

Mae Gareth yn drist. is Gareth PRT sad 'Gareth is sad.'

Suppletion in copulas (45)

89

Mae'r bachgen yn dal. is-the boy PRT tall T h e boy is tall.'

The facts of Welsh should be contrasted with those of Irish as described in Doherty (1992). Doherty, like Rouveret, argues that there is variation in the copula that reduces to the stage-/individual-level distinction. Yet the argument in the Irish case is considerably stronger for the reason that the interpretation of the predicates arguably plays a role in the variation of the copula. It is precisely this kind of evidence in Welsh that we lack. Rouveret avoids this problem by suggesting that the stage-/individual-level distinction is not a property of the predicate itself but of the particle that embeds it. From this point of view the particle yn has a semantic structure that introduces a spatio-temporal argument and makes the predicate embedded under it a stagelevel predicate. Now Carlson in part used perception verbs in English to motivate the individual-/stage-level distinction. Carlson's generalization is that under certain perception verbs like see, only a stage-level gerund, one that is transitory, is acceptable. If the particle yn effectively turns all predicates into stage-level predicates by adding a spatio-temporal argument, the two semantic classes of predicates should behave in a uniform manner under perception verbs like see. Yet (46) is acceptable but (47) appears to be semantically odd, much as it is in English, although judgements here are subtle and must be handled cautiously.24 (46)

(47)

Gwelaf i see + PRES I 'I see John Gwelaf i see + PRES I 'I see John

John yn dod ataf i. John PRT come to me coming to me.' John yn dal. John PRT tall being tall.'

If correct and systematic, these kinds of fact would pose a difficulty for Rouveret's syntactic/semantic account for the alternation between mae and ydy.

5 'Anomalous' instances of sydd My treatment of the Celtic copulas outlined in section 2 assumed that subjects of unergative verbs are projected external to the VP while subjects of predicate adjectives are projected in small clauses internal to the VP. 25 The unergatives lead me to deny the correctness of Koopman and Sportiche's (1991) VP Internal Subject Hypothesis. Instead I believe that subjects are given a thematic role compositionally from the verb phrase and the aspectual phrase, in the spirit of Tenny (1987). On this view subjects raise from the highest aspectual phrase to be licensed by Case Theory. 26

90

Randall Hendrick

The terms of the analysis I am advocating allow three specifier positions in which subjects could conceivably occur: specifier of ASPP, the (functional) aspectual verb, specifier of T and specifier of AGR. The left-edge requirement of T prevents the specifier of AGR from having a subject at PF. 27 Of the three positions that remain, two (specifier of ASPP and specifier of the aspectual verb) typically are not positions for Case-licensing with the result that a clause has a single subject. This is purely a morphosyntactic fact, not a semantic one. Other theories, such as Diesing (1990) and Rouveret (this volume), also recognize multiple subject positions, but these positions are semantically distinguished with the lower position reserved for indefinites. If there were a way to Case-license an NP in the specifier of the other functional head, we could have more than one subject given our assumptions about clausal structure, and, I suggest, they would have no distinguishing semantic properties. As it happens, Welsh has the means to license a subject in the specifier of ASPP; in such structures we have clauses with two 'subjects', and there is no observable restriction on the definiteness of those subjects. Awbery (1977) describes what she terms some 'anomalous' occurrences of sydd. The sentences in (48) and (49) exemplify the construction that Awbery has in mind. (48)

(49)

y dyn y mae Wyn wedi bod yn son amdano the man PRT be.PRES Wyn PRT be PRT talking about.3SG 'the man that Wyn has been talking about' y dyn sydd a Wyn wedi bod yn son amdano the man be.PRES with Wyn PRT be PRT talk about.3SG 'the man that Wyn has been talking about'

Given our discussion of the distribution of sydd in section 3, we can infer with some confidence that the example in (49) must have a structure in which a trace is to the right of sydd in subject position. Yet Wyn is the logical/agentive subject of this sentence and appears to the left and outside the VP bod yn son amdano. We can make sense of these by giving this sentence the structure in (50).

Suppletion in copulas

91

(50)

bod

Wyn,

ASP ASP

wedi v bod

VP ASPP ASP

VP

To produce the standard (48) Wyn raises from [SPEC, ASPP] to [SPEC, TP] to receive Case. Bod (realized as mae) raises to T and AGR, where it satisfies its leftedge requirements at PF. An operator co-indexed to the NP y dyn binds a variable functioning as the object of the preposition amdano. The anomalous relative in (49) results if another Case-assigner, the preposition a, is inserted, to Case-license Wyn in [SPEC, ASPP]. This leaves [SPEC, TP] vacant. The object of the preposition is then free to raise to [SPEC, TP] and then to the [SPEC, CP]. Because there is a trace in the subject position, bod takes the form sydd. Essentially (49) involves a structure with two 'subject positions' made possible by the presence of the extra Case-licenser. The logical subject in such structures appears as the specifier of the aspectual phrase where it is projected and receives its thematic role. Yet crucially this subject is a proper name and definite. If this is the correct analysis of sentences like (49), then the anomalous relative

92

Randall Hendrick

construction poses a challenge to theories that claim that the subject positions are semantically distinguished, with the lower position reserved for indefinites.28

6 Conclusion In the Standard Theory of Chomsky (1965) and Chomsky and Halle (1968), readjustment rules mediated the interface between syntax and phonology and accounted for idiosyncratic changes in a particular lexical item when concatenated with affixes, such as the replacement of go+PAST by went. In general, it has been assumed since the highly influential work of Aronoff (1976) that such irregular forms are lexically listed and not the result of re-adjustment rules. Nevertheless the spirit of the Standard Theory's claim that irregularity of these forms is syntactically inert has generally been preserved. I have argued that phenomena like suppletion, which re-adjustment was held to account for, could potentially have syntactic effects. In particular I have defended the claim that the suppletion in the third-person form of the Welsh copula has a strong syntactic effect, contributing to the licensing of empty categories for the purposes of the ECP. Of course, this conclusion contrasts sharply with a variety of views that take syntax to be immune from the morphology of words, as in, for example, Di Sciullo and Williams (1986). It is also worth pointing out that it stands at odds with theories such as Halle and Marantz's (1992) that relegate suppletion to the spelling-out of abstract syntactic elements at some postsyntactic, morphological level. Instead I have argued that suppletive forms must be present at LF, where principles like the ECP are checked. This conclusion about suppletion is part of a larger working hypothesis that identifies morphosyntactic variation as the locus of differences in the syntax of copulas both between languages and within a single language. This position is forced on us if we maintain that the copula is semantically vacuous; any variation it exhibits must come from the interface of syntax with PF. Evidence for the importance of morphosyntactic variation can be drawn from the behaviour of the Welsh copula as contrasted to its Breton counterpart with respect to adjacent complementizers and traces. Morphosyntactic variation is also at the root of the suppletion within a language as my discussion of the Welsh forms of bod (ydy, mae and sydd) has tried to show. Even the number and the character of the subjects of bod is reduced to morphosyntactic variation.

Notes 1 Without tarring them in any way with the positions defended here, I would like to thank my friends Gwen Awbery, Bob Borsley, Melynda Dunigan, Joe Emonds, Steve Harlow,

Suppletion in copulas

93

Alec Marantz, Jim McCloskey, David Pesetsky, Alain Rouveret and Gert Webelhuth for their thoughts on many of the issues touched on here. Carnie (1993), in a fascinating treatment of Irish, follows a similar objective. His work places the morphosyntactic variation in features of the Irish tense system rather than in the lexical/functional distinction I appeal to here. I leave to future work which of these avenues is to be preferred. Doherty's treatment reduces the syntactic differences between these two copulas to a semantic difference in stage- and individual-level predicates. Carnie (1993) shows that important empirical problems dog such a reduction, just as Schmitt (1992) does for similar attempts in Spanish and Portuguese. I take the sentence-initial particle to be an instance of C. See Rouveret (1990) for a different point of view. The examples in (3) and (4) are adapted from Williams (1980) and Sproat (1985) respectively. The verb gweithio in (5) is termed a 'verbal noun' in traditional descriptions. For the purposes of this chapter I will follow Sproat (1985), Sadler (1988) and Borsley (1993) in taking these elements to be verbs. The preverbal particle yn is homophonous to the preposition yn as exemplified in (i). (i)

Mae'r ferch yn y ty. be.PRES-the woman in the house 'The woman is in the house.'

See Rouveret (this volume) for another view. Harlow (1989) shows that a set of phrasal domains in Welsh also trigger mutation: essentially nouns adjacent to phrasal boundaries undergo consonant mutation, as in (i). (i)

Mae [PP yn yr ardd] [NP gi mawr]. is in the garden dog big 'Here is a big dog in the garden.'

(ci)

The phenomenon is pursued further in Borsley and Tallerman (1996). I set it aside here as tangential to my main concern. More traditionally it is claimed that a head can only Case-license to its right in Welsh. The line of thought I am pursuing factors out the linear order requirement from the licensing requirement. We may also have to say that the AGR + T + V complex raises to C, or that C lowers. I will assume the latter hypothesis here. I follow Pesetsky in assuming that X counts as standing on the left edge of a constituent YP so long as X is part of some W° that is adjacent to [yp. T in Breton is not specified LE(T, CP) and as a consequence appears to be marked --LE(T, CP). It does appear to carry the specification LE(T, AGRP), however. For expository convenience I assume throughout this chapter the version of the ECP outlined in Chomsky (1986), where the concepts L-mark, barrier and inheritance are introduced and defined. The verb mae is a functional verb in the sense of Abney (1987) embedding ASPP in this structure.

94

Randall Hendrick

10 One could frame this analysis in terms of the system sketched in Chomsky (1993) by letting yn mawr be a complex lexical form that must have the features of yn checked by raising to ASP. 11 If this assumption turns out to be incorrect, we would conclude that the asymmetry between predicative adjectives and verbs reduces to the asymmetry in compounding familiar from traditional grammars of Welsh of proper compounds (that induce consonant mutation) and improper compounds (that do not). See Thorne (1993) on this distinction. In Hendrick (1993) I provide some evidence that the asymmetry at issue here is structural. See note 23 for a brief summary of that argument. As Bob Borsley points out to me, the behaviour of unaccusative verbs and raising verbs will prove decisive on this score. I set aside the proper treatment of unaccusative verbs for reasons of space; I hope to address this topic in a subsequent paper. Raisingto-subject predicates should parallel the predicate adjectives. If raising predicates appear in the periphrastic construction with bod yn, we would expect the raising verb to move up to the aspectual particle yn, just as predicate adjectives do. The evaluation of this prediction is complicated by the fact that there are few clear cases of raising predicates in Welsh and even fewer that appear in the periphrastic construction. Apparent counterexamples to the prediction would have to be reanalysed as involving control of PRO, movement to an A' position, or as simplex clauses, unless one decides to adopt the type of analysis of raising predicates in Chomsky (1986), where the raising verb itself serves as a derived antecedent of the raised subject. This latter option would treat the raising predicates as parallel to the Welsh suppletive form sydd (be) that I discuss in the text below. The situation is more straightforward in Breton where there are clear raising verbs that appear in periphrastic constructions. In such structures the raising predicate undergoes head movement to fuse with the aspectual particle that is an affix. The Breton raising predicate seblant (seem) has the periphrastic form bezan (be) seblant + et (seemed). 12 This position is motivated in Hendrick (1991) and Rouveret (this volume). A generally similar view is sketched in Borer (1993). 13 These examples are adapted from Trepos (1980: 109-10). 14 I assume that Breton has the specification LE(T, AGRP). Carnie and Harley (1994) makes a similar argument that predicate adjectives and nominals raise to tense in Irish. 15 Grimshaw (1993) explores subject-auxiliary inversion in a similar spirit. 16 Here Pesetsky follows the lead of work in prosodic morphology; see, for example, McCarthy and Prince (1993). 17 In Hendrick (1994) I attribute this preference to the Elsewhere Principle: because the synthetic agreement has more grammatical features, such as person and number, defining it the deletion operation affects it before the a in C. In the structures carrying the suppletive sydd and a complementizer, a will not surface in C because it has more grammatical features, such as Case, than the copula which I take to be vacuous, predictable from its co-occurrence with the preverbal particles yn and wedi. The Elsewhere Principle leads the grammar to prefer the elimination of a rather than sydd. While attractive, this approach says nothing about the behaviour of the complementizer y that I discuss in the next note.

Suppletion in copulas

95

18 In formal varieties of Welsh one finds sydd co-occurring with the complementizers. As (22) would lead us to expect, the complementizer that sydd appears with is y rather than a (see Williams 1980: 96). This suggests that we should integrate y into our hierarchy by the minimally different (i) or (ii). Hierarchy (i) will yield the dialect in which sydd precludes both complementizers a and y; (ii) will produce the dialect which permits sydd to co-occur with y but not with a. (i) (ii)

Recoverability Principle < LE(sydd, CP) < LE(a, CP), LE(y CP) < LE(SAGR, CP) Recoverability Principle < LE(sydd, CP) V LE(y CP) < LE(a, CP) < LE(SAGR, CP)

19 This example is taken from Gramadeg Cymraeg Cyfoes (1985: 78). 20 This paradigm is adapted from Press (1986: 146). 21 It only co-occurs with indefinites, parallel to Welsh oes (although eus is regularized so that it alternates with eo in all clauses). 22 I assume with Kayne (1989b) that first person = [+ speaker, + person], second person = [-speaker, + person] and third person = [—speaker, -person]. Tense = [ + PAST] and present = [-PAST].

23 The Welsh mae does not co-occur with topicalized predicate adjectives; instead the form ydy appears (e.g. (i)). Topicalized predicate adjectives and VPs behave differently; the latter surface with mae (e.g. (ii)). In Hendrick (1993) I attribute this contrast to the impossibility of the complementizer yr in structures with predicate adjectives for ECP reasons. (i)

(ii)

(iii)

Hir oedden nhw. long be.PAST.3pL they 'They were long.' Darllen mae Sian. read be.PRES Sian 'Sian is reading.' Wedi adrodd roedd e? PRT recite be.PAST he 'Had he recited?'

Mae replaces (y)r + bod in the third-person singular. There is a trace of a raised subject to be licensed in (i) but not in (ii)-(iii), in line with the hypothesized structural asymmetry in how subjects are projected. In (i) the complementizer must delete to avoid making the C' category a barrier by minimality. The AP itself is not a barrier, Adj having raised to the particle yn and L-marked the AP. (The particle yn systematically deletes in fronted constructions as illustrated by the contrast (ii)-(iii) for reasons I cannot explain in a principled fashion.) The trace is licensed by chain composition under strict government (not m-command), as discussed in Chomsky (1986b). Because yr deletes in (i) ydy surfaces; no such deletion occurs in (ii) and we find mae. See Rouveret (this volume) for another line of explanation. 24 Example (46) is borrowed from Jones and Thomas (1977).

96

Randall Hendrick

25 I ignore here the correct analysis of unaccusative verbs in Welsh, a topic I hope to address in future work. 26 Rouveret (this volume) adopts a very similar analysis. 27 Here I assume that the finite verb raises to AGR but not to C, as I argue in Hendrick (1994). Several of the arguments to the same conclusion for Irish offered in McCloskey (forthcoming) can be duplicated in Welsh. 28 Of course, one could avoid this puzzle by claiming that there is yet another position within the VP that is reserved for indefinite subjects. The problem noted in the text only arises if one accepts the claim that the lowest position that subjects are projected in is outside VP and within ASPP. For reasons of space I do not treat the possessive construction that Rouveret analyses as evincing a privileged position for indefinite subjects. I hope to treat them in a subsequent article.

3

Fronting constructions in Welsh

Maggie Tallerman

Like all the Celtic languages, Welsh very commonly displays word orders other than the canonical VSO, the unmarked surface word order.1 Traditional Welsh grammar distinguishes between two constructions which are non-VSO: the socalled 'mixed' and 'abnormal' word orders. Both display what, in a pretheoretical sense, we might call 'fronting' of some constituent, although the pragmatic function of this fronting is typically claimed to be very different in each case (see, for example, Fife and King 1991; Watkins 1991; and other papers in Fife and Poppe 1991). In this chapter I consider the derivation of each of these constructions from the point of view of the principles-and-parameters framework of generative grammar. I will show that the 'mixed' construction, which I refer to as the cleft construction in the text, is best analysed in terms of the CP-recursion analysis of Rizzi and Roberts (1989) and Cardinaletti and Roberts (1991). We will see that the clefted constituent is sited in the specifier position of a CP which is itself the complement to a higher complementizer. On the other hand the 'abnormal' sentences (a term I keep for the sake of convenience) actually involve adjunction to a matrix CP, and no CP-recursion is involved. The chapter is structured as follows: section 1 outlines the data and illustrates the major differences between the two construction types. Section 2 discusses some previous generative treatments of one of the types of fronting under discussion, the cleft, and shows that these analyses share a common problem when more data are taken into consideration. Section 3 presents an alternative analysis of both the cleft and the abnormal constructions, and section 4 defends this analysis in detail.

1 The data Sentences which do not have the unmarked VSO word order in Welsh occur in what are traditionally considered to be two distinct clause types, known as the mixed sentence (here referred to as the cleft type) and the abnormal sentence. The examples in (1) and (2) illustrate these two clause types respectively, in each case showing a fronted subject NP: 97

98

Maggie Tallerman

(1)

a. Myfi a gafodd anrheg. me PRT got.3sG gift 'It was me that got a gift.' b. Y dynion a werthodd y ci. the men PRT sold.3sG the dog I t was the men who sold the dog.'

(cleft)

(2)

a. Myfi a gefais anrheg. me PRT got. ISG gift 'I got a gift.' b. Y dynion a werthasant y ci. the men PRT sold.3p the dog T h e men sold the dog.'

(abnormal)

The clause type in (2) is commonly claimed to have no emphatic value: it appears to be a topicalization rather than a focalization of some constituent, and is typically translated into unmarked SVO order in English. See, for example, the discussion of these constructions in Fife and King 1991, and the remarks made by Williams (1980: 168). On the other hand, the construction in (1) is typically translated as a cleft, and results in contrastive focus on the fronted constituent. The major properties of the two constructions are outlined in (3): (3)

Cleft sentences a. Major function is to focus fronted constituent, although take on function of NP topicalization since demise of the abnormal sentence. b. Productive in all recorded periods of the language. c. [-agreement] - a fronted subject does not agree with the verb in either person (la) or number (lb): the verb always appears in its thirdperson singular form. d. Most constituents may occur in initial position; not restricted to NP. e. The negator is placed before the fronted constituent. f. The construction is not iterative: only one constituent can be fronted. g. Fronting can occur in subordinate clauses. h. The fronted XP is typically preceded by a complementizer (etymologically a copula) in subordinate clauses, and in earlier periods, in main clauses too. Abnormal sentences a. Exist to 'topicalize' some constituent(s): no focus value (Fife and King 1991). b. Only productive in Middle Welsh (although possibly existed up to this century; see Williams 1980: 168). c. [ +agreement] - a fronted subject triggers full agreement in person (2a) and number (2b) on the inflected verb.

Fronting constructions

99

d. Any constituent may occur in initial position; not restricted to NP. e. The negator is placed between the fronted item and the main verb. f. The construction is iterative: as many as five constituents have been found to be fronted. g. Fronting cannot occur in subordinate clauses. h. The fronted XP is never preceded by a complementizer or copula. Of course, from a purely formal point of view, there is no superficial distinction between the two constructions in many cases: if the fronted constituent is singular, if there is no negation, and if the constituent is fronted from a main clause, then it is only on functional grounds that there will be anything to choose. However, where there are surface points of divergence, then we can clearly tell that there is something to be accounted for. I do not intend to investigate the pragmatic functions of the two constructions; see many of the papers in Fife and Poppe (1991) for such discussion. It does seem clear, however, that the abnormal construction, which was so startlingly prevalent and productive in Middle Welsh literature - to the extent that the majority of sentences are not verb-initial - was a stylistic device which did not generally produce contrastive focus on the fronted constituent(s). As an example of this consider (4), a typical example of Biblical Welsh: (4)

A'i ddisgyblion a ddaethant ato ac a'i deffroasant. and.3sG.M disciples PRT came.3PL to.3sG.M and PRT-3SG.M woke.3PL 'And his disciples came to him and woke him.' (Matthew viii, 25)

Although no construction with the particular syntactic form of the abnormal construction exists in Modern Welsh, the 'mixed' sentence has a dual role: it exists mainly to cleft a focused constituent, but also can convey non-emphatic topicalization. To that extent the modern cleft construction seems to have taken over the function of the abnormal construction. This covers points (a) and (b) in (3); I will briefly illustrate the remaining points of similarity and divergence. Point (c) is one of the major distinctions between the two constructions, and must clearly be accounted for in any analysis of them: why do we find subject agreement in the abnormal type but not in clefts? Since the cleft construction has exactly the same superficial form as the relative clause in Welsh (at least in the case of clefted NPs) it seems likely that the same syntactic processes underlie both constructions. In terms of point (d), there seems to be little distinction: not only NP can be fronted, but also PP, VP and AP (although the latter is not good for all speakers in modern clefts).

100

Maggie Tallerman

(5)

a. [PP Ym Mangor] y siaradais i llynedd. in Bangor PRT spoke. ISG I last.year 'It was in Bangor that I spoke last year.' b. [yP Pori'r comin a'r cloddiau] a wnaeth browse-the common and-the hedges PRT did.3sG Ifas am y lleill. Ifas for the others 'Ifas browsed the common and the hedges for the others.' c. A [Ap chalet] uu yr arueu ereill. and hard was the arms other. 'And the other arms were hard.' (Ystorya de Carlo Magno,2 85.8)

In (5a) and (5b) we have wide focus of PP and of VP from Modern Welsh, both cleft examples, and in (5), a fronted AP from Middle Welsh, considered to be an abnormal sentence by Fife and King (1991: 82), who cite the example. Point (3e) concerns the different placings of negators in the two constructions. Example (6a) illustrates the negation placed before the fronted constituent in the cleft construction, and (6b), the negation placed after the fronted constituent in the abnormal construction. We will see in section 4.4 that the only negation which is possible in the latter case is clausal negation, whilst the cleft construction can have not only the cleft constituent itself negated, as here, but also may have clausal negation. This distinction is another important point of divergence between the two constructions, and is predicted by the analysis in the main body of the chapter. (6)

a. Nid/dimy NEG

dyn a

ddaeth.

(cleft)

the man COMP came.3sG

I t wasn't the man who came.' b. Y dyn ny daeth. the man NEG came.3sG T h e man did not come.'

(abnormal)

(Evans 1989: 180) Point (30 concerns whether or not there can be more than one constituent in the fronted position. In the Middle Welsh abnormal construction, there could be, as (7) shows: (7)

Ac [wrth henny] [er rey ereyll] [en kyflavn o ofyn] and at that the ones other PRED full of fear [adav e dynas] a orugant. leave the city PRT did.3PL 'And at that, those others, full of fear, left the city.' (Brut y Brenhined, Llanstephan MS. 1 Version,3 712-13)

Fronting constructions

101

However, in contrast, only one constituent can be clefted, as (8) shows: (8)

a. Gwnaeth y rhai eraill adael y ddinas. did.3sG the ones other leave the city Those others left the city.' b. * Dywedodd mai [y rhai eraill] [adael y ddinas] a wnaeth. said.3sG PRT the ones other leave the city PRT did.3sG ('He said that those others left the city.') c. Dywedodd mai [NP y rhai eraill] a wnaeth said.3sG PRT the ones other PRT did.3sG e adael y ddinas. leave the city 'He said that it was those others who left the city.' d. Dywedodd mai [Vp gadael y ddinas] a wnaeth y said.3sG PRT leave the city PRT did.3sG the rhai eraill e ones other 'He said that it was leave the city that those others did.'

Example (8a) shows normal VSO order. Example (8b) is the cleft equivalent of (part of) (7), and two constituents have been clefted, namely the subject and the VP (non-finite verb plus object), giving the ungrammatical result shown. If either one of these constituents is clefted, the result is grammatical, as (8c, d) show. Not to pre-empt discussion, I indicate the canonical position of the cleft constituent in unmarked word order with e. Clearly, any analysis of the fronting constructions must account for the facts in (7) and (8): why should it be that the abnormal construction allows several fronted constituents, but the cleft construction only one? Point (3g) concerns the occurrence or non-occurrence of fronting in embedded clauses. The abnormal construction appears to be a root construction, only occurring in main clauses, whilst the cleft construction occurs in both main and embedded clauses. Of course, it is difficult to be certain that a construction which is no longer productive has or does not have a particular grammatical property, but since the abnormal construction is so prevalent in Middle Welsh texts, and does not occur anywhere in subordinate clauses there, we can be fairly certain that it would have been ungrammatical. (See also Watkins 1977 and Fife and King 1991: 137.) On the other hand, the cleft construction occurs freely in embedded clauses: (8c, d) illustrate with a fronted NP and VP respectively, and (9) shows a clefted PP in an embedded clause:

102

Maggie Tallerman

(9)

Mi wn i mai [PP yng Nghymru] y mae Gwent. PRT know I PRT in Wales PRT is Gwent 'I know that it's in Wales that Gwent is.'

The examples in (8) and (9) also illustrate the particle mai, which occurs in embedded clefts in the position immediately preceding the focused constituent. However, as Jones and Thomas (1977) point out, this particle is not necessarily overt in Colloquial Welsh: (10)

Mae Mair yn gwybod John oedd yn chwerthin. is Mair PROG know John was PROG laugh 'Mair knows that it was John who was laughing.' (Jones and Thomas 1977: 295)

In earlier periods of the language, there could also be an item preceding the fronted constituent in main-clause clefts, although this was optional even in Old Welsh. This example is a cleft from Middle Welsh: (11)

Ysmi a'e heirch, is me PRT-her seeks.3SG 'It's me who seeks her.' (The White Book Mabinogion* 479.29-30)

The item ys which precedes the fronted constituent is a copula, as indeed was the modern cleft particle mai (and also taw, the cleft particle used in South Welsh) in earlier stages of the language. Although no particle precedes the cleft constituent in declarative main clauses in the modern language, we will see in section 4.3 that reflexes of it appear in other cleft clause types, and we will discuss the role of the 'copula' particle in permitting adjunction in embedded clauses. This concludes the illustration of the major types of data which form the basis for our discussion. To summarize, the most important properties to be accounted for in any analysis of these two constructions are their agreement patterns; the appearance of recursively fronted constituents in the abnormal construction, but not the cleft; the appearance of the copula (or the complementizer) in the cleft; the restriction of the abnormal construction to root clauses; and the negation facts.

2 The cleft construction: previous analyses 2.1 An outline of some generative treatments As far as I am aware, previous analyses of the Welsh fronting constructions within a generative framework have not discussed the distinction between the abnormal and the cleft sentence types (see, for example, Harlow 1981; Sproat 1985; Sadler 1988; Hendrick 1988, 1990a; Rouveret 1991, this volume). Instead,

Fronting constructions

103

authors confine their attentions to the cleft construction, which of course has a parallel in relative clauses, at least in terms of clefting of NPs: so, for example, the cleft sentence in (lb) has the same superficial form as the relativized subject NP in (12), which displays the same lack of number agreement as the cleft construction: (12)

y dynion a werthoddy ci the men PRT sold.3sG the dog 'the men who sold the dog'

However, we can also cleft PP, AP and VP, so the constructions are not identical.5 With the possible exception of Rouveret 1991, previous accounts have also failed to discuss the embedded clefts, and in particular the position and analysis of the mai constituent. This is, however, crucial to the analysis of the construction presented here.6 What I will call the standard Government-Binding (GB) analyses of clefts have mostly followed Harlow's (1981) outline rather closely, and have the fronted constituent in Spec, CP (or some earlier analogue, such as Topic position in S"), the preverbal particle a or y in complementizer position, and, in the case of fronted subjects, a gap in canonical subject position in the clause: (13)

S'VXP

(Harlow 1981: 222; Hendrick 1988: 201) (14)

PRT

+

V...

ef

(Sproat 1985: 192; Sadler 1988: 188)

104

Maggie Tallerman

(15)

S"

TOP

(Hendrick 1990a: 147) The authors cited typically consider the fronted constituent to be base-generated in the extraclausal position, where it is followed by what is usually considered to be a complementizer a or y which binds a trace in the canonical subject, etc. position: this is assumed by Harlow (1981) and by Hendrick (1988). Stump (1989) also has essentially this analysis of the topicalized subject in Breton, a construction very similar in form, if not in pragmatic function, to the Welsh clefted subject. An alternative analysis assumes that the particles a or y are not complementizers, but simply verbal particles adjoined to the verb (as in Sadler's account). In this case there is an empty operator in C. The constituency facts concerning the 'complementizers' are notoriously hard to test, since these particles tend not to occur naturally in spoken Welsh. Perhaps because of this, not much evidence to support either position has appeared in the literature. However, in a later paper, Harlow (1983: 83) argues against the complementizer analysis by claiming that it is not possible to detach the particle from the rest of the clause, so giving no analogues to the right-node raising type of construction in English. / think that, but I'm not sure whether, John will come, nor clausal co-ordinations where only one clause has an overt complementizer: Jan said that Mark is clever and May is stupid. If such results obtain, we could certainly assume that the ungrammaticality in Welsh is due to the clitic status of a and y: since the particles must be cliticized to a following finite verb then Harlow's findings are predictable. We can also add that no adverbial may intervene between the complementizer and the following clause, as shown in (16), but this restriction is equally compatible with an analysis where the particles are situated in C° but must be able to cliticize to an adjacent verb.7 (16)

a. *Dywedoddef y, gyda Haw, bydd yn gadael. said.3sG he coMPwith handbe.FUT.3sG PROG leave 'He said that, by the way, he will be leaving.'

Fronting constructions

105

b. *Mae hi'n siwr y heddiw bydd y plant is she.PRED sure COMP today be.FUT.3sG the children yn swnllyd. PRED noisy

'She's sure that today the children will be noisy.' In what follows I adopt what I see as the recent consensus over the position of the particles in the Celtic languages, and analyse them as complementizers; see also Hendrick (1988: 144), Rouveret (1991: 376f.) and McCloskey (1992b). I move on now to discussion of the main problem shared by the analyses discussed in this section.

2.2 A problem and a proposed solution Each of the foregoing analyses suffers from the same drawback when taken at face value: they are incapable of accounting for the embedded cleft constructions illustrated in (8) and (9), because in each case the fronted material, whether it is analysed as Topic, daughter of S", or whatever, follows rather than precedes the complementizer mai. In other words, if mai is in the C-slot, where is the fronted material? Exactly the same problem emerges if we translate the analyses into more recent GB terms: if the fronted material is in Spec, CP, then the natural assumption is that the complementizer mai should be in C: but this, of course, is the wrong way round, as we see in (8) and (9). This observation is taken by Rouveret (1991: 372f.) as evidence that a clefted NP cannot be in Spec, CP, but instead should be in a position internal to the clause, namely [Spec, AgrP]. We return to this point in sections 4.2 and 4.4.8 There are various indications that mai is indeed a complementizer. It occurs only in embedded contexts, in complement clauses to verbs, and these clauses must be + finite and -wh. The counterparts to mai found in negative clefts, nid, used in main clauses, and nad, in embedded clauses, are parallel to the ordinary clausal negators found in Literary Welsh, ni(d) in main clauses and na{d) in embedded clauses.9 These are uncontroversially complementizers, and the formal identity between the cleft and ordinary negators suggests that both are members of the same category. Mai is in complementary distribution with nad, and also with the interrogative cleft complementizer of Literary Welsh, ai (see section 4.3). It seems clear, then, that mai is another member of the set of complementizers. Returning to topicalization, note that standard GB analyses of English topicalization have similarly failed to predict the word order facts there, as recently outlined by Authier (1992). As is well known, English also has the topicalized constituent in postcomplementizer position, as in (17):

106

Maggie Tallerman

(17)

a. I knew that Irish, she couldn't speak, b. *Iknew Irish, that she couldn't speak.

This has led to suggestions in the literature that rather than having a topic in the C-projection, or external to it, instead we must treat embedded topicalizations as adjoined to IP in English (see, for example, Baltin 1982; Lasnik and Saito 1992). However, an alternative analysis has recently been proposed by several researchers: this is to allow CP to be an iterative projection, so that a complementizer that might also take another CP as its complement. The fronted constituent would then be sited in the specifier position of the lower complement CP. This proposal is supported by data from embedded verb-second constructions in English and other Germanic languages (see Rizzi and Roberts 1989; Cardinaletti and Roberts 1991; Authier 1992; McCloskey 1992; Rouveret, this volume). In the Germanic and other European languages discussed in the references cited, it appears that the CP-recursion phenomenon is limited to the complements of certain verb classes such as bridge verbs: see McCloskey 1992 for extensive discussion. The higher C°-position in English is distinguished from the lower C°head in that the former always contains the complementizer that. The CP which is the complement to that can contain either a topicalized constituent or a negative polarity item such as never, only, under no circumstances; the latter type of phrase always triggers the subject-auxiliary inversion which is subsumed under the heading of embedded verb-second. A typical example is shown in (18), with the proposed derivation in (19): (18)

She said that under no circumstances could she learn Irish.

The proposal I will defend in section 4 is that the cleft phenomenon in Welsh also constitutes an example of CP-recursion, appearing in both embedded and main clauses. On the other hand the abnormal sentences do not have this structure: they are formed via adjunction of a phrase to CP. Section 3 sets out the proposed structures of each sentence type. I turn first to the analysis of the cleft construction.

107

Fronting constructions (19)

V

could

she learn Irish

3 The derivation of the cleft and the abnormal constructions 3.1 The cleft sentence: proposed derivation The basic clause structure which I am assuming for Welsh is as in (20): (20)

CP Spec

Subj

I leave aside here the question of which functional projection, Agr or Tense,10 should be higher in an analysis which separates these inflectional morphemes, since for the present purposes it is sufficient to simply consider the position of the fronted constituents in relation to the IP. In (20) the verb raises to 1° in order to pick up (or check) its inflectional features; see McCloskey's (forthcoming) discussion of Irish for arguments against the further movement of 1° to C° in that

108

Maggie Tallerman

language, and comments disfavouring such an approach for VSO languages generally. I assume, following Koopman and Sportiche (1991), that the subject is basegenerated in the VP projection. A further question not considered here is whether the subject is Case-marked in situ in VSO languages, as has been proposed for Welsh by Rouveret (1991) and Tallerman (1993a), and by McCloskey (1991b, forthcoming) for Irish, or whether it raises to the Spec position of a higher functional head (see Rouveret, this volume; and Bobaljik and Carnie, this volume). Given the word order variations illustrated by Rouveret, movement of the subject under certain circumstances seems the most promising approach. The structure I propose for an embedded cleft sentence such as that in (21) is shown in (22). I concentrate on fronted subjects as it is in these cases that we see most clearly the contrasts between the cleft and the abnormal sentence, as discussed in section 1. (21)

Dywedais i mai'r dynion a werthodd y ci. said.lsG I coMP-the men coMPsold.3sG the dog 'I said that it was the men who sold the dog.'

(22)

Here, the cleft constituent y dynion 'the men' is moved by w/z-movement from its underlying position in the specifier of VP to the cleft position in the specifier of the lower CP, possibly via some higher Spec position. Alternatively, we might consider a cleft constituent to be adjoined to CP, so that w/j-movement would move an empty Operator from the subject position into [Spec, CP]. However, as we will see in section 4.1, there is in fact evidence against this latter analysis of the cleft construction.

Fronting constructions

109

Either version of the CP-recursion analysis would overcome the problem raised earlier in connection with traditional generative analyses of clefts, namely that the position of the embedded cleft could not be in [Spec, CP] if the complementizer mai follows the fronted constituent. Recall that this is taken by Rouveret (1991) as evidence that the clefted XP must be in a specifier position internal to the clause. The problem concerning word order outlined in section 2.2 is overcome by the analysis in (22). We need to ask, though, whether there are further good reasons to adopt the CP-recursion analysis rather than the account proposed by Rouveret (1991). This is taken up in section 4. I will continue to illustrate largely with embedded cleft constructions, since the overt clefting complementizer mai in embedded clauses aids the reader's memory: only cleft and not abnormal sentences have such a complementizer. As we will see, the presence of the complementizer is not fortuitous; instead, its presence allows us to predict the full range of divergent behaviours in the two types of fronting. Matrix cleft constructions have no overt complementizer in the modern language, at least in declarative affirmative clauses. The structure I propose for these is the same as that in (22), but where the higher C°-position is normally empty. I return to these constructions in section 4.3. Two questions arise in connection with the derivation in (22). Firstly, what ensures that the recursion of CP is not unlimited? As Cardinaletti and Roberts (1991: 6) propose in connection with CP-recursion in Germanic, it is clear that the two expositions have different properties. In Welsh only the higher C° is filled by mai (or its counterparts in interrogative (section 4.3) and negative clauses (section 4.4)) and only mai and no other complementizer selects CP as a lexical property. The lower C°, like other complementizers, selects IP-complements. The second question concerns the possibility of moving a constituent to the higher specifier position via w/z-movement, that is, [Spec, CPi]. In principle we might expect that either a cleft XP might move to both positions, giving double clefting, or that one XP might move to the higher specifier position only. Both possibilities give a completely ungrammatical result: (23)

a. *Dywedais said.lsG b. *Dywedais said.lsG

i y ci mai'r dynion a werthodd. I the dog coMP-the men COMP sold.3sG i'r dynion mai a werthodd y ci. I-themen COMP COMP sold.3sG the dog

In (23a), movement would clearly lead to an ECP violation: under Relativized Minimality the cleft XP in the higher [Spec, CP] would be unable to antecedentgovern its trace; see Cardinaletti and Roberts (1991: 39, fn. 3). We therefore account for the ungrammaticality of (23a) via a general principle. Ian Roberts (p.c.) points out that (23b) is parallel to */ said under no circumstances that would he do that, and suggests that the ungrammaticality is due to the fact that neither that nor mai can license anything in their specifier position.

110

Maggie Tallerman

Before looking at the predictions made by the CP-recursion analysis of clefts, I will propose an analysis of the abnormal sentences; we will then be in a position to compare the predictions made concerning the two constructions. 3.2 The abnormal sentence: proposed derivation If we take the most salient feature of the abnormal sentence to be its restriction to root clauses, what analysis might account for this? Since any account of the abnormal construction should predict the possibility of multiple frontings, as shown in (7), it seems reasonable to assume that the fronted constituent does not have a 'dedicated' position such as some specifier, or a Topic node, but instead occupies an adjoined position. One hypothesis which seems to make exactly the right predictions concerns the inadmissibility of adjunction to argument categories. Following the proposal made by Chomsky (1986b) to this effect, McCloskey (1992) formulates what he calls the Adjunction Prohibition: (24) Adjunction Prohibition Adjunction to a phrase which is s-selected by a lexical head is ungrammatical. (McCloskey 1992: 11) The result of this general restriction is that no adjunction can occur to any argument of a lexical head. Since VP and IP do not constitute such arguments, being the complements of 1° and C° respectively, adjunction to these projections is predicted to be possible. A matrix CP is also a possible adjunction site, since of course it is not an argument of any head. Adjunction to an embedded CP, on the other hand, is predicted to be impossible, since embedded CPs are s-selected by lexical heads (V, A, etc.). Thus we have an account of the non-occurrence of the abnormal sentence in embedded clauses. Unlike the cleft construction, the abnormal sentence never contains a copula or complementizer such as mat cf. points (3h). This is good evidence that the abnormal sentence cannot be analysed via CP-recursion, since it does not contain a CP-selecting C°. Otherwise, even if such a complementizer were phonetically null, the abnormal sentence would be predicted to occur in embedded contexts.

Fronting constructions

111

We can now propose that the structure of the abnormal sentence is as in (25): (25)

CP XP

CP

Recursive adjunction to CP predicts that several constituents can appear in the fronted position in an abnormal sentence, thus giving a straightforward account of one of the main differences between the cleft and abnormal constructions. A question that immediately arises is whether ^-movement takes place in the derivation of the abnormal construction as it does in the cleft sentence. In section 4.2 we will see that there is evidence that it does not.

4 Defending the analysis 4.1 Adjunction to CP If the abnormal type of fronting cannot occur in embedded clauses, why is it licit for clefts to occur in an embedded context? The answer to this comes from the CP-recursion analysis of clefts proposed in section 3.1. In (22) the cleft constituent is not in an adjoined position, but in [Spec, CP], a position to which it has been moved by w/z-movement. Adjunction in the embedded clause in (21) is, however, perfectly possible: adverbials can freely adjoin to CP 2 in (22), since this is an embedded clause which is not selected by a lexical head but by a functional head, the complementizer mai. Thus the grammatical examples in (26) are not ruled out by the Adjunction Prohibition in (24): (26)

a. Mae hi'n ymddangos [CPi mai [CP2 fel arfer is she-PROG appear COMP as usual [CP2 yr is-ganghellor sy'n llongyfarch y myfyrwyr the vice-chancellor REL.is-PROG congratulate the students dosbarth cyntaf.]]] class first 'It appears that as usual it's the vice-chancellor who is congratulating the first-class students.'

112

Maggie Tallerman b. Mae'n debyg [CPi mai [CP2 oni cheir deddf eiddo is-PRED likely COMP unless obtain.PASS act property newydd [CP2Meibion Glyndwr a wneith achosi problemau.]]] new COMP make.FUT cause problems 'It's likely that unless a new property act is obtained, it's Meibion Glyndwr that will cause problems.'

Note, however, that any strings of constituents in a fronted position in an embedded clause must be able to be interpreted as follows: the final constituent is a cleft XP, and any preceding constituents are adjoined to CP 2 , the complement of mai. This is the case in (26). Compare this situation with that in (8b), repeated here as (27), where more than one constituent is moved by w/2-movement into the cleft position: (27)

*Dywedodd mai [y rhai eraill] [adael y ddinas] a wnaeth. said.3sG COMP the ones other leave the city COMP did.3sG 'He said that those others left the city.'

The result is ungrammatical. On the other hand the data in (26) are grammatical because they do have only one constituent in the cleft position, yr is-ganghellor 'the vice-chancellor' in (a), and Meibion Glyndwr in (b). The adverbials are not clefts here, but are adjoined to the lower CP, as shown. The analysis of the cleft construction presented in section 3.1 therefore makes the correct predictions about these data: since there is only one cleft position in (22), namely the specifier of CP 2 , this predicts that only one constituent can be clefted. CP 2 , selected by the clefting complementizer mai, can, however, have adverbials recursively adjoined to it. As we have seen, this does not violate the Adjunction Prohibition, and since the adverbials are not moving into the cleft position, [Spec, CP], in (26), such examples are fully grammatical. The difference between (26) and (27) is that in (27), we have two major constituents of the embedded clause both competing for the same [Spec, CP] landing site. Why is it not possible to interpret the first constituent as an adjoined XP, as in (26)? The answer must be that in (26) the adverbials are not moved into their adjoined position by w/j-movement but are base-generated11 in the position adjoined to CP; in (27), however, the two fronted constituents are part of the predicate-argument structure of the clause, and cannot be base-generated outside it. Such constituents from within the clause can only front via w/z-movement, and since there is only one [Spec, CP] landing-site for such constituents, it is entirely predictable that only one is allowed to be fronted. It is, however, possible to get one adverbial in the cleft position [Spec, CP], that is, moving to that position via ^-movement. Note that any strings of adverbials in an embedded clause must in fact be interpreted in the following way: since they are prevented from adjoining to a lexically selected CP, CPi, by the Adjunction

Fronting constructions

113

Prohibition, all adverbials except the final one are adjoined to CP 2 , a nons-selected CP, and the final one is interpreted as the cleft constituent. So in (28), the clefting complementizer mai, which creates a non-argument CP for the noncleft adverbial to adjoin to, is obligatory: (28)

a. Dywedodd o ddoe *(mai) [CP2 yfory [CP2 yn yr adran said.3sG he yesterday COMP tomorrow in the department y bydd y cyfarfod.]] COMP be.FUT.3sG the meeting 'He said yesterday*that it's in the department that the meeting will be tomorrow.' b. Dywedodd hi neithiwr *(mai) [CP2 eleni [CP2 mwy na said.3sG she last night COMP this year more than thebyg y bydd y plant yn gweithio'n galed.]] likely COMP be.FUT.3sG the children PROG work-PRED hard 'She said last night that this year, it's more than likely that the children will work hard.'

Without mai there would be no indication that the CP-recursion structure existed, and the examples would read as if the adverbials were simply adjoined (illicitly) to the s-selected CPi: in such cases the judgements of native speakers follow exactly the predictions made by the Adjunction Prohibition. We might wonder why it is that in sentences like (28) the strings of adverbials are not all adjoined to CP 2 , rather than the final one being interpreted as a cleft XP. A plausible answer is that mai selects a + wh- CP and so requires that some phrases undergo w/^-movement into the Specifier of its complement. This forces a cleft interpretation for the final adverbial. Where there is only one adverbial, such as in (29), it is possible to omit the clefting complementizer mai, and still interpret the adverbial as being a cleft constituent; compare (10). (29)

Mae hi'n siwr (mai) heddiw y bydd y plant is she-PRED sure COMP today COMP be.FUT.3sG the children yn swnllyd. PRED noisy

'She's sure that it's today that the children will be noisy.' None the less the structure of (29) must still employ CP-recursion, even if mai is not overtly present in the higher C° position, since there would otherwise be no explanation for the badness of the examples in (28) without the mai. It appears that one adverbial can be interpreted as a cleft XP, even without this being signalled by mai, but any additional adverbials must be identified, by the presence of mai, as being recursively adjoined to a non-s-selected CP.

114

Maggie Tallerman

Note that we now also have an answer to the question raised in section 3.1 as to whether the cleft XP might be adjoined to the lower CP rather than in its specifier position. If the cleft constituent was indeed in an adjoined position, there would be no way to distinguish between (26) and (27), since both would simply involve a string of adjoined constituents. In the current account of (26), however, one XP is legitimately in [Spec, CP2] and any other constituents must be adjoined adverbials. Since (27) cannot have that structure, it is correctly predicted to be ungrammatical. I therefore conclude that a clef ted XP is moved into [Spec, CP2] by w/z-movement, and that this is the only position targeted as a landing-site. The analysis given in this section of strings of adverbials in cleft constructions makes an obvious prediction, to which I now turn. If all adverbials (except clefted adverbials) are in a position adjoined to CP 2 , and the cleft constituent is in [Spec, CP2], then the adverbials ought to precede the cleft XP consistently. Such is indeed the case, in both matrix (30a) and embedded cleft clauses (30b): (30)

a. ??Y myfyrwyr ddoe oedd yr is-ganghellor yn eu cyfarch. the students yesterday was.3sG the vice-chancellor PROG 3PL greet '??It was the students yesterday that the vice-chancellor was greeting.' b. ??Dywedodd hi mai'r plant yn ffodus y said.3sG she coMP-the children PRED fortunate COMP bydd yn gweithio'n galed. be.FUT.3sG PROG work-PRED hard '??She said that it was the children fortunately that will be working hard.'

Finally in this section I return briefly to the analysis of clefted XPs in Rouveret (1991). Recall that Rouveret's observations concerning the position of the complementizer mai with respect to a cleft NP lead him to conclude (1991: 372) that the cleft NP cannot be in [Spec, CP]. Rouveret proposes that the IP-internal position [Spec, AgrP] is the landing-site for [ + N] categories only, whilst other cleft phrases are moved to [Spec, CP] as in the present account. The motivation for this distinction is the appearance of a cleft PP before the complementizer y, as in (31): (31)

[ppYn yr ardd] y mae Hefin. in the garden COMP is Hefin I t ' s in the garden that Hefin is.' (Rouveret 1991: 374, fn. 37)

Certainly, the presence of the PP before the complementizer seems to be good evidence that the cleft phrase is indeed in [Spec, CP], as I have also proposed here. However, Rouveret's 1991 account cannot handle the fact that a cleft PP in an embedded clause also follows the clefting complementizer mai, just as cleft APs and NPs do:

Fronting constructions (32)

115

Mi wn i m a i [ P P y n y r ardd] y mae Hefin. PRTknow.lsG I COMP in the garden COMP is Hefin 'I know that it's in the garden that Hefin is.'

In an analysis which does not employ CP-recursion there is no readily apparent way of accounting for the data in (32): Rouveret himself recognizes both mai and y as complementizers, in his 1991 account, but as he proposes only one C° position, there is no position available for mai. Embedded VP clefts are analogous to clefted PPs, with mai preceding the clefted constituent, and a complementizer, a, following it. We can therefore conclude that the CP-recursion analysis adopted both in the present account and by Rouveret in this volume makes the correct predictions for all types of clefted XP. In the next section I turn to agreement patterns in the two constructions under discussion.

4.2 Agreement patterns in the cleft and abnormal sentences Recall from the data in (1) and (2) that the two constructions have a different agreement pattern when subjects are fronted: the cleft type displays no subject agreement on the verb; on the other hand, the abnormal construction of Middle Welsh had full verbal agreement with the fronted subject:12 both number and person agreement. Since (2) is an artificial example just intended to illustrate the distinction between the two construction types, I repeat here as (33) example (4), and cite a further attested abnormal sentence for completeness: (33)

A'i ddisgyblion a ddaethant ato . . . and-3sG.M disciples COMP came.3PL to.3sG.M 'And his disciples came to him

(34)

Ni a

(Matthew viii, 25) yfwn.

we COMP drink. 1 PL

'We will drink.' (Chwedlau Odo,13 18, 19-20) Example (33) displays number agreement between a plural subject NP and the verb, and (34) displays person agreement: the fronted constituent is a first-person plural pronoun, and the verb has the first-person inflection. Note that in Modern Welsh there is no number agreement between a full lexical subject NP and the verb: the verb is singular even when the subject is plural: (35)

Gwerthodd y dynion y ci. sold.3sG the men the dog 'The men sold the dog.'

116

Maggie Tallerman

Such was also the case in Middle Welsh, at least in the majority of instances. For example: (36)

Yna y doeth y kennadau. then COMP came.3sG the messengers Then the messengers came.' (Pedeir Keinc y Mabinogi,14 79, 27)

Evans (1989: 179) does mention that the verb is 'occasionally' plural before a plural lexical NP, but this is clearly not the norm in Middle Welsh. We can therefore reject at once the possibility that the verb in the abnormal sentence might agree with the subject NP because that was the usual agreement pattern: it was not. In all periods of the language, however, a verb has agreed in full with a pronominal NP subject, as this Modern Welsh example shows: (37)

Gwerthasant /?ro/nhw'r ci. sold.3PL they-the dog They sold the dog.'

As (37) shows, the verb is plural when the subject is either an overt or an empty pronominal. Parallel facts hold of course for person agreement. We can now give a straightforward account of the verbal agreement in the abnormal sentences in (33) and (34): the subject of the clause is the null pronominal pro, and the verb agrees with that NP. The appearance of full verbal agreement is good evidence that no w/z-movement is involved in the derivation of a subject fronted under the abnormal construction, since variables, like the full lexical NPs for which they are place-holders, are not associated with full agreement in Welsh: (38)

[Pa ddynion]/ a werthodd t, y ci? which men COMP sold.3sG the dog 'Which men sold the dog?'

Here the verb is in the third-person singular form although the w/z-phrase is plural. Since full agreement occurs standardly in the abnormal constructions, they are unlikely to involve w/z-movement. We have proposed that the cleft construction, on the other hand, is formed by w/z-movement; see (22), for example. Since, as we saw in (12), relative clauses are exactly parallel to clefts, and display the same lack of agreement with a moved subject, it is hardly surprising to find a further w/z-movement construction having the same properties. Given that the empty NP in the canonical subject position is a w/z-trace, the present analysis correctly predicts the lack of agreement when subjects are clefted. In this respect the current account appears to make more accurate predictions than that of Rouveret (1991). Recall that in Rouverefs

Fronting constructions

117

account of cleft sentences, the cleft NP is not in [Spec, CP] but rather in an A'-position internal to IP ([Spec, AgrP] to be precise). There is no w/j-movement in the derivation of the clefts, and Rouveret suggests that the verb agrees with what he terms the 'silent pronoun' (1991: fn. 33) (and elsewhere the 'null pronominal' (1991: fn. 41)) in the subject A-position. This can only be pro. But this cannot be the right answer, since as we have seen in (37) there is full agreement with pro, yet there is no number agreement in cleft constructions. I take this to be evidence for the correctness of the analysis of clefts presented here, and continue to maintain that the cleft sentence is formed via ^-movement. This conclusion is exactly parallel to that drawn by Borsley and Stephens (1989: 424 ff.) concerning the parallel SVO 'topicalization' structures in Breton, where the agreement patterns are the same as in the Welsh cleft construction. I turn next to the status of the complementizer mai and the other particles which can, or could, introduce a cleft construction.

4.3 The copula or complementizer in the cleft construction As we have seen in section 1, the matrix cleft construction, unlike the embedded clefts, does not have a particle introducing the cleft constituent. None the less, as we saw in section 4.1, the same restrictions on the order of adverbials and clefted XP obtain in matrix clauses; it seems likely, then, that the matrix cleft construction also employs CP-recursion, and has the cleft constituent in [Spec, CP] as proposed in section 3.1. Additional evidence for this comes from the occurrence of an overt clause-introducing particle in certain circumstances. The first is the negative cleft construction, illustrated in (6a): this is investigated fully in section 4.4. Secondly, in interrogative cleft clauses in the literary language the cleft is introduced by the item ai:ls (39)

Ai

ceffyl a

werthodd hi?

COMP horse COMP sold.3sG

she

'Is it a horse that she sold?' Lewis and Pedersen (1989: 320) show that ai is etymologically the interrogative particle a plus a form of the copula. Cliticization and grammaticalization appear to have occurred: where once the particle a would have been in the C° position, and the copula in a lower position, they subsequently formed a morphologically simple item in the C° position.16 As pointed out in section 1, both mai and the dialectal variant taw are etymologically copulas; compare the related (and homophonous) mae, the thirdperson singular present tense of 'be'. In earlier periods of the language, both mai and taw were also able to be used in conjunction with a distinct complementizer particle, y, as the Middle Welsh example (40) shows:

118

Maggie Tallerman

(40)

Mi a gredwn . . . y taw ti oed Bown. me PRTbelieve.coND.lsGCOMPTAw you were Bown 'I would believe . . . that thou wert Bown.' {Ystory a Bown de Hamtwn,11 24. 1541-2)

In earlier periods of Middle Welsh the copulas in cleft constructions even agreed with the main verb in tense, but this agreeing copula was eventually replaced by the single form ys in main clauses (shown in (11)) and then lost from the root position altogether. I have outlined the situation in the current period and earlier periods in some detail in order to illustrate the development of what is clearly a complementizer in the modern language. I suggest that the copula was once a finite verb which moved productively to C°, where it sometimes fused morphologically with the complementizer in that position (as in the case of ai). Over time, the system of copula tense agreement with the main verb broke down, and the copula became the complementizer: the modern equivalent of (40) is now ungrammatical. Grammaticalization of mai and the other complementizers legitimizes the CPrecursion which we proposed for the structure of the clefts: if mai were still a lexical verb, it would not permit adverbials to adjoin to its complement CP, since this would lead to a violation of the Adjunction Prohibition in (24). This lends support to the theory that mai has indeed undergone grammaticalization and (permanent) movement to C°. Since an overt complementizer such as mai signals the CP-recursion structure, it is unsurprising to find that it is usually retained in embedded clause clefts, whereas there is typically no overt complementizer in (Modern Welsh) matrix clause clefts. Adverbial adjunction to a matrix clause will not violate the Adjunction Prohibition, but of course adjunction to an embedded clause will do so unless it is adjunction to a CP which is not lexically selected, as in the case of CP complements to mai. An overt CP-selecting complementizer therefore has a clear role to play in embedded clauses which it does not in main clauses. None the less, we do of course assume that cleft constructions in Welsh always involve CP-recursion, even if the CP-selecting complementizer itself is phonetically null. We turn in the next section to the final set of facts which differentiate the cleft and abnormal constructions: the negatives.

4.4 Negation in fronting constructions In any discussion of negation in fronting constructions there are two main facts to be accounted for: firstly, why do cleft negations always have the fronted constituent itself negated, whilst abnormal sentences never do, relying solely on negation of the rest of the clause? Secondly, when cleft sentences display clausal

Fronting constructions

119

negation (as well as negation of the fronted XP) why do they have a different negative complementizer than the abnormal construction? I turn first to a brief outline of the data. Recall from section 1 that abnormal sentences, when negated, always involve clausal negation: the fronted constituent itself is not negated. On the other hand the cleft type of fronting always involves negation of the fronted XP, although, as we will see, there may be clausal negation as well. I repeat (6b,a), the examples which illustrated these facts, as (41) and (42):18 (41)

Y dyn ny

daeth.

(abnormal)

the man NEG-COMP came.3sG

T h e man did not come.' (Evans 1989: 180) (42)

Nid/dim y

dyn a

ddaeth.

(cleft)

NEG-COMP the man COMP came.3sG

'It wasn't the man who came.' In the analysis presented here, these negation facts are entirely predictable, since they hinge on the presence or absence of a complementizer position before the fronted constituent. The analysis of the abnormal construction in section 3.2 proposes adjunction of XP to matrix CPs. There is no complementizer position before the 'abnormally' fronted XP: hence no negative (or any other) complementizer can appear in initial position in examples like (41). In clefts, on the other hand, our analysis proposes a clefting complementizer in all cases, and the whfronting of XP to the specifier position of that complementizer's CP complement. It is therefore not surprising to find overt negative complementizers in initial position as shown in (42). Clearly, unlike in the case of affirmative or interrogative clefts, the negative complementizer needs to be overt in all clauses, since there will otherwise be no sign of negation. Hence the overt appearance of nid, dim in (42). In embedded clefts, we usually find the negative complementizer nad:19 (43)

Mi wn

i [nad

y

dyn a

ddaeth]

PRTknow.lsG I NEG-COMP the man COMP came.3sG

'I know that it was not the man who came.' In (42) and (43) only the cleft XP is negated, but it is also possible to have clausal negation as well: a negative complementizer can occur in the lower C° position in a structure such as (22), giving examples like (44) with the structure in (45):20 (44)

Nid

y

dyn na

ddaeth

(ddim).

NEG-COMP the man NEG-COMP came.3sG NEG

'It was not the man who didn't come.' (Watkins 1991: 332)

120

Maggie Tallerman

(45)

ddaeth

ddim

Example (45) is exactly parallel to the structure for affirmative clefts proposed in section 3.1. The analysis presented here captures straightforwardly an otherwise puzzling distinction between the abnormal and cleft clauses in the negative. Compare (41) and (44). In both cases we have a negated clause: . . . ny doeth (Modern Welsh ni ddaeth) in the abnormal sentence (41), but . . . na ddaeth in the cleft construction (44). Why should there be this distinction between the negative complementizers? The solution is straightforward: the contrast is between negation of matrix clauses versus negation of embedded clauses: ni is the main-clause negator and na the embedded clause negator. I illustrate with unmarked VSO word order.21 (46)

a. Ni welais i'r dyn. NEG-COMP saw.lsG I-the man 'I didn't see the man.' b. Mi wn i [na welaist ti'r dyn] PRTknow.lsG I NEG-COMP saw.2SG you-the man fc I know you didn't see the man.'

With these general facts about Welsh negation established, we can see what predictions are made about negative clauses by the analysis in section 3. Abnormal sentences are always root clauses, and involve adjunction to the matrix CP (see (25)). There is only one complementizer and it must be a matrix complementizer: hence the appearance of ni/ny in (41). Clefts, on the other hand, always involve CP-recursion, so the complementizer which negates the clause itself, C 2 in (45), is perforce the complementizer found in embedded clauses. The analysis presented here captures the fact that the clause which the cleft XP moves

Fronting constructions

121

out of is actually a type of embedded clause. Not surprisingly, it has the embedded negative complementizer na. Admittedly, a CP which is a complement not to a verb or other lexical head in a finite main clause, but merely to a higher complementizer, is not what first comes to mind when the term 'embedded clause' is used. But if we allow that term to include CP complements to C° - as the Welsh facts indicate we must - then a range of facts are elegantly accounted for. Note that again the same conclusions are not available in Rouveret's 1991 analysis of the cleft construction. In that account, the cleft XP is situated in a position within IP (the specifier position of the higher functional head projection, AgrP, in Rouveret's terms). If a negative clause follows it, as in (44), the presence of the embedded negative complementizer na is entirely unpredictable, since the cleft XP is not in a higher C-projection. I therefore conclude that both the agreement facts noted in section 4.2 and the negation facts of this section lead to a refutation of Rouveret's earlier (1991) hypothesis, in favour of the CP-recursion analysis presented here and by Rouveret in this volume. 4.5 Conclusion

We have seen throughout this section that an elegant account of the differing syntactic properties of the cleft and abnormal constructions can be provided, given the assumption of CP-recursion in the case of the cleft, and recursive adjunction to matrix CPs in the case of the abnormal sentence. Since there are two overt complementizers in cleft constructions, at least under certain circumstances, an analysis involving CP-recursion is an obvious one to adopt in order to account for the appearance of a cleft XP following, rather than preceding, the clefting complementizer. It is not clear how the adjunction to IP account of embedded topicalization would handle the same data, nor how any distinction could be made between the cleft and the abnormal constructions. Although the CP-recursion hypothesis was originally intended to account for the verb-second phenomena in Germanic languages, and other 'embedded matrix clause' phenomena (McCloskey 1992: 19) in a variety of languages, it can easily be extended to a wider variety of data, as we have seen here. The cases discussed by McCloskey involve bridge verbs allowing root phenomena to occur in an embedded context. Clefts in Welsh are not restricted to the complements of bridge verbs; CP-recursion has occurred instead as a result of the grammaticalization of the copula verb in cleft constructions, so that it is always sited in the C°-position, where it selects a CP complement.

122

Maggie Tallerman Notes

1 An earlier version of this chapter appears as Tallerman (1993b). Preliminary results were presented at the Conference on Comparative Celtic Syntax, University of Wales, Bangor, in June 1992, and at the second Plenary Conference of the European Science Foundation programme in Language Typology, EUROTYP, at Donostia/San Sebastian, 2-5 September 1992, in a paper entitled 'The "abnormal" and "cleft" word orders in Welsh'. Thanks are due to the European Science Foundation for the chance to pursue this research, and I am grateful to both the above audiences for useful suggestions. The present version has benefited from helpful discussion with the editors, Ian Roberts and Bob Borsley, and from the comments of an anonymous reviewer. In addition, especial thanks to Lewis Davies for his patient help as a consultant. Remaining errors are my own. 2 S. Williams, 1930. Cardiff: Gwasg Prifysgol Cymru. 3 B. Roberts, 1971. Dublin Institute for Advanced Studies. This example is cited in Fife and King (1991: 90). 4 J. G. Evans, 1907. Cardiff: Gwasg Prifysgol Cymru. This example is cited by Fife and King (1991: 84). 5 As Harlow (1981: 225) points out, there is no pied-piping in Welsh relative clauses, so only relativized NPs and not PPs have a parallel in the cleft construction. 6 The account of cleft sentences outlined by Rouveret (this volume) rectifies many of the problems of the earlier analysis in Rouveret (1991), and also includes a treatment of embedded clefts. 7 Both the examples in (16) will be grammatical if the adverbial is interpreted as belonging to the matrix clause, and the embedded clause complementizer y follows the adverbial: (i)

a. Dywedodd ef gydallaw, y bydd yn gadael. said.3sG he with hand COMP be.FUT.3sG PROG leave 'He said, by the way, that he will be leaving.' b. Mae hi'n siwr heddiw y bydd y plant yn swnllyd. is she-PRED sure COMP today be.FUT.3sG the children PRED noisy 'She's sure today that the children will be noisy.'

Unlike the constructions in Irish discussed by McCloskey (forthcoming), there is no evidence that examples such as (i) involve complementizer lowering in Welsh. Crucially, the adverbial cannot be interpreted as part of the embedded clause. It may, however, be contrastively clefted from the embedded clause, a construction we return to in section 4.1. 8 Rouveret's 1991 discussion concentrates on the analysis of the copula constructions shown in (i), which, he argues, involve a clefted NP in (a) and a fronted predicate NP in (b). I will leave aside such data here (but see Rouveret, this volume): (i)

a. Arthur yw'r brenin. Arthur is-the king 'Arthur is the king.'

Fronting constructions

123

b. Y brenin yw Arthur, the king is Arthur T h e king is Arthur.' (Rouveret 1991: 372) 9 Amongst the ordinary negative complementizers, ni, na are usually used preconsonantally, and nid, nad prevocalically. The cleft negative complementizers are nid, nad in all environments. Example (i) illustrates ordinary clausal negation, and (ii) negation in clefts, with a focalized PP: (i)

(ii)

a. Nid edrychaist amy llyfr. NEG looked.2SG for the book 'You didn't look for the book.' b. Gwn nad edrychaist amy llyfr. know. ISG NEG looked.2SG for the book 'I know that you didn't look for the book.' a. Nid am y llyfr yr edrychaist. NEG for the book COMP looked.2sG 'You didn't look for the book' b. Gwn nad am y llyfr yr edrychaist. know.lsG NEG for the book COMP looked.2sG 'I know that you didn't look for the book.'

10 For some discussion of these issues see Hendrick (1991), Rouveret (1991) and Tallerman (1993a). 11 'Base-generated' may perhaps be a shorthand for 'not moved by w/z-movement': see McCloskey (1992: 12ff.) for some discussion of the possibility that there are no basegenerated adjunctions. The only crucial feature of the present discussion is that the adjoined adverbials do not target [Spec, CP] as their landing-site. 12 See Evans (1989: 180): there are cases which are usually taken to be abnormal constructions but which have no agreement. A well-known example is shown in (i): (i)

Gwyr a aeth Gatraeth gan wawr. men COMP went.SG Catraeth by dawn 'The men went to Catraeth at dawn.' (Canu Aneirin 4.84. I. Williams, 1938. Cardiff: Gwasg Prifysgol Cymru)

Ian Roberts (p.c.) suggests that such examples might be unintroduced clefts; this seems unlikely, though, as the pragmatic effects (i.e. focus) associated with clefts are absent. I leave aside such data as (i) in what follows. 13 Example (34) is from I. Williams, 1958. Cardiff: Gwasg Prifysgol Cymru. 14 I. Williams, 1930. Cardiff: Gwasg Prifysgol Cymru. 15 The colloquial language simply has a non-overt complementizer, and intonation alone signals the interrogative: (i)

Ceffyl a werthodd hi? horse COMP sold.3sG she 'Is it a horse that she sold?'

124

Maggie Tallerman

16 Negative interrogative clefts in Literary Welsh are introduced by a complementizer onid, a variant of the set of negative complementizers discussed in the following section. 17 Morgan Watkin, 1958. Cardiff: Gwasg Prifysgol Cymru. 18 In (42), nid is more likely to be found in Literary Welsh, whilst dim is exclusively Colloquial Welsh. 19 Although in some dialects we find mai with a co-occurring negator, giving mai dim. 20 Literary Welsh typically has the na/d complementizer without the medial negator ddim, whereas Colloquial Welsh typically just retains the mutation properties of the complementizer, or has cliticized *d, and has an overt medial negator. 21 In Colloquial Welsh the negative complementizers are usually visible mainly by their mutation effects; the chief mark of negation is the medial negative ddim following the verb. For example: (i)

a. Welais i ddim o'r dyn. NEG.saw.lsG I NEG of-the man 'I didn't see the man.' b. Mi wn i [welaist ti ddim o'r dyn] PRT know.lsG I NEG.saw.2sG you NEG of-the man 'I know you didn't see the man.'

The same analysis obtains as for Literary Welsh: the superficial differences relate to which negators are phonetically overt and which are not.

4

Bod in the present tense and in other tenses

Alain Rouveret

1 Be and bod

When it is followed by an adjective or by an indefinite or bare noun phrase, English be confers a predicational interpretation on this term: the object designated by the grammatical subject is subsumed under the property or the concept denoted by the predicate.1 Definite noun phrases can also be used predicationally (non-referentially), as (lb) shows: (1)

a. Stanley is smart / a doctor. b. Horatio is Hamlet's best friend.

It is well known, however, that the postverbal element in ^-sentences cannot be uniformly characterized as a predicate. First, the realization site of the referential subject and of the term functioning as the predicate may be reversed when the latter is a definite noun phrase. Examples (2a) and (2b) illustrate what Higgins (1973) calls the specificational type: (2)

a. Hamlet's best friend is Horatio. b. What I don't like about John is his tie.

Second, identificational statements involving deixis such as (3) and equative sentences like (4) are not interpreted predicationally, but as identity sentences, asserting that two objects that are denotationally distinct within the structural description of the sentence correspond to the same entity in the real world: (3) (4)

That man over there is Alfredo Funoll. The Morning Star is the Evening Star.

In the philosophical tradition inherited from Frege, predication and identity are posited as irreducibly distinct notions; identificational/equative sentences are analysed as structures with two arguments. Recent syntactic research, however, has essentially confirmed the semantic insight expressed by Jespersen (1924: 153-4) that 'the linguistic "copula" does not mean or imply identity, but subsumption in the sense of the old Aristotelian logic.'2 Ruwet (1982) establishes that linear order is not the only characteristic distinguishing predicative and specificational sentences. The predicational predicate in (1) may be pronomina125

126

Alain Rouveret

lized, relativized, topicalized, while the second term in a specificational construction may not be. The former has a syntactic behaviour analogous to that of the direct complement of a transitive verb; the latter behaves in some respects like a preverbal subject. Longobardi (1983) reaches the conclusion that the Logical Form of copular sentences never contains two real thematic arguments, but always a non-referential expression functioning as a predicate with respect to a syntactic subject. He shows that specificational sentences should be analysed as inverse predicational sentences. Combining this proposal with the view that be is a lexically unaccusative verb selecting a small clause, hence external to the predicative connection itself, Moro (1988, 1991) suggests that the defining property of copular constructions is that be not only allows for the raising of the subject to SpecIP, as ordinary raising verbs do, but also for the fronting of the predicate, if it is nominal. In the first case, a canonical predicational construction is derived; in the second, an inverse predicational construction, that is, a specificational sentence.3 Note that, if we follow Jespersen's insight, we must assume that identity sentences are also semantically asymmetric and admit a parallel derivation.4 This chapter argues for a unitary characterization of the verb bod in Welsh. Just as there is only one be, there is only one bod, occurring in predicational, specificational, identiflcational and equative constructions. The diverging semantic interpretations of the relevant constructions can be shown to follow from the properties of the terms linked by bod and from the semantics of the predicates involved: they need not and should not be encoded in the meaning of bod itself. One feature, however, distinguishes my analysis of copular sentences from the standard one: extending to Welsh a proposal made by Travis (1992) for English be, I will assume that bod is not specified in the lexicon for the selection of a small clause, but is inserted into the lower V-position of a Larsonian VP-structure, a position from which it governs the predicate, not the subject. This decision will be motivated in section 2.3. The claim that there is only one verb bod at first may appear controversial. The reason is that bod has several forms in the third person of the present indicative. An accepted practice in Welsh linguistics is to distinguish between an existential meaning and a copular meaning of bod. In this view, the morphological diversification of the forms of bod marks a semantic distinction, corresponding to the existential/copula divide (see (5)/(6)). Closer examination, however, reveals that the picture is more complex. The existential form mae also appears in positive predicational sentences (see (7)). The so-called copula, yw, is the form used in negative and interrogative predicational sentences (see (8)), as well as in constructions in which an adjectival or indefinite nominal predicate occupies the clause-initial position (see (9)).

Bod (5)

(6)

(7)

(8)

(9)

127

Mae bleiddiau yn Rwsia. is wolves in Russia There are wolves in Russia.' Y brenin yw Arthur, the king is Arthur 'Arthur is the king.' Y mae yn ysgrifennwr rhagorol. PRT is PRED writer excellent 4 He is an excellent writer.' A ydyw Mair yn athrawes dda? Q is Mair PRED teacher good 'Is Mair a good teacher?' Ysgrifennwr rhagorol yw ef. writer excellent is he 'He is an excellent writer.'

I intend to argue that this morphological complexity, far from calling the unity of the verb bod into question, can be traced back to the interaction of three independent factors: the different lexical representations associated with stagelevel and individual-level predicates; the syntactic behaviour of bod, distinct from that of ordinary verbal predicates; how the temporal reference of sentences in the morphologically simple present tense is determined in Welsh. The following claims will be presented and justified: (10) (11) (12) (13)

The argument structure of stage-level predicates includes an abstract spatio-temporal argument, that of individual-level predicates does not. 5 Bod is the only verb in Welsh which can incorporate a locative clitic. Bod is the only verb in Welsh which can raise to C in overt syntax, both in root and embedded clauses. The morphological simple present tense in Welsh cannot function as a Tense operator at LF.

I intend to show that the form mae exclusively appears in sentences in the simple present tense containing a stage-level predicate and is required only if no operator able to bind the spatio-temporal argument of the predicate is made available by the context. This situation presents itself in affirmative predicational sentences. The form yw is legitimate when there is no spatio-temporal argument to bind, that is, with individual-level predicates, and when an extra operator is available to license the spatio-temporal argument of stage-level predicates, namely in interrogative and negative predicational sentences. No alternation between the forms of bod is observable in the past, simply because the morphological past tense has quantificational force and is able to license the spatio-temporal argument of stage-level predicates. Independent principles ensure that the sentences which

128

Alain Rouveret

contain a definite nominal (individual-level) predicate are necessarily realized as verb-second structures. If this analysis is correct, the following picture emerges: there is only one verb bod selecting a complement able to function as a predicate; the morphological alternation and the semantic divide between existential bod and copular bod have a strictly syntactic and logical basis.6

2 Predication and existence: the form mae 2.1 Predicational mae and existential mae The aim of this section is to characterize the basic syntax and semantics of mae. Two uses of this form must be distinguished. 2.1.1 Predicational mae Mae is the form used in ordinary predicational sentences which assert the existence of a relation between an object and a property or concept. (14)

(15)

(16)

(17)

a. Mae'r ci mawr yn yr ardd. is-the dog big in the garden T h e big dog is in the garden.' b. Mae ci mawr yn yr ardd. is dog big in the garden 'A big dog is in the garden.' a. Y mae'r ffermwr wedi cau y glwyd PRT is-the farmer PERF shut the gate T h e farmer has shut the gate.' b. Mae dyn yn siarad efo Mair. is man PROG speak to Mair 'A man is speaking to Mair.' a. Mae Sion yn ddedwydd. is Sion PRED happy 'Sion is happy.' b. Mae rhywun yn glaf. is someone PRED sick 'Someone is sick.' Y mae Sion yn feddyg. PRT is Sion PRED doctor 'Sion is a doctor.'

(PP)

(AspP)

(yn + k¥)

(yn + bare NP)

Although the identity of the lexical projection functioning as a predicate varies, these constructions share one characteristic: the relation between the nominal

Bod

129

argument and its predicate is mediated by an explicit marker. The predicate is a locative PP in (14), a verb-noun phrase ( = VNP) preceded by an aspect marker in (15), an adjective phrase or a bare noun phrase preceded by the predicative marker yn in (16) and (17). A second property shared by all raae-constructions is that the verb may be preceded by the affirmative particle y (see (15a), (17)).7 I will assume that, like other sentence particles, the affirmative particle is realized in C and is not an affixal element requiring a morphological support. This characterization, if correct, implies that mae is not itself realized in C, but in the higher inflectional head, Agrs. Finally, the nominal argument can be either definite or indefinite (see (14a/b)). This property crucially distinguishes predicational mae from existential mae.

2.1.2 Existential mae The following sentences in which the nominal argument is indefinite illustrate the existential use of mae: (18)

(19)

a. Mae ci mawr yn yr ardd. is dog big in the garden 'There is a big dog in the garden.' b. Mae yna gi mawr yn yr ardd. is there dog big in the garden 'There is a big dog in the garden.' a. Mae bleiddiau yn Rwsia. is wolves in Russia 'There are wolves in Russia.' b. Mae yna fleiddiau yn Rwsia. is there wolves in Russia 'There are wolves in Russia.' c. Mae bleiddiau. is wolves 'There are wolves.' d. Mae yna fleiddiau. is there wolves 'There are wolves.'

These examples are not semantically homogeneous. Example (18a) can be interpreted in one of three ways. We know that it can have a simple predicational reading (see (14b)). In this case, the indefinite argument receives a specific interpretation. It can also specify the value of a variable, the range of which has been previously defined: the sentence asserts that a big dog belongs to the class of x's

130

Alain Rouveret

which are in the garden. In this case, the interpretation is specificational. Lastly, it can signal that there is a big dog in the garden and denote the pure emergence of a phenomenon. In this case, the interpretation is presentational. The most natural translation of these presentational/specificational sentences makes use of the locative-existential expression there is in English or // y a in French. In Welsh, the deictic locative adverbial yna can also be added to the constructions containing a locative PP. If it is, it forces the presentational/specificational reading, provided that the nominal argument is indefinite (see (18b)). Examples (19) differ from (18) in that the nominal argument is a bare plural NP. As Jones and Thomas (1977: 51-2) observe, (19a) and (19b) assert 'the existence of a particular object as opposed to its location, even though its existence occurs at a particular place and in a particular time.' This interpretation of the data is confirmed by the fact that these existential sentences are also well formed in the absence of a locative PP, as (19c) and (19d) show. These examples illustrate the absolute use of bod: they are ontological assertions, stating the existence of entities, usually designated under the name of wolves. The presence of yna, although not obligatory, greatly improves the acceptability of these sentences: (19a) and (19c) are felt as slightly marginal. They are more natural if, in addition to yna, a locative PP is present, anchoring them more firmly: (19b) is preferred to (19d). All these sentences are instances of the existential use of mae. This value is not in doubt when bod is used absolutely. But it is also present when the predicate is a locative PP. The close connection between existential and locative sentences has often been emphasized: all existential sentences are, at least implicitly, locative sentences, provided that the term 'locative' is given a broad meaning encompassing both locative and temporal reference.8 Note that the verb bod can be assumed to have its full existential value also in the predicational construction which, as observed in section 2.1.1, can be analysed as a locative construction. The occurrence of existential mae is not restricted to sentences containing a locative PP. It can also be combined with an aspectual or adjectival predicate: (20)

a. Mae yna ddyn yn siarad efo is there man PROG speaking to 'There is a man speaking to Mair.' b. Mae yna rywun yn glaf yn is there someone PRED ill in 'There is someone ill in the house. c. *Mae yna rywun yn glaf.

Mair. Mair y ty the house

In these examples, mae cannot stand alone, but must be reinforced by yna: the deictic locative adverbial makes available the presentational interpretation which is otherwise excluded. In sentences containing an adjectival predicate, both yna

Bod

131

and a locative PP are necessary to give rise to a well-formed presentational sentence. Example (20c) is not acceptable. Paradigms (14)—(17), (18)—(19) and (20) show that the same form of bod is used in existential sentences and in canonical predicational sentences, a fact which suggests that these two constructions share at least one property. We know that they differ in that the nominal argument is necessarily indefinite in the former, while it may be either definite or indefinite in the latter. I intend to show that their common characteristic lies in the status of the predicate: it is semantically construed and syntactically marked as a stage-level predicate.

2.2 Stage-level predicates and the mat-construction As mentioned above, the relation between the nominal argument and its predicate in the ra^e-construction is mediated by an explicit marker which, in some cases, may be identified as a locative or temporal preposition. Let us try to be more precise about the syntactic status and the semantic import of these markers. Following a recent practice in Celtic linguistics, the various particles appearing in the aspectual periphrases (see (5), (7), (15)) have been analysed as instances of the functional category Aspect and the Asp + VNP sequences as AspP projections. This labelling is relatively neutral with respect to the different treatments of the construction proposed in the literature. Two types of approaches may be distinguished.9 Some authors view the aspectual periphrases as special predicative complexes in which the verb-noun is a non-finite verb, the verb bod an auxiliary and the particle an aspect marker. This line of thought is illustrated by Jones and Thomas (1977), Sproat (1985), Hendrick (1991), Borsley (1993). Another approach, starting with Anwyl (1899)and developed by Awbery (1976) and Fife (1990), assigns to the aspectual periphrasis the status of a normal predicative construction, in which bod is the verb of existence and the aspect marker a genuine preposition functioning as an aspect marker. This characterization relies on the observation that these markers are homophonous with locative or temporal prepositions: progressive yn with the preposition meaning 'in', perfective wedi with the preposition meaning 'after', ar with the preposition meaning 'on'. 10 The relevant elements clearly fulfil a specific function when they occur in the aspectual construction. The claim made by this analysis is simply that they do preserve a number of prepositional properties in their aspectual use. The status assigned to the verb-noun also differs in the two approaches: in the first one, it can only be identified as a non-finite verb; in the second one, a very different characterization is available: the VN, which can function as the object of a preposition-like element, is a nominal form of the verb. This analysis is in accordance with the morphological make-up of verb-nouns: most of them are derived through the adjunction of a nominalizing affix to the verb stem.

132

Alain Rouveret

Syntactically, VNP domains, which have the internal syntax and sometimes the external distribution of noun phrases, can plausibly be analysed as DPs having a null D head, more precisely as nominalizations of a particular kind: they are nominalized verbal structures embedded in a DP. 11 The functional structure associated with verbo-nominal DPs is the following: (21)

fop

D [NomP . . . Norn [Vp . . . V . . . ]]]

This analysis makes Celtic VNPs akin to Romance participial structures, which Kayne (1993) identifies as nominalizations contained in a DP. In this view, the Asp + VNP sequences are prepositional predicates of spatial and temporal location. Although much more justification needs to be presented in favour of this treatment and the detailed arguments produced by Sproat (1985) against the prepositional analysis of the aspectual markers and by Borsley (1993) against the view that VNPs have the internal syntax of NPs must still be refuted, I will assume that aspectual periphrases are normal predicative constructions, in which bod is the verb of existence, not an auxiliary.12 If this analysis is on the right track, the mae . . . yn . . . 'progressive' construction in Welsh can be viewed as structurally analogous to the following English sentences, cited by Smith (1991: 230): (22)

a. I am in the midst of writing a report, b. The murderer is in the act of escaping.

The two constructions are also semantically analogous. In order to determine whether the semantic interpretation of the ra#e-construction can be related to its syntactic make-up, it is necessary to digress a bit and clarify the distinction between stage-level and individual-level predicates. Carlson (1977) shows that in English, two classes of predicates may be distinguished on both semantic and distributional grounds. A stage-level predicate has a transitory value and denotes some accidental temporally bound manifestation of the subject's being (e.g. being angry, laughing); an individuallevel predicate refers to a permanent characteristic which is an integral part of the subject's being or identity (e.g. being intelligent, knowing Welsh). It is thus natural to ask whether the forms mae and yw can be distinguished on the basis of the type of predicate they select. Dechaine (1993) observes that the stage-level/individual-level distinction cuts across lexical categories. Some categories are prototypically stage-level, others prototypically individual-level. Locative prepositional phrases are instances of the former type, (bare) nominals are instances of the latter. As for adjectives and verbs, they seem to divide equally into the two types. In the m#e-construction, locative PPs and bare NPs can function as predicates (see (14), (17)). Both stagelevel and individual-level adjective phrases are legitimate:

Bod (23)

133

a. Mae'r ffenestr yn agored. is-the window PRED open T h e window is open.' b. Mae'r mor yn las. is-the sea PRED blue T h e sea is blue.'

In English, stative verbs, which clearly belong to the individual-level type, cannot appear in the progressive construction. Examples (24) show that the corresponding periphrastic construction in Welsh is not restricted in this way: stative verbs can occur in the ZW-construction with the aspect marker yn in the present tense and in other tenses:13 (24)

a. Y mae yn pwyso pedwar pwys a hanner. PRT is PROG weigh four pound and half 'He/she/it weighs four pounds and a half.' b. Mae Mair yn adnabod Sion. is Mair PROG know Sion 'Mair knows Sion.' c. Yr oedd y wraig honno yn gwybod Saesneg a Chymraeg. PRT was the woman that PROG know English and Welsh This woman knew English and Welsh.'

These distributions could be taken to show that the stage-level/individual-level distinction plays no role in the choice of the form mae. However, this analysis would miss an important generalization: the relation between the subject and its adjectival, nominal or verbo-nominal predicate is systematically mediated by an explicit marker. I propose to view predicative yn, aspectual yn and wedi as elements exclusively introducing stage-level predicates. More precisely, the combination of the marker with the predicate gives rise to a stage-level predicate, exactly as the combination of a locative preposition with its object does. Indefinite nominals denoting individual-level properties can also be turned into stage-level predicates, as the well-formedness of (17) shows. Kratzer (1989) proposes to derive the difference between stage-level predicates and individual-level predicates from a difference in argument structure. Stagelevel predicates have an abstract 'Davidsonian' spatio-temporal external argument, whereas individual-level predicates lack it. Being null, this argument must be bound at LF by an element functioning as an operator. This idea can be implemented in one of two ways. Either the spatio-temporal argument is taken to be a characteristic of the argument structure of the lexical predicates AP, NP, VNP, PP themselves, or it is represented as a property of the element which introduces them. The Welsh data clearly favour the second option. I will take the argument structure of the locative preposition yn, which is typically stage-level,

134

Alain Rouveret

and that of aspectual yn and predicative yn, which are prepositional-like elements, to be endowed with a spatio-temporal argument.14 As stated above, this argument must be bound at LF by an element functioning as an operator. My proposal is to analyse mae as 'augmented' bod, that is, as the third person of bod in the present tense incorporating a locative adverbial clitic, which is the equivalent of Italian ci and of French y: (25)

mae

= [CL L O C -/W.PRES.3SG]

At LF, the incorporated adverbial clitic functions as an operator binding the spatio-temporal argument of the stage-level predicate.15

2.3 A preliminary analysis of mai-constructions The characterization of mae as 'augmented' bod immediately raises the following questions: what is the derivational history of this form? In what type of configuration is it inserted? What is the structural origin of the locative clitic it incorporates? The answer to these questions largely depends on the position one adopts concerning the interaction between morphology and syntax. Following Baker (1988), let us assume that morphologically complex words are built up in the syntactic component by the variant of the rule move-a applying to heads and manipulating basic elements such as roots, stems and affixes. In this view, the initial representation associated with raae-constructions must include a specific category harbouring the clitic. In order to implement this idea, I propose to abandon the small-clause analysis of be/bo ^-constructions and to extend to Welsh bod the Larsonian VP-structure which Travis (1992) associates with English be. The D-structure representation generated by this hypothesis is shown in (26): (26)

[VPl [ v CL Loc ] [Vp2 [spec NP ] [V' [v bod\ [XP]]]]

Bod is inserted into the head of the lower YP-shell, VP2; the locative clitic occupies the higher V position;16 the theme argument is generated as the external argument of VP 2 ; the predicate XP is the syntactic complement of bod. On its way to T and Agrs, bod moves through the higher V position and adjoins to CL Loc . Later incorporation of bod-CL to T and to Agrs gives rise to the form mae. In section 2.4, I show that the definiteness effect observable in existential constructions lends further support to analysis (26). 2.3.1 Focalization of aspectual predicates The properties of cleft-constructions in which the focused element is an aspectual predicate provide additional evidence in favour of the idea that bod is inserted into a Larsonian VP-structure and also in favour of the analysis of aspectual

Bod

135

predicates as prepositional phrases. Paradigm (27) shows that in Welsh, selected PPs and adjunct PPs have the same behaviour with respect to extraction. Both can be clefted.17 The fronted constituent is followed by the subordinating particle y and binds a trace in its original position.18 And it bears the contrastive/identifying interpretation associated with focused elements. (27)

a. I'r bachgen y rhoddodd y dyn lyfr. to-the boy COMP gave the man book 'It is to the boy that the man gave a book.' b. Yn y nos y mae'r dyn yn gweithio. in the night COMP is-the man PROG work T h e man works in the night.'

Interestingly, the aspectual predicates appearing in the predicative construction show a similar behaviour with respect to extraction. They can be clefted (as in (28a)) or extracted from a finite complement clause (as in ((28b)), but they cannot be extracted from an indirect question (see (28c)): (28)

a. Wedi cau y glwyd y mae'r ffermwr.19 PERF shut the gate COMP is-the farmer T h e farmer has shut the gate.' b. ?Wedi cau y glwyd y credwn y mae'r ffermwr PERF shut the gate COMP we-believe COMP is-the farmer c. *Wedi cau y glwyd tybed a yw'r ffermwr. PERF shut the gate (I) wonder whether is-the farmer

If AspPs are PPs, the grammaticality of (28a) follows from the same principles that account for the well-formedness of sentences (27). The marginal status of (28b) mirrors the marginal status of long-distance extraction of ordinary PPs. The ungrammaticality of (28c) can be traced back to the w/i-island condition. The fact that this process leaves the subject behind is also expected in the analysis schematized in (26), since the nominal argument is basically inserted into a position external to the Asp/PP projection.20 Suppose instead that the subject originates in the AspP complement of bod, more precisely in the specifier position of the VNP domain, complement of Asp. This alternative option is schematized in (29): (29)

[yp [y [v bod ] [ ASP P Spec Asp [ V N P N P S [VN' VN ( N P O ) ]]]]]

If (29) is adopted, the fact that predicate-clefting leaves the subject behind could be taken to indicate that, at the point in the derivation where this process occurs, the subject has been moved out of VNP and AspP. Schema (29) must be rejected, however, since the corresponding derivation contains a problematic step. It crucially supposes that no violation occurs when the subject argument of a verb-noun is extracted out of VNP across the Asp head. If Asp is a functional

136

Alain Rouveret

head, the VNP domain should be opaque, since Asp is not 'lexicalized' (the verbnoun does not move into it). If, as I propose, Asp is a preposition, extraction of the subject should trigger the realization of a resumptive pronoun or clitic in the position it governs, as is the case in Welsh whenever the moved element is the object of a preposition.21 The difficulty disappears if (26) is adopted. The subject should be able to move freely since it is head-governed by bod after the adjunction of the verb to [v CL]. 22

2.4 The distribution of definite and indefinite subjects The existential construction and the predicational construction differ in that the nominal argument is necessarily indefinite in the former, while it may be definite or indefinite in the latter. The aim of this section is to introduce additional data, which provide evidence in favour of the following generalizations: (30)

a. In the mtfe-construction, the subject may be realized in SpecTP or in a position internal to VP. b. Only indefinite subjects may be realized inside VP. c. Indefinite subjects inside VP yield the existential reading.

2.4.1 Mai-constructions In the mae-construction, the subject argument need not be adjacent to the inflected verb. It can follow a locative PP. But this option is available only with indefinite expressions. If the subject is a definite noun phrase, it obligatorily precedes the locative complement. (31)

a. Mae ci mawrynyr ardd. is dog big in the garden 'A big dog is in the garden.' b. Mae yn yr ardd gi mawr. is in the garden dog big There is a big dog in the garden.' c. Mae'r ci mawr yn yr ardd. is-the dog big in the garden. T h e big dog is in the garden.' d. *Mae yn yr ardd y ci mawr. is in the garden the dog big

As the translations of the above examples show, the position of the indefinite subject makes a difference with respect to the availability of the existential reading. Example (31a), in which the subject precedes the locative PP, only has the predicational reading; (31b), where the subject follows the locative, only has the

Bod

137

existential reading; (31c), which contains a definite noun phrase, can only be interpreted predicationally. These observations suggest that the S-structure position of the subject correlates with the existential/non-existential reading of the sentence. In (31c), the definite subject precedes the locative predicate. It is plausible to assume that it is realized in the same site as the subject of verb-initial sentences containing an 'ordinary' verb, that is, SpecTP.23 The well-formedness of (31a) shows that indefinite subjects are not ruled out from this position. Given the D-structure (32), we expect the movement of NP,- first to SpecVPi, then to SpecTP, to be free, since no intervening potential landing-site is crossed in the process: (32)

[AgrsP Agrs [TP Spec T [VP, Spec [v CL ] [VP2 NP, [v bod ] [ PP ]k ]]]]

Example (31b), in which the predicate precedes the subject, can be viewed as an instance of Locative Preposing.24 The question which arises is: which position hosts the nominal argument at S-structure? Two options are available: the specifier of the lower VP or the specifier of the higher VP. If the first one is correct, the locative predicate is necessarily realized in SpecTP. If the second one is correct, it may occupy either SpecTP or the specifier of the higher VP. Whatever the correct analysis, the nominal subject is realized in a position internal to the higher VP-shell and the predicate necessarily crosses the specifier filled by the subject on its way to its surface position. We see that the derivation of Zwd-sentences with inverted predicates contains a problematic step - the predicate is moved across the position containing the subject - which should lead to a violation of the Minimality Condition. In fact, the problem disappears once we adopt the theory of derivations outlined in Chomsky (1993). Chomsky's basic claim is that the cyclic raising of a lexical head to the functional heads which dominate it allows a XP to raise over a filled specifier position if and only if it raises to the next higher specifier position. More precisely, the effect of head-raising is to create a minimal domain containing the extraction position, the filled intermediate position and the position targeted by movement. Inside this domain, the target of the movement and the intermediate position are equidistant from the original site of the moved constituent and count as non-distinct for the purposes of the Minimality Condition. Once a Larsonian VP-structure for the verb bod is adopted, this theory provides the necessary framework for a satisfactory analysis of the construction under consideration. Raising and adjunction of bod to [v CL] in (32) will render SpecVP2 and SpecVPi equidistant from the complement position, allowing the predicate to move to SpecVPi across the subject without violating Minimality. The predicate may stop there or move to SpecTP. The corresponding S-structure representation is thus either (33) or (34):

138 (33) (34)

Alain Rouveret [Agrsp Ugrs maey ] [TP Spec T [VPl PP* [v e ] v [VP2 NP,- [ V e ] v [PP e ]k]]]] UgrsP Ugrs maew ] [TP PP* T [Vp, [pp e ]*[ve]v[vp2 NP, [v e ] v [PP e ]k]]]]

We see that Chomsky's (1993) framework makes available an analysis of examples like (31b) which is in conformity with the Minimality Condition. 25 Note that under the small-clause analysis of /^//^-constructions, it would be very difficult to explain why the intervening small-clause subject does not block the antecedentgovernment relation between the moved predicate and its trace. In (35), there is no minimal domain that contains both SpecVP, the subject NP, and the locative predicate, since no head is moved from inside the small clause.26 (35)

[VP Spec [y [v bod ] [sc NP, PP*]]]

In brief, paradigm (31) provides initial evidence in favour of the generalizations stated in (30). It was shown that the definite or indefinite subject in predicational constructions is realized in SpecTP, while the indefinite subject in existential constructions occupies the specifier of the lowest VP. The ungrammaticality of (3Id) shows that definite subjects cannot remain inside VP, contrary to indefinite subjects.

2.4.2 Bod-initM constructions The same distributional asymmetry between definite and indefinite subjects can be observed in the embedded domains containing the uninflected form bod. When the predicational/existential construction is embedded under a declarative or epistemic predicate, the verb takes the form of the verb-noun bod and occupies the initial position of the clause. If the nominal argument is indefinite, it can be adjacent to the verb, as in (36a), or be separated from it by a locative PP, as in (36b). In the first case, the embedded sentence has a predicational reading; in the second case, it has an existential reading. When the subject is definite, it obligatorily follows the verb-noun bod (as in (36c)) and cannot be separated from it by a locative complement (as in (36d)) and the embedded sentence has a predicational interpretation. (36)

a. Dywedodd Mair fod ci mawrynyr ardd. said Mair be dog big in the garden 'Mair said that a big dog was in the garden.' b. Dywedodd Mair fod yn yr ardd gi mawr ffyrnig. said Mair be in the garden dog big ferocious 'Mair said that there was a big ferocious dog in the garden.' c. Dywedodd Mair fod y ci mawr yn yr ardd. said Mair be the big dog in the garden 'Mair said that the big dog was in the garden.'

Bod

139

d. *Dywedodd Mair fod yn yr ardd y ci mawr ffyrnig. This parallelism between finite and non-finite clauses would be surprising if no movement of the verb was involved in embedded domains. It can be shown, however, that frod-initial structures have the internal syntax of finite domains. Harlow (1992) provides strong arguments for the view that a movement of bod is involved in their derivation. As for the uninflected form bod itself, Awbery (1976) proposes to analyse it as a finite form, with its tense and aspect features deleted. I will assume that the internal structure of these clauses includes the same functional projections as embedded finite domains, that is, TP, AgrsP and CP. The predicate being stage-level, the presence of a locative clitic binding the spatiotemporal argument of the predicate is required. This means that bod is inserted into a structure with two VP shells, exactly as in raae-constructions. But in this case, the clitic realized in the higher V-position is silent and does not give rise to a specific form. This analysis assigns the following D-structure to /wd-initial constructions (0 denotes the silent locative clitic): (37)

. . . [CP C [AgrsP Agrs [TP Spec T [VP, Spec [v 0 ] [VP2 NP,- [ V bod\ [PP]]]]]]

In the course of the derivation, bod first adjoins to [v 0 ], then the complex head [v bod-®] moves to T and Agrs. In the predicational constructions (36a) and (36c), the subject NP, raises to SpecTP. In the existential construction (36b), the indefinite subject keeps its original position; the predicate is realized either in SpecVPi or in SpecTP. The raising of the predicate across the subject does not give rise to a violation of the Minimality Condition, because the positions involved in this process are included in the same minimal domain at the relevant level of derivation. 27 Paradigm (36) thus provides additional evidence in favour of the generalizations (30). Indefinite subjects have the option of appearing in SpecVP or in SpecTP, while definite subjects can only be realized in SpecTP. The S-structure position of the subject correlates with the existential/non-existential reading of the sentence.

2.4.3 Possessive absolute constructions Possessive absolute constructions, while lending additional support to the generalizations (30), show that they must be slightly modified. The general pattern for expressing temporary or permanent possession is the one given in (38). This periphrastic construction combines the existential verb bod with the preposition gan (or gyda) meaning 'with'.

140

Alain Rouveret

(38)

a. y mae N P g a n N P b. y mae g a n N P N P

The two options do not alternate freely. Construction (38a) is legitimate with both definite and indefinite possessees (see (39a/b)); (38b) is not available when the possessee is definite (see (39c/d)). In the paradigm (39), gan appears in its inflected form: its pronominal object has been incorporated into it. When gan is followed by a full noun phrase, the distributional pattern is basically the same definite subjects cannot follow the predicate - but (38a) is also preferred when the subject is indefinite (see (40a/b)). (39)

(40)

a. Y mae'rllyfr ganddo. PRT is-the book with-him 'The book belongs to him.' b. ?Y mae llyfr ganddo. PRT is book with-him 'He has a book.' c. Y mae ganddo lyfr. PRT is with-him book 'He has a book.' d. *Y mae ganddo'r llyfr. PRT is with-him-the book a. Y mae pibell gan eich tad. PRT is pipe with your father 'Your father has a pipe.' b. ??Y mae gan eich tad bibell.28 PRT is with your father pipe

As the proposed translations show, the two constructions are not semantically equivalent. The difference between them can be examined in the light of the distinction made by Benveniste (1960) between the notions of 'possession' (ownership or temporary possession) and 'appartenance' (permanent possession). A predicate of'appartenance' is used to define an object (Ce livre est a lui, This book belongs to him, Hie liber patris mei est), while with a predicate of possession, the object is unspecified {Ilaun livre, He has a book, Est patri meo liber). In Welsh, (38a) expresses 'possession' or 'appartenance', (38b) exclusively expresses 'possession'. We do not know whether the predicate, when it precedes the subject, is realized in SpecTP or in the specifier of the higher VP (see the discussion of (31b)). The syntax of ddim in dialects making use of the medial negation provides a clue as to its position. Assuming that ddim is inserted between T and VP, the predicate should follow ddim if realized in SpecVP, precede it if realized in SpecTP. The contrast between (41a) and (41b) indicates that only the inflected preposition has the option of moving to SpecTP and suggests that gan + NP has to remain inside

Bod

141

VP. Example (41c) confirms that definite subjects are fronted to SpecTP. (41)

a. 'Does ganddo ddim ceffyl. not-is with-him not horse 'He has no horse.' b. *'Does gan y ffermwrddim ceffyl. not-is with the farmer not horse c. Dydy'r ceffyl ddim gan y ffermwr. not-is the-horse not with the farmer 'The horse doesn't belong to the farmer.'

The asymmetry between (41a) and (41b) could indicate that only nominal projections can be fronted to SpecTP. The constituent headed by ganddo qualifies as a nominal projection since inflected prepositions are analysable as projections of the prepositional Agr category;29 ordinary prepositional phrases do not. 30 If this analysis is correct, paradigms (39)-(40) and (41) provide additional evidence in favour of the generalizations (30). The important fact is that similar distributions are observed in absolute possessive constructions, where no form of bod is present. The absolute construction usually contains a subject which, if pronominal, takes the form of an independent pronoun, and an aspectual, prepositional or adjectival predicate, occurring in this order. (42) (43)

a/ac + NP + XP a. A mi yn ofnus, ni ddywedais ddim. and I PRED shy NEG said nothing 'Since I am shy, I said nothing.' b. A hwy yn yr eglwys, ysbeiliwyd eu ty. and them in the church was-looted their house 'While they were in the church, their house was looted.'

The structures in (43) qualify as predicational constructions. When the predicate is locative-possessive, however, two options are available: either the nominal argument precedes the predicate, as in other absolute constructions, or the inflected prepositional phrase comes first. In the latter case, the reading is that of temporary possession and the definiteness effect emerges: the nominal argument is necessarily indefinite. (44)

a. A'r llyfr ganddo, . . . and-the book with-him, . . . 'The book belonging to him, . . .' b. A llyfr ganddo, . . .

142

Alain Rouveret c. A chanddo lyfr, . . . and with-him book 'Him having a book, . . .' d. *A chanddo'r llyfr, . . .

In order to evaluate the theoretical significance of these examples properly, it is necessary to clarify their structural analysis. First, absolute possessive constructions must be analysed as full clauses (otherwise there would be no available landing-site for the preposed predicate in (44c)). Second, the asymmetry between definite and indefinite subjects also manifests itself in contexts in which no form of bod is present. On the basis of considerations I will not fully develop here, I believe that an approach invoking a null verb bod or a dropped copula (present at D-structure, deleted or unrealized at S-structure) is not to be pursued. If this stand is correct, the internal structure of absolute constructions includes no VP-projection. To shorten the discussion, I will assume without further arguments that the category Tense is universally projected in all propositional domains and adopt Dechaine's (1993:296) claim that 'a morphological tense implies the presence of Tense as a syntactic position [but] the converse does not hold'. Following Gueron (1989) and Dechaine (1993), I will also assume that contrary to morphological tense which categorically selects V, to the exclusion of N, P and A, null Tense imposes no c-selectional restrictions.31 In this view, absolute constructions are propositional domains which include no VP projection. But they must contain a silent locative clitic able to bind the spatio-temporal argument of their stage-level predicate at LF. Since no VP is present in the structure, I will assume that this element originates in a specific category, say Z, determining its own projection, and that the nominal subject and the prepositional predicate are generated in the specifier and the complement positions of ZP. Null Tense, which imposes no c-selectional restrictions, can take ZP as its complement. The D-structure representation of absolute constructions can be schematized as follows: (45)

a [TP Spec [T 0 ] [ZP NP \r\z CL] [XP]]]]

Since there is no verb bod into which CL could incorporate, I will assume that it is adjoined to null Tense in the course of the derivation. In 'ordinary' absolute constructions, NP raises to SpecTP (see (43), (44a), (44b)). In predicate-initial absolute constructions (see (44c)), it is the locative-possessive PP which moves to SpecTP.32 The raising of the predicate across the subject is possible because the adjunction of CL to T renders SpecTP and SpecZP equidistant from XP. 33 The generalizations stated in (30), which crucially refer to the category VP, must be modified to cover paradigm (44). Let us refer to the domain governed by Tense in Zwd-sentences and in absolute constructions as the Lexical Predicative Domain (henceforth LPD). The LPD coincides with the higher VP in (32) and

Bod

143

(37), with ZP in (45). Generalizations (30) can be reformulated as follows: (46)

a. In predicative propositional clauses, the subject may be realized in SpecTP or in a position internal to LPD. b. Only indefinite subjects may be realized inside LPD. c. Indefinite subjects inside LPD yield the existential reading.

We are now in a position to provide an analysis for the distributional asymmetry between definite and indefinite subjects.

2.5 The definiteness effect One possibility explored by many researchers is that the definiteness effect in existential constructions arises as a result of the impossibility of Case-marking the postverbal nominal argument. A more promising approach is to try to derive the distributional asymmetry between definite and indefinite subjects from the general semantic principles which determine the interpretation of noun phrases. However, even in a semantically based account of the definiteness effect, it is necessary to clarify the way in which the postverbal argument is Case-licensed. 2.5.1 Existential closure In the semantic approach, the level of analysis at which an explanation for the definiteness effect may be found is not S-structure, but LF or the semantic representation of sentences. This can be done only within a framework which provides a way to relate the semantic properties of noun phrases to the logical and the syntactic representations of the sentences which contain them. Diesing (1992) has recently developed a theory which has the required property. Diesing adopts Heim's (1982) claim that indefinites introduce a variable that needs to be bound in order to be interpreted and argues that in sentences containing an indefinite, but no explicit quantifier, the variable is bound by an implicit existential quantifier that existentially closes off the 'nuclear scope'. And she shows that the domain of 'existential closure' for indefinite subjects in English and in German should be defined in sentential terms as the VP of the sentence. The Welsh data discussed in 2.4 basically confirm this view. In maeconstructions and Z?o
144

Alain Rouveret

interpretation when they are realized in SpecTP. Definite subjects, which only admit a presuppositional reading and resist an existential interpretation, are excluded from the domain of existential closure and cannot remain inside the LPD.

2.5.2 The Case of indefinites Let us now ask the important question of how the indefinite NP in existential constructions is Case-licensed. This question must be given a principled answer even if a semantic account of the definiteness effect is adopted. In a recent paper, Lasnik (1992) argues that in English, the expletive there and the nominal argument must be assigned Case independently. The expletive is Case-marked in SpecIP, like ordinary subjects; the nominal argument is Case-marked by the verb be itself under government. The Case involved is an inherent one, which Lasnik identifies as the partitive Case, postulated by Belletti (1988) to account for the properties of unaccusative"constructions in Italian. The relevant assumptions are summarized in (47): (47)

a. Be is endowed with a partitive Case feature. b. Inherent Case is assigned under government. c. A chain headed by a Case-assigner has the Case properties of its head.

The claim that existential be assigns partitive Case directly explains the definiteness effect observable in the existential construction. If partitive Case inherently requires non-definiteness and can only be assigned to indefinite expressions, the postverbal nominal argument is expected to be necessarily indefinite. At the same time, since the nominal argument may be preposed to SpecIP, in the absence of there, and assigned nominative in this position, it must be assumed that the assignment of partitive Case is optional. In fact, Lasnik proposes that Case assignment is generally optional. A stipulation of obligatoriness of Case assignment would be entirely redundant with the Case Filter. Lasnik's account is not without problems. First, as Travis (1992) observes, his claim that be assigns an inherent Case is not compatible with the analysis of be as a verb selecting a small clause, which he adopts. In the standard view, the assignment of an inherent Case is crucially linked to a thematic property of the verbal head and, as such, cannot take place into a small clause (the subject of a small clause is not a thematic dependant of the verb). Travis' solution is to abandon the small-clause analysis and to adopt a Larsonian VP-structure for the verb be in English. I showed that an analysis of Zwd-constructions along similar lines is also well motivated in Welsh.

Bod

145

Second, possessive absolute constructions make it clear that, in Welsh at least, it must be possible for indefinite subjects to be Case-licensed inside the LPD, even when no form of bod is present. This can be taken to indicate that in maesentences and in ^^-constructions, bod is not the (only) element responsible for the Case-marking of indefinite subjects internal to VP. If the analysis developed in the previous sections is correct, the element present in all existential constructions is the locative affix which is either incorporated into bod or directly adjoined to null Tense. We are led to assume that the complex heads [bod-CL]] and [CL-[T 0]] function as Case-assigners licensing an indefinite argument in the position they govern. This property can be interpreted in one of two ways: either the locative affix itself is endowed with a Case feature, or the incorporation of the locative affix to bod or to null Tense 'activates' a Case feature borne by these elements. The second option, which assigns to the locative affix a mediating function, analogous to the role played by Agrs and Agr o in the assignment of the nominative and accusative Cases supplied by finite Tense and V, appears to be more plausible than the first and I will adopt it. This analysis straightforwardly accounts for the Case-marking of indefinite subjects in existential constructions. The relevant structures are given in (48): (48)

a. [Agrs bod-CL ]y. . . [ T e ] v . . . [VP, • • • [v e ] v [VP2 NP,- [v e ] v • . . ]] b. [TP . . . [T CLj [T 0 ]] [ZP NP, [ z e ]j,. . . ]]

In (48a) and in (48b), NP, occupies the specifier position associated with the lowest member of the chain headed by the Case-assigner ([bod-CL] and [CL-T], respectively). It is also governed by the next higher member of this chain. Whether Case assignment exclusively involves a specifier-head relation or can also take place under government, it is safe to assume that NP/ is licensed in these structures. The fact that NP, can also be moved to SpecTP and be Case-marked in this position shows that the assignment of the 'existential' Case inside the LPD is optional. 34 Up to now, nothing has been said about the identity of the 'existential' Case. We know that it is compatible with indefinite nominal expressions and also that definite expressions cannot remain in the site where it is assigned. If this asymmetry is given a purely Case-theoretic account, the 'existential' Case must be identified with Belletti's partitive Case, as Lasnik proposes. Note however that if this Case is a property of the complex heads incorporating the locative clitic, as suggested above, it cannot be characterized as an inherent Case. Another analysis is available if the semantic account of the definiteness effect sketched in 2.5.1 is adopted. Definite noun phrases can also be marked for the 'existential' Case, if they remain inside the LPD. But the resulting configuration is excluded at LF or in the semantic representation of the sentence, as argued in 2.5.1. In this view, the 'existential' Case does not exclusively affect indefinites. It is not necessarily partitive Case, but could as well be accusative Case. This second option seems to

146

Alain Rouveret

me preferable. It is in accordance with our previous conclusion that the Case assigned to indefinites in existential constructions is not an inherent property of the verb bod, but a characteristic of the heads incorporating the locative clitic. An important result of this analysis is that mae can receive a unitary characterization, as the third person of bod in the present tense incorporating a locative adverbial clitic. In both existential sentences and predicational sentences, the clitic binds the spatio-temporal argument incorporated into the stage-level predicate. The D-structure configuration underlying the two sentential types is the same. Mae is not directly involved in the definiteness effect, only the implicit existential quantifier defining the domain of existential closure is. This suggests that the locative adverbial clitic and the existential quantifier should be considered as separate entities at LF.

3 Predication and identity: the form yw/ydyw 3.1 Copula yw

The examples cited in section 1 make it clear that the form yw/ydyw has two uses. It functions as a copula in sentences containing two definite nominals (see (6)).35 It also appears in negative and interrogative predicational structures (see (8)), as well as in constructions in which an adjectival or indefinite nominal predicate occupies the clause-initial position (see (9)). The analysis of ^-constructions encounters two questions. The first concerns the syntactic profile of sentences having a definite predicate: why is the copula not realized in sentence-initial position, contrary to what happens in ^^-constructions and in finite sentences containing an ordinary verb? The second question concerns the form yw itself: why is the same form appropriate in positive sentences having a definite predicate and in negative and interrogative existential constructions? These two questions are independent of each other. On the one hand, the constraint barring the verb bod from the sentence-initial position is also operative in tenses other than the present. On the other hand, the formal coincidence between the verb of identity and the verb of existence in negative and interrogative sentences manifests itself only in the present tense, since the diversification of the forms of bod only exists in this tense. Let us consider the copular use first. Grammatically, two constructions must be distinguished, which differ as to the mutual position of the two terms connected by yw: in (49a), the initial position is occupied by the definite predicate; in (49b), the subject comes first. In both cases, yw is the only option; existential mae is not a possible choice (see (49d)). The copula is never realized in sentence-initial position (see (49c)). Whatever its position, the predicate is never preceded by the particle yn.

Bod (49)

147

a. Y brenin yw Arthur. the king is Arthur 'Arthur is the king.' b. Arthur yw'r brenin. Arthur is-the king 'It is Arthur who is the king.' c. *Yw Arthur y brenin. d. *Y mae Arthur yn y brenin. PRT is Arthur PRED the king

Semantically, the copula expresses a relation of identity or equivalence between the two terms it connects. Under the ordinary interpretation of (49a) and (49b), the definite description y brenin picks out a role and the proper name Arthur identifies a particular individual.36 Not surprisingly, yw is found in sentences containing both a definite description and a term with no predicational aptitude, such as a proper name, a first- or a second-person pronoun, or a demonstrative. (50)

a.

Hwn yw'r athro. this-one is-the teacher 'The teacher is this man.' a'. Yr athro yw hwn. the teacher is this-one 'This man is the teacher.' b. A hyn yw'r perygl, fod y gwynt yn ffyrnig. and this is-the danger be the wind PRED violent 'And this is the danger, that the wind is violent.' b'. A'r perygl yw hyn, fod y gwynt yn ffyrnig. and-the danger is this be the wind PRED violent 'And the danger is this, that the wind is violent.' c. Myfi yw. I is 'It is I.'

As these examples make abundantly clear, copular sentences give rise to inversion effects. And it is often not a trivial task to determine which term is the subject and which one is the predicate. In one case at least, the answer is straightforward: in (50c), the initial term is necessarily the predicate since there are well-formed sentences with a null subject, but no well-formed predicational sentences with a null predicate.37 Prosodic considerations confirm that the normal order in the ^-construction is predicate-jw-subject. Jones (1993:14) observes that in this case, 'the intonational pattern generally falls over the clause, and no constituent is given special prominence'. When the subject comes first, it 'becomes more prominent -

148

Alain Rouveret

this prominence is due to greater stress and the use of a higher pitch level than is found on the equivalent constituent in normal order'. In (49b), Arthur bears a contrastive accent and the overall interpretation of the sentence is that of a cleft structure with a uniqueness presupposition associated with the focalized constituent (= it is Arthur who is the king, not Peredur).38 It is plausible to assume that in the normal order, the initial predicate, which bears no contrastive accent, also functions as the grammatical focus of the sentence.

3.2 Predicate-initial copular structures 3.2.1 Copular sentences are verb-second clauses

The presence of a definite predicate has the same effect in the present tense and in other tenses. Either the predicate or the subject must be fronted. The result is a structure in which the verb occupies the second position, as examples (49) and (50) show. Looking at predicate-initial structures first, it appears that the fronted predicate is realized in an A'-position, most probably in SpecCP, rather than in SpecAgrJP. Contrary to what happens in English and in French, there is no restriction on the type of the fronted predicate, which may be an indefinite nominal expression.39 (51)

a. Arwr yw Sion. hero is Sion 4 Si6n is a hero.' cf. *A hero is John.

b. Un twp yw Sion. one silly is Sion 'Sion is someone silly.' cf. *A bore is John.

It is tempting to relate this asymmetry between Welsh and English or French to a structural difference. In these languages, predicate-initial sentences are well formed only if the predicate is a definite expression. It can be assumed that such constructions are inverse predicational constructions in which the subject is realized in SpecIP (= SpecAgrsP), a position which unequivocally qualifies as an A-position in English and in French.40 In Welsh, the lack of restriction on the predicate suggests that it occupies an A'-position, probably the specifier of CP which hosts the elements functioning as operators, whatever their status with respect to the definite/indefinite distinction. Binding phenomena lead to the same conclusion. The postverbal subject may function as an antecedent for an anaphor contained in the fronted predicate:

Bod (52)

149

a. Ei elyn pennaf ei hun yw Sion. his enemy chief himself is Sion 'Sion is his own worst enemy.' cf. *Himself's worst enemy is John. b. Rhybarod i wthio'i hunan ymlaen yw Sion. too ready to push himself forward is Sion 'Sion is too ready to push himself forward.'

This property suggests that at LF, a reconstruction process restores the fronted predicate to its original position. It is well known that reconstruction processes apply exclusively to operator-variable constructions and affect elements which are realized in an A'-position, usually in SpecCP. As for the copula, it can be shown that it is realized in C. Neither the relative particle a nor the complementizer y may intervene between the initial predicate and the verb. With respect to this property, there is a sharp contrast between copular sentences and predicational/existential sentences with a fronted predicate. It was established in section 2 that with respect to movement and extraction phenomena, PPs and AspPs behave as a natural class. In the examples illustrating this parallelism, the particle y occurs between the fronted element and the form of bod. In the present tense, y is frequently omitted before mae, but it is obligatorily realized in the past tense: (53)

a. Yma *(yr) oedd Arthur. here COMP was Arthur b. Wedi gweithio *(yr) oedd Sion. PERF work

COMP was Sion

A brief inspection of copular sentences containing the form oedd reveals that y is obligatorily absent in these constructions: (54)

a. (Y) brenin(*yr) oedd Arthur. (the) king COMP was Arthur b. Athrawes (*yr) oedd Mair. teacher COMP was Mair

This datum can be interpreted in one of two ways: either the copula is realized in C or, for some reason still to be determined, C must remain empty. The first analysis, which I will adopt, accounts for the complementary distribution between the subordinating particle and the copula in a straightforward way. In this account, two movements are involved in the derivation of predicate-initial copular sentences: the predicate is fronted to SpecCP, the copula is moved to C. Welsh copular sentences thus admit a derivation and a structure similar to those of verbsecond clauses in Germanic languages. In the standard analysis of these clauses, two movements are also involved: the movement of the inflected verb to C and the

150

Alain Rouveret

movement of an XP-constituent to SpecCP. On the other hand, the comparison between sentences (53) and sentences (54) strongly suggests that predicate-initial copular sentences and cleft sentences are two instances of the same propositional type: the former are concealed cleft sentences. Let us summarize the result obtained so far. Two movements take place in the derivation of predicate-initial copular sentences: the predicate is fronted to SpecCP; the copula raises to C. I intend to show that these two movements are necessary to license the predication relation between the nominal predicate and the subject.

3.2.2 Licensing the predication relation

We know that, when the two nominal expressions connected by bod are definite, the fronting of the predicate to SpecCP is one of the strategies giving rise to a well-formed output. The question that arises is why the definite predicate, if moved at all, can only be raised to SpecCP.41 The answer lies, we submit, in the functional structure of copular constructions. In Welsh, only indefinite bare noun phrases may be syntactically marked as stage-level predicates; definite noun phrases cannot be preceded by the predicative particle yn: (55)

*Y mae Arthur yn y brenin. is Arthur PRED the king

PRT

This property is probably related to another general characteristic of definite predicates: they cannot function as adjunct (secondary) predicates. In French, (56a) is legitimate, (56b) excluded: (56)

a. II est parti simple soldat he left ordinary soldier b. *I1 est parti simple soldat he left ordinary soldier

et and et and

est revenu capitaine. came-back captain est revenu le general, came-back the general

It can be assumed that the locative, aspectual or nominal/adjectival predicate in predicational/existential constructions, which is licensed by the locative clitic incorporated into mae, has the semantic status of a secondary predicate: it behaves as an adjunct with respect to the primary predicate mae. The expectation is therefore that definite predicates, which cannot function as adjuncts, should be excluded from the predicational/existential construction. This prediction is correct, as (55) shows. Welsh is not an isolated case among languages. The English example (57), cited by Hoekstra and Mulder (1990), is also ungrammatical:

Bod (57)

151

*In this country, there is a woman the president.

I will nevertheless maintain the hypothesis that copular constructions relating two definite expressions are structures of predication. The above restrictions only show that the way structures containing a definite predicate are licensed is different from the way indefinite-predicate structures are licensed. If definite nominals cannot be turned into stage-level predicates, it must be assumed, on grounds of economy, that the locative clitic is not present in the syntactic representation of the corresponding sentences. In what structural configuration is the copula inserted? My proposal is that the VP-domain headed by bod at D-structure is the direct complement of T and contains the two nominal expressions involved in the predication relation: the subject occupies its specifier, the predicate its complement position. The relevant substructure is given in (58): (58)

[TP . . . T [VP NP, [y [v bod ] [ NP \ ]]]

Note that, given (58) and the theory of derivations outlined in Chomsky (1993), the predicate NP7 cannot be moved to a higher (L-related) A-position across the subject NP, without violating the Minimality Condition. After the adjunction of bod to T, SpecTP and SpecVP are equidistant from the original site of the predicate. But it is the subject which must move to SpecTP, since it is the designated position for the assignment of nominative Case. Later raising of bod-T to Agrs creates a new minimal domain, but the complement position of VP is not part of this domain. The only available option is for the predicate to move directly to a (non-L-related) A'-position, the specifier of the CP-domain minimally containing TP. This analysis correctly predicts that in copular sentences the predicate cannot intervene between the verb and the subject, contrary to what happens in existential sentences: (59)

*Yw'r brenin Arthur. is-the king Arthur

The absence of the locative clitic in copular constructions also implies that only one Case is available for the licensing of the subject, the nominative Case assigned in SpecTP. In other words, the subject NP must leave the VP in order to be Casemarked. How is the predication relation licensed in predicate-initial structures? There is no doubt that the propositional domain AgrJP in (60) is construed in relation with the fronted element: it functions as a predicate with respect to the initial term, v/hich fulfils the subject role in the relation. (60)

[ C P [spec y brenin]j [ c yw]y [AgrsP • • • Ugrs e ] v [TP [spec Arthur^ [T e ] v [VP [NP e ]/ [V' [v e ] v [NP e ],]]]]]

152

Alain Rouveret

Williams (1980) claims that propositional domains may function as predicates, that is, may be construed as being related to a nominal expression functioning as the subject in the relation, only if they contain an element - the predicate variable - lacking independent reference. A property of predicate variables is that they must be present at S-structure. In (60), Agr s P clearly qualifies as a complex predicate since it is an unsaturated domain containing a variable bound by the fronted nominal expression realized in SpecCP. It is easy to check that the null element occupying the original position of the predicate in (60) is properly licensed. The fact that this element is not governed by a clitic indicates that it is not a pronominal variable and strongly supports the view, assumed so far, that a movement is involved: the predicate raises from its original position to SpecCP, leaving a trace that functions as a variable. This trace is properly head-governed by the verbal chain headed by yw. It is also minimally antecedent-governed by the predicate in SpecCP.

3.2.3 Focus interpretation Let us now address the second question raised by these structures: how is the focus interpretation of the initial predicate derived? It is plausible to assume that the process moving the predicate to SpecCP can be characterized along the same lines as focus fronting in other languages. If syntactic movement is exclusively driven by morphological necessity, as Chomsky (1991, 1993) claims, a term may be moved to SpecCP only if it is endowed with a feature [ + wh] or [ + topic] or [ +focus]. I will make the hypothesis that in the structures under consideration, the predicate is marked [ + focus] at D-structure. The well-formedness of example (9) shows that indefinite predicates may also be marked [ +focus]. Following Brody (1990), I will also assume that preposed focus phrases are realized in the specifier position of a head having an operator feature [ + f] indicating identificational interpretation, that is, 'focushood'. The fact that only one focused element may be preposed to sentence-initial position confirms the view that a substitution operation is involved. The relation between the [ + f] head and the content of its specifier must be checked at some level. In Welsh, this relation is governed by principle (61): (61)

The specifier of a [ + f] head must contain a focus phrase at S-structure.

To account for the obligatoriness of copula-raising to C, one can think of [ 4- f] as a feature which must be 'strong' when it coexists with a focus phrase and which is strong only when it is morphologically supported. This condition can be satisfied in one of two ways: either C contains a particle - ordinary cleft sentences instantiate this possibility - or the copula is adjoined to it. The relevant configuration is schematized in (62):

Bod (62)

[ C p [spec y brenin]i [ c yw]w [ A g r s P . . . [ A g r s e ] v [ T P [spec Arthur [ +focus] [ + f]

153 ] . . . [ N P e ], ]]]

3.3 Subject-initial copular structures

In the preceding analysis, only predicate-initial structures were taken into account. Subject-initial copular clauses have basically the same properties. In particular, they do not allow the insertion of the particle y between the initial subject and the copula: (63)

Mair(*yr) oeddyr athrawes. Mair COMP was the teacher 'Mair was the teacher.'

Another characteristic common to the two structures is the nature of the element bound by the initial term: it is identifiable as a variable. In order to establish this point, it is necessary to digress a bit and to clarify the syntax of agreement in copular sentences. In predicate-initial sentences, agreement obligatorily involves the postverbal subject position: (64)

(65)

a. Y brenin ydw i. the king be.lsG I 'I am the king.' a'. *Y brenin yw myfi. the king be.3sG me b. Plant diog ydych chwi. children lazy be.2PL you 'You are lazy children.' Pethau cas ydyw'r brain, things dreadful be.3SG-the crows 'Crows are dreadful things.'

In (64 a, b), the inflected verb incorporates the person and number features of the postverbal subject. The co-occurrence of the third person of the copula with a first-person pronoun in postcopular position gives rise to ungrammaticality (see (64 a')). When the postverbal subject is a noun phrase, the copula bears the poor agreement inflection (65). In this respect, the copula behaves exactly as ordinary verbs in verb-initial declarative sentences. The fact that the alternation between rich and poor agreement manifests itself with yw, and that the associated morphology is clearly verbal and not pronominal, unequivocally shows that the copula is a verbal form.

154

Alain Rouveret

In subject-initial structures, the copula bears the mark of poor inflection, whatever the features of the preverbal subject. This property manifests itself clearly when the focalized element is a pronoun: the presence of a first-person singular pronoun in preverbal position does not induce the realization of a firstperson feature on the verb. (66)

a. Fi yw'r brenin. I be.3sG-the king 'I am the king.' cf. French Le roi, c'est moi. b. Os ti yw'r marchogwr gorau, fi yw'r rhedwr gorau. if you be.3sG-the rider best I be.3sG-the runner best 'If you are the best rider, I am the best runner.'

The fact that the pronoun is an independent pronoun, not a dependant one, supports the conclusion that the precopular position does not define an agreement context. This behaviour recalls what is observed in cleft structures and in relative clauses in which the focalized/relativized element is the local subject of an ordinary verb: (67)

a. chwi a ddaeth you REL came 'You came' OR 'you who came b. fy mrodyr a saethodd y petris my brothers REL shot the partridges 'my brothers shot the partridges' OR 'my brothers who shot the partridges

In both cases, the 'anti-agreement' effect manifests itself:42 the null element realized in SpecTP is not governed by a rich agreement inflection. In Rouveret (1991), it is argued that the derivation of synthetic forms (the forms which are specified for a rich inflection) involves the incorporation of a number affix into the Agrs category containing the person affix and that no incorporation into Agrs takes place in the derivation of the analytic form. The fact that the copula in (66a, b) and the ordinary verb in (67a, b) are third-person singular and are not specified for number can be taken as evidence that no incorporation from SpecTP into Agrs is involved. Rather, the focalized or relativized pronominal subject is directly moved from SpecTP to SpecCP. The null element realized in the postverbal subject position and bound by the moved element must be identified as a variable, not as pro. I will assume that in subject-initial copular sentences, the subject has been moved first to SpecTP, then to SpecCP and that the copula has raised to C. The S-structure representation corresponding to (68a) is (68b):

Bod (68)

155

a. Arthur yw'r brenin. b. [Cp [Arthur ], [ c yw]v [AgTSP . . . [Agrs e ]

v

[TP [spec e ],- [T e ] v [VP [NP e ],- [V' [v e ] v [NP y brenin ],-]]]]] The trace realized in SpecTP is properly head-governed by yw and antecedentgoverned by its antecedent. Principle (61) is satisfied at S-structure by the raised subject, which is marked [ +focus] in the initial representation. The adjunction of the copula to C in overt syntax renders the [ + focus] feature of the latter strong.

3.4 Embedded copular sentences In Welsh, cleft sentences may be embedded. When they are, the particle following the focused element is the same as in non-embedded contexts, the embedded domain is introduced by a special complementizer: mai.43 Predicational sentences with a focalized prepositional or aspectual predicate follow this general pattern: (69)

a. Clywsom mai yn y neuadd yr oedd y cyngerdd. we-heard coMPin the hall COMP was the concert 'We heard that the concert was in the hall.' b. Gwn mai wedi gadael y glwyd ar agor y mae'r ffermwr. I-know COMP PERF leave

the gate

on open COMP is-the farmer

'I know that the farmer has left the gate open.' Tallerman (this volume) convincingly shows that embedded cleft constructions should be analysed as CP-recursion structures in which the lower C category is occupied by the particles y and a.44 In Welsh, the marked property of C to select a CP domain is signalled by the special complementizer mai, which can be assumed to be exclusively specified in the lexicon for the selection of a CP complement.45 Let us now consider the embedded copular constructions exemplified in (70): (70)

a. Cred Peredur mai'r brenin yw Arthur. believes Peredur coMP-the king is Arthur 'Peredur believes that the king is Arthur.' b. Cred Peredur mai Arthur yw'r brenin. believes Peredur COMP Arthur is-the king 'Peredur believes that it is Arthur who is the king.'

The fact that the embedded domain is headed by mai, not y, suggests that these structures are also instances of the C CP configuration. But mai does not coexist with the subordinating particle y. As in matrix copular sentences, the insertion of y between the initial predicate or the fronted subject and the verb bod gives rise to ungrammaticality. This asymmetry between embedded structures with a fronted

156

Alain Rouveret

aspectual or prepositional predicate and embedded copular sentences with a nominal predicate strongly suggests that in the latter, the lower C category is occupied by the copula itself.46 If the fronted predicate or the fronted subject was realized in SpecAgrsP, rather than in SpecCP, a single complementizer would be involved and it would be difficult to explain why the embedded domain is headed by mai rather than by y.41 From the above considerations, it can be safely concluded that embedded copular structures are embedded verb-second clauses and can be analysed along the same lines as root copular sentences.

3.5 Predicational yw The data examined so far could lead to the conclusion that the mae/yw distinction is the lexical reflection of the semantic difference between the sentences in which bod expresses a qualification or a class membership and those in which it establishes a relation of equivalence or identity between two terms. However, a closer examination reveals that the mae/yw partition does not coincide with the predication/identity divide. The obligatory choice of yw in structures containing a definite predicate was related to the fact that definite nominals are canonically individual-level and, as a consequence, cannot be governed by the particle yn and do not require the projection of the locative clitic in the structure. But the ^^-construction is a possible choice when the predicate is a bare noun or an indefinite nominal expression. The overall interpretation of the corresponding sentence is clearly predicational. (71)

a. Meddyg yw Sion doctor is Sion 'Sion is a doctor.' b. Un twp yw Sion. Someone stupid is Sion 'Sion is someone stupid.'

The copula never occupies the clause-initial position. But in contrast to what happens with definite predicates, indefinite ones obligatorily precede the copula. The construction Subject-^w-Predicate is not available.48 (72)

a. *Yw Sion feddyg. b. *Sion yw meddyg.

We know that indefinite nominal expressions can function as complements of the particle yn and appear in the mae-construction. This suggests that they can be construed as stage-level predicates. The fact that they are also legitimate in the jw-construction can be taken as evidence that bare nominals are unspecified with

Bod

157

respect to the stage/individual distinction and that it is the combination yn + NP which qualifies as a stage-level predicate and imposes the projection of the locative clitic. This analysis is in accordance with the claim made in section 2.2 that the stage-level/individual-level asymmetry is not encoded in the lexical categories themselves.49 If it is on the right track, nothing should prevent a bare NP from co-occurring with the form yw and occupying the SpecCP position in the ywconstruction. Yw also functions as a substitute for mae in predicational sentences headed by an interrogative or a negative particle: (73)

a. A ydyw Ifan yn bregethwr? Q is Ifan PRED preacher 'Is Ifan a preacher?' b. Nid yw plant yn dweud y gwir bob amser. NEG is children PROG tell the truth always 'Children do not always tell the truth.'

The constructions exemplified in (73) have a clearly predicational interpretation and are not instances of the identificational or equative type. They show that yw substitutes for mae in structures in which existential bod would be immediately preceded by a negative or an interrogative particle. Yw seems to be imposed by the presence of the particle, not by the nature of the predicate which, as in predicational sentences, is preceded by yn, when adjectival or (indefinite) nominal. It is plausible to assume that negative and interrogative predicational sentences have the same structure as the corresponding affirmative constructions: the initial C is occupied by the interrogative or negative particle and the verb is adjoined to Agrs. These distributions teach us that yw is not always realized in C. The occurrence of yw in negative and interrogative predicational constructions tells us very little about the form yw itself, since the inflected forms of ordinary lexical verbs in all tenses and the past forms of bod can also be immediately preceded by these particles. It is their incompatibility with mae which calls for an explanation. This restriction can be made to follow from the principle barring vacuous quantification, combined with the assumption that mae is an 'autosaturated' form, incorporating a locative adverbial clitic, which functions as an operator at LF. As a result, mae cannot be bound by a quantifier and is not appropriate in negative and interrogative structures. In order to account for the legitimacy of yw in the same contexts, one can assume that the lexical/syntactic representation of yw includes an unsaturated place, which can only be bound by an operator. In (73), the operator is the interrogative or negative quantifier underlying the initial particle. This characterization trivially accounts for the fact that the constraint which bars yw from the sentence-initial position also operates in these structures.51 But it must

158

Alain Rouveret

be kept in mind that the 'quantificational' behaviour of yw does not differ from that of ordinary inflected verbs in the present tense. If yw includes an unsaturated place, ordinary inflected verbs should also contain one. I will show in section 4 that this is indeed the case and that the solution lies in the deficient status of the morphological present tense in Welsh. This characterization of yw is compatible with the other uses of this form. A quantification relation is also involved in predicate and subject cleft constructions (see (49a, b), (50)). The operator in this case is the focus feature in C, which licenses the empty place contained in the verb when the latter moves to C. 52 The properties of negative/interrogative predicate-initial sentences exemplified in (74) also fit in with our analysis: (74)

a. Ai wedi gweithio y Q

PERF work

mae Sion?53

COMP is

Sion

'Has Sion worked?' b. Ai yn y simnai y mae'r aderyn yn nythu? Q in the chimney COMP is-the bird PROG nest I s it in the chimney that the bird is nesting?' c. Nid canu y mae Sion. NEG sing COMP is Sion 'It is not singing that Sion is.' The structure associated with these constructions is schematized in (75): (75)

[ c Q/Neg ] [CP [spec XP ] [ c y ] [AgrsP [Agrs mae ] . . . ] ] where XP = PP, AspP, PredP, Adv

The form of bod appearing in examples (74) is the same as the one used in the corresponding affirmative constructions. The occurrence of mae can be taken to indicate that the initial particle has no access to the proposition introduced by y: it does not bind the verb; semantically, only the focused constituent is included in its scope. Similarly, the initial particle has no effect on the form of bod selected in negative/interrogative copular sentences. As expected, it is the form yw which is found in this case: (76)

a. Ai pregethwr yw Ifan? Q preacher is Ifan 'Is Ifan a preacher?' b. Nid athro yw ef. NEG teacher is he 'He is not a teacher.'

In the corresponding structure, yw is realized in the lower C category and, as in affirmative copular sentences, incorporates the focus feature which licenses the

Bod

159

empty place it contains. The initial particle itself has no access to the lower C category. Only the focused constituent is included in its scope.

4 The status of the present tense For the analysis to be complete, we must explain why the distinction between predicational/existential bod and copular bod only manifests itself in the present tense. It seems natural to look for an explanation of this fact in the properties of the present tense in Welsh. 4.1 The present-tense effect It is well known that the temporal reference of sentences in the morphologically simple present tense is not uniform across languages and that this variation crucially manifests itself with eventive predicates. In French, eventive verbs in the present tense have either a continuative interpretation - they denote an ongoing, uncompleted event {Marie chante en ce moment) - or a generic/habitual interpretation {Marie chante des cantiques), while only the latter interpretation is available in English {Angelo builds houses, *Angelo builds a house). Stative verbs in the present tense license a normal present-time interpretation in both French and English. Campbell (1991: 161) notes that 'the availability of specific present time interpretation in the morphologically simple present tense is a diagnostic for the stative/nonstative distinction in English' and refers to this feature of the present tense as the present-tense effect. Eventive predicates in Welsh also show the present-tense effect. As in English, the morphologically simple present tense cannot have a specific present-time interpretation and disallows reference to a specific event. The future meaning is prevalent in Colloquial Welsh, the present sense being expressed by the aspectual periphrasis bod . . . yn . . . But the present tense can also encode the so-called historical present (interpreted as a scheduled future or as an aorist referring to past events), mark an action or a state as recurring habitually in the present or denote what is true at all times. (77)

(78)

a. Ysgrifenaf. I-write 'I will write.' b. Gwnaf fynd I-do go 'I will go.' Ysgrifena'r awdwr hwn bob amser yn dda. writes-the author this always PRED well This author always writes well.'

160

Alain Rouveret

(79)

Yna gwelir ef yn rhedeg. then is-seen him PROG run 'Then, he is seen running.'

Welsh differs from English, however, in that the vast majority of stative predicates do not license a specific present-time interpretation, only a very limited subclass does. Fife (1990: 105) gives the list in (80) as exhaustive. (80)

bod gweld tybied teimlo

'to 'to 'to 'to

be' be' see' imagine' feel'

gallu, medru clywed gwybod, adnabod credu, coelio

'can' 'to hear' 'to know' 'to believe'

These predicates, which denote a mental process or an intellectual operation, a sensory perception or various notions of ability, are used in the present tense with a present-time value. The same is true for the verb bod.54 The problems raised by the reference of verbal tenses across languages are extremely complex and it is not possible here to mention all the approaches which have been developed to deal with them. Some are quantiflcational in nature and extend the notions and concepts used in the analysis of nominal reference to the analysis of verbal and temporal reference. The general idea underlying Enc's (1991), Campbell's (1991) and Stowell's (1993) treatment of the interpretive properties of the English tenses is that the present tense has no quantiflcational force. The sentences containing a form in the present tense are quantiflcational structures in which the range of the temporal variable is determined by a nonovert operator distinct from T, GENERIC or HABITUAL to be precise. The temporal reference of sentences in the past is determined by the Past Tense element itself, functioning as an operator (or as a referential element). I will extend this characterization to Welsh and assume the following: (81)

The morphological present tense in Welsh cannot function as a Tense operator at LF.

The range of interpretations associated with sentences in the present tense in Welsh suggests that the morphological present does not have the capacity to license a temporal variable at LF. I will assume that in the eventive constructions showing the future or habitual reading, the binder is a non-overt operator FUTURE Or GENERIC/HABITUAL.55

If (81) is correct, why do stative and eventive predicates not have the same temporal reference in the simple present tense? The answer probably lies in the different aspectual properties of states and events, as Dechaine (1993) proposes. States are properties of moments; events, which have an internal structure and crucially occur over intervals, are not. Hence, in languages showing the presenttense effect, only states can be interpreted as holding at a time which coincides or

Bod

161

overlaps with the utterance situation, giving rise to a non-past interpretation. In order to represent this difference syntactically, I will simply assume that the event argument of eventive predicates is locally bound by PAST in the past tense,56 by the non-overt operator GENERIC/HABITUAL or FUTURE in the present tense.57

4.2 Bod in the present tense

Like the other stative predicates listed in (80), the verb bod in the simple present tense licenses a present-time interpretation. I wish to claim, however, that the mae/yw alternation should be viewed as another correlate of the property of the simple present tense stated in (81). Bod being a stative predicate, its argument structure includes no event argument. However, the locative/aspectual/ 'predicative' stage-level predicate selected by bod in its predicational/existential use contains a spatio-temporal argument which has to be licensed. Since the morphological present has no quantiiicational force, it is not able to bind this argument. The recourse to the form mae, which incorporates a locative clitic, is necessary. No specific form of bod is required in the past tense, since the morphological past can function as an operator at LF, binding the spatio-temporal arguments of stage-level predicates. As for the form yw, it is legitimate (i) when there is no spatio-temporal argument to bind, for example when the predicate is a definite nominal denoting an individual level property; (ii) when the context makes an extra operator available (NEGATION, QUESTION), able to license the spatio-temporal arguments of stage-level predicates. If this analysis is correct, the following picture emerges: there is only one verb bod c-selecting a complement able to function as a predicate; the morphological alternation and the semantic divide between existential bod and copular bod have a strictly syntactic and logical basis. The three contributing factors in this alternation are the stage-level/individual-level distinction, the deficient quantificational status of the present tense in Welsh, and the specific syntax of the verb bod, which can incorporate a locative clitic or raise to C in overt syntax. Notes I am extremely grateful to Gwen Awbery, Bob Borsley, Rose-Marie Dechaine and Ian Roberts for extensive written comments on an earlier draft of this chapter. I also wish to thank Marcel den Dikken, Randy Hendrick, Bob Morris Jones, Lea Nash, Georges Tsoulas and Laurie Tuller for discussing various aspects of the analysis with me, as well as two anonymous referees for Cambridge University Press for observations that forced some much-needed clarification. Cited by Moro (1991).

162

Alain Rouveret

3 Moro (1991) shows that this analysis opens the way to a representation of the syntactic, referential and interpretive asymmetries between the former and the latter. It is not without problems, however. In order to explain that the raising of the predicate across the small-clause subject does not give rise to a Principle A violation, Moro assumes that SpecIP is an A'-position. As observed by Heycock (1991), the ability of this movement to feed subsequent A-movement shows that this claim cannot be correct: (i) (ii)

4

5 6

7 8 9 10 11

The real problem is the prime minister. The real problem seems to be the prime minister.

She proposes that the predicate of the small clause raises first to the VP-subject position and then to the IP-subject position - a possibility which Moro does not consider - and suggests that in the resulting structure, the domain in which the trace of the predicate must be bound is the VP headed by the copula. Note that if the distinction between A- and A'-positions is subsumed under the distinction between L-related and nonL-related positions, as Chomsky (1993) proposes, SpecIP unequivocally qualifies as an L-related position, since I is a functional category morphologically related to a lexical category. In these structures, the definite nominal expression functioning as a predicate is not used referentially, although it has an independent denotation. On the complex relations between 'reference' and 'denotation', see Vergnaud and Zubizarreta (1992). This proposal, due to Kratzer (1989), will be clarified in section 2. This chapter only tackles the problems raised by the mae/yw alternation and says nothing about the other forms of bod in the present tense: sydd and oes. It is to be hoped that the approach based on (10)—(13) will carry over to these forms. This holds only for reasonably formal registers. In less formal Welsh and in the Southern dialects, the particle is often omitted. On this point, see Benveniste (1960), Lyons (1967), Freeze (1992), among others. See Fife (1990) and Borsley (1993) for detailed discussions. The particle newydd, which is clearly adverbial, not prepositional, constitutes an exception to this generalization. In Rouveret (1994), I argue that the D head of a verbo-nominal DP can only be filled by a pronominal clitic binding the genitive position governed by the verb-noun: (i)

Mae Mair wedi ei weld [e]. isMair PERF CLsee 'Mair has seen him.'

The prefixed and infixed pronouns in nominal DPs should be viewed as 'autonomous' clitics realized in D, moving directly from SpecNP (the genitive position) to D in overt syntax. For a different characterization, see Roberts and Shlonsky (this volume), who analyse the clitic in (i) as a base-generated Agr element lacking v-features, hence blocking v-movement to Agro and Agrs; see also Roberts (1994c). 12 One of the arguments against the prepositional analysis is that yn, in its different uses, triggers different mutations: the concrete preposition takes nasal mutation, the aspectual marker has a null mutation effect. As for the predicative marker yn governing the adjectival and nominal predicates in (16), it takes soft mutation. Whether this difference

Bod

13 14

15

16 17 18

19

163

must be interpreted as an indication that three distinct elements are involved, I will leave open. On this question, see Watkins (1957, 1960), Fife (1990) and Hendrick (this volume). Note that an analysis of the yn + NP/AP sequences as PPs, rather than as projections of a one-member functional category Pred, should not be excluded a priori. For clarity, I will continue to refer to the constituents headed by aspectual particles as AspPs and to the domains headed by predicative yn as PredPs. Bob Borsley (p.c.) rightly observes that the status of aspectual markers and the status of VNPs are separate matters. The former could be prepositions without the latter being DPs. It must be noted, however, that the characterization of VNPs as non-finite VPs is compatible with the prepositional analysis of aspectual markers only if non-finite verbs - as distinct from both non-finite clauses and nominal forms of the verb - can be shown to function as objects of prepositions. See Awbery (1976: 121^) on stative predicates in Welsh. This analysis raises an obvious question: how is the difference between stage-level and individual-level predicates to be represented in languages which have only one verb be and make no recourse to a predicative marker? English is a case in point. I will leave the question open. See Stump (1985), Diesing (1992). This proposal forces us to analyse constructions like (19c) as containing a locative predicate denoting a stage-level property. This unexpressed locative predicate means something like 'in the universe'. In Travis' analysis, the higher V position is empty at D-structure. On the syntactic properties and semantic function of cleft constructions in Welsh, see T. A. Watkins (1991). Cleft constructions involving the local subject or direct object show the 'relative particle' a, instead of y\ those involving a non-local subject or direct object use the particle y, but have a resumptive pronoun in the original position of the focalized element. When the focused constituent is the aspectual predicate of a progressive construction, yn has no realization: (i)

Gweithio y mae Sion. work COMP is Sion 'It is working that Sion is.'

This could be taken to indicate that the particle yn, which heads the aspectual predicate, incorporates into bod prior to predicate-raising and that it is the beheaded AspP/PP which is fronted to the specifier of CP. Whatever its merits, this hypothesis does not explain why the realization of yn is obligatory in non-inverted progressive constructions, that is, why the incorporation of the particle cannot take place in this case. I will leave the question open. 20 This implies that in (i), the subject Gwyn has not been extracted from the AspP containing the 'raising verb' dechrau: (i)

Mae Gwyn wedi dechrau darllen y llyfr. is Gwyn PERF begin read the book 'Gwyn has begun to read the book.'

164

Alain Rouveret

(I owe this example to Bob Borsley.) The well-formedness of (i) can be accounted for only if specific claims concerning the thematic behaviour of verbo-nominal domains are introduced. For a proposal relevant to the problem under consideration, see Rouveret (1994). Analysis (26) also has obvious implications for the proper analysis of reconstruction effects in fronted predicate constructions. I will not explore them here. 21 As shown by the following example: (i)

Gwelais y ty y buoch ynddo. I-saw the house that you-have-been in.3sG.M 'I have seen the house in which you have been.'

22 A small-clause analysis of 6o
23

24 25

26

[VP [vlv bod ] [AspP NPS [AspPAsp [VNp VN (NPO) ]]]]]

Following Chomsky and Lasnik (1993), one can analyse the higher AspP projection as a small clause, the position occupied by the subject as its specifier and the lower AspP projection as its head. The small-clause analysis, however, meets non-trivial problems which will be discussed in section 2.4. Empirical arguments supporting the view that in Welsh the finite subject is realized in SpecTP at S-structure (except in presentational mae-sentences) are given in Rouveret (1991, 1994). Bobaljik and Carnie (this volume) reach the same conclusion for Irish. For different analyses of this phenomenon, see Emonds (1976), Hoekstra and Mulder (1990). Ian Roberts points out to me that, if one sticks to the strict interpretation of Minimality developed in Chomsky (1993), the recourse to equidistance in (32) is neither legitimate nor necessary. Equidistance can be invoked only for positions which are potential landing-sites for feature-checking. In the Minimalist Program, only the specifiers of functional heads define checking positions. The specifier of the higher VP in (32) does not fulfil this condition. If this strict interpretation is adopted, one of the arguments against the small-clause analysis of the complement of bod disappears. But this conception might turn out to be too restrictive, precisely when the specific properties of Larsonian VP-structures are taken into account. For example, Chomsky (1994) observes that in these structures, the adjunction of V to the higher head position (which he takes to be occupied by a light verb v) does not by itself allow the satisfaction of a morphological property of V. Only the ulterior movement of the V-v complex to the higher functional heads does. Similarly, one could suggest that the specifier of VP] in (32) counts as a potential landing-site with respect to equidistance, although it does not qualify as a checking position. The fact that the phenomenon of predicate inversion raises a serious problem for the small-clause analysis has also been noticed by den Dikken (1994). His solution is to

Bod

165

analyse small clauses as projections of a functional head Agr, which can only be incorporated into a higher verb be. 27 When the subject is pronominal, it is realized as a proclitic affixed to bod. (i)

Dywedodd Mair ei fod yn yr ardd. said Mair CL be in the garden 'Mair said that he was in the garden.'

The clitic can be shown to be realized in C (see Rouveret 1994). 28 However, Gwen Awbery observes that pattern (38b) gives a well-formed result when the postverbal subject is heavy. She provides the following examples: (i) (ii)

??A oes gan eich tad bibell? 'Has your father a pipe?' A oes gan eich tad bibell hen ffashiwn fel yr un yn y llun? 'Has your father an old-fashioned pipe as the one in the picture?'

29 See Rouveret (1991) for an analysis of prepositional (and verbal) inflection in Celtic in terms of incorporation into Agr. Roberts and Shlonsky (this volume) argue that Celtic prepositional and verbal agreement are pronouns diachronically reanalysed as Agr heads. 30 This means that in example (ii), note 28, the PP headed by gan is realized in the specifier of the higher VP and the subject in the specifier of the lower one. 31 In this respect, the Tense present in bod-initia\ constructions behaves as a finite, morphological Tense: it requires the projection of a VP, the head of which hosts bod. 32 But this option exists for a small subset of prepositional predicates only, including gflfl-predicates. 33 This analysis recalls den Dikken's (1994) account of predicate inversion inside small clauses. Den Dikken identifies Z with Agr. 34 In mfle-constructions and in bod-'mitia\ structures, the Case assigned to definite and indefinite NPs in SpecTP is nominative. Whether the Case assigned in SpecTP in absolute constructions is also nominative or a default Case, I will leave open. 35 This holds for identificational, equative, specificational and predicational sentences. 36 But, as Fauconnier (1992) observes, a proper name may also be used to denote a (theatrical) role: (49a) and (49b) also have a reading in which the king plays the role of Arthur. 37 This example, which contemporary speakers judge a bit marginal, is given by Anwyl (1899: 157). 38 This value of the initial term manifests itself clearly in question-and-answer structures: (i)

a. Pwy yw eich ffrind? who is your friend 'Who is your friend?' b. Pwy yw Myfanwy? who is Myfanwy 'Who is Myfanwy?'

-

Myfanwy (yw fy ffrind). Myfanwy is my friend 'Myfanwy is.' Fy ffrind (yw Myfanwy). my friend is Myfanwy - 'She is my friend.'

166

Alain Rouveret

39 For a semantic explanation of the ungrammaticality of the corresponding English examples, see Partee (1987) and Gueron (1992). 40 On this point, I differ from Moro (1991); see n. 3. 41 Recall that the locative-possessive predicate in the raae-construction has access to SpecTP. 42 Example (67a) is discussed in Awbery (1977). On the anti-agreement effect, see Ouhalla (1993) and Rouveret (1994). 43 Taw is also found in the Southern dialects. Anwyl (1899 : 177) observes that mai and taw are etymologically related to the verb bod. Although these elements exclusively function as complementizers in contemporary Welsh, embedded cleft sentences can be taken to provide additional evidence in favour of the idea that bod is the only verb in Welsh which can raise to C in overt syntax. 44 A phenomenon similar to the one described here could be observed in Old French (the following example is taken from Le Roman de Guillaume de Dole dating back to the thirteenth century): (i)

semble Que cil qui a fet le romans Qu'il trovast toz les moz des chans 'it seems that the one who told the story that he found all the words of the songs'

45 It should be noted that the CP-recursion option seems to have a much larger extension in Welsh than in the languages for which it was initially devised. In Germanic, it is restricted to a specific context, the complement of bridge verbs. In Welsh, it is generalized to all subordinate clauses. If embedded clefts involve a C CP structure, then one must admit that the CP-recursion option is available in indirect questions introduced by the particle ai, as well as in negative complement clauses introduced by the negative complementizer nad: (i)

a. Gofyn ai eich brodyr a ddaeth ddoe. he-asks whether your brothers REL came yesterday 'He asks whether it was your brothers who came yesterday.' b. Dwed nad eich brodyr a ddaeth ddoe. he-says NEG your brothers REL came yesterday 'He says that it wasn't your brothers that came yesterday.'

These examples clearly show that the selection of a CP domain is not an exclusive property of the complementizer mai. Embedded copular sentences (see below) are also found in indirect questions and negative complement clauses: (ii)

Dwed y dyn nad athrawes yw Mair. says the man NEG teacher is Mair 'The man said that Mair is not a teacher.'

Moreover, the occurrence of mai is not restricted to predicate complement clauses. It may also head a clause which is the object of a preposition. This linguistic variation raises a difficult problem, which I will not consider here.

Bod

167

46 In this respect, they resemble the embedded verb-second or auxiliary-second structures in Germanic languages and in English, which, according to Rizzi and Roberts (1989), involve CP-recursion: (i)

He said that under no circumstances would he do it.

They also show the island behaviour typical of C CP structures (see Rizzi and Roberts 1989), as the following examples show: (i)

a. *Arthur y Arthur that b. *Y brenin the king

credaf mai'r brenin yw. I-believe* that-the king is y credaf mai Arthur yw. that I-believe that Arthur is

47 This analysis differs from the one developed in Rouveret (1991), where the initial element in embedded copular sentences was assumed to occupy the Agr s P position, yw being realized in Agrs. This account was embedded in a framework in which any position not included in a Case-marked chain was identified as an A'-position. In the framework adopted here, the L-related position Agr s P cannot be an A-position and cannot serve as a landing-site for focalized constituents (see n. 3). The distributional arguments given in the text show that this theoretical claim is also empirically motivated. 48 A second difference between definite and indefinite predicates is that the form mae may be retained when an indefinite predicate is fronted. (i)

?? Yn feddyg y mae Sion.

The result is judged marginal by many speakers, but Jones (1993) reports that this construction is used in the dialects. 49 The constructions in which an adjectival phrase occupies the initial position are also basically predicational sentences with a fronted predicate. (i) (ii)

Tal ydy'r athro. tall is-the teacher Ffeind wrth bawb ydy Mair. kind to everyone is Mair.

Not surprisingly, AP predicates are excluded from the postcopular position. Examples (i) and (ii) are provided by Jones (1993). There is some variation among native speakers concerning the status of this construction. According to Awbery (p.c), (iii) is excluded; only (71b), where the adjective modifies an indefinite pronoun, is grammatical: (iii)

*Twp yw Sion.

50 When yw co-occurs with an indefinite predicate, the latter is obligatorily realized in SpecCP; the subject-initial construction is not available (see (72b)). This restriction could be related to the fact that the normal way of focalizing the subject of an indefinite predicate in Welsh is to resort to a structure containing the form sydd:

168

Alain Rouveret

(i)

a. Sion sydd yn feddyg. Sion who-is PRED doctor 'It is Sion who is a doctor.' b. Ef sydd yn athro. he who-is PRED teacher 'He is a teacher.'

In the sydd-construction, the relation between the predicate and the rest of the clause is mediated by the same markers as those which occur in the mfle-construction. (ii)

(iii)

Tegell sydd ar y tan. kettle that-is on the fire 'There is a kettle on the fire.' Mair sydd wedi ennill. Mair who-is PERF win 'It is Mair who has won.'

This suggests that sydd incorporates a locative clitic. Recourse to the syafaf-construction is thus impossible when the predicate is a definite (individual-level) nominal. The fact that sydd also appears in relative clauses formed on the local subject of a predicational sentence suggests that it is a suppletive form, 'standing for' the combination a + mae, which is unattested. According to Hendrick (this volume), sydd substitutes for V + AGR, not for C + V + Agr, and is required to ensure the proper government of the trace in subject position. In this view, the absence of the relative particle a in the corresponding construction can be taken as evidence that sydd raises to C in overt syntax, as yw does. Contrary to indefinite nominals, aspectual predicates are not legitimate in affirmative jKvv-constructions. Example (i) is ungrammatical. (i)

*Wedi gweithio yw Sion. PERF work is Sion

This restriction is not unexpected (ordinary PPs are also excluded from the pre-copular position). Suppose that a verbo-nominal DP may function as a predicate only if (a) it is governed by null Tense, or (b) it is governed by a preposition functioning as an Aspectmarker. Verbo-nominal clauses instantiate the first possibility (detailed arguments in favour of the analysis of these clauses as CP domains containing a null Tense are produced in Rouveret 1994): (ii)

Dywedodd Sion i Mair fynd i Gaerdydd. said Sion to Mair go to Cardiff 'Sion said that Mair had gone to Cardiff.'

Aspectual periphrases illustrate the second possibility. The obligatory choice of mae in aspectual constructions follows from the stage-level status of the Aspect-marker (see section 2.2).

Bod

169

51 This generalization holds for reasonably formal registers of the language. In spoken Welsh and in the dialects, it is obscured by the fact that the initial particle may be omitted. But the fact that the initial verb is affected by the same mutations when the particle is phonetically realized and when it is omitted suggests that it is syntactically present in the latter case (the mutation only affects verbs whose initial letter is a stop consonant): a. A welaist Fred? Q you-saw Fred 'Did you see Fred?' b. Welaist Fred? c. Ni welais Fred. NEG saw Fred 'I didn't see Fred.' d. Welais i ddim Fred, saw I NEG Fred

(radical form: gwelaist)

(radical form: gwelais)

52 Contrary to yw, mae does not move to C in overt syntax. The C-position in predicational cleft sentences is filled by the subordinating particle y (see (28a, b)). It is necessary to suppose that, in this case, the focus feature does not count as an operator at the level at which the choice between mae and yw occurs, that is, S-structure. This result follows if one assumes that the focus feature is strong (counts as an operator) only when a verb adjoins to it in overt syntax. For a proposal along these lines, see Chomsky (1993). 53 Interrogative ai has a status analogous to that ofmai: ai is an instance of C, specified in the lexicon for the selection of a CP complement. 54 This distinctive semantic property has a syntactic reflex in question-and-answer constructions. The predicates listed in (i) are the only ones which can be used in answers referring to present reality. (i)

a. A weli di y ty? 'Do you see the house?' b. A glywi di'r swn? 'Do you hear the noise?' c. A alii di ddatod y cwlwm? 'Can you untie the knot?' d. A ydyw 'r tren yn dod? 'Is the train coming?'

-

Gwelaf. '(Yes,) I see [it].' Clywaf. '(Yes,) I hear [it].' Gallaf. '(Yes,) I can.' Ydyw, yn wir. 'It is, actually.'

Eventive predicates, which are semantically future in the present-tense inflection, do not occur in answers. The auxiliary gwneud is used instead. (ii)

a. A ddarlleni di'r rhan arall? Q read you-the part other 'Will you read the other part?' b. A orffenni di'r gwaithgolchi i mi? Q finish you-the washing for me 'Will you finish the washing for me?'

-

Gwnaf. I-do 'Yes (, I will).' Gwnaf. I-do - 'Yes (, I will).'

170

Alain Rouveret

55 According to Enc (1991) and Campbell (1991), the asymmetry between the present tense and the past tense in English essentially follows from a morphological property: finite verbs inflect for tense only in the past tense; in the present tense, they are morphologically specified for agreement, but not for tense. This analysis cannot be correct, since Welsh finite verbal forms are specified for tense and agreement both in the present and in the past tense. The property stated in (81) is thus an empirical characteristic which has to be stipulated. 56 The well-formedness of the sentences containing a past verbal form preceded by a negative or interrogative particle indicates that the PAST operator and NEGATION or QUESTION count as a single operator at LF. 57 This is essentially the approach adopted by Enc (1991). Enc's claim that the GENERIC operator binds the event argument of eventive verbs only in the absence of a Tense operator is not without problems, as observed by Dechaine (1993). In French, (81) does not hold, yet eventive verbs in the present tense allow a generic interpretation. I will ignore this difficulty here.

5

Pronominal enclisis in VSO languages

Ian Roberts and Ur Shlonsky

1 Introduction In this chapter we present evidence that both Semitic (Arabic and Hebrew) and Welsh clitic systems bear striking similarities to each other, and are significantly different from those of Romance or Germanic. We motivate an analysis of both systems which treats them as base-generated syntactic affixes in Agr. Hence these clitics are not in fact pronouns, that is, XPs. 1 Our central theoretical claim is that this type of clitic system is non-trivially connected to the (full or residual) VSO nature of the languages in question. Adopting and adapting a proposal for English auxiliaries in Chomsky (1993), we propose that weak/clitic pronouns 2 must check features with an Agr head with strong nominal features. However, we argue that the nature of VSO systems is such that Agr heads with strong features are largely absent. It follows that weak/ clitic pronouns cannot be licensed in a VSO system. The functional role of such pronouns - which we will argue to involve licensing pro - is then carried by the Agr heads themselves. We thus tie together two apparently unrelated properties of these languages, namely word order and the nature of the clitic system. We also explain the pervasiveness of agreement marking that these languages show; where a Romance or Germanic language has a pronoun, these languages have agreement, hence it is not a surprise to find agreeing prepositions, for example. Moreover, the apparent preference for enclisis that these languages show is a consequence, in our terms, of the fact that the apparent clitics are really affixes; enclisis thus follows from the Right-Hand Head Rule (Williams 1981b). This result is consistent with the proposal in Kayne (1995) that heads can only adjoin to the left of other heads. Put more simply, our account of Celtic and Semitic clitics is consistent with the idea that clitics are always and only on the left of their hosts and affixal heads are always on the right, as Kayne's theory predicts. Our results thus lend support to Kayne's theory, over a theory which makes no predictions about the direction of head adjunction or affixation such as the one proposed in Chomsky (1994). In section 3 we present the properties of Semitic clitics and contrast them with Romance. Here we also motivate the analysis of the clitics as base-generated syntactic affixes in Agr. In section 4 we show how the same analysis carries over 171

172

Ian Roberts and Ur Shlonsky

to Welsh. In section 5, we present our view of the connection between the clitic typology and VSO typology. Section 6 extends the discussion to look at analytic and synthetic agreement paradigms in Welsh, and the alternation between VS and SV order in Standard Arabic. We show how natural assumptions in the minimalist system can explain what appear to be three different abstract feature values for Agr. Finally, section 6.1 deals with doubling phenomena: we briefly analyse the 'echo pronouns' of Welsh, and show how this system is consistent with the general characterization of 'VSO clitics'. First, however, we present our assumptions about the extension of standard Government-Binding Case theory which has become known as Checking Theory and outline the central assumptions that we adopt. For full technical details, see Chomsky (1993: 6-19).

2 Checking Theory

The central idea is that the earlier notion of Case assignment is to be subsumed and generalized to a theory of how functional heads license lexical categories and cause the movement of lexical categories (both heads and maximal projections). The functional heads include Agr (of various kinds), T, C, D, etc. The content of these heads includes abstract features that enter into checking relations with features of lexical categories. The checking operation takes place in the checking domain of a functional head; the checking domain of a head H can be characterized for our purposes as the specifier of H or a head adjoined to H (see Chomsky 1993: 12 for a full definition). For example nominative Case assignment in a language like English is construed as a checking relation between a DP in Spec of AgrSP and AgrS. The element checked is a Case feature (or N feature, in Chomsky's terminology) associated with the functional head AgrS. The presence of a strong N feature on AgrS (in finite clauses; Chomsky suggests that this feature is associated to AgrS by head movement of finite T) causes the subject DP to raise from its VP-internal position to Spec of AgrS. Strong features are those that must be checked prior to the mapping into the phonology (Spell Out). On the other hand, where a given feature is weak, nothing requires movement in the overt syntax (although it must take place in the mapping to LF (Logical Form)). The 'delay' in movement for checking of weak features is ensured by the Procrastinate Principle, which we can formulate as follows: (1)

Movement is delayed whenever possible.

It is a property of PF (Phonetic Form) that strong features must be checked prior to Spell Out; hence, movement cannot be delayed where strong features must be checked. On the other hand, PF is insensitive to the presence of weak features, and hence movement for checking of weak features can be delayed until

Pronominal enclisis

173

after Spell Out. The effect of the Procrastinate Principle is that this kind of movement must be delayed, all other things being equal. More generally, the abstract features of functional heads are taken to be the elements that trigger movement, both of heads and of maximal projections. Economy principles ensure that movement will take place for no reason other than to check features; whether movement is overt or not is determined by the interaction of strong features with the Procrastinate Principle. The strength or weakness of the abstract features of functional heads is taken to be the sole locus of parametric variation. In VSO systems, the N-feature associated with nominative Case assignment is weak, and so the subject does not raise overtly; we will discuss this approach to the characterization of VSO typology in detail below. The interaction of Checking Theory and economy principles also guarantees the strong locality conditions that constrain head movement and A-movement. The relevant economy principle requires movement to be as short as possible (the shortest movement condition), hence it must always move a category just to the nearest potential checking position; movement to a more distant position violates Economy, leading to ungrammaticality. An important notion in this connection is that of equidistance, defined as follows. (2)

If a and /3 are in the same minimal domain, they are equidistant from 7.

The minimal domain of a head H is the smallest set of nodes contained in the maximal projection of H that are not themselves projections of H. In effect, then, the minimal domain of H includes H's specifier, complement and anything adjoined to H or to HP. If H undergoes movement to a higher head G, then the minimal domain is defined with respect to the resulting chain (H, t). This domain includes the specifier of G and anything adjoined to GP, in addition to the specifier of H and anything adjoined to HP; it does not include HP itself, since HP is a projection of H. It follows from this characterization of minimal domain combined with the definition of equidistance in (2) that the specifier of HP and the specifier of GP are equidistant positions. Movement to either of these two positions has the same status with respect to the shortest movement requirement. Consider, for example, the derivation of a simple transitive clause in English. Owing to the presence of a strong nominative feature, the subject raises to Spec of AgrSP overtly, as we mentioned above. Assuming that Spec of TP is either absent or not an A-position, this position does not count as a potential landing-site. Similarly, Spec of AgrOP is absent prior to LF (since specifier positions are created by movement and AgrO in English has a weak N-feature which therefore does not trigger overt movement). Hence, movement to Spec of AgrSP satisfies the shortest movement condition. At LF, however, the object must raise to Spec of AgrOP from its base position in VP. This movement involves crossing the base position of the subject, Spec of VP. The verb also raises to AgrO in LF, creating a

174

Ian Roberts and Ur Shlonsky

chain (V, t) with a minimal domain which includes both Spec of VP and Spec of AgrOP. Thus, Spec of AgrOP and the base position of the subject are equidistant from the base position of the object and so the object can move directly to Spec of AgrOP, 'skipping' the subject position. In this way, the notions of equidistance and minimal domain derive the pattern of crossing A-chains in transitive clauses. In addition to Case features, or N-features, Chomsky proposes that functional heads have V-features. These are Tense and Agreement features, and are associated with the functional heads which make up the clausal system. V-features can be weak or strong, triggering covert or overt verb movement respectively. In this way the well-known differences among languages regarding verb placement can be captured: English Agr and T have weak V-features, while French Agr and T (in finite clauses) have strong V-features (see in particular Emonds 1978: Pollock 1989). In VSO languages, the functional heads that make up the clausal system typically have strong V-features, as we shall see. See also Bobaljik and Carnie (this volume). In Chomsky's presentation, N-features are implicitly taken to trigger DPmovement exclusively, and V-features are taken to be the only trigger for head movement. We would like to add two provisos to these assumptions. First, we take it that it is possible in principle for D-movement to check N-features; that is, a DP may come to be licensed by movement of its head D to a higher head which bears an N-feature. This idea is the analogue in checking terms of the idea that incorporation of the head of a nominal can satisfy that nominal's Case requirement (see Baker 1988; Everett 1989; Rizzi and Roberts 1989). Second, it is clear that outside the clausal domain there may be Agr heads whose 'V-features' trigger movement of the lexical head the Agr is associated with, where that lexical head is something other than V. For example, in languages where PPs show overt agreement we are led to postulate a prepositional Agr, AgrPrep. The 'V-features' of this Agr trigger P-movement (this would be the only possibility, given the locality constraints on head movement). Given these two considerations, it would be preferable to refer to N-features as 'argument-checking features' and Vfeatures as 'predicate-checking features', where arguments are categories that receive thematic roles and predicates categories that assign them. Nevertheless, we will retain Chomsky's terminology for ease of exposition. With the above elements of Checking Theory as background, we can now turn to the analysis of word order and clitics in Semitic and Celtic.

Pronominal enclisis

175

3 Properties of Semitic clitics Our primary concern are examples such as those in (3) with the clitic in bold print. (3)

a. kaan bixayyt-ha. (he) was-sewing-3sG.F 'He was sewing it.' b. S'asaan-ha bitxayyt 1-fistyaan . . . because-3sG.F sews the-dress . . . 'because-she sews the dress c. tmunot-eha tluyot Tal ha-kir. picture-3sG.F hang on the-wall 'Her pictures hang on the wall.' d. xasavnu Tal-eha. (we) thought about-3sG.F 'We thought about-her.'

(Palestinian Arabic)

(Palestinian Arabic)

(Hebrew)

(Hebrew)

An adequate analysis of these clitics must account for the fact that they are manifested on all lexical categories (i.e. on V in (3a), on P in (3b, d) and on N in (3c)). The analysis must also explain why Semitic clitics are without exception enclitics and why they manifest no overt distinctions of Case. These three properties, and others to be presented below, sharply distinguish Semitic clitics from Romance ones, as can be immediately discerned by comparing the expressions in (3) above with the French example in (4) below. (4)

Elle l'a cousu. she it-has sewn 'She has sewn it.'

French and Romance clitics generally appear only on verbs, they manifest Case distinctions and they are either proclitics or enclitics (in French they are proclitics in all cases except positive imperatives; in Italian they are enclitic to infinitives, except optionally in negative imperatives). The French object clitic in (4) is attached to the auxiliary, the Semitic one in (3a) to the main verb; if the clitic ha in (3a) attaches to the auxiliary kaan, the result is severely ungrammatical. With some exceptions, Romance clitics are attached to the auxiliary in compound tenses. This is never the case in Semitic. Semitic clitics are invariably attached to the main verb in periphrastic constructions. In so far as the generalization underlying clitic placement in French or Italian is that the clitics are attached to the highest verbal head in the clause, that is, the auxiliary in compound tenses, the main verb in simple ones, the generalization holding for Semitic should be stated as in (5). (5)

Clitics are always attached to the closest c-commanding head.

176

Ian Roberts and Ur Shlonsky

A second difference evident from the comparison of (3) and (4) is that Romance clitics may appear to the left of their hosts (proclisis) while in Semitic they are, as noted above, without exception enclitics. In Romance, the choice between proclisis and enclisis is determined by the tense and mood of the verb. No such pattern can be found in Semitic. A third important difference is that while Romance clitics appear on verbs and auxiliaries, Semitic clitics appear on all lexical and some of the functional categories, as in (3) above or in the Palestinian Arabic paradigm in (6) below. (6)

a. Verb + Object fhimt l-nrfalme. (I) understood the-teacher b. Noun + Possessor beet 1-mTalme house the-teacher 'the teacher's house' c. Preposition + Object min l-nrfalme from the-teacher d. Complementizer + Subject ?innu l-m^alme that the-teacher e. Quantifier + DP kull l-nrfalmaat all the-teachers

fhimt-ha. (I) understood-her beet-ha her-house

min-ha from-her ?in-ha that-she kull-hin all-them

A fifth property of Semitic clitics which distinguishes them from Romance ones is that there are no clitic clusters in Semitic. The relevant case involves clitics in the double-object construction. Arabic causative verbs and a small number of non-causative verbs rather regularly alternate between a double accusative and an accusative dative complement alignment (see Mouchaweh 1986; Hoyt 1989; Hazout 1991). Consider the following paradigm from Cairene Arabic (CA, Kenstowicz and Wahba 1980). (7)

a.

b. c. d.

?il mudarris fahhim 1-dars li 1-bint. the teacher CAus-understandthe-lessonto the-girl 'The teacher explained the lesson to the girl.' ?il mudarris fahhim 1-bint 1-dars. the-teacher CAUs-understand the-girl the-lesson ?il mudarris fahhim-u li 1-bint. the-teacher CAUs-understand-it to the-girl ?il mudarris fahhim-ha 1-dars. the-teacher CAUs-understand-her the-lesson

Pronominal enclisis e. f.

177

?il mudarris fahhim-u laa-ha. the-teacher CAUs-understand-it to-her *?il mudarris fahhim-ha-u/u-ha. the-teacher CAus-understand-her-it/it-her

Examples (7a, b) illustrate the dative alternation. In the first sentence in this pair, there is a direct object and an indirect (prepositional dative) one. The second, dative-shifted sentence, illustrates the double-object variant. The second pair of sentences, (7c, d), illustrate pronominalization of one of the complements. In (7c), the theme is cliticized on the verb and in (7d) the causee, that is, the shifted object. The crucial data are found in the third pair in (7). When both complements are pronominalized, only one can show up as a clitic on the verb, the other one must find another host. In the first member of this pair, (7e), the direct object is cliticized on the verb while the indirect one appears as a clitic on the preposition. Example (If) shows that cliticizing both complements on the verb in any order is unacceptable. Put differently, only the prepositional dative construction can be the source for pronominalization of both complements. If clitic clusters are disallowed in CA, then (7e) is the only option when both objects are pronominalized because only in the prepositional dative construction is there a second host for the second clitic. A final property distinguishing Semitic clitics from Romance ones to be noted at this point is that Romance clitics have a morphological affinity with determiners. No such affinity can be observed in Semitic. To conclude this descriptive section, we list the relevant properties of Semitic clitics discussed above, noting that for each and every one of them, Romance clitics have the opposite property. (8)

Properties of Semitic clitics a. They occur on the right of their host, never on the left. b. They are always attached to the closest c-commanding head. c. They appear on all lexical categories and on certain functional ones. d. They do not manifest case distinctions. e. They never cluster, that is, there is a single clitic per host. f. They bear no morphological resemblance to nominal determiners.

3.1 Analysis Assume that Semitic clitics are Agr° elements and consider first object clitics. Suppose that, in Semitic, AgrO° may contain an affix. Thus, when the verb raises out of VP, for example over negation, as in (9), it must adjoin to AgrO° on its path upwards.

178

Ian Roberts and Ur Shlonsky

(9)

bitaxayyt-o-s.

(Palestinian Arabic)

3F.IMPERF-SeWS-3SG.M-NEG

'She does not sew it.' The strict locality on clitic attachment, that is, the fact that object clitics and Semitic clitics in general are manifested on their 'closest' host is a consequence of the Head Movement Constraint. The contents of AgrO0 must appear on the main verb in periphrastic constructions and not on the auxiliary because AgrO c-commands the main verb and not the auxiliary. One major difference between Romance and Semitic clitics thus emerges: Romance clitics are XPs at D-Structure while Semitic clitics are affixal (Agr) heads at all levels. A problem arises since object clitics lie on the outside of subject agreement. This is not obvious from the examination of (9), since third-person subject agreement on the imperfect form of the verb is represented by means of a prefix and not a suffix. The perfect (or past-tense) verbal form, however, has only suffixal agreement. In (10) we see that object agreement must occur to the right (on the outside) of subject agreement. In the imperfect, the plural morpheme is also suffixal, and once again, object agreement must occur to its right, as in (11). (10)

a. xayyat-at-o. SeW-PERF-3SG.F-3SG.M

'She sewed it. b. *xayyat-o-at. SeW-PERF-3SG.M-3SG.F

(11)

a. bitxayyt-u-ha. 2-IMP-SeW-PL-3SG.F 4

YOU(PL) sew it(FEM).'

b. *bitxayyt-ha-u 2-IMP-SeW-3SG.F-PL

This leads us to abandon the assumption that AgrS° in Semitic contains a syntactic affix. Rather than taking subject agreement morphology on a verb to constitute the affixal contents of AgrS°, we suggest that it is base-generated on the verb. Movement of V to Agr° is motivated by the need to check the appropriate features, as in Chomsky (1993). This claim is mirrored in the verbal morphology. Agr heads have different forms from the lexically incorporated subject agreement markers. The choice between lexical and syntactic affixation of subject agreement seems to be a rather low-level difference. Welsh, as we shall see in section 4, treats subject agreement as an AgrS affix, thus regularizing the pattern observed in other categories. Proceeding, let us suppose that whenever an affix appears in one of these Agr heads, a DP bearing the appropriate features appears in its Spec at some level of representation, possibly LF. Take this DP to be a referential pro, as in the schema

Pronominal enclisis

179

in (12) below, where X is a member of the class of clitic-bearing heads.

02)

AgrP DP

Ag

1

1

pro

Agr

1 1

clitic

XP X1 X

Let us further assume, with Chomsky (1993), that AgrSP, AgrOP, etc. are heuristic labels. AgrSP is the Agr category accessed by the subject while AgrO is targeted by the direct object. The association of AgrS° with subjects and of AgrO° with objects is due to the operation of derivational constraints (e.g., the need to respect Equidistance, see above, §2) and is not a consequence of some intrinsic connection between the Agr category and the nominal expression in its Spec. The content of Agr° is constituted of a particular choice of agreement or cp features, but the labels AgrSP, AgrOP etc. do not refer to different categories, but to different instantiations of AgrP. This explains the absence of Case distinctions on these Agr heads. One expects Case distinctions to show up on nominals, i.e. on XPs, and Standard Arabic indeed manifests a robust system of morphological Case, but on DPs, not on clitics. The fact that clitics appear on all lexical and some of the functional categories should now be taken to indicate that all lexical categories have Agr projections, so that a PP or a CP is actually dominated by an Agr phrase. The formal similarity between clitics and determiners in Romance suggests that they belong to one and the same category, namely D°. By a similar token, the absence of any such similarity in Semitic is just what one expects if clitics are Agr° elements and determiners belong to the category D°. Turning to the absence of clitic clusters, note that this does not follow from anything yet. The absence of both subject and object clitics showing up on a verb simply follows from the hypothesis that AgrS does not contain an affix, so that a verb base-generated with subject agreement can move through AgrO, pick up the object agreement affix and move further up to AgrS. The relevant paradigm to examine is one where a verb has more than a single object. Such a test case is provided by the double-object construction, manifested in numerous Arabic dialects. Recall the Cairene Arabic paradigm in (7). Example (70 in particular illustrates the ungrammaticality of clusters. The view that Arabic clitics are Agr heads allows us to formulate succinctly the restriction violated in this example in the following terms.

180

Ian Roberts and Ur Shlonsky

(13)

There is only a single AgrP associated with VP.3

Consideration of (7c, d) shows that the single Agr projection associated with VP can be headed by an affix corresponding to either the direct object (7c) or the indirect object (7d). We take this to be related to the absence of Case distinctions on the Agr heads. Why should (13) hold? A straightforward and principled answer is provided by Chomsky's notions of domain and equidistance discussed in section 2. Consider a possible structure where (13) does not hold, that is, where VP is dominated by two Agr projections (as in Ouhalla 1994).4 (14) AgrP^

Agr 2

V

VP Subject

VV VP 10 DO

Chomsky's proposal is that A-movement of a DP can skip the closest ccommanding Spec position if V has moved to a higher head and extended its domain. This is so since the extraction site and the intermediate Spec position are rendered equidistant from the target Spec. This system allows for objects to raise to Spec of AgrO crossing over the base subject position and for subjects to raise over objects in Spec of AgrO. Crucially, however, only one Spec can be skipped. Thus, if one object moves to Spec of Agr2 in (14) and the other to Spec of Agr1? the subject would be stranded in VP. Similarly, if one object moved to Spec of Agr2 and the subject to Spec of Agr i, the second object would be stranded. Pronominalization of both objects in Cairene Arabic requires the prepositional dative construction. One clitic shows up on the verb, indicating that the single AgrP above VP has been triggered. The other clitic is manifested on the preposition, indicating that the preposition's associated AgrP has been used and that P has raised to its respective Agr. Such a derivation is fully consistent with Chomsky's approach, since the two AgrPs lie within the domains of two distinct heads.5 To conclude, we have shown that the major properties of Semitic clitics listed in (8) can be uniformly accounted for by the assumption that these clitics are in fact

Pronominal enclisis

181

Agr° categories to which a lexical head adjoins. The argument associated with the Agr head is a pro, occupying the specifier position of the respective Agr head. In these respects, Semitic clitics are of a very different sort from those of the Romance languages. In the next section, we shall see that Arabic and Hebrew are not unique in having syntactically active Agr heads where other languages have pronouns. The Celtic languages, and in particular Welsh, manifest a similar system. 4 Properties of Welsh clitics

Traditional treatments of the Welsh personal pronouns distinguish several classes. Lewis and Pedersen (1989) distinguish independent, enclitic, suffixed, infixed, dependent and independent genitives. Both the independent and the enclitic pronouns have three separate forms each, namely, a simple form, a reduplicative form and a conjunctive one. Distinguishing also two types of dependent genitive, this gives sixteen different forms for each pronoun. Morris-Jones (1913: 27If.) distinguishes independent from dependent pronouns, each of which then falls into three further subclasses. Doing justice to this rich and complex system would go beyond the scope of our limited investigation. We therefore restrict our attention to four classes of elements that might be considered pronominal clitics. First, there is a class of infixed pronouns (the traditional term). In Modern Literary Welsh, these pronouns are enclitic to sentential particles, but they are largely absent from the spoken language. Sentence (15a) gives an example of an infixed pronoun and (15b) gives the full paradigm (D = soft mutation on a following consonant, H = aspirate mutation). (15)

a. Mi'ch gwelais (i). PRT-you saw (I). 'I saw you.' b. -m, -th + D, -i + H (m.)/-i +H (f.)/-J, -«, -ch, -/ +H/-s6

A second class is a set of historically genitive forms that appear (a) in possessive constructions, (b) as an object-agreement marker on the verbal noun and (c) as an agreement marker on the complementizer/auxiliary bod;1 (16d) gives the full paradigm (+ N indicates a nasal mutation on the following consonant): (16)

a. ei wraig his wife 'his wife' b. Mae Megan wedi ei weld, is M. after his see 'M. has seen him.' c. Dywedodd y bachgen ein bod wedi cyrraedd. said the boy our being after arrive The boy said that we had arrived.' d. fy + N, dy + D, ei + D (m.) ei (f.), ein + H, eich, eu.

182

Ian Roberts and Ur Shlonsky

Third, there are subject-agreement and prepositional-agreement markers, which bear a certain formal resemblance to pronouns. In (17) we present (a) the Modern Literary Welsh present-tense forms of the regular verb caru 'to love', (b) the forms of the Modern Welsh inflected preposition ar ('on') and (c) the suffixed pronouns of Lewis and Pedersen: (17)

a. caraf, eery, car, carwn, cerweh, carant b. arnaf, arnat, arno, ami, arnon, arnoch, arnon c. -f, -d/t, -o/-ddo, -i/-ddi, -m/n, -ch, -ynt/-ddynt.

'love' 'on'

It seems clear that the agreement forms are related to the pronouns. In their discussion of the origin of these forms Lewis and Pedersen (1989: 282) say that they derive from 'the habit of suffixing a subject pronoun, after the uncompounded verbal forms. Some of the personal endings derived from IndoEuropean which resembled to an extent the Celtic subject pronouns became of use in the new system.' Morris-Jones (1913: 333) makes essentially the same observation for the firstand second-person plural endings; here the endings are clearly derived from the pronouns ni and chwi. Similarly, Morris-Jones (p. 397) says: 'Personal pronouns forming objects of prepositions in Brit. [Brythonic] and Goidelic came to be agglutinated to the prepositions, and ultimately developed into mere inflections.' Finally, there are the echo pronouns. These are the simple enclitics of Lewis and Pedersen; their forms are (/)/, ti, o/e, hi, ni, chwi, nhw. These elements are historically nominative. They can double both verbal and prepositional agreement, as well as the ei forms: (18)

a. Clitic + verbal noun Mae Megan wedi ei weld o. is M. after his see he. 'M. has seen him.' b. Agreeing verb + subject Canon nhw. sang they c. Possessor + noun ei wraig o his wife he d. Agreeing preposition: arno fo on + 3sc him e. Complementizer Dywedodd y bachgenein bod ni wedi cyrraedd. said the boy our being us after arrive 'The boy said that we had arrived.' (Awbery 1994: 2)

Pronominal enclisis

183

In simple tenses with VSO order and no preverbal particle to host an infixed pronoun, this kind of pronoun functions as a direct object, appearing after the subject, giving, for example, Gwelodd Megan o = 'saw Megan he' 'Megan saw him' (see Morris-Jones 1913: 279). The verbal and prepositional agreement markers share the following properties with Semitic clitics: (19)

a. They occur on the right of the host. b. They attach to the nearest c-commanding head. c. They attach to more than just V, that is, P - note that the ei forms are associated with N and C. d. They do not manifest case distinctions, despite the fact that verbal agreement goes with a nominative DP and prepositional agreement with a non-nominative. e. They never cluster. f. They bear no morphological resemblance to determiners - the definite article is y, yr or V, depending on the phonological context.

The e/-forms clearly do not show property (19a), but they do show the other properties. In fact, the d-forms are not proclitic, since numerals and a few adjectives (notably hen 'old' and annwyl 'dear') can intervene between ei and a possessee. {Hen and annwyl are among the very few adjectives that can ever precede the noun they modify in Welsh.) (20)

ei hen dy 'his old house'

The e/-forms thus seem to be separate from their 'host'. Notice, though, that eiforms never 'climb' in that they are always associated with a verbal noun in a compound tense, never with an auxiliary; they associate with a range of categories (if verbal nouns are in fact non-finite verb forms, as argued by Sproat (1985) and Borsley (1993), then they associate with N, C and V); they do not manifest case distinctions, they do not resemble determiners and they do not cluster. The last point is illustrated in (21): (21)

Mae Emrys yn ei roi iddo. Is E. in his give to + 3sG 'Emrys gives it to him.'

One clitic is manifested by an e/-form; the other by prepositional agreement. If we take the e/-forms and the verbal/prepositional agreement together, then, we see a remarkable similarity with the Semitic facts. The obvious anomaly of these forms vis-a-vis the other types of Celtic and Semitic clitics concerns the placement of the ei-forms with respect to their 'hosts'.

184

Ian Roberts and Ur Shlonsky

We argue that the ez-forms and the verbal/prepositional agreement forms constitute a single agreement system. The formal differences are thus essentially a matter of suppletion. Evidence for treating e/-forms and verbal/prepositional agreement as a single system comes from three sources. First, both may appear on C, depending on the nature of C. Where C is occupied by bod, e/-forms can appear, as illustrated in (16c). Prepositional agreement appears on the prepositional complementizer i (which may be rather similar to English infinitival for); here, as elsewhere, an echo pronoun is possible: (22)

Roeddent hwy wedi clywed [iddynt ennill y gadair]. were they after hear for + 3PL win the chair They had just learned that they had won the chair.' (Hendrick 1994: 180; the minor morphological differences reflect regional variation)

The obvious interpretation of these facts is to assume that there is an Agr-position associated with C, AgrC, as proposed by Shlonsky (1994b). The fact that the Agr-element precedes the C-position in (16c) strongly suggests that AgrC is structurally higher than C; see Roberts (1994b). Second, both types of agreement are incompatible with a full non-pronominal DP following the 'host'. (23)

a. *Mae Megan wedi ei weld Emrys. is M. after, his see Emrys. b. *Canon y plant. sang.3PL the children. c. *ei wraig Emrys his wife Emrys d. *arno Emrys on + 3sG Emrys e. *Dywedodd y bachgenei bod Megan wedi cyrraedd. said the boy her being Megan after arrive

Third, as (18) shows, they are both compatible with an echo pronoun following the host. Furthermore, the echo pronouns pattern largely as in (19).8 Example (18) illustrates all of them except for the lack of clustering; this is shown in (24). (24)

Mae Emrys yn ei roi o iddo fo. is E. in his give he to + 3SG he 'Emrys gives it to him.'

We see that Welsh has two systems of clitic/agreement, elements that have properties that are strikingly similar to those of Semitic.9 Prepositional and especially verbal agreement in Celtic and clitics in Semitic have often been treated as incorporated pronouns (Anderson 1981 on Breton;

Pronominal enclisis

185

Doron 1988 on Hebrew and Irish; Hale 1987 for Irish; Fassi-Fehri 1993 on Arabic). Irish has no echo pronouns of the Welsh type, and so agreement is in complementary distribution with any nominal or pronominal element. While this kind of analysis might work for Irish (but see McCloskey and Hale (1983) for some arguments against it), it does not carry over naturally to Welsh, given the existence of echo pronouns. Moreover, it is difficult to maintain that the echo pronouns are themselves incorporated, given that they can appear as objects following a non-pronominal subject in simple VSO clauses, they can be coordinated with full DPs, and they follow a postnominal adjective in possessives. The last two facts are illustrated in (25): (25)

a. iddo fo a'i frawd to-3sG he and-his brother 'to him and his brother' b. ei gar newydd o his car new he 'his new car'

Like the Welsh echo pronouns, the existence of clitic doubling in many Semitic languages argues against the incorporation approach. The following examples illustrate clitic doubling in Palestinian Arabic. (26)

a. fhimt-ha la 1-mTalme. (I) understood-3sG.F to the-teacher 'I understood the teacher.' b. beet-ha la-l-nrfalme house-3sG.F to-the-teacher 'the teacher's house' c. min-ha la 1-nrfalme from-3sG.F to the-teacher 'from the teacher'

Furthermore, as pointed out in section 3, Kayne's (1995) approach to phrase structure makes right-adjunction of a head to another head impossible. For these empirical and theoretical reasons, we reject the incorporation approach. The analysis proposed in section 3 for Semitic clitics carries over to the Welsh agreement + ei system, if we make two provisos. First, unlike Semitic, it seems correct to treat Welsh subject agreement as an affixal head and not as a set of features on V since it patterns with the rest of the system. In a sense, Welsh is more symmetrical than Semitic here. Second, the ^/-clitics are clearly distinct from what we find in Semitic. We treat ei, like all the other clitic elements, as a basegenerated Agr. Unlike the other elements, however, ei lacks V-features. Thus ei is incompatible with V-raising. For this reason, ei cannot appear in clauses lacking an auxiliary as it will block V-raising to AgrS. 10

186

Ian Roberts and Ur Shlonsky

Similarly, the e/-forms do not trigger N-raising in possessives. With a fully nominal possessor, the order is always Possessed-Possessor, but with the ei-fovms this order is reversed: (27)

a. car John 'John's car' b. ei gar 'his car'

We account for this by saying that N-to-D raising takes place in possessives with full DPs like (27a); here car is in D (see Ritter 1988, 1995; Siloni 1991b, 1994 on N-to-D raising in Semitic constructs). In (27b), ei occupies the AgrD position and, owing to its lack of features triggering movement, car does not move to SpecDP. The parallel with what happens in clauses is clear; ei occupies AgrO and blocks movement of the verb. Given these considerations, we must say that e/-forms lack predicate features (rather than V-features, as was stated above)? since they block both N-raising and V-raising. Treating the Welsh clitic system as parallel to Semitic directly accounts for properties (19c, d, f). Properties (19a, b) depend on the feature specification of the Agr head, as we have just seen. As for the ban on clitic clusters, the account in section 2 carries over perfectly. In fact, the following general prediction emerges: (28)

A language has Agr clitics iff it has no clitic clusters.

Given the analysis presented in section 3, (28) essentially follows from Relativized Minimality since it is this principle that determines that only a single AgrP can be associated with each VP. Conversely (since the prediction is biconditional), if a language has clitic clusters this shows that the clitics are not Agr-heads, a conclusion that tells us something about Romance, Germanic and Slavonic clitics.11 Given the remarkable similarity of the Welsh and Semitic agreement or clitic systems, the set of questions we turn to now revolves around the identification of the parameter(s), the setting of which gives rise to the patterns of agreement and cliticization we have discussed. A major tenet of this work is that the similarity between Welsh and Semitic must be stated on a level more abstract than that of the phenomena themselves. Put plainly, our goal is to associate the similarity of agreement manifested by these language types with another familiar point of similarity, their word order.

5 Pronouns, checking and VSO The Celtic languages as a whole are, as is well known, fully VSO. This order is characteristic of Semitic too. Modern Standard Arabic is less of a pure VSO language than the Celtic ones, since the order SVO is also an option. The

Pronominal enclisis

187

Modern Arabic dialects to a large extent, as well as Hebrew, are best characterized as 'residual' VSO languages.12 Historically, the elements that we have analysed as base-generated Agr in these languages are derived from pronouns. This is clear from the forms of verbal and prepositional agreement in Welsh. The following Semitic pattern attests to a similar diachronic development. Examples (29a-c) show the forms, (29d) recapitulates the diachronic development. (29)

a. huwwa ?akala he (free form) ate b. ?akaltu-hu (I) ate-it c. ?akalt-u (I) ate-it d. huwwa —• hu —• u

(Classical/Standard Arabic) (Classical/Standard Arabic) (Palestinian Arabic)

It is also clear that the verbal noun was nominal in Middle Welsh (see Clack 1993) and that the ^/-clitics derive from genitive pronouns (see Lewis and Pedersen 1989: 203-7). The development of these elements into Agr-heads was not due simply to some diachronic 'weakening' process, but rather was driven by syntactic constraints connected to the nature of VSO systems. Thus the Romance clitics, although also 'weakened' forms (ultimately of the Latin demonstrative Me), are distinct in distribution, since the Romance languages are SVO. Even the northern Italian subject clitics - which are synchronically agreement heads and diachronically weakened forms of pronouns - are not suffixal. The following Veneto example (from Poletto 1993: 104) shows this; the fact that the subject precedes the verb indicates that it is not a suffix (and, given Kayne (1995), it cannot be a verbal affix): (30)

Nane el vien. John SCL comes.

Morphophonological weakening alone cannot explain the differences between Celtic/Semitic on the one hand and Romance on the other: why should the weakened pronouns develop into preverbal clitics in the latter languages but into postverbal affixes in the former? Confirmation that word order is relevant comes from northern Italian dialects. Here the subject clitics are affixal in contexts of I-to-C movement, for instance in interrogatives: (31)

Cossa galo fato? what has-he done

(Padovano) (Poletto 1993: 299)

188

Ian Roberts and Ur Shlonsky

Poletto presents morphological evidence that the inverted subject clitics are affixes (they form a single morphological and phonological unit with the verb; they have a different form from the preverbal clitics; they generally form a more complete paradigm than the preverbal clitics), and so have a morphosyntactic status quite distinct from that of non-inverted subject clitics. In fact, she proposes an analysis of these elements which treats them effectively as Agr-heads (in AgrC, which, following Shlonsky (1994b), she takes to be lower than C). We see then that, where we have inversion in an SVO language, we find 'clitics' that behave like the Celtic and Semitic ones. It seems that VS order is connected to the appearance of this kind of clitic. The basic property that derives VSO order, we assume, is that AgrS (and/or T, the precise decision on this point is not crucial for our purposes; see Chomsky 1993; Bobaljik and Carnie this volume; McCloskey this volume) has weak N-features. This has two consequences: Spec of AgrS is not a checking position at S-structure and clitic adjunction to the left of AgrS is impossible on the assumption that both kinds of movement are driven by feature checking. Cliticadjunction to the right of AgrS is impossible on general grounds (Kayne 1989a, 1995). (AgrS though, has strong V-features in VSO languages, triggering syntactic verb-raising.) Generalizing across categories, we arrive at (32): (32)

Agr does not allow S-structure checking of N-features in XSO configurations.

Suppose, following Koopman (1993), Cardinaletti and Starke (1994) and others, that atonic pronouns must check their N-features at S-structure. This idea explains why pronouns very often move overtly where full DPs do not: cf. Scandinavian object shift, Romance clitics and, arguably, Germanic and Slavonic Wackernagel clitics. Following Roberts (1993), we could say that weak or clitic pronouns cannot procrastinate since their content is exhausted by phi-features. Another way to put this, following the spirit (though not the letter) of Chomsky's (1993) account of why English auxiliaries cannot procrastinate, is to say that pronouns are bundles of phi-features in D whose only role is to identify the content of a pro elsewhere in DP (presumably in NP; see Torrego 1988; Uriagereka, 1995). Since they have no content, they cannot survive to LF and so cannot procrastinate. The referential/semantic content of a pronominal DP is borne by the pro it contains, not by the morphological features of D. On this view, the structure of a DP containing a weak pronoun is as follows. (33)

pronoun

Pronominal enclisis

189

Now, the view of VSO languages encapsulated by (32) implies that S-structure checking of weak pronouns/clitics will be impossible.13 We thus derive our central claim (see Ouhalla 1989): (34)

Weak pronouns are impossible in a VSO system.14

Instead, the functional role of checking, identifying the content of pro, falls on the only other possible element: Agr. So, in VSO systems base-generated Agr corresponds to the clitic systems of non-VSO languages. Hence the striking parallels between Welsh 'clitics' and Semitic 'clitics', and the general extreme richness and pervasiveness of agreement in these languages. It also explains why 'enclisis' is so prevalent in these languages. This is a consequence of the fact that the 'clitic' is the actual incorporation host of the lexical category, combined with the general requirement that head-to-head adjunction can only be to the left of the host head (Kayne 1989a, 1995). In this sense, the prevalence of enclisis in VSO languages follows from the lack of directionality parameters. 15 Diachronically, we take it that VSO languages have reanalysed clitics as Agr: as we have seen, this is a traditional observation for Welsh. Also, a VSO language may turn into an SVO one but retain its 'VSO'-style clitic/agreement system; this is what has happened in the case of Hebrew and non-VSO dialects of Arabic. Our suggestion, then, is that Spec of Agr cannot be filled by an overt pronoun in a VSO language. None the less, we assume that Spec of Agr is filled by pro at some level of representation. One might suggest that pro differs from overt pronouns in not requiring N- or Case-feature checking; the requirement on pro being that it occur in an A-position (for formal licensing) and enter into a Spechead agreement relation for the recovery of featural content. This in turn would imply that agreement can be distinct from N-feature-checking; an idea that is independently motivated on the basis of Romance past-participle agreement and the existence of agreeing complementizers. On this view, pro, like weak pronouns, cannot procrastinate, perhaps because formal licensing is only possible in the domain of a functional head. An alternative would be to say that pro differs from overt pronouns in that it does not raise in the syntax, that is, that it can procrastinate (assuming that pro, unlike weak/clitic pronouns, has referential content). We will not choose between these two alternatives here, but the account of echo pronouns in Welsh that we propose in the following section is compatible with the latter view but not the former. In this section, we have presented our central claim: weak/clitic pronouns are impossible in VSO systems. Hence the functional role of such pronouns identification of DP-internal pro - is taken over entirely by Agr. This explains (a) the pervasiveness of agreement in these languages, (b) the pervasiveness of apparent 'enclisis', (c) the lack of Romance- or Germanic-style weak/clitic pronouns.

190

Ian Roberts and Ur Shlonsky 6 Analytic and synthetic agreement and the position of the subject

An important distinction must be made between 'analytic' and 'synthetic' Agr in VSO languages. In Welsh, third-person singular agreement is synthetic, in that it is compatible with a following non-pronominal DP, while non-third-person singular agreement is only compatible with/?ro in Spec of Agr. This is shown in (35). (In (35a), pro may be on either side of V, depending on what turns out to be the correct position of V. See below.) (35)

a. (pro) *canodd/canon (pro) They sang.' b. canodd/*canon y plant The children sang.'

(synthetic) (analytic)

We propose the following account of the difference between analytic and synthetic Agr: (36)

a. Analytic Agr has weak N-features. b. Synthetic Agr has no N-feature.

Analytic Agr has features which are checked by DP-raising to Spec at LF in the standard way. Synthetic Agr is a kind of 'pure' agreement head, perhaps like Romance participle agreement or agreeing complementizers. Since pro does not require checking of N-features, it is the only type of DP compatible with synthetic Agr (whether pro raises to Spec of AgrS at SS or at LF), whence the ungrammaticality of the examples in (23) with the full DP following the agreement head. Proposition (36b) raises a conceptual issue having to do with the possibility of functional heads which entirely lack features of a certain type; this clearly extends the range of UG possibilities in an a priori undesirable way. In fact, it is possible to maintain that synthetic Agr has a Case feature, contrary to (36b), which is effectively checked by itself. We can do this by pushing the idea that synthetic AgrS are syntactic affixes. Let us generalize the idea, which is a consequence of the system for movement and checking put forward in Chomsky (1993), that inserting a by the generalized transformation (GT) entails checking a feature of a. Suppose that syntactic affixes are substituted into their head positions, where head substitution is equivalent to lexical insertion. Then, substitution automatically checks the Case feature of the head, since the Agr element inherently has a Case feature and Agr inherently has a Case feature. For this to work, the notion of checking domain should be regarded as derivative of the GT; that is, we must in general take targets of GT to be coextensive with checking domains. This is certainly feasible in the framework of Chomsky (1993), and clearly desirable.16 Thus we do not need to weaken UG by allowing functional heads to lack certain feature options. We also derive the general, and non-trivial, result that syntactic affixes effectively 'absorb Case'. This seems like a good result for the treatment of clitic systems generally.17

Pronominal enclisis

191

Another consequence of our approach is that we can account very simply for 'impersonal passives' that violate the Motivated Chomage Law of relational grammar (see Comrie 1977; Perlmutter and Postal 1984): (37)

a. Fe'i

gwelwyd.

(Welsh)

PRT + him see + 'PASS'

'He was seen.' b. lo hay a katuv sam ?et ha-sa^a. NEG was written there ACC the-time 'The time wasn't written there.'

(Colloquial Hebrew) (Shoshani 1980)

We treat the ending -wyd as an indefinite pronoun like English one, French on, Italian si, etc., with the morphosyntax of an agreement affix because it is a basegenerated syntactic affix in Agr. So there is nothing to say about Motivated Chomage, because the examples in (37) are not passives at all; there is an arbitrary pro, licensed by the verbal affix, in subject position. Welsh provides direct evidence that there is no movement here in that an echo pronoun is possible in object position. Contrast the other passive construction in Welsh, where this is not possible (we return to what underlies the ungrammaticality of (38b) in section 6.1 below): (38)

a. Fe'i PRT + him

gwelwyd

o.

saw +'PASS'

he

'He was seen.' b. Cafodd y ferch ei gweld*(hi). got the girl her see (she) 'The girl was seen.' Arabic differs from Welsh in that SVO orders coexist alongside VSO ones. Alongside (39a), where the verb does not agree with the postverbal subject in number, and thus patterns with the Celtic analytic forms, we find (39b), where full agreement is manifested: (39)

a. 7anna ?al sangM.SG the 'The children b. ?al ?awlaad the children 'The children

?awlaad. children sang.' 7annuu. sang-3M.PL sang.'

Example (39b), however, is not equivalent to the Celtic synthetic Agr for the simple fact that this Agr in Arabic forces the subject to appear in its Spec. Hence, it cannot lack N-features or check them by incorporation. In the framework of our proposal, this should be taken to mean that Arabic has two types of Agr, one with weak N-features - giving rise to VSO order - and one with strong N-features, much like what is found in familiar SVO languages.

192

Ian Roberts and Ur Shlonsky

Pronouns can move to Spec of Agr where Agr has a strong N-feature, while VSO order is ruled out where the subject is a pronoun: (40)

a. Hum 7annuu. they sang-3PL.M They sang.' b. *7anna hum. sang.3sG.M they

This is so because pronouns cannot procrastinate. Thus Arabic nominative pronouns must either occur preverbally, that is, when strong-N Agr is selected, or not occur at all, if the VSO-type, weak-N Agr is selected. It is not entirely correct that nominative pronouns cannot occur to the right of a verb. They are licit in postverbal position but only when the verb fully agrees with the pronoun. (41)

7annuu hum. sang.3PL.M they

The fact that the verb must fully agree with the pronoun in this order should be taken to mean that the configuration in (41) is derived by selecting a strong-N Agr, moving the pronoun to Spec of AgrS and then raising the verb higher than Spec of AgrS. 18 Interestingly, free-standing pronouns are only found in subject position in Arabic. In all other environments, only the null pronoun + Agr option is available (Syrian Arabic is apparently different; see Borsley, 1995). Put differently, only AgrS has two variants; all the other Agr-heads are pure VSO weak-N heads. It is tempting to relate this to our earlier claim that subject agreement in Arabic is not (always) an affix in AgrS but is lexically incorporated into the verb. We can thus relate the SVO option of Arabic and its absence in Welsh to the availability of lexical affixation of subject agreement in the former but not in the latter. We summarize the distinctions between Welsh AgrS, which we take to be a syntactic affix, and the Arabic Agr characteristic of SV orders in the columns below: (42)

Welsh No clustering No weak nominative pronoun No DP in Spec (obligatory VS) Syntactic affixation

Arabic Clustering Weak nominative pronoun DP in Spec of AgrS Lexical affixation

In this section we have developed an account of the analytic and synthetic agreement paradigms in terms of a simple difference in N-features. In these terms, we were able to capture the salient differences between Welsh and Arabic subject agreement very neatly. We also entertained, somewhat inconclusively, the

Pronominal enclisis

193

possibility that inverted subject clitics in Romance are like the Celtic and Semitic ones.

6.1 Welsh echo pronouns We saw above that echo pronouns can appear following any Agr-head, whether it hosts incorporation or not (i.e. whether it is a verbal/prepositional agreement affix or an ei-fovm); compare (18) with (23), repeated here: (18)

a. Clitic + verbal noun Mae Megan wedi ei weld o is M. after his see he. k M. has seen him.' b. Agreeing verb + subject Canon nhw. sang they c. Possessor + noun ei wraig o his wife he d. Agreeing preposition arno fo on + 3sG him e. Complementizer Dywedodd y bachgen ein bod ni wedi cyrraedd. said the boy our being us after arrive T h e boy said that we had arrived.' (Awbery 1994: 2)

(23)

a. *Mae Megan wedi ei weld Emrys. is M. after his see Emrys. b. *Canon y plant. Sang.3PL the children. c. *ei wraig Emrys his wife Emrys d. *arno Emrys on + 3sG Emrys e. * Dywedodd y bachgen ei bod Megan wedi cyrraedd. said the boy her being Megan after arrive

It is implausible to treat all these cases as involving 'inversion' of some kind, and so an analysis along the lines of that proposed in the previous section for (41) does not seem promising. In fact, such an analysis is impossible with the e/-forms, since we can see that the head (the verbal noun in (16a) or the possessee in (16c))

194

Ian Roberts and Ur Shlonsky

has not raised over ei. This implies that we must seek another account of echo pronouns. We claimed in section 5 that weak/clitic pronouns generally cannot procrastinate since they lack semantic properties: they have only the formal features necessary for licensing pro. In VSO languages, since Agr-heads are unavailable for S-structure checking, weak/clitic pronouns cannot exist. Implicit in this account is the idea that the internal structure of a pronominal DP in a VSO language is simpler than that of a non-VSO language, containing perhaps only pro, where in non-VSO languages the D-head is frequently realized (and perhaps, if Cardinaletti (1994) is right, other parts of the DP are realized too). The starting point for an analysis of echo pronouns is to question this conclusion. Suppose the echo pronoun is in the pronominal DP. What are the options for an analysis along these lines? We could take the echo pronoun to be the entire DP. However, this idea goes against the letter of the account of clitics in VSO languages in section 5 in that we then expect echo pronouns not to procrastinate, which they appear to do (we return to this point below), and it goes against the spirit of the account in that we now have pronominal DPs that do not contain pro (and which are not 'strong'). It is clear that echo pronouns do not move to Spec of the head they are associated with from their relative order with respect to this head. It is also clear that they do not incorporate to that head, for the same reason. We could always posit an abstract incorporation, but in the case of 'echo pronouns' in VSO clauses we would be led to posit excorporation of V: (43)

Gwelodd Megan o. saw Megan him

It seems undesirable to posit abstract incorporation if that entails also positing excorporation (whatever one's view of the possibility of excorporation in general): this suggests that something is being missed. An alternative might be to assume that echo pronouns are heads of pronominal DPs, rather like the Torrego and Uriagereka proposal for some Romance clitics (see Torrego 1988; Uriagereka 1995). This proposal still falls foul of our general analysis of clitics in VSO languages, as we would expect such a DP to be required to incorporate, which it apparently does not do. Moreover, as we pointed out in section 4, echo pronouns are like.typical VSO clitics in that they do not resemble determiners and do not show case distinctions. This disfavours analysing them as Ds. A third possibility, which appears to be the best, is to say that echo pronouns are generated in the Agr-position inside the pronominal DP (see Koopman 1993 for a similar analysis). In this way, they fit the general VSO clitic pattern, and the drawbacks just mentioned for a D-analysis do not hold. We assume, then, that echo pronouns are optional realizations of the Agr of a pronominal DP (Agr occupies a position between D and N - see note 19 and Duffield this volume). It is

Pronominal enclisis

195

unlikely that these elements play a role in licensing pro since they are optional; pro's features are thus recoverable elsewhere.19 On this view, Irish, Breton and Scots Gaelic - all of which lack echo pronouns - simply lack this expression of AgrD. In Pembrokeshire Welsh, on the other hand, these are the only pronominal forms in some contexts (see Awbery 1994). This view has an important consequence for the analysis given above. The position of echo pronouns indicates that, in fact, pronominal DPs procrastinate in (pure) VSO systems. Thus we assume that at LF the echo pronoun disappears (since all Agrs disappear), and pro raises to the specifier of its licensing head and checks N-features there. This does not substantially change any of the details of what has gone before, but it suggests that we should not stipulate that pro is licensed in a checking position at SS; pro, like any substantive element, can be licensed at LF. The proposed analysis also implies, interestingly, that even infixed pronouns have a weak feature, since echo pronouns are compatible with them: (44)

Mi'ch gwelais chwi. PRT-you saw (I) you T saw you.'

We saw in section 4 above that echo pronouns are also obligatory in transitive clauses with a simple finite verb. This suggests that pro requires morphological features for identification; where there is no overt Agr head, the echo pronoun is required. However, echo pronouns are impossible in the object position of compound passives (with the auxiliary cael; these are distinct from the 'impersonal passives' in (37)), although the preverbal ei form is obligatory. (45)

Cadoffy ferchei gweld(*hi). got the girl her see (she) 'The girl was seen.'

In terms of the analysis being proposed here, this is a consequence of the fact that the entire object DP has moved, as is standard for objects in passives. Since procrastination is always possible in Welsh, it is not clear why the object should move in passives. It seems that we have to appeal to the Extended Projection Principle (EPP). This is interesting because the EPP does not at first sight appear to hold in VSO languages (see McCloskey this volume). PF/z-movement behaves similarly to passive. An echo pronoun cannot appear at the extraction site (in cases of short extraction; longer extractions involve various kinds of resumptive pronoun strategies - see Hendrick 1988): (46)

a. *Pwy who b. *Pwy who

welodd y ferch o? saw the girl he mae'r ferch wedi ei weld o? is-the girl after AgrO see he

196

Ian Roberts and Ur Shlonsky

These facts are straightforwardly predicted by our analysis. The entire object DP is moved and so no echo pronoun can appear in object position.

7 Conclusion

The central point of this chapter has been to tie together two apparently disparate, but strikingly similar, aspects of the syntax of the Celtic and Semitic languages. We have argued that both VSO typology and the prevalence of enclisis in these languages are a consequence of the property in (28). This analysis entails the idea that weak pronouns are essentially a means of identifying pro, an idea which merits further investigation and development. More generally, we see that two salient common features of these languages can be subsumed under a single generalization, and moreover a generalization which fits into the format for parametric variation proposed in Chomsky (1993). It has often been observed that the Celtic and Semitic languages show striking similarities in their syntax (their phonologies, morphologies and lexicons are quite different); these similarities cannot be attributed to contact or to a genetic relationship, 20 and so they must reflect some accidentally shared common property or properties. If the range of parametric variation allowed by UG is very limited, a desirable assumption on general grounds, then we should expect to find cases of historically and geographically unrelated languages which apparently have many syntactic features in common. Indeed, as our knowledge of the possible dimensions of syntactic variation increases, we expect to find cases of 'parametric isomorphy', where several major parameters have the same value in given pairs of languages. The comparison of the Celtic and Semitic languages is a very fruitful area in which possibilities of this type can be explored and developed, as we hope our work has shown.

Notes This chapter developed out of separate work (Roberts 1994c; Shlonsky 1994a, forthcoming). A preliminary version was presented at the seventeenth GLOW colloquium, Vienna, April 1994. We are grateful to the audience for their questions and comments. We would also like to thank our colleagues in Geneva and Bangor for discussion. Special thanks to Bob Borsley, Guglielmo Cinque, Alain Rouveret and the participants at the third Welsh Syntax Seminar. There is good evidence for the distinction between weak and clitic pronouns in many languages - see Cardinaletti and Starke (1994). However, for the purposes of this chapter, the two can be conflated and both terms are used interchangeably. We assume that AgrSP is not associated with VP but with TP.

Pronominal enclisis

197

4 Unlike Larson (1988), however, the indirect object in (14) is mapped onto a hierarchically higher position than the direct object. See Belletti and Shlonsky (1995). 5 We put aside the question of the derivation of (7b), namely the double-object construction within the minimalist framework. 6 The different third-person forms correspond to historical genitives and accusatives respectively. In all the other persons, the case forms were syncretic. Synchronically, there is no reason to consider these as anything other than morphologically conditioned variants. There are also other variants, such as ~w (3sg.) after the preposition/complementizer i 'to/for'. 7 It is unclear whether bod here is a complementizer or an auxiliary - a problem that arises elsewhere in Welsh. Hendrick (1994) argues that this is a kind of Aux-to-C construction (see also Rouveret, this volume) in which case we can maintain that the clitic is attached either to the moved I or to C. 8 When an echo pronoun is a direct object of a finite verb, this is not the case. See the discussion surrounding (43). 9 A potential case of clustering of echo pronouns arises in simple VSO clauses with a pronominal subject and object: Gweloddhi o 'Saw she he' ('She saw him'). However, the echo pronouns are in different places here, as section 6 will reveal. 10 In these cases, there can be an infixed pronoun in the relevant register, as (15) shows. As we saw above (viz. the discussion following (18)), if the object is pronominal and there is no infixed pronoun, then the echo pronoun is obligatory. Sometimes the infixed pronoun only shows up as mutation leading to some revealing minimal pairs: (i) (ii)

Mi welodd (hi). 'She saw.' Mi gwelodd (hi). 'He/she saw her.'

The lack of soft mutation in (ii) (whose phonological reflex is the presence of initial /g/) indicates that there is an infixed pronoun between the particle and the verb. Hence in (i) the optional echo pronoun is the subject, doubling the agreement. In (ii), there is an infixed pronoun and the optional echo pronoun is interpreted as doubling this element; the subject is null. It is significant that in (i), where there is neither an infixed nor an echo pronoun, no object is recoverable. This shows that Welsh is not a null-object language; that is, pro cannot occur in object position in this kind of context. This also shows that the generalization in Hendrick (1988) that echo pronouns only appear where pro is licensed is not entirely correct. 11 The obvious conclusion to draw as regards the Romance, Germanic and Slavonic clitic systems is that the clitics are not in Agr. Two possibilities suggest themselves for the treatment of these systems. Either the clitics move to Agr, perhaps by some process combining DP-movement and D-movement of the kind originally proposed in Sportiche (1990), and elaborated in differing ways by Belletti (1993), Cardinaletti and Starke (1994) and Rizzi (1993), or they are base-generated in special Voice Phrases of the kind proposed in Sportiche (1992) with movement of pro to Spec/Voice for checking purposes. To choose between these possibilities, and to consider even a fraction of the relevant data from these languages, would take us too far afield.

198

Ian Roberts and Ur Shlonsky

12 See Shlonsky (forthcoming) for further discussion. 13 We can allow for second-position clitics like Welsh infixed pronouns and Berber clitics (Ouhalla 1989), if we assume that the 'higher' head that these elements attach to has strong features; alternatively, we could treat this position like VoiceP in the sense of Sportiche (1992) and allow only pro to move into its Spec for reasons discussed in connection with synthetic Agr below. Either way, the second-position case clearly falls outside the generalizations we are making here; see also section 6 on the interaction of infixed and echo pronouns in Welsh. 14 Since Arabic has both a VSO system and an SVO one, (32) holds only of the former. The free pronoun in (29a) thus belongs to the SVO system where it can be checked in Spec/AgrS. The occurrence of free-standing (weak) pronouns in Modern Hebrew is similarly related to the fact that Hebrew is predominantly SVO, in particular in the clausal system. 15 Although we adopt the idea behind Chomsky's account of why English auxiliaries raise, we execute it slightly differently. Chomsky proposes that English auxiliaries raise overtly because, despite the fact that English AgrS has only weak features, the auxiliaries cannot survive to LF and so their features must be checked overtly. Effectively, then, Procrastinate is overridden by the fact that the auxiliaries cannot be present at LF. Our approach to checking the features of weak pronouns implies, however, that even if there are no strong features Procrastinate cannot be overridden. Although a stronger approach, our account creates the empirical problem that there can be no raised auxiliaries in English. Notice, however, that Chomsky assumes that modals and do are not base-generated in V, and so arguably do not move (or do not move in the same fashion as, for example, French main verbs, and hence arguably do not move for the same reasons). Moreover, Chomsky's account runs into difficulties with infinitival have-be raising in French (see Pollock 1989), and with the lack of a distinction between auxiliaries and main verbs in the mainland Scandinavian languages (Vikner 1994). It is thus probably the case that Chomsky's account of why English auxiliaries raise should be abandoned on both empirical and conceptual grounds. As an alternative, we suggest that have and be raise overtly because they contain a richer specification of agreement than other English verbs, and the agreement features must be checked before LF. Effectively, then, we treat the auxiliaries like pronouns. 16 It is not feasible in any straightforward way in the rather different framework of Chomsky (1994), however. 17 One could also interpret Sportiche's (1992) doubly filled Voice filter in these terms. The doubly filled Voice filter prevents both a head and a specifier from being phonologically realized. This kind of ban on the co-presence of a head and a specifier can be understood if the head is a syntactic affix which obligatorily checks the relevant feature of the category. 18 This construction is then comparable to cases of subject-clitic inversion in French or in northern Italian dialects (see (31)). The similarity with subject-clitic inversion is all the more telling since (40) is ungrammatical when the subject is not a pronoun but a full DP. The contrast between (40) and (i) below is thus reminiscent of the familiar French contrast in (ii).

Pronominal enclisis (i)

(ii)

199

*7annuu ?al ?awlaad. sang.3pL.M the children T h e children sang.' a. Chantent-ils? sing. 3PL-they 'Do they sing?' b. *Chantent les enfants? sing.3PL- the children 'Do the children sing?'

19 The structure of a DP containing an echo pronoun is thus

pronoun

pro

20 A genetic relationship may exist between Indo-European and Afro-Asiatic, but this relationship could not account for the syntactic similarities given that VSO appears to be a Celtic innovation within Indo-European.

Aspect, agreement and measure phrases in Scottish Gaelic David Adger

1 Introduction The goal of this chapter1 is to defend an extended version of the Visibility Criterion (Chomsky 1986a) whereby the link between Theta Theory and Case Theory is generalized. We will argue that the Case-theoretic requirement in the formulation of Visibility should be replaced with a requirement of co-indexation with a functional head; the relevant functional heads being Agreement (Agr) and Tense (T). We will argue that Tense involves two functional heads which are composed into a single chain via the mechanism of selection. This allows the incorporation of aspectual information into the licensing condition. By aspectual we mean the indirect relationship between the utterance time and the time of the event denoted by the verb. Scottish Gaelic (SG) is particularly relevant to both the extended version of Visibility and to the composition of tense information into a single chain because of two factors: firstly, co-indexation with Agr in SG correlates with word order; and secondly, composition of tense information is transparently reflected in the morphological form of tense and aspect particles. Given these two factors, the extended version of Visibility predicts that certain word orders in SG will be forced or ruled out. This phenomenon is particularly clear with respect to the syntax of measure phrases in the language, which we show to be sensitive to aspectual information.

2 Generalized Visibility Typically, argument NPs have to be licensed in two ways: they must be licensed by Theta Theory and they must be licensed by Case. The link between these is made explicit in the Visibility Condition, credited to Aoun by Chomsky (1986a): (1)

Visibility An NP chain can be interpreted as an argument iff it has structural Case. 200

Aspect, agreement, measure phrases

201

This condition can be read as a licensing condition, given that an NP chain must be licensed at LF in order to receive an interpretation: (2)

Visibility (revised) An NP chain can only be interpreted if it is Theta-Theory licensed and Case-Theory licensed.

Quasi-Arguments (see Chomsky 1981) generally seem to engender problems for this conjunctive formulation of Visibility, since they are theta-marked but do not seem to need Case. We can illustrate this with the behaviour of measure phrases. These appear to be licensed by argument structure, since they are obligatory: (3)

(4)

a. b. c. a. b. c.

Anson weighs 70 kilos. The conference lasted two weeks. The book cost a dollar. * Anson weighs. T h e conference lasted. T h e book cost.

and they are NPs in as much as they are projections of N. Note, however, that they seem to be Case-resistant, so they can never be marked by the inserted Casemarking preposition of in gerundive nominals. Thus compare: (5) (6)

Anson's constant devouring of cakes. *Anson's constant weighing of 70 kilos.

Further evidence comes from the behaviour of measure phrases in agreement contexts. Chomsky (1993) argues that Case-checking is a reflex of being in a Spec-head relationship with an Agreement head. In an earlier paper (Adger 1994b) we showed that measure phrases never raise to this position; we review some of this evidence below. Given these facts, the question arises as to how measure phrases escape Visibility. Two options are possible: either they are only theta-licensed, and need no Caselicensing to count as legitimate LF objects or they are licensed in some other fashion. The former option involves showing that measure phrases are exceptions to Visibility for some principled reason, or that Visibility is in some way falsified by the behaviour of these elements. The latter option involves generalizing Visibility so that it takes into account other modes of licensing, rather than just Case Theory. It is this latter option we will defend here. Consider the first option. To argue that measure phrases are exceptions to Visibility we could show that they project only to NP rather than DP and provide evidence that abstract Case is assigned only to DPs, reformulating Visibility minimally to take account of this. Measure phrases do in fact only allow a restricted set of determiners, as we show below (see the examples in (36)). However, it would seem theoretically more attractive to allow all nominals to

202

David Adger

project the full functional structure associated with them and to rule out certain projections to D for independent reasons. Also allowing certain verbs to select NP rather than DP seems to place too great a burden on the theory of selection since it allows selection for functional as well as lexical categories (see Grimshaw 1991 for discussion and for a theory that rules this out in principle). Discarding this option then, we would like to show that measure phrases are licensed in some other way. We provided an argument to this effect in Adger (1994b), where we showed that measure phrases never raise to the specifier of AgrP to receive structural Case. We argued that they are instead unselectively bound by the Tense head of the sentence, which also binds the event argument of the verb (Higginbotham 1985). This has empirical support in that it predicts that measure phrases are not extractable from weak islands and cannot occur with quantificational determiners. We review how these predictions follow below. This argument actually contains a hidden assumption: that is that measure phrases must be licensed in some other way than by Case Theory. It therefore implicitly assumes a disjunctive version of Visibility, along the lines of: (7)

Visibility An NP chain can only be interpreted if it is Theta-Theory licensed and either Case-licensed or co-indexed with Tense.

This formulation of Visibility resolves the problem engendered by measure phrases, but we would like to eliminate the disjunction in the consequent since it undermines the generalization. Consider what 'Case-licensed' means. The type of Case we are interested in here is structural. Chomsky (1993) has proposed that structural Case assignment for subjects and objects is unified under specifier-head agreement (see Mahajan 1990 for empirical argument to this effect). Essentially, the idea is that an NP in the specifier of an Agreement head at LF will count as structurally Case-licensed. Specifier-head agreement is a standard co-indexation relationship, which allows us to reformulate Visibility as: (8)

Visibility An NP chain can only be interpreted if it is Theta-Theory licensed and either co-indexed with Agr or co-indexed with Tense.

We can now eliminate the disjunction in this formulation of Visibility by defining a notion of F-licensing:2 (9)

F-licensing An NP chain is F-licensed if it is co-indexed with a functional head.

Aspect, agreement, measure phrases

203

We then define a generalized version of Visibility as: (10)

Generalized Visibility An NP chain can only be interpreted if it is Theta-Theory licensed and F-licensed.

So far the motivation for this formulation of Visibility has been to provide a missing step in an argument about the licensing of measure phrases. The rest of this chapter is devoted to showing that this condition also explains some peculiar further facts about the licensing of measure phrases.

3 Licensing measure phrases in Scottish Gaelic 3.1 Fronted Object Phrases Scottish Gaelic is a VSO word order language related to Modern Irish (most closely to the Ulster dialects) spoken by about 80,000 speakers, mainly on the west coast and isles of Scotland. We will assume that the underlying order is SVO and that the VSO order is derived via V-raising to T or some higher functional head (see Sproat 1985). We therefore expect to see SVO orders where this raising has not taken place; for example, in infinitival constructions, or where an auxiliary verb realizes tense. This is the case in the progressive construction below. (I will gloss aspectual particles as SIM (simple), PER (perfect) and PRO (prospective). The semantics of these particles is discussed in detail in section 4.3): (11)

Tha

mi a'

be.PRES I

bualadh a'

SIM hit.VN

chait.

DET.GEN Cat-GEN

T am striking the cat.' Here, an auxiliary appears in the initial Tense slot and the theta-role-assigning verb is nominalized and occurs as the complement of an aspectual particle a\ (See Cram 1981 for detailed discussion of this nominalization in SG. The form used here is traditionally termed a 'verbal noun' (hence VN in the glosses).) The complement of the verb occurs postverbally and is marked with the genitive Case when it is definite.3 However, there are also cases where we find an OV order and where the VO order is ill formed: (12)

a. Tha mi air an cat be.PRES I PER the.coM cat.coM 'I have struck the cat.' b. *Tha mi air a bhualadh be.PRES I PER PRT close.vN 'I have struck the cat.'

a bhualadh. PRT close.vN an cat. the.coM cat.coM

204

David Adger

In these examples the object is obligatorily preposed, a particle a appears and the object is marked with the common case (the case assigned to canonical subjects and objects in non-periphrastic constructions). We will term the part of this construction from the object an cat to the nominalized V bhualadh a Fronted Object Phrase (FOP). This same construction is found as the complement of modals, attitudinal predicates, certain adjectives and in a passive construction. In each case the fronting is obligatory: (13)

a. Feumaidh mi [Daibhidh a bhualadh]. must I David PRTStrike.vN 'I must hit David.' b. 's toigh learn [Daibhidh a bhualadh]. COP liking with.lsG David PRT strike.VN 'I like hitting David.' c. Tha e doirbh [Daibhidh a bhualadh]. be.PRES it difficult David PRT strike.vN 'It is hard to hit David.' d. Chaidh [Daibhidh a bhualadh]. go.PAST David PRT strike.vN 'David was struck.'

There is evidence that FOP is a maximal phrasal constituent since it may rightnode-raise, cleft and pseudo-cleft (we give only the adjectival constructions here): (14)

(15)

(16)

's e a' cheist sin a fhreagairt a tha doirbh. it's that question PRT answer.vN COMP be.PRES difficult 'It's answering that question that's difficult.' 's etha doirbh ach a'cheist sin a fhreagairt. it's be.PRES difficult but that question PRT answer.vN 'What's difficult is to answer that question.' Tha e doirbh ach tha e math a'cheist sin a fhreagairt. be.PRES it difficult but be.PRES it good that question PRT answer.vN 'It is hard, but it is good, to answer that question.'

We argued in Adger (1994b) that the relationship between the preposed DP and the particle a is one of agreement. This claim was based on the behaviour of subject agreement in SG. We provide here an argument to the same effect based on the prepositional agreement construction in SG.

3.2 Fronted Object Phrase as AgrP It is well known that the Celtic languages display a predicate-argument agreement phenomenon whereby there is a complementarity between overt agreement

Aspect, agreement, measure phrases

205

and overt arguments. 4 The following paradigm for objects of prepositions is illustrative: (17)

a. 's toigh leam/leat/leis/leatha/leinn/leibh/leotha coffaidh. COP liking with.lsG/2sG/3sG.M/3sG.F/lPL/2pL/3PL coffee 'I/you/he/she/we/you/they like coffee.' b. 's toigh le Mairi coffee. COP liking with Mary coffee c. *'s toigh leatha Mairi coffaidh. COP liking with.3SG.F Mary coffee d. *'s toigh leatha i coffaidh. COP liking with.3sG.F she coffee

The generalization we would like to make here is the following. (For further defence of this generalization, its application to agreement in Celtic generally and how it derives from Checking Theory (Chomsky 1993), see Adger 1994a.) (18)

The phi-feature set of the agreeing element is the complement set of the phi-feature set of the argument that is agreed with.

The phi-feature set referred to here is {person, number, gender}. We can assume that the phi-features of X are determined by the morphological paradigm of X. This has obvious advantages in terms of learnability over simply stipulating phifeature specifications as part of UG. We can reify this generalization by appealing to the internal structure of the functional head Agr in SG. Whereas most languages appear to require or allow the specification of agreement features on both the predicate and the argument, SG allows each feature to appear only once. Thus Agr in SG differs from Agr in other languages in that it has slots for phi-features which can only be specified once. Further specification leads to ill-formedness. Nouns in SG are therefore specified for the phi-features number and gender since they inflect for only these categories. Pronouns are specified for this same set of features plus person. Under this view pro would have a null phi-feature set, since it has no morphological paradigm. The generalization in (18) predicts, then, that pro will show up with full agreement, and that pronouns will show up with no agreement.5 It also predicts that overt DPs will show up with agreement for only person features, but since overt DPs are always third person we expect no variation in inflection. This generalization seems to capture perfectly the situation in the above SG prepositional paradigms. If non-agreeing forms of these prepositions are not morphologically available, then pronouns will be unable to co-occur with them. In contrast, pro will force full agreement and DPs will show an uninflecting form reflecting the generalization in (18). The structures with pro are as follows:

206

David Adger

(19)

a. 's toigh leam pro coffaidh. COP liking with.lsG pro coffee 4 1 like coffee.' b. *'s toigh le pro coffaidh. COP liking with.3 pro coffee 'I like coffee.'

Here (b) violates (18), since le is specified for person features, but pro is specified as having a null phi-feature set. Overt DPs also follow (18): (20)

a. *'s toigh leatha Mairi coffaidh. COP liking with.3sG.F Mary coffee b. 's toigh le Mairi coffaidh. COP liking with Mary coffee 'Mary likes coffee.'

Here (a) violates (18), since leatha is specified for person, number and gender, but Mairi is specified for number and gender. Example (b), in contrast, conforms to (18) since le contains just person features. Finally, (18) explains the ill-formedness of pronominals: (21)

a. *'s toigh leatha i coffaidh. COP liking with.3sG.F she coffee b. *'s toigh le i coffaidh. COP liking with.3 she coffee

Both are ruled out since pronouns are specified for person, number and gender, but both le and leatha are specified for at least one of these features. Now note that the generalization in (18) predicts the behaviour of the FOP in precisely the same way: (22)

a. Feumaidh Daibhidh mo/do/a/a/ar/ur/am b(h)ualadh. must David 1SG/2SG/3SG.M/3SG.F/1PL/2PL/3PL/strike.VN 'David must hit me/you/him/her/us/you/them.' b. Feumaidh Daibhidh am balach a bhualadh. must David the boy PRT strike.VN 'David must hit the boy.' c. *Feumaidh Daibhidh na balaich am bhualadh. must David the boys PRT strike.VN 'David must hit the boys.' d. *Feumaidh Daibhidh tu a bhualadh. must David you PRT strike.vN 'David must hit you.'

Aspect, agreement, measure phrases

207

The empty object pro results in a marker that inflects for person, number and gender; an overt DP object results in a marker that inflects for only person (so a is analogous to le) so overt DPs cannot occur with agreeing forms and pronominals are ill formed. The only difference between the paradigms is that the particle a serves a dual purpose as an Agr-head with a full phi-feature set and as an Agr-head with only person features. It might be possible to derive this stipulation from a deeper understanding of the notion of default features (see Andrews 1990). Space precludes discussion of this here (see Adger 1994a: ch. 3) so we will leave this as a stipulation. Given this, the following structure is licensed: (23)

Feumaidh Daibhidh pro a bhualadh. must David pro PRT strike.VN 'David must hit him.'

The relationship between a and the preposed DP is governed by the generalization about agreement in (18), which provides motivation that a is Agr, the preposed DP is in its specifier and FOP is AgrP. There is, however, another way of interpreting these data. We could assume that a is agreement but that it is merely a feature on V and does not project syntactically. The preposing of the object is then either movement to the specifier of VP or adjunction to VP and FOP is of the category VP. However, there is strong empirical evidence that such an interpretation is incorrect. In SG there are two clefting particles: 's e and 's arm. The former of these typically clefts DP or CP (non-predicates) while the latter clefts predicates such as AdvP or AspP: (24)

a. 's e/*'s ann am ministear a tha mi a' ceilidh a-nochd. it's the minister that be.PRS I SIM visit.vN tonight 'It's the minister that I'm visiting tonight.' b. 's e/*'s ann gun do bhuail thu an cat a tha it's that PAST strike.PAST you the cat that be.PRES mi a' ciallachadh. I SIM meaning 'It's that you struck that cat that I mean.' c. *'se/'s ann a-nochd a tha mi a' ceilidh ministear. it's tonight that be.PRES I SIM visit.vN minister 'It's tonight that I'm visiting a minister.' d. *'se/'s ann a' ceilidh ministear a tha mi a-nochd. it's SIM visit.vN minister that be.PRES I tonight 'It's visiting a minister that I am tonight.'

It is possible to cleft FOP, as we saw earlier, and in these cases we get 's e:

208

David Adger

(25)

's e/*'s ann a' cheist sin a fhreagairt a tha doirbh. it's that question PRT answer.vN COMP be.PRES difficult 'It's answering that question that's difficult.'

However, if we cleft a bare verbal noun with no agreement marker then we get 's ann, which is a clear indication that FOP differs categorially in some way from VP: 6 (26)

*'s e/'s ann falbh a tha doirbh. it's leave.vN that be.PRES difficult 'It's leaving that is difficult.'

Under the agreement as a feature of V story there is no easy explanation for this contrast, whereas if Agr projects syntactically then we have a clear categorial difference that we can appeal to. The structure of FOP is then:7 (27)

a. Feumaidh Daibhidh [AGRP prot [AGR> mo [bhualadh tj\]]. must David ISG strike.VN 'David must hit me.' b. Feumaidh Daibhidh [AGRP [ a m balach]/ [AGR' a [bhualadh /,]]]. must David the boy Agr strike.vN 'David must hit the boy.'

3.3 Measure phrases and Fronted Object Phrase Turning to measure phrases, we note that these never raise into the SpecAgrP position: (28)

(29)

a. Feumaidh a' cho-labhairt mairsinn seachdainn. must the conference last.vN week 'The conference must last a week.' b. *Feumaidh a' cho-labhairt seachdainn a mhairsinn. must the conference a week AGR last.VN a. Feumaidh a' cho-labhairt cosg tri mile not. must the conference cost three thousand pounds 'The conference has to cost 3,000 pounds.' b. *Feumaidha' cho-labhairt tri mile not a chosg. must the conference three thousand pounds AGR cost.vN

In these examples with modals the measure phrase remains in its base position and preposed measure phrases are ungrammatical. In the following examples the progressive construction which allows VO order is well formed but the perfective construction which forces OV order is ungrammatical.

Aspect, agreement, measure phrases

209

(30)

a. Tha a' cho-labhairt a' mairsinn seachdainn. be.PREs the conference SIM last.VN week T h e conference is lasting a week.' b. *Tha a' cho-labhairt air seachdainn a mhairsinn. be.PREs the conference PER a week AGR last.VN T h e conference has lasted a week.' (31) a. Tha a' cho-labhairt a' cosg tri mile not. be.PREs the conference SIM cost three thousand pounds T h e conference is costing 3,000 pounds.' b. T h a a' cho-labhairt air tri mile not a chosg. be.PREs the conference PER three thousand pounds AGRO COSLVN T h e conference has cost 3,000 pounds.' Under the account we argued for in Adger (1994b), the measure phrase in the grammatical instances is part of a T(ense)-chain (Gueron and Hoekstra 1988, Bennis and Hoekstra 1989, Gueron 1992). T-chains are formed by the binding of an event variable in the VP by a temporal operator, thus temporally anchoring the event denoted by the verb: (32)

Op,...V(e ; )

We argued that measure phrases are syntactically part of a T-chain and are therefore mapped denotationally into events. A corollary of this is that measure phrases cannot denote individuals, which are participants in events. This proposal immediately accounts for the well-known fact that measure phrases are not extractable across weak islands (Rizzi 1990): (33) (34)

*What/ do you wonder whether Anson weighs tt *What/ do you regret that the book costs f,

These data are explained since there is substantial agreement in the literature that only individual-denoting elements may extract from weak islands (see Cinque 1990; Dobrovie-Sorin 1990; Rizzi 1990; Frampton 1991; Manzini 1992a; Szabolcsi and Zwarts 1993 among others).8 The extraction data discussed in the above references are replicated in SG: (35)

a. De bha thu ag aicheadh gun do bhuail Daibhidh? what be.PAST you SIM deny.vN COMP PAST-strike David 'What did you deny that David struck?' b. *De bha thu ag aicheadh gun do chosg an leabhar? what be.PAST you SIM deny.vN COMP PAST-cost the book 'What did you deny that the book cost?'

Here the weak island is induced by the factive verb aicheadh, and extraction of the measure phrase is much worse than extraction of a canonical argument.

210

David Adger

Given that the formation of a Tense chain is contingent on the ability of Tense to govern the measure phrase, we expect that Tense chains are unable to be formed when we have a closer governor. This is the case when the measure phrase has a quantificational determiner: (36)

a. b. c. d.

The conference lasted many weeks. The conference lasted three weeks. T h e conference lasted every week. T h e conference lasted most weeks.

Many and three here are cardinal determiners while every and most are quantificational (see Milsark 1977; Partee, Bach and Kratzer 1987; and references therein). Only the latter create a Minimality barrier by counting as closer governors. This means that (c) and (d) are ruled out because no Tense chain can be formed but also the measure phrase cannot raise to SpecAgrP for Case. Visibility in its generalized form is then violated. Again, these data are replicated in SG: (37)

a. Tha a' cho-labhairt a' mairsinn grunn seachdainnean. be.PRES the conference SIM last many weeks T h e conference lasted many weeks.' b. T h a a' cho-labhairt a' mairsinn a h-uile seachdainn. be.PRES the conference SIM last all/every week T h e conference lasted every week.'

3.4 A puzzle We noted above that preposing of measure phrases into the Spec AgrP position is ungrammatical in SG, but that they can remain in base position when the AgrP is a complement of a modal or of progressive aspect. We also noted that preposing of measure phrases is ill formed in the perfective construction. Here, however, an interesting difference arises: the measure phrase is also ungrammatical in base position in the perfective construction. Thus: (38)

a. T h a a' cho-labhairt air seachdainn a be.PRES the conference PER a week AGR T h e conference has lasted a week.' b. T h a a' cho-labhairt air a mhairsinn be.PRES the conference PER AGR last.vN T h e conference has lasted a week.'

mhairsinn. last.vN seachainn. week

That such an aspectually sensitive phenomenon occurs seems to confirm that measure phrases are in some way licensed by being incorporated into a Tense chain, but obviously the notion of Tense chain is not articulated enough to predict this effect. The rest of this chapter is devoted to refining the notion of Tense chain

Aspect, agreement, measure phrases

211

on independent grounds. This independent refinement will then be shown to account for this puzzle.

4 Aspectual chains In this section we will consider whether the notion of Tense chain can be made more articulate. We will argue that Tense chains are actually composed of smaller aspectual chains that instantiate the relationship between Reichenbachian temporal reference points and morphosyntactic structure. We build on the work of Giorgi and Pianesi (1992), Drijkoningen and Rutten (1991), Stowell (1992), Hornstein (1990), and Zagona (1990).

4.1 Lexical specification, selection and indexation We will follow much recent work (Williams 1981a; Higginbotham 1985; Zwarts 1991, 1992) and assume that lexical categories have both an argument structure (Grimshaw 1990) (or theta-grid) and a special distinguished argument that acts as a syntactically accessible variable and specifies the denotational type of the category. We will term elements of the former theta-arguments and we will term the latter the denotational argument. We represent this type of information enclosed in angled brackets, with the denotational argument to the left and the thetaarguments, structured as a nested list (following Grimshaw 1990), to the right. Thus the lexical specification of a verb like kiss has the following structure: (39)

kiss

Here e specifies that kiss denotes an eventuality, and (a(b)) specify that there are two major participants. The notion of theta-marking relevant for the theta-criterion involves matching elements of the argument structure with XP sisters of the theta-marking head. We will follow Zwarts (1992) and represent this as co-indexation between the appropriate theta-argument and the denotational argument of the complement XP. Thus kiss Anson involves the structure: (40)

kiss

Anson

where x ; is the denotational argument of the head of the NP Anson, which has a null-argument structure. We will refer to this type of indexing as 'selection indexing'. Note that this type of indexing involves the denotational argument of the N-head, intervening functional categories are irrelevant. But it is well known that semantically N determines the range of the dominating determiner or quantifier,

212

David Adger

thus we expect there to be another type of indexing where the operator associated with the determiner or quantifier also binds the denotational argument of N. We will refer to this type of indexing as 'binding indexing' and represent it as superindexation. Thus: (41)

kiss

every man

Selection indices are established at whatever point in the derivation lexical insertion takes place (DS in conventional theories and on application of the generalized transformation GT in Chomsky 1993), while binding indices are established atLF.

4.2 Morphosyntactic tense and interpretation

Reichenbach (1947) provides an analysis of the tense/aspect system of English which uses three temporal reference points: the speech time (S), the event time (E) and a reference time relating the two (R). Reichenbach argues that R is implicated in the analysis of all the tenses. If R precedes S then some variety of the past tense is involved; if R and S are contemporaneous then we have a present-tense variety and if R follows S then we have a variety of the future. Which particular variety is involved depends on the relationship between R and E: E preceding R results in a perfect; E following R results in a prospective and E contemporaneous with R results in a simple tense. We provide some examples below (>, <, and = can be glossed as 'follows', 'precedes' and 'is contemporaneous with', respectively):9 (42)

a. b. c. d.

Anson Anson Anson Anson

sings (S = R, E = R). sang (S > R, E = R). has sung (S = R, E < R). had sung (S > R, E < R).

Recently, a number of authors have argued that this system of relationships is directly instantiated in the morphosyntax. Giorgi and Pianesi (1992), for example, argue that past-participial morphology (the functional head Asp) in Italian represents the E < R relation while the finite tense (the functional head T) represents the relation between S and R (see also Drijkoningen and Rutten 1991; Stowell 1992). Thus: (43)

avevo

mangiato.

have.PAST eat.PPART

S> R E< R 'I had eaten.'

Aspect, agreement, measure phrases

213

Here T is realized by an auxiliary verb while the main verb has moved into Asp. Giorgi and Pianesi (1992) argue that the denotational event arguments of the auxiliary verb and the main verb must be co-indexed for the structure to be interpreted. This means that a Tense chain is established via head movement of Aux to T and V to Asp. We will follow the thrust of Giorgi and Pianesi's ideas, but we shall be more explicit about how the Tense chain is composed. Giorgi and Pianesi provide no way of ensuring that R in T and R in Asp are co-indexed, as they must be to ensure proper interpretation. Likewise, they offer no insight as to how E in Asp is co-indexed with the event argument in V, another prerequisite for correct temporal interpretation. The mechanisms we have outlined above to deal with theta-marking are in fact all that is necessary. Asp governs (and is sister to) VP. 10 If the relationship between them is one of selection, then Asp takes VP as its internal argument and co-indexes the denotational argument of VP with its internal theta-argument: (44)

Asp

VP (eh(a(b)))

We thus establish the co-indexation between E in Asp and e in VP via selection indices. Now the co-indexation relationship between R and E in Asp can be read off the semantics of Asp. If Asp is simple (R = E) then R and E are lexically coindexed. Otherwise they are contra-indexed. The relationship between R in Asp and R in T is likewise established by selection. AspP is the internal argument of T, and therefore T co-indexes its internal theta-argument with the denotational argument of Asp, establishing the co-indexing relationship via selection indices. Thus we have the following representation for a simple present tense: (45)

T

Asp

VP

The indexing relationship between S and R can again be read off the semantics of S. If S is present tense then S = R and co-indexing occurs; otherwise we have contra-indexing. We will refer to the substructures formed by Asp and V and by T and Asp as aspectual chains (trivially all three heads form aspectual chains singly). These are composed into a single domain via selection. The Tense chain for the clause is formed, as seems semantically plausible, by binding from a temporal operator. The temporal operator associated with the utterance time is projected in the specifier of TP at LF (along the lines of Stowell 1992) and this operator binds the denotational arguments that it governs: (46)

Op

OrV

T

Asp

VP

<5>, (*,-)> ( ^

M

214

David Adger

This system allows the temporal interpretation to run directly off the morphosyntactic structure with no further stipulations than are already required for canonical theta-marking. The Tense chain is composed of smaller aspectual chains via the mechanism of selection. One interesting point about this system is that it dissociates thematic structure (associated with lexical heads) from argument structure (associated with both lexical and functional heads). This is a position recently argued for by Adger and Rhys (1993), Rhys (1993) and Cann (1993) among others.

4.3 Tense and aspect in Scottish Gaelic 4.3.1 Compound tenses SG appears to reflect Reichenbach's analysis rather directly. The most common way of marking the difference between present and past tenses is to use a form of the verb bith 'be' as an auxiliary with a nominalized form of the main verb which occurs with an aspectual particle. The verb bith marks the relationship between S and R: (47)

a. Tha Daibhidh a' falbh. be.PRES David SIM leave.vN 'David is leaving.' b. Bha Daibhidh a' falbh. be.PAST David SIM leave.vN 'David was leaving.' c. Bithidh Daibhidh a' falbh. be.FUT David SIM leave.vN 'David will leave.'

The particle d here marks that E = R. This becomes even clearer if we inspect the perfect paradigm: (48)

Tha/Bha/Bithidh Daibhidh air falbh. be.PRES David PER leave.vN 'David has/had/will have left.'

Here air marks that E < R. The paradigm is completed by the prospective, shown in (49) where gu marks that E > R: (49)

Tha/Bha/Bithidh Daibhidh gu falbh. be.PRES David PRO leave.vN 'David is/was/will be about to leave.'

Aspect, agreement, measure phrases

215

The analysis of the SG tense/aspect system we propose is simply that the auxiliary verb in T marks the S, R relation while the particle in Asp marks the E, R relation. The domain for the Tense chain is composed as discussed above, by selection, from the smaller aspectual chains. An example of this for the present perfect is given below: (50)

Tha Daibhidh air be.PRES David PER (5,, (*,)) (Rh(Ej)) 'David has left.'

falbh. Ieave-VN (eJt(a))

4.3.2 A bhith As well as the simple compound tenses discussed above, SG allows the use of an auxiliary to carry the aspectual particle: (51)

(52)

Tha/Bha Daibhidh air a bhith a' falbh. be.PREs/PAST David PER be.vN SIM leave.vN 'David has/had left.' Tha/Bha Daibhidh gu bhith a' falbh. be.PREs/PAST David PRO be.vN SIM leave.vN 'David is/was about to leave.'

This device allows the composition of aspectual particles: (53)

a. Tha Daibhidh gu bhith air a bhith a' falbh. be.PRES David PRO be.vN PER be.vN SIM leave.vN 'David is about to have left.' b. Tha Daibhidh air a bhith gu bhith a' falbh. be.PRES David PER be.vN PRO be.vN SIM leave.vN 'David has been about to leave.'

One constraint that emerges here, though, is that the simple aspect marker must come finally (this was first noted by Cram 1981): (54)

a. Tha Daibhidh be.PRES David 'David has left.' b. T h a Daibhidh be.PAST David

air a bhith a' falbh. PER be.vN SIM leave.vN a' a bhith air falbh. SIM be.vN PER leave.vN

Giorgi and Pianesi claim that, in order for a structure with an auxiliary and main verb to be interpreted, the denotational arguments of the auxiliary and main verb must be co-indexed. If we accept this claim, we immediately provide

216

David Adger

an explanation for the contrast in (54). The relevant structure for the wellformed (a) is: (55)

Tha be.PRES

a bhith a' be.vN SIM

Daibhidh air PER David

/)> (et)

E.-)} (*«(*«)>

'David has left.'

falbh. leave, VN

Here the denotational arguments of the auxiliary and the main verb are coindexed via the normal selection process which passes up the index through the simple aspect marker (where the denotational argument and the internal thetaargument of a' are specified lexically to be co-indexed because of the meaning of a'). We can assume that the auxiliary verb bith adds no aspectual information, nor does it add information that there is another event taking place; it simply serves to morphologically carry the aspectual marking of the particle air. The auxiliary, then, basically requires that its internal argument and its denotational argument carry the same information, so that the internal argument of bith and its denotational argument are co-indexed. This is a characteristic property of auxiliaries. Now consider the structure for the other example: (56)

*Tha Daibhidh be.PAST David

a' SIM

a bhith air be.vN PER

falbh. leave.vN

(RJ,(EJ))

(ej{aj))

(e,)

(Rj, (£,)>

Here, because the lexical specification of air marks that E < R, E and R cannot be co-indexed. A contra-indexation is then passed up to the auxiliary, which will then be contra-indexed with the main verb, in violation of Giorgi and Pianesi's constraint. 11 We also predict that gu, the prospective marker, will behave in the same way as air. This is the case: (57)

a. Tha Daibhidh be.PRES David 'David has left.' b. *Tha Daibhidh be.PAST David

gu bhith a' falbh. PRobe.VN SIM leave.vN a' a bhith gu falbh SIM be.vN PRO leave.vN

4.4 Summary This section has motivated the idea that the Tense chain used in the licensing of measure phrases is composed of smaller aspectual chains via the mechanism of selection. Once the separate aspectual chains have been composed via selection indices, the temporal operator of the clause can bind all of the denotational arguments within its selection domain via binding indices to form the T-chain proper. The advantage of this system is that it allows temporal interpretation

Aspect, agreement, measure phrases

217

to run directly from morphosyntactic structure. The system also predicts an unexpected constraint in the tense/aspect system of SG.

5 Consequences for licensing measure phrases Recall that a Tense chain is essentially formed via the mechanism of selection, encoded as co-indexation of the internal theta-argument of a head with the denotational argument of that head's XP sister. This immediately predicts that internal arguments of verbs are possible elements of a Tense chain. Note, however, that the selection-driven process of aspectual chain composition relies on a lexical indexing of the denotational argument and internal theta-argument of either aspectual particles or auxiliaries. This lexical indexing is read off the semantic properties of the head (see (45)). It follows that the internal theta-argument of a verb will only be part of an aspectual chain if there is a lexical co-indexing of the denotational argument of the verb and its internal theta-argument that follows from the meaning of the verb. Any such verb we then expect to behave much like an auxiliary with respect to the kinds of complements they select, since this type of lexical indexing is characteristic of auxiliaries. In fact this is exactly the kind of property we would like to attribute to verbs that take measure phrase complements. These can almost always be paraphrased by a copular verb. Crucially, this same copula is used as an auxiliary in the languages concerned: (58) (59) (60)

a. b. a. b. a.

Anson weighs 70 kilos, Anson is 70 kilos. The book cost 12 dollars, The book is 12 dollars. Tha da clach deug de chudrom ann an Daibhidh. be.PRES two stone teen of weight in(REDUP) David 'David weighs 12 stone.' b. Tha tri not air an leobhar seo. be.PRES three pounds on the book that T h a t book costs 3 pounds.'

We therefore specify the lexical entry of a verb like cost as: (61)

cost (etMbi)))

We can now actually derive the observation that measure phrases do not raise. If the phi-features of arguments are specified on the positions in the argument structure then the internal theta-argument of cost will have no nominal phifeatures, since it is co-indexed with the denotational event argument, which has

218

David Adger

no nominal phi-features. This means that the XP complement will never raise to the AgrO projected by cost, since there are no phi-features to check.12 Such a situation is not the case for most complements which have phi-features and are therefore required to be in the specifier of AgrO at LF for reasons of Casechecking. Measure phrases, since they do not raise to AgrO, are therefore potential Tense chain members. Consider, then, the licensing of measure phrases in postverbal position in SG: (62)

Tha a' cho-labhairt a' be.PRES the conference SIM

mairsinn last.vN

seachdainn. week

T h e conference is lasting a week.' Here the denotational argument of the measure phrase is co-indexed with that of the verb and that of Asp. It therefore is part of a composed aspectual chain via selection indices. Note that it also conforms to the generalized version of Visibility if we take 'co-indexed' in this definition to refer to selection indices. Now consider the structure with the perfective particle: (63)

Tha a' cho-labhairt air be.PRES the conference PER

mairsinn last.vN

seachdainn. week

T h e conference has lasted a week.' Here the measure phrase is co-indexed with the event argument of the verb but the semantics of the aspectual particle means that the denotational argument of the air carries a different index. This means that the measure phrase is not co-indexed with a functional head and that the generalized version of Visibility is violated since seachdainn is not F-licensed. The same explanation extends to the analogous structure with the prospective marker gu, which is also ill formed. The question now arises as to why the corresponding English structures are grammatical: (64)

a. The conference has lasted a week, b. ?This book has cost twenty pounds.

We will attribute the slight ill-formedness of the (b) example here to a semantic tension between the perfect, which requires a final bound on an event, and the measure verb cost, which is lexically specified as some kind of unbounded state. Recall that we require the measure phrase to be co-indexed with a T functional head. This requirement is satisfied by the head movement of V into the aspectual functional head realized by -ed, under the standard assumption that head movement leaves a trace.

Aspect, agreement, measure phrases

219

But now we seem to have lost some of the empirical coverage of the original approach in Adger (1994b). Recall that quantificational determiners were ruled out in measure phrases because they create a Minimality barrier for T-chain formation. Selectional indices, however, ignore functional structure, so a measure phrase with a quantificational determiner would be able to be part of a T-chain. This problem ignores, however, the difference we have already drawn for independent semantic reasons between selectional and binding indices. Selectional indices compose aspectual chains into a single domain, but the actual formation of T-chains is stated in terms of binding indices. We repeat the relevant structure: (65)

Op T

Asp

VP

( V (Si(Rk)) (Ri(Et)) (ei(a(b)))

The Tense operator of the sentence binds all the denotational arguments it governs within its selection domain. Government of the denotational argument of a quantified measure phrase is blocked by the quantifier, which counts as a closer binder: (66)

Op T j

Asp

VP

Q

k

DP

Op (SU*i)) ( 4 W > (ei(a(b)))Q <**) So, while the measure phrase is in the appropriate selectional domain to be part of a T-chain, whether or not it is actually part of a T-chain depends on binding. One final prediction is made by the system we have outlined in this chapter. We predict that measure phrases in structures with the auxiliary verb bith and the R = E aspectual particle a' should be well formed, even if the auxiliary carries the perfective particle. This follows since the measure phrase will be co-indexed with the aspectual head a\ as above. This proves to be the case, and such structures are actually a paraphrase for the ill-formed bare perfective structures: (67) Tha a' cho-labhairt air a bhith a' mairsinn be.pRES the conference PER be.vN SIM last.vN <*,,(£,)) (*„(«,)> (%,&)) (e,(a(b,))) seachdainn. week (x,,0) The conference has been lasting a week.'

220

David Adger 6 Conclusions

This chapter has sought to defend a generalized version of Visibility by showing that it is implicated as a premise in the basic analysis of measure phrases and that it has interesting and unexpected empirical consequences when combined with an independently motivated refinement of the notion of Tense chain. We argued that potential Tense chains were composed of smaller aspectual chains via a mechanism of selection indices and that the actual Tense chain is formed from binding indices. Crucial evidence for this theoretical standpoint derives from the interaction between word order, agreement and aspect in Scottish Gaelic.

Notes Many thanks to Bob Borsley, Ronnie Cann, Elisabet Engdahl, Laura Joosten, Martin Mellor, Catrin Sian Rhys, Jeff Runner, two anonymous reviewers and the participants of SCIL V for comments on earlier drafts/presentations of this chapter. F-licensing, as it is defined here, would predict that a w/j-phrase in the specifier of CP is independently licensed, and thus does not need to be co-indexed with Agr or T. It is empirically unclear whether w/z-phrases actually do need abstract Case, given examples like: (i) (ii)

Who/ do you believe sincerely /, to be a genius. Someone [O,7, to fix the sink] is coming this afternoon.

In the first case we have the trace of the w/z-phrase in a position where it cannot be assigned Case by the lower Infl, since it is non-finite, nor by the higher ECM verb, since there is an adverb intervening. The second case involves movement of an empty operator from a non-Case-marked position to the specifier of the C of the infinitival relative. Such examples suggest that this definition of F-licensing may be correct. On the other hand, if it turns out that w/z-phrases do need Case, then the definition here can be revised so that it applies only to A-chains. Indefinites are marked with what appears to be the common case (nominative/accusative alternations are neutralized in SG, unlike Irish). We will not discuss this here, although two possible explanations come to mind: either the postverbal slot is actually assigned partitive Case and the partitive paradigm is an amalgam of the common and genitive paradigms (this is suggested by Ramchand 1993a); or indefinites and definites are licensed in different ways, the latter by an inserted Case-marking (similar to English ^/-insertion) and the former by Tense-binding. We will not take up this question here. This does not appear to be the case for Welsh, which appears to have pronouns with agreement. See Hendrick (1988) for a reanalysis of the Welsh data which makes it amenable to this generalization; Hendrick claims that the 'pronouns' that appear with agreement in Welsh are actually a reduplicated form of agreement themselves, and that they occur with a pro. See also section 6 of Roberts and Shlonsky's chapter in this volume for relevant discussion of these facts.

Aspect, agreement, measure phrases

221

5 Bob Borsley has pointed out that claiming that pro has no phi-features will result in different generalizations for antecedent agreement having to be made for overt and nonovert pronouns. Overt pronouns will be required to have compatible phi-features with their antecedents, while pro will be required to be in an agreement relationship with a head that has compatible phi-features. It might be possible to state these generalizations with reference only to agreement with a head, since overt pronouns will have to be in an agreement relationship with a head for reasons of checking/Case anyway. 6 These data are actually more complex than they seem from this presentation. Some speakers have very weak judgements here, often preferring 's e to 's ann and justifying this judgement with some statement to the effect that the verbal noun is a noun and therefore must be used with "s e. Other speakers find no contrast, deeming both to be marginally acceptable. Finally, the speakers who agree with the judgements given here typically have no realization for the aspectual particle a' and allow a range of complement types after adjectives like doirbh, suggesting that the appearance of 's ann may be attributed to the clefting of an aspectual phrase that happens to be homophonous with FOP. However, the point still holds that there is a contrast between (25) and (26) which is difficult to explain if agreement is a feature of V. 7 It may be objected that the elements which we have glossed here as agreement markers are in fact possessive pronouns, and indeed this is the traditional analysis. However, whereas we have a fair amount of evidence for the syntactic realization of agreement features and a clear way of incorporating them into our theory, the same situation does not hold for possessive pronouns. In fact, we would argue that possessive pronouns in SG are simple manifestations of agreement features in the standard cases as well. See McCloskey and Hale (1984) for discussion of the similar Irish facts. Bobaljik and Carnie (this volume) argue for theory-internal reasons that these elements are manifestations of agreement in Irish, as do Roberts and Shlonsky (this volume) for Welsh. 8 We shall not address the question of the correct logical form of these elements here. The important point is that the weak island constraint is a constraint on the denotation of an extracted element. Krifka (1990) has pointed out that DPs are typically ambiguous between what he terms an Event-Related and an Object-Related reading. The former reading derives from the DP in question being counted as part of the event in the interpretative domain. On our account measure phrases have obligatory EventRelated readings, since they are part of a T-chain. As noted by T. Hoeskstra (see also Honcoop 1992), Event-Related readings are never possible across weak islands. 9 This analysis ignores phenomena such as imperfectivity, for which is is necessary to view the Reichenbachian points as temporal intervals with internal structure. 10 Asp is actually probably sister to AgrP and only governs VP through AgrP, but we shall abstract away from Agr in the discussion here. This abstraction is derivable from Relativised Minimality (see Roberts 1991b) if we assume that Agr does not block government of VP by Asp since it is a different type of functional head. 11 We might think of the be and have auxiliaries in English as having in part the same semantics as the SG particles a' and air and rule out examples like *Anson is having left in the same way. My feeling is that the English system is more complex, however, in conflating the information that SG spreads between the aspectual particles and the auxiliary bith.

222

David Adger

12 A general problem for the checking view of raising is provided by examples like: (i) (ii)

Under the bed seems to be a good place to hide, Very quickly seems to be the best way to do it.

Presumably neither the PP nor the AdvP have features that need to be checked. It may be that subject-raising differs from object-raising due to the interference of some version of the Extended Projection Principle. Webelhuth suggests that such subjects contain an empty Determiner on the basis of the fact that they form DP chains in a number of constructions (1989: 40). I have no alternative explanation.

7

A minimalist approach to some problems of Irish word order

Jonathan David Bobaljik and Andrew Carnie

1 Introduction Recent work in the principles-and-parameters approach to syntactic theory has been concerned with the range of word order variation in the world's languages.1 It is a tenet of this approach that such variation can be derived from a highly constrained set of simple parameters, interacting with universal principles of natural language. We would like to investigate here how the facts of Irish word order may best be represented within the framework argued for in Chomsky and Lasnik (1993) and Chomsky (1993), and discussed in Marantz (1995). The first section will provide a brief overview of the relevant notions and mechanisms of this framework and the analysis we propose. After this, we will turn to the relevant Irish data, moving on to a discussion of the predictions of our analysis and potential problems it raises. In the final section, we will provide a more detailed refinement of our initial analysis, discussing its import for a featuredriven theory of syntactic variation as in Chomsky (1993).

2 The framework and an initial analysis Over the last half-decade, much work in syntax has been devoted to motivating and supporting the claim that all arguments of a verb, and in particular the subject, are base-generated within the maximal projection (VP) of that verb (i.e. the VP-internal subject hypothesis of Fukui and Speas (1986), Kitagawa (1986), Koopman and Sportiche (1991) among many others). This approach entails that in a language such as English, the subject must raise to somewhere within the maximal projection of an Inflectional category to receive abstract nominative Case. Extending a proposal by Pollock (1989), Chomsky (1991, 1993) has suggested that both structural cases (i.e. nominative and accusative) are realized in a parallel manner, via movement (either overtly: before Spell Out (formerly Surface Structure), or covertly (at Logical Form, the semantic component), of the arguments to positions within the inflectional complex. Specifically, it is suggested that all structural case and agreement is the realization of a specifier-head 223

224

Jonathan David Bobaljik and Andrew Carnie

relationship with an appropriate functional (Agr) head. Thus the inflectional complex includes a Tense phrase (TP), and two (non-distinct) agreement phrases (AgrS, AgrO): (1)

[AGRSP [AgrS [TP [T [AGROP [AgrO [VPsubj [V[obj

While some movement may be at LF, the verb must eventually raise to AgrS, adjoining to each of the intervening head positions via head-to-head movement (Travis 1984). NPs must receive Case, so all arguments must raise to the specifier position of one of the agreement phrases at some point in the derivation. As the agreement heads are non-distinct,2 the Case with which each is associated is determined by the nature of the element which adjoins to it. The accusative Case, being in some sense a verbal attribute, must be realized in the specifierhead relationship with the complex head [V, AgrO] derived via the first step of the head-to-head movement when the verb is transitive. (2) Accusative Case

AgrOP

By similar logic, head movement of Tense (T) to AgrS will create the complex head [T, AgrS] and nominative Case will be realized in a specifier-head relationship to this head. AgrSP

(3) Nominative Case

Subj

Irish word order

225

Within the bounds of this approach, there are a number of possible derivations which will result in a surface VSO word order. The first possibility is that the verb (or verbal complex) has 'fronted' to an initial complementizer position, in which case the surface positions of the subject and object are not a priori evident. This approach is taken in Stowell (1989) and Doherty (1992) and by Carnie, Pyatt and Harley (forthcoming) for Old Irish. A second approach, suggested by Chomsky (1993), is that the VSO order is derived if the verb raises overtly to some inflectional head, but the subject and object remain in situ, raising covertly at LF. In what follows, we will suggest that empirical evidence points against both of these approaches for Modern Irish. Specifically, we claim that there is evidence (a) that the initial V in Irish is not in C and (b) that nominal arguments in Irish raise in the overt syntax. The remaining possibility, then, is that Irish tensed V raises overtly to AgrS, 3 but that the subject raises only as far as specifier of the Tense phrase. 4 Determination of the position of the object under this analysis raises issues which will be addressed towards the end of the chapter.

3 Irish The basic word order in Irish tensed clauses is VSO. The nominative subject follows the verb, and the accusative object follows the subject. Oblique NPs and adverbs 5 generally follow the arguments of the verb. This is illustrated in (4): (4)

a. Rith run. PAST

siad.

They ran.' b. Chonaic Sean see.PAST

(VS)

they.NOM

an madra.

(VSO)

J.NOM the dog

'John saw the dog.' c. *Sean chonaic an madra.

(*SVO)

J.NOM see.PAST the dog

'John saw the dog.' d. Chonaic se i. (VSO showing case) saw.PAST he.NOM her.ACC 'He saw her.' e. Thog se an teach leis an ord. (VSO Obi) build.PAST he the house with.DEF the hammer 'He built the house with the hammer.' f. Beidh Nora ag an ndroichead amarach. (VSO Adv) be.FUT N. at the bridge tomorrow 'Nora will be at the bridge tomorrow.'

226

Jonathan David Bobaljik and Andrew Carnie

Following work by McCloskey (1983), we will assume that VSO languages do not have an underlying 'flat structure' but are derived from SVO order by movement.

3.1 Against verb to comp raising

An obvious analysis within the bounds of the present framework would be that the subject (and possibly object) raise overtly to the specifiers of respective agreement phrases for Case checking, and the verb obligatorily raises through the inflectional complex and on to C (5). (5)

The well-known 'verb-second' (V2) phenomena (6) of Germanic languages have frequently been analysed as movement of the verb to C with a concomitant restriction that the specifier of the complementizer phrase be filled by some constituent (e.g. den Besten 1990). Part of the motivation for this analysis is the fact that in subordinate clauses V2 order is not permitted, and the order C-SOV is standard (Koster 1975). The hypothesis is that the verb may raise to an empty complementizer position in matrix clauses, but that in embedded clauses, the complementizer position isfilled(possibly with a phonologically null complementizer), and the verb cannot raise to it (7) (data from Haegeman 1991). (6)

a. Karl kaufte dieses Buch gestern. Karl bought this book yesterday 'Karl bought this book yesterday.' b. Dieses Buch kaufte Karl gestern. this book bought Karl yesterday 'Karl bought this book yesterday.' c. Gestern kaufte Karl dieses Buch. yesterday bought Karl this book 'Karl bought this book yesterday.'

Irish word order

(7)

227

d. Was kaufte Karl? what bought Karl 'What did Karl buy?' . . . dass Karl gestern das Buch gekauft hat. . . . that Karl yesterday the book bought had '. . . that Karl had bought the book yesterday.'

If Irish were to have a comparable analysis, that is, obligatory fronting of the verb to C in matrix clauses (although without the requirement that the specifier of the complementizer phrase be filled), then we would expect similarly the order C-SOV or C-SVO in embedded clauses.6 This prediction is immediately falsified by the facts of Irish. In fact we only get C-VSO order. The verb still must raise. It cannot move to C since that position is already filled: (8)

Ceapaim [go bhfaca se an madra] think.PRES.lsG [that see.PAST.DEP he.NOM the dog] COMP V Subj Obj 'I think that he saw the dog.'

There is conceptual motivation for rejecting the obligatory V-to-C analysis. Challenging the 'standard' analysis of V2 in Germanic, Travis (1991) and Zwart (1993) suggest independently that V-to-C raising only occurs when some morphological property of C (such as 'topic' or '±wh') must be satisfied. Both authors adduce empirical evidence to support their claim that matrix SVO order in Germanic languages with a non-topicalized subject cannot be derived from V-to-C raising. The conceptual point that these papers (and earlier work along these lines) raise is the following: if movement is motivated solely by morphological properties, then V-to-C raising can be motivated for topicalization or question formation (cf. English 'Aux to Comp Inversion'), but cannot be motivated for non-topicalized declarative clauses. In Irish, there is no property such as topicalization or interrogation which would force overt raising of the verb to C. An analysis involving obligatory V-to-C in simple declaratives is conceptually untenable. In addition, there is empirical evidence against the V-to-C raising approach. McCloskey (forthcoming) presents comprehensive evidence from the behaviour of IP adjoined and CP adjoined adverbials that the verb in Irish cannot have raised beyond the left edge of the inflectional complex (thus is lower than C). The data are quite complicated and for reasons of space we will not repeat them here. Roughly speaking, McCloskey argues that since IP adjoined adverbs appear to the left of verbs, verbs cannot be higher than the left edge of the inflectional complex.

228

Jonathan David Bobaljik and Andrew Carnie 3.2 Against subject and object in situ

The approach suggested by Chomsky (1993) is that the verb raises to some head within the articulated inflectional complex before Spell Out. The subject and object remain in situ in the verb phase in the overt component, raising covertly atLF. (9)

UgrSptAgrS [ TP [T [ AgrO p[AgrO[ VP subj [V [ obj ]]]]]]]]]

t

I

I

I

In this section, we will argue that there is empirical evidence that such an approach is also untenable for Irish. The analysis suggested by Chomsky is not obviously falsifiable by looking solely at tensed clauses, both matrix and embedded. In both cases, the order is C-VSO. As adverbial elements in Irish generally occur following the verb and its arguments, adverb placement cannot be used as a diagnostic for the structural positions occupied by the arguments and the verb.7 Turning to non-finite clauses,8 however, one immediately notes that VSO order is impossible. In all dialects nonfinite clauses may show the surface order OV (where there is a strong pragmatic preference for sentences without overt subjects):9 (10)

Ba mhaith Horn [(e) an teach aL thogail]. COP good with.lsG him.ACCthe house.ACC TRANS build 'I would like him to build the house.'

When there is an overt NP or pronoun, the non-finite verb is preceded by both the object and the transitive particle aL.w Note that both the subject and object are marked accusative.11 This OV order is also found in the recent perfective12 (also called the 'after perfect'). (11)

Ta me tar eis an teach aL thogail. be.PRES I ASP the house TRANS build 'I have just built the house.'

While SOV order in infinitives is obligatory in the northern dialects (Connacht and Ulster), in the southern Munster dialect there is an alternative to (10): a marked SVO order. In this construction, the object may appear postverbally (12b). This option is available only with an overt subject (which takes accusative Case). In either case, the 'transitive' particle aL is present also. (12) Southern: Munster a. Ba mhaith Horn, [ PRO, an abairt aL scriobh]. (PRO O V) COP good with.lsG the sentence.ACC TRANS write 'I want to write the sentence.'

Irish word order b. Ba mhaith liom[cp Sean aL scriobhna habairte]. COP good with.lsG John.ACC TRANS write the sentence.GEN *I want John to write the sentence.' (formal)

229 (SVO)

Note that there is a difference in Case-marking of the direct objects in (10) and (12). In the SOV order (10) and (12a), the object bears accusative Case. In the SVO example in (12), the object bears genitive Case. As McCloskey (this volume) points out, while pedagogic grammars mandate the genitive for postverbal objects in this construction, 'this rule is implemented sporadically at best, even by speakers who preserve the genitive regularly in other contexts.' This variation in overt case is curious, though largely orthogonal to present concerns. In particular, the possibility of accusative Case for postverbal objects indicates that these NPs are arguments and that this (postverbal) position is the structural object position. Genitive postverbal objects are also evidenced in progessives in formal registers, accusative ones in colloquial ones (see Noonan 1993 for discussion): (13)

Ta si ag scuabadh an urlair. be.PRES she.NOM PROG sweep the floor.GEN 'She is sweeping the floor.'

Returning to the infinitives, it would appear that accusative Case is available for the object to the left of the non-tensed or participial verbs. It, thus, may raise to the specifier of AgrOP (Duffield 1991). The Munster dialects allow a construction whereby the object may remain in situ. All dialects have postverbal (genitive) objects in progressives. Our claims so far are: (14)

a. The object (along with the verb) is in situ in progressives and Munster non-finite clauses. b. The preverbal object in all dialects is raised to the specifier of AgrO. 13

Extending this somewhat, if the object has raised overtly to the specifier position of AgrOP yet the subject still precedes the object, then the subject must have raised past the object. Taken together, the data and arguments of this subsection entail that the Chomsky-style Subject-and-Object in situ analysis cannot be maintained for Irish. In the previous subsection, we ruled out the V-to-C analysis. Let us now present an alternative analysis, for the most part compatible with the framework developed by Chomsky (1991, 1993) and Chomsky and Lasnik (1993), which does capture the basic facts of Irish word order, and also sheds some light on the similarities between Munster SVO infinitives and progressive constructions in all dialects.

230

Jonathan David Bobaljik and Andrew Carnie 4 Our analysis

Consider again the structure of transitive clauses given in (1), (1)

[AGRSP [AgrS [TP [ T [AGROP [AgrO [VPsubj [V [ obj

In section 3.1, we showed that derivation of Irish VSO by obligatory V-to-C movement was untenable on both empirical and conceptual grounds, In section 3.2, we argued that there is evidence in Irish pointing to overt raising of the object to specifier position of AgrOP, which would rule out an analysis of raising of arguments at Logical Form. In addition, this entails that the subject must raise past the object in the specifier of AgrO, as the subject linearly precedes the object. Finally, as finite clauses have the verb preceding both the subject and the object, the verb must raise to some position higher than the subject, but lower than COMP. This analysis is detailed in the following sections. In section 4.1, we will derive VSO order in tensed clauses. In section 4.2, we will look at the derivation of non-finite clauses. 4.1 Irish VSO in finite clauses The analysis we shall pursue here is that the overt movement in Irish consists of head movement V —» AgrO —• T —> AgrS, and of NP movement of the object to the specifier of AgrOP and the subject to the specifier position of the Tense phrase (TP). (15)

Ugrsp [AgrS'[AgrS + T + AgrO + V], [rpsubj* [T> tt [Agrop objw UgrO' h [VP h [ v U tm]]]]]]]]

Let us look at the derivation proposed above in more detail. For the sake of simplicity, we will discuss this in terms of a step-by-step derivation. The first step in the derivation is head movement of the verb to AgrO, creating the complex head [AGRO V + AgrO]. The chain created by this step allows the object to raise over the subject to the specifier of AgrO - the next-highest specifier position.

Irish word order (16)

231

AgrOP

Informally, in order to raise over the specifier of the VP which contains the subject, the verb must raise and adjoin to AgrO. This follows from the Minimality effects discussed by Rizzi (1990) which ultimately can be derived from considerations of Economy (Chomsky 1991, 1993). In particular, this is related to Holmberg's (1986) generalization that verb-raising is required for overt object-raising, and likewise provides a principled account of Baker's (1988) 'Government Transparency Corollary'. Next, the (complex) head AgrO (containing the verb) raises to Tense (T), creating the complex head [T AgrO, T], and the subject raises to the specifier of the Tense phrase:

(17)

232

Jonathan David Bobaljik and Andrew Carnie

Again, considerations of economy require the head movement in order to permit raising of the subject to 'skip' the intervening specifier of AgrO containing the object. The last overt step is raising of the head T (Tense, containing Tense, AgrO and the verb) to adjoin to AgrS, creating [AgrS T + AgrS ]. (18)

'Spell Out' occurs at this stage, resulting in 'surface' VSO order. Finally, covert movement occurs at Logical Form to check agreement features and assign nominative Case to the subject. The subject raises from the specifier of the Tense phrase to the specifier of the AgrSP. Note that this movement only occurs in the semantic component and is never realized in the phonological output:

Irish word order

233

(19) Covert movement (i.e. at LF) AgrSP Subj

While this analysis derives the correct word order, it appears somewhat ad hoc. Now let us consider how such a derivation might be motivated, using the theory of syntactic features.

4.2 Features Within the framework being explored here, Chomsky (1993) proposes that each of the heads (Tense and the two Agrs) have N[ominal] and V[erbal] features which may be parameterized with either a 'strong' value or a 'weak' one. 14 Strong features are required to be checked in the derivation by Spell Out (i.e. in the overt syntax), while weak features need not be. The interaction of these features with independent principles (for example, the Procrastinate Principle (Chomsky 1993) requires that if movement is not required to be overt, it will be covert) will dictate whether certain steps of the derivation occur overtly (prior to Spell Out) or covertly (at Logical Form). The N-features correlate with the specifier positions, governing NP movement, and the V-features with the heads, governing head movement. Consider, for example, how the differences between English and French, discussed in Pollock (1989) and Chomsky (1991), are to be represented on this approach. Their proposed feature specifications are given in (20): (20) AGR Tense

N V N V

English weak weak strong strong

French weak strong strong strong

234

Jonathan David Bobaljik and Andrew Carnie

Strong features must be checked in the overt syntax. As N-features are correlated with the specifier-head relationship, the specification strong for the N-feature of Tense in both languages requires that an NP argument raise to check its features in the specifier-head configuration with Tense. This, in essence, is what ultimately derives the requirement that all sentences have a subject (i.e. the 'Extended Projection Principle' of Chomsky 1981). By hypothesis (Chomsky 1993), both English and French require that Tense raise overtly to AgrS to check its N features. We indicate this by a strong valence for the V-features of Tense, requiring overt raising (head movement) of T to AgrS to check these features. This raising will mean that the structural specifier of the Tense phrase is not licensed for feature-checking, despite its strong N feature. In order for the strong N-features of Tense to be checked, then, an NP-argument (the subject) will have to raise overtly to the specifier of the complex head [Agrs T + AgrS] resulting from the head movement of Tense to AgrS. This is illustrated schematically in (21).15 (21)

AgrSP

Subj

There are three distinct head-movement processes in English and French: (1) T moves to AgrS; (2) V moves to AgrO; and (3) [V + AgrO] moves to AgrS. The first movement is overt in both languages as required by the strong V-features of Tense. The remaining movements are governed by the V-features of the Agr nodes. In English, the V-features of Agr are weak and thus only the raising of Tense to AgrS occurs overtly, whereas in French, the V-features of Agr are strong and both of the remaining head movements occur overtly, with all (finite) verbs raising in the visible syntax. Following Pollock (1989) and Chomsky (1991, 1993), this accounts for the differences between the two languages. The only relevant difference between the two languages, then, is in the specification for the V-features of AGR. In our analysis Irish, like French, has strong V-features of AGR (requiring the verb to raise overtly), and strong N-features of T (requiring that the subject check its Case features in the specifier-head configuration with T), but its remaining

Irish word order

235

features, including the V-feature of Tense, are weak. This last is the key. In French (and English), we showed that strong V-features for Tense entailed overt raising of Tense to AgrS, rendering the specifier of TP unavailable, and requiring that the strong N-features of Tense be checked in the specifier of the complex Head [T + AgrS] (21). By hypothesis, Irish has weak V-features and thus T need not (and so cannot) raise independently to AgrS. As the N-features of Tense are strong, the NP-argument which will check these features, the subject, thus only need raise as far as the specifier of TP in the overt syntax for checking of the features to be satisfied. Summarizing, the crucial difference between French, which displays SVO order, and Irish, which displays VSO, is that in Irish there is a difference in the valence of the V-features of Tense which correlates with whether or not Tense must raise overtly to AgrS (i.e. independently of the raising of V —• AgrO —> TENSE —• Agrs). Note that in Irish T does, in effect, raise overtly to AgrS, but only as a step in the sequence of head movements V —» AgrO —> T —> AgrS. This difference correlates with the possibility of checking the N-features of Tense in the specifier of TP (Irish) as opposed to in the specifier of AgrSP (with the complex head [T + AgrS] (French)). The features of English, French and Irish are thus: (22) AGR Tense

N V N V

English weak weak strong strong

French weak strong strong strong

Irish weak16 strong strong weak

To summarize then, the weakness of the V-feature on the Tense node indirectly licenses the specifier of TP as a possible subject position, unlike the specifier of TP in English and French. VSO order, therefore, results from the interaction of two facts: firstly, AgrS's N-features are weak and Tense's N-features are strong, thus allowing NPs to raise only as far as the specifier of TP overtly; secondly, and more interestingly, the specifier of TP is made available by the Tense node's weak V-features.

4.3 Non-finite clauses As discussed above, we assume that the object in the Munster SVO non-finite clauses and all dialects' progressives is in its base position. AgrO in both progressive and non-finite clauses is unavailable as a Case position. Consider SOV infinitives, with an accusative object. In section 3.2 we claimed that these involved overt raising of the object. It was pointed out in the preceding section that overt raising of the object to specifier of AgrO is only possible if the verb has raised

236

Jonathan David Bobaljik and Andrew Carnie

overtly to AgrO. Thus we must claim that the verb is at least as high as AgrO in this construction. Following Duffield (1990, 1991) for Irish, and Adger (1994a, 1994b) for a related phenomenon in Scots Gaelic, we assume that the landing-site of the object shift is the specifier of AgrOP, and that the aL 'transitive' particle on the verb is a realization of AgrO, and marks the licensing of accusative Case. Ramchand (1993a) and Guilfoyle (1993) have claimed that this particle is a realization of some VP-internal Aspect head, and that the locus of object shift is to the specifier of this position. We assume that this approach is misguided for the following reasons. Firstly, following work by Adger (this volume), this particle behaves like an agreement morpheme. As discussed in McCloskey and Hale (1984), agreement and overt nominal arguments in Irish and Scots Gaelic are in complementary distribution. Except under very specific circumstances, the presence of an overt nominal argument precludes the appearance of agreement. Interestingly, in the speech of older speakers, the 'transitive' particle behaves in exactly the same way as overt subject agreement. When an overt object NP is present, it takes the form of the default third-person possessive pronoun aL (23a). When no overt object NP is present it is inflected for person and number 17 (23b). When agreement is present no overt NP may surface (23c). (These data are the Irish equivalents to Adger's Scots Gaelic examples.) (23)

a. Ba mhaith Horn na buachailli a L bhualadh. COP good with.me the boys TRANS.3.SG strike 'I would like to strike the boys.' b. Ba mhaith liommo L /do L /a L /a/ar N /bhur N /a N bualadh. COP good with.me 1SG/2SG/3M.SG/3F.SG/1PL/2PL/3PL strike. 'I would like to strike me/you/him/her/us/you/them.' c. *Ba mhaith Horn na buachailli a N mbualadh. COP good with.me the boys TRANS.3.PL strike 'I would like to strike the boys.'

This strongly suggests that this is an agreement rather than an aspectual particle (see Roberts and Shlonsky, this volume, for discussion of a similar phenomenon in Welsh). The second argument against this being an aspectual particle comes from the fact that it can co-occur with other aspectual particles. In particular it occurs in conjunction with the recent perfective particle tar eis, as seen above in (11), repeated here: (11)

Ta me tar eis an teach a L thogail. be.PRES I ASP the house TRANS build 'I have just built the house.'

Assuming that these constructions are monoclausal, the requirement that two particles be present to indicate a recent present would be quite surprising (see

Irish word order

237

Adger, this volume, for more discussion). For these reasons, then, we assume that the landing-site of object shift is the specifier of AgrOP. The next question we must consider is: how is the subject allowed to raise past the object in specifier of AgrOP if the verb has raised no higher than AgrO in the northern dialects?18 That is, if the verb has not raised past the shifted object, then the specifiers of TP and AgrO should not be equidistant from the base position of the subject, thus the latter should not be able to raise overtly. Watanabe (1993a) offers one solution to this problem, that AgrO excorporates and raises overtly to non-finite T, stranding the main verb in AgrO. Bobaljik (1994) suggests that tying the equidistance clause to overt verb movement is problematic even in the languages for which it was developed and rather that equidistance (or more accurately domains) should be seen representationally, for example, at LF. There are undoubtedly other solutions to this problem, but we leave the matter open for further research. 5 Summary and conclusion

In this chapter we have attempted to account for the facts of Irish word order in the framework of Chomsky (1993). We have shown that both an in situ analysis and a V-to-C analysis are inadequate. We claim that Irish, like French, is a verbraising language, but that, unlike French, it does not require the overt movement of the subject NP to the specifier of AgrSP. Licensed by a weak V feature, Tense does not raise to adjoin with AgrS, thus allowing the subject to remain in Tense's specifier position at Spell Out. It is our hope that this work will stimulate further research into word order phenomena and their link to parametric variation in the features of functional categories. Notes An earlier and shorter version of this chapter was presented at the twelfth Annual Harvard Colloquium in 1992 and appears under the same title in the proceedings of that conference. We would like to thank David Adger, Donall 6 Baoill, Robert Borsley, Tony Bures, Maire Ni Chiosain, Noam Chomsky, Nigel Duffield, Danny Fox, Eithne Guilfoyle, Ken Hale, Heidi Harley, Dianne Jonas, Alec Marantz, Jim McCloskey, Maire Noonan, David Pesetsky, Colin Phillips, Elizabeth Pyatt, Ian Roberts, JanWouter Zwart and the participants of the 1994 VSO workshop at MIT for helpful discussions on this material; usual disclaimers apply. Jonathan Bobaljik's work has been supported in part by a Mellon Fellowship and an SSHRC doctoral fellowship, Andrew Carnie's work by a grant from the Alberta Heritage Fund and by a grant from the Social Sciences and Humanities Research Council of Canada. In this framework, they are only a collection of relevant phi-features such as person, number and gender. More exactly, raising is: V to AgrO; AgrO to T, T to AgrS.

238

Jonathan David Bobaljik and Andrew Carnie

4 A related analysis has been independently reached for Arabic VSO by Ouhalla (1994). 5 There is an exception to this claim where an adverb intervenes between the subject and the object (see McCloskey, this volume). We will discuss this in more detail below. 6 However, see Carnie, Pyatt and Harley (forthcoming), who claim that VSO order in Old Irish (Irish from the eighth century AD) involved at least some verb-raising to C. 7 But see note 5. 8 The argument has been raised, by both Nigel Duffield (1994) and James McCloskey (this volume), that the NP movement available in non-finite clauses need not be identical to that in finite clauses. Furthering their claim, it might be claimed that since NP movement in this framework is directly related to the verb movement properties and tenseness of a clause, we should predict that the NP movement in non-finite clauses will not be like that of finite clauses. We believe there are two main problems with this kind of objection: the first is metatheoretical, the other empirical. First, the metatheoretical problem: by Occam's razor, we should not complicate the grammar any more than necessary. The null assumption then will be one where the NP-movement properties in finite and non-finite clauses are identical. By claiming that the NP-movement properties 'need not' be identical we are simply complicating the grammar without cause. The second objection comes in the form of an incorrect prediction that might be made by those who claim that NP-movement in tensed and non-finite clauses are different (although it is not made by either McCloskey or Duffield). Given that in non-finite clauses (as will be seen below) there is less verb movement than in finite clauses (i.e. the verb only moves to AgrO, if it moves at all), we would predict less movement in nonfinite (SOV) clauses than in tensed (VSO) clauses. This is precisely the opposite of what is evident on the surface: there is obvious NP-movement in the non-finite (SOV) clauses, whereas there is no obvious (i.e. non-string vacuous) movement in VSO clauses. From a theoretical perspective, then, given that there is less movement in non-finite clauses than there is in finite clauses, any movement that appears in a non-finite clause will necessarily appear in finite clauses as well. Any overt movement that occurs in non-finite clauses will necessarily be a subset of the movement that can occur in finite clauses. Thus we can, without hesitation, use the evidence from non-finite clauses for determining the (minimum) NP movement in finite clauses. 9 Irish apparently always allows subjects of non-finite clauses which surface with accusative Case-marking. Chung and McCloskey (1987) show convincingly that this subject is not receiving Case from the higher clause: in no respect does the embedded subject behave as a matrix object, and with respect to, for example, binding phenomena, it clearly behaves as if it is in the embedded clause at all levels of the derivation. In addition, this accusative Case is always available for the subject, regardless of the matrix predicate. 10 This particle also surfaces as do in some dialects and registers. 11 Full NPs, like those in the examples below, do not show a morphological distinction between nominative and accusative Cases; however, pronouns do. 12 See Ramchand (1993a) and Adger (1993 and this volume) for discussion of the related construction in Scots Gaelic. We will discuss their analyses in more detail below.

Irish word order

239

13 There is one set of data in the literature which might be construed as a strong argument against this approach. McCloskey (this volume) points out that there is a very limited set of adverbials that may appear after the subject but before the object: (i)

Nior shaothraigh Eoghan ariamh pingin. NEG earned Owen ever penny 'Owen never earned a penny.' (data from McCloskey, this volume)

McCloskey assumes these are VP-adjoined adverbs. He takes this as evidence in favour of our approach since it shows that subjects must have raised outside (to the left of) VP. It can, however, be taken as evidence against our overt object shift in finite VSO clauses - since the so-called VP adverb is to the left of the object. Note, however, that there are two issues at hand here: the position of the object and the adjunction site of the adverb. We suggest that it is not in fact the first of these issues which should be questioned here; rather, it is the second (following a suggestion by Pilar Barbosa (p.c.) and taken up in Harley 1993). McCloskey's list of adverbs that appear in this position consists entirely of temporal adverbs: ariamh/riamh 'ever', go mink 'often', tamallfada 'long time'. We suggest that these adverbs are not adjoined to VP, rather are adjoined to some segment of TP (probably an AspP (see Adger, this volume, for discussion)) that is dominated by TP and dominates AgrO {contra Ramchand 1993a and Guilfoyle 1993), thus are adjoined higher than AgrO. This is confirmed by the fact that these adverbs appear higher than (and seem to take scope over) aspectual particles: (ii)

Bhi

na sealgairi tamall fada ag ADVERB

ASP

amharc

orthu

V

be.PAST the hunters long-time PROG watching them 'The hunters were watching them for a long time' (data from McCloskey, this volume) We suggest that the structure of a sentence like (i) would be something like: (iii)

[AGRSP V [TP Subj [ASPp Adverb [ASpp ASP UGROP Object [VP tsuhJ tvcrhj tobj]]]]]]

See Harley (1993) for more discussion. 14 See McCloskey (this volume) for an alternative, but related, view of syntactic features for Irish. 15 Note that we are using 'strong V-features' somewhat loosely here. If only features of targets can vary in strength as proposed in Chomsky (1993), and not features of the heads which undergo movement (as in the text here), then 'strong V-features of tense' should be taken to mean that whatever set of features conspire to force T to raise to Agr in English 'independently', their make-up is different in Irish. For more on the difference between independent raising of T to AgrS, and such raising as a part of the head chain raising, and in particular an explanation of how such raising renders the specifier of TP unavailable, see Bobaljik and Jonas (forthcoming: section 5). 16 We have claimed here that the N features of Agr are weak. We have done this so that there is no requirement that the subject raise overtly to the specifier of AgrSP. In doing this, the reader may have noticed, we have eliminated the trigger for object shift. This is

240

Jonathan David Bobaljik and Andrew Carnie

a problem we do not attempt to resolve here. However, the behaviour of these NPs is suggestive that the assumption the two agreement projections are identical may need revision. We will not pursue this here but leave it for future research. For some ideas, see Ouhalla (1991a, 1994), McCloskey (this volume) and Carnie (forthcoming). 17 Duffield (1991) claims that this option is not available for Irish. To our knowledge, he is incorrect in this regard. According to 6 Siadhail (1989), use of agreement is available in the speech of older speakers, especially in the Ulster dialect. Younger speakers tend to prefer using an overt pronominal and the default aL, but both forms are found. In prescriptive grammars and formal registers, the form with no overt nominal and an agreement particle is preferred. 18 Recall from section 3.2 above that in the southern dialects one only gets overt subjects in transitives when the object is postverbal and genitive (i.e. one gets either OV or SVO, but never SOV). We thus limit the discussion here to the northern SOV dialects.

8

Subjects and subject positions in Irish

James McCloskey

1 Introduction There now exists a consensus of sorts concerning the analysis of VSO clause structure of the kind found in Celtic languages. The consensus maintains that in this VSO type, the subject occupies a specifier position lower than (and therefore to the right of) the head position occupied by the finite verb. In this respect such languages contrast with the well-studied SVO languages in which the subject occupies the specifier position of the head which hosts the finite verb. SVO

FP

VSO

Sub]

Within this broad consensus, three principal strands of disagreement emerge: 1 the issue of what head position the finite verb occupies; 2 the issue of what lower specifier position the subject occupies; 3 the issue of what mechanism makes the difference between VSO languages of this type and SVO languages. The present chapter 1 is part of a larger effort to provide answers to these questions for Irish. It assumes the results of two earlier papers (McCloskey 1991b, forthcoming) which tried to establish an answer to the first issue in establishing that verb movement in Irish is not to C°, but rather to the highest inflectional position only. If this is granted, then the range of answers which can be given to question 2 narrows - the subject must occupy either the internal subject position (within VP), or else the specifier position of one of the inflectional projections that occupy the space between V° and C°. This second alternative has the interesting property that it becomes available only if one assumes that INFL is phrase-structurally complex (in the way argued for by Pollock (1989) and in much subsequent work). If the second alternative is right, then something like Pollock's view must 241

242

James McCloskey

be right. My purpose here is to try to provide an answer to question 2 and to explore the implications of that answer for question 3. Almost all recent work on Irish assumes that the correct answer to question 2 is that the subject (in Irish) occupies the VP-internal position at all visible levels of representation (apart from the papers already cited, see Chung and McCloskey 1987; Koopman and Sportiche 1989, 1991; Guilfoyle 1990, 1993; Duffield 1991; Chomsky 1993). An alternative view is considered by Koopman and Sportiche (1991: 232-5) and defended explicitly for Welsh. A similar view is presented in Bobaljik and Carnie (1992). They argue that in finite clauses, V° raises to the highest inflectional head position but that the subject also raises specifically, that it raises out of VP into the specifier position of one of the lower inflectional projections. I argue here on a number of grounds that this latter position is closer to being correct.2 This chapter argues that if this view is accepted, then an explanation becomes available for certain aspects of Irish syntax which in other perspectives seem rather eccentric. The (apparent) eccentricities in question centre on constructions involving null (non-thematic) subject positions. In theoretical terms, the issue most at stake in this discussion is the status of the (second clause of the) Extended Projection Principle (Chomsky 1982, 1986a) - the principle which requires that subject positions be structurally realized. With the advent of the Internal Subject Hypothesis, it became crucially unclear what position this principle referred to. Is it properly construed so as to require that the internal subject position always be structurally realized (as suggested, for instance, by Sigurdsson 1991: 348-51)? Or is the requirement that there be a structural subject to be construed rather as a property of functional projections, as suggested in Chomsky (1993)? The picture becomes muddier still in the context of Pollock's proposals. If there are many inflectional projections between C° and V°, which one of them, if any, is privileged by the Extended Projection Principle? I hope to show that the Irish data shed some light on these questions.

2 Unaccusatives 2.1 Salient unaccusatives

It is a distinctive feature of the lexical structure of Irish that it possesses a large class of verbs which appear in the kind of structure seen in (1). Some typical examples are presented in (2). (D

(2)

[IP(HN]

[PPPDP]]

a. Laghdaigh ar a neart. decreased on his strength 'His strength decreased.'

Subjects and subject positions

243

b. Mheadaigh ar a neart. increased on his strength 'His strength increased.' c. Bhreisigh ar an ghluaiseacht. increased on the movement The movement increased.' d. Chuirar an stoirm. put on the storm The storm increased (in fury).' That is, these are verbs which s-select a single argument, which mark that argument with what is traditionally taken to be a preposition and which seem to entirely lack a structural subject. The verb itself appears in the so-called 'analytic' form (McCloskey and Hale 1984), the finite form which encodes no information about person, number or gender. There are some fifty verbs of this type in the language that I know of. A listing is presented in appendix II.3 How are structures such as (2) to be analysed? The crucial initial observation is clearly that all the predicates of the type in (1) belong to the semantic categories characteristic of the class of unaccusative verbs. The largest class represented is the class of involuntary changes of state; there are also many versions of 'succeed' and 'fail'. These are both semantic types whose syntactic unaccusativity is well established for other languages.4 It will be convenient to have a name for this class of unaccusatives. I will for present purposes call them the salient unaccusatives, since, as I hope to show, the crucial property of these verbs is that they exhibit the characteristic properties of unaccusativity very clearly in their surface syntax. How is the semantic property of unaccusativity to be linked with the formal characteristics shown by these verbs? To answer this question, I would like to establish two central analytical points about the salient unaccusatives:5 1 Their single oblique argument is internal at all levels of representation; that is, the oblique argument does not raise to subject position. 2 Traditional grammars are right in taking the oblique arguments to be PPs and the morphemes which precede the nominal argument to be prepositions (rather than, for instance, Case-marking particles). As to the second issue, we can point to such unaccusatives as those seen in (3): (3)

a. D'eirigh idir na fir. rose between the men The men quarrelled.' b. Thosaigh idir na fir. began between the men The men quarrelled.'

244

James McCloskey

Idir is the preposition meaning 'between'. As such, it is subject to the kinds of selectional restrictions standardly associated with this element - that is, it must take either a plural or a co-ordinate complement: (4)

a. idir na bailte between the towns 'between the towns' b. idir Corcaigh agus Baile AthaCliath between Cork and Dublin 'between Cork and Dublin' c. *idir an teach between the house 'between the house'

The predicates in (3) are, unsurprisingly, subject to exactly the same restriction: (5)

Tosaionn idir me begins between me 'I quarrel.'

To explain this observation, it is not enough to appeal to the selectional restrictions associated with the semantic predicate 'quarrel'; this would still leave as a lexical accident the ungrammaticality of (6): (6)

Thosaigh idir Eoghan le Ciaran. began between Owen with Ciaran 'Owen quarrelled with Ciaran.'

(7)

Throid Eoghan le Ciaran fought Owen with Ciaran 'Owen fought with Ciaran.'

These observations, however, are entirely expected if idir in (3) simply is the preposition 'between'; if idir is a kind of Case-marker, they remain either mysterious or accidental. Consider, in the same light, cases such as (8): (8)

Theigh fa dtaobh don ghirseach. warmed about-the girl 'The girl became agitated.'

Irish has a large class of compound prepositions which are phonologically, morphologically and syntactically complex. Like compound prepositions in English (in spite of with reference to and so on), such elements have internal syntactic structure. The element/
Subjects and subject positions

245

assumption that Case-marking particles do not have internal syntactic structure, we have perhaps further reason for believing that the crucial selected elements in these unaccusative structures are prepositions and not Case-marking particles.6 There are in fact other (and probably stronger) reasons of a more theoretical character for believing that the oblique arguments of this class of unaccusatives are in fact true prepositional phrases. We will consider those reasons presently. For the moment, let us take it that there is here at least initial reason to judge that position plausible. The second analytical point has to do with the S-structure status of those oblique arguments. If we accept the consensus view (since Perlmutter and Postal 1984; Burzio 1986) that unaccusative predicates are predicates which take only internal arguments, then the oblique arguments of the salient unaccusatives will originate in internal argument positions.7 Should it be assumed that those arguments remain in their internal position, or should it be assumed rather that they undergo movement to become derived subjects? Case requirements would not force this movement, since if the conclusions of the preceding discussion are accurate, then the oblique argument will be Case-marked by its governing preposition. Even if the oblique arguments are taken to be DPs which are marked for an oblique Case (against the spirit of the discussion above), that Case is an inherent Case lexically governed by the verb, and movement would still not be forced by Case considerations. The only principle, it would seem, which would force movement would be the Extended Projection Principle - the requirement that subject positions be structurally realized. Pursuing the question of whether movement does in fact apply to the single internal argument of the salient unaccusatives, then, should reveal something about how the Extended Projection Principle operates in Irish, if it does. The possibility of taking the oblique argument in (2) or (3) to be a surface subject arises because of a fundamental fact about Irish word order - in finite clauses subjects appear in the immediate postverbal position. In simple cases such as (2) there is no obvious positional clue to tell us whether the oblique nominal is in subject position or in complement position (perhaps following an empty subject position). To be able to tell whether the single argument of a clause like (2) or (3) is a subject or a complement we must turn to other clause types and more complex constructions. As is well known, Irish is VSO only in its finite clauses. Small clauses (as in (9)) and non-finite clauses (as in (10)) both exhibit obligatory subject-initial orders (for detailed discussion and analysis, see McCloskey 1980, 1986; Chung and McCloskey 1987; McCloskey and Sells 1988; Guilfoyle 1990, 1993; Duffield 1991).

246

James McCloskey

(9)

a. Chonaic me [sc n a gasrai ag caoineadh]. saw I the boys cry.PROG 4 1 saw the boys crying.' b. Ba mhinic [Sc na gasrai ag caoineadh]. COP.PAST often the boys cry.PROG T h e boys were often crying.' c. Agus [sc e ag tarraingt ar an bhaile]. and him draw.PROG on home 'as he approached home' a. Nior mhaith Horn [Cp iad imeacht]. I-wouldn't-like them leave[-FiN] 'I wouldn't like them to leave.' b. I ndiaidh [Cp iad eisteacht liom] after them listen[-FiN] with-me 'after they had listened to me'

(10)

It then becomes interesting to ask where the oblique arguments of impersonal unaccusatives appear in such clause types. The answer is that they always and obligatorily appear in postverbal (complement) rather than preverbal (subject) position. This is illustrated for small clauses in (11), for non-finite clauses in (12). (11)

a. Braithim [sc ag teacht as fheithleoga]. I-feel come.PROG out-of sinews 'I feel sinews stretching.' (GLL 259) b. B'

fhada [Sc ag cailliuint ar a

COP.PAST long

(12)

IOSCPROG

mhisneach].

on his courage

'His courage had long been waning.' c. Agusgan [sc ag eiri leis sa phost nua]. and NEG rise.PROG with-him in-the job new 'And him not doing well in the new job' a. Caithfidh eiri leis. must rise[-FiN] with-it 'It must succeed.' (GLL 225) b. I ndiaidh fealladh air fiche uair. after fail[-FiN] on-him twenty time, 'after he had failed twenty times' (MO 071) c. N ' fheadfadh rith leo treis mo ghui-se. NEG could run[-FiN] with-them after my prayer 'They couldn't prosper after my prayer.' (NChAN 83)

Subjects and subject positions

247

The alternative order, in which the oblique argument precedes the verb, is strongly ungrammatical. Notice that it is manifest for such examples that they do not involve overt subjects, since the word order ambiguity characteristic of finite clauses is not in play here. It is clear, then, that at least in small clauses and in non-finite clauses, there is no movement of the oblique argument to subject position. The argument can be pushed a little further by considering the syntax of the progressive aspect. In a full clause (finite or non-finite), the progressive is expressed by means of the verb td ('be') and the aspectual particle ag (McCloskey 1983; Guilfoyle 1990; Duffield 1991): (13)

a. Ta na daoine ag pilleadh ar an bhaile. be.PRES the people return.PROG on home T h e people are returning home.' b. Tan luchair orm na daoine a bheith ag pilleadh ar an bhaile. be. PRES gladness on-me the people be[-FiN] return.PROG on home 'I am glad that the people are returning home.'

An analysis of the syntax of such constructions is presented in McCloskey and Sells (1988) which is built on the assumption (drawing on Stowell 1981) that the complement to td is always a small clause, whose subject raises to the subject position governed by td. In the case of small clauses headed by passive verbs, we have, presumably, a two-step movement - the internal argument moves first to the subject position within the small clause and from there to the subject position of the verb td. This is illustrated for a verb in the progessive passive in (14): (14a) illustrates movement within the small clause alone; (14b) illustrates the further,, movement which applies when the small clause is complement to td. (14)

a. Chonaic me [Sc na tithe sin dha dtogail t\. saw I those-houses raise.PROG.PASS 'I saw those houses being built.' b. Ta na tithe sin [Sc t dha dtogail t\. be.PRES those-houses raise.PROG.PAss Those houses are being built.'

I will not repeat the arguments for this analysis here, but I will assume its basic correctness. For the particular case of the progressive construction, we can assume that the small-clause complement of td is a projection of the aspectual particle ag or else of the verb itself. The subjects of (13a) and (13b) then (na daoine) originate in the subject position of the small-clause complement of td and raise to the higher subject position in the domain of td.s With this in mind, consider how the salient unaccusatives interact with the progressive. The crucial observation is that the oblique argument remains to the

248

James McCloskey

right of the progressive verb and that the subject position associated with td is null: (15)

a. Bhi ag neartu ar an nglor. was strengthen.PROG on the noise T h e noise was getting louder.' b. Bhi ag eiri reasunta maith Horn i liniocht. was rise.PROG reasonable well with-me in drawing 'I was doing fairly well in drawing.' (AThig 49) c. go bhfuil ag feabhsu ar eifeacht an teagaisc COMP is improve.PROG on effectiveness the teaching.GEN 'that teaching effectiveness is improving.' (DD 84)

If the oblique arguments of salient unaccusatives were subject to a raising requirement in finite clauses, then the pattern we would expect is that seen in the ungrammatical (16): (16)

a. *Bhi aran nglor ag eiri b. *Bhi liomag eiri reasunta maith i liniocht. c. *Ta ar eifeacht an teagaisc ag feabhsu.

The pattern seen in (16) is that which is found in other cases in which the internal argument of the verb of a small-clause complement to td must undergo raising. We have already seen this in the case of the progressive passive ((14) above). We can also see it for certain other predicate types. Consider the alternation seen in (17): (17)

a. Rinne siad ceoltoir brea de. made they musician fine of-him They turned him into a fine musician.' b. Rinne ceoltoir brea de. made musician fine of-him 'He became a fine musician.'

The obvious analysis of the alternation seen in (17) is to assume that the two predicates differ only in whether or not they have an external argument (AGENT or CAUSE). The verb in (17b) is, on this analysis, unaccusative in that it has only internal arguments (perhaps one internal argument, if the sequence of constituents following the verb is taken to be a small clause). This case differs from the salient unaccusatives in that the postverbal DP has no independent means of satisfying the Case Filter (in particular, there is no governing preposition to perform this function). We expect, then, that the postverbal argument in (17b) must have raised to the subject position of rinne, even though there is, again, no positional cue to indicate whether the phrase ceoltoir bred is a surface subject

Subjects and subject positions

249

or a surface complement. That being so, we expect this predicate to behave quite differently from the salient unaccusatives in its interaction with the progressive construction - the postverbal argument of (17b) should appear in the higher subject position. This is in fact what is observed: (18)

go raibh cabaire brea ag deanamh diom COMP was chatterer fine make.PROG of-me 'that I was becoming a fine chatterer' (FBF 33)

These contrasts fall into place naturally on the analysis being developed here, according to which the single argument of a salient unaccusative need not (and therefore within the framework of assumptions developed in Chomsky 1991, 1993, must not) raise out of its internal argument position. One final observation provides further support for this conclusion. It is possible to cleft the verb of a progressive construction along with its complements (McCloskey 1983): (19)

(Is) [ag togail tithe] a bhi siad /. COP.PRES raise.PROG houses COMP be.PAST they 'It was building houses that they were.'

Unsurprisingly, it is not possible to cleft the progressive verb along with its subject: (20)

*Is

na daoine ag imeacht a

bhi t.

COP.PRES the people leave.PROG COMP was

'It's the people leaving that were.' If the subject in a progressive clause has raised out of the small-clause complement of td, this contrast is expected. Compare now the behaviour of the salient unaccusatives. These verbs show the pattern of (19) rather than that of (20): (21)

a. Ag eiri

ar an leanbh a

rise.PROG on the child

bhi t.

COMP was

'It was becoming more agitated that the child was.' (LG 248) b. agus is ag teacht ann a bhi t. and COP.PRES come.PROG in-himcoMP was 'It was growing that he was.' (1/75) c. Ni ag dul den mheas a ta /. NEG go.PROG of-the respect COMP is 'It's not decreasing that this admiration is.' (DA 33)

250

James McCloskey

These facts suggest strongly that the oblique arguments of salient unaccusatives are S-structure complements of their verbs. To summarize, I hope to have established in the discussion so far that two statements hold true of the salient unaccusatives: (i) that their single internal argument is a prepositional phrase; (ii) that that argument remains in the internal argument position at S-structure (or before Spell Out in the framework of Chomsky 1993). There is a sense in which all of this is profoundly unsurprising. Given the existence of unaccusative verbs, as classically construed (in Burzio 1986, for instance), there is no reason why such verbs should not mark their single internal argument with a preposition. If prepositions are Case-assigners, then there is no Case-theoretic reason why the internal arguments of such verbs should raise to a position where they can be assigned Case. Saying that the internal argument need not raise is tantamount to saying that it may not raise, given the framework of Chomsky (1991, 1993). Even if this framework is not assumed, we are dealing here with a language in which preposition-stranding is not grammatical (McCloskey 1979). Given that this is so, raising of the DP-argument would be impossible. Nevertheless, some important questions obviously remain. There is, first of all, an analytical question internal to Irish. What is the status of the subject position of a clause built around a member of the salient unaccusative class? The canonical subject position in an Irish finite clause is the (specifier) position to the right of the head position occupied by the finite verb. For a case like (22), we have established so far that no overt element occupies this position (since the oblique argument of the unaccusative continues to occupy a complement position within VP): (22)

Mheadaigh ar luas na naomhoige. increased on speed the.GEN curragh.GEN 'The speed of the curragh increased.'

But what, then, is the status of the canonical subject position in (22)? Is it occupied by a null pleonastic (as argued by Travis 1984), or is it truly (syntactically) empty? If it is occupied by a pleonastic, is that pleonastic linked with an argument? If the pleonastic is linked, what is it linked with - PP or DP? Ultimately, these are questions about the Extended Projection Principle (Chomsky 1982, 1986a), since that is the principle which would force the existence of a (null) pleonastic element in such cases. A second question grows out of comparative concerns. The class of verbs we have been examining here is not a type whose existence has so far been reported in other languages. Why should this be, and why should verbs of this type exist in Irish? Before addressing these questions, we should deepen the puzzle a little.

Subjects and subject positions

251

2.2 Putative unaccusatives We have dealt in the preceding section with a class of unaccusative predicates which exhibit (from a typological perspective) a rather unusual cluster of properties. But Irish also possesses a class of unaccusatives which from a comparative perspective are more familiar. These predicates belong to the semantic categories typical of the unaccusatives; as far as their surface syntax is concerned, however, they appear to be indistinguishable from simple intransitives. Many of these verbs exist in systematic alternation with the salient unaccusatives: (23)

a. Neartaigh ar a ghlor. strengthened on his voice 'His voice strengthened.' b. Neartaigh a ghlor. strengthened his voice 'His voice strengthened.'

The alternation between the syntax of the salient unaccusative type and the syntax of a simple intransitive is characteristic of the verbs listed in (24).9 (24)

laghdaigh meadaigh breisigh lagaigh neartaigh treisigh gearaigh maolaigh

'decrease' 'increase' 'increase' 'weaken' 'strengthen' 'strengthen' 'sharpen' 'become blunt'

brostaigh leathnaigh moilligh tit claochlaigh feabhsaigh tromaigh ciunaigh

'quicken' 'widen' 'delay' 'fall/decrease' 'deteriorate' 'improve' 'become heavy 'become quiet'

I will call this class of predicates the putative unaccusatives. The name is designed to suggest that, although clearly unaccusative by semantic or notional criteria, these verbs bear no obvious formal mark of being different syntactically from any other intransitive. In particular, it is clear that in a simple intransitive use like (23b) the single argument of the verb is a surface subject. All the tests which yielded the opposite conclusion for the salient unaccusatives yield this conclusion for the putative unaccusatives. The salient unaccusatives and the putative unaccusatives contrast systematically with respect to all these criteria. In small clauses and in non-finite clauses, the single argument of a putative unaccusative appears in the canonical (preverbal) subject position: (25)

B' fhada [Sc a shaibhreas ag meadu]. COP.PAST long his wealth increase.PROG 'His wealth had been increasing for a long time.'

252

James McCloskey

(26)

I ndiaidh a shaibhreas meadu after his wealth increase[-FiN] 'after his wealth had increased'

In the progressive construction, the single argument of the putative unaccusative appears in the higher subject position (the subject position associated with td): (27)

Ta mo shaibhreas ag meadu. is my wealth increase.PROG 'My wealth is increasing.'

It is impossible to cleft the single argument of a putative unaccusative in the progressive along with the verb itself: (28)

*Is

mo shaibhreas ag meadu a ta /. my wealth increase.PROG COMP is 'It's increasing that my wealth is.'

COP.PRES

Notice that in every case the putative unaccusative contrasts in its syntactic behaviour with the salient unaccusative. These contrasts suggest strongly that the single argument of a putative unaccusative verb undergoes raising to a position where it can be assigned Case. Furthermore, this movement must apply before S-structure (before Spell Out). A final observation reinforces this conclusion. One of the ways in which subjects can be distinguished from complements in Irish is with respect to the distribution of resumptive pronouns. A resumptive pronoun may not appear in the highest subject position of a CP which contains the operator which binds it. This effect has been known as the Highest Subject Restriction, and is illustrated in (29). For discussion, see McCloskey (1979), Borer (1984b), McCloskey (1990), Varlokosta and Hornstein (1993). (29)

*ah fear a the

bhfaca se Una

man COMP saw

he Una

'the man that he saw Una' No such restriction holds of objects: (30)

na daoine ar mhairbh na robaluithe iad the people COMP killed the robbers them 'the people that the robbers killed'

In this respect also, the two groups of unaccusatives contrast: (31)

*an cnapan ar laghdaigh se the lump COMP lessened it 'the lump that decreased in size'

Subjects and subject positions (32)

253

an cnapan ar laghdaigh air the lump COMP lessened on-it 'the lump that decreased in size'

The contrast between (31) and (32) again suggests strongly that the internal argument of the putative unaccusative raises to subject position but that the internal argument of the salient unaccusative does not.10 There is nothing particularly surprising about this result. In the absence of a Case-assigner, the internal argument of an unaccusative must find some means to satisfy the demands of the Case Filter (however interpreted). One way in which this can happen is that the internal argument will raise to the canonical subject position, where it is assigned nominative Case in the routine way. Verbs which avail themselves of this option are those which we have called here the putative unaccusatives. For the class of verbs in (24), which show an alternation between the two patterns, we will assume that they optionally 1-select (in the sense of Pesetsky 1991: 10-13) the preposition ar. Optional 1-selection of a preposition is a feature of Irish lexical structure which is attested quite independently of current concerns. The verb comhairligh ('advise'), for instance, may take either a direct object, or it may 1-select the dative preposition do: (33)

a. Chomhairligh me do an teach adhiol. advised I to-him the house sell[-FiN] 'I advised him to sell the house.' b. Chomhairligh me e an teach a dhiol. him 'I advised him to sell the house.'

More needs to be said, however. To make sense of the data we have seen so far in this section, it is not enough to say that the internal argument of a putative unaccusative may raise. The result that must be secured is rather that raising to subject position must apply. In particular, we must rule out the possibility that the internal argument could remain in situ and form an expletive-argument CHAIN with a pleonastic element in subject position. That is, the expletiveargument linking structures familiar from much other work on the phenomenon of unaccusativity since Burzio (1986) must be unavailable in Irish. If this possibility is not ruled out, the observations we have documented in this section will be unaccounted for, since what they cumulatively indicate is that the single argument of a putative unaccusative must appear in subject position and may not appear in a complement position. To summarize, what we have seen so far is that unaccusative verbs in Irish appear in just two surface syntactic frames: (i) the internal argument may be realized as a PP, in which case it remains in complement position and the subject position is obligatorily non-overt; or (ii) the internal argument may raise to the

254

James McCloskey

canonical postverbal subject position. One unaccusative pattern which is systematically unattested is the type seen in the French and Italian examples in (34): (34)

a. II est arrive trois hommes. it has arrived three men Three men have arrived.' b. pro ne sono cadute mold, of-them are fallen many 'Many of them fell.'

In (34) a pleonastic pronoun (possibly null as in Italian) fills subject position and is linked with the unaccusative argument which remains in complement position. Before considering these patterns further and examining the questions raised by them, I want to show that they have a generality beyond the class of unaccusatives.

3 The perfective passive

One of the constructions in Irish which might deserve the name 'passive' is the construction I will call the perfective passive. This is illustrated in (35): (35)

a. Beidh an trachtas criochnaithe agam amarach. be.FUT the thesis finished at-me tomorrow 'I'll have finished the thesis/have the thesis finished tomorrow.' b. Ta teach ceannaithe agam. be.PRES house bought by-me 'I have bought a house.' c. nuair a bhi an meid sin deanta when COMP be.PAST that-much done 'when that much had been done'

This is a 'passive' construction in formal terms only. What I mean by this is that this construction has the syntactic characteristics familiarly associated with passives in many European languages. Specifically, the familiar pattern of alternations between active and passive variants concerning surface realization of arguments is found in this construction. The argument corresponding to the object of the active verb appears as a surface subject of the passive form; the argument corresponding to the subject of an active verb appears, in the perfective passive, as the object of a preposition (in Irish ag 'at'). Expression of this PP is entirely optional (see 35c). The passive verb itself appears in a non-finite form which in morphological terms closely resembles certain stative deverbal adjectives. In a full clause, this participial form is supported by the verb be

Subjects and subject positions

255

used as an auxiliary; in a small clause, the participle appears without any auxiliary: (36)

a. agus an meid sin deanta acu and that-much done by-them 'when they had done that much' b. Ba mhinic [Sc na rudai ceanna raite roimhe aige]. COP.PAST often the things same said before by-him 'He had often said the same things before.'

So far, we could be describing an English passive. The Irish perfective passive, however, has none of the rhetorical or discourse functions commonly associated with the passive in, for example, English. The perfective passive is simply the formal means used to express a particular aspectual category - a recent perfective or completive aspect.11 The basic characteristics of the construction can then be summarized as in (37): (37)

notional subject => object of preposition ag notional object =>• subject aspect =>• recent perfective The passive ag-phrase is systematically optional.

Given the extensive formal similarities between the Irish perfective passive and familiarly studied passives, I will assume here that essentially the same analysis is appropriate for both. That is, I will assume that derivation of the perfective passive involves the licensing of a syntactic structure essentially like that licensed by an unaccusative verb - one involving a non-thematic subject position and the suppression of the transitive verb's normal ability to assign Case to its DPcomplement. It will not matter for present purposes which theory of the licensing of such structures is adopted. What will be crucial is the assumption that perfective passive structures always involve non-thematic subject positions. In the typical case, the direct object of a transitive verb so passivized will raise to the nonthematic subject position. The fact that the direct object in a frozen or idiomatic phrase may appear in the subject position of a perfective passive, as in (38), suggests that it is correct to regard perfective passives as being derived by movement (rather than by a lexical process of some kind). (38)

a. Rinne se a chota ban. made he his white coat 'He made his fortune.' b. Ta do chota ban deanta agat. is your coat white made by-you 'You have made your fortune.' (FBF 222)

256

James McCloskey

The perfective passive is a productive category in all dialects of Modern Irish,12 but there are important differences among the dialects with respect to the range of verbs which support the construction. The examples presented in (35) and (36) all exemplify passivization of a basically transitive verb. Such cases are found in all dialects. In northernmost varieties, they are the only kinds of cases found. That is, these varieties incorporate a restriction like the familiar English restriction which bars passive from applying to verbs other than transitive (Case-assigning) verbs. Southern varieties, however (Munster varieties especially), show a rich array of impersonal forms in the perfective passive. In particular, perfective passives are formed from verbs which take a PPcomplement. Some attested examples of this type are presented in (39) (for full clauses) and (40) (for small clauses): (39)

a. Ta labhartha aige le cupla duine cheana. is spoken by-him with a-few people already 'He has spoken to a few people already.' (OTh 53) b. Bhiodh scriofa chuige roimh re. used-to-be written to-him in-advance 'He used to have been written to in advance.' (NChAN 76) c. Bhi cuinithe agam rot air. was thought by-me before-you on-it 'I had thought of it before you' (SAIL 276)

(40)

a. nior thuisce [Sc feachta aige air] no-sooner looked by-him on-it 'no sooner had he looked at it' (ALA 98) b. chonac [Sc beirthe aige ar laimh fir eigin eile]. I-saw gripped by-him on hand man some other 'I saw that he had gripped some other man's hand.' (AThig 137) c. agus [sc dearbhtha aige go cruaidh air fein] and accused by-him hard on-himself 'and he had harshly accused himself (F201)

It is perhaps worth stressing that the derivation of such examples13 is completely productive. There is, as far as I know, no verb which selects an external argument and a PP-complement which fails to form a perfective passive. In such cases, we have, in essence, the derived equivalent of a salient unaccusative, with an obligatorily empty subject position and a PP-complement.

Subjects and subject positions

257

Munster varieties also allow the free formation of perfective passive forms from a number of other intransitive verb types - from verbs which take optional internal arguments, for example: (41)

a. Agus sula raibh ite acu agus olta acu . . . and before was eaten by-them and drunk by-them 'And before they had eaten and drunk (OTh 100) b. Nuair a bhi criochnaithe againn when COMP was finished by-us 'when we had finished' (P68) c. do bhi faiscithe cho cruaig sin aige. was squeezed so-hard by-him 'He had squeezed so hard.' (SAIL 250)

More surprisingly, perhaps, perfective passives are also freely derived from salient unaccusatives: (42)

a. go bhfuil teipithe ar an rinceoir mor COMP is failed on the dancer great 'that the great dancer has failed' (BB 133) b. Bhi briste ar a fhoighid. was broken on his patience 'His patience had given out.' (FFF 337) c. go bhfuil eirighthe leis sa n-obair COMP is rose with-him in-the work 'that he has done well in the business' (SmBN \:40) d. Bhi rite leis fein. was run with-himself 'He had done well.' (SAIL 235)

and also from putative unaccusatives: (43)

nios faide . . . na mar bhi dulta aige farther than as was gone by-him 'farther than he had gone.' (OTh 14)

258

James McCloskey

(44)

ta tagaithe os bhur gcomhair agam. is come before-you by-me 'I have come before you.'

(45)

go raibh teite lena n-anam acu COMP was fled with-their soul by-them 'that they had fled for their lives'

(46)

ni fada a bhi siulaithe aige. was-not long COMP was walked by-him 'It wasn't long that he had walked.'

(47)

nuair a bhi bogaithe ag an la when COMP was softened by the day 'when the day had become milder'

(48)

nuair a bhi traite sios uaidh when COMP was ebbed down from-it 'when the tide had ebbed down from it'

(49)

go bhfuil trialtha cho maith agat COMP is proved so good by-you 'that you have proved (to be) so good'

(50)

conus ata iompuighthe amach aige how is turned out by-him 'how he has turned out'

(51)

ta culaithe ag anam naisiunta na ndaoine. is receded by soul national the people 'The national spirit of the people has receded.'

(52)

Cheapas aris go raibh seasta agam ar bhobghaiste. I-thought again COMP was stood by-me on booby-trap 'I thought again that I had stood on a booby-trap.'

(53)

nuair a bheadh preamhaithe i gceart aiges na crainn when be[coND] rooted properly by the trees 'when the trees would have rooted properly'

(L9)

(SCh 163)

(LA 36)

(cp SAIL 228)

(LA 36)

(SAIL 270)

(L 120)

(DD 33)

(SD 11)

(IC 61)

Subjects and subject positions (54)

259

Ach ni raibh tosuighthe i gceart fos ag an ngeimhreadh. but NEG was begun properly yet by the winter 'But the winter had not yet properly begun.' (U 265)

These properties of the perfective passive raise interesting questions about the proper understanding of passive-like operations in cross-linguistic perspective. We will not address those questions here, but concentrate rather on issues surrounding the status of non-thematic subject positions. The pattern just documented essentially recreates the pattern we have seen earlier for the unaccusatives. Two patterns emerge: 1 a pattern in which there is a necessarily non-overt subject; 2 a pattern in which the internal argument DP has undergone A-movement. Once again, what is unattested is the familiar pattern in which a pleonastic (null or overt) occupies subject position and is linked with an argument in complement position. If such a pattern existed it would be of the form in (55): (55)

[Ta pro, YPass DP, ]

Such structures are, however, completely ungrammatical: (56)

a. *Ta pro ceannaithe teach agam. be.PRES bought a-house by-me 'I have bought a house.' b. *Ta se ceannaithe teach agam. be.PRES it bought a-house by-me 'I have bought a house.'

The perfective passive, then, poses a set of questions very similar to those posed by the unaccusatives and has the further interest of demonstrating that those questions are not questions that concern unaccusative verbs alone, but reflect, rather, broader structural issues.

4 Expletives and the Extended Projection Principle Two questions arise, then, which I take to be related: 1 Why is the mechanism of expletive-argument CHAINS absent? 2 Why does Irish have the comparatively rare salient unaccusative type?

260

James McCloskey

The problem represented by question (2) was in fact anticipated by Luigi Burzio (1986: 73, note 9; 209-10, note 4). He points out that the general framework he developed might lead one to expect the existence of unaccusatives which subcategorize for a PP (i.e. verbs of the Irish type which we have called salient unaccusatives) but points out also that such verbs 'do not seem to exist' (209).14 He proposes that such verbs cannot in general exist because of the interaction of a number of factors. The Extended Projection Principle requires that subject positions must be filled, either with a pleonastic element or a raised argument. Movement of the internal argument will in the general case be impossible for verbs of the salient unaccusative type, given that most languages do not allow preposition-stranding. Therefore for a verb of this type to surface in legal structures, a pleonastic element would have to be inserted in the position privileged by the Extended Projection Principle. But pleonastic elements must be linked, and this linking will always be problematical for the case of verbs which subcategorize (only) for a PP. The class is unattested, then, because verbs so configured in the lexicon could never appear in legitimate syntactic structures. I want to maintain here that Burzio was essentially right in his proposals, that they account for the relative rarity of the salient unaccusative type, but that their occurrence in Irish is related to the distinctive character of its clause structure. Burzio's explanation for the non-existence of the prepositional type among unaccusatives depends crucially on the obligatory presence of an expletive element (whose presence is mandated by the Extended Projection Principle). I will assume that such elements must, in the general case, be linked with an argument. There are two principal theories available about how that linking can be achieved. Either an expletive-argument CHAIN can be formed at S-structure (Burzio 1986; Chomsky 1986), or else one can maintain that the expletive must be eliminated by the level of Logical Form (since it has, by definition, no function at that level), and that such elimination is accomplished by way of LF movement of the associated argument into the position occupied by the expletive (Chomsky 1986,1991, 1993). On either interpretation, no linking would be possible given an unaccusative which selected a PP-complement. It would seem to be a minimal requirement on CHAINS (in the sense of Chomsky 1986, which unifies A-movement chains and expletive-argument CHAINS) that each member be of the same syntactic category. If this is so, then the PP-complement of an unaccusative could not form a c H A i N with the expletive element and the expletive would fail to be linked. In the context of the theory of LF expletive replacement, the result is the same movement of a PP into a DP-position is ruled out by the structure-preserving character of the substitution operation. If expletives are required to be linked, then, the non-existence of the salient unaccusative type is expected.15 Or rather, it is expected in any language in which the appearance of expletive elements is mandated. If we can assume for a given language that their presence is not required, then a fortiori no linking is required and the existence of the

Subjects and subject positions

261

prepositional type (what we have called in Irish the salient unaccusatives) should be unproblematical. Specifically, if the canonical subject position in Irish is not a position which is privileged by the EPP, then there will be no requirement that it be filled by an expletive element. Elementary considerations of economy will further require that it not befilled.In the absence of an expletive element, there is no reason why the salient unaccusative type should not exist. Similarly, the various impersonal uses of the perfective passive which we considered in the previous section cease to be problematical. Passives of verbs which take PP-complements can be analysed in essentially the same way as the salient unaccusatives (they are, from a certain perspective, simply derived unaccusatives).16 Passives of intransitives are also unproblematical. Consider again one of the unaccusative cases: (57)

nuair a bhi traite sios uaidh when COMP was ebbed down from-it 'when the tide had ebbed down from it' (LA 36)

There seems to be little possibility of holding for such examples that there is an expletive subject and that this expletive is linked. If no expletive is mandated, however, then the question of linking cannot arise and examples such as (57) are not troublesome. We also have an explanation for the completely general unavailability of expletive-argument CHAINS for unaccusative and passive structures. If there are no expletives, there can be no expletive-argument CHAINS. Another expectation is generated if we pursue this line of analysis. It is well known that passive and unaccusative structures which do not involve promotion to subject typically show a sensitivity to the definiteness of the internal (unmoved) argument of the passive or unaccusative. Indefinites are unproblematical in the VP-internal position; definites sit less easily in that position. Now if, as seems likely, definiteness effects derive from properties of expletive-argument CHAINS (Safir 1985, Reuland and ter Meulen 1989, among many others), and if expletiveargument CHAINS are absent from Irish (because expletives are absent) then our expectation will be that there should be no definiteness restriction associated with passive or unaccusative structures in Irish - even in those cases in which the internal argument of the passive or unaccusative remains in VP-internal position. The observant reader will already have noticed that this expectation is in fact fully realized. None of the passive or unaccusative structures which we have been concerned with shows any trace of a definiteness restriction, even in structures where it seems clear that their internal arguments remain in VP-internal position. Definites, names and quantificational DPs appear freely in the internal argument position of salient unaccusatives and impersonal passives, as illustrated by many examples already cited to make other points.17

262

James McCloskey

To summarize the discussion so far: we have suggested that if it can be maintained that the Extended Projection Principle does not privilege the canonical subject position in Irish (if the Extended Projection Principle, so to speak, 'does not apply' in this language), then we can make sense of some aspects of Irish syntax which otherwise seem eccentric from a theoretical perspective and unusual from a comparative perspective. But of course the important task of making sense of this claim remains. What could it mean to say that a principle like the Extended Projection Principle 'does not apply' in a given language? Before addressing that question, we need to consider some data that might pose problems for the partial analysis so far developed.

5 Clausal complements and the internal subject position The potential difficulty I have in mind concerns structures involving clausal complements. Consider the case of verbs which have no external argument but which take clausal complements. As illustrated in (58)—(61), the situation we find here is that the subject position may be either empty or filled with the pronoun se (meaning 'it' or 'he'). 18 (58)

Caithfidh(se)go must

(59)

(60)

(61)

bhfuilse breoite.

it COMP is

he ill

'It must be that he is ill.' Tharla (se) go raibh siad ann. happened it COMP were they there 'It happened that they were there.' Thit (se) amach go bhfaca Sean Anna, fell it out COMP saw Sean Anna 'It happened that Sean saw Anna.' Ta (se) ar mo chumas tu a shabhail. is it on my ability you save[-FiN] 'It is within my power to save you.'

An identical pattern is found in the case of the perfective passive: (62)

Bhi (se) geallta agam do an teach adhiol. was it promised by-me to-him the house sell[-FiN] 'I had promised him to sell the house.'

If what we have said so far is roughly correct, then the pronoun se must not be an expletive in the sense in which we have been using the term so far in this discussion. There is, in fact, a fairly straightforward interpretation of these observations which is consistent with the analysis as developed so far, one that is very much in

Subjects and subject positions

263

the spirit of Bennis' (1986) exhaustive treatment of the corresponding question for Dutch. It is a feature of clausal complements in Irish that they must always be absolutely right-peripheral ( 6 Siadhail 1989: 270-1). It is clear too that this positioning must often be achieved by way of movement. This is indicated, for instance, by the following kinds of facts. Irish has a rule which postposes pronouns to right-peripheral positions (Chung and McCloskey 1987, O Siadhail 1989: 207-10). It applies, for instance, to small-clause subjects in the way seen in (63), where the relevant pronoun is in bold face: (63)

a. Chuala me iad ag beicfi le cheile sa dorchadas. heard I them roar.PROG to-one-another in-the darkness 'I heard them roaring to one another in the darkness.' b. Chuala me ag beicfi iad le cheile sa dorchadas. c. Chuala me ag beicfi le cheile iad sa dorchadas. d. Chuala me ag beicfi le cheile sa dorchadas iad.

In structures involving a clausal complement, however, when pronoun postposing applies, the clause must still appear to the right of the shifted pronoun: (64)

a. Chuala me e raite go mbiodh se ann. heard I it said COMP be.PAST.HABiT he there 'I heard it said that he used to be there.' b. Chuala me raite e go mbiodh se ann. c. *Chuala me raite go mbiodh se ann e.

The positioning of the clause in (64b) and the ungrammaticality of (64c) suggest in combination that clauses may not remain in complement position if that position is not right-peripheral, and furthermore that rightward positioning may be achieved by movement. If Chung and McCloskey (1987) are right in taking pronoun postposing to be a right-adjunction, then presumably the complement clause in (64b) is adjoined higher than the position to which the pronoun is itself adjoined. This displacement of clausal complements frequently results in a pronominal copy of the clause being left in its original position: (65)

(66)

Ta me cinnte de go mbeidh si i lathair. be.PRES I sure of-it COMP be.FUT she present 'I am sure that she will be present.' Bhi siad dha ra ar an nuacht g o . . . be.PAST they it say.PROG on the news COMP They were saying on the news that . . . '

264

James McCloskey

But appearance of the pronoun is entirely optional when other principles (whatever principle, for instance, lies behind the ban on preposition-stranding) are respected. Examples (66) and (67), for instance, are equally grammatical: (67)

Bhi siad ag ra ar an nuacht go . . . be.PAST they say.PROG on the news COMP They were saying on the news that . . . '

I suggest that the pronoun se which appears in subject position in the examples originally of concern to us (i.e. (58)-(62)) is to be identified with the pronouns in (65) and (66). That is, these are pronouns which mark the displacement of a clausal complement to right-peripheral position.19 Such pronouns occur regularly in rightward dislocations: (68)

Bhi se ina sheasamh os mo chomhair amach pataire do be.PAST he standing before-me out chubby person of bhuachaill a bhi chomh ramharle banbh. boy COMP be.PAST as fat as young-pig 'There stood in front of me a chubby little boy who was as fat as a young Pig' (FBF 208)

(69)

mar bhi si ar muineadh maistreas a bhi chomh Hath for was she. AGR:P1 teach.PROG teacher COMP was as grey le broc with badger 'For there was teaching us a teacher who was as grey as a badger.' (FBFU)

Many questions remain open here. However, it seems plausible that the phenomenon represented by (58)-(62) is to be understood in the context of a theory of rightward dislocations and the pronominal copies associated with them. If that is so, they need not stand as obstacles to the more general line of thought developed in this chapter. Other constructions also need to be investigated before we can be sure that the general claim made here (that Irish lacks expletives) is tenable - existentials such as (70), for instance: (70)

Ta daoine ann nach mbeannoinn doibh. be.PRES people in-it NEG COMP I-would-greet to-them 'There are people that I wouldn't say hello to.'

The existential construction is considered in some detail in Chung and McCloskey (forthcoming). It is shown there that this construction involves no expletive subject and that the element ann is in fact an existential predicate. The predicate

Subjects and subject positions

265

status of the element emerges clearly in small-clause structures, in which it unambiguously occupies the position reserved for the (obligatory) predicate: (71)

a. Is fada [Sc daoine ann nach ngeilleann do COP.PRES long people in-it NEG COMP yield to phiseoga]. superstitions There have long been people who don't give credence to superstitions/ b. 6 tharla [Sc daoine ann nach ngeilleann do since happened people in-it NEG COMP yield to phiseoga]. superstitions 'Since there are people who don't give credence to superstitions.'

If analyses such as these survive scrutiny, then we are fairly close to being able to conclude that expletives of the familiar kind and their associated syntax are absent from the grammar of Irish.

6 Interpretation If this is so, how might the patterns documented here be accounted for? The central task is to say why Irish should so thoroughly lack Extended Projection Principle effects and thereby allow the range of constructions (the salient unaccusatives in particular) whose syntax we have investigated here. In this final section, I would like to develop an account of this facet of Irish syntax. I want, furthermore, to take the analytical risk of assuming that the relevant property is closely connected with the particularities of Irish clause structure and its interaction with verbal morphology. There is, to begin with, an excellent reason why the internal subject position at least should never be implicated in EPP effects. Being the specifier of a lexical projection, it will be projected only when thematically required. In the theory of Chomsky (1993), the Extended Projection Principle reduces to the requirement that the agreement features and tense features of a verb be licensed by an appropriate nominal with which it is in the specifier-head relation. Checking of this kind is a property of functional, not lexical, projections and will never, therefore, require the realization of a specifier in a lexical projection. Safir (1992, 1994) provides important evidence for this conclusion by examining the distribution of expletive elements in small clauses in English, arguing that one finds expletive there in small clauses exactly when there is independent evidence for the presence of functional structure in that small clause.

266

James McCloskey

In the context of a set of assumptions like these, then, the absence of expletives within lexical projections is entirely expected. We must then consider the status of functional projections. The relevant question is this: what property of the inflectional system of Irish could produce the range of effects we have considered here? The account of Irish clause structure developed in much earlier work (McCloskey 1991b, for instance) is not of much help here. The assumption behind this work is that there are two modes of Case assignment; Case can be assigned either under government (downward and to the right), or else in the Spec-head configuration (see also Rizzi and Roberts 1989). Languages may 'choose' one method of assignment or the other for the assignment of nominative. If the government option is chosen, nominative is assigned downward and to the right from INFL, yielding a VSO language of the Irish type. If the alternative option is chosen, nominative is assigned upwards and to the left from INFL and an SVO language like English results. (72)

Complements

This view of Irish clause structure says essentially nothing about the set of problems we have been examining here. There is no reason why a structural difference in the mode of nominative assignment should be reflected in any difference in the way the Extended Projection Principle functions. If, for instance, the Extended Projection Principle reflects the obligatory character of the agreement relation, then one might expect different things depending on whether or not agreement also has a 'choice' about which structural relation it uses - the government relation or the Spec-head relation. If the former, then the internal subject position should show Extended Projection Principle effects; if the latter, then the specifier of IP should be obligatorily projected and filled, yielding structures such as (73):20 (73)

[Expy ^,yNj Sub, Complements]

But as we have seen, neither of these expectations is realized.

Subjects and subject positions

267

The proposal of Chomsky (1993) is more promising. On this view, the subject also remains within the lexical projection of V°. But this is not because of any permitted variation in the mode of Case assignment. In this conception, nominative Case, being a structural case, is licensed exclusively in the specifier position of the relevant (functional) head. The difference between Irish and English turns on whether the features implicated in the licensing of nominative are 'strong' or 'weak'. If the relevant features are strong, then the nominative DP which they license must appear in their specifier in visible syntax. If the relevant morphosyntactic features are weak (as would be the case in Irish), then they must be checked only in the covert part of the derivation, and the DP which will check those features must remain in its VP-internal position in overt syntax. So in (72), the nominative DP must remain in the internal subject position in overt syntax and must raise to the specifier of one of the functional projections in the covert portion of its derivation. We thus have VSO order in the visible syntax (since the verb, it is assumed, raises out of the VP in the overt syntax). What does this account imply for the distribution of expletives and for the potency of Extended Projection Principle effects in the language? Let us assume that the morphosyntactic features associated with subjecthood (nominative Case and agreement) are associated with some functional head F. These features must sooner or later be checked by way of the presence of a DP of the appropriate form in the specifier position of F. If, in a language like Irish, the features of F need not and therefore must not be checked in the visible part of a derivation, then projection of a specifier for F and the insertion of an expletive in that specifier position will never be required. Considerations of economy will therefore forbid the projection of the specifier position and the insertion of the expletive element. The morphosyntactic requirements of F can only, on this view, be satisfied by way of LF movement.21 This account has a number of merits. It makes available a connection between the EPP puzzles we have been concerned with here and the distinctive character of Irish clause structure, and it appeals only to parameterization in the setting of morphosyntactic features. This kind of difference among language-types is understandable in a way that a statement such as 'The Extended Projection Principle does not apply in Irish' simply is not. There is a difficulty, however. On this view (as on the first alternative we considered) the subject remains within the lexical projection of V° throughout the observable part of a derivation. This VP-internal position plays no role in featurechecking; rather it exists only to meet the thematic requirements of the verbal head. Feature-checking (both for Case features and for Agreement features) is accomplished in functional projections. Since we have just established that the relevant features must be weak (to account for the absence of expletive elements), then movement to satisfy the needs of the morphosyntactic features associated

268

James McCloskey

with subjecthood should only apply in the covert position of the derivation and its application should not be directly observable. But this cannot be right. The principal result that emerges from our discussion of the putative unaccusatives and the perfective passive is that Case-driven movement is just as obligatory and just as overt as it is, say, in English. (74)

a. Neartaigh ar a ghlor. strengthened on his voice 'His voice strengthened.' b. Neartaigh a ghlor. strengthened his voice 'His voice strengthened.'

In (74a), the internal argument remains within VP (because of the availability of the Case-assigner P°), but in (74b), the internal argument has undergone obligatory raising to the canonical subject position and is assigned nominative Case. Similarly in (75): (75)

a. Rinne sin leannan dinn. made that couple of-us 'That made us lovers.' b. Rinne leannan dinn. made couple of-us 'We became lovers.'

and also for the perfective passive: (76)

a. Ta se criochnaithe t againn. is it.NOM finished by-us 'It has been finished by us.' b. *Ta criochnaithe se againn. is finished it.NOM by-us 'It has been finished by us.'

The observation is quite general: movement driven by Case considerations is obligatory and overt in Irish, just as it is in English.22 These observations are fundamentally at odds with the theory just sketched in at least two distinct ways: 1 We have just been driven to the paradoxical position that the relevant morphosyntactic features are simultaneously strong (to force movement to the canonical subject position when that is required for Case purposes) and weak (to disallow the appearance of an expletive).

Subjects and subject positions

269

2 Movement into a lexical projection is in any case anomalous within the terms of reference of this theory. The various cases in which movement to the subject position is forced seem to have in common that they are Case-driven. But if we accept the terms of reference of Chomsky (1993), then such movement should never have as its target a position within a lexical projection. The feature-checking mechanisms which drive such movements have as their domain of application the functional projections rather than the lexical projections. Consider the second worry first. If the general conception of grammar presented in Chomsky (1993) is right, then the target position for the movements we have seen in this chapter should be within a functional projection above VP. That is, what we have been calling the 'canonical subject position' (the postverbal position in a finite clause) must be the specifier of a functional projection above VP, rather than being VP-internal. Interestingly, there is independent evidence that this is the correct view. The evidence comes from the distributional properties of a class of adverbials. It has usually been assumed that adverbs in Irish must always appear in rightperipheral positions and that they are as a consequence of little use as probes for determining what positions are occupied by the various major clausal constituents (McCloskey 1983; Ernst 1992). This is largely true. There is, however, a class of adverbs which have as one of their characteristic positions the position immediately following the subject but preceding all complements. Examples illustrating this possibility are presented in (77)-(80):23 (77)

(78)

(79)

a. Ni bhfuair aon bhean riamh roimhe greim laimhe air. NEG took any woman ever before-it grip hand.GEN on-him 'No woman had ever before taken his hand.' (CC 17) b. Nior smaoinigh me ariamh 6 shin air go dti anois. NEG.PAST thought I ever since on-it until now 'I never thought about it since until now.' (CC 148) c. Ni chluinfeadh aon duine choiche aris Ciaran ag gabhail cheoil. NEG hear.coND any person ever again Ciaran making music 'No-one would ever again hear Ciaran making music' Deireann siad i gconai paidir roimh am lui. say they always prayer before time lie[FiN] 'They always say a prayer before bed-time.' Chuala Roise go minic roimhe an t-amhran sin. heard Roise often before-it that-song 'Roise had often heard that song before.'

270

James McCloskey

(80) Nior

bhuail aon fhear amhain fos liom a bhfuil a chuid struck any man one yet with-me COMP is his share eadaigh ghlain air. NEG.PAST

clothes.GEN clean.GEN on-him

'I haven't yet met one single man who has his clean clothes on.' (AI6) Notice that the adverbs which appear in these examples are all of a type which are plausibly analysed (on semantic and on comparative grounds) as being adjoined no lower than VP. Since they appear to the left of complements of V°, it is plausible to believe that they are left-adjoined to VP (or to a projection higher than VP). It follows that the nominative subject must be outside VP. These observations suggest that Case-driven movement does in fact target a functional rather than a lexical projection. If this is so, then certain decisions are forced concerning the derivation of VSO order in Irish. We have just concluded that the subject must be in the specifier position of a functional projection which contains VP. But the finite verb must occupy a head position higher than this again. It must not, though, occupy the exposition, given the arguments of McCloskey (forthcoming). We are therefore dealing with two inflectional heads above VP: a higher one which hosts the finite verb, and a lower one whose specifier position is the nominative position. Movement into the nominative position is obligatory and overt. Therefore the lower head must bear a strong N-feature. The specifier of the higher head is never occupied in overt syntax; it must therefore have a weak N-feature: (81)

Complements

This much is dictated by basic design principles of the theory we have adopted here (essentially that of Chomsky 1993). But if it is right, then the EPP puzzles with which we have been struggling are also solved. It is a general truth about expletives, I believe (however it is to be derived), that they may not occupy a

Subjects and subject positions

271

position lower than the highest inflectional projection.24 What this means is that expletives would have to appear (if they were to appear) in the specifier position of the higher head in (81). But this is what we have just argued to be impossible, because of the weak specification of the N-feature of that head. This is the essence of the account I want to propose. The intuition which drives it will remain constant across a variety of possible implementations, I believe. Implementing it in a serious way will involve providing answers to at least the following questions: 1 Why may expletives not appear in the lower position? 2 What becomes of the lower projection (which, recall, is headed by strong features and therefore requires a filled specifier in visible syntax) in those impersonal constructions in which no nominative element appears (the salient unaccusatives, for instance, or the various impersonal passives considered earlier)? The answer to these questions is, I believe, intimately linked with the answer to a third: 3 What is the identity of the two functional heads between V° and C°? I would like to close by suggesting some (very tentative) answers to these questions. Duffield (1991) has presented evidence that (in Irish if not in general) the Tense projection properly contains the projection implicated in subject-verb agreement. That is, Fi in (81) is Tense; F2 is Agr (the head which hosts the subject-agreement element).25 Groat (forthcoming) has pointed out that it is a fully general property of expletive elements that they are invariant in form - that their form, in particular, does not vary by person, number or gender. Groat concludes from this observation that expletives are always placed in the specifier of the Tense projection, where sensitivity to phi-features is irrelevant. If this is correct, and if Duffield is also correct about the relative order of embedding of functional projections (in Irish), then we have an answer to question (1). Consider now question 2. The issue here centres on examples like (82): (82)

D' eirigh go maith leofa. TENSE rise.PAST well with-them 'They did well.'

What becomes of the strong features on F 2 in such cases? There is no overt nominative in (82), and we have argued at length that it does not contain a covert expletive either. How then can the relevant features of F 2 be licensed? The answer, I think, is that the lower projection (the F 2 projection of (81)) is not part of the structure in such circumstances. The question of how its features

272

James McCloskey

are checked therefore does not arise. The verb in such cases appears in the socalled 'analytic' form (McCloskey and Hale 1984). This is a form of the verb which is specified for Tense (arguably also Mood in some cases and under some interpretations), but not for any agreement features. Note that this is not a thirdperson singular form; it may appear with any kind of subject at all, as shown in (83). A 'synthetic' form of the same verb, which is specified for person and number features, is shown in (84): (83)

(84)

D' eirigh Ciaran/ na girseachai/me/muid. PAST rose Ciaran the girls I we 'Ciaran/the girls/I/we rose.' D' eiriodar pro. PAST rise.PAST.AGR:p3

'They rose.' If one takes seriously what the morphology seems to say about such forms, one would conclude that they are specified for Tense features only, but not for subject agreement features. That being so, the presence of an agreement projection is not required to license them. Further, since all arguments of the verb are independently Case-licensed in a salient unaccusative construction such as (82), no feature of a nominal complement will require the presence of an agreement projection. That is, there is no morphological checking function which would be served by the presence of such a projection. Finally, note that it is widely assumed that agreement projections must not appear in LF representations, since they serve no function at that level (Chomsky 1991). Putting all this together, what we see is that no grammatical damage will be done if F 2 of (81) simply does not figure at all in the derivation of a salient unaccusative or of an impersonal passive. Nor will any function be served by its presence. There is nothing particularly mysterious about this. A derivation begins with a particular set of lexical choices - with the selection from the lexicon of a 'numeration' in the terminology of Chomsky (1994). If F 2 (identified as the subject agreement element) is not so selected, no undesirable consequences for feature-checking or for interpretation ensue. If F 2 is part of the initial numeration (is selected from the lexicon), then its features (both verbal and nominal) will remain unchecked and the derivation will not converge.26 I conclude that in the case of the impersonal constructions we have been dealing with, only the higher projection of (81) (Tense by assumption) is ever present. If, however, the initial numeration includes a nominative noun (as part of a transitive structure, for instance), then F 2 must also be selected to license the nominative features of the subject; otherwise, convergence will be impossible. What this means ultimately is that the particular configuration of properties we find in Irish derives from two sources - the way in which VSO order is forced in

Subjects and subject positions

273

finite clauses (a weak specification on Fj, a strong specification on F2), and the very particular properties of its analytic inflection.27

7 Afterword We began this discussion by asking why the salient unaccusatives (and related impersonal constructions) can exist as a productive lexical class in Irish, given that they represent, in comparative perspective at least, something of an oddity. Our aim has been to show that the general theoretical principles which militate against their existence (in essence, those articulated by Burzio) are valid and exceptionless, but that the particularities of Irish clausal organization and verbal morphology interact in such a way that the effects of those principles are largely invisible in the grammar of Irish. These particularities have been traced ultimately to details of the ways in which morphosyntactic features are realized and checked in the language. This study can thus be seen as a contribution to the larger research effort which seeks to trace word order variation across languages to fundamentally morphosyntactic origins. The proposals made here are very different indeed from proposals that have been made concerning the analysis of other VSO languages. It is not the case, furthermore, that the properties we have documented for Irish here are properties of VSO languages in general. This observation is in harmony with the trend of recent work on VSO languages, which has shed great doubt on the idea that they might form a unitary class. Chung (1990) shows clearly that the entire line of analysis we have pursued here is not right for the VSO language Chamorro. Borer (1995) makes a strong case that the alternation between VSO and SVO orders in Modern Hebrew is a function of optional verb movement out of VP. Even within the Celtic family, different analyses seem to be right for different languages. Welsh and Breton are two of the more closely related languages within that family; yet, for Colloquial Welsh, the evidence that the subject is relatively high is strong (Awbery 1990) and, for Breton, the evidence is equally strong that the subject is always lower than negation and lower than the target position for Object Shift (Schafer 1994). The evidence for Modern Irish that the verb does not raise to C° is strong, in my opinion (McCloskey forthcoming); yet Carnie, Pyatt and Harley (1994) have presented interesting evidence that there is raising to C° in Old Irish (Irish of the eighth century). There is no 'VSO parameter', then, and no reason to expect that even relatively closely related languages should fall the same way with respect to such a parameter. Current theorizing includes a multiplicity of functional heads and specifier positions and as descriptive work proceeds, analyses accumulate which account for word order variation among languages by way of fine-grained distinctions concerning exactly how high in the extended projection of V° the verb

274

James McCloskey

raises, or in which of the various specifier positions within VP or above VP subjects, complements and other clausal constituents appear. If this work is on the right track at all, we are led to expect that there will be many, many different ways of arriving at a surface order of Verb-Subject-Complements. Current indications are that this expectation is realized.

275

Subjects and subject positions Appendix I

Data sources

The following are the sources from which examples have been quoted. In each case, the title, the author (or translator) and the major dialect represented by the language of the text are indicated. AI ALA AThig BB CC DA DD DM F FBF FFF G GLL IC LA L

Allagar na hlnise, Tomas Criomhthain. Kerry. Ar Leitheidi Aris, Padraig Ua Maoileoin. Kerry. A Thig Na Tit Orm, Maidhc Dainin 6 Se. Kerry. Bride Bhan, Padraig Ua Maoileoin. Kerry. Cora Cinniuna, Seamus 6 Grianna. Donegal. Deora Athias agus Dreachta Eile, Shan 6 Cuiv. Cork. Dochas agus Duaineis, Aindrais O Muimhneachain. Kerry. An Druma Mor, Seosamh Mac Grianna. Donegal. Fdnai, Sean Og O Caomhanaigh. Kerry. Fiche Bliain ag Fas, Muiris O Suilleabhain. Kerry. Fear Fior agus Fealltoir, Niall O Domhnaill. Donegal. An Gabhar Sa Teampall, Micheal Ua Ciarmhaic. Kerry. An Gealas i Ldr na Leithe, Padraig 6 Ciobhain. Kerry. Iomairi Criche, Padraig Ua Maoileoin. Kerry. Leoithne Aniar, ed. Padraig Tyers. Kerry. Lucidn, An tAthair Peadar 6 Laoghaire. Cork.

LG MO

NChAN NhAOTh

on P SAIL SCh SD SM

SThOC

SmBN

U

Le Gealaigh, Padraig O Ciobhain. Kerry. Muintir an Oiledin, Peadar O'Donnell, trans. Seosamh Mac Grianna. Donegal. Ndr Chios Ar Namhaid, Ger 6 Ciobhain. Kerry. Na hAird 6 Thuaidh, Padraig Ua Maoleoin. Kerry. 6 Thuaidh, Padraig Ua Maoleoin. Kerry. Peig, Peig Sayers, ed. Maire Ni Chinneide. Kerry. Seanchas Amhlaoibh I Luinse, ed. Sean 6 Croinin. Cork. Sean an Chota, Sean O Luing. Kerry. Sionnach Ar Mo Dhudn, Breandan 6 hEithir. Galway. Silas Marner, George Eliot, trans. Aindrias 6 Baoighill. Donegal. Scealta O Thomas O Cathasaigh, ed. Dubhglas de hide. Sligo. Sgealaidheacht as an mBiobla Naomhtha, Cuid a hAon, An tAthair Peadar Ua Laoghaire. Cork. Unaga, translated by Eoghan 6 Neachtain. Galway.

276

James McCloskey Appendix II

Salient unaccusatives

(1)

Laghdaigh ar a neart. decreased on his streng 'His strength decreased.'

(12) Mhoilligh ar a rith. slowed on his run 'His pace slowed.'

(2)

Mheadaigh ar a neart. increased on his strength 'His strength increased.'

(13) D'athiompaigh air. turned-back on-him 'He had a relapse.'

(3)

Bhreisigh ar an ghluaiseacht. increased on the movement The movement increased.'

(14) Thit ar lion na macleinn. fell on number the students The number of students fell.'

(4)

Chuirar an stoirm. put on the storm The storm increased (in fury).'

(15) Chlaochlaigh ar a neart. deteriorated on his strength 'His strength waned.'

(5)

Lagaigh air. weakened on-him 'He weakened.'

(16) Bhris ar a fhoighid. broke on his patience 'His patience gave out.'

(6)

Neartaigh ar a sciuch. strengthened on his voice 'His voice strengthened.'

(7)

Threisigh ar m' iarrachtai. strengthened on my attempts 'My attempts strengthened.'

(8)

Ghearaigh ar a choisceim. sharpened on his pace 'His pace quickened.'

(9)

Mhaolaigh ar a sail, blunted on their pace Their pace eased/slowed.'

(10) Bhrostaigh ar a siul. hurried on their pace Their pace quickened.' (11) Leathnaionn ar usaid na widens on use the Gaeilge. Irish The use of Irish spreads.'

(17) Chaill ar a mhisneach. lost on his courage 'His courage failed.' (18) Matheann ar an airgead. if goes on the money 'If the money gives out.' (19) Thrafadh ar an spiorad sin. would-ebb on that-spirit That spirit would wane.' (20) Feabhsaionn air. improves on-it 'It improves.' (21) Chruaigh orthu. hardened on-them They hardened.' (22) Thromaigh ar a n-anacair. became-heavy on their distress Their distress became harder to bear.'

Subjects and subject positions

277

(23) D'eirigh air. rose on-him 'He became agitated.'

(34) Leanannda cumhacht. follows of-her power 'Her power persists.'

(24) Scoilfidh ort. will-burst on-you 'You'll burst.'

(35) Ni thainig de. NEG came of-it 'Nothing came of it.'

(25) Chiunaigh ar ghibrisc na quietened on chatter the bpaisti. children The children's chatter died down.'

(36) Tagann fen urlar. comes under-the floor 'The floor is drawing damp.'

(26) Theiporthu. failed on-them 'They failed.'

(38) Ghormaigh aige. became-blue at-him 'He became depressed.'

(27) Chlis orthu. failed on-them 'They failed.'

(39) Chuaigh agam. went at-me 'I succeeded.'

(28) Sharaigh orthu. overcame on-them 'They failed.'

(40) D'eirigh go maith leis. rose well with-him 'He did well.'

(29) D'fheallorm. betrayed on-me 'I failed.'

(41) Rith leis. ran with-him 'He did well.'

(30) Chinn orthu. surpass on-them 'They failed.'

(42) Shroich leis. reached with-him 'He succeeded.'

(31) Mabheireann orm. if catches on-me 'If I'm in a fix.'

(43) Tiocfaidh as na bristi. will-come out-of the trousers 'The trousers will stretch.'

(32) Chuaigh da radharc. went of-his sight 'His sight faded.'

(44) D'eirigh eatarthu. rose between-them 'They fell out (quarrelled).'

(33) D'imighda neart. left of-his strength 'His strength gave out.'

(45) Thoisigh eatarthu. began between-them 'They fell out (quarrelled).'

(37) Dhubhaigh aige. became-black at-him 'He became depressed.'

278

James McCloskey

(46) Theigh fa dtaobh don warmed about-the ghirseach. girl The girl became hot and bothered.'

(47) Thainig insa ghasur. came in-the boy The boy grew up.'

Notes Because this work deals in large part with constructions found only in dialects which I do not myself speak, I have been more dependent even than usual on the help and expert advice of many friends and colleagues over the years in establishing the relevant data. I am grateful in particular to Breandan Feirtear, Brenda Ni Shuilleabhain, Liam 6 Murchu, Marian Ni Chiobhain, Breandan 6 Buachalla, Malachy McKenna, Liam Mac Con Iomaire, Liam Breatnach, Micheal 6 Siadhail, Angela Bourke and Diarmaid 6 Se. Sadly, three of those who provided help are now dead - Aindrias O Muimhneachain, Liam Bodhlaeir and Diarmuidin O Suilleabhain. For help and encouragement I am grateful to Sandy Chung. Discussions with Mark Baker, Andrew Carnie, Noam Chomsky and Ian Roberts in the later stages of writing were especially useful, as was a set of comments on an earlier version by John Frampton. The work reported on here was supported in part by the National Science Foundation, Grant No. BNS-9021398 to the University of California, Santa Cruz. Bobaljik and Carnie (1992) also claim that direct objects in Irish undergo obligatory Object Shift (to the specifier of the object agreement projection). I will not consider that issue here. Verbs of the type in (2) occur in all dialects, but not all of the verbs listed in appendix II are found in this pattern in all of the dialects. The type is much more productive in southern than in northern varieties. Thus, many of the verbs listed in appendix II occur in this usage only in southern varieties. This is part of a larger pattern of differences having to do with Null Subject Parameter properties. Essentially, southern dialects show such properties much more completely than do northern varieties (McCloskey 1991a). However, a small number of the collocations listed in appendix II (Teigh fa dtaobh do in (46), for example) are found only in northern dialects. The single case in the list for which this is perhaps not clear is the pair of verbs {tosaigh idir ('begin between') and eirigh idir ('rise between')) which both mean 'to quarrel, have an argument'. But it should be stressed that both of these collocations are strongly nonagentive in their semantics. Quarrels described with these verbs happen; they are not initiated. The possibility that these unaccusatives might be subject to an Icelandic-style raising of the oblique was suggested by Noam Chomsky in informal presentations some years

Subjects and subject positions

6 7

8

9

279

ago. Stowell (1989) has argued that at least some of what are called prepositions in Irish are actually case prefixes. For the particular cases we have discussed at least. But the syntactic points to be established below can be established on the basis of these cases alone. There is a set of issues concerning how the internal/external argument distinction should be captured in a framework which includes the Internal Subject Hypothesis. Certainly, the original assumptions of Williams (1980) on the issue cannot be maintained. For the purposes of this chapter, I will simply assume that internal arguments are those which originate in complement positions (daughters of X 1 ) and that external arguments are those which originate as sisters to X 1 . There is certainly more to be said about the issue than this, but these questions are not, as far as I can tell, crucial for present purposes. For a different view of the syntax of td, see Guilfoyle (1990), who takes it to be a 'light verb'. The main point we wish to establish here (concerning the analysis of the salient unaccusatives) can also be established in the context of her proposals, I believe. All of the verbs listed in (24) also have a transitive use. The full paradigm then consists of four possible syntactic frames in which the verbs may appear: (i)

(ii)

(iii)

(iv)

Mheadaigh increased 'My wealth Mheadaigh increased 'My wealth Mheadaigh increased 'I increased Mheadaigh increase 'I increased

ar mo shaibhreas. on my wealth increased.' mo shaibhreas. my wealth increased.' me ar mo shaibhreas. I on my wealth my wealth.' me mo shaibhreas. I my wealth my wealth.'

What produces this paradigm is the interaction between two possibilities: that the internal argument may or may not be marked with a preposition (usually ar 'on'); and that there may or may not be an external argument (AGENT or CAUSE). If there is an external argument, then the verb is a Case-assigner (is transitive) in accord with Burzio's Generalization (Burzio 1986). If there is no external argument, we have either a salient unaccusative (in which case there is no raising of the oblique argument) or an unaccusative of the more familiar kind, with obligatory raising of the internal argument. 10 There is another difference between the two classes of unaccusatives which is perhaps worth mentioning. Subjects in Irish must be adjacent to the INFL/V complex which is responsible for the assignment of nominative Case (McCloskey 1991b: 260-4; 291-2). This requirement holds for the putative unaccusatives: (i)

*Laghdaigh de reir a cheile se decreased gradually it 'It gradually decreased.'

280

James McCloskey

but not for the salient unaccusatives: (ii)

11

12 13

14

Laghdaigh de reir a cheile air. decreased gradually on-it 'It gradually decreased.'

This contrast too is unsurprising given the general framework developed here, since, in terms of their position and in terms of Case assignment, there is no difference between the subject of a putative unaccusative and the subject of any other verb. For the salient unaccusatives, on the other hand, the internal argument, we have argued, remains in complement position throughout. There is a sizeable body of philological work on the perfective passive. This work has been largely preoccupied with defining exactly what the aspectual force of the construction is and what its historical sources are. We will have nothing to say about these questions here. For discussion, see Zimmer (1901), Sjoestedt-Jonval (1931), O'Rahilly (1932/72), Dillon (1941), Wagner (1959), Hartmann (1974), Greene (1979), 6 Se (1983, 1993). Contrary to Greene (1979: 140-1), where it is claimed that it is marginal in Ulster Irish. Examples like (40b) are close to untranslatable in English. The lack of impersonal passives and the marginality of expletives in complement small clauses mean that one has to resort to a translation using a //^/-complement, as in (40b). It is important to stress, however, that the Irish example has the semantics of direct perception usually associated with perception verbs in their use with small-clause complements. The interpretation of an example like (40b) is not the indirect perception or propositional interpretation associated with perception verbs when they have a full CP-complement. See also (1 la) above. The existence of this example type has interesting implications for the analysis of perception verbs and their complement syntax. It would take us too far afield, however, to explore those questions here. Burzio (1986) mentions just one verb in Italian which might be of this type: (i)

Si tratta di suo figlio. REFL treats of your son 'It's about your son.'

French similarly has s'agir de and German es geht urn, all meaning 'it's a question of or 'it's about'. There remains, though, an important difference between these isolated examples and the systematic lexical class represented by the Irish salient unaccusatives. Giuseppe Longobardi further points out for the German idiom that the es survives in subordinate clauses (Ich glaube dafi es um . . . geht), indicating perhaps that this predicate does not involve a true non-thematic subject position. 15 The logic of the argument implies that an unaccusative could select a PP-complement only if it also selected a bare DP-complement (which could be linked with a required expletive). Such unaccusatives are in fact known to exist (Baker 1993). 16 A difficulty, though, is that such passives do not seem to be as rare as the salient unaccusatives. If passive structures always involve small clauses, then perhaps what happens in such cases is that the small clause is associated with a pleonastic, as proposed in Shlonsky (1987). If this view is right, then Irish continues to contrast with

Subjects and subject positions

17

18

19

20

281

other languages in permitting impersonal passives of the type in (39) in small clauses (as shown in (40)), even though there is in this case no possibility of linking an expletive within the small clause with another element. See Chomsky (1991), however, for arguments against this view. Vikner (1991) suggests that in such passives, the expletive is linked with the passive morpheme itself, in the context of a general analysis of passives along the lines of that presented in Baker, Johnson and Roberts (1989). Neither of these alternatives seems to be entirely satisfactory. There is an important difference between the Irish case and others, however, that must be relevant: Irish forms impersonal passives of this type completely productively. In all other cases known to me, there are considerable lexical restrictions and considerable variability from speaker to speaker. There is an alternative view of the interaction between Case-assignment and definiteness which grows out of the work of Adriana Belletti (1988) and is now widely adopted. According to this view, unaccusatives have the ability to optionally assign 'partitive' Case to their complements. This Case is incompatible with definite arguments and hence, on this view, when we see the definiteness effect we are seeing the effects of the assignment of partitive Case. The presence of an expletive in unaccusative and passive structures is forced by concerns quite independent of the Case requirements of the unaccusative argument; specifically, it is forced by the Extended Projection Principle. If one wanted to adopt this general view for the Irish data we have been dealing with, one would have to say, I think, that unaccusatives (and passives) in Irish systematically lack the ability to assign partitive Case. If this statement were not included in the grammar of Irish, then we would expect to find structures in which an unaccusative or passive verb assigned partitive Case to its complement (giving rise to a definiteness effect) and in which that complement remained in the internal argument position. No expletive need be generated (as in our earlier discussion) and we would have structures indistinguishable from structures involving an expletive-argument CHAIN in which the expletive is null. But these are just the structures which we have tried to show here do not exist. I take this to be an argument against the partitive Case approach to the analysis of the definiteness effect. Both options are available in all dialects. There is a small but important dialect difference, though, in that expression of the pronoun is much more common in northern varieties (McCloskey 1991a). This difference too seems to be related to the larger pattern of differences concerning null-subject properties alluded to earlier (compare note 3); see also 6 Siadhail (1989: 272-5). It is unclear whether such examples as (58)-(62) should be taken as derived by way of movement of the CP to subject position followed by rightward displacement, or if it should be assumed that only the pronoun raises to the subject position (like any bare DP argument of the unaccusative in need of Case) 'following' rightward displacement of the clause. Structures like (73) are attested in the transitive and unergative impersonal constructions of many Germanic languages (see Vikner 1991 and references cited there). Structures like (73) would also be expected if the EPP simply required the specifier of IP to be filled.

282

James McCloskey

21 Alec Marantz and John Frampton point out that this is not the standard interpretation of what it means for a feature to be weak. The standard interpretation would allow an expletive to appear in the specifier of a head bearing a weak feature. I will assume the stronger interpretation here: if the N-feature of a head is weak, then its specifier may not be occupied in overt syntax. 22 This is true even of raising, initial appearances notwithstanding. See McCloskey (1986) for detailed discussion. A conceptually very similar problem is considered by Chomsky and Lasnik (1993) in their discussion of accusative Case assignment in English. If assignment of accusative involves LF raising of the object into the specifier of a functional projection, then that movement is, they assume, an LF phenomenon. But then the kind of partial movement seen in (i) seems mysterious: (i) (ii)

I consider [him to have been treated / very badly], *I consider [to have been treated him very badly].

Chomsky and Lasnik's response to this dilemma is to appeal to the Extended Projection Principle. Non-finite clauses in English are as much subject to the requirement that they have structural subjects as are finite clauses. Expletives appear regularly in such structures: (iii)

I consider [there to have been too much trouble already].

It is perfectly reasonable, then, to conclude for such cases that it is the Extended Projection Principle which forces movement in the lower clause of (i). But in Irish, of course, we can appeal to no such solution since the burden of our discussion so far has been that the 'canonical subject position' in Irish (the target of movement) is not privileged by the Extended Projection Principle. In particular, expletives may not appear in that position. 23 One must be careful in using this test to allow for the interference of certain rightward displacements. Two such processes are relevant: pronominal objects in Irish displace rightwards (Chung and McCloskey 1987); and 'heavy' constituents are similarly subject to a tendency towards rightward shifting. The examples below are chosen so as to minimize the interfering effects of these phenomena. The material which appears to the right of the crucial adverb is not heavy, nor is it pronominal. This makes it less likely that rightward displacements are at work here. If rightward movements do not exist at all (Duffield 1994; Kayne 1995), then these issues need to be rethought. 24 See Chomsky (1994: 40-1) for a possible derivation. 25 See Ouhalla (1991a, 1994) for related discussion and a different view; see also Schutze (1993). If one adopted the proposals of Watanabe (1993b), and the 'Three-Layered Case Theory' in particular, which involves the postulation of an additional functional head above AgrSP but below CP, then one might identify F 2 as the subject agreement projection (as here) and F } with Watanabe's Y-projection. (One would also have to assume verb movement to Y°.) This interpretation has the consequence that one could maintain the now widely accepted view that the Tense projection is properly contained within the subject agreement projection.

Subjects and subject positions

283

26 It seems that something similar must be assumed for the case of intransitive structures in languages which have overt Object Shift (i.e. languages which have strong nominal features on the object agreement element). Non-finite clauses in Irish present very direct evidence that the interpretation suggested in the text is the correct interpretation for this case (McCloskey and Sells 1988). 27 If the view of finite-clause structure developed here is on the right track, then the ellipsis construction discussed in McCloskey (1991b) must involve not VP-ellipsis, but rather ellipsis of a higher functional projection. If antecedent-contained ellipsis is resolved along the lines suggested in Hornstein (1994), Takahashi (1993) and Lasnik (1993), then we must assume further raising of the subject in the covert part of the derivation. The analysis developed here also has implications for the analysis of small clauses in Irish (they cannot be lexical projections) and for the analysis of non-finite clauses. These questions I leave for future work.

Negation in Irish and the representation of monotone decreasing quantifiers Paolo Acquaviva

1 Introduction A noteworthy feature of Irish syntax is that a full-fledged negative sentence is required in order to express propositions which in other languages may be qualified as negative by the means of formally non-negative quantificational expressions, such as few people or only two friends in English.1 Thus, sentences like (la-b) must be translated as (2a-b), with a significantly different syntactic structure: (1)

a. John made few mistakes. b. John made only two mistakes.

(2)

a. Ni NEG

dhearna Sean moran dearmaid. made Sean many mistakes

b. Ni NEG

dhearna Sean achdha dhearmad. made Sean but two mistakes

The aim of this chapter is to relate in a principled fashion this observation with the apparent lack of a separate paradigm for negative quantifiers in Irish: (3)

a. John made no mistakes, b. Nobody will stop her.

(4)

a. Ni dhearna Sean aon dearmad arbith. NEG made Sean a mistake any ( = (3a)) b. Ni choiscfidh duine arbith i. NEG will-stop people any her ( = (3b))

These two characteristics receive a unified explanation by a general theory of sentential negation, according to which negatives like no friends and quantified NPs of the type of few friends or only two friends have essentially the same syntactic representation. The hypothesized phrase-structural uniformity under284

Negation in Irish

285

lying (2)-(4), moreover, affords a simple way to account for the illustrated crosslinguistic divide, without making recourse to language-specific stipulations about the structural representation of semantically homogeneous sentences. 2 Expressions of sentential negation 2.1 Negative markers and negated quantifiers To indicate that a sentence lies within the scope of a negative operator, languages make use of several types of overt marking. The most common strategy involves introducing a negative marker, which in English has two different forms depending on whether or not it is suffixed to the inflected auxiliary: (5)

a. John did not turn up. b. John didn't turn up.

Sentential negation can also be expressed by morphologically negative arguments or adverbials: (6) (7)

Nobody turned up. John never turns up.

As is well known since the classic study of Klima (1964), a sentence may be qualified as negative also by quantified arguments like few friends or by adverbials like seldom or scarcely, which lack an explicitly negative morphological characterization. Unless these phrases fail to have sentential scope, the sentence containing them behaves on a par with overtly negative sentences with respect to a number of diagnostics. The clearest shared properties are the availability of affirmative question-tags and of negative either-tags: (8)

a. Your friends said nothing, did they? b. *didn't they?

(9)

a. Few of your friends said anything, did they? b. *didn't they?

(10)

a. John seldom forgets, does he? b. *doesn't he?

(11) (12) (13)

John stayed quiet, and the others said nothing, either. Jack is very distrustful, and few people would trust him, either. John would attend all the meetings, and I seldom missed one, either.

For many speakers, in addition, a neither-idig can be appended to a sentence qualified as negative by expressions such as few N, scarcely or seldom: (14)

John said nothing, and neither did Jack.

286

Paolo Acquaviva

(15) (16)

Few actors really liked the play, and neither did many critics. I seldom missed a meeting, and neither did John.

The sort of diagnostics originally devised by Klima (1964) for English thus appear to single out a syntactic natural class of negative adverbials and quantifiers that can have scope over the whole sentence. However, some evidence indicates that the elements thus isolated are part of a broader class, defined by a common behaviour with respect to other tests. The expressions gathered together as 'negative' by Klima's tests are included in the class of so-called affective operators, which are able to license polarity items in an appropriate syntactic context: (17) (18) (19)

Nobody said anything. Few people said anything. John seldom admires anyone.

Apart from occurring in yes-no questions and after some 'inherently negative' expressions, polarity items are licensed by adverbials like only, even though the structural requirements in this case are different from those holding of other adverbials: (20) (21)

a. Only John had anything to say. a. John had only one thing to do. b. *John had only anything to do.

The inclusion of only in the class of affective operators is made particularly interesting by the observation that the same element patterns on a par with scarcely, barely, seldom and all other negatives in other respects as well: firstly, all these phrases can be preposed, triggering auxiliary-subject inversion: (22) (23) (24) (25) (26)

Nobody could he legitimately blame for the disaster. Never did I hear from him anymore. Scarcely anyone would John ever trust. Seldom did I forget an appointment. Only then did John realize what lay in store for him.

Secondly, when any one of the members of this class is embedded in a complement clause from where an adverbial phrase is extracted, the resulting sentence is marginal or downright unacceptable. In general, research in this domain since Ross (1984) has shown that an embedded negative sentence acts as an island with respect to the antecedent-government relation which links a moved phrase (typically an adverbial) to its trace; as argued by Rizzi (1990), referential arguments alone can overcome this induced opacity by establishing a binding relation with their traces. Thus, the extracted manner adverbials in (27)-(28) fail to

Negation in Irish

287

establish the required link with their traces, because the latter are embedded within a clause over which a negative operator has scope: (27)

a. *[How long] do you think that nobody could tolerate this ft b. * few people c. ?* only John

(28)

a. T h e tact [with which] no one but Jack could behave t is amazing. b. ?* only Jack c. * few other people

In conclusion, a single cluster of syntactic properties groups together elements qualified as negative by Klima's tests, as well as other affective operators like only. The result is that negatives proper (that is, elements qualified as negative by their morphology and/or by their interpretation) are not the only expressions which count as negative in the syntax.

2.2 Irish negatives

Unlike in English, it is not generally possible in Irish to qualify a sentence as within the scope of negation simply by the means of a negative morphological characterization of an adverb or of an argumental NP. Negation must instead be expressed as a clause-initial particle attached to the main verb, and indefinite NPs or adverbials within its scope may be marked as polarity items: (29)

Nior

iarr me gar ar bith. ask I favour at all 'I asked for no favour.'

NEG.PAST

(30)

Ni thiocfadh le haon duine a chloisteail. NEG could-come with any person his hearing 'Nobody could hear him.'

(31) Nil cead agat dul amachas an seomra seo ar fhath arbith. NEG.is permission at-you go out of the room here on reason any 'For no reason must you leave this room.' In the case of infinitival clauses, the verbal noun (which takes up the functions typically associated with the infinitive in other languages) is preceded by the particle gan, formally identical to the preposition corresponding to 'without'.2 (32)

Duirt se leis gan labhairt. said he at-himNEG. speak 'He told him not to speak.'

288

Paolo Acquaviva

As McCloskey (forthcoming) points out, phrases like X ar bith or aon X allow free-choice interpretation in the appropriate affirmative context, which highlights their similarity with polarity items like anyone or anything in English: (33)

D'fheadfadh rud arbith tarlu. could thing any happen 'Anything could happen.' (McCloskey's (103))

The descriptive generalization which emerges is therefore that, when sentential negation has a morphological realization, it cannot be expressed on adverbials or argument NPs, but rather appears as a negative prefix attached to the inflected verb or to the verbal noun. Interestingly, much the same state of affairs holds for the counterpart of those English sentences where sentential negation is not expressed by a morphologically negative phrase: (34)

Nil moran cairde aige. NEG.are many friends at-him 'He has few friends.'

(35)

Nior leigh se moran leabhair. NEG.PAST read he many books 'He read few books.'

(36)

Nil aige ach tri chara. NEG.are at-him but three friend 'He has only three friends'.'

(37)

Nior

leigh se ach deich leabhar.

NEG.PAST read he but ten

book

'He only read ten books.' An immediate extension of our descriptive generalization to all the sentential negative operators discussed in section 2.1, however, would amount to a gross oversimplification. In fact, it is possible to translate literally sentential negative operators like, for instance, only, scarcely and hardly? (38)

Is fanach a thagann se. is scarcely that comes he 'He scarcely ever comes.'

(39)

Is ar eigin ata fhios agam cad a dearfas me. is scarcely that-is knowledge at-me what that will-say I 'I scarcely know what to say.'

Negation in Irish (40)

289

Is ar eigin a bhi aon duine ann. is scarcely that was one person there There was hardly anyone present.'

Nevertheless, a common distributional restriction on amhain, ar eigin and fdnach makes it exceedingly unlikely that the kind of asymmetry between Irish and English shown in (29)—(31) and (34)-(37) should be caused by accidental lexical gaps in the Irish paradigm of negative operators. It is revealing, in fact, that such phrases may only occur with the function of sentential negative adverbials in cleft sentences, and never in the sentence-final position which is usual for other sentential adverbs: (38)

Is fanach a thagann se. is scarcely that comes he 'He scarcely ever comes.'

(41)

Tagann se go minic. comes he often 'He often comes.'

(42)

T a g a n n se fanach. comes he scarcely 'He scarcely ever comes'

(43)

a. Feicim anois e go minic. see-I now him often 'I often see him now.' b. *Feicim anois e ar eigin. see-I now him scarcely 'I scarcely see him now.'

Amhain, on the other hand, has a more liberal distribution, but when used as a nominal modifier it is more appropriately glossed as 'alone' than 'only'; as expected, in this meaning it cannot modify a plural NP. (44)

D'61 me aon pionta amhain. drank I one pint alone 'I drank only one pint.'

(45)

Nior 61 me ach aon pionta amhain. NEG.PAST drink I but one pint alone 'I only drank one pint.'

(46)

*D'61 me dha phionta amhain. drank I two pints alone 'I only drank two pints.'

290

Paolo Acquaviva

(47)

Nior 61 me achdha phionta (*amhain). NEG.PAST drink I but two pints alone 'I only drank two pints.'

Finally, the adjective annamh 'rare' can be employed as an adverbial with the meaning of 'rarely', 'seldom', both in cleft and non-cleft constructions; in the latter case, the aspirating particle go is prefixed, as for any other adjective used adverbially: (48)

Is annamh a labhrann Sean, is seldom that speaks Sean 'Sean seldom speaks.'

(49)

Labhrann Sean go hannamh. 4 speaks Sean rarely 'Sean speaks rarely.'

To recapitulate, sentential negation in Irish must be uniquely marked on the verb. Not unexpectedly, this extends to sentences which translate those English sentences containing adverbials of the type of scarcely or hardly, or quantified nominals of the form few N'. Those English sentences revealed as negative by their behaviour in connection with Klima's tests, therefore, must be translated as morphologically negative in Irish, with an overt realization of negation on the verbal complex. However, the picture is considerably complicated by the fact that the requirement that negation to be overtly realized is apparently suspended in the case of Irish cleft sentences; as examples like (38)-(40) show, it is possible to have adverbials corresponding to the English scarcely or seldom following the affirmative form of the copula. Furthermore, annamh 'seldom' represents an important exception to both generalizations, because it is not confined to cleft constructions, despite being semantically homogeneous with other 'negative' adverbs such as ar eigin.

3 The syntactic representation of negation 3.1 Evidence for LF movement Setting aside for the time being the issue of morphologically non-negative expressions, let us turn to negatives proper first. In a principles-and-parameters framework which incorporates Logical Form as a separate level of representation where (some) scope relations are expressed as relations of structural prominence, negative sentences are associated with an LF representation in which the negative phrase occupies an operator position having the whole sentence in its scope. This fundamental idea may be implemented by hypothesizing that expressions like nobody are raised from their S-structure position to a higher operator position

Negation in Irish

291

in the mapping to LF, by the same mechanism of quantifier raising which affects other quantifiers as well. However, while standard applications of quantifier raising are clause-bounded, negative quantifiers are able to take scope over the predicate of a superordinate clause. This point is illustrated by the following ambiguous sentence, taken from Klima (1964): (50)

I'll force you to marry no-one.

Independent evidence shows in fact that the scopal properties of negatives are obtained by abstract movement in the mapping of S-structure to LF; the trace left behind is predicted to trigger ungrammaticality in case it violates the licensing principle for empty categories (ECP). An embedded negative quantifier which has scope over the whole sentence is thus subject to the same requirements holding of traces. This explains the subject-object asymmetry illustrated in the following examples, originally pointed out by Kayne (1981): (51)

a. Je n'exige que la police arrete personne. 'I don't require that the police should arrest anybody.' b. *Je n'exige que personne soit arrete. 'I don't require that anybody should be arrested.'

The negative quantifier personne has to be interpreted with sentential scope, as indicated by the scope-marker particle ne. This is possible when it is in object position, as in (51a), but not from the subject position, because the latter fails to be properly governed and a trace in that position would violate ECP. The conjunction of these two observations, namely that negatives are subject to abstract movement and that this movement is in principle unbounded, has led to the natural conclusion that the type of syntactic movement instantiated by negatives is the same as that instantiated by w/j-phrases: in both cases, a phrase reaches by LF a non-argumental scope position, creating a possibly multimembered chain, whose links are connected by the relation of antecedent government.

3.2 NegP The conclusions outlined in the preceding section are largely motivated by the distribution of negative quantifiers. Negative markers like not in English or ne in French, as we have seen, cannot easily be assimilated to quantified expressions, and their fixed position in the clause suggests that they are a sort of scope marker. At any rate, the questions arise: what position is the target for the raising of negative quantifiers, and what structural relations hold between this position and the one occupied by negative adverbials and by negative markers?

292

Paolo Acquaviva

A unified answer is provided by the hypothesis that a negative sentence may involve the projection of an autonomous NegP phrase, which is the functional projection occupied at S-structure by those expressions previously qualified as scope markers, and to which negative quantifiers are raised at LF. This hypothesis has been put forth in several different forms in the recent literature (the main references being Moritz 1989; Pollock 1989; Laka 1990; Zanuttini 1991), and has been substantiated by a number of theoretical and empirical arguments. I will assume without further argument that this approach is substantially correct; my analysis will thus rely on the assumption that, in SVO languages like English, a functional projection NegP is generated in negative sentences between the two projections which express the inflectional features, AgrP and TenseP: (52)

AgrP Spec

A number of implicit phrase-structural assumptions deserve to be made clear in this connection. The most obvious is that two inflectional nodes are posited for the agreement and tense features (cf. Pollock 1989), where AgrP is regarded as the higher node (see Belletti 1990 and Chomsky 1991). As a consequence, AgrP is regarded as the topmost clausal projection (apart from CP), and is therefore claimed to be structurally represented in infinitival clauses as well as in tensed ones.5 Following the analysis of Belletti (1990) and Moritz (1989), I will also assume that the intervention of a functional head Neg° between Agr° and T° does not prevent the verb from raising from V° to Agr° to pick up its inflectional features in languages like Italian and, in certain contexts, French: (53)

[AgrP Agr° [NegP Neg° [TP T° [VP V . . . ]]]]

Notice that, in a structure like (53), it is highly unlikely that the verb could raise to Agr° by head-to-head movement via T° and Neg°, because there is ample evidence that the morpheme expressed on Neg° has a distribution essentially different from that of pure verbal affixes like tense and agreement.6 Hypothesizing the existence of a NegP projection between AgrP and TP is therefore problematic, because the raising verb would have to skip across the Neg° head on its way up, in violation of Relativized Minimality:

Negation in Irish (54)

293

a. Relativized Minimality X a-governs Y only if there is no Z such that (i) Z is a typical potential a-governor for Y; (ii) Z c-commands Y and does not c-command X. b. Z is a typical potential head governor for Y = Z is a head m-commanding Y. (Rizzi 1990: 7)

This problem is overcome by Moritz (1989) and Belletti (1990) with the observation that, if Neg° is cliticized to the complex Agr° head into which the inflected verb is incorporated, the two head-chains share the same head: (55)

Neg°-Agro-To-V°

*

1

[NegP t

'

[TP

t

It

[VP

|

t]]]

Since the two chains overlap on Agr°, Relativized Minimality is representationally respected.7 It will be noted that the raising of Neg° to Agr° is crucial in ensuring the wellformedness of verb raising. However, the process by which the expression of negation merges with the verbal inflection is apparently a widespread typological tendency, not just a by-product of verb raising; this is most clearly shown by French infinitivals, where ne is supposed to raise to an empty Agr°. The hypothesis that NegP is generated between the two inflectional nodes Agr° and T° partially makes explicit the common intuition that negation is associated with inflection (see, for example, Stowell 1989); however, it is still necessary to think that Neg° raises further up, in order to merge with the topmost inflectional node. This is the insight underlying the approach to Irish negation of Duffield (1991), who argues that, much as in other better-studied languages, sentential negation is expressed in Irish by a NegP projection generated between the two inflectional nodes. In his account, which I essentially follow here, the two inflectional nodes are ordered with TP higher than AgrP (see also McCloskey, this volume); the finite verb raises only as far as Agr°, while T°, if lexicalized by a [PAST] formative, is either cliticized on a higher C°, or surfaces as a potentially silent trigger of lenition: (56)

Silim go dtuigeann se think-I that understands he 'I think that he understands.'

(57)

Silim gur thuig se. think-I that.PAST understood he 'I think that he understood.'

294

Paolo Acquaviva

Negation, on the other hand, obligatorily raises to the highest inflectional node, be it C° or T°. (58)

Silim nach dtuigeann se. think-I NEG.that understands he 'I think that he doesn't understand.'

(59)

Silim nar thuig se. think-I NEG.that.PAST understood he 'I think that he didn't understand.'

Despite the considerable superficial differences between a VSO language like Irish on the one hand and SVO languages like English and French on the other hand, therefore, the syntax of negation converges in two important respects: evidence points to the existence of a separate NegP projection, and Neg° is attracted towards the highest inflectional node for reasons which are only in part traceable to morphosyntactic requirements. It has been suggested several times that the expression of negation needs to raise as far as possible in order to take scope over the whole clause. We will see in the next section that a reconsideration of some aspects of the NegP hypothesis leads to a more precise qualification of the 'semantic' origin of generalized Neg°-raising.

3.3 NegP and quantifier raising One of the main pieces of evidence in favour of a full-fledged NegP projection for negative sentences is the selective opacity displayed by clauses in the scope of negation: while A-binding relations such as those connecting a raised subject NP to its trace are generally not affected by the intervention of sentential negation, A-bar government relations are generally blocked in the same context: (60)

A misprint, appears [tt not to have been detected tt for years].

(61)

a. How strongly, do you believe [that the inflation will rebound tfil b. *How strongly, do you not believe [that the inflation will rebound /,]? (Rizzi 1990: 16)

In the Relativized Minimality framework, this effect is traced back to the intervention of an A-bar specifier between the trace and its antecedent in an A-bar position. The NegP hypothesis straightforwardly accounts for this effect, in that the NegP posited in negative sentences blocks A-bar antecedent government in virtue of its intervening A-bar specifier: (62)

. . . [ how strongly], . . . [Negp not Neg° . . . / , . . . ]

Negation in Irish

295

This approach to negative islands therefore involves the additional assumption that the Spec NegP position is filled even when it contains no lexical material. Such an assumption, for instance, is required by the English suffixed negation -n 7, or by clitic-like negative formatives in Romance languages like French ne and Italian non, all of which have been categorially interpreted as heads. The same holds in cases where sentential negation is not expressed by a negative marker at all, but rather by a negative quantifier or by some sentential adverbial. The question therefore arises how to account for the syntactic behaviour of such negative elements in terms of the NegP hypothesis. The analogy of negative quantifiers with quantifiers proper (like every student) on the one hand and with w/z-phrases on the other hand has led several researchers to claim that negative quantifiers are raised at LF in order to reach the scope position corresponding to Spec NegP, thus patterning on a par with whphrases in situ, which are likewise LF-raised to Spec CP in order to reach the scope position necessary for their interpretation.8 Thus, in a sentence like (63) below, the negative quantifier no secret is assumed to raise to Spec NegP at LF in order to reach its scope position; in the resulting configuration, the filled Spec NegP acts as an A-bar antecedent governor for the trace of the raised adverbial, so that the latter cannot be connected to its trace via government, because it is not the closest potential A-bar antecedent governor. (63)

a. I believe [that John can keep no secret for very long]. b. For how long do you believe [that John can keep a secret /]? c. *For how long do you believe [that John can keep no secret f]?

The NegP hypothesis thus provides an elegant way to unify the issue of the structural representation of negative markers (for which the head and specifier positions of NegP are in principle available) and the issue of the precise characterization of the raising undergone by negatives qua quantifiers by LF. This appealing approach, however, raises conceptual problems in connection with negative quantifiers in subject positions, like no-one in (64): (64)

a. No-one spoke. b. [Agr no-one [NegP [Vp spoke]]]

The subject quantifier must be deemed to undergo a downgrading movement from Spec AgrP to Spec NegP in order to reach their scope position; this assumption is problematic, because the trace left behind in Spec AgrP is not properly governed for the purposes of the ECP.

296

Paolo Acquaviva 3.4 Against negative quantifiers

The problematic treatment of subject negative quantifiers is not the only reason for reconsidering the issue. Another difficulty is raised by the Negative Criterion, a well-formedness condition which applies at LF and, in certain languages, at S-structure as well: The Negative Criterion (65) a. a negative operator must be in a Spec-head agreement configuration with a [ + Neg] head. b. a [ + Neg] head must be in a Spec-head agreement configuration with a negative operator. (Haegeman and Zanuttini 1991: 244) This representational constraint matches the JfTj-Criterion, which has been proposed in slightly different forms by May (1985) and Rizzi (1991a); since it can be unambiguously shown that the JF/z-Criterion holds as early as at S-structure in languages like English, it would be desirable for the Negative Criterion to have the same domain of application. However, this runs counter to the observation that Spec NegP is empty in negative sentences containing a negative quantifier, like / saw nothing. The solution that suggests itself is to allow Spec NegP to be filled by a silent operator, thus making an assumption which is independently motivated for other contexts (notably, yes-no questions). But, once this approach is taken, it is difficult to see what forces us to posit a raising (or lowering) of a negative quantifier to a Spec NegP position already filled by a presumably coindexed operator. There is in fact an implicit tension between the assumptions that Spec NegP is the scope position reached by negative quantifiers and that it is filled by an abstract operator. In the light of observations of this kind, it is argued in Acquaviva (1993) that the technical problems associated with the representation of negative quantifiers in terms of the NegP hypothesis are all related to the undisputed assumption that negatives are syntactically quantifiers, that is to say expressions which need to end up in a scope position by LF. This circumstance is all the more significant in that, from an interpretive viewpoint, all negative 'quantifiers' are semantically decomposable into negated NPs with a weak determiner, in the sense of Barwise and Cooper (1981);9 thus, even a quantified NP of the form neither N, where the determiner is semantically 'strong', can be analysed as a negated indefinite NP. 10 Combining these interpretive characteristics with the noted difficulties in the assumption of a systematic quantifier raising applying to negatives, it is proposed in Acquaviva (1993) to adopt the approach to indefinites inaugurated by Kamp (1981) and Heim (1982), in which indefinite expressions are taken to be associated with semantic variables. In this perspective, negative expressions like no man are indefinites unselectively bound by a negated existential quantifier, and not

Negation in Irish

297

quantifiers by themselves. This intuition is technically implemented by adopting the suggestion that the (higher) inflectional node in a clause is associated with an abstract existential operator, which by default closes the variables which are left open in its scope (cf. in particular Heim 1982 and Higginbotham 1985); the projection NegP is then assumed to express a Boolean negative connective ->, which by itself is not a variable-binding operator; the effects of sentential negation are derived by hypothesizing that this Boolean negative operator merges with the existential operator expressed in the inflectional node. The resulting negated existential acts as an unselective binder, which can close variables in its scope regardless of referential co-indexing. This approach raises several questions, which are addressed in detail in the cited reference. It is, however, necessary to briefly mention two relevant issues here. The first is the compelling evidence that negatives are subject to LF movement, and therefore do not simply behave like other indefinites.11 On the basis of the asymmetry between subject negatives and H'/z-phrases in situ I argue in Acquaviva (1993) that negatives do not raise in order to reach a scope position, but rather that movement is enforced in all and only those cases where a negative is not structurally close enough to the NegP projection. More precisely, it is claimed that a negative indefinite receives sentential scope if it is governed by a member of the NegP projection; this is accomplished either through head government by the raised Neg° head, for subject negatives, or by antecedent government in other cases. This is expressed by the following principle, which applies at LF: (66)

An indefinite morphologically marked [ + negative] is assigned scope by a governing [ + negative] operator.

Principle (66) is essentially an alternative descriptive generalization; it can be given explanatory force by noting that the government relation which is claimed to hold between NegP and negative indefinites is the fundamental local relation made available by the grammar. Moreover, the two relations of head government and antecedent government have been independently claimed to be involved in the other major operator construction, namely that involving w/j-phrases: on the one hand, McDaniel (1989) and Rizzi (1991b) have highlighted the role of government as local relation connecting links of an 'operator chain' in a particular instance of w/z-construction; on the other hand, Nishigauchi (1990) has argued that interrogative w/z-phrases are given quantificational force by the interrogative C° head under head government. If these converging suggestions are on the right track, then principle (66) may be regarded as a particular instance of a general mechanism for operator composition made available by the grammar. Secondly, this approach to sentential negation affords a different interpretation of the widespread tendency noted above for Neg° to incorporate into the highest inflectional node. Apart from morphosyntactic reasons, this general requirement

298

Paolo Acquaviva

may in fact be regarded as a consequence of the merging of the negative Boolean operator with the abstract existential closure, which I take to be expressed on inflection. This contention is made more precise by incorporating into the analysis the assumption that every verb is associated with an open event position, as argued in Higginbotham (1985, 1989). In this perspective, existential closure is not seen just as a default closure for indefinites which are not bound by any other quantifier (as in Heim 1982), but is rather a constant feature of any sentence. Sentential negation can therefore be defined as follows: (67)

Sentential negation = closure of the event variable by a negated existential operator.

The defining characteristic of sentential negation, namely that of having scope over the whole sentence, is thus phrased in precise structural terms; there is no need to postulate that an expression of negation should further raise at LF in order to reach a scope position structurally higher than the rest of the sentence.

4 Extending the NegP hypothesis 4.1 NegP without negative morphology Any analysis of negation which is not purely semantic faces the question raised by the apparent similarity between negatives proper and elements which, like scarcely ox few Nin English, can be shown to trigger the same effects as negatives without having their interpretation nor their morphological characterization. In concrete terms, the problem arises whether the manifold parallelism illustrated in the first section between these two classes is a consequence of their phrase-structural uniformity, and, if not, whether such a parallelism has a syntactic basis at all. The first thing to note in this connection is that a choice has already been made (implicitly at least), in so far as the island effect triggered by negatives has been explained in structural terms as a consequence of the NegP projection characterizing negative clauses. It is easy to see that, since the same kind of opacity characterizes both classes, NegP must be invoked in both cases; compare (68)-(69) below with the corresponding sentences in (61)—(63) above: (68)

a. How strongly, do you believe [that the inflation could rebound fj? b. *How strongly, do you believe [that the inflation could scarcely rebound //]? c. *How strongly, can you scarcely believe [that the inflation could rebound /,]?

Negation in Irish (69)

299

a. I believe [that John can keep few secrets for very long]. b. For how long do you believe [that John can keep a secret t]l c. *For how long do you believe [that John can keep few secrets t]l

Despite this straightforward parallelism, an extension of the NegP hypothesis to all the environments which are characterized as negative by Klima's tests is not immediately feasible. An autonomous NegP projection was originally proposed in order to provide a structural slot where negative markers could be generated as heads or specifiers of their own projection. Languages like French and Italian, in particular, make it clear that adverbials or negative quantifiers having sentential scope involve a negative characterization of the verbal inflection, which is explicitly marked by a negative particle. Nothing of this sort happens with the counterparts of seldom or few friends, however: (70)

a. John seldom makes mistakes. b. Jean fait rarement des erreurs. c. Gianni fa raramente degli sbagli.

(71)

a. John has few friends. b. Jean a peu d'amis. c. Gianni ha (ben) pochi amici.

To summarize, it would seem that the NegP hypothesis cannot be extended beyond the domain of negatives proper; yet a shared syntactic property such as the blocking of antecedent government indicates that positing two unrelated types of representation for negatives proper and for the other expressions which qualify as negatives by Klima's tests would miss a significant generalization.

4.2 Monotone decreasing quantifiers It is a familiar generalization since Ladusaw (1979) that the class of affective operators, defined by their ability to license polarity items in the appropriate syntactic configuration, is semantically homogeneous:12 all the members of this class are monotone decreasing quantifiers. These are characterized by the property of preserving the truth of a proposition (hence, allowing a correct inference) if the main predicate in the sentence is replaced by another, whose reference is a subset of the set referred to by the original one (see Barwise and Cooper 1981; Hoeksema 1986; Keenan and Stavi 1986). More precisely, determiners may be downward-entailing in either of two ways: left downward-entailing, if the truth is preserved whenever the left-hand argument A is replaced by another argument A' whose extension is a set smaller (or equal) in size to that referred to by A:

300

Paolo Acquaviva

(72)

D(A)

B

->

D(A')

B

A typical example for this type of determiner is every, which allows the inference from (73a) to (73b): (73)

a. Every student understood the question. b. Every good student understood the question.

More common are right downward-entailing determiners, where truth is preserved when the set referred to by B shrinks in size: (74)

a. No student understood the question. b. No student understood the question immediately.

Adopting the terminology of Barwise and Cooper (1981), I will refer to those quantified noun phrases with a right downward-entailing determiner as 'monotone decreasing quantifiers'. The inferential pattern is further illustrated by the following examples: (75)

a. Few people knew that story, b. Few people liked that story.

(76)

a. Only John could read Latin. b. Only John could read Latin easily.

In (74) through (76), the (a) sentence implies the (b) sentence, but the converse does not hold. Notice that the same applies to adverbials like seldom or scarcely: (77)

a. Jack seldom speaks. b. Jack seldom speaks clearly.

(78)

a. I could scarcely read your article. b. I could scarcely understand your article.

Monotone decreasing quantifiers are much rarer than monotone increasing quantifiers, which have the reverse inferential pattern: (79)

a. Some students understood the question immediately, b. Some students understood the question.

(80)

a. A few people liked that story, b. A few people knew that story.

(81)

a. John often speaks clearly, b. John often speaks.

Note also that every, which is a left downward-entailing determiner, forms a monotone increasing quantifier with the NP it qualifies:

Negation in Irish (82)

301

a. Every student understood the question immediately, b. Every student understood the question.

Both negatives proper and other expressions behaving alike with respect to Klima's tests, therefore, are monotone decreasing quantifiers. This semantic uniformity, by itself, is not particularly revealing, given that all expressions syntactically behaving as 'negatives' are also commonly qualified as semantically negative in some intuitive sense. However, it is important for our approach because of the systematic connection which can be established between indefinites and the whole class of monotone decreasing quantifiers, not just negatives proper. The 'monotonicity correspondence universal' proposed in Barwise and Cooper (1981: 186), in fact, states that 'there is a simple NP which expresses the monotone decreasing quantifier ->Q if and only if there is a simple NP with a weak noncardinal determiner which expresses the corresponding monotone increasing quantifier Q'. Thus, there do not appear to exist in natural languages basic determiners semantically analysable as 'negation - [strong determiner]', such as 'not every' or 'not most'. In turn, this generalization makes it possible to extend to all monotone decreasing quantifiers my proposal that negative expressions are not syntactic quantifiers, but rather indefinites combined with a negative operator expressed on NegP. 13 The following is an example of the proposed structure for a sentence characterized as negative (in Klima's sense) by a monotone decreasing quantifier: (83)

a. John has few friends. b. [Agrp John Agr°-Neg° [NegP OP t [TP T° [VP has [VP few friends]]}]

Since every monotone decreasing quantifier can be analysed as the negation of an indefinite, my proposal amounts to claiming that the syntactic effects shared with negatives proper are caused by the fact that NegP is projected in both cases. This analysis does not entail that monotone decreasing quantifiers (and negatives in particular) should simply be regarded as a sort of polarity items, which receive their interpretation only through binding from NegP; it is a trivial observation that such expressions maintain their negative interpretation independently of the presence of a NegP (for instance, when used as answers to constituent questions: Who came? Nobody I Few people). What is actually claimed is that sentential scope must be mediated through a sentential operator, realized on NegP. The function of the latter projection is that of making negated indefinites dependent on a single negative operator with sentential scope.14 This re-interpretation of the NegP hypothesis has therefore the crucial property of applying to cases in which a sentence including a monotone decreasing quantifier (like scarcely) cannot be immediately paraphrased by a formally negative sentence, with the schematic form -ip. A sentence like John could barely speak, for instance, may be paraphrased as It-is-almost-not-the-case that John could speak,

302

Paolo Acquaviva

but not as It-is-not-the-case that John could almost speak. In an approach that links NegP directly with the semantics of a negative sentence of the second form, this suffices to show that barely does not involve the projection of NegP. Such a conclusion is not necessary, on the other hand, if we regard NegP as the syntactic means by which barely receives sentential scope. In this perspective, the very name 'negative phrase' given to the hypothesized projection is misleading, in that it suggests a negative semantic value which is only appropriate with negatives proper. In Acquaviva (1993), in fact, I put forth several reasons for thinking that the appropriate feature characterizing NegP is not [ +negative], but rather [ +polarity reversing]; negatives proper, which are alone in requiring a special morphology and often an overt negative marker on NegP, are a subset of [ + polarity reversing] expressions, which are given sentential scope through local binding from NegP. The [ +polarity reversing] feature, in a sense, is thus the unifying feature of all those expressions which intuitively count as syntactically negative, including negatives proper. The foregoing analysis has shown that the set of expressions qualified as syntactically negative on the basis of their distributional similarities with negatives proper can receive a unified phrase-structural representation. This amounts to claiming that the uniform pattern considered in section 2 above has a basis in the syntactic representation; more specifically, in the LF representation, where it is held that government must connect NegP to a monotone decreasing quantifier. The definition of (syntactic) sentential negation given in (67) is therefore confirmed.

4.3 Monotone decreasing quantifiers in Irish It is important to emphasize that the conclusions arrived at in the preceding section are based on the syntax of languages such as English or French, where negatives proper and other monotone decreasing quantifiers constitute two clearly related but separate classes. If we take into consideration languages such as Irish, the evidence is at first sight contradictory. On the one hand, of course, our hypothesis that all monotone decreasing quantifiers are like negatives in involving a NegP projection is directly supported: English sentences of the form / saw only John or / saw few people are expressed as overtly negative sentences in Irish, with the form / didn 't see but John or / didn 't see many people', in the light of our previous analysis, we may simply account for this cross-linguistic difference by arguing that the abstract NegP hypothesized for English is spelt out in Irish. On the other hand, to claim that Irish lacks the entire class of nonnegative monotone decreasing quantifiers will be factually incorrect, as we have seen above in (38)-(40) and (48)-(49): adverbs like annamh 'seldom', ar eigin 'scarcely' or amhdin 'only' do not require the verb to be overtly marked for

Negation in Irish

303

negation, even though, with the exception of annamh, they only occur in cleft sentences (unless their interpretation is not of monotone decreasing quantifiers, as for amhdin; see (44)-(47) above). In order to understand what enables monotone decreasing quantifiers to appear in cleft sentences and in non-clefts containing annamh, it is necessary to focus on the core case first. A typical negative sentence has the following form: (84)

Ni lochtaionn Cait duine ar bith. NEG blames Cait person any 'Cait doesn't blame anyone at all.'

Adopting the view argued for by Duffield (1991), (84) has the following representation: (85)

Neg1

Spec

„.—•—-^^^

Neg° \,

ni

AgrP ^— Spec

Agr' -—-^ VP

Agr° lochtaionn Spec

1 i

Cait

V

v°

NP

t

duine ar bith

In all negative sentences like (84), Neg° must be overtly realized as an affix on the inflected verb. Notice that this requirement is different in nature from that displayed by languages like Italian or Standard French, where Neg° has likewise to be overtly expressed; the difference is that negation in Irish is uniquely expressed on NegP, and it must always be expressed there. In French, on the other hand, since personne is inherently characterized as negative, the overt realization of Neg° as ne is dispensed with in most (but not all) syntactic contexts, except in formal registers: (86)

II (n') admire personne. 'He admires nobody.'

Neg° in Irish, therefore, has to be overt in a stronger sense than in French or Italian: a null realization is never allowed. The relevance of this qualification becomes apparent when we turn to examine those negative Irish sentences whose English translation contains a non-negative monotone decreasing quantifier.

304

Paolo Acquaviva

(87)

Ni fhaca me ach Sean. NEG saw me but John 'I saw only John.'

(88)

Nior thainig moran daoine. NEG.PAST come many people 'Few people came.'

(89)

a. Ni mor a d'athraigh an ait. NEG.is big that changed the place 'The place has changed a lot.' b. Is beag a d'athraigh an ait. is little that changed the place 'The place has changed little.'

On the basis of the revised NegP hypothesis illustrated in section 3.4,1 claim that the structural representation of both the English and Irish sentences involves a projected NegP. The proposal, as is easy to see, seems to be immediately confirmed by the fact that NegP is overt in Irish. However, one important question remains: why is a negative marker on Neg° obligatorily present in Irish, but absent in English and, more importantly, French and Italian? In order to provide an answer, it must be noted that in neither language group is a monotone decreasing quantifier paired to a negative marker in NegP. In Italian, for example, where Neg° is overtly realized much more often than in French, the marker non is interpretively independent of any monotone decreasing quantifier; in more formal terms, we may say that negation identification is not allowed in such cases: (90)

Gianni non ha pochi amici. Gianni NEG has few friends 'Gianni doesn't have few friends. (He's got many.)'

The issue is further complicated by the fact that pochi may mean either 'few' or 'a few';15 thus, it can co-occur with negation in a construction like (91): (91)

Gianni non ha che pochi amici. 'Gianni doesn't have but a few friends.'

Monotone decreasing quantifiers may be built with adverbial modifiers like appena or si e no, 'barely' and 'scarcely': (92)

a. Gianni non ha appena tre amici. b. Gianni non ha si e no tre amici. Gianni NEG has barely/scarcely three friends Tt is not true that Gianni has scarcely three friends.'

Negation in Irish (93)

305

*Gianni non ha che appena/si e no tre amici. 'Gianni doesn't have but barely/scarcely three friends.'

The impossibility of negation identification between a negative marker in NegP and a monotone decreasing quantifier appears to be a general property, not subject to cross-linguistic variation (as usual, a claim of this sort is subject to verification). Given the invariant character of this incompatibility, in Acquaviva (1993) it is traced back to an interpretive mismatch between the function associated with the negative marker and that associated with the determiner of the monotone decreasing quantifier. Sentences like (87)-(89) show that Irish conforms to this generalization: negation is only expressed once, on the inflected verb. We never find the equivalent of 'I didn't see only John' or 'Few people didn't come'. Note that moron (XP) is not by itself a monotone decreasing quantifier, but rather the form taken by the corresponding monotone increasing quantifier ('many', 'much') under the scope of an affective operator; that is, either an interrogative operator in.yes-no questions or a negative marker. Recall that, as shown in (44)-(47), the behaviour of the modifier amhdin is only seemingly different; in particular, the impossibility of (46)-(47) above illustrates in fact that the modifier does not have the same function as the English only. On the other hand, amhdin is predicted to be acceptable in the sense of 'only' if it is not construed with sentential scope. This accounts for the common constructions in which amhdin has scope over a non-sentential constituent such as a verbless (or nominal) sentence, regardless of grammatical number: (94)

a. Don scrudaitheoir amhain. to-the examiner only 'For examiner's use only.' b. Mna amhain. 'Women only.' c. Seandaoine amhain. 'Old people only.'

A nominal sentence lacks both verb and inflection, and therefore there is neither an event variable nor an existential operator closing it. This means that the examples in (94) are not instances of sentential negation (in the sense specified by (67)), and amhdin is accordingly free to take scope over the whole expressions. As shown by the admissibility of a plural NP in (94b-c), here amhdin is a genuine adverbial, as opposed to a modifier interpreted as a cardinality predicate 'one'. The theory here advocated thus directly accounts both for the existence of this construction of amhdin with plurals and for its limitation to nominal sentences otherwise a very puzzling pattern. We can summarize our conclusions as follows: general independent reasons prevent negation identification from obtaining between a negative marker in

306

Paolo Acquaviva

NegP and a monotone decreasing quantifier; recall that this requirement has been posited independently of Irish data. In Irish, on the other hand, whenever NegP is projected Neg° must be lexically realized. As a consequence of these two conflicting conditions, what in other languages may be expressed by the means of monotone decreasing quantifiers (with an abstract NegP) is expressed in Irish as an overtly negative sentence.

5 Cleft sentences and annamh 5.1 Lack of parallelism with other sentential adverbials

Cleft sentences represent a systematic exception to the generalization reached in the preceding paragraph. Even though Irish lacks altogether lexical entries corresponding to determiners like few or little, adverbials like barely or scarcely do have their counterparts: (95)

Is ar eigin ata aithne agam air. is scarcely that-is acquaintance at-me on-him 'I barely know him.'

(96)

Is fanach afheicim anois e. is scarcely that-see-I now him 'I seldom see him now.'

As has been mentioned above in section 2.2, however, such expressions trigger ungrammaticality if they are embedded in the sentence in the positions usually occupied by adverbials: (97)

*Ta aithne agam air ar eigin. is acquaintance at-me on-him scarcely

The very existence of these adverbials and the strong constraints on their distribution make it apparent that the generalized lack of monotone decreasing quantifiers cannot be simply interpreted as an accidental lexical gap. We will now see that the seemingly puzzling distribution of such expressions is in fact accounted for by an approach to sentential negation along the lines of the theory illustrated above. An adverbial like ar eigin is sentential, in the simple sense that by its lexical meaning it qualifies propositions. This means that, were it generated in the canonical sentence-final position, it should be somehow connected to an operator having scope over the sentence, expressed by the autonomous NegP projection. The attested impossibility of sentences like (97) can therefore be regarded as a subcase of the general principle banning monotone decreasing quantifiers from being linked to an abstract NegP in Irish. In other words, ar eigin must have scope

Negation in Irish

307

over the event variable; this requires the presence of a NegP projection, as argued above; since the projection of a null Neg° is impossible in Irish, ar eigin cannot occur in sentence-final position.

5.2 Go hannamh The distribution of tags is, among Klima's tests, the only one which can be used to identify negative sentences in Irish. As is well known, some care is needed in applying diagnostics originally devised for English to other languages (see Ultan 1978 for a typological survey). However, it is a well-established generalization, clearly illustrated in O Siadhail (1973), that in Irish an affirmative declarative clause may be accompanied either by a positive or by a negative tag, depending on the illocutionary characterization of the sentence.16 (98)

a. Beidh tu ag imeacht amarach, an mbeidh? will-be you at leaving tomorrow Q will-be 'You're leaving tomorrow, are you?' b. Beidh tu ag imeacht amarach, nach mbeidh? will-be you at leaving tomorrow NEG.Q will-be 'You're leaving tomorrow, aren't you?'

However, only a positive tag may be attached to a negative declarative: (99)

a. Ni bheidh tu ag imeacht amarach, an mbeidh? NEG will-be you at leaving tomorrow Q will-be 'You're not leaving tomorrow, are you?' b. *Ni bheidh tu ag imeacht amarach, nach mbeidh? NEG will-be you at leaving tomorrow NEG.Q will-be 'You're not leaving tomorrow, aren't you?' ( 6 Siadhail 1973: 144^5)

These clear-cut judgements provide us with an unambiguous way to identify truly negative sentences. As it appears, when go hannamh is attached at the end of the sentence, a negative tag is perfectly grammatical: (100)

Labhrann Sean go hannamh, nach labhrann? speaks Sean rarely NEG.Q speaks 'Sean speaks rarely, doesn't he?'

Some informants actually find the positive tag somewhat deviant: (101)

?Labhrann Sean go hannamh, an labhrann? speaks Sean rarely Q speaks 'Sean speaks rarely, doesn't he?'

308

Paolo Acquaviva

I take this pattern of acceptability to indicate that the problematic sentences are not 'negative' in the relevant syntactic sense; both the full grammaticality of negative tags and the somewhat dimmer acceptability of positive ones are exactly what is expected of affirmative clauses. Despite its meaning, go hannamh must not be considered to have sentential scope in (100). As illustrated in Acquaviva (1993), such a conclusion is forced in several cases by certain monotone decreasing quantifiers, which must be taken as optionally (or preferably) co-indexed with a NegP projection. The most important observation is that the lack of a coindexed NegP projection hosting a negative sentential operator is a lexically conditioned option; that is to say, it cannot be invoked generally for all monotone decreasing quantifiers.17

5.3 Cleft sentences are not negative Let us now turn to the challenge posed to our theory by the grammaticality of cleft sentences containing monotone decreasing quantifiers. In this case it is not possible to simply assume that clefted adverbials may have non-sentential scope, as for go hannamh, because it would then remain unexplained why ar eigin is ungrammatical in sentence-final position. It is plausible, instead, that monotone decreasing quantifiers are allowed to occur in cleft sentences for structural reasons. Given that the adverbials under consideration qualify the whole proposition, any approach in which a sentence is simply defined as 'negative' when it is in the scope of a negative or [ + polarity reversing] operator would be unable to find a solution to the paradox; some assumption would have to be given up. In this connection, however, the definition of sentential negation proposed above proves more adequate: (67)

Sentential negation = closure of the event variable by a negated existential operator.

A definition like (67) has the essential advantage of distinguishing the property of qualifying a whole proposition from the formal property of having the event variable bound by a negated existential operator. On the basis of (67), I will now argue that adverbials like ar eigin are allowed in cleft sentences simply because such constructions are not, in the relevant sense, instances of sentential negation. The decisive piece of evidence in favour of my contention is the observation that the sentences under scrutiny clearly differ from unambiguously negative sentences in allowing negative tags:

Negation in Irish (102)

(103)

309

a. Ni labhrann Sean go minic, an labhrann? NEG speaks Sean often Q speak 'Sean doesn't speak often, does he?' b. *Ni labhrann Sean go minic, nach labhrann? NEG speaks Sean often NEG.Q speak 'Sean doesn't speak often, doesn't he?' Is annamh a labhrann Sean, nach annamh? is seldom that speaks Sean NEG.Q seldom 'Sean seldom speaks, doesn't he?'

If cleft sentences containing annamh, ar eigin or fdnach were instances of sentential negation, we would expect a negative tag to trigger ungrammaticality, or at least some deviance. This is plainly not the case: (104)

Is annamh a theann tu go Sasana, nach annamh? is seldom that go you to England NEG.Q seldom 'You seldom go to England, don't you?'

(105)

Is annamh a fheiceann Tomas anois e, nach annamh? is seldom that sees Tomas now him NEG.Q seldom 'Tomas seldom sees him now, doesn't he?'

Again, most of the informants consulted tended to judge a negative tag more acceptable, or more plausible, than a positive one: (106)

?Is annamh a theann tu go Sasana, an annamh? is seldom that go you to England Q seldom 'You seldom go to England, do you?'

The definition of sentential negation provided above in (67) crucially refers to the event variable associated with the main verbal predicate. According to (67), a sentence containing an adverbial like annamh or ar eigin would be negative if its event variable were closed by a negated existential. In the normal case, this is impossible because the existential operator binding the event variable would have to merge with an abstract Neg° - and this is not allowed in Irish. The only possibility, lexically restricted, is that the adverbial may be construed without sentential scope, that is to say, without being linked to a dominating NegP. The picture is different in a cleft sentence: here the main verb is not a lexical verb, but the copula. If the latter were associated with an event open position, on a par with lexical verbs, it should be closed by a negated existential operator, suitably coindexed with annamh or ar eigin. The resulting structure would not be different from that obtaining in non-cleft sentences, and ungrammaticality would be predicted. Since the structures are instead grammatical, our theory forces us to assume that the copula is not linked with an event variable. This forced conclusion is both intuitively motivated and semantically defensible (see, for

310

Paolo Acquaviva

instance, Moro 1991). The most interesting observation, however, is that it overlaps with the results arrived at by a completely independent path by Doherty (1992), who presents strong evidence for regarding the Irish copula as an inflectional particle, categorially well distinct from verbs. Example (107), adapted from Doherty's (94), illustrates the proposed structure, where the focused material is the complement of Agr° and the rest of the sentence occupies the specifier of AgrP on the right: (107)

Agr°

J annamh I | ar eigin j

The copula is thus analysed, syntactically as well as interpretively, as the spell-out of Tense features, connecting two expressions in a predication relation. In order to be interpreted as a sentential adverbial, annamh does not need to be linked to an operator (NegP) having scope over the copula; in this case, therefore, and only in this case, the semantic requirements of ar eigin do not enforce sentential negation as defined in (67). This does not mean that the copula is irrelevant to the negative or affirmative status of a sentence, of course. After all, if the copula itself is negated, the tag questions show that the sentence is negative: (108)

a. Ni annamh a theann tu go Sasana, an annamh? NEG-is seldom that go you to England Q seldom 'You don't go seldom to England, do you?' b. *Ni annamh a theann tu go Sasana, nach annamh? NEG-is seldom that go you to England NEG.Q seldom 'You don't go seldom to England, don't you?'

This circumstance, however, poses no difficulty to our approach: it suffices to note that the negative Boolean connective and the abstract existential operator are associated with two different nodes, namely the top T° (as the negative form

Negation in Irish

311

of the copula) and the embedded inflected verb, respectively.18 Such a split realization of negation and existential operator is not exceptional, as (109) shows: (109)

Not that I should be remotely interested in what you think.

To summarize, a monotone decreasing quantifier present in a cleft sentence, irrespective of its structural location, does not suffice to qualify the sentence as negative. Cleft sentences in which a [ + polarity reversing] adverbial is predicated of a CP are shown to be syntactically non-negative by the pattern of tag questions; this is neatly accounted for by our theory with the independently justified assumptions that the copula is not associated with an event variable, and that the Irish cleft construction has the structure illustrated in (107).

6 Conclusion

The evidence considered in this study points to the conclusion that negatives proper and other monotone decreasing quantifiers receive essentially the same syntactic representation when they have sentential scope. I have argued that an approach to syntactic negation which posits the existence of a separate projection NegP is able to account for this underlying uniformity, provided that negative 'quantifiers' and other monotone decreasing quantifiers are not regarded as syntactic quantifiers, but rather as indefinites locally bound by an operator at LF. An approach along these lines can account for the distribution of negative sentences in two typologically very different languages such as English and Irish, in that the contrast between the two negative systems has been traced back to the availability of a null Neg°. Finally, it has been shown that the theory of sentential negation here proposed manages not only to accommodate the syntax of Irish overtly negative sentences, but also to predict that sentential [ + polarity reversing] adverbials may occur in cleft sentences only. Notes I would like to thank all the people who helped me in the preparation of this chapter, and in particular James McCloskey and Maire Ni Chiosain. All errors and shortcomings are mine. But see Duffield (1991) for an account of the difference between the two with respect to lenition. In order to ensure the homogeneity of the data, I have based my analysis on the dialect of Irish spoken in Connemara. Some speakers perceive a contrast between the cleft construction and the less acceptable non-cleft construction. This contrasts with the position held by Ouhalla (1991b) and Zanuttini (1991).

312

Paolo Acquaviva

6 In particular, the independently motivated assumptions that pas occupies Spec NegP and that the infinitival verb does not reach Agr° in French enforce the conclusion that ne itself raises to Agr° in the infinitival construction: (i) (ii)

*Pas ne souvent lire Joyce est comprehensible. 'Not to read Joyce often is understandable.' Ne pas souvent lire Joyce est comprehensible. 'Not to read Joyce often is understandable.'

7 In a similar vein, Borsley, Rivero and Stephens (this volume) propose in connection with Breton that a head cannot move over Neg° as long as it moves to an L-related head and not another non-L-related head. 8 That Spec NegP is the scope position reached by negatives at LF is explicitly advocated by Zanuttini (1991), Haegeman and Zanuttini (1991), Moritz and Valois (1992) and Haegeman (1992); the suggestion had been put forth by Moritz (1989) and Rizzi (1990). 9 Barwise and Cooper (1981) simply define as 'weak' those determiners which are not 'strong'; D is a 'strong' determiner if a sentence of the form D N is a N is tautological. This simple test defines every and most as strong determiners, while some or a are weak. In fact, the truth value of A man is a man depends on the truth value of There is a man; if the latter is false, the former cannot be true. Barwise and Cooper's approach is sufficient for our purposes; it should be noted, however, that subsequent investigations have strongly argued for an autonomous definition of the notion of weak determiner (see especially Keenan 1987 and Higginbotham 1987). 10 As Barwise and Cooper (1981) point out, this empirical generalization illustrates a nontrivial, logically contingent property of the semantics of natural language, which is therefore susceptible of falsification. 11 In this connection see the debate between Laka (1990) and Zanuttini (1991) about the similarity of Romance negatives with polarity items. 12 Recall, however, that polarity items are licensed also in yes-no questions. 13 In Acquaviva (1993) the issue of non-sentential scope for such expressions is addressed in some detail. 14 The implementation of this approach requires positing an interpretive procedure for 'negation identification', by which if negated expressions are unselectively bound by a negative operator (a negated existential in the case at hand), the two instances of negation are identified, and treated as two expressions of a single operator. See Acquaviva (1993), where it is pointed out that such a procedure is available to languages irrespective of the ± Negative Concord typological divide. 15 The unambiguous variant is ben pochi 'few', but it appears to be incompatible with a negative marker, under any interpretation: (i)

Gianni non ha (*ben) pochi amici

See the discussion in Acquaviva (1993). 16 'Q' expresses the so-called 'interrogative particle' expressed by an and its negative counterpart nach. 17 Another possibility would be to think that annamh is not actually a monotone decreasing quantifier, and that it should be glossed as 'a few times' instead of 'few times'. It is

Negation in Irish

313

certainly plausible, in principle at least, that the addition of the particle go may interfere with the monotonicity of annamh, much as in the case of few I a few. However, the [ + polarity reversing] characterization of the bare annamh is made apparent in cleft sentences by the licensing of the polarity expression ar chor ar bith 'at all': (i)

Is annamh a labhrann Sean ar chor ar bith. is seldom that speaks Sean at all 'Sean seldom speaks at all.'

See section 5.3 below. 18 Other approaches may be feasible in principle. One possibility is to think that the copula is indeed associated with an existential operator, which closes the event variable expressed by the embedded verbal predicate. This, however, would force us to assume that the embedded inflectional node is not associated with an existential (or else the upper existential would quantify vacuously). The solution adopted in the text appears to be preferable.

10

On structural invariance and lexical diversity in VSO languages: arguments from Irish noun phrases

Nigel Duffield 1 Introduction

This chapter offers a comparative analysis of noun phrases in Irish, Hebrew and Maltese, with particular focus on the derivation of the possessor construction illustrated in (1) below, known to Semitic grammarians as the Construct State.1 I will use the term Construct State Nominal, henceforth CSN, to refer to this construction in all three languages:2 (I)

a. teach an fhir house DET man.GEN 'the man's house' b. ca'if ha-yalda scarf DET-girl 'the girl's scarf c. ras 1-mara head DET-woman 'the woman's head'

(Irish)

(Hebrew) (Maltese)

The present analysis is guided by two theoretical assumptions, both adopted from the recent Minimalist Programme (Chomsky 1993; see also Bobaljik and Carnie, this volume). The first of these is that cross-linguistic word order variation reduces entirely to variation within the lexicon, specifically to the (abstract) morphological properties of lexical items in particular languages. Under this approach, there is no strictly syntactic 'parametric' variation: in particular, functional categories are taken to be universally projected in a fixed hierarchical order; and the organization of phrase structure is assumed to be uniform across languages, with specifiers universally projected to the left of their heads (see Chomsky 1994; Kayne 1995). Observed word order contrasts are then explained in terms of the 'timing' of syntactic movement within a given derivation - either pre- or post-Spell Out - rather than in terms of phrase-structure parameters or alternative mechanisms of syntactic licensing. To illustrate with a relevant example, consider the derivation of VSO vs. SVO word order. In previous analyses, this distributional contrast was accounted for by positing two distinct mechanisms of (nominative) Case assignment: Spec-head 314

Structural invariance and lexical diversity

315

agreement and lexical government (see, inter alia, Koopman and Sportiche 1989). In VSO structures, the in situ subject was assigned Case under government (2a); in SVO languages, the subject raised to receive Case via Spec-head agreement (2b). Grammars were parameterized according to which mechanism of Case assignment was operative. (2)

a. Ugrsp V. + Agr b. UgrsP

k

NP,- [Vf + Agr

NP

[V, tf

O]]]]

[VPmax tj [v/tf O]]]]]

In Minimalism, the Case Requirement is uniformly satisfied through the Spechead relation. The observed distributional contrasts result from the timing of subject-raising: SVO order results from raising prior to Spell Out (3a); if raising occurs only at LF, (surface) VSO order is obtained (3b).3 The timing of movement is not arbitrary: rather, it is assumed to follow from the strength of the syntactic features associated with particular lexical elements; if the relevant features are strong, early movement is forced; if weak, raising may be postponed until LF. 4 NP-features of Agrs = weak: subject-raising at LF

NP-features of Agrs = strong: subject-raising prior to Spell-Out

316

Nigel Duffield

The second theoretical assumption adopted here is that syntactic movement is motivated exclusively by 'self-interest': the Minimalist principle of GREED (Chomsky 1993, 1994) excludes explanations for movement, in which some element x moves to satisfy the syntactic requirements of another element y: in (2), for example, the verb may not move (altruistically) to assign Case to the subject through 'lexicalization' of Agr; rather, if it moves in the overt syntax, it is out of self-interest - here, to check some strong V-feature under Agr. Again, if the subject NP does not move overtly, this is because it does not have to check strong features, hence, it can postpone movement until LF. With these theoretical assumptions in mind, we may turn to the analysis of the CSN constructions in (1) above. Before doing so, however, it is instructive to compare this structure with an alternative means of expressing possession in the three languages, namely through the use of a prepositional phrase (4). Borer (1984a, 1988) and Ritter (1991) term this latter construction the Tree Genitive' (FG); we will adopt their terminology here. In all three languages, the same idea can often be expressed using either construction: whenever both constructions are available, the FG is often considered the less formal alternative:5 (4)

a. an phictiur den fhear picture de + DET man.DAT 'the picture of the man' b. ha-bgadim shel ha-yalda (*bgadim ha-yalda) clothes shel DET-girl (*ha-bigdey shel ha-yalda) 'the girl's clothes' c. il-ktieb ta' Pawlu DET-book ta Paul 'Paul's book'

There are, however, certain important differences in the form and availability of CSNs in the three languages. First, Maltese differs from the other two languages in that - with some interesting exceptions - productive use of CSNs is semantically restricted to 'inalienable' head nouns, those denoting body parts and kinship relations.6 In Hebrew, as in many varieties of Arabic, CSNs are distinguished from FGs both phonologically and lexically (see Borer 1984a, 1988; Glinert 1989). In contrast to the prepositional construction which involves two primary stresses - on bgadim and ha-yalda in (4b) - the CSN in Hebrew functions as a single stress-bearing unit, with primary stress falling on the possessor hayalda. The lexical contrast found in Hebrew is clearly illustrated in (lb) vs. (4b): many Hebrew nouns display a distinct Construct form that must be employed in CSNs, but whose use in other contexts is highly restricted.7 In contrast to the Semitic languages, Irish exhibits neither of these lexical (morphological or

Structural invariance and lexical diversity

317

semantic) constraints; in almost all contexts involving underived nominals, the CSN is preferred over the FG in Irish, and is fully productive.

2 Deriving CSNs 2.1 Rightward specifiers The simple fact that possessor noun phrases in CSN constructions appear to the right of the head noun immediately forces a movement analysis of N°-NP order, given the theoretical assumption that specifiers are universally projected to the left of their heads, and that the 'd-structure' relationship between possessor NPs and the head noun is uniformly that schematized in (5) below:8 (5) Spec

Just such an analysis was first proposed (independently) by Guilfoyle (1988) for Irish, by Ritter (1988) for Hebrew, and by Mohammad (1988) for Arabic. In all of these analyses, it was assumed that the head noun moved to D° to assign Case to the possessor noun phrase. This was also claimed to account straightforwardly for the complementary distribution of CSNs and prenominal determiners, illustrated in (6) below. The examples in (7) show that no similar restriction is found in the corresponding FG constructions: (6)

a. (*an) pictiur an fhir DET picture DET man.GEN.SG 'the man's picture' b. (*ha) bigdey ha-tinok DET clothes DET-baby 'the baby's clothes' c. (*is-)sieq il-mara DET foot DET-woman 'the woman's foot'

318

Nigel Duffield

(7)

a. an phictiur den fhear DET picture de + DET man 'the picture of the man' b. ha-xaver shel ha-rofim friend shel DET-doctors 'the friend of the doctors' c. in-nanna ta' Ganni DET-grandmother ta John 'John's grandmother'

The apparent parallelism between nominal and clausal word order (VSO ~ NSO) was also appealed to as evidence for N°-to-D° movement in (6). Given the principle of GREED, however, both the Case-theoretic motivation for movement as well as the appeal to 'cross-categorial harmony' are called into question: if the head noun moves, then this must be self-serving; and there is no reason to suppose that nouns and verbs should share the same strong features. Hence, we conclude that although movement must have taken place to derive N°-NP order in (6), it is not necessarily movement to D° in all cases.

2.2 Adjective placement 2.2.1 The facts of adjective placement in Semitic CSNs provided the main empirical motivation for the original head-raising proposals. Both Hebrew and Maltese forbid placement of adjectival modifiers between the head noun and the possessor noun phrase (8b); instead, the AP must appear to the right of the possessor noun phrase, as the Maltese examples illustrate. This leads to potential ambiguities as in (10b), whenever both nouns agree in number and gender. The crucial point to observe here - as noted in Duffield (1992) - is that this constraint does not obtain in Irish: APs must immediately follow the noun they modify, including the bare head noun of CSNs (8a); hence, no parallel ambiguities can arise (10a). 9 (8)

a. guth laidir an tsagairt voice strong DET priest.GEN 'the priest's powerful voice' b. *sieq 1-eminij-a Willi foOt.F.SG DET-right.F.SG W. 'Willy's right foot'

(9)

a. guth an tsagairt laidir voice DET priest.GEN strong 'the powerful priest's voice' (*'the priest's powerful voice')

Structural invariance and lexical diversity b. sieq

319

Willy 1-leminij-a

foot.F.SG W.

DET-right.F.SG

'Willy's right foot' (10)

a. teach an tsagairt chiuin house DET priest.GEN quiet.GEN 'the quiet priest's house'/(*4the priest's quiet house') b. riu ir-ragel il-kbir brother.MSG DET man.M.SG DET-big 'the man's big brother'/'the big man's brother'

Following Cinque (1993), I will assume that adjective phrases are universally generated to the left of the noun they modify, and that cross-linguistic contrasts in adjective placement can be attributed to differences in the extent of overt N°-movement (see also, inter alia, Giorgi and Longobardi 1990; Crisma 1990; Lobel 1990; Valois 1991). Since nouns appear to the right of adjectival modifiers in all three languages under investigation, even in simple noun phrases, the conclusion is that the contrast in (8)—(10) between Irish on the one hand, and Maltese on the other, is attributable to overt specifier-movement of the possessor noun phrase: this is operative in Semitic, but not in Irish. In Minimalist terms, this means that the NP-features of F 2 in (11) below are strong in Semitic, but weak in Irish:

Spec

NP-features of F 2 = weak in Irish —> subject-raising at LF NP-features of F 2 = strong in Heb/Mal subject-raising prior to Spell-Out

As observed in Duffield (1992), the contrast between Hebrew/Maltese and Irish can only be derived by hypothesizing a second functional projection above AP as the landing-site for specifier movement; in Duffield (1992), it was claimed that this functional projection was AgrP. A very similar analysis was independently proposed in Ritter (1991) for Hebrew, in which the hypothesized second projection was Num(ber) phrase; we return to this proposal shortly. Notice that

320

Nigel Duffield

whereas the Maltese and Hebrew facts force the derivation schematized in (11) in which the head noun has raised to F 1 - the Irish data in (8)—(10) require N°-movement only as far as F2 . There is, in fact, some independent empirical evidence to support the generalized head movement out of NP schematized in (11). This comes from the ordering of adjective phrases in Irish. Sproat and Shih (1991) and Cinque (1993) argue for a universal precedence hierarchy for adjective placement with respect to the head noun - which, for Cinque, subsumes quantifier, numeral and classifier phrases. Crisma (1990) proposes the hierarchy in (12a) for underived nominals, illustrated by the Italian and German examples in (12b) and (12c) respectively. With respect to Irish, Sproat and Shih (1991) argue that the observed order of adjectival modifiers is consistent with this universal hierarchy only if partial N°-movement is assumed: the examples in (13) illustrate relevant Irish examples, with t indicating the proposed source of the head noun. 10

(12)

a. possessive > cardinal > ordinal > quality > size > shape > colour > nationality b. (i) suoi due altri bei grandi quadrij tondi grigi/, the his two other nice big pictures round grey 'his two other nice big round grey pictures' c. meine drei interessante dreieckige rote franzosische Bticher 'my three interesting triangular red French books'

(13)

a. clabhsur a n-ochtu hostanx mbx daorluachach tt closure POSS-3PL. eighth hotel big expensive 'the closure of their eighth big expensive hotel' b. liathroidi bheag bhui tt ball small yellow 'a small yellow ball' c. feari beag eadrom tt man small slight 'a man of slight build'

Although this constitutes good evidence of partial N°-movement, it is nevertheless questionable whether it is correct to lump together' quantifier and numeral phrases - functional modifiers, as it were - with other attributive adjective phrases: it is probably not coincidental that functional modifiers invariably precede, and other adjectives invariably follow, the head noun in the three languages under investigation; see Duffield (1995) for discussion.

Structural in variance and lexical diversity

321

2.2.2

The fact that Hebrew and Maltese behave similarly with respect to adjective placement in CSNs does not mean that they always display identical patterns. One important difference shows up in the corresponding FG constructions: whereas Hebrew (and Irish) disallow adjectival modifiers of the head noun to the right of the possessor noun phrase, this is permitted in Maltese, at least for underived nominals (see Fabri 1993: 156):11 (14)

a. *ha-bayit shel ha-mora ha-gadol DET-house shel DET-teacher DET-big 'the teacher's big house' b. ha-bayit ha-gadol shel ha-mora DET-house DET-big shel DET-teacher 'the teacher's big house' c. ir-rota ta' Pawlu 1-gdid-a DET-bicycle.F.SG ta Paul DET-new.F.SG 'Paul's new bicycle' d. ir-rota 1-gdid-a ta' Pawlu DET-bicycle-F.SG DET-new.F.SG ta Paul 'Paul's new bicycle'

Ritter (1991: 45) proposes a uniform structure for CSNs and Free Genitive constructions in Hebrew: under her analysis, reproduced in (15) below, the difference between the two constructions is a simple function of the extent of N°-movement. Although this is conceptually a very attractive way of capturing the CSN/FG contrast, and, I believe, basically on the right track, this analysis encounters certain problems when confronted with Irish and Maltese data, including the Irish adjective-placement facts and Maltese FG facts just mentioned: facts that suggest, on the one hand, that the additional functional projection (AgrP) must be involved; and, on the other, that the genitive possessor noun phrase in CSNs occupies a different position at Spell Out from the prepositional possessor phrase in FG constructions. (15)

a. axilat Dan ha-menumeset et ha-uga eating Dan DET-polite et DET-cake 'Dan's polite eating of the cake'

(CSN)

[Up[axilati] [NumP Darij [Numo tt] [ NP [ A p ha-m...] [NP tj [N' h et ha-uga]]]]]

322

Nigel Duffield b. ha-axilat ha-menumeset shel Dan et ha-uga DET-eating DET-polite shel Dan et DET-cake 'Dan's polite eating of the cake'

(FG)

[DP[ ha ] [NumP [Num° axilatH [ N P[AP ha-m . . . ][Np shel D. [N' /,- et ha-uga]]]]]

Before considering these apparent counterexamples to Ritter's proposal in more detail, we will examine some further cross-linguistic evidence in support of the basic idea, developed independently in Ritter (1991), Duffield (1992) and Lyons (1992); viz. that all nouns raise via head movement to a functional projection associated with (grammatical) number of features, and that - in Hebrew, at least - the head of CSNs raises further to some higher head position. What is interesting about this evidence is that there are subtle, but significant, contrasts among the three languages.

2.3 Possessive clitics 2.3.1 All three languages show alternations in CSN constructions between possessive clitics (POSS) and possessor noun phrases. In all cases, the possessive clitic shows agreement in person, number and gender with the possessor; in addition, possessive clitics and determiners are in complementary distribution, as illustrated for Irish and Maltese in (16):12 (16)

a. (*an)a

phictiur

DET 3.SG.M picture

'his picture' b. (*1-) omm-u, DET mother-3sG.M 'his mother'

(*an) a

pictiur

DET 3SG.F picture

'her picture' (*iz-)ziju-ha DET uncle-3sG.F 'her uncle'

(*an)a

bpictiur

DET 3.PL picture

'their picture' (*iz-)zijt-ek DET uncle-2sG 'your aunt'

The languages diverge from one another, however, in at least two important respects. First, as is obvious from the examples above, POSS is enclitic in Maltese (and Hebrew) but proclitic in Irish. More significantly, Irish POSS is syntactically separable from the head noun: this is shown by the fact that classifier and numeral phrases may intervene between POSS and the head noun; the Irish patterns illustrated in (17) are quite ungrammatical in either Maltese or Hebrew: (17)

a. a cuigdteach 3.PL five house-SG 'their five houses'

(*cuig a dteach)

Structural invariance and lexical diversity b. mo chead bhroga ISG first shoe-PL k my first shoes' c. do chuid leabhar your CL book-SG 'your books'

323

(*chead mo bhroga)

(*chuid do leabhar)

These facts suggest two things. First, it is clear that Irish POSS has inflectional properties quite different from its Semitic counterpart. Although phonologically proclitic, Irish POSS cannot be treated as an inflectional affix, since it exhibits no selectional restrictions (see Fabb 1984; Jaeggli 1986); therefore, it is free to raise independently in the syntax, if it has strong features to check. Second, if one assumes a uniform 'D-structure' representation for all three languages, then the contrastive distributions in (16) - including the complementary distribution of POSS and determiner elements - can be obtained only either if the head noun has raised higher in Maltese/Hebrew than in Irish (18a), or if POSS has moved higher in Irish than in Hebrew (18b). (18)

a. [ty + poss* b. [poss*.

tk N,-] N f tk]

In light of the formal problems associated with the type of long head movement involved in (18b),13 we will assume that although both of these operations N°-movement and POSS-raising to D° - take place, in all three languages POSS heads its own functional projection above NP, namely AgrP. The distributional contrast in (16) then results from the fact that in Maltese and Hebrew, the head noun raises through Agr° to D° to check strong features under D° (19a), whereas in Irish, POSS raises independently to D°, as in (19b). By hypothesis, POSS is able to check the same features; given Economy considerations, POSS, rather than N°, must check these features if it can, since it is closer to D°. 14

NP; Agr°

h

NumP Spec

Num' Num°

N?

AP*

NP Spec NPy-

in Irish -*• subject-raising at LF

"T

N° t:

D°

AgrP

Nt+posskSpec

Agr'

Np. Agr° \

NumP Spec

Num'

NP-features of Num = strong in Hebroy/Maltesc - • subject-raising prior to Spell-Out

The derivations schematized in (19) also deliver an account of the second property shared by the languages with respect to POSS clitics. This apparently distinguishes Maltese on the one hand from Irish and Hebrew on the other. In CSNs in both Hebrew and (most dialects of) Irish, there is an obligatory complementary distribution between POSS and possessor noun phrases:15

Structural in variance and lexical diversity (20)

325

a. *a, phictiur an fhir, 3.SG.M picture DET man.GEN 'his picture of the man' b. *beyt-o, ha-mora, house-3.SG.M DET-teacher 'the teacher's house'

In McCloskey and Hale (1984), this distribution is viewed as a type of radical pro-drop effect: 'rich' agreement on the head noun licenses pro. In Chomsky (1993), it is proposed that pro is licensed only in the Spec-Head relation to [Agr a Agr], where a is [ + tense] or V, Agr strong or V = V*. (Chomsky 1993: 10) Since we are dealing here with Agr within noun phrases, we suppose that Tense and V-features are irrelevant: translating Chomsky's proposal to the nominal domain, we assume instead that strong 'N'-features in Agr license pro. In Hebrew, then, these strong N-features are checked by raising the head noun to Agr; in Irish, the lexical POSS automatically checks its own features. Notice that, as in the clausal domain, the strength of N-features is independent of the strength of the NP-features responsible for Case-checking. Therefore, we assume that pro is licensed in all three languages by a strong Agr, but whereas in Hebrew and Maltese pro raises overtly - albeit inaudibly - to [Spec, Agr'], in Irish pro does not raise until LF, since the NP-features of Agr are weak. This assumption permits a uniform treatment of Case-checking for all possessor NPs, whether or not they are phonetically realized; it accounts simultaneously for the distribution of pro as well as the surface position of overt possessors vis-d-vis adjective phrases (see section 2.2.1 above). 16 At this point, it is necessary to discuss the contrast between Maltese and Hebrew/Irish alluded to above. The Maltese examples in (21) contrast with the examples in (20) in (apparently) permitting the co-occurrence of the POSS and an overt possessor noun phrase: (21)

a. omm-u, Pawlu, mother-3sG.M Paul 'Paul's mother' b. xagnar-ha, Marija, hair-3sG.F Mary 'Mary's hair'

From one point of view, this might seem to be a desirable fact, since nothing in the analysis so far - or, for that matter, in any previous 'Agreement-based' analysis of this construction - enforces the complementarity observed in Hebrew

326

Nigel Duffield

and Irish (see, inter alia, McCloskey and Hale 1984; Jaeggli 1986; Dooley Collberg 1991). However, as Fabri (1993: 178) clearly demonstrates, closer inspection reveals ^that cases are anomalous: in contrast to the Maltese examples in (8b)-(10b) above, any modifying adjective phrase in these latter cases must appear between the head noun and the possessor noun phrase, apparently as in Irish: (22)

a. sieq-u

1-leminij-a

Willi

(cf. (8b))

f00t(F.SG)-M.SG DET-right-F.SGW.

'Willy's right foot' b. *sieq-u Willy 1-leminij-a

(cf. (9b))

From this minimal contrast and other intonational evidence, Fabri (1993) concludes that the overt possessor NPs in (21)—(22) should be treated as adjuncts to the phrasal projection immediately below AP. Although the details of Fabri's account differ considerably from the current proposal, we assume here that his general conclusion, namely that 'doubled' NPs in CSN constructions are really adjuncts (to NP), is fundamentally correct.

2.3.2 Poss in FG constructions In fact, there are good reasons to believe, pace Ritter (1991) - see (15) above that in Hebrew FG constructions in which POSS clitics co-occur with co-referential shel-NP phrases, these latter phrases should also be treated as adjuncts, at least in some instances. The examples in (23) are adapted from Borer (1984a: 81): (23)

a. beyt-o, shel ha-mora7 house-3.SG.M shel DET-teacher 'the teacher's house' b. ktivat-o, shel Dan/ et ha-ma'amar writing-3.SG.M shel D. et DET-article 'Dan's writing of the article'

(24)

a. ktivat Dan et ha-ma'amar writing D. et DET-article 'Dan's writing of the article' b. ktivat ha-ma'amar writing DET-article 'the writing of the article'

To do justice to Ritter's analysis of FG constructions would take us beyond the scope of the present chapter (see Duffield 1992, 1995, for discussion); here, it must suffice to mention two important Hebrew vs. Irish contrasts. Notice, first, that in the Irish example in (25a), the POSS clitic a may co-occur with, but may not be

Structural invariance and lexical diversity

327

co-referential with, the following de + NP; also, in both (25a) and (25b), the de + NP phrase is obligatorily interpreted as the thematic object of pictiur. (In this respect, Irish de patterns more closely with the Hebrew accusative Case-marker et, than with shel (see (26a).) Second, in CSNs involving the verbal noun in Irish, the thematic constraint is even stricter: the POSS clitic must bind the (phonetically null) complement NP; it cannot be interpreted as the thematic subject (26b); the result of this is the near-minimal contrast between (26a) and (24a) above:17 (25)

a. a*//y phictiur den fliear, 3.SG.M picture de + DETman 'his picture of the man' / *'the man's picture' b. an phictiur den fhear, DET picture de + DET man 'the picture of the man' / 'the man's picture' (theme/*possessor)

(26)

a. *scriobh Sheain den write.VN S.GEN

litir

de + DET letter

'Sean's writing of the letter' b. *a scriobh na litreach 3SG.M write.VN DET letter.GEN 'his writing of the letter' c. scriobh na litreach write.VN DET letter.GEN 'the writing of the letter' If we now consider the examples in (27) - from Borer (1984a: 81) and Ritter (1988: 922) - it can be observed, first, that shel may be used to Case-mark either NP; and, more significantly perhaps, that the two arguments of ktivat may occur in either order (NOS ~ NSO). This freedom of word order is not observed in Irish: it is-much more typical of adjuncts than of NPs in fixed argument positions (though see note 15). (27)

a. ktivat Dan shel ha-ma'amar writing D. shel DET-article 'Dan's writing of the article' b. ktivat ha-ma'amar shel Dan writing DET-article shel D. 'Dan's writing of the article' c. tmunat ha-yalda shel ha-mora picture DET-girl shel DET-teacher 'the teacher's picture of the girl' / % 'the girl's picture of the teacher' 18

328

Nigel Duffield

Therefore, pending further investigation of this question, we conclude that at least in cases such as those in (23) and (27) above, shel-NP phrases in Hebrew should be treated on a par with the Maltese cases in (21)—(22) above; that is, as adjuncts to NP. This still leaves unresolved the problem of the complementarity of 'rich' agreement and overt possessor noun phrases. In Duffield (1993), we proposed a possible account based on the notion of Relativized Minimality: for a number of reasons, this account now appears rather unsatisfactory (see Duffield, 1995). Notice that the alternative analysis that has traditionally been offered to explain this type of complementarity in the clausal domain - namely an Tncorporation'type account (see, inter alia, Hale 1989; Guilfoyle 1990) - will not work here either, at least in Irish, given the syntactic independence of POSS from the head noun, or from any other lexical head; see section 2.3.1 above. Of course, within Minimalism, there is a technical solution to this problem: one could simply stipulate that in Irish, Hebrew and Maltese - though not, say, in Italian - the presence of strong N-features in Agr precludes the occurrence of any NP-features (weak or strong) whatsoever. If this were so, then NPs could not be checked at any point in the derivation, hence, they could not be inserted; at LF, POSS itself would be interpreted as bearing the thematic role otherwise associated with the overt possessor noun phrase. Although it would achieve the desired result, I will not adopt this quite ad hoc solution here, first, because it is non-explanatory at anything but the most superficial level and, second, because it appears to undermine the principle of GREED. It would be quite unfortunate, I believe, if this turned out to be the best explanation available (see Roberts and Shlonsky, this volume, for an interesting alternative account of this complementarity).

2.4 Definiteness and demonstrative interpretation: 'D-features' 2.4.1 Thus far, we have claimed that the derivation of CSNs in all three languages involves N°-movement: in all cases, the head noun raises overtly at least as high as Num°; 19 in Hebrew and Maltese, N° raises higher in the overt syntax through Agr° to D°. For Irish POSS constructions, we claimed that N° remains in Num° prior to Spell Out, and that POSS raises independently to check strong features in D°. At this point, two questions arise: the first concerns the nature of these latter features in D; the second concerns Irish CSNs where no POSS clitic is involved. In respect of the first question, we propose - essentially following Ritter (1988) - that the relevant features that motivate raising to D° are those associated with the (semantic) property of definiteness: let us call these D-features, for the sake of

Structural invariance and lexical diversity

329

discussion. Let us further assume that in those contexts where head nouns raise overtly to D°, this is due to the fact that these nouns are inherently (lexically) specified as [ +definite]. Once again, the empirical evidence that supports this proposal reveals interesting contrasts among the three languages. Consider first the following examples: (28)

a. pictiur an fhir picture DET man.GEN 'the/*a picture of the man' b. xaver ha-roPim friend DET-doctors 'the/*a friend of the doctors' c. oht ir-ragel sister DET-man 'the/*a man's sister'

(29)

a. pictiur den

fhear

picture de +DET man 4

a/*the picture of the man' b. xaver shel ha-rof im friend shel DET-doctors 'a/*the friend of the doctors' c. wieried minnnut otit Pawlu one of siblings Paul 'one of Paul's siblings' These examples illustrate the fact that CSNs must be interpreted as definite if the possessor subject is definite: compare the corresponding FG constructions in (29). However, Maltese (and apparently Hebrew)20 are distinguished from Irish in requiring that CSNs be always inherently definite.21 This is illustrated by the minimal contrasts between Irish and Maltese illustrated in (30) and (31) below: in Irish, CSNs may be indefinite and may be directly modified by prenominal numeral phrases; in Maltese, both of these options are excluded:22 (30)

a. cineal airithe cait kind certain cat.GEN 'a certain kind of cat' b. racht iontach feirge outburst great anger.GEN 'a great outburst of anger' c. tri charr deag Sheain three cars ten S.GEN 'Sean's thirteen cars'

330

Nigel Duffield

(31)

a. *otit tifla sister girl 'a girl's sister' b. *wafid-a otit Pawlu one sister Paul 'one of Paul's sisters' c. *iz-zewg dufr-ejn Pawlu DET-two nail-PL Paul 'Paul's two nails' d. iz-zewg dufr-ejn ta' Pawlu DET-two nail-PL ta Paul 'Paul's two nails'

(cf. dufr-ejn Pawlu 'Paul's nails')

This contrast can be accounted for if it is supposed that head nouns in Maltese (and Hebrew), but not those in Irish, are inherently lexically specified with strong D-features: the syntactic consequence of this being that they must move overtly to D° to check these features. Under such an analysis, the impossibility of directly modifying the head of CSNs by numeral phrases in Maltese would follow if it were assumed that the presence of intervening numeral phrases blocked N°-movement (thus causing the derivation to crash due to the resulting unchecked D-features at Spell Out). 2.4.2 Demonstrative licensing and interpretation The evidence presented in the two previous sections appears to suggest that in Irish CSNs, whenever POSS clitics or numeral phrases are involved, there is no further N°-raising past Num°. However, it is not clear from these data what happens whenever POSS is not inserted in Irish. Notice that it may still be that N- and D-features are projected - under Agr° and D° respectively - even when POSS is absent: if this were the case, then overt N°-movement from Num° to Agr° to D° would be required to check these features, and to prevent the derivation from crashing at PF ((32), irrelevant structure omitted).23 In this final section, we present some evidence from the placement and interpretation of demonstrative elements in Irish which suggests that this movement does indeed take place.

Structural invariance and lexical diversity

331

(32)

'+D'

As the examples in (33) illustrate, Irish demonstratives show a three-way proximity contrast, seo, sin, siud, meaning roughly 'this', 'that' and 'yon' (or 'yonder') respectively. Notice, first, that in simple noun phrases - that is, in noun phrases without full NP possessor subjects - these demonstrative elements are subject to a strict co-occurrence constraint: they must be licensed either by an overt prenominal determiner, as in (33a), or by a POSS clitic (33c). In Hebrew, by contrast, demonstratives appear not to be able to co-occur with POSS at all (34): (33)

(34)

an

a. an mac seo DET man DEM! 'this man' b. *mac seo c. a mac seo

cailin sin

DET girl

DEM2

'that girl' *cailin sin a theach sin

an

bhean

siud

DET woman DEM3

'yon woman' *bhean siud a dteach siud

3SG.FSOn DEM]

3SG.M house DEM2

3PL house DEM3

'this woman's son'

'that man's house'

'yon people's house'

a. xag ze festival DEM 'this festival' b. *rishum-o

ha-xag

ha-ze

DET-festival DET-DEM

'this festival' ha-ze

drawing-3.SG.M DET-DEM

'that drawing of his' In addition to this syntactic condition, there is a more interesting constraint on the interpretation of demonstrative elements, just in case they are licensed by possessive clitics. As the translations in (33c) show, demonstrative elements are obligatorily interpreted with respect to the possessive clitic, rather than with respect to the head noun; hence in (33c), a theach sin can only be interpreted as 'the house of that man over there', not 'his house over there'. For this reason, *rno mhac siud, lit. 'my' son over there'', although not strictly ungrammatical, is excluded on pragmatic grounds. It is possible to account both for the distributional and the interpretive constraints illustrated here if it is assumed that demonstratives are licensed by, and construed with, whatever lexical element checks the strong D-features of the

332

Nigel Duffield

phrase. Given that strong features must be checked prior to Spell Out, determiner-less cases such as those in (33b) will be ungrammatical, since no element has raised to D°, and the features are left unchecked. Whenever POSS is projected (33c), it will automatically license and identify the demonstrative element by raising to D° in the overt syntax. In these cases, since D° is then occupied by the licensing POSS element bearing a different referential index from the head noun, the demonstrative cannot receive the index of the head noun, and thus cannot be construed with it. In simple noun phrases involving a separate determiner (33a), the demonstrative element will be referentially associated with the head noun, but this relation will be indirect, mediated by the determiner head (which will necessarily bear the same referential index as the head noun). These derivations at Spell Out are illustrated in (35): (35)

a. *[ D P ' + D ' [AgrP [NumP fearj siud\ b. [DP 4 Ugrp'* UmP theachj siud1] 1 c. [UP an [Agrp [NumP fearj siud1] [Np[tj]]]]]

Such an explanation of these syntactic and interpretive constraints is naturally only possible if the head-raising account of CSNs proposed here is adopted. Notice that this hypothesis about demonstrative licensing makes a prediction with respect to the behaviour of demonstratives in CSNs containing full possessor noun phrases. If overt N°-to-D°-movement is possible in Irish, whenever (i) the head noun is lexically specified as [ + definite], and (ii) whenever no POSS element is involved, then we expect that just in these structures, demonstrative elements should still be permitted even in the absence of prenominal determiners. We would further expect that in this context DEM should inherit the reference of the head noun, since the noun itself will act as the demonstrative licenser (in virtue of raising to D°, checking strong D-features). As the examples in (36) show, both of these predictions are borne out: just in CSNs, determiner-less demonstratives are licensed, and interpreted as expected. Hence, the head-raising analysis immediately explains the minimal contrast between the phrases in (36) and those in (33b) above. 24 (36)

a. mac' sin* an fhir SOn

seo

'that son of this man' b. cota* seo' Phadraig coat DEMi Patrick.GEN 'this coat of Patrick's' (37)

*bney

(*mac sin)

DEM2 DET man.GEN DEMi

ha-ele

ha-kibuts

members DEM-DET DEM-DET

'these members of the kibbutz'

(*cota seo)

Structural in variance and lexical diversity

333

Notice, further, that in contrast to Semitic, the head noun of Irish CSNs does not have to be specified as [ + def], as demonstrated by the examples in (30) above. If, however, the [-def] form were selected for insertion into the derivation, then, by the Procrastinate Principle, it would not raise to D° until LF; if indeed DP is projected in such instances - see below. In most cases, the difference is harmless, since one could allow the corresponding D-features of DP to be weak. However, if demonstrative elements are dependent upon a strong D° for their licensing and interpretation, then in just these cases, a potential problem would arise if the [-def] form of the head noun were inserted. It is interesting to note, therefore, that CSNs in Irish may (optionally) be modified by prenominal determiners, just in case demonstrative elements are also involved: this is illustrated in (38), from Ernst (1992: n.17) and Christian Brothers (1990: 8): (38)

a. (an) mothu sin an tsaighdiura DET feeling DEM 2 DET soldier 'that feeling of the soldier' b. (na) gnoithe seo an eallaigh DET matters DEM! DET livestock 'these matters of the livestock' c. (sa) ghleann seo na ndeor in + DET valley

man

(*na gnoithe an eallaigh)

(*sa ghleann na ndeor)

DEMI DET tears.GEN

'in this valley of tears' d. (an) fear sin an airgid DET

(*an mothu an tsaighdiura)

(*an fear an airgid)25

DEM2 DET money

'that rich man' That is, alongside the minimal contrast between (33b) and (36) above *[N DEM] VS. [N DEM NP.GEN] - there is a second minimal distinction: *[DET N NP] vs. [DET N DEM NP.GEN]. From a Minimalist perspective, in which optionality is essentially excluded from the grammar by Economy considerations, the apparently optional nature of determiner placement in (38) must be explained in terms of differing lexical properties: we are suggesting here that the lexical contrast should be traced to the [idefiniteness] features associated with the head noun.

2.4.3 Definiteness and Case-checking Finally, related to the previous point, there is some independent evidence which shows that indefinite CSNs in Irish behave rather differently from their definite counterparts, and which suggests that in the case of indefinites, N° raises overtly only as high as Agr° (see Ritter 1991; Ernst 1992). This evidence has to do with an interaction between definiteness and Case assignment in what are traditionally

334

Nigel Duffield

termed 'Common in Form, Genitive in Function' contexts (see Christian Brothers 1990). In descriptive grammars of Irish, it is pointed out that in verbal-noun constructions - which have essentially the internal structure of CSNs - complements are assigned genitive Case, unless these complements are indefinite - and, typically, adjectivally modified - in which case they may be assigned 'common' (nominative) Case. This contrast is illustrated by (39) from 6 Huallachain and 6 Murchu(1981: 193):26 (39)

a. (Bionn se ag) [deisiu na ngluaistean] (be.HAB he PROG) repair.VN DET car.GEN 'He repairs the cars.' b. (Bionn se ag) [deisiu gluaisteain Ghearmanacha] (be.HAB he PROG) repair.vN cars.NOM German.NOM 'He repairs German cars.' c. *(Bionn se ag) [deisiu na gluaisteain Ghearmanacha] be.HAB he PROG repair.VN DET.NOM cars.NOM German.NOM 'He repairs the German cars.'

The present analysis offers a way to derive these case alternations in a principled fashion. Let us assume first that 'genitive Case' is simply the phonological realization of Spec-head agreement with D°, whereas 'nominative' is the realization of a Spec-head relation to Agr°; in fact, this seems to be a standard, and rather uncontroversial, assumption (see Abney 1987). Now let us suppose -pace Ritter (1991) and Ernst (1992) - that DP is only projected at all whenever some lexical element has D- or DP-features to check: since Irish has no indefinite articles, this means that indefinite noun phrases will be AgrPs, rather than DPs. With these two assumptions, we suggest that the Case alternations in (39) are a direct function of the extent of specifier movement at LF: in (39a) na ngluaistean raises at LF to [Spec, Agr'] for Case-checking purposes; it then raises further to [Spec, D'] to check its deflniteness features; whereas in (39b), gluaisteain Ghearmanacha raises only as high as [Spec, Agr'], and this shorter movement is spelt out as nominative Case. The option of having genitive Case in both contexts is accounted for, once again, by assuming that the head noun can be inserted with or without deflniteness features. (For the related idea that T° is higher than Agrs° in Irish - hence, that nominative Case results from this 'shorter movement' at the clausal level - see McCloskey (this volume).27)

3 Summary In this chapter, it has been argued that many of the observed and previously discussed parallelisms among VSO languages can be accounted for in a restrictive fashion by assuming strict structural invariance, and by deriving word order

Structural invariance and lexical diversity

335

contrasts through movement. Rather than focusing on clausal word order, as is more usual, we have concentrated here on syntactic patterns within noun phrases. Although the structural commonalities among these VSO languages are striking, what is perhaps more interesting is the extent of lexical contrast, and of the subtle, often rather intricate, differences that exist even between relatively closely related languages, such as Hebrew and Maltese. Given the present space constraints, this chapter can do no more than draw attention to some of the more obvious contrasts among these languages (that have nevertheless often been glossed over in previous work). However, it is hoped that this chapter can make a contribution to research in this area by showing that although some theoretical insights can be gained by treating VSO languages as a homogeneous set, the goal of descriptive adequacy requires greater attention to the idiosyncrasies of particular languages. Notes This chapter revises and extends previous analyses, including Duffield (1992, 1993). That work in turn was based on earlier proposals by Guilfoyle (1988), Ritter (1988) and Mohammad (1988). The revisions here are prompted in response to work that has since come to my attention, including that of Lobel (1990), Ritter (1991, 1993), Ernst (1992), Fabri (1993) and Cinque (1993), as well to the comments of reviewers, especially those of Bob Borsley. Due to space limitations, the analysis set out here is inevitably rather promissory in nature: the proposal is developed and argued for in much greater detail in Duffield (1995). I am extremely grateful to all those who provided helpful comments on previous versions of this proposal, particularly to my former colleagues at Heinrich-HeineUniversitat, Dusseldorf. I would also like to thank Donall P. 6 Baoill, Grainne Gallagher and Maire Ni Chiosain for their help with the Irish data. All remaining errors and inaccuracies are, of course, mine alone. This research was supported in part by an FCAR team grant (no. 94-ER-0578). Almost all of the data presented here are drawn from recent primary and secondary sources. For Hebrew, the examples and judgements are drawn from Borer (1984a), Glinert (1989) and Ritter (1988, 1991); for Maltese, from Fabri (1993); for Irish, from Christian Brothers (1990), 6 Huallachain and 6 Murchu (1976), and 6 Siadhail (1989). Throughout the chapter, in example sets where all three languages are presented, Irish examples are given first, or to the left, followed by Hebrew and Maltese, respectively. The following abbreviations appear in the glosses: d e m ^ = demonstrative element (degree of proximity), DET = definite determiner, COM = non-genitive Case (nominative or accusative), GEN = genitive Case, NOM = nominative Case. In the Irish examples, the initial consonant mutations, 'lenition' and 'eclipsis', are indicated by bold type. Although I will argue here for structural invariance within NP(DP), I would take issue with the standardly assumed order of functional projections at the clausal level, in

336

4

5

6

7

8

Nigel Duffield which AgrP dominates TP (see Chomsky 1991; Bobaljik and Carnie, this volume). In Duffield (forthcoming), it is argued in detail that (at least in northern dialects of Modern Irish) VSO order involves verb movement to Agr, where Agr is below T°. The less radical conclusion to be drawn from this is that I adopt structural invariance as a heuristic only: this chapter can be viewed as showing how apparently diverse word order patterns can be accounted for whilst assuming a restrictive, uniform syntax. (The reader is hardly expected to adopt the more radical conclusion on the basis of this promissory note; however, see McCloskey (this volume) for some arguments in favour of projecting TP above AgrP.) The Procrastinate Principle ensures that if movement can be postponed until LF, then it must be (see Chomsky 1993). I will ignore the quite serious questions of learnability that are raised by the Minimalist assumptions adopted here. One particular difficulty is that since 'feature strength' is taken to be independent of overt morphological properties, it is difficult to see how the notion is explanatory in any deep sense - in so far as it simply recapitulates the observed word order distributions - nor how the strength of features is learnable. See Schmidt (1994) and Duffield (1995) for further discussion of this question, which is clearly beyond the scope of the present chapter. It is not clear to me why the CSN should be semantically restricted in this way in Maltese; although the fact that it is provides further evidence of the lexical basis of the construction. None the less, as the examples show, neither Irish nor Hebrew exhibits this constraint, at least in more formal registers. One consequence of this is that the FG alternative to (lc), e.g. ir-ras ta Pawlu ('Paul's head'), is interpreted as referring to a decapitated (alienated) head; use of the CSN carries no such obligatory interpretation, and is preferred even in the latter context (see Fabri 1993: 160). Fabri (1993) notes that whilst some of the apparent exceptions to the 'inalienability' condition can be shown to be compounds by a variety of syntactic tests, there is a significant set of CSNs with 'alienable' head nouns that cannot be explained away, and which must be explicitly marked in the lexicon, for instance xewqet it-tfal (the children's wish), zmein it-tadam (the time of tomatoes). For further discussion of the compound vs. CSN contrast, see Borer (1988) and Duffield (1995); for a discussion of defmiteness and inalienability, see Vergnaud and Zubizarreta (1992). It is interesting to note that although CSNs are more productive in Hebrew than in Maltese, their use with low-frequency head nouns is associated with more formal registers. In acquisition, children do not acquire these constructions until relatively late, greatly preferring the alternative shel strategy; children also show production errors with CSNs at earlier stages of acquisition, especially in incorrect use of the normal form where the Construct form is required, and in the (ungrammatical) insertion of prenominal determiners (see Berman 1987). The idea that specifiers universally precede their heads is derived from more basic principles in Kayne (1995) and Chomsky (1994). Previously, it had been assumed as a stipulated property of X-bar Theory (see Johnson 1991; Emonds 1992). Even if specifier-head-complement order were parameterized on a language-particular basis,

Structural in variance and lexical diversity

9

10

11

12

13 14

15

16

337

I should still want to argue for a movement analysis here, given that the specifiers of all other maximal projections in Irish can be shown to be to the left of their heads (see Duffield 1994, 1995). Adjectival modification of the head noun in Irish CSNs in cases such as (8a) is not restricted to single adjectives, thus not so easily amenable to an account where N° raises through A 0 . Nevertheless, the facts of demonstrative placement in these constructions (see below; Duffield, 1995) suggest that it is necessary to differentiate between AP vs. PP (or other adjunct) modification of the head noun. For reasons of space, we ignore the important question of whether adjective phrases should be treated as adjuncts or higher specifiers, and the related issue of whether the partial N°-movement proposed here involves movement through A0. See Cinque (1993) and Duffield (1995) for more detailed discussion. The differences between derived and underived nominals is an important point of contrast among the three languages: one striking difference between Irish on the one hand and Hebrew and Maltese on the other is that Irish never allows the possessor noun phrase to be a thematic subject in CSN constructions involving the verbal noun. See below. It will be noticed that, in Irish, the differences between the third-person agreement morphemes are expressed not by their surface form - they are uniformly realized as schwa - but rather by the initial consonant mutation which they induce on the head noun. For a different approach to the problem of enclitic pronouns, see Roberts and Shlonsky (this volume). See Borsley, Rivero and Stephens (this volume). An analysis more in keeping with the spirit of Minimalism would treat the Semitic cases slightly differently: if POSS is inflectional in these languages, then it should be inserted already attached to the head noun under N°: the 'inflected' noun would then raise via Agr° to D°, checking strong features as it moved. Here, it does not greatly matter which technology is employed, provided that the relevant - lexical - distinction is made between Semitic and Irish POSS. This is quite reminiscent of the Minimalist analysis of Germanic 'verb-second' proposed in Zwart (1992). I am grateful to Sheila Dooley Collberg for pointing this out to me. McCloskey and Hale (1984) discuss one dialectal exception to this complementarity: the Cois Fhairrge dialect (see de Bhaldraithe 1975) appears to allow a pronominal copy of POSS postnominally: for example, a muirin sise 'her family'. See also the Welsh 'echo pronouns' discussed in Roberts and Shlonsky (this volume). It is tempting to relate the distributional and case properties of overt possessor NPs to a remark in Chomsky (1993) which immediately precedes that quoted above: It is natural to expect less-marked Case to be compensated (again, as a tendency) by more-marked agreement. (Chomsky 1993: 10) This comment refers to nominative-accusative vs. ergative-absolutive Case-marking, rather than to the licensing of pro; nevertheless, by distinguishing between N- and

338

Nigel Duffield

NP-features, it is no longer accidental that Irish with no overt movement to [Spec, Agr'] should show morphologically distinct genitive Case, whereas Hebrew shows movement but no morphologically overt Case distinctions. 17 In fact, the picture is considerably more complex than this. There is evidence that both Hebrew and Maltese possessor noun phrases headed by deverbal nouns - CSNs and FGs respectively - should be treated differently from possessor noun phrases headed by underived nouns. Once again, the main evidence comes from adjective placement: in both languages, adjective phrases modifying the head noun may occur to the right of the possessor noun phrase in underived contexts, but this is not possible whenever the deverbal noun ('action nominal') is involved. The examples in (i) and (ii), from Glinert (1989: 32) and Fabri (1993: 168) respectively, illustrate the constraint; once more, see Duffield (1995) for discussion: (i)

(ii)

a. *hitpatxut ha-tsafron he-mehira development DET-north DET-rapid 'the rapid development of the north' b. ha-hitpatxut ha-mehira shel ha-tsafron a. Tiskoperta ta' 1-Amerika importanti DET-discovery ta' DET-America important 'the important discovery of America' b. l'iskoperta importanti ta' 1-Amerika

(cf. (14c))

18 Ritter (1988) implies that the latter reading is excluded for some speakers. She claims that even for speakers for whom this reading is available, 'shel does not appear to be the realization of case assigned by the noun to its object': however, she does not elaborate on what this might mean. What is important for present purposes is that the 'NOS' reading is both available, and preferred. 19 In addition to the adjective-ordering facts discussed above, there is a good deal of other empirical evidence - from expressions involving numeral and classifier phrases in Irish which shows clearly that overt N°-Num° raising is a completely general operation, occurring in all noun phrases, irrespective of whether or not they contain a possessor subject. Due to space constraints, this cannot be elaborated here; see Duffield (1995), also Ritter (1993). 20 Once again, the issue is complicated by the existence of superficially very similar N-NP compounds which may be indefinite in both Hebrew and Maltese. See Borer (1988) and Fabri (1993) for discussion and analysis. 21 This is certainly true of multiply embedded CSNs in Hebrew, as Ritter (1991: n. 8) acknowledges in a footnote: (i)

(ii)

ben xaver ha-mora friend son DET-teacher 'the teacher's friend's son' ???ben xaver mora son friend teacher 'a teacher's friend's son'

Structural in variance and lexical diversity

339

22 It seems not implausible that the distinct Construct form found in many Semitic varieties should be analysed as the phonological 'spell-out' of an inherent (lexically specified) definiteness. See Ritter (1991: 40ff.), however, for a different interpretation of this effect. 23 It may appear as though this movement (to D°) does not take place whenever numerals are present: since in Irish - though not in Semitic - CSNs may be modified by numeral phrases which surface, in most cases, to the left of the head noun; hence tri charr Shedin 'Sean's three cars'. However, it is not clear that this prenominal position is the base position of numerals, since some numeral phrases, notably 'teens', realized in Irish as deag, obligatorily appear between the head noun and the possessor, as would be expected if head-raising to D° had taken place: for example, tri charr deag Shedin 'Sean's thirteen cars'. It is possible, therefore, that numeral phrases themselves raise independently across D°, to give the usual (surface) distribution. Clearly, more requires to be said about these phrases: however, numeral phrases raise more general, complex questions that are beyond the scope of this chapter, though see Duffield (1995: chapter 5) for some discussion. Thanks once again to Bob Borsley for drawing my attention to the relevance of these cases. 24 Nevertheless, this analysis raises a potentially serious question about demonstrative placement: although it can provide a systematic account of demonstrative licensing and interpretation, it seems unable to account for the placement of demonstrative elements, which seem to require a complementary, and quite different, syntax. The correct observation about the distribution of demonstrative elements seems to be that they attach to the right of the head-noun complex, that is, either to the right of the head noun, or to the right of the head noun plus any modifying adjectives. If it is assumed that demonstratives project their own structural position, then this is certainly a problem, since the present analysis cannot easily represent the constituency of [N AP], unless head-raising through A° is assumed (see note 9). However, it is less of a problem if demonstratives are treated as having no direct syntactic role. See Duffield (1995) for further discussion. 25 As discussed in Duffield (1995), there are cases of grammatical structures with [DET N DET N] word order; however, in almost all cases, these can be shown to be compounds, rather than CSNs; once again, see Borer (1988) and Fabri (1993) for discussion of similar cases in Hebrew and Maltese, respectively. 26 That verbal-noun constructions should be considered a type of DP internally is not uncontroversial: Borsley (1993), for example, claims that the corresponding verb-noun constructions in Welsh have a verbal (infinitival) rather than nominal status in the synchronic grammar. It is true that Irish verbal-noun constructions do differ from CSNs in certain respects, for example in requiring adverbial, rather than adjectival modification of the verbal noun, and in not permitting these modifying adverbials to intervene between the head noun and the complement. This notwithstanding, it remains true that with respect to the Case- and theta-assigning properties at issue here, in particular with respect to the interaction between Case and Definiteness, the two constructions show extremely similar behaviour. Whether verbal-noun constructions involve some additional (external) dominating functional projection, perhaps Object Agreement, as suggested in Duffield (1991),

340

Nigel Duffield

should be regarded as a separate issue. See Ramchand (1993b) and Adger (this volume) for related proposals for Scots Gaelic. 27 Assuming parallelism in the relevant respects between verbal-noun constructions and CSNs (see note 25), we would expect under this account that indefinite CSNs which do not have a verbal noun as their head should none the less display identical Case properties, that is, the indefinite possessor should optionally be Case-marked nominative. For some reason, for which I have no interesting formal explanation, indefinite CSNs seem to be dispreferred with either genitive (teach fhir 'a man's house') or nominative (teach fear 4a man's house') Case; nevertheless, both are possible, as predicted.

References

Abney, S.P. 1987. The English noun-phrase in its sentential aspect', doctoral dissertation, MIT. Acquaviva, P. 1993. 'The logical form of negation: a study of operator-variable structures in syntax', doctoral dissertation, Scuola Normale Superiore, Pisa. Adger, D. 1994a. 'Functional heads and interpretation', doctoral dissertation, Centre for Cognitive Science, University of Edinburgh. 1994b. 'The licensing of quasi-arguments', in P. Ackema and M. Schoorlemmer (eds.) Proceedings of the Conference of the Student Organisation of Linguistics in Europe (CONSOLE), University of Utrecht, 1-18. Adger, D. and C. S. Rhys 1993. 'Eliminating disjunction in lexical specification', unpublished ms., November 1993, to appear in P. Coopmans, M. Everaert and J. Grimshaw (eds.) Lexical Specification and Lexical Insertion: Selected Papers from a Workshop at the University of Utrecht, Hillsdale, NJ: Lawrence Erlbaum. Anderson, S. 1981. 'Topicalization in Breton', Proceedings of the Berkeley Linguistics Society 7: 27-39. Anderson, S. and S. Chung 1977. 'On grammatical relations and clause structure in verbinitial languages', in P. Cole and J. Sadock (eds.) Syntax and Semantics, vol. VIII: Grammatical Relations, New York: Academic Press, 1-25. Andrews, A. 1990. 'Unification and morphological blocking', Natural Language and Linguistic Theory 8: 507-57. Anwyl, E. 1899. Welsh Grammar, part II: Syntax, London: Swan Sonnenschein. Aronoff, M. 1976. Word Formation in Generative Grammar, Cambridge, MA: MIT Press. Authier, J. Marc 1992. 'Iterated CPs and embedded topicalization', Linguistic Inquiry 23: 329-34. Awbery, G. 1976. The Syntax of Welsh: a Transformational Study of the Passive, Cambridge University Press. 1977. 'A transformational view of Welsh relative clauses', Bulletin of the Board of Celtic Studies 27: 155-206. 1990. 'Dialect syntax: a neglected resource for Welsh', in Hendrick (ed.), 1-25. 1994. 'Echo pronouns in a Welsh dialect: a system in crisis?', in I. Roberts (ed.) Research Papers in Welsh Syntax (Bangor Research Papers in Linguistics 5), 1-29. Baker, M. 1985. 'The mirror principle and morphosyntactic explanation', Linguistic Inquiry 16: 373^16. 1988. Incorporation: a Theory of Grammatical Function Changing, Chicago University Press. 1993. 'Why unaccusative verbs cannot dative shift', in A. J. Schafer (ed.) Proceedings of NELS 23: 33-^7.

341

342

References

Baker, M., K. Johnson and I. Roberts 1989. 'Passive arguments raised', Linguistic Inquiry 20: 173-251. Baltin, M. 1982. 'A landing site theory of movement rules', Linguistic Inquiry 13: 1-38. Barwise, J. and R. Cooper 1981. 'Generalized quantifiers and natural language', Linguistics and Philosophy 4: 159-219. Belletti, A. 1988. 'The case of unaccusatives', Linguistic Inquiry 19: 1-34. 1990. Generalized Verb Movement, Turin: Rosenberg and Sellier. 1993. 'Case checking and clitic placement: three issues in Italian/Romance', unpublished ms., University of Geneva. Belletti, A. and U. Shlonsky 1995. 'The order of verbal complements: a comparative study', Natural Language and Linguistic Theory 13: 489—526 Bennis, H. 1986. Gaps and Dummies, Dordrecht: Foris. Bennis, H. and T. Hoekstra 1989. 'Why Kaatje wasn't heard sing a song', in W. Jaspers (ed.) Sentential Complementation and the Lexicon: Studies in Honor of W. de Geest, Dordrecht: Foris, 21^0. Benveniste, E. 1960. 'Etre et avoir dans leurs fonctions linguistiques', originally in Bulletin de la Societe Linguistique de Paris 55 (1960); repr. in Benveniste, Problemes de linguistique generate, Paris: Gallimard, 1966. Berman, R. 1987. 'A developmental route: learning about the forms and use of complex nominals in Hebrew', Linguistics 25: 1057-85. Besten, H. den 1983. 'On the interaction of root transformations and lexical deletive rules', in W. Abraham (ed.) On the Formal Syntax of the Westgermania, Amsterdam: John Benjamins, 47-138. 1990. 'Studies in West Germanic syntax', doctoral dissertation, University of Tilburg. Bhaldraithe, T., de 1975. Gaeilge Chois Fhairrge: an Deilbhiocht, Dublin Institute for Advanced Studies. Bobaljik, J. D. 1994. 'What does adjacency do?', in H. Harley and C. Phillips (eds.) The Morphology-Syntax Connection, MIT Working Papers in Linguistics 22: 1-32. Bobaljik, J. D. and A. H. Carnie 1992. 'A minimalist approach to some problems of Irish word order', presented to the twelfth Harvard Celtic Colloquium. Revised version published in the present volume. Bobaljik, J.D. and D. Jonas forthcoming. 'Subject positions and the roles of TP', Linguistic Inquiry. Borer, H. 1984a. Parametric Syntax, Dordrecht: Foris. 1984b. 'Restrictive relative clauses in Modern Hebrew', Natural Language and Linguistic Theory 2: 219-60. 1988. 'On the morphological parallelism between compounds and constructs', Yearbook of Morphology 1: 45-66. 1993. 'On the projection of arguments', in E. Benedicto and J. Runner (eds.) Functional Projections (University of Massachusetts Occasional Papers 17), University of Massachusetts, Amherst. 1995. 'The ups and downs of Hebrew verb movement', Natural Language and Linguistic Theory 13: 527-606. Borsley, R.D. 1986. 'Prepositional complementizers in Welsh', Journal of Linguistics 20: 277-302.

References

343

1990. 'A GPSG approach to Breton word order', in Hendrick (ed.), 81-95. 1992. 'Celtic clause structure', in K. Borjars and N. Vincent (eds.) Functional Categories in Complementation {Eurotyp Working Paper 111,3), European Science Foundation, 3 21. 1993. 'On so-called verb-nouns in Welsh', Journal of Celtic Linguistics 2: 35-64. 1995. 'On some similarities and differences between Welsh and Syrian Arabic', Linguistics 33: 99-122. Borsley, R.D. and J. Stephens 1989. 'Agreement and the position of subjects in Breton', Natural Language and Linguistic Theory 7: 407-27. Borsley, R.D. and M.O. Tallerman 1996. 'Phrases and soft mutation in Welsh', Journal of Celtic Linguistics 5. Brandi, L. and P. Cordin 1989. 'Two Italian dialects and the null subject parameter', In Jaeggli and Saffir (eds.), 111-42. Brody, M. 1990. 'Some remarks on the focus field in Hungarian', Working Papers in Linguistics 2: 201-25. University College, London. Burzio, L. 1986. Italian Syntax, Dordrecht: Kluwer. Campbell, R. 1991. 'Tense and agreement in different tenses', The Linguistic Review 8: 159— 83. Cann, R. 1993. 'Syntactic projection and argument structure', unpublished ms. presented at the spring meeting of the Linguistics Association of Great Britain, Birmingham, England. Cardinaletti, A. 1994. 'On the internal structure of pronominal DPs', The Linguistic Review 11: 195-220. Cardinaletti, A. and I. Roberts 1991. 'Clause structure and X-second', to appear in W. Chao and G. Horrocks (eds.) Levels, Principles and Processes: the Structure of Grammatical Representations, Berlin: Foris/de Gruyter. Cardinaletti, A. and M. Starke 1994. 'The typology of structural deficiency', unpublished ms., Universities of Geneva and Venice. Carlson, G. 1977. 'Reference to kinds in English', doctoral dissertation, University of Massachusetts, Amherst. Carnie, A. 1993. 'Nominal predicates and absolutive case marking in Irish', MIT Working Papers in Linguistics 19: 89-129. Forthcoming. 'On the theory of movement and non-verbal predication', doctoral dissertation, MIT. Carnie, A. and H. Harley 1994. 'Predicate raising and the Irish copula', unpublished ms., MIT. Carnie, A., E. Pyatt and H. Harley forthcoming. 'The resurrection: raising to comp? Some evidence from Old Irish', to appear in Studies in the Linguistic Sciences 24. Cavar, D. and C. Wilder 1992. 'Long head movement? Verb movement and cliticization in Croatian', Arbeitspapier Nr. 7 (Institut fur Deutsche Sprache und Literatur II). Johann Wolfgang Goethe-Universitat, Frankfurt-on-Main. Chomsky, N. 1965. Aspects of the Theory of Syntax, Cambridge, MA: MIT Press. 1975. Reflections on Language, New York: Pantheon. 1980. Rules and Representations, New York: Columbia University Press. 1981. Lectures on Government and Binding, Dordrecht: Foris.

344

References

1982. Some Concepts and Consequences of the Theory of Government and Binding, Cambridge, MA: MIT Press. 1986a. Knowledge of Language: its Nature, Origin and Use, New York: Praeger. 1986b. Barriers (Linguistic Inquiry Monograph 13), Cambridge, MA: MIT Press. 1988. Language and Problems of Knowledge: the Managua Lectures, Cambridge, MA: MIT Press. 1991. 'Some notes on economy of derivation and representation', in R. Freidin (ed.) Principles and Parameters in Comparative Grammar, Cambridge, MA: MIT Press, 417-54. 1993. 'A minimalist program for linguistic theory', in K. Hale and S. J. Keyser (eds.) The View from Building 20, Cambridge, MA: MIT Press, 1-52. 1994. 'Bare phrase structure', MIT Occasional Papers in Linguistics 5. Chomsky, N. and M. Halle 1968. The Sound Pattern of English, New York: Harper & Row. Chomsky, N. and H. Lasnik 1993. 'Principles and Parameters theory', in J. Jacobs, A. von Stechow, W. Sternefeld and T. Venneman (eds.) Syntax: an International Handbook of Contemporary Research, Berlin: Walter de Gruyter, 506-69. Christian Brothers 1990. New Irish Grammar, Dublin: C. J. Fallon. Chung, S. 1990. 'VPs and verb movement in Chamorro', Natural Language and Linguistic Theory 8: 559-620. Chung, S. and J. McCloskey 1987. 'Government, barriers and small clauses in Modern Irish', Linguistic Inquiry 18: 173-237. Forthcoming. 'Existentials in two VSO languages', Linguistics Research Center, University of California, Santa Cruz. Cinque, G. 1990. Types ofA-bar Dependencies, Cambridge, MA: MIT Press. 1993. 'On the evidence for partial N-movement in the Romance DP', unpublished ms., University of Venice. Clack, S. 1993. 'A more nominal verb-noun in Middle Welsh', paper presented at the autumn meeting of the Linguistics Association of Great Britain. Comrie, B. 1977. 'In defense of spontaneous demotion', in P. Cole and J. Sadock (ed.) Syntax and Semantics, vol. VIII: Grammatical Relations, New York: Academic Press, 25-55. Cram, D. 1981. 'Scottish Gaelic syntax: a transformational approach', unpublished ms., University of Aberdeen. Crisma, P. 1990. 'Functional categories inside the noun phrase: a study on the distribution of nominal modifiers', Tesi di Lauria, University of Venice. Dechaine, R.-M. 1993. 'Predicates across categories: towards a category-neutral syntax', doctoral dissertation, University of Massachusetts, Amherst. Delsing, L.-O. 1990. 'The Scandinavian noun phrase', Working Papers in Scandinavian Syntax 42: 57-79. Di Sciullo, A. and E. Williams, 1986. On the Definition of Word, Cambridge, MA: MIT Press. Diesing, M. 1990. 'Verb movement and the subject position in Yiddish', Natural Language and Linguistic Theory 8: 41-79. 1992. Indefinites, Cambridge, MA: MIT Press. Dikken, M. den 1994. 'Predicate inversion and minimality', in R. Bok-Bennema and C.

References

345

Cremers (eds.) Linguistics in the Netherlands, Amsterdam: John Benjamins, 1-12. Dillon, M. 1941. 'Modern Irish atd se deanta agam "I have done it"', Language 17: 49-50. Dobrovie-Sorin, C. 1990. 'Clitic doubling, WH-movement, and quantification in Roumanian', Linguistic Inquiry 21: 351-97. Forthcoming. The Syntax of Romanian: Comparative Studies in Romance, Berlin: Mouton de Gruyter. Doherty, C. 1992. 'Clause structure and the Modern Irish copula', Syntax at Santa Cruz 1: 65-91; to appear in Natural Language and Linguistic Theory. Dooley Collberg, S. 1991. 'The status of agr in Modern Irish', Working Papers 37: 129-60, Lund University, Department of Linguistics. Doron, E. 1988. 'On the complementarity of subject and subject-verb agreement', in M. Barlow and C. A. Ferguson (eds.) Agreement in Natural Language: Approaches, Theories, Descriptions, Center for the Study of Language and Information, Stanford, 201-18. Drijkoningen, F. and J. Rutten 1991. 'Government and temporal reference', in P. Coopmans, B. Schouten and W. Zonneveld (eds.) OTS Yearbook 1991, Dordrecht: ICG Printing, 1-18. Duffield, N. 1990a. 'Movement and mutation in Modern Irish', unpublished ms., University of Southern California. 1990b. 'Movement and the case of the Irish verbal noun', in B. Bird, K. Hunt and V. Samilan (eds.) Proceedings of the Western Conference in Linguistics, vol. Ill, 79-87. 1991. 'Particles and projections', doctoral dissertation, University of Southern California. 1992. 'The construct state in Celtic and Semitic: part one', in A. Henry (ed.) Belfast Working Papers in Linguistics, University of Ulster, Jordanstown, 1-67. 1993. 'Irish construct state nominals and the radical pro-drop phenomenon', Proceedings of NELS 23: 113-27. 1994. 'Are you right? On pronoun-postposing and other problems of Irish word order', to appear in Proceedings of the Thirteenth West Coast Conference on Formal Linguistics, Stanford: Stanford Linguistic Association. 1995. Particles and Projections in Irish Syntax, Dordrecht: Kluwer. Emonds, J. 1976. A Transformational Approach to English Syntax, New York: Academic Press. 1978. 'The Complex V - V in French', Linguistic Inquiry 9: 151-75. 1981. 'Word order in generative grammar', Journal of Linguistic Research 1: 33-54. Enc, M. 1991, 'On the absence of present tense in English', unpublished ms., University of Wisconsin, Madison. Ernst, T. 1992. 'Phrase structure and directionality in Irish', Journal of Linguistics 28: 415— 43. Evans, D. S. 1989. A Grammar of Middle Welsh, Dublin Institute for Advanced Studies. Everett, D. 1989. 'Clitic doubling, reflexives and word-order alternations in Yagua', Language 65: 339-72. Ey{)6rsson, T. 1994. 'Functional categories, cliticization, and verb movement in the early Germanic languages', paper presented at the ninth Comparative Germanic Syntax Workshop, Harvard. Fabb, N. 1984. 'Syntactic affixation', doctoral dissertation, MIT.

346

References

Fabri, R. 1993. Kongruenz und die Grammatik des Maltesischen, Tubingen: Max Niemeyer. Fassi-Fehri, A. 1993. Issues in the Syntax of Arabic Clauses and Words, Dordrecht:Kluwer. Fauconnier, G. 1992. 'Roles and values: the case of French copula constructions', in C. Georgopoulos and R. Ishihara (eds.) Essays in Honour of S.-Y. Kuroda, Dordrecht: Kluwer, 181-206. Fife, J. 1990. The Semantics of the Welsh Verb: a Cognitive Approach, Cardiff: University of Wales Press. Fife, J. and G. King 1991. 'Focus and the Welsh "abnormal sentence": a cross-linguistic perspective', in Fife and Poppe (eds.), 81-153. Fife, J. and E. Poppe (eds.) 1991. Studies in Brythonic Word Order, Amsterdam: John Benjamins. Fillmore, C. 1968. 'The case for Case', in E. Bach and R. Harms (eds.) Universals in Linguistic Theory, New York: Holt, Rinehart and Winston, 1-88. Frampton, J. 1991. 'Relativized minimality, a review', The Linguistic Review 8: 1-46. Freeze, R. 1992. 'Existentials and other locatives', Language 68: 553-95. Fukui, N. and M. Speas 1986. 'Specifiers and projection', Papers in Theoretical Linguistics: MIT Working Papers in Linguistics 8: 129-72. Gazdar, G., E. Klein, G. K. Pullum and I. A. Sag 1985. Generalized Phrase Structure Grammar, Oxford: Basil Blackwell. Giorgi, A. and G. Longobardi 1990. The Syntax of Noun-Phrases, Cambridge University Press. Giorgi, A. and F. Pianesi 1992. 'From semantic representations to morphosyntactic structures', unpublished ms. presented at Going Romance, Utrecht. Glinert, L. 1989. The Grammar of Modern Hebrew, Cambridge University Press. Gramadeg Cymraeg Cyfoes. Contemporary Welsh Grammar. 1985. Cowbridge, Wales: D. Brown. Greenberg, J. 1963. 'Some universals of grammar with particular reference to the order of meaningful elements', in J. H. Greenberg (ed.) Universals of Language, Cambridge, MA: MIT Press, 73-113. Greene, D. 1979. 'Perfects and perfectives in Modern Irish', Eriu 30: 122-41. Grimshaw, J. 1990. Argument Structure, Cambridge, MA: MIT Press. 1991. 'Extended projection', unpublished ms., from the LSA Summer Institute, Santa Cruz, CA. 1993. 'Minimal projection, heads and optimality', unpublished ms., Rutgers University. Groat, E. forthcoming. 'English expletives: a minimalist approach', Linguistic Inquiry. Gueron, J. 1989. 'Subject, tense and indefinite NPs', Proceedings of NELS 19: 142-60. 1992. 'Types syntaxiques et types semantiques: la phrase copulative comme palimpseste', unpublished ms., Universite de Paris X. Gueron, J. and T. Hoekstra 1988. 'T-chains and the constituent structure of auxiliaries', in A. Cardinaletti, G. Cinque and G. Giusti (eds.) Constituent Structure, Dordrecht: Foris, 35-99. Guilfoyle, E. 1988. 'Parameters and functional projection', Proceedings of NELS 18: 193— 207. 1990. 'Functional categories and phrase structure parameters', doctoral dissertation, McGill University.

References

347

1993. 'Nonfinite clauses in Modern Irish and Old English', proceedings of the twentyninth Regional Meeting of the Chicago Linguistic Society, Chicago, IL, 199-214. Haegeman, L. 1991. Introduction to Government and Binding Theory, 1st edn, Oxford: Basil Blackwell. 1992. 'Sentential negation in Italian and the negative criterion', unpublished ms., Universite de Geneve. 1994. Introduction to Government and Binding Theory, 2nd edn, Oxford: Blackwell. Haegeman, L. and R. Zanuttini 1991. 'Negative heads and the Neg criterion', The Linguistic ReviewS: 233-51. Hale, K. 1987. 'Incorporation and the Irish synthetic verb forms', unpublished ms., MIT. 1989. 'Some remarks on agreement and incorporation', unpublished ms., MIT. Hale, K., L. Jeanne and P. Platero 1977. 'Three cases of overgeneration', in P. Culicover, T. Wasow and A. Akmajian (eds.) Formal Syntax, New York: Academic Press, 374416. Halle, M. and A. Marantz. 1992. 'Distributed morphology and the pieces of inflection' in K. Hale and S. J. Keyser (eds.) The View from Building 20: Essays in Honor of Sylvain Baumberger. Cambridge, MA: MIT Press, 111-176. Harley, H. 1993. 'The case of the Icelandic experiencer', unpublished ms., MIT. Harlow, S. J. 1981. 'Government and relativization in Celtic', in F. Heny (ed.) Binding and Filtering, London: Croom Helm, 213-54. 1983. 'Celtic relatives', York Papers in Linguistics 10: 77-121. 1989. 'Syntax of Welsh soft mutation', Natural Language and Linguistic Theory 7: 289316. 1992. 'Finiteness and Welsh sentence structure', in H. Obenauer and A. Zribi-Hertz (eds.) Structure de la phrase et theorie du liage, Saint-Denis: Presses Universitaires de Vincennes, 93-119. Hartmann, H. 1974. 'Distribution und Funktion der Expanded Form in einigen Dialekten der irischen Sprache von Co. Galway', Zeitschrift fur Celtische Philologie 33: 140— 284. Hazout, I. 1991. 'Verbal nouns: theta-theoretic studies in Hebrew and Arabic', doctoral dissertation, University of Massachusetts, Amherst. Heim, I. 1982. 'The semantics of definite and indefinite noun phrases', doctoral dissertation, University of Massachusetts, Amherst. Hendrick, R. 1988. Anaphora in Celtic and Universal Grammar, Dordrecht: Reidel. 1990a. 'Breton pronominals, binding and barriers', in Hendrick (ed.), 121-65. (ed.) 1990b. Syntax and Semantics, vol. XXIII: The Syntax of the Modern Celtic Languages, San Diego: Academic Press. 1991. 'The morphosyntax of aspect', Lingua 85: 171-210. 1994. 'The Brythonic copula and head-raising', in D. Lightfoot and N. Hornstein (eds.) Verb Movement, Cambridge University Press, 163-88. Hewitt, S. 1990. 'The progressive in Breton in the light of the English progressive', in M. J. Ball, J. Fife, E. Poppe and J. Rowland (eds.) Celtic Linguistics: Readings in the Brythonic Languages, Amsterdam: John Benjamins, 167-88. Heycock, C. 1991. 'Layers of predication: the non-lexical syntax of clauses', doctoral dissertation, University of Pennsylvania.

348

References

Higginbotham, J. 1985. 'On semantics', Linguistic Inquiry 16: 547-93. 1987. Indefiniteness and predication', in Reuland and ter Meulen (eds.), 43-70. 1989. Elucidations of meaning', Linguistics and Philosophy 12: 465-517. Higgins, F.R. 1973. The pseudo-cleft construction in English', doctoral dissertation, MIT. Hoeksema, J. 1986. 'Monotonicity phenomena in natural language', Linguistic Analysis 16: Hoekstra, T. and R. Mulder 1990. 'Unergatives as copular verbs: locational and existential predication', The Linguistic Review 7: 1-79. Holmberg, A. 1986. 'Word order variation in the Scandinavian languages and English', doctoral dissertation, University of Stockholm. Holmberg, A. and C. Platzack 1991. 'On the role of inflection in Scandinavian syntax', in W. Abraham, W. Kosmeijar and E. Reuland (eds.), Issues in Germanic Syntax, Berlin: Mouton de Gruyter, 93-118. Honcoop, M. 1992. 'ER or OR: quantification, intervention and fission', Master's dissertation, Department of Linguistics, University of Leiden. Hornstein, N. 1990. As Time Goes By: Tense and Universal Grammar, Cambridge, MA: MIT Press. 1994. 'An argument for minimalism: the case of antecedent-contained deletion', Linguistic Inquiry 25: 455-80. Hoyt, K. 1989. 'Verb raising in Lebanese Arabic', MIT Working Papers in Linguistics 11. Jaeggli, O. 1982. Topics in Romance Syntax, Dordrecht: Foris. 1986. 'Three issues in the theory of clitics: case, doubled NPs, and extraction', in H. Borer (ed.) Syntax and Semantics, vol. XIX: The Syntax of Pronominal Clitics, New York: Academic Press, 15-42. Jaeggli, O. and K. Safir (eds.) 1989. The Null Subject Parameter, Dordrecht: Kluwer. Jespersen, O. 1924. The Philosophy of Grammar, London: George Allen and Unwin. Johnson, K. 1991. 'Object positions', Natural Language and Linguistic Theory 9: 577-636. Jones, B. M. 1993. 'Ascriptive and equative sentences in children's Welsh', Papurau Addysg Aberystwyth. Jones, B. M. and A. R. Thomas 1977. The Welsh Language: Studies in its Syntax and Semantics, Cardiff: University of Wales Press. Kamp, H. 1981. 'A theory of truth and semantic representation', in J. Groenendijk, T. Janssen and M. Stokhof (eds.) Formal Methods in the Study of Language, Mathematical Center, Amsterdam, 277-322; reprinted in J. Groenendijk, T. Janssen and M. Stokhof (eds.) 1984. Truth, Interpretation and Information. Dordrecht: Foris, 277-322. Kayne, R. S. 1975. French Syntax, Cambridge, MA: MIT Press. 1981. 'Two notes on the N I C , in A. Belletti, L. Brandi and L. Rizzi (eds.) Theory of Markedness in Generative Grammar, Pisa: Scuola Normale Superiore, 317-46. 1989a. 'Facets of Romance past participle agreement', in P. Beninca (ed.) Dialect Variation on Generative Grammar, Dordrecht: Foris, 85-104. 1989b. 'Notes on English agreement', CIEFL Bulletin, Hyderabad, 1: 41-67. 1991. 'Romance clitics, verb-movement and PRO', Linguistic Inquiry 22: 647-86. 1993. 'Toward a modular theory of auxiliary selection', Studia Linguistica 47: 3-31. 1995. The Antisymmetry of Syntax, Cambridge, MA: MIT Press.

References

349

Keenan, E. 1987. 'A semantic definition of "Indefinite N P " ' , in Reuland and ter Meulen (eds.) Keenan, E. and J. Stavi 1986. 'A semantic characterization of natural language determiners', Linguistics and Philosophy 9: 253-326. Kenstowicz, M. and W. Wahba 1980. 'Clitics and the double object construction in Cairene Arabic', Studies in the Linguistic Sciences 10.2: 149-63. Kitagawa, Y. 1986. 'Subject in Japanese and English', doctoral dissertation, University of Massachusetts, Amherst. Klima, E. 1964. 'Negation in English', in J. A. Fodor and J. J. Katz (eds.) The Structure of Language, Englewood Cliffs, NJ: Prentice-Hall, 246-323. Koopman, H. 1993. 'The internal and external distribution of pronominal DPs', unpublished ms., University of California, Los Angeles. Koopman, H. and D. Sportiche 1989. 'Subjects', unpublished ms., University of California, Los Angeles. 1991. 'On the position of subjects', Lingua 85: 211-58. Koster, J. 1975. 'Dutch as an SOV language', Linguistic Analysis 1: 111-36. 1987. Domains and Dynasties: the Radical Autonomy of Syntax, Dordrecht: Foris. Kratzer, A. 1989. 'Stage-level and individual-level predicates', unpublished ms., University of Massachusetts, Amherst. Krifka, M. 1990. 'Four thousand ships passed through the lock: object-induced measure functions on events', Linguistics and Philosophy 13: 487-520. Kuroda, S.-Y. 1988. 'Whether we agree or not', Lingvisticae Investigationes 12: 1-47. Ladusaw, W. 1979. 'Polarity sensitivity as inherent scope relations', doctoral dissertation, University of Texas, Austin. Laka, M. I. 1990. 'Negation in syntax: on the nature of functional categories and projections', doctoral dissertation, MIT. Larson, R. 1988. 'On the double object construction', Linguistic Inquiry 19: 335-91. Lasnik, H. 1992. 'Case and expletives: notes towards a parametric account', Linguistic Inquiry 23: 381-405. 1993. 'Lectures on minimalist syntax', University of Connecticut Occasional Papers in Linguistics 1. Lasnik, H. and M. Saito. 1992. Move Alpha, Cambridge, MA: MIT Press. Lema, J. and M.-L. Rivero 1989a. 'Long head movement: ECP vs H M C , Proceedings of NELS 20: 333-47. 1989b. 'Inverted conjugations and V-second effects in Romance', in C. Laeufer and T. Morgan (eds.) Theoretical Analyses in Contemporary Romance Linguistics: Selected Papers from the Nineteenth Symposium, on Romance Linguistics (LSRL XIX), 21-23 April 1989, Amsterdam: John Benjamins, 311-28. 1991. 'Types of verbal movement in Old Spanish: modals, futures, and perfects', Probus 3: 237-78. Lewis, H. and H. Pedersen. 1989. A Concise Comparative Celtic Grammar, 3rd edn, Gottingen: Vandenhoeck and Ruprecht. Lobel, E. 1990. 'Typologische Aspekte funktionaler Kategorien in der Nominalphrase', Zeitschrift fur Sprachwissenschaft 9: 135-69. Longobardi, G. 1983. 'Le frasi copulari in italiano e la struttura della teoria sintattica',

350

References

Annali delta Scuola Normale di Pisa 13. 4: 1151-64. 1994. 'Reference and proper names', Linguistic Inquiry 25(4): 609-66. Lyons, C. 1992. T h e construct: VSO genitive structures', unpublished ms., University of Salford. Lyons, J. 1967. 'A note on possessive, existential and locative sentences', Foundations of Language 3: 390-6. MacAulay, D. (ed.) 1992. The Celtic Languages, Cambridge University Press. McCarthy, J. J. and A. S. Prince 1993. 'Prosodic morphology 1: constraint interaction and satisfaction', unpublished ms., University of Massachusetts, Amherst. McCloskey, J. 1979. Transformational Syntax and Model Theoretic Semantics, Dordrecht: Reidel. 1980. 'Is there raising in Modern Irish?', Eriu 31: 59-99. 1983.'A VP in a VSO language', in G. Gazdar, E. Klein and G. K. Pullum (eds.) Order, Concord and Constituency, Dordrecht: Foris, 9-55. 1983. 'Raising, subcategorization and selection in Modern Irish', Natural Language and Linguistic Theory 1: 441-85. 1986. 'Case, movement and raising in Modern Irish', in J. Goldberg, S. MacKaye and M. Wescoat (eds.) Proceedings of WCCFL 4: 190-205. 1990. 'Resumptive pronouns, A-bar binding and levels of representation in Irish', in Hendrick (ed.), 199-248. 1991a. 'Dialects and parameters: the case of Irish', paper presented to the thirteenth Annual Celtic Studies Conference, University of California, Berkeley, 15-17 March. 1991b. 'Clause structure, ellipsis and proper government in Irish', Lingua 85: 259-302. 1992. 'Adjunction, selection and embedded verb second', unpublished ms., University of California at Santa Cruz. Forthcoming. 'On the scope of verb-movement in Irish', to appear in Natural Language and Linguistic Theory. McCloskey, J. and K. Hale 1984. 'On the syntax of person-number inflection in Modern Irish', Natural Language and Linguistic Theory 1: 487-533. McCloskey, J. and P. Sells 1988. 'Control and A-chains in Modern Irish', Natural Language and Linguistic Theory 6: 143-89. McDaniel, D. 1989. 'Partial and multiple wh-movement', Natural Language and Linguistic Theory 7: 565-604. Mahajan, A. 1990. 'The A/A-bar distinction and movement theory', doctoral dissertation, MIT. Manzini, M. R. 1992a. Locality: a Theory and its Empirical Consequences, Cambridge, MA: MIT Press. 1992b. 'Second position dependencies', paper presented at the eighth Workshop on Germanic Syntax, Tromso. Marantz, A. 1995. 'A reader's guide to the minimalist program for linguistic theory', in G. Webelhuth (ed.) The Principles and Parameters Approach to Syntactic Theory: a Synopsis, Oxford: Basic Blackwell. May, R. 1985. Logical Form, Cambridge, MA: MIT Press. Milsark, G. 1977. 'Toward an explanation of certain peculiarities in the existential construction in English', Linguistic Analysis 3: 1-30.

References

351

Mohammad, M. 1988. 'On the parallelism between IP and DP', in Proceedings of the Seventh West Coast Conference on Formal Linguistics, Stanford Linguistics Association. Moritz, L. 1989. 'Apercu de la syntaxe de la negation en Francais et en Anglais', memoire de licence, Universite de Geneve. Moritz, L. and D. Valois 1992. 'French sentential negation and LP-pied piping', Proceedings of NELS 23: 319-33. Moro, A. 1988. 'Per una teoria unificata delle frasi copulari', Rivista di Grammatica Generativa 13: 35-57. 1991. 'The raising of predicates: copula, expletives and existence', MIT Working Papers in Linguistics 15: 119-81. Morris-Jones, J. 1913. A Welsh Grammar: Phonology and Accidence, Oxford University Press. Mouchaweh, L. 1986. 'De la Syntaxe des petites propositions', These de doctorat, Universite de Paris VIII. Nishigauchi, T. 1990. Quantification in the Theory of Grammar, Dordrecht: Kluwer. Noonan, M. 1993. 'Statives, perfectives and accusativity: the importance of being have\ in J. Mead (ed.) Proceedings of Eleventh West Coast Conference on Formal Linguistics, Stanford Linguistics Association. O Huallachain, C. and M. O. Murchu 1976. Irish Grammar, Coleraine: New University of Ulster. O'Rahilly, T. 1932/1972. Irish Dialects Past and Present. First published by Browne and Nolan, Dublin 1932, re-issued, Dublin Institute for Advanced Studies, 1972. O Se, D. 1983. 'Gaeilge Chorea Dhuibhne: an fhoneolaiocht agus an deilbhiocht', doctoral dissertation, University College Dublin. 1993. 'The perfect in Modern Irish', Eriu 43: 39-67. O Siadhail, M. 1973. 'Abairti freagartha agus mireanna freagartha Sa Nua-Ghaeilge', Eriu 24: 134-59. 1989. Modern Irish: Grammatical Structure and Dialectal Variation, Cambridge University Press. Ouhalla, J. 1989. 'Clitic movement and the ECP: evidence from Berber and Romance languages', Lingua 79: 165-215. 1991a. Functional Categories and Parametric Variation, London: Routledge. 1991b. 'Sentential negation, relativized minimality and the aspectual status of auxiliaries', The Linguistic Review 7: 183-231. 1993. 'Subject-extraction, negation and the anti-agreement effect', Natural Language and Linguistic Theory 11: 477-518. 1994. 'Verb movement and word order in Arabic', in D. Lightfoot and N. Hornstein (eds.) Verb Movement, Cambridge University Press, 41-72. Partee, B. 1987. 'Noun phrase interpretation and type-shifting principles', in J. Groenendijk, D. de Jongh and M. Stokhof (eds.) Studies in Discourse Representation Theory and the Theory of Generalized Quantifiers, Dordrecht: Foris, 115-44. Partee, B. H., E. Bach and A. Kratzer 1987. 'Quantification: a crosslinguistic perspective', unpublished ms. (National Science Foundation proposal), University of Massachusetts, Amherst.

352

References

Perlmutter, D. and P. Postal 1984. 'The 1-Advancement Exclusiveness Law', in D. Perlmutter and C. Rosen (eds.) Studies in Relational Grammar, vol. II, University of Chicago Press, 81-125. Pesetsky, D. 1991. 'Infinitives', unpublished ms., MIT. 1994. 'Class lectures', unpublished ms., MIT. Platzack, C. 1987. 'The Scandinavian languages and the Null-Subject Parameter', Natural Language and Linguistic Theory 5(3): 377-402. Poletto, C. 1993. 'La sintassi dei clitici soggetto nei dialetti italiani settentrionali', doctoral dissertation, Universities of Venice and Padua. Pollock, J.-Y. 1989. 'Verb-movement, Universal Grammar, and the structure of IP', Linguistic Inquiry 20: 365-424. Press, I. 1986. Grammar of Modern Breton, Berlin: Mouton de Gruyter. Ramchand, G. 1993a. 'Aspect and argument structure in Modern Scottish Gaelic', doctoral dissertation, Department of Linguistics, Stanford University. 1993b. 'Aspect phrase in Modern Scottish Gaelic', Proceedings of NELS 23: 415-29. Reichenbach, H. 1947. Elements of Symbolic Logic, New York: Macmillan. Reuland, E. and A. G. B. ter Meulen (eds.) 1987. The Representation of (In)definiteness, Cambridge, MA: MIT Press. Rhys, C. S. 1993. 'Functional projections and thematic role assignment in Chinese', doctoral dissertation, Centre for Cognitive Science, University of Edinburgh. Ritter, E. 1988. 'A head-movement approach to construct state noun-phrases', Linguistics 26: 909-29. 1991. 'Two functional categories in noun-phrases: evidence from Modern Hebrew', in P. Cole and J. Sadock (eds.) Syntax and Semantics, vol. XXV: Perspectives on PhraseStructure, New York: Academic Press, 37-62. 1993. 'Cross-linguistic evidence for number phrase', unpublished ms., University of Calgary. 1995. 'On the syntactic category of pronouns and agreement', Natural Language and Linguistic Theory 18: 405-443 Rivero, M.-L. 1991. 'Long head movement and negation: Serbo-Croatian vs Slovak and Czech', The Linguistic Review 8: 319-51. 1993. 'Finiteness and second position in long head movement languages: Breton and Slavic', unpublished paper, University of Ottawa. 1994. 'Clause structure and V-movement in the languages of the Balkans', Natural Language and Linguistic Theory 12: 63-120. Rizzi, L. 1986. 'Null objects in Italian and the theory of pro\ Linguistic Inquiry 17: 501-57. 1990. Relativized Minimality, Cambridge, MA: MIT Press. 1991a. 'Residual verb-second and the wh-criterion', unpublished ms., Universite de Geneve. 1991b. 'Argument/adjunct (a)symmetries', unpublished ms., Universite de Geneve/SISA, Trieste. 1993. 'Some notes on Romance cliticization', unpublished ms., Universite de Geneve. Rizzi, L. and I. G. Roberts 1989. 'Complex inversion in French', Probus 1: 1-30. Roberts, I. G. 1991a. 'Excorporation and minimality', Linguistic Inquiry 22: 209-18. 1991b. 'Head government and the local nature of head movement', GLOW

References

353

Newsletter 26: 48-9. 1994a. 'Two types of head movement in Romance', in D. Lightfoot and N. Hornstein (eds.) Verb Movement, Cambridge University Press, 207-42. 1994b. 'Agreement in Comp and the second position', unpublished ms., University of Wales, Bangor. 1994c. 'Agreement, clitics and case in Welsh', Research Papers in Welsh Syntax 5: 92-106, University of Wales, Bangor. Rognvaldsson, E. and H. Thrainsson 1990. 'On Icelandic word order once more', in J. Maling and A. Zaenen (eds.) Syntax and Semantics, vol. XXIV: Modern Icelandic Syntax, New York: Academic Press, 3-^0. Ross, J. R. 1969. 'Auxiliaries as main verbs', in W. Todd (ed.) Studies in Philosophical Linguistics, Series 1. Evanston: Great Expectations Press, 77-102. 1984. 'Inner islands', unpublished ms., MIT. Rouveret, A. 1990. 'X-bar theory, minimality and barrierhood in Welsh', in Hendrick (ed.), 27-79. 1991. 'Functional categories and agreement', The Linguistic Review 8: 353-87. 1994. Syntaxe du gallois: principes generaux et typologie, Paris: CNRS Editions. Ruwet, N. 1982. 'Les phrases copulatives', in N. Ruwet (ed.) Grammaire des insultes et autres etudes, Paris: Editions du Seuil. Sadler, L. 1988. Welsh Syntax: a Government-Binding Approach, London: Croom Helm. Safir, K. 1985. Syntactic Chains, Cambridge University Press. 1992. 'Thematic purity', paper presented to the Workshop on Specifiers, University of California, Santa Cruz, March. 1994. 'Perception, selection and structural economy', Natural Language Semantics 2:47-70. Santorini, B. 1989. 'The generalisation of the Verb-Second Constraint in the history of Yiddish', doctoral dissertation, University of Pennsylvania. Schafer, R. 1992. 'Negation and verb-second in Breton', Working Paper 92-02, Syntax Research Centre, University of California, Santa Cruz. 1994. 'Nonfinite predicate initial constructions in Modern Breton', doctoral dissertation, University of California, Santa Cruz. Schmidt, C. 1994. 'Satzstruktur und Verbbewegung: eine minimalistische Analyse zur interen syntax der IP Inflection-Phrase im deutschen', doctoral dissertation, Cologne University. Schmitt, C. 1992. 'SER and ESTAR: a matter of aspect', Proceedings of NELS 22. Schutze, C. 1993. 'Towards a minimalist account of quirky case and licensing in Icelandic', in C. Philips (ed.) Papers on Case and Agreement II, Cambridge, MA: MIT Working Papers in Linguistics, 321-75. Shlonsky, U. 1987. 'Null and displaced subjects', doctoral dissertation, MIT. 1992. 'Resumptive pronouns as a last resort', Linguistic Inquiry 23: 443-68. 1994a. 'Semitic clitics', Geneva Generative Papers 2.1: 1-11. 1994b. 'Agreement in Comp', The Linguistic Review. 11: 351-375 Forthcoming. Clause Structure and Word Order in Hebrew and Arabic, Oxford: Oxford University Press. Shoshani, R. 1980. 'The object marker in intransitive contexts', unpublished ms., University of Tel Aviv.

354

References

Sigurosson, H.A. 1991. 'Icelandic case-marked PRO and the licensing of lexical arguments', Natural Language and Linguistic Theory 9: 327-63. Siloni, T. 1991a. 'Hebrew noun phrases: generalized noun raising', unpublished ms., University of Geneva. 1991b. 'Noun raising and the structure of the noun phrase', MIT Working Papers in Linguistics 14. 1994. 'Noun phrases and nominalizations', doctoral dissertation, University of Geneva. Sjoestedt-Jonval, M. L. 1931. Description d'un parler irlandais de Kerry, Paris: Ernest Leroux. Smith, C. 1991. The Parameter of Aspect, Dordrecht: Kluwer. Sportiche, D. 1988. 'Conditions on silent categories', unpublished ms., UCLA. 1990. 'Movement, agreement and case', unpublished ms., UCLA. 1992. 'Clitic constructions', unpublished ms., UCLA. Sproat, R. 1982. 'Redundancy rules and Welsh mutation', unpublished ms., MIT. 1985. 'Welsh syntax and VSO structure', Natural Language and Linguistic Theory 3: 173— 216. 1986. 'More on Celtic consonant mutation: a case for a non-autosegmental analysis of CM in Welsh', unpublished ms., MIT. Sproat, R. and C. Shih 1991. 'The crosslinguistic distribution of adjective ordering restrictions', in C. Georgopoulos and R. Ishihara (eds.) Interdisciplinary Approaches to Language: Essays in Honor of S. Y. Kuroda, Dordrecht: Kluwer, 565-94. Stephens, J. 1982. 'Word order in Breton', doctoral dissertation, University of London. Stowell, T. 1981. 'Origins of phrase structure', doctoral dissertation, MIT. 1989. 'Raising in Irish and the Projection Principle', Natural Language and Linguistic Theory 7: 317-59. 1992. 'Aspects of tense theory', unpublished ms., presented at GLOW, Lisbon. 1993. 'Syntax of tense', unpublished ms., UCLA. Stump, G. T. 1984. 'Agreement vs. incorporation in Breton', Natural Language and Linguistic Theory 2: 289-348. 1989. 'Further remarks on Breton agreement', Natural Language and Linguistic Theory 7: .429-71. Szabolcsi, A. and F. Zwarts 1993. 'Weak islands and an algebraic semantics for scopetaking', Natural Language Semantics 1: 235-84. Takahashi, D. 1993. 'On antecedent-contained deletion', unpublished ms., University of Connecticut, Storrs. Tallerman, M. 1993a. 'Case assignment and the order of functional projections in Welsh', in A. Siewierska (ed.) EUROTYP Working Papers (European Science Foundation Programme in Language Typology, Group 2), vol. V, Strasburg: European Science Foundation, 1-41. 1993b. 'Fronting constructions in Welsh', in I. Tsimpli (ed.) Newcastle and Durham Working Papers in Linguistics 1: 251-75. Taraldsen. 1990. 'D-projections and N-projections in Norwegian', in M. Nespor and J. Mascaro (eds.) Grammar in Progress, Dordrecht: Foris, 419-31. Tenny, C. 1987. 'Grammaticalizing aspect and affectedness', doctoral dissertation, MIT.

References

355

Thorne, D. A. 1993. A Comprehensive Welsh Grammar. Gramadeg Cymraeg Cynhwysfawr, Oxford: Basil Blackwell. Timberlake, A. 1982. 'The impersonal passive in Lithuanian', in Monica Macaulay et al. (eds.) Proceedings of the Eighth Annual Meeting of the Berkeley Linguistics Society, Berkeley Linguistics Society, 508-24. Tomaselli, A. 1989. 'La sintassi del verbo finito nelle lingue germaniche', doctoral dissertation, University of Pavia. Torrego, E. 1988. 'A DP analysis of Spanish nominals', unpublished ms., Boston University. Travis, L. 1984. 'Parameters and effects of word order variation', doctoral dissertation, MIT. 1991. 'Parameters of phrase structure and verb-second phenomena', in R. Freidin (ed.) Principles and Parameters in Comparative Grammar, Cambridge, MA: MIT Press, 339-64. 1992. 'Notes on case and expletives: a discussion of Lasnik's paper', unpublished ms., McGill University, Montreal. Trepos, P. 1980. Grammaire Bretonne, Rennes: Ouest France. Ultan, R. 1978. 'Some general characteristics of interrogative systems', in J. Greenberg, C. Ferguson and E. Moravcsik (eds.) Universals of Human Language, vol. IV: Syntax, Stanford University Press, 211^8. Uriagereka, J. 1995. 'Aspects of the syntax of clitic placement in Western Romance', Linguistic Inquiry 26: 79-124. Valois, D. 1991. 'The internal syntax of DP and adjective placement in French and English', Proceedings of NELS 21: 367-82. Varlokosta, S. and N. Hornstein. 1993. 'A bound pronoun in Modern Greek', Natural Language and Linguistic Theory 11: 175-95. Vergnaud, J.-R. and M. L. Zubizarreta 1992. 'The definite determiner and the inalienable constructions in French and in English', Linguistic Inquiry 23: 595-652. Vikner, S. 1991. 'Verb movement and the licensing of NP-positions in the Germanic languages', doctoral dissertation, University of Geneva. 1994. Verb Movement and Expletive Subjects in the Germanic Languages, Oxford University Press. Vikner, S. and B. Schwartz forthcoming. 'The verb always leaves IP in V2 clauses', to appear in A. Belletti and L. Rizzi (eds.) Parameters and Functional Heads, New York/ Oxford: Oxford University Press. Wagner, H. 1959. Das Verbum in den Sprachen der britischen Inseln (Buchreihe der Zeitschrift fur Celtische Philologie, vol. 1), Tubingen: Niemeyer. Watanabe, A. 1993a. 'Agr based case theory and its interaction with the A-bar system', doctoral dissertation, MIT. 1993b. 'Case absorption', unpublished ms., University of Tokyo. Watkins, T. A. 1957. 'Yr arddodiad HG (h)i, in, CC y, yn\ Bulletin of the Board of Celtic Studies 17.3: 137-58. 1960. 'CC y/yn berfenwol', Bulletin of the Board of Celtic Studies 18.4: 362-72. 1991. 'The function of the cleft and non-cleft constituent orders in Modern Welsh', in Fife and Poppe (eds.), 329-51. Webelhuth, G. 1989. 'Syntactic saturation phenomena and the modern Germanic

356

References

languages', doctoral dissertation, University of Massachusetts, Amherst. Williams, E. 1980. 'Predication', Linguistic Inquiry 11: 203-38. 1981a. 'Argument structure and morphology', Linguistic Review 1: 81-114. 1981b. 'On the notions "lexically related" and "head of a word"', Linguistic Inquiry 12: 245-74. 1994. 'A reinterpretation of evidence for verb movement in French', in D. Lightfoot and N. Hornstein (eds.) Verb Movement, Cambridge University Press, 189-206. Williams, S. J. 1980. A Welsh Grammar, Cardiff: University of Wales Press. Wojcik, R. 1976. 'Verb-fronting and auxiliary do in Breton', Proceedings of NELS 6: 25978. Zagona, K. 1990. 'Times as temporal argument structure', unpublished ms., read at the conference 'Time in Language', Massachusetts Institute of Technology, March. Zanuttini, R. 1991. 'Syntactic properties of sentential negation: a comparative study of Romance languages', doctoral dissertation, University of Pennsylvania. Zimmer, H. 1901. 'Grammatische Beitrage 2: uber verbale Neubildungen im Neuirischen', Zeitschrift fur Celtische Philologie 3: 61-98. Zwart, J.-W. 1992. 'Verb movement and complementizer agreement', GLOW Newsletter 26: 58-9. 1993. 'Verb movement and complementizer agreement', in J. Bobaljik and C. Phillips (eds.) Papers on Case and Agreement, I {MIT Working Papers in Linguistics 18), 296341. Zwarts, J. 1991. 'The RP-hypothesis', in J. van Lit, R. Mulder and R. Sybesma (eds.) Proceedings of the Leiden Conference for Junior Linguistics, University of Leiden, 259-73. 1992. 'X'-syntax-X'-semantics: on the interpretation of functional and lexical heads', doctoral dissertation, Research Institute for Language and Speech, University of Utrecht. Zwicky, A. 1984. 'Welsh soft mutation and the case of object NPs', CLS 20: 387^02. Zwicky, A. and G. Pullum 1983. 'Cliticization vs. inflection: English n't', Language 59: 502-13.

Index

a bhith, 215-16 A'-binding constructions, 47-9, 297 Abney, S.P., 12-13, 93n, 334 abnormal sentences: adjunction to matrix CPs, 110-11, 119-21; agreement patterns in, 115-17; and cleft sentences, 27-8, 97-102; derivation proposed, 110-11; negation of, 99, 100, 118-21; properties, 98-9 Accusative Case, realization of, 224 Acquaviva, Paolo, 46, 284—313 Adger, David, 17, 31, 43, 200-22, 236, 237, 238n, 340n adjectives, 38-9, 77-8, 79 adjunction, 65, 66-7; to CP, 111-15 Adjunction Prohibition, 110, 111-13, 118 adverbials, 269-70, 306-7 adverbs, and attributive adjectives, 30 affective operators, licensing polarity items, 286, 299 Afro-Asiatic languages, 199n agreement: analytic and synthetic and the position of the subject, 82, 190-6; 'antiagreement' effect, 154; clitics and, 16-17; clitics and null arguments, 40-3; Scottish Gaelic, 200-22; and Tense, 9-10, 23, 200 Anderson, S., 54, 73n, 184 Andrews, A., 207 Anglo-Saxon, 3 annamh, 290, 302-3, 312-13n; cleft sentences and, 290, 306-11 Anwyl, E., 131, 165n, 166n Aoun, J., 200 appartenance, and possession, 140 Arabic, 171, 172, 185, 186-7, 191-2, 198n, 238n; agreement, 192; Cairene, 176-7, 179, 180; CSNs and FGs, 316, 317;

357

Palestinian, 175, 176, 178, 185, 187; Syrian, 39-40, 192 Aronoff, M., 82, 92 article: definite, 38; indefinite in Breton, 38; postnominal, 14 aspect, Scottish Gaelic, 200-22 aspectual chains, 211-17 aspectual periphrases, 131-2, 168n aspectual predicates, focalization of, 134-6 aspectual properties, different of states and events, 159-61 Authier, J. Marc, 105, 106 Aux node, 5-6, 8 auxiliaries, 9, 54, 63, 71-2 Avoid Resumptive Pronoun (ARP), 84-5 Awbery, Gwen, 19-20, 33, 46, 90, 92n, 131, 139, 161n, 163n, 165n, 166n, 167n, 182, 193, 195, 273 Bach, E,, 210 Baker, M., 6, 15, 134, 174, 231, 278n, 280n, 28In Baltin, M., 106 Barbosa, Pilar, 239n Barwise, J., 46, 296, 299, 300, 301, 312n be: equative use, 125; identificational use, 125; as a lexically unaccusative verb selecting a small clause, 126; partitive Case feature of, 144; predicational use, 125-6; small-clause analysis, 138, 144; specificational use, 125-6 Belletti, Adriana, 9, 144, 145, 197n, 28In, 292, 293 Bennis, H., 209, 263 Benveniste, E., 140, 162n Berber clitics, 198n Berman, R., 336n

358

Index

Besten, H. den, 6, 11, 226 bezan, 85-7 Bhaldraithe, T. de, 337n binding indexing, 212, 219 Bobaljik, Jonathan David, 20, 23, 31, 33, 108, 164n, 174, 188, 221n, 233-240, 242, 278n, 314, 336n bod, 28, 29, 181, 197n; -initial constructions, 138-9; copular, 126-8, 159, 161; existential, 126-8, 159, 161; and locative clitics, 127, 145-6, 161; and predicative particle yn, 76-8; in the present tense, 161; small-clause analysis, 138, 144, 164-5n; suppletion in, 79-85; unitary characterization of, 125-70 Bodhlaeir, Liam, 278n Borer, H., 94n, 252, 273, 316, 326, 327, 335n, 336n, 338n, 339n Borsley, Robert D., 1-52, 53-74, 87, 92n, 93n, 94n, 117, 122n, 131, 161n, 162n, 163n, 164n, 183, 192, 196n, 220n, 221n, 237n, 312n, 335n, 337n, 339n Bourke, Angela, 278n Brandi, L., 16 Breatnach, Liam, 278n Breton, 2, 184, 195, 273; A'-binding construction, 48-9; agreement and clitics, 40-1; auxiliaries, 53-4, 64, 69-72; clefting, 27-9; copula, 45-6, 75, 79; description, 3; indefinite article, 38; interrogative complementizer, 28; long head movement in, 53-74; mutation, 49; negation, 46, 67-8; nominals, 38^40; non-finite clauses, 29-34; ober sentences, 68-71; participial constructions, 35-8; particles, 26, 27; similarities with Slavonic and Romance languages, 7; speakers, 3; topicalization, 27, 54-6, 60, 68, 104, 117; VAuxSO clauses, 23-5, 53, 54-7; verb-fronting, 11, 55-6, 58-61; and the VP-internal subject hypothesis, 22-3, 56-7; VSO clauses, 19-23 bridge verbs, 106, 121 British, see Brythonic Brittonic, see Brythonic

Brody, M., 152 Brythonic, 2, 3, 76, 182 Bulgarian, 24; long head movement in, 53, 57, 58-9, 60, 66; renarrated mood, 64-5 Bures, Tony, 237n Burzio, Luigi, 245, 250, 253, 260, 273, 279n, 280n

Campbell, R., 159, 160, 170n Cann, Ronnie, 214, 220n Cardinaletti, A., 11, 97, 106, 109, 188, 194, 196n, 197n Carlson, G., 88, 89, 132 Carnie, Andrew, 20, 23, 31, 33, 93n, 94n, 108, 164n, 174, 188, 221n, 223-40, 242, 273, 278n, 314, 336n Case: assignment, see Checking Theory; and definiteness, 333-4; of indefinites, 144-6; realization of, 223-4 Case Filter, 144 Case Theory, 78, 79, 89; link with Theta Theory, 200-3 Cavar, D., 65 Celtiberian, 2 Celtic languages; copulas, 75-96; description, 2; issues in syntax, 19-46; possessives, 14, 39, 314-40; similarities with Semitic languages, 7, 314-40 CHAINS, minimal requirement, 260 Chamorro, 273 Checking Theory, 6-7, 171, 172-4, 178-9, 180, 186-9, 205; see also features, theory of syntactic Chomsky, Noam, 9-10, 21, 56, 93n, 94n, 95n, 162n, 164n, 169n, 212, 237, 249, 250, 260, 272, 278n, 281n, 292, 336n, 337n; on adjunction, 110; Checking Theory, 6-7, 171, 172-4, 178-9, 180, 186-9, 205; derivation of VSO order, 225, 228-9; Economy principle, 231; English auxiliary raising, 198n; Extended Projection Principle, 242, 265, 282n; and Head Movement Constraint, 19, 6 3 ^ ; on Minimalist Programme, 137-8, 151, 314, 316; on pro, 325; Standard Theory,

Index 92; theory of syntactic features, 6-7, 152, 172-4, 190, 223, 233-5, 239n, 267-9, 270; Universal Grammar, 4, 196; Visibility Condition, 200-2 Christian Brothers, 333, 334, 335n Chung, S., 26, 31, 54, 238n, 242, 245, 263, 264, 273, 278n, 282n Cinque, Guglielmo, 196n, 209, 319, 320, 335n, 337n Clack, S., 187 clausal complements, and the internal subject position, 262-5 clause structure, and functional categories, 7-12 cleft constructions: embedded, 101-2, 105-6, 109, 111-15, 155; matrix, 109, 114, 117, 118-21 cleft sentences: and abnormal sentences, 27-8, 97-102; agreement patterns in, 115-17; and annamh, 290, 306-11; copula or complementizer in, 117-18; and CP-recursion analysis, 97, 106, 108-9, 110-21; derivation proposed, 107-10; generative approaches, 102-5; Government-Binding analysis of, 103-4; with monotone decreasing quantifiers, 308-11; negation of, 98, 100, 118-21; properties, 98 clefting, 27-9, 97-102 clitic doubling, 16-17, 172, 185 clitic systems, and word order, 171-99 clitics, 15-19; agreement and null arguments, 40-3; base-generation of, 18, 171-99; derivation of, 17-19; placement, 15-16, 17-18, 171, 185; pronominal, 15-19; relationship with agreement, 16-17, 40-3; syntactic, 15-19; Wackernagel, 188 co-indexation, 200, 202, 211; of denotational arguments of auxiliaries and main verbs, 213, 215-16, 217 Cois Fhairrge dialect of Irish, 337n 'Common in Form, Genitive in Function' contexts, 334

359 complementizers, or verbal particles, 25, 26-7, 104-5 Comrie, B., 191 Connacht dialect of Irish, 228 Connemara Irish, 31 In Construct State construction, Semitic, 14, 314, 339n Construct State Nominal (CSN), 314, 317-35; adjective placement in, 318-22; possessive clitics (POSS) in, 322-8; rightward specifiers, 317-18 Continental Celtic, 2; see also Celtiberian; Gaulish; Lepontic contra-indexation, 213, 216 Cooper, R., 46, 296, 299, 300, 301, 312n copular sentences: embedded, 155-6; functional structure, 150-2; predicateinitial, 148-53; subject-initial, 153-5; syntax of agreement in, 153-4; as verbsecond clauses, 148-50, 156 copulas, 43-6; Breton, 45-6, 75, 79, 87; English, 125-6; functional, 76, 79; morphosyntactic variation in, 75-92; nature of the Irish, 43-^, 310-11; substantive, 76-9, 87; syntactic effects of suppletion in Celtic, 75-96; Welsh, 44-5, 76-9, 125-70 Cordin, P., 16 Cornish, 2, 3 CP-recursion, 106, 108-10, 117-18, 121, 155, 166n, 167n Cram, D., 203, 215 Crisma, P., 319, 320 Czech: auxiliaries, 71; long head movement in, 24, 53, 57, 59, 60

D-features, 328-34 Davies, Lewis, 122n Dechaine, Rose-Marie, 132, 142, 160, 161n, 170n definite subjects, distribution of, 136-43 definiteness; and Case-checking, 333-4; and demonstrative interpretation, 328-34; effect, 143-6, 261, 281n Delsing, L.-O., 14

360

Index

demonstratives: licensing and interpretation, 330-3; placement 339n denotational argument, 211, 217, 218 derivations, Chomsky's theory of, 137-8, 151 Determiner Phrase (DP), 12-13, 39-40; Event-Related and Object-Related, 221n determiners: left downward-entailing, 299-300; Minimality barrier on quantificational, 210, 219; right downward-entailing, 300 Di Sciullo, A., 92 Diesing, M., 11, 88, 90, 143, 163n Dikken, Marcel den, 16In, 164-5n Dillon, M., 280n Dimitrova-Vulchanova, Mila, 73n Dobrovie-Sorin, C , 65, 209 Doherty, C , 20, 43-4, 76, 89, 93n, 225, 310 domain, 173-4, 180; checking, 63, 190; functional, 8; internal, 63; minimal, 173-4 Dooley Collberg, Sheila, 326, 337n Doron, E., 185 Drijkoningen, F., 211, 212 Duffield,Nigel, 7, 14, 17, 23, 26, 40, 43, 194, 229, 236, 238n, 240n, 242, 245, 247, 271, 282n, 293, 303, 31 In, 314-40 Dunigan, Melynda, 92n Dutch, 55, 56, 263

echo pronouns, Welsh, 172, 182-4, 189, 193-6, 337n eclipsis, 47 Economy Principle, 173, 231, 333 e/-forms, 183-4, 185, 186, 1 9 3 ^ Elsewhere Principle, 94n Emonds, Joe, 4-5, 8, 20, 92n, 164n, 174, 336n Empty Category Principle (ECP), 48, 64, 67-8, 82, 84, 92, 109, 291 En?, M., 160, 170n enclisis, 171-99, 322 endocentric categories, 13 Engdahl, Elisabet, 220n

English, 70, 174, 294, 295; definite predicates, 150-1; expletives in small clauses, 265; negation compared with Irish, 284-313; negative markers and negated quantifiers, 285-7; Nominative Case, 223; northern dialect around Edinburgh, 3; position of inflected main verb in finite clauses, 4-7; POSS-ing gerundives, 12-13; stative verbs, 133, 159-60; strong and weak features compared with French, 233-4, 235; tense/aspect system, 211, 212; topicalization, 105-6 equidistance, 137, 164n, 180; definition of, 173^ Ernst, T. 269, 333, 334, 335n ethical dative, 18 Evans, D.S., 100, 116, 119, 123n Evans, J.G., 122n Everett, D., 174 exceptional Case-marking clauses, in Irish, 32 excorporation, 66-7, 194 existence, and predication, 128-46 'existential' Case, 145 existential closure, 143-4, 298 existential construction, in Irish, 264—5 exocentric categories, 13 expletive-argument CHAINS, 259, 260, 261

expletives: absence in Irish, 261, 264-5, 270-1; and the Extended Projection Principle, 259-62, 267-8; position of, 270-1 Extended Projection Principle (EPP), 195, 222n, 234, 242, 250, 262, 265, 282n; and expletives, 259-62, 267-8

F-licensing, 202, 220n Fabb, N., 323

Fabri, R., 321, 326, 335n, 336n, 338n, 339n Fassi-Fehri, A., 185 Fauconnier, G., 165n

Index features, theory of syntactic, 223, 233-5, 239n, 267-9, 315-16, 328-34; see also Checking Theory Feirtear, Breandan, 278n Fife, J., 97, 98, 99, 100, 101, 131, 160, 162n, 163n Fillmore, C , 11 finite verbs, position in French compared with English, 4-7 focus interpretation, of initial predicate in copular constructions, 152-3 formatives, lexical and functional, 75 Fox, Danny, 237n Frampton, John, 209, 278n, 282n Free Genitive (FG) constructions, 316, 317-18, 321; possessive clitics (POSS) in, 326-8 Freeze, R., 162n Frege, Gottlob, 125 French, 20, 21; clitics, 175; definite predicates, 150; eventive verbs, 159; infinitives, 8-9; negation, 46, 292, 294, 295, 299, 303, 304; object clitics, 17-18; position of inflected main verb in finite clauses, 4-7, 174; strong and weak features, 174, 233-4, 235; subject-clitic inversion, 198n; unaccusatives, 23, 254, 280n; y, 134; see also Old French Fronted Object Phrases (FOP), 203-10 fronting: constructions, see abnormal sentences; cleft constructions; in embedded clauses, 101-2; negation in, 118-21; in Welsh, 97-124 Fukui, N., 11, 12, 223 Full Interpretation, Principle of, 6 functional categories, 7-12, 314 functional heads, relationship with lexical categories, 172-4, 265-73

Gaelic, see Goidelic; Scots Gaelic Gallagher, Grainne, 335n Gaulish, 2 Gazdar, G., 73n Generalized Phrase Structure Grammar (GPSG), 73n

361 generalized transformation (GT), 190, 212 Generalized Visibility, 200-3, 218, 220; defined, 203 generative grammar, 97; cleft constructions, 102-5 Genitive Case, 334 German, 10, 20, 280n; adjective placement, 320 Germanic languages, 28In; clitics, 186, 188; verb-second phenomena, 10, 121, 149, 226, 337n Giorgi, A., 211, 212-13, 215-16, 319 Glinert, L. 316, 335n, 338n Goidelic, 2, 182 Government Transparency Corollary (Baker), 231 Government-Binding: analysis of clefts, 103^4; analysis of English topicalization, 105-6; Case Theory, see Checking Theory GREED principle, Minimalist Programme, 316, 318, 328 Greek, 2 Greenberg, J., 27 Greene, D., 280n Grimshaw, J., 94n, 202, 211 Groat, E., 271 Gueron, J., 142, 166n, 209 Guilfoyle, Eithne, 20, 21, 236, 237n, 239n, 242, 245, 247, 279n, 317, 328, 335n

Haegeman, L., 1, 4, 226, 296, 312n Hale, Ken, 42, 185, 221n, 236, 237n, 243, 272, 325, 326, 328, 337n Halle, M., 92 Harley, Heidi, 94n, 225, 237n, 238n, 239n, 273 Harlow, Steve J., 20, 26, 50-1, 92n, 93n, 102, 103, 104, 122n, 139 Hartmann, H., 280n Hazout, I., 176 head movement, 14-15; types of, 65-6; see also long head movement Head Movement Constraint (HMC), 6, 15, 19, 57, 63-4, 178

362

Index

head-to-head movement, 14-15; see also incorporation heads, L-related and non-L-related, 53, 64-6, 67-8 Hebrew, 40, 43, 171, 175, 185, 187, 198n, 273; Colloquial, 191; CSNs and FGs, 316, 317; noun phrases compared with Irish and Maltese, 7, 314-40 Heim, I., 143, 296, 297, 298 Hendrick, Randall, 1, 23, 26, 33, 36, 48, 56, 75-96, 102, 103, 104, 105, 123n, 131, 161n, 163n, 168n, 195, 197n, 220n Hewitt, S., 69 Heycock, C , 162n Higginbotham, J., 202, 211, 297, 298, 312n Higgins, F.R., 125 Highest Subject Restriction, 252 Hoeksema, J., 299 Hoekstra, T., 150, 164n, 209, 221n Holmberg, A., 51n, 231 Honcoop, M., 22In Hornstein, N., 211, 252, 283n Hoyt, K., 176

Icelandic, 11, 21, 278n identity, and predication, 125-6, 146-59 inalienability condition, 316, 336n incorporated pronouns, 184-5 incorporation, 174, 328; see also head-tohead movement indefinite subjects, distribution of, 136-43 indefinites: the Case of, 144-6; and monotone decreasing quantifiers, 301, 311; negation and, 296-8 indexation, selection and lexical specification, 211-12 Indo-European, 2, 182, 199n infinitives, short movement of main-verb, 9 inflection, and negation, 2 9 3 ^ Insular Celtic 2; see also Brythonic; Goidelic inversion constructions, 10-11 Irish, 185, 195; A'-binding construction, 47-8, 49; clefting, 27-9; clitics, 41, 43; copulas, 4 3 ^ , 75, 76; CSNs in, 316-17;

description, 2-3; minimalist approach to problems of word order, 223^0; mutation, 49; negation in, 46, 284-313; negatives, 287-90; nominals, 38-40; nonfinite clauses, 29-34, 235-7, 238n, 245, 246-7, 251; northern dialects, 256; noun phrases, 314-40; participial constructions, 35-8; particles, 26; similarities with Hebrew and Maltese, 7, 314-40; small clauses, 245, 246-7; southern dialects, 256, 278n; speakers, 3; subjects and subject positions, 241-83; verb raising in, 225-37; and the VPinternal subject hypothesis, 21, 22, 23; VSO in finite clauses, 19-23, 225-9, 230-3, 238n, 245; see also Cois Fhairrge; Connacht; Connemara; Munster; Old Irish; Ulster isomorphy, parametric, 196 Italian, 46, 134, 254, 280n, 292, 295, 299, 303, 304; adjective placement, 320; clitics, 175, 187-8; Fiorentino, 16; northern dialects 16, 187-8, 198n; Padovano, 187; unaccusative constructions, 144; Veneto, 187 Italic languages, 2 Jaeggli, O., 4, 16, 17, 323, 326 Jespersen, O., 125, 126 Johnson, K., 281n, 336n Jonas, Dianne, 237n Jones, Bob Morris, 20, 95n, 102, 130, 131, 147, 161n, 167n Joosten, Laura, 220n Kamp, H., 296 Kayne, R.S., 10, 15, 17-18, 52n, 72, 95n, 132, 171, 185, 187, 188, 189, 282n, 291, 314, 336n Keenan, E., 299, 312n Kenstowicz, M., 176 King, G., 97, 98, 100, 101 Kitigawa, Y., 223 Klima, E., tests for negatives, 285, 286, 287, 290, 291, 299, 307

Index Koopman, H., 11, 21, 56, 89, 108, 188, 194, 223, 242, 315 Koster, J., 55, 226 Kratzer, A. 43, 133, 162n, 210 Krifka, M. 22In Kuroda, S.-Y., 11 Ladusaw, W., 299 Laka, M.I., 292, 312n Larson, R., 197n Larsonian VP-structure, 126, 134^5, 137, 144, 164n Lasnik, H., 64, 106, 144, 145, 164n, 223, 229, 282n, 283n Latin, 2; loan words in Welsh, 3 Lema, J., 53, 57 lenition, 47, 77, 31 In Lepontic, 2 Lewis, H., 117, 181, 182, 187 lexical diversity, structural invariance in VSO languages and, 314-40 lexical indexing, 217 Lexical Predicative Domain (LPD), 142-4 lexical specification, selection and indexation, 211-12 LHM, see long head movement Lobel, E., 319, 335n Locative Preposing, 137 locative sentences, and existential sentences, 130 Logical Form (LF), 126, 172, 223, 233, 260, 290-1 long head movement, 7, 24, 57-61, 323; in Breton, 53-74; definition, 53; and licensing of Tense, 24, 61-3, 72; nature of, 65-7; and negation, 59, 67-8; properties of, 58-61; restrictions on, 67-8 Longobardi, Giuseppe, 14, 126, 280n, 319 Lyons, C , 322 Lyons, J., 162n Mac Con Iomaire, Liam, 278n MacAulay, D., 2 McCarthy, J.J. 94n

363 McCloskey, James, 20, 21, 22, 23, 26, 31, 32, 37, 42, 47-8, 56, 93n, 96n, 105, 106, 107, 108, 110, 121, 122n, 123n, 185, 188, 195, 221n, 226, 227, 229, 236, 237n, 238n, 239n, 240n, 241-83, 288, 293, 31 In, 325, 326, 334, 336n, 337n McDaniel, D., 297 McKenna, Malachy, 278n mae, 126-7: as an 'auto-saturated' form, 157; analysis of constructions with, 134-8; distribution of, 87-9, 95n; existential, 129-31; predicational, 128-9; and stage-level predicates, 75, 88-9, 131-4; suppletive form, 80, 87-9 mae/yw distinction, 44-5, 156-9, 161, 169n Mahajan, A., 202 mai, 28, 29, 102, 103, 105, 113-15, 117-18, 155-6 Maltese, 40, 43; inalienability condition, 316, 336n; noun phrases compared with Irish and Hebrew, 7, 31440 Manx, 2 Manzini, M.R., 65, 209 Marantz, Alec, 92, 93n, 223, 237n, 282n May, R., 296 measure phrases, 200, 201; and Fronted Object Phrases, 208-10; licensing Scottish Gaelic, 203-11, 217-19 Mellor, Martin, 220n Middle Welsh, 28, 98, 99, 100, 101, 102, 115, 116, 117-18, 187 Milsark, G., 210 Minimalist Programme: assumptions of, 314-17, 333; GREED principle, 316, 318, 328; Irish word order, 223-40 Minimality Condition, 137-8, 151, 164n, 210, 219, 231; see also Relativized Minimality mixed sentences, see cleft sentences Mohammad, M., 317, 335n monotone decreasing quantifiers, 46, 299-306,308-11 monotone increasing quantifiers, 300-1, 305

364

Index

monotonicity correspondence universal, 301 Moritz, L., 292, 293, 312n Moro, A., 126, 161n, 162n, 166n, 310 morphology, and syntax, 75-96, 134, 152 morphosyntactic structure, and Reichenbachian temporal reference points, 211-17 Morris-Jones, J., 181, 182, 183 Motivated Chomage Law, 191 Mouchaweh, L., 176 Mulder, R., 150, 164n Munster dialect of Irish, 228, 229, 235, 256, 257 mutation, 47, 49-51, 77

N-features, as 'argument-checking features', 174, 233-5, 239n, 270-1 Nash, Lea, 161n Natural Language and Linguistic Theory, 1-2 negation: evidence for LF movement, 290-1; in fronting constructions, 118-21; and inflection, 293-4; in Irish, 284-313; in non-finite clauses, 32-3; the realization of in Celtic languages, 46; sentential, see sentential negation; the syntactic representation of, 290-8 Negative Criterion, 296 negative markers, 295; in Irish, 285-7, 304-5 negatives, Irish, 287-90 NegP hypothesis, 291-5, 298-306, 311 Ni Chiobhain, Marian, 278n Ni Chiosain, Maire, 237n, 31 In, 335n Ni Shuilleabhain, Brenda, 278n Nishigauchi, T., 297 nominals, 38-40, 314-40; relationship between Celtic and Semitic, 14, 3 9 ^ 0 ; the structure of, 12-14 Nominative Case, 223-4, 266-7 non-finite clauses, word order, 29-34, 235-7 non-finite verbs, see verb-nouns Noonan, Maire, 229, 237n

noun phrases, comparison of Irish, Hebrew and Maltese, 39-40, 314^*0 null argument languages, agreement and clitics, 40-3 numerals, precede noun, 38-9 numeration (Chomsky), 272

6 O O 6 6 6 6 O 6 6

Baoill, Donall P., 237n, 335n Buachalla, Breandan, 278n Dochartaigh, Cathair, 5In Huallachain, C , 334, 335n Muimhneachain, Aindrias, 278n Murchu, Liam, 278n Murchu, M., 334, 335n 'Rahilly, T., 280n Se, Diarmaid, 278n, 280n Siadhail, Micheal, 240n, 263, 278n, 28In, 307, 335n 6 Suilleabhain, Diarmuidin, 278n Object Shift, 188, 273, 278n, 283n Ogam script, 2 Old French, 166n Old Irish, 2, 225, 238n, 273 Old Spanish, 24, 53, 58 Old Welsh, 3, 102 optionality, in Minimalist perspective, 86, 333 Ouhalla, J., 49, 166n, 180, 189, 198n, 238n, 240n, 282n, 31 In

P-Celtic, 2 parameters of variation, see principles-andparameters theory Partee, B., 166n, 210 participial constructions, 35-8 particles: aspectual in Scottish Gaelic, 214-16; clause-initial, 25-7, 47; clefting in Scottish Gaelic, 207-8; negative root and subordinate in Welsh, 26; status as complementizers, 25, 26-7, 104-5 partitive Case, 144, 145 passive constructions, 37-8; see also perfective passive; progressive passive Pedersen, H., 117, 181, 182, 187

Index Pembrokeshire Welsh, 46, 195 perfective constructions, 36-7 perfective passive, 37-8, 254-9, 261, 262, 268, 280n Perlmutter, D., 191, 245 Pesetsky, David, 75, 78, 82, 86, 93n, 94n, 237n, 253 phi-features, 188, 205, 217-18 Phillips, Colin, 237n Phonetic Form (PF), 172 Pianesi, F., 211,212-13, 215-16 Pictish, 3 Platense Spanish, 16 Platzack, C , 51n pleonastic elements, 250, 254, 259, 260; in small clauses, 280-1 n Poletto, C , 187-8 Pollock, J.-Y., 4, 7, 56, 67, 174, 198n, 223, 233, 234, 241-2, 292 Pollock-Emonds tests, 5In Poppe, E., 97, 99 Portuguese, long head movement in, 53, 58 POSS-mg gerundives, English, 12-13 possession, 39; and appartenance, 140; with prepositional phrase, see Free Genitive (FG) possessive absolute constructions, in Welsh, 139^3, 145 possessive clitics (POSS), in Construct State Nominal (CSN), 322-8 possessor construction, see Construct State Nominal (CSN) Postal, P., 191, 245 predication: and existence, 128-46; and identity, 125-6, 146-59; licensing the relation, 150-2 preposition-stranding, 250, 260, 264 prepositions, inflected, 41-2 present tense: GENERIC/HABITUAL and FUTURE operators, 160-1; status in Welsh, 158, 159-61 Press, I., 87, 95n Prince, A.S., 94n principles-and-parameters theory, 1, 4-7, 97, 196, 223, 290, 314

365 pro, 16-17, 188-9, 205-7, 22In /?ro-drop effect, 325 proclitics, 42-3, 49, 175-6, 322 Procrastinate Principle 7, 172-3, 233, 333, 336n progressive constructions, 35-6, 37, 247-9, 252 progressive passive, 37, 247-8 pronominal enclisis, in VSO languages, 171-99 pronoun postposing, right-adjunction, 263 pronouns: atonic, 188; checking and VSO, 186-9; phi-features, 188; weak as a means of identifying pro, 189, 196 propositional domains, as predicates, 142, 152 Pullum, G., 17 Pyatt, Elizabeth, 225, 237n, 238n, 273

Q-Celtic, 2 quantifier raising, 291; NegP and, 294-5 quantifiers, negative, 284, 285-7, 296-8; see also monotone decreasing quantifiers; monotone increasing quantifiers Quasi-Arguments, 201

Ramchand, G., 220n, 236, 238n, 239n, 340n re-adjustment rules, 92 reconstruction processes, 149 Recoverability of Deletion Principle, 82-5 Reichenbach, H., 212, 214 relational grammar, 191 relative clauses, 47-9 Relativized Minimality, 64-5, 109, 186, 221n, 292-3, 294, 328 resumptive pronouns, 47, 48, 84-5, 252 Reuland, E., 261 Rhys, Catrin Sian, 214, 220n Right-Hand Head Rule, 171 Ritter, E., 14, 186, 316, 317, 319, 321-2, 326, 327, 328, 333, 334, 335n, 338n, 339n

366

Index

Rivero, Maria-Luisa, 7, 15, 23, 28, 53-74, 312n, 337n Rizzi, Luigi, 6, 16, 64, 65, 73n, 97, 106, 167n, 174, 197n, 209, 231, 266, 286, 293,

296, 297, 312n Roberts, B., 122n Roberts, Ian, 1-52, 53, 64, 65-7, 97, 106, 109, 122n, 123n, 161n, 162n, 164n, 165n, 167n, 171-99, 220n, 221n, 237n, 266, 278n, 281n, 328, 337n Rognvaldsson, E., 11 Romance languages, 7, 16, 19, 43, 53, 187; auxiliaries, 71, 72; clitics, 67, 186, 188, compared with Semitic clitics, 175-81; negative formatives, 295; participial structures, 132 Ross, J.R., 11, 286 Rouveret, Alain, 46, 73n, 75, 93n, 94n, 95n, 96n, 106, 108, 196n; on bod, 125-70, 197n; cleft sentences, 27, 28-9, 102, 103, 105, 109, 114-15, 116-17, 121, 122-3n; on mae/yw, 44-5, 156-9, 161; on particles, 26, 36; on subject positions, 21, 22, 23, 56, 88, 90 Rumanian, long head movement in, 53, 58 Runner, Jeff, 220n Rutten, J., 211, 212 Ruwet, N., 125

Sadler, L., 26, 34, 48, 93n, 102, 103, 104 Safir, K., 4, 17, 261, 265 Saito, M., 106 Santorini, B., 11 Scandinavian languages, 14, 5In, 198n; object shift, 188 Schafer, R., 54, 55, 73n, 273 Schmidt, C , 336n Schmitt, C , 93n Schiitze, C , 282n Schwartz, B., 11 Scots Gaelic, 2, 31, 195, 203, 236, 340n; aspect, agreement and measure phrases in, 200-22; clitics, 43; compound tenses, 214-15; description, 3; speakers, 3; tense and aspect in, 214-16

selection indexing, 211-12, 219 selection theory, 200, 202, 211-20 Sells, P., 31, 245, 247, 283n Semitic languages, 39-40, 43; constructstate construction 14, 314, 316, 339n; similarities with Celtic languages, 7, 314^40 sentential negation, 284—313 Serbo-Croatian: auxiliaries, 71; long head movement in, 24, 53, 57 Shih, C , 320 Shlonsky, Ur, 7, 14, 16, 17, 40, 84, 162n, 165n, 171-99, 220n, 22In, 236, 280n, 328, 337n Shoshani, R., 191 SigurSsson, H.A., 242 Siloni, T., 14, 186 Sjoestedt-Jonval, M.L., 280n Slavonic languages, 7, 53; auxiliaries, 71; clitics, 63, 186, 188 Slovak, long head movement in, 53, 57 small clauses, pleonastic elements in, 280-ln small-clause analysis of be/bod constructions 138, 144, 164-5n Smith, C , 132 Spanish: Latin American dialects, 16; a null subject language, 16-17; see also Old Spanish; Platense Spanish Speas, M., 11, 12, 223 Specified Subject Condition (SSC), 18-19 Spell Out, 172, 223, 233 split-Infl hypothesis, 8-9, 22 Sportiche, D. 11, 18, 19, 21, 56, 89, 108, 197n, 198n, 223, 242, 315 Sproat, R., 77, 78, 93n, 102, 103, 131, 132, 183, 203, 320 stage-level/individual-level predicates, 44-5, 75, 88-9, 127, 131-4, 161 Starke, M., 188, 196n, 197n Stavi, J., 299 Stephens, Janig, 7, 15, 23, 26, 27, 28, 51n, 53-74, 117, 312n, 337n Stowell, T., 20, 31, 160, 211, 212, 213, 225, 247, 279n, 293

Index Structure Preservation Condition, 9 Stump, G.T., 26, 27, 56, 104, 163n Subjacency, 47 subjecthood, morphosyntactic features of, 267-9 subjects: internal position, and EPP, 265-7; internal position and clausal complements, 262-5; and subject positions in Irish, 241-83 subsumption, 125 superindexation, 212 suppletion: syntactic effects in Celtic copulas, 75-96; third-person present form, 79-87 SVO languages, 20, 187 sydd 45, 80-5, 83-4, 89-92, 95n, 167-8n syntax, and morphology, 75-96, 134, 152 Szabolcsi, A., 209

tags, distribution of and negation in Irish, 285-6, 307-8, 310 Takahashi, D., 283n Tallerman, Maggie 23, 27-8, 51, 93n, 97-124, 155 Taraldsen, K.T. 14 taw, 28, 102, 117, 166n Telegraph, Principle of, 45, 82, 84, 86, 88 Tenny, C , 78, 89 Tense: and Agreement, 9-10, 23, 24, 200; and aspect, in Scottish Gaelic, 200, 214-16; morphosyntactic and interpretation, 142, 212-14 Tense chains (T-chains): and aspectual chains, 210-17, 220; measure phrases and, 209-10 tense/aspect system of English, temporal reference points (Reichenbach), 211, 212 ter Meulen, A.G.B., 261 Theta Theory, link with Case Theory, 200-3 theta-arguments, 211, 217 Thomas, A.R., 20, 95n, 102, 130, 131 Thorne, D.A., 94n Thrainsson, H., 11 Three-Layered Case Theory, 282n

367 Tomaselli, A., 10 topicalization, 98, 227; in Breton, 27, 54-6, 60, 68, 104, 117; GB analyses of English, 105-6; remnant, 55, 56 Torrego, E., 188, 194 traces, 47 Travis, L., 6, 15, 52n, 126, 134, 144, 163n, 224, 227, 250 Trepos, P., 94n Tsoulas, Georges, 161n Tuller, Laurie, 161n Turkish, 13

Ulster dialects of Irish, 3, 203, 228, 240n, 280n Ultan, R., 307 unaccusatives, 23, 126; in Irish, 242-50, 259-61; putative, 251-4, 257-9, 268, 279-80n; salient, 242-50, 257, 259-61, 271, 272, 273, 280n; salient, listed, 276-8 Universal Grammar (UG), 4, 82 Uriagereka, J., 188, 194

V-features, as 'predicate-checking' features 6-7, 174, 233-5, 239n Valois, D., 312n, 319 variation: morphosyntactic origins, 273; parameters of, 4, 196, 314; synchronic, 7; theory of syntactic, see features, theory of syntactic Varlokosta, S., 252 verb + auxiliary sentences, in Breton, 54-7 verb raising: and subject and object remaining in situ in Irish, 225, 228-9; to Comp in Irish, 225, 226-7, 230-7 verb-fronting, Breton, 55-6 verb-nouns, 29 verb-second clauses: in Germanic languages, 149; in Welsh proposed, 29, 148-50 verb-second (V2) phenomena, 10-11, 20-1, 121, 226, 337n verbal particles, or complementizers, 104-5

368

Index

verbs, analytic and synthetic forms, 272; see also finite verbs; verb-nouns Vergnaud, J.-R., 162n, 336n Vikner, S., 11, 51n, 198n, 281n Visibility Condition, 200-2; see also Generalized Visibility Visibility Criterion, extended version, 200-22 VP-internal subject hypothesis, 11-12, 21-3, 56-7, 88, 89, 223, 241, 242, 262-5, 279n VSO clauses, 19-23, 225-33, 241-83; in Breton, 23-5, 53, 54-7; root clauses, 19; subordinate clauses, 20-3 VSO languages, 14, 53, 203; analytic and synthetic agreement in, 190-6; basic or derived order, 19-20; doubt about unitary class, 273; long head movement in, 53-74; pronominal enclisis in, 171-99; pronouns, checking and, 186-9; 'residual', 187; structural invariance and lexical diversity in, 314—40 Wagner, H., 280n Wahba, W., 176 Watanabe, A., 237, 282n Watkin, Morgan, 124n Watkins, T.A., 97, 101, 119, 163n weak island constraint, 209, 22In Webelhuth, Gert, 93n, 222n Welsh, 2, 21, 46, 178, 236; A'-binding construction, 48; abnormal sentences, 27-8, 97-124; agreement and clitics, 40, 43, 192; Biblical, 99; clefting, 27-9, 97-124; clitics, 171-99, properties of, 181-6; Colloquial 26, 45, 102, 124n, 159, 273; copulas, 44-5, 75, 76-9; description, 3-4; echo pronouns 182-3, 185, 189, 193-6, 337n; fronting constructions in, 97-124; impersonal passives, 191;

interrogative complementizer, 28; Literary 3, 25, 26, 45, 105, 124n, 181, 182; mutation, 49-51; negation, 22, 46; nominals, 38-40; non-finite clauses, 29-34; in North Wales, 25; participial constructions, 35-8; particles listed, 25; personal pronouns, 181-2; in South Wales, 25, 28, 102; speakers, 4; VSO clauses, 19-23; see also Middle Welsh; Old Welsh; Pembrokeshire Welsh ^-Criterion, 296-7 w/z-questions, 47-9 Wilder, C , 65 Williams, E., 5, 92, 171,211 Williams, I., 123n Williams, S., 122n Williams, S.J., 93n, 95n, 98, 152, 279n Wojcik, R., 70 word order: and clitic systems, 171-99; contrasts over timing of syntactic movement, 314-40; cross-linguistic variation as variation within the lexicon, 314-15 X-bar Theory, 7, 13, 336n Yiddish, 11,21 yn, 30, 45, 76-8, 162-3n yw, 44-5, 126-7, 146-8; predicational, 156-9 Zagona, K., 211 Zanuttini, R. 292, 296, 31 In, 312n Zimmer, H., 280n Zubizarreta, M.L., 162n, 336n Zwart, Jan-Wouter, 11, 52n, 227, 237n, 337n Zwarts, F., 209, 211 Zwicky, A., 17, 50