Grammatical Categories: Variation in Romance Languages (Cambridge Studies in Linguistics)

This page intentionally left blank

G r a m m at ical Cat e gorie s

Grammatical categories (e.g. complementizer, negation, auxiliary, case) are some of the most important building blocks of syntax and morphology. Categorization therefore poses fundamental questions about grammatical structures and about the lexicon from which they are built. Adopting a ‘lexicalist’ stance, the authors argue that lexical items are not epiphenomena, but really represent the mapping of sound to meaning (and vice versa) that classical conceptions imply. Their rule-governed combination creates words, phrases and sentences – structured by the ‘categories’ that are the object of the present inquiry. They argue that the distinction between functional and non-Â�functional categories, between content words and inflections, is not as deeply rooted in grammar as is often thought. In their argumentation they lay the emphasis on empirical evidence, drawn mainly from dialectal variation in the Romance languages, as well as from Albanian. m . r i ta m a n z in i a n d l e o na r d o m. s avoia are both Full Professors of General Linguistics at the University of Florence.

In this series 94. 95. 96. 97. 98. 99. 100.â•… 101. 102. 103. 104. 105. 106. 107. 108. 109. 110. 111. 112. 113. 114. 115. 116. 117. 118. 119. 120. 121. 122. 123. 124. 125. 126. 127. 128.

j oan b yb e e : Phonology and language use l aur i e b aue r : Morphological productivity t homas e r ns t : The syntax of adjuncts e l i zab e t h c l os s t r augot t and r i c h a r d b. d a sh e r : Regularity in semantic change maya hi c kmann: Children’s discourse: person, space and time across languages di ane b l ake mor e : Relevance and linguistic meaning: the semantics and pragmatics of discourse markers i an r ob e r t s and anna r ous s ou: Syntactic change: a minimalist approach to grammaticalization donka mi nkova: Alliteration and sound change in early English mar k c . b ake r : Lexical categories: verbs, nouns and adjectives c ar l ota s . s mi t h: Modes of discourse: the local structure of texts r oc he l l e l i e b e r : Morphology and lexical semantics hol ge r di e s s e l : The acquisition of complex sentences s har on i nke l as and c he r yl zo l l : Reduplication: doubling in morphology s us an e dwar ds : Fluent aphasia b ar b ar a dancygi e r and e ve s w e e t se r : Mental spaces in grammar: conditional constructions he w b ae r man, duns tan b r ow n a n d g r e v i l l e g . c o r b e t t : The syntax–morphology interface: a study of syncretism mar c us t omal i n: Linguistics and the formal sciences: the origins of generative grammar s amue l d. e p s t e i n and t. dani e l se e ly: Derivations in minimalism paul de l acy: Markedness: reduction and preservation in phonology ye huda n. fal k: Subjects and their properties p. h. mat t he w s : Syntactic relations: a critical survey mar k c . b ake r : The syntax of agreement and concord gi l l i an c at r i ona r amc hand: Verb meaning and the lexicon: a first phase syntax p i e t e r muys ke n: Functional categories j uan ur i age r e ka: Syntactic anchors: on semantic structuring d. r ob e r t l add: Intonational phonology second edition l e onar d h. b ab b y: The syntax of argument structure b. e l an dr e s he r : The contrastive hierarchy in phonology davi d adge r , dani e l har b our a n d l au r e l j. wat k i n s: Mirrors and microparameters: phrase structure beyond free word order ni i na ni ng zhang: Coordination in syntax ne i l s mi t h: Acquiring phonology ni na t op i nt zi : Onsets: suprasegmental and prosodic behaviour c e dr i c b oe c kx, nor b e r t hor ns t e i n a n d ja i r o n u ň e s: Control as movement mi c hae l i s r ae l : The grammar of polarity: pragmatics, sensitivity, and the logic of scales m. r i ta manzi ni and l e onar do m . savo i a : Grammatical Â�categories: variation in Romance languages Earlier issues not listed are also available

CAMBRIDGE STUDIES IN LINGUISTICS General Editors: p. austin, j. bresnan, b. comrie, s. crain, w.dressler, c. j. ewen, r. lass, d. lightfoot, k. rice, i. roberts, s. romaine, n. v. smith

Grammatical Categories: Variation in Romance Languages

Grammatical Categories Va r i at i on i n Ro m ance L anguag e s

M. Rita M a n zin i University of Florence

Leo na r d o M . S avo ia University of Florence

cambrid ge uni ve r s i t y p r e s s Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, São Paulo, Delhi, Tokyo, Mexico City Cambridge University Press The Edinburgh Building, Cambridge CB2 8RU, UK Published in the United States of America by Cambridge University Press, New York www.cambridge.org Information on this title: www.cambridge.org/9780521765190 © M. Rita Manzini and Leonardo M. Savoia 2011 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 2011 Printed in the United Kingdom at the University Press, Cambridge A catalogue record for this publication is available from the British Library Library of Congress Cataloguing in Publication data Manzini, Maria Rita Grammatical categories : variation in romance languages / M. Rita Manzini, Leonardo Maria Savoia. â•… p.â•… cm. – (Cambridge studies in linguistics ; 128) ISBN 978-0-521-76519-0 (hardback) 1.╇ Grammar, Comparative and general–Grammatical categories.â•… 2.╇Language and languages–Variation.â•…I.╇Savoia, Leonardo Maria, 1948–â•…II.╇Title. P240.5.M36 2011 415–dc22 2010052183 ISBN 978-0-521-76519-0 Hardback Cambridge University Press has no responsibility for the persistence or accuracy of URLs or external or third-party internet websites referred to in this publication, and does not guarantee that any content on such websites is, or will remain, accurate or appropriate.

Contents

List of tables Acknowledgements

page x xi

Introduction:€grammatical categories and the biolinguistic perspective

1

1

 he structure and interpretation of (Romance) T complementizers

13

1.1

Romance complementizers are nominal and head their own noun phrase Structure of the complementizer phrase 1.2.1â•… Combining a left periphery in the complementizer phrase and in the embedded sentence; combining two complementizers 1.2.2â•… Some potential problems The left periphery beyond complementizers 1.3.1â•… Is order dictated by interpretation€– or interpretation by order? 1.3.2â•… Embedded contexts Conclusions

1.2

1.3

1.4

2

14 19

23 30 37 38 43 47

49

2.2 2.3

Variation in Romance k-complementizer systems Systems with two k-complementizers 2.1.1â•… Definite and indefinite complementizers€– and alternative analyses 2.1.2â•… Generalized wh–complementizers  ‘If’ The interaction with (non-)finiteness

3

Sentential negation:€adverbs

80

3.1

Sentential negation adverbs are nominal and argumental

83

2.1

49 54 61 65 73

vii

viii╅ Contents 3.1.1╅ Further evidence 3.1.2╅ Sentential negation adverbs as nominal arguments 3.2 Ordering sentential negation with respect to other adverbs 3.2.1╅ The order of negation with respect to aspectual adverbs 3.2.2╅The order of negation with respect to quantificational and manner adverbs 3.2.3╅ General discussion 3.3 The interaction of adverbial and verbal positions:€the participle

88 92 94 106

4

128

Sentential negation:€clitics

111 118 120

4.1 Interactions of negation clitics and subject clitics 4.2 Interactions of negation clitics with object clitics 4.2.1â•… Non-negative n 4.3 Negative concord and negative doubling

131 138 145 152

5

The middle-passive voice:€evidence from Albanian Data 5.1.1â•… Middle-passive morphologies 5.1.2â•… The interpretation of the middle-passive morphologies 5.1.3â•… The Arbëresh varieties The u clitic Specialized inflections 5.3.1â•… Be–participle

159

The auxiliary:€have/be alternations in the perfect Evidence 6.1.1╅ Theoretical background Auxiliary selection independent of transitivity/voice 6.2.1╅ Auxiliary selection according to person Splits according to transitivity/voice 6.3.1╅ Auxiliary selection according to voice 6.3.2╅ Auxiliary selection according to transitivity 6.3.3╅ Irreversibility Finer parametrization 6.4.1╅ Interactions between auxiliary selection according to transitivity/voice and according to person 6.4.2╅ The third auxiliary Some conclusions

196

5.1

5.2 5.3

6 6.1 6.2 6.3

6.4

6.5

160 160 164 169 172 184 188

196 203 208 209 216 216 218 222 223 224 228 233

Contentsâ•… ix 7 7.1

7.2

7.3

7.4

8 8.1 8.2 8.3 8.4 8.5

 he noun (phrase):€agreement, case and definiteness T in an Albanian variety Theoretical and empirical background 7.1.1â•… Nominal inflections in Albanian 7.1.2â•… Generative approaches to case Analysis of Albanian nominal inflections 7.2.1â•… Consonantal inflections 7.2.2â•… Vocalic case inflections and lack of inflections 7.2.3â•… Prepositional contexts 7.2.4â•… Summary The Albanian noun phrase 7.3.1â•… The genitive 7.3.2â•… The adjective 7.3.3â•… Adjectives as heads of the noun phrase Concluding remarks ( Definite) denotation and case in Romance:€history and variation The Latin case system Romance case systems:€Romanian Other Romance case systems€– and alternative accounts Loss of case in Romance:€Romansh –s Pronouns€– and some conclusions Notes References Index

236 237 239 244 246 250 255 259 261 262 262 266 272 275

276 277 286 295 302 308

312 331 345

Tables

6.1

6.2 7.1 7.2 8.1

x

 istribution of be (E) and have (A) according to person D in the present perfect (in Central and Southern Italian varieties) page 212 Distribution of be, have and syncretic forms in the present perfect in Piedmontese and Lombard varieties 231 Distribution of nominal inflections in Albanian 244 Denotational properties of Albanian nominal inflections 262 Denotational properties of Latin nominal inflections 286

Acknowledgements

The research reported in this book has been financed largely through PRIN grants from the MURST/MIUR, namely Per una cartografia strutturale delle configurazioni sintattiche:€ microvariazione nei dialetti italiani (1997–1999), La cartografia strutturale delle configurazioni sintattiche e le sue interfacce con la fonologia e la semantica. Parametri morfosintattici e fonosintattici (1999–2001), Categorie linguistiche:€Categorie di flessione nominale e verbale (Accordo, Aspetto); Nome e Verbo (2001–2003), I sistemi linguistici ‘speciali’ (apprendimento, disturbi) e la variazione tra i sistemi linguistici ‘normali’. Categorie funzionali del nome e del verbo (2003–2005), Strutture ricorsive in sintassi, morfologia e fonologia. Studi sulle varietà romanze. slave e albanesi (2005–2007), Morfosintassi e lessico:€ Categorie della flessione nominale e€verbale (2007–2009). Special thanks go to all our informants, both Romance and Albanian, though space limitations prevent us from mentioning all of them here. Our debt to the friends and colleagues whose work inspired ours should be obvious from the references. However, we take this opportunity to thank Neil Smith, as a (former) general editor of the series, for helping our project along.

xi

Introduction:€grammatical categories and the biolinguistic perspective According to Chomsky (2000b:€ 119), ‘the human language faculty and the (I–)languages that are manifestations of it qualify as natural objects’. This approach€– which ‘regards the language faculty as an “organ of the body”’€– has been labelled the ‘biolinguistic perspective’ by Chomsky (2005:€1). Hauser, Chomsky and Fitch (2002:€1570) base their discussion of the key biological question of evolution on the ‘biologically and individually grounded’ use of the term language ‘to refer to an internal component of the mind/brain (sometimes called “internal language” or “I-language”)’. They distinguish two conceptions of the faculty of language, one broader (FLB) and one narrower (FLN): FLB includes FLN combined with at least two other organism-internal systems, which we call ‘sensory-motor’ and ‘conceptual-intentional’ … A key component of FLN is a computational system (narrow syntax) that generates internal representations and maps them into the sensory-motor interface by the phonological system and into the conceptual-intentional interface by the (formal) semantics system … Most, if not all, of FLB is based on mechanisms shared with nonhuman animals … FLN€ – the computational mechanism of recursion€ – is recently evolved and unique to our species.â•…â•… (Hauser, Chomsky and Fitch 2002:€1571)

The conception of the language faculty and of (I-)languages as ‘natural’, ‘biologically grounded’ objects corresponds to specific theories concerning their internal articulation: the I-language consists of a computational procedure and a lexicon. The lexicon is a collection of items, each a complex of properties (called ‘features’) … The computational procedure maps an array of lexical choices into a pair of symbolic objects, phonetic form and LF [logical form] … The elements of these symbolic objects can be called ‘phonetic’ and ‘semantic’ features, respectively, but we should bear in mind that all of this is pure syntax and completely internalist.â•…â•… (Chomsky 2000b:€120)

The internal articulation of the FLN is crucial to the biolinguistic programme, no less than its applications to domains such as language evolution, genetics 1

2â•… The biolinguistic perspective and neurology. Here we address some points concerning this; specifically, we concentrate on the issue of language variation, starting with the idea that ‘the diversity and complexity can be no more than superficial appearance … the search for explanatory adequacy requires that language structure must be invariant’ (Chomsky 2000b:€7), and ‘There is a reason to believe that the computational component is invariant, virtually … language variation appears to reside in the lexicon’ (Chomsky 2000b:€120). From this perspective, a central aim of our work is to provide empirical support for what we may call the lexical parametrization hypothesis (Manzini and Wexler 1987), and thus to make more precise the sense in which it holds. Without a doubt ‘one aspect is “Saussurean arbitrariness”, the arbitrary links between concepts and sounds … However, the possible sounds are narrowly constrained, and the concepts may be virtually fixed’ (Chomsky 2000b:€120). In the present study, we address the issue of how the linguistically relevant conceptual space yields different (I-)languages beyond the obvious aspect of ‘Saussurean arbitrariness’. Before proceeding to the empirical core of the argument, we briefly introduce some of the conceptual underpinnings of the framework we adopt, beginning with the thesis that language ‘is a system that is, as far as we know, essentially uniform. Nobody has found any genetic differences … since its emergence there has not been any significant evolution. It has stayed that way’ (Chomsky 2002:€147). This view is shared by much current work on human cognitive and linguistic evolution (Lieberman 1991; Jackendoff 2002). The conclusion holds both for living languages and for ancient ones (whether documented and no longer spoken or merely reconstructed); as argued by Labov (1994), the same mechanisms of (surface) variation and change affect all of them. To take a comparative typological perspective: no evidence of anything like speciation has been found … Languages from typologically very different areas have the same latent structural potential … this survey has uncovered no evidence that human language in general has changed since the earliest stage recoverable by the method used here. There is simply diversity, distributed geographically.â•…â•… (Nichols 1992:€227)

As for this geographically distributed diversity: a residual zone or a set of residual zones will contain a good deal of the world’s possible linguistic diversity in microcosm, and both the existence of internal diversity and its actual profile are stable and obviously very natural situations. Diversity of a particular kind may even be regarded as the state to which a group of languages will naturally revert if left undisturbed … Spread zones, in contrast, are typically highly divergent from one another, but each is internally quite homogeneous … Just which language spreads in a

The biolinguistic perspective╅ 3 spread zone is a matter of historical accident, and this historical accident can distort the statistical distribution of linguistic types in an area.╅╅ (Nichols 1992:€23)

The set of languages considered in this work presents the kind of variation that we expect in natural languages in the absence of external constraints. Because of the political and cultural factors which, for centuries, have kept the Italian peninsula in conditions of great administrative and social fragmentation, dialectal differentiation in Italy has been preserved for longer (i.e. up to the present day) than in other areas of Western Europe, including Romance-speaking ones. Thus Italian varieties provide a rich and articulated picture of language variation that contrasts with that of other intensively studied varieties such as those of English. The view we take is that it is linguistic situations such as those in Britain, for example, that represent a somewhat misleading picture of variation, reflecting not only the internal shaping forces of language development, but also external mechanisms of social and political standardization. The variation seen in Albanian, including the major Gheg vs. Tosk divide in mainland Albania, and Arbëresh varieties of Southern Italy, has the same general character as that observed in Romance varieties. In the internalist (i.e. ‘biologically, individually grounded’) perspective that we adopt, variation between two or more varieties (linguistic communities) is in fact not qualitatively different from variation within the same variety (community), or even within the production of a single speaker. For example, to the extent that a speaker alternates between stylistic levels according to the situation of use, s/he will have a ‘bilingual’ competence of sorts€ – which, given the lexical parametrization hypothesis adopted here, can be accounted for as the co-existence of different lexicons with a single computational component (MacSwan 2000). Suppose, then, that the lexicon is the locus of linguistic variation€ – in the form of a uniform (i.e. invariant) computational component, and of an invariant repertory of interface primitives, both phonological and conceptual. Non-trivial questions arise at this point:€how can the lexicon vary on the basis of a universal inventory of properties (or ‘features’), and why does that variation in the lexicon result in variation in order, agreement, selection, and other relations that are computationally determined? These questions are amply debated in current linguistic theory. Our empirical discussion aims to support certain positions emerging from the debate, as opposed to others which are in principle equally possible. In particular, the answer to the preceding questions is mediated for various scholars by the notion that there is a fundamental distinction between functional and non-functional elements. Thus, within the Distributed Morphology framework, Embick (2000:187) assumes a ‘distinction between the functional and lexical vocabularies of a language … functional categories merely

4â•… The biolinguistic perspective instantiate sets of abstract syntacticosemantic features’, on which the derivational component operates. The actual phonological terminals corresponding to these abstract categories are inserted only after a level of morphological structure, where readjustment rules apply (Late Insertion). It is evident that the overall architecture of the grammar implied by this model is considerably more complex than one in which ‘the formal role of lexical items is not that they are “inserted” into syntactic derivations, but rather that they establish the correspondence of certain syntactic constituents with phonological and conceptual structures’ (Jackendoff 2002:€131). Kayne’s (2006, 2008a) parametrization model, while avoiding recourse to Late Insertion, is close to Distributed Morphology in assuming that functional items correspond to a universal lexicon of sorts. Lexical and hence grammatical differences depend on whether the elements of this functional lexicon are overtly realized or ‘silent’. Interestingly, for Kayne (2006), even variation in the substantive lexicon can be reduced to variation in functional structure in the sense just defined, as can be seen in his construal of shallow as ‘LITTLE deep’, that is, essentially as the specialized lexicalization of deep in the context of the silent functional category ‘little’. Manzini and Savoia (2005, 2007, 2008a) pursue a model under which, again, there is a unified conception of lexical variation€– however, this is of the type traditionally associated with the substantive lexicon:€there is a conceptual and grammatical space to be lexicalized and variation results from the distinct partitioning of that space. There is no fixed functional lexicon which varies along the axis of overt vs. covert realization€– so-called functional space is just like all other conceptual space, and all lexical entries are overt. Thus, the distinction between functional (i.e. grammatical) contents and conceptual ones is an external one; as such it may very well be useless, and at worst it may obscure the real underlying linguistic generalizations. Our conception of variation within the so-called functional lexicon is consistent with current conclusions regarding the conceptual space and the different ways in which it surfaces in natural languages. Fodor (1983) and Jackendoff (1994), among others, develop the Chomskyan theme that concepts, like other aspects of language, must have an innate basis€– largely because of the poverty of stimulus argument. It has already been observed by Lenneberg (1967) that lexical items are the overt marks of a categorization process through which human beings carve out an ontological system from the perceptual continuum of the external world. This process of categorization is of course only indirectly connected with the objects of the external world. Jackendoff (1994:€195) notes that the lexical forms employed to express spatial location and motion (e.g. The messenger is in Istanbul; The messenger went from Paris to Istanbul;

The biolinguistic perspectiveâ•… 5 The gang kept the messenger in Istanbul) typically also express possession (e.g. The money is Fred’s; The inheritance finally went to Fred; Fred kept the money), the ascription of properties (e.g. The light is red; The light went from green to red; The cop kept the light red), etc. This suggests that thought has a set of precise underlying patterns that are applied to pretty much any semantic field we can think about. Such an underlying ‘grain’ to thought is just the kind of thing we should expect as part of the Universal Grammar of concepts; it’s the basic machinery that permits complex thought to be formulated at all.â•…â•… (Jackendoff 1994:€197)

Dehaene, Izard, Pica and Spelke (2006) study geometrical concepts in an isolated group of Amazonian people whose language, Mundurukú, ‘has few words dedicated to arithmetical, geometrical, or spatial concepts’. They conclude that geometrical knowledge arises in humans independently of instruction, Â�experience with maps or measurement devices, or mastery of a sophisticated geometrical language … There is little doubt that geometrical knowledge can be substantially enriched by cultural inventions such as maps, mathematical tools, or the geometrical terms of language … however, the spontaneous understanding of geometrical concepts and maps by this remote human community provides evidence that core geometrical knowledge, like basic arithmetic is a universal constituent of the human mind.â•…â•… (Dehaene, Izard, Pica and Spelke 2006:€385, our italics)

In a similar vein, Hespos and Spelke (2004) study the acquisition of the conceptual distinction between ‘tight’ and ‘loose’ fit of one object to another in English-speaking children, which is not lexicalized in English, though it is in other languages like Korean. Their conclusion is that ‘like adult Korean speakers but unlike adult English speakers, these infants detected this distinction … Language learning therefore seems to develop by linking linguistic forms to universal, preexisting representations of sound and meaning’ (Hespos and Spelke 2004:€453). In short, the building blocks that are combined to make up the potentially infinite variety of human lexicons are innate. The lexicons of different languages are formed on this universal basis, covering slightly different extensions of it and in slightly different ways. The view we advocate here is simply that ways of representing the event, such as transitivity or voice (chapters 5–6), ways of connecting arguments to predicates (or to one another), such as cases (chapters 7–8), and more, are to be thought of as part of this general system. There is no separate functional lexicon€ – and no separate way of accounting for its variation. We started with the general Chomskyan biolinguistic, or internalist, picture of language, and of its basic components, both broadly and narrowly construed. Variation is crucial to establishing this model for the obvious reason that the uniformity thesis, as laid out above, requires a suitably

6â•… The biolinguistic perspective restrictive account of observed cross-linguistic differences. But, even more fundamentally, the lexical parametrization hypothesis that we adopt means that questions of variation will inevitably bear on the form of the lexicon, as one of the crucial components of the I-language. The other main component of the I-language is ‘the computational procedure’, which ‘maps an array of lexical choices into a pair of symbolic objects, phonetic form and LF’ (Chomsky 2000b, quoted above). As for the latter, Culicover and Jackendoff (2005:€ 6) aptly characterize a particularly popular conception of the relation of LF to the syntax (i.e. the computation) as ‘Interface Uniformity’, which holds that ‘the syntax-semantics interface is maximally simple, in that meaning maps transparently into syntactic structure; and it is maximally uniform, so that the same meaning always maps onto the same syntactic structure’. This bias inherent in much current theorizing provides a standardized way of encoding the data, but does not appear to have any strong empirical motivation; nor is the encoding it provides a particularly elegant or transparent one. Conceptually it corresponds to a picture where syntax ‘includes’ interpretation, in the sense that all relevant semantic information finds itself translated into syntactic structure. In contrast, we agree with Culicover and Jackendoff (2006:€ 416) on the idea that interpretation is ‘the product of an autonomous combinatorial capacity independent of and richer than syntax’, ‘largely coextensive with thought’, which syntax simply restricts in crucial ways. Linguistic meanings are merely an input to general inferential processes; the linguistic categorization of the conceptual space encoded by lexical items does not correspond to ‘meaning’ itself but rather to a restriction of the inferential processes producing it. Sperber and Wilson (1986:€174) provide a particularly compelling discussion of the point that linguistic expressions only denote because of their inferential associations:€ ‘Linguistically encoded semantic representations are abstract mental structures which must be inferentially enriched’. In such a model, the well-known indeterminacy of linguistic meanings becomes a key property of successful communication: A linguistic device does not have as its direct proper function to make its Â�encoded meaning part of the meaning of the utterances in which it occurs. It has, rather, as its direct proper function to indicate a component of the speaker’s meaning that is best evoked by activating the encoded meaning of the linguistic device. It performs this direct function through each token of the device performing the derived proper function of indicating a contextually relevant meaning.â•…â•… (Origgi and Sperber 2000:€160)

Note that we disagree with Culicover and Jackendoff (2005) on the model of syntax to be adopted. Our analysis depends on a representational version of minimalism, roughly in the sense of Brody (2003). Crucially, the LF primitives

The biolinguistic perspectiveâ•… 7 we employ are independently available within a minimalist grammar as defined by Chomsky (1995), and in this sense the approach we take is compatible with Chomsky’s model. In fact, we would argue that our views on lexical variation and on interpretation are the simplest construal of Chomsky’s (2000b) proposals, as summarized above€– much simpler than other current approaches, and in this sense closer to the core of minimalism and of the biolinguistic programme. Therefore, any theory maintaining a functional/lexical divide must define the boundary between the two€– which is a far from trivial task. The domain of spatial relations and of events involving them is a case in point. Spatial relations are covered by prepositions (or particles in their intransitive use), among other items. In particular, prepositions/particles can combine with elementary verbs to lexicalize events with a spatial component; for instance, English has put down (the book), Northern regional Italian has mettere giù (il libro). At the same time, Tuscan and literary Italian has a verb posare ‘put down’, and the examples could be multiplied (go in and enter in English, etc.). Particles in Germanic languages (but also in Romance, for instance in Northern Italian varieties) also allow for aspectual interpretations. If, on the basis of these, of the role they play in case systems, etc., we treat prepositions/particles as part of the functional lexicon, what should we infer about spatial primitives? Are they functional? If so, how is their relation to posare, enter, etc. (i.e. canonical lexical verbs) expressed? As mentioned above, the answer envisaged by authors such as Kayne (2006) is that apparent variation in the substantive lexicon reduces to variation in the pronunciation of functional categories; hence the substrings lexicalized by what would traditionally be thought of as lexical categories consist in reality of a number of functional specifications€– which may surface in some languages and not in others, or surface to different extents in different languages. In this way, the functional lexicon effectively spreads over considerable portions of the substantive lexicon; taking this to the extreme, one may want to say that lexical categories are but an epiphenomenon of abstract functional structure. Since the proposal we are putting forward is that lexicons are merely ways of partitioning an abstract categorial space, we are in a way suggesting theories close to those we are taking issue with. At the same time, we consider it significant that we take the step of calling the lexical/functional divide into question, while they typically don’t. To begin with, the different approaches make different empirical predictions in the data domains they both address. Thus, we have specifically referred to Kayne (2006, 2008a, 2009) and Distributed Morphology, since we can directly compare our respective approaches with regard to such domains as fine variation in clitic structures, where we believe our model to be preferable on grounds of descriptive as well as explanatory adequacy (Manzini and Savoia 2009b, 2010).

8â•… The biolinguistic perspective The lexical/functional issue seems to us particularly noteworthy, because at heart it concerns the distinction between the narrow and broad language faculty (FLN and FLB). Let us assume that there is a universal inventory of concepts, and that the lexicon represents a way of realizing it. In theories in which there are in fact two inventories, one for functional categories and one for non-functional ones, it seems to us that the functional and non-functional lexicons are implicitly or explicitly apportioned to the language faculty narrowly construed and broadly construed, respectively. The reduction of the divide that we are proposing has implications not only for the more technical aspects of the theory of grammar, but also opens up the possibility that the universal conceptual repertory which is partitioned by language-particular lexicons is part of the broadly construed language faculty in its entirety. In fact, we see no reason why the grammatically relevant categories investigated here should not constitute categorizations in a domain of general cognition. In other words, what we are saying is that the existence of a functional lexicon associated with the FLN is not a matter of logical or factual necessity€– and as such it should be open to scrutiny. Given the position that we tentatively take on the matter€– namely that eliminating the divide does not imply any empirical problem, and on the contrary allows for a certain simplification of the architecture of language€– we may wonder why such a distinction is so prominent in linguistics. Neuropsychological literature provides much evidence, based both on recent brain imaging techniques and on more traditional language disorders and acquisition studies, that different brain areas are implicated by different conceptual clusters. The prediction is that manipulable objects such as tools are strongly linked to motor behaviour and therefore their representational networks should comprise a significant amount of neurons in motor contexts. Animals, which are most of the time (visually) perceived rather than manipulated, should be represented by networks that partly reside in the visual cortex.â•…â•… (Bastiaansen et al. 2008)

Conversely, ‘assemblies representing function words remain limited to the perisylvian cortex and strongly left-lateralized in typical right-handers’ (Pulvermüller 1999:€260–1). This appears to underlie, in particular, the differential treatment of different sublexicons by aphasic patients (anomics, agrammatics, etc.). Given such results, it does not seem to us to be necessary to draw the conclusion that there is a functional lexicon associated with the computational system of natural language and distinguished on these grounds from a contentive lexicon. Another possibility is that there is a continuum of meaning complexity between the ‘simple’ concrete content words that have clearly defined entities they can refer to … more abstract items that may or may not be used to refer to objects and actions

The biolinguistic perspectiveâ•… 9 and function words … According to the present proposal, the important criterion is the strength of the correlation between the occurrences of a given word form and a class of non-linguistic stimuli or actions.â•…â•… (Pulvermüller 1999:€261)

In other words, it is not so much the functional lexicon that has a special status within the architecture of the mind-brain, but rather certain concrete contents as opposed to more abstract ones. Once freed from the burden of highly articulated inventories and hierarchies of functional categories, we can entertain a simpler syntax, much in the sense of Culicover and Jackendoff (2005). As already mentioned, on the other hand, we do not believe that levels of representations of the type proposed by Culicover and Jackendoff (2005), including rich notions such as grammatical functions, linking rules etc., are required by such a simpler syntax. Rather, the grammar implemented here is a representational version of current minimalist theories (cf. Brody 2003). The relation of the syntax, and more precisely its LF component, to interpretation, as outlined above, is crucial in our view to understanding the role of language variation in the overall economy of the faculty of language. If our construal of syntax and its relation to interpretation is correct, the syntax restricts interpretation, but does not ‘contain’ it (Culicover and Jackendoff 2006). Thus the boundary between syntax and interpretation is a loose one, allowing for a number of different matchings of syntactic form to (inferentially determined) meaning. The looseness of this relation seems to be an essential design feature of the faculty of language, in the sense that it permits the invariant constructs of syntax to cover changing meanings. Lexical items are at the core of language variation simply because they represent the core unit of this interface between syntax and interpretation. In this sense, variation is not an accidental property of the faculty of language, and neither are the characteristics of variation that we try to outline in this study. Rather, they pretty much represent a by-product of the general design of the language faculty. The aspect of our work which provides the title for this book (‘grammatical categories’) has to do with the redefinition of the grammatically relevant classes (i.e. the ‘categories’) of natural language. In general, we take it that the lexicons of natural languages are learnable in that lexical entries individuate natural classes. We apply this logic in particular to Romance complementizers which have the same form as wh–items (Italian che and the like) and to Romance sentential negations which have the same form as negative polarity arguments, in particular ‘nothing’ (Piedmontese nen etc.). In both cases we conclude that lexical identity of form is not a matter of homophony but reveals the sharing of deeper categorizations. This calls into question, among

10â•… The biolinguistic perspective other things, the classical functional categories of C(OMP) (chapters 1–2) and NEG (chapters 3–4). Elsewhere in this book, we find no reason to entertain a functional category status for the so-called AUX(iliaries) have and be, which are argued just to be main verbs selecting a participial clause (chapter 6). In chapter 5 the cluster of meanings associated with Romance si and its Albanian counterpart u are reduced to a unified characterization which also holds of other morphological instantiations of middle-passive voice. Even syncretisms involving case morphology€ – and the functional category K(ASE) according to some (cf. Fillmore 1968; Giusti 1995), are analysed in chapters 7–8 as instances of ambiguous interpretation of the same underlying category, rather than as instances of default lexicalization. This, in turn, requires a revision of the categorizations provided by standard morphological feature systems. It should be kept in mind that the functional structure that this book calls into question (COMP, NEG, AUX, K) is quite independent of recent cartographic proposals (see Cinque and Rizzi (2008) for an overview) which aim to provide a fine-grained picture of functional categories and the way in which they map to syntactic hierarchies. The result is an increase in the number of functional categories, yielding hierarchies of considerable complexity, which have been objected to on the grounds that they enrich the grammar by introducing a great number of new categories and orderings. Yet the same concern regarding the expressive power of the theory could be voiced for standard approaches to functional structure, since the creation of a new functional category or a new feature annotation of an existing category is not subject to any formal or substantive constraints. In this book we propose a take on the problem which goes back to the very first models of exploded structures (Larson 1988), and even further to the very first approaches to ‘functional’ structure in generative grammar (Rosenbaum 1967 on complementation). We argue that structures are indeed atomized, in the sense that a wealth of differentiated head positions are projected under Merge. At the same time, our contention is that a considerable amount of this atomization (perhaps all) does not derive from the introduction of novel categories, but simply from the recursion of certain elementary, identical cells. Thus, the complementizer (chapters 1–2) is not introduced as a specialized head C(OMP); rather, the clearly nominal nature of the complementizer in Romance languages (as in Germanic ones) suggests that the complementizer is the N complement of the matrix verb; in turn, this N takes the embedded sentence as its complement. This structure is as internally articulated as that of Rizzi (1997), but its internal articulation does not depend on a functional hierarchy. Rather, it depends on the recursion of ordinary nominal and sentential

The biolinguistic perspectiveâ•… 11 embeddings. Similarly, negation (chapters 3–4) is one of the earliest functional categories proposed under an articulated view of phrase structure, dating back at least to Pollock (1989). Based on evidence from Romance varieties, we propose, however, that so-called negative adverbs and heads are negative polarity elements, and, even more radically, that they participate in the argumental structure of the verb, coinciding specifically with the individuation of the internal argument position. Thus, in Romance languages, there is neither evidence for a lexicalized negative operator nor for a functional position hosting it. The final case study to be introduced here concerns the internal structure of nouns and noun phrases (chapters 7–8). Following an established trend in generative grammar, we argue for the conclusion that noun phrases (as well as adjective phrases) have the same internal organization as sentences. From this perspective, we take up the classical proposal of Higginbotham (1985) that the D(eterminer) saturates the obligatory (internal) argument of the nominal predicate; in this sense, D properties yet again represent an instantiation not of functional structure, but of predicate–argument structure. Case, in turn, is not construed as a (functional) primitive of grammar, but rather as a label covering much more elementary properties, relating again to the saturation of predicate– argument structures. Throughout the discussion, the emphasis is very much on empirical evidence. We repeatedly argue that our model not only fares better with respect to fairly reasonable simplicity metrics, but also that it has descriptive advantages. In fact, and quite strikingly in our opinion, less powerful theories are better suited to capturing complex (micro)variation data of the type we consider than theories potentially capable of greater descriptive power. In particular, we subscribe to the simplicity argument in favour of representational grammars advanced by Brody (2003). This implies abandoning derivations, including the notions of a cycle (phases) and an asymmetric search space (feature checking). What we retain is representational relations:€chains, agreement, etc. Simplicity is paramount, to the extent that existing empirical evidence does not provide any support for the more complex grammar. In particular, complex data concerning agreement (and variation in agreement patterns) are accounted for in Manzini and Savoia (2005, 2007, 2008a) by abandoning phifeature checking in favour of identity (or better, compatibility) of referential properties; uninterpretable and unvalued features are also eliminated under this approach. At no point is there any evidence that a derivational approach would have empirical advantages€ – on the contrary, the complexity of the variation effectively requires the simpler representational approach. In general, representational grammars are simpler than derivational ones in that the latter postulate

12â•… The biolinguistic perspective purely computational processes whose results are LF-relevant and hence redundant with LF constructs. Movement and the LF-relevant notion of chain are the obvious cases, but this also holds of the computational operation of agreement and its LF reflexes relevant for coreference etc. By contrast, the representational model views LF-relevant relations as determined directly by the interpretive calculus at the LF interface (chains by the theta-calculus, and so on). This adoption of a representational model goes hand in hand with the present take on functional structure. In the minimalist grammar of Chomsky (1995), functional categories, or features, bear the burden of computation, since they crucially enter into feature checking. To the extent that functional categories are reduced to ordinary predicates and arguments, functional feature checking is also replaced by ordinary selectional restrictions. In turn, empty functional heads and/or Specs, needed only for the checking of functional features, can be eliminated. As for the highly articulated hierarchies proposed in cartographic research, if functional and lexical properties of lexical entries are not formally different, as argued here, the most natural and economical assumption is that their ordering is restricted entirely by interpretive principles, including closure requirements, scope, etc. Needless to say, in the absence of any theoretical distinction between functional and lexical elements, we do not expect that they will be inserted at two different points in the derivation, namely in syntactic and morphological structure respectively (as in Distributed Morphology). On the contrary, we predict that all syntactic structure will be projected directly from lexical entries, independently of their properties. In other words, we propose a view in which morphological-level structures and relations are entirely unified with syntax. Lexical entries, in turn, are entirely characterized in terms of positive properties. Another theme of general significance is the question of how the present approach to functional structure relates to the issue of ‘grammaticalization’. In the terms of, say, Roberts and Roussou (2003), grammaticalization is essentially the reanalysis of a lexical category as a functional category, often with the result that a lexical entry can have both a lexical and a functional construal. In such cases we typically propose that there is a single lexical item with a unified (lexical, not functional) characterization; the che ‘that’ complementizer/wh–phrase (chapters 1–2) is a case in point. It is evident that, to the extent that this latter treatment can be generalized, the distinction between functional and lexical dissolves; if so, the problem of why exactly lexical categories would turn into functional ones (i.e. the problem of ‘grammaticalization’) simply does not arise.

1 The structure and interpretation of (Romance) complementizers

In this chapter we start from the observation that in Romance languages, Â�complementizers are nominal, belonging to the same argumental series as wh–phrases, although current theories treat them as functional projections of verbs, filling the same positions as verbs do. We argue that the Romance chetype complementizer is not a functional category of the verb, but rather a nominal head, which satisfies an argument slot of the matrix verb and which takes the embedded sentence as its complement (section 1.1). Both as a wh–phrase and as a complementizer, Italian che introduces a variable. If it Â�introduces an individual variable, the wh–phrase reading arises; if it introduces a propositional variable (ranging over situations or possible worlds) it is read as a so-called complementizer. In section 1.2, we argue that our proposal is compatible with fine distributional evidence relating to the left periphery of the sentence, in particular with the fact that elements such as topics or foci can occur both below the complementizer and above it. We capture this distribution by allowing them to occur at the left periphery of the embedded sentence (below the complementizer) or at the left periphery of the complementizer phrase (above the complementizer). This structural hypothesis predicts that occurrences of the relevant material below and above the complementizer can combine€ – a fact that requires quite complex functional hierarchies in alternative accounts. It should be kept in mind that, insofar as this chapter proposes an alternative to the generally adopted view that the complementizer is a functional projection of the sentence, it is not aimed specifically at so-called cartographic models. These, however, are discussed in some detail in section 1.3 (especially Rizzi 1997, 2001, 2004), because they explicitly consider the same type of data as we do. In section 1.2 we also briefly consider possible general objections to our proposal, for instance concerning the distinction between complementizer phrases and conventional noun phrases:€how is this distinction made for the purposes of selection? Extraction facts are also relevant:€ how are sentences 13

14â•… Romance complementizers:€structure & interpretation introduced by complementizers different from complex NPs? For the purposes of selection, complementizer phrases can be distinguished from conventional noun phrases in that they introduce a propositional and an individual variable respectively. As for extraction, the question why complementizer phrases, in contrast to conventional noun phrases, are not islands is the same as the question why complementizer phrases have a left periphery that is an ‘escape hatch’ (Chomsky 1973), and conventional noun phrases do not. This remains very much an irreducible primitive of all generative approaches (up to Chomsky 1995, 2008), since it is not the case that noun phrases lack a left periphery altogether (Szabolcsi 1994). We construe this primitive not as a structural, but as an interpretive one, contrasting propositional denotations with individual denotations. A different question concerns the fact that, although Romance complementizers belong to the wh–series, they typically have a non-interrogative interpretation. We return to this question in chapter 2, where we show that this property is parametrized. In particular, we illustrate Romance systems where the ‘that’ complementizer€– or an element of the wh–series€– lexicalizes the ‘if’ (interrogative/ hypothetical) complementizer as well. In general, the argument in favour of the present theory is based on explanatory adequacy. First, it simplifies the lexicon, allowing for a unified lexical entry for elements like Italian che, the only possible alternative being homophony. It also solves the mystery of why C would host such disparate categories as verbs and complementizers, reserving what we might continue to call C for verbs. Perhaps most interestingly, it reduces what would otherwise be potentially complex functional hierarchies to the recursion of simple predicate–argument structure (i.e. the complementizer is an argument of the matrix predicate, taking the embedded sentence as its argument). Furthermore, the burden of proof is on alternative theories to prove that they can account in an explanatory way for the fine variation we observe in complementizer systems, a matter to which we return in chapter 2.

1.1

Romance complementizers are nominal and head their own noun phrase

Current theories hold that complementizers are functional projections of the verb, that is, their position is essentially one which could in other circumstances be filled by a verb. This is true independently of whether there is a set of C positions (Rizzi 1997), or a single C position with a rich specifier structure (Chomsky 1995). But if complementizers fit into the same type of position as

Romance complementizers are nominalâ•… 15 verbs, it is unclear why they so clearly coincide with functional projections of the noun:€for instance with demonstratives in Germanic, or with wh–elements in Romance. Even if we were to accept that the relation of complementizers to the wh–system of Romance languages, to the demonstrative system of English etc. is a purely historical one, the question would arise of how nominal, quantificational elements could come to fit verbal specifications. In other words, the form of the problem would change, but not its substance. Consider for instance Italian, in which che, like English that, introduces finite declaratives, as in (1); since Kayne (1976) this has been identified with the relative clause introducer in (2). The same element also introduces interrogatives with the meaning of ‘what’ as in (3). Furthermore, che can appear as the wh–determiner of complex interrogative NPs as in (4). (1)â•… So che fai questo I.know that you.do this ‘I know that you do this’ (2) Il lavoro che fai è noto the workâ•… thatâ•… you.doâ•… is â•… known ‘The work you do is well-known’ (3) Che fai? what you.do? ‘What are you doing?’ (4) Che lavoro fai? which job you.do ‘Which job do you do?’

One possibility that we can reject is that Italian che simply has two lexical entries, one of which corresponds to the ‘that’ complementizer and the other to the ‘what’ wh–element. This solution does not have any explanatory value, given that the pattern that it describes is not an accidental coincidence observed in one or even a few languages, but a systematic phenomenon in Romance, as can be gleaned from the data to follow. What is more, Caponigro and Polinsky (2008) find the same formal identity between the wh–system and the complementizer system in a completely unrelated language, Adyghe (a NW Caucasian, Abkhazo-Adyghean language), showing that in Adyghe the same syntactic structure can be mapped to four different meanings, namely relative clause, complement clause, wh–interrogative and yes–no interrogative. Uncontroversially, in sentences like (3) che heads its own noun phrase, which in turn fills a position in the C field of the sentence, as in (5a). From an interpretive point of view, che introduces a variable corresponding to the internal argument of the predicate fare ‘to do’, as in (5b).

16╅ Romance complementizers:€structure & interpretation (5) a. [che] [C b. che x, fai x

[I fai

In examples of the type in (4), the interrogative noun phrase che lavoro ‘what job’ is associated with a structure of the type in (6a), where che represents a quantificational position Q within the noun phrase. Assuming that the structure of the noun phrase is parallel to that of the sentence, the inflected head noun is positioned in I€ – to parallel the position of the inflected verb within the sentence. From an interpretive point of view, che again introduces a variable restricted by the predicate lavoro ‘job’, as in (6b). (6) a. [Q cheâ•… [I lavoro]] [Câ•… [I fai b.â•… che x:€x lavoro, fai x

Apart from the fact that both complementizer and wh–phrase take the morphophonological form che, from a syntactic point of view the complementizer che in (1) and the wh–phrase che in (3) both occupy a position at the left periphery of the sentence. Since under current theories, complementizer che fills precisely the kind of position that we questioned at the outset, namely C, in (7a) it is assigned to an unnamed position above C€– the latter independently known to be a possible position of the verb in the sentence. From an interpretive point of view, complementizer che could be an operator introducing a variable, again like wh–phrase che€– the main difference between them being the nature of the variable. This ranges over individuals for the wh–phrase che in (5)–(6), while for complementizer che it ranges over situations/possible worlds, as in (7b).1 (7) a. [che [Câ•… [I fai questo b. che x: x fai questo

The fact that the operator-variable structure is interpreted as a question in (5)–(6), but not in (7), need not stand in the way of their unification, as we know that wh–phrase che can also have non-interrogative interpretations (very much like its English counterpart what). Thus che can be the head of a free relative, as in (8), or an exclamative, as in (9). The most natural interpretation of this range of data is that question, declarative (relative) and exclamative values are not intrinsic to the wh–phrase, but rather contextually determined.2 (8) Fai che ti pare do what to.you pleases ‘Do what you like’ (9) Che non farei! what not I.would.do ‘What I wouldn’t do!’

Romance complementizers are nominalâ•… 17 Given the discussion so far, if complementizer che appeared in the appropriate context, there is no reason why it shouldn’t assume the interrogative value of, say, English if or whether. Now, in well-known Romance languages like Italian, ‘if’ has a specialized lexicalization, se in Italian, in which the hypothetical and interrogative values overlap. In chapter 2, however, we will show that this state of affairs does not necessarily hold€– in other words, that there are languages in which the lexicalizations of ‘that’, ‘if’ and ‘wh–’ overlap, as we expect. Similarly, recall that the Adyghe morphology discussed by Caponigro and Polinsky (2008) is ambiguous between a ‘yes–no interrogative’ and a ‘complement clause’ interpretation (among others). Let us assume, then, that complementizer che is to be identified with wh– –phrase che (as the morphology but also the interpretation suggest) and that, as anticipated in the discussion of (7a), it cannot be hosted in C, since C is a verbal position, while che is nominal. An alternative structure is suggested by what is perhaps the earliest approach to finite complementation in generative grammar (Rosenbaum 1967) as well as by analyses of special subsets of complement sentences, notably factives (Kiparsky and Kiparsky 1970) and unselected questions (Adger and Quer 2001). These have in common the postulation of a nominal structure of some sort for sentential complementation. What we propose here is that every sentence introduced by che has a nominal layer, represented by a noun phrase headed by che itself. Thus, in a sentence like (1) the verb heading the matrix sentence takes as its complement a noun phrase headed by the che complementizer, which in turn takes the embedded sentence as its complement, yielding a structure of the type in (10). Following the assumptions we have already introduced in (6) concerning the position of head nouns in noun phrases, che is in I.3 (10)

I che C I fai V

questo

18â•… Romance complementizers:€structure & interpretation It should also be noted that the theoretical literature does contain treatments in which sentential introducers are removed from the C projection(s) of the embedded sentence and are made to depend directly on the main sentential tree. Notably, Kayne (1994), in analysing the infinitival introducers of Romance languages, such as Italian di ‘of’ etc., treats them not as complementizers, but as functional projections of the main verb, triggering movement of the embedded sentence to their Spec position. What we propose here is different, in that we are rejecting entirely the idea that complementizers are functional heads (either of the embedded or of the main sentence) and we are arguing that they are true arguments of the main verb, in turn taking the embedded sentence as their argument. For reasons of space we will not be able to deal with di–type introducers here. Manzini (1982) and Manzini and Savoia (2005) treat them as prepositions. A more direct comparison can be established with the work of Arsenijevic (2009), according to whom ‘the variable denoted by a wh–element gets bound by a question operator in questions or by an appropriate head in relatives. Its presence in F[inite] C[omplement] C[lause]s in a significant number of languages signals that the denotation of these clauses involves a variable’. This conclusion converges with those in the text, yet the syntax proposed by Arsenijevic (2009) for English markedly differs from ours. Thus, that, which semantically introduces a lambda operator, syntactically occupies the conventional C position. The variable over which the lambda operator abstracts corresponds to the Spec of a Force head generated immediately under the C head. This position is occupied either by a nominal expression with Force content such as claim or by its abstract incorporated counterpart for verbs such as to claim (analysed as make claim). In these terms the overall structure assigned to claim that John kissed Mary is a relativization headed by claim:€[N claim [C that [Force claim [John kissed Mary]]]]. Similarly, Kayne (2010) revises his (1976) idea that the relative que of French is really the que complementizer, by proposing instead that ‘that isn’t [a complementizer]. The that that introduces sentential complements is really a relative pronoun … The claim that English sentential that is a relative pronoun must be taken to extend … for example, to Italian che, to French que … From the present perspective that is not a Force in Rizzi’s (1997) sense nor the head of a CP phase in Chomsky’s (2001) sense’. The actual implementation that Kayne provides for these ideas differs markedly from ours, while bearing a considerable resemblance to Arsenijevic’s. In particular, for Kayne ‘factive sentences … have a deleted or silent FACT. If so … factives too must involve relative clause structures.’ Assume, then, that sentential complements and

Structure of the complementizer phraseâ•… 19 sentential subjects are always accompanied by a head noun, even if they are not factive … Either it will be raised from within the relative in a way largely parallel to what happens with overt fact … or a silent noun will be so raised’. We have insisted more than once on the empirical reasons, but also on the general theoretical problems that lead us to reject the silent categories approach of Kayne, for instance in the domain of clitic phenomena (Manzini and Savoia 2008a, 2009a, 2010; Savoia and Manzini 2010). These effectively apply to Arsenijevic (2009) as well. On the other hand, it seems to us that the extra assumptions concerning the (silent) Force phrase of Arsenijevic, or equivalently the silent head nouns of Kayne, are unnecessary, since both the semantics for propositional embedding and the identity of so-called complementizers and wh–phrases follow from the simpler syntax proposed here. We will return to this comparison in chapter 2. In the next section, we argue that the structure in (10) is compatible with the distributional evidence concerning the so-called left periphery of Romance languages€– in fact, it provides a particularly economical way of dealing with it. 1.2

Structure of the complementizer phrase

Given a structure like (10), we predict that material related to the quantificational and informational structure of the sentence (wh–elements, topic, focus, etc.) should be hosted by the left periphery of the embedded sentence and hence should follow the complementizer. Indeed the presence of such material under the complementizer is familiar from English as well as from Romance languages. In (11) we provide some examples from Italian, where (11a) illustrates a topic and (11b) a focus in the left periphery of the embedded sentence. (11)â•… a.â•… b.

So che questo non l’hanno preso I.knowâ•… that â•… this not it they.have taken ‘I know that they haven’t taken this’ So che questo hanno preso (non quello) I.know that this they.haveâ•… taken (not that) ‘I know that they have taken this one, not that one’

As expected, wh–phrases can also appear under the complementizer. In fact, in our own judgement, this option is open in Italian, as in (12). In any event it is fully productive in Southern Italian varieties such as Arena in (13), where any wh–phrase can appear either in the left periphery of the matrix sentence€– i.e. in its scope position, as in (13a’) and (13b’)€– or in the left periphery of the embedded sentence, hence under the complementizer, as in (13a) and (13b).

20â•… Romance complementizers:€structure & interpretation (12)â•… Credi che con chi se la prenderanno you.think that with whom MP it they.will.take ‘Who do you think they will take it out on?’ (13) Arena (Calabria) a.â•… ti kriði ka ðuvi ðrmi you thinkâ•… thatâ•… whereâ•… he.sleeps ‘Where do you think he sleeps?’ a’. duvi ti kriði ka ðrmi where you think that he.sleeps ‘Where do you think he sleeps?’ b. ti kriði ka pEkki vEni you thinkâ•… thatâ•… why he.comes ‘Why do you think he is coming?’ b’. pEkkiâ•… ti kriði ka vEni why you think that he.comes ‘Why do you think he is coming?’

In turn, the verb can invert with the subject in sentences embedded under the complementizer, yielding instances of embedded V2. In particular, among the (present-day) Romance languages, many Romansch and Ladin varieties, like La Pli and Scuol in (14) and (15) respectively, exhibit V2 in embedded sentences, hence under a che-type complementizer. On the standard assumption that V2 depends on the positioning of the verb in C, this means that the left periphery of sentences embedded under the complementizer, i.e. its C field, can host a verb as well. (14)â•… La Pli de Mareo (Alto Adige/South Tyrol) a.â•… i te diZi ke le liber a-i lit I you tell that the book have-I read ‘I am telling you that I have read the book’ b. i te diZi â•… ke endomaN/ gonot/ magari ve-el I you â•… tell thatâ•… tomorrow/ often/ may be comes-he ‘I am telling you that he is coming tomorrow/ often/ perhaps’ (15)

Scuol (Grisons) i m an dit tSa frs drmaS-t/ drm-al they me have told that perhaps sleep-you/ sleeps-he ‘They told me that perhaps you are/ he is sleeping’

The data that have been presented are merely compatible with the structure in (10) and do not in themselves provide any argument in its favour. An adequate analysis of the same data is available within the articulated theory of the C field proposed by Rizzi (1997, 2001, 2004). In particular, the distribution of Italian che and other Romance che-like complementizers in (11)–(15)

Structure of the complementizer phraseâ•… 21 corresponds to that predicted by Rizzi (1997) for the highest complementizer.4 The embedded V2 facts follow, as already proposed by Schwartz and Vikner (1996), if the verb occupies a different C position (in fact the lowest C position for Rizzi (1997)).5 By contrast, the existence of embedded V2 is problematic in frameworks like Chomsky’s (1995), where only one C position is available. In such a framework, one possibility is to derive embedded V2 from the positioning of the verb in I (Santorini 1989). The clitic nature of the inverted subject in (14)–(15) tends to discount this possibility, on the assumption that subject clitics are associated with a high inflectional domain€– hence if the verb were in I, subject clitics should precede, rather than follow, the verb. The only alternative that we can see to Rizzi’s (1997) articulated view of the C field is precisely the one we are advocating here:€namely, treating the complementizer as the head of an independent noun-phrase-like projection. The discussion so far only addresses the conditions that make structures like (14)–(15) possible; a different question altogether is what makes them necessary. According to McCloskey (2004) there is a particular interpretive value associated with embedded V2, which is revealed by the selectional restrictions discussed in the literature on Germanic languages. Thus, the highest layer of an embedded V2 structure€– i.e. the one filled by the complementizer€– realizes illocutionary force, and allows for inversion of the verb in the lower layer. In languages/contexts that do not allow for embedded V2, a single layer is present, hosting the complementizer and excluding V2; this does not realize illocutionary force. We are not aware of any data on Romance embedded V2 that would allow us to decide whether it obeys semantic restrictions. But suppose it does. The general schema of explanation proposed by McCloskey (2004) need not be tied to the double CP structure that he adopts; rather, it can be implemented, as far as we can see, by the structure that is at the heart of the present proposal. Thus, properties of the matrix sentence select a particular set of properties on the embedded complementizer€– which in turn select for V2 in the embedded sentence. Languages/contexts without embedded V2 simply do not select for the relevant properties on the complementizer head, and so the V2 position of the verb is not selected in the embedded sentence.6 While in (12)–(13) we have seen some cases where the che-type complementizer precedes a wh– phrase, there are many Romance varieties in which the che-type complementizer follows a wh–phrase in both main and embedded interrogatives. This pattern is quite robust in Northern Italian varieties; in (16) we exemplify just one of them.

22â•… Romance complementizers:€structure & interpretation (16) â•… Castellazzo Bormida (Piedmont) a.â•… kwan k i madZi when that you.pl eat ‘When do you eat?’ b. dallwO k i d4wOmi where that you.pl sleep ‘Where do you sleep?’

Now, precisely because the complementizer is the head of an independent constituent, a left periphery can be postulated for it as well. In this schema of explanation, therefore, the wh–phrase that precedes che belongs to the left periphery of the complementizer itself, roughly as in (17). Note that the wh–phrase dallw ‘where’ is simply categorized according to its relation to the predicate, as Loc(ative); we will return to the theory of left periphery elements that this labelling implies. (17) Castellazzo Bormida Loc dallw

c C I k C D i I drw mi c

The analysis in (17) predicts that the entire set of focus and topic elements should be able to appear to the left of the complementizer. Examples like those in (18)–(19) show that the left periphery of the complementizer can host not only wh–elements (presumably foci) but also topicalized material, including both adverbs and the lexical subject, doubled by the subject clitic following the complementizer.

Structure of the complementizer phraseâ•… 23 (18)

Castellazzo Bormida marjo dal’lw k u dRwm Mario where that he sleeps ‘Where does Mario sleep?’

(19)

Cerano (Piedmont) lo:/ adme� inda k i ve� they/ tomorrow where that they go ‘Where are they going (tomorrow)?’

Our structural hypothesis therefore accounts both for data in which the complementizer has topic and focus material to its right and for data in which it has topic and focus material to its left, and thus matches the descriptive power of an articulated C field of the type in Rizzi (1997). At this point the important question arises of whether a left periphery of the type in (16)–(19) is restricted to complementizer-headed noun phrases or whether it can associate with noun phrases in general. It is evident that, to the extent that the structure of noun phrases parallels that of sentences, noun phrases are predicted to host leftÂ�peripheral material exactly as the sentence does. In fact, the left periphery of the noun phrase in (6) hosts the wh–item che, paralleling the left periphery of the sentence in (5). Similarly, the ability of the left periphery of Hungarian DPs to act as an escape hatch for a possessor phrase was the key to Szabolcsi’s (1994) proposal of a parallel structure for noun phrases and sentences. Nevertheless, an asymmetry between sentences/complementizer-headed noun phrases and other noun phrases does in fact exist€– namely, that operators hosted at the left periphery of ordinary noun phrases must bind variables within the noun phrase itself. By contrast, sentences and complementizerheaded phrases can host operators binding long-distance variables. Within the analysis that we are suggesting, this asymmetry can be captured by saying that only propositions and nouns introducing propositional variables€– i.e. complementizers€– can support the relevant set of operators. This distinction is stipulated€– i.e. it is an apparently irreducible primitive of natural languages. But exactly the same is true of Chomsky’s (1973) original proposal of a C(OMP) node providing an ‘escape hatch’ for Subjacency in S(entences) but not in NPs, and its successors up to the present (Chomsky 1995, 2001, 2008). 1.2.1

Combining a left periphery in the complementizer phrase and in the embedded sentence; combining two complementizers Suppose we accept that left-peripheral material occurring after the complementizer is associated with the embedded sentence, while left-peripheral material occurring before the complementizer is associated with the left periphery of

24â•… Romance complementizers:€structure & interpretation the complementizer itself. We then predict that the two left peripheries should combine. The simplest verification of this prediction is that topics can not only precede the cluster of wh–phrase and che, as in (18)–(19), but they can also follow it, as in (20). In present terms, this means that in (20) the wh–phrase occurs in the left periphery of the complementizer, while the topic appears in the left periphery of the embedded sentence.7 (20)â•…

Castellazzo Bormida dallw kə  mat  drm where that the girl she sleeps ‘Where is the girl sleeping?’

Furthermore, a left periphery associated with the complementizer, in the shape of a wh–phrase and/or topics preceding it, can be combined with a left periphery in the embedded sentence in the shape of a verb in C. In other words, it is possible to embed V2 (as revealed by subject–verb inversion) under a sequence of wh–phrase and che complementizer, as in (21)–(22). Note that in the Cantoira example (22b), the complementizer is preceded not only by the wh–phrase, but also by the topicalized subject. (21)â•… Mezzenile (Piedmont) a. əndua k u wnt-i where that they go-they ‘Where are they going?’ b. kaŋ k u viunt-i when that they come-they ‘When are they coming?’ (22) Cantoira (Piedmont) a. ənduə k u dyərt-e whereâ•… thatâ•… theyâ•… sleep-they ‘Where do they sleep?’ b. lu: andua k u vEnt-e they where that they go-they ‘Where are they going?’

Data of the type in (21)–(22) are associated by the present theory with structures of the type in (23), in which the left periphery of the complementizer combines with the left periphery of the sentence embedded under it. The position of the verb, although we keep the conventional C label for it, is now characterized by exclusively verbal properties. Note also that two copies of the subject clitic D are present in (23), on either side of the verb; we assume that while the inverted subject clitic is in the ordinary subject (clitic) position

Structure of the complementizer phraseâ•… 25 immediately above I, the higher copy is in the C domain.8 In keeping with the labelling introduced in connection with (17), the wh–phrase is categorized as Loc(ative) by its intrinsic content€– as is the topicalized subject, notated as D like its clitic counterparts. We shall return to the lack of Top (Focus etc.) labels in section 1.3. (23) Cantoira D lu Loc andua (C) I ke D u C ��nt D e

Poletto (2000), working essentially within Rizzi’s (1997) framework, argues that in sentences like (21)–(22), che is inserted in an intermediate position of the C field. This is distinct from the higher C position (Force for Rizzi (1997)) involved in examples like (11)–(15), and from the lower C position (finiteness for Rizzi (1997)) involved in (16)–(20). Thus, theories of CP recursion can match the descriptive power of the present theory (with a single complementizer position combined with two left peripheries) by postulating three separate complementizer positions. Despite this apparent descriptive equivalence, the two theories are clearly different. Consider the fact that in several varieties interrogatives introduced by a wh–phrase alternate with interrogatives introduced by a wh–phrase and che. In at least some of them, the former have subject clitic inversion€– i.e. V2,

26â•… Romance complementizers:€structure & interpretation as in (24a’ and b’)€– but the latter don’t, as in (24a and b). In many more varieties, for instance Castellazzo Bormida, wh–questions introduced by the wh–phrase and che without inversion, as in (17), alternate with yes–no questions with inversion, as in (25). Both facts lead to the conclusion that in the relevant languages the cluster of wh–phrase and che is in complementary distribution with V2, which is otherwise required by interrogatives. (24)â•… Viguzzolo (Piedmont) a. indE k u drmæ where that he sleeps ‘Where does he sleep?’ a’. ind u drmæ-l where he sleeps-he ‘Where does he sleep?’ b. kwænt k u drmæ how.much that he sleeps ‘How much does he sleep?’ b’. kwand u drmæ-l when he sleeps-he ‘When does he sleep?’ (25)

Castellazzo Bormida 4 d4wm-4 she sleep-she ‘Does she sleep?’

A theory such as Poletto (2000) predicts the data in (24) by assuming that the che complementizer sits in the lowest C position; if so, verb movement will not be able to target that position€ – nor, according to Poletto (2000), will it be able to target any higher position because of minimality. However, under the schema of explanation adopted here, the wh–phrase is in the left periphery of the complementizer che and the verb is in the C position of the embedded sentence, exactly as discussed for (23). Hence the two positions are different, and neither interferes with the movement paths of the other, so the fact that they cannot both be filled in (24) must be explained on other grounds. In present terms, in varieties like (24), when the wh–phrase is introduced in the left periphery of the sentence, interrogative modality is lexicalized by the verb in C, as in (24a’ and b’). By contrast, if the wh–phrase is introduced in the left periphery of the complementizer, V2 is no longer necessary and is in fact excluded, as in (24a and b); we assume that this is so because of a selectional constraint, whereby the complementizer selects the declarative

Structure of the complementizer phraseâ•… 27 modality (i.e. the V in I rather than in C) in the embedded sentence. In the (rarer) languages of the type in (23), interrogative modality is lexicalized by the verb in C even when the sentence is embedded under the complementizer. There is no need, and indeed no evidence, for a different categorization of the complementizer which ultimately yields its different position, as in Poletto (2000). Let us mention also that, along with languages which form interrogatives with a wh–phrase and che (no V2), as in (16)–(20), languages which form them with a wh–phrase and V2 (no che), as in (24a’ and b’), and languages that form them with both che and V2, as in (21)–(23), there are languages that form them with neither. Thus, in (26) the wh–phrase introduces an interrogative sentence alone, i.e. without che-type complementizer, while at the same time the verb does not realize interrogative modality in C, but remains in I, as can be seen by the lack of inversion with the subject (clitic). (26)

Filattiera (Tuscany) ke kamiza t yi what shirt you want ‘Which shirt do you want?’

Our examples so far involve main sentences. Leaving aside embedded questions introduced by the ‘if’ complementizer (to which we return in chapter 2), in embedded wh–questions V2 is generally not found. This is the case in languages in which no inversion is found in main wh–questions, as in the Castellazzo Bormida example in (16’), in languages that do have inversion, as in the Cantoira example in (22’), and also in the Viguzzolo example in (24’) in the absence of a complementizer. In other words, V2 is generally limited to root contexts. We conclude that, contrary to the Romansh/Ladin varieties with embedded V2 in (14)–(15), matrix predicates in Northern Italian languages do not select for V in C in the embedded sentence€– nor for properties of the che-type complementizer that in turn select for embedded V2. (16’)â•…

Castellazzo Bormida di-m ki k i4 vi6N tell-me who that he comes ‘Tell me who is coming’

(22’)

Cantoira di-me ki (k) e vint tell-me whoâ•… thatâ•… heâ•… comes ‘Tell me who comes’

28â•… Romance complementizers:€structure & interpretation (24’)

Viguzzolo a ŋ s2 k a f@ŋ Iâ•… notâ•… knowâ•… whatâ•… theyâ•… do ‘I don’t know what they do’

At the same, there is no structural impossibility of having V2 in embedded questions€ – a fact which is duly reflected by the existence of data like (27) attesting the possibility of the relevant patterns. In this case we assume that the verb indeed lexicalizes interrogative modality in C. It is this interpretive property that discriminates between the pattern in (16’), (22’) and (24’), and the (much rarer) one in (27). From a strictly syntactic point of view, they are equally possible. (27) â•… Castiglione d’Adda (Lombardy) a. di-m sa tSam-ot tell-me who call-you ‘Tell me who you are calling’ b. di m sa sE -t a dre a fa tell me what are you in the process of doing ‘Tell me what you are doing’

A final set of examples which have been considered in the literature in the context of Rizzi’s (1997) proposals concerning an articulated left periphery involve the lexicalization of left-peripheral material, typically a topic, between two copies of the same che-type complementizer, as in (28). Examples like (28) are predicted within a theory of complementizers as functional heads, on the assumption that more than one complementizer position in the hierarchy is instantiated, along the lines of Paoli (2007). (28)â•… Castellazzo Bormida a. l E mii ke nuiâ•… k a l lavu itâ•… isâ•… betterâ•… thatâ•… we thatâ•… weâ•… itâ•… wash ‘It is better that we wash it’ b. l E mii ke vujautSâ•… k i m la dagi it is better that you.pl that you.pl me it give ‘It is better that you give it to me’

In the structures that we have laid out so far there is no room for two complementizers co-occurring. This is not to say that structures cannot be provided for (28). On the contrary, the relevant examples can be adequately described through recursion of the complementizer phrase. In other words, as shown in (29), the matrix predicate takes as a complement the higher complementizer, which in turn selects the lower complementizer, hosting the topic in its left periphery, and embedding the complement sentence.9

Structure of the complementizer phraseâ•… 29 (29) Castellazzo Bormida I ke D vujautʃ I ke D i

... I dagi

Needless to say, the question is not only whether a structure, say (29), can be assigned to a sentence like (28), but also what forces this structure. McCloskey (2004) entertains two alternatives. The first is that double complementizer structures semantically differ from simple complementizer structures; the other possibility is that the higher layer of complementizer structure ‘exists solely to facilitate the adjunction’ of the embedded topic material (McCloskey 2004:€ fn.€ 30). As it turns out, Paoli (2007) argues that both possibilities are instantiated in Romance. Thus, she finds that the double complementizer structures of Turinese (but ‘not the “mainstream” type spoke in Turin’) and of Ligurian are semantically restricted, in that the lower complementizer must select the subjunctive. By contrast, our data from Castellazzo in (28) display the double complementizer phenomenon both with the indicative in (28a) and the subjunctive in (28b). In this respect they seem entirely comparable to the data that Paoli (2007) quotes from Medieval Romance (Tuscan and other varieties). Since the evidence we have does not point to a semantic distinction between double and simple complementizer structures, they must be distinguished on structural grounds. Specifically, based on the evidence that we have, we conclude that matrix predicates in Castellazzo can immediately embed complementizers, as in (29), or wh–phrases as in (16’), but not topic/focus material. This state of affairs can be captured in terms of a selectional constraint imposed by the matrix verb on any complementizer phrase it embeds€– roughly to the effect that it can contain only clause-typing material (in the sense of Cheng (1991)), i.e. only declarative che or the interrogative cluster of wh–phrase and

30â•… Romance complementizers:€structure & interpretation che. Two alternatives are then open for the insertion of focus/topic material. First, one can position such material in the left periphery of the embedded sentence, as in (20). Alternatively, it can be positioned in the left periphery of the embedded complementizer€– but then a higher complementizer must be added, satisfying the selectional constraint just defined, as in (29). In this respect, therefore, we reject the view of Paoli (2007:€ 1075), according to whom the lower complementizer ‘is not a subordinating particle, but overt realization of Top°’ in medieval Romance.10 1.2.2 Some potential problems Before we proceed with our discussion, we will try to clear the ground of a number of quite general questions that may be raised against our approach. Perhaps the most basic such question involves selection of an embedded complement by a matrix predicate. If a complementizer head in Romance is nominal and its projection is a noun phrase, how can we state the distinction between selecting for a complementizer phrase, i.e. a sentence, and selecting for an ordinary noun phrase? The answer is that in terms of the interpretive categories adopted here, complementizers (like bare sentences) correspond to propositions, while conventional noun phrases correspond to individual terms. Therefore selection can adequately be stated on the basis of interpretive categories.11 Selection may even provide an argument in favour of the present articulation of the left periphery, as opposed to the cartographic one. Suppose that a higher predicate embeds a sentence whose leftmost and highest element is, say, a topic, as would be the case, for instance, in (18) and (19). Under Rizzi’s (1997) theory the predicate effectively embeds a topic phrase€– which, according to Newmeyer (2005), provides no clear grounds for selecting the interrogative force in lower functional projections. Whether this turns out to be a serious problem or not, it does not arise in the theory that we are building. A verb selecting for a complementizer-headed noun phrase selects for the properties of the complementizer head. Alternatively, if the higher predicate selects a bare sentence€– i.e. one without a complementizer€– it selects for properties of the embedded verbal head. Another property that has been consistently used to explain differences in distribution between noun phrases and sentences is Case. In the present framework, Case cannot be used to this end precisely because complementizers, which introduce sentences, are nominal€– and should therefore have the same Case properties as ordinary nouns. The evidence with which Stowell (1981) introduces his Case Resistance Principle concerns the fact that noun phrases, including gerunds, can be the object of a preposition, but sentences cannot, as in (30a) vs. (30b).

Structure of the complementizer phraseâ•… 31 (30) a. We talked about the Marines going to Iraq b. *We talked about that the Marines went to Iraq

Now, elements such as before, after and without would seem to be prepositions since they can select noun phrases. But their Italian counterparts also embed sentences introduced by che, as shown in (31). Hence there appears to be no general selectional constraint against sentences as objects of prepositions or against complementizers in such a position. To the extent that more specific constraints are real, they can be stated again in terms of the interpretive notions of individual vs. propositional variable. (31) a. b.

Sono arrivato prima/dopo che I.am arrived before/after that ‘I arrived before/after you left’ Me ne sono andato senza me away am gone without ‘I went without you noticing it’

sei partito you.are left che te ne accorgessi that you of.it noticed

A different kind of question has to do with the fact that theories in which a sentential complement is contained within a nominal layer have been proposed before€– but typically differentiate between various types of sentential complements. Specifically, Kiparsky and Kiparsky (1970) differentiate complements of factive verbs, which are endowed with a nominal layer, from complements of non-factive verbs, which do not have such a layer. Crucially, different syntactic behaviours are argued to follow from the different structures. Thus, it is suggested that factive verbs are islands for extractions, since their structure makes them into complex NPs. Our judgement for Italian is that there really is no difference between extraction from a non-factive context like (32a) and extraction from a factive one like (32b)€– and even extraction from ‘the fact that’ in (32c) is not severely degraded. In contrast, there is a strong contrast with extraction from a relative clause, as in (32d). (32) a. b. c.

Chi pensi che non abbiamo visto? Who you.think that not we.have seen ‘Who do you think that we didn’t see?’ Chi ti dispiace che non possiamo vedere? who you regrets that not we.can see ‘Who do you regret that we can’t see?’ Chi ti dispiace il fatto che non possiamo vedere? who you regrets the fact that not we.can see ‘Who do you regret the fact that we can’t see?’

32â•… Romance complementizers:€structure & interpretation d. d’.

*Chi ti dispiace il motivo perâ•… cui non possiamo vedere who you regrets the motiveâ•… for whichâ•… not we.can see ‘Who do you regret the reason why we can’t see?’ Chiâ•… ti chiediâ•… perchèâ•… nonâ•… abbiano potuto vedere? who youâ•… ask why not they.haveâ•… been.ableâ•… to.see ‘Who do you wonder why they haven’t been able to see?’

The fact that (32b) patterns with (32a) may simplify our task with respect to the distinction between factive and non-factive complements, since we may not need to worry about providing different structures for them. But the clear contrast between (32a) and (32d) raises the question of whether they shouldn’t both be blocked as instances of complex NP islands. In fact, this question can be reduced to the one we discussed at the end of section 1.2.1 concerning the admissibility of long-distance operator material in the left periphery of sentences vs. noun phrases. There we concluded that complementizers and sentences (because of their common propositional content) differ from other noun phrases in being able to host such material€ – i.e. in behaving like ‘escape hatches’. If so, we expect that ordinary noun phrases may give rise to a complex NP island effect, whereas complementizers do not. In other words, there is nothing in the present proposal contradicting standard accounts of islands in terms of subjacency/ phases€– though we remain strictly non-committal with respect to them.12 Italian does display some limited sensitivity to factive islands with adjuncts, which can have both matrix and embedded scope in sentences like (33a), while in (33b) embedded scope appears to be quite hard to obtain. This suggests that the adjunct cannot be extracted from the factive sentence€ – though this extraction is possible from the non-factive one. In other words, factives show an asymmetry between arguments, as in (32b) and adjuncts, as in (33b) of the type well known from Cinque (1990), and accounted for since Rizzi (1990) as a Minimality effect. (33)â•… a.â•… b.

Perchèâ•… pensi che siano venuti? why you.think that they.are come ‘Why do you think that they are coming?’ Perchè ti dispiace che siano venuti? why you regrets that they.are come ‘Why do you regret that they came?’

Now, we know that in some languages the factive vs. non-factive distinction involves the choice of different complementizers, one such language being Greek, as studied by Roussou (1994). Therefore it is natural to propose that the embedded complementizer in (33b) has some property (selected by the higher predicate) that the complementizer in (33a) does not have, for instance a

Structure of the complementizer phraseâ•… 33 definiteness property, which, as suggested by Roussou (1994), is consistent with the interpretation of factivity. It is reasonable to think that this property triggers a Minimality effect in (33b) which is absent from (33a). In other words, the present analysis of complementizers need not interfere with current theories of extraction (with respect to which we remain non-committal, as before). Rather, the two issues are orthogonal. Summing up so far, the structure that we propose for complementizers is motivated in section 1.1 on the basis of their nominal nature in Romance, as seen in the fact that they have the same form as wh–elements. Section 1.2.1 shows that this structure is compatible with the distribution of the complementizer with respect to other elements of the so-called left periphery. But note that this latter result depends on the complementizer being the head of its own projection€– and does not depend on it having the same form as the wh–operator or even a nominal nature. In other words, though we argued that the nominal, wh– nature of Romance complementizers requires the structure we propose for them, nothing in the structure we propose requires nominal, let alone wh– properties in the complementizer. Complementizers which do not coincide with wh–elements can easily be seen in Romance systems. A simple example is provided in (34), from a variety from Sardinia which distinguishes the declarative complementizer ki, as in (34a), from kiE ‘who’ in (34b) and kalE ‘which’ in (34d), as well as from ittE ‘what’ in (34c–d). Thus, though the complementizer can be assigned to the k-series to which ‘who’ and ‘which’ belong, it does not coincide with any wh–item. For languages like (34), we of course maintain the same analysis as for Italian che. In fact, given the morphological relatedness of ki to the ksystem of wh–elements, a language like Luras can be described as having a specialized wh–operator for propositional variables. (34)â•… Luras (Sardinia) a.â•… m ana naDuâ•… ki enis kraza to.meâ•… they.haveâ•… told thatâ•… you.comeâ•… tomorrow ‘They told me that you are coming tomorrow’ b. kiE eniDi whoâ•… comes ‘Who is coming?’ c. ittE znâ•… fattEN whatâ•… they.are doing ‘What are they doing?’ d. ittE/ kalE libbru t a llEaDu what/ whichâ•… book to.youâ•… he.hasâ•… brought ‘What/which book did he bring you?’

34â•… Romance complementizers:€structure & interpretation We know from Germanic languages that complementizers can belong to a non-wh– nominal series, namely that of demonstratives, like English that. We maintain the idea that the complementizer introduces a propositional variable for these languages as well. On the other hand, a wh–complementizer of the type generally instantiated in Romance (but also in Adyghe according to Caponigro and Polinsky (2008)) is very naturally construed as a lambda abstractor; the question of how best to construe the demonstrative complementizer of English is left open here. A related open question is whether the different nature of the complementizer has reflexes in the syntax and in the interpretation of Romance vs. Germanic complementation. In chapter 2 we introduce Romance complementizer systems that have no counterpart in Germanic languages known to us, possibly indicating that these are possibilities open to wh–complementation systems but not to demonstrative ones. Further afield, there are languages in which so-called complementizers are verbal, rather than nominal in nature. A case in point is the Buru language of Eastern Indonesia, as discussed by Roberts and Roussou (2003) on the basis of Klamer (2000). In Buru, fen can function as a main verb followed by a quotation, as in (35a), or can combine with another verb of saying introducing direct speech as in (35b) or indirect speech as in (35c); the data are from Klamer (2000). (35)â•… a.â•… b. c.

Nak ana-t feneâ•… “Ng-ina nau daholo” 3sgPossâ•… child-Nomâ•… say 1sgVoc-motherâ•… 1sgPossâ•… bunch-head ‘Her child said, “Mother, the hand (of bananas) at the top of the stalk is mine”’ Da prepaâ•… fenâ•… “Siraâ•… ruaâ•… kaduk” 3sgâ•… speak say 3pl two arrive ‘She said “The two of them came”’ Da prepa fene ringe mata haik 3sg speak say 3sg die Prf ‘He said that he was already dead’

Under the present approach, in all three examples in (35) fen can be treated as the I head of its own projection€– in this case not a noun-phrase-like projection, but a verb phrase-/sentence-like projection. In the absence of other verbal specifications, fen is read as a main verb, as in (35a)€– which we predict to be possible precisely on the basis of the fact that it heads a verb phrase-/sentencelike constituent. On the other hand, the combination of fen with another verb in (35b–c) recalls so-called serial verb constructions. In Manzini and Savoia (2005) we consider serial verbs in connection with constructions in Southern Italian varieties in which aspectual/modal/motion verbs with reduced or absent inflectional properties embed sentential complements, on condition that their

Structure of the complementizer phraseâ•… 35 temporal reference and their EPP (Extended Projection Principle) arguments coincide. We conclude that these constructions involve complex predicate formation; this implies the coincidence of temporal reference and argumental structure€ – and in turn these properties will typically result in a lack of inflection on one of the two verbs. The same general properties seem to hold in (35b–c), in which the so-called complementizer status of fen can then be described in terms of complex predicate formation€– again on the basis of the usual structure in which it heads a verb phrase-/sentence-like projection. Data of the type in (35) are worth bringing up not only for their intrinsic interest€– but also because of an issue that we raised in passing in section 1.1 and then abandoned in subsequent discussion:€namely, that (35) and the like are routinely described in terms of processes of historical change. Thus, according to Klamer (2000:80), ‘we can explain the synchronic distributional restrictions on fen if we assume that historically … the report verb fen(e) has developed an alternative interpretation as a quote marker fen … In contexts where fen is preceded by another verb which reports words, thoughts, or perceptions, it has developed a complementizer interpretation’. To quote just one similar case, for Whitman (2000:€222) in Ewe ‘the categorial feature of bé “say” changes from V to C’. A closer term of comparison with the present theory is provided by Roberts and Roussou’s (2003) discussion of English that€– an element which we have cited more than once as providing evidence for the connection of complementizers with bona fide nominal heads. In a nutshell, when it comes to that ‘we are dealing with one and the same lexical item which can surface as either D or C’ (Roberts and Roussou 2003:€115). This analysis is compared to Davidson’s (1997 [1968]:€828–9) idea that ‘sentences in indirect discourse, as it happens, wear their logical form on their sleeves … They consist of an expression referring to a speaker, the two place predicate “said”, and a demonstrative referring to an utterance’. According to Roberts and Roussou (2003:€113–14), in Davidson’s analysis complementizer that ‘is actually the demonstrative’; by contrast, ‘it is possible to argue that that in terms of its position in the sentence has been grammaticalized as a C element’. In present terms, the explanation for the range of interpretations of, say, Italian che or English that has to do with conditions internal to a single grammatical competence system; in other words, there is no necessity for invoking several competence systems in a relation of historical change to one another. Specifically, no grammaticalization is implied, understood roughly as reanalysis from lexical to functional (Roberts and Roussou 2003). In this respect the present analysis is closer to the syntactically naive one of Davidson (1997

36â•… Romance complementizers:€structure & interpretation [1968]). The gist of our proposal is that Italian che and its Romance counterparts are nominal quantificational heads in all cases€ – with a syntax typical of nominal heads, i.e. that of serving as arguments of predicates, and with a semantic content typical of quantificational heads, i.e. that of introducing variables. Similarly, we surmise that both demonstrative that and complementizer that are nominal heads and arguments of verbs; the only difference is that complementizer that has a propositional restrictor and demonstrative that a nominal one. Before proceeding, we will consider just one more issue. Nothing that we have said so far leads us to expect that complementizers appear only in embedded contexts. In fact we have already seen a complementizer occurring in matrix sentences:€ namely, the che-type element following the wh–phrase in (16)–(24).13 The che complementizer also introduces yes–no matrix questions in a variety like Florence in (36), though in the same variety wh–questions are introduced simply by a wh–phrase (as in standard Italian). We surmise that the insertion of ke in yes–no questions corresponds to the presence of a focalization bearing on the main verb, paralleling the focalization on the wh–constituent in wh–questions, but consistent with the yes–no interpretation. If so, we can maintain that the yes–no complementizer proper (i.e. what we will call the polarity complementizer in chapter 2) is the se ‘if’ element that occurs in embedded yes–no questions (in Florence as in standard Italian). (36)

Firenze (Tuscany) ke lla viEne la maria that she comes the Mary ‘Is Mary coming?’

Finally, matrix instantiations of the complementizer are also possible in standard Italian, subject to a modal split, since matrix subjunctives can be introduced by the complementizer, as in (37), while matrix indicatives are not. In other Romance varieties, the che-type complementizer can introduce ordinary indicative sentences, as reported in the literature for Provençal (Ronjat 1937:€536 ff.) and Guascon (Rohlfs 1977:€205); an example from Sardinian is provided in (38). (37) Che entrino that they.enter ‘Let them enter’ (38)

Àllai (Sardinia) ka dZai ɖɖ a ffattu that alreadyâ•… it he.hasâ•… done ‘He has already done it’

The left periphery beyond complementizersâ•… 37 1.3

The left periphery beyond complementizers

In the discussion of distributional evidence in section 1.2, we argued that this evidence is also accounted for by our analysis when tested against the alternative view of complementizers as functional categories of the verb. The argument in favour of our hypothesis does not have to do with empirical coverage per se, but rather with the way in which this empirical coverage is achieved. In our view, there are features of the present proposal that make it more conceptually perspicuous (more explanatory) than available alternatives. In particular, the cartographic model, which aims at an empirical coverage comparable to ours, is associated with a trend towards finer-grained phrase structure. In a sense, such an approach is simply what the empirical evidence requires€– and in this sense we share it. At the same time, in the specific implementation of the cartographic programme presented, say, by Rizzi (1997, 2001, 2004), the fine articulation of syntactic structure derives from the postulation of specialized hierarchies of functional categories. The model of the left periphery that we have defined here is clearly different. Thus, where Rizzi (1997) has a hierarchy of functional projections for complementizers, we have a single complementizer position€– and this position is not functional. Rather, on the evidence of its morpholexical form as well as its interpretation, the complementizer is the head of a noun phrase that selects the embedded sentence. In other words, a certain amount of the articulation of the so-called left periphery is achieved through the recursion of predicate–argument structures€– the superordinate verb selecting the complementizer noun phrase as its argument, and this in turn selecting the embedded sentence as its argument. Critics of cartography note that functional hierarchies are potentially unrestricted devices, since a new position or set of positions in the hierarchy can always be introduced to meet new empirical evidence. The present approach exempts at least complementizer structures from this potential problem. On the other hand, the present chapter so far only deals with the complementizer itself. Nothing that we have said touches on the independent issue of whether there is a single (conventionally C) position to which the verb can move or more than one. In previous work we have sided in favour of more than one verb position above I€– based notably on the comparison of finite verbs in questions and V2 with imperatives and infinitives. Evidence concerning their distribution with respect to clitics (enclisis vs. proclisis), negation and other material leads us to the conclusion that at least two different (conventionally C) positions of the verb are involved. In adopting multiple head positions above I, we therefore follow Rizzi (1997) and the related cartographic literature.

38â•… Romance complementizers:€structure & interpretation At the same time, in the model of Rizzi (1997, 2001, 2004), the possible head positions above I are also used to order topic, focus and other phrasal material, for it is one of the postulates of standard cartography, based on Kayne’s (1994) anti-symmetry, that there is a single Spec position for each head; therefore a head hierarchy automatically defines a hierarchy of phrasal material€ – and in fact hierarchies of phrasal material are stated as hierarchical orderings of the heads to which they attach (cf. Cinque 1999 on adverbs). The issue of the ordering of phrasal material in the left periphery is logically independent of the status of complementizers. It is evident, however, that the present reanalysis of complementizers leads us to expect that a considerable amount of currently postulated functional structure may be dispensed with in that respect as well. In this respect, potentially more promising views are held by Chomsky (1995, 2000a, 2008), according to whom each head supports any number of Specs (as for Brody (2003)), while topic, focus and similar notions correspond to interface interpretations€– not to features (or categories) entering syntactic computations. 1.3.1 Is order dictated by interpretation€– or interpretation by order? In what follows, we concentrate on the respective position of wh–phrases and of lexical subjects, specifically in Northern Italian varieties with subject clitics. The topic(-like) nature of the lexical subject in Romance languages is supported precisely by their position in questions. Thus, in subject-clitic languages the subject clitic follows finite verbs in questions, including the auxiliary, as in (39). This corresponds to the position of the lexical subject in Germanic languages, and can be analysed accordingly, as a result of the clitic remaining in the same position as in declarative sentences€– while the verb alternates between I in declaratives and C in questions. By contrast, lexical subjects in Romance do not appear between the auxiliary and the participle, but only after the participle or before the auxiliary, i.e. in the right or left periphery of the sentence, as in (39) again. (39)â•… Oviglio (Piedmont) a.â•… E -l amni marju is he come Mario ‘Has Mario come?’ b. marju E -l amni Mario is he come ‘Has Mario come?’

When wh–phrases are brought into the picture, it remains true that lexical subjects are generally positioned in the right or left periphery of the sentence;

The left periphery beyond complementizersâ•… 39 however, in this latter case they are found before the wh–phrase, as in (40a), and not between the wh–phrase and the verb, as in (40b). The examples in (40c-d) illustrate the parallelism of other topics with lexical subject in this respect. Both are normally excluded from the position between the wh–phrase and the verb in C. (40)â•… Modena (Emilia) a. lo ki tSam -el he who calls he ‘Who does he call?’ b. *ki lo tSam -el whoâ•… he calls he c. la torta indo l E -t the cake where it have you ‘Where have you put the cake?’ d. *indo la torta l E -t where the cake it have you

mesa put mesa put

According to Rizzi (1997:€ 299), the positioning of the lexical subject is determined essentially by the Wh–Criterion of Rizzi (1996), whose satisfaction requires the verb and the wh–phrase to be in a head-Spec configuration in C(Focus)P, which forces their adjacency and hence the impossibility of a lexical subject (or other material) intervening between them. At the same time, even for Rizzi (1996:€87), the unacceptability of a sentence like (41a) in Italian contrasts with the grammaticality of (41b), in which the lexical subject intervenes between the wh–phrase perchè ‘why’ and the verb. (41)â•… a.â•… b.

*Dove Gianni è andato? where G. is gone ‘Where has Gianni gone?’ Perché Gianni è partito? why G. is left ‘Why has Gianni left?’

Working in a model with a single C position, Rizzi (1996) proposes that in sentences like (41b), it is the wh–phrase perchè that occupies this position, so that the verb is in the ordinary I position and the subject precedes it. However, in Northern Italian varieties the position of the lexical subject between ‘why’ and the verb combines with inversion of the subject clitic and the verb, as in (42), indicating that the verb is in C. (42)â•…

Modena perkE al putEin e l parti why the child is he left ‘Why has the child left?’

40â•… Romance complementizers:€structure & interpretation In fact, it is not just ‘why’ that allows for the lexical subject (or other topical material) to appear between the wh–phrase and the verb, as Rizzi (1996:€87) also remarks. In his terms, the obligatoriness of verb movement to C in interrogatives is weakened when a D-linked wh–phrase is involved, for reasons that remain unclear, so that the verb is (again) in I. Data from Northern Italian varieties confirm Rizzi’s (1996) intuition about the possibility of positioning the subject after a D-linked wh–phrase. At the same time, this can combine with inversion of the subject clitic after the verb, as for instance in (43), leading us to conclude that the verb is in C. Hence the phenomenon is not connected with the verb remaining in I. (43)â•… Corte (Veneto) a. kEl de kis marjo ljezara-lo which of these Mario will.read he ‘Which of these will Mario read?’ b. *ula marjo va-lo where Mario goes he ‘Where does Mario go?’

Incidentally, note that if we combine the previous generalizations about the relative order of wh–phrases and lexical subjects/topics with the proposal that the complementizer is a specialized nominal head with its own left periphery, we obtain a straightforward prediction concerning interrogatives introduced by wh–phrase and complementizer. Namely, we predict that we will find not only the order in (23), in which the topic precedes the wh–phrase and complementizer sequence, but also the order in which the topic appears after the wh–phrase and hence before the complementizer. This prediction is confirmed by data like (44a); similar evidence is noted by Poletto and Vanelli (1995:€153) for a variety from the Turin area. Crucially, the order in (44a) appears to be restricted by the same factors that we considered for questions introduced by a simple wh–phrase€– so that a D-linked wh–phrase favours the order in (44a), while a non D-linked wh–phrase tends to exclude it, as in (44b, b1). (44)â•… S.Maria Maggiore (Piedmont) a. kwal ad kwi guit ul dZua k u tSama which of those children the John that he calls ‘Which of those children does John call?’ b. ki k a tSama lu:r who that they call they ‘Who do they call?’ b’. *ki lu:r k a tSama who they that they call

The left periphery beyond complementizersâ•… 41 Proceeding now with the theoretical discussion, a new take on the ‘why’ problem is offered by Rizzi (2001). In addition to the three C positions (finiteness, Focus and Force, cf. fn. 3–5) proposed by Rizzi (1997), Rizzi (2001) introduces a further C position, namely Int(errogative), located between the highest C(Force) position and the C(Focus) one. This C(Interrogative) position is meant to host elements like ‘why’ as well as interrogative ‘if’, se in Italian. Crucially, in Rizzi’s (1997) model, Top phrases can be freely interleaved between any C projections€– therefore topicalized material is predicted to occur between ‘why’ in C(Interrogative) and the verb, even if the latter is in C(Focus). Note that an eventual unification of ‘why’ with D-linked wh–phrases requires the latter to be moved to (or inserted in) Spec of C(Interrogative) as well. Abstracting away from the theoretical postulates of the cartographic approach€– essentially the idea that all phrasal and head movement must be supported by the presence of a matching functional head€– the empirical generalization seems to be that wh–phrases can move into two different positions, namely a slightly higher one that can precede a topic and a lower one that cannot. The question is whether the codification of this generalization in terms of functional structure yields predictions or insights that are otherwise unavailable. Incidentally, the labels ‘interrogative’ and ‘focus’ proposed by Rizzi (2001) for the relevant positions, while clearly interpretive in nature, are not sufficient to characterize the interpretation in full, for there is obviously no sense in which the wh–phrase is a Focus in one case but not in the other (or interrogative in one case but not in the other). As already mentioned, the alternative is essentially the one suggested by Chomsky (1995, 2000a, 2008), namely that Topic, Focus and the like simply name interpretations of left-peripheral material. Suppose that the left periphery considered in (40)–(43) is freely ordered above the verb€– in the absence of any functional hierarchy (of categories or features) forcing it. If so, sentences of the type in (41a), which represent the crucial case of ill-formedness, cannot be excluded on formal, computational grounds; on the contrary, their illformedness must be interpretive in nature. It seems to us that a potential argument in favour of this theoretical stance comes from the fact that the unacceptability of sentences like (41a) is far from a matter of absolute judgement in particular, there is no connection between the intrinsic lexical shape of the wh–phrase and the range of positions that the lexical subject or other topic material can take with respect to it. Thus, Benincà (2001) notes the acceptability of sentences like (45), in which the wh–phrase does not appear to be D-linked. Benincà (2001) further connects the well-formedness of (45) to a ‘rhetorical question’ interpretation, implying

42â•… Romance complementizers:€structure & interpretation ‘a negative answer’. We are not sure that this is the correct characterization of rhetorical questions, which seem to correspond more generally to questions that presuppose a certain answer, not necessarily a negative one (‘Who wants more money? Everybody does!’). Even so, it seems to us that (45) need not be interpreted as a rhetorical question at all€– rather, it can be a genuine request for information. (45)â•… Chi la fisica la sa bene? Who the physics it knows well ‘Who knows physics well?’

Independently of what its correct characterization may be, we may agree that the interpretation of (45) differs from that of (41a). This leads us to predict that in the out of the blue context implied by (41a), the utterance in (45) is equally ill-formed, while (41a) becomes grammatical if the range of interpretations that make (45) grammatical is forced. In a theory where the relative position of wh–phrases and topics/lexical subjects is governed by functional hierarchies of categories/features, we must assume that the very same wh–phrases can be associated with one or the other of these functional specifications and can be placed according to them. The other way of thinking about the relevant data is that the computational component allows for any positioning of the relevant elements (in the left periphery); however, crucially, different orderings yield different interpretations. To the extent that both views can be used to characterize contrasts like those considered here, they are notational variants. However, one of them is arguably simpler€– in more than one respect. To begin with, notions such as topic and focus are encoded only once (as interpretations) in the theory we are upholding here€– while they are (redundantly) encoded twice (as interpretations and as grammatical properties) in the alternative theory. This simplicity argument has a counterpart when it comes to the single lexical entries. Assuming the maximally restrictive Inclusiveness principle of Chomsky (1995), according to which only intrinsic properties of lexical items (i.e. properties associated with that item in the lexicon) enter syntactic computation, we would have to admit that topic, focus and similar properties can be (optionally) associated with any nominal head. The alternative is to consider that these notions, like other notions that are relational in nature, do not correspond to features/categories at all, but rather to configurations (cf. Chomsky 2000a on theta-roles)€ – defined in this case at the LF interface. These simplicity considerations, in the absence of empirical evidence, prove decisive in our view.14

The left periphery beyond complementizersâ•… 43 1.3.2 Embedded contexts In the preceding section, we considered the relative order of wh–phrases and lexical subjects in matrix interrogatives. A further element of complexity is introduced by embedded interrogatives. Here the judgement of Rizzi (1997:€289) is that in Italian the topic can precede the wh–phrase, as in (46a), while the configuration in (46b), in which the topic follows the wh–phrase, is ‘slightly marginal’. For us, both examples are equally well-formed. In fact, Rizzi’s (1997) structural schemas allow the topic to either precede or follow the wh–phrase, assuming its position to be the same in (46a) and in (46b), namely C(Focus). (46)â•… a.â•… b.

Mi domando il premio Nobel a chi lo potrebbero dare myself I.ask the prize Nobel to whom it they.could give ‘I wonder to whom they could give the Nobel prize’ Mi domando a chi il premio Nobel lo potrebbero dare myself I.ask to whom the prize Nobel it they.could give

Recall from the previous section that for Rizzi (1997), movement of the verb to C(Focus) normally excludes the wh–phrase–topic order in matrix questions. Therefore he automatically predicts that the latter resurfaces in embedded questions, as in (46b), because of the absence of verb movement. Data from Northern Italian varieties, however, call his explanation into question. In a null-subject language like Italian, it is hard to detect the position of the verb in the sentence on independent grounds, but in Northern Italian varieties the position of the subject clitic provides a reliable independent test. In many of these languages, inversion of the verb with the subject clitic does not take place in either matrix or embedded questions, providing evidence that the verb maintains its I position in all cases. Yet in the same languages, lexical subjects and other topics may appear in either right-peripheral or left-peripheral positions in matrix interrogatives, while the occurrence of the topic or lexical subject between the wh–phrase and the verb is not attested. This contrasts with embedded questions, in which the lexical subject/topic can either precede or follow the wh–phrase, as shown in (47). (47)â•… Fontanigordaâ•… (Liguria) a.â•… k2lu li duve u druome that.oneâ•… thereâ•… whereâ•… he sleeps ‘Where does he sleep?’ b. ne suo k2lu li duve u druome not I.know that.one there where he sleeps ‘I don’t know where he sleeps’

44â•… Romance complementizers:€structure & interpretation b’.â•… ne suo duve k2lu li u druome not I.know where that.one there he sleeps ‘I don’t know where he sleeps’

In other words, the contrast between matrix and embedded questions in a language without subject-clitic inversion is exactly the same as the one observed in a language with subject-clitic inversion, as in (48) vs. (40). (48)â•… Modena a. a n sO briza to fradEl indo al I not know not your brother where he ‘I don’t know where your brother goes’ b. a n sO briza ki lo l a I not know not who he him has ‘I don’t know who he has called’

va goes tSamE called

The conclusion that the ordering of the lexical subject/topic with respect to the wh–phrase in matrix questions is not determined by the position of the verb with respect to the wh–phrase is supported by at least one independent argument, from subject-clitic languages with interrogative inversion. In several relevant Northern Italian varieties, the subject clitic is not simply inverted after the verb but doubled before and after the verb. The relevant examples are (23) for questions introduced by a combination of wh–phrase and complementizer and (24a’, b’) for questions introduced by a simple wh–phrase. Therefore the supposed Spec-head adjacency of the wh–phrase and the verb is systematically disrupted by the preverbal subject clitic. In terms of the structure already provided in (23), in (24a’, b’) a D position intervenes between the wh–phrase in the left periphery of the sentence and the verb in C. If one wanted to claim that the clitic represented adjoined material, one would at the very least have to clarify how the double adjunction (of a proclitic and an enclitic) comes about. In short, for the various reasons reviewed, the contrasts relating to the positioning of the lexical subject in matrix and embedded questions cannot be due to the Wh–Criterion (or its variants). In other words, the relative ordering of wh–phrases and topics/lexical subjects seems to be determined by their intrinsic properties, without the position of the verb playing any role. If so, then it is no longer clear that the general acceptability of wh–phrases in front of lexical subjects/topics in embedded questions, and their acceptability in certain matrix questions (introduced by ‘why’, D-linked wh–phrases etc.) should not be accounted for in the same way. Answering the question of what would unify these various contexts is beyond the scope of the present chapter, which aims simply at establishing a

The left periphery beyond complementizersâ•… 45 theory of complementizers€– including of course its compatibility with a more general model of the left periphery. Nevertheless, the solution envisaged by Manzini and Savoia (2005) is worth mentioning, since it represents an extension of sorts of the approach to complementizers themselves. The idea is that embedded interrogatives, in which the wh–phrase can precede topic material, are effectively (free) relatives, of which the wh–phrase represents the nominal head. In other words, just as the complementizer is not in the left periphery of the sentence but rather projects its own noun phrase, the wh–phrase can project its own noun phrase, which embeds a sentence, eventually including topical material. This proposal can equally be applied to matrix questions, as Manzini and Savoia (2005) effectively do for matrix questions introduced by ‘why’. In other words, ‘why’ is not in the left periphery of interrogative questions; rather, it systematically introduces them as a sort of propositional operator (complementizer). In any event, we may assume that ‘why’ projects a nominal constituent€– which could then be taken to be more akin to a (free) relative. The latter proposal could further be extended to D-linked wh–phrases, and more generally to all wh–phrases that embed a topic. Other extant proposals in the literature seem to rest on intuitions compatible with the analysis just sketched. We have already mentioned Rizzi’s (1997) treatment of ‘why’ as a complementizer head. Kayne and Pollock’s (2001) discussion of pourquoi ‘why’ in French is also relevant. They consider the fact that ‘why’ and other wh–phrases like en quel sens ‘in which sense’ do not trigger inversion of the verb with the subject clitic in French matrix questions, in contrast to other wh–phrases. What they suggest is that questions like en quel sens les fleurs parlent ‘in which sense flowers speak?’ ‘include an abstract verb corresponding to say’, as if one were to say ‘In which sense are you saying that flowers speak?’. This proposal shares with ours the intuition that ‘why’ and other items of the same class belong to a different sentential(-like) constituent with respect to the sentence they introduce. At the same time, for reasons of restrictiveness of the theory, but also for strictly empirical reasons, we reject what Kayne (2006, 2008a) calls silent categories (Manzini and Savoia 2008a, 2009a, 2010; Savoia and Manzini 2010), including the ‘abstract verb’ of Kayne and Pollock (2001). In our intuition, the right interpretation for a sentence like (45), with the order wh–phrase–topic, is more likely to be simply ‘who (is it that) knows physics?’. Similarly, the interpretation of en quel sens les fleurs parlent? could be reconstructed as ‘in which sense (is it that) flowers speak?’. In other words, in both cases the embedded sentence is predicated of the wh–phrase€– more or less as we expect for (free) relatives, or clefts.15

46â•… Romance complementizers:€structure & interpretation For the sake of completeness, we note that in the discussion that precedes, the position of lexical subjects is assimilated to that of topics. However, this does not necessarily hold in all languages (witness the case of English), nor even in all Romance languages. Thus, in French, the lexical subject appears between the wh–phrase and the verb in so-called complex inversion examples, as in (49), independently of interpretive factors. The fact that the verb is inverted with respect to the subject clitic means that it is in a relatively high (conventionally C) position, so that the lexical subject must be in a higher position still€– yet not a topic position. The present analysis already provides a non-topic position for subjects above C, namely the position taken by the preverbal subject clitic in the structure in (23). The ability of the lexical subject in French to appear in a position which in Northern Italian varieties is only available for subject clitics correlates with the fact that while lexical subjects are obligatorily doubled by subject clitics in Northern Italian varieties, they lexicalize the subject alone in French€– and determine complementary distribution with subject clitics. (49) Quand Jean vient-il? When J. comes-he ‘When is John coming?’

At the same time, French, like Northern Italian varieties, reserves the lowest subject position, internal to the I domain, for clitics; therefore only subject clitics can be found inverted after the verb in questions. This contrasts with a language like English, in which the lexical subject ordinarily appears after the verb in questions€– and hence presumably occupies the same position as subject clitics in Romance languages. The English-type position of the subject can also be seen in some Romance languages, namely Romansh varieties in which the lexical subject can appear between the auxiliary and the participle, as in (50).16 (50)â•… Trunâ•… (Grisons) a. ain ilts taLO:ʀ kuʀdai pEʀ tiaʀa are the dishes fallen to ground ‘Have the dishes fallen to the ground?’ b. an iLts afOnts duʀmiu have the children slept ‘Have the children slept?’

There is a final set of data from the work of Rizzi (1997) that goes potentially unpredicted by the present approach. According to the data presented there, a topic can precede a wh–phrase not only in matrix questions, but also in embedded questions. However, a che-type complementizer cannot be preceded by a topic. The relevant data for embedded questions are of the type in (46);

Conclusionsâ•… 47 by contrast, Rizzi (1997) judges (51) and the like to be ill-formed. Benincà (2001), however, does not exclude topics in front of declarative complementizers altogether, since according to her an example like (51b) is well-formed. However, (51b) involves a so-called ‘hanging topic’; for Benincà (2001), Â�so-called clitic left dislocation, as exemplified in (51c), remains ill-formed. For us, the two sentences in (51b–c) have the same status; i.e. they are both acceptable. The same holds for a sentence like (51a), which is ambiguous between the descriptive categories of hanging topic and clitic left dislocation. (51)â•… a.â•… b. c.

Credo il tuo libro che loro lo apprezzerebbero molto I-believe your book that they it would.appreciate a.lot ‘I believe that they would appreciate your book a lot’ Sono certa questo libro che non ne ha mai parlato nessuno I.am certain this book that not of.it has ever spoken anybody ‘I am certain that nobody has ever talked about this book’ Sono certa su questo tavolo che non ci hanno messo niente I.am certain on this table that not there they.have put anything ‘I am certain that nobody put anything on this table’

The theory of Rizzi (1997) is constructed in such a way as to exclude examples of the type in (51). In particular, the che complementizer in these examples is identified with the highest C position, namely C(Force), which closes off the C field (cf. fn. 4, 6); therefore the prediction is that no leftperipheral material can precede it. In the present approach, however, in which che is the head of an independent nominal projection, the possibility must be open for its left periphery to host topical material. In fact, this is the structure we have postulated for sentences like (23) in section 1.2. Therefore examples like (51) are predicted to be grammatical. 1.4

Conclusions

In a nutshell, the present proposal represents a viable alternative to theories of C as a functional projection of the verb€– and possibly a better one in that it allows us to simplify functional architectures, treating complementizers in terms of the recursion of predicate argument structures. Another advantage resides in the simplification of the lexicon, to the extent that the complementizer can be given a unified lexical entry with the wh–phrases with which it is (often) homophonous. Though this would appear to be an even smaller gain than the previous one, consider that in a minimalist model the lexicon is all there is to language variation€– therefore simplifications in the lexicon are simplifications of the only learning task children have in front of them.

48â•… Romance complementizers:€structure & interpretation In the next chapter, we will pursue this second theme, providing some case studies in language variation and illustrating the account available for them under the present model. We argue that the model we are proposing is capable of accounting for the intricate parametrization of Romance varieties in a transparent way, effectively predicting that certain parametric values should be instantiated. In our view, these predictions depend on treating the complementizer as an argument, and specifically as an argument belonging to the same wh–series as classical wh–phrases. The burden of proof is on other theories to show that the same results can be mimicked if complementizers are functional projections of the verb. This may very well be possible€– but almost certainly through additional assumptions.

2 Variation in Romance k-complementizer systems

As observed in chapter 1, finite complementizers in Germanic and Romance are clearly nominal, belonging to argumental series such as demonstratives and wh–elements respectively. Therefore we propose that the complementizer is not a functional category of the verb, but rather an independent nominal head, which satisfies an argument slot of the matrix verb and which takes the embedded sentence as its complement. The aim of the present chapter is to investigate the range of variation in the lexicalization of complementizers in Romance varieties. We argue that only the treatment of the complementizer sketched in chapter 1 allows for a transparent account of the observed variation. First, the overlapping of Romance complementizers with the wh–system follows precise patterns, which excludes the possibility that we are merely dealing with homophony. Second, an account of the variation internal to the complementizer system requires the complementizer to be interpreted as introducing a propositional variable€– which is natural if it is a nominal head, but not if it is a functional projection of the verb. In section 2.3, we also consider the finiteness restriction that complementizers are subject to, arguing that they do not select for embedded temporal/modal properties (this being a potential argument in favour of their status as a functional projection of the verb), rather they select for the properties of the embedded EPP argument€– hence of the proposition as a whole.

2.1

Systems with two k-complementizers

The system of standard Italian, which we have considered in chapter 1, is characterized by a single complementizer introducing finite declarative clauses€– which is also the distribution familiar for English that. However, there are many Central and Southern Italian varieties which have two finite declarative complementizers, as in (1)–(2). One of the two complementizers, generally ka, is systematically found to introduce complements to verbs 49

50â•… Variation in Romance k-complementizer systems of ‘saying’ and ‘knowing’. The other complementizer, generally ke/ki (k), is found to introduce complements to verbs of ‘believing’ and ‘wanting’, though in some varieties (Guglionesi) some predicates alternate between the two complementizers. In other varieties, where all these predicates select the same complementizer ka, a second complementizer is found to introduce complements to ‘before’ and ‘after’ (Montenerodomo). Another important common property of these systems is that they involve some overlap between the complementizer system just described and the wh–quantifier system. Crucially, this overlap never affects the ka complementizer; instead it is the ke/ki complementizer that quite generally overlaps with the wh–quantifier for ‘what’. (1)â•… Guglionesi (Molise) a.â•… i. m ɔnnə dəttə ka vE krE to.me they.have said that he.comes tomorrow ‘They told me that he will come tomorrow’ a. ii. pEndzə ka vvE krE I.think that he.comes tomorrow ‘I think that he will come tomorrow’ b. i. vujjə kə vi krE I.want that you.come tomorrow ‘I want you to come tomorrow’ b. ii. pEndzə kə vvE krE I think that he.comes tomorrow ‘I think that he will come tomorrow’ b. iii. sNg aSSeutə prəmə kə tteu mənəssə I.am gone.out before that you would.come ‘I went before you came’ b. iv. sNg aSSeutə dppə kə ssi məneutə I.am gone.out after that you.are come ‘I went after you came’ c. kə ffi what you.do ‘What are you doing?’ (2) Montenerodomoâ•… (Abruzzi) a. i. m om ditt ka vi dum:n to.me they.have said that you.come tomorrow ‘They told me that you will come tomorrow’ a. ii. pEndz ka iss ve dum:n I.think that he comes tomorrow ‘I think that he will come tomorrow’ a. iii. vuless ka mniSS I.would.want that he.would.come ‘I would want him to come’

Systems with two k-complementizersâ•… 51 b. i. b. ii.â•… c.

so SSeut preim k mneiv I.am gone.out before that he.came ‘I went before he came’ so SSeut dop k tu avi mneut I.am gone.out after that you had come ‘I went after you came’ k ffi what you.do ‘What are you doing?’

The existence of double complementizer systems in varieties of Central and Southern Italy has been noted in the literature. However, Rohlfs (1969 [1954]:€190) puts it in the same bracket as another form of split lexicalization for sentential introducers, found in so-called Balkan languages as well as in some Southern Italian and Sicilian varieties. The relevant systems are characterized either by the lack of morphological infinitives or by their very reduced presence. Therefore the embedded infinitival clauses of English or Italian, associated with control and raising interpretations, are rendered in these Balkan (-like) languages by the embedding of finite clauses introduced by a specialized particle, which in Calabrian varieties, for instance, is mu. Thus, for Rohlfs, Neapolitan pèns ca vèn ‘I think that he come’ vs. vògli k mmang ‘I want that he eats’ and Calabrian pensu ca vèni ‘I think that he comes’ vs. vogghiu mu (mi) mangia ‘I want that he eats’ represent strictly comparable systems. The strongest argument for the independence of the two phenomena is that they actually combine. Thus, the Calabrian variety from Arena in (3), which has control and raising complements introduced by mu, also independently presents two complementizers of the k-series, roughly with the distribution described above for Montenerodomo, as in (2a–b). Of the two complementizers, it is the ki one, selected notably by ‘before’ and ‘after’, that has the same form as the wh–quantifier for ‘what’, as in (3c). (3)â•… Arena (Calabria) a. i. mi Dissiru ka vEni dmani to.meâ•… they.said thatâ•… he.comes tomorrow ‘They told me that he will come tomorrow’ a. ii. kriju ka vEni I.believe that he.comes ‘I think that he will come’ a. iii. E mmiu ka vini it.is better that you.come ‘It is better for you to come’ b. nESSivi duppu ki vinni I.went.out after that he.came ‘I went after he came’

52â•… Variation in Romance k-complementizer systems c. d.

ki ffatSi ijju what does he ‘What is he doing?’ vuu mu niï†fiï†fiu I.want Prt I.go.out ‘I want to go out’

We assume that introducers of the type of mu in Calabrian (or të in Albanian) are not complementizers in the sense defined in the preceding paragraph€– i.e. they are not (nominal) heads taking the embedded proposition as their (sentential) complement. Rather, they are internal to the embedded clause, to which they contribute modal properties (Rivero 1994; Roberts and Roussou 2003) and/or EPP properties involved in control and raising interpretations. Thus, we will put mu and the like aside without further discussion. Ledgeway (2003a, 2005, 2009) also argues against the identification of double complementizer systems with systems including a complementizer and a so-called subjunctive particle. It is possible that the three-complementizer split (ca, cu and che) described by Ledgeway (2005:€367 ff.) on the basis of Sgrilli (1983) for Early Salentino represents a system comparable to Arena in (3)€– with cu representing the ‘subjunctive particle’, very much like ku in contemporary Salentino varieties. Let us go back, then, to (1). The distribution of ka and k in varieties such as Guglionesi is reminiscent of the fact that standard Italian employs the indicative in complements of ‘to say’ and the subjunctive in complements of ‘to think’ or ‘to want’, with regional variants allowing for the indicative at least under ‘to think’, as in (4). (4) a. b. c.

Dico che viene I.say that he.comes ‘I say that he will come’ Penso che venga/ viene I.think that he.come/ he.comes ‘I think that he will come’ Voglio che venga I.want that he.come ‘I want him to come’

This similar distribution (and the somewhat similar distribution of indicatives vs. infinitives in the English translations) seems to suggest that some common property lies at the core of the complementizer split and the modality split. However, it should be stressed that it is not the case that one phenomenon depends on the other. In particular, in the examples in (1)–(2) the two

Systems with two k-complementizersâ•… 53 complementizers can embed the same indicative verbal forms, while ka in the Montenerodomo example in (2aiii) can embed the one form of subjunctive that appears to be present in the language, which can be roughly characterized as the counterfactual. Further examples of this pattern in Guglionesi are provided below in (9). In other words, there is no obligatory selection of the verb modality by the complementizer or vice versa. The evidence at our disposal also allows us to exclude the possibility that the complementizer split can be linked to the relatively impoverished modal system of varieties like (2); in other words, not only is it the case that there is no selection relation between complementizers and subjunctives, but there is no complementary distribution (functional equivalence) either. A case in point is represented by Sardinian varieties such as Paulilàtino in (5), in which a full-blown indicative€– subjunctive system of the standard Italian type combines with a double complementizer system of the type exemplified in (2) with Abruzzese varieties. (5) Paulilàtino (Sardinia) a. i. m anta nau ka bbeni kkraza to.me they.have said that he.comes tomorrow ‘They told me that he will come tomorrow’ a. ii. pEnts ka bbenizi I.thinkâ•… that you.come ‘I think that you will come’ b. i. pEnts ki bbEndzEDE I.think that he.come ‘I think he will come’ b. ii. kErdz ki bbEndzEzE I.want that you.come ‘I want you to come’ b. iii. sE bissiu appustisâ•… ki ze Benniu I.am gone.outâ•… after that you.areâ•… come ‘I went after you came’ b. iv sE bissia primma ki EssE Benniu DuE I.am gone.out before that be come you ‘I went before you came’ c. ittE faEzE what you.do ‘What are you doing?’

It will be noted that the ka complementizer combines with morphological indicatives in (a) while the ki complementizer combines with morphological subjunctives. However, the correlation between complementizer and verb modality breaks down in connection with complements to ‘before’ and ‘after’

54â•… Variation in Romance k-complementizer systems in (4biii) and (4biv) respectively. Following the pattern of standard Italian, the complement of ‘before’ is a subjunctive (a counterfactual) while the complement of ‘after’ is an indicative. Nevertheless, in both cases the complement is introduced by the ‘modal’ complementizer ki. This should of course be taken together with the existence of languages, like Montenerodomo in (2) or Arena in (3), in which complements to ‘before’ and ‘after’ maintain a specialized complementizer while other declarative sentential complements are uniformly introduced by ka. The lack of a one-to-one correlation between complementizers and modality is discussed at length by Ledgeway (2003a, 2005), who studies double complementizer systems in Early Neapolitan and other Southern Italian texts. He concludes that indicative clauses are frequently introduced by ‘CHE’. He also notices that selection for complementizers cannot be directly imputed to matrix predicates, since ‘it is not difficult to find minimal pairs … where the same main clause predicate selects in one case for an indicative clause headed by CA and in another for an indicative clause headed by CHE’ (Ledgeway 2005:€348). In other words, the historical varieties studied by Ledgeway have a complementizer system strictly comparable to that illustrated here by Guglionesi in (1) and in fact also by Paulilàtino in (5); the similarity between Ledgeway’s varieties and Sardinian ones is independently noted by Damonte (2006), who extends Ledgeway’s analysis to them. Nevertheless, Ledgeway subscribes to Formentin’s (1998:€ 432) conclusion that ‘the conjunction ca never introduces subjunctive clauses’, thus establishing a one-way implication between subjunctive and CHE complementizers. This oneway implication is not upheld by our data. Recall that we have already reviewed systems, such as Montenerodomo in (2), in which ka introduces a subjunctive (2aiii). More to the point, Sardinian varieties, which have fully productive subjunctives, show that both complementizers combine with both moods (indicative and subjunctive). Relevant data are found below in (11) for the variety from Làconi. Similarly the free alternation of ka and k with subjunctives is exemplified for Guglionesi in (9). Evidence that the subjunctive does not imply chi independently emerges in the Baunei corpus of Damonte (2006:€92), who quotes examples of ca with the subjunctive such as Mi pare ca custas cadirasa siente meda comodasa ‘it seems to me that these chairs are (lit:€be) very comfortable’. 2.1.1

Definite and indefinite complementizers€– and alternative analyses According to the conclusions of chapter 1, the finite k-type complementizer in Romance languages is an independent nominal head introducing a propositional

Systems with two k-complementizers╅ 55 variable, whose content is restricted by the embedded sentence. If so, the difference between the two complementizers in (1) is most naturally construed as a difference between two types of variable/quantification. We could assume, for instance, that the ka complementizer is a definiteness element, effectively the counterpart to a definite determiner, so that the LF of the Guglionesi example in (1a.ii) would be of the type in (6a). By contrast, the LF of a sentence like the Guglionesi example in (1b.ii) could include an indefinite quantification, corresponding to a free variable bound by existential closure, as sketched in (6b). (6) (Guglionesi) a. I think the x:€x he comes tomorrow b. I think for some x:€x he comes tomorrow

Because the evidence and the results discussed in this paper are mostly morphosyntactic in nature, we must content ourselves with a sketchy account of the interpretation. The evidence presented above is, however, sufficient to yield the generalization that of the two complementizers present in languages like Guglionesi in (1), it is always the indefinite complementizer that overlaps with a wh–quantifier. This is because the other complementizer effectively introduces a definite description, which is hardly compatible with wh–quantification€– while the indefinite complementizer introduces a propositional variable subject to existential closure, more or less like the argumental wh–variable. Manzini and Savoia (2005) provide examples from a considerable number of languages which behave like Guglionesi, and in all cases the same generalization holds, as shown by the summary table reproduced in (A) in the Appendix. Of course it is also logically possible to have double complementizer systems of the type in (6) in which no overlap with the wh–system is found, as summarized in table (A’) in the Appendix; the Paulilàtino sentence in (5) exemplifies this type of system.1 The preceding results are important for the present discussion for more than one reason. A preliminary point is that the identity of form shown by complementizers and wh–elements covers grammars that are not just different in general terms, but are specifically different in terms of the finite complementizer system€– having either a single complementizer or a split between complementizers. In other words, it cannot be objected that the formal identity of the complementizer with the wh–system is in a sense the property of just one grammar, and therefore to be treated as accidental. More importantly, the fact that in two complementizer systems it is the indefinite complementizer that coincides with the wh–system supports the semantics for complementizers sketched here, and indirectly the syntax in chapter 1.

56â•… Variation in Romance k-complementizer systems Now the question is:€under the alternative view, which has complementizers as C functional heads, could one provide characterizations for the two different complementizers, and capture the overlap with wh–quantifiers (or lack thereof)? Ledgeway (2003a, 2005), dealing with data closely comparable with ours, does provide an answer to the first part of the question. Before evaluating it, let us note that the two complementizer systems of languages like those in (1) may have a reflex even in a language like Italian in complementizer deletion phenomena; we will therefore consider first what the existing literature says about those. In fact there appears to be a good match between matrix predicates that require the indefinite complementizer in (6b) and those which, according to the literature, allow for the deletion of the che complementizer in Italian, as in (7b–d). By contrast, contexts which require the definite complementizer, as in (6a), seem to coincide with those that do not admit complementizer deletion in Italian, as illustrated in (7a). (7)â•… a.â•… b. c. d.

So *(che) viene I.know that he.comes ‘I know that he is coming’ Penso (che) venga I.think that he.come ‘I think that is coming’ Vorrei (che) venisse I.would.want that he.came ‘I would like him to come’ E’ meglio (che) venga It.is better that he.come ‘It is better for him to come’

The issue is made more complex by the existence of at least one variety in which effectively any matrix predicate admits complementizer deletion. The latter is identified by Cocchi and Poletto (2002) with the variety spoken in Florence. As in many cases involving Italian, the question can legitimately be asked whether, on the contrary, the judgements attributed to the standard are not unduly constrained by normative considerations; in fact even the judgements in Cocchi and Poletto (2002) appear to be unnecessarily restrictive for the Italian spoken in Florence.2 Be that as it may, the reason to address the (potential) pattern in (7) is that phenomena of complementizer dropping have been prominent in the theoretical literature since the study of the that-t filter in English by Chomsky and Lasnik (1977). Therefore distributions of the type in (7) have been discussed more than once (Poletto 2001; Giorgi and Pianesi 2004), providing us with a possible alternative to the analysis advocated here.

Systems with two k-complementizers â•… 57 On the basis of the preceding discussion of double complementizer systems, we would be led to characterize the alternation in (7) by saying that in order to introduce a propositional definite description the complementizer is necessary, as in (7a); while other types of propositional complementation admit the bare embedding of a sentence under the matrix verb, as in (7b–d). As for the question of why contexts obligatorily introduced by the complementizer feature the indicative, while subjunctive contexts admit complementizer dropping, we can assume that the indicative vs. subjunctive split is itself connected with the definite or indefinite binding of a propositional variable. In fact, Manzini (2000) notices that mood selection is not exclusively determined by the matrix predicate. Thus, ‘to know’ in (8a) normally requires the indicative. The latter remains possible when the matrix predicate is questioned, as in (8b), in which case the reading is still a factive one presupposing the truth of the embedded proposition. Questioning the matrix sentence also renders the subjunctive possible, in which case, however, the embedded proposition is in the scope of the question operator, yielding a non-presuppositional reading. (8)â•… a.â•… b. c.

So che è/ *sia venuto I.know that he.is/ he.be come ‘I know he came’ Sai che è venuto? you.knowâ•… that he.is come ‘Do you know that he came?’ Sai che sia venuto? you.know that he.is come ‘Do you know if he came?’

Alternations like (8b–c) can be described by saying that the indicative (like ka–type complementizers) introduces a propositional definite description, read outside the scope of polarity operators like the question operator. By contrast, the subjunctive (like ke–type complementizers) introduces an indefinite propositional variable, interpreted within the scope of polarity operators. This accounts for the parallelism between the complementizer split and the indicative vs. subjunctive split apparently present in (7). Let us now consider Giorgi and Pianesi (2004) and Poletto (2001). They assume, as we do, that so-called complementizer deletion does not correspond to the actual deletion of lexical material nor to the zero instantiation of the C position. For Giorgi and Pianesi (2004), while the sequence of che and a subjunctive provides separate instantiations for modality (through the che complementizer) and for agreement (through the I position of the verb), in complementizer deletion contexts, the Mood and Agr properties are conflated in the

58â•… Variation in Romance k-complementizer systems I head, and simply lexicalized by the verb. The reason why che cannot delete in indicative contexts has to do with the fact that in indicative contexts T and Agr properties cannot conflate, but must be independently lexicalized. To be more precise, ‘the T-features are duplicated in che/C in such a way that the indexical component can be evaluated outside the clause’ (Giorgi and Pianesi 2004:€200). According to Giorgi and Pianesi (2004) the satisfaction of this latter condition is independently required in order to obtain the so-called Double Access Reading, under which the situation denoted by the embedded sentence is taken to hold both at the time of the matrix event and at the time of utterance. What is relevant here is that Giorgi and Pianesi’s (2004) framework could in principle account for double complementizer systems, as the result of a split between exponents of C endowed with Tense (ka) and exponents of C endowed with Mood (ke/ki). But since there is no (direct) connection between the whâ•fi system and subjunctive Mood (e.g. wh–questions do not necessarily select the subjunctive) the overlapping of modal complementizers and the wh–system (to the exclusion of the non-modal complementizers) would not be predicted. The present discussion provides the missing link; but crucially it does so by extricating the complementizer from the Mood and Tense system to which Giorgi and Pianesi (2004) confine it€– and by assuming that it is an argumental element. Correspondingly it is removed from the C position and construed as an independent (nominal) head. Let us now turn to Poletto (2001), who draws a parallel between complementizer deletion and Germanic V-to-C movement; the latter, at least in some languages, is in complementary distribution with the lexicalization of the complementizer (and thus seems to result in its deletion). The idea is that ‘the class of verbs selecting C[omplementizer] D[eletion] complements is exactly the same as in Germanic embedded V2 contexts’ (Poletto 2001:€267). Thus both phenomena can be captured by the movement of the verb to the C position. On the assumption that ‘CD is possible only when the embedded verb is a subjunctive, a future or a conditional form’ and that ‘these forms all have a modal quality’, Poletto (2001:€278) proposes that ‘a [–realis] feature … is realized on the head of the complement and attracts the verb into the CP domain’. Let us leave aside the question of whether the verb does or does not move to the C position in Italian complementizer deletion contexts, what is relevant here is Poletto’s (2001) characterization of these contexts in terms of a [–realis] feature in C. In these terms, it is not difficult to recognize the common conceptual core of Poletto’s (2001) analysis and Giorgi and Pianesi’s (2004), namely that complementizer deletion involves modal properties, absent from indicative contexts. The point on which we differ from both analyses is once again made clearer if we try to extend Poletto’s (2001) account to the double

Systems with two k-complementizers â•… 59 complementizer systems considered here. One could assume, for instance, that what we have called the indefinite complementizer lexicalizes the€–realis feature, while the other complementizer satisfies the +realis feature in C. But if so, there is no reason why the former should overlap with the wh–system (to the exclusion of the latter), since wh–questions do not appear to have any (immediate) connection with€–realis (not being restricted, say, to the subjunctive). As before, the present theory has the advantage that it can make direct predictions on this point€– but these crucially depend on complementizers being extricated from the C position, and treated as argumental, nominal heads. We consider next the proposals put forth by Ledgeway (2003a, 2005). The empirical generalization that Ledgeway proposes concerning the distribution of the two complementizers depends on a further set of data concerning the distribution of CHE and CA with respect to embedded left-peripheral material. In his corpus of eleven early texts, he finds that out of a total of 327 examples of CA-clauses … a mere 10.1% were found to contain one or more elements in the left periphery, whilst from a total of 1,061 examples of indicative clauses introduced by CHE … 41.8% were found to host one or more elements in the left periphery … Secondly even if there do occur some examples of the complementizer CA preceding one or more elements of the left periphery … with very few exceptions, all such examples involve elements of the Focus field, namely … foci … and fronted indefinite quantifiers.â•…â•… (2005:€360)

Ledgeway’s characterization of CA and CHE is in fact partially based on mood selection, since he assumes that ‘CA and CHE are invariably merged in … Fin° … as an overt reflex of the different modal specifications (indicative vs. subjunctive) they check in Fin°, from which automatically follows the claim that CA only occurs in indicative clauses and CHE only in subjunctive clauses’ (2005:€372). However, ‘the appearance of topics and foci forces the complementizers CA and CHE to move out of Fin° to target Force°’ (2005:€376); crucially ‘this movement operation is spelt out morphologically only in the case of the indicative complementizer CA which invariably surfaces as CHE’ (2005:€374–5). The empirical generalization, namely that the presence of an embedded left periphery forces the complementizer to be CHE, deserves to be investigated for contemporary varieties as well. What we have found is that our varieties do not observe such a restriction, since both foci (e.g. (9c)) and topics (including clitic left dislocation in (9b)) appear under ka in Guglionesi in (9) and in the Sardinian varieties in (10)–(11).3 We also exemplify topics and foci under k and ki respectively, in order to establish that our varieties do not retain the asymmetry between the two complementizers, simply reversing their respective positions (i.e. ka higher and k/ki lower).4

60â•… Variation in Romance k-complementizer systems (9) Guglionesi a.â•… m ann  ka krE  to.me they.have said that tomorrow you.came ‘They told me that tomorrow you would come’ b. m ann  ka u kES ts l annâ•…    to.me they.have said that the cheese Refl it they.have taken the boys ‘They told me that the cheese was taken by the boys’ c. m ann  ka ï†fi ts ann tt i ajjEun to.me they.have said that the cheese Refl they.have taken the boys ‘They told me that it was the cheese that the boys took’ d. vujj ka/k pur tEu mnss I.want that also you come(subj) ‘I want you too to come’ e. vujj ka/kâ•… krE pur lrâ•… mnssn I.wantâ•… that tomorrowâ•… also they come(subj) ‘I want them as well to come tomorrow’ (10)â•… Paulilàtino a. m anta nau ka raza bbenizi to.meâ•… they.haveâ•… saidâ•… thatâ•… tomorrowâ•… you.come ‘They told me that you will come tomorrow’ a’. m anta nau ka u libru ɖ aza leddzju to.me they.have said that the book it you.have read ‘They told me that the book you read’ b. EstE mmendzus ki fintsaza juanni bEndzEDEâ•… kraza it.is better that even John comes tomorrow ‘It is better that John as well comes tomorrow’ (11)â•… Làconi (Sardinia) a. dEɔ krEɔ ka/tɔi issu Buru/ kraza I believe that he too/ tomorrow ‘I believe that he will come as well/tomorrow’ b. dεɔâ•… krEɔ tSi/kaâ•… issu Buru/â•… kraza I believeâ•… that he too/ tomorrowâ•… ‘I believe that he will come as well/tomorrow’

eniDi he.comes EdZaDa he.comes

Let us now consider Ledgeway’s (2003a, 2005) analysis, which, as we have seen, consists of a more conservative assumption, namely that complementizer choice depends on mood, and an innovative proposal, namely that movement of CA results in it being spelled out as CHE. The first part of the proposal is in essence the same as that reviewed above for complementizer deletion. Thus our generalization that the ‘subjunctive’ complementizer is the one that overlaps

Systems with two k-complementizersâ•… 61 with a wh–element (if such an overlap is present in the system), remains as inaccessible to Ledgeway as it does to Giorgi and Pianesi and Poletto, and for much the same reason. Indeed, connecting complementizers directly to modality (as is natural in a treatment where both belong to the functional spine of the verb) leaves us without any basis for associating the complementizer with the argumental, nominal system of wh–phrases. The novel part of Ledgeway’s (2003a, 2005) proposal is that movement of CA from the Fin to the Force position results in its morphological spellout as CHE. Now, there is no syntactic and/or morphological theory that can literally change a lexical item into another (because of Chomsky’s (1995) Inclusiveness, among others). Nor is it possible to have a lexical entry specified for a context defined by movement (since only intrinsic properties of lexical items belong in the lexicon). Therefore, Ledgeway’s analysis seems to imply that double-complementizer languages have a spell-out for indicative complementizers in Fin€ – and another spell-out for subjunctive complementizers as well as for complementizers in Force. This means that the unification achieved by Ledgeway’s analysis is only apparent€– as it requires a disjunctive lexical entry for the CHE complementizer. Alternatively, one could try to unify the entry for CHE by recourse to underspecification€ – i.e. to the idea that CHE is simply the default complementizer of the relevant languages, inserted as the Elsewhere case. But if so, the generalization noted here concerning its overlapping with the wh–system would become impossible to explain, in the absence even of modal properties on the complementizer. 2.1.2 Generalized wh–complementizers Section 2.1.1 discusses systems with a split between definite and indefinite complementizers of the type of Guglionesi in (1) or in Paulilàtino in (5). A different grammar, however, seems to be exemplified by Montenerodomo in (2) or by Arena in (3), in which the k/ki complementizer overlapping with a wh–question word is restricted to a few contexts, which in our data coincide with complements to ‘after’ and ‘before’. These contexts do not correlate with the verb modality€– which can be subjunctive under ‘before’ but is consistently indicative under ‘after’. Interestingly, all of the languages tabulated in (A’) of the Appendix, in which there is no overlap between the wh–system and the complementizer system, show the definite vs. indefinite complementizer split of the type in section 2.1.1. By contrast, the distribution of k/ki complementizers in languages like Arena in (3) or Montenerodomo in (2) appears to be necessarily connected with their wh–nature.

62â•… Variation in Romance k-complementizer systems In our view the key to the distribution observed resides in the fact that prepositions like ‘before’ or ‘after’ can easily embed complements introduced by wh–arguments/adjuncts, as in the comparative structures in Italian (12). In turn, the comparatives in (12) appear to be interpreted in relation to a temporal argument. Thus (12a) corresponds to an LF roughly of the form ‘I arrived at a time t before/earlier than a time t such that I thought I would arrive at t’, where the most embedded (italicized) proposition undergoes ellipsis.5 (12)â•… a.â•… b.

Sono arrivato prima di I.am arrived before of ‘I arrived earlier than I thought’ Sono arrivato dopo di I.am arrived after than ‘I arrived later than planned’

quanto how.much

pensassi I.thought

quanto stabilito how.much planned

On the basis of (12) we propose that languages that embed a wh–complementizer under ‘after’ and ‘before’ effectively select a (comparative-like) wh–structure. As noted before, the evidence we are considering is essentially morphosyntactic in nature and therefore hardly insightful when it comes to the semantics; hence semantic analyses are sketched here essentially for the sake of falsifiability. We suggest that the wh–complementizer acts in this case as a sort of generalized quantifier, binding the propositional variable, restricted by the following sentence and, at the same time, the temporal variable. We then expect that the generalized quantifier that we have hypothesized also turns up as a whâ•fi binder of arguments; in other words that it is not specialized as a propositional introducer, but will overlap with wh–arguments€– as is indeed the case. It is worth noting that Ledgeway (2009) notices the existence of doublecomplementizer systems that cannot be assimilated to those in section 2.1.1 in relation to the Calabrian variety of Cosenza. He reports that even in the speech of younger generations chi is obligatory in optative expressions such as chi ti vò affucà ‘(that you) go drown’; in a more conservative variety he finds chi to occur in complement sentences when the embedded verb is in the subjunctive (counterfactual). His generalization is that ‘ca is a passe–partout complementizer compatible both with indicative and with subjunctive, while chi is a modally marked complementizer’. Needless to say, this characterization cannot be extended to varieties like Arena in (3), in which ki is selected by ‘after’.6 Let us nevertheless consider his analysis€– namely that ca realizes the Force head while chi lexicalizes the Fin head in Rizzi’s (1997) schema, i.e. the head ‘responsible for marking modal distinctions’. In this schema, therefore, the Cosenza variety would involve the opposite distribution to the varieties of section 2.1.1, in which che is assigned to the Force position, while ca is restricted to Fin. Now, the point being made here is that the k of languages like Guglionesi

Systems with two k-complementizersâ•… 63 and the ki of languages like Arena have in common the fact that they overlap with the wh–system. From this perspective, Ledgeway’s (2003a, 2005, 2009) analysis would be saying that in double-complementizer systems, the wh–complementizer is sometimes a specialized Fin and sometimes a default. It seems to us that a theory that captured the continuity between the various double-complementizer varieties would have a clear explanatory advantage.7 This leads us to the next point to be examined, namely whether our theory can in fact capture this continuity. Crucially, in all languages of the same type as Guglionesi in (1), the indefinite complementizer, in the sense of (6b), turns up in ‘before’/‘after’ contexts as well. Now, if the wh–complementizer in languages like Arena in (3) or Montenerodomo in (2) is to be able to function as some sort of generalized binder of propositional and wh–variables, it cannot be definite (which would exclude wh–quantification). Therefore, in languages with definite vs. indefinite complementizers like Guglionesi, the contexts relevant for its lexicalization will be picked up by the indefinite complementizer, acting as a generalized wh–quantifier. This conclusion only depends on the indefinite nature of the complementizer and not on its actual formal identity with a wh–argument. Therefore we predict that the indefinite complementizer will also pick up ‘before’/‘after’ contexts in systems where it does not overlap with a wh–argument, such as Paulilàtino in (5).8 The final fact worth bringing up is that in all of the languages tabulated in (A–A) in the Appendix, what we have called the indefinite and generalized wh–complementizers also introduce relative clauses. Relevant examples are given in (13)–(16). These involve that relatives, and not who relatives, since, although the relevant complementizers overlap with ‘what’, as illustrated in (1)–(3) and (5), ‘who’ has a distinct lexicalization, as illustrated in (13b) and in (14c)–(16c). We can extend to (13)–(16) the analysis that we have already proposed for (12), treating ki/k as some sort of generalized binder for the propositional variable and the argumental one (actually a resumptive pronoun in the examples in (14b)–(16b). The treatment of ‘after’ and ‘before’ complements in what precedes also implies that we should find the indefinite/generalized wh–complementizer introducing comparatives. The data in Paulilàtino’s (16d) confirm this prediction. (13)â•… Guglionesi a.â•… E kkull k vvad sEmbr he.is that that I.see always ‘He is the one that I see all the time’ b. ki vE who comes ‘Who comes?’

64â•… Variation in Romance k-complementizer systems (14)â•… Montenerodomo a.â•… e kkwillâ•… k vve:tâ•… sEmbr it.isâ•… that thatâ•… I.see always ‘It is the one I see all the time’ b. e kkwill k i so dat l sld it.is that that to.him I.am given the money ‘It is the one I gave the money to’ c. ki ve whoâ•… comes ‘Who comes?’ (15) Arena a. su kkijjiâ•… ki vviju sEmpri they.areâ•… those thatâ•… I.see always ‘They are the ones that I see all the time’ b. su kkijji ki tSi dEttsi li srdi they.areâ•… thoseâ•… thatâ•… thereâ•… I.gave the money ‘They are the ones I gave the money to’ c. ku vni whoâ•… comes ‘Who comes?’ (16) Paulilàtino a. Es kussu ki bbi zEmpErE he.isâ•… that thatâ•… I.see always ‘He is the one that I see all the time’ b. Es kussu ki  appɔ jau z inari he.is that that him I.have given the money ‘He is the one that I gave the money to’ a’. juanniâ•… ki Estâ•… ammiu meu John thatâ•… is friend mine ‘John, who is my friend …’ c. kiE bbeniDi who comes ‘Who comes?’ d. tEndzâ•… pru llibrsâ•… ki (n)â•… inarE I.have moreâ•… books thanâ•… not money ‘I have more books than money’

In a language like Ardaùli in (17), of the Paulilàtino general type, the indefinite complementizer ki introduces restrictive relatives and the definite ka complementizer introduces appositive relatives. We suggest that in this language the choice of the complementizer is sensitive to the nature of the embedded variable. In particular, ki is incompatible with appositive relatives, in that they contain an individual variable; thus, the language resorts to the definite complementizer ka.9

‘If’â•… 65 (17)â•… Ardaùli (Sardinia) a.â•… Es kussu ki bbi sEmpErE he.isâ•… that that I.see always ‘He is the one that I see all the time’ b. Es kussu ki  app jau zu inarE he.is that that himâ•… I.have given the money ‘He is the one that I gave the money to’ a’. dzuanni ka Estiâ•… ammiuâ•… meu John, thatâ•… is friend mine ‘John, who is my friend …’ c. kiE eniDi whoâ•… comes ‘Who comes?’ d. tEndzâ•… pru llibrɔsâ•… ki (n)â•… inarE I.have moreâ•… books thanâ•… not money ‘I have more books than money’

As discussed in chapter 1, Arsenijevic (2009) and Kayne (2010) propose theories of complementation as relativization. We may wonder how these fare with respect to the relativization facts in (13)–(17). Consider for instance Paulilàtino€ – i.e. a language that has a two-complementizer system, but in which relative clauses are formed without exception by the indefinite complementizer. If the definite ka complementizer is nothing but a relative pronoun, why doesn’t it ever overtly appear in relatives? In fact, the preliminary question arises of how Kayne or Arsenijevic would deal with double-complementizer systems. In Arsenijevic’s terms, different complementizers could select for different types of ‘Force’ arguments. But even so, it is difficult to see why what we have characterized as the indefinite complementizer has the same form as the relative complementizer, but the definite one does not. For it does not seem possible to say that relative clauses associate with a particular type of Force. Kayne’s proposal would have to deal with the facts in a similar way; therefore the same critique applies (cf. also the discussion on conditionals in the next section).

2.2

‘If’

The complementizer system of Italian, as discussed in chapter 1, and that of the varieties considered so far, have a specialized lexicalization for the interrogative/hypothetical complementizer, se in Italian, corresponding (more or less) to English ‘if’. Thus the Italian data in (1)–(4) of chapter 1 are completed by (18).

66â•… Variation in Romance k-complementizer systems (18)â•… a.â•… b.

Se piove (spesso) esco if it.rains often I.go.out ‘If it rains I (often) go out’ Non so se viene not I.know if he.comes ‘I don’t know if he’ll come’

As it turns out, this state of affairs does not hold of necessity, but only as one of several possible parametric options. In some Sardinian varieties with a double declarative complementizer system, the indefinite complementizer also lexicalizes hypothetical and interrogative ‘if’, as exemplified by Làconi in (19). Note that the indefinite complementizer tʃi belongs to the wh–lexical series, exemplified by tʃinni ‘who’ in (19ci). Note also that the same distinction holds between the complementizer for restrictive and appositive relatives as we discussed for Ardaùli in (17); as before, it is the complementizer introducing restrictive relatives that overlaps with the indefinite/‘if’ complementizer. (19)â•… Làconi a. b. i. b. ii b. iiiâ•… c. i. c. ii. d. i. d. ii e. i

m anti nau ka ennis kraza to.me they.have said that you.come tomorrow ‘They told me that you come tomorrow’ bZ tSi EdZas kraza I.wantâ•… that you.comeâ•… tomorrow ‘I want you to come tomorrow’ E mmedZuzu tSi bbEdZaza it.is better that you.come ‘It is better for you to come’ sE essia prima Zi fRssaz arribbau I.am gone.out before that you.were come ‘I went out before you came’ a ttSinni tserriaza to whom you.call ‘Who are you calling?’ itta faizi whatâ•… you.do ‘What are you doing?’ funti gussuzu tSi tserriu zEmprE they.were those that I.call always ‘They were the ones that I always call’ srrE Dua ka EstE ammia mia sister yours that is friend mine ‘Your sister, who is a friend of mine …’ tSi llEzE BEdï†ı that you.want I.come ‘If you want, I will come’

‘If’â•… 67 e. iiâ•… f.

tSi rEDE n bbEss thatâ•… it.rains not I.go.out ‘If it rains, I won’t go out’ n iï†fiï†fiiu tSi u tserriu not I.know that him I.call ‘I don’t know if I shall call him’

Another major pattern of lexicalization of ‘that’ and ‘if’ is exemplified in (20) by Miglionico. This has a single declarative complementizer and a separate complementizer for hypotheticals and questions€– reproducing the conditions of a language like Italian. However, the hypothetical/question complementizer actually belongs to the wh–system, in fact overlapping with the wh–element for ‘who’, namely tʃi. Because of this, the embedded sentence in (20e) is ambiguous between the wh–question reading ‘who’ and the yes–no question reading ‘if’. (20)â•… Miglionico (Lucania) a. m vnn ditt ka vin kra to.me they.have said that he.comes tomorrow ‘They told me that he comes tomorrow’ b. s assut prim ka tu vniss I.am gone.out before that you came ‘I went out before you came’ c. i. tSi ve:n whoâ•… comes ‘Who is coming?’ c.â•… ii.â•… tSe ffa ï†fin what they.do ‘What are they doing?’ d. tSiâ•… vvuo ve if you.wantâ•… I.come ‘If you want, I come’ e. nanâ•… sattS tSi ven not I.know who/if he.comes ‘I don’t know who comes/if he comes’

In short, in the Làconi examples in (19) there is a single lexicalization for the indefinite declarative complementizer and for the question/hypothetical complementizer. Table (B) of the Appendix summarizes varieties for which this distribution holds; all of them are Sardinian, of a type in which there is no overlap between the complementizer and the wh–system. Crucially, from the table in (B) it can be seen that the hypothetical/interrogative complementizer never has the same form as the ka definite complementizer. In turn, in a language like Miglionico in (20) the lexicalization for the hypothetical/interrogative

68â•… Variation in Romance k-complementizer systems complementizer overlaps with that of a wh–element, while the (single) declarative complementizer has a different lexicalization. The data summarized in table (C) of the Appendix show that it is always the hypothetical complementizer that coincides with the wh–system (as opposed to the declarative one). The complementizer system of languages like (19)–(20) closes a potential gap in the discussion so far. As discussed in chapter 1, we know that wh–items are not intrinsically interrogative; in Italian, for instance, argument che can head a free relative as in (21a) or an exclamative as in (21b). We interpreted these data as showing that question, declarative or exclamative values are not intrinsic to the wh–element, but rather contributed by contextual operators. This is far from an isolated case in natural languages. For instance, so-called N-words in Romance are not intrinsically negative, but rather introduce a variable which can receive a negative interpretation in the scope of a negative operator€– and can equally well be licensed in non-negative contexts (cf. chapters 3–4). (21)â•… a.â•… b.

Faiâ•… che ti pare do whatâ•… youâ•… please ‘Do as you please’ Che non farebbe! what not he.would.do ‘What wouldn’t he do!’

On these grounds we expect that a complementizer belonging to the wh–series will not require a question interpretation. What does represent a potential problem for the present theory, however, is that none of the complementizers considered in section 2.1 allows for such an interpretation. Nothing in what we have said so far blocks the possibility that a k–complementizer could be interpreted as interrogative in the scope of an appropriate question operator, in a similar way to specialized interrogative wh–complementizers like English whether. From this perspective, systems like (19)– (20) are interesting in the first instance because they instantiate precisely the possibility we predict. Let us consider, then, what the characterization of the hypothetical/interrogative complementizer may be, beginning with a language like Italian (or indeed English) in which it has a specialized lexicalization. Hypothetical sentences are interpreted much in the same way proposed previously for complement sentences, that is, as the restriction of a propositional variable. According to Lewis’s (1975) classical treatment, the latter is bound by an adverb of quantification, or by a generic quantifier in the absence of other overt quantifications. Thus the sentence in (18a) has an LF of the type in (22a), if the adverb of

‘If’â•… 69 quantification is computed. Otherwise the propositional variable is closed by a universal quantification, as in (22b). (22)â•… a.â•… For many situations/possible worlds x:€x it rains, I go out (in x) b. For all situations/possible worlds x:€x it rains, I go out (in x)

An element which introduces a propositional variable (restricted by the embedded propositional content) is associated in the present grammar with structures of the type argued for in chapter 1 for k-type complementizers€ – which we will then extend to Italian se ‘if’.10 The latter can in turn be interpreted not only in a hypothetical context as in (15), but also in an interrogative context€– exactly like English ‘if’. The embedded interrogative interpretation simply follows if the variable introduced by se and the like (and restricted by the proposition following it) can be licensed in the scope of a question operator. We provisionally assume that this is what the interpretation of an interrogative complementizer amounts to, as in (23).11 (23)â•… I wonder for which situations/possible worlds x:€x John comes

At this point the characterization of the se complementizer is fundamentally the same as for the che complementizer, and more specifically for the indefinite complementizer in (6b), where the variable that the complementizer introduces is closed quantificationally. Therefore we fully expect that ‘if’ and the indefinite declarative complementizer can be lexicalized by the same item, yielding the Làconi system in (19). In view of the proposals reviewed in section 2.1.2 to the effect that complement sentences are relatives, it is particularly worth remarking on the independent literature treating conditionals as relatives€– specifically as free relatives. Thus, for Bhatt and Pancheva (2006), ‘turning to conditionals, our proposal that they are interpreted as free relatives amounts to the claim that they are definite descriptions of possible worlds’. As for the syntax, ‘the null operator in Spec, CP of if-clauses and likely the when itself in e.g. German conditionals, is a definite binder of the possible world variable’. As Bhatt and Pancheva note, the proposal that there is a covert operator in the Spec, CP of conditional if clauses goes back to Larson (1985), where the focus of the discussion is interrogative if clauses. For the latter, Larson posits the presence of a covert whether, which he extends to conditional if clauses. For Bhatt and Pancheva, ‘the fact that if functions in many languages as both a conditional and an interrogative complementizer makes sense within the general proposal that conditionals are free relative clauses. In English and in many other languages, this syncretism would be part of a more general structural parallelism between questions and free

70â•… Variation in Romance k-complementizer systems relatives’. However, if ‘free relatives are interpreted as definite descriptions, i.e. with the variable abstracted over being bound by a definite operator’ (Bhatt and Pancheva 2006), the lexicalization patterns observed in Làconi, in which ‘if’ overlaps with the indefinite complementizer, and not with the definite one, seem to be difficult to account for. More generally, the present account does not require us to postulate zero world operators in the syntax. Quite simply, the so-called complementizer is itself the element that introduces a (situation/ possible world) variable in the syntax of LF.12 If anything, our problem is how to provide a sharper characterization for languages like Italian that have two distinct complementizers, or languages like Paulilàtino in (5) that end up having three. Let us begin by considering a Â�language with three distinct forms, including the definite and indefinite declarative complementizers, as well as the hypothetical/interrogative complementizer, as exemplified for Paulilàtino in (5) and (24). Following the discussion in section 2.1, the definite complementizer introduces a propositional complement akin to a definite description, while the indefinite complementizer introduces an indefinite variable. The latter is also true of the proposition introduced by ‘if’, which therefore needs to be further differentiated. (24)â•…

Paulilàtino si ‘prEDEâ•… n Ess ifâ•… it.rains notâ•… I.go.out ‘If it rains, I won’t go out’

In introducing the contexts triggering subjunctive in Italian in section 2.1.1, we mentioned that they are not necessarily lexically selected; rather, subjunctive can be triggered in the scope of the question operator in (8c). In the scope of the question operator, the declarative complementizer, ordinarily embedded by a verb like to know, alternates with the ‘if’ complementizer, as in (25)€– a phenomenon discussed in the literature under the name of ‘unselected questions’ (Adger and Quer 2001). Needless to say, the reading of the embedded sentence in (25) is always non-factive, i.e. within the scope of the matrix question operator, independently of whether the indicative or the subjunctive is selected. As shown in (26), a matrix negation has very much the same effect as a question operator, both with respect to the triggering of the subjunctive in (26a) and to the possibility of the ‘if’ complementizer in (26b). In (26) only the co-occurrence of the che complementizer with the embedded indicative in (26a) yields a presupposed (factive) reading of the embedded sentence; both the subjunctive and the ‘if’ complementizer force the sentence to be read within the scope of the negation operator.

‘If’â•… 71 (25)â•… Sai se è/ siaâ•… venuto? you.knowâ•… ifâ•… he.is/â•… be come ‘Do you know if he has arrived?’ (26) a.â•… Nonâ•… sa che sono/siaâ•… guarito not he.knowsâ•… thatâ•… I.am healed ‘He doesn’t know that I am healthy again’ b. Nonâ•… sa se sono/siaâ•… guarito not he.knowsâ•… that/ifâ•… he.be healed ‘I don’t know if he is healthy again’

The fact that ‘if’ is triggered in the scope of a matrix question or negative operator with an otherwise declarative (factive) verb, as in (25)–(26), points the way to a solution for the problem we are concerned with, i.e. the characterization of the contexts specialized for ‘if’. Negations and questions are two of the fundamental contexts triggering (negative) polarity items€– the third one being the hypothetical one. Thus, for instance, in Italian nessuno ‘anybody’, niente ‘anything’ etc. are licensed in the scope of negation (yielding the equivalent of the negative quantifiers nobody, nothing etc. in English), as well as in the scope of a question or of a hypothetical. We suggest that the complementizer system of languages like Italian or Paulilàtino in (24) (or, for that matter, English) is sensitive to polarity, so that complementizers, i.e. nominal heads responsible for introducing propositional variables, are lexicalized by two different items in non-polarity and polarity contexts, surfacing in the latter as se in Italian, if in English etc. Correspondingly, we shall refer to these as the polarity complementizers. In section 2.1 we concluded that it is to be expected that the complementizer system, i.e. the system of propositional variable introducers, and the wh–system may coincide. More to the point, in cases of split complementizer systems the coincidence is predicted to involve the indefinite complementizer, i.e. the one introducing an indefinite variable, rather than the definite complementizer, which introduces a (propositional) definite description. This mode of reasoning can now be extended to the ‘if’ complementizer. In languages of the type of Làconi in (19) there is no overlap between the complementizer and wh–systems. However, languages like Miglionico in (20) verify this prediction for the systems with a polarity complementizer ‘if’ and a nonpolarity complementizer ‘that’. For in Miglionico, as in the other varieties tabulated in (C) of the Appendix, it is the ‘if’ complementizer that belongs to the wh–system€– and not the (single) declarative complementizer. A further parameter brought to the fore by varieties like Miglionico has to do with which element of the wh–system the complementizer identifies with. In

72â•… Variation in Romance k-complementizer systems Italian, as reviewed in chapter 1, the overlap is between the declarative complementizer and the wh–element for ‘what’. By contrast, in Miglionico in (20) the overlap is between the ‘if’ complementizer and the wh–item for ‘who’. This parameter is independent of the others considered. Thus, in Ruvo di Puglia in (27) the ‘if’ complementizer has the same form as ‘what’, and not ‘who’ as in Miglionico. By contrast, in a variety like Secinaro, otherwise entirely comparable to Guglionesi in (1), the modal complementizer has the same form as ‘who’, and not ‘what’, as in (28). (27)â•… Ruvo di Puglia (Apulia) a.â•… tS cɔ:v nn iss that it.rainsâ•… not I.go.out ‘If it rains I won’t go out’ a’. nn tsattï†fi tï†fi fratt m cɔm not I.know if brother-yours me calls ‘I don’t know if your brother is calling me’ b tSi vEn whoâ•… comes ‘Who comes?’ c. tSe ffɔS whatâ•… you.do ‘What are you doing?’ (28) Secinaro (Abruzzi) a. vujj k vvi I.want that you.come ‘I want you to come’ b. k vvE who comes ‘Who comes?’ c. ku ffE what you.do ‘What are you doing?’

In terms of the discussion in chapter 1, the wh–quantifier for ‘what’, such as Italian che, is characterized by the absence of any lexical restriction. Thus che questions can be answered by an animate or inanimate noun phrase or by a proposition, as in (29). Of course a lexical restriction can be added by a noun phrase, when che ‘what’ appears as its specifier€– or by a sentence, when che appears as the ‘that’ complementizer. (29)â•… a.â•… Che c’ è? what there is ‘What is it?’

Interaction with (non-)finitenessâ•… 73 b. Il campanello ‘the doorbell’ b’. Il postino ‘the postman’ b’’. Suona il postino rings the postman ‘The postman is ringing’

By contrast, the wh–quantifier for ‘who’ is characterized by a restriction to humans, which poses an interesting question as to its ability to appear as a complementizer at all. In fact, the occurrence of, say, Miglionico tʃi in (20) as a complementizer excludes the possibility that the restriction to humans is encoded in the lexicon. What we suggest instead is that it is a lexical property of tʃi that it requires a restriction. In complementizer contexts the latter is provided by the proposition that tʃi introduces. In wh–contexts, on the other hand, the restriction to humans may represent an interpretive closure in the absence of lexical restrictions. If this line of reasoning is correct, we should be able to find languages in which the wh–quantifier for ‘who’ also appears as the wh–determiner, restricted by a following noun phrase. A case in point is Fontanigorda in (30), which, like Italian, has a single declarative complementizer overlapping with a wh–quantifier€– except that the overlap involves ‘who’, rather than ‘what’. As expected, the same element also appears as the wh–determiner, as shown in (30d). (30)â•… Fontanigorda (Liguria) a.â•… m aŋ ittu ke te vie dɔpu to.meâ•… they.haveâ•… saidâ•… thatâ•… youâ•… comeâ•… afterwards ‘They told me that you are coming afterwards’ b. ke te tSammi whoâ•… youâ•… call ‘Who do you call?’ c. kuɔseâ•… te fE what youâ•… do ‘What are you doing?’ d. ke kamiï†ıaâ•… te te bEtti which shirt youâ•… yourselfâ•… put.on ‘Which shirt are you putting on?’

2.3

The interaction with (non-)finiteness

The last question we shall consider in this chapter concerns yet another asymmetry between wh–quantifiers and complementizers, observed even in

74â•… Variation in Romance k-complementizer systems languages in which they have the same lexicalization. Thus, in standard Italian che ‘what’ introduces both finite and infinitival questions, as in (31a). In contrast, complementizer che is restricted to finite contexts, and infinitival embeddings involve, if anything, prepositional introducers like di ‘of’ in (31b). (31)â•… a.â•… b.

Che faccio/â•… fare? whatâ•… I.do/ to.do ‘What should one/I do?’ Mi hanno dettoâ•… di/*cheâ•… essereâ•… venuti to.meâ•… they.haveâ•… said to/that be come ‘They told me that they had come’

The problem is not a descriptive one, since complementizer che can be distinguished from wh–che on the basis of both the syntactic context of insertions and its semantic characterization. From an explanatory point of view, however, the objection may legitimately be raised that the connection between complementizers and finiteness crucially links them to the functional projections of the verb. Now, finiteness has two components€– one relating to tense, and another relating to agreement. As is well known, some Romance languages allow these two components to be distinguished, since they feature so-called inflected infinitives. A case in point is provided by Sardinian varieties, including Paulilàtino in (32). Both inflected and non-inflected infinitives can be introduced by a prepositional complementizer, as in (32b)€ – but inflected infinitives also allow for the finite complementizer, as independently observed by Jones (1993). The finite complementizer, then, is sensitive not to the temporal/modal/aspectual properties of the verb, but to the presence of an agreement inflection. As we have seen, Paulilàtino, like many Sardinian varieties, has two declarative complementizers, a modal and a non-modal one€– and it is the modal complementizer that co-Â�occurs with inflected infinitives. However, Manzini and Savoia (2005) also provide examples from varieties which have only one declarative complementizer, which also appears in front of the inflected infinitive (for instance, Siniscola). (32)â•… Paulilàtino a.â•… l appu attu innantis dE/ ki TrrarE- s tuE it I.have done before to/ that come.back-2sg you ‘I did it before you came back’ b. l an fattu innantis dE EnnErE-â•… (nE) /ki nnErE-nE i it they. have done before to come- 3pl that come-3pl they ‘They did it before they came’

In our theory (cf. in particular chapter 5) the so-called agreement inflection is a lexicalization of the EPP argument of the sentence internal to the verb. If

Interaction with (non-)finitenessâ•… 75 no lexicalization of the EPP argument is present, either by an inflection or by a lexical subject, as is the case for non-inflected infinitives, we assume that the EPP argument is introduced as a variable at the interpretive interface. This variable can then be given a bound reading (control/raising) or a generic reading (arbitrary control). In such terms, the restriction of the che-type complementizer to finite sentences€– including inflected infinitivals€– can be restated as an incompatibility between it and the EPP variable. The question then becomes why the finite complementizers would have this crucial property. We may surmise that the presence of an EPP variable within the sentence defines an open predicate, rather than a proposition. Since we have proposed that the finite complementizer of Romance, and presumably that of English as well, takes a proposition as its complement, we may conclude that what the complementizer is actually incompatible with is the open expression resulting from the presence of the EPP variable. Similarly, for Roussou (2010), the fact that in English that requires a propositional complement ultimately yields the that-t filter, precisely because it excludes the possibility that the EPP argument is a variable. In Roussou’s terms, the absence both of that and of the EPP gives rise to a case of predicate embedding€ – which allows the base-generated wh–phrase in the matrix left periphery to bind a variable in the argument structure of the embedded predicate. In fact, Roussou (2010) analyses both the *for-t and the *for–to filters as results of the same restriction, to the effect that for, like that, embeds a proposition, barring EPP variables. Though a discussion of the that-t filter is beyond the scope of the present work, it is worth recalling that, according to the classical conclusion of Taraldsen (1978), languages like Italian lack that-t filter effects (whence the apparent lack of correlation with the finiteness constraint on the complementizer) in that they are null-subject languages. We propose that what is at stake is simply the fact that while in non-null-subject languages the EPP must be satisfied by a syntactic-level argument (the subject), in null-subject languages a morphological-level argument (the inflection) suffices. The latter therefore closes off the proposition even in the case of wh–extraction (effectively acting as a resumptive pronoun). Similarly, non-null-subject Romance languages, in contrast to English, generally have subject clitics and resolve the propositionality requirement on chetype complementizers by lexicalizing one of them. On the basis of a comparison with Romansh varieties, Taraldsen (2002) and Manzini and Savoia (2005) in particular conclude that the€–i of French qui is a subject clitic of this type. A further twist on the finiteness problem is introduced by the ‘if’ complementizer. In a language like standard Italian, hypothetical se ‘if’ is restricted to

76â•… Variation in Romance k-complementizer systems finite sentences, as in (33a), while interrogative se normally occurs in front of infinitival questions, as in (33b). Now, recall that in Làconi in (19) the same tʃi form lexicalizes both the modal declarative complementizer and the polarity (i.e. hypothetical/question) one. As it turns out, while declarative tʃi is restricted to finite sentences, interrogative tʃi combines with infinitivals, as in (34). In the discussion surrounding (23), we interpreted the interrogative complementizer as a wh–element ranging over propositions in the scope of a question operator. We must conclude that this interpretation makes it compatible with embedded EPP variables. In this respect, ‘if’ in the wh–construal behaves exactly like other wh–elements, including the English wh–complementizer whether. (33)â•… a.â•… b. (34)

*Seâ•… piovere, … if to.rain Non so se uscire not I.know if to.go.out ‘I don’t know whether to go out’

Làconi n iSSiu tSi u tsErriai not I.know if him to.call ‘I don’t know whether to call him’

The issue is further complicated by the fact that while Italian se or Làconi tʃi can introduce infinitival questions, as in (33)–(34), neither English if nor French si can do so. In present terms, this means that French si and English if must embed complete propositions, very much like che or that, even if they are construed in the scope of an interrogative operator. The present analysis can be usefully compared with Kayne’s (1991). For Kayne, the finiteness constraint on complementizers reflects their incompatibility with the PRO subject of infinitivals; in other words, he substantially agrees with the view advocated here that the finiteness requirement relates to the EPP rather than to the temporal properties of the sentence. However, for him, what is responsible for this constraint is the structural government relation holding between the complementizer and PRO. The lack of such a constraint on Italian se correlates with a further structural property, i.e. with the relatively high position of the infinitive, which protects PRO from government by the complementizer. That the position of the infinitive is higher in Italian than in French is shown, according to Kayne, by the fact that the clitic precedes the infinitive in French, while it follows the infinitive in Italian. However, he acknowledges that his theory ends up not accounting for the incompatibility of Italian che with the infinitive, since on the evidence of cliticization the infinitive must be in the same high position in declaratives as in questions (1991:€95). In Làconi

Interaction with (non-)finitenessâ•… 77 the problem is even starker, since the same tʃi complementizer cannot combine with infinitival declaratives€– as predicted by Kayne on the basis of the preverbal position of the clitic seen in (19)€– but can combine with infinitival questions. Furthermore, in the account of Kayne (1991), whether is oblivious to the finiteness of the sentence it embeds because it is not a complementizer at all, but a wh–phrase€– where these two notions are given a structural characterization once more, as a C head and a C Specifier respectively. However, a structural characterization as a C Specifier cannot be applied to Làconi tʃi€– on pain of not being able to predict that (as a C head) it excludes non-finite sentences in declarative contexts.13 A different type of evidence concerning definiteness restrictions on Â�complementizers of the k-series comes from Northern Italian vÂ�arieties Â�(including some that display the standard Italian identity of form between the wh–quantifier for ‘what’ and the complementizer), which form questions by combining whâ•fi quantifier and complementizer, as discussed in detail in chapter 1 (especially section 1.2). Most of the relevant languages alternate between wh– che in finite sentences and the wh–element alone in infinitivals, as shown in (35) for Zoldo Alto. However, there are (a few) otherwise entirely comparable languages in which the wh– che cluster introduces both finite and infinitival questions, like Civate in (36)€– though the che-type complementizer is otherwise restricted to finite complements. (35)â•… Zoldo Alto (Veneto) a.â•… di-me ke ke te faTe tell-meâ•… whatâ•… thatâ•… youâ•… do ‘Tell me what you are doing’ b. no sai ke fa notâ•… I.knowâ•… whatâ•… to.do ‘I don’t know what to do’ (36) Civate (Lombardy) a. tSe ke te tSamet whoâ•… thatâ•… youâ•… call ‘Who are you calling?’ b. so mia tSe ke tSa’ma I.know not who that to.call ‘I don’t know who to call’

The pattern of Zoldo Alto is what we expect on the basis of the assumption that in the wh–che sequence, the che-type element is the ordinary complementizer; as such we predict it to be sensitive to the finiteness of the embedded clause and in particular to exclude an infinitival sentence, whose EPP

78â•… Variation in Romance k-complementizer systems argument is a variable under present assumptions. Let us consider, then, the case of Civate. Evidently the presence of the wh–quantifier in the left periphery of the ke complementizer is sufficient for the latter to also be interpreted in the scope of a question operator€– that is, as a wh–element of sorts ranging over propositions. If so, we expect it to have the property that we have independently reviewed for such elements above, that is, that of being insensitive to the finiteness restriction. Appendix (A)

Comp def

Comp indef/wh–

rel

who

what

Sonnino Pontecorvo Colledimacine Montenerodomo Torricella Peligna Secinaro Civitaluparella Vastogirardi Capracotta Guardiaregia Guglionesi Frigento Nocara Albidona Terranova Pollino Morano Orsomarso Conflenti Platania Gizzeria Sorbo S.Basile S.Pietro a Maida Iacurso Arena Umbriatico Gerace S.Agata del Bianco

ka ka ka ka ka ka ka ka ka ka ka ka ka ka ka ka ka ka ka ka ka ka ka ka ka ka ka

ke k k k k k k k k k k ke k k k ki ki ki ki ki ki ki ki ki ki ki ki

ke k k k k k k k k k/ka k ki k k k ki ki ki ki ki/ka ki ki ki ki ki ki ki

ki ki ki ki k k ki kia kia ki ki ki k kwE ku ku ku kinE kinE kina kinE ku ku ku kini ku ku

ke k k k k ku k k k k k ke k k k ki ki ki ki ki ki ki ki ki ki ki ki

(A’) Dorgali Aritzo Ardaùli Paulilàtino Gallo

ka ka ka ka ka

ki tSi ki ki ku

ki tSi ki€– ka ki ku

kiE (e)tSinE kie kiE ki

ittE ittE (e)itte ittE kwe

Interaction with (non-)finitenessâ•… 79 Tufillo

ka

k

k

kia

Canosa Sannita

ka

k

k

ki

(k) dEkk ki

(B) Làconi Orroli Settimo S.Pietro Sìliqua

comp ka ka ka ka

comp/ if tSi ki ki ki

rel tSi€– ka ki€– ka ki ki

who tSinni (e)kini kini (ak)kini

what itta itta itta itta

(C) Miglionico Minervino Murge Gravina in Puglia Bitetto Ruvo di Puglia Canosa Bisceglie Martina Franca Molfetta Mesagne Putignano Brindisi Grottaglie Carmiano Uggiano Copertino Melissano

comp ka ka ka ka ka ka kE ka ka ka ka ka ka ka ka ka ka

if tSi tS tSi tS tS tS tS tS tSi tSi tS tSi tSi Si tSi tSi tSi

rel ka ka ka ka ka ka kE ka ka ka ka ka ka ka ka ka ka

who tSi tS tSi tS tSi tS tS tS tSi tSi tS tSi tSi Si tSi tSi tSi

what tSe tSk attSok tS tS tSE tS tS tSE tSe tSE tSE tSe tSE tSE tSe tSi

3 Sentential negation:€adverbs

The currently standard approach to the structure of negation (in Romance) was inaugurated by Pollock (1989), who proposed that negation adverbs such as pas in French fill the Spec position of a NegP projection generated below the I position targeted by the verb. The head of NegP can in turn be filled by a negative clitic like ne in French, whose higher inflectional position depends on movement, i.e. cliticization. In other languages, which include colloquial French, no negative head is present. Belletti (1990) applies the same theory to Italian, which only has a negative head; the latter originates in the Neg position and moves higher as a result of cliticization. A more complex set of data, involving Northern Italian varieties, is considered by Zanuttini (1997), who proposes that there are several Neg positions. Specifically, a Neg position is generated above I, while below I there are three Neg positions. The inflectional Neg position hosts negative clitics in languages like Italian which do not require a sentential negation adverb. In contrast, languages which require a sentential negation adverb generate it in one of the lower Neg positions; if a clitic combines with the adverb, it is generated in the head of the relevant Neg position and moves to the inflectional domain via cliticization. In other words, for Zanuttini (1997) preverbal clitic negations are associated with two different structures, according to whether they negate alone or combine with a negative adverb. In turn, the lower Neg positions are defined in relation to the general hierarchy of adverbs proposed by Cinque (1999); according to Zanuttini (1997), her Neg2, Neg3 and Neg4 positions occur within the aspectual adverbial series, while they do not interact in any significant way with either the temporal or the modal series. Neg1 corresponds to the negative clitic position in the inflectional domain. According to Cinque (1999:€106), the aspectual adverbial series is ordered according to the hierarchy partially reproduced in (1a). The three adverbial negation positions proposed by Zanuttini (1997:€99) are ordered with respect to this hierarchy as in (1b). The three different Neg positions correspond to three different types of sentential negation adverbs, which Zanuttini individuates in 80

Sentential negation:€adverbsâ•… 81 Northern Italian varieties on the basis of their distribution. The higher adverbial Neg position Neg2 corresponds to sentential negation adverbs such as pa in Piedmontese varieties or the optional mica of standard Italian. In the same Piedmontese varieties in which pa occurs as Neg2, the Neg3 position is filled by sentential negation adverbs of the nen type. In turn, the lowest adverbial negation position Neg4 corresponds to the Lombard type no. (1)â•… a.â•… [Tanterior already [AspTerminative no longer [AspContinuative still [AspPerfective always…[Voice well b. [Neg2 [already [Neg3 [no longer [still [always … [well [Neg4

In closely related literature, the functional category Neg also plays a role in explaining interpretive facts, specifically ‘negative concord’, whereby two negative elements in the sentence do not give rise to a double negation, but rather are interpreted as instantiations of a single logical negation. Consider, for instance, a Romance language like Italian which has a negative clitic; the latter combines with a negative argument or adverb (niente ‘nothing’, mai ‘never’, etc.) to yield a single logical negation. Haegeman and Zanuttini (1991) explain this phenomenon on the basis of a Neg Criterion, modelled on the Wh–Criterion of Rizzi (1996). On the basis of the Neg Criterion, the negative clitic in the head position of a NegP requires a Neg operator in its Spec and vice versa. The head-Spec configuration is in turn read as an agreement relation, yielding a single negation interpretation. This analysis presupposes that niente, mai and the like€– i.e. n-words€– are negative quantifiers. A consistent body of literature on Romance languages argues for a different conclusion, namely that Romance n-words are negative polarity items (Rizzi 1982; Laka 1990; Longobardi 1992; Acquaviva 1994). In other words, n-words have no intrinsically negative properties, but are simply existentials, or free variables in the terms of Heim (1982), which are interpreted in the scope of the negation or another polarity operator. The most immediate argument in favour of the polarity status of n-words in Romance comes from the fact that they occur in modal environments (questions, hypotheticals) without any implication of negative meaning. The argument that is often advanced in favour of a negative quantifier status for n-words, namely that they appear in fragments (Zanuttini 1997), depends on the fact that fragments are the result of ellipsis€– and that they interpretively correspond to a full sentential structure, obviously capable of hosting an abstract negative operator. In what follows we assume that n-words are negative polarity items (and not negative quantifiers). This entitles us to (provisionally) disregard the role played by NegP in negative concord. We shall return to negative concord in chapter 4, where we shall also consider the sentential negation clitics which enter into it.

82â•… Sentential negation:€adverbs In this chapter we focus on sentential negation adverbs. Specifically, in Â� section 3.1 we note that sentential negation adverbs either take the same form as negative arguments such as ‘nothing’ or are bare nouns. We argue that treating them as nominal elements, specifically connected to the internal argument of the predicate, allows for an explanation of data such as the fact that they trigger the partitive, or their sensitivity to the person (1st/2nd vs. 3rd) of the internal argument itself. Correspondingly, we abandon the Neg categorization in favour of a reassignment of sentential negation adverbs to a nominal, argument-related category, tentatively labelled Q.1 In section 3.2 we examine the distributional phenomena targeted by the hierarchy in (1) – keeping in mind also the fact that recent literature (Ernst 2002; Svenonius 2002; Nilsen 2003) argues that they are sensitive not to syntactic hierarchies but to semantic constraints on selection and/or scope. In particular, we reassess the sub-hierarchy in (1) in the light of our treatment of sentential negation adverbs as both nominal and argumental. We conclude that the adverbs of the sub-hierarchy in (1) are attached to the eventive domain of the sentence in accordance with LF selection requirements and that they are ordered within this domain according to the same macrocategories encountered in the argumental and inflectional domains€– i.e. roughly by quantificational (Q), deictic (Loc) and nominal/argumental (N) properties. In section 3.3 we briefly reconsider another major distributional pattern concerning the position of adverbs with respect to verbs. Cinque (1999) and Zanuttini (1997) predict that any verbal head appearing internally to the hierarchy of adverbs in (1) merely cuts it into two segments, without any reordering of the adverbs relative to one another. We argue that if the ability of non-finite verbs to appear in the middle of the sequence in (1) is construed as an indication of their low position, then the fact that they notably precede pronominal clitics (i.e. inflectional material) remains unexplained. We conclude that the non-finite verbs are quite high in the sentence (participles also define independent sentences as in Kayne (1993)). The adverbs in (1) attach even higher because the lack of tense specifications allows for an eventive treatment of the higher projections of the sentence (along the lines of the flexible matching of syntactic and interpretive projections proposed by Svenonius (2002) and Ernst (2002)). Overall, then, we devote this chapter to providing a recategorization of sentential negation adverbs; the alternative account of their distributional properties is meant to establish the compatibility of this recategorization with known facts. As anticipated, we leave the recategorization of negation clitics and the reappraisal of negative concord for chapter 4.

Sentential negation adverbs are nominalâ•… 83 3.1

Sentential negation adverbs are nominal and argumental

In several Northern Italian varieties, notably Piedmontese ones of the type also studied by Zanuttini (1997), the sentential negation adverb has the same form as the negative argument for ‘nothing’. This is true of nEŋ as in (2), nEint(a) as in (3), nota as in (4), and rEŋ as in (5), as well as nia in Ladin varieties, as in (6). As a result of the variable valency of the verb, the (b) examples are ambiguous between the argumental and sentential negation interpretations; specifically, since the verb can be construed both transitively and intransitively, the negative item can fill the internal argument slot, or it can have a reading equivalent to a sentential negation. Conversely, it is only the valency of the verb that disambiguates the sentential negation readings in (a) and the argumental readings in (b). (2) Montaldo (Piedmont) a.â•… i u tSam nEŋ Iâ•… himâ•… callâ•… not ‘I don’t call him’ b. i mdZâ•… nEŋ I eat nothing/not ‘I don’t eat (anything)’ (3) Oviglio (Piedmont) a. a n El vig nEinta I notâ•… himâ•… seeâ•… not ‘I don’t see him’ b. u n maJdZaâ•… nEinta heâ•… notâ•… eats nothing/not ‘He doesn’t eat (anything)’ (4) Quarna Sotto (Piedmont) a. tSamuâ•… not -t I.call notâ•… you ‘I am not calling you’ b. j vOg nota I see no/nothing ‘I don’t see (anything)’ (5) Stroppo/Macra (Piedmont) a. lu tSamuâ•… rEŋ himâ•… I.call not ‘I don’t call him’ b. al fai rEŋ he does nothing ‘He does nothing’

84â•… Sentential negation:€adverbs (6)â•… La Pli de Mareo (Alto Adige/South Tyrol) a.â•… i ne le kErdiâ•… nia Iâ•… notâ•… himâ•… call not ‘I don’t call him’ b. ju ne li nia I notâ•… readâ•… not/nothing ‘I don’t read (anything)’

The common lexicalization of the sentential negation adverb and negative arguments is recognized in the historical literature, specifically in connection with what is taken to be a ‘grammaticalization’ process changing the negative argument into a functional category Neg. Thus, Roberts and Roussou (2003) mention ‘negative quantifiers/pronouns’ as one of the ‘principal sources for clausal negators’ (English not is an example of this). In fact, there are many Northern Italian varieties in which the types nen, neinta, etc. for the sentential negation do not have the same form as ‘nothing’, so that the connection between the former and negative arguments appears to be purely etymological. However, varieties of the type in (2)–(6) seem to point to a bona fide lexical identity of negation and negative argument. Therefore we may wonder whether the best theory is one which posits two lexical entries for them, with the negation associated with the category Neg and the argument with the category N/Q. In fact, only a subset of the negative elements reviewed so far contains the negative morphology n-; thus, the type rEŋ is etymologically connected to a bare noun (
Trun (Grisons) Elts dOrmŋâ•… bk(a) theyâ•… sleep not ‘They don’t sleep’ Pramolloâ•… (Piedmont) a drm pa he sleeps not ‘He doesn’t sleep’ Premanaâ•… (Lombardy) elâ•… fo: miE it I.do not ‘I don’t do it’ Finale Emiliaâ•… (Emilia) i n dramâ•… briʑa theyâ•… notâ•… sleep not ‘They don’t sleep’

Sentential negation adverbs are nominalâ•… 85 Interestingly, in relation to sentential negations of the type in (7) MeyerLübke (1899:€§693–4) proposes that what we describe as sentential negation adverbs originate in a partitive construction. In support of his proposal he quotes Old French examples such as (8), in which the ‘negative adverb’ mie, a bare noun ‘minimizer’, overtly co-occurs with the partitive. (8)â•… de s’espee ne volt mieâ•… guerpir of his swordâ•… notâ•… he.wanted not to abandon ‘He didn’t want to abandon his sword’

(Chanson de Roland 465)

This type of data recalls the phenomenon described by Pesetsky (1982) for Russian whereby the accusative object in non-negative contexts alternates in negative contexts with an object morphologized either in the accusative or the partitive, as in (9). (9)â•… a.â•… b.

Ja ne polucal pis’ma I not received letters(acc.pl.) ‘I didn’t receive letters’ Ja ne polucal pisem I not received letters(gen.pl.) ‘I didn’t receive letters’

Considerable interaction between the negation and case assignment to the internal argument of the verb is also found in Northern Italian varieties. Given its potential relevance for the status of the negation (purely functional or still nominal), we shall review it next. In one group of languages, which includes the Piedmontese varieties in (10), the negation triggers the partitive even in the presence of a definite interpretation. Thus, the partitive clitic in (10) doubles the proper name in topic position, and indeed can normally alternate with an accusative, as in (10a). It will be noted that the languages exemplified are characterized by enclitics in declarative finite sentences (cf. Tortora 2002). However, the variety from Quarna Sopra exemplified in (15) combines the same phenomenon with proclisis on the finite declarative verb, showing that it cannot depend on enclisis. (10)â•… a.â•… b.

Trecate (Piedmont) (a mmarju) tSamumâ•… -ru/â•… -na mija the Mario we.call him/of.himâ•… not ‘We are not calling Mario’ Cerano (Piedmont) Marjoâ•… tSamaâ•… -n mea Mario call of.him not ‘Don’t call Mario!’

The data in (10) shed a new light on a different, but fairly obviously connected phenomenon that has long been known for French. Thus, as discussed

86â•… Sentential negation:€adverbs by Kayne (1984), in negative environments French allows for indefinite noun phrases (bare plurals and bare mass singulars) introduced by de; these same forms are not allowed in the absence of negation, as illustrated in (11). The same phenomenon is fairly widespread in Northern Italian varieties, as shown in (12). (11)â•… Jeâ•… *(ne)â•… veuxâ•… pas de cadeaux I not want notâ•… ofâ•… gifts ‘I (don’t) want gifts’ (12) a. b. c. d.

Casorezzo (Lombardy) al maJdʒaâ•… miŋgaâ•… da biskOti heâ•… eats not of biscuits ‘He doesn’t eat biscuits’ Casei Gerola (Lombardy) u bev mi d vi he drinks not of wine ‘He doesn’t drink wine’ Cravagliana (Piedmont) al maJɟa mia da buskuti heâ•… eats not of biscuits ‘He doesn’t eat biscuits’ Stroppo/ Macra al beu rEŋ de viŋ heâ•… drinksâ•… notâ•… of wine ‘He doesn’t drink wine’

For the object noun phrase in (11), Kayne (1984) proposes a structure including a non-lexicalized negative quantifier Q followed by the partitive de cadeaux, yielding a structure of the type je ne veux pas [ Q [de cadeaux]]. The presence of the empty Q quantifier means that the noun phrase as a whole is subject to the Empty Category Principle (ECP) of Chomsky (1981); this in turn predicts a generalization that appears to hold at least of French, namely that noun phrases of the type under consideration are restricted to the object position. This proposal of Kayne (1984) can be considered a predecessor of his recent ideas about ‘silent’ categories (Kayne 2006, 2008a). Manzini and Savoia (2009a, 2010) and Savoia and Manzini (2010) argue in some detail against this conception on grounds of restrictiveness of the theory, as well as on specific empirical grounds. Here we note only that Chomsky’s (1981) ECP is formulated for empty categories whose content is determined in that framework by the computation, not by the lexicon€– so that the extension of the ECP to empty Qs of the type in Kayne (1984) in fact represents a substantial revision of the theory.

Sentential negation adverbs are nominalâ•… 87 In at least one of the languages in (12), namely Stroppo in (12d), the sentential negation adverb rEŋ has the same form as the negative argument, as illustrated in (5). This suggests an alternative analysis of the partitive in (12). Suppose that we take the fact that ‘not’ and ‘nothing’ in languages like Stroppo have the same form not to be just a matter of homophony (eventually to be explained on diachronic grounds), but rather as an indication of the fact that a single lexical entry (with non-disjunctive properties) is involved. If we take the negative argument content to characterize the lexical entry of rEŋ as a whole, we obtain the Logical Form in (13) for (12d); this is endowed with the interpretive properties indicated by Kayne (1984), but without any recourse to silent Q elements. Quite simply, we can assume that the role of the negative quantifier licensing the partitive is played by rEŋ itself. To be more precise, rEŋ introduces the polarity variable existentially closed within the scope of the polarity (here negation) operator. (13)â•… Stroppo/ Macra al beuâ•… [Q rEŋ]â•… [de viŋ]

For the data in (10), we envisage an analysis along the lines of that provided for (12). In fact (10) represents a generalization of the structures already proposed for (12), where no definiteness restriction applies. On the contrary, we must assume that the elements that can enter into the partitive construction in (12) are sensitive to the same constraints that generally restrict the occurrence of (object) bare nouns in Romance to plurals and mass singulars. The proposal in (13) raises many questions, some of which can be settled on the basis of fairly standard assumptions. For instance, the fact that rEŋ de viŋ in (13) is not a constituent can be viewed as a result of the fact that rEŋ as a sentential negation is in a Q-floated position. Other questions require more articulated answers. In particular, one may wonder what the difference is between ‘He doesn’t drink’ and ‘He drinks nothing’ if ‘not’ and ‘nothing’ coincide in the lexicon; we shall return to this question in section 3.1.2. Genitive of negation in Russian (9) arises with a negative clitic, rather than with a negative adverb; the reason will become apparent when we discuss negation clitics in chapter 4. In short, we have argued that the lexical formal identity between negative arguments and sentential negation adverbs in (2)–(6) supports a single lexical entry for these elements. We have further argued that this analysis provides an explanation of why sentential negations license partitives, as in (10)–(12). More generally, our idea is that the same analysis applies to all sentential negation adverbs€ – and not only to those of which the relevant phenomena hold.

88â•… Sentential negation:€adverbs If negations of the type of nEn, nEinta, nuta illustrated in (2)–(4) have the referential content of the negative argument ‘nothing’ (with which they may lexically coincide), one may wonder what the referential content may be for mia in (12a–c), (10) and for similar minimizers (briza, pa, etc). In fact, they have the only reference independently known to be compatible with bare singular count Ns, i.e. again that of negative (polarity) items. To illustrate, bare singular Ns, though generally excluded in Romance, become possible in the scope of a negative operator, as illustrated in (14) for standard Italian. Note that what is crucial for the argument in the text is that bare count singulars are allowed only in negative (polarity) contexts; it is not crucial that all negative (polarity) contexts should license all bare count singulars.2 (14)â•… a.â•… b. c.

*(Non)â•… si muoveâ•… foglia not M/Pâ•… moves leaf ‘Not a leaf stirs’ *(Non)â•… alzò ciglio not he.raisedâ•… eyebrow ‘He didn’t raise an eyebrow’ *(Non) proferì parola not he.said word ‘He didn’t say a word’

3.1.1 Further evidence Before we proceed with a discussion of the proposal we are putting forward, and the various issues it raises, we will review some more complex evidence relating to the interaction of sentential negation adverbs with the argument structure of the predicate. The varieties of Quarna Sopra and Quarna Sotto both alternate between a ‘nothing’-type adverb, i.e. nota, and a bare-N adverb, namely mia, as illustrated in (15) and (16) respectively. Consider first Quarna Sopra in (15). In this language, nota appears in intransitive (unergative) contexts, as in (15a), as well as in the middle voice (reflexive), as in (15b). In transitive contexts, nota appears when the internal argument is a 1st or 2nd person pronoun, as illustrated by (15c). With 3rd person internal arguments, nota Â�co-occurs with accusative lexical objects and clitics, as in (15d–d’), whilst with partitive clitics and with lexical objects introduced by the partitive preposition di ‘of’, we find mia, as in (15e). Note that the partitive (15e) has an interpretation that makes it equivalent to the accusative (15d). (15)â•… Quarna Sopra (Piedmont) a. i drm/ t drum/ G drm/ i drumma/ i drumi/ i drmn I sleep/ you sleep/ he sleeps/ we sleep/ you sleep/ they sleep ‘I don’t sleep’ etc.

nt not

Sentential negation adverbs are nominalâ•… 89 b. m lau nt/ t lavt nt/ s lava nt me I.wash not/ you you.wash not/ himself he.washes not ‘I don’t wash myself’ etc. c. m/t vgn nt me/you they.see not ‘They don’t see me/you’ d. i camnâ•… ntâ•… u te fril theyâ•… call not the your brother ‘They don’t call your brother’ d’. u vgn nt him they.see not ‘They don’t see him’ e. n camn miaâ•… d u te fril of.himâ•… they.callâ•… not ofâ•… theâ•… yourâ•… brother ‘They don’t call your brother’ e’. n vgn mi of.it/of.them they.see not ‘They don’t see (of) it/them’

In the variety of Quarna Sotto in (16), nota appears with intransitive (unergative) verbs, as in (16a), and in the middle (reflexive), as in (16b). Interestingly, unaccusatives, as in (16c), can split between nota and mia according to person, so that 1st and 2nd person co-occur with nota, while 3rd person co-occurs with mia. In transitive contexts, mia and nota again split according to person. Thus, nota occurs with 1st and 2nd person objects, as in (16d), while mia co-occurs with 3rd person objects, either in the form of a partitive, as in (16e) and (16f’), or of an accusative lexical object, as in (16f). Accusative clitics do not occur in negative contexts. For enclisis of the clitic to declarative finite verbs, we refer the reader to the brief discussion of this phenomenon in relation to the examples in (10). (16)â•… Quarna Sotto a.â•… j drmma/ d drumma/ G drmma/ j drumoma/ j drumi/ drmu nota I sleep/ you sleep/ he sleeps/ we sleep/you sleep/â•… they.sleep not ‘I don’t sleep’ etc. b. jâ•… lau not-m/ d lEu not-t/ G lava not-s I washâ•… not-me/ you wash not-you/ he washes not-himself ‘I don’t wash myself/ you don’t wash yourself/ he doesn’t wash himself’ c. sum/ t i/ soma/si Ju nota I.am/you are/â•… we.are/you.are come not ‘I haven’t come’ etc. c’. i mia Ju/ Ju nota he.is not come/ come not ‘He hasn’t come’

90â•… Sentential negation:€adverbs d. â•… vgâ•… notɐ -m/â•… -t/â•… -u he sees not me/ you/ you(pl) ‘He doesn’t see me/you’ e.  vg mi-n he sees not-of.it/them ‘He doesn’t see it/ them’ f. lavu mia  kamiz they.washâ•… not the shirts ‘They don’t wash the shirts’ f’.╇  beu mi d viŋ he drinks not of wine ‘He doesn’t drink wine’

Person split phenomena are pervasive in the Romance languages. For instance, in the clitic system, 1st and 2nd person forms distinguish at most between nominative and non-nominative; 3rd person forms are the only ones to register any difference between accusative and oblique, both in the morphology and in the position of the clitic.3 Similarly, it will be seen from (15b) and (16b) that only the 3rd person has the specialized middle (reflexive) marker si. We surmise that, although a 1st or 2nd person element and a 3rd person one can equally serve as arguments of a predicate, they do so through different syntactic means. In particular, the position and the morphology of 1st and 2nd person arguments are not necessarily sensitive to their anchoring in the event structure (i.e. whether they are the first or second internal argument of a ditransitive, etc.). In this sense, we speak of ‘discourse-anchored’ elements, i.e. elements whose position and morphology are sensitive only to their denotational content. By contrast, the morphosyntactic properties of 3rd person elements reflect their argumental role (accusative vs. dative marking, specialized middle marking by si, etc.). In this sense, we speak of ‘event-anchored’ elements.4 Let us consider Quarna Sotto in (16). In present terms, mia selects for event-anchored internal arguments. The latter include 3rd person accusatives and partitives, as well as the 3rd person nominative arguments of unaccusatives. In turn, nota appears in the complementary set of environments, i.e. those where there isn’t an event-anchored internal argument. Therefore it combines with discourse-anchored (i.e. 1st and 2nd person) internal arguments and with predicates, i.e. unergatives, that either have no internal argument or, as we assume here, have an incorporated one (Hale and Keyser 1993). Reflexives are of particular interest; if the distribution of mia and nota was simply sensitive to a person split, we should expect the reflexive to have nota in the 1st and 2nd and mia in the 3rd€– whether reflexives are unaccusative structures (Marantz 1984) or transitive structures, with the reflexive clitic playing the role

Sentential negation adverbs are nominalâ•… 91 of the internal argument (Burzio 1986) (cf. chapters 5–6). On the other hand, many independent (for instance distributional) facts point to the conclusion that si patterns with 1st and 2nd person object clitics rather than with 3rd person accusatives. The notion of discourse anchoring is meant to capture this fact as well; specifically, si is characterized as the free variable of the system (as in chapters 5–6 here) and is dissociated from the event-anchored set on the basis of its indefinite/quantificational nature. Therefore we account for the fact that 3rd person reflexives will pattern with 1st or 2nd person internal arguments in being selected by nota. The interaction of the negation adverb with the person split is hardly expected if the adverb corresponds to a functional category Neg€– whose content presumably is that of the logical connective of negation. In other words, it is hard to find a reason why the logical operator of negation would be lexicalized in two different ways according to the argument structure of the verb. We are not saying that the interaction of the negation adverb with the argument structure of the verb cannot be stated in terms of a conventional theory treating the adverb as an instance of logical negation. What we are saying is that any such connection would remain a pure stipulation. Suppose, however, that, as outlined in section 3.1, the sentential negation adverb is in reality a nominal element bearing a particularly close relation to the internal argument of the verb (with which it can lexically coincide). Suppose, furthermore, as also suggested in section 3.1, that the negative adverb can be interpreted to the extent that it forms a partitive structure with the internal argument€– eventually triggering an overt partitive case on it, as seen in Quarna Sotto itself. From this perspective, connecting the so-called sentential negation to the nominal, argumental set of categories, its interaction with the argumental structure of the predicate is not surprising and in fact expected. Consider Quarna Sopra in (15), then. In this language, mia is found in structures with 3rd person partitive internal arguments, while nota occurs in all other structures, including those with 3rd person accusatives, 1st/2nd person objects, intransitives (unergatives) and middles (reflexives). Because partitive structures are restricted to 3rd persons, it is evident that the distribution found in Quarna Sopra presupposes a person split analogous to the one examined for Quarna Sotto; in other words, Quarna Sopra further restricts the distribution of mia observed in Quarna Sotto to event-anchored (i.e. 3rd person) partitives. The distribution of mia in Quarna Sopra again argues in favour of the nominal, argumental status of the sentential negation. Thus, although it can be stipulated that two different Neg functional categories, whose content is presumably that of logical negation, are inserted depending on the presence of a partitive, 3rd

92â•… Sentential negation:€adverbs person internal argument or on its absence, there is no connection between the two sets of facts, and hence no explanation why such a state of affairs holds. By contrast, an explanation of this state of affairs follows from the view of negation that we are developing here, which connects the so-called sentential negation to the nominal, argumental set of categories. 3.1.2 Sentential negation adverbs as nominal arguments On the basis of the preceding evidence, so-called negation adverbs are nominal categories.5 Furthermore, the interactions of the sentential negation with the internal argument of the verb reviewed above (the fact that the negation is lexicalized differently according to the person of the internal argument, the partitive under negation, the ambiguity between adverbial and argumental reading of ‘nothing’) point to the conclusion that the negation is connected to the internal argument slot. This conclusion is supported by the fact that from a purely truth-functional perspective, negating the internal argument, as in, for example, I ate nothing, is equivalent to negating the sentence, e.g. I didn’t eat. At the same time, negating the internal argument and negating the sentence as a whole cannot simply be identified. Consider the simple case in which the sentential negation co-occurs with lexicalization of the internal argument by a noun phrase, as in I didn’t eat the apple. The analysis that we propose is based on languages in which the negation selects a partitive rather than an accusative internal argument. In this case it is evident that the sentential negation can be construed as introducing a quantification over the internal argument€ – which correspondingly is lexicalized as a partitive. As already sketched in section 3.1, in connection with (13), we extend this analysis to all cases where the sentential negation co-occurs with an overt lexicalization of the internal argument. Thus we take it that examples such as (16d–e) from Quarna Sotto, independently of the actual presence of a partitive, have the same Logical Form, as schematized in (17a) and (17b). In (17), N labels the partitive and P labels 1st /2nd person. In accordance with reasonable assumptions about the nature of elements selecting the partitive, we impute to the negation the (nominal) categorization Q, which characterizes weak quantifiers in the structure of DPs, as we have already done in (13). The logical negation operator in whose scope the polarity elements not and mi  are interpreted is notated by ⌐.. (17)â•… Quarna Sotto a.â•… [⌐â•… [D  [I vg [Q not [1/2Pm]] b. [⌐ [D â•… [I vg [Q mi[N n]]

Sentential negation adverbs are nominalâ•… 93 On the basis of the analysis in (17), we also expect that the sentential negation interpretively combines with the internal argument of unaccusatives, despite the fact that this is lexicalized as the EPP argument. As for as the contrast mentioned above between I ate nothing and I didn’t eat, the crucial observation is that two different argumental frames of to eat are involved. The nothing example implies a transitive argumental frame, where nothing satisfies the internal argument. On the other hand, the not example involves unergative to eat, which, following Hale and Keyser (1993), we construe as a concealed transitive with an incorporated internal argument. From this perspective, not again introduces a quantification over the internal argument, which in this case is incorporated. We can make this more precise if we assume, along the lines of Hale and Keyser (1993) and Chomsky (1995), that an unergative predicate and more generally a transitive predicate (of which the unergative is a subtype) consists of two layers. The most embedded layer (headed by V in Chomsky (1995)) corresponds to an elementary event, while the outer layer (headed by v in Chomsky (1995)) corresponds to a causation imputed to the external argument. Thus the incorporated internal argument of unergatives is the elementary event to which the external argument applies. But if so, transitive structures also include an elementary event to which the external argument applies, which can be construed as an internal argument as well. The only difference with respect to unergatives is that the internal argument of the elementary predicate becomes the internal argument of the transitive predicate as a whole, much as in overt incorporation structures such as Romance (or better Bantu) causatives. This allows us to refine the account proposed for sentences like (17) or English I didn’t eat the apple€ – which we now predict to be ambiguous. If the negation quantifies over the internal argument of the elementary event, we obtain the reading ‘It was not the apple that I ate’ or ‘No situation of me eating anything was a situation of me eating the apple’. If the negation quantifies over the elementary event, we obtain the reading ‘It was not eating the apple that I did’ or ‘No situation of me doing anything was a situation of me eating the apple’. As we shall see in section 3.2.2, this type of ambiguity characterizes quantificational adverbs in general, and, following de Swart (1993), quoted by Ernst (2002:€347), it is connected to the focus Â�structure of the sentence. Since unaccusatives reduce to elementary events, according to Hale and Keyser (1993) and Chomsky (1995), the latter is of course what is quantified over€– or rather, its internal argument. In the present approach, as a rule, we avoid building interpretive structure into the syntax in the absence of independent morphosyntactic evidence. Now, no such evidence seems to us to be forthcoming in the case of the v-V structure

94â•… Sentential negation:€adverbs of Chomsky (1995) or the incorporation of V into v. For instance, all evidence pertaining to the postverbal subject in Italian, i.e. the in situ subject under the VP-internal subject hypothesis, seems to indicate that it shares the same predicative low domain of the sentence with the internal (and other) arguments. In other words, we object to the functional category v very much on the same grounds as we object to (much) functional structure in the rest of this work€– namely that it simply provides a syntactic encoding for what is in reality an interpretation. In this sense it is at best redundant€– and at worst it potentially obscures the real nature of the phenomena it seeks to capture. Therefore we maintain a syntax in which the elementary event corresponds to the verb, or the verb plus its internal argument€– while the complex predicate (causative) reading simply arises by composition of the external argument with the elementary predicate. 3.2

Ordering sentential negation with respect to other adverbs

Suppose that sentential negations are nominal in nature and attached to the internal argument position of the verb€– as argued in section 3.1. The question still arises of how the present theory can accommodate the evidence presented by Zanuttini (1997) and Cinque (1999) concerning the relative order of several types of adverbs and other adverbial material. As mentioned at the outset, Zanuttini and Cinque seek to account for the data by postulating three Neg positions ordered within the aspectual adverb (sub)hierarchy as in (1). The evidence is somewhat complicated by the fact that many varieties have at least two different sentential negation adverbs, which Zanuttini characterizes as presuppositional vs. non-presuppositional. In general, it seems to us that such a distinction is difficult to maintain, given that all instances of negations, and more generally of focus, must have a presupposition as part of their interpretation. In what follows we will simply refer to what Zanuttini calls the presuppositional adverb as the mica-type adverb, since mica lexicalizes this type of adverb in standard Italian. We begin by reviewing the data€– the reader interested in the analysis can skip directly to the table in (38) and to section 3.2.1. In some varieties, negation adverbs precede the aspectual series including ‘already’, ‘any longer’, ‘still/yet’ and ‘always’. A case in point are varieties like Castellazzo in (18), in which this position characterizes the mica-type adverb mEj in (18b) as well as the ordinary sentential negation adverb nEit in (18a). Other varieties have a single sentential negation adverb to cover the two contexts€– and once again this occurs before the entire aspectual series of adverbs.

Ordering sentential negation and other adverbsâ•… 95 This is illustrated in (19) for a variety in which the negation adverb is mia and in (20) for a variety in which it is pa. Quite straightforwardly, all of these adverbs would correspond to the Neg2 position of Zanuttini (1997) and Cinque (1999). (18)â•… Castellazzo Bormida (Piedmont) a.â•… a ŋ l u nEitQ zQ/ pi/ aŋkurQ/ d lNk I not it have not already/ any longer/ yet/ always ‘I haven’t done it already/ any longer/ yet/ always’ b. a ŋ l u mEjQ zQ/ pi/ aŋkurQ/ d lNk I not it have not already/ any longer/ yet/ always ‘I haven’t done it already/ any longer/ yet/ always’

fa:tS done fa:tS done

(19) Pozzaglio (Lombardy) a. El drma mia p/ Jam/ sEmper/ bE: he sleeps not any longer/ yet/ always/ well ‘He doesn’t sleep any longer/ yet/ always/ well’ b. El g a mia bEle durmi:t he Loc has not already slept ‘He hasn’t already slept’ (20) Pomaretto (Piedmont) a. a drm pa bŋ/ pi/ ŋkar/ sampre he sleeps not well/ any longer/ yet/ always ‘He doesn’t sleep well/ any longer/ yet/ always’ b. al a pa dZO dyrmi he has not already slept ‘He hasn’t already slept’

The evidence becomes less straightforward when we consider sentential negation adverbs that appear inside the aspectual series. The hierarchy in (1) leads us to expect that, insofar as they instantiate Neg3, they will follow ‘already’ and precede the lower adverbs. In reality, in a language like Cantoira in (21), in which the sentential negation adverb follows ‘already’, we observe that it also follows ‘any longer’ in (21a) and (21c), and ‘still/ yet’ in (21a’) and (21c), while preceding ‘always’ and ‘well’. The mica-type adverb is unproblematic, since it has the distribution already observed in (18)–(20); i.e. it could correspond to Neg2 in Zanuttini’s (1997) terms. The relevant examples are given in (22). (21)â•… Cantoira (Piedmont) a. u dyrt dï†ı pi Jiŋ he sleeps already any longer not ‘He already doesn’t sleep any longer’ a’. u miJdZunt aŋku Jiŋ they eat yet not ‘They don’t eat yet’

96╅ Sentential negation:€adverbs b.╅ c. d.

u miJdZunt Jiŋ biŋ/ sEmp they eat not well/ always ‘They don’t eat well/ always’ u l ont pi/ aï†−ku Jiŋ tSama: they him have any longer/ yet not called ‘They haven’t called him yet/ any longer’ u l ont Jiŋ sEmpe tSama: they him have not always called ‘They haven’t always called him’

(22)â•… Cantoira a.â•… u l ont pa pjy/ Jku/ sEmpe tSama: they him have not any longer/â•… yet/ alwaysâ•… called ‘They haven’t called him yet/ any longer/ always’ a’. u l ont pa dï†ı miï†−dZa they it have not already eaten ‘They haven’t already eaten it’

Zanuttini (1997:€74) considers the ordering of nen ‘not’ after pi ‘any longer’, arguing that pi nen ‘no longer’ is a single constituent, placed in the position otherwise occupied by ‘any longer’ alone. However, even if Zanuttini (1997) was right about pi nen, we would be led to conclude that the sentential negation adverb is lower in the hierarchy in (1) than Neg3€– on the basis of the fact that it also follows (a)nku ‘still/yet’. This conclusion somewhat diminishes the interest of discussing the issue of whether pi nen is in fact a constituent, because, on the basis of a lower ranking of nen in the hierarchy, we would automatically predict that it follows pi. However, Zanuttini (1997) deploys a number of arguments in favour of the single constituent status of pi nen, which deserve some discussion. A classical constituency test applied by Zanuttini involves the impossibility of inserting any lexical material between pi and nen. But in infinitival contexts, where the negation adverb is in preverbal position, pi and the like can actually follow the verb, as seen in (23) for Cantoira itself, as well as other varieties. Of these, Margarita and Mezzenile have the same pattern as Cantoira in finite sentences.6 As for Montaldo, a comprehensive set of data is provided in (36)–(37) below. Note that our argument cannot simply be dismissed by noting that the string in (23) is nen … pi rather than pi … nen, as we are arguing for the possibility that these two elements represent separate constituents, irrespective of their relative order. (23)â•… a.â•…

Cantoira dZi t e dit t Jiŋ tSama-lu I to.youâ•… have said to not call-him ‘I told you not to call him any longer’

pjy any longer

Ordering sentential negation and other adverbsâ•… 97 b. c. d.

Mezzenile (Piedmont) t Ei di t Jint tSamaâ•… pjy Jy:n to.youâ•… I.haveâ•… saidâ•… to not call any longer anybody ‘I told you not to call anybody any longer’ Margarita (Piedmont) j  di-te t nEŋ tSamE-u pi I I.have told-you to not call-him any longer ‘I told you not to call him any longer’ Montaldo i sJ kuntEnta yd nEŋ avEi pi palQ I am happy to not have any longer spoken ‘I am happy not to have spoken any longer’

Furthermore, it is true that some varieties feature an incorporated form of pi nen; Zanuttini cites pin for Bollengo, while our data include, for instance, pjiŋ in Piverone (Manzini and Savoia (2005)). However, the argument could be turned against Zanuttini’s proposal. Thus, only a pin-type form can be rightly considered as a single lexicalization of ‘no longer’. By contrast, a language which has the combination pi nen can be argued to lexicalize the negative adverb ‘no longer (no more)’ literally by combining two constituents, namely ‘more’ (pi or più in standard Italian) and the negative adverb. Working within the framework of Zanuttini (1997), it is also possible to describe the fact that the negative adverb precedes anku ‘still/yet’ by saying that anku is incorporated into the negation. The proposal is no less plausible than the idea that pi incorporates, since in many Northern Italian varieties there exists a negative counterpart of ‘still/yet’, represented here by am in Pozzaglio, as in (19) (to be compared with non-negative am in S. Angelo, as in (27d)). Yet one may wonder why Zanuttini (1997) doesn’t simply reformulate the hierarchy in (1) by positioning Neg3 further down, below ‘still/yet’. The answer has largely to do with the distribution she observes in infinitives. In her data, the negation adverb of the nen type can only precede the infinitive, while pi nen can either precede or follow. She takes this to indicate that nen is higher in the adverb hierarchy than pi nen (treated as a single constituent). Thus, if the infinitive is positioned between Neg3 and the ‘any longer’ Asp position, the order nen€– infinitive€– pi nen is derived. The fact that pi nen can also precede the infinitive simply means that the verb can move slightly lower. In proposing this analysis, Zanuttini (1997) extends to Piedmontese varieties the schema of explanation proposed by Pollock (1989) for French infinitives; for, according to Pollock, the fact that French infinitives follow pas means that they move to a relatively low position, lower in any event than the position of the finite verb (which precedes pas). One difficulty with this extension, as

98â•… Sentential negation:€adverbs already noted by Kayne (1991), lies in the fact that while French has proclisis on the infinitive, Piedmontese varieties have enclisis, as illustrated in (23) for Cantoira and Margarita. The simplest account for this fact is that the infinitive (like other modal forms of the verb, e.g. the imperative) fills a relatively high, C-related position. But if so, the ordering of adverbs with respect to non-finite verbs cannot be explained by the low position of the verb within the adverb hierarchy. We return to these matters in section 3.3. What is also directly relevant here is that in our data, for instance the Margarita examples, nen can occur not only before the infinitive, as noticed by Zanuttini (1997), and illustrated here in (24a), but also after the infinitive, as in (24b). Since pi nen can also occur before the infinitive, again as noticed by Zanuttini and illustrated here in (24c), this means that even if we adopt Zanuttini’s (1997) analysis of the position of the infinitive, the language provides no evidence in favour of pi nen occurring in a lower position than nen. (24)â•… Margarita a.â•… j  di-te t nEŋ tSamE-u I have told-you to not call-him ‘I told you not to call him’ b. j  di-te t tSamE-u nEŋ I have told-you to call-him not ‘I told you not to call him’ c. j  di-te t pi nEŋ tSamE-u I have told-you to any longer not call-him ‘I told you not to call him any longer’

In short, the evidence reviewed seems to provide no support for Zanuttini’s (1997) conclusion that the position of nen inside the aspectual string corresponds to the Neg3 position in (1). As far as we can see, it can be made compatible with such a hypothesis, given, for instance, a high enough degree of optionality in the movement of the non-finite verb, in order to account for the data in (24). But, of course, making the movement of the non-finite verb highly optional reduces the predictive power of the model, eventually turning it into a series of descriptive statements. An equally (or more) problematic case is represented by the third class of negations individuated by the hierarchy in (1)€– which, according to Zanuttini (1997), corresponds to no adverbs found in the Milan and Pavia varieties. In particular, in varieties like Milan or Casorezzo, the latter of which is illustrated in (25), the sentential negation no precedes relatively high aspectual adverbs such as ‘always’ and ‘still/yet’ in (25a–c’), while it follows an adverb like ‘well’ in (25a), which is much lower in the hierarchy in (1). The logic of the

Ordering sentential negation and other adverbsâ•… 99 hierarchy, of course, is that an element like no which follows ‘well’ should follow other aspectual adverbs. This unexpected distribution combines with another property not observed before, namely that no is not positioned between the auxiliary and the participle as in the varieties considered so far (cf. for instance (18), (22)). Rather, it follows the participle, as in (25c). (25)â•… Casorezzo a. al drmi ben n:€ se:mpr he sleeps well not always ‘He doesn’t always sleep well’ b. ly la vedi ben n ŋkamo he it sees well not yet ‘He doesn’t see it well yet’ c. l a drmiâ•… ben n: he has slept well not ‘He hasn’t slept well’ c’. l a drmiâ•… n ŋkamo he has slept not yet ‘He hasn’t slept yet’

Zanuttini (1997) argues that the overall pattern of Milanese (or Casorezzo, as illustrated here) can be explained by assigning no to the Neg4 position, lower than ‘well’. An argument in favour of this conclusion is provided by the Â�participle. According to Zanuttini, this moves to a position immediately higher than ‘well’, leaving the low Neg4 negation to its right. Other adverbs, which are higher in the hierarchy, appear to the left of the participle, i.e. between the participle and the auxiliary. Zanuttini acknowledges the problem represented by the positioning of no in front of ‘always’ and ‘still/ yet’ as in (25a–b). In these examples, she proposes that no actually fills a different position, namely the Neg2 position associated with the negation adverbs in (18)–(20) and with the mica-type adverb in varieties like (22). Interestingly, she argues that a no appearing in Neg2 is interpreted as a mica-type adverb. Yet this analysis would lead us to expect that when no precedes ‘always’ or the aspectual adverbs, it also precedes ‘well’. Data of this type are indeed attested, as in (26). But the data already reported in (25a–b) show that the orderings of no after ‘well’ and before ‘still/yet’/‘always’ can also be combined. (26) Casorezzo a. al drmi n se:mpr be: he sleeps not always well ‘He doesn’t always sleep well’ b. ly la vedi n ŋkamoâ•… be: he it sees not yet well ‘He doesn’t see it well yet’

100â•… Sentential negation:€adverbs Furthermore, we note that in at least some of the varieties with a no negation of the relevant type, a separate mica-type negation is also present. A case in point is S. Angelo in (27), in which the negation follows ‘well’ as in (27a) and the participle as in (27c), but a mia negation is also present, as in (d). This tends to discount Zanuttini’s (1997) idea that the no negation may play the role of the mica-type negation in contexts like (27b), where it is ordered in front of other low adverbs, as opposed to preceding them. (27)â•… S. Angelo Lodigiano (Lombardy) a.â•… El drme bEŋ n he sleeps well not ‘He doesn’t sleep well’ b. El drme n sEmper he sleeps not always ‘He doesn’t always sleep’ c. i m aŋ tSamade n theyâ•… meâ•… haveâ•… called not ‘They haven’t (yet) called me’ d. El tSami mia nn am himâ•… I.call not yet ‘I am not calling him yet’

A further set of varieties which display a no negation is exemplified here by Viguzzolo in (28). The distribution of no in these varieties overlaps with that described here for nen, mia, etc. in the languages in (18)–(20). Thus, no precedes all adverbs in the aspectual series, as in (28a), and it normally appears together with these between the auxiliary and the participle, as in (28b–b’). The language of Viguzzolo also has a mia negation which has the same distribution as no, as illustrated by (29). (28)â•… Viguzzolo (Piedmont) a.â•… a nn al tSameŋâ•… n ŋku/ empr/ be theyâ•… notâ•… him call not yet/ always/ well ‘They don’t call him yet/ always/ well’ b. a l aŋ n (semper) tSama theyâ•… himâ•… haveâ•… not always called ‘They haven’t always called him’ b’. aŋ n ŋku maï†−dʒa they.haveâ•… not yet eaten ‘They have not yet eaten’ c. u nn a n drumi be heâ•… notâ•… hasâ•… notâ•… slept well ‘He hasn’t slept well’

Ordering sentential negation and other adverbsâ•… 101 (29)â•… Viguzzolo a.â•… a nn al tSameŋâ•… mej py they not him call not any longer ‘They don’t call him any longer’ a’. u n drm mej ŋku he not sleeps not yet ‘He doesn’t sleep yet’ b. u n l a mej py det he not it has not any longer said ‘He hasn’t said it any longer’ b’. a nn aŋ mej ŋku/ʑa maï†−dZa they not have not yet/already eaten ‘They have not already/ yet eaten’

Zanuttini (1997) is aware that in some varieties, the no-type negation appears before the participle, and illustrates this distribution for Pavia. In her analysis, however, no in Pavia still occupies the Neg4 position. Her independent evidence in favour of this low position resides in the fact that in Pavia, no can also appear after the participle. The fact that no precedes the aspectual series is analysed, as before, as an effect of its optional appearance in the mica-type negative position, i.e. Neg2. As for sentences in which no precedes rather than follows the participle, this is analysed as a consequence of the fact that the participle moves to a lower head position. A variety present in our data which illustrates the variable order of the no negation with respect to the participle is Castiglione in (30b–c), in which (as in Viguzzolo) the negation also precedes ‘well’, as in (30a–b). Furthermore, Castiglione has a separate mica-type negation, illustrated in (30d). (30)â•… Castiglione d’Adda (Lombardy) a.â•… el drm n beŋ/ sEmper he sleeps not well/ always ‘He doesn’t always sleep/ sleep well’ b. i aŋ durmi:d nâ•… (beN) they have slept not well ‘They haven’t slept (well)’ c. i aŋ n (semper) durmi:d they have not always slept ‘They haven’t (always) slept’ d. el drm miGa beŋ he sleeps not well ‘He doesn’t sleep well’

102â•… Sentential negation:€adverbs Since in both Viguzzolo and Castiglione the negation precedes ‘well’, as it does other adverbs, the simplest analysis for these languages is one that assimilates their no negation to the high negations in (18)–(20). But this, in turn, implies that the post-participial position of the negation in Castiglione is entirely unconnected to its supposedly low position in the adverb hierarchy. Indeed, the order participle€– no€– ‘well’ in (30b) is not predicted by Zanuttini’s (1997) analysis, since, if the post-participial position of no depends on its filling Neg4, then it should precede ‘well’ and not follow it. Similarly, the order no€– participle€– ‘well’ in (28c) cannot be explained by the claim that no is pre-participial because the participle is even lower than no€– for in that case no ought to follow ‘well’. Zanuttini’s (1997) analysis of the post-participial position of no proves equally problematic for Milanese-type varieties, represented here by Casorezzo. In Zanuttini’s own terms, though no is supposed to be quite low when it follows ‘well’, it is also supposed to raise to a higher position when preceding aspectual/quantificational adverbs. But now observe that no follows the participle independently of which adverbs it combines with and in which orders, as in (25c–c’). Thus, by Zanuttini’s own logic it must be the case that the participle is high enough to precede the whole adverbial string in (1). But if so, of course, the post-participial position of the negation does not tell us anything about whether it in fact appears low in the adverbial string€– which is, quite strikingly, exactly the same conclusion we reached for Margarita in (24). We provide our own analysis of the position of adverbs with respect to verbal heads in section 3.3, and we disregard this issue altogether in the rest of this section. More questions concerning Zanuttini’s (1997) model arise when we consider that the patterns reviewed so far are not the only ones created by the interaction of negation with the aspectual series. The nen pattern in our data that provides the best match with Zanuttini (1997) is illustrated in (31). This has ‘already’ and ‘any longer’ preceding the negation, and the other relevant adverbs following it. The pattern is predicted by Zanuttini on the basis of the incorporation analysis of pi nen. The data in (32) provide a comparison with the mica-type negation in the same language. (31)â•… Mombercelli (Piedmont) a.â•… u d42m nEŋ aŋkua/â•… bE he sleeps not yet/ well ‘He doesn’t sleep yet/ well’ b. u d42m ʑa/ pi nEŋ he sleeps already/â•… any longerâ•… not ‘He doesn’t sleep already/ any longer’

Ordering sentential negation and other adverbsâ•… 103 c. i lu tSmu nEŋâ•… sEmp theyâ•… himâ•… call not always ‘They don’t always call him’ (32)â•… Mombercelli a. i d42mu pa py they sleep not any longer ‘They don’t sleep any longer’ b. i aŋ pa ŋku/ sEmp/ ʑa dumi they have not yet/ always/ alreadyâ•… slept ‘They haven’t slept yet/ already/ always’

But now consider Pamparato in (33), in which ‘already’ precedes the nen(t)type negation, which is therefore a candidate for Neg3, as in (33d). As expected under the hierarchy in (1), nen(t) precedes ‘any longer’, as in (33a) and (33c). What is not expected is that nen(t) follows ‘still/yet’, as in (33b) and (33d). The data in (34) provide a comparison with the mica-type negation. (33)â•… Pamparato (Piedmont) a.â•… i d24muâ•… naint tSy they sleep not any longer ‘They don’t sleep any longer’ a’. u drm naint saimp he sleeps not always ‘He doesn’t always sleep’ b. i d24muâ•… ŋku naint they sleep yet not ‘They don’t yet sleep’ c. i aŋ naint tSy d24mi they have not any longerâ•… slept ‘They haven’t slept any longer’ d. i aŋ ŋku/ ʑQ naint d24mi they have yet/ already not slept ‘They haven’t yet/already slept’ (34) Pamparato (Piedmont) a. i d24muâ•… pQ tSy / ŋku theyâ•… sleep notâ•… any longer/â•… yet ‘They don’t sleep yet/any longer’ b. i aN pQ Nku/ ʑQ d24mi they have not yet/ already slept ‘They haven’t slept yet/ any longer’

Other languages that share with Pamparato in (33) the property of having a negation that precedes ‘no longer’ while following ‘still/yet’ include Romansh varieties. A case in point is Mustér in (35), in which the negation is of the

104â•… Sentential negation:€adverbs bu(ka), bare-N type. In other words, the ordering pattern is not tied to the morphological ‘nothing’-type negation. As shown in (35a, c), bo in Mustér precedes ‘any longer’€– but, as shown in (35b, d), it follows ‘still/yet’. (35)â•… Mustér (Grisons) a.â•… jauâ•… drmlâ•… boâ•… pli/ adina/ bain I sleep not any longer/â•… always/â•… well ‘I don’t sleep any longer/ always/ well’ b. jau drml aun bo I sleep yet not ‘I don’t sleep yet’ c. i aŋ bo duʀmiu pli they have not slept any longer ‘They haven’t slept any longer’ d. i aŋ auŋ bo duʀmiu they have yet not slept ‘They haven’t slept yet’

A further pattern attested in our data is illustrated in (36) (the data in (37) provide a comparison with the mica-type negation). In (36), nEŋ precedes ‘already’€– which, in terms of the hierarchy in (1), would have to mean that we are in the presence of an exponent of Neg2. Yet again nEŋ follows both ‘any longer’ and ‘still/yet’, which would once again have to constitute cases of incorporation into negation. (36)â•… Montaldo a.â•… i u tSam pi nEŋ I him call any longer not ‘I don’t call him any longer’ b i u vEg nEŋ sEmp I him see not always ‘I don’t always see him’ c. i  ŋku nEŋ fQ-u I have yet not done-it ‘I haven’t yet done it’ d. i  nEŋ dʒQ fQ-u I have not already done-it ‘I haven’t already done it’ e. i  nEŋ drmi bEŋ I have not slept well ‘I haven’t slept well’ (37)â•… Montaldo a.â•… i u vEg pæ pi/ sEmp I him see not any longer/ always ‘I don’t see him any longer/ always’

Ordering sentential negation and other adverbsâ•… 105 b.â•… i  pQ Nku/ dZa fQ-u I have not yet/ already done-it ‘I haven’t done it already/ yet’

In the table in (38), we summarize the data we have provided concerning the relative position of the negation and the other adverbs in the sub-hierarchy in (1). It seems to us, from looking just at the distribution of the negation adverb with respect to these other adverbs, that there are two major patterns, rather than three. Thus, in varieties of the type in (38a), the negation always precedes the relevant subset of adverbs. On the other hand, in varieties of the type in (38b), their relative order is somewhat variable.7 It will be noticed that when it comes to the patterns in (38b), practically any relative order of the negation adverb with respect to ‘already’, ‘still/yet’ and ‘any longer’ is attested€– with one striking exception. What is not attested in our data is the order which should be the basic one given the hierarchy in (1), with ‘already’ preceding the adverb, and all other members of the hierarchy following it. This absence is significant, and would be so even if we wanted to follow Zanuttini’s (1997) line of explanation with respect to the possible permutations€– i.e. that the reordered substring roughly reflects cases of incorporation. Indeed, we notice that sometimes ‘still/yet’ is reordered, and sometimes ‘any longer’€– in other words, in neither case is reordering necessary, which means that there is no independent reason why the supposedly basic order should be excluded. Therefore the fact that it is not found should at least cause some perplexity. (38)

Relative position of negation and other adverbs

a. Castellazzo B.â•› Pozzaglioâ•› Pomarettoâ•› Viguzzoloâ•› Castiglione d’A.â•› b. Cantoiraâ•› Mombercelliâ•› Pamparatoâ•›

already no longer still

always

nɛ̃â•›itQâ•› (18a) mEâ•›jQâ•› (18b) mia (19) pa (20) nâ•› (28) mejâ•› (29) nâ•› (30) mIGâ•›a

– zQâ•› – zQâ•› – bEâ•›le – dʒâ•›â•›

– pi – pi – pâ•› – pi

– ʑa

– py

– aŋ↜kurQâ•› – aŋ↜kurQâ•› – Jâ•›amâ•› – ŋ↜karâ•› – ŋ↜ku – ŋ↜ku

– d lâ•›ŋ↜kâ•› – d lâ•›ŋ↜kâ•› – sEâ•›mper – bɛ̃â•›: – sampre – bâ•›ŋ↜ – ɕâ•›empâ•›r – be – sEmper

– beâ•›ŋ↜ – beŋ↜

Jiŋ↜(21) pa (22) nŋ↜(31) pa (32) naint (33)

dZâ•›â•›

aŋ↜ku€– – ŋ↜ku aNku4aâ•› – Nkuâ•› ŋ↜ku€–

– sEâ•›mp – sEâ•›mp – sEmpâ•› – sEmpâ•› – saimp

– biŋ↜

– py

ʑâ•›a€– – ʑâ•›Qâ•› ʑâ•›Qâ•›–

pi€– – pjy pi – py – tSâ•›y

well

– bEâ•›

106â•… Sentential negation:€adverbs already no longer still Mustérâ•› Montaldoâ•› Casorezzoâ•› S.Angelo L.â•›

pQâ•› (34) –zQâ•› bo(35) nEŋ↜(36) –dZQâ•› pQâ•› (37) – dZQâ•› nâ•› (25)–(26) nâ•› (27) mia

– tSyâ•› – pli pi€– – pi

– Nkuâ•› aun€– Nkâ•›u€– – Nkâ•›u – ŋ↜kamâ•›

always

well

– adina – sEmpâ•› – sEmpâ•› – sEâ•›mper – sEâ•›mper

– ain –bENâ•› beâ•›(ŋ)↜€– bEâ•›ŋ↜–

3.2.1 The order of negation with respect to aspectual adverbs Summarizing so far, the functional hierarchies for negation adverbs, and adverbs more generally, proposed by Zanuttini (1997) and Cinque (1999) do not seem to be sufficient to account for the spread of variation observed in Italian varieties. A separate question is whether they are necessary. The alternative generally suggested is that ‘adverb attachment is driven by interpretation’ (Svenonius 2002:€209), essentially as in the earliest treatments of adverb placement in generative grammar (Jackendoff 1972). Specifically, Ernst (2002:€92) claims that ‘the most important determinant of adjunct licensing is an adjunct’s scope (and other selectional) requirements, encoded as lexical requirements and verified at LF, rather than syntactic feature licensing, as in Cinque (1999) and other current work’. As Nilsen (2003:€13) observes, according to Ernst (2002) ‘adverbs can attach freely to any functional projection as long as the semantic requirements of the adverbs (FEO-calculus) are respected. Hence Ernst has it that, while adverbs are ordered by his FEO-calculus, the categories that the adverbs attach to are ordered by an orthogonal relation’, i.e. the basic functional sequence. Nilsen (2003:14) argues against this approach on the grounds that (among other things) ‘this set-up seems to force us to abandon the view that T has semantic import’. Indeed, according to Ernst (2002:€97), ‘events and propositions are not necessarily mapped to any one particular projection … A relatively high projection like TP may represent an event as long as nothing forces it to be converted to a proposition (such as a modal auxiliary or adverb); likewise a low projection like PredP can denote a proposition if no element above it requires an event’. What is more, Nilsen (2003:€18) objects to Ernst (2002), as well as Svenonius (2002), on the grounds that ‘although … the facts discussed by Cinque (1999) should ultimately be derived from more fundamental considerations, … the primary tool to do so can[not] be enrichment and manipulation of ontological

Ordering sentential negation and other adverbsâ•… 107 categories’€– i.e. the categories (fact, event, etc.) in terms of which the selectional properties of adverbs are stated. While it is evident that a limited repertory of conceptual entities is generally desirable, it seems to us that the flexible interpretation of syntactic categories required by Ernst (2002) is not in itself worse than a rigid interpretation of them. In fact, it is one of the overarching themes of the present work that the interface between syntax and interpretation is not rigid€– in the sense both that ambiguity is not necessarily resolved syntactically and that the same interpretation can be conveyed by different syntaxes. From this perspective, we are not surprised to find that a given syntactic domain€– here the I domain€– potentially maps to different semantic entities, or that the opposite may happen:€the same semantic entity may map to different syntactic domains. Let us consider, then, the relative orders illustrated above in which sentential negation adverbs appear internally to the substring of aspectual/manner adverbs individuated by Cinque’s (1999) and Zanuttini’s (1997) work. We assume, as is standard, that the lowest verbal domain of the sentence is characterized by the closure of thematic roles by arguments. The fact that negation adverbs, and adverbs in general, precede the argumental string suggests that they are inserted outside this lower verbal domain. At the same time, several adverbs, and in particular those in the substring in (1) that directly interests us here, typically follow the finite verb, which appears in the I head of the inflectional domain. Therefore, between the lower thematic domain and the higher inflectional one there must be at least an intermediate domain, which in Manzini and Savoia (2005) we name E (suggesting Event). It is in this domain that we find the adverbial substring in (1). We begin with the aspectual adverbs già ‘already’, ancora ‘still/yet’ and più ‘any longer’. Ernst (2002:€341–7) construes the interpretation of ‘already’ in the following terms (based on work by Michaelis (1996)):€‘the immediate scope of already denotes a state S, located at reference time, and whose inception precedes the time of an expected possible state S’ of the same type as S’. Thus a sentence like Karen has already performed is interpreted as saying that ‘S, the result state of Karen’s performing … precedes another interval at which she was expected to finish performing’. As for still, ‘the state scoped by still holds at reference time, held at a previous time and its expected end e was before reference time’. Thus, in a sentence like They still were doing it yesterday, ‘yesterday identifies the reference time, and still indicates that the process was ongoing at an earlier event-time as at reference-time’. As for ‘no … longer’, a widely held analysis takes it to be the negative counterpart of ‘still’ (Löbner 1989).

108â•… Sentential negation:€adverbs In the varieties of Montaldo, Pamparato and Mombercelli in (38b), the negation adverb precedes only one of the aspectual adverbs, respectively ‘already’, ‘any longer’ and ‘still/yet’. The rigid extrinsic ordering that characterizes Cinque’s (1999) hierarchies is a clear liability here, since there is no way of interspersing Neg positions and adverb positions so as to make the different possible orders emerge. The incorporation of single adverbs (in particular pi ‘any longer’) into the negation, suggested by Zanuttini (1997) as a way of making Cinque’s (1999) hierarchy compatible with the data, hints at a different type of explanation, whereby the properties of the single lexical item become relevant; in other words, the relative ordering of adverbs is conditioned by lexical idiosyncrasies. This also seems unsatisfactory to us; as we indicated when presenting the data, there is no clear evidence that the sequences that Zanuttini (1997) treats as single constituents are in fact constituents. Our idea is that the syntactic grids determining the relative order of adverbs are much coarser grained than the functional hierarchies of Cinque (1999), and refer to the same nominal categories that are relevant for the ordering of argumental material. It is in this connection that the conclusion of section 3.1€– that the so-called negation adverb is in reality a nominal category connected to the internal argument of the predicate€– becomes relevant for ordering as well. Let us begin with structures of the type in (13) or (17), where the negation adverb occupies a Q position. As we have just seen, this can be taken to be internal to an intermediate domain between V and I, i.e. E. We model the nominal arrays of verbal domains on inflectional-level clitic structures, where we independently argue for Q elements (represented in the clitic domain by the si variable, cf. chapters 5–6). In the inflectional clitic domain there is a restricted range of elements above Q, which includes, in particular, what we may take to be the left edge position of the theory, notated R (with a nod to Referentiality). Evidently, this left edge position could be imputed to the adverbs that precede the negation in Q, as illustrated in (39) for the Piedmontese varieties in (38b). (39)â•… a.â•… b. c. d.

Cantoira [R dZ6/ pi/ aŋkuâ•… Mombercelli [R ʑa/ pi Pamparato [R ʑQ/ Nku Montaldo [R pi/ Nku

[Q Jiŋ [Q nEN [Q naint [Q nEŋ

If the Q position of the negation remains constant, the aspectual adverbs ordered after it occur in a lower position. Our general schema for nominal

Ordering sentential negation and other adverbsâ•… 109 positions recognizes a series of positions below the left edge R position and the quantificational Q position, which are essentially linked to deixis, including 1st and 2nd person arguments and locatives. We may want to assign the aspectual adverbs that follow the negation in Q to this set of positions, adopting, for instance, the label Loc in (40)€– but keeping in mind that deictic relations in time, rather than in space, are being denoted. (40)â•… a.â•… b. c.

Mombercelli [Q nEŋ [Loc aNku4a Pamparato [Q naint [Loc tSy Montaldo [Q nEŋ [Loc dZQ

In a nutshell, this analysis amounts to saying that negations of the type being investigated generally have a Q position. There they are mostly followed by adverbs of the aspectual series in a deictic position (conventionally Loc). In some varieties, some aspectual adverbs are shifted to the left edge of the domain (here R) on a lexical basis. In a sense this is the analysis that Zanuttini (1997) gives for the pi nen order:€namely, that the relatively high position of the negation is masked by ‘any longer’ incorporating with it. The difference is that in the analysis that we are now developing, the reordering of the aspectual series around the negation requires no incorporation. At the same time, the existence of truly (i.e. morphologically) incorporated forms of ‘no longer’ and ‘not yet’ supports the idea that selection for particular lexical items is at stake in the reordering. At this point we can bring into the picture another major type of language, namely the type in (38a), in which the negation adverb systematically precedes the aspectual series. In terms of the analysis being considered, they can be described as languages with the negation adverb relatively high in the R or Q position, with other aspectual adverbs following it in the deictic reference positions (i.e. Loc), as shown for Castellazzo in (41). Manzini and Savoia (2005), who do not entertain the possibility of a deictic (Loc) position for the aspectual adverbs, assign them to the Q position, on the basis that they imply a quantificational closure over events. At least as far as the schema in (41) is concerned, the Q and Loc alternatives yield equivalent results. (41)â•… Castellazzo Bormida [R/Q mEjQ/ n˜itQâ•… [Q/Loc zQ/ pi/ aŋkurQ

That two different positions, presumably R and Q as in (41), are needed for the negation is a conclusion motivated by independent evidence. In many

110╅ Sentential negation:€adverbs varieties the mica-type negation can actually combine with the ordinary sentential negation adverb, as in (42). In this case, the mica-type negation generally takes the higher position, presumably R, and the ordinary negation takes the lower position, presumably Q. (42)╅

Margarita u Q pa nEŋ tSama-u he has not not called-him ‘He has not called him’

In cases where two aspectual adverbs combine with the negation, both can precede it, as illustrated in (43) for varieties (Cantoira and Mombercelli) in which we have shown that each of the aspectual adverbs involved can separately precede the negation. (43)â•… a.â•… b.

Cantoira u dyrt dZ pi Jiŋ he sleeps already any.longer not ‘He doesn’t already any longer sleep’ Mombercelli u d42m ʑa pi nEŋ he sleeps already any.longer not ‘He doesn’t already sleep any longer’

A straightforward application of the structural schemas already proposed leads us to assume that while the negation regularly appears in Q, the left edge position R of the system can be iterated for aspectual adverbs. Alternatively, adopting Manzini and Savoia’s (2005) idea that aspectual adverbs can appear in Q, it is Q that turns out to be iterated, equivalently for present purposes, as shown in (44). (44)â•… Cantoira [R d [R/Q pi [Q Jiŋ

If we put together the pattern in (42), where the mica-type negation precedes the ordinary negation, with the fact that aspectual adverbs precede the ordinary negation, we may expect orders like (45a), where the mica-type negation is followed by an aspectual adverb and then by the ordinary negation. Similarly, since aspectual adverbs can also follow the ordinary negation, we may expect the pattern in (45b), where the mica-type negation is highest and the aspectual adverbs are on either side of the ordinary negation. (45)â•… a.â•…

Margarita u Q pa pi/ ŋku nEŋ tSama-u he has not any.longer/ yet not called-him ‘He hasn’t called him yet/ any longer’

Ordering sentential negation and other adverbsâ•… 111 b.

Montaldo i  pQ ŋku nEŋ dZa fQ-u I have not yet not always done-it ‘I haven’t yet done it already’

The distribution of the two aspectual adverbs on either side of the ordinary negation in (45b) can be predicted for Montaldo simply on the basis of the union of the structures (39d) and (40c). Assuming that the position of the micatype negation, in turn, is R, we obtain the structure in (46). (46)â•… Montaldo [R pQâ•… [R/Q ŋkuâ•… [Q nEŋâ•… [Loc/Q dZa

Before considering in more detail what degree of mobility the aspectual adverbs as well as the negation have, in the next section we will analyse the order of the negation with respect to a quantificational adverb like ‘always’, as well as with respect to the low manner adverb ‘well’. 3.2.2

The order of negation with respect to quantificational and manner adverbs Accounting for the ordering of negation with respect to ‘always’ and ‘well’ implies some preliminary understanding of these adverbs. Let us first consider ‘always’€ – and quantificational adverbs more generally. According to Ernst (2002:€ 347), who quotes work by de Swart (1993), ‘these adjuncts quantify over subsets of events within the set denoted by their sister constituent; the precise delimitation of these sets depends on the focus structure of the sentence’. So a sentence like John always eats pasta can mean ‘In all situations in which John eats something, he eats pasta’ or ‘In all situations in which John does something, he eats pasta’. When it comes to order, according to Ernst (2002:€363) ‘frequency adverbs … have fewer restrictions than aspectual adverbs’. Yet the fact to be explained is that ‘always’ is found obligatorily after the negation adverb in all varieties studied here or that we are aware of. It seems evident that this obligatory ordering is dictated by semantic requirements. In fact ‘always’ is obligatorily read in the scope of the logical negation. The Romance languages, like English, have a separate lexical item for a universal ‘always’ taking scope over the logical negation, i.e. never (standard Italian mai etc.). Obviously the reading in the scope of the logical negation requires ‘always’ to be lower than the negation adverb, i.e. the element which implies the logical negation itself. Given the ordering of aspectual adverbs with respect to the negation adverb sketched in section 3.2.1, and the rigid ordering of the quantificational adverb with

112╅ Sentential negation:€adverbs respect to negation which we have just analysed, we predict that the three combine in a fixed order, which is what can be seen in the examples in (47). (47)╅ a.╅ b. c.

Cantoira uj t pi Jiŋâ•… sEmp dyrmi he has any.longer not always slept ‘He hasn’t always any longer slept’ Margarita j  pa pi nEŋâ•… sEmpre vist-je them I.haveâ•… not any.â•… longer not always seen-them ‘I haven’t always seen them any longer’ Pamparato i aŋ ʑQ naint saimp d24mi they have already not always slept ‘They haven’t always already slept’

If we keep to the idea that the negation lexicalizes Q properties in the E domain, quantificational adverbs like ‘always’, which follow it, must appear in a lower position in the same domain. However, the deictic position(s) that we have assigned to aspectual adverbs are not consistent with the interpretive properties of quantificational adverbs, which simply denote relations between sets of events very much independent of temporal deixis. This then leaves the lowest position of the nominal sequence, namely N, which we connect to the internal argument of the predicate (cf. chapters 7–8). In fact, we have seen at least one way in which ‘always’ is related to the internal argument€– namely by implying a focalization on it, in one of the possible interpretations of John always eats pizza, namely ‘whenever John eats something, he eats pizza’. Alternatively the entire elementary predicate can be in the scope of the focus operator as in the interpretation ‘all situations of John doing anything are situations of John eating pizza’. As discussed in section 3.1.2, the same ambiguity is present with the quantificational adverb we are mainly concerned with here, i.e. the negation. Thus John doesn’t eat pizza can be understood as ‘It is not pizza that John eats’ or as ‘It is not eating pizza that John does’. In section 3.1.2 we concluded that in the latter reading the negation quantifies over the elementary event that can be construed as the internal argument of the causation (or other) property introduced by the application of the external argument. In the former reading the negation quantifies over the internal argument of the elementary predicate or equivalently of the whole transitive predicate, namely pizza. If a parallel analysis is applied to non-negative quantificational adverbs, it is natural to conclude that these adverbs have at their disposal the N position connected to the internal argument. This in turn yields a straightforward structure for

Ordering sentential negation and other adverbsâ•… 113 the sequences in (47), as illustrated in (48) for (47c). Remember at the same time that another possibility is suggested by structures like (46) in which the negation in R or Q precedes elements (potentially) in the same types of positions. On this basis we may assume that the negation in Q precedes the quantificational adverb also in Q, where it can attach because of its quantificational content. (48)â•… Pamparato [R ʑQâ•… [Q naintâ•… [Q/N saimp

Before we proceed further, let us introduce the last major type of adverbials included in the hierarchy in (1), namely manner adverbs like ‘well’. According to Ernst (2002) manner adverbs are part of a larger class of predicational adverbs; ‘the manner adverbial restricts the denotation to a set of events of V-ing characterized by their property of (manifesting) rudeness, dimness, strangeness and the like’ (Ernst 2002:€ 259), for instance in She left rudely or The bulb shone dimly. The low point of attachment of manner adverbials reflects the fact that what they are predicated of is events (as opposed to situations/sentences). However, Ernst (2002) does not consider the strict relation that an adverb like ‘well’ has not only with the event, but also with the internal argument of that event. For instance, in several Southern Italian varieties ‘well’ (bene in standard Italian) translates as the adjective for ‘good’ (buono in standard Italian). In (49a–c) we report examples in which this adjective can be seen to overtly agree with the accusative internal argument. In the same varieties, the internal argument of unaccusatives also agrees with ‘well’ as shown in (49a’–c’). As for unergatives, either ‘well’ takes an invariant form corresponding to the masculine singular, as in the Marzano example in (49a’’), or it agrees with the external argument, as in the Gizzeria example in (49b’’). (49)â•… Marzano Appio (Campania) a. u/a veru bbuonu/ bbna him/her I.see well.m/ well.f ‘I see him/her well’ a’. issu/ essa se lava sEmbe bbuonu/ bbna he/sheâ•… him/herself washesâ•… alwaysâ•… well.m/well.f ‘S/he always washes well’ a’’. issu/essaâ•… a rurmitu bbuonu he/she has slept well ‘S/he slept well’ Gizzeria (Calabria) b. u/â•… a/ i lavu bbnu/â•… bbna/bbni him/her/themâ•… I.wash well.m/well.f/well.pl ‘I wash him/her/them well’

114â•… Sentential negation:€adverbs b’. b’’.

iɭɭu/ iɭɭa s a he/sheâ•… him/herselfâ•… hasâ•… ‘S/he has washed well’ iɭɭu/ iɭɭa a ddrmutu He/sheâ•… hasâ•… slept ‘S/he slept well’

ɭɭavatu bbnu/ bbna washedâ•… well-m/well.f bbnu/ bbna well/ well.f

Orsomarsoâ•… (Calabria) c. (a kammisa/ i kavutsuni) aɟɟu lavta/ lavtiâ•… bbna/â•… bbuni the shirt.f/ the hose.pl I.have washed.f/pl. well.f/ well.pl ‘The shirt/ the hose I washed well’ c’. arâ•… arrivtu/arrivta bbunu/ bbna s/he.has arrived.m/f. well.m/well.f ‘S/he has arrived well’

The same connection between manner adverbs and the internal argument can be seen in English sentences like I ate a quick pizza, meaning I quickly ate a pizza/I ate a pizza quick(ly). Similarly, it is possible to say that John has done a job well, or equivalently that John has done a good job. Therefore the manner adverb is connected to the internal argument position insofar as it is predicated either of an elementary event or of its internal argument€– which is also adopted as the internal argument of the transitive predicate as a whole. This is also what makes it possible for an adjective agreeing with (or embedded under) an accusative argument to modify an entire (elementary) event. The preceding discussion leads not only to the conclusion, shared with Ernst (2002), that manner adverbs attach to a relatively low domain in the sentence (namely the E domain here), but also that they are connected in particular to the N position of the domain. The nominal nature of manner adverbs is evident in languages like (49), but also in standard Italian and Northern Italian varieties in which bene is a noun, as in il bene ‘the good’ (e.g. il bene comune ‘the common good’) and so is male ‘badly’, as in il male ‘the evil’ (cf. il bene e il male ‘good and evil’). In the light of our hypotheses it is particularly interesting to consider data that show the position of ‘well’ not only with respect to negation, but also with respect to aspectual and quantificational adverbs, as in (50). (50)â•… a.â•… b.

Cantoira u dyrt Jiŋ sEmp biŋ heâ•… sleepsâ•… notâ•… alwaysâ•… well ‘He doesn’t always sleep well’ Margarita u parla pi nEN bEN he speaksâ•… any.longer not well ‘He doesn’t speak well any longer’

Ordering sentential negation and other adverbsâ•… 115 c.

Piverone (Piedmont) al drm ŋku Jiŋ sEmpe bEŋ he sleeps yet not always well ‘He doesn’t always speak well yet’

In each case ‘well’ closes the adverbial string, which is what we expect if it takes the lowest N position in the domain. The other adverbs will then take the positions that we have suggested for them in the preceding discussion, yielding the structures in (51) for the examples in (50). Recall that in the discussion of quantificational adverbs, in particular ‘always’, we suggested that the Q position may be recursive, while aspectual adverbs that appear before negation are inserted in the left-edge position R. (51)â•… a.â•… b. c.

Cantoira [Q Jiŋâ•… [Q sEmpâ•… [N biŋ Margarita [R pi [Q nEŋ [N bEŋ Piverone [R ŋku [QJiN [Q sEmpe [N bEŋ

With this background, we can consider the final question implicit in the hierarchy in (1), namely the position of the no-type negation that follows manner adverbs. Recall that Zanuttini (1997) is led to propose the low Neg4 position in (1) for these elements on the basis of two pieces of evidence:€ first, their post-participial position, and second, the fact that they follow low manner adverbs like ‘well’. In reality, the evidence concerning the relative ordering of adverbs also shows that the no-type negation precedes quantificational and aspectual adverbs. Crucially, when negation is combined both with ‘well’ and ‘always’/‘yet’, two orders are attested in our data. One of them has the negation preceding both the quantificational/aspectual adverb and the manner adverb, as in the Casorezzo example in (26a); this is consistent with Zanuttini’s (1997) proposal that the relative ordering of Neg4 in front of other adverbs depends on its taking a higher position. What is not consistent with her proposal is the alternative order, which has the manner adverb before the negation and the other adverbs following it, as in the Casorezzo example in (25b). In reviewing the data, we further indicated that the existence of a dedicated mica-type negation in varieties like S.Angelo makes it implausible that no should take on the role of the mica-type negation when it is positioned higher. The alternative we propose is therefore simply that the no-type adverb has the very same position in the eventive domain string as the other negation adverbs reviewed so far, namely Q, as indicated in (52). This position is

116â•… Sentential negation:€adverbs compatible with quantificational, aspectual and manner adverbs following it, assuming that they take their canonical quantificational/deictic/nominal positions as in (52b). It is also compatible with the alternative order in (52a), in which only the manner adverb precedes no, on the assumption that in this case ‘well’ occurs in the left-edge R position of the system. Under this account, there is no difference in the positioning of Milanese-type no and Piedmontese nen; rather, there is a difference as to which types of adverbs are found in the ‘left-edge’ of the domain (here R), namely the manner adverb in Milanese varieties and aspectual adverbs in Piedmontese ones. (52)â•… Casorezzo a.â•… [R ben [Q nâ•… [Q/ Loc se:mpr/ ŋkamo b. [Q n [Q/ Loc se:mpr/ ŋkamo [N ben

Note that it does not follow from (52) that the relative ordering with negation higher than ‘well’ is restricted to cases where there is an adverb of another type. For some varieties this is as it should be, since there is evidence of variation in order even in simpler strings, as in (53). (53)â•…

Arconateâ•… (Lombardy) a maJdʒu be:n no/ no beŋ I eat well not/ not well ‘I don’t eat well’

For varieties like Casorezzo itself in which no optional order is attested in simple adverbial pairs, we of course predict the complex order in (52a)€– i.e. precisely the one that is problematic under other analyses. As for the order in (52b), we note that at least the quantificational adverb ‘always’ systematically precedes ‘well’ in the absence of negation, as in (54). We conclude that the optionality in (52) depends on whether the relative order ‘well’€– negation or ‘always’€– ‘well’ dictates the overall pattern; we shall elaborate on this alternative directly below. (54)â•… Casorezzo a.â•… Eŋ sempr maï†−dZa:â•… they.haveâ•… alwaysâ•… eaten ‘They have always eaten well’ b. Eŋ maï†−dZa:â•… semprâ•… they.haveâ•… eaten always ‘They have always eaten well’

be: well be: well

Assuming that the account in (52) is on the right track, it raises the question of why there should be variation of the kind observed, with some languages putting ‘well’ before negation (to the exclusion of aspectual adverbs) and other

Ordering sentential negation and other adverbsâ•… 117 languages (i.e. the Piedmontese varieties of section 3.2.1) showing the reverse distribution. An obvious observation is that all languages that allow the order in (52a) have a negation adverb of the no type. This belongs to the n- negative polarity series but does not have the same form as any negative polarity argument. Rather, it has the same form as the deictic negation. On the other hand, the varieties that allow aspectual adverbs before negation have negation adverbs either of the ‘nothing’ type or of the bare N type reviewed in section 3.1. From this perspective, the Piedmontese ordering in (39)–(40) can be described by saying that nominal negation adverbs (‘nothing’ or bare N type) only allow deictic adverbs out of their domain (i.e. aspectual adverbs under the characterization provided here). Since ‘well’ appears to be lexicalized by clearly nominal elements, such as ‘good’ in the examples in (50) or bene in standard Italian and Northern varieties, we may equivalently describe Piedmontese varieties by saying that nominal adverbs must remain in the domain of the nominal-type negation. By contrast, deictic-type, i.e. no-type, negations in Milanese varieties, only allow nominal adverbs, such as low manner adverbs, out of their domain. Equivalently, we may describe Milanese varieties by saying that deictic (i.e. aspectual) adverbs must remain within the domain of the deictic negation. The generalizations just stated suggest that the parameter differentiating between Milanese-type and Piedmontese-type varieties may not need to be (entirely) stipulated. Rather, the different inherent properties of the negations involved may determine their placement relative to other adverbial subseries. As for the general principles under which the parallelism indicated may fall, they could perhaps be identified with a requirement for the negation (of a certain type) to close off certain types of adverbial subdomains. This may interact with other requirements on the adverbial sequence, such as the one in (54), whereby the manner adverb ‘well’, which is in the logical scope of the quantificational adverb ‘always’, is also in its syntactic domain. In particular, the complex order in (52b) takes this latter requirement into account. In contrast, in (52a) the composition of the negation with the quantificational adverb ‘always’ (read in its logical scope) evidently means that the negation determines the overall order, in particular with respect to the manner adverb ‘well’. In any event, the principal aim of this discussion has been to indicate how an account of the variation between Piedmontese and Milanese-type varieties is possible if we abandon the account provided by the various Neg positions in the hierarchy in (1)€– which we have shown to be descriptively inadequate. Finally, in analysing Milanese-type varieties we disregarded the fact that no appears after the participle, though Zanuttini (1997) and Cinque (1999) regard this as crucial. The reason why we did not discuss this is that the post-participial

118â•… Sentential negation:€adverbs position of no is also found in languages like Castiglione in (30) in which no precedes ‘well’ and the other adverbs considered here. In other words the two phenomena are (at least partially) independent of one another. Yet the issue deserves to be considered in more detail, since it still seems to be true that there are no negations that both follow ‘well’ and precede the participle. Before we consider the issue in section 3.3, however, we will try to draw some general conclusions concerning the present discussion of the relative ordering of adverbs. 3.2.3 General discussion Cinque (1999), Ernst (2002) and Nilsen (2003), from opposing points of view, argue that a theory of adverb ordering that uses both semantic and syntactic requirements is less economical than one restricting itself to a single component€ – though in semantically based models like Ernst (2002) or Svenonius (2002), a certain degree of freedom is observed in adverb placement because it is sensitive to semantic primitives (events, facts, etc.) that are not necessarily isomorphic with syntactic ones (TP, VP, etc). More clearly, for Nilsen (2003) the positioning of sentential-level adverbs is determined entirely by the fact that they are positive polarity items that must be outside the scope of non-veridical operators. However, Nilsen (2003) does not consider the event/ predicate-level adverbs studied here, while Ernst (2002) explicitly excludes negation adverbs from his discussion. We do not doubt that notions of positive and negative polarity (Nilsen 2003) or selection for events vs. propositions (Ernst 2002) play a role in adverb placement. Yet it seems to us that Cinque (1999) is correct in concluding that purely semantic constraints are not sufficient to explain the distribution of adverbs. This is particularly clear in the case of the negation adverbs which represent the focus of the present discussion, because the scope of the logical negation is entirely independent of their surface position with respect to aspectual and manner adverbs. On the other hand, Cinque’s (1999) conclusion that syntactic constraints are sufficient for adverb ordering works (to the extent that it does) only because interpretive facts are encoded in the syntax in the form of functional hierarchies. This encoding introduces a degree of rigidity absent from semantic accounts. In this respect, the discussion in this section can be read as an elaborate argument that the variation in the position of the negative adverb cannot be captured by (1)€– not even allowing for the various readjustments proposed by Zanuttini (1997). Our general diagnosis of the kind of difficulties encountered by previous syntactic analyses is the same as that presented in chapters 1–2 for complementizers, namely that categories lifted from formal semantics accounts or

Ordering sentential negation and other adverbsâ•… 119 descriptive (functional, typological, etc.) accounts do not necessarily yield the correct primitives on which to build the syntax. Our proposal differs from those put forward in previous literature in assuming that the positioning of adverbs in what we have called the eventive E domain of the sentence is sensitive to much the same categorizations (Definiteness, Quantification, deixis, etc.) as the placement of arguments in the inflectional domain of the sentence (clitics) or of the noun phrase. From the present perspective, manner adverbs close the adverbial string in that they are connected to the N-internal argument€– understood as the elementary event that the higher-level specifications of the sentence select, or as the internal argument of this elementary event, which is construed as the internal argument of the sentence as a whole. Next, we have argued that the aspectual series of adverbs is essentially connected to deixis, in the same way as demonstratives in the structure of noun phrases, or locatives as arguments of the elementary predicate€– though in the case of aspectual adverbs the connection is to temporal rather than spatial deixis. On the basis of what we know about deictic categories in noun phrases and in arguments, we expect aspectual adverbs to appear between quantificational adverbs and manner/nominal adverbs, which they do. Alternatively or complementarily, aspectual adverbs may be construed as belonging to a macrocategory of quantificational adverbs (Manzini and Savoia 2005), in which case they are predicted to commute in position with quantificational adverbs (narrowly construed). Incidentally, ‘already’, ‘still’ and ‘any longer’ typically have verbal counterparts, such as ‘to continue’ in the case of ‘still’, ‘to stop’ in the case of ‘no longer’ etc., which for Cinque (1999, 2006) also instantiate the same functional categories as their adverb counterparts. Yet we have independent reasons to believe that so-called auxiliaries in general are just ordinary verbs selecting sentences, as argued in chapters 5–6. It is also worth noting that in standard Italian there is an alternative lexicalization of già ‘already’, namely bell(o/i/ etc.), which is clearly adjectival/nominal in nature, as in (55); in many Northern varieties this is the only lexicalization of ‘already’. This adjectival lexicalization connects the aspectual adverb to the manner adverb (if anything)€– and through it in turn to the elementary event/internal argument specifications. (55)â•… L’ ho (sempre) bell’ e fatto it I.have always nice and done I have (always) already done it’

Proceeding from bottom to top, quantificational adverbs close (or may close) the aspectual–manner range; by and large, all that needs to be said about the

120â•… Sentential negation:€adverbs sentential negation adverb is that it belongs to the quantificational series. At the same time, the (re)positioning of certain adverbs to its left, namely aspectual adverbs in Piedmontese varieties and manner adverbs in Milanese-type varieties, provides evidence for a left-edge position of sorts within the adverbial field. This (re)positioning seems to depend (in part) on the lexical shape of the negation adverb; thus, deictic negation adverbs of the Milanese type co-occur with left-edge positioning of manner (i.e. nominal) adverbs, while the nominal-type negations of Piedmontese varieties co-occur with the repositioning of aspectual (i.e. deictic) adverbs. In other words, the interplay of lexical terminals, such as adverbs in the case at hand, depends on their actual lexical make-up and cannot be predicted solely on the basis of abstract arrays of features making them up. To take other examples, it matters for the purposes of ordering whether the content of ‘already’ is carried by già, which appears at the beginning of the sequence in (1), or by bell(o) as in (55), which instead appears at the end of the sequence; it matters whether ‘well’ is the nominal element bene or the adjectival element buono, and so on. The theories of adverb ordering reviewed in this section, even when they ostensibly aim to account for a variation spread, as does Zanuttini (1997), ignore this dimension of the problem. In a nutshell, the distribution of adverbs depends on the interaction of interpretation with the syntactic and lexical structures that restrict it in natural language. Shaping the adverbial space purely through interpretive notions of scope (Nilsen 2003) or selection for events, facts, etc. (Ernst 2002) seems to be insufficient to account for the fine variation in the positioning of the verbal range which is the focus of the discussion in Zanuttini (1997) and partly also in Cinque (1999). At the same time, a syntactic theory along the lines of Cinque (1999) or Zanuttini (1997) contains elements of rigidity that prevent it from accounting for various data€– except perhaps by allowing for languageparticular variations in the hierarchy whose universal nature represents the core of the theory, and therefore defeating its very purpose. 3.3

The interaction of adverbial and verbal positions:€the participle

In the discussion of the relative positioning of adverbs in section 3.2, we omitted the evidence that according to Cinque (1999) and Zanuttini (1997) comes from the positions of adverbs with respect to non-finite verbal heads, including the infinitive and the participle. One reason that led us to factor out this evidence is that no correlation holds, say, between the placement of adverbs after the participle and their placement with respect to manner or other low adverbs,

Interaction of adverbial and verbal positionsâ•… 121 as discussed for Castiglione in (30). Similarly, the conclusion of Cinque (1999) and Zanuttini (1997) that verbal heads (participles and infinitives) preceded by negation and other adverbs are particularly low in the structure is contradicted by evidence from the placement of non-adverbial material, notably (en)clitics, as seen in (23)–(24). At the same time, it appears that constraints are observed in the placement of adverbial material with respect to verbal heads€– subject to cross-linguistic variation. Therefore, in this section we briefly address the nature of the observed constraints and variation, concentrating on the participle. In section 3.2 we saw that negation adverbs are generally positioned between the auxiliary and the perfect participle in perfect tenses; yet there are languages in which the negation appears after the participle. Zanuttini (1997) and Cinque (1999) account for this pattern by assuming that the negation occupies a very low position, namely Neg4, so that the participle, moving towards the left periphery of the low sentential domains, systematically leaves it to its right. However, we have already shown that the post-participial negation of Casorezzo, although it follows ‘well’ as in (25c), precedes quantificational/ aspectual adverbs, as in (25c’), so that the latter are in turn preceded by the participle. This means that by Zanuttini’s (1997) and Cinque’s (1999) own reasoning there are no empirical grounds for claiming that the post-participial position of the negation is due to its low position in the adverb hierarchy; it could be due instead to the relatively high position of the participle. If so, we expect to find that negations that Zanuttini (1997) considers to occupy a (relatively) high position can also follow the participle. Indeed, recall that Zanuttini (1997) attributes the no–aspectual adverb order seen in Milanese-type varieties, as in (25c’), to a repositioning of the negation in Neg2; yet in (25c’) the negation follows the participle. Another case in point is Quarna Sotto, which was introduced in section 3.1.1. Both nota and mia can follow the participle, as in (56a), though they are typical Piedmontese negations of the ‘nothing’ and bare N types assigned by Zanuttini to Neg3. Furthermore, at least mia also appears before the participle as in (56b); and the relative order ‘still/yet’–mia is possible with either placement of the participle. Following Zanuttini’s (1997) and Cinque’s (1999) own reasoning, alternations like those in (56) would prove that two slightly different positions are available for the participle, a higher one in (56a) and a lower one in (56b). (56)â•… Quarna Sotto a.â•… i ŋ ku tSamaâ•… mi- n/ not- m theyâ•… haveâ•… yet called notâ•… of.him/â•… not-â•… me ‘They haven’t yet called him/ me’

122â•… Sentential negation:€adverbs b. i n ku mi/ mi ku Ju they are yet not/ not yet come ‘They haven’t come yet’

In general, Cinque (1999) must invoke relatively high or low positions for the participle to account for the fact that adverbs which (unlike Neg) do not have several positions at their disposal can occur to the left or right of the participle. Thus, in the examples involving auxiliary–participle structures provided so far, the adverbs in (1) mostly appear between the auxiliary and the participle, with the exception of ‘well’, which appears after the participle, as do the negation adverbs just discussed. Yet in Southern Italian varieties the preferred order has the adverbs in (1) following the participle, as illustrated by one example among many in (57); note that in (57a) the lexicalization of the negation is carried entirely by the clitic, on which more will be said in chapter 4. The logic of Cinque’s (1999) explanation would be that in (most) Southern Italian varieties the participle occupies a higher position than in (most) Northern Italian ones. (57)â•… Celle di Bulgheria (Campania) a.â•… nunn addZu vistu ccu/ aŋkra not I.have seen any.longer/ yet ‘I haven’t seen him yet/ any longer’ b. l addZu vistu ddZa him I.have seen already ‘I have already seen him’

Yet, as with the infinitive briefly mentioned in section 3.2 (cf. (24)), the analysis of the relative order of adverbs and participles proposed by Cinque (1999) is contradicted by evidence concerning the order of the participle with respect to other clausal material. There is evidence (cf. here chapter 5) that perfect participles define independent sentences, as is the case for the so-called absolute (i.e. adjunct) participles of standard Italian (Belletti 1990). These participles seem to be in quite a high position in their own sentence, since enclitics can attach to them. This also shows that the sentences they define are provided with a full inflectional layer€– in fact, if the high position of the participle is identified with the modal C position, they are also endowed with the C layer. In the Piedmontese varieties studied by Zanuttini (1997) and in this chapter, clitics in auxiliary–participle constructions do not typically procliticize to the auxiliary, as in standard Italian and most other Italian (or Romance) varieties, but rather appear post-participially. The data in (45) and (47c), introduced for independent reasons and reproduced below, illustrate this state of affairs. If, following Cinque (1999, 2006), we adopt the traditional point of view that perfect tenses are monoclausal, the positioning of adverbs between the auxiliary

Interaction of adverbial and verbal positionsâ•… 123 and the participle suggests a relatively low position for the participle€– at least, lower than the position of the finite verb, which precedes them. Yet the classical account just described leaves clitics completely out of the picture. Recall that in finite sentences these appear quite high before the finite verb; hence their post-participial position would suggest that the participle is higher than the finite verb€– clearly a contradiction. (45)â•… a.â•… b.

Margarita u Q pa pi/ ŋku nEŋ tSama-u he has not any.longer/ yet not called-him ‘He hasn’t called him him yet/ any longer’ Montaldo i  pQ ŋku nEŋ dZa fQ-u I have not yet not always done-it ‘I haven’t yet done it already’

(47) c.

Margarita j  pa pi nEŋ sEmpre vist-je them I.have not any.longer not always seen-them ‘I haven’t always seen them any longer’

The fact that the post-participial clitics of Piedmontese represent a problem for the general schema of explanation ultimately going back to Pollock (1989) is pointed out by Kayne (1991) in relation to infinitives. If, as already suggested, we take the clitic position as an indication that the participle is in a high position, say the C position, then this forces the conclusion that it defines an independent sentence. The auxiliary in turn defines an independent sentence whose complement is the participial sentence, while the monoclausal interpretation must be the result of incorporation processes applying between the matrix and embedded predicate. This general idea can be fleshed out along the lines of chapter 5 here, but allows for very different implementations, such as that of Kayne (1993). Though the bi-sentential construal of perfect tenses yields a straightforward analysis of clitic placement, a problem now arises with adverbs, specifically those in the sub-hierarchy in (1), which would be expected to appear systematically after the participle in C, given that they appear systematically after the finite verb in the lower I position. Here the theory of semantic selection for adverbs argued for by Svenonius (2002) and Ernst (2002) will help us to resolve the paradox. Participles may define full sentences on the evidence of the inflectional material they are able to associate with, yet traditional monoclausal treatments of perfect tenses embody an important interpretive intuition about participles,

124â•… Sentential negation:€adverbs namely that they lack independent temporal reference (though they may have aspectual/modal properties corresponding to their placement in C). The lack of temporal properties, while not interfering with the inflectional argumental domain (represented by clitics), allows very high syntactic projections like C to preserve a semantic characterization akin to that of lower predicational/eventive projections in tensed sentences. If so, while post-participial adverbs can be connected to the positions they occupy in finite clauses, pre-participial adverbs can be construed as attaching to domains higher than C, which are unavailable to them in ordinary finite clauses. In short, we assume that the participle is inserted in a fixed C position on the basis of equally fixed aspectual/modal properties. The variable order with respect to adverbs, both within a given variety and cross-linguistically, becomes a question about adverbs themselves, some of which will appear in the low eventive domain (after the participle) and others will attach to the higher modal C field (before the participle). One of the arguments that Cinque (1999) deploys in favour of his construal of the evidence is that while the position of the participle (or other verbal head) varies with respect to the adverbs, the latter maintain their order with respect to one another. A good example of this line of argumentation is provided by the relative position of ‘well’ in the hierarchy, following aspectual and quantificational adverbs. Though in Northern Italian varieties negation, aspectual and quantificational adverbs are generally found before the participle, as seen for instance in the data in (45) and (47c) reproduced in this section, ‘well’ follows the participle, as shown in (58). (58) Castellinaldo (Piedmont) a. u  piâ•… / aŋkura naŋ maï†−dZ baŋ he has any.longer/ yet not eaten well ‘He hasn’t eaten well yet/ any longer’ a’. u  naŋ sampre maï†−dZ baŋ he has not always eaten well ‘He hasn’t always eaten well’ Fontane (Piedmont) b. j aŋ piâ•… / ŋk Jent dymi bEŋ they have any.longer/ yet not slept well ‘They haven’t slept well yet/ any longer’ b’. j aŋ Jent sEmp dymi bEŋ they have not always slept well ‘They haven’t always slept well’

For Cinque (1999) the contrast between (45) and (58) implies that the participle is inserted between ‘always’ and ‘well’ in the fixed hierarchy in (1). For

Interaction of adverbial and verbal positionsâ•… 125 us, the data require a different analysis€– namely that the very same properties that typically associate ‘well’ with the lower position in the adverb rank also associate it with the core eventive domain. In other words, in order to adequately describe the split between the ‘well’ adverb positioned after the participle and other adverbs positioned before the participle, all we need is the notion that manner adverbs like ‘well’ hook on to a nominal N position while other adverbs appear in deictic/quantificational slots. Yet nothing in principle prevents an N adverb from appearing before the participle in the high C domain, which counts as an eventive domain in the absence of temporal specifications. This way of putting things is therefore less restrictive than Cinque’s (1999), for any rearrangement of the hierarchy in (1) is disallowed (at least in principle) by the fact that it is part of the computational system. As far as we can tell, the flexibility of the present theory is precisely what is required by the data. Specifically, ‘well’ is described by Cinque (1999) as occupying a Voice projection. This predicts that ‘well’ does not appear before active participles, while it will appear in front of passive participles that can move to the Voice head. In reality, our intuitions on standard Italian are that both examples in (59) are acceptable, so that pre-participial ‘well’ is independent of voice. In present terms, pre-participial occurrences of ‘well’ need not be restricted to low projections, but may equally appear in higher projections, as long as selection for events is respected. (59) a. b.

La commissione ha bene accolto la nostra proposta The committee has well received the our proposal ‘The committee received our proposal well’ La nostra proposta è stata bene accolta dalla commissione The our proposal is been well received by the committee ‘Our proposal has been well received by the committee’

As for quantificational and deictic (i.e. aspectual) adverbs, Southern Italian varieties of the type in (57) seem to require the fewest stipulations, since all adverbs of the low range in (1) follow the participle exactly as they follow the finite verb, in both cases appearing in the low eventive domain. In Northern Italian varieties of the type in (58), what seems to be happening is that quantificational/deictic adverbs are preferably inserted in the highest possible eventive domain, i.e. the C domain delimited by the participial projection. This parameter in turn seems to be just a manifestation of a very general variation scheme in natural languages. To take another example close at hand, in the perfect tenses of Piedmontese varieties like (45) and (47), the clitic occupies the post-participial position

126â•… Sentential negation:€adverbs corresponding roughly to its lowest possible position (in the inflectional domain of the predicate with which it has a direct selectional relation). In contrast, in standard Italian, as in most Italian and Romance varieties (French, Spanish, etc.), the clitic in the same structures occupies the highest position available to it (in the inflectional domain of the highest verb of the verbal complex formed by ‘restructuring’, i.e. roughly event unification). As far as we can tell, there is a complete parallel in all other respects between the Piedmontese perfects and, say, the standard Italian ones. In general, certain lexical items select certain domains of lexicalization in certain languages€– without any correlation with meaning, or other properties of the grammar. This is precisely the situation with (quantificational and deictic) adverbs, which are restricted to the postverbal lower range of the sentence by the presence of tense specifications in finite clauses, but in untensed clauses can range from lower to higher domains. In itself, our construal of the adverb placement parameter with respect to non-finite (and finite) verbs is neither simpler nor more complex than Cinque’s (1999) account in terms of the placement of the participle, for the lower or higher functional head to which the participle moves does not appear to be determined by general principles. We have seen that for Cinque (1999) the crucial argument in favour of the participle moving€ – and adverbs staying in place€– is that when a participle interrupts a sequence of adverbs, so that some are to its right while some are to its left, the two subsequences are still predicted by the hierarchy in (1). However, we have presented more than one counterexample to this prediction. Thus, the examples relating to Castiglione in (30) or to Quarna Sotto in (56) show that the negation can appear postparticipially, and aspectual adverbs before the participle, though aspectual adverbs are (or can be) preceded by the negation when they all occur in the same domain. Further discussion would require us to increase the already rich repertory of facts in this chapter, at least by adding data concerning the position of adverbs, including the negation, in infinitival sentences. At the same time it is evident that for infinitival sentences we could adopt the same principles of analysis explored here for participles, since the latter are treated here as independent sentences as well. Thus, consider the data in (23)–(24), of which we reproduce below just the Margarita examples. The post-infinitival position of the clitic (shared with many Romance varieties, including standard Italian) leads us to associate the infinitive with a high C projection (Kayne 1991). If so, the ordering of the negation and aspectual adverb before the infinitive represents a shifting of their position with respect to the tensed sentence€– parallel to the one discussed for participial sentences.

Interaction of adverbial and verbal positionsâ•… 127 (23)â•…

c.â•… Margarita j  di-te t nEŋ tSamE-uâ•… pi I I.haveâ•… told-youâ•… to not call-him any longer ‘I told you not to call him any longer’

(24) c.

Margarita j  di-te t pi nEŋ tSamE-u Iâ•… have told-youâ•… to any longer not call-him ‘I told you not to call him any longer’

Under the present analysis, we expect the relative order of adverbs to reproduce the hierarchy observed in finite sentences when they occur in the same domain, as in (24). But nothing prevents them from occurring in different domains€– and this will create what appear to be reorderings of the (underlying or superficial) hierarchy, as in (23). This type of evidence represents a direct counterexample to Cinque’s (1999) and Zanuttini’s (1997) strongest prediction€ – and is in fact difficult to account for within their analysis, except by stipulation.

4 Sentential negation:€clitics

Many Romance languages have a negative clitic which stands alone (for instance standard Italian) or is doubled by a negative clitic (for instance French). Since Pollock (1989) these have been modelled by a NegP projection in which the negative clitic and the negative adverb occupy the head and Spec position respectively. According to Zanuttini (1997), who extends this model to Italian dialectal variation, languages which require a sentential negation adverb gen­ erate it in one of the lower Neg positions; if a clitic combines with the adverb, it is generated in the head of the relevant Neg position and moves to the inflec­ tional domain by cliticization. By contrast, negative clitics in languages like Italian which do not require a sentential negation adverb are hosted by a Neg position generated above I, Neg1. In other words, for Zanuttini (1997) preverbal clitic negations are associ­ ated with two different structures, according to whether they stand alone or combine with a negative adverb. Moreover, she detects a correlation between these structures and the ordering of the negative clitic with respect to subject clitics in languages that have both of them. In her terms Neg1, hence the stan­ dalone negative clitic, will precede inflected subject clitics, though it may fol­ low vocalic uninflected subject clitics. Counterexamples to this generalization are Tuscan varieties in which the standalone negative clitic precedes 1st/2nd person subject clitics but follows 3rd person ones (cf. Càsola in (7) below). Zanuttini (1997:€37) recognizes that these varieties require two different pos­ itions for subject clitics€– not only one following Neg1 but also one preceding it. Yet this is not sufficient to account for a more direct counterexample like Vagli in (2) below, in which the standalone negation follows all inflected sub­ ject clitics. In turn, according to Zanuttini (1997) the negative clitic doubled by a negative adverb follows all subject clitics. Yet there are several examples of negative clitics doubled by negative adverbs in which the negative clitic also precedes the 1st/2nd person subject clitic while following 3rd person ones; Oviglio in (20) below is a case in point. Zanuttini (1997:€39) suggests 128

Sentential negation:€cliticsâ•… 129 that this should be explained in terms of two different positions occupied by (inflected) subject clitics, one following and one preceding the negative clitic. But again there are more direct counterexamples in which the negative clitic doubled by the negative adverb precedes the entire series of inflected subject clitics.1 From this evidence we conclude that there is no necessary correlation between the order of the negative clitic with respect to subject clitics and whether it doubles a sentential negation adverb or not. Correspondingly, in the rest of the chapter we discuss negation clitics without taking into account whether they stand alone or combine with an adverb. At the same time, there is a lot of literature which takes the clitic to instan­ tiate the logical negation operator (Rizzi 1982; Longobardi 1992 on Italian); this makes negative clitics natural exponents for the functional category Neg, on the assumption that its content coincides with the logical negation itself. Yet in chapter 3 we concluded that sentential negation adverbs (which can also stand alone) are nominal constituents and negative polarity items; this led us to recategorize them from exponents of the Neg functional category to quantificational (Q) elements in the eventive (adverbial) domain of the sen­ tence. In this chapter we argue that clitic negations are also best construed as nominal elements, anchored in the argument structure of the verb, and specific­ ally the (nominal or eventive) internal argument. Correspondingly they do not instantiate the specialized functional category Neg, but rather they instantiate nominal categories independently necessary for hosting pronominal clitics in the inflectional domain of the sentence. Like negation adverbs, they are sim­ ply negative polarity items, which need to be read in the scope of the negation (or other polarity operator)€– and therefore imply the introduction of a logical negation operator at the interpretive interface. If we are correct, the latter is never lexicalized in the languages at hand (and possibly in natural languages in general). The evidence to be presented concerns first of all the distribution of negation clitics with respect to subject and object clitics. In section 4.1 we consider data which involve copying of the negative clitic on either side of a subject clitic, as well as the classical distributions studied by Zanuttini (1997) and Poletto (2000). In section 4.2 we consider copying of a negation clitic on either side of an object clitic, and then we move on to simple instances of a negation clitic inside the object string€– both phenomena also observed by Parry (1995, 1997). In both cases the doubling of the negation clitic is sensitive to the socalled person split, roughly between 1st/2nd person and 3rd person. On these

130â•… Sentential negation:€clitics grounds we argue that clitic doubling could not be a result of purely phono­ logical processes (of the type envisaged by Harris and Halle (2005)), because the notion of person split that it is sensitive to is a syntactic one. Similarly, readjustment rules at Morphological Structure (in the sense of Halle and Marantz (1993)) would have to import this same syntactic notion (and others) into a separate morphological component€– which, even if possible, would amount to a description, rather than an explanation of the phenomena. As for analyses of doubling based on head-Spec configurations and subsequent movement, we note that a base-generated head-Spec configuration is not suf­ ficient to account for multiple copies of the clitic, since there is a unique head position that the clitic can fill. The surface distribution of the clitic copies, sensitive only to the nature of other elements of the clitic string, also casts doubt on the idea that they are spell-outs of clitic traces, since it remains to be shown that general principles of movement and spell-out could determine such a distribution. Our own analysis is based on the assumption that clitics are base-Â�generated in the positions where they surface, being connected to their copies or other lexical material by the interpretive calculus at the LF interface, as in so-called representational models (Brody 2003). The sensitivity of negation clitics to the person split is explained in accordance with the conclusions of chapter 3 on the similar sensitivity displayed by negation adverbs€ – namely as a result of their nominal character and their connection to the overall argument structure of the sentence. Section 4.2.1 considers potentially problematic cases where one of the apparently negative copies also surfaces in positive contexts. Because we treat the so-called sentential negations as nominal elements, connected with the argument structure of the sentence, in section 4.3 we are led to conclude that doubling of sentential negations is an instance of neg­ ative concord. We deal with the latter by assuming, as anticipated, that nega­ tions are negative polarity items, so that negative concord amounts to closure of two or more negative polarity items by the same logical negation (or by the same existential quantifier in the scope of the logical negation). We also briefly consider what lack of negative concord amounts to€– or, equivalently, mutual exclusion of negative polarity elements in the negative concord read­ ing. We suggest an account modelled on mutual exclusions elsewhere in the grammar, for instance between pronominal clitics. This account correctly predicts that mutual exclusion (lack of negative concord) facts are domain sensitive.

Interactions of negation with subject cliticsâ•… 131 4.1

Interactions of negation clitics and subject clitics

Let us consider a case of doubling of the clitic negation on either side of a sub­ ject clitic, as seen in the Northern Tuscan variety of Viano in the 2nd person singular in (1ii). In the other persons, reported in (1) under the correspond­ ing roman numbers, the negative clitic simply follows the subject clitic. The examples in (1ii) show that the doubling of the negative clitic is entirely indif­ ferent to the composition of the object clitic string. (1)â•… Viano (Tuscany) i. a n drm I not sleep ‘I don’t sleep’ ii. a.â•… a n t n drm ClS not you ClS sleep ‘You don’t sleep’ b. n t (n) m cam not you not me call ‘You don’t call me’ c. n t n l cam not you not him call ‘You don’t call him’ d. n t n t lav not you not you wash ‘You don’t wash yourself’ e. n t ŋ g l dE not you not there it give ‘You don’t give it to him’ iii. i/ la n drm he/she not sleeps ‘S/he doesn’t sleep’ iv. a n dormjaŋ we not sleep ‘We don’t sleep’ v. n durmi not you.sleep ‘You don’t sleep’ vi. i/ la n dɔrmn they.m/they.fâ•… not sleep ‘They don’t sleep’

The pattern in (1) is connected to the fact that cross-linguistically a negative clitic can appear either before or after the subject clitic, as illustrated here in (2)–(3), again with varieties from Northern Tuscany. To be more precise, in a

132â•… Sentential negation:€clitics variety like Vagli in (2), the negative clitic follows the subject clitic, while in a variety like Sillano in (3) it precedes the differentiated subject clitic and it follows invariable e. (2)â•… Vagli di Sopra (Tuscany) i nun dɔrmE tu n dɔrmE i/ Eâ•… nun dɔrmE nun sâ•… dɔrmE nun durmitE i nun dɔrmn I not sleepâ•… etc. ‘I don’t sleep’ etc. (3)

Sillano (Tuscany) (e) n(o) i dɔrma (e) non tu dɔrma (e) no ll dɔrma (e) non dormjaŋ (e) non durmidd (e) no ll dɔrmŋ ClSâ•… not I sleepâ•… etc. ‘I don’t sleep’ etc.

Poletto (2000) accounts for data similar to (2)–(3) by assuming that both subject and negation clitics have more than one position available to them. To begin with, in (3) the invariable subject clitic is higher than the inflected one, since the negation occurs between them. Furthermore, the negation in (3) is higher than the negation in (2) since the former precedes the inflected subject clitic, while the latter follows it. This yields a hierarchy roughly of the type:€SCl (invariable)€– Neg€– SCl (inflected)€– Neg. We adopt the conclusion that the two subject clitics in (3) correspond to two different subject clitic positions. Similarly we accept that there are two different positions for the negations, preceding the inflected subject clitics in (3) and fol­ lowing them in (2). Following the intuition of Chomsky (1995) as to the nature of the EPP argument, we notate the subject clitic as D. More importantly, we do not identify the position of the negative clitic with a dedicated functional projection Neg, though nothing would prevent us in principle from doing so. Rather, given the existence of an independent hierarchy for pronominal clitics, we simply assume that the negation fits into it€– and specifically corresponds to the R slot (generically suggesting Referentiality), which occurs immediately

Interactions of negation with subject cliticsâ•… 133 after the subject clitic and immediately before object clitics. Furthermore the two different positions of the subject clitic and the two different positions of the negation clitic do not correspond to different functional positions, but sim­ ply to two different instantiations of the same D and R positions€– internal to the I and C domains respectively. On these grounds we assign the structures in (4) and (5) respectively to, say, the 2nd person singular of the paradigms in (2) and (3). (4)â•… Vagli di Sopra D tu/i/ε R n(un) I dɔ rmε

(5)â•… Sillano D e R no (C)

e

D i/tu/l

I dɔ rmε

The structures in (4)–(5) amount to the proposal that Vagli and Sillano dif­ fer as to whether the negation is inserted in the domain immediately above C, as in Sillano, or in the domain immediately above I, as in Vagli.2 If the Cand I-domain positions of the negation in (4) and (5) are both instantiated, we derive the doubling of Viano in (1ii), as illustrated in (6).

134╅ Sentential negation:€clitics (6)╅ Viano D a R n (C) R t

R n

e

I dɔ rm

e

e

The question now arises of why the pattern in (6) is restricted to the 2nd person singular. A connection can be established with another pattern which singles out 2nd person clitics in their interaction with the negation. As illus­ trated in (7) with Càsola, again a Northern Tuscan variety, invariable subject clitics precede the negation, as expected; on the other hand, inflected subject clitics split, in that only 2nd person clitics follow the negation, while 3rd per­ son clitics precede it. Notice that both Càsola in (7) and Viano in (1) lack inflected clitics for the 1st person. Therefore we may equally well describe the facts by saying that in Càsola in (7) the negation precedes all inflected 1st and 2nd person subjects and follows 3rd person subjects; similarly, in Viano in (1) doubling opposes all inflected 1st and 2nd person subjects to 3rd person sub­ jects. In other words, both types of language can be described as instantiating a classical person split between 1st /2nd person and 3rd person. (7)â•… Càsola (Tuscany) a n Dɔrm n t Dɔrm i/la nâ•… Dɔrm a n Durmiaŋ a n v Durmi i/la n Dɔrmn ClS Neg ClS sleep etc. ‘I don’t sleep’ etc.

The structures proposed for the negation clitic that follows the inflected subject clitic, as in the Vagli example in (4), or precedes it, as in the Sillano example in (5), suggest an account of the alternating pattern of a language

Interactions of negation with subject cliticsâ•… 135 like Càsola along the lines of (8). In (8a) the negation is inserted within the I domain and therefore follows the 3rd person subject clitic. In (8b), assuming that the position of 1st and 2nd person subject clitics remains constant, the neg­ ation precedes them in that it is inserted in the C domain. (8)â•… Càsola a. [D i/a [R n b.â•… [R n [C [D t

[I Dɔrm [I Dɔrm

In chapter 3 we analysed the interaction of the person split with the adverbial negation on the basis of the assumption that the so-called adver­ bial negation is neither adverbial nor, strictly speaking, negative; rather, negative adverbs are nominal elements interpreted as negative polarity items. Furthermore, we suggested a characterization of 1st and 2nd per­ son arguments as ‘discourse-anchored’ and 3rd person arguments as ‘eventÂ�anchored’. Now, the pattern in (8), far from appearing arbitrary, can be seen as a more abstract version of the Quarna split in chapter 3. In other words, the negation (here the clitic) has two different lexicalizations (here, posi­ tions) depending on the presence of a 1st/2nd or 3rd person argument (here the EPP argument).3 The Viano example in (6), repeated here in (9b), is a variant of the Càsola example in (8b). Thus, in (9b) a copy of the C-domain negation is inserted when the subject clitic is 1st or 2nd person. By contrast, the I-domain copy is the sole lexicalization of the negation with the 3rd person subject clitic in (9a). From another perspective, Càsola and Viano in (8)–(9) are sensitive to the same parameter between the I-domain and C-domain lexicalizations of the negation as Vagli and Sillano in (4)–(5), except that one of the two negations is generalized to the whole paradigm in (4)–(5), whereas they split according to person in (8)–(9). (9)â•… Viano a. [D i/la [R n [I Dɔrm b. [D aâ•… [R nâ•… [C [D t [R n [I Dɔrm

In the theory proposed by Poletto (2000), where data of the type in (7) are also considered, the relative order of subject clitics and the negation depends on a clitic hierarchy of the type described above, roughly ClS(invariable)€ – Neg€– ClS(3rd)€– Neg€– ClS(2nd)€– Neg. Under this account, the negation of Càsola as in (7) would be the middle one, since it precedes ClS(2nd) but follows ClS(3rd). This hierarchy could also correctly derive the doubling of Viano as in (1), where ClS(2nd) is flanked by two copies of Neg. Poletto’s (2000) model predicts the impossibility of the reverse order to the one found in Càsola, with

136â•… Sentential negation:€clitics the negative clitic preceding the 3rd person subject clitic and following the 1st and 2nd person clitic. This is because her clitic hierarchy, roughly ClS€– Neg€– ClS(3rd)€– Neg€– ClS(2nd)€– Neg, forces any negation that precedes ClS(3rd) to also precede ClS(2nd). However, when it comes to doubling, the same hier­ archy can yield the unattested string Neg€– ClS(3rd)€– Neg, assuming that the highest and lowest Negs of the hierarchy are instantiated. Here, starting with the two different positions of the negation in (8), we are able to predict the doubling in (9), as opposed to the reverse order. However, we have not yet explained why the negation and the subject clitic pattern as in (8), as opposed to the reverse€– since, contrary to Poletto (2000), the position of the negation with respect to 1st/2nd person clitics and 3rd person ones is not written into a hierarchy. In other words, the question is why the negation cannot be forced into the C domain by an event-anchored (3rd person) EPP argument, while remaining within the I domain in the presence of a discourseanchored (2nd person) argument. If we construe the negative clitic as a nom­ inal element and a negative polarity item, the generalization is that the presence of a discourse-anchored argument provokes the lexicalization of argumental material (the so-called negation clitic) in the C domain, thus indirectly linking discourse-anchoring and the C domain. On the contrary, event-anchoring is connected to the I domain.4 Independent evidence suggests that the connection just established is cor­ rect. Thus, in the imperatives of Southern Italian and Albanian varieties, dis­ course-anchored and event-anchored object clitics are found in the C and I domains respectively€– the former in mesoclisis (Manzini and Savoia (2009a), and references quoted there), and the latter in enclisis. The higher position of discourse-anchored elements with respect to event-anchored ones is also con­ firmed for non-related languages (for instance Salish, as discussed by Davis (1999)). This casts further doubt on Poletto’s (2000) hierarchy, in which the 3rd person subject clitic is higher than the 1st or 2nd person one. Crucially, the interactions of the negative polarity clitic with the person split require that it also participates in the definition of predicate–argument structure. If it was simply a logical connective, the reason for such an interaction would remain mysterious€– or at the very least could not fall under the generalization just proposed. Other interactions between the so-called negative clitic and subject clitics concern mutual exclusion phenomena which affect the combinations of sub­ ject and object clitics in Romance varieties. We observe that in some of these varieties the exclusion of the subject clitic (specifically 3rd person) is induced not only by object clitics, but also by the clitic negation. The data in (10a) show that Agliano has subject clitics that are obligatorily lexicalized in the absence

Interactions of negation with subject cliticsâ•… 137 of either negation or object clitics. The negation in (10b) excludes the subject clitic, with the possible exception of discourse-anchored clitics (in practice the 2nd person singular). The examples in (10c–c’) show that object clitics have the same effect of excluding the subject clitic€– though again they optionally combine with a discourse-anchored one. (10)â•… Agliano (Tuscany) a. (i) Drm tu Drmi i/la ddrma/Drma s Drma i Durmit i/la ddrmn/ Drmn ClS sleep ‘I sleep’â•… etc. b. nun drm nunâ•… (tu) drmi nun drma nun s drma nun durmit nun drmn Neg (ClS) sleep ‘I don’t sleep’â•… etc. c. (tu) l/la/ m cami youâ•… him/her/me call ‘You call him/her/me’ c’. l/â•… la/ɟi / m/ tâ•… camn/camat him/her/them/â•… me/us they.call/ you(pl).call ‘They/ you call him/her/them/me/us’

Mutual exclusions between subject and object clitics are similar to a mutual exclusion more often discussed in the literature on Romance, namely that between two object clitics€– for instance that between the dative and the accusative clitic in the Spurious se of Spanish. Current analyses in Distributed Morphology, Optimality Theory, etc. are based on the idea that there is a con­ straint against the co-occurrence of certain forms, and that one of them is even­ tually substituted for by a ‘default’. We propose a different model, which does not have recourse to notions of competition or default. Rather, only one of two clitic forms is lexicalized in that this form subsumes the crucial properties of the other (the non-lexicalized one). In cases of generalized exclusion of the subject clitic by the object clitic, as in (10c–c’), we propose that any pronominal clitic, independent of its denotation, is sufficient to lexicalize D(efiniteness) proper­ ties for the entire clitic string/domain and hence to exclude the subject clitic, a pure instantiation of such properties.5 The interest of data such as the Agliano

138â•… Sentential negation:€clitics examples is that the negative clitic in (10b) behaves exactly like an object clitic in excluding the subject clitic. This is expected if the so-called negation is nothing more than a nominal element, and specifically a negative polarity item. If the negative clitic was the lexicalization of the negative operator, it would be very difficult (or impossible) to see why it should interact with the argumental clitic series in the way it does. Another relevant property of Agliano is the different treatment of eventÂ�anchored and discourse-anchored subject clitics in (10b) and in (10c) vs. (10c’). As just proposed, the pure D(efiniteness) properties of an event-an­ chored (3rd person) subject clitic are subsumed by the lexicalization of any other element of the clitic string, yielding for instance (10c’); however, this is not necessarily the case for the deictic reference associated with speaker/ hearer, which can therefore be independently lexicalized, as for instance in (10c). Incidentally, since the negation precedes the 2nd person clitic when they co-occur in (10b), we conclude that Agliano has the same structure as Càsola in (8b). 4.2

Interactions of negation clitics with object clitics

In the examples presented so far, the negative clitic systematically precedes object clitics, as is usual in Italian varieties. However, cases in which the nega­ tive clitic appears inside the object clitic string are noted in the literature, in particular by Parry (1997) for Cairo Montenotte, in which the negative clitic precedes 3rd person objects but follows 1st and 2nd person ones. What is more, Parry (1997) observes that in some varieties of Liguria and Piedmont the nega­ tive clitic can appear both to the right and to the left of the 1st and 2nd per­ son clitic. Zanuttini (1997:€18) suggests that only languages with doubling of the negative clitic by an adverbial negation allow this doubling. In reality, the Northern Tuscan varieties that we exemplify below are a counterexample to this generalization. Let us begin by considering the relatively simple case of Bedizzano in (11), in which only one instance of the negation appears, following 1st and 2nd per­ son object clitics and preceding 3rd person ones. This positioning of the nega­ tive clitic within the object clitic string is insensitive to the person of the verbal paradigm, as can be seen from the comparison between 2nd and 3rd person in (ii) and (iii) respectively. (11)â•… Bedizzano (Tuscany) ii.â•… a.â•… t m n cam you me not call ‘You don’t call me’

Interactions of negation with object cliticsâ•… 139 b. c. d. iii. a. b. c.

t nâ•… l cam you not him call ‘You don’t call him’ t m n l/n da youâ•… me not it/of.itâ•… give ‘You don’t give me it/any of it’ t n i l da youâ•… notâ•… to.himâ•… it give ‘You don’t give it to him’ i tt/ss/vv n l/n da (pr ent) heâ•… you.sg/us/you.plâ•… notâ•… it/of.itâ•… givesâ•… at all ‘He doesn’t give it/any of it to you/us’ i n i  da he not to.him it gives ‘He doesn’t give it to him’ i n i/s da kwEst he not to.him/to.us gives this ‘He doesn’t give this to him/us’

Doubling data are provided in (12) for Colonnata. The negative clitic Â� follows all subject clitics and precedes 3rd person object clitics, both accusa­ tive and dative (as in (iiib)). At the same time, the negative clitic both pre­ cedes and follows 1st and 2nd person clitics as well as the se-type clitic (as in (iiia)). (12)â•… Colonnata (Tuscany) i. a. a n t n we I not you not see ‘I don’t see you’ b. a n t n l ‘dag I not you not it give ‘I don’t give it to you’ c. a n l ve I not it see ‘I don’t see it’ ii. a. t n t n lav you not yourself not wash ‘You don’t wash yourself’ b. t n l ve you not it see ‘You don’t see it’ iii.â•… a. i n s n lav he not himself not washes ‘He doesn’t wash himself’ b. i n i l da he not to.him it gives ‘he doesn’t give it to him’

140â•… Sentential negation:€clitics c. i n t n l da he not you not it gives ‘He doesn’t give it to you’

We approach the doubling data of Colonnata in (12) assuming that the nega­ tive clitic is doubled because it is inserted both in the C and in the I domain; the possibility of a negative clitic being inserted in either domain has been independ­ ently motivated in section 4.1, and its doubling in both domains allowed us to account for the doubling on either side of a subject clitic. We can account for the Colonnata data if we assume additionally that 1st and 2nd person clitics (P in the present notation, to suggest Person) insert in the C domain, so that they precede the lower copy of the negation and follow the higher copy, as in (13). This ana­ lysis furthermore requires all subject clitics to be generated within the C domain. Note that we notate 3rd person accusative clitics as N (cf. here chapters 7–8). (13)

Colonnata

D i R n P t

e (C) R n

e N l I da

If we apply the approach in (13) to the non-doubling data of Bedizzano, we obtain structures like (14), which display the person split between 3rd person (N) clitics in the I domain and other clitics in the C domain, so that the former follow and the latter precede the negation in the I domain.

Interactions of negation with object cliticsâ•… 141 (14)

Bedizzano

i

D P t

e (C) R n

e N l/n I da

The question is what explanatory value the structures in (13)–(14) have. It appears that the relevant generalization is the same as that formulated in the conclusions of section 4.1, namely that event-anchored clitics are associated with the I domain, while discourse-anchored ones are associated with the C domain. In (13), the negation splits along the same lines as object clitics, namely one copy occurs in the I domain and one copy in the C domain. Note that in this analysis, all subject clitics are in the C domain€– we may take this to be due to their EPP nature, potentially extraneous to the event- vs. discourse-anchored contrast. Recall that even in the varieties of section 4.1, all inflected subject clit­ ics occurred in the same domain (the I domain in that case)€– and it was the neg­ ation that eventually split between two domains. In other words, the behaviour of the negation is closer to that of object clitics than that of subject clitics. In Southern Italian and Albanian varieties in which discourse-anchored and event-anchored object clitics occur in imperatives in the C and I domain respectively, locatives and the si-type clitic pattern with 1st and 2nd person. The same is true of se in the Bedizzano example in (11iiia). Descriptively speaking, this is the 3rd person reflexive, but in present terms it is the free variable of the argumental clitic system (cf. here chapters 5–6). Therefore the split must involve something like the present notion of discourse- and event-anchoring rather than just the opposition of 1st and 2nd person vs. 3rd person. It is important to recall that what we are seeking to analyse is finely struc­ tured parametric space. What we find is a set of universal categorial distinc­ tions. One such distinction is that between the C and the I domain. This interacts with the person split either directly, so that event-anchored complement clitics have an I lexicalization and discourse-anchored ones have a C lexicalization, or

142â•… Sentential negation:€clitics indirectly, so that the C negation associates with the discourse-anchoring of the EPP argument and the I negation with its event-anchoring. Despite the unpre­ dictable way in which the relevant distinctions show up in a given language, what we find is that the patterns they give rise to may not be reversible. Thus it is always discourse-anchored elements that are higher than event-Â�anchored ones€ – more precisely, that trigger the lexicalization of the C domain (for instance by the negative polarity clitic) as opposed to the lexicalization of the I domain in combination with event-anchored elements. The fact that negative clitic doubling interacts with LF interface notions such as that of person split excludes the possibility that it could be a purely morpho(phono)logical phenomenon. For instance, the copying and displace­ ment processes targeted by Distributed Morphology (Halle and Marantz 1993), or more recently by Harris and Halle’s (2005) theory of metathesis, can affect clitic material entirely comparable to negative n. In other words, since the negative clitic is clearly part of the clitic cluster, there is no reason why its copying (or displacement) should not be handled at the morpheme structure/ phonological form (MS/PF) interface as that of argumental clitics routinely is. However, if the preceding discussion is correct, the notions necessary to account for the present data are syntactic/LF notions, such as those of person split or I vs. C domain. To be more precise, it is possible to import the primi­ tives necessary to state the correct distribution of the negative clitic (1st/2nd vs. 3rd person, and so on) at the MS/PF interface. The problem, however, is one of explanatory adequacy. On the one hand, a theory in which the relevant notions are available in more than one component must surely count as more complex than a theory in which they are handled in a unified (morphosyntac­ tic) component. More importantly, if the interaction between the person split and the placement of the negation is to be understood in terms of the lexicali­ zation of the relevant clitics in I and C domains, then this essential aspect of our explanation cannot be reproduced at MS/PF at all. In our view, this type of consideration argues against any attempt to reduce negative doubling to an MS/PF process, and it also indirectly casts doubt on MS/PF accounts of pro­ nominal clitics€– whose parallels with negative clitics seem to call for a unified analysis. Next, in (15) we report data similar to those in (11), but taken from a variety (Càrcare) of the type studied by Parry (1997), with the negative clitic doub­ ling a negative adverb identical to the argument for ‘nothing’. It will be noted that the sentential negation adverb is in complementary distribution with other negative adverbs, as in (15iib–c), and arguments, as in (15iiib), and in fact appears to be altogether optional, as in (15iiic). Regarding this last example, we assume that (true) optionality simply reflects the presence of different

Interactions of negation with object cliticsâ•… 143 grammars in the competence of the same speaker. As for the question of nega­ tive concord, i.e. the possibility (or impossibility) of combining two or more given negative forms with a single negation meaning, we return to it in section 4.3€– in connection too with the more general problem of how to interpret a doubled clitic, or the doubling of phrasal material by a clitic. What is relevant here is that in (15) the negation is expressed by a triplet consisting of a negative adverb or phrasal argument doubled by a pair of clitics. The position of these is the same as that observed in (12); thus, they precede and follow P clitics and si-type clitics, in all cases preceding accusative/Â�partitive ones. Subject clitics generally precede the higher copy of the negative clitic. However, they can either follow or precede the 2nd person singular subject clitic, as in (15iib) and (15iic) respectively. What is more, these two potential positions of the negative clitic can combine, much as observed in (1); thus we obtain sentences of the type in (15iia), in which three copies of the negative clitic are present alongside the negative adverb. (15)â•… Càrcare (Liguria) ii. a. Eŋ t Eŋ t Eŋ lɔvi nE:nt notâ•… youâ•… notâ•… yourselfâ•… notâ•… washâ•… nothing ‘You don’t wash yourself’ b. Et Eŋ m Eŋ tʃɔmi mɔi you not me not call never ‘You never call me’ c. Eŋ t Em Eŋ lE dɔi mɔi not you me not it give never ‘You never give it to me’ iii. a. u ŋ s Eŋ lɔva nE:nt he not himself not washes nothing ‘He doesn’t wash himself’ b. u ŋ m Eŋ dɔ nE:nt he not me not gives nothing ‘He gives me nothing’ c. u ŋ m Eŋ lE/nuŋ dɔ he not me not it/of.it gives ‘He doesn’t give it/any of it to me’

We can extend to varieties of the Càrcare type the same treatment already proposed for Bedizzano or Colonnata. Specifically, the lower negative clitic appearing after the P object clitic can be lexicalized within the I domain, as in (16) below, while the object P clitic itself and the copy of the negation pre­ ceding the P clitic are found in the C domain. As we have seen, an interesting property of Càrcare is that the negative clitic can in fact be trebled, with its highest copy appearing in front of the subject clitic. Since the doubling of the negation on either side of the subject clitic is sensitive to a person split, as in

144â•… Sentential negation:€clitics Viano (discussed in section 4.1), it is natural to extend to the cases at hand the analysis proposed in section 4.1. The problem is that in Viano the discourseanchored subject clitic was in the I domain, so the copy of the negation that preceded it could be in the C domain. But since, for the reasons just reviewed, the discourse-anchored subject clitic is in the C domain in (16), the highest copy of the negation must be lexicalized in a still higher domain. More specif­ ically, we can assign it to the domain immediately above C in a split-C analysis of the type proposed by Rizzi (1997); this is CI in (16) below. (16)â•… Càrcare R εn (C1) D t

R εŋ

P t

(C) R εŋ

I l vi Q nεnt

c

It seems to us that there is no way of accounting for the data (or for their variation) in terms of a base-generated head-Spec configuration to which move­ ment subsequently applies. Apart from general concerns regarding the status of the head-Spec configuration (Starke 2004; Chomsky 2008), important coun­ terevidence is represented by the possibility of having more than one negative clitic, hence more than one potential head of the construction. It is true that doubling could be analysed as the result of multiple spell-out of the copies that a single clitic head leaves behind in the course of the derivation. However, the distribution of negative clitics is constrained exclusively by the distribution of other material in the clitic string. In no case is there any evidence that the sur­ face distribution depends on the presumed base-generated head-Spec config­ uration. The latter therefore represents an empirically unmotivated enrichment of the theory. Similarly, an explanation of doubling in terms of copying and

Interactions of negation with object cliticsâ•… 145 multiple spell-out presupposes that the placement of the clitic copies can be determined on the basis of general constraints on movement, and their morp­ holexical properties on the basis of PF rules. It seems to us that the burden of proof falls squarely on the proponents of such a theory€– on both counts. 4.2.1 Non-negative n A potential problem for the analysis of negative clitic doubling comes from the fact that in varieties of Liguria and Piedmont, P clitics can be followed by n morphology even in positive contexts. As it turns out, there are indications that in these contexts as well, the distribution of n is syntactically determined. Thus, in Dego in (17), the 1st person singular alternates between m if an accusative clitic is present and m-ŋ in accusative-less contexts. The alternation cannot be phonologically governed, since both the verb in (17a) and the accusative in (17b) are monosyllabic forms with an initial consonant. A similar contrast is noted by Parry (1997) for Rocca d’Arazzo. (17)â•… Dego (Liguria) a.â•… t m-Eŋ tmi you me call ‘You call me’ b. u m li/ a/ i d he me it-m./it-f./them gives ‘He gives it/them to me’

Furthermore, the n morphology that is in complementary distribution with the accusative in (17) appears after the P clitic in negative contexts, irrespect­ ive of the presence or absence of an accusative. Thus, in negative contexts, m followed by Eŋ combines with the accusative as well as with the partitive, as in (18). If we take Eŋ in these contexts to instantiate the negation, then this repro­ duces the distribution studied for the varieties of section 4.2 (without doub­ ling), for instance Bedizzano. (18)â•… Dego a.â•… u m/t Eŋ li d he me/you not it gives ‘He doesn’t give it to me/you’ b. u m  naŋ d ClS me not of.it gives ‘Nobody gives any of it to me’

nE:nt nothing niSyŋ nobody

Following the conclusions of the preceding section, we assume that the nega­ tive (polarity) clitic in Dego is inserted in the R position of the pronominal clitic string, as shown in (19a), where it yields no mutual exclusion effects. On

146â•… Sentential negation:€clitics the other hand, the mutual exclusion between the  segment and the accusa­ tive in the positive contexts in (17) can be accounted for by assigning it to the lowest position in the clitic string, otherwise occupied by the elements with which  is in complementary distribution, i.e. the accusative and partitive. In our notation this position is N, as in (19b). (19)â•… Dego a. D u P m/t (C) R εŋ N li I d�

Q nε:nt

b. D t P m (C) N εŋ I tʃ� mi

If the occurrences of the  morphology in the two contexts in (19) involve two different lexical entries, then not only their homophony but also their simi­ lar interaction with P clitics remain purely coincidental. Thus, negative  yields person split structures like (19a), where P clitics are inserted in the C domain and N clitics in the I domain. In the same way, positive  appears to be selected by P clitics. Assuming, then, that there is a single lexical entry for , its differing distribution and interpretation in negative and positive contexts need to be accounted for.

Interactions of negation with object cliticsâ•… 147 Before considering this question, we introduce evidence from the variety of Oviglio, in which the complementary distribution between n morphology and the accusative is observed in negative contexts. In Oviglio, what appears to be the higher copy of the negative clitic follows an already familiar pattern, appearing after subject clitics, except for the 2nd person one, which it pre­ cedes, as in (20). In turn, what appears to be a lower copy of the negative clitic, nu, is inserted after P clitics. However, this differs from the negative clitics considered in section 4.2 in that it is in complementary distribution with the accusative and partitive, as in (21). What is more, the data from, say, Bedizzano show that it is the lower negative clitic (the one inside the object string) that is obligatory in the absence of doubling. By contrast, in the Oviglio variety it is the higher n clitic that is obligatory in non-doubling examples. In all cases the negative clitic or clitic pair is obligatorily doubled by the sentential negation adverb næint(a) ‘nothing’. (20)â•… Oviglio (Piedmont) a n drm a n t drmi u n drm a n drumuma i n drmi i n drmu ClSâ•… not ClS sleep ‘I don’t sleep’â•… etc.

næinta næinta næinta næinta næinta næinta not

(21)â•… Oviglio i.â•… a.â•… a n t nuŋ tSam næinta I not you not call nothing ‘I don’t call you’ b. a n t El dag næint I not you it give nothing ‘I don’t give it to you’ ii. a. a n t um nu tSami næinta ClSâ•… not you me not call nothing ‘You don’t call me’ b. a n t um El dai næint ClS not you me it give nothing ‘You don’t give it to me’ t nu tSama næinta iii a. u n he not you not calls nothing ‘He doesn’t call you’ b. u n El tSama næinta he not him calls nothing ‘He doesn’t call him’

148â•… Sentential negation:€clitics The distribution of the n clitic is sensitive to the familiar person split, whereby it appears before P subject clitics but after 3rd person ones. Under the analysis of section 4.1, the position following 3rd person clitics reflects the lexicaliza­ tion of the negation clitic within the I domain, as in (22b). The position before the P subject clitic reflects the lexicalization of the negation in the C domain, as in (22a). In both cases the complementary distribution between nu and accusative clitics can be accounted for if nu is inserted in the N position. (22)â•… Oviglio a. D a

R n (C) D t

P um

N nuŋ I t ʃami Q næinta

b. D u

R n (C) P t

N nuŋ

I t ʃama Q næinta



A further twist on this pattern is that the nuŋ negative form has the same form as the partitive, as seen for instance in (23a). To be more precise, partitive nuŋ

Interactions of negation with object cliticsâ•… 149 has an n alternant appearing in front of (auxiliary) verbs beginning with a vowel, as in (23a’). Similarly but not identically, while nuŋ lexicalizes the lower neg­ ation in (23b–c) before a verb beginning with a consonant, it does not surface in front of (auxiliary) verbs beginning with a vowel, as in (23b’–c’). Incidentally, (23c) establishes that the middle-passive clitic si, despite its association with the 3rd person, behaves like a discourse-anchored clitic, followed by nuŋ. In other words, it is not a pure person split that is relevant, but a more abstract split, such as the one encoded here through the notions of discourse and event anchoring. (23)â•… Oviglio a.â•… u nu da doi a pr e he of.themâ•… gives two to each ‘He gives two of them to each one’ a’. aâ•… n o dahtS du a pr e I of.themâ•… have given two to each ‘I have given two of them to each one’ b. aâ•… m nu o næint lava I myself not am nothing washed ‘I haven’t washed myself’ b’. aâ•… n t o næint tSama I not you have nothing called ‘I haven’t called you’ c. u  nu drɔm næinta it M not sleeps nothing ‘One doesn’t sleep’ c’. u   E næint lava he not M is nothing washed ‘He hasn’t washed himself’

There are several reasons not to treat the formal identity of the negation and the partitive as a pure case of homophony. One of them is the complementary distribution between the negation and the partitive (more generally the N argu­ ment, including the accusative). Another reason is the otherwise unexpected restriction to contexts before verbs beginning with a consonant. Suppose then that we provide a unified lexical entry for nu. All of the evidence we have presented suggests that we should characterize it as the partitive. This charac­ terization is supported by the fact that€– in contrast to the varieties of sections 4.1€– it is the higher copy of the negation that is obligatory in the absence of doubling; thus, nu is not necessary for the negative interpretation of the sen­ tences in which it occurs. If nu is the partitive, its distribution in negative contexts can be accounted for by assuming that it must be lexicalized in the scope of the negation, if the internal argument is a discourse-anchored element. Indeed, in chapter 3 we

150â•… Sentential negation:€clitics observed that the negative adverb mia of Quarna selects for N objects€– lexi­ calized as partitives when they are clitics. Similarly, in Oviglio the negation requires the object to be an N€– inserted in the form of the partitive nu when otherwise only discourse-anchored elements are present. In non-negative con­ texts, nu is inserted under the ordinary circumstances that require a partitive form. On the basis of Oviglio, we can also return to the slightly more complex case of  in Dego. In negative contexts,  has the same form as the nega­ tive polarity clitic. Now, as discussed in more detail in section 4.3, a negative polarity item is essentially an indefinite, i.e. a free variable, which gets exist­ entially closed, and assumes its negative value by being interpreted in the scope of a negation (or other modal) operator. Suppose  is such an element. By definition, in positive contexts the negative operator is not present, and  in Dego is read existentially. Following the discussion of Oviglio, we fur­ ther assume that in positive contexts Dego must lexicalize  in N when the internal argument is discourse-anchored, simply because the lexicalization of the internal argument in this language requires N material€ – essentially as it does in Oviglio in negative contexts. Structurally, the negative reading of  appears to correlate with the R position and the positive reading with the N position. Evidently the negative reading requires  to be in a quantifi­ cational/left edge position, while N receives the ordinary internal argument interpretation. The various dimensions of language variation implicit in the accounts for Dego or Oviglio can also be shown to be realized independently of one another, which is an important argument in favour of the present analysis. Thus, P clitics ending in n morphology in non-negative contexts are found in varieties which, at least descriptively, do not have any negative clitics, as in S. Bartolomeo in (24). The distribution of n is syntactically determined by the presence vs. absence of an accusative, as shown in (24a) vs. (24b). The sentential negation, involving only a negative adverb, is exemplified in (24c). (24)â•… S. Bartolomeo Pesio (Piedmont) a.â•… u m-â•… d susi he me gives this ‘He gives me this’ b. u m lu d he me it gives ‘He gives it to me’ c. i lu tSam Je I him call nothing ‘I don’t call him’

Interactions of negation with object cliticsâ•… 151 The way we accounted for the complementary distribution of the  morph­ ology with accusatives in a variety like Dego was to say that  is inserted in N in sentences where the internal argument is otherwise lexicalized by a P element. Data like those of S. Bartolomeo indicate that this type of distri­ bution is independent of  expressing the negation. In other words, though the two interpretations can coexist in the same lexical item, as in Dego, one is independent of the other. Not only are there very many languages in which only the negative interpretation is attested, but there are also varieties like S. Bartolomeo in which only the non-negative one arises.6 Similarly, the formal identity of what are descriptively the partitive and the negation is not an isolated phenomenon. Thus, in Càrcare the partitive, illustrated in (15iiic), has the same form as the lexicalization of the negation in modal contexts, such as the negative imperative (morphosyntactically an infinitive) in (25). On the basis of the proposals advanced so far, it is natural to assume that the same lexical element, associated with the internal argument, is involved both in (25) and in the partitive. We may further speculate that when it is lexicalized in the I domain, as in (15iiic), its interpretation is partitive. In modal contexts, on the other hand, it must be lexicalized in one of the domains of the articulated C field, since it precedes the verb, which itself is in C or higher. In this position it will be read as a polarity specification, as in (25). (25)â•…

Càrcare nu St-lE a tSamE not stay him toâ•… call ‘Don’t call him’

In short, the presence of elements identical to what we have characterized as copies of the negation in positive contexts might at first suggest that they correspond to uninterpretable material, whose relevance is purely prosodic or computational. In reality, the data reviewed in this section, far from support­ ing this conclusion, provide evidence in favour of the idea that the negation is nominal (and argumental) in nature. Thus we account for the fact that what appears to be a lower copy of the negation clitic can have the same form as the partitive (Oviglio, Càrcare) or can lexicalize the same N slot as it does (Dego, S. Bartolomeo). If instead the negation copies corresponded to the mul­ tiple spell-out of the head of a specialized Neg category, their presence in nonnegative environments, and more to the point, their formal identity (lexical or distributional) with the partitive could not be accommodated. Similarly, a prosodic analysis of doubling could not account for the cases in which the socalled negation copy has the same form as a clearly contentful element, such

152╅ Sentential negation:€clitics as the partitive. In either case, all of the lexical coincidences examined in this section would have to be treated as cases of homophony, obscuring the patterns that connect them. 4.3

Negative concord and negative doubling

Our analyses so far motivate the conclusion that a negative clitic is a nom­ inal element, specifically a negative polarity item; hence it introduces not the negative operator, but simply a variable read within the scope of the negative (or other modal) operator. On this point we differ from much of the literature (Rizzi 1982; Longobardi 1992 on Italian), which takes the clitic to instantiate the negative operator. If the negative clitic is a negative polarity item, it is evi­ dent that it must itself be read in the scope of a sentential negation operator. The latter is therefore not introduced by any morpholexical constituent, but rather is semantically implied by the presence of the negative polarity clitic (or other negative polarity material). As mentioned in chapter 3, the most imme­ diate argument in favour of the polarity status of n-words in Romance (Rizzi 1982; Longobardi 1992; Acquaviva 1994 on Italian) comes from the fact that they occur in modal (irrealis) environments without any implication of nega­ tive meaning. Strikingly, the argument holds not only for n-phrasal units but also for n-clitics whose non-negative occurrences have been studied in the lit­ erature as instances of ‘spurious’ or ‘expletive’ negation (e.g. Belletti 2000).7 Let us now return to simple examples of the type in (2)–(3) or (7), in which the sentential negation is represented by a single negation clitic. Their LF includes a negative operator, i.e. ⌐. in whose scope the negative polarity item represented by the n-clitic is interpreted, as in (26). If we consider negative polarity items to be indefinites, i.e. free variables (Heim 1982), we may very well assume that the variable introduced by the so-called negative clitic is existentially closed (Acquaviva 1994) in the scope of ⌐. (26)â•… a.â•… b.

Vagli [⌐ [∃x [D i/ E [R nun (x) [I drmE Sillano [⌐ [∃x [R no (x) [D ll [I drm

Consider, then, the case routinely described in terms of the doubling of a clitic negation by an adverbial one, as illustrated here by the paradigm in (20), for instance. Since the so-called negative clitic has exactly the same status as any other polarity argument in the sentence, it is evident that each time it �co-occurs with one of them, the resulting configuration is one of negative concord under which all variables are read in the scope of a single negation

Negative concord and negative doublingâ•… 153 operator. Under the set of assumptions introduced here, negative concord is in fact the expected state of affairs. Quite simply, the variables introduced by the clitic and by the adverb are both interpreted in the scope of the same ⌐ and existential closure operators, as in (27). Thus we predict that there is a single instance of the negation at the interpretive level. (27)

Oviglio [⌐â•… [∃x,yâ•… [D aâ•… [R n (x)â•… [D tâ•… [I drmiâ•… [Q næinta (y)

In turn, the doubling of a negative clitic by another is identical in all relevant respects to its doubling by an adverb. Thus, consider (1) again, with simple negation in the 3rd person (1iii) alternating with the doubling of the clitic in the 2nd person (1ii). The two relevant LFs are provided in (28). (28a) contains a single variable, very much like (26), while (28b) is comparable to (27). In all cases the presence of a single negation operator returns a single negation (negative concord) reading. (28)â•… Viano a.â•… [⌐ [∃x [D i/ la [R n(x) [I drm b. [⌐â•… [∃x,yâ•… [D aâ•… [R n (x) [D t [R n (y) [I drm

Negative concord, though necessary to the interpretation of sentences like (27) or (28b), is not sufficient to derive it. There is another crucial component that enters into the reading for sentences like (27) or (28b), as opposed to the ordinary negative concord reading, which is that the two negations (whether they are both clitics or a clitic and an adverb) are understood as ‘doubling’ one another. In Manzini and Savoia (2007) we deal with doubling in connection with pronominal clitics. In particular, if doubling is not a morphophonological or computational (i.e. multiple spell-out) effect, then the fact that the differ­ ent instances of a doubled pronominal clitic express the same argument must depend on interpretation at the LF interface. In fact, the doubling interpret­ ation can be formalized through the notion of chain, which in representational models, in the sense of Brody (2003), is an LF primitive, and not a product of the derivation. Thus the theta-calculus at the LF interface will force all of the different instances of a doubled pronominal clitic (or of a clitic and its doubling full noun phrase) to be in a chain relation€– i.e. to fill the same argument slot. If the so-called negation is a nominal, argumental element, the same solution can be applied to the cases at hand. Specifically, the interactions of the sentential negation adverb with the internal argument of the verb reviewed in chapter 3 (the differing lexicaliza­ tion of the negation according to the person reference of the internal argu­ ment, the genitive of negation, and the ambiguity between the adverbial and

154â•… Sentential negation:€clitics argumental readings of ‘nothing’) motivated the assignment of the negation to the internal argument slot. There is a distinct tradition in linguistic studies which identifies the sentential negation with a negative quantification over the Davidsonian event argument of the sentence (Acquaviva 1994). However, we do not conceive of the event argument as an ordinary argument slot in the argument structure of the predicate (Higginbotham 1985), nor do we construe the so-called negation as a visible instantiation of the event argument. Rather, we have proposed that negative adverbs can introduce a variable restricted by the elementary subevent to which the external argument applies or by its internal argument, which is construed as the internal argument of the sentence as a whole. This line of thought is compatible with what we have said so far about the negative clitic, and is applicable to it. Pursuing this, the two negative polarity items in (27), i.e. the clitic and the adverb negation, or in (28b), i.e. the two clitic negations, introduce a variable restricted by the same argument€– i.e. the elementary event of the sentence, or its internal argument. This means that the two variables x and y in (27) and (28b) are effectively identified€– i.e. they enter into a chain(-like) relation. Summing up so far, the (rough) semantics that we have postulated for the so-called sentential negation delivers negative concord as a consequence, without any need for further assumptions. In turn, the latter is a prerequisite for the chain(-like) reading which underlies phenomena usually described in terms of doubling. In the present approach there is no copying (either morpho­ phonological or syntactic), but rather independent insertion of lexical items€– which are then identified by the argument calculus at the LF interface. The two approaches are equivalent in the interpretive component, as far as we can tell, since the chain relation, whether derived through movement or primitive, holds at the LF interface. The reason why the present analysis is to be preferred is the one we have tried to advance throughout this chapter€– in short, it remains to be shown that the complex interactions of doubling with interpretation-based notions such as that of person split etc. can be captured under alternative mod­ els. What is more, our argument is that when these interactions are dealt with at the level at which they belong, i.e. the LF interface, morpholexical-level facts can also be handled in a more revealing fashion€– specifically, without recourse to systematic homonymy. It is not any one fact that is compelling in our view, but rather the accumulation of apparently unrelated facts all pointing in the same direction. For the sake of completeness, we also consider briefly how the present approach can account not only for negative concord but also for its impossibility,

Negative concord and negative doublingâ•… 155 i.e. for cases where two n-words cannot combine under a negative concord reading. These cases are potentially problematic to the extent that, as argued in relation to (27)–(28), negative concord is the state of affairs normally pre­ dicted by the present theory. Specifically, negation clitics generally combine with other negative elements; however, negation adverbs are often in comple­ mentary distribution with negative arguments and adverbs, as has been noted in passing for Càrcare in (15ii). Very much the same conclusion can be drawn with respect to the 2nd person singular of Oviglio in (20), in which ‘never’ is in complementary distribution with the ‘nothing’-type sentential negation adverb present in all other persons. The first relevant observation is that this pattern is by no means necessary. For ease of exemplification, we stick to the ‘nothing’-type sentential negation and the combination of this negation with argumental ‘nothing’ and ‘nobody’. In (29) we report data from the Mezzenile variety, in which the ‘nothing’-type negation can always combine with a negative argument, whether this is ‘noth­ ing’ and adjacent to the ‘nothing’-negation, as in (29a), or ‘nobody’ and not adjacent to the ‘nothing’-negation, as in (29b). (29)â•… Mezzenile (Piedmont) a.â•… u fi (Jint) ï†−Ente he does not nothing ‘He doesn’t do anything’ b. u j nt Jint tSama JyN theyâ•… Locâ•… haveâ•… not called nobody ‘The haven’t called anybody’

In other varieties, more restrictive patterns are observed. Thus, there are lan­ guages in which the combination of the ‘nothing’-type negation with the nega­ tive argument ‘nobody’ is possible, as in (30a), though the combination with ‘nothing’, as in (30b), is not attested. This in turn holds when the two elements are, or would be, adjacent. Once adjacency is removed, all combinations are possible, as in (30a’–b’). (30)â•… Fontane (Piedmont) a. u i vEN Jent Jy it Loc comes not nobody ‘There doesn’t come anybody’ b. e mdZu Jent they eat nothing ‘They eat nothing’ a’.   a Jent vŋgy Jyŋ it Loc has not come nobody ‘There hasn’t come anybody’

156â•… Sentential negation:€clitics b’. j aN Jent madʒa Jent they have not eaten nothing ‘They haven’t eaten anything’

Other data point to a contrast based purely on adjacency. Thus, in several varieties the sentential negation adverb is in complementary distribution with a negative argument in simple tenses, but not in perfect tenses. Incidentally, because this pattern is relatively frequent we can document it easily both in languages which have only a sentential negation adverb, like S. Bartolomeo, and in languages, like Dego, which have a negation clitic as well. The rele­ vant contrast in S. Bartolomeo is between (31a–b) and (31a’–b’), and in Dego between (32a) and (32b). (32a’) from Dego provides a concrete illustration of the pattern not found in these languages.8 (31)â•… S. Bartolomeo Pesio a. i t:m JyN I call nobody ‘I don’t call anybody’ b. i mï†−dZ Jente I eat nothing ‘I don’t eat anything’ a’. i  JeN tSamQ Jy I have not called nobody ‘I haven’t called anybody’ b’. i  JeN maï†−dï†ıâ•… Jente I have not eaten nothing ‘I haven’t eaten anything’ (32) Dego a. i J maJduâ•… JEnte theyâ•… not eat nothing ‘They eat nothing’ a’. *i J maJdu nEnt JEnte they not eat not nothing ‘They eat nothing’ b. i n aJ nEnt maï†−d JEnte they not haveâ•… not eaten nothing ‘They have eaten nothing’

Let us consider, then, how the present theory can account for Dego (32a’). Recall that the doubling of a negative clitic by another negative clitic or adverb has been treated as akin to the doubling of a pronominal clitic by another clitic or a noun phrase. If so, the mutual exclusion of the sentential negation adverb and the negative polarity argument in Dego (32a’) is comparable to a mutual

Negative concord and negative doublingâ•… 157 exclusion between pronominal clitics€ – perhaps most famously the dative– accusative mutual exclusion that gives rise to the Spurious se phenomenon in Spanish. Therefore, we propose to account for Dego (32a’) and the like along the lines suggested by Manzini and Savoia (2007) for mutual exclusions between clitics. In particular, we propose that inserting a negative polarity argument associated with the internal argument, such as nte in Dego (32a’), is sufficient to satisfy all of the properties that would otherwise be lexicalized by the sentential negation adverb, effectively excluding its insertion. This pro­ posal is given theoretical content by our conclusion in chapter 3 that a senten­ tial negation introduces a variable restricted either by the internal argument of the sentence or by the elementary event that includes it. From this perspective, the negative polarity argument, i.e. nte, subsumes the contribution of the negation adverb to the interpretation of the sentence because it already intro­ duces a variable corresponding the internal argument of the sentence. As we noted, in languages like (31) and (32), the mutual exclusion holds only when the negation and the negative argument are within the same predica­ tive/eventive domain, for it is perfectly possible to have the same properties lexicalized twice in two different domains, e.g. by the clitic negation in the inflectional domain and by the negative polarity argument in the predicative domain in Dego (32a). Similarly, according to the conclusions of chapter 3, a negative adverb preceding the participle is inserted in a high C-type domain within the participial sentence, and is thus predicted to combine with a postparticipial negative argument inserted in the predicative/eventive domain, as in Dego (32b). We also expect that the absence of mutual exclusion between elements in different domains of the sentences (as opposed to those in the same domain) has a counterpart with pronominal clitics. In fact, mutual exclusion between subject and object clitics is obviated in interrogatives, where the sub­ ject clitic remains in the postverbal (enclitic) domain while object clitics are found in the preverbal domain (as proclitics). We further expect that, along with languages of the type in (31), there might be languages like (29) that have no mutual exclusion€– pretty much because mutual exclusion between clitics is a strictly language-specific phenomenon. More to the point, we predict that, given the account we have provided for it, mutual exclusion between sentential negations and negative arguments is constrained by lexical properties. Thus, it is expected that there should be lan­ guages like Fontane in (30) in which the mutual exclusion is only between the sentential negation adverb and the ‘nothing’ argument, while the ‘nobody’ argument (endowed with a human restriction) does not necessarily subsume the adverb.

158â•… Sentential negation:€clitics Delving further into the patterns of mutual exclusion is beyond the scope of this chapter, since it would require us to introduce a considerable amount of additional data. Before concluding, however, it is worth noting that if the dis­ tribution of sentential negation adverbs with respect to negative polarity argu­ ments is sensitive to their domains of insertion, as in (31) and (32), we expect a similar effect to be observable with negative clitics. Indeed, complementary distribution with negative arguments is found in many Italian varieties, includ­ ing the standard language, in the configuration in which a negative argument would precede the clitic, as in the case of a preverbal subject in (33a). This phenomenon is again parametrized; thus, the co-occurrence of the negation clitic with a negative polarity preverbal subject is attested in Old Italian texts, as noted by Meyer-Lübke (1899:€§695), who quotes the example in (33b). In (33a) we can apply essentially the same analysis proposed for Dego (32a’). We assume that the position of the preverbal subject in null subject and clitic subject languages is in the C field (cf. here chapter 1). Inserting the negative polarity subject in the C domain or higher subsumes all modal properties other­ wise lexicalized by the negation clitic. Therefore the clitic need not, and must not, be inserted.9 (33)â•… a.â•… b.

Nessunoâ•… (*non)â•… dorme nobody not sleeps ‘Nobody sleeps’ gente neuna non v’ arrivava (Novellino 55) people none not there arrived ‘Nobody arrived there’

It is evident that the line of reasoning deployed so far can in principle be extended to mutual incompatibilities between any two negative polarity items, including two arguments, two adverbs or an argument and an adverb. Thus, we can say that the lexicalization of one subsumes (relevant properties of) the lexicalization of the other (within a given domain) and hence excludes it (i.e. a re-lexicalization of its relevant properties). In any event, the general interpret­ ive mechanism of negative concord laid out in (26)–(28) remains in place.

5 The middle-passive voice:€evidence from Albanian

Fifty years ago, Chomsky (1957:€§5.4) argued that passive sentences should be excluded from phrase structure grammar and introduced instead by a transformational rule applying to active sentences. This was because introducing passives through rewriting rules would mean doubling the selectional restrictions independently imposed on actives, while a transformational rule would allow them to be stated only once. Chomsky (1965:€103–4) provides what has remained the standard conceptualization of this transformational process by proposing that the Manner Adverbial should have as one of its realizations a ‘dummy element’ signifying that the passive transformation must obligatorily apply. That is, we … may formulate the passive transformation … with an elementary transformation that substitutes the first NP for the dummy element passive and places the second NP in the position of the first NP.

In current practice, the by-phrase is independently generated by Merge; but the analysis whereby passive is defined by ‘substitution’ of an internal argument for the EPP position (second or internal Merge) remains at the core of generative transformational grammar. In this chapter we propose to evaluate this analysis in the light of data from Albanian, which provides two separate and complementary phenomena of interest. On the one hand, the passive (i.e. promotion of the internal argument to the EPP position with the external argument independently interpreted) has the same lexicalization as the reflexive, the anticausative and the impersonal. The question then is whether all of these different interpretations are associated with the same underlying syntax. On the other hand, there is no single lexicalization of this cluster of meanings; rather, it varies according to tense and aspect specifications. Thus, in standard Albanian the middle-passive voice is lexicalized by a specialized (agreement) inflection in the present or imperfective past, by the clitic u combined with the active forms of the verb in the perfective past, and by the periphrasis be–participle in the present perfect and in the pluperfect. The question in this respect is whether (in the passive and in 159

160â•… The middle-passive in Albanian the other interpretations) there is a single underlying syntax corresponding to all of these different morpholexical formats. The discussion in this chapter also provides a background for the discussion of the functional category ‘auxiliary’ in chapter 6, since the notion of voice (together with notions of transitivity and person split) crucially enters into the determination of auxiliary selection patterns. Correspondingly, after introducing the Albanian data in section 5.1, we devote section 5.2 to the clitic lexicalization of middle-passive voice and section 5.3 to the inflectional lexicalization; however, we leave the auxiliary–participle construction till chapter 6. In section 5.3.1 we briefly examine the possibility, also attested in Albanian, of lexicalizing the passive interpretation (in all tenses) with a be–adjectival participle construction. 5.1

Data

5.1.1 Middle-passive morphologies In the present indicative, Albanian has a specialized inflection for the nonactive voice, as exemplified in (1) for Gjirokastër (a Tosk variety, essentially like the standard) and for Shkodër (a Gheg variety). In (1a) we illustrate verbal bases ending in a vowel, while in (1b) we exemplify verbal bases ending in a consonant; the comparison with the active is provided in (1a’) and (1b’) respectively. The most complex instantiation of the middle-passive morphology can be seen in the vocalic bases of Shkodër, which are followed by the h affix, which is in turn followed by inflections for person. In the consonantal bases, as well as in a possible pronunciation of the vocalic bases in Gjirokastër, the affix preceding the person inflections is simply  (Trommer 2005). The comparison with the active present allows us to establish that the person inflections are themselves specialized for the middle-passive voice€– very clearly so in the singular, where the active and middle-passive forms bear no relation to one another. (1)â•… Gjirokastër a.â•… la- (h)- m/ S/â•… t/â•… mi/â•… ni/â•… n wash- MP- 1sgâ•… etc. ‘I washed myself’ etc. a’. la- i/ n/â•… n/â•… im@/â•… ni/â•… in@ wash- 1sg etc. ‘I wash (something)’â•… etc. b. viS- - m/ S/ t/â•… mi/â•… ni/â•… n@ dress- MP- 1sg etc. ‘I dress (myself)’ etc.

Dataâ•… 161 b’. vS vS vS vS- im vS- ni vS- in dress- 1sgâ•… etc. ‘I dress (somebody)’ etc. Shkodër a. l- - m/ S/ t/ na/ ni/ n wash- MP- 1sg etc. ‘I wash myself’ etc. a’. l- i/ n/ n/ im/ ni/ in wash- 1sg etc. ‘I wash (something)’ etc. b. veS- - m/ S/ t/ na/ ni/ n dress- MP- 1sg etc. ‘I dress (myself)’ etc. b’. veS- i ve:S ve:S veS- im veS- ni veS- in dress- 1sgâ•… etc. ‘I dress (somebody)’â•… etc.

In the simple past, Albanian resorts to a different morphosyntax for the formation of the middle-passive voice, preposing the clitic u to the verb, as illustrated in (2). The clitic can be taken to correspond roughly to Romance se; it is associated with all the different forms of the paradigm, as is also the case in some Romance varieties (in particular Romansh ones) for se. As for the morphology of the verb, no specialized middle-passive affix is present; furthermore, the person inflections are identical to those of the active, except for the 3rd person singular, whose active form is provided in (2a’) and (2b’). Even there, the middle-passive voice is characterized simply by the omission of the inflection present in the active paradigm, not by a different inflection. We interpret affixes like€–it in the vocalic paradigm of Gjirokastër or€–v in the vocalic paradigm of Shkodër as connected to the expression of the perfective past. (2)â•… Gjirokastër a.â•… u la- it- a/ / -/â•… @m/â•… @t/â•… @n MP wash- past- 1sg etc. ‘I washed myself’ etc.

162â•… The middle-passive in Albanian a’.  la- it- i it wash- past- 3sg ‘He washed it’ b. u vS- a/ /â•… -/â•… @m@/â•… @t/â•… @n@ MP dress- 1sg etc. ‘I dressed (myself)’ etc. b’.â•…  vS- i him dress- 3sg ‘He dressed him’ Shkodër a. u l- v- a u l- v-  u l: u l- m u l:- t u l- nE MPâ•… wash-â•… past-â•… 1sgâ•… etc. ‘I washed myself’ etc. a’. E la- u itâ•… wash-â•… 3sg ‘He washed it’ b. u veS- a u veS-  u vS u veS- m u veS- t u veS- n MPâ•… dress- 1sg etc. ‘I dressed (myself)’ etc. b’.  veS- i itâ•… dress-â•… 3sg ‘He dressed him’

The perfective past in (2) differs from the present in (1) both in temporal properties and in aspectual ones, under the natural assumption that the present is essentially an imperfective form. Therefore the lexicalization of the Â�middle-passive voice could in principle be sensitive to tense or to aspect. In Tosk varieties, the imperfective past follows the pattern of the present, with specialized middle-passive morphology, as in (3); thus, the split between present and imperfective past on the one hand and perfective past on the other appears to be based on aspect. The morphological analysis of the verb shows that, as in the present, the middle-passive voice is carried by the affix€–, which is followed by a€–S morpheme carrying the past specification; the latter is specialized for the middle-passive, as can be seen by comparing it with the active.

Dataâ•… 163 (3)â•… Gjirokastër a.â•… la- (h)-â•… S- a/ E/ -/â•… im/â•… it/â•… in wash-â•… MP- past- 1sgâ•… etc. ‘I washed myself’ etc. a’. E la- j- a j  n- tE n- im n- it n- in itâ•… wash- past- 1sgâ•… etc ‘I washed it’â•… etc. b. viS- - S- a/ E/ -/â•… im/â•… it/â•… in dress- MP- past- 1sg etc. ‘I dressed (myself)’ etc. b’. viS- j- a j-  t n- im n- it n- in dress-â•… past-â•… 1sgâ•… etc. ‘I dressed (somebody)’ etc.

On the other hand, in Gheg varieties, the specialized morphology and clitic realizations of the middle-passive voice split according to tense; thus, while the present has specialized middle-passive morphology, not only the perfective past but also the imperfective past in (4) have the u clitic. The data in (4) show that substituting an  accusative clitic for the u middle-passive voice clitic yields the active reading with no change in verb morphology. In other words, what the u clitic combines with is the ordinary active morphology of the verb. (4)â•… Shkodër a. u / l- ï†fi-â•… a ï†fi-  t ï†fi- im ï†fi- it ï†fi- in MPâ•… /itâ•… wash-â•… past. impf-â•… 1sgâ•… etc. ‘I.washed myself/ it’â•… etc. b. u/  veS- S- a ï†fi-  t ï†fi- im

164â•… The middle-passive in Albanian ï†fi- it ï†fi- in MPâ•… /himâ•… dress- past.impf-â•… 1sgâ•… etc. ‘I dressed (myself)/ it’â•… etc.

The middle-passive conjugation in Albanian also includes forms consisting of an auxiliary followed by the participle, as illustrated in (5) for the present perfect. In particular, the auxiliary jam ‘I am’ followed by the participle is sufficient to yield the middle-passive voice. The data for comparison in (5a’) and (5b’) show that the active is formed with the same participle but with the kam ‘I have’ auxiliary. Thus, in this case it is the switch from kam ‘I have’ to jam ‘I am’ that yields the switch from active to middle-passive voice. Similarly, the pluperfect is formed with the imperfective past of the two auxiliaries followed by the participle. As for the morphology of the participle, Tosk varieties (including the standard) have a participial ending –r, which is not present in Gheg varieties; it is worth noting that the vocalic bases in the Gjirokastër examples in (5a–a’) also include the perfective€–it morphology. (5)â•… Gjirokastër a.â•… @St la- it- ur he.isâ•… wash-â•… prf-â•… prt ‘He has washed himself’ a’.  ka la- it- ur itâ•… he.hasâ•… wash-â•… prf-â•… prt ‘He has washed it’ b. @St vS- ur he.isâ•… dress-â•… prt ‘He has dressed (himself)’ b’.  ka vS- ur him he.has dress- prt ‘He has washed him’ Shkodër a. ï†fi l:/ ve:S he.is washed/ dressed ‘He has washed/dressed (himself)’ b.  k l:/ ve:S him he.has washed/ dressed ‘He has washed/ dressed him’

5.1.2 The interpretation of the middle-passive morphologies In the preceding discussion, we have illustrated the three basic morphologies for middle-passive voice in Albanian. In each case we have chosen to illustrate the middle-passive voice with verbs which make the reflexive interpretation

Dataâ•… 165 particularly salient, and we have glossed our examples accordingly. In reality, each of the forms that we have exemplified is multiply ambiguous, allowing for a range of meanings that is independently attested, for instance, for the Romance counterpart of the u clitic, e.g. Italian si. In what follows, we review the various meanings, showing that they are associated with all morphological instantiations of the middle-passive voice. The reflexive reading prominent with a verb like ‘to wash’, for instance, implies a single participant in the event, which is both its theme (patient, etc.) and its causer (agent, etc.). This can easily be distinguished from another reading which equally involves a single participant in the event€– which we shall refer to as anticausative. This is the reading where the single participant is the theme (patient, etc.) and no external agency (cause, etc.) is expressed or implied in the event. This is evidently a salient meaning for the middle-passive predicate in (6), which we correspondingly glossed as ‘to wake up’. Of course, although ‘to wake oneself up’ is also a possible predicate, the reading is less salient for pragmatic reasons. What is important to note is that the anticausative reading, like the reflexive reading in the previous section, is associated with all lexicalizations of the middle-passive voice, namely the specialized inflection in the present (6a), the clitic in the perfective past (6c) and the jam–(perfect) participle formation in the present perfect (6d); the imperfective past (6b) has the specialized inflection or the clitic, depending on the variety. (6)â•… Gjirokastër a. zu- (h)-â•… t wake-â•… MP- 3sg ‘He wakes up’ b. zu- (h)- S wake-â•… MP-â•… past ‘He woke up’ c. u z- it MP wake- prf ‘He woke up’ d. ï†fit zu- a he.is wake- prt ‘He has woken up’ Shkodër a. tSo- h- t wake MP- 3sg ‘He wakes up’ b. u tSo- t MP wake- 3sg ‘He woke up’

166â•… The middle-passive in Albanian c. d.

u tSu: MP woke ‘He woke up’ ɐï†fit tSu: he.is woken ‘He has woken up’

The reading of the middle-passive voice that implies two participants in an event, including the theme (patient, etc.) and an external argument (agent, cause, etc.), is what is known as the passive. Again, the passive meaning is available independently of the particular morphology instantiating the middlepassive voice, as illustrated in (7). In these examples, it is really the by-phrase that distinguishes the passive from the other possible readings. Of course, the passive reading implies an agent, a so-called implicit argument, even when no by-phrase is lexicalized. Here we exemplify the present (7a), the perfective past (7c) and the present perfect (7d); the imperfective past reflects the morphology of the present (Tosk) or of the perfective past (Gheg). (7)â•… Gjirokastër a.â•… kt kmiï†fi la- (h)- n ga aj these shirts wash- MP- 3pl by him ‘These shirts are washed by him’ c. ata u z- it- @n ga t@ tir@t they MP wake- prf- 3pl by the others ‘They were woken by some people’ d. kt kmiï†fiâ•… jan la- it- u ga aj these shirts are wash- prf- prt by him ‘These shirts have been washed by him’ Shkodër a. fmia veS- - t pei nns the.childâ•… dress- MP- 3sg by the mother ‘The child is dressed by his mother’ a’. at tï†fi- h- n Q tiErt they wake- MP- 3pl by the others ‘They are woken up by some people’ d. jan tSu Q tiErt they.are woken by the others ‘They have been woken up by some people’ d’. ɐï†fit l: pei nɐns he.is washed by the mother ‘He has been washed by his mother’

It comes as no surprise that middle-passive voice morphology can attach to unergative verbs in Albanian, since these are construed by current theories

Dataâ•… 167 (Hale and Keyser 1993) as concealed transitives, where the verb effectively incorporates an object (cf. also the discussion in chapter 3). Therefore we may expect that the combination of the middle-passive morphology with such a verb will yield an impersonal meaning, essentially as a by-product of passivization, as in (8). Note, however, that in an impersonal passive we would expect it to be possible to lexicalize the external argument independently with a by-phrase. However, this does not seem to be possible in Albanian. (8)â•… Gjirokastër a.â•… ati fl- (h)- t mi there sleep- MP- 3sg well ‘There one sleeps well’ c. ati u fit mi there MPâ•… sleptâ•… well ‘There one slept well’ Shkodër a. atj fl- (h)- t mi there sleep- MP- 3sg well ‘There one sleeps well’ b. atjE u flE- tE mi there MP sleep 3sg well ‘There one slept well’ c. ktu u fje:t mi hereâ•… MPâ•… slept-3sgâ•… well ‘Here one slept well’ d. atjE ɐï†fit fie:t mi (*N ata) there it.isâ•… sleptâ•… well (by them) ‘There one has slept well’

More importantly, the middle-passive voice can also attach to unaccusative predicates€ – i.e. intransitive predicates once again, which cannot reasonably be construed as concealed transitives. Rather, the only argument of such predicates corresponds to their theme, e.g. the element that undergoes the change of location with motion verbs such as ‘to go’, exemplified in (9). With these verbs, therefore, the middle-passive voice cannot be analysed as yielding a sort of passive, albeit an impersonal one. Rather, it yields an impersonal tout court, which must then be entered among the possible interpretations of the middle-passive voice in Albanian; this is confirmed by the impossibility of associating such structures with a by-phrase. Once again, there is a parallelism with Italian si; however, it must be emphasized that the impersonal interpretation in Albanian is not restricted to the clitic morphology, but is equally found with specialized inflection or with jam–participle formations.

168â•… The middle-passive in Albanian (9)â•… Gjirokastër a. ga ati dil- - t from there exit- MP- 3sg ‘One exits from there’ a’. ai del he exits ‘He exits’ b. ga ati u dD from there MP exited ‘One exited from there’ b’. ai dD- i he exited- 3sg ‘He exited’ c. ga ati ɐï†fit dal mi from there it.is exited well ‘One has exited well from there’ c’. ka dal he.hasâ•… gone ‘He has exited’ Shkodër a. pej ktj dl- - t throughâ•… thereâ•… go.out-MP- 3sg ‘One goes out that way’ a’. ai dl heâ•… goes.out ‘He goes out’ b. pej ktj u dl- t through there MP go.out- 3sg ‘One went out that way’ b’. ai dl- t heâ•… go.out-3sg ‘He went out’ c. pej ktj u do:l through there MP went.out ‘One went out that way’ c’. ai dol- i heâ•… went.out- 3sg ‘He went out’ d. pej ktj ɐï†fit d:l mi through there it.is gone.out well ‘One has gone out well that way’ d’. kæ d:l he.has gone.out ‘He has gone out’

(* ata)

Dataâ•… 169 5.1.3 The Arbëresh varieties The Arbëresh varieties of Albanian, spoken in Southern Italy, belong to the Tosk group, and indeed reflect the conditions of the standard (or of Gjirokastër here) in many respects. In particular, with vocalic bases, these varieties lexicalize the middle-passive voice with the specialized verb inflection in the present indicative and in the imperfective past, as illustrated respectively in (10a) and (10b) from Portocannone, in which the middle-passive morphology is€–x; the perfective past has the clitic u, as in (10c). Microvariation is present at various points between the mainland varieties and Arbëresh, as well as within the Arbëresh fold. We note in particular that in Portocannone the perfective past, despite the presence of the u middle-passive clitic, maintains the middleÂ�passive affix€–x– of the present and imperfective past. As for the person inflections, it is not only the 3rd singular that distinguishes active and middle-passive in the perfective past, but also the 1st person singular. (10)â•… Portocannone (Molise) a.â•… la- xE- m/ S/ t/ mi/ ni/ n wash-â•… MP-â•… 1sgâ•… etc. ‘I wash myself’ b. la- x- S- a/ E/ i/ m/ t/ n wash- MP- impf- 1sg/ etc. ‘I washed myself’ c. u la- x- tS/ E/ -/ m/ t/ n MP wash- MP- 1sg etc. ‘I washed myself’ etc. c’. E la- v- a/ E/ i/ m/ t/ n it wash- past- 1sg etc. ‘I washed it’ etc.

A major point of variation between the middle-passive voice of Arbëresh varieties and that of mainland varieties concerns auxiliary–participle formations in the perfect. In Arbëresh varieties these involve the auxiliary kam ‘I have’, exactly as in the active, rather than jam ‘I am’; therefore middle-passive voice is lexicalized by the u clitic. The Portocannone variety displays an interesting further parameter concerning participial morphology. In the active voice in (11b) the participle has recognizably the same form as in the Gjirokastër example in (5), with the verbal base la- followed by the perfective morphology€–it and the participial ending –ur. By contrast, in the non-active voice in (11a), the participle is formed through suffixation of the middlepassive morpheme€–x, followed by the ordinary participial ending€–ur. Thus, in Portocannone and similar varieties the vocalic verb bases bear specialized

170â•… The middle-passive in Albanian morphology throughout the paradigm, even when a u clitic is present, as in the pluperfect in (11), but also in the perfective past in (10c). (11)â•… Portocannone a. at kiï†fin u la- x- they had MP wash- MP- ‘They had washed themselves’ b. at kiï†fin E la- it- they had it wash- prf- ‘They had washed it’

ur prt ur prt

With this morphological background, we are now in a position to consider the readings that are associated with the various forms. In the examples that we provided above, it is of course the reflexive reading that is salient. The possibility of what we have called the anticausative reading is evident in the examples in (12). In the case of auxiliary–participle formations, we provide a comparison of the middle-passive (12d) with the active (12d’)€ – which displays the difference between the two participial morphologies. A further point of variation between Portocannone and other varieties (both mainland and Arbëresh) emerges in the data in (12d)–(12d’), namely that the participle can be introduced by a coordinating/subordinating particle, literally ‘and’. This parameter is essentially irrelevant here (but cf. chapter 6). Another property which singles Portocannone out (and which is largely irrelevant for present purposes) is that the clitic is not positioned before the auxiliary, but immediately before the participle, even in the absence of the particle, as can be seen in (11). (12)â•… Portocannone a.â•… z- - Em wake- MP- 1sg ‘I wake up’ b. z- x- S- a wake- MP- past- 1sg ‘I woke up’ c. u z- tS MPâ•… wake- 1sg ‘I woke up’ d. ai kiS E u tSa- it had and MP break- ‘It had broken’ d’. ai kiS E E tSa- he had and it break- ‘He had broken it’

x- ur MP- prt it- ur prf- prt

Dataâ•… 171 Next, the impersonal meaning is available both with unergative predicates and with unaccusative ones, exemplified here in (13). As usual we provide the contrast between the middle-passive participle formation in (13d) and the active one in (13d’). (13)â•… Portocannone a. ktu vE-â•… xE- t te ha hereâ•… go- MP-â•… 3sgâ•… to the village ‘This way one goes to the village’ c. ktu u va- x te ha here MP go- MP to the village ‘This way one went to the village’ d. ktu kiS u va- x- ur te ha here it.had MP go- MP- prt to the village ‘This way one had gone to the village’ d’.â•… kiS va- t- ur he.hasâ•… go-â•… prf-â•… prt ‘He had gone’

As we fully expect, the range of morphologies that we have considered so far can be associated with a passive reading€ – i.e. a reading characterized, like the transitive one, by the presence of two roles and two event participants, except that of course the theme is found in the EPP position. Relevant examples are provided in (14). It should be noted that, while in mainland Albanian by-phrases normally co-occur with the middle-passive morphology, as illustrated in (7), the MP-passives of Arbëresh are normally impersonal, in the sense that they only allow for an impersonal (i.e. generic) reading of the agent€– that is, they do not normally combine with the by-phrase. (14)â•… Portocannone a. atiE la- xE- n kmiS-t here wash- MP- 3pl shirt-the.pl ‘Here shirts are washed’ c. atiE u la- x- n kmiS-t here MP wash- MP- 3pl shirt-the.pl ‘Here the shirts were washed’ d. atiE kiï†fin u la- x- ur kmiS-t here had MP wash- MP- prt shirt-the.pl ‘Here the shirts had been washed’ d’.â•… kiï†fin i la- it- ur they.hadâ•… themâ•… wash-â•… prf-â•… prt ‘They had washed them’

172â•… The middle-passive in Albanian 5.2

The u clitic

We shall begin our discussion with the structures formed with u, which we have described throughout as comparable to Romance si/se. Just as the distributional properties of Italian si require it to be treated as a pronominal object clitic (as opposed to an affix, a subject clitic, etc.), we can argue that the same holds for Albanian u. Consider, for instance, enclisis–proclisis alternations in Arbëresh varieties. These show that u is sensitive to exactly the same conditions as other object clitics, such as accusative E ‘him/her’ or dative i ‘to him/to her/to them’, down to very fine variation. Thus, we have seen that in a variety like Portocannone, the auxiliary precedes both, as in (11). Another variety where the same holds is S.Benedetto, as in (15), which shows that there is no correlation with the possibility of the participle being introduced by E ‘and’. By contrast, in a variety like Civita, as in (15), the u clitic and the E accusative clitic precede the auxiliary. We refer the reader to Manzini and Savoia (2007) for an analysis of the relevant parameter(s); what is relevant here is the complete parallelism between the two clitics. (15)â•… S.Benedetto Ullano (Calabria) a. kia  pa I.had him seen ‘I had seen him’ b. kia u aitu I.had MP washed ‘I had washed myself’ Civita (Calabria) a. u kiï†fi zua MP I.had woken ‘I had woken up’ etc. b.â•…  ki pa: him I.had seen ‘I had seen him’

Consider also the imperative 2nd person singular. As exemplified in (16) with the Arbëresh variety of Civita, in the positive forms pronominal clitics are found in enclisis (as a reflex of the high position of the verb); this is true both of accusatives, as in (16a), and of the u clitic, as in (16a’). By contrast, the presence of the negation induces proclisis (as a reflex of the verb staying in its inflectional position), and this affects the accusative and u clitics alike, as in (16b–b’). (16)â•… Civita a. zɟj  wake.upâ•… him ‘Wake him up’

The u cliticâ•… 173 a’. b. b’.â•…

zɟj u wake.up MP ‘Wake up’ ms  zɟ not him wake.up ‘Don’t wake him up’ ms u zɟ not MP wake.up ‘Don’t wake up’

Given this distribution, it is evident that any adequate theory of Albanian u must take into account the fact that it is an object clitic. Within the object clitic string itself, the u clitic appears to be found in a relatively low position. In particular, as shown in (17), u follows the 3rd person dative, as well as the 1st person clitic. Incidentally, these are examples of what we have called the anticausative interpretation; the dative adds a benefactive/malefactive specification. (17)â•…

Gjirokastër m/ i u i gta to.me/to.him MP broke the.glass ‘The glass broke on me/him’



Shkodër m i/ u y: gota to.me/to.him MP broke the.glass ‘The glass broke on me/him’



Portocannone m/ i u tSa- x  bukjer to.me/to.him MP break- MP a glass ‘A glass broke on me/him’

The object clitic nature of u corresponds to a rather natural treatment of at least the reflexive interpretation. Thus, we could say that, exactly like accusative E, reflexive u is a lexicalization of the internal argument of the verb€– with the difference that while E is pronominal, u is anaphorically dependent on the EPP argument. This is the theory proposed by Burzio (1986) for Italian reflexive si. However, by analysing reflexive si as just described, Burzio (1986) is forced to postulate the existence of at least one other homophonous si€– i.e. an impersonal si, which lexicalizes the external argument of the verb as a generic. Its effect is that the EPP position is vacated and the internal argument can and must move into it, yielding the classical movement derivation for middle-passives. The problem, of course, is that if there were two different si’s, one would expect to be able to tell them apart through their syntactic behaviour (and not just their interpretation). In reality, Manzini and Savoia (2005, 2007) show that

174â•… The middle-passive in Albanian Italian si behaves homogeneously under distributional tests, even when object and subject clitics otherwise show a split. What is more, the postulation of two si’s forces other principles to have a disjunctive formulation, as is notably the case for auxiliary selection. As if this were not enough, data like those from Albanian show that the cluster of meanings associated with si forms a natural class, showing up in many diverse languages€– and with many diverse morpholexical instantiations. The main alternative analysis available in the literature involves unifying the various si’s under the movement derivation classically associated with passives. Thus, let us assume that in the passive, si becomes associated with the external theta-role of the predicate, and this forces the internal argument to externalize, yielding the typical promotion of object to subject. Because a reflexive predicate is by definition symmetric, reflexive si could in principle correspond to the internal or to the external argument of the verb. Suppose that, just like passive si, reflexive si is associated with the external argument of the predicate; the derivation that ensues is then identical to that of the passive, with promotion of the object to subject position providing for a unification of the two si’s (Marantz 1984). Needless to say, this derivation not only unifies various interpretations of si, but also does so by extending the classical movement approach to all of them. Despite what may appear to be its theoretical advantages, this analysis fails for empirical reasons. In particular, we note that authors who have espoused the unaccusative theory of middle voice have systematically considered languages like French, which does not have the impersonal (non-passive) reading of the middle voice. For a language like Italian, a theory associating si with movement from the object to the subject position is directly contradicted by the existence of examples like (23a) below in which si co-occurs with an overtly lexicalized accusative. Another important disadvantage of this analysis is that it does not predict that the morphosyntax of si is consistently that of an object clitic. That this is not an idiosyncratic property of si can be seen precisely in the context of cross-linguistic comparison, for instance with Albanian. There is no doubt that the properties of u are in some respects quite different from those of Italian si, in that, for instance, u is associated with all persons as opposed to si, which is only associated with 3rd person (though there is great variability in Romance languages, and in Romansh the si-like form can be associated with all persons). Another difference is that si is associated with all temporal and aspectual specifications of the verb, while, as we have seen, u is restricted to the perfective (standard Albanian) or to the past (Gheg varieties). Precisely because of

The u cliticâ•… 175 this variation, it is all the more striking that what remains constant in the morphosyntax of u and si is that they behave like object clitics; evidently this is a central property of such forms and not merely an accidental one. But although their distribution suggests that elements like si or u are just the middle-passive counterpart of the accusative clitic, this fact cannot be captured by the treatment of middle-passive morphology in terms of movement. Note also that si or u cannot be treated as subject clitics in languages like Italian or Albanian which do not otherwise have such elements€– nor can they be subject clitics in imperatives, which consistently lack such elements even in subject clitic languages. Reinhart and Siloni (2005), basing their work on Chierchia (2004) and Reinhart (1997), introduce as many separate operations on argument structure as there are basic meanings of si€– namely reflexive bundling, which bundles the external theta-role with some other theta-role; saturation/arbitrarization (responsible for passives/impersonals), which saturates the external theta-role through existential closure; and decausativization (responsible for anticausatives), which reduces (i.e. suppresses) an external [+cause] theta-role. The unification of these various rules by a single morphology is imputed to Case theory. The assumption is that the arity reduction operations just mentioned do not affect the Case properties of the verb, leaving an accusative (or a nominative in arbitrarization contexts) potentially unchecked; ‘the clitic (or its equivalent) reduces Case’ (Reinhart and Siloni 2005:€402). As far as we can see, Chierchia (2004), and consequently also Reinhart (1997) and Reinhart and Siloni (2005), are oblivious to the existence of impersonal unaccusatives. The latter exclude the possibility that arbitrarization can simply be construed as reduction of the external theta-role, for there is no external theta-role to be reduced in unaccusatives. Furthermore, Reinhart and Siloni (2005) argue that si is not an ‘object clitic’, but their evidence really shows that si is not an accusative clitic (for instance not triggering faire–à constructions when embedded under a causative). On the contrary, Italian si and its Albanian u counterpart behave like object clitics with respect to their distribution€– contrasting with subject and other clitics. In other words, any theory that does not treat si (or u) as an object clitic is forced to state all of the relevant generalizations (distributional, etc.) twice, once for object clitics and once for si/u. Here, the point is that if si/u is a bona fide object clitic, we expect it to be like any other object clitic (with which it patterns) in having not only Case properties, but also denotational and argumental ones. In this respect, Reinhart and Siloni (2005) represent a step backwards even with respect to Chierchia (1995), who identifies si with the variable existentially bound in passives/impersonals.

176â•… The middle-passive in Albanian A clue that Reinhart and Siloni’s (2005) theory is on the wrong track regarding the status of si is that it yields incorrect predictions as to the connection between si and auxiliary selection. In order to explain the parameter differentiating Italian, in which the middle voice is associated with be, and Dutch, in which it is associated with have, they invoke a distinction between the thematic and structural components of Case. In their account, Dutch zich, ‘though referentially defective, occupies the complement position’ where it can check structural Case, while Italian si cannot, leaving ‘the structural accusative residue to be checked … The auxiliary “be” is used whenever there is such an accusative residue’ (Reinhart and Siloni 2005:€432–3). Leaving aside any other considerations, languages like Soazza, reviewed in chapter 6, show that the clitic or non-clitic status of the middle morphology is completely irrelevant for the auxiliary selection parameter€– thus eliminating a potential argument for treating si as anything but a bona fide clitic counterpart of zich. As stressed by Culicover and Jackendoff (2005), current generative theorizing is strongly biased in favour of what they call Interface Uniformity, i.e. the principle that ‘the syntax–semantics interface is maximally simple, in that meaning maps transparently into syntactic structure; and it is maximally uniform, so that the same meaning always maps onto the same syntactic structure’. From such a perspective, the objections just raised may be considered of little import when weighed against the possibility of maintaining a ‘uniform’ movement analysis for passive. The argument pursued here is that loss of predictive power with respect to the actually observed morpholexical forms is to be taken as seriously as loss of predictive power at the LF interface. Hence the difficulty in predicting the object clitic behaviour of si or u cannot be discounted even in the face of apparent gains in ‘Interface Uniformity’.1 We take the conclusion that si is a bona fide pronominal clitic as our starting point, assuming that as such it fills an argument slot of the predicate (as does its Albanian counterpart u, etc.). Another premise of our analysis is that the denotation of si/u differs from that of other pronominal clitics in that it is an indefinite, i.e. a free variable. This is proposed by Chierchia (1995) for impersonal si and by Manzini (1986) for all si; the various readings of si simply depend on the possible ways of closing the variable. Consider the reflexive interpretation. In the absence of evidence to the contrary, we can assume that the structure of the reflexive sentence matches that of its transitive counterpart; in other words, u in the reflexive sentence occupies an object clitic position, as E does in the active sentence; in (18) it is categorized as Q (as si is) in accordance with its variable interpretation. In turn, the EPP argument, i.e. D in present terms, is lexicalized in (18) by the verb inflection€–in,

The u cliticâ•… 177 as is normally the case in null-subject languages like Albanian. Crucially, the referential properties of u, which by hypothesis are those of a variable, set it apart from other pronominal clitics. In order for the variable to be valued, it must be bound by a referential element. The reflexive interpretation is simply the consequence of the binding of the u variable by the closest available referring element, i.e. the EPP argument. The construal of reflexivization that we have now provided is essentially the traditional one, with the reflexive element (u in this case) associated with an internal argument position and bound by the EPP argument. (18)â•… Shkodër Q u

I I l� ʃ (x,y)

D in

Consider now the passive interpretation, which could equally be associated with the sentence in (18). Our proposal is that the structure of the passive sentence is identical to that of the reflexive sentence, with the u clitic inserted in the internal argument position. Indeed there is no evidence that the reflexive and the passive readings correspond to different underlying structures. Rather, all morphosyntactic evidence points to the conclusion that structures like (18) are genuinely ambiguous, allowing for both readings under consideration. If the same structure underlies both the reflexive and the passive readings, then passives must differ from reflexives only interpretively. Let us assume that in the passive, the dependency between the u variable and the EPP argument corresponds to a chain. If so, the passive is treated exactly as in classical generative grammar, as an instance of chain formation between the internal argument and the EPP argument. The only difference is that instead of a trace (i.e. an empty category or a copy), the analysis we propose has an overtly lexicalized internal argument, whose semantics is that of a variable. There are several respects in which this analysis is at variance with standard generative frameworks, including notably the minimalist programme of Chomsky (1995). While these take a derivational view of movement, the present analysis is representational. Thus, since the two positions related by the chain in (18) under the passive reading are each independently lexicalized, there cannot be a derivational process of movement relating them; rather, movement must reduce to the notion of chain at the LF interface (Brody 2003).

178â•… The middle-passive in Albanian Chomsky (1995) assumes, in a manner consistent with the overall derivational outlook of minimalist theory, that the verb inflection is a cluster of uninterpretable features which drive processes such as movement (agreement, etc.) because of the need for such features to be checked (valued, deleted, etc.). By contrast, our discussion of the structure in (18) presupposes a treatment of the verb inflection as an interpretable element€– specifically as the morphologicallevel lexicalization of the EPP argument of the sentence. Finally, under (minimalist) movement, chains can be motivated by feature-checking requirements or by interpretive requirements. In present terms, they can be motivated only by interpretive requirements, i.e. in the case of the passive chain, by the need to provide a value for the variable internal argument. The representational construal illustrated here for movement affects all transformational processes. Thus, agreement, which is the rule specifically responsible for feature checking in Chomsky’s (1995) framework, is taken to be a relation which must hold if various interpretations (including the chain one) are to hold in turn. For instance, the chain in (18) requires agreement (or, to be more precise, compatibility in referential properties) between the EPP argument and the element lexicalizing the variable, though in this case the requirement is trivially met because of the presence of a variable. In short, we part ways with standard models of generative transformational grammar in adopting a representational model with the properties outlined in the preceding discussion. At the same time, it should be clear that in this representational form, our analysis of Italian si or Albanian u includes standard generative ideas about passive as involving a chain between the internal argument and the EPP argument. Thus, while we share the concern of Culicover and Jackendoff (2005) about Interface Uniformity, we certainly do not subscribe to their view of chain interpretations as mediated by a Grammatical Functions (GF) level of representation. The analysis we propose is no more (and no less) than a representational version of generative transformational models (specifically of minimalist ones). We contend that, at the LF interface, the present theory maintains all of the explanatory power of conventional theories of movement. At the same time, crucial motivation for it comes precisely from the kind of considerations pertaining to morpholexical structure that we have detailed above. In the present theory, it is perfectly possible to maintain that elements such as Italian si or Albanian u are exactly what they appear to be, i.e. normal object clitics. In this respect, the crucial assumption is simply that their content is that of a variable€– in other words, that the variable status is not restricted to traces (i.e. empty categories or copies) created by movement. This latter restriction seems

The u cliticâ•… 179 to us to be an artifact of strictly derivational approaches, while representational approaches can easily handle specialized lexical items with the content of variables. In short, we can maintain what appears to be the transitive structure of sentences like (18), with the clitic instantiating an object, while at the same time incorporating the core generative insight that the passive interpretation involves the chain construal of the internal argument with the EPP argument. Consider now what we have called the anticausative reading, salient in examples like (6). This reading can be obtained on the basis of structures like (19), entirely parallel to (18) above, through the formation of a chain between the variable internal argument, lexicalized by u, and the EPP argument, represented by the verb inflection. This leads to the interpretation under which the EPP argument is interpreted as the internal argument of the verb (roughly the theme undergoing the waking up). Note that the argumental frame of the verb is in itself transitive; quite simply, in the anticausative reading the external argument is not interpreted. Needless to say, the passive interpretation, roughly ‘he was woken up’, is equally predicted to be possible in (19), on the basis of chain formation. Thus, exactly as in standard movement models, passives and anticausatives share the same core syntax. The difference is that in the passive reading, the implication is preserved that the event takes place through an external agency or cause, corresponding to the external argument of the transitive argument frame of the verb. The latter can receive independent lexicalization through a by-phrase, or it can be interpreted through generic binding of the argument variable€– yielding a so-called ‘implicit argument’, while in the anticausative reading the implication is that the theme is not acted on by another agent/cause. (19)â•… Shkodër Q u

I I tʃo (x,y)

D tε

Going back now to the reflexive interpretation, the maximally simple assumption about the nature of the dependency between the u variable and the EPP argument is, again, that it is a chain. This assimilation of the reflexive to the passive/anticausative does not prevent their respective meanings from being clearly differentiated. As in the anticausative interpretation, in the reflexive there is no implication of an external agency or cause, differentiating both from

180â•… The middle-passive in Albanian the passive. At the same time, the reflexive and the anticausative are differentiated in that, in the reflexive, some degree of intentionality is associated with the argument of the si sentence; thus, reflexive readings are available only with EPP arguments capable of a mental state. In other words, pragmatic knowledge about the event of ‘waking up’ excludes the reflexive reading in (19). The preceding discussion implies that we agree with Chierchia (2004) and Reinhart and Siloni (2005) that the external argument configurations in the middle are responsible for differences between anticausative, passive and reflexive readings. Yet it seems to us that they correspond to all and only the logically possible such configurations€– and as such they do not need to be stated, but follow as a matter of logical necessity. Thus, in the anticausative the external theta-role is not interpreted. In the passive, it is interpreted through quantificational closure (passive); alternatively, it is interpreted through assignment to an adjunct, i.e. the so-called by-phrase. As for the reflexive, it seems to us that the arguments put forth by Reinhart and Siloni (2005) are not decisive as to whether the external argument is to be imputed to the si chain (reflexive). For instance, ne cliticization from postverbal subjects is not a reliable test of the internal or external argument status of the latter, depending instead on the presentational (focus) properties of the sentence (Belletti and Rizzi 1981; Belletti 1988; Saccon 1992, pace Burzio 1986). Thus, under the right conditions we accept ne-extraction from the inverted subjects of reflexives (varying pragmatic contextualizations are in fact a better explanation for the variability in grammaticality judgements noted by Reinhart and Siloni (2005) among Italian speakers). Therefore we side instead with Chierchia (2004) in assuming that reflexives are a subclass of anticausatives, in which agency is imputed to the sole lexicalized argument€– by a meaning postulate in Chierchia (2004). Note now that there are also verbs which display the middle-passive conjugation without having a transitive counterpart; an example is ulm ‘I sit down’ in (20a–c). One may legitimately wonder how these differ from active unaccusatives, i.e. predicates whose only argument is a theme (an internal argument), and which are not formed through middle-passive morphology. An example is provided by a motion verb like dal ‘I go out’, exemplified in (9) and here in (20a’–c’), which combines with the active person ending and in the present perfect with the auxiliary kam ‘I have’. (20)â•… Shkodër a.â•… ul- E- t sit-â•… MP- 3sg ‘He sits down’

The u cliticâ•… 181 b. c. a’. b’. c’.

ai u ul he MP sit-3sg ‘He sat down’ ɐï†fit u:l he.is sat ‘He has sat down’ ai dl he goes.out ‘He goes out’ ai dol- i he went.out- 3sg ‘He went out’ kQ dQ:l he.hasâ•… gone.out ‘He has gone out’

For verbs like dal, we simply assume that their single argument slot (a theme), as in (21a), is assigned to the obligatory argument of the sentence, i.e. the EPP argument. This yields an unaccusative reading comparable to the anticausative reading in (19)€– but does not imply the presence of non-active morphology. At the same time, the grammar must provide a way to distinguish between verbs like dal ‘I go out’ and verbs like ulEm ‘I sit down’. A way to formalize this distinction is simply to associate the latter with an argument frame of the same type found on tSo- in (19), as shown in (21b). The fact that, in (21b), the potentially transitive frame is restricted to the anticausative reading will have to be learned as a lexical property.2 (21) a. dEl (x) b. ulEt (x,y)

The final reading of Albanian u sentences that remains to be considered is the impersonal one, most clearly present in sentences involving unaccusative predicates such as (9). Under the line of explanation pursued throughout this section, one may be led to conclude that, in the absence of distributional or morphological evidence to the contrary, the structure underlying (9) is the same as that already indicated for the other interpretations of u in (18)–(19), as in (22). The crucial difference is that (18)–(19) contain not only the variable u clitic, but also some independently referring EPP argument€– even if only represented by the inflection of the verb. By contrast, (22) contains no independently referring EPP argument. Indeed, the obvious construal of the generic (or ‘impersonal’) interpretation associated with the EPP argument in (22) is that the u variable itself supplies it, through closure by a generic operator. This interpretation, in turn, can correspond to a syntax in which, just as in the

182â•… The middle-passive in Albanian other cases considered before, the u clitic forms a chain with the EPP argument represented by the D inflection of the verb. We assume that a generic interpretation cannot simply be associated with the 3rd singular inflection of the verb; this necessitates the introduction of the variable, i.e. u, which can be bound by the generic operator, as detailed above. (22)â•… Shkodër Q u

I I dεl (x)

D tε

Recall, as we noted above, that the arbitrarization rule of Chierchia (2004) and Reinhart and Siloni (2005) (i.e. existential closure of an external argument) could not yield impersonals with unaccusative verbs. If our discussion is on the right track, this problem can be avoided simply by taking arbitrarization to correspond not to an operation on argument structure, which needs to be stated in terms of such primitives as external argument, but to an interpretation at the LF interface, which can apply to any variable not independently closed. We also noted above that the parallel between Italian si and Albanian u is all the more interesting because the two elements also display important points of variation. A relevant observation in this respect concerns the fact that impersonal si in Italian can combine with accusative objects, as in (23a). This possibility is not available in Albanian€– and in fact it is excluded in many Italian varieties as well, where the only possible combination between si and a transitive predicate is the counterpart of standard Italian (23b), i.e. a passive. (23) a. b.

Li si chiamerebbe volentieri them MP would.call gladly ‘One would gladly call them’ Si chiamerebbero volentieri MP would.call gladly ‘They would gladly be called’

In present terms, the crucial difference between the impersonal in (23a) and the passive in (23b) is that in (23b), the si clitic is construed as an instantiation of the internal argument slot, bound by the independently lexicalized EPP argument. On the other hand, in (23a), si itself satisfies the EPP position, while the internal argument is independently lexicalized by the accusative

The u cliticâ•… 183 clitic. The impossibility of the Albanian counterpart of (23a) can then simply be described in terms of a necessary association of Albanian u with the internal argument slot. On the assumption that the only argument slot of unaccusatives is an internal argument slot (albeit assigned to the EPP position), we still derive the impersonal reading of u with unaccusatives. In conclusion, we have now provided an answer to one of the key questions raised at the beginning€– namely whether a single syntax underlies all of the different meanings of the middle-passive voice. At least in the case of si/u sentences, what gives rise to all of the different readings is the property of si/u that it is a variable. As a result of the presence of this variable, one of the argumental positions of the predicate remains unassociated (in anticausatives/reflexives) or is interpreted only in that it is existentially closed (passives, impersonal) and/or associated with an argument which is external to the structure of the predicate (the by-phrase). This is not the characterization associated with (middle-)passive by standard generative theory€– according to which (middle-)passive is reduced to movement from object to subject position. However, the construal of the notion that we suggest is in a way equally traditional€– having to do with the non-closure (or generic closure) of the argument structure of the verb. Importantly, the grounds for this switch are entirely empirical, in particular that the standard generative construal in terms of object to subject promotion stands no chance of unifying the passive/ reflexive reading with the impersonal one.3 Needless to say, (the representational counterpart to) object to subject movement still characterizes the bound readings of the variable when construed as an internal argument (in passives, anticausatives and reflexives). This latter fact also provides us with a key for understanding languages, like those routinely considered in the literature, in which the impersonal reading illustrated here in Italian and Albanian is not available (except perhaps in the shape of impersonal passives, with unergatives). A language like French, in which the impersonal has the specialized subject clitic lexicalization on (‘one’), provides a good minimal contrast with Italian.4 A good minimal contrast with Albanian is provided by Greek (Roussou 2008; Manzini, Roussou and Savoia forthcoming), in which there are specialized middle-passive inflections which split according to aspect as in Albanian, but these do not yield the impersonal reading. In terms of the present approach, the difference between the se of French and the si of Italian can be straightforwardly described by assuming that French se is a bound variable which does not admit of simple quantificational closure. We shall return to Greek and Albanian in the next section.

184â•… The middle-passive in Albanian 5.3

Specialized inflections

On the basis of the conclusions reached in section 5.2, we are in a position to consider the second key question raised at the beginning, namely whether there is a single underlying syntax corresponding to all of the different morpholexical instantiations of the middle-passive voice€– or whether instead what we are dealing with is merely the interpretive equivalence of (slightly) different syntaxes. Consider, then, the lexicalization of middle-passive voice in Albanian with specialized verb inflections, as in (1) and (3). In Gjirokastër and in the vocalic bases of Shkodёr, this non-active morphology can be seen to include an invariable affix –hE. With the consonantal bases of Shkodёr, we can take the€ –E extension of the base to represent the middle-passive morphology (Trommer 2005); the same will hold for the reduced form with€–E in Gjirokastër. As for the person inflections, these differentiate the active and the middle-passive in the present, and more specifically in the singular; in the plural it is only the 1st person that appears to be sensitive to voice. In the imperfective past of Gjirokastër in (3), there is substantial identity of the person endings in the active and middle-passive voices, the only difference being that the 3rd person singular is not lexicalized in the middle-passive. An insight into the nature of the middle-passive€ –h infix is provided by the fact that in an Arbëresh variety like Portocannone, illustrated in section 5.1.3 above, –hE (or –x in Portocannone) systematically co-occurs with the u clitic in all forms of the verb in which the latter is present, hence not only in the perfective past (10c), but also in the perfect, which in Arbëresh varieties is formed with kam ‘I have’ and the u clitic, as in (11a). One possible conclusion suggested by this pattern is that it represents a case of doubling, whereby the u clitic and the€–hE infix lexicalize essentially the same properties, at the syntactic level and at the morphological level respectively. If so, given the characterization provided for the u clitic in section 5.2, we are led to suppose that the –hE morphology represents a Q-type variable attaching to the verbal base I, as illustrated in (24) for the present indicative. The specialized€–t person ending in turn lexicalizes the D argument of this verbal constituent. (24)â•… Shkodёr D t I l�(x,y)

Q hε

Specialized inflectionsâ•… 185 Given the structure in (24), the computation of the different meanings associated with it proceeds in the same way as detailed in the previous section for u clitic structures. If the€–(h) variable, filling the internal argument slot of the predicate, is bound by the EPP argument, represented in (24), by the€–t inflection, the passive, reflexive and anticausative readings are obtained, depending on the interpretation of the external argument. If this is generically closed, we obtain the passive interpretation, while the anticausative/reflexive interpretation depends on the external argument remaining unlinked. In the case of a predicate like ‘to wash’ in (24), the salient meaning is reflexive rather than anticausative because agency/intentionality is normally attributed to the only lexicalized argument. Generic closure of the u variable leads in turn to the impersonal interpretation, seen in connection with unaccusative predicates like dl in (25). Overall, the effect of the€–(h) variable, just as with the u variable, is to associate the verb with an unsaturated or generically closed argument position€– namely the external argument with transitives and the only argument with unaccusatives. (25)â•… Shkodёr D t I dε l (x)

Q ε

As already noted, in the Arbёresh variety of Portocannone the€–x morphology combines with the u clitic to yield an instance of doubling, illustrated in (26) for the past perfective. The distribution of the u clitic in Portocannone, as in Arbёresh varieties in general, is determined by the perfect/imperfect split, so that u co-occurs with the perfective forms of the verb, including auxiliary–Â�participle constructions, formed with the kam ‘I have’ auxiliary. By contrast, the€–x morphology enters into the formation of the entire middle-passive paradigm, including the existence, noted in section 5.1.3, of a specialized middle-reflexive participial form. These separate distributions for€–x and u, when combined, yield the correct distribution of the doubling configuration (as their intersection). (26)

Portocannone Q u D tʃ I la(x,y)

Q x

186â•… The middle-passive in Albanian Assuming that the discussion of the€–h morphology is on the right track, it still remains for us to clarify the nature of the specialized agreement endings that characterize middle-passive inflectional formations. In particular, the singular –m/–ʃ/–t inflections of the present bear no relation to their active counterparts. In her discussion of Greek middle-passive voice, Roussou (2008) notices that in imperfective tenses, the middle-passive voice of Modern Greek is entirely realized by a series of specialized agreement inflections, which can be analysed into a thematic vowel –e/–o and specialized person endings –me, –se, –te, etc. for the present in (27a) or –mun, –sun, etc. for the past in (27b) (Philippaki-Warburton 1973; Ralli 1988). (27)â•… a. a’. b. b’.â•…

plen- ome/â•… ese/â•… ete/ omaste/ osaste/ onde Wash-â•… MP.1sgâ•… etc. ‘I wash (myself)’ etc. plen- o / is/ i/ ume/ ete/ un wash-â•… 1sgâ•… etc. ‘I wash (something)’â•… etc. plen- omun/â•… osun/ otan/ omastan/ osastan/ ondan wash-â•… MP.1sgâ•… etc. ‘I was washing (myself)’ etc. eplen- a/ es/â•… e/â•… (plen-) ame/â•… (plen-) ate/â•… (eplen-) an wash-â•… 1sgâ•… etc. ‘I was washing (something)’ etc.

The middle-passive perfective tenses are formed instead by the agreement inflections of the active voice combined with the affix –th for the present tense,5 as in (28a) and the affix –th–ik for the past tense, as in (28b); in this second case –th can be taken to realize middle-passive voice again, and –ik can be taken to realize past tense (Philippaki-Warburton 1973). (28)â•… a.â•… a’. b. b’.

pli -th o/ is/ i/ ume/ ite/ un wash-â•… MP-â•… 1sgâ•… etc. ‘I wash (myself)’ etc. plin o/ is/ i/ ume/ ete/ un wash- 1sg etc. ‘I wash (something)’ etc. pli- th- ik- a/ es/ e/ ame/ ate/ an wash MP past 1sg etc. ‘I washed (myself)’ etc. eplin- a/ es/ e/ (plin-) ame/ (plin)-ate/ (eplin-) an wash 1sg etc. ‘I washed (something)’ etc.

Roussou (2008) notices that the split in lexicalization between the imperfective and the perfective has some of the same properties in Greek and in

Specialized inflectionsâ•… 187 Albanian. In particular, in both Greek and Albanian, the imperfective has specialized middle-passive agreement inflections, while the perfective has the ordinary active agreement inflections, which combine with the –th affix in Greek and with the u clitic in Albanian. On the basis of this parallelism, Roussou (2008) suggests that Greek perfective tenses involve a structure along the lines of the Albanian u formations, where the –th affix plays the same role at the morphological level as the clitic plays at the syntactic level€– as a variable lexicalization of the internal argument. Crucially, Roussou (2008) goes on to propose that while the active agreement inflections associate with the EPP argument, the middle-passive agreement inflections associate with the internal argument. In other words, while active agreement is nominative, middle-passive agreement is absolutive. Extending this analysis to Albanian, we would conclude that the difference between the active agreement inflection (in the perfective tenses) and the middle-passive agreement inflections (in the imperfective tenses) is that while the former corresponds to a nominative case configuration, the latter correspond to an absolutive one, picking up the internal argument. At this point a potential problem arises. In Greek, specialized middlepassive agreement inflections are incompatible with –th morphology:€ thus, *pli–th–ome or *pli–th(ik)–omun are ruled out. Yet in Albanian, specialized middle-passive agreement co-occurs with the€–hE affix, which we construe as a morphological-level counterpart of the middle-passive clitic u, exactly as Roussou (2008) suggests for the –th affix in Greek. Now, note that Roussou (2008) does not provide an analysis of the thematic vowel specialized for the middle-passive agreement endings; in fact, she adopts a segmentation that does not separate the thematic vowel from the person ending with which it combines (following Ralli 2005). However, Manzini and Savoia (2005, 2007) treat Â�so-called thematic vowels as variables closing off the internal argument of the verb. From this perspective, the specialized thematic vowel of Greek middlepassive agreement inflections could be the real counterpart of Albanian –(h)E. This could also mean that the mutual exclusion between the middle-passive agreement inflections and the€ –th affix of Greek should be understood as a mutual exclusion between the€–th affix and the thematic vowel of these inflections, rather than with their (absolutive) person endings.6 We can now return to the distribution of the different lexicalizations of the middle-passive voice according to aspect (Tosk Albanian and Greek) or tense (Gheg Albanian). In Tosk Albanian the imperfective (including the present and the imperfective past) lexicalizes the middle-passive voice with specialized morphology, while the perfective lexicalizes it by syntactic means, i.e. through the u clitic. In Gheg varieties of Albanian, a split is also found, except that it

188â•… The middle-passive in Albanian is temporally based, distinguishing the present (with specialized morphology) from the past (with u). The importance of the aspectual/temporal split is underscored by the comparison with Greek, which also has a middle-passive voice whose lexicalization is differentiated according to aspect. Roussou (2008) suggests that this distribution involves an ergativity split, with a set of nominative inflections (the active ones) and a set of absolutive inflections (the middle-passive ones) apportioned to the imperfective and perfective tenses respectively. Nevertheless, in other attested ergativity splits, the nominative pattern tends to associate with imperfective tenses, and the ergative pattern with perfective ones. We therefore leave this matter open, having noted its interest. We started the discussion in this section by remarking that comparison between the lexicalization of the middle-passive with an u clitic, as analysed in section 5.2, and its lexicalization with specialized inflections could shed some light on the question of whether a single syntax underlies all of the forms allowing for the cluster of middle-passive interpretations. Since we have assumed that the€–(h)E represents a morphological-level counterpart to the u clitic, the answer is in a sense positive:€in both cases a crucial role is played by the variable instantiation of the internal argument. Yet while one of the structures has a syntactic-level variable (in the shape of the u clitic), the other does not; crucially, no abstract structure, specifically no movement structure, intervenes to make these two morpholexical instantiations of middle-passive isomorphic. 5.3.1 Be–participle The question of whether the same syntactic structure underlies the same (cluster of) interpretations, i.e. the uniformity question in the sense of Culicover and Jackendoff (2005), can equally be asked with respect to the third possible lexicalization of Albanian middle-passives, namely the combination of the jam ‘I am’ auxiliary with a participle. We delay our account of these structures till chapter 6, where they are considered in relation to other instantiations of perfect tenses. In the final part of this chapter, we consider a different lexicalization available in Albanian for the passive reading, again involving the jam ‘I am’ auxiliary. This auxiliary, however, embeds a participial form which is different from those reviewed in section 5.1, yielding passive for all of the various tenses of the paradigm. Both in being restricted to the passive reading, and in not being restricted temporally/aspectually, the relevant construction in Albanian is close to the English be–en passive and its Romance counterparts (essere V–to in Italian). It is useful to begin by reviewing what is perhaps the most basic occurrence of jam, i.e. as a copula. In (29), we provide various examples of this configuration in the variety of Shkodёr. The paradigm of adjectival embedding in (29a–b) is the same as in the standard. This requires the full adjectival

Specialized inflectionsâ•… 189 inflection, including a preposed article and a postposed inflection, both agreeing for number, gender, definiteness and Case. (29)â•… Shkodёr a. t i kutS/  kutS- it.is m.sg red/ f.sg red-f. ‘It is red’ b. jan t kutS/ kutS- they.are pl red/ red-f ‘They are red’

The examples in (30) illustrate the embedding of participles (consonantal base) in the copular construction, yielding a structure identical to that seen for adjectives in (29). It is interesting that in the variety of Shkodёr, the adjectival inflection is not simply added to the bare bases with which participles have been seen to coincide in (5) above. Rather, the base shows a participial inflection€–m or€–un depending on verbal class (vocalic and consonantal respectively). Verbal adjectives are regularly formed from transitive predicates (‘to dress’) as in (30a–a’), from intrinsically middle-passive ones (‘to sit down’) as in (30b), and from active unaccusative ones (‘to come’), as in (30c). The adjective in (30d), from unergative ‘to sleep’, does not have the meaning of ‘slept’, but rather of ‘asleep’. The example in (30e) establishes that the participial form with adjectival inflection is found not only in copular constructions, but also in other predicative constructions, here in a small clause embedded under kam ‘I have’; incidentally, the same example also allows us to illustrate participles formed from vocalic bases. As shown in (30a–a’), copula–participle formations based on transitive verbs admit of the passive meaning, disambiguated here by the presence of a by-phrase. A comparable example is provided in (31) for Gjirokastër. (30)â•… Shkodёr a. t i veS-un/  veS-un s/he.is m.sgâ•… dress-ed/â•… f.sgâ•… dress-ed ‘S/he is dressed up (by his/her mother)’ a’.â•… jena t veS-un/ t veS-un-a we.are pl dress-ed/ pl dress-ed-pl ‘We are dressed up (by our mother)’ b. guja ɐï†fitâ•…  ulun the.womanâ•… is f.sg seated ‘The woman is seated’ c. ï†fit i arDun he.is m.sg arrived ‘He has arrived’ d.â•… ai ɐï†fit i fjetun he is m.sg asleep ‘He is asleep’

(pi s Qms) (by gen mother) (pi s Qms) (by gen mother)

190â•… The middle-passive in Albanian e. i kam kmiSa-t /t lQ-m- themâ•… I.haveâ•… shirt-s Art wash-ed-f ‘I have the shirts washed’ (31)â•… Gjirokastër jan t@ vu/ vua they.are pl dressed.up/dressed.up-f. ‘They are dressed up’

The data in (29)–(31) can be usefully integrated with those of Arbёresh varieties, which also have the construction in which the copula is followed by the inflected participle. This is illustrated for Portocannone in (32a) with a transitive predicate and in (32b–c) with unaccusative predicates. In this construction, transitive predicates associate with by-phrases, as in (32a); by contrast, in the discussion of (14) we saw that the middle-passive voice is normally restricted to a generic agent; i.e. it excludes the by-phrase. The participle that is involved in the copular construction is the ordinary active form, despite the existence in the Portocannone variety of a specialized middle-passive participle, as discussed in section 5.1. As shown in (32a’–b’), the same variety also admits of copular (non-perfect) jam–participle constructions in which the participle is uninflected. (32)â•… Portocannone a. kt kmiS jan/ kjetn t@ la- it- these shirts are/ were pl wash- prf- ‘These shirts are/were washed by them’ a’.â•… kt kmiS jan/ kjetn la- it- ur theseâ•… shirts are/ were wash- prf- prt ‘These shirts are/were washed by them’ b. iSt i uj- ur he.is m.sg seat- prt ‘He is seated’ b’. iSt uj- ur he.is seat- prt ‘He is seated’ c. iSt i vdEk- ur he.is m.sg die- prt ‘He is dead’

ur- a (tE a’ta) prt fpl (by them) (tE a’ta) (by them)

For bases ending in a vowel, as in (33a), the adjectival participles of Shkodër shows an –m suffix which can be analysed as the bearer of the aspectual, perfective properties of the participle. Thus, the –m inflection is an I head which takes the verbal base as its complement. The verbal base, including the thematic vowel, appears independently of further inflectional material as the invariable participle in (5), i.e. lQ. Verbal bases ending in a consonant, in turn, form the

Specialized inflectionsâ•… 191 participle with a suffix –(u)n, as in (33b). We assume an analysis parallel to that for bases ending in a vowel, so that the –n inflection is an I aspectual, perfective head, which takes as its complement the verbal consonantal base combined with a thematic vowel€–u, selected in this case by the perfective aspect itself. Entirely parallel structures can be provided for Tosk varieties (including both the standard and Arbëresh dialects, like Portocannone), with the only difference that –r is the participle-forming I inflection in these languages with both vocalic and consonantal bases. (33)â•… Shkodër a. e�

I m

veʃu

I n

b.

The agreement inflections following the verbal subconstituents in (33) are the same as those found in adjectives and nouns. Following the analysis of nominal inflections in Albanian in chapter 7, we assume that at least vocalic endings correspond to N inflections, whether they lexicalize so-called accusative, as in (34a) or so-called nominative, as in (34b). (34)â•… Shkodër a. N ε e�

I m

b. N a veʃu

I n

192â•… The middle-passive in Albanian The article preceding the participle can be treated, as in standard analyses of the noun phrase, as a D element at the phrasal level, giving rise to structures like (35). In present terms, as detailed once again in chapter 7, the agreement between the determiner and the Case inflection of the participle is a reflex of the fact that they form a chain€– in other words, together they satisfy the same argument slot(s). (35)â•… Shkodër a. D ε

I l� mε

D t

I veʃuna

b.

From the point of view of the present discussion, the crucial question concerns the passive interpretation of sentences like (30a–a’). Given the complete formal similarity to (29), we may once again begin by asking how an ordinary copula–adjective construction is interpreted. In the case at hand, it seems evident that the embedded predicate kutʃ ‘red’ has a single argument, lexicalized by its N inflection and by the determiner that it agrees/forms a chain with. Agreement and chain formation in turn hold with the D (i.e. EPP) argument of jam ‘I am’, i.e. its inflection. The overall result is, correctly, a raising interpretation, whereby the EPP argument of the matrix sentence is assigned the argument slot of the embedded (adjectival) predicate. The same raising interpretation characterizes examples like (30a–a’), the only difference being that they involve transitive, i.e. two-place, participial predicates. In this case, the determiner and the agreeing inflection of the participle pick up its internal argument slot; this means that participial structures are absolutive, in the sense that the D argument which closes them (the determiner) is associated with their internal argument slot. The latter ends up being associated with the EPP argument of jam ‘I am’ by the same mechanism just detailed for the copular sentences in (29)€– so that, from this perspective, the passive interpretation of sentences like (30a–a’) (with the internal argument of the participle assigned to the EPP argument of the copula) is simply an

Specialized inflectionsâ•… 193 ordinary adjectival interpretation. The external argument slot of the participial predicate is in turn closed by an adjunct by-phrase or by generic quantification, in this latter case yielding the implicit agent reading. We noted in passing that in Portocannone, the copular passive can either have the agreeing form in (32a) or a non-agreeing form as in (32a’). In other words, the structure of (32a’) appears to be entirely analogous to that found in, say, the Gjirokastër example in (5) for the perfect middle-passive. Nevertheless, we maintain the conclusion that all copular passives in (30)–(32) result from a nominal embedding of the predicate; the lack of agreement in the Portocannone example in (32a’) represents an independent parameter. By contrast, in chapter 6 we pursue the conclusion that the structure of embedding in the perfect tenses is not nominal, but sentential€– so that examples like (5) have a bi-clausal, restructuring syntax. Once again, Greek presents a constellation of facts strictly comparable to those of Albanian. The middle-passive voice of Greek includes a perfect formed by an auxiliary followed by a participle, as in (36a); (36b) provides the comparison with the active. In this case, the middle-passive configurations of Greek are strictly comparable to those of Arbёresh varieties, since the auxiliary is exo ‘have’ both in the active and in the middle-passive, and the difference between the two voices is found in the participle. In (36a) it bears the middle-passive morphology€–th, coinciding in particular with the 3rd person singular of the middle-passive present perfective, while in (36b) it coincides with the 3rd person singular of the active present perfective. What is immediately relevant for present purposes is that Greek has another participial form which is created by attaching to the verb base the affix€–men followed by the adjectival (nominal) agreement for gender, number and Case. The Greek copular (‘periphrastic’) construction formed by this participle and the copula has a passive interpretation, as in (37), under which, as discussed at length by Anagnostopoulou (2003), it systematically co-occurs with by-phrases. (36)â•… a. b.

exo, exis, exi, etc. pli-th-i have.1sg, 2sg, 3sg, etc. washed ‘I have been washed’ (‘you have/he has been/ etc. washed’) exo, exis, exi, etc. plin-i have.1sg, 2sg, 3sg, etc. washed ‘I have washed (something)’ (‘you have /he has etc. washed’)

(37) Afta ta pedhia ine dolofoni-mena these the children are murder-ed ‘These children are murdered’

194â•… The middle-passive in Albanian A question that we have so far left implicit is how the copular structure relates to the generative notion of adjectival passive. All the evidence at our disposal suggests that the adjectival, i.e. stative, reading of passives can be associated with the Albanian examples in (30)–(32) involving transitive predicates, both with and without a by-phrase. On the other hand, the verbal, i.e. eventive, passive reading seems to be responsible for the fact that Arbёresh speakers routinely offer these structures€ – rather than structures like (14) above€– when passive is elicited. An important line of thought in generative grammar, dating back at least to Wasow (1977), associates verbal passives with a syntactic derivation, and adjectival passives with a lexical one. The latter analysis is adopted by Terzi and Wexler (2002) for Greek. We take it that the evidence from Albanian runs counter to the traditional account of verbal vs. adjectival passives in terms of syntactic vs. lexical derivation, as also argued for Greek by Anagnostopoulou (2003). Instead, we conclude that the so-called adjectival and verbal passive are just interpretations attached to the same copular structures€– which, everything else being equal, are systematically ambiguous between the two. Anagnostopoulou (2003), basing her work on Kratzer (2000), introduces in the structure of adjectival (i.e. stative) passive a category ‘Stativizer’ differentiating it from the verbal, i.e. eventive passive. But this at best succeeds in translating the interpretation into the syntax, without any explanatory gain, while at worst it obscures the fact that the stative/eventive opposition (i.e. in traditional generative terms the adjectival/verbal one) is precisely a matter of ambiguity at the LF interface, not corresponding to any independently observable morphosyntactic difference. Finally, the fact that in Arbёresh, by-phrases, which normally occur in the copular construction (32), are incompatible with middle-passive voice forms, as in (14), is reminiscent of a restriction found in Italian (and generally in Romance) against by-phrases in si-passives. In both instances, the external argument is interpreted, but the only possible reading is through binding by a generic operator; in other words it cannot be linked to a referential noun phrase within an adjunct noun phrase. In Romance, one may be tempted to relate this restriction to the presence of si, which is sometimes construed in the literature as an absorber of the external theta-role. In Arbёresh, however, the impersonal reading of the passive characterizes not only the perfective forms with the u clitic, but also the imperfective forms with the specialized verb morphology. What is more, mainland Albanian varieties seem to allow by-phrases in contexts that are morphosyntactically

Specialized inflectionsâ•… 195 identical to that of Arbёresh. Therefore we conclude that there is an independent parameter defining the possibility of middle-passive forms co-occurring with a by-phrase€ – i.e. whether the external argument of a middle-passive allows quantificational closure only, or can also be closed by an adjunct (essentially a form of predication, external to the core event of the sentence).

6 The auxiliary:€have/be alternations in the perfect

The discussion in this chapter is based on a set of data which reflect microvariation in a closely related set of languages (Romance varieties) as well as variation between more distant languages (Romance and Albanian). In section 6.1 we present the basic evidence and review the notions of person split, transitivity and voice, in terms of which we analyse the data in sections 6.2 and 6.3. We discuss person-split systems in detail in section 6.2, and we concentrate on auxiliary selection according to transitivity/voice in section 6.3, arguing in particular that the notion of transitivity should be split from that of voice, and a characterization not based on movement should be adopted for the latter (cf. chapter 5). In section 6.4, we introduce some Italian varieties in which selection according to transitivity/voice and the person split cross-cut, as well as varieties in which the have/be distinction is neutralized in parts of the paradigm. As discussed in section 6.1.1, the account we provide is crucially based on the assumption that the embedded participle does not select the auxiliary, in the sense in which a lexical category could be said to select its functional projections. Rather, the auxiliary and the participle define two independent sentences (Kayne 1993), and the selectional relation is the ordinary one from matrix predicate (the so-called auxiliary) to embedded sentence. Adopting the bi-clausal analysis just outlined for perfects amounts to eliminating auxiliaries, i.e. functional verbs, from the present grammar. 6.1

Evidence

We begin with the well-known evidence concerning standard Italian. As argued by Burzio (1986), auxiliary selection in Italian is sensitive to the distinction between transitive/unergative and unaccusative verbs; the former take have as in (1a), and the latter take be, as in (1b). (1)â•… a.â•… Ho lavato (la camicia) I.haveâ•… washed the shirt ‘I have washed (the shirt)’ 196

Evidenceâ•… 197 b. Sono arrivato/arrivata I.am arrived.m/f. ‘I have arrived’

The most notable complication concerning Italian is that the presence of the si element correlates with the selection of be, as in (2), independently of the many readings available for si:€namely reflexive, as in (2a), anticausative, as in (2b), passive, as in (2c) or impersonal, as in (2d). Most importantly, in the impersonal in (2d), the auxiliary be induced by si co-occurs with an overt accusative clitic€– i.e. with a transitive frame. The fact that si cannot be treated simply as a valency-reducing morpheme (an ‘intransitivizer’) is underscored by examples like (2d’), where it can be seen that si (impersonal again) combines with an unaccusative predicate. (2)â•… a.â•… b. c. d. d’.

Gianniâ•… si è lavato G. MPâ•… is washed ‘Gianni has washed himself’ Gianniâ•… si è svegliato G. MP is woken ‘Gianni has woken up’ Le camiceâ•… si sonoâ•… lavate (a secco) The shirts MPâ•… are washed dry ‘The shirts have been dry-washed’ Li si è mangiati themâ•… MPâ•… isâ•… eaten ‘One has eaten them’ Si è arrivati One is arrived ‘One has arrived’

In (2) we have glossed si as MP to suggest ‘middle-passive’. In fact, the closest morphology to si we know of is the so-called middle-passive morphology of languages like Greek or Albanian analysed in chapter 5. In (3) we reproduce the basic data showing that the middle-passive morphology of Albanian has the same range of interpretations as the si morphology of Italian, including reflexive (3a), anticausative (3b), passive (3c) and impersonal (3d). Note that, crucially, in (3d) the middle-passive morphology attaches to an unaccusative verb, just as si does in Italian. (3)â•… Gjirokastër a.â•… la- (h)E- m wash- MP- 1sg ‘I wash myself’ b. zu- (h)E- t wake- MP- 3sg ‘He wakes up’

198╅ The auxiliary:€have/be alternation c. d.

kt kmi la- (h)E- n ŋga aj these shirts wash- MP- 3pl by him ‘These shirts are washed by him’ ŋga ati dil- E- t from there exit- MP- 3sg ‘One exits from there’

The comparison between the middle-passive voice of Albanian and Italian si is made particularly direct by the fact that, whereas in the present, the middlepassive voice of Albanian is lexicalized by the specialized inflection in (3), in the perfective past it is lexicalized by a clitic€– namely u€– which combines with normal active inflections, as shown in chapter 5 and below in (4). (4)â•… Gjirokastër a.â•… u la- it- a MP wash- Prf- 1sg ‘I washed myself’ b. u z- it MPâ•… wake- Prf ‘He woke up’ c. ata u z- it- n ŋga t tiErt they MP wake- prf- 3pl by the others ‘They were woken by some people’ d. ŋga ati u d from thereâ•… MP exited ‘One exited from there’

There is a third way of instantiating middle-passive voice in Albanian through auxiliary selection€– this is the one that concerns us directly, and we will come to it shortly. Returning to the issue at hand, namely the distribution of have and be in the Italian perfect, the descriptive conclusion is that be is found with unaccusatives and with middle-passives. Have is restricted to the complementary set, i.e. transitive/unergative actives. Looking back to the data in (2), it appears that the notions of transitivity and voice in Italian cannot be collapsed; surely, example (2d) is not intransitive, even though it is middle. Indeed it is precisely because of si that Burzio (1986) could not provide a unified characterization of be-selection, but ended up with a disjunctive statement. As indicated at the beginning, here we shall consider the problem of auxiliary selection in a parametric context. Hence, before evaluating whether selection frames for have and be can be unified in Italian, we shall turn to other parametric settings. One of them is especially well-known from Germanic languages like Dutch or German, but is also robustly attested in Romance varieties such as Soazza in (5)–(6). Like Dutch, and like Italian, Soazza has be with

Evidenceâ•… 199 unaccusatives as in (5a) and have with unergatives/transitives as in (5b). The interesting point is that exactly the same distribution holds in the presence of si morphology, as in (6). Thus, the reflexive in (6a), the anticausative in (6b) and the impersonal/passive in (6c), which are formed with transitive/unergative verbs, have auxiliary have. Only the impersonal in (6d), formed with an unaccusative verb, has auxiliary be. (5)â•… Soazza (Grisons) a. som rivo/ rivada I am arrived.m/f ‘I have arrived’ b. o dormi:t I.have slept ‘I have slept’ (6)â•… Soazza a. El/la s a lavo/lavada he/she MP has washed.m/f ‘He/she has washed himself/herself’ b. alâ•… s a Zmortso al tSar it MP has gone.off the light ‘The light has gone off’ c. s a sempro dormit beŋ MP has always slept well ‘It has always been slept well’ d. s e sempro rivo tart MP is always arrived late ‘One has always arrived late’

Reference to a Romance variety like Soazza, as opposed to better known languages like Dutch, allows us to establish a few important points. First, in comparing Italian with Dutch, one may be tempted to conclude that the different auxiliary selection with si and zich respectively is due to their different clitic vs. non-clitic status; the account of Reinhart (1997) and Reinhart and Siloni (2005) is built on this premise. However, comparison of sufficiently close languages like Italian and Soazza shows that different auxiliary selection properties can correlate with essentially stable properties of the middle-passive morphology. Thus, there is no doubt that the s morpheme in (6) has exactly the same clitic prosody, distribution etc. as its Italian counterpart. Second, Dutch zich and Italian si also differ in that zich does not combine with unaccusative verbs to yield the impersonal reading (again we take this to be an independent parameter). But s in Soazza is like Italian si in this respect€– which allows us to show that s simply does not influence auxiliary selection. Rather, selection

200â•… The auxiliary:€have/be alternation is determined by verb class (transitive/unergative vs. unaccusative) in (6) as well. Albanian is in a way the mirror image of Soazza, in that it displays no sensitivity to transitivity alternations, but selects have and be according to voice. Recall from chapter 5 that the present perfect of Albanian, like that of Romance and Germanic languages, is formed by a combination of auxiliary and perfect participle. In the active voice, the auxiliary is have, both with transitives and with unaccusatives, as in (7a) vs. (7b). However, in the middle-passive€– i.e. as part of the conjugation that includes the present in (3) and the simple past in (4)€– the perfect is formed with be. In fact, be followed by the very same perfect participle found in the active (i.e. without the support of either suffixal or clitic material) yields the set of middle-passive interpretations, namely the reflexive (8a), the anticausative (8b), the passive (8c) and the impersonal (8d). (7)â•… Gjirokastër a.â•… E ka la- it- ur it he.has wash- prf- prt ‘He has washed it’ b. ka dal he.has gone.out ‘He has gone out’ (8)â•… Gjirokastër a. t la- it- ur he.is wash- prf- prt ‘He has washed himself’ b. t zu- a he.is wake- prt ‘He has woken up’ c. kt kmi jan la- it- u ŋga aj these shirts are wash- prf- prt by him ‘These shirts have been washed by him’ d. ŋga ati ɐt dal mi from thereâ•… it.is exited well ‘One has exited well from there’

Incidentally, it may be noted that the perfect participle in Albanian never agrees with any of its arguments€– though a full agreement paradigm is available for it and emerges in adjectival contexts (cf. chapter 5). By contrast, Soazza has the same perfect participle agreement system as standard Italian, even if the auxiliary is have, for instance in the reflexive in (6a). In other words, perfect participle agreement is entirely independent of the selection of have or€be.

Evidenceâ•… 201 Another parametric choice attested by well-known Germanic and Romance languages (such as English or Spanish) is formation of the present perfect with the auxiliary have independently of transitivity and voice. In (9) we illustrate this parametric option with a Southern Italian variety which (like Spanish) shows the insensitivity of auxiliary selection to si; and (unlike Spanish) provides an example of the independence of participle agreement from auxiliary selection, since unaccusative participles as in (9b) agree with their argument. (9)â•… Verbicaro (Calabria) a.â•… aj laßa:t (a makn) I.have washed the car ‘I have washed (the car)’ b. a mmurt / mmɔrt s/he.hasâ•… died.m/f ‘S/he has died’ c. s a llaßa:t MPâ•… he.hasâ•… washed ‘He has washed himself’

Another logical possibility, i.e. selection of the be auxiliary in all of the present perfect, appears to be more rarely instantiated€– but can also be found in Romance varieties such as Pescolanciano in (10). While the selection of be with the unaccusative in (10a) or si in (10c) may be familiar from Italian, this is a language that also has be with transitives, as in (10b). Incidentally, it will be seen that the agreement pattern is the one familiar from Italian whereby the feminine morphology in (10b) singles out the internal argument, lexicalized by the accusative clitic. (10)â•… Pescolanciano (Abruzzi) a. sŋg mnu:t I.am come ‘I have come’ b. la   her they.are called ‘They have called her’ c. ts E lava:t@ MP he.is washed ‘He has washed himself’

Finally, there is another major pattern of the have vs. be split instantiated by Romance languages, since in some varieties auxiliary selection is sensitive to the reference of the EPP argument. The most widely known split opposes the 1st and 2nd person with be to the 3rd person with have (Kayne 1993; Cocchi 1995; D’Alessandro and Roberts 2010), as in the paradigm in (11) (in

202╅ The auxiliary:€have/be alternation the examples A indicates the selection of have auxiliary and E the selection of be auxiliary). This classical person split is oblivious to verbal class, as shown in (11), as well as to si.1 (11)╅ S.Benedetto del Tronto (Marche)

s vnu:t/ vist E Si E a A mâ•… E t E a A ‘I have come/seen’ etc.

Note that there is no intrinsic association of be with 1st and 2nd person and of have with 3rd person. Thus, varieties are found in which it is the 1st and 2nd person that are associated with have, while the 3rd person is associated with be, as in (12). This parametric choice is less robustly attested, so that in Morcone in (12) it only characterizes the singular, while the plural has have in all persons. However, the limitation of the person split to the singular is an independent parameter, since it can also be seen in a variety like (13) which in the singular has be in the 1st and 2nd person and have in the 3rd. Incidentally, the morphophonology of Morcone (with non-neutralized final vowels) allows us to see that the participle agreement once again follows exactly the same lines as standard Italian€– namely, agreement with unaccusatives (‘come’) and lack of agreement with unergatives (‘slept’). (12)â•…

Morcone (Campania) addo menuto/ durmuto a E emo menuti/durmuto ete ao ‘I have come/slept’ etc.

A A E A A A

(13)â•…

Bisceglie (Apulia) s drmmi:t/ vni:t si a mm av:tâ•… onn ‘I have slept/come’ etc.

E E A A A A

Evidenceâ•… 203 The varieties in (11)–(13) prompt a few general remarks. First, we have seen that although the pattern in (11) may be more robustly attested, the reverse pattern in (12) is equally possible. However, not all parametric choices are reversible. Thus, we are aware of no language in which auxiliary selection is the reverse of, say, Soazza, so that be is associated with transitive and have with unaccusatives. Also, we do not know of any language which patterns in the opposite way to Albanian, selecting have with middle-passives and be with actives. This lack of reversibility will have to be accounted for as part of the overall parametric picture. Second, the picture of auxiliary selection according to person provided in (11)–(13) is greatly simplified with respect to the actually observed variation. A fuller picture will be presented in section 6.2. Importantly, it is not only auxiliary selection according to person that is subject to this fine variation, but also transitivity and voice alternations, as discussed in section 6.4. 6.1.1 Theoretical background Our study of parametrization in auxiliary selection presupposes a number of assumptions concerning the structure and the interpretation of perfects. Perhaps the single most important assumption is that perfects are not monoclausal, consisting of a verb associated with an auxiliary functional projection€ – as in the conception of English auxiliaries of Chomsky (1957, 1981, 1995). Rather, the embedded participle and the matrix auxiliary each define a separate sentential unit (Kayne 1993). The importance of this point cannot be overstated. A decade after Kayne’s (1993) discussion, Bentley and Eythórsson (2003:€460) maintain that perfective auxiliaries are morphological exponents of tense/aspect features on a par with past tense markers in English … On the basis of the above analysis, the selection of perfective auxiliaries in Italian, Dutch and other languages (‘auxiliary selection’) can be considered to be an instance of allomorphy. Â�Accordingly, ‘have’ and ‘be’ are allomorphs of a tense/aspect morpheme which in combination with the past participle forms the analytic perfect. This type of selection might be taken to be comparable to the formation of the past with weak and strong verbs in English. This would imply that ‘have’ is added by a rule and that the smaller class of verbs selecting ‘be’ would have to be memorized … we propose, in the first instance, that auxiliary selection in languages like Italian is due to a morphological rule which is sensitive to the lexical semantics of verbs.

Their evidence includes the fact that ‘agreement markers (qua clitics) attach to the tense/aspect markers (qua auxiliaries)’. They argue that on a ‘clitic climbing/argument composition’€– i.e. restructuring€– analysis, the ‘obligatoriness’ of the positioning of clitics must be stipulated.

204â•… The auxiliary:€have/be alternation Now, in some Romance languages, enclisis on the participle is actually observed. Belletti (1990) notices that standard Italian has ‘absolute’ (i.e. adjunct) participle constructions, in which the participle is associated with sentence-level inflectional properties such as the ability to support an enclitic. If we adopt Kayne’s (1991) theory of enclisis, in which enclisis depends on the high position of the verb in the sentence, we conclude that the participial clause has a C layer as well, hosting the participle itself in enclitic contexts. What is more, the Piedmontese varieties reviewed in chapter 3 show enclisis on the participle in the perfect as well, as in (14). We take this to provide evidence that the participle is a full clause, with the verb in C. In particular, Forno in (14) displays not only enclisis on the participle, but also proclisis on the auxiliary, as in (14a’–b’). This optionality of clitic positioning in Forno parallels that of clitic climbing in Italian ‘restructuring’ constructions, in the sense of Rizzi (1982)€– suggesting that they admit of the same explanation.2 (14)â•… Forno Valle Strona (Piedmont) a.â•… l a viStu -n he has seen us ‘He has seen us’ a’. a n a viSt he us has seen ‘He has seen us’ b. l E lava -s heâ•… isâ•… washed -MP ‘He has washed himself’ b’. a s E lava he MP is washed ‘He has washed himself’

Albanian varieties provide even stronger evidence that the embedded participle syntactically represents an independent sentence (Savoia and Manzini 2007). Thus, in the Arbëresh variety of Portocannone the participle can be introduced by the sentential connective  ‘and’, and pronominal clitics are associated with the participial clause rather than with the matrix clause, as illustrated for the active in (15a) and for the middle-passive in (15b). Note that, while in Romance varieties like (14) a clitic associated with the participle appears in enclisis (showing that the participle is in a relatively high position, i.e. C), in Albanian the clitic precedes the participle, as with any other finite verb. This leads us to conclude that participles and finite verbs occupy exactly the same position in the sentence, i.e. I.

Evidenceâ•… 205 (15)â•… Portocannone a.â•… ai kiS E E he had and it ‘He had broken it’ b. ai kiS E u it had and MP ‘It had broken’

tSa- it- ur break- prf- prt tSa- x- ur break- MP- Prt

We can associate with the active in (15b) a structure of the type in (16). Both the kam auxiliary and the participle head their own sentential projection, each of which is associated with a full argument structure (witness the clitics in the participial clause). We assimilate the coordinating particle to a subordinating one, as suggested by Kayne (1994, and references quoted there). In accordance with the theory of complementizers in chapters 1–2, we then construe  as an autonomous nominal head, which is selected by the matrix predicate and takes the participial clause as its complement. (16)â•… Portocannone D ai I kiʃ I ε Q u I tʃaxur

Besides the conclusion that perfect participles are bi-sentential structures, we share with Kayne (1993) the conclusion that the have and be auxiliaries are exactly the same verbs that appear in possessive or copular constructions respectively, descriptively as main verbs. The main argument in favour of this conclusion is that the coincidence of the different contexts in which the same lexical forms appear is systematic€– and cannot therefore be imputed to simple homophony. Bentley and Eythórsson’s (2003) approach is incapable of capturing this obvious and important fact. On the other hand, Kayne (1993) also argues that have and be are transformationally related. His idea, which is based on the possessive construction, is that have is essentially an applicative of be, derived through the incorporation

206â•… The auxiliary:€have/be alternation of a (dative) preposition; in other words, possessive have is be–to, as proposed by Benveniste (1966). Similarly, auxiliary have is derived from auxiliary be through the incorporation of a prepositional complementizer. The problem is that, as far as we can see, there is no Romance variety, of the many examined, that provides any morphophonological clue in favour of the proposed derivation (nor any other language that we know of).3 Therefore we simply assume that have and be are two independent lexical entries. In other words, have and be are just what they appear to be€– namely the independently known possession and copular verbs, eventually embedding a participial clause. There is one major piece missing in the structural sketch of participles provided above concerning the EPP argument of the participle. As in all non-finite clauses (including, notably, infinitivals) the EPP position of the participial clause is not overtly realized. According to the theory of empty categories in Chomsky (1981) maintained by Chomsky (1995), the EPP position of nonfinite sentences corresponds either to a trace€– i.e. a copy€– in raising constructions or to a base-generated empty category, conventionally PRO, in control constructions (whether obligatory or arbitrary control). On the other hand, some authors have sought a unified characterization of these environments in terms of the minimalist notion of trace, i.e. copy. Thus, for Hornstein (1999), (obligatory) control as well as raising is derived through movement; raising and control are simply the names for different interpretations of the same syntactic object, i.e. a chain. Because this theory seeks to unify only the bound readings of the embedded subject of non-finite sentences, so-called arbitrary control is excluded from it€– being represented presumably by an empty category again (say pro). Another possibility that has been explored in the literature is that in fact there is no embedded EPP position in non-finite sentences. This stance is explicitly taken by Epstein and Seely (2006), but is also implicit in Manzini and Roussou (2000). It is interesting, therefore, that both Roussou (2009) and Manzini (2009) reject this possibility on the basis of control and raising from finite complements in Greek and Albanian respectively. Since finite complements involve a nominative position, it is evident that control and raising interpretations cannot be linked to the absence of an EPP position. We therefore adopt the standard assumption that all sentences have an EPP argument. At this point, it also becomes relevant that we adopt a representational model of grammar, as defined by Brody (2003), in which chains are an interpretive construct at the LF interface€– and do not depend on a computationally driven process of movement. At the LF interface, the reading of a trace (i.e. a copy) is that of a variable (bound by its antecedent, i.e. the wh–operator

Evidenceâ•… 207 in wh–chains and so on). Similarly, the base-generated empty category (i.e. PRO/pro) is generically closed in the arbitrary reading, so is once again a variable (Lebeaux 1984). Therefore we simply assume that in non-finite clauses, in which the EPP argument is not provided by the morphosyntactic (agreement) structure, it enters the LF interface computation in the form of a variable. In raising and obligatory control, the EPP variable has an antecedent-bound reading; in arbitrary control, it is interpreted by generic closure (in the absence of an antecedent). Finally, we need to say a few words on the notions of voice and transitivity. Descriptively, in standard Albanian have forms the perfect in the active voice, as in (7), and be forms the perfect in the middle-passive voice, as in (8); in present terms, this means that have selects for an active embedded sentence and be for a middle-passive one. Similarly, in Italian (1)–(2) the middle-passive voice, as instantiated by si, is associated with be independently of the transitivity of the verb, while in the active have and be alternate according to transitivity. The latter is the only alternation present in Soazza in (5)–(6) independently of voice. Evidently, explaining these patterns requires a preliminary account of what exactly is meant by the descriptive labels of active and middle-passive or transitive and non-transitive. Furthermore, the two splits, according to voice and transitivity, cannot be unified, since Albanian or Romance varieties like Soazza are sensitive to one but not the other. What appears to be the more complex and therefore problematic of the two notions, i.e. voice, has been studied in chapter 5 in relation to Albanian. Our conclusion regarding the Albanian clitic u and its Italian counterpart si was that what ultimately gives rise to all of the different readings of the middle-passive voice is the property of si/u being a variable. As a result of the presence of this variable, one of the argumental positions of the predicate remains unassociated (in anticausatives/reflexives) or is interpreted only by quantificational closure (passives, impersonal) or is finally associated with an argument which is external to the core predicate (the by-phrase). Therefore middle-passive voice is defined by the lack of closure (or the generic closure) of the argument structure of the verb. By contrast, the theoretical characterization for the notion of transitivity would appear to be a settled question on the basis of Burzio (1986) and the relational grammar work on which it is based. According to Burzio (1986), transitive verbs have both an internal and an external argument€– linked to the sister of V position and to the sister of I (EPP) position respectively. Unergatives only have an external argument, which is linked to the EPP position. Finally, unaccusatives have only an internal argument, which is promoted to the EPP

208â•… The auxiliary:€have/be alternation position. Regarding the latter, it should not be forgotten that the movement derivation of unaccusative verbs is meant to unify them with middle-passives for the purposes of be-selection. Yet we have just seen that what is classically taken to be the strongest argument in favour of a uniform movement structure for middle-passives and unaccusatives in reality works against it. For in languages like Albanian, it is only middle-passives that combine with be, to the exclusion of unaccusatives€– and the reverse is true in Soazza. We have now adopted a characterization of middle-passives as predicates with an open variable in their argument structure, either not interpreted or interpreted through quantificational closure. But the standard construal of unaccusatives as predicates with a single argument slot, saturated by the EPP argument, leaves no room for non-interpreted or quantificationally closed argument slots. Therefore this simple characterization is sufficient to distinguish unaccusatives from middle-passives for present purposes€ – as already suggested in chapter 5 in relation to Albanian. The same characterization is also sufficient to distinguish them from transitives and unergatives€– for transitives have two arguments, and, following Hale and Keyser (1993) and Chomsky (1995), unergatives can be treated essentially as concealed transitives in which the internal argument has incorporated into the verb. 6.2

Auxiliary selection independent of transitivity/voice

Consider Pescolanciano in which the auxiliary of the perfect is be, as in (10). From the present perspective, Pescolanciano can simply be described as a language in which be embeds any participial clause without any restrictions. By contrast, in Verbicaro, the perfect is systematically formed with the auxiliary have, as in (9). However, we cannot simply say that have in Verbicaro embeds just any participial clause, with no restrictions, since passives in this language, as in standard Italian or English, are formed by the participle embedded under be. Recall that in the middle-passive perfect of Albanian varieties, jam ‘be’ is followed by an invariable (non-agreeing) participle, as in (7)–(8). By contrast, in the passive it is followed by a participle featuring both an agreement inflection and a preposed article, as discussed in chapter 5; a relevant example is reproduced in (17). (17)â•…

Gjirokastër jan t@ vESu/ vESua they.are Art dressed/dressed-f. ‘They are dressed’

Auxiliary selection independent of transitivity/voiceâ•… 209 We take this contrast to imply that the present perfect in (7)–(8) and the passive in (17) involve two different embedding structures. While the evidence reviewed in section 6.1 leads to the conclusion that the embedding in (7)–(8) is sentential, a nominal embedding is involved in (17), as discussed more in detail in chapters 5 and 7. We tentatively propose that the same distinction between sentential and nominal embedding of the participle holds (in some form or other) in Romance as well. Indeed, a contrast between lack of agreement in the perfect and agreement in the copular (passive) construction can be found in some Romance varieties. Thus, in (18), the free alternation between have and be in the perfect does not have any effect on the participle, which maintains its invariable, non-agreeing format. As shown in (18b–b’), on the other hand, in the same languages be requires an agreeing adjective in the copular construction. (18)â•… Montebello Ionico/ Saline Ioniche (Calabria) a. Era/ Eri/ Era/ Erumu/ Eruvu/ Erunuâ•… durmutu/ vinutu I.was etc. slept/ come a’. aiva/ aivi/ aiva/ aivumu/ aiuvu/ aivunuâ•… durmutu/ vinutu I.hadâ•… etc. slept/come ‘I had come/ slept’ etc. b. Era staŋku I.was tired b’. Erumu staŋki we.were tired(pl)

We thus factor out passive as involving a different structure of participial embedding with respect to perfects. If so, the comparison between Pescolanciano in (10) and Verbicaro in (9) yields a simple parameter whereby participial clauses are selected by either have or be to the exclusion of the other. 6.2.1 Auxiliary selection according to person The patterns of auxiliary selection according to person in (11)–(13) can be obtained at this point by letting the parameter just formulated€– i.e. the selection of participial clauses by have or be€– interact with the person split between 1st/2nd person and 3rd person. The latter is independently motivated in the typological literature. Thus, DeLancey (1981) argues that languages with so-called ergativity splits, i.e. alternations between the ergative (/absolutive) case system and the nominative (/accusative) system, most commonly oppose 1st and 2nd person to 3rd. Person splits are also found in Romance languages. Restricting ourselves to clitics, 1st and 2nd person object clitics differ

210â•… The auxiliary:€have/be alternation from 3rd person ones with respect to their distribution, their morphological make-up (gender and Case distinctions), their agreement properties, and more (for instance, the drop phenomenon studied by Savoia and Manzini (2010)). According to Manzini and Savoia (2007), the person split, in its various manifestations, depends on the fact that the speaker and the hearer (and the sets including them) are anchored directly in the universe of discourse, independently of their role within the event; on the other hand, non-participants in the discourse depend directly for their characterization on the position assigned to them within the structure of the event.4 The same split can be defined in more traditional frameworks by a feature [±discourse participant] (e.g. Legendre (2010) has [±local]). The reason we do not adopt such a characterization is that in our morphosyntax, there are only positively specified properties. In other words, the 3rd person cannot simply be characterized in negative terms, as excluding speaker and hearer reference (Benveniste 1966). Rather, 3rd person reference is characterized in positive terms, by its necessary anchoring in the event. In this sense, we also differ from Harley and Ritter (2002), who assume monovalent (privative) features, but at the same time adopt the default view of 3rd person, as characterized by the absence of the Participant feature node. Let us now cross the have vs. be parameter defined with respect to Pescolanciano and Verbicaro with the person split€– so that have and be can be sensitive to the reference of the EPP argument. The crossing should in principle yield two systems. In one system, be is associated with a 1st and 2nd person EPP argument while the 3rd person requires have€– while in the mirrorimage system, it is the 1st and 2nd person that require have while the 3rd person is associated with be. In other words, the crossing of the two parameters yields the systems observed, at least in the singular, in S.Benedetto in (11) and Morcone in (12) respectively. It is important to stress that we take the person split to interact with the selectional properties of auxiliaries at the core syntactic level. By contrast, for Bentley and Eythórsson (2001:€ 71), ‘alternation according to person is part of a grammatical person marking system on verbs’. Similarly Loporcaro (2007:€ 186) argues that ‘where precisely (in which verb persons) the morphemes ‘have’ and ‘be’ occur … is however a matter of morphology, not syntax’. What seems crucially to motivate their conclusions is the fact that there exist many more possible distributions than the 1st/2nd vs. 3rd exemplified so far. Thus, for Bentley and Eythórsson (2001:€71), ‘the mixed paradigm does not appear to be attributable to syntactic or semantic principles associating each auxiliary with a particular grammatical person. This possibility is ruled

Auxiliary selection independent of transitivity/voiceâ•… 211 out by the variety of existing patterns … Thus we would not subscribe to the view that there is a rule of auxiliary selection according to person’. Similarly, for Loporcaro (2007:€186), ‘what is inconceivable … is for this empirical variability to be directly encoded into structural categories in a one-to-one correspondence’. Here we will argue that, on the contrary, a syntactic account of this considerable variation is exactly what is needed. Quite intricate patterns can be gleaned from the tables in Manzini and Savoia (2007) and Legendre (2010). The number of languages described by Manzini and Savoia (2005, 2007) is obscured by the fact that several varieties display one or more alternations, both free and not free (such as alternations according to transitivity/voice and according to whether the following verb starts with a consonant or a vowel). In order to make the possible patterns emerge more clearly, in Table 6.1 we have reformatted the data in Manzini and Savoia (2007, Appendix to chapter 6). Wherever an alternation was present, we have written down two separate paradigms. Free alternations have been treated so as to minimize the resulting pattern€– i.e. where several of them were present, we chose be or have throughout. Even a cursory look at Table 6.1 reveals that the singular and the plural do not necessarily pattern alike€– in fact they most often display separate patterns. This has already been illustrated above by the contrast between S.Benedetto in (11) and Bisceglie in (13)€– where the same person split occurs throughout the paradigm in S.Benedetto and only in the singular in Bisceglie. Let us now consider the singular. The richest possible grammar is one in which not only the 1st/2nd vs. 3rd person split is allowed, but also 1st person vs. 2nd/3rd and 2nd person vs. 1st/3rd. Once these three possibilities are crossed with the two auxiliaries, six patterns emerge. All of these are attested in Table 6.1. Now the splitting of 1st from 2nd person€– i.e. of the two members of the discourse-anchored set€– is also a long-standing issue in the typological and formal literature. In typological approaches, where the discourse- vs. non-discourse-anchored split is taken to be a cut along an animacy/agenthood hierarchy (Dixon 1994), the 1st singular is taken to precede the 2nd singular in the hierarchy in that ‘the speaker is for him- or herself, the quintessential agent’ (Dixon 1994:€90). Here, of course, we have discounted the idea that animacy and/or agenthood enter into the determination of person-split phenomena, since our system does not contemplate the possibility of hierarchies, but only of discrete splits. Similarly, the formal feature hierarchy of Harley and Ritter (2002) yields ‘speaker’ as the unmarked member of the ‘discourse participant’ set. As mentioned above, we cannot have recour