Clitics in Greek
Linguistik Aktuell/Linguistics Today (LA) Linguistik AktueU!Linguistics Today (LA) provides a platform for original monograph studies into synchronic and diachronic linguistics. Studies in LA confront empirical and theoretical problems as these are currently discussed in syntax, semantics, morphology, phonology, and systematic pragmatics with the aim to establish robust empirical generalizations within a universalistic perspective.
General Editors
VVerneri\braharn
Elly van Gelderen
University ofVienna I Rijksuniversiteit Groningen
Arizona State University
Advisory Editorial Board Josef Bayer University of Konstanz
Christer Platzack University of Lund
Cedric Boeckx ICREA/Universitat Autonoma de Barcelona
Ian Roberts Cambridge University
Guglielmo Cinque University of Venice
Lisa deMena Travis McGill University
Liliane Haegeman University of Ghent
StenVikner University of Aarhus
Hubert Haider University of Salzburg
C. Jan-Wouter Zwart University of Groningen
Terje Lohndal University of Maryland
Volume160 Clitics in Greek. A minimalist account of proclisis and enclisis
by Marias Mavrogiorgos
Clitics in Greek A minimalist account of proclisis and enclisis
Marias Mavrogiorgos University of Cambridge
John Benjamins Publishing Company Amsterdam I Philadelphia
The paper used in this publication meets the minimum requirements of American National Standard for Information Sciences - Permanence of Paper for Printed Library Materials, ANSI z39.48-1984.
Library of Congress Cataloging-in-Publication Data
Mavrogiorgos, Marias. Clitics in Greek: a minimalist account of proclisis and enclisis I Marias Mavrogiorgos. p. em. (LinguistikAktuell!Linguistics Today, ISSN 0166-o829; v.16o) Includes bibliographical references and index. Greek language, Modem--Ciitics. 2. Greek language, Modem--Enclitics. I. Title. PAlo61.M38 2010
1.
489'·359--dC22 ISBN 978 90 272 5543 3 (Hb ; alk. paper) ISBN 978 90 272 8803 5 (Eb) © 2010 -John Benjamins B.V. No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any other means, without written permission from the publisher.
John Benjamins Publishing Co.· P.O. Box 36224 · 1020 ME Amsterdam -The Netherlands John Benjamins North America· P.O. Box 27519 ·Philadelphia PA 19118-0519 · usA
Table of contents
Preface
IX
CHAPTER 1
Introduction
1
CHAPTER 2
The properties of Greek ditics and their structural analysis 5 2.1 Introduction 5 2.2 Greek personal pronouns: Clitics vs. strong pronouns 5 2.2.1 Some basic properties of Greek personal pronouns 5 2.2.2 Strong vs. clitic/weak pronouns: The issue of deficiency 12 2.2.2.1 The partition of pronouns into types: Kayne (1975), Cardinaletti & Starke (1999) 12 2.2.2.2 Partition criteria applied to Greek 13 2.2.3 The morpho-syntactic status of clitics: affixes or words? 41 2.3 A dynamic approach to the structure of personal pronouns 52 2.4 Conclusions 55 CHAPTER 3
Internal structure of clitics and cliticization 57 Introduction 57 3.2 Analyses proposed for personal pronouns 57 .. ·3 Type A: DPs 58 3.:3.1 Sub-type (a): Pronouns as intransitive D0-heads projecting a DP 58 3.3-2 Sub-type (b): Pronouns as transitive/complex DPs 63 3.3-2.1 Uriagereka (1995): Pronouns as transitive- DPs which differ in their internal structure 63 3.3-2.2 Panagiotidis 2002 a uniform transitive D0 approach. 76 3·3-3 Sub-type C: Pronouns as heads in the extended verbal projection 91 3·3-4 Sportiche's account 102 3·4 Type B: Pronouns as transitive XPs 103 3.4-1 Cardinaletti & Starke (1999) 104 3.4-2 Dechaine & Wiltschko (2002) 107 3·5 Summary 111 3-1
VI
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
CHAPTER4
Derivation of prodisis 4.1 4.2
4·3
4·4·
4·5· 4.6 4·7
Introduction 113 Properties of clitics 113 4-2.1 Clitics: Movement vs. base generation 113 4.2.2 XPvs.X0 us 4.2·3 Clitic placement: Adjunction or incorporation? 119 Clitics as non-phases 127 4-3.1 Derivation of proclisis 127 4-3.2 The Syntax-Morphology mapping: Words as m-phases and clitics 4-3·3 The effect on the outcome: LF and PF saturation of clitics 134 4·3·3-1 PF valuation 134 4·3·3-2 LF valuation 134 Have - constructions 134 4-4.1 Other clitic climbing constructions 137 Indirect object constructions 137 Clitic doubling 138 Conclusions 143
113
131
5 Patterns of prodisis/endisis and the role ofV-movement 5-1 Introduction 145 5.2 De-limiting the phenomenon 145 5·3 Different types of enclisis: Syntax vs. PF 146 5·4 Finiteness: A rather problematic notion 157 5·5 V-movement 168 5.6 Conclusions 191 CHAPTER
145
CHAPTER 6
Endisis in Greek: The role of CM in deriving endisis 6.1 Introduction 193 6.2 The finiteness gradation: From indicatives to gerunds/ absolutives 193 6.2.1 Some preliminary remarks on verbal marking in Greek 194 6.2.2 Mood types and their finiteness properties 195 6.2.2.1 Indicative forms 195 6.2.2.2 Subjunctive clauses 203 6.2.2.3 Imperatives 206 6.2.2.3-1 Positive imperatives 2CYJ 6.2.2.3-2 Negative imperatives 217 6.2.2.4 Gerunds/ Active participles as absolutives/free adjuncts 220 6.2.3 Summary of empirical points 224 6.3 EPP and head movement: Extending the V domain 225
193
Table of contents vn 6.3.1 Preliminary remarks 225 6.3.2 The analysis 231 6.3.2.1 Assumptions 231 6.3.2.2 CM and enclisis 236 6.3.3 Further instances of the proclisis-enclisis alternation in Greek 6.3.3.1 Negated imperatives/gerunds in Greek 250 6.3.3.2 Surrogate imperatives 254 6.3·3·3 Clitic switch 255 6.4 Conclusions 266 CHAPTER?
Concluding remarks References Index
250
Preface
This book is based upon an original version finalized in 2009 (Mavrogiorgos 2009). Several changes have been made. Chapter 3 has been added from scratch, while minor revisions have been made in Chapters 2, 4, 5 and 6. This book aims at providing a principled analysis for two interrelated phenomena in the morpho-syntax of Greek clitic pronouns: proclisis (1) and enclisis (2), respectively: (1) When the verb is in the indicative or the subjunctive, the clitic pronoun precedes the verb, and nothing may intervene between it and the verb. a.
To katharisa. Ilcl cleaned.lsg 'I cleaned it'
b. Thelo na to kathariso. Want.lsg subj. itcl clean.1sg 'I want to clean if
(indicative)
(subjunctive)
(2) When the verb is in the imperative or the gerund, the clitic pronoun follows the verb, and nothing may intervene between it and the verb. a.
Katharise to! Clean.2sg.imp ilcl 'Clean it!'
oti... b. Katharizondas to katalava Cleaning ilcl understood.lsg that 'While I was cleaning it, I realized that.:
(imperative)
(gerund/ab solutive)
It is argued that clitic pronouns are topicalizers, namely optional determiner heads that are merged in the left periphery of the (direct or indirect) object DP of the clause. They encode familiarity/prominence/topicality, as opposed to the lower/internal determiner which encodes definiteness (see Alexi.adou, Haegeman & Stavrou 2007 on the internal and external determiner). In doubling structures the two determiners appear together in the numeration, while in non-doubling structures only the lower D-head appears. By being at the edge of a minimal phase, the external D-head is probed by v*- transitive, to which it moves and incorporates, giving rise to a proclitic doubling structure. Given that the clitic acts as a scope marker, movement of the clitic to v* equals to movement of the whole D P to the periphery of v*. Accordingly, the clitic pronoun is analyzed as the highest/external D-head that may merge within the DP. which marks the DP as
x
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
familiar, and which connects it with the outside world both semantically/pragmatically and syntactically. As far as the semantic restrictions on doubling are concerned, I assume that these follow from the interaction of the two D-heads. I further argue that syntactic cliticization follows from syntactic agreement between the clitic pronoun and a phase head. For Greek and for other languages with adverbal clitics which are sensitive to finiteness with respect to the proclisis/enclisis alternation I propose that the relevant phase head is v*-transitive and not Tense (T) (contrary to Kayne 1991, Philippaki-Warburton 1995, Terzi.1999a, among many others). The derivation proceeds as follows: following Kayne's (1975) movement hypothesis, I take the clitic to be merged as/within a DP/D in the complement position, its phi-features being visible to appropriate probes higher up in the clause. V (or some higher v head), being a phase head by inheritance, attracts the clitic to its (external) specifier, building an A-chain. At the same time, an optional EPP feature on v*, which is linked to a familiarity/focus/old information D feature, probes the clitic forming a parallel A'-chain. The clitic moves as an XP/X to the left edge of v*, where it incorporates, yielding proclisis. The incorporated clitic moves along with the v*-V complex to T. v* in this case is a minimal morpho-syntactic phase which is transparent at the edge. In this sense it allows nonphases (such as clitics for example) to incorporate into it under certain conditions (e.g. sisterhood and proper subset-hood in terms offeatures). Incorporation is allowed for two interrelated reasons. On the one hand, the clitic contains only a subset of features of those contained by v*. This allows incorporation, according to an incorporation law which claims that incorporation is allowed only if the features of the incorporee are not different from that of the incorporator (Roberts 2009). On the other hand, incorporation into v* is allowed, even though v* is a minimal phase, because its edge is still accessible, due to the fact that v* has not checked all its features as yet In enclisis the generalization is that person agreement on T is restricted/ defective, while the verb must check verbal inflection on the higher Complementizer Modal head ( CM). The elitic pronoun targets the v* head, as in proclisis, however it does not incorporate into it, because CM is the new phase head by being the highest inflectional verbal head in a chain of verbal heads (phase sliding). The verb moves to CM and the clitic merges with it from the lower specifierofvP/TP. In this way we get the generalization thatenclisis obtains when the verb moves across the cliticization site to a V-related site, i.e. to a site where a verbal inflectional head is found. This allows us to differentiate between imperatives which have V-to-C movement and take enclitics, from interrogatives, which in some languages have V- to- C movement, but take proclitics. The advantage of this analysis is that it can be generalized across constructions and across languages without great difficulty, since it is based on the general principles of (i) subset of features, and (ii) edge availability/accessibility. Moreover, by linking cliticization to agreement with phase heads, and given Chomsky's theory on phase heads, it is flexible enough to be able to account for a plethora of distinct clitic constructions both within a single language and cross-linguistically, which is something previous theories lacked.
CHAPTER 1
Introduction
16 years ago Luigi Rizzi (see Rizzi 1993) claimed that the study of Romance cliticization must, at least, deal with the following four questions: (1) a. b. c. d.
What is the categorial status of clitics? What makes clitics move? What is the landing site of cliticization? What determines encliticization and procliticization?
Since then, but even before (see e.g. Kayne 1975), these questions have been challenging the minds of linguists investigating the different aspects of the syntax of elitic pronouns, not only in Romance but also in other languages. The aim of this monograph is to try to give an answer to each one of these questions for Standard Modern Greek (henceforth, Greek), and even to a few more. But before we go into what this monograph is all about, let me briefly point out what it is not about. This book does not contain a literature review of different theories of cliticization, clitic structure and the proclisis-enclisis alternation. Readers are referred to van Riemsdijk's (1999a) Eurotyp edition, and to the (2006) Blackwell Compa.nion to Syntax (see especiallyvol. I) (but see also Chapter 3). Moreover, I have not included a detailed comparison between Romance and Greek, as such a comparison was not among my aims. Rather, I mainly focused on Greek, and used Romance for comparison only in minor cases. Finally, I assume that the reader is acquainted with Chomsky's papers from 2000 onwards (Chomsky 2000, 2001a,b, 2005, 2006), and I therefore introduce his theoretical notions with little if any explanation. Having discussed what this monograph is not about, let us move to what it is about This monograph aims at providing a principled and novel analysis for two interrelated phenomena in the morpho-syntax of Greek clitic pronouns, namely proclisis (2) and enclisis (3) respectively, and as a result, at providing an answer to the 'whaf, 'whY, 'where' and 'how' questions asked by Rizzi in (1) above: (2) Proclisis: when the verb is in the indicative or the subjunctive, the clitic pronoun precedes the verb, and nothing may intervene between it and the verb. (3) Enclisis: when the verb is in the imperative or the gerund, the clitic pronoun follows the verb, and nothing may intervene between it and the verb. Starting with proclisis, it is argued that the clitic pronoun is a topicalizer, namely a definite head that is optionally merged in the left periphery of the (direct or indirect) object DP of the clause rendering it a topic. v*/T (that is v* in T. following Gallego's
2
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
2006 analysis of phase sliding in Null Subject Languages) probes into the v*p/TP phase and finds the clitic pronoun with which it agrees in phi-features. The clitic moves to v·•"/T due to an EPP feature that comes along with it. When the clitic moves to the left edge ofv*/T, it incorporates into it, yieldingproclisis. This happens for two interrelated reasons. On the one hand, the clitic contains only a subset of features of those contained by v* IT. This allows incorporation, according to an incorporation law which claims that incorporation is allowed only if the features of the incorporee are not different from that of the incorporator (Roberts 2009). On the other hand, incorporation into v"'/T is allowed, even though v*/T is a minimal phase, because its edge is still accessible, due to the fact that v*/T has not checked all its features as yet A minimal phase is defined as a minimal lexical category that is a lexical phase (i.e. v, n, a, etc.) or that contains one. Being a phase, it also has an edge which remains accessible until all its features have been checked off. Then, it can be sent to Spell-Out ln enclisis the generalization is that restricted person agreement on T is associated with [+ imperative]/[+ gerund] paradigm and an inflectionaVV-related CM(odal) head. This means that in enclitic environments the verb must target the higher CM head in order to check the special verbal morphology. The clitic pronoun targets the v"'IT head, as in proclisis, however it does not incorporate into v*/T, because the edge of the phase is not accessible; the phase in this case includes the CM head, given that the latter is inflectional/verbal, part of the INFL domain of the clause. As a result, the edge will be transferred to the CM head, and that is where the clitic will be able to incorporate from the specifier of v*/T, yielding enclisis. The advantage of this analysis is that is can be generalized across constructions and across languages without great difficulty, since it is based on the general principles of (i) subset of features, and (ii) edge availability/ accessibility. This means that the host/edge and the features involved may be different allowing us in this way to capture clitic structures from a number of different languages, as well as different structures from within one and the same language. The structure of the remaining chapters is as follows: In Chapter 2 I present the main properties of Greek personal pronouns, following the criteria and the observations made by Kayne (1975) and Cardinaletti & Starke (1999). Their criteria are applied to the Greek pronouns, and it is shown that fuWtonic pronouns are strong pronouns while weak/atonic pronouns are clitic pronouns. The morpho-syntactic status of clitics is also discussed and it is argued that Greek clitics are words. Finally, the status of Greek clitics as determiners/definite heads is argued for. In Chapter 3 I present a number of distinct proposals that have been put forward in the literature with regard to the internal structure of clitics and its relation to cliticization. I show that most of these analyses are inadequate on both empirical and theoretical grounds, a fact which further supports my proposal that Greek clitics are DPs/ Ds that move from the object position, incorporating into their host ln Chapter 4 the derivation of proclisis is discussed. It is argued that clitics A-move as XPs/Xs to the specifier of v*IT into which they incorporate because on the one hand
Chapter 1. Introduction the edge is accessible while on the other they are minimal non-phases, i.e. they only have a subset of the features contained within the phase head. Incorporation has an effect on the outcome, because it satisfies the LF and PF properties of the clitics. It satisfies their PF properties by providing them with a PWd into which they can incorporate. It also satisfies their LF properties by providing them with a fiuniliarity/topic interpretation. Periphrastic constructions with the auxiliary echo ("have") are also discussed where it is argued that echo basically copies the v-features of the v* on the participle, which results in the clitic incorporating into echo. Indirect object constructions are also discussed, where I follow the analysis proposed by Anagnostopoulou (2003a), as well as doubling constructions, in which it is argued that the doubling clitic is an external determiner on the doubled DP that is probed by v*ITs phi features and the EPP feature. ln Chapter 5 I discuss the proclisis-enclisis alternation, as it appears in Greek and other languages that follow the so-called '(non-)finiteness' pattern. I attempt to define the level of finiteness which appears to be relevant to the alternation. Also, the V-movement theory is discussed (see Kayne 1991), and it is shown that movement across the cliticization site is neither a necessary nor a sufficient condition for enclisis. Instead, 1 claim that movement to the cliticization site along with resolution of proclisis-enclisis on the basis of finiteness at the clitic incorporation site are necessary conditions for the derivation of proclisis or enclisis. In Chapter 6 I discuss the role of finiteness, and in particular, of CM (which corresponds to Rizzi's 1997 CFin) in the proclisis-enclisis alternation. More specifically, I investigate the finiteness properties of a series of mood structures in Greek., and I formulate the following two generalizations: Generalization A: Absence of a syntactically active (Le. unvalued) person feature on the T/ Agr head (i.e. restriction on person agreement) correlates with enclisis, + imperative mood and V- to-CM movement (obligatory). Presence offull agreement on the T/Agr head correlates with proclisis, - imperative mood and V-to-T movement (obligatory). Generalization B: Proclisis correlates with non-restricted/full person agreement on T, while enclisis correlates with restricted person agreement on T. Non-restricted agreement is associated with the [-imperative] paradigm and a non-inflectional CM head (and, thus, with non-obligatory movement to C), while restricted agreement is associated with the [+ imperative] paradigm and an inflectional CM head (and, thus, with obligatory movement to C). In other words, for enclisis to obtain, the verb must move to CM, a V-related site, across the initial cliticization site (in this case v* /T), while the clitic moves to the edge of CM, the new cliticization site. At the end of the chapter I also discuss negated imperatives, which I claim are impossible in Greek due to the fact that the negation morpheme is merged in CM. I also discuss surrogate imperatives which I claim may have an imperative syntax since they realize a subset of the features of a true imperative form. Finally, I discuss free clitic ordering in enclisis in Greek., where I claim that in accusative genitive orders, the genitive pronoun is a weak pronoun rather than a clitic. Finally, Chapter 7 contains the conclusions of this monograph.
3
CHAPTER 2
The properties of Greek elitics and their structural analysis 2.1
Introduction
In this chapter I will present the main properties of Greek personal pronouns, following the criteria and the observations made in Kayne (1975) and Cardinalletti & Starke (1999). I will apply their criteria to the Greek pronouns and I will show that fulVtonic pronouns are strong pronouns while weak pronouns are clitic pronouns. I will also discuss the morpho-syntactic status of clitics as words, as well as their categorial and functional status as DPs.
2.2
Greek personal pronouns: Clitics vs. strong pronouns
2.2.1
Some basic properties of Greek personal pronouns
Standard Modern Greek 1 (henceforth, Greek) has two series of overt personal pronouns, namely strong and weak pronouns (also known as tonic vs. atonic or non-clitic vs. clitic pronouns; see Holton, Mackridge & Philippaki-Warburton 1997).2 As the names themselves suggest, one of the main (and also mostly widely known) differences between these two pronominal sets is that strong, although not weak, pronouns carry inherent lexical stress. According to traditional grammar weak pronouns, as opposed to strong ones, cannot stand alone in the sentence, but need to attach/cliticize to a phonological host in order to be felicitously pronounced (hence the name clitic). Greek is surely not the only European language that makes such a distinction in the personal pronoun paradigm, as similar parallel pronominal series are found in Romance, Germanic and Slavic languages (as well as in Semitic languages, or even in Chichewa) (see Cardinaletti & Starke 1999, van Riemsdijk 1999b and references therein for European languages, Ritter 1995 for Hebrew, and Bresnan & Mchombo 1987 for Chichewa). Nevertheless, it also differs from them in that, for example, it does not employ any other types of clitic category, such as auxiliary and complementizer clitics
1.
Standard Modern Greek refers to the official form of the language that is spoken in Athens
and a big part of Southern Mainland Greece (e.g. the Peloponnese) and which is taught in school 2. Besides strong and dttic pronouns Greek also has pro subject pronouns. I am going to discuss their properties in relation to the strong and weak series later on in this chapter.
6
Cities in Greek: A Minimalist Account ofProclisls and Enclisis found in Slavic languages, or pro- PP eli tics of the en/y and ne type, which are found in French, Italian, Barceloni among others (see Vas & Veselovska 1999 for details).3 Both strong and weak personal pronouns are fully marked for nominal inflectional features such as number (singular/plural), person (1st, 2nd, 3rd) and gender (masculine, feminine and neuter, but only in the third person), as well as for case (nominative, accusative and dative, which surfaces as genitive4 ).1he forms of the third person singular and plural are in essence demonstrative forms (which are declined as adjectival forms ending in -os, -i, -o) (similar to what we find e.g. in Romance languages like Spanish or French). These (main) properties are illustrated in the tollowing table, which contains the two pronominal series 5 (cf. Drachman 1997: 221. See also Vas & Veselovska 1999:915 and Holton, Mackridge and Philippaki-Warburton 1997): (1)
Table 1: Paradigm of the two series (strong vs. weak) of the personal pronoun in Greek
CASE
STRONG/ FULL PRONOUNS
nom gen
ego esi emena esena (ne) (ne)
ace
emena esena (ne) (ne)
aftos aftu (nu) afton (e)
afti afti (ni)s
afto aftu (nu)
em is em as
esis esas
afti afton (on)
aftes afton (on)
afta afton (on)
aftin (e)
afto
em as
esas
aftus
aftes
afta
ti
tes
ta
mas mas 1 pl
sas sas 2pl
tus
tus
tus
tus
tis/ (tes) ta
3pl masc
3pl fern
WEAK/ SHORT PRONOUNS
nom gen mu ace me PER 1 sg NUM GEND
tos su se 2sg
ton(e) 6
ti to tis tu ti(n)(e) to
3sg masc
3 sg fern
tu
3sg neut
3pl neut
3· Note that Greek also has so-called PF clitics, such as the definite determiner o, I, to, or the preverbal particles like ea, na, as, dhen, min. For more information see Holton, Mackridge & Phllippaki - Warburton ( 1997). 4 In Standard Greek. the genitive form has taken over the uses of the old dative, as opposed to Northern Greek dialects, in which accusative has taken over this role. 5· Horrocks (1997) actually claims that third person clitics are historically descendant from the ancient Greek deictic pronouns (c£ a(u)tos). The alternative forms with a final (e) are used in every day, non-careful speech, both with enclitics as well as with proclitics (in the latter case certain morpho-phonological conditions must be met).
6.
Chapter 2. The properties of Greek clltics and their structural analysis A few things should be noted here. First of all, as the highlighted parts within the cells7 illustrate, strong and weak/ clitic pronouns are morphologically related, in the sense that the weak forms are morphologically reduced in relation to the strong ones. Typically, and putting aside the additional endings within the parentheses which are added for metrical reasons only (see Drachman & Malikouti-Drachman 1999), strong pronouns are (at least) bisyllabic, as opposed to weak ones which are in most cases monosyllabic. Second, the nominative clitic pronouns, which are only possible in the third person singular and plural, are extremely rare, used with the demonstrative forms pu.(n) (meaning 'where is?'; pun < pu + 'n < pu ine (literally, 'where is') and na. (meaning 'there/here; depending on person) (see Holton, Mackridge & Philippaki-Warburton 1997, Joseph 1994 for both, and Christides 1990 for deictic na): (2) a.
na -tos! There-he.nom 'There he is'
b. pun -ta8? Where-them.nom.neut 'Where are they?'
Greek
Greek
Pun and na may also take accusative clitics, nominative full Noun Phrases (NPs) (i.e. pronouns and NPs accompanied by the definite article), as well as nominative elitics doubling full NPs9: (3) a.
Pun-tin(a)? Where- her.clacc 'Where is she?'
b. Na-me!
Greek
Greek
Here-me.clacc 'Here am I!'
7· Cells here refer to cells of a paradigm (that of personal pronouns). Note, however, that no theoretical implications with regard to the notion of paradigm are intended by thls discussion: paradigms are used only for informal expository reasons. Optionally, the initial It/ of the clltic may be voiced with or without subsequent assimilation of the final/ n/ ofpun, so that the whole [wh-word-ditic] duster would be pronounced as: pu(n) da?. Here, I will abstract away from such optional morpho-phonological processes (but see main text below for some discussion, and Revithiadou 2006 for sandhi (and other morpho-phonological) operations taking place within the host-ditic duster). 8.
9· Although not all combinations are possible (cf for example the ungrammaticality of*J~a mas!*na-sas, or pu(11)-mas!* pu(11)-sas, as against to the grammatical ones na-maste ('here we-are') or pu-saste ('where you-are')).
7
8
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls Pun(e)-/ na o Nikos?/! Where-/there the Nick..nom 'Where is Nick?/ There is Nick coming!'
Greek
d. Pun-ti i Maria ? /Na-ti i Maria! Greek Where-she.clnom the Maria.nom ?/there-she.clnom the Maria.nom! 'Where is Maria? I There is Maria!' Third, some cells share the same form, as e.g. the nominative and accusative forms of singular and plural, neuter, 3rd person strong and weak pronouns (afto vs. afto! afta vs. afta/ to vs. to/ ta vs. ta), a phenomenon which is quite widespread, and is typical for adjectives and nouns taking the -os, -i, -o endings. The same appears to apply for emena and esena (namely, the accusative and genitive/dative, 1st and 2nd person singular forms of the strong personal pronoun), as well as for emaslesas (the respective forms of the plural), mas/sas (namely, the corresponding clitic forms), or tus/tus/tus (genitive plural, 3rd person, masculine, feminine and neuter clitic pronouns). The default assumption for all these cases is that they are syncretic forms, i.e. forms that spell-out more than one cells in the paradigm (based on such features as case, or gender). Afl syncretism is widespread in pronominal paradigms across languages, this is not an unusual phenomenon (see e.g. Cysouw 2003 on possible pronominal paradigms on a cross-linguistic basis). On the other hand, syncretism does not appear to make the right predictions in all cases. To give one example, Tsakali (2004, 2006) has convincingly shown that the emenalesena forms are not syncretic for case; rather, they are accusative forms, since they are not grammatical with verbs or prepositions taking a genitive complement: (4) a.
Miluse *emena Was-talking.3sg *me.gen/acc.str
I-./ tu Petru. I-./ the Peter.gen
..esena I -./ tu Petru. b. Itan iper Was.3sg in-favor-of *you.gen/acc.str I-./ the Peter.gen 'He/ she was in favor of you/Peter: c.
Idhe -./emena I-./ ton Petro. Saw.3sg me.gen/acc.str I-./ the Peter-ace 'He/ she saw me/ Peter:
Greek Greek
Greek
On the other hand, the corresponding (plural) clitic forms, although homonymous, appear to be syncretic, given that they are equally acceptable with accusative and genitive taking predicates: (5)
-./mas
milise I -./mas idhe. Us.cl.acc/gen talked.3sg I us.cl.acc/gen saw.3sg 'He/she talked to us/ saw us:
Greek
Chapter 2. The properties of Greek clltics and their structural analysis
Although the actual facts are more complex than what is suggested here, 10 it remains a fact that in some cases we have syncretism while in others we don't (simply there is an empty cell in the paradigm). Accordingly, the table in (1) could be restructured as follows (the change in the order of cases as well as of the 3rd person forms has been made in order to facilitate the illustration of the syncretic cases, as well as those of unoccupied cells):
First of all, as Tsakali (2004, 2006) points out, the ungrammatical instances of strong genitive 1st and 2nd person pronouns are rendered grammatical if doubled by a clitic pronoun (so that 'mu milise emena' is ok, as opposed to '* milise emena' (recall that milao ('to talk to someone') takes a genitive complement); in fact only when an appropriate clitic is present is the structure acceptable). In these cases, her claim is that the doubled pronoun does not take case. Putting aside the problems her analysis seems to face, her observation is essentially correct. Furthermore, she also argues that the corresponding plural forms of the strong personal pronoun are accepted as complements of genitive taking predicates (a fact which suggests that these forms, namely emas/esas, are (case) syncretic). However, I do not agree with her judgements, since I find examples like 'milise emai ('talked.3sg. us.gen/acc') or 'xarise emas lefta' ('gave.3sg us.gen/acc money') equally ungrammatical (a doubling clitic or the use of a periphrastic PP would render these examples grammatical- cf. 'mas milise emas'; 'milise me emas' (us.cl.gen!acc talked.3sg us.strong.gen/acc; talked.3sg with us.strong.gen/acc) (the preposition me takes an accusative complement - cf. me ton Petro (with the Peter.acc))). Similar effects with those of emena/esena (or emas/esas) are also found with the non-augmented 3rd person genitive forms of the strong pronoun (both in the singular and in the plural): cf- * milisa aftis/aftu/afton vs. -1 milisa aftinis/aftunu/aftonon. Cliticization or the use of a PP-per.lphrasls renders the examples much more acceptable in most cases: 'tis milisa aftis'; 'milisa me aftin', but '??* tus milisa afton/ «]to non/ton pedhlon'. Furthermore, if the offending pronoun is focalized (i.e. moved to the left per.lphery of the clause (see Tsimpli 1995)), the sentence is acceptable: 'EMENA mllise' (note that ditics are not allowed with focalized constituents for independent reasons, while the PP-per.lphrasls sounds better than the focalized non-diticized strong pronoun). In addition, not all predicates taking a genitive complement are equally bad with non-cliticized genitive strong pronouns. For example, the verb dhino ('to give') sounds much better to my ear than e.g. milao ('to talk to') or the more archaic ipertero ('to be better than'), although in both cases the use of an appropriate ditic or a PP-per.lphrasls is much better: (i) ??Edhose emena ta lefta. Gave.2sg me.gen/acc.str the money.acc 'He/she gave the money to me.' (ii) *?Mllai emena/iperteri emena. Talks.3sg me.gen/acc.str/ is better than me. 'He/she talks to mel is better than me:
10.
9
10
Ol:tics in Greek: A Minimalist Account ofProdisis and Endisis (6) Table 2: Strong and weak personal pronouns in Greek (revised)
CASE
STRONG/ FULL PRONOUNS
nom ace
ego emena (ne) 0
gen
esi esena (ne) 0
aftos afton (e) aftu
afto
afti aftin
em is emas
esis esas
0
0
afton (on)
0 sas
ti
afti
afta aftes
aft us
(e) afti (ni)s
(nu)
WEAK/ SHORT PRONOUNS nom ace gen PER NUM GEND
0 mu me 1 sg
tos to ton( e) tu 3sg 3 sg masc neut
0 su se
2sg
ti 0 tis mas ti(n)(e) 3sg 1 pl fern
ta
tes tis/(tes)
tus
2pl
3pl masc
3pl 3pl neut fern
A fourth point. which has often been used as an argument in favor of the hypothesis that Greek 3rd person clitics are categorically determiners (see, among others, Drachman 1997; Anagnostopoulou 1999; Tsimpli & Stavrakaki 1999; Alexiadou 2001; Panagiotidis 2002; Alexiadou, Haegeman & Stavrou 2007) is also the fact that the biggest part of the third person clitic paradigm (including both dative/genitive and accusative forms) is homophonous with that of the definite article (only the genitive plural, and the nominative masculine/feminine singular and plural forms are realized differently). Note, that (partial or full) homonymy is also frequently found in many Romance languages, although in most cases it seems to be restricted to the accusative paradigm only (see e.g. Uriagereka 1995; Bleam 1999 for discussion). The definite article paradigm is given below: (7) Table 3: Greek definite article Singular
Masruline
Nominative Genitive Accusative
0
Feminine
Neuter to
tis ti(n)
tu
to(n) i
i
ta
ton tus
ton tis
ta
tu
to
Plural Nominative Genitive Accusative
ton
Chapter 2. The properties of Greek clltics and their structural analysis Moreover, as Drachman (1997) has pointed out, sub-parts/morphemes of the personal pronouns (especially the lstand the 2nd person) are also found in the present tense of the copula, as well as in media-passive (i.e. non-active) present forms. This is shown below (see also Drachman 1997: 223) 11 : (8) Table 4: Copula (present tense of'to be') media-passive/non-active form of'to be written'
lme ise lne lmaste is(a)ste lneTheyare
lam You are He/she/it is Weare You are
grafome grafese grafete grafomaste grafosaste/grafeste grafonte
I am written you are written he/she/it is written we are written you are written they are written
Although a detailed and successful decomposition of personal pronouns and the relevant verbal agreement endings into morphemes is not a trivial task at all (raising important issues such as the internal morpho-syntactic structure of these constituents (e.g. whether they involve the projection of multi-head constructions, or alternatively of single heads containing feature bundles (at least for the case of clitics, which do not seem to license specifiers- see discussion below))), the strong resemblance between these items lends support to the claim that they share (some) morpho-syntactic features. For example, all first person forms share the /m/ segment, second person ones the /s/ segment. while third person ones the /t/ segment. 12 As a result, it becomes quite
Note that the present tense of the copula verb seems to have similar endings with the medio-passive/non-active form (only 3rd person singular and plural are different). Moreover, in Greek the same passive morphology is used for passive, refiexive (both inherent and afto- reflexives, lnchoative/anticausative, and even active meanings (in the case of deponents)). For example,plen-ome ('to be washed'- cf active 'pleno' ('to wash')) has a refiexive or passive meaning, dex-ome ('to accept') has an active meaning, while kegh-ome has an lnchoative/anticausative meaning. See Alexiadou & Anagnostopoulou (2004) for this kind of syncretism in Greek and other languages, as well as Theophanopoulou-Kontou (2000).
11.
u. Here, I abstract away from the well-known debate about the status of third person in the grammar (see Nevins 2007 for a recent discussion, and references therein). I will come back to this later on. Also note that the analysis of !ml, Is/ and /t/ as person morphemes is further corroborated by the fact that the same morphemes are found in possessive pronouns, which also encode person. Note, on the other hand. that the active forms of the verb (cf the verb grafo ('to write'): graf-o, graf-is, graf-l, graf-ume, graf-ete, graf-un) do not seem to contain the same morphology (in fact the presence of It/ in the second person is not expected. given the corresponding non-active and pronominal forms). Nevertheless, given the striking resemblance between the personal pronoun forms and the medio-passive endings, as well as the fact that the
11
12
Cities in Greek: A Minimalist Account ofProdisis and Endisis
plausible to assume that these segments are morphological exponents of a person feature. Judgements for other features (e.g. number, gender or case) are much trickier, given the fusional character of Greek (see, though, Drachman 1997 for a morphological decomposition of these forms into all the relevant features).
2.2.2
Strong vs. clitic/weak pronouns: The issue of deficiency
The pattition ofpronou.ns into types: Kayne (1975), Cardinaletti & Starke (1999) Besides these fairly obvious differences, strong and weak personal pronouns also differ in other important respects, namely syntactically, semantically, phonologically, pragmatically and possibly also structurally (depending on whether their differences imply a structural difference or not). The general observation is that weak pronouns are more restricted or more deficient (with regard to their properties) in relation to strong ones. This is a fairly well established phenomenon, and it has been extensively discussed in the generative literature. Kayne ( 1975) and Cardinaletti & Starke (1999) are two of the most widely cited works in this respect Kayne (1975) has shown that French distinguishes between two series of subject and object personal pronouns, namely strong pronouns and clitics, based on a number of distributional, syntactic, and morphophonological criteria, such as coordination, modification, focalization, and distribution. His main claim was that strong pronouns pattern like full NPs in all these respects, as opposed to clitics, which appear to have a special status. 13 Since the publication of Kayne's seminal work on French pronouns, many researchers have observed that other Romance languages also make a distinction between strong and weak personal pronouns, with similar (if not identical) properties to those observed in French (e.g. Spanish, Italian, and Romanian). Furthermore, systematic research into the properties of pronouns in Germanic languages in the late '80's and early '90's (see e.g. Holmberg 1986; Cardinaletti & Starke 1996, 1999 and references therein for more references on Germanic; Cardinaletti 1999) has revealed that they also distinguish between (at least) two classes of pronouns, although there also appear to be some differences in relation to Romance. For example, in some Germanic
2.2.2.1
same segments mark person across paradigms, any other hypothesis looks less plausible and less interesting. Although the distinct active and medio-passive endings could probably be amenable to a diachronic explanation (possibly involving incorporation of nominative vs. accusative forms of the personal pronoun- cf the -o ending of 1st singular vs. ego; and see Giv6n 1976, Roberts & Roussou 2003: Chapter 4 for an account of how personal endings arise in languages), the question remains how one can capture this difference in synchronic terms. A morphological approach (e.g. the presence of an active feature may regulate the exponence of the person feature) would probably work, however this issue is beyond the scope of this discussion. 13. In fuct, Kayne (1975, 1983) argued that subject clitics are in reality strong pronouns that undergo P(honological) F(orm) cliticlzation, as opposed to object clitics which are true syntactic clitics. In this way he was able to account for the XP vs. X0 properties of subject and object clitics respectively. See also Cardinaletti & Starke (1996) for a critique of Kayne's approach.
Chapter 2. The properties of Greek clltics and their structural analysis
languages (e.g. Mainland Scandinavian languages, Standard German, and English) two distinct series of pronouns have been distinguished (namely strong and weak), however their weak series seems to share properties with both strong and weak pronouns in languages like French (cf. e.g. the properties of German es, according to Cardinaletti & Starke 1996). In addition, some Germanic languages, like Colloquial Standard German (see e.g. Grohmann 1997), West Flemish (see Haegeman 1993), and Tirolese German (see Cardinaletti & Starke 1996) have been shown to have a third type of pronoun (besides the two 'standard' strong/weak types), which has such clitic properties that make it look quite similar to weak pronouns in Romance languages. Based on these properties of personal pronouns in Germanic languages (as well as on the properties of pronouns in Romance and other languages), Cardinaletti & Starke (1996, 1999), have argued that cross-linguistically pronouns can be classified in three distinct sub-types, namely strong, weak and clitic pronouns, which form a gradation of structural deficiency (with morphological syntactic, phonological and semantic repercussions), in the sense that clitic pronouns are more deficient than weak pronouns which are more deficient than strong pronouns. Not all languages display all three types (e.g. Standard German appears to have only strong and weak pronouns, while Greek has only strong and clitic (object) pronouns- although for Greek see also discussion below and Chapter 6), while on the other hand distinct types are not necessarily marked differently. The latter fact is quite important, given that morphology had been taken up to that point as one of the most salient criteria for distinguishing between distinct pronoun types (cf. Cardinaletti 1999 on the delay in the discovery of distinct pronoun types in Germanic languages, especially with regards to the strong-weak distinction). Despite all this, it seems that each type has a predictable behavior which is shared up to a large extent across all languages. I will come back to their theory when I will discuss the issue of the internal structure of clitic pronouns as opposed to strong pronouns in Greek. 2.2.2.2
Partition criteria applied to Greek
Greek strong and weak pronouns appear to display the properties of strong and clitic pronouns respectively, following the tripartite typology of Cardinalettl & Starke (1999 ), or those of strong pronouns and syntactic X0 clitics in Kayne's (1975) analysis. In particular, the following criteria have been proposed in order to distinguish between different types of pronouns: a. Morphological criteria (morphological deficiency) b. Syntactic criteria (distribution, coordination, modification, X0 vs. XP properties) c. Semantic criteria (animacy/human, referential properties, impersonal, expletive and non-referential uses) d. (Morpho)phonological criteria (sandhi rules, prosodic restructuring, inherent stress) A. Coordination (syntactic) and human reforence (semantic) First of all, strong and elitic pronouns fall into the two distinct pronoun classes (class 1 and class 2) proposed by Cardinalettl & Starke (1999) on the basis of coordination and
13
14
Cities in Greek: A Minimalist Account ofProclisis and Enclisis
human reference. In particular, Cardinaletti & Starke (1999) have pointed out that in many languages pronouns appear to be divided into two distinct classes (even if they are morphologically non-distinct): class 1 pronouns may be coordinated but can only have human referents, as opposed to class 2 pronouns which cannot be coordinated but which may refer to both human and non-human entities. This distinction becomes fully transparent when we look at third person pronouns, and is able to distinguish between strong vs. weak/elitic pronouns 14: (9) a.
Quiero a ella y a estas dos aqui. Spanish (Spain) Want.lsg her and those two here Ella: class 1 pronoun (strong); coordination: ..J; +human reference only
Spanish (Spain) b. *Pedro la y estas dos quiere. Peter her and those two wants.3sg La: class 2 pronoun (clitic); coordination:*;± human reference 'Peter wants her and those two here? Hu gadol. (hu = Dani/ shulxan =table) Hebrew (Israel) He big.masc 'He is big? Hu: class 2 pronoun (weak); coordination:-;± human reference)
d. Hu vezeh gdolim. (hu = Dani) Hebrew (Israel) He and-this-one.masc big.pl.masc 'He and this one are big: Hu: class 1 pronoun (strong); coordination: ..J; +human reference only) e. *II et celui de Jean sont beaux. French (France) He and the.one of John are pretty Il: class 2 pronoun (weak); coordination: *; ± human reference f.
Lui et
celui de Jean sont beaux. French (France) He and the.one of John are pretty Lui: class 1 pronoun (strong); coordination: ..J; +human reference (examples taken from Cardinaletti & Starke 1999: 149 (their (lOb) and (lOd))
Greek clitic pronouns certainly fall into class 2 (no coordination;± human). (3rd person) strong pronouns, on the other hand, may be coordinated but do not appear to be restricted in terms of reference. This is not peculiar, since 3rd person pronouns are really demonstratives, 15 which typically do not exhibit a non-humanness restriction on a 14. I.e. class 2 pronouns may be either weak or ditk pronouns, whereas class 1 pronouns are exclusively strong pronouns. 15. Demonstratives may be used either as pronouns, or as adjectives within a noun phrase. In both cases they may have a deictic or an anaphoric usage (see also Alexiadou et al. 2007). Typically, in the former case they pick up their referent through the extra-linguistic context, while in the latter the referent is provided by previous linguistic context. In their dektic use they
Chapter 2. The properties of Greek clltics and their structural analysis
cross-linguistic level (cf. Cardinaletti & Starke 1999 for a discussion). For this reason, they seem to share properties from both classes. Cardinaletti & Starke argue that demonstratives do not fall into the same paradigm with (true) personal pronouns. Although, strictly speaking, this is correct, here I will use a demonstrative pronoun in its personal pronoun use in order to illustrate its difference from clitic pronouns in terms of coordination: (10) a.
Idha afton ke aftus eki. Greek Saw.lsg him.str and them.str there 'I saw him and those over there: Afton: demonstrative pronoun (strong); coordination: ...J; ±human reference
b. *I Maria ton ke aftus eki idhe. The Mary.nom him.cl and them.str there saw.3sg Greek 'Maria saw them and those over there: Ton: class 2 pronoun (clitic); coordination:*;± human reference Of course, 1st and 2nd person strong pronouns can only have a[+ human] reading (for obvious reasons - cf. the fact that they realize the speaker and the addressee respectively (putting the issue of number aside for the moment)), and they can also be coordinated. The corresponding clitic pronouns also have (only) the [+human/animate] reading but cannot be coordinated: (11)
a.
Alexandhra
kerase emas
ke
tus ghitones
tis. Greek The Alexandra.nom treated.3sg us.str and the neighbours.acc her.cl
b. *I Alexandhra mas ke tus ghitones tis kerase. The Alexandra.nom us.cl and the neighbours.acc her.cl treated.3sg Greek ':Alexandra treated us and her neighbours: Mas: class 2 pronoun (clitic); coordination:*;± human reference B. Morphological deficiency (morphological) A second major difference between full and clitic pronouns is that they are morphologically distinct In particular, clitics and strong pronouns are morphologically related in an asymmetric way, in the sense that clitics typically are reduced morphological variants of the corresponding strong forms. This is immediately obvious if one looks at the table of personal pronouns given in (1). Here, I give an example from the 2nd person singular: (12) Esas dhen sas ksero kala. You.pl.str not you.pl.cl know.lsg well 'You, I don't know well:
Greek
can be accompanied by so-called reinforcers, like edho ('here') and eki ('there'). Finally, in their anaphoric usage as pronouns they typically have an emphatic reading, as is the case for all other strong pronouns in languages with (subjecUobject) ditics and null pro subjects.
15
16
Cities in Greek: A Minimalist Account ofProdisis and Endisis
This is a so-called Clitic Left Dislocation (CLLD) construction (see Anagnostopoulou 1994 for Greek, and more generally Cinque 1990), which involves a strong personal pronoun, presumably adjoined to IP, and a clitic pronoun within the IP constituent (note that this is only one possible way of analyzing this construction - see Iatridou 1991 for a dislocation analysis of CLLD, and Cecchetto 1999, 2000; Belletti 200 5 for an alternative, movement analysis for Italian). 16 It is fairly obvious that the clitic is morphologically reduced in comparison with its strong counterpart Note that the same asymmetric relationship between strong and clitic pronouns is quite common cross-linguistically. For example, in French moi - me (to me - me) is such a minimal pair (see also Kayne 2000 for an analysis which takes moi to be hi-morphemic as opposed to me which is mono-morphemic), and the same applies to the Spanish pair (a) nosotros- nos (to us- us). Similar remarks have also been made for the morphological relationship between strong and weak pronouns, i.e. weak pronouns may also be morphologicallyreduced with regard to strong pronouns (cf e.g. a loro -loro (to them- them) in Italian). Finally, a clitic may be more reduced than a weak pronoun, as e.g. 'm - ihm (him-him) in Colloquial Standard German. These two facts are illustrated below: (13) a.
Non diro mai tutto a loro. Not will-say.1sg. never everything to them
b. Non diro mai loro tutto. Not will-say.1sg never them.weak everything 'I will never say everything to them: c.
Italian (Italy) Italian (Italy)
DaB Maria ihm gestem das Buch gegeben hat. That Maria him.weak yesterday the book.acc given has
d. DaB'm Maria gestern das Buch gegeben hat That him.cl Maria yesterday the book.acc given has 'That Mary has given him the book:
German German
In this respect, one can find examples of strong-weak-clitic triplets, where each pronoun type is morphologically (more) reduced with respect to the other one (so that strong is less reduced than weak which is less reduced than clitic). This is, tor example, the case of a loro - loro - gli in Standard Italian. As a final point. we should add that morphological asymmetry is only a tendency: although elitics tend to be more reduced than weak and strong pronouns, and although weak pronouns tend to be more reduced than strong ones, there are many cases where this is not true, in that some or all pronouns within a tripartition/bipartition may be identical in form. This is especially quite common between strong and weak pronouns, although itis less common for clitic-strong 16. Adam Ledgeway asks why one would want to argue that esas in (12) is adjoined to IP, rather than in the left periphery. As I also point out in the text, adjunction to IP is one plausible account of the position of the Clitic Left Dislocated pronoun. Movement to a Topic position within the Left Periphery is another possible analysis. Which one is to be chosen will depend on further evidence, however it is not important for the point being made here.
Chapter 2. The properties of Greek clltics and their structural analysis
pairs. Examples include the French strong-clitic pair nous - nous, or the German strong-weak pair ihn - ih" (for more examples see Cardinaletti & Starke 1996, 1999). C. Modification (syntactic) A third difference is that strong pronouns can be modified, whereas clitics (and weak pronouns in languages that have them) cannot This is illustrated in the following example: ( 14) a.
Skeftete [esena mono] I [emas tus dhio] I [aftin me to kokino forema] . Thinks.3sg[onlyyou.str] I [us.str the two] I [her.str with the red dress]
b. * [Se mono] I [mas tus dhio] I [tin me to k.okino forema] skeftete. [You.cl only] I [us.cl the two] I [her.cl with the red dress] thinks.3sg 'He/ she only thinks about you I the two of us I the one with the red dress: Greek In this respect, strong pronouns pattern with full noun phrases: (15) Vlepi [ti Theopula mono]. Sees.3sg [the Theopula only] 'He/she sees only Theopula:
Greek
D. Distribution (syntactic) Moreover, clitic pronouns appear to have a quite restricted distribution, since they may only appear in particular positions within the clause, whereas strong pronouns and full noun phrases are much freer in this respect In particular, clitics (and clusters thereof) must immediately precede or tallow the verb (or the auxiliary, if the sentence contains an auxiliary), largely depending on the finiteness properties of the latter: they precede indicatives and subjunctives and follow imperatives and gerunds. Nothing may intervene between the clitic and its host: (16) a.
Karina ton ithele san treli. The Korina.nom him.cl wanted.3sg like crazy.nom 'Karina wanted him like crazY-
b. Mu to iche pi to koritsi su. Me.cl itcl. had.3sg. said.inf. the girl.acc your.cl 'Your girl had said it to me: c.
0 Bilis (*to) tha (.V to) (*tara) The Bilis.nom (it.cl) FUT (itcl) (now) (*to) (,Y tara) to Mitsaki. dhiri beat.3sg. (it.cl) (now) the Mitsaki.acc 'Bilis will beat Mitsaki noW.
d. (*To) kane (*tara) (,Y to) (tara) to pastitsio Eli! (It.cl) do.2sg.imp. (now) it.cl now the pastitsio.acc Eli.voc 'Do it now Eli the pastitsio!'
Greek
Greek
Greek
Greek
17
18
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
(Strong) object pronouns and full noun phrases do not need to be adjacent to the verb, and moreover they cannot appear in the same position as proclitics: (17) a.
To pire o Kostas. ltd took.3sg the Kostas.nom
Greek
'Kostas took it' b. Pire o Kostas to kinito. Took.3sg the Kostas.nom the mobile.acc
Greek
c. *To kin ito pire o Kostas. The mobile.acc took.3sg the Kostas. 'Kostas took the mobile phone:
Greek
Given that clitics can only immediately precede or follow their host, it follows quite straightforwardly that they also cannot appear in the theta-position (external merge position), as well as in peripheral positions, such as clitic left dislocated positions, and left and right dislocated positions (i.e. hanging topic positions) (see Anagnostopoulou 1997, 2005a, Espaftol-Echevarria & Ralli 2000 for some properties of clitic dislocated and hanging topic positions). Full pronouns and noun phrases can appear in these positions without any problems. However, note that a clitic is required in conjunction with CLID and Left Hanging Topics. Topicalized indefinite DPs do not have such a requirement (see also Giannakidou & Merchant 1997 for indefinite topicalization in Greek): (18) a.
[Tin Katina/emena/*tne] i ghonis tu Marku Greek [The Katina.acc/me.str/me.cl] the parents.nom the Mark.gen *(tin/me) proskalesan sto parti gia ta sarandaxronia ghamu tus. (her.cVme.cl) invited.3pl to-the party for the fourty years marriage.gen their.cl 'Mark's parents have invited Katinalme to their fourty years of marriage party. [Clitic Left Dislocation]
b.
*(tin/ton) [I Katinalaftos/*tos], i Marina [The Katina.nomlhe.str.he.cl] the Marina.nom (her.cl/him.cl kales e. invited.3sg 'Katina/he, Marina invited her/him: Greek [Hanging Topic Left dislocation]
c.
Marina ..(tin/ ton) kalese, tin The Marina.nom (her.cVhim.cl) invited.3sg the Kalina /afton /*ton. Katina.acc/him.str/him.cl 'Marina invited, Kalina/him:
Greek [Right dislocation]
Chapter 2. The properties of Greek clltics and their structural analysis
d. Fraules/*tis i Adonia dhen pulai. Greek Strawberries.acc/them.cl the Adonia.nom not sells.3sg [Indefinite topicalization] e.
Tis milise o Markoras tis Marinas/aftis/*tis. Greek Her.cl talked.3sg the Markoras.nom the Marina.dat/her.dat.str/her.cl 'Markoras talked to Marina/her: [Clitic doubling]
Clitics cannot focalize to the left periphery of the sentence, as opposed to strong pronouns and full nouns phrases (definite and indefinite ones) 17: (19) Tis VALITSES sas / APTES /*TES parte! The BAGGAGE.acc your.cVTHEM.str/THEM.cl take.2pl.imp. 'Take your LUGGAGE/THEM!'lS
Greek
A further difference between clitics and strong pronouns/noun phrases is also the fact that the former, but not the latter, cannot be used in isolation (for example, as answers to (narrow focus) questions) 19: (20) Pjus zografise i Flora? -Emas/*mas/tus filus tis. Greek Who.acc painted.3sg the Flora.nom?- Us.str/us.cVthe friends.acc her.cl 'Who did Flora paint?'- Us/Her friends:
17· Note that typically ditics are not compatible with a focalized pronoun/noun phrase, in that they cannot double the latter (see Tsimpli 1995): (i)
TON JANI ("ton) xastukise i mana mu. THE JOHN.acc (him.d) slapped.3sg the mother.nom me.geiLd 'It was JOHN that my mother slapped.'
Greek
18. Note that strong pronouns and full noun phrases may also receive (contrastive) focus in situ (see Tsimpli 1995), and, as expected, ditics are equally ungrammatical in this position: (i)
Parte tis VALITSES sas/APTES/..TES.
Greek
19. Cardinaletti & Starke (1999) take isolation to be a peripheral position as well, possibly a case of dislocatioiL Here, I do not intend to make any comments on the structural nature of isolated constituents, although it seelllS fuirly dear to me that at least answers to narrow focus questions must involve some form ofPF deletion, as shown by the existence of question-answer pairs. What matters is that ditic pronouns cannot be uttered on their own, contrary to other word constituents. This could be related to the inherent incompatibility of ditics with focus (given that these constructions typically involve focus), although the impossibility of ditics in non-focalized peripheral (and non-peripheral) positions makes this assumption much less plausible. See also Cardinaletti & Starke ( 1999) and discussion in main text below.
19
20
Cities in Greek: A Minimalist Account ofProdisis and Endisis
E. Setnantic properties of eli tics (semantic) Turning now to their semantic properties, clitic pronouns can be referential or nonreferential. In their referential use 20 they are typically anaphoric, in the sense that they pick up their reterent via a prominent antecedent, which is provided by the previous immediate linguistic context (discourse). 21 However, quite rarely they may be also used deictically (or, more appropriately, ostensively), in which case they may pick up their referent by ostension22 (although pointing is not a necessary condition if the antecedentis salient enough in the discourse): (21) a.
Chtes sinandisa ti Merii. Yesterday met.lsg the Mary.acc. Tisi milisa jia ti mitera tis. Her.cl talked-to.3sg for the mother.acc her.cl 'Yesterday, I met Mary. I talked to her about her mother:
Greek
b. Kala, dhen vlepis oti tui milao? (accompanied by pointing) Well, not see.2sg that him.cl.dat talk.lsg Greek 'Can't you see that I am talking to him?' c. Na min toni Subj not him.cl ksanadho! (uttered immediately after a certain man has left) again-see.lsg 'I would not like to see him again:
Greek
10. Or, more precisely, their co- referential use, given that (ditic) pronouns refer by sharing the same referential index with a prominent/salient antecedent (either via simple co-reference, or via syntactic/semantic binding- cf the semantic distinction between free and bound pronouns (see Helm & Kratzer 1998 for an overvlew) ).
Prominence is a necessary condition for the antecedent of (referential) ditic pronoWlS (see also Helm 1982 for the Prominence Condition which pronominal definite NPs must obey). This is, for example, obvious from the fact that they cannot be accommodated: in '0 Nikos d.hiavase ena vivlio ghia ton Perikll. ke tu eghrapse ena gramma ('The Nikos.nom read.3sg a book. ace for the Pericles. ace, and him.d wrote.3sg a lette.t:acc'II 'Nick read a book about Pericles, and wrote him a letter'), tu can only refer to Pericles, the most salient antecedent in the discourse, and not to the author of the book about Pericles. Finally, note that doubling is not compatlble with a novel definite reading (see Anagnostopoulou 1999 for extensive discussion). 22. For the difference between deixis (which involves the use of linguistic expressions/categories that relate what is talked about to a spatio-temporal centre) and ostension (which involves pointing, i.e. expressions that direct the addressee to a particular antecedent) see Lyons (1999). Note that as Helm & Kratzer (1998) correctly point out, anaphoric and deictic uses do not seem to involve an actual distinction, given that in both cases a (referential use of a) pronoun picks up its referent via correfence with a prominent antecedent. 11.
Chapter 2. The properties of Greek clltics and their structural analysis
Clitics can also have a number of non-referential uses. The most typical case is when they are variables bound by a non-referential antecedent such as a quantifier: (22) Kathe anthropos pistevi oti ton aghapane. Every man.nom thinks.3sg that him.cllove.3pl 'Every man thinks that they love him: [For every x, x a man, x thinks that they love x]
Greek
Since variable binding is contingent on syntactidsemantic binding (Reinhart 1983), clitic pronouns may receive a bound variable interpretation even when their antecedent is referential (e.g. a proper noun), as long as the latter binds it This is illustrated below, where in the second elided conjunct tu (a possessive clitic) can receive a sloppy identity reading according to which rn is bound by (the index of) Kostas23 : (23) 0 Janis pire tilefono ti jineka tu ke o Kostas The John.nom took.3sg phone the woman.acc his.cl and the Kostas.nom episis. also 'John called his wife, and also did John: Greek = Janis 1 called his 1 wife, and Kostas 2 called his 2 wife, too. (sloppy identity) = Janis 1 called his 1 wife, and Kostas 2 called his 1 wife, too. (strict identity) 3rd person clitics may also be used non-referentially in idiomatic constructions, which are very productive in Greek (see Bibis 2002 for details): (24) Mu tin espase. Me.cl her.cl broke.3sg 'He/she really got on my nerves/annoyed me:
Greek
13. Possessive pronouns involve ditic genitive pronouns, which are homophonous with those used in the verbal domain in dative and (other) applicative constructions (e.g. benefactives, malefactives, and non-argumental datives). However, there is evidence that these are not the same underlying element, since in Northern Greek dialects, which use accusative instead of the dative, only genitive ditics/pronouns are possible with nominals. Moreover, as I show in Chapter 6, there is additional evidence that these are not ditics, but weak pronouns. If this ls true, then it would be more appropriate to use an unambiguous example, such as the following:
(i)
ton parakoluthi i astinomia, o 0 Janis nomize oti The John.nom thought.3sg that him.d watch.3sg the police.nom, the Kostas episis. Greek Kostas.nom also = John 1 thought that he 1 was being watched by the police, and Kostas 2 thought that he 2 was being watched by the police, too. (sloppy identity) = John 1 thought that he 1 was being watched by the police, and Kostas 2 thought that he 1 was being watched by the police, too. (strict identity)
:n
22
Cities in Greek: A Minimalist Account ofProdisls and Endisis This example is ambiguous between a literal and an idiomatic interpretation. In the latter case the 3rd person singular feminine elitic does not refer to anything; rather, it forms an idiom along with the verb. lf the clitic or the verb changes, a literal and referential interpretation arises 24 : (25) a.
Mu tis/to espase. Me.cl them.fem.pl.cVitcl broke.3sg 'He/she broke them/it on me:
b. Mu ti dhanise. Me.cl her.cllent3sg 'He/ she lent it to me:
Greek
Greek
1st and 2nd person clitics cannot be used non-referentially in idioms. However, they may be non-referential when used as bound variables. Kratzer (1998, 2008) calls these pronouns zero pronouns, because their phi-features are not semantically interpretable; rather, they are added in the post-syntactic component for PF-interpretability reasons.25 .M a result, these pronouns may only involve an indexical feature in the syntax, and nothing else. An example of a 1st person variable reading is given below, where the pronoun mas is bound by more than one antecedent (cf. Partee 1989 for this test, and Rullmann 2004 for variable readings of 1st and 2nd person pronouns in English): (26) Kathe jineka pu ghnorizo nomizi oti mas zilevun. Greek Every woman-nom that meetlsg thinks.3sg that us.clacc are-jealous-of.3pl = Every woman I meet thinks that they are jealous of us ( = her and me) = Every woman I meet thinks that they are jealous of us (= me and someone else) A further example with a possible sloppy identity reading of a 2nd person pronoun is the tollowing one: (27)
[Esi] 1 nomizis oti i Roksani [se] 1 e:xi. erotefthi [You] think.2sg that the Roxanne.nom you.cl has.2sg fhllen-in-love-with.3sg
14. Note that further restrictions apply: the dative referential ditic must be present in order for the non-literal reading of the accusative ditic to go through: 'tin espase' ('her.d broke.3sg') can have either a literal or a metaphorical reading depending on the animacy of the clitic referent, but crucially cannot be interpreted idiomatically. 15. Not all bound variables are zero pronouns, though, in llght of such examples as
(i) 0 n1y you remember our first appointment (taken from Kratzer 2008: 82 (her 85)) which could be paraphrased as: 'only you are an x such that x remembers the first appointment ofx and me: In this particular case 'our' must start its life in the derivation with a 1st person feature, given that no antecedent ls present in the structure that could (locally) bind the zero pronoun and 'transmit' its person feature to it. This ls, however, orthogonal to our discussion, given that what interests me here ls that 1st and 2nd person ditics can be variables bound by a non-referential expression, and not the exact derivation of such constructions.
Chapter 2. The properties of Greek clltics and their structural analysis
ala to idio nomizi ke [o Tasos]z Greek but the same thinks.3sg and the Tasos.nom = 'You1 think that Roxanne has fallen in love with youl' but Tasos2 thinks that Roxanne has fallen in love with you 1' (strict identity) = 'You1 think that Roxanne has fallen in love with youl' but Tasos2 thinks that Roxanne has fallen in love with himz' (sloppy identity) Although bound variable readings are possible with 1st and 2nd person elitics (provided all conditions on syntactic binding have been met), these readings are less frequent than in the case of 3rd person eli tics, a fact which is not surprising, since 1st and 2nd person clitics (and pronouns or expressions more generally) tend to be interpreted referentially and deictically, in the sense that they point/refer to an individual that satisfies the role of the speaker or hearer in the extra-linguistic context. 26 Accordingly, it is far less common to find inherently deictic elements being interpreted as non-referential. This brings forward the issue of the internal structure of pronouns and the relation of this structure to their syntactic and semantic properties. For example, in the case of the 1st person clitic pronouns, do these involve exactly the same structure and/ or features in their distinct uses/interpretations? Or, is what we essentially see a single (elsewhere) form spelling out distinct features/structures? And if the latter is the case, would we need to postulate distinct 1st person clitics (in terms of feature bundles or structures) in the Lexicon, or could we simply assume that these are built in the syntax by means of feature combination? I will come back to these issues later on.
F.
Semantic properties of strong object pronouns (semantic)
When compared with clitic pronouns, strong (personal) pronouns seem to have somewhat different semantic properties, and certainly different uses up to some degree (which is definitely related to the 'choice issue' discussed in Montalbetti 1984, Lujan 1986, and Cardinalettl & Starke 1999 on more general grounds). Starting with object strong pronouns, these typically get a referential reading, and are mainly used for emphasis, contradiction/antithesis to a previous salient referent, and more generally as the non-presupposed part of the clause. They can be used deictically (this is trivial for 3rd person, given that the demonstrative form is used), or they can have anaphoric uses (in the sense that they refer back to a previous linguistic antecedent (not necessarily overt), however in any case their antecedent must be prominent in the discourse, on a par with clitic pronounsP Usually. they are focused (e.g. emphatic/contrastive focus, 16. In other words, 1st and 2nd person pronouns refer to an individual that is assigned the role ofthe speaker and addressee respectively. Accordingly, the referent of a 1st personal pronoun e.g. may differ according to the (spatio-temporal) context, but the speaker role remains constant.
27. This follows from the fact that one cannot say for example: # Chtes pigha sena parti
('yesterday went.lsg to-a party'). ldha aftus na chorevun (saw.1sgthem subj dance.3pl) (note that here aftus is not (although it can be) used deictically). Obviously, 1st and 2nd person pronouns would be felicitous in such contexts, however given the fact that they are inherently indexical/
23
24
Cities in Greek: A Minimalist Account ofProdisis and Endisis narrow/new information focus, or nuclear stress if they are the most deeply embedded constituent in the non-presupposed part/phrase of the clause), although this is notessential and they may also appear in positions which do not receive any kind of focus. 28 (28) a.
Pjon ithele? Who.acc wanted.3sg ~ Esena ithele/# Se ithele. You.str wanted.3sgl you.cl wanted.3sg
Greek
[narrow focus]
~ EMAS aghapai,
b.
oxi esena/ # Mas/MAS aghapai, oxi esena. 29 us.str loves.3sg, not you.str/us.cl/us.clloves.3sg, not you.str 'It is us that she loves, not you~ Greek [contrastive focus]
c.
Athina theli emena ghia to rolo tis Julia. The Athina.nom wants.3sg me.str for the role.acc the Julia.gen ~thina wants me for the part of Julia: Greek [flat intonation]
d. To ksero oti o Savas 1 ine kalos dhaskalos, ltd know.lsg that the Savas.nom is.3sg good teacher
deictic, their 'antecedent' is discourse prominent in a trivial way. Similar remarks apply to deictic uses of third person pronouns, since ostension makes a referent salient in the discourse. 18. This seems to be the case also for subject strong pronouns. Similar effects have been observed in other Null Subject Languages, at least for subject pronouns (see e.g. Rigau 1986 for Spanish and Catalan, and references therein). In reality, what appears to happen in these cases is that the pronoun receives some kind of weak emphasis. For example, lfl say: 0 skinothetls dhlalekse EMENA gia to rolo ('the director chose. 3sg ME.str for the part'), the most salient reading is that of contrastive focus, namely that he chose me and no-one else (or me, instead of you, her, them, etc., depending on context) for the particular part. On the other hand, the same sentence uttered without focus on emena (namely: '0 skinothetis dhialekse emena gia to rolo') would probably be interpreted as follows: the director chose me for the part, you for the dresses, him for... , etc. In particul~ in such a construction the pronoun seems to have a distinctive interpretation (see Rigau 1986), in that is weaker than contrastive (or more generally identificational) focus: it does not imply that no one else was chosen for the part; rather, it implies that individuals that are related to me within the particular discourse context will do something else/were chosen for something else. Accordingly, it might be the case that the director has actually chosen a second person for the same part (perhaps, in case I get sick, or something like that).
19. Note that contrastive stress is possible also with ditics, but only in corrective focus constructions (i.e. in order for the sentence in (i) below to be felicitous, the immediately previous context must involve an utterance which contains the 'su milise' subpart, namely the corrected portion of speech):
(i)
milise, oxi su milise. Me.d talked-to.3sg, not you.d talked-to.3sg 'He/she talked to ME, not to you:
MU
Greek
Chapter 2. The properties of Greek clltics and their structural analysis
ala egho dhen nomizo oti tha ithela afton1 jia but I.nom not think.lsg that fut wantlsg him.str for dhaskalo tu pedhiu mu. teacher the child.gen me.cl.gen 'I know that Savas is a good teacher, but I don't think Ia want him as a Greek teacher for my child: [anaphoric use; contrastive focus] e.
f.
Tsipi 1 mu pe oti ghustara aftin5. The Tsipi.nom me.cl told.3sg that liked.3sg her.str. 'Tsipi told me that I liked her:
Greek [focused)
Ego imun eki ke milusa me ton Kosta. I.nom was.lsg there and was-talking.lsg with the Kosta.acc Otan i Dhanai idhe emena, mu edhose ena fill. When the Dhanai-nom saw.3sg me.str, me.cl gave.3sg a kiss.acc 'I was there talking to Kostas. When Dhanai saw me she came forward and gave me a kiss: Greek [nuclear stress]
g. Sigha pu tha dhulepsete esis afton/emena. Greek Silently that fut work.2pl you.str.nom him.str/me.str 'You will not cheat on him/me: [stressed/focused; deictic use; emphasis] h.
lei oti tha ferun afti1 sto parti. Maria 1 The Maria.nom says.3sg that fut bring.3pl she.str to-the parti.acc Greek 'Mary says that it is her that they will bring to the party. [stressed/focused; emphasis]
Note that in (27h) afti can only have a co-referential reading3° with the c-commanding subject if it is focused (so that, for example, a possible meaning of the sentence could be something like: Mary says that they will bring HER (and not for example me or Joanna) to the party). Flat intonation would render co-reference ungrammatical.31 Interestingly enough, binding of demonstratives is not normally possible, a fact which has been traditionally attributed to a Principle C violation. Therefore, it is not entirely a straightforward matter to establish how binding is possible in the above cases, assuming that afti is a
30. Note that strictly speaking afti, ifbound, is a (bound) variable, given that syntactic binding entails semantic binding (see Kratzer & Heim 1998). Co-reference could also follow from accidental co-reference (that is, assignment of the same index from the context). Here, I use coreference in the traditional (non strictly semantic) sense, i.e. afti, being anaphoric, picks up the referent of its antecedent (here, I Maria). 31. Similar remarks apply to (27e) which also involve an 'anaphoric' use of the third person pronoun.
25
26
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
demonstrative.32 One could claim that these are not cases of true binding, but simply the result of accidental co-reference. Although this is possible, it can be easily shown that afti (and other 3rd person pronouns in similar environments) are bound variables and are subject to Prindple B ofbinding theory. This is illustrated in the following paragraph. Then, if it is true that 3rd person strong pronouns, which are prototypically demonstratives (or can be used as demonstratives), can also be bound, we either have to assume that a demonstrative can be both an R-expression and a pronoun/variable, or that 3rd person pronouns are not demonstratives in these particular uses, even though they share the same form with a true demonstrative. In other words, it is either the case that these are R-expressions which somehow escape a Principle C violation (possibly via syntactic or semantic means), or that they are bound variables (and hence are not subject to it in the first place). I will discuss this point later on However, let me point out here that these facts are (also) related to the so-called Choice Prindple (see Cardinaletti & Starke 1999, and main text below), namely the fact that a strong pronoun is used in a particular environment with the interpretation of a clitic (or a null pronoun in the case of subjects) when the latter cannot appear in that environment for independent reasons. For instance, 3rd person strong pronouns can be bound when they are complements to a preposition that does not allow for a clitic pronoun. or when they must be focused, i.e. in contexts where clitics are not possible independently. This means that variable status/interpretation is constrained by the Choice Prindple. An example of a prepositional object is given below: (29) 0 Vasilis 1 nomJZl oti milaghane ghi' aft.on1 sto parti. The Valisis.nom thinks.3s g that were-talking.3pl about him.str in-the party 'Vasilis thinks that they were talking about him at the party. Greek Note that in this example afton does not need to take on any spedal intonation or focus. This is the case because no other pronoun (clitic or null) is possible in the particular position. I will come back to strong subject pronouns later on. As a final note, it should be pointed out that 1st and 2nd strong object pronouns can also be bound by a c-commanding antecedent (on a par with 3rd person ones), however in that case focus is not required. This would follow trivially from the binding status of these pronouns, if we assume they are subject to Prindple B, on a par with clitics (as opposed to demonstratives, which are subject to Principle C). Whether this is actually true or not will be discussed later on in this chapter, since it could also be the case that these are R-expressions that for some reason escape Principle C. In any case, it is crucial to point out that 1st and 2nd person pronouns 32· Note for example that if the demonstrative pronoun ls used as a real demonstrative (i.e. ostensively, and encoding proximity to the speaker - see Lyons 1999 for these notions), coreference between the demonstrative and a c-commanding constituent ls obviously not possible: (i)
I Dhomna lpe oti ithele aftin. [stressed, demonstrative use] The Dhomna.nom said.3sg that wanted..3sg her.acc Greek 'Dhomna said that he wanted her (showing someone else in the room): This shows that aftos,-1,-o may have anaphoric/pronominal uses apart from true demonstrative uses.
Chapter 2. The properties of Greek clltics and their structural analysis
differ from 3rd person ones in that they can be bound in any syntactic (object) position, i.e. they are not constrained by the Choice Principle. This is shown in the following example, where an adjunct preposition phrase is focalized, which ensures that the 2nd person strong pronoun is not focused (cf. that two focuses are not possible in a single clause - see Tsimpli 1995).33 Note that the sloppy reading test confirms that this is a bound variable: (30) Les pos dhialeksan esena ghia ta MATIA su mono, Say.2sg that chose.3pl you.str for the EYES.acc your.cl only, ke to idhio lei ke i Lena. and the same.acc says.3sg and the Lena.nom Greek 'You are saying that they chose you for your EYES only, and so does Lena~ =You 1 are saying that they chose you 1 for your1 eyes, and Lena2 is saying that they chose you 1 for your1 eyes, too. (strict reading) = You 1 are saying that they chose you 1 for your1 eyes, and Lena2 is saying that they chose her2 for her2 eyes, too. (sloppy reading) 33· Note that it is not dear to me whether esena is not focalized at all in this example, given that the most salient reading of thls sentence is that the addressee has been chosen (among other individuals that are related to the addressee contextually) for a particular reason. If that is the case, then esena ls also subject to the Choice Principle (since a ditic pronoun would not be posslble with the same interpretation). 0 n the other hand, recall that strong pronouns (both objects and subjects) may also have a dlstinctive interpretation, which does not require contrastive focus/emphasls. In this particular case, it appears that esena does have a distinctive interpretation, therefore it is plausible to assume that the Choice Principle ls not applicable. Nevertheless, note that although a clitic and esena would both be possible in terms of focus/stress (assuming that thls ls not a focused position), they would dtlfer in terms of interpretation, since they would not render the same distinctive reading (recall that ditics are topicality markers, and no contrast/ dlstinction between their referent and other salient (possible) referents is implied). This suggests that the Choice Principle ls also operative here, but on semantic/pragmatic grounds. Moreover, note that 3rd person strong pronouns cannot be bound when there is another focus in the sentence. For example, cf the ungrammaticality/grammaticality of * I Olgha 1 lei pos (0 KOSTAS) aghapai tifti 1 (O KOSTAS)I(o Kostas) II ..J tin 1 aghapai (0 KOSTAS) I (o Kostas): ('the Olgha.nom says.3sg that (THE KOSTAS.nom) loves he.t:str (fHE KOSTAS.nom) I (the Kostas. nom) I !her.d loves') [capital letters stand for contrastive focus, while lower case letters for the postverbal subject stand nuclear stress related focus], as opposed to a 1st person pronoun: I Olgha lei pos (O KOSTAS) aghapai emena (0 KOSTAS) I (o Kostas) (where emena = me.str). This suggests that 1st and 2nd person pronouns dtlfer in this respect from 3rd person ones. Note also that strong object pronouns are not in principle incompatible with foci within a clause. In fact, they are perfectly possible: 0 NIKOS kalese emena/esena/afton sto parti (fHE NIKOS. nom invited..3sg me.str/yOILstr/him.str to-the party.acc) [both contrastive and narrow focus posslble]. What seems to not be allowed is for the 3rd person pronoun to be interpreted as a bound variable when another focus is present, not the actual co-occurrence of pronouns and foci (cf. Egho leo pos 0 KOSTAS simbathi emena, keto idhio les ki esi ('I.nom say.lsg that THE KOSTAS.nom likes.3sg me.str, but the same.acc say.2sg and you.nom'), which may have a sloppy identity reading, showing that this restriction does not apply to 1st person pronouns).
27
28
Cities in Greek: A Minimalist Account ofProdisis and Endisis
As I have already pointed out above, a 3rd person pronoun can also be interpreted as a bound variable (as shown by the sloppy identity test below), but in this case it must be (contrastively) focused (whereas this is not necessary for 1st and 2nd person pronouns)34 : (31) Kathe mafiozos 1 nomizi oti (mono) afton1 kinighai i Every mafia boss.nom thinks.3sg that (only) him.str chases-3sg the astinomia police.nom I i astinomia kinighai (mono) afton1. /the police.nom chases-3sg (only) him.str 'Every mafia boss thinks that the police are chasing only him: ( 32) Kathe dhimosioghrafos 1 lei oti oli afton1 akune Every journalist.nom says.3sg that all. nom him.str listen-to.3pl /oli akune afton 1• /all.nom listen-to.3pl him.str 'Every journalist says that everyone listens to him:
Greek
Greek
Alternatively, it must be complement to a preposition which does not allow a clitic pronoun. In that case no special intonation is required.3 5 36: (33) Kathe piitis nomlZI oti oli milane jia afton!* jia ton Every poet.nom thinks.3sg that all talk.3pl for him.str/for him.cl
34· A bound variable reading is also possible when the strong pronoun is conjoined with some other noun phrase/pronoun, since a ditic is not possible in this position for independent reasons (ditics cannot be coordinated, presumably because they are heads - see Cardinaletti & Starke 1999, Kayne 1994 for discussion). 35· The same trivially applies to 1st and 2nd person object pronouns that are complements to prepositions: (i) Ego nomizo oti oli milane ghia mena, I think.1sg that all.nom talk.3pl about me.str, ke to idhio nomizis ki esi. Greek and the same thinks.2sg and you 'I think that everyone talks about me, and so do you.' = 11 think that everyone talks about mel' and you 2 think that everyone talks about you 2• (sloppy reading) = 11 think that everyone talks about mel' and you 2 think that everyone talks about me,. (strict reading) 36. Here I have not said anything about ditk left dislocated and doubled strong pronouns in object positions. A major difference between them and non-doubled/dislocated pronouns is that the former are necessarily interpreted as topics (see also Anagnostopoulou 1994, 1997, 1999, Phlllppaki-Warburton et a1. 2004, and references therein). As a result, they cannot be focalized and moreover they loose their emphatic use. A second difference is that they can be
Chapter 2. The properties of Greek clltics and their structural analysis
me kala lojia. with good words.acc 'Every poet thinks that everybody talks about him positively:
Greek
G. Semantic properties of strong su.bject pronouns (semantic) Finally, in what follows I will briefly discuss the (semantic) properties of strong subject pronouns, however I will not dwell much on them, given that they are not the centre of my research. Recall that subject clitics are extremely rare in Greek, with a highly restricted distribution, which drastically differs from that of object clitics (to give one example, both subject clitics and full pronouns are possible in the rare environments they appear in), and as a result, no fruitful comparison between them and their corresponding strong pronouns is possible. Subject strong pronouns appear to pattern in many respects on a par with their object counterparts in terms of their semantic and pragmatic properties. That is, they are used emphatically, contrastively, as new information (for example as answers to narrow focus questions), and for switch-reference/introduction of new topic (i.e. with a distinctive interpretation). This use and interpretation of overt subject pronouns is quite common among so-called Null Subject Languages, in which the subject pronoun is typically null being resumed by the subject agreement morphology on the verb. Presumably, the null pronoun is preferred over the overt one, unless the latter is not possible for independent reasons (for example because there is no such thing as null focus, or because some additional interpretation is offered by the overt pronoun - see also discussion about pronoun choice later on). (34) a.
Pjos to ekane? Who.nom it.cl did.3sg 'Who did it?' 'EGHO to ekana/ to ekana EGHO. I.nom it.cl did.3sg/itcl did.3sg I.nom 'I did if
Greek
[narrow focus]
sinandise AFTOS (ke ox1 1 Nadja) Greek b. (AFTOS) tus (HE.nom) them.cl met3sg HE.nom (and not the Nadja.nom) 'It was HE that met them (not Nadja): [contrastive focus] Se prosvale o Alekos. You.acc insulted.3sg the Alekos.nom. me prosvales (ESI). Oxi, (ESI) No, (You.nom) me.cl insulted.2sg (You) ~lekos insulted you. No, YOU insulted me~
Greek
[emphasis/ correction]
interpreted as bound variables without any focus requirements. For example, (i) below is grammatical, whereas its non-doubled or left-dislocated counterpart must be focalized/stressed: (i)
I Eleni nomizi oti afti dhen tin simpathi o Kostas/dhen ti simpathi afti
Greek
29
30
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
Greek d. Chtes idha tin Andhromachi ke ton Pavlo. Yesterday saw.lsg the Andhromachi.acc and the Pavlo.acc 'Yesterday, I met Andromache and Pavlos: Aftos fenotan poli kurasmenos, san arostos. He.nom seemed.3sg very tired.nom, like sick.nom 'He looked very tired, as ifhe were sick: [switch reference) Strong subject pronouns can be used either deictically or anaphorically (in the simple sense that they may pick up their referent either by a linguistic or a non-linguistic antecedent).This is obvious for 3rd person pronouns (recall that they are demonstrative forms), although 1st and 2nd person pronouns may also be used anaphorically (for example as identificational foci): afto. (35) a. Eftichos pu den idha Gladly that not saw.1sg this.acc 'It's good that I didn't see this:
Greek
b. Chtes pigha sto sinema. Greek Yesterday went.lsg to-the cinema.acc. Egho dhen ithela na dho to ergho, I.nom not wanted.l sg subj see.1 sg the movie.acc ala me epise o Xristos. but me.cl convinced.3sg the Xristos.nom 'Yesterday I went to the cinema. I didn't want to see the movie, but Xristos convinced me to the contrary. They are typically referential (cf. their inherent deictic nature), however they may also be interpreted as bound variables. This is trivial for 1st and 2nd person subject pronouns, while 3rd person pronouns must be contrastively focused in order to be bound37: (36) a.
Ksero oti tha pezo egho xartia ow TO KALOKERI, Know.1st that fut be.playing.1sgl.str cards.acc ALL THE SUMMER.acc ke to idhio kseris ki esi. Greek and the same.acc know.2sg and you.str 'I know that I will be playing cards all summer, and so do you: = 11 know that 11 will be playing cards all summer, and you2 know that 11 will be playing cards all summer (strict identity reading) = I1 know that 11 will be playing cards all summer, and you2 know that you 2 will be playing cards all summer. (sloppy identity reading)
37· Again, as for object strong pronouns, subject strong pronouns may be bound also when they are coordinated, modified, etc., Le. whenever they appear in an environment that does not allow d.itics (which are prototypically bound) for independent reasons.
Chapter 2. The properties of Greek clltics and their structural analysis
b. Kanenas 1 dhen ipe pas (aftos 1) grafi musiki. (aftos 1). Noone.nom not said.3sg that (he.str) writes.3sg music.acc (he.str) 'Noone said that he writes music: Greek c.
Kathe fititis 1 ipe oti (*aftos 1) TON NEKRO iche Every student.nom said,3sg that (he.str) THE DECEASED.acc had.3sg dhi (*aftos 1). seen (he.str) 'Every student said that he had seen the DECEASED? Greek
Another way for a 3rd person subject pronoun to be bound is to be bound by a c-commanding bound variable (e.g. a pro- see Montalbetti 1984): (37) Kanenas 1 den ipe pas pro 1 nomizi. oti (aftos 1) ine eksipnos (aftos 1). Noone.nomnot said that pro think.3sgthat (he.str)is smart.nom(he.str) Greek 'Nobody said that he thinks that he is smart: On the other hand, a null subject can be interpreted as a bound variable without any such restriction: pistevi oti pro 1 ine eksipnos. (38) Kathe fititis 1 Every student.nom thinks.3sg that pro is.3sg smartnom 'Every student thinks he is smart:
Greek
Finally, it should be pointed out that strong subject pronouns cannot be used expletively, or impersonally, although they may be interpreted as [± human] when they are 3rd person (recall that they are demonstratives). Only pro subjects can be expletive or impersonal, or take an arbitrary interpretation (see Holton. Mackridge & PhilippakiWarburton 1997 for more details; see also Spyropoulos 2002 for arbitrary pro in Greek). Moreover, strong subject pronouns cannot be used in idioms (although this appears to be the case for subjects more generally- see Marantz 1984), as opposed to clitic object pronouns. Summarizing what I have said above regarding the semantic properties of personal pronouns in Greek, clitic object pronouns can be [± referential], [± human] (for 3rd person), can participate in idioms as expletive constituents, and must have a prominent antecedent. With regard to the Binding Theory, they can be bound variables or free pronouns (cf. the± referential readings). On the other hand, strong objects pronouns can be [±human], can be[± referential], since they may be interpreted as free or bound (although certain restrictions may apply for the bound reading with 3rd person pronouns), and they cannot be used expletively, impersonally or participate in idioms. Finally, subject strong pronouns can also be [± human], [± referential] (i.e. bound or free), and they cannot be used expletively, impersonally or participate in idioms. Pro, on the other hand, may receive expletive, impersonal, and arbitrary readings, can be[± human] and[± referential] (i.e. bound or free).
31
32
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
H. 1he "Choice Principle" Besides their semantic properties, a further property of Greek pronouns, strong and clitic, is that they appear to be subject to what I have been informally calling the Choice Principle. According to this 'principle, which has been proposed by Cardinaletti & Starke (1999), different types of pronouns in terms of deficiency (i.e. strong vs. weak, strong vs. clitic, weak vs. clitic) enter into some type of competition where the most deficient form wins, provided the relevant position of pronoun insertion does not clash with the inherent properties of the inserted type. If the latter case arises, the next less deficient pronoun is chosen. Put differently, a strong(er) pronoun is only possible when a weak(er) one is impossible, or vice versa the most deficient pronoun must be chosen if possible. As far as Greek pronouns are concerned, we have already seen that there are two series of object pronouns, strong and clitic, as well as two series of subject pronouns, namely strong and null (I leave subject clitics aside). According to the above-mentioned principle, clitics would be preferred over strong object pronouns, and null pro over strong subject pronouns38 in those environments where both would be possible in principle. However, this seems not to be the case in some particular contexts. For example, prepositions taking accusative complements do not accept clitic pronouns (why this is the case is an independent issue I am not going to discuss here - see Cardinaletti & Starke 1999 for a possible account, see also Abels 2003). As a result, only strong object pronouns are possible in this environment. In that case strong pronouns may take on properties that would normally be found only with elitic pronouns: for example 3rd person pronouns can be bound (as variables), although in other (neutral) contexts (i.e. in contexts where both clitics and strong pronouns are in principle possible) this is not possible. Similar remarks could be made for focalized environments, i.e. positions where focus must be assigned for independent (interpretational) reasons, or for coordination environments: in both contexts, a clitic pronoun is not possible and therefore a strong pronoun is used instead. Bound variable interpretation is also possible exactly in these cases. Null and strong subject pronouns also seem to be subject to this general principle. For example, null subjects are systematically chosen over strong pronouns in neutral 38. The reason for that ls that subject pro of Null Subject Languages is taken (by Cardinaletti & Starke 1999, and others) to be a weak subject pronoun This assumption is based on the semantic properties of pro, which I have referred to rather briefiy in the previous paragraph with regard to Greek, as well as on the fact that pro cannot refer to a non-prominent antecedent (for example via ostension), like clitics, or the fact that it ls an XP that must appear only in a particular (case-related) position, namely [spec, AgrSP], which ls exactly what weak object pronouns do with AgrOP. Of course, pro cannot be conjoined. focused. modified. etc., which ls a property of defident pronouns, and which follows trivially from the fact that it is a null element. However, the very fact that it also has properties that are particular to weak pronouns suggest that its defidency follows from the fact that it ls a weak pronoun (structurally), and not from its null/non-overt nature.
Chapter 2. The properties of Greek clltics and their structural analysis contexts (i.e. in contexts where both pronoun types are possible in principle), however this preference breaks down when the null pronoun is impossible for independent reasons. This would be for example the case with focused positions, structures involving coordination or modification, and so on. Here again we find that binding of a strong pronoun is possible exactly in those environments where independent reasons prohibit a pro subject. Although I do not intend to offer here a full review of Cardinaletti & Starke's principle (different versions of which have been around in the literature for some time- cf. Chomsky's 1981 Avoid Pronoun Principle, or Montalbetti's 1984 Overt Pronoun Constraint, and see Cardinaletti & Starke 1999: §7 for a comparison between those principles and the Minimise Structure/Choose the most deficient pronoun principle), two related issues arise with regard to the Choice phenomenon: first, the question remains whether true neutral contexts really exist. This is important, since if those are impossible, then pronoun choice cannot depend on favoring the least deficient pronoun; rather, choice of pronoun will be imposed by the distinctive features/ structure of each relevant position independently. To illustrate what I mean by this, for example in Greek, as I have already pointed out, clitic pronouns and strong object pronouns may both appear in a non-focalized context with a prominent antecedent. Since both pronouns may appear in such a context in principle, the prediction made by Cardinaletti & Starke (1999) is that the clitic should always appear. However, a strong pronoun is also possible, under the same conditions. A possibility would be to assume that in the strong pronoun case there is some additional feature/property that is not compatible with the clitic, and which is the reason why the strong pronoun surfaces (as for example appears to be the case with ostension of non -prominent antecedents). In fact, there seems to be some kind of semantic property that is characteristic of un-focused strong pronouns, namely the distinctive interpretation. Assuming that such an interpretation is not compatible with a clitic (a fairly natural assumption), everything would follow almost trivially. However, the main problem with this approach is that it implies that the strong pronoun would never compete with a clitic, at least in Greek, since there doesn't seem to exist another neutral context for them to compete (and similar remarks would apply to null pro vs. strong subject pronouns, the latter also having the distinctive interpretation). However, if there is no neutral context then there is no evidence that the weakest form is preferred. Maybe, the reason why the weak form shows up is simply because it is the only form that is compatible with that particular context, and not the result of competition among equally possible competitors. The second issue, which is related to the first one, is that the actual way a particular pronoun type is selected over another one is not entirely clear. For instance, in those cases where a strong pronoun is selected over a weaker one, because the latter is not possible for independent reasons, it is not obvious how the two pronouns ever compete, given that the independent reasons should already imply a non-neutral context: recall that modification, coordination or focalization are structural properties, imposed on the pronoun by its structural context. Even if we assume that the actual
33
34
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
choice takes place at the base position (following in this Cardinaletti & Starke 1999), any additional feature or property should already be present at the point of first merge (potentially an independent problem, since it could imply look-ahead at least in some cases), which in turn suggests that the derivational system would never put a clitic pronoun in that context (i.e. no competition would be possible). Additionally, if the choice takes place so early, i.e. at the first merge state, it follows that we should expect pure strong vs. weak pronoun properties in those cases where the weak pronoun is not possible for independent reasons. 'This in turn would lead us to expect that a strong pronoun 'substituting tor an impossible weak/clitic pronoun would only display properties pertaining to strong pronouns. Unfortunately, this is not true. To take one example, Greek 3rd person strong pronouns can be bound variables, exactly in those contexts where a clitic pronoun would be impossible, i.e. in contexts where the strong/ weak distinction is blocked. Moreover, they do not even need to get the distinctive interpretation in these particular uses. Finally, they lose their ability to encode the proximaVdistal distinction. In other words, they seem to abolish some of their strong properties, and take on some of the weak properties that characterize clitics. Accordingly, it is not clear to me how the Choice Principle, as presented in Cardinaletti & Starke (1999), would account tor this mixed status. On more general grounds, although the Choice Principle seems intuitively correct, in the sense that it seems to be true that languages generally tend to use less structure whenever possible, it still faces a series of problems, mostly because it is rather vague as a proposal, especially with respect to the actual grammaticallevel(s) thatregulate(s) this choice.
I.
Morpho-phonological criteria
Moving on to the phonologicaVprosodic properties of strong pronouns and clitics in Greek, it is quite well known that Greek clitics prosodically restructure into their host (see e.g. Nespor & Vogel1986, Drachman & Malikouti-Drachman 1999, Mavrogiorgos 2004, Revithiadou &Spyropoulos 2007a), on a par with clitic (and weak) pronouns (as well as other deficient functional elements, like for example articles) cross-linguistically. This basically follows from their prosodic deficiency. In particular, clitics are morphemes that consist of a single syllable and hence do not torm a foot. Given the basic tenet of Metrical Stress Theory (see Liberman 1975 and Liberman & Prince 1977), namely that stress is assigned to a syllable within a foot (essentially to the strong syllable/head of the foot), the lack of inherent/lexical stress that characterizes clitic pronouns follows. Moreover, since according to the Prosodic Hierarchy (ct~ Selkirk 1995) the Prosodic Word (PWd) is the minimal prosodic constituent that a morpho-syntactic constituent may be mapped onto, and since a PWd must minimally contain a toot, it also follows that a clitic, which is not a foot, cannot be mapped onto a PWd. As a result, a clitic needs to be parsed within an adjacent PWd (on its left or right side), following basic premises of the end-based mapping algorithm (see Hale & Selkirk 1987). This explains why a clitic must (at least at PF) cliticize to a host, and why this is not a
Chapter 2. The properties of Greek clltics and their structural analysis
requirement for strong pronouns (assuming that these take inherent/lexical stress and hence can be mapped onto a full foot and a PWd). Many researchers have argued that Greek clitics either adjoin or incorporate into their (adjacent) prosodic host (see e.g. Zec & Inkelas 1991; Selkirk 1995; Drachman & Malikouti-Drachman 1999; Tzakosta 2003; Mavrogiorgos 2004; Revithiadou & Spyropoulos 2007).39 1hat is, they are either affixal (yielding a recursive PWd/adjunction structure) or internal (yielding a single PWd, the clitic being dominated by the verbal mother-PWd) functional heads, following proposals made by Selkirk (1995). According to her, functional words, such as articles, pronominal clitics, particles and auxiliaries may be organized into PWds (i.e. be parts of a PWd) in four distinct ways, depending on their prosodic status. Putting the case of 'a functional word as a PWd' aside, prosodic elitics (i.e. morp hosyntactic words that are not themselves a PWd - see Selkirk 199 5) may be mapped onto p- structure as follows: ( 39) i.
free clitic
ii. internal clitic iii. afli.'Cal clitic fnc = function word lex = lexical word
( fnc ) ( lex ) PWd ) PPh ((fnc lex) PWd) PPh ((fnc (lex) PWd ) PWd ) PPh
In (i) the prosodic clitic is adjoined to the PWd, torming a Prosodic Phrase (PPh). In (ii) the clitic incorporates into the PWd, being sister to the lexical category. Finally, in (iii) the clitic is adjoined to the PWd, which is iterated and forms an adjunction construction. The view that clitics are part of the PWd of their host (namely, the verb, at least for clitics within the verbal domain), is supported by sandhi rules, which take place between the clitic and its host, 40 and which have the PWd as their application domain. For example, in Greeks-voicing takes place within the PWd, and is found both within 39· I.e. pace Nespor & Vogel (1986), who argue that ditics are not internal or affixal elements; rather, they constitute a separate constituent known as Clitic Group (CG), which is part of the prosodic hierarchy. For a critique and dismissal of the CG see Booij (1996) and Peperkamp (1996), among others. 40. Sandhi rules in Greek generally involve assimilation (place or manner of articulation, for example, voicing, or turning a nasal consonant into a labial in front of a labial, or to a labiodental in front of a labiodental, etc.) dissimilation or vowel deletion. See Holton, Mackridge & Philippaki-Warburton (1997) for detailed discussion. Note that such rules take place within words (obligatorily), between ditics and their host, between articles and their host, as well as between preverbal particles and the verbal host (all optionally). In a parallel fashion, vowel deletion may also occur between verbal forms and disyllable PWds, such as the first and second personal pronouns 'egho' ('I') and 'esi' ('you') or certain adverbials like 'edho' ('here') and 'eki' ('there'): /'kse.ro e. 'gho/ ~ ['kse.ro. 'go] ('I know') ///'ki. ta. e. 'kif ~ ['kita. 'ki] ('look there'). Apparently, in these cases we have no incorporation and the formerly weak syllable of the first foot (from the right) is re-metrified as the strong head of a one-syllable foot. Note that it is always an I e/ that is deleted.
35
36
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
words and within clitic-host combinations (see Mavrogiorgos 2004; Revithiadou & Spyropoulos 2007): (40) a. zmolikas (s ~ z in front of m) [name of a mountain) Greek b. maz miristike (s ~ z in front of m) [ us.cl smelled.3sg = she found us out] Greek Similar effects follow from other cases of consonant assimilation (such as place of articulation assimilation, which may or may not be combined with voicing), as for example the k ~ g alternation following I nl, or an n~ m alternation preceding If/: (41) a. ton katalava ~ to(n) gatalava [him.cl understood.3sg] b. ton fonakse ~ to(m) fonakse [him.cl shouted-at.3sg]
Greek Greek
In addition, hiatus resolution (i.e. vowel coalescence) may take place (optionally) between a proclitic and its verbal host. This phenomenon involves vowel deletion of the initial vowel of the host under specific conditions (basically, it is constrained by a vowel sonority hierarchy that is specific to Greek, as well as by further morpho-syntactic conditions; see Mavrogiorgos 2004 for more details on these conditions, as well as on other contexts where hiatus resolution takes place between a weak/deficient element and a strong host). Vowel deletion normally leads to prosodic restructuring of the clitic into its host (cf. the fact that the clitic ends up carrying the lexical stress of the verbal PWd): (42) mu.to. 'e.dho.se ~ mu. t6.dho.se [me.cl itcl-gave.3sg]
Greek
(Optional) vowel (e-) deletion (e is taken to be one of the less sonorant vowels in Greek - see Condoravdi 1990 and Holton, Mackridge & Philippaki-Warburton 1997) is also found in imperatives, first of all within the imperative PWd itself, but also between the final stress-less e-ending of 2nd person singular imperatives and the following clitic, however it does not appear to be the same process as vowel hiatus in proclisis (cf. for example that it is not triggered by hiatus, given that Greek clitics always have consonantal onsets, but rather it is related to the weakening of the unstressed final (e) - see Mavrogiorgos 2004 and references therein for details, as well as Revithiadou & Spyropoulos 2007). Crucially e-deletion is not possible across PWds: (43) a. dh6se mu ~ dh6zrnu [give.2sg.imp me.cl] (e-deletion and s-voicing) b. kanete ~ kande [do.2pl.imp] (e-deletionand t-voicing) pare taper!~* partaper [take.2sg.imp taper.acc] Greek A further diagnostic that enclitics attach to their host/PWd is that they trigger secondary/rhythmic stress in order to avoid violation of the so-called three-syllable window rule, according to which no PWd may be stressed beyond the third syllable from the right (i.e. the farthest possible stress from the rightis on the antepenultimate syllable) (cf. Nespor & Vogel1986):
Chapter 2. The properties of Greek clltics and their structural analysis
(44) Dhfdhakse: dhidhakse to (teach.2sg.imp. it.cl)
Greek
What is more revealing about the prosodic restructuring of clitics is that all the above mentioned processes (i.e. deletion, voicing, assimilation, dissimilation and stress shift/ change) also take place within 'lexical' PWds (i.e. PWds that correspond to a morphological word, a terminal element taken fully inflected from the Lexicon - see Chomsky 1995), however many of them behave differently within a lexical PWd. For example, most sandhi rules are obligatory within lexical PWds, whereas they are only optional within a 'clitic + host' complex (although see the e-deletion rule within positive imperatives, which seems to be optional). Moreover the addition of a suffix to an antepenultimate stressed word forces the stress to move to a lower syllable, whereas a clitic triggers a secondary stress: (45) mathima: mathimatos (class.nom: class.gen), * mathimatos (but: mathima tu - class.nom his.gen) Greek Based on these and other facts (e.g. resyllabification of the final consonant of elitics with the first consonant of the verbal host is not possible in proclitics, as opposed to resyllabification of the first consonant of clitics with the last consonant of the imperative host (after e-deletion), which is possible), Mavrogiorgos (2004), Revithiadou & Spyropoulos (2007) have argued that proclitics adjoin to the PrW of their host, yielding an adjunction structure (Selkirk's 1995 atfixal clitic structure), whereas enclitics incorporate into the PWd of their host (Selkirk's 1995 internal clitic structure)41 : (46) a. b.
[ cl [V]PWd lPwd (proclisis) [V cl]Pwd (enclisis)
Details aside, the main point here is that elitics restructure into their host obligatorily, exactly because they lack inherent stress (and hence cannot form a PWd on their own). On the other hand, strong pronouns have inherent stress and theretore they do not need to restructure (and as a result they don't). A final point about the phonological and prosodic properties of clitics and strong pronouns has to do with their (in)ability to take (contrastive) focus. Cardinaletti & Starke (1999) explicitly argue that the traditional conception that strong pronouns are typically focused, while clitic pronouns can never be focused, is empirically incorrect 41. Revit:hiadou & Spyropoulos (2007) have also proposed the following prosodic phrasing for preverbal clitics preceded by stress-less functional heads (such as e.g. the preverbal particles Sa, na, as): [fc d [V]PWd] PWd" In this case the functional syllable and the clitic syllable adjoin to the verbal PWd (cf. Sa tuz maghepswne ['fut them.cl mesmerize.lpf]). Finally, proclitics may also incorporate into the verbal PWd in those cases where the initial vowel on the verb is deleted and the ditic takes on the lexical stress of the verb. In that case proditks are also internal clitics. Similar remarks apply to enclitics that have not incorporated (c£ for example grafe to! vs. grafto! ('write.2sg.imp.imperf itcf)).
37
38
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
As a result. it becomes impossible to claim that clitics do not appear in certain contexts (e.g. coordination, modification, ostension) because they cannot bear contrastive stress. Rather, it seems that weak/clitic pronouns typically do not take focus because they require a prominent antecedent. and the latter is not compatible with focus/ stress. ln this way, pronoun choice remains a purely structural issue. The main points Cardinaletti & Starke make with regard to strong and weak pronouns also apply to Greek. For instance, as we have already seen strong pronouns are not necessarily stressed (in prosodic terms) in all cases (e.g. after a preposition, or when left-dislocated, or even when the pronoun has the distinctive reading). On the other hand, clitic pronouns may bear lexical stress (due to prosodic incorporation), or rhythmic stress (due to the 3 syllable window rule). However, note that in all these cases this is not an inherent lexical stress of the clitic itself (otherwise we wouldn't be able to account for the need of clitics to prosodically restructure) (pace weak pronouns, which may have their own lexical stress). Semantically focused clitics are only possible with emphatic stress under the repair/correction interpretation (which requires previous utterance of the clitic without the stress). 42 This kind of focus is not contingent on the availability of inherent stress, as opposed to other focus types. This is tor example evident from the tact that it may apply to sub-parts of words (cf. e.g. Pitsa, oxi Kitsa ('Pizza, not Kitsa'). Finally, it is also possible to find semantically unfocused strong pronouns, e.g. in coordination, in left-dislocation structures, or after a preposition. The fact that in these cases contrastive focus on a separate constituent is possible shows that strong pronouns do not have to be contrasted everywhere: (47) a. Emena I FLORA maresi (ochi i Karina). Me.str THE FLORA.nom me.cl-like.3sg (no the Korina.nom) 'It is FLORA that I like, not Karina: dhen tha po jia afton. b. TIPOTA NOTHING.acc not fut say.lsg about him.str 'I will say NOTHING about him:
Greek
Greek
This distribution shows that the basic difference between clitics and strong pronouns is not so much the presence vs. absence of focus, but the presence vs. absence of inherent lexical stress. This is in accord with the main conclusion of Cardinaletti & Starke (namely that focus is only indirectly related to the distribution of strong and clitic pronouns), as well as with our previous remarks regarding the prosodic deficiency of clitics. Finally, it is in agreement with the observation that weak pronouns may only optionally restructure (namely, in those contexts where they become reduced and loose their inherent stress). 42. In other words, this ls more like a metagrammatical type of focus, and as a result differs from contrastive stress which seems to have semantic properties as well (see Zubizarreta 1998 on the difference between emphatic and contrastive stress).
Chapter 2. The properties of Greek clltics and their structural analysis
].
XP vs. X status (syntactic)
A last property of clitic and strong pronouns regards their phrase structure status. More specifically, a standard assumption in the transformational-generative syntax approach has been that strong pronouns are maximal projections, i.e. XP constituents,43 whereas clitics are syntactic heads, i.e. X0 elements.44 Cardinaletti & Starke (1999) offer a number of arguments in favor of the X0 status of clitics cross-linguistically. For instance, in Germanic languages clitics cannot be found in the V2 initial position (where, according to standard assumptions, anXP is required in Spec CP, see den Besten 1983), and in languages like Spanish doubling is only possible by a clitic, but not by a strong pronoun (similar remarks apply for left dislocation in Spanish and Italian). Finally, clitics often seem to count as constituting a single complex head along with their host for movement, coordination, and deletion operations, such as e.g. Aux-to-Comp in Italian, coordination in many Romance languages, or auxiliary deletion in French (see also Kayne 1975 for arguments that deletion involves cl-V deletion). Putting here aside the complex issue ofwhether clitics are base generated on their host (or some functional head that is part of the Extended Verbal projection- see Grimshaw 1991 for this notion), or move to it (which would complicate the head status of clitics considerably), we can say that Greek clitics also have head properties, according to the above mentioned criteria: they can participate in doubling and Left Dislocation constructions (48), they seem to move along with their verb (49), they cannot move across another head of the higher functional domain of the clause (like for example negation or verbal particles) (50),45 and moreover they get deleted along with the auxiliary verb (51) (provided the surface form of the clitic in the first conjunct is compatible with that required by the verb in the deleted part of the second conjunct in terms of case- see Kayne 1975 for French): ( 48) a.
Ton idha ton Niko. Him.cl saw.lsg the Nick.acc
Greek
43· Weak pronouns are also considered to be XPs, however I will leave them here out of the dlscussion since I am mainly concerned with the dlstinction between ditics vs. strong pronouns.
44 Note that I do not assume that ditics are only heads. On the contrary, and as discussed in detail in Chapter 3, I take them to have both XP and X0 properties, following Chomsky (1995). However, here only the head propertles of clitics are being dlscussed 45· Elena Anagnostopoulou points out that thls example looks like a ditic climbing case, which is impossible in Greek for independent reasons (hence the ungrammaticality). Note, however, that Terzi (1997) has analyzed these structures as instances ofblocking of ditic movement by an intervening head, namely the particle. A simpler example illustrating the blocking effect of a head with regard to diticization would be the following: (i) *To dhen idha. ltd not saw.1sg 'I didrlt see tt:
In thls particular case the ditic has moved across the negation particle, leading to ungrammaticality.
39
40
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
b. Ton Niko ton idha. 'I saw Nick/ Nick I saw?
Greek
(49) a. Ta kerdhi tu ...J schedhon/...J idhi ta triplasiase. Greek The winnings his- gen almost/ already them-ace tripled-3sg 'He almost/already has tripled his winnings: b.
Ta kerdhi tu ta triplasiase schedon!idhi.
Greek
c. *Ta kerdhi tu ta schedhon! idhi triplasiase.
Greek
d. *Ta kerdhi tu triplasiase schedhon!idhi ta.
Greek
(SO) *To theli na aghorasi. It.cl wants.3sg subj buy.3sg 'He/she wants to buy it: (51) a.
*Ton
ke tin idha. Him.cl and her.cl saw.lsg 'l saw him and hd
Greek
Greek
b. Ton echo dhi ke aghapisi. Him.cl have.lsg seen and loved 'I have seen and loved him?
Greek
*Ton echo dhi ke echo aghapisi.
Greek
d. *Ton echo dhi ke ton aghapisi.
Greek
e. *Ton echo dhi ke milisi. Him.cl have.lsg seen and talked-to 'I have seen him and talked to him:
Greek
f.
Mas exi dhi ke milisi. Us.cl has.3sg seen and talked-to 'He/she has seen us and talked to us:
Greek
Some explanatory remarks are in order here. First of all, (49) uses the relevant position of the elitic and the verb with regard to the position ofadverbs (in the spirit of Alexiadou 1997, Cinque 1999) to show that Greek clitics attach syntactically to their finite host low in the clause (at the vP periphery, but not necessarily in T), however they may move along with the verb as proclitic 'free riders' to a higher position.46 (50) shows that
46. Obviously, here it is asswned that adverbs are merged following a rigid hierarchy, while main verbs and auxlliaries may optionally move to higher clausal positions (see Ledgeway & Lombardy 2005 for the same intuition). Elena Anagnostopoulou asks how we know that the verb moves and not the adverb shifts. I refer the reader to Cinque (1999) for more details, however let me point out that a simple piece of evidence against adverb shifting ls the rigidity of the adverb hierarchy: I.e., if adverbs were shifting, we would expect to find variation in their ordering, which is nevertheless not what we find For example, in (49a-b) if we add the next lower adverb to schedhlm or idhi, the ordering between these two adverbs will not change, irrespective
Chapter 2. The properties of Greek clltics and their structural analysis
na, a preverbal particle marking subjunctive mood and merged in a head above INFL (=Inflection) (see Terzi 1992, 1997; Giannakidou 1998, 2007; Philippaki-Warburton 1998; Philippaki-Warburton & Spyropoulos 1999; Roussou 2000, 2006, among many others), blocks head movement to the higher clause (see Terzi 1997, and also Kayne 1989b for French and Italian clitics which cannot move over an embedded C head47 ). Finally, (51) shows a number of deletion facts, which mirror those reported in Kayne ( 1975) for French, and which have been argued to offer evidence in favor of the head properties of clitics (see also Kayne 1994, and discussion in Chapter 4). In this respect note that similar deletion facts have been noticed for preverbal particles in Greek, which are also heads (cf. Philippaki-Warburton & Spyropoulos 1999): (52) ea miliso ke *(ea) figho. Put talk.1sg and (fut) leave.1sg 'I will talk and (then) leave:
Greek
Although the position of 8a (and that of other preverbal markers) does not seem to be the same as that of clitics (see e.g. Roussou 2000), in the sense that it does not seem to be incorporated into the verb in the syntax, as clitics have been reported to do (but see Kayne 1994 and discussion in Chapter 4), on the other hand the parallelism (in terms of deletion) between preverbal heads and clitics strongly suggests that the latter may also be heads.48 2.2.3 The morpho-syntactic status of clitics: affixes or words?
In this section I will discuss a closely related issue of clitic pronouns, namely their morpho-syntactic status. The morpho-syntactic status of clitics has been one of the major issues in the theory of cliticization as well as of the syntax-morphology interface. The hallmark of clitics is that they are both word-like and affix-like. For instance, they seem to be visible to syntactic rules, which typically apply to words (e.g. movement, they feed binding), but also to morpho-phonological rules which apply within words (e.g. stem+ affix combinations) (for Greek see Joseph 2002, Malik.outi-Drachman & Drachman 1988, and references therein). Therefore, clitics appear to share properties of both words and (lexical) affixes, in other words they are mixed or hybrid
of the position of the adverbs with regard to the verb. This would not be possible with adverb shifting without further stipulations. 47· Although clitic movement across the C head is allowed with restructuring infinitives such as ji11ire di, cominciare a, riuscire a, and provare a. How restructuring structures are to be analyzed is beyond the scope of this discussion (see Cinque 2004). 48. This assumption may be further corroborated by the fact that preverbal markers undergo sandhi rules (like assimilation, etc.) which are normally possible between adjacent heads, and which are also found with clitics and clitic clusters, but not between strong pronouns (see also Cardinaletti & Starke 1999 on this point).
41
42
Cities in Greek: A Minimalist Account ofProdisis and Endisis
categories. Interestingly enough they also exhibit properties that are not typical either of words or affixes proper. For instance, as morphemes/affixes they undergo certain operations that are similar but not identical to those applying to other affixes (e.g. allophonic assimilation). Clitics are also syntactically non-typical: e.g. they appear in positions which are not available to non-clitics (movement to special positions), or they may appear in more than one special positions (see e.g. Ledgeway & Lombardi 2005), and moreover they can have both phrasal and head properties (see Sportiche 1996; Kayne 1989b, 1991; Belletti 1999).49 All in all then, clitics share properties with both words and affixes, but on the other hand they also seem to be somewhat special when compared to other words and affixes. Given this highly hybrid status of clitics, two main trends have appeared in the literature: one was to reject the notion of'hybridness' within the grammar as such and categorize clitics either as words (e.g. Kayne 1975) or as affixes (e.g. Joseph 1988, 2002; Drachman & Malikouti-Drachman 1999), albeit with atypical properties (cf. the notion of atypical affix or word). The other was to take the special properties of clitics at face value and categorize them as separate entities, namely clitics, which should be viewed as real mixed elements, lying between affixes and words (see e.g. Zwicky 1977, 1985; Zwicky & Pullum 1983; Gerlach 2002 ). Because Zwicky's categorization and criteria were proposed as descriptive tendencies/properties, independently of a particular theoretical framework, they were used by various researchers as arguments in favor of a lexical or a syntactic view of clitics (i.e. the first trend subsumed in essence the second trend). As a result, a basic distinction between strictly lexical(ist) and syntactic accounts of clitics (especially special elitics, given their syntactic and phonological/affixal properties) was created. Lexical accounts of (special) clitics became very popular in the generative tradition, given their strong parallelism with morphological affixes. Very generally, lexical analyses take clitics to be affixes (either word-level (i.e. they are added to the constructed word in the pre-syntactic Lexicon) 50 or phrasal-level (i.e. they are added to a phrase (e.g. VP) via a number of operations in the post-syntactic morphology - see Klavans 1985, Anderson 1992, Miller 1992, Monachesi 1995 for phrasal affixes, and 49· These apply basically to so-called special ditics (for the distinction between simple and special clitics see Anderson 1992, 2005, Zwicky 1977).
so.
Here, 'word-lever is a more general term which intends to cover all cases of ditics/atfixes that are attached to or within the word This would involve for example stem-level affixation and word-level affixation. See also Condoravdi & Klparsky (2001) for more details on this. Note that Condoravdi& Klparsk.y (2001) correctly point out that most of the phonological and morphological arguments that have been used in the literature against the lex:ical/affixalstatus of ditics, and which show that clitics have distinct morpho-phonological properties from other affixes, do not really prove that ditics cannot be affixes. This has been anyway the major approach of Lexical Phonology to this problem. On the other hand, positing a separate word-to-word or phrasal morphology and phonology does not prove that ditics are affixes either (unless it can be shown that other affixes may also have similar properties).
Chapter 2. The properties of Greek clltics and their structural analysis
Drachman & Malikouti-Drachman 1999, Condoravdi & Kiparsky 2001 for word level affixes)); inflectional or derivational) which are morphological parts of their host (or domain, more generally). Syntactic accounts, on the other hand, argue that clitics are independent words/ morphemes and their a:ffixal status stems from their phonological or syntactic properties, or a combination thereof. Their common important characteristic, which differentiates them from lexical approaches, is that they do not consider clitics as morphological parts of their host (this is so, independently of whether clitics have moved to their surface position, or have been base generated attached to their host (see ]aeggli 1986)). The major criticism of lexical(ist) approaches towards syntactic ones is that the latter cannot account for (some of) the a:ffixal properties of elitics, and more generally that a syntactic approach would violate the Lexical Integrity Hypothesis, given these properties. To give an example, Miller (1992) argues that deletion under identity (e.g. in cases of coordination of two verbs and one object clitic) is a reliable test for syntactic constituency, given that syntactic adjunction (assuming that clitics adjoin to a head within the extended verbal projection) should allow deletion of the clitic independently of the verb (i.e. contrary to what Kayne's 1975 theory predicts), due to the fact that an adjoined constituent (here the clitic) is able to c-command over its mother node/ segment. Since clitics cannot be deleted separately from their host, they must be internal morphemes of the verb, and as a result non-visible to syntax (and hence, unable to c-command outside the V node). On the other hand, for those syntactic approaches which assume that there is a pre-syntactic generative Lexicon the major problem that lexicalist proposals seem to face (among others) is that clitics can appear attached to verbs/hosts with which they bare no lexical relation. Since, under standard lexicalist assumptions, affixes (derivational or inflectional) attach to a stem that subcategorizes/ selects for them, it is not straightforward how this kind of'climbing' can ever take place. 51 Here I will assume that a syntactic approach to cliticization, which takes (special) clitics to be independent (bound) morphemes that are added from the lexicon to the numeration and then are inserted into the syntactic derivation separately from their host, is the correct one. There are three main reasons for this: (a) First of all, I think that the problems that appear to challenge syntactic theories are much less severe than those challenging the lexicalist approaches. In particular, on the one hand, most, if not all, of the morpho-phonological/a:ffixal properties of clitics This is espectally so, given that in most cases of climbing the actual host and the selecting verb do not constitute a single constituent morphologically (c£ for example have-participle constructions in many Romance languages and Greek, or so-called restructuring constructions, in which constituents like adverbs or quantifiers may intervene between the two verbs). As a result, one could not simply assume that in fact we are dealing with one (morphologically complex) predicate. See Philippaki-Warburton & Spyropoulos (1999) for a detailed criticism of the lexicalist approaches. See also Monachesi (1995, 2005) for an alternative, lexicalist interpretation of climbing constructions.
51.
43
44
Cities in Greek: A Minimalist Account ofProdisis and Endisis that have been reported in the literature can be accounted for on independent grounds, without needing to have recourse to a lexical affixation of clitics.52 Even more interestingly, this would be the case even if we assumed the strict version of the Lexicalist Hypothesis (as e.g. in Chomsky's 1995 theory). For example, it has been pointed out by various researchers that the morpho-phonological sandhi rules, stress rules and vowel deletion rules that apply within the clitic + host combinations can be attributed to the prosodic deficiency of clitics and in particular to the fact that they have to adjoin or incorporate into their adjacent PWd postsyntactically. The combination of prosodic and alignment constraints, in the spirit of Selkirk (1995), would make sure that the deficient clitic would be integrated into the prosodic structure and hence be fully interpreted by the Sensorimotor systems of the brain (assuming that all constituents must be integrated into the prosodic structure) (see also Zec & Inkelas 1991; Peperkamp 1996; Vigario 1999; Tzak.osta 2003; Revithiadou 2006; Revithiadou & Spyropoulos 2007a). 53 Deletion facts could be equally accounted for within a syntactic approach. Recall that deletion facts have been one of the major and most serious problems posited to the syntactic approaches by the lexicalist approaches. This is so because clitics appear to behave on a par with other (inflectional) affixes in this respect. For this reason, showing that these facts can be accommodated within a syntactic approach would give extra credence to the latter, and deprive lexicalist theories of a strong argument. First of all, it is important to point out that c-command does not seem to be a sufficient condition for wide scope over a conjunction. This is shown by the fact that preverbal particles in Greek must be repeated under conjunction, although they are merged in distinct heads which appear to be related to each other in terms of c-selection54 (cf. for example that distinct types of mood heads match distinct types of 52. Elena Anagnostopoulou points out: "except, perhaps, from clitic ordering and related prop-
erties (see Bonet (1991 )." 53· Note that the fact that all these properties can be accounted for via a post-syntactic morphological component, does not really prove on its own that ditics are not affixes (cf the notion of post-cyclic affixes in lexicalist approaches). However, it shows that given all the other arguments in favor of a syntactic approach, these particular morpho-phonological properties of clitics could be accounted for without any major problems, stripping in this sense the lexicalist approaches from a basic argument. 54 Elena Anagnostopoulou asks what explains the restriction that particles have to be repeated under conjunction, if it is not related to c-command. One possibility would be to assume that a (post-syntactic) morpho-phonological rule requires that the two conjuncts contain the same set of morpho-phonological features, so that If one conjunct contains e.g. negation, the other conjunct must also contain it. Alternatively, we could assume that these particles are in fact subparts of the verb from the start of the derivation, on a par with elements like the Germanic zu. Given the rule that only similar elements may be conjoined. the repetition of particles in conjunction would follow trivially. Note, however, that according to this analysis V-movement to higher heads, such as CModal (see Chapter 5), becomes V-related (since it involves checking of
Chapter 2. The properties of Greek clltics and their structural analysis
negation heads, or that distinct types of complementizers match distinct types of mood heads - see e.g. Terzi 1992; Philippaki-Warburton 1998; Philippaki-Warburton & Spyropoulos 1999; Roussou 2000): tin tenia. (53) Dhen ea dhume ke *(dhen) *(9a) sxoliasume Not fut see.1pl and not fut comment-on.1pl the movie.acc 'We are not seeing and commenting on the movie:
In the above example deletion of the future particle is impossible.55 Deletion of negation is possible, but it would yield a non-negative interpretation for the second conjunct (which shows that the higher negation cannot have scope over the conjunction). Note that although these heads are strictly ordered with each other and although they do not allow a phrasal constituent to interfere between them and the (d)-Verb, it does not necessarily follow that they constitute a single constituent/complex head in the syntax, or in the lexicon. This follows from the fact that preverbal particles have been argued to block verb movement to a higher head, which can only be the result of moving a lower head across an immediately dominating head (i.e. a typical violation of the Head Movement Constraint (HMC)- see Rivero 1994, and Terzi 1992).56 Moreover, the assumption that preverbal particles along with the verb and object clitics are listed in the Lexicon as lexical entries is quite implausible, given that on the one hand it would impose a huge burden on the Lexicon, while on the other it would miss a number of facts, such as e.g. the blocking effect of preverbal particles to V-movement, or the selectional restrictions observed between these particles (see also PhilippakiWarburton & Spyropoulos 1999). All in all, then, if c-command does not guarantee wide scope, the argument of the lexicalist theories weakens: clitics could still be
features that are part ofthe verbal head), creating thus problems to the analysis of encllsis (which is taken to be related to V-related movement). For this reason, it seems more adequate to assume that these particles are independent morphemes (an assumption which is further supported by the fact that some of them (like e.g. na or min) may appear on their own) that are merged in the left periphery of the clause and which clitidze on the verb only at the post-syntactic morphophonological component. 55· In fact, and in order to be more precise, deletion of ea ls impossible in both conjuncts, given that the perfective aspect verb that is used in the example is one ofthe so-called dependent forms, i.e. it cannot appear on its own without the presence of a verbal particle such as 8a, na or as (note crucially, that negation cannot 'save' the dependent form on its own; rather a modal(ity) head is required (see Roussou 2000 for details)). 56. Note in this respect that negation does not seem to have wide scope over a conjunction also in Italian (see Belletti 1994), however in that case negation is analyzed as a ditic. This is so because negation can move along with the verb (and an object ditic) as one complex head in a number of structures, such as for example the Aux-to-Comp construction. In Greek, though, such possibility never arises, suggesting that there the negation particle is not a ditic. See Roussou (2000) for further evidence in favor of the separate head status of preverbal particles in Greek.
45
46
Cities in Greek: A Minimalist Account ofProdisis and Endisis assumed to be able to c-command over the conjunction. but their obligatory repetition in both conjunctions would be attributed to a different source. That this might be the case is further suggested by the properties of incorporated adverbs in Greek. Rivero (1992) and Alexiadou (1997) have shown that in Greek light manner and aspectual adverbs may syntactically incorporate into a c-commanding verb. 57 (Syntactic) incorporation constructions have been traditionally analyzed as involving head movement/adjunction, subject to the HMC (and in reality to the Empty Category Principle (ECP)- see Baker 1988 for extensive argumentation).lnterestingly enough, incorporated adverbs behave like clitics in many respects (although not in all respects - see Alexiadou 1997, Mavrogiorgos 2006 for discussion), and crucially in terms of wide scope.58 In particular, incorporated/adjoined adverbs need to be repeated in each conjunct. unless the adverbial has moved to an auxiliary verb and the conjunction involves the participial clauses/parts: (54) a.
Kalomasisa ke efagha to komati to kotopulo. Well-chewed-lsg and ate-lsg the piece-ace the chicken-ace 'I chewed the piece of chicken carefully and then ate
Greek
it:
b. Masisa ke ghrighorokatapja to k.omati to kotopulo. Chewed-lsg and quickly-swallowed-lsg the piece-ace the chicken-ace Greek 'I chewed and then quickly swallowed the piece of chicken: c.
Masisa ke efagha grighora to k.omati to kotopulo. Chewed-lsg and ate-lsg quickly the piece-ace the chicken-ace Greek 'I chewed and ate the piece of chicken quickly:
57· Note that this does not exclude in principle the existence oflexically derived 'adverb + verb' combinations (as for example argued by Smirniotopoulos & Joseph 1997), in the same way that the existence of lexically derived constructions of this type would not exclude the posslbility of syntactically constructed constructions. This is corroborated by the fact that syntactic adverbial incorporation is restricted by a number of syntactic constraints which are met only by a subset of adverbs, always gives rise to a compositional reading, and moreover allows what has been called as adverb climbing (i.e. the adverb may appear incorporated in the participle or the auxiliary in complex tense constructions) (see Rivero (1992) and Alexiadou (1997) for more details with regard to these and other arguments). In this respect, note that for a Distributed Morphology approach, all incorporated adverbs would be syntactically derived, with any differences being accounted for in the spirit of Marantz (1997, 2001), namely via incorporation at distinct levels (root vs. functional level). 58. Note that although conjunction of ditics appears to be acceptable only if the clitic is repeated (as opposed to the adverbial cases where the absence of adverbs merely renders a distinct reading), this should follow from the simple fact that some verbs require a complement, whereas with adverbials this is not normally the case. Accordingly,. whereas * tin idha ke chastoukisa (her.cl saw.lsg and slappedlsg) is out, since "slap" requires a complement, tin idha ke chtlpisa (her.cl saw.lsg and got.hitlsg) is in under the non-transitive/inchoative version of the verb chtipao (to hit- to get hit). This suggests that ditic repetition cannot be (only) related to a morphological requirement.
Chapter 2. The properties of Greek clltics and their structural analysis
d. Kalomasisa ke kaloefagha to komati to kotopulo.
Greek
e. Exo psilofai k.e pji. Have-lsg little-eaten-part and drunk-part 'I have eaten a bit and I have drunk:
Greek
f.
tru
HPsiloexo ke pji. 'I have eaten a bit and drunk a bit:
Greek
While in (54a-b) the adverbial cannot have wide scope over the conjunction, this is possible in ( 54c ), where it is not incorporated, and in ( 54d) where it is repeated in each conjunct Finally, in (54e-f) the adverbial is incorporated into the auxiliary, and hence can have wide scope over the conjunction. Now, if it is correct that these particular cases of adverbial incorporation are syntactically derived by means of head adjunction/ movement of the adverb to the verb (main or auxiliary), two things follow: (a) that syntactic head movement/adjunction does not always yield wide scope (this applies independently of whether we assume that an adjoined head may c-command outside the complex head or not5 9); (b) that the lack of wide scope exhibited by elitics does not necessarily entail that they cannot be syntactically adjoined to their host Combined with the previously mentioned facts on preverbal particles (which doc-command the verb, and moreover are not adjoined to it syntactically), we may conclude from these points that c-command (independently of whether it involves a head adjunction construction or not) is neither a necessary nor a sufficient condition for wide scope. Although this does not prove on its own that clitics adjoin as heads to their host, it definitely shows that head adjunction would not necessarily imply wide scope; hence, lack of wide scope cannot be attributed to lack ofhead movement/adjunction.60 59· Although note that this is exactly what the lexicallst theories assume. Moreover, the same conclusion follows from the requirement that the adjoined head must c-command its trace (see Baker 1988). 6o. Note that in other languages higher heads may be deleted from a conjunction without any ungrammaticality (see for example deictic subject clitics in many Northern Italian dialects - see Poletto 2001). Moreover, in Greek nominals the definite article may have wide scope over a nominal conjunction, provided the conjoined nouns modify the same referent. However, given that conjunction oftwo nouns requires that they denote a set ofproperties predicated of a single individual, we can safely conclude that in these cases the conjunction involves two noun phrases embedded under a single article that introduces the (single) referent (see also Alexiadou, Haegeman & Stavrou 2007 for a dlscussion of these constructions; see also Longobardi 1994 for similar data in Italian). This means that although the article c-commands the two nouns (or noun phrases), wide scope is not possible in thls case (for evidence that the definite article ls in Din Greek, see Alexladou 2001, 2005, and references therein). Again, c-command does not seem to correlate with wide scope; rather, what appears to be crucial is that conjunction involves two noun phrases. In this respect, note that conjunction of two DPs with definite articles does not yield a single referent, but two dlstinct referents. If we delete one of the two articles, the single referent interpretation kicks in (provided the two nouns denote properties that can be predicated of one individual felicitously). This confirms that D-NP+ NP involves NP conjunction
47
48
Cities in Greek: A Minimalist Account ofProdisis and Endisis
As a final point, note that coordination facts are quite camp lex, and I do not intend to offer an analysis of coordination here, since this is not the goal of this section. However, as a way of summarizing the above discussion, we could point out that the few coordination data we have looked into suggest that conjunction involves phrasal conjunction, combined with some general notion of identity, which may include structural, morphologicaVphonological as well as referential (i.e. semantic) properties. Identity would in essence ensure recoverability of deleted information. C-command seems to play a role, however this must be the case only when the relevant head c-commands the conjunction of two structurally lower phrasal constituents. This would be the case for example in NP + NP conjunction c-commanded by a singleD head (namely. the definite article). In such constructions identity seems to be also semantically or pragmatically constrained: for example the two noun predicates must be able to denote properties of one and the same referent. Similar constraints seem to apply to IP + IP conjunctions that are embedded within a single sentential subordinator: (55) a. *?Mu ipe oti tha pai sti dhulia Me.cl said.3sg that fut go to-the job.acc ke tha pernuse to apoghevma apo ti leschi. Greek and fut pass.3sg the afternoon from the club.acc 'He/she told me that he/she would go to his/her job and that he/she would drop by the club in the afternoon: b. ?Mu ipe oti pighe stin paralia Me.gen said.3sg that went.3sg to-the beach.acc ke k.olimbise olo to apoghevma. Greek and swam.3sg all the afternoon. 'He/ she told me that he/ she went to the sea and was swimming all the afternoon: It seems that this kind of restriction has a semantic source, and hence is applicable to elements like determiners and complementizers. However, it is not clear whether similar restrictions apply to all instances of conjunction. On the other hand, within the conjuncts themselves, one possibility would be to assume that only phrasal constituents can be deleted. Although this sounds a bit (and not DP conjunction and deletion of one D, assuming that the morphology cannot have access to the semantic features that correspond to the deleted D, so as to be able to delete them too), and moreover that c-command and categorial identity are not enough to secure deletion under conjunction. It seems that the referential/binding property of the definite article, or more generally its subordinating use disallows deletion under identity. In fact we find similar effects with (some uses of) complementizers/subordinators like oti (that), an (if) and pote (when) (although not e.g. withpjo11 (whom), and not with all instances ofthese complementizers-it appears that they could be deleted if they c-commanded two IPs which could be taken as involving two related events/propositions). This issue is quite complicated and requires further research. I will leave it here.
Chapter 2. The properties of Greek clltics and their structural analysis
controversial, it seems to make a number of predictions. First of all, a clitic would not be able to delete unless it is a full phrase, i.e. a maximal constituent. In this way elitics would be treated on a par with other noun phrases which can be deleted under identity. In fact, in many Tobler- Mussafia languages, in which clitics appear to project to the phrasal level, clitics may be deleted in conjunction (see for example European Portuguese, or Medieval Spanish),61 and the same seems to apply to subject clitics in French. All of these constituents have been claimed to be XPs. On the other hand, in languages like Greek or Italian and Spanish, objectclitics behave as heads in their landing site/surtace position. This means that they are adjoined to a head within the Extended Verbal projection. Assuming that only phrasal constituents can be deleted, it would be predicted that whenever the verb or the clitic get deleted, the other one also gets deleted, simply because in reality it would be the whole phrase containing them that would be deleted in the first place. As a result. we can capture the fact that parts of complex heads cannot be normally deleted under identity, without abandoning the c-commanding property of clitics, and their syntactic status. All in all, the assumption here is that a head will be deleted as part of a larger constituent If a clitic is part of this larger constituent (for head clitics this will be trivial), it will always be deleted together with the verb contained in this constituent. Ifboth the morpho-phonological and the deletion problems can be resolved in an adequate manner, then not much remains for the lexicalist theories to argue with against syntactic theories. On the contrary, they seem to face serious problems with regard to the syntactic properties of clitics. Recall that it has been pointed out in the literature that special clitics in Romance and in Greek exhibit mixed X0 and XP properties. Kayne (1975) was the first one to note that French object clitics cannot be base generated on the verb via some lexical redundancy rule (which would add a subcategorization property like 'take a direct/indirect object clitic'), because such an approach would fail to account for all the empirical evidence that shows that clitics move syntactically from the object (XP) position. This evidence includes the complementary distribution of clitics and full noun phrases/pronouns (in French), the blocking effect of embedded subjects to clitic movement in restructuring constructions (as well as the fact that the clitic may trigger participle agreement in the matrix clause- see Sportiche 1996), or the clitic climbing properties of clitics. Additional evidence for XP movement is offered by past participle agreement with direct object clitics in French and Italian (see Kayne 1989a), or by restrictions on extraction out of PPs or DPs (see Sportiche 1996 for detailed discussion). On the other hand, clitics appear to have X0 properties in their surface position, as is evidenced from the fact that they (or the 61. Note, however, that as Adam Ledgeway correctly has pointed out to me this is true only in prodisis; in enclisis the ditic has to be repeated The reason why this happens is an issue for further research. One possibility would be that enclitics are XPs/X 0s, contrary to proditics, which are XPs. This would lead to incorporation of enclitics to the verbal host and as a result to lack of wide scope in conjunction. Another possibility would be to assume that the enclitic head moves out of the XP into the verbal host.
49
50
Cities in Greek: A Minimalist Account ofProdisis and Endisis
head they attach to) cannot move across a higher head in restructuring constructions (see e.g. Kayne 1989b for the blocking effect of negation and complementizers in these contexts, although see also Sportiche 1996 for opposing arguments), or from the fact that they may move along with their host to a higher head position (see discussion in Chapter 5, Section 5.5 on V-to-C movement). If clitics have properties that are akin to phrases and heads, i.e. syntactic words, and if they are constrained by rules that refer to syntactic chains member of which is the clitic, it is not clear how a lexicalist approach would account for their syntactic visibility. In this respect, note that postulating a pro object is not enough, since the clitic itself must also be syntactically visible in order to be able to bind the null pro. Similar remarks have been made with regard to clitics in other Romance languages, as well as in Greek. For example, Alexiadou & Anagnostopoulou (1997) argue that (doubling) clitics pass a number of diagnostics that have been proposed in the scrambling literature for the detection of A-properties of chains 6z: they feed or bleed binding (a hallmark of A-chains) (e.g. they create or destroy Weak Crossover Effects, they obviate Principle C violations), and they are compatible with floating quantifiers. To these we could add that they do not induce Weak Crossover Effects, they do not license parasitic gaps and they are subject to the Specified Subject Condition. As far as their X0 properties are concerned, we have already seen that Greek clitics do have such properties, on a par with Romance clitics. All in all, it appears that a strictly lexicalist approach to cliticization would not be able to accommodate these facts in a natural way. This would be true independently of how one analyzes the phrasal duality/ambiguity of clitics (e.g. via movement, via base generation and binding of an XP in the complement position, etc. - see Chapter 4), since the point here is that the clitic must be visible to the syntactic computation. This conclusion, combined with our previous conclusion that the problems used by the lexicalist theories against the syntactic theories can be accounted for in a straightforward way, leads us to the more general conclusion that syntactic theories of cliticization can account both for the affixal and the syntactic/word properties of clitics. Besides the fact that a syntactic approach faces less problems in comparison to a lexicalist approach, an additional reason why the former would be more preferable is that clitics can be a syntactic constituent (and hence cliticization can be a syntactic operation) even within a strict Lexicalist framework. Recall that structurally speaking 61. In fact, their more general goal is to assimilate ditic doubling in Greek with scrambling in languages like German, using syntactic, semantic/interpretational and phonological evidence. In their analysis they use Sportiche's (1996) proposal on doubling and scrambling. Finally, note that in order to account for the lack of A-bar properties in doubling constructions (as opposed to scrambling which appears to show mixed properties) they follow Mahajan (1994) and Deprez (1994), according to which scrambling involves first movement to an A-position, followed (in some cases) by A-bar movement to a higher position. Presumably, Greek clitics will only be able to take the first A-movement step. See discussion in chapter 4 for more details on this as well as on other A-properties of Greek clitics.
Chapter 2. The properties of Greek clltics and their structural analysis clitics are ambiguous between phrases and heads. Crucially, however, they exhibit head properties in their surface position, i.e. only with regard to their structural relation with their host The common way to capture this effect in the Principles & Parameter's framework was to argue that clitics attach to their host as heads (independently of whether they attach onto their host, the head containing their host or a higher empty head c-commanding their host). 63 If this is true, any affix-like properties of clitics would follow from the highly restrictive nature of head movement (adjunction), and the fact that clitics form a complex head with their host (which contains the verb). In other words, the fact that clitics move along with their host, or the fact that they cannot be deleted nor be replaced independently of their host, would simply follow from their being X0 elements and the independent restrictions posited by the theory for head movemenVadjunction. As for the morpho-phonological properties of clitics, which are similar to those of affixes, this would also follow from their head status in combination with a post-syntactic morphologicaVphonological component that assigns prosodic structure and applies a number of morpho-phonological operations to the already built syntactic structure.64 Other affix-like properties of elitics (like e.g. ordering restrictions, or selectional restrictions) would follow from a combination of syntactic
63. Here, I abstract away from problems such as e.g. the violation(?) of the Uniformity Condition on chains that a movement approach would incur (recall that the clitic starts as an XP and lands as an X0). See main text below for discussion on this and other relevant issues. 64. Note that strictly Lexicallst approaches to morphology typically assume a linear, unidirectional relation between morphology and syntax, according to which morphology is an independent generative component of the grammar that operates on morphemes, yielding words, and which precedes syntax (i.e. Its output ls the input to the syntactic component). The Atomicity Thesis greatly restricts the relation between the two components, as It does not allow to the syntax to see into the internal structure of the constructed words/morphological Items. The idea that is inherently implied by this restriction is that syntax and morphology are two independent components with distinct operations and rules/constraints. Their interface takes place via features only. Presumably, in thls way the independency of morphology is highlighted: recall that in reality morphology and syntax have both common and distinct properties, so that the real picture is more complicated than what the lexicallst theories accept. Despite all this, a lexicalist approach does not in principle exclude the presence of a post-syntactic morpho-phonological component, although in such a case the Atomicity Thesls would probably have to be weakened (in the sense that It could hold but need not have to hold). In fact, Lexical Phonology was a lexicalist framework that assumed the existence of a post-syntactic phonological component, which involved a number of so-called 'post-lexical' rules. Accordingly, post-lexical rules may apply both within words and across word boundarles (i.e. they are phrasal, acrossthe-board rules). Note that although Lexical Phonology called these rules phonological, in reality many of these rules had to make reference to morpho-syntactic properties of the syntactic structure, and hence were in essence morpho-phonologicaL Taking this approach to Its logical end, one could also assume that phonological features are added post-syntactically, in the PF, even though words are constructed in the pre-verbal Lexicon. Whether something like this could work successfully is an independent question that is surely beyond the scope of this discussion.
51
52
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
and post-syntactic operations (see Chapter 4 for a detailed account of cliticization). In any case, and abstracting away from the details of the actual implementation of such an approach (for which there has been abundant evidence in the literature), my point here is that even if we adopt Lexicalism as a working hypothesis, we are not obliged to treat clitics as lexical affixes due to the fact that they have affix-like properties: a (particular) syntactic approach can do the same job, and possibly even better, judging from the empirical facts mentioned in earlier paragraphs. The third reason I follow a syntactic approach is simply because I adopt the Distributed Morphology framework (see Halle & Marantz 1993, 1994, Embick & Noyer 2005), according to which the pre-syntactic Lexicon simply contains a list of morphemes (i.e. terminal heads or bundles of syntactic/semantic (functional) features and roots) which are syntactic objects, in the sense that they are subject to the operations of syntax. Accordingly, words are not constructed in a separate morphological component, but they are created in the syntax via morpheme concatenation (basically via head movement/adjunction) and Merger (see Marantz 1988). A post-syntactic morphologicaVphonological component contains a number of relevant operations which precede or follow the so-called Lexical Insertion operation. The latter inserts phonological teatures/exponents (taken from a post-syntactic Lexicon) into the terminals/ heads of the (altered) syntactic structure. Insertion is constrained by Competition, a notion that more or less corresponds to the well-known Panini principle. As becomes immediately obvious, in such a framework the question of whether clitics are lexical affixes or syntactic words does not even arise/cannot be stated. Clitics are syntactic morphemes/heads like any other morpheme/head. The fact that they are similar to but at the same time different from other affixes/heads simply follows from their inherent properties (in featural and/or phrase structural terms) and the interaction of these features with the more general principles of syntax. The reason I put this point last was simply because I wanted to show that a syntactic approach is favorable independently of the particular theory of the syntax-morphology interface one adopts.65
2.3 A dynamic approach to the structure of personal pronouns
As far as the morpho-syntactic status of clitics and their internal structure is concerned, a plausible idea could be that the clitic starts off as a cpP (cf. Dechaine & Wiltschko 2002) but only becomes a DP when it reaches its verbal host, i.e. v""-T (see also Chapter 4). More particularly, the ditic would merge with aD- head/feature in the v*-T position, becoming a proper D head. The D-head/feature in this position would be related to [-focus ]/topicality, accounting for the derived scope of cliticization 65. Of course, any differences and/or simllarities between ditics and agreement markers at all grammatical levels (i.e. syntactic, semantic, morphological, phonological, pragmatic) would still need to be accounted for in a principled way, as in any other theory.
Chapter 2. The properties of Greek clltics and their structural analysis
in Greek. Although this is quite an insightful idea and one which I am going to use later on, the main problem with it as it stands here is that the clitic and the doubled DP do not share the same features although they are in an agreement chain (see Sufier 1988).66 This is so, since the clitic is a cpP, while the doubled phrase is a DP. This means that as long as DPs get doubled by clitics, the latter cannot be cpPs; rather, they must be DPs, so as to match their doubled DPs in features. So, how should we proceed? Here, I will follow two general approaches, namely the dynamic approach to pronoun construction and the fragmentation approach to pronoun construction. According to the dynamic view, pronouns are constructed on line out of features/heads, which are combined together by the derivational system in the spirit of Kratzer (2008). The fragmentation view, on the other hand, argues that for example direct object clitics are subject to distinct restrictions from indirect object clitics both within and across languages and as a result they should be analyzed separately (see Anagnostopoulou 2006 citing Bleam 1999). Starting from direct object clitics, I will assume that 3rd person clitics have gender features as their basis (cf. Kratzer 2008 ). Gender is number neutral it denotes a property. and it is a descriptive feature. Syntactically it resides in the nominalizer n, and on its own it cannot form a pronoun, since it is of type <e,t>. Gender combines with number, which is an operator, with its own syntactic projection, as well as with [def] which resides in D. [Def] turns it into a definite description, therefore into type <e>, and hence into a proper pronoun. In other words, 3rd person clitics do not have any lexical content (root), they only have AgrN (i.e. nand num) and D67:
(56)
D
~ NumP
D [def]
~
NUM
nP
[sg, pl] n
[GEND] 66. There is plenty of evidence that there is an agreement relation between the clittc and its doubled DP in clitic doubling constructions. For instance, the feature sharing property is characteristic of agreement chains. Moreover, the strict locality that characterizes these chains, as well as the fact that there is no Principle C violation by the clitic, and the fact that the clitic position offers a new scope and binding position for the doubled DP, speak in favor of an A- agreement chain. 67. In fact, Kratzer (2008) asswnes that the nwnber node [plural] c-commands the D-node, being an operator that pluralizes its DP-sister constituent Here, I will follow Ritter (1995) and assume that nwnber heads its own projection in syntax, which is lower than D. Also, note that in case of bound variables a nwnerical feature would be present in the structure, in the form of an 11 (or, alternatively, it would be located inn, where gender also resides),and it would be of type <e>.
53
54
Cities in Greek: A Minimalist Account ofProdisis and Endisis
As for 1st and 2nd person clitics, I will assume that these have participant features, 1st and 2nd person, which are of type <e>, therefore they cannot combine with [def]. As a result. they are directly referential, referring to the speaker and hearer respectively. They also contain the group feature, and/or number, although not gender, since gender cannot combine with features such as speaker/hearer (the result would be a pronoun of type
).68 Number does not project a phrase on its own; rather, it combines with person to give the interpretation of the intersection of two sets, namely the set of number and the set of person (see also Anagnostopoulou 2003a: Chapter 5). As Kratzer herself points out (see Kratzer 2008: 66), morpho-phonological reasons may require that 1st and 2nd person pronouns are Ds,69 even though they do not require a [Def] feature semantically. In that case they would also end up being D Ps. DP
(57)
~ PERP
D
~ NUM PER [sg, pl] [1, 2] As far as indirect object clitics are concerned, for the time being I will assume that they are also DPs, on a par with direct object elitics. 70 The main reason for this is that on the one hand 3rd person clitics are identical to the definite determiner (besides the genitive plural) and that on the other hand both 3rd as well as 1st and 2nd person clitics obviate locality effects because they check the D feature off an EPP-T head (see Anagnostopoulou 2003a). I will present more evidence in favor of the DP-status of indirect object elitics later on in the following chapter as well as in Chapter 4. The immediate question that is raised here is about binding theory. That is, if clitics are also DPs, why are they not subject to Principle C? Well, the simplest answer to this question is because they do not have anN PI root complement, i.e. they do not have a nominal range, which is in essence the difference between a pronoun and an R-expression. As for the D-head of the elitic, borrowing the 'words' from Borer (2005), it is
68. Note, however, that this seems to be only a tendency, as we find languages where 1st and 2nd person pronouns are marked for gender (e.g. Semitic languages). Kratzer (2008) argues that in these cases descriptive features such as gender only contribute conventional implicatures. 69. Evidence for the presence of D in 1st and 2nd person pronouns may be cases ofN-to-D raising (see Longobardi 1994). Elena Anagnostopoulou asks how expressions like 'we linguists' are to be analyzed according to this analysis. I assume that in that case the structure in (57) takes a complement nP phrase that contains the noun linguists. 70. The question arises whether the 3rd person indirect object dttic has person features or not
Following Anagnostopoulou (2003a), I will assume that it does have a person feature, which it checks against v*. This accounts for the PCC effects ofiii-1/II clitic orders.
Chapter 2. The properties of Greek clltics and their structural analysis
only an open D 'range/value' that is contained in the D-head. Range to this value will be assigned externally via EPP-checking with v*.
2.4 Conclusions
Summing up what I have said so tar, in the preceding sections I presented the main properties of strong and clitic personal pronouns in Greek, following criteria and observations made in Kayne (1975) and Cardinaletti & Starke (1999). I showed thatfull personal pronouns are strong pronouns (in the sense of Cardinaletti & Starke 1999), while weak pronouns are clitics. I also discussed the morpho-syntactic structure of clitic pronouns, their phrase structure properties as well as their categorial and featural properties.
55
CHAPTER3
Internal structure of elitics and diticization
3.1 Introduction
Within the generative literature, the investigation of structural and other properties of personal pronouns has a long tradition. Already before the '70's (see e.g. Postal1969), linguists have been investigating such properties as the categorial status of pronouns (e.g. whether they are nouns or determiners), their phrasestructure status (e.g. whether they have simplex or complex internal structure), or their exact feature specifications (both lexical and/or functional). As we have already seen above for Greek (and which also appears to be a valid generalization at the cross-linguistic level), personal pronouns differ in some important respects, although they also appear to share a number of properties. For example, pronouns may be overt or covert, they may have distinct binding properties, they may occupy distinct positions within the clause, or they may contain/realize distinct sets of features. On the other hand, the majority of pronouns do not involve an overt nominal root, which would explain why pronouns do not denote a lexical concept, although they typically denote an individual and pick up their reference either through syntactic binding or via co-reference (see Grodzinsky & Reinhart 1993 on how this could work; see also Abney 1987, and Panagiotidis 2002 for a more general discussion). Moreover, when they are overt they are typically marked for different types of features, such as e.g. phi-features or case, although neither the quality nor the quantity or combination of these features has to be the same both within and across distinct types of pronouns (within a single language or across languages).
3.2 Analyses proposed for personal pronouns
The different major analyses that have been proposed in the literature for personal pronouns and for their various sub-types can be distinguished into two basic types. Type A analyses assign a DP status to pronouns (with distinct structures in many cases both within and across pronominal types), while type B analyses assign a different categorial status to different pronouns (e.g. DPs, q>Ps, etc.).
58
Cities in Greek: A Minimalist Account ofProdisis and Endisis
3·3 Type A: DPs The first type ofanalysis analyzes different types of pronouns as determiners which project a nP. This type can be further divided into three sub-types: (a) (some or all types of) pronouns are intransitive n°-heads projecting a nP (which is merged in a 9-position); (b) (some or all types of) pronouns are transitive n°-heads, taking at least an NP complement and projecting a nP (which is merged in a 9-position); (c) clitic pronouns are n°-heads merged in designated clitic heads within the clausal functional projection. In this respect they strongly resemble other clausal heads (and agreement heads in particular). 3.3.1 Sub-type (a): Pronouns as intransitive D 0-heads projecting a DP
In the '80's Abney analyzed pronouns as intransitive n°-heads that project a n P; when they co-occur with an overt nominal head, an NP is also projected. The same analysis is given for other (optionally) intransitive determiners. This is illustrated below (specifiers omitted): (1) a.
nP
b.
nP
~ no
no
NP
~ we/he/ she/those/these
we/those
linguists
Although Abney does not specifically talk about clitics, neither does he distinguish pronouns into different types, many researchers in the clitic literature in the '80's and the '90's simply assumed that clitics are intransitive nPs, on a par with strong pronouns.1 Some of them, however, tried to link the special distribution of elitic pronouns to the very fact that they were analyzed as intransitive nPs. For instance, Rizzi (1993) argues that clitics are special intransitive determiners (projecting a nP), 2 which are merged in the complement position of the verb, and which contain NP- features that can only be checked nP-externally by the corresponding features on V. This is so, t. In reality, a lot of researchers were not really interested in the internal structure of ditics; they either didn't specify at all the categorial and internal structure of ditics, or they simply ass wned a DP analysis without any further discussion. Later on, when the distribution of ditics as compared to that of strong or weak pronouns was linked to their internal structure, the latter became much more cruciaL 2. Rizzi analyzes clitics as a (type of definite) determiner, mainly based on a number of similarities that have been noticed in the literature and which involve different grammatical levels (e.g. morphological, semantic or syntactic similarity). Although he acknowledges the fact that clitics and determiners are not identical (see main text below, and IDeam 1999 on this issue), he assumes that their differences lie in their featural make-up, rather than their categorial or phrase-structure properties.
Chapter 3. Internal structure of ditics and cliticlzation 59
because clitics do not take an (overt) NP complement. which would check its NPfeatures DP-internally. Furthermore, the reason this checking must be overt is either linked to the assumption that clitic features are strongl (which, in turn, is linked to the observation that Romance clitics overtly realize case and phi-features (and possibly other features) which are not realized by other determiners within a full noun phrase), or alternatively to the assumption that items with no lexical content are not visible at LF (an observation that is attributed to Chomsky 1993). In any case, the argument is that clitic placement is forced by the lack of a lexical NP complement within the cliticDP.4 Accordingly, the exact derivation of a monotransitive accusative clitic structure like la conosco ('I her know') would involve overt movement of the whole DP to the specifier of AGRoP for case checking (a 'V-related landing site, as Rizzi calls it), followed by movement of the D/clitic head out of the DP to an immediately higher recursive AGR head, AGRcl.s,6 3· The idea that ditic movement may be linked to the strong case features of ditics (as opposed to full DPs and other DP pronouns) has also been proposed byBelletti (1999). According to her analysis, clitks in Romance are intransitive D0-heads that move from their theta-position (through the specifier of AGRPastParticple) to the AGRo projection (either as XPs or as heads, given their structure and depending on the particular language), after which they move/adjoin to V in AGRs (for PF-reasons, namely because AGRo is weak and hence cannot host a strong case element). 4· In filet. although initially Rizzi assumes that clitks are intransitive D0-heads, later on he points out that a transitive analysis according to which the ditic takes a null nominal pro (see Uriagereka 1995, Corver & Delfitto 1993/1999), would be equally fine, if one assumed that such a noun cannot check the ditic features DP-internally. Since LF movement is contingent on lexical content, and given that ditics move because they do not have such content, it follows that his assumption about the null pronominal NP complementing a clitic could only work if lexical content received an abstract interpretation. The question, of course, is why we would need to postulate such an element if an .Intransitive determiner analysis would give the same results (although see Panagiotidis 2002 for arguments in favor of the null pro; see also the discussion in the main text below). Moreover, given that under current minimalist assumptions case/phi-feature checking can also take place by means ofLong Distance Agree (see Chomsky 2001a), which does not require LF- (feature) movement (if such a movement exists at all), it is not clear why the clitic would be required to move in the first place, assuming the above analysis. Finally, note that other intransitive determiners (such as e.g. demonstratives) may appear in situ, without any problems (and the same applies to intransitive quantifiers like olos ('all') in Greek, as opposed to French tout)). Although Corver & Delfitto (1993/1999) argue that these have a deictic element which licenses the null pro NP, it is not clear how such an analysis would work for Rizzi, given that the particular lexical content (= deixis) must be a property of D, and not of the pronominal noun. 5· Note that the second head movement step presumably takes place in order to account for the head properties of clitics (e.g. the fact that they act like heads in their landing site- see Kayne 1989b, 1991). As I will discuss in the next chapter, this local head movement was initially proposed in Sportiche (1990) and Roberts (1993) in order to account for the apparent head movement violations invoked by diticization. 6. Cardinaletti (1994) and Laenzlinger (1998) follow a similar line of analysis.
6o
Cities in Greek: A Minimalist Account ofProdisis and Endisis
a.
b.
c.
Rizzi's proposal, however, faces a number of problems: The proposed phi-/case-feature checking of the intransitive clitic D 0 -head by an appropriate external V- head (here V = part of the verbal extended projection, such as AGRo or v) should not be possible according to the definition of Agree and the standard minimalist assumptions on feature interpretability, as presented from Minimalist Inquiries (MI) (Chomsky 2000) onwards? This is the case, because the appropriate verbal head contains uninterpretable features, and as a result requires a set of interpretable phi-features in order to value its own features and assign (structural) case. Since the intransitive D 0-head does not contain interpretable phi-features (a standard assumption in minimalism, and the main motivation for elitic movement according to these analyses), it follows that Agree should fail, and hence no case would be assigned to the clitic DP (leading to derivational crash). 8 If phi-features on the D0 -head are uninterpretable, it would be expected that they get deleted and erased after checking, following standard minimalist assumptions (see Chomsky 1995 ). This would predict, among others, that they are not visible to LF. However, this is empirically incorrect, since in many (although not all) instances phi-features on clitics are semantically interpretable.9 One way of avoiding the problem described in (b) above would be to assume that in some way checking recovers the interpretable/semantic content that the clitic head inherently lacks. However, it is not at all straightforward how this result could be achieved via checking, at least under standard assumptions. Moreover, even if checking may lead to recoverability, the latter is not a panacea: it would lead to a uniform semantic interpretation of clitics, which is not what we find
7· Note that within the MP (=Chomsky 1995) model, checking of the phi-feature set on the intransitive D0-head by an appropriate verbal head (such as e.g. AGRo for Rizzl. Cardinaletti and Laenzlinger) would be possible in princ.lple, since according to this theory syntactic heads enter the derivation fully inflected, i.e. with values. One could assume that checking between these two phi-feature sets in a spec-head configuration would suffice to license/check the features on the intransitive D0-head, even though both sets involve uninterpretable phi-features. Note that - as Holmberg (2005) has correctly pointed out - a similar problem arises for subject pro under current minimalist assumptions. This follows from the GB analysis of pro (see Rizzi 1986), according to which pro must both license and identifY its phi-feature content with that of an appropriate governo.t: Put in minimalist terms, pro would be taken to contain a phifeature set that needs to be checked by phi-features on T, even though the latter are also uninterpretable. Hence one would have to assume that pro contains interpretable features after all, as any other pronominal does. 9· Obviously, similar problems would arise even if we adopted the by now widely accepted assumption that D contains both interpretable and uninterpretable phi-features (see e.g. Ritter 1995), since the latter would still be required to survive by LF, contrary to what the theory prescribes. Note that the postulation that uninterpretable features could optionally/in some cases be visible at LF would seriously undermine the basic tenet of minimallsm, namely that LF and PF interpret the syntactic derivation, which therefore has to satisfY their particular requirements. 8.
Chapter 3. Internal structure of ditics and cliticlzation
d.
empirically (cf. the discussion below regarding animacy/human features in clitics). As a result, what would be required is an operation that gives rise to purely formal recoverability with or without concomitant semantic repercussions, a result which is not easy to obtain, especially within theY-model of derivation (see Chomsky 1995). Finally, even if (feature) deficiency were connected to (feature) recoverability, the latter being presumably attained via checking/movement of the clitic to an appropriate verbal host, it is not at all clear why such a movement would have to be overt: it is easily conceivable that checkinglrecoverability takes place either at LF or by Agree (following MP or MI/Derivation by Phase (DbP), respectively). Rizzi argues that overt movement is forced either by strong case features on the elitic or due to the lack of any lexical content/an NP complement (assuming that it is a prerequisite for covert movement). Nevertheless, both arguments are seriously flawed. On the one hand, the idea that eli tics contain strong case features (and hence move overtly to their verbal host/head, given the early minimalist assumption that rich morphology somehow correlates with strong features and therefore with overt movement (see Chomsky 1993, Roberts 1993, Rohrbacher 1994 and Vikner 1995)) is based on the empirical fact that clitic pronouns in Romance inflect for case as opposed to strong D Ps/pronouns, which do not. Although this may be true in most Romance languages (although not all of them, as Romanian, e.g., has kept case distinctions across the whole (pro )nominal paradigm but has the opposition clitics vs. strong pronouns), it definitely wouldn't work for Greek, which marks case across the whole nominal paradigm: Tu ipa. (2) a. Ton idha vs. Him.cl saw.1sg vs. him.gen.cl said.3sg 'I saw him: vs. 'I told him: b. Idha ekinon/ton Petro. vs. Ipa ekinu/tu Petru. Saw.lsg him.str/the Peter.acc vs. Told.1sg him.gen.str/the Peter.gen 'I saw him/Peter: vs. I told him/Peter:
In (2) we see that even though (strong) case morphology is marked on clitics, strong pronouns and full D Ps, it is only clitics that appear on the left side of the verb. As for the second proposal, namely that clitics have no lexical content and hence would be invisible for LF movement, it hinges on Chomsky's (1993) reasonable speculation that semantically empty constituents are not visible to LF rules such as movement, and therefore they are forced to move overtly. 10 Putting aside the obvious problem oflookahead that this explanation is facing, the main issue is that (at least some) clitics are not semantically vacuous elements. For instance, they contain interpretable phi-features, such as person (deictic category) and number, semantic features such as specificity or A typical example of such a constituent are French and English auxiliaries which have to move overtly, being semantically vacuous (see Chomsky 1995: 198).
10.
61
62
Cities in Greek: A Minimalist Account ofProdisls and Endisis
animacy and definiteness, as well as features such as topicality. All these would surely count as semantic content and hence would secure their visibility at LF and as a result their possible LF movement For this reason, the attribution of cliticization to the lack ofNP or oflexical content does not preclude the possibility of clitics moving at LF, and therefore does not explain why clitics must move overtly to their verbal host The second important issue that Rizzi's approach has to face and which it shares with transitive D0-approaches is the issue of recoverability, namely the idea that elitics are lacking something (features that are syntactic or semantic in nature) that needs to be recovered via cliticization. The problem with that is that it would irrevocably link cliticization with particular syntactic/ semantic properties that clitics would have to possess in a uniform way, since cliticization would lead to recovery of the missing features. As a result, we would not be able to find clitics with uninterpretable (i.e. deleted at LF) phi-features, or with no definiteness feature, or no person feature, because these would always be recovered due to cliticization. Unfortunately, this is not what we find empirically. For this reason I think that the notion of recoverability is problematic, at least in the way it is used by the intransitive D0-approaches. A more recent incarnation of the intransitive D0-approach to the internal structure of clitics is that of Chomsky (1995) .11 He points out that his revised Bare Phrase Structure Theory makes the prediction that an item that is both minimal (i.e. an X0) and maximal (i.e. an XP) is in principle possible (cf. Muysken 1982), and that clitic determiners in Romance languages are one such example. Let's see how this is possible. According to Chomsky (1995), bare output conditions (i.e. minimalist assumptions) make available to the computational system only minimal and maximal projections (and the features contained in these projections). Moreover, the status of projections cannot be determined absolutely (as some kind of inherent feature), but only relationally, i.e. in relation to the structure they appear in. According to this theory, "a category that does not project any further is a maximal projection XP, and one that is not a projection at all is a minimal projection xmin; any other is an x: invisible at the interface and for computation". (Chomsky 1995: 242-243).12 Following this line of reasoning, before merger the clitic pronoun is both minimal and maximal, and the same applies to the verb (since on the one hand they have not projected at all, while on the other they do not project any further). Upon merger, V projects, so that the initial V becomes u. Although Chomsky (1995) does not discuss strong pronouns specifically, presumably he would treat them as projecting DPs. Note also that Phillppaki-Warburton et alios (2004) analyze Greek clitics as DPs/D 0s, following in this Chomsky (1995). A somewhat different, though essentially similar, approach to phrase structure ls offered by Speas (1990). Note that according to Chomsky ( 1995, 2000, 2001 a, 2001 b, 2005) phrase markers (Le. syntactic objects) are constructed by the generalized operation Merge, which ls necessary on conceptual grounds, since recursion is a unique characteristic of natural languages. Merge takes two syntactic objects a and j3, and forms a new object K such that K = {y {a, ~}}.where y ls the label of the newly formed object. Under standard assumptions, y is either a or~. Le. the label is the head of a or ~12.
Chapter 3. Internal structure of ditics and cliticlzation
minimal, and the projected one maximal, while the clitic remains minimaVmaximal. This is illustrated in the following tree diagram (English glosses are used here- the tree diagram is taken to be a notational variant of a set theoretical operation): (3)
see
~ see
it.cl
Movement of the clitic to some higher position within the clause does not change its maximaVminimal status, given that it does not project. According to Chomsky, this fact trivially captures the mixed X0/XP properties of cliticization, namely the fact that they are initially merged in the complement'e-position (a phrasal category by definition in the X-bar theory), although eventually they move across lower head positions, adjoining in their final step to a higher functional head within the Extended Verbal projection. All in all, then, Chomsky>s approach differs from the previously mentioned analyses in that he links the intransitive DP structure of clitics to their mixed phrasal status properties, and not to the actual motivation of clitic movement.B 3.3.2 Sub-type (b): Pronouns as transitive/complex DPs
A number of researchers have argued against the intransitive n°-hypothesis. Although they do not always share the same assumptions or the same goals, nevertheless they all argue for the basic thesis that (at least some) pronouns (strong, weak or clitic ones) are DPs with a complex internal structure, which involves at least two heads, and which may or may not be shared among all different pronominal types. Accordingly, two further sub-types may be distinguished within the transitive n°-approach: (a) pronouns, more generally, and/or clitics in particular, are all transitive DPs, however they differ in terms of their internal structure and/or their features (this includes both differences across types and within types (as e.g. different types of clitics)) (see e.g. Cardinaletti 1994, Laenzlinger 1998, or Uriagereka 1995, Co rver & nelfi tto 199 3/1999 ); (b) all pronouns have exactly the same internal structure and/or features (e.g. Panagiotidis 2002). 3.:3.2.1 Uriagereka (1995): Pronouns as tra.nsiti11e- DPs which
differ in their internal structure Here I will discuss Uriagereka (1995), who argues for a separate structure for 3rd (i.e. weak) vs. 1st and 2nd (i.e. strong) person direct object'accusative clitics in Romance 13. Note, however, that this mixed phrasal status may be responsible for ditic movement.
E.g. Chomsky (1995) points out that clitic movement may be related to linearization requirements, namely to the fact that clitics and their selecting predicate do not form an anti-symmetric structure. In this respect see also Moro (2000).
63
64
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
languages. 14 His main goal is to account for the movement properties of clitics and their specific landing site (which may be different depending on person as well as on the particular language), while also accounting for the possibility vs. impossibility of doubling in a number of Romance languages. Following the insights of Postal (1969) and Torrego (1988), he argues that 3rd person clitics are determiner heads taking a null NP complement and potentially a specifier, in which a doubled DP may occur in certain languages. In this respect, definite articles and clitics become structurally similar, and the DP-hypothesis is extended to 3rd person clitics (see Uriagereka 1995: 81): (4) a.
b.
DP
~ (double) D'
DP
~ D' Spec
~ N
D
~ NP
D
I
6.
I
6.
clitic
pro
def article
lexicalN
Uriagereka's claim that 3rd person clitics are determiner heads is based: (a) on the fact that both determiners and clitic pronouns have emerged from the Latin accusative demonstratives illum, illa.m, illud; (b) that synchronically they form identical or almost identical paradigms; and (c) that in some languages they may share more properties, such as e.g. in Galician, where definite articles may (optionally) cliticize from an NP onto a c-commanding verb. Furthermore. the existence of a null pro NP is supported by the fact that in some Romance languages a definite article may license a null pro N P which is modified by a relative clause or a PP (as e.g. the Spanish la que vino = the.fem who came.3sg). The same structure with an overt NP complement is found in English (cf. e.g. the one who came), and in fact it was used as an argument by Postal to propose that the one = he. Uriagereka argues that 3rd person clitics move to a VP external site in order to identify pro: because they lack person features (being 3rd person determiners- see Carver & Delfitto 1993/1999 for the licensing of pro, and Benveniste 1971 on 3rd person), they cannot license the pro NP complement on their own (contrary to strong pronouns which have person features and hence may do so). As a result, (only) elitics must target the p0 head (in those languages where F0 is active), which encodes the speaker's point of view. Since person features are standardly linked to referentiality, his claim reduces to saying that pro forces a referential reading on the determiner when the latter is related to a person bearing item. The syntactic/ structural expression of this relation (i.e. movement), as well as the actual mode of movement (e.g. head movement or 'pseudo' -affixation) can be parametrized depending on the language (e.g. on whether 14. Urlagereka does not discuss dative ditics, however he points out (Uriagereka 1995: 87) that they are agreement heads, on a par with AGRs and AGRo (following in this Sportid!e 1992/1998).
Chapter 3. Internal structure of ditics and cliticlzation
or not F is present or active). Being VP-external by LF, 3rd person clitics are interpreted as specific DPs, since they fall under the Restrictive Clause of the Mapping Hypothesis. In this way he captures the fact that 3rd person direct object clitics in all languages move outside ofVP and are necessarily interpreted as specific. As for 1st and 2nd person (strong15) clitics, Uriagerek.a argues that they are minimaVmaximal DPs which may take an adjoined doubled DP: (5)
DP
/"--.... (double)
DP/D clitic
These move/ adjunction-scramble as XPs to AGRoP, from which they may further head-adjoin to F0 (in those languages where F0 is active; head-movement may be related to PF reasons). Strong clitics do not have a (null) pro, but are inherently referential due to their person features. In this way, Uriagereka can account for a number of differences between 1st/2nd vs. 3rd person clitics (e.g. the former typically occur before the latter, and they are not necessarily doubled when followed by a strong pronoun). All in all, tor Uriagereka the combination of F0 (and its distinct properties across languages) with the internal structures of clitics (depending on person), which appear to be shared among Romance languages (at least at the synchronic level), gives rise to distinct placement possibilities both across languages and within a given language. 16 However, there are some problems with this type of analysis, which pose serious challenges on both theoretical and pragmatic grounds: (i) First of all, it uses the standard arguments in favor of the postulation of a null N-pro within the clitic, namely the Extended Projection argument (i.e., functional categories must close off lexical categories- and see Panagiotidis 2002 and discussion below), the DP hypothesis (i.e. the hypothesis that DPs contain a D 0 head taking an NP complement- see Abney 1987), and the parallelism between elliptical DPs and pronominals.
15. Note that Uriagereka argues that weak ditics (i.e. 3rd person ones) move as heads directly to F0 (or to AGRo in French), as opposed to strong clitics (i.e. 1st/2nd person ones) which move
as DPs to the periphery of AGRoP, from whence they may move to the F0 head (in some languages), being maximal/minimal categories. Uriagereka (1995: ft. 3) points out that weak ditics are dependent heads, while strong clitics are independent heads. Strong pronouns are independent non-head elements. 16. Bleam (1999), following Uriagereka, argues that direct object Spanish ditics are determiners. On the other hand, following a fragmentation approach to cliticlzation (see Anagnostopoulou 2006), she claims that dative ditics are agreement markers heading an applicative-type/dative clitic type head, which gives rise to the double object construction (see Demonte 1995).
65
66
Cities in Greek: A Minimalist Account ofProdisls and Endisis
However, the Extended Projection argument does not go through. On the one hand, the Extended Projection theory, at least as formulated by Grimshaw (1991), does not require functional heads to take complements, or lexical heads to be complements of functional heads. Rather, it defines the possible combinations of functional and lexical heads on the basis of an algorithm that handles categorial features such as N or V, and values thereof. Moreover, under a constructionalist realizational approach to morphology, such as e.g. Distributed Morphology, insertion of a lexical item into a functional head is possible, without requiring the presence of a lexical root. This is empirically supported by so-called light verbs, or other grammaticalized elements that do not contain any lexical (in the sense oflexical vs. functional) content. Second, the DP hypothesis proposes that determiners are separate heads projecting a functional phrase within the DP, but does not have anything to say about whether determiners do or do not take a lexical complement, since this is an independent issue. Finally, with regard to the parallelism between elliptical DPs and (strong) pronominals, although it is true that in languages like English both constructions appear to contain a covert NP (which may be the result of PF deletion, or be simply merged as such from the Lexicon, depending on the analysis adopted- see Lobeck 1995 and Merchant 1999), itis not clear whether all kinds of pronouns contain such a phrase. For example, we have already discussed some empirical evidence that no N-head is present in weak pronouns. As a result, although theoretically speaking the postulation of the same structure for all D Ps (both pronominal and non-pronominal ones) may look parsimonious in the first place, it is not always supported empirically, neither is it theoretically necessary. Also, the null N-pro does not seem to have the same properties in elliptical DPs and pronominals, a difference which would be unexptected under this theory17: e.g. while elliptical DPs cannot take a c-commanding antecedent (i.e. they must refer to a separate entity from the one introducing the corresponding non-ellided DP, behaving in this sense as an R-expression), pronouns may do so in most cases (see also Panagiotidis 2002 and Anagnostopoulou 2003b): (6) Xtes pira ena prasino forema stin Eladha ala telik.a Yesterday bought.l sg a green dress to-the Eladha, but finally to jirise piso/ala te1ika ithele to kokino mprosta sti vitrina it.cl returned back/but finally wanted. 3sgthe red in-front to-the window. 'Yesterday, I bought a green dress for Eladha, but in the end she took it back to the shop/ she wanted the red one in the front of the shop:
17. Possibly, one could attribute these differences to differences in the featural content of the N-pro (e.g. presence vs. abser1ce oflexical features sud! as human), however it is not dear to me how exactly this would work. Another possibility would be to assume a separate derivation for ead! analysis (e.g. PF-deletion for elliptical DPs, and merger of a null N-pro for pronominals), nevertheless this would violate both the letter and the essence of the Co rver & Delfitto argument
Chapter 3. Internal structure of ditics and cliticlzation 67 In the above example a clitic pronoun may refer back to 'the specific green dress I bought for Eladhi, while the elliptical D P may only refer to a different red dress that is in the front of the shop where 1 bought the green dress. In addition, elliptical DPs and pronouns are subject to different licensing conditions: the former must be semantically licensed by a local antecedent (namely a DP containing an overt lexical N-head (i.e. a root)), and also formally licensed by means of an appropriate antecedent within the DP itself. Pronouns, on the other hand, may but do not have to be semantically/syntactically bound (i.e. have a c-commanding (A-) antecedent), although they must be assigned a reference value via a contextual assignment (which will be extra-linguistic/discourse centred ifno semantic binding has taken place). However, pronouns do not require a formal antecedent within their structure, or if they do, as some theories have argued for some kinds of pronouns (e.g. clitics ), the relevant feature involved is a lexical one([+ human]), or, according to some analyses, a [person] feature (for 3rd person clitics - see Uriagereka 1995), specificity or referentiality, but not number or case, which are some of the typical formal features related to N P-pro licensing (see Lobeck 1991, and Alexiadou, Haegeman & Stavrou 2007 for references): (7) a.
*the red(-s) vs . ..J the red ones vs . ..,J these/that
b. ..ta_ vs . ..J ta kokina_ ('the.neutpl vs. the.neutpl red.neut.pf) vs . ..J afta ('these') c.
To aghorasa xtes. ltd bought3sg yesterday 'I bought it yesterday:
(7a) shows that in English the elided N P may be licensed by an antecedent with number features (which are carried by one or by a determiner, since adjectives do not inflect), (7b) shows that in Greek an (agreeing) adjective is required to license an elided NP (note that it carries a full set of phi-features- see e.g. Lobeck 1991 for English, Giannakidou & Stavrou 1999, Alexiadou, Haegeman & Stavrou 2007 for Greek), and (7c) shows that no antecedent within the clitic phrase is required. A semantic antecedent is offered via contextual information. (ii) This approach links eli tic movement to two basic factors: (a) the presence of a null NP-pro which needs licensing; (b) the lack of certain features on the D0-head, which therefore cannot license the NP-pro DP-intemally. For Uriagereka the relevant feature is [referentiality] (which is ultimately linked to person). The main idea behind this proposal, namely that clitic deficiency is linked to clitic behavior, is intuitive and empirically promising, since it allows it to account tor basic differences between strong and clitic pronouns. However, because it typically defines clitic deficiency in terms of one or more particular features, which are not necessarily shared by all clitics types, this analysis fails to predict that these types pattern on a par with respect to their syntactic position and other common properties. Uriagereka
68
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
argues that movement of 3rd person clitics takes place in order to license the NP-pro, since the clitic determiner does not contain a person feature that would render pro (and therefore the whole DP pronoun) re.terentia1. 18 The problem with this analysis is 18. The obligatory external-VP position of ditics accounts for the fact that they are specific and
that non-specific ditic determiners are impossible cross-linguistically. Note, howeve~ that specificity cannot be related to the obligatoriness of ditic movement: ditics are definite (i.e. strong) determiners, however strong determiners do not have to move out ofvP (overtly) in order to receive a specific/strong/familiar reading (see Corver & Delfitto 1993/1999, de Hoop 1992, 1996, 2000 and van der Does & de Hoop 1998, pace Diesing & Jelinek 1995, Diesing 1996 (although see Diesing 1997: 419 :If.)). For example, de Hoop (1992) argues that the following strong DP may receive a strong/referential reading in the absence of movement: (i)
alle krakers heb gesproken. dat ik gisteren that I yesterday all squatters have spoken 'that I talked to all squatters yesterday.'
Definite DPs also do not have to scramble, although they may do so (example from van der Does & de Hoop 1998 (their 39)): (ii)
dat ik (de president) gisteren that I (the president) yesterday 'that I saw the president yesterday.'
(de president) (the president)
heb gezien. have seen
Similar remarks apply to weak determiners, which may have a strong reading even within the vP (although they cannot have a weak reading when scrambled/ object shifted/doubled). The following examples are from Icelandic (taken from ThraJ.nsson 2001: 191 (his 92b) ): (iii)
a.
Nemandinn las prjar bmkur; ekki tL Student-the read three books not 'There are three books that the student didn't read (namely... ).'
b. Nemandinnhefur ekki lesil'l prjar bmkur. Student-thehas not read three books 'It is not the case that the student has read three books: or: 'There are three books that the student hasn't read (namely... ).' Direct object ditidzation has the same effect with scrambling/object shift, in that it forces a strong reading in most (if not all) cases: (iv)
Ths idha dhio fitites su na pernane apo edo. Them.cl saw.1sg two students.acc yours.d subj. pass.3pl from here 'There are two students of yours that I saw passing by.'
Note, however, that in Greek doubling is rarely allowed also with non-specific and/or non-referential (l.e. attributive) DPs, if the latter are introduced by the definite article: (v)
xaidhevo tin pjo megh.ali gh.ata Egho pote dhen ti I never not her.d petl sg the most big cat.acc (opote pao sta spitia filon mu). (whenever go.1sgto-the houses friends.gen my.d) 'Whenever I go to my friends' houses I never pet the biggest cat:
(vi)
Pandos Notwithstanding
pire to leoforio i Maria telika to the Mary.nom finally it.d took3sg the bus.acc
Chapter 3. Internal structure of clitics and cliticlzation 69 that there are 3rd person clitics which are not referential, but which appear in the same position as their referential counterparts. These include bound variables, expletive and predicative elitics, as well as direct object clitics used in idioms in Greek, French and other languages (see Bibis 2002 for extensive discussion). Uriagereka (1995: 88-89) (and also see Carver & Delfitto 1993/1999) explitly recognizes the problems posed by (some of) these clitics, however he argues that they are variables which are referentially dependent in the same way that referential and specific clitics are. 19 Although it is true that both free and bound variables pick up their referent via a semantic assignment, it is not at all clear to me why they should involve the same syntactic derivation, namely movement to the 'point of view' fO head. This is so, because free variables are assigned a referent via contextual assignment, a fact which could be linked to fO, if the latter is assumed to be the clausal centre of speech which interfaces LF with contextual (pragmatic) information. However, bound variables pick up their referent indirectly, via structural binding by a c-commanding antecedent (typically a determiner phrase containing a binder) whose index is assigned a referent via contextual assignment This would mean that they would not have to establish a direct relation with F0 (assuming the latter is the syntactic expression of context assignment, not a straightforward assumption); rather the index bearing element (i.e. the binder) would. Moreover, expletive elements may move. although they clearly do not establish any referential dependency relation; recall that according to standard assumptions in the literature
ghia na pai mexri tis Sulas. for subj. go.3sg until the Sula.gen Notwithstanding, Maria took the bus in the end, in order to go to Sula's place: In (v) and (vi) the most salient reading is a strong/referential reading, however, whenever the context allows it, a non-specific (for vi) or a non-referentiaVattributive reading for (v) is also available. However, in order for these clauses to be felicitous 'the bus' and 'the biggest cat' must have been mentioned previously in the discourse (in a non-referential and non-specific reading respectively), i.e. they must be familiar. If that is true, then one would either have to conclude that specificity and!or referentiality are not necessary or sufficient conditions for clitici.zation of 3rd person clitics, or that the clitic pronouns in these particular rare cases do not involve the same features or the same derivation with specific ones. However, the latter assumption would make it totally accidental that both ditic types have exactly the same properties, especially given the fact that different ditic types typically differ from each other in terms of features, syntactic position, semantic properties, etc. (e.g. see Grimshaw 1997). Therefore, specificity and/or referentiality cannot be triggering the actual ditic movement, neither can they constitute inherent properties of clitic determiners; rather, the exact interpretation of direct object ditics has to be related to the doubled element (overt or covert). This conclusion is also supported by dative clitics, which are not nessecarily interpreted as specific, depending on the semantics of their double, nevertheless they also appear to ditic climb on the verb (see also Uriagereka 1995 and Bleam 1999 for discussion). 19. For Uriagereka, ditics and null subject pronouns receive the same readings (i.e. bound
variables, or specific/referential) because both contain a null pro. According to him, it is pro that may be interpreted either way.
70
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
expletives do not have any semantic content but merely act as PF place holders. Movement may apply to both clitic and non-clitic expletives across languages. For instance, in English there may move to a higher finite clause, although it is nota clitic (presumably, movement is triggered by an EPP feature on T, which is obligatory in English - see Chomsky 1981 ), and the same applies to the neuter singular weak pronoun, which also moves to some higher position (see Carver & Delfitto 1993/1999:844 ff.): (8) a. There seems [(there) to be a man in the room]. b. I figured it out quite easily that you didn't want to come. Similar examples are also found with weak object pronouns in Dutch, expletive pronouns in Swedish, and with expletive clitics in Greek (tor Greek see Tsakali 2006): (9) a.
Ik heb het toen betreurd dat Jan ziek was. I have it then regretted that Jan ill was 'I regretted it that Jan was ill: (Dutch) [example taken from Carver & Delfitto 1993/1999: 844 (their 92d)]
b. Han tar det8 mycket sallan [8c t9 lugnt] He takes it very seldom easy 'He takes it easy very seldom: c.
iksera (to) oti o Janis ine plusios Itcl knew.lsg (the) that the John.nom is rich ke (to) oti i Beth ine apo tin Ameriki. and (the) that the Beth.nom is from the America.acc
(Swedish)
To20
(Greek)
All these examples show that both referential and non-referential elements may move, and in fact to the same position (both for clitics/weak pronouns and non-clitics). In addition, there are referential DPs which do not move (overtly) to F0, such as e.g. strong pronouns, full DPs and weak pronouns. The same applies to strong pronouns which are interpreted as bound variables (e.g. in environments where clitics and weak pronouns are not available for independent reasons): although they appear to be referentially dependent, they do not move. As a consequence, referentiality cannot account for all instances of 3rd person pronouns, and most importantly for the fact that they invariably appear attached on the verb in exactly the same position, even though movement to this position should be linked to referentiality. Moreover, the vP-external position of 3rd person clitics does not always entail a specific reading for them, contrary to what Uriagereka (1995) claims. Rather, specificity, as I also pointed out in footnote 18, seems to be dependent on a
20. The expletive nature of the ditic ls supported by the fact that its phi-feature values are fixed and default (unmarked), i.e. it has to be neuter, singular and 3rd person. Note that the ditic itself ls not obligatory.
Chapter 3. Internal structure of ditics and cliticlzation
combination of factors, such as the internal content of the clitic pronoun itself (e.g. on properties of its double), and the external (or internal) vP position in the clause. 21 Problems of a similar nature occur with other types of clitics. For example, 1st and 2nd person clitics contain person features. Since they are inherently referential, Uriagereka argues that 1st and 2nd person clitics are maximal phrases which do not contain a pro complement, that they scramble-adjoin to AGRoP for specificity related issues and that they may further PF-move to F0, depending on the properties of each language. However, 1st and 2nd person clitics do not necessarily have the above properties in all instances. For example, they may have non-reterential and non-specific readings, when they are used as bound variables (either as pronouns or as anaphors, depending on the language), but nevertheless they seem to appear in exactly the same position as referential and specific 1st and 2nd person clitics: (10) a.
Kathe ghineka pu gnorizo nomtzl oti mas Every woman.nom that know.1sg thinks.3sg that us.cl zilevun. be-jealous-of. 3p1 'Every woman I have dated thinks that they are jealous of us:
(Greek)
b. Me idha ston kathrefti. Me.cl saw.1sg in-the mirror.acc 'I saw myself in the mirror: Also, strong 1st and 2nd person pronouns do not have to scramble adjoin to AGRoP or to move to fO, even though they have person features and presumably no pro complement, and moreover they are also phrasal projections, like 1st and 2nd person clitics. And although strong pronouns may receive a novel reading (in which case they are typically stressed/emphasized), and hence be bound within the vP by an existential
11. That the vP-external position is involved in spectfityhas been shown by Dlesing (1992) for both subjects and objects, but also by the empirical fact that inherently indefinite/non-specific pronouns like English one or Spanlsh uno (and the same holds for Greek enas) cannot be scrambled, object shifted or doubled:
(i)
Dhen echo palto, ala tha ("to) aghoraso Not have.1sg coat.acc, but fut (it.d) buy.1sg 'I do not have a coat, but I will buy one tomorrow.
(ii)
Nel, jeg har ingen paraply, men jeg k0per (*en) muligens (V en) imorgen. No, I have no umbrella, but I buy (one) possibly one tomorrow 'No, I have no umbrella, but I will possibly buy one tomorrow: [example taken from Diesing 1997:413 (her 74-75)]
(iii)
Dat ik (* een plas) nog ( ...J een plas) moet doen. That I (a plss) still (a plss) must do 'that I still have to take a piss: [example taken from der Does & de Hoop: 1998 (their 1Oc-d)]
ena avrio. one.acc tomorrow
71
72
Cities in Greek: A Minimalist Account ofProdisis and Endisis
closure operator, they do not have to receive such a reading. 22 In fact, they may also receive a strong, familiar reading (especially when they are used in the place of weak pronouns in a number of environments where the latter are not possible for independent reasons), without cliticizinglmoving. This shows again that specificity does not trigger overt movement, although movement to a higher position typically correlates with a specific/strong/familiar reading. Consequently, it remains unclear why according to this theory 1st and 2nd person clitics must scramble-adjoin to a vP external position, when compared with the corresponding strong pronouns. Dative clitics appear to raise similar issues. Uriagereka argues that they are agreement markers, and hence directly merged on the extended verbal projection. The problem with this kind of approach is that it may be true for some languages, but not for all. For example, Bleam (1999) argues that Spanish dative clitics are agreement markers based on such evidence as their morphology, their obligatory presence in the dative alternation construction, and the lack of specificity effects (see also Jaeggli 1986, Suiier 1988, and Sportiche 1992/1998).23 Similar remarks have been made with regard to dative clitics in other languages, such as Romanian (see Dobrovie-Sorin 1990), as well as with regard to object clitics more generally in languages like Bulgarian and Albanian (see Franks & King 2000). On the other hand, in Greek, for example, dative clitics are morphologically similar to the corresponding definite determiner forms, exactly like direct clitic objects. Moreover, their presence does not correlate with the double object construction (see Anagnostopoulou 2006), and as a result their presence does not seem to be obligatory (in most cases- see Anagnostopoulou 2003a, 2006 for an extensive discussion of those environments where dative cliticization is obligatory). Rather, they mark their double as topical (i.e. old information) and familiar/prominent (see Anagnostopoulou 1999, 2006), which is also the function of accusative clitics. Finally, although dative clitics are compatible with certain types ofDPs which are either ungrammatical or less grammatical when combined with accusative (determiner) clitics, the presence or lack of specificity effects (or definiteness or referentiality effects according to other researchers - see Anagnostopoulou & Giannakidou 1995, Anagnostopoulou 1999 for arguments in favor of a referentiality approach to Clitic Doubling on a cross-linguistic level24 ) cuts across the clitic-agreement distinction. For This does not have to be the case in all languages which have strong and weak/ditic pronoun series. For example, Greek and Spanish appear to allow strong, non-focused personal pronouns (see e.g. my extensive discussion on Greek personal pronouns), as opposed to French, which generally does not allow for this option (besides some well defined environments (e.g. when strong pronouns obviate a person-case constraint - see discussion in Rezac 2007)). 23. See also Roca (1992) for a detailed presentation of differences between accusative and dative ditics in Spanish, which support the assumption that dative ditics in Spanish are agreement markers. 22.
24· Note that different researchers use notions such as definiteness, specificity and referentiality in distinct ways (e.g. semantic or morpho-syntactic), a fact that may cause some confusion. This confusion is made worse by the met that these notions only partially overlap with each other, which makes a
Chapter 3. Internal structure of ditics and cliticlzation
example, definite/specific direct objects may be marked by different strategies or a combination thereof in different languages. These strategies include clitid:zation, case marking, movement or agreement Object agreement is used for example in Hungarian in order to mark definite/specific accusative objects, which project the highest D position within the DP (the latter is related to deixis, referentiality and/or specificity -see S:zabolcsi 1994 and Alexiadou, Haegeman & Stavrou 2007, and Bartos 2001 for Hungarian). Palauan, Swahili, and Hebrew are also languages which use agreement in order to mark definite, spedfic or topical/familiar accusative objects (see Danon 2001 for Hebrew, Georgopoulos 1991 tor Palauan, Seidl & Dimitriadis 1997 tor Swahili). It appears, then that specificity/definiteness is neither a necessary nor a sufficient condition for agreement status. This is so, because on the one hand specificity/ definiteness may be marked via agreement (i.e. it is not true that agreement is always insensitive to specificity/definiteness), while on the other hand clitics may not mark specificity/definiteness (as it appears to be the case with Greek datives). As a result, specificity/definiteness differences between accusative and dative clitics are not sufficient to establish a morpho-syntactic distinction between the agreement and determiner-head status of a eli tic; additional criteria are required. Now, if dative clitics in Greek are not agreement markers but determiners, it is not at all clear how Uriagereka's account could explain their movement properties and clausal position. (iii) The third major point with regard to the transitive D0 analysis is the issue of feature recoverability and its relation to overt clitic movement Uriagereka argues that the pro NP complement cannot be licensed by the D 0-head, since the latter lacks lexical or person features (i.e. features that are presumably related to the semantic properties of the clitic, such as referentiality and± human interpretation). Movement of the clitic to the specifier position of a head that belongs to the extended verbal projection allows the D0-head (via checking) to recover the missing feature that will license pro.
The theoretical correlation between clitic movement and recoverability of lexical features faces a number of serious challenges. First of all, there are some technicality issues: (a) e.g. the role of features and feature checking in the licensing of pro is notalways clear. ln Uriagereka's analysis the relevant feature is person, and is presumably present in the F-head, but absent in the D-head. Under standard assumptions, person on the verb (or any functional projection that is part of the extended verbal projection) is uninterpretable (i.e. it has no value- see Chomsky 2001a) and hence deletable, in true, cross-linguistic generalization regarding DOM effects even harder (see Anagnostqmulou & Giannakidou 1995 fur a discussion on this point). Despite this, there seems to be a general concensus in the literature that only definite and/or specific direct objects. which are interpreted as familiar/topics (ie. as not bound by the existential operator within the vP) are differentially marked in different languages. What counts as definite and/or specific may differ across languages. Moreover, in some languages animacy also seems to play a role in specifying which accusative object is differentially marked (see e.g. Armenian or Leista dialects of Spanish).
73
74
Cities in Greek: A Minimalist Account ofProdisis and Endisis
the sense that it cannot be interpreted at LF. If that is true, it is not clear how the D0head is assigned/inherits a person feature that licenses pro, given that on the one hand the determiner does not have a value slot for person to inherit it, while on the other there is no value on the verb to be passed on to the clitic determiner. (b) The second technical issue is related to the actual interpretation of the NP-pro after checking. In particular, Uriagereka assumes that the NP-pro in clitic pronouns does not have any lexical content (or 'range: following Cardinaletti & Starke 1999) before movement, however it is recovered/ assigned by means of movement'checking (lexical content is person). Now, according to the analysis lexical content on N equals semantic content, and therefore has to be interpretable at LF. Since NP-pro must recover/acquire lexical/semantic content (such as person) in order to be eligible at LF, it follows that such an interpretation would be available to (some) clitic pronouns. However, we have already seen that there is extensive evidence showing that clitics are not necessarily interpreted as [+person]. In fact. the original motivation for elitic movement has been - according to this analysis - the fact that clitics are underspecified for such content. The problem, of course, is that recoverability oflexical content on NP-pro may result to an interpretation which is not confirmed by empirical data. & a result, the linkage of clitic movement with recoverability of lexical! semantic content on NP-pro does not make the right empirical predictions, and therefore appears to be rather problematic. Besides these technicalities, the main problem with the recoverability approach is that feature recoverability cannot be uniquely linked to clitic movement This is so, because according to this analysis recoverability depends on feature checking, which presumably involves a structural spec-head relation (see Chomsky 1995). However, at least in early minimalism (see Chomsky 1995), feature checking via a head-spec relation could take place either in the overt or the covert syntax (i.e. at LF), depending on the strength of the checked feature( s). This would mean that recoverability via checking could possibly take place by means of covert movement. i.e. at LF. In other words, within such a framework, feature recoverability (as feature checking) on its own cannot guarantee elitic movement in all relevant instances. Someone could argue that recoverable features are always strong, thereby forcing generalized overt movement of the clitic. However, this would not explain much, being a mere stipulation. Moreover, we would expect at least one language to have LF clitic movement, 25 on a par with movement of other pronominals and DPs, i.e. we would expect some cross-linguistic variation in this regard. This, however, does not appear to be the case (see Cardinaletti & Starke 1999). Another idea, which has been used by a number of researchers,
25. This is based on the long standing empirical observation that object ditic pronouns do not typically appear in the same vP-internal position as strong pronouns and full DPs; rather, they tend to occur in derived positions, even ifthe latter are not identical across languages. Note that this point is relevant only for those analyses which assume that ditics reach their derived position via movement (and not e.g. by base-generation).
Chapter 3. Internal structure of ditics and cliticlzation
including Rizzi (1993), Cardinaletti (1994), Uriagerek.a (1995) and Carver & Delfitto (1993/1999), among others, would be to argue that clitics must move overtly, because they do not contain lexical/semantic content, the latter being a prerequisite for being visible at LF. That is, if clitics did not move overtly, the derivation would crash, since they would be invisible for movement at LF, and hence no spec-head checking relation and no recoverability would take place. However, there seem to be two problems with this account: first, lack of lexical content does not entail lack of semantic content (although the opposite appears to be true in most cases- see e.g. expletives). For instance, phi-features on DP are typically considered to be LF-interpretable features, although they are not lexical (in the sense that they are functional; presumably, they can be lexical features, i.e. features assigned in the Lexicon). If it is natural to assume that LF sees semantically interpretable features, independently of their functional or lexical status, a fact which trivially follows from the Full Interpretation principle and which is the actual suggestion originally made by Chomsky (1995) (see Chomsky 1995: 198), clitic pronouns would be expected to be LF-visible provided they had some semantic content (i.e. semantically relevant features), and independently of whether they had lexical content or nol As we have already seen, at least some clitic pronouns (and weak pronouns, more generally) may have such features (as e.g. phi-features, or deictic features for 1st and 2nd person pronouns); accordingly. at least these should be LF-visible, and hence able to move at LF (in fact, they would be required to move at LF. given the Procrastination Principle, according to which movement that may take place at LF must take place at LF). The second problem with this account is that in later phases of the Minimalist Program (see e.g. Chomsky 2000, 2001a and 2005) feature checking is dissociated from movement to the target/checking head. According to this analysis, long-distance Agree between a probe and a local goal within the former's c-command domain suffices to check features on both heads, and as a result movement to the minimal domain of the target (i.e. probe) is not required. If movement takes place, it is overt and attributed to a separate EPP feature. 26 Under this analysis the assumption that clitic pronouns do not have lexical and/or semantic content could not force overt clitic movement in all cases, contrary to fact All in all, depending on the exact assumptions of the particular version of minimalist theory followed, the actual movement of a clitic pronoun to the minimal domain of an agreeing probe head may (early minimalism, which allows for LF-movement) or must (late minimalism, which allows for long distance Agree) be dissociated from the valuation/ checking process, the latter being sufficient to ensure recoverability.
26. According to this analysis LF-movement for valuation/checking of syntactic/formal fea-
tures would not exist, although LF-movement for semantic/interpretational reasons could be possible in princtple. Note also that the LF-movement argument would not go through in a theory in which movement is always overt, LF-movement being reinterpreted as pronunciation of the lowest chain link (see e.g. Bobaljlk 2002 and references therein).
75
76
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
As a result, movement should be better viewed as a distinct PF-property, even in those cases it seems to piggy-back on feature checking. 3<J.2.2 Panagiotidis 2002 a uniform
t,.ansitive J:ll approa.ch.
Panagiotidis (2002) also follows the transitive D 0-hypothesis, and on a par with Carver & Delfitto (1993/1999) applies it to all types of pronouns. However, the actual implementation of his approach differs in important respects, on both theoretical and empirical grounds. His main claim goes as follows: all pronouns, strong, weak. clitics and demonstratives have the following internal structure: DP
(11)
--------------
D [person:v]
[numb :u]
NumP
Num [number:v]
NP
~
N (pronominal) [gender:v] (where u = unvalued/uninterpretable and v =valued /interpretable) That is, pronouns are transitive determiners. The presence of particular features on particular heads follows from the assumption that LF-interpretable features must appear on certain heads cross-linguistically. 27 Panagiotidis uses a number of theoretical and empirical arguments to support his proposal. For example, he argues that functional heads are typically dependent on lexical ones, possibly on phonological (depending on the given language) and abstract/ syntactic grounds (they have uninterpretable categorial features that can only be checked by an appropriate lexical head), and as a result the intransitive D 0-hypothesis cannot be maintained. He further points out that the existence of optionally transitive determiners (such as e.g. demonstratives or plural indexicals in English- see Abney 1987) is problematic, since selection is obligatory (hence, an NP is required in all cases, even when they are covert). For Panagiotidis pronominal referencelpronominality (i.e. the essence of pronouns) derives from the fact that the null or overt N does not denote a lexical concept
27. This does not prevent uninterpretable counterparts of the same features to be located on other heads. E.g. D ls assumed to have both gender and number features, which are uninterpretable (on D) and which are checked and deleted. Two further points should be made here: (a) D also contains definiteness when person ls absent (i.e. in 3rd person pronouns), since person entails definiteness but not vice versa (see Ritter 1995); (b) gender ls taken to be an interpretable feature that resides inN (it ls an idiosyncratic property ofN - see Ritter 1995, and Ralli 2003 for Greek).
Chapter 3. Internal structure of ditics and cliticlzation 77
(lack of descriptive content). As far as clitics are concerned, he argues that clitic movement involves two steps: (a) an XP movement step to [Spec, AGRoP] (for case checking); and (b) a local X0 movement step to a higher functional head (e.g. T) (following a quite standard approach in the literature -see Sportiche 1990). He points out that a complex DP approach can capture this derivation in a principled fashion. Moreover, it is conceptually more adequate than a D 0/DP analysis, since it allows different types of features to be interpreted at specific points within the clause (phi-features at [Spec, AGRoP]; specificity/topic-hood at T). This is so, because the D 0-head, which contains definiteness/person, but not phi-features, moves out of the DP in the second X0-step. Panagiotidis's approach seems to be the least able one to account for the differences found across pronominal types both within and across languages, given that he postulates a single structure for all types (as well as for certain types of D Ps ). Since according to current theorizing variation in structure (presumably, in terms of features, on the basis of which structure is projected) reflects variation in the actual syntactic/ semantic behavior of syntactic constituents, Panagiotidis's uniform approach would predict that no such differences should exist among pronouns, which is obviously not true, both for Greek, as I pointed out previously, but also for other languages (see e.g. the discussion in Cardinaletti & Starke 1999). Panagiotidis argues that clitic movement takes place stepwise, each step being related to some (separate) feature or property P that is not present in strong and only partially present in weak pronouns. 28 However, there are some problems with this kind of approach. First of all, it fails to be explanatory in two ways: (a) it does notreally solve the problem of the syntactic distribution ofpronominals; rather, it restates it by simply postulating two arbitrary features/properties, which need further motivation and investigation, not an easy task at all, taking into consideration the failure of such features as 'specificity, 'topicality, 'case, 'persol\ 'referentialitY, which have been
18. In fact, he argues that there are two features, [+ X] and [+ Y], related to clitic movement.
[+ X] triggers overt XP movement to [Spec, AGRoP], while [+ Y] triggers further X0 movement (of the D0 head) to T0• These features do not appear to depend on the presence of the empty pronominal N, given that strong pronouns do have one, but do not move overtly. On the other hand, they must be interrelated, in the sense that [+ Y], the feature triggering the second step, ls dependent on the presence of [+X], the feature triggering the first step. Two things follow from these remarks: according to Panagiotidis's analysis: (a) the only wa:y to capture the distinct behavior of ditics and weak pronouns as related to strong ones, ls to postulate these two additional features that are not present in the structure of strong pronouns; (b) clitic movement cannot take place unless a first XP step to [Spec, AGRoP] has taken place. Panagiotidis further argues that a two-step approach to diticization is independently required, given its properties, and that a complex no/DP approach to the internal structure of clitics, ala Chomsky (1995), is not compatible with the two-step derivation, because it cannot capture the fuct that the same element, a no/DP, moves as XP to check feature X and as X0 to check feature Y, while on the other hand it predicts that both phi- and n-features would be interpreted both at AGRoP and at TP, which is not the case, and hence should not be allowed.
78
Cities in Greek: A Minimalist Account ofProdisls and Endisis
widely used for the same reason in the past literature. 29 As a result. unless we find a principled way to isolate and motivate these extra features that are only specific to some types of pronouns (at least in some languages), any attempt to attribute the special properties of weak and clitic pronouns to such features will be vacuous; (b) it does not explain how the 'addition' of the same extra features to the common pronominal structure, triggering in this way the same two steps, seems to be found in so many different languages. This might, in fact, suggest that these feature/properties are so central to cliticization, that they should be considered as permanent features of the pronominal structure of clitics and weak pronouns. Essentially, this would mean that clitics and weak pronouns do not have the same structure with strong pronouns, contrary to Panagiotidis's original hypothesis. As far as the difference between pronouns/elliptical DPs and full DPs with regards to the possibility of taking arguments is concerned, it appears to be problematic, as well The main reason is that, empirically speaking, such a difference is not absolute, since there are languages where (some) pronouns and elliptical DPs may take an argument For example, Ticio (2003, 2005) argues that clitici.zation out of complex DPs in Spanish is allowed under certain structural (locality) conditions, even in the presence of argumental DPs. An example is given below (e stands for the base/first merge position of the clitic): (12) Vendimos [los libros [de Cervanteslagl y losi compramos We-sold the books of Cervantes and cl we-bought [ei [de Chomskylagl· of Chomsky 'We sold the books by Cervantes and we bought the ones by Chomsky. [adapted from Ticio 2005: 280 (her (95))] The same sentence is not grammatical in Greek (unless the clitic doubles the DP, in which case it is not covert),30 even though extraction out ofDPs is possible in principle: (13) a. *Pulisame ta vivlia tu Cervantes ke ta aghorasame Sold.lpl the books.acc the Cervantes.gen and them.cl boughtlpl tu Chomsky. the Chomsky.gen We sold the books by Cervantes but we bought the ones by Chomsky: b.
pulisame tu Chomsky. ochi ta arthra. sold.lpl the Chomsky.gen, not the articles.acc 'It is the books of Chomsky that we sold, not his articles: TA
VIVUA
THE BOOK
29. This follows mainly from the fact that not one of these syntactic/semantic features can successfully pick up only weak and/or ditic pronouns. 30. The relevant sentence would appear as 'Pulis arne ta vivlia tu Cervantes ke ta aghorasame ta vivlia tu Chomsky: although in this case the contrastive focus reading, which is found in elllptical DPs, would not be available.
Chapter 3. Internal structure of ditics and cliticlzation 79
On the other hand, demonstrative pronouns do co-occur with argument D Ps in Greek,31 and the same applies to (definite/indefinite) elliptical DPs3 2: (14) Dhen vrika tin kokini tsanta tis Marias ala tha para tin mble/mia tis Sulas. Not found.1sg the red purse.acc the Mary.gen but Fut take.1sg the blue/one.acc.fem the Sula.gen 'I did not find Mary's red purse but I will take Sula's blue one/one from Sula instead: (15) a.
Kita afta tis Sulas! Look.2sg these.plneuter the Sula.gen 'Look these ones that are Sula's:
b. Kratisa merika/kapja/ ena tu Kosta. Kepllsg some.acc /one.acc the Kosta.gen 'I kept some/one that belonged to Kostas:
31. Note that in reality these constructions do not involve a demonstrative pronoun containing a covert noun. Rather, as it has been discussed quite extensively in the literature (and see Giusti 1997 and Alexiadou, Haegeman & Stavrou 2007 for more details and references), demonstratives and 1st and 2nd person pronouns are maximal projections that may be merged in a specifier position within some larger DP. Similar remarks apply to indefinite quantifiers, although there is a high chance that these appear in some lower position within the DP. An approximation of the relevant structure is given below (for evidence against N-movement to a higher position in Greek see Alexiadou 2001 ):
(i)
[DP [DP afto] [D' [D 0 to] [XP kokino [X' [YPafte [NP forema [DP tis Maras]]]]]]]
(ii) [DP [D' [XP merika [YP k.okina [NP vivlia [DP tis Maras]]]]]]
The corresponding English construction would be as follows: (ill) [DP [D' [D0 those] [XP friendly [NP students [DP ofPhysics]]]]]
This structural difference does not change the argument, given that in both languages the N-head would be covert (i.e. an empty pronominal noun would be merged): (iv) -.J [DP [DP afto] [D' [YP afto [NP [NO EN] [DPtis Maras]]]]] (v) -./ [DP [DP emis] [D' [NP [NO~] [DP tis fisikis]]]] ('we (students) of Physics') (vi) ~ [DP [D' [D 0 those] [NP [N° ~] [DP ofPhysics]]]]]
32· The same appears to be true also for indefinite drop constructions, although the result may sound less acceptable than when the relevant DPs have an overt indefinite determiner: (i) Aghorases metafrasis tu Kakridhi? Bought2sg translations.acc the Kakridhl.gen 'Did you buy translations made by Kakridhis?' ?? Oxi, ala aghorasa
tu Seferi. No, but bought.lsg the Seferi.gen 'No, but I bought translations made by Seferis:
so
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
English on the other hand does not allow argument DPs, both with overt and covert pronominal nouns: (16) a. *You will meet the ones of Physics. b. *Take those/some/one ofJohn. c. .. I will buy it of Chomsky. lfall these pronouns and DPs contain a null pronominal noun with no lexical content, and if such a noun cannot license argumental DPs, it is not clear how the differences noted above between Greek, Spanish and English would be accounted for in the first place. If we would like to stick to this analysis, we would either have to assume that~ may differ in its properties across languages or across structures within a single language (cf. demonstratives and clitics in Greek), or, alternatively, that each case should be analyzed differently (both within and across languages). In any case, we would be led to conclude that not all kinds of pronouns and elliptical DPs share exactly the same structure, a conclusion which is against the basic tenet of Panagiotidis's analysis, and which shows that adherence to a parallel structure for all these constructions is not empirically tenable. Another problem that Panagiotidis's approach faces is that it cannot predict in a principled way the correlations that appear to exist between different types of properties (i.e. properties of different grammatical levels) that have been observed both within and across pronominal types (see Anagnostopoulou 2003b for this argument). In addition, if the main arguments used to argue in favor of Panagiotidis's approach can be refuted, and the common properties shared among pronoun types be accounted for by these other approaches, then the 'same structure for all pronouns' approach would have to be abandoned. This is what I am going to discuss below. Panagiotidis makes the following points with regard to the existence of~ and its properties: a. Pronominality, namely the fact that pronouns are interpreted as semantic variables in LF and beyond and hence do not denote a concept, is due to the presence of an empty pronominal noun, Bw which is a noun without semantic content b. The existence of overt empty/grammatical nouns ~ (i.e. nouns with formal/functional features only), such as one in English, supports the existence of a null counterpart, both in English (cf. those vs. those ones) and in other languages in which ~may never appear overtly. c. Pronouns encode gender cross-linguistically; since gender has been independently shown to be an inherent property ofN (given that it is a derivational feature) and under the assumption that LF-interpretable features only appear on certain heads cross-linguistically, it follows that also pronouns must contain anN-head encoding gender. Most, if not all, of these arguments can be refuted quite easily, which would make Panagiotidis's analysis much more problematic. For example, with regards to point
Chapter 3. Internal structure of ditics and cliticlzation
(a) Panagiotidis himself admits that EN is not a necessary condition for pronominality: in his analysis of null subjects he argues that pro does not exist and that AGRs be split into a D(v) head and Numb head. Moreover, no~ head is present, however this does not prevent Dv from being interpreted pronominally, which clearly shows that pronominality cannot be used as an argument in favor of the presence of ~ in other pronominal elements. If the semantic variable interpretation of pronouns does not stem from EN' the latter ends up having no LF effect, and since it does not have a PF effect either (for obvious reasons), there should be no reason for its existence in minimalist terms. In order to deal with this problem, Panagiotidis argues that there is some other LF role for ~: it semantically denotes a (trivial) open predicate of type <e, t>, which saturates a determiner (which is of type <<e,t> e>) (see Higginbotham 1985, Heim & Kratzer 1998 for discussion). Although this is generally true, it has been independently argued for in the literature that pronouns are of type <e>, which would make the semantic contribution of~ dubious. A possible counterargument is that the correlation between D and <e> in pronouns in the semantic literature basically stems from the assumption that pronouns are syntactically intransitive D0- heads. If there were independent arguments in favor of a complex syntactic structure, the interpretation of D as of type <e> would be problematic, since under standard assumptions regarding functional application. the meaning of the whole DP pronoun (i.e. <e>) would be the result of the meanings of the elements that make up the pronoun plus the meaning of their (mode of) combination. This would render the interpretation of D as <e> rather unlikely (unless one assumed that all other sub-constituents of the pronoun are also of type <e> and their combination would be a function that takes <e> type elements as arguments and yields <e> type constituents).33 In fact, Kratzer (2008) has convincingly shown that pronouns may consist of one or more sub-features (which may or may not be realized by distinct syntactic heads), each of which has its own semantic import, reducing in this way the possible combinations of these features on a cross-linguistic level and giving rise to a typology that seems to be both empirically and psycho-linguistically adequate (see also Chapter 2). Some of these features, such as e.g. nominal numerical indexes or participant features, are of type <e>, while others, like e.g. gender, are of type <e,t >.To illustrate how this would work in principle, Kratzer argues that bound variable pronouns are in essence minimal pronouns, in that their phi-features are not interpretable at LF; rather, they are simply copied onto the pronoun for PF reasons, so that the only syntactic feature that is present in the derivation and which is visible at LF is the nominal index. Other pronouns, on the other hand, do have interpretable features, which are part of the
33· See Elbourne (2005), who analyzes full DPs and a number of(semantically) different types ofpronouns (referential, bound variables, D-type pronouns) as having the same syntactic structure: they consist of a D head taking an index argument and an NP (ONE) argument, the latter being null (and of type <e, e> ). In this sense, his analysis is very close to that of Panagiotidis (2002), although approached from a semantic angle.
81
82
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
numeration, take part in syntactic operations and are interpreted at LF (either as predicates or as individuals). If such a dynamic approach is on the right track, it becomes obvious that the semantic role Panagiotidis assigns to ~· even if plausible for some pronouns, cannot be maintained for all of them. In this respect. the rigidity of his account and its tailure to cater in a straightforward way for the differences among pronouns parallels the same problems we discussed above with regard to the syntactic/ structural aspects of his analysis. Although the lack of semantic evidence in favor of an <e,t> ~in some pronouns is not sufficient to rule it out in other pronouns, it nevertheless shows that there is no deep LF tor postulating ~ in all pronominals. As a result, its presence, at least in those pronouns where no independent evidence for it is available, would have no PF and no LF effect. and thus would have no reason for existence, following standard minimalist assumptions. With regards to point (b), although there are indeed languages which have overt pronominal nouns, such as e.g. English one or Japanese kare (see Dechaine & Wiltschko 2002 and Panagiotidis 2002 ), and possibly covert ones, it does not necessarily follow from that that such a noun should be available in all languages, even in those which do not offer phonological evidence tor it For instance, English appears to have an overt empty/grammatical noun (i.e. a lexical noun containing only grammatical/ formal features), one, and possibly also a covert one, at least according to a non-deletion approach to ellipsis (see Lobeck 1995), as shown in the following examples: (17) this one vs. this 0 Panagiotidis assigns the following structures to the DPs in (20), following proposals by Postal (1969 ): (18) [DP [D 0 this] [NP [N° one)]] vs. [DP [D0 this] [NP [N° 0]]] According to Panagiotidis, both one and 0 are N-heads, which D 0 agrees with. In this respect, the syntax of DPs containing such empty N-heads parallels that of DPs with lexical (non-closed class) N-heads. However, in English there seems to be syntactic evidence in favor of a zero N-head in constructions like those in ( 17) and ( 18). For example, a restrictive relative clause can modify (most) personal pronouns or demonstratives: (19) a. He who must not be named gave orders to the Death Eaters. b. She who must be obeyed has made her entrance. (example taken from Elboume 2005: 120 (his 114)) Moreover, personal pronouns may be followed by an overt lexical N-head (although there also some restrictions depending on person/number): (20) a. b. c. d.
Wemen You linguists. You (men) who wish to escape. Youhere.
Chapter 3. Internal structure of ditics and cliticlzation
In Greek. on the other hand, only strong pronouns and full nouns may be modified by a (restrictive or non-restrictive) relative clause: (21) a. Milisa me esena pu ise ghlosologhos. Talked-lsg with you.str that are.2sg linguist 'I saw you who are a linguist:
(Greek)
b. ldha afton ton andhra pu xtipise tin Eleni. Saw.lsg him.str the man.acc that hit3sg the Eleni.acc 'I saw the man that hit Helen? c.
Idha afton pu xtipise tin Eleni. 'I saw him who hit Helen?
The same sentences are ungrammatical with clitic pronouns, unless they double a DP: milisa pu/o opios ise ghlosologhos. (22) a. *Su You.cl talked.lsg that/the who.nom are.1sg linguist 'I talked to you who are a linguist:
(Greek)
b. *Ton sinandisa pula opios xtipise tin Eleni. Him.cl metlsg that/the who.nom hit3sg the Helen.acc 'I met him who hit Helen? c.
Su milisa esena pula opios ise ghlosologhos.
d. Ton sinandisa afton/ton andhra pu!opios ine ghlosologhos. Also, clitics cannot be followed by an N-head, contrary to strong pronouns (unless, again, they double a DP): (23) a. *Tin thelo edho. Her.cl want.lsg here 'I want this one here:
(meaning 'I want this one here')
idha (ton) ghlosologho. b. *Se Her.cl saw.1sg the linguist.acc 'I saw you the linguist: c. Tin thelo aftin edho. d. Se idha esena ton ghlosologho. Interestingly enough, the English pronoun it, which has been argued by many linguists to be a weak pronoun (cf. Carver & Delfitto 1993/1999 and Cardinaletti & Starke 1999, and the following examples: ..J give it away vs. * give away it, I1-.J It is him that I like vs. *It is it that I like), is also ungrammatical with a relative clause or a following N-head (see also Elbourne 2005): (24) a. *It that rolls tastest gathers no moss. b. .. It which rolls fastest gathers no moss.
83
84
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
c. That which rolls fastest gathers no moss. d. Those who hesitate are lost (all examples taken from Elboume 2005: 121 (his 116)) (25) *I wantithere. Finally, note that in Greek a bare definite article behaves on a par with clitics and weak pronouns, as opposed to the indefinite article, and contrary to what is the case in Romance. In English a bare definite determiner without one is also ungrammatical: (26) a. ..Egho idha tin pu filise o Kostas. I saw.1sg the.acc that kissed.3sg the Kostas.nom b. Egho idha mia pu filise o Kostas. I saw.1sg one.acc that kissed.3sg the Kostas.nom c.
Quiero la que Juan quiere. Love.1sg the that John wants.3sg 'I love the one that John loves:
d. The *(one) who hesitates is lost. (taken from Elbourne 2005: 122 (his 120)) These data are quite revealing. As far as relative clause modification is concerned, it is obvious that in Greek an overt N-head is possible with modification. Strong pronouns are also compatible with modification, supporting in this way Panagiotidis's analysis that pronouns contain a (covert) nominal head. However, clitics are not grammatical with it, which renders his uniform approach to the internal structure of pronouns rather problematic: given that ~ has the same features in both strong and clitic pronouns, it is not straightforward to me why clitics would not allow modification of their covert empty noun. On the other hand, a theory that would postulate anN-head for strong pronouns only would naturally account for this difference. With regards toN-predicate complementation, it seems that Greek clitics do not allow it (i.e. they cannot take an overt N-head), as opposed to strong pronouns, as well as to English pronouns. However, it is not clear whether the Greek and the English examples have similar structures. This is so, because English pronominals seem to share the same distribution with other determiner heads, like e.g. definite articles, while in Greek the strong pronoun (or demonstrative) may co-occur with other determiners-3 4: (27) a. We (*the) linguists b. Emis *(i) ghlosologhi. We.str (the) linguists.nom 'We linguists? 34· In fact, they must co-occur with an overt determiner when anN-head or some other predicative head (like e.g. an adjective) ls included within the larger DP (cf esis *(i) prasini ('yotLpl the. pl green. pf)// emis *(I) fitites tis fisikis ('we the students of physics'). When no overt head follows them no determiner ls allowed, ahhough a DP complement may be present: emis (*i) tis fisikis.
Chapter 3. Internal structure of ditics and cliticlzation One possibility would be to analyze strong pronouns as being in a specifier position within a larger DP (possibly, [Spec, DP] ). This is supported by the fact that demonstratives, which are used in the place of 3rd person personal pronouns, also appear in a specifier position within the DP, and may also co-occur with a definite determiner (see Alexiadou, Haegeman & Stavrou 2007; Panagiotidis 2000): (28) Aftos .. (o) anthropos. This the man.nom 'This
man:
If this analysis were on the right track, theN-head test would not really tell us much about the plausible presence of anN-head within the pronominal tree ofboth strong and clitic pronouns, given that the overt N-head heads the larger DP: (29) a. b.
[DP [DPaftos] [D' [D 0 o] ... [NP anthropos]]] : [DP [DP emis] [D' [D 0 i] ... [NP ghlosologhi]35 [DP [DP mas] [D' [D 0 tus] [NP ghlosologhus]])
The only thing we could conclude would be that clitic pronouns are not licensed in the specifier position of a larger DP which contains an overt N-head (as well as an overt D-head, presumably due to interpretational reasons related with the inherent properties of the determiner - see Panagiotidis 2000 on the checking of a definite feature in Spec DP), although they are ok if these heads are either non-overt or not projected at all. In this respect. they would contrast with strong pronouns, which give rise to grammatical results in both cases. Of course, both pronominal types would be ungrammatical with an overt bare N-head, strongly resembling the ungrammaticality of multi -definite constructions in Greek when the modifier does not contain an overt definite article (cf. tin tsanda *(tin) prasini ('the bag the green') - see Alexiadou, Haegeman & Stavrou 2007 for extensive discussion and references).
35· Note that this analysis may not be without problems: it seems that the personal pronoun behaves as the head of the construction in terms of subject agreement and doubling: (i) Mas/*tus antipathi emas tus ghlosologus. ('us /*them dislikes us.str the linguists. ace') (ii) Emls i ghlosologhi imaste/"" ine atsldhes. ('We.str the linguists are.l pl/*3pl clever')
Since in Greek DPs in specifier positions do not seem to be able to percolate their person or definiteness features (or values of features) to the head of the DP they are embedded in, as shown from the examples below, it would follow that the examples in (i) and (ii) suggest that the strong pronouns are heads of the whole DP. If that ls true, then these examples could involve appositive structures. (iii) Echi vivlia edho. ('Has books here'= There are books here) (iv) *Echi ta vivlia enos pedhiu edho. ('Has the books one.gen chlldgen here') (v)
Den ta/*mas simpathun ta pedhia emas ton ghlosologhon. ('Not them/us llke.3pl the children.acc us.gen the linguists.gelf)
85
86
Cities in Greek: A Minimalist Account ofProdlsis and Endlsls
An exact analysis of these constructions is beyond the scope of this section. However, it is important to stress the point that strong pronouns can be followed by an overt N-head, as opposed to clitic pronouns which cannot do so. This is not a straightforward difference, especially given the fact that the determiner article, which is largely homophonous with the 3rd person clitic paradigm, does allow (in fact, requires) an overt N-head. Even though it is not clear to me how the presence vs. absence of an N-head inside the pronoun seated in the specifier of the larger DP would play any role in this difference, the latter also remains a question for Panagiotidis's analysis. All in all, it is only the relative clause test that offers some evidence against the postulation of an empty N-head in clitic pronouns. Note, however, that this test does not show that modification of a null N-head by a relative clause is not allowed; rather, it only shows that such modification is not allowed unless some N-head, overt or covert, is present in the structure. One of Panagiotidis's major arguments in favor of the presence of an empty lexical N-head for all pronouns and elliptical DPs is based on the presence of gender on 3rd person pronouns. His argumentation goes as follows: it has been convincingly shown in the literature (see Corbett 1991, Ralli 2002, Ritter 1993, among many others- and see Alexiadou, Haegeman & Stavrou 2007 tor extensive reterences) that gender is an inherent lexical property of N-heads that has to be learned separately for each (gender carrying) lexical item.36 Moreover, as Panagiotidis argues, gender cannot be a feature on determiners, given that on the one hand gender agreement is possible on determiner-less adjectival phrases, while on the other hand gender is also found in languages with no overt articles (such as e.g. Slavic languages). These facts, combined with his theoretical assumption that interpretable features are found on specific heads, support the hypothesis that pronouns, and more generally any nominal with no overt N-head, contain a grammaticaVempty noun with inherent gender teatures. Accordingly, gender agreement on adjectives, determiners or other heads within the DP is the spell-out of an uninterpretable feature on these heads that has been valued via Agree (in the sense of Chomsky 2000, 2001a) from the interpretable gender feature on N. Although gender marking on pronouns is among the strongest arguments in favor of the postulation of a covert N-head in their structure, it nevertheless faces a number of problems. First of all, it seems to be too rigid in that there are languages where gender is not a property on N (see Anagnostopoulou 2003b on this argument). Ritter (1993) for example argues, quite convincingly, that a number of differences in the properties of nouns in some Romance languages and in Hebrew can be accounted for 36. This ls supported by a number of facts in different languages. For example, in Hebrew gender morphology is used derivationally in order to derive new nouns from existing nouns (c£ e.g. magav 'wiper' vs. magev-et 'towel' (Ritter 1993: 796 (her 2a) ), and in addition the exact gender suffix attached to a noun ls fairly unpredictable. Also in Greek gender morphology may be used derivationally (d. e.g. jitonlsa 'neighbor-fern' from jiton - 'neighbor-masc' (see Ralli 2002: 522 (her 4b) ), but more importantly it can be shown that gender ls an inherent, unpredictable property oflexical stems (see Ralli 2002 for extensive argumentation).
Chapter 3. Internal structure of ditics and cliticlzation
adequately if one takes gender to be specified on Numb 0 vs. N° respectively. If Ritter is correct in her analysis, then not all pronouns in every language would need an empty N-head due to gender marking. And, since according to Panagiotidis's approach intepretable features on separate heads do not interact with each other, it is not clear to me how his analysis would capture the fact that e.g. in Hebrew gender does not interact with number, as opposed to Spanish where it does. A second problem with the assumption that gender may only be specified on N is that there are cases where gender is not an inherent property of N. For example, Alexiadou (2004) and Ralli (2002 ), among others, argue that in a set of nouns in Greek (mainly those ending in -os which denote professions) the nominal stem is underspecified for gender (i.e. it has no inherent value for it), the specification for the latter being assigned at the p hrasallevel by the determiner: (30) a.
o/i dhikighoros (the.masc/the.fem lawyer)
b. 0 Mitsos II Roksani ine ghiatros. The.masc Mitsos.nom.masc /The.fem Roksani.fem is.3sg doctor 'Mitsos/Roxani is a doctor: ln this case the gender feature on the determiner cannot be uninterpretable, as it would be impossible for it to get valued from the gender feature on N (given that the latter does not have a value). We may conclude two things from these cases: (a) gender on N does not always come with a value from the Lexicon; (b) gender on D may be inherent and interpretable. Accordingly, the presence of gender on a pronoun would not necessarily mean that the latter must contain an empty N-head. Although the link between gender and~ is not as straightforward as Panagiotidis would like it to be, it is nevertheless true that gender is an inherent property of nominal elements (see Chomsky 1995 and Corbett 1991, 2006). Moreover, although gender may be inherent or non-inherent, nevertheless, when inherent, it may only appear on some constituent that is nominal in terms of syntactic category.37 I.e. gender appears to be contingent on the presence of some nominal feature in the structure. Consequently, if we assume that EN is not present in pronouns, the presence of gender (and possibly other inherent nominal phi- features, such as person and number) would remain unexplained and largely coincidental which is counter-intuitive, given that gender co-occurs with person and number on nominal constituents quite systematically.38 This is the essence of the empirical side of the Extended Projection issue Panagiotidis points out with regard to the bare DP analysis of pronouns. We could follow 37· Recall that there are languages where verbal elements may inflect for gender, however in that case gender is uninterpretable, as any other phi-feature on verbal related heads. Inherent gender, on the other hand, may only be found with nominal elements. 38. Note in this respect that pronominals in Greek always show up with nominal morphology, which further supports the hypothesis that there must be some nominal head within the pronominal structure.
87
88
Cities in Greek: A Minimalist Account ofProdlsis and Endlsls
Marantz (2001, 2006) and assume that categorial status is not an inherent property of lexical heads, but rather is syntactically assigned via special phase heads, which are merged with (appropriate) lexical roots that are not specified for category.39 According to this theory, a noun would be derived by merging a lexical root with a functional nominalizer head n. The outcome of this merger would be sent to Spell Out, ensuring that whatever merges next will have to contain features that make it compatible with the complex n-root-head. These could possibly include nominal/categorial features, capturing in this way the notion of Extended Projection.40 According to this approach, then, category defining heads such as v0, n° and a0 (where a0 stands for adjective) have a special status in a number of respects (syntactic, morphological, categorial, semantic) (see Marantz 2001, 2006 for extended discussion). Since gender is cross-linguistically an inherent feature of nouns, and since a noun may only be derived by means of an n- head (i.e. roots are not specified for category, and hence would not be expected to be specified for gender), it follows that n° would also carry the gender feature. Accordingly. gender would not be dependent on a lexical root, but rather on the presence of a functional nominal head. Two questions arise from the above discussion: (a) if gender is a feature on a nominalizer head, how can we capture the fact that in purely morphological terms gender may be dependent on the lexical stem, rather than e.g. the inflectional class of the endings added to the stem? (see e.g. Ralli 2002, 2003 for arguments that gender in Greek is an inherent property of lexical stems); (b) 'shift' of gender to a functional nominalizer head leaves the empty N-head in pronominals (and possibly in elliptical DPs) without LF-support, however it is not at all clear that it can capture the structural differences that can be observed among the different kinds of pronouns and DPs (and which were discussed above-cf. e.g. modification by a relative clause, or by anN-head), since according to this analysis all pronouns would end up with a nominalizer, but with no (empty) root head. With regard to point (a), one could argue that the actual value of the gender feature on the nominalizer is specified upon its merger with a root The value could be the function of the semantic properties of the root (e.g. in those cases where gender is defined on the basis of such features as animacy/human- see Corbett 1991), or simply the result of some arbitrary convention (i.e. as an inherent property of the nominalizer-root combination). In both cases gender would be viewed as a 'lexical' property, since it would be negotiated within the root domain. Moreover, note that the lexical stem (i.e. the root in Distributed Morphology terms) is only one of the possible factors
39· These heads can be realized by category-changing derivational morphology (like e.g. -ity (n°) or - ify (v0) in English, and - ik (n°) and -ev (v0) in Greek), or by a null element 40. Following a root-nominalizer approach, an Extended Projection would be formed by means of a series of heads related with each other via c-selection. The latter would be technically derived via merge/agree between these heads on the basis of categorial features (see Julien 2002, Panagiotidls 2002, Matushansky 2006, and references therein, for this analysis).
Chapter 3. Internal structure of ditics and cliticlzation
which define gender. For example, Ralli (2002, 2003) has convincingly shown that in Greek derivational affixes specify the value of the gender of the whole word. These affixes realize the nominalizer head, giving direct support to the assumption that the latter hosts the gender feature. With regard to point (b), we could argue that strong pronouns and elliptical DPs have a lexical root. as opposed to weak pronouns which do not Although this hypothesis may seem rather ad hoc, and moreover it forces us to abandon the analysis of pronominality in terms of an empty pronominal Bw on the other hand there seems to be some evidence in favor of it For one thing, 3rd person strong pronouns in Greek can be decomposed into a stem (root) 41 which remains constant, and a set of endings which change shape depending on the set of phi-features expressed42: (31) ajtns, aftu, ajtnn, afti, ajtnn, aftus (he.sg, his.sg, him.sg, they.masc, etc.) afti, aftis, aftin, aftes, ajtnn, ajtes (she.sg, her.sg, her.sg, they.fem, etc.) afto, aftu, ajto, afta, afton, afta (it.sg, its.sg, itsg, they.neu, etc.) 3rd person clitics, on the other hand, lack the invariant morpheme, and only contain the inflecting part, which denotes gender, person and number distinctions. Even though one could argue that the clitic contains a null root (instead of an overt one), it would not be straightforward why clitic pronouns typically lack an overt morphological stem across different languages. Moreover, the postulation of such a difference between strong and clitic pronouns would account both for the presence of gender on both pronominal kinds as well as for the lack of structural evidence in favor of an empty root in clitics. Note, however, that a similar analysis is not as straightforward for 1st and 2nd person strong pronouns. For example, Ralli (2000) points out that egho ('I.nom.sg') does not seem to be amenable to a morphological decomposition into a stem and an inflectional ending. Drachman (1997), on the other hand, analyzes the same form as e-gho, where e- is the stem (and the same applies to other 1st person types, namely e-mas, e-mis, and 2nd person types, such as e-si, e-sas, etc.). Even if a stem-inflectional ending decomposition of all strong pronouns is not possible, we could take these forms to have a null N-head. Elliptical D Ps could also be analyzed as containing a nominalizer and an empty lexical root This is supported by the fact that such D Ps have the same s-selectional restrictions with DPs containing an overt lexical N-head. To illustrate, in the following example,
41. For the relation between stem and root and the status of stem in Greek as the basis of words see Ralli (2003). 42. Note that the inflectional ending may be further decomposed into sub-morphemes, so that t- can be analyzed as encoding third person (cf 3rd person ditics and the definite article, where t- is also present) while -os-, -1-, -o-, etc. can be analyzed as encoding number, gender and case. The same analysis may apply to (other) deicticpronominals such as tutos, -i, -o ('this one'), and ekbws, -1, -o ('that one').
89
90
Cities in Greek: A Minimalist Account ofProdisis and Endisis
(32) I metafrasi tu Kakridhi ki afti [.V tu Kazantzaki]/ The translation the Kakridhi.gen and that one [.V the Kazantzakis] I [# tis ftochias]. [#the poverty] the second conjunct. which consists of a 3rd person strong pronoun and a genitive DP argument (which is possible in such constructions in Greek. as opposed to English, presumably for independent reasons; for genitive arguments within the DP in Greek see Alexiadou 2001, Alexiadou, Haegeman & Stavrou 2007), is acceptable only if the non-ellided DP is semantically compatible with the lexical N-head of the first conjunct. One way to capture this fact would be to analyze the second conjunct as containing a null N-head which copies the lexical content of the N-head in the first conjunct. The exact nature of this N-head (whether e.g. it is an empty lexical head, or involves PF-deletion) and the copying/licensing process itself are not important here. Rather, what matters is that an elliptical DP without a lexical root would be predicted to impose no semantic restrictions on the DP argument.43 All in all, it seems that gender on clitics, strong pronouns and elliptical DPs does not necessarily force the presence of an empty ~ in all these constructions. On the contrary, there is some evidence that weak pronouns do not contain a null root. while on the other hand a theory decomposing the traditional N into a nominalizer and a root can capture the structural differences but also the other similarities and/or differences between pronominal kinds in a more adequate way. If all these arguments can be decomposed, and if Panagiotidis's theory is both too strong and too weak in empirical respects (in the sense that on the one hand it overgeneralizes, while on the other it may be too restrictive), we can safely conclude that the hypothesis that all pronoun types have one uniform structure cannot be maintained on empirical and theoretical grounds.
43· Interestingly enough, ditic pronouns may allow coercion of predicates into a metaphorical reading, which is not possible with strong pronouns and full DPs. For example, the verb anapodhoghtrizo ('to turn upside down') receives a literal reading which involves a [± animate] theme: (i) I Paresa anapodhoghirise aftus /ta adhelfia tis. The Paresanom turned-upslde-down.3sg them.str /the siblings.acc his.d 'Paresa turned them/her siblings upside down.' (literal reading) However, a clitic alongside the literal reading also allows a coerced non-literal/metaphorical (psychological) reading: (ii) I Paresa tus anapodhoghirise. The Paresa.nom them.d turned-upside-down.3sg 'Paresa upset them (metaphorical) /turned them upside down (literal)? Although it is not entirely dear to me why only a ditic pronoun allows semantic coercion, one plausible hypothesis would be to link this possibility to the lack of any lexical/root content in the ditic.
Chapter 3. Internal structure of ditics and cliticlzation
3·3·3 Sub-type C: Pronouns as heads in the extended verbal projection Another type of analysis, which has been quite influential in the clitic literature, especially for languages displaying the phenomenon ofClitic Doubling (CD), is that clitics are functional heads merged in (the extended projection of) the verb. This type of approach, also known as the 'base generation' approach to cliticization (see Anagnostopoulou 2006 for an extensive overview of approaches to Clitic Doubling and cliticization more generally; see also Philippaki-Warburton & al. 2004) was developed in the late '70's and early 'SO's as an attempt to account for languages exhibiting the phenomenon of Clitic Doubling. The latter refers to a construction in which a clitic co-occurs with a DP in the complement position of the verb (i.e. within the bounds of a certain syntactic domain- see Tsak:ali 2006 on this), with which it agrees in a number of features/ properties. The possibility of clitic doubling, at least in some Romance languages, was taken as a major argument against Kayne's (1975) movement approach, which was developed to account for languages like French, which do not seem to have clitic doubling (see however Kayne 2000, who claims that direct object strong pronouns in French must be doubled). Although base generation approaches have undergone major changes throughout the years, both in terms of their empirical coverage and their particular theoretical orientation/goals, most (if not all) of them analyze clitics as heads merged directly on their host (or some functional projection related to their host). In the early instances of the base generation approach, for instance, Jaeggli ( 1982) argues that clitics are syntactic affixes with insertion frames, which are generated directly on V and which absorb government of the verb (Jaeggli 1986 drops this option). This tact combined with the basic assumptions of the G B case theory, as well as the proposal that clitics do not saturate the El-role of the verb (so that an N P position must be projected in the com plement of the verb even when clitics are present in the structure44 ), accounts for Kayne's generalization, namely the observation that at least Romance languages allowing clitic doubling have an independent case marker (like Spanish a or Rumanian pe) that can assign case to the doubled DP instead of the verb. Borer (1984) follows a similar approach, in that she also links Kayne's generalization (and hence the availability of doubling in a given language) to case-theoretic concerns. However, for her clitics are agreement morphemes, parts of the verbal head onto which they attach, which spell-out and absorb its case features,45 limiting in this way doubling to those languages which have independentcase markers/ special prepositions. 44 The NP complement ls PRO, when covert (since it ls ungoverned). In hls 1986 revlsion of thls theory, he argues that it ls a pro. 45· In particular, she argues that ditks realize/absorb the case feature of the verb (and not its government feature (pace Jaeggli 1982, but see Jaeggli 1986 who adopts Borer's proposal), since she provides extensive argumentation that ditics in Hebrew do govern their complement), however an inflectional rule that may operate in the syntax inserts non-inherent phi-features of the verb, such as person, number and gender, into thls case feature/position.
91
92
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
Although clitics do not spell-out inherent features on the verb (such as e-roles and categorial features), so that the latter have to be assigned to/be realized by an overt or covert complement, 46 they must be co-indexed with the doubled N P-complement, and agree with it in phi-features. Co-indexing is forced by the so-called Complement Matching Requirement, which is related toe-role assignment Sufter (1988) builds on Borer's (1984) proposal that clitics are inflectional elements. In particular she argues that Spanish elitics are (object) agreement endings which are listed in the Lexicon as separate items/morphemes, and generated as part of the verb (on a par with subject agreement morphemes). Clitics form an (agreement/ binding) chain with the doubled constituent, however they do not absorb case. The immediate result of this, and one of her main claims, is that the (im)possibility of clitic doubling in a given language is not determined by case; rather, it depends on language particular properties such as specifidty. Moreover, as Sufier convindngly shows, in certain dialects of Spanish doubling does not appear to be dependent on the presence of the special preposition a. Rather, the latter is analyzed as an animacy marker (and cf. Rumanian pe, which has been shown to be sensitive to the feature [+ human]). Apparently, although Spanish generally prefers doubling to co-occur with animacy marked DPs, this is only a tendency and not a necessary condition.47 Finally, Sufier points out two further things: (a) clitics, being agreement markers, contain phifeatures/case and the features [animate] and [specific]. Due to the matching principle that applies to agreement markers, they must agree with the doubled DP in all these features.; (b) direct object/accusative clitics differ from indirect object/dative clitics in terms of the restrictions they impose on doubling (hence on the agreement chain they build with the doubled DP): the former are specified as spedfic (and/or animate, depending on the given language) in the Lexicon, and as a result can only double a spedfic DP, as opposed to the latter, which do not have this spedfication, and hence are not constrained by spedfidty in this respect These two points purport to explain the semantic and other restrictions imposed on Clitic Doubling. Sportiche (1992/1998) also analyzes clitics as functional heads. However, he differs from Borer, Jaeggli and Sufier, in that he takes them to be heads within the extended projection of the verb, heading their own projections, and not to be directly generated on the verb (following the general proliferation of functional clausal heads postulated by the syntactic theory of the time- see e.g. Ouhalla 1991, Pollock 1989). Sportiche locates these heads in the INFL domain of the verb (right above AGRs) and gives them the name 'Clitic Voices'. These clitic heads involve checking of some property P under a Spec- Head configuration with their double XP (which is overt or covert (pro), merged in the a-position), whereby the doubled XP moves overtly or covertly to 46. In this sense, ditics are not derivational but inflectional affixes. Note that fur Borer the covert NP complement ls undefined in terms of Chomsky's (1981, 1982) anaphor/pronominal repertoire. 47· Put in different words, doubling occurs only with a subset of a-/pe- marked DPs (see also Borer 1984 on Rumanian pe).
Chapter 3. Internal structure of ditics and cliticlzation
the clitic specifier. Movement of the XP is forced by the Clitic Criterion (subsumed under a generalized licensing criterion), which requires checking of some feature [F] under a Spec-Head configuration (at LF). For Sportiche, accusative clitics involve specificity checking/licensing. In this respect, they resemble wh- or other operator dependencies, and their specifiers are A-bar specifiers. Dative clitics, on the other hand, do not check any interpretative/ quanti.ficational features, but only agreement (cf. Suiier's 1988 point that dative clitics do not yield specificity effects in doubling), therefore their specifiers are A-specifiers. In this theory, clitic doubling and its crosslinguistic (non)availability tallows from a doubly filled voice filter, which disallows a clitic voice phrase to have both its specifier and its head position filled by overt material that encodes some property P licensed by the head of the phrase. This predicts that doubling will be possible only in those languages which delay (overt) XP-movement to LF. 48 The general schema he assumes is given below
(33)
CL Voice
~ XPA CL' ~ CU VP ~
XP* Anagnostopoulou (1994, 1999) and Anagnostopoulou & Giannakidou (1995) use Sportiche's theory for Greek direct object elitics, however they differ from him in that they take the accusative clitic head to license referentiality (and not specificity). However, in later work of hers (see Anagno stopoulou 200 3a and references therein), Anagnostopoulou argues against the assumption that elitics are base generated heads in the extended projection of the verb, based on evidence from locality. On the contrary, she claims that clitics (both direct and indirect ones) in Greek are D-bundles of features
48· Sportiche (1992/1998) argues that his general ditic schema allows for a number of options which give in a rather straightforward way the parametrization of ditic constructions crosslinguistically. These options are: (a) Movement of the XP to the specifier of the Clitic Voice may be overt or covert; (b) the ditic head may be overt or covert; (c) the moved XP may be overt or covert. Accordingly,. ditic doubling constructions arise when an overt XP moves covertly to the specifier of an overt head, while undoubled constructions arise when a covert XP (a pro) moves overtly or covertly to the specifier of an overt head Finally, scrambling, which is unified with accusative ditic doubling, since both license a specificity feature, arises when an overt XP moves overtly to the specifier of a covert ditic head
93
94
Cities in Greek: A Minimalist Account ofProdisis and Endisis
(i.e. sets of formal properties of DPs, including case) which move overtly from the a-position (where the double is merged) to a verbal functional head (v and/or T).49 These are the major analyses in the literature which have proposed a functional head status for clitics.50 The main aspect that seems to differentiate these approaches from others which also assign a head status to clitics is that they analyze them as being base generated as heads, and not as landing to a head position (excluding Anagnostopoulou 2003a). This is not surprising, given that the main goal of these analyses was to account for clitic doubling in a number of languages; postulating a binding chain between a base generated clitic (receiving case) and a base generated doubled DP (receiving theta role) could capture the availability of doubling as well as the agreement-like properties of clitics. lt is important to point out here that the head/base generation approach and the DP/movement approach are not in principle incompatible with each other. In fact, in recent years a number of proposals have been made that try to combine them. These proposals are part of the so-called fragmentation view on clitics (see Anagnostopoulou 2006 citing Bleam 1999) according to which clitic pronouns, both within and across languages, may receive a distinct (structural) analysis depending on their properties. In this respect, analyses of this type differ from more traditional ones, which typically assign either head or phrasal status to all clitics, with any differences among the latter being attributed to the internal or external structure/properties of the relevant head or phrase.
49· Other researchers who assume a head analysis of Greek ditics are Tsakali (2006), and Tsimpli (1999), Tsimpli & Stavrakaki (1999), Roussou & Tsimpli (2006), among others. Tsakali follows Sportiche's approach, whereby the ditic and its associate are generated in distinct positions and form an agreement chain that ensures feature matching. Moreover, she argues that although case is assigned to the head ofthe chain (i.e. to the clitic head) by v (in T), the 6-role is assigned to the foot (i.e. to the associate). Oitic-associate chains resemble in this sense expletive-associate chains (e.g. in English) (note that she assumes the dissociation of case and agreement, following Chomsky 1995). Tsimpli (1999) and Tsimpli & Stavrakaki (1999) claim that ditics are resumptive pronouns that spell-out the D-, phi- and case features of the correspondingv heads (see also Torrego 1998 for Spanish). Roussou & Tsimpli (2006) basically follow the same intuition, however they apply it in the framework of Manzini & Savoia (2005), according to which clitics are generated in ditic voices in distinct domains (lexical, INFL and C domain).
so.
A head status analysis, at least for dative clitics in Spanish, has also been proposed by Bleam (1999) (for Leista dialects of Spanish), as well as by Cuervo (2003). According to Bleam the doubled dative DP moves overtly or covertly to the specifier of the dative head, whereas Cuervo argues that it is directly merged there (c-commandingthe theme). Thrrego (1998) has also argued that ditics are heads (in particular, X0s and XPs, following Chomsky 1995), i.e. bundle of features (including case, D- and phi-features, as well as a specificity feature for direct objects), which are added onto v (accusative clitics) or p (dative clitics). The addition of an accusative clitic on v triggers overt raising of the doubled DP to the specifier of v, yielding the well known specificity effects of marked accusatives in Spanish.
Chapter 3. Internal structure of ditics and cliticlzation
To give an example, Bleam (1999), based on a number of differences observed among direct/accusative and indirect/dative object clitics in Spanish, has argued that the former should be analyzed as determiners in the sense of Uriagereka ( 199 5), while the latter should be analyzed as agreement heads, base generated in the extended verbal projection. On the other hand, Uriagereka (1995) e.g. suggests a structural distinction between 1st and 2nd person clitics, as opposed to 3rd person ones, although both are taken to be phrasal categories, while Sufier (1988) and Sportiche (1992/1998) take both accusative and dative clitics to be heads, with differences in their teatural content and/or their status as heads (e.g. agreementvs. specificity markers for Sportiche). In the Greek literature both head and phrasal-type analyses have been used, although a fragmentation approach has not been proposed to my knowledge up to this point, mainly because the main arguments for such an approach, as those have been used in languages like Spanish e.g., do not seem to hold for Greek (see Anagnostopoulou 2006, and main text below). ln the next section, as well as in Chapter 4, I will argue that a base generation approach is not the best analysis for Greek clitics (which of course does not preclude the possibility that clitics of other languages might be analyzed as agreement markers). On the contrary, a view that takes clitics to be pronouns can capture the facts more adequately, while on the other hand it is preterable also for theoretical reasons. First of all, before we start this discussion let me point out the obvious, namely that the fact that clitics behave as heads does not necessarily imply that they are AGR heads. On the contrary, they could be specificity heads, topicality heads or some other type of head that is merged within the extended verbal projection, or even a pronominal suffix head that moves into the projection. This means that although we know that clitics are (also) heads, we cannot be sure that they are AGR heads, unless we have proven that it is so. So, why would so many people have assumed that object clitics in Greek (and in other languages, like e.g. Romance) are AG R heads? There are two main reasons. On the one hand, object clitics have morphosyntactic properties that make them look like subject agreement morphemes: (a) they are heads (cf. V-to-C in imperatives (34), lack of interpolation (35), repetition under conjunction (36)), exactly like subject agreement (34) [ep [c Pes tu to]i [TP [ T (Pes tu to)i] ...]] Tell.2sg.imp him.gen.cl it.acc.cl 'Tell him that!' (35) to (.. makria) eriksa (makria) II .. eriks- (makria) -a vs. eriksa (makria) It.cl (far) threw-lsg (far)l/ thre-(far)-w vs. threw (far) 'I threw it (the ball) furl/ I threw far~ (36) Irthe ki efig*(e) vs. To pira ke *(to) dhosa. Came.3sg and left.3sg vs. Itcl took.lsg and (itcl) gave.lsg 'She came and left~ /11 took it and gave it back:
95
96
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls (b) they have phi features (including gender, as opposed to subject agreement morphemes which, by being verbal, typically do not have gender at least in most forms) even though they also include other semantic/syntactic features such as specificity or topicality. (c) they may form topic chains with coindexed DPs (like e.g. clitic doubling or CLLD chains), on a par with subject agreement morphemes: (37) a.
Irthe o Janis. Came.3sg the John.nom 'John came:
idha ton }ani. b. Ton Him.cl saw.1sg the John.acc 'I saw him, John: c.
[ Clitic Doubling]
0 Janis ghnorise ti Lisa. The John.nom met.3sg the Lisa.acc ~s for John, he met Lisa:
ti ghnorisa xtes. d. Ti Lisa The Lisa.acc her.acc metl sg yesterday ~s for Lisa, I met her yesterdaY.
[Clitic Left Dislocation]
On the other hand, the formation of clitic doubling chains, or in other words the lack of complementary distribution between full DPs (pronominal and nonpronominal ones) and elitic pronouns has been among the strongest arguments in tavor of the base generation hypothesis of clitics and of their agreement status (see Borer 1984, 1986 and Jaeggli 1982, 1986 and references therein). This is so because the doubled DP was analyzed as being the actual argument of the clause, so that the only available position for the clitic within the clause would be the object agreement head position. Two counterarguments could be proposed against the two basic points stated above. More particularly, as far as doubling is concerned, let me point out that there have been other alternative analyses in the literature which have been proposed and which do not need to postulate that object clitics are agreement heads in the extended verbal projections. One of the most famous examples of such an approach is the so-called doubled DP hypothesis by Torrego (1988) and Uriagereka (1995), according to which the doubled DP and the clitic start off in the derivation as subparts of a larger DP which is the argument of the verb. Moreover, besides the fact that there are alternatives to the clitic as agreement head account the question arises whether the latter is also an adequate account More specifically, the clitic as agreement account may account for the crosslinguistic variation in terms of doubling on the basis of Kayne's generalization, according to which a preposition is available in the structure that may assign an extra case to the argument, since case (or case government) has already been absorbed by the clitic agreement head. In this way one can explain the difference between Spanish which allows clitic doubling and Italian which doesnt. Now, the problem is that Kayne's generalization is empirically false, as Anagnostopoulou (1994, 2005a), Suiter (1988) and Tsakali
Chapter 3. Internal structure of ditics and cliticlzation 97 (2006) clearly show: Greek, Portefio/Rioplatense Spanish and Macedonian. as well as other languages, double their D Ps without a preposition, and in fact some of them disallow doubling of a prepositional phrase. lf that is true, an account that takes clitics to be Agr heads and is based on Kayne's generalization in order to capture the crosslinguistic variation in doubling does not seem to be able to account for the latter, at least by using Kayne's generalization in combination with the base generation approach. As far as the second counterargument is concerned, despite their morpho-syntactic similarities object clitics also differ from subject agreement heads in certain important respects. First of all, object clitics are nominal as opposed to subject agreement morphemes which are not (contra Alexiadou & Anagnostopoulou 1998). In particular, object clitics morphologically belong to the nominal paradigm, they have a full set of nominal phi features, namely person, number and gender and they inflect for case (nominative, genitive and accusative). Subject agreement heads, on the other hand, are verbal in the sense that they only inflect for number and person, but not for gender or for case (unless of course we want to say that they form a unique series of exponents for nominative case that is used only with verbs to denote the subject but not the object). In this respect. it is quite unsurprising that object clitics look extremely similar to the definite article, whereas no such resemblance exists between the article and the subject agreement morphemes. However, besides morphology, the most convincing evidence that subject agreement is not nominal as opposed to object clitics which are, is the fact that Iatridou & Embick (1997) have put forward, namely that pro cannot take a CP/IP antecedent. because the latter- being not a nominal- does not have the right (i.e. nondefault) phi-features that will license pro. In this respect. note that although subject agreement cannot license pro, object clitics can do so, and hence make a CP/IP antecedent fully grammatical. This is so, because clitics are nominal and have a full set of person/number and gender features, as opposed to subject agreement which is verbal and hence lacks certain features (see also Iatridou & Embick 1997: ft. 22): (38) a.
*[An ftasume argha], pro tha tromaksi tin Maria. If arrive.1 pl late pro fut scare.3sg the Mary.acc 'If we arrive late, it will scare Mary.
[their (2b))
b. An o Kostas tilefonisi i Maria tha mas to pi pro. If the Kostas.nom calls.3sg the Maria.nom fut us.cl it.cl say.3sg pro. 'If Kostas calls, Mary will tells us that' [their (iia)] Now, if clitics are nominal heads, we wouldn't expect to find them as agreement heads within the extended verbal projection, as the latter only contains verbal heads (see Grimshaw 1991). As a result, the nominal character of object clitics, namely the fact that they contain a set of interpretable phi-features and case, renders the agreement hypothesis much less likely. Another major difference between object clitics and subject agreement is that the former are optional in many cases (although not all cases) while the latter are obligatory. Moreover, subject agreement may only refer to a single case role (nominative), which
98
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
is a typical characteristic of agreement morphemes, whereas clitics may refer to more than one case roles (genitive and accusative and even nominative in particular contexts), which is found with clitics but not with agreement (see Corbett 2003). In this respect, one could point out correctly that optionality is a more general property of direct object marking in a number of languages which cuts across such distinctions as agreement and cliticization. For instance, Hungarian may or may not have object agreement depending on the properties of the direct object (see Bartos 2001, and Szabolcsi 1994), and something similar appears to apply in Peninsular Spanish with regard to cliticization in relation to such features as animacy and specificity. For this reason. it would appear that optionality and obligatoriness do not really tell us much about the nature of the relation established between a verb and an argument, e.g. whether we have agreement or cliticization, apart from the structural relation of subject vs. object: subjects are obligatory as opposed to direct objects which aren't Note, however, that there is a slight difference between agreement and cliticization, and a qualitative one: while agreement is optiona~ it is selectively so. This means that agreement involves only certain types ofDPs (typically defined morpho-syntactically, e.g. definite ones) but not others, but agreement with these types is obligatory not optional. So, for example in Hungarian agreement with a definite direct object DP is not optional; rather, it is obligatory. The optionality lies in the selection of the kind of D Ps the verb agrees with: not with all types of DPs, but only with definite ones. Cliticization, on the other hand, appears to work differently, in that in some languages it may be truly optional. For instance, in Greek cliticization is optionally allowed with certain types ofDPs (e.g. definites), at least in most cases. This means that a definite object DP may or may not be cliticitized depending on whether it is interpreted as a topic/old information. This is a distinct type of optionality from that found with optional agreement As a result, although optionality may be found with direct objects cross-linguistically, we may still be bothered to distinguish between agreement and cliticization. A third difference between object clitics and subject agreement is that the latter does not impose any semantic restrictions on any doubled/coreferential DPs, as opposed to the former which do. 51 For example, direct object elitics in Greek cannot double a non-referential quantifier, while the same quantifier is absolutely fine with a subject agreement morpheme: (39) a.
*Ton idha kapjon. Him.cl saw.1sg someone.acc 'I saw someone:
b. Kapjos idhe ton Niko. Someone.nom saw.3sg the Nick.acc 'Someone saw Nick:
51. Although note that there seems to be a continuum, so that indirect object ditics Impose less
restrictions than direct object clitics, resembling in this respect subject agreement morphemes.
Chapter 3. Internal structure of ditics and cliticlzation 99 Note. however, that it is more the actual structural position of the doubled argument rather than its status as cliticized or agreed with that is important here. This is so. because semantic restrictions appear to cross-cut the clitic-agreement barrier, so that in e.g. Hungarian object agreement similar semantic restrictions apply. In other words, subjects would be more freely doubled than direct objects, independently of whether doubling involved agreement or cliticization. If that is true, the above difference does not really tell us much about the status of clitics as agreement heads or not Also, subject agreement morphemes may or may not act as topic markers (i.e. as non-focus, old information, prominence markers), as opposed to object clitics which are inherent topic markers. As a result, clitics are not compatible with focused constituents. This is illustrated by the examples below: (40) a.
o Vasilis/o VASILIS fonakse. The Vasilis.nom/THE VASILis.nom shouted.3sg 'Vasilis/VASILIS shouted:
b. Ti sinandise tin Elena!* TIN ELENA. Her.cl met3sg the Elena.acc/THE ELENA.acc 'She met Elena: Note however that indirect object eli tics appear to be more acceptable in focused constructions, thus forming some kind of a continuum with subject agreement morphemes being on the one end and object clitics on the other: (41) ?*ns ELENAS tis edosa ena mpukali krasi. THE ELENA.gen her.cl gave.lsg a bottle.acc wine.acc 'It was to ELENA that I gave a bottle of wine: tu edhoses to periodhiko? (42) ?*PJANU WHOSE.gen his.cl gave.2sg the magazine.acc Who did you give the magazine to?' (43) subject agreement > indirect object elitic > direct object eli tic If topicality/focus structure applies on the scale given in (43), then the difference between clitics and agreement does not seem to play any role at all. Instead, what we have is a continuum which is again based on the structural position of the constituent and not on its nature as agreement or clitic. Finally, as we have already discussed in this chapter, there are certain morphophonological differences between subject agreement morphemes and object clitics when they attach to the verbal stem, which would not be expected if both were agreement heads. For instance, whereas subject agreement will force the accent to move one
100
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
syllable to the right if it is the fourth syllable, a clitic will trigger the addition of a second accent: (44) 'a.na.pse vs. a. 'na.pse.te vs. 'a.na. 'pse. to Light.2g.imp. light.2pl.imp. light2sg.imp. it.cl 'Light! vs. Light it!' One way to go around this problem would be to assume that the subject agreement morpheme is added to the T node via fusion post-syntactically at PF, since it is a semantically empty/redundant node, contrary to clitics which have interpretable phifeatures and hence must be present in the derivation from the start, i.e. selected from the numeration and added to the derivation prior to PF. According to such an analysis, clitics will be syntactic constituents/heads that at some point of the derivation come to be adjacent to their host (to which they attach morpho-phonologically at PF), whereas subject agreement involves fusion of an agreement head into the T node at the postsyntactic level. One could assume that different morpho-phonological rules apply in each case, leading to the distinct morpho-phonological behaviour. All in all, in this section we have seen enough evidence to suggest that subject agreement and object clitics differ in important respects to be considered the same syntactic item. In the following section I will pose the question the other way around: what if clitics were agreement heads, then what would we expect of them? Do they behave like proper agreement heads? Or, are they different? And, in what ways? Well, first of all if object clitics were agreement heads, they would be added and/ or end up into the corresponding AGRo heads within the extended verbal projection either syntactically (through the derivation) or post-syntactically (by fusion to the relevant v-nodes), depending on the exact theory we follow. ln any case, by being uninterpretable, their phi-features would either be deleted/erased or not interpreted, and the PF would give the order V-obj.cl.-obj.cl-subj.agr., since structurally the subject agreement node lies higher than the object agreement nodes. The problem with these hypotheses is that they are empirically falsified: the phi-features of object clitics are interpretable most of the times (although not all the times), which means that they are not deleted or erased or are uninterpretable, while the right PF order is obj.cl. obj. clV- subj.agr.. Moreover, if object clitics were agreement morphemes, they would be subject to defective intervention (see Chomsky 2001a), according to which a probe cannot agree with a c-commanded goal if there is another goal structurally closer to the probe. According to Preminger (2009), Agree and Clitic Doubling behave differently as far as defective intervention is concerned: while in the case of Agree we get default phi-features, in the case of Clitic Doubling the elitic simply disappears giving rise to ungrammaticality. Preminger uses this difference in behaviour as a potential diagnostic tool by looking into whether it can correctly distinguish between Agree and Clitic Doubling cases (based on independent well established criteria). To illustrate, in Greek
Chapter 3. Internal structure of ditics and cliticlzation ditransitives an accusative DP cannot be clitic doubled across a structurally higher dative DP (see also Anagnostopoulou 2003a: Chapter 4): (45) f*O Kostas toni sistise [tis Marias]i The Kostas.nom him.cl introduced.3sg [the Mary.gen] [ton adelfo mu]i. [the brother.acc my.gen] 'Kostas introduced my brother to Mary. There are two ways to save this structure: one is to drop the clitic: (46) 0 Kostas sistise tis Marias ton adelfo mu. The second way is to cliticize the genitive DP tis Marias, so as to get it out of the way as an intervener: (47) 0 Kostas tis ton sistise tis Marias ton adelfo mu. The dropping criterion, as well as the doubling criterion (i.e, the fact that doubling of the indirect object clitic obviates an intervention effect, which is a standard cliticization criterion- see Preminger 2009 and Anagnostopoulou 2003a), clearly show that here the relationship is clitic doubling and not Agree, so we are dealing with clitics and not with object agreement morphemes. Finally, if object clitics were agreement heads, we would expect them to strictly correlate with a particular morpho-syntactic configuration. For example, a dative clitic would signal the presence of an applicative head and as a consequence of an applicative construction, whereas an accusative clitic would signal the presence of an accusative head and as a consequence of a direct object construction. Although this may be true for some languages, it is not for Greek, though. In Greek a sentence can have an app licati ve construction or a direct object construction without requiring the presence of a dative or an accusative clitic respectively (see Anagnostopoulou 2006 on this). Moreover, correlation of a cliticwith a particular AGR head (e.g. of an accusative clitic with the accusative/ direct object AG Ro) would not be able to explain how a clitic with no phi features (e.g.) may appear in the same position as a clitic with phi-features (for instance in the case ofbound clitics). Besides the above arguments, there are two further arguments against the agreement analysis of object clitics as agreement heads: (a) their movement properties; (b) their coindexation properties. As tar as (a) is concerned, it is well known that clitics appear on predicates they have not been selected by, that they may move out of participles or out of Prepositional Phrases, and that they are subject to the Specified Subject Condition, among other things (see Kayne 1975, Sportiche 1992/1998). The question that pops up is whether these movement properties could be accounted for within an agreement analysis. The answer is negative for two reasons: (a) they must involve movement since they obviate locality effects (see Anagnostopoulou 2003a and Preminger 2009 ), and (b) clitics give rise to higher scope and binding possibilities for
101
102.
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
their doubled DPs, which is not possible for agreed with DPs (see Anagnostopoulou 2003a on expletive-associate chains). As for (b), coindexation between the clitic and the DP is neatly captured via movement and multiple copy spell out (resumption). A third argument against the agreement analysis of object clitics is that they do not move as X0s, rather they land as X0 s. More particularly, it has been a well known fact that clitics are heads, however there has been some controversy whether clitics move as heads or simply are heads. Kayne (1975) for example analyzes moved clitics as heads although they are XPs prior to movement. In the '90's it was a very common approach to assume that clitics moved as heads. However, in 1995 Chomsky first proposed that this cannot be true: given his Bare Phrase Structure elitics must always be both heads and maximal projections. Moreover, recall that clitics are the only heads that may bypass a head or may be bypassed by a head (e.g. a past participle or an imperative), i.e. they are the only heads that do not obey the Head Movement Constraint (HMC). This would be plausible if they simply moved as non-heads, i.e. as maximal projections (or as both maximal and minimal projections). Also, Sportiche (1996) has convincingly shown that elitics move across a restructured clause to a higher clause as an XP and not as an X (since they trigger agreement on a higher clause participle). This would imply that e.g. in V-to-C movement in questions what we see is simply the clitics having landed onto the complex verb and moving along with it to a higher clausal head position. In other words, clitics land but do not move as heads. 52 How this is possible will be discussed later on in Chapter 4. Summing up, in this section I have tried to show that the hypothesis that object clitics are agreement heads faces some serious challenges and that the alternative that they are object clitics fairs better. Towards this goal, I presented evidence that undermines their similarity to subject agreement morphemes, I argued against their possible Agr properties and 1 also presented some arguments against an Agreement analysis. In what follows I discuss Sportiche's account which attempts to combine an Agr/base generation and a movement/ clitic analysis together. 3·3·4 Sportiche's account Here I will discuss some problems that Sportiche's analysis taces. First of all he puts a series of nominal heads within the extended verbal projection, although this should not be possible in principle (see Grimshaw 1991 ). Moreover, he claims that at least the accusative clitic voice specifiers are A-bar specifiers, even though there is enough evidence from binding and scope that in Greek these are A specifiers formed via Agree (see Anagnostopoulou 2003a and Chapter 4). Third, the voice filter sounds as an ad hoc solution that guarantees the existence of doubling and its cross-linguistic (non) availability, in other words the Doubling Parameter: only a language in which the XP 52. In fact, they do not even land as heads: they land as heads/XPs, and they move on as heads, after having incorporated.
Chapter 3. Internal structure of ditics and cliticlzation 103 movement (of an overt XP) can delay up to LF can have clitic doubling. One way around this problem would be to take this filter to be a PF filter, like other doubly filled filters in grammar (e.g. doubly filled camp filter): for instance, within an overt movement framework such as the one proposed by Bobaljik (2002) we could assume that the XP always moves overtly to the specifier of the clitic voice, however the voice filter only allows the derivation in which the lower copy of the XP is pronounced, since a doubly filled projection is not acceptable for PF reasons. The question of course would be why French or Italian does not have doubling, since the latter does not depend on parameterization in terms of LF any more, but rather on the general avoidance of a malformed pattern. In other words, we would expect this doubly filled pattern to be avoided/marked in every language, leading to doubling as well in French and in Italian. However, this is not what we see. It seems then that the PF filter is not the right solution to this problem either. A final argument against Sportiche's analysis is offered by Anagnostopoulou (2003a), who claims that the clitic must move overtly from the object DP position to the clitic voice position (i.e. v) since movement takes place for locality reasons and is EPP-driven. I refer the reader to Anagnostopoulou (2003a) for more details. On the other hand, there are two positive points in Sportiche's analysis. First, he tries to unify the syntax of cliticization/clitic doubling with the syntax of scrambling, which is quite insightful, given that there are many similarities between the two (see Alexiadou & Anagnostopoulou 1997, Leonetti 2004). Second, although the voice filter itself is ad hoc, the intuition behind it is pointing to the right direction: for Sportiche doubling will be possible eventually only in those languages which delay overt XP movement to LF. Alexiadou & Anagnostopoulou (1998, 2001) have pointed out exactly this, that clitic doubling, i.e. the formation of clitic/agreement chains is a way of moving constituents without actually moving them at all. This is a basic property of Greek, Spanish and other languages that have pro drop, clitics and more generally of languages that can form these clitic chains: it is the clitics that move, while the doubles remain in situ. In this respect, Sportiche's analysis is pretty insightful.
3·4 Type B: Pronouns as transitive XPs The second major type of analysis of pronouns is that they are complex structures of different degrees and types of complexity. The main difference between this type of analysis and the D-transitive analysis (with distinct internal structures) is that whereas the latter assumes that all pronoun types are DPs uniformly, but may differ in their internal structure, the former claims that different types of pronouns realize different portions/chunks of a pre-determined hierarchical layered structure. According to this type of analysis, different pronoun types have a distinct categorial status, which yields their distinct properties. Cardinaletti & Starke (1999) and Dechaine & Wiltschko (2002) are the two main examples of this analysis. They differ in that the former uses
104 Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
categorial status to distinguish between strong, weak and clitic pronouns, while in the latter the same distinction cuts across pronoun types (i.e. all types of pronouns may have more than one sub-types/categories, so that both strong and clitic pronouns e.g. can be of the same category). 3·4·1 Cardinaletti & Starke (1999)
Cardinaletti & Starke (1999) (and Cardinaletti 1994) postulate a transitive determiner analysis, but for strong pronouns only. This is in line with their more general theoretical approach to the three distinct pronominal types they independently postulate on the basis of empirical data. In particular they argue that strong, weak and clitic pronouns form a deficiency hierarchy, where strong pronouns are less deficient than weak pronouns, which are less deficient than clitic pronouns. This deficiency hierarchy spreads across all levels of grammatical description and is ultimately related to structural deficiency, which is analyzed as the actual cause of all types of deficiency: each class contains (one) more (or less) morphological head/projection than the next one in the hierarchy, so that clitics are the least complex structures compared to weak and strong pronouns. For strong pronouns they give the following structure (see Cardinaletti & Starke 1999: 195). In their theory this is the fullest possible structure a pronoun can have: (48) Strong pronouns CLP
~ CLO
~Lp
(L =any lexical category) In (48), the traditional D-category is decomposed into two distinct heads, namely C and 1.53 The former contains referentiality/definiteness features, while the latter contains phifeatures. These two feature sets may be realized as a single morpheme or as two separate morphemes, depending on the given language. Moreover, C can host dummy (case) 53· I stands for INFL, and is considered to be the parallel head to INFL in the verbal domain. Note that Cardinaletti (1994) uses a singleD head, which dominates a Support head (:E here).
Chapter 3. Internal structure of ditics and cliticlzation markers like the Spanish marker a, or the Romanian pe (which encode a human feature), and more generally complementizers (i.e. subordinators ), similar to those found in complex prepositions or complex complementizers. Cardinaletti & Starke further propose that the content of C can be successfully reduced to a single feature, namely (abstract) case, since the latter entails a referential index (and hence a range, which for strong pronouns will be a default [+ human] range, yielding in this way all the semantic properties of full pronouns, namely referentiality/definiteness and a human interpretation). Moreover, other properties of strong pronouns are equally attributable to the presence of the C-head: modification, coordination, non- phonological reduction/restructuring, case assignment in situ, and impossibility of an expletive, non-referential or impersonal reading. In other words, it is precisely the presence of this particular morpheme/head (along with its features) that (partly) makes a strong pronoun what it is, and distinguishes it from weak and clitic pronouns (which presumably do not have a C head). 1: is considered to be a dummy support morpheme, a head which in some languages may also appear in the verbal domain (cf. Lakas 1990 1: projection, which is merged above INFL and which hosts emphatic or negative support morphemes, and contains polarity (assertion/negation) and focus features). Cardinaletti & Starke propose that 1: is the locus of the prosody-related features of L; as a result, its presence entails the presence of inherent/lexical stress. Again, absence of this head entails absence of word stress, and subsequently obligatory restructuring. According to their theory, this is what happens with clitics, which lack 1: (as opposed to weak pronouns, which may but do not have to restructure, since they have the 1: head in their structure). I(NFL) contains phi-features (person, number and gender), on a par with the corresponding head in the verbal domain, while L contains the lexical N head (which is typically null and semantically dummy (i.e. range-less) in pronouns). Note that both IN FL and L are assumed to be shared by all types of pronouns. Accordingly, the structure they propose for weak and clitic pronouns, as opposed to that of strong pronouns, is the following: ( 49) weak pronouns
clitic pronouns
LLP
ILP
~ Lo L
~
LP ~
As we can see, weak and clitic pronouns are more deficient than strong pronouns, in that they have less structure/heads (peeled-off structures). For Cardinaletti & Starke, this difference is enough to account for all the shared and non-shared properties among the different pronominal types, including syntactic distribution. In particular,
105
106
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
they argue that only strong pronouns can be inserted in the e- position. where deletion of the Cor the L head is possible (in fact, preferred), up to crash. However, the deleted structure/features (either a weak or a clitic pronoun) must be recoverable at all levels of representation (due to the Projection Principle). Recoverability takes place in the syntactic derivation under a local configuration between the deficient pronoun and an appropriate X0 head (where local configuration = Spec- Head relation or incorporation to X0 ). Accordingly, a weak pronoun moves to [Spec, AGRoP] overtly, so as to establish a local Spec-head relationship with AGRo (which will allow the pronoun to recover a case teature), while a clitic will have to move to [Spec, AGRoP] tor the same reason. but also to take an extra head step to a higher functional head (Lor V), in order to recover the prosody-related features. The reason why head movement is required (and not for example XP movement) is that the clitic must maintain a local relationship with two heads simultaneously (given that recoverability is dependent on the local relationship), so that the only possible derivation is to have an XP step followed by an X0 step (i.e. a combination ofboth plausible types oflocal configuration). All in all, Cardinaletti & Starke (1999) assume a transitive determiner analysis only for strong pronouns, exactly because their analysis capitalizes on structural differences (i.e. degrees of deficiency) among pronoun types in order to explain their different properties. In this respect. their approach differs both from that of Panagiotidis and Carver & Delfitto, according to which all pronouns have the same internal and external structure and! or features (transitive DPs ), and from that of Cardinaletti ( 1994) or Ritter (1995), according to which pronouns are all DPs although they may differ with regard to their internal structure/features. As far as I can see, there are two main problems with the Cardinaletti & Starke (1999) approach: recoverability and structural deficiency with its one to one repercussions on morphology, syntax and semantics. Starting with recoverability, according to Cardinaletti & Starke (1999), pronouns start out strong but subsequently are peeled off, by loosing structure, to their weak, clitic or null counterparts via "Minimize a up to crash". Since minimization involves actual deletion of syntactic structure, the pronouns need to compensate for the lost information by moving within the built tree in order to be in a local structural relation with a particular head (e.g. AGR). The idea behind these movements is that they are triggered by some kind of interpretation of the Projection Principle according to which "Information of Level R must be present at R + n': which forces the recovery of the erased features. The question that arises is how recoverability allows any differences in the syntax, semantics and morpho-phonology of weak, clitic and strong pronouns to survive at LF and PF, if recovery exists. In other words, if recovery is a mechanism in syntax that ensures that any feature that gets erased and hence makes a pronoun different from a strong pronoun gets recovered/non-erased by LF/PF, we would expect to find only strong pronouns at the interface levels, which is however empirically wrong. That is somehow, either PF/ LF must be able to see behind the recovered features or the syntactic movements of weak and clitic pronouns are not related to feature recovery.
Chapter 3. Internal structure of ditics and cliticlzation As for the second problem, recall that in Greek as well as in other languages with clitics and null pronouns (e.g. Spanish) a strong pronoun may be used instead of a clitic in a construction whenever a clitic cannot be used for independent reasons. For instance, it is known that clitics cannot be used after certain prepositions (for independent reasons which will not concern us here) either alone or when coordinated: (SO) To podhilato perase ksista apo aftin ke ti fili tis. The bike.nom passed.3sg closely from her.str and the friend.acc her.cl 'The bike passed extremely close by her and her friend:
gia aftin ke to mora tis. (51) Milusane Were.talking.3pl about her.str and the baby.acc her.cl 'They were talking about her and her baby.' The interesting thing about these strong pronouns is that they seem to behave like weak pronouns in other respects. For example, they have weak semantics, in that they can be bound variables: (52) Kathe fititis gnonZI oti i gonis tu milane gi afton Every student knows.3sg that the parents.nom his.cl talk.3pl about him.str stan kathigiti tu. to-the professor.acc his.cl 'Every student knows that his parents talk to his professor about him: Normally, this is not possible for a strong 3rd person pronoun in Greek. For Cardinaletti & Starke's account, such a hybrid pronoun, a strong form with weak semantics, is a problematic form, since their theory only allows strong structures with strong syntax and strong semantics, or weak/clitic structures with weak/clitic syntax and weak/clitic semantics, but no combinations. Now, if the torm is strong, this means that a strong torm with strong syntax and strong semantics and morpho-phonology has been chosen from the syntactic lexicon, i.e. from the start of the derivation, and as a result no weak semantics would be predicted to be possible. Although it is not my purpose here to give a theory ofhow such hybrid forms arise in the first place, I think that something towards the syntax-PF interface would suffice to account for what's going on. For instance, one could assume that what we have here is a clitic form that get's spelled out as the strong form, which is the unmarked form, due to independent reasons. In any case, what is important to keep in mind here is that Cardinaletti & Starke (1999) would predict that these forms would be strong both structurally and semantically, which is not empirically correct
3.4.2 Dechaine & Wiltschko (2002) Dechaine & Wiltschko (2002) also offer a transitive XP approach to pronominals. More particularly, they argue that the notion pronoun is not a primitive and they
107
108
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
militate against an analysis that treats pronouns as DPs uniformly, as well as against analyses that account for the differences found across types of pronouns by postulating distinct internal structures for each type (cf Ritter 1995). Instead, they suggest that these differences should be attributed to the distinct categorical status that different types of pronouns display. In particular, they propose that pronouns (strong, weak and clitic) across languages may be distinguished into three kinds: pro-DPs, pro-
b.
DP
---~
/ D
c. NP
~ NP
NP
I N
N
I N In this way they draw attention to the fact that there are important correlations between the binding properties of each type and their categorical status, in the sense that the latter determines their internal semantic and syntactic properties which in tum determine their binding-status. Pro-DPs are morphologically complex and have the internal syntax of determiners and the external syntax/distribution of arguments. They have definite interpretation and are subject to Principle C, i.e. they are R-expressions. ProPs, whereas English 1st and 2nd person pronouns are analyzed as pro-DPs and 3rd person pronouns as pro-q>Ps. The question is how Greek clitics would be analyzed according to this proposal. Recall, as we have said before, that in Greek the pronominal system is based on the distinction strong vs. clitic pronouns. Strong pronouns typically refer to new referents (cf. 54), may not be interpreted as bound variables (cf. 57) and are subject to Binding Principle C (cf. 55-56), i.e. they are R-expressions.54 Clitics 54· This does not hold for 1st and 2nd person strong pronouns which are allowed to be A- bound (i) pro; stenoxorieme otan kalun eme11a1 (# apo lkto) feel bad-lsg when invite-3pl me-ace (from pitty-acc) 'I feel bad when they invite me (out of pitty):
Chapter 3. Internal structure of ditics and cliticlzation
on the other hand refer to old/prominent referents (cf. 54), may be interpreted as bound variables (cf. 58) and are subject to Binding Principle B (cf. 55-56, 58): (54) Xtes mu milise o Kostas gia Yesterday me-clgen talked-to-3sg the Kostas-nom about the tin Elenii, ke malista mu [" tini edikse] I [# edikse Helen-ace, and in fact me-clgen her-dace showed-3sg I showed-3sg aftini] se mia fotografia. her-accstrong in a photograph-ace (55) Otan [*kalune afto~] /["to~ kalune],o Janisi enthusiazete. John-nom gets-3sg excited When invite-3pl him-acclhim-acccL the *'When they invite himi, Joh~ gets excited: (56) 0 Janisi enthusiazete, otan .. [kalune aftoni] I "[toni kalune]. 'Joh~ gets excited, when they invite himi:
(57) *Kanenasi den ipe oti pistevi oti i Dorita antipathi aftoni Nobody-nom not said-3sg that believes-3sg that the Dorita-nom dislikes him-ace 'Nobody said that he believes that Dorita dislikes him: (58) Kathe mathitisi pistevi oti tonili psaxnun. Every student-nom thinks-3sg that him-dace look-for-3pl 'Every student thinks they are looking for him:
Based on the Binding Theory facts we could propose, following Dechaine & Wiltschko that strong pronouns are DPs, while clitics are
(ii)
pro1 stenoxorieme feel bad-lsg
otan me1 kalun apo ikto when me-dace invite-3pl from pitty-acc
(iii)
Ego1 nomizo oti l!MENA1 agapail agapai emenar 1-nom think-lsg that ME-ace loves-3sg 'I think it ls ME that he/she loves:
The same applies to focalized 3rd person pronouns. (iv)
0 Janls 1 .lpe oti APTOS1 itan 0 The John-nom sald-3sg that HE-nom was-3sg the kaliteros mathitis/o kaliteros mathitls itan aftos1• best student- nom 'John said that HE was the best student:
In non-argumental positions, strong pronouns dont even need to be stressed (familiar reading): (v)
Eknevrizome1 otan milane gia mena, me kakia. get-irritated-lsg when talk-3pl about me-ace with malevolence-ace 'I get irritated when they talk about me with malevolence:
Similar facts appear to hold in other languages, like e.g. English. See Decb.aine & Wiltschko (2002) for a plausible analysls of these facts based on Strong Crossover and Quantifier Raising (QR).
109
uo Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
All in all, it appears that if we follow Dechaine & Wiltschkds diagnostics, we end up categorizing strong pronouns as DPs and clitics as q>Ps. Although this picture may look neat and tidy, there are some issues that need to be addressed before we consider the whole matter closed. First of all as it has been noted in the literature, some of the diagnostics that Dechaine & Wiltschko use in their paper seem to lead to opposing views once more data are taken into consideration. To illustrate, 1st and 2nd person pronouns in English cannot take a complement (like in 'we linguists') if they are in the singular (such as * 'I linguist' or * 'me linguist'). As a result, one and the same diagnostic would categorize them as DPs and q>Ps at the same time, which would be quite problematic. Another issue with their analysis would be their claim that 1st and 2nd person English pronouns may not be bound variables. As Rullman (2004) has very convincingly shown, this is not true, and in fact one can show it quite easily with the following example: (59) You need someone to put you to bed, and Dina does too. =You need someone to put you to bed, and Dina needs someone to put her to bed. =You need someone to put you to bed, and Dina needs someone to put you to bed. If that is true, then according to this test 1st and 2nd person pronouns should be q> Ps and DPs (since there are other sentences which do not allow a sloppy reading, at least according to Dechaine & Wiltschko). Similar problems would arise with the 'predicate' diagnostic, since they have to argue that sentences like 'That's me/us/you' are not predicate uses, but equative structures, involving two DP expressions. Besides the fact that it is not clear why these are not predicate uses, again they fail to explain why the same examples are not ok with nominative case (cf. e.g.* That's I/he/she/we). That is, data from the same diagnostic give rise to distinct classifications, since when they are in the accusative they are classified as q>Ps, whereas when they are in the nominative they are classified as DPs. A further and quite important issue involves the characterization of Greek elitics as q>Ps. This classification entails that clitics have no inherent semantics and that they are bound variables. Although the latter is true, i.e. although they are bound variables, the former is not true, at least for 3rd person clitics, which are definite determiners. This is evidenced first of all by their morpho-syntax, namely by the fact that the paradigm of the 3rd person clitics (sg and pl) is almost identical with that of the definite article. Secondly, by the fact that the interpretation of these clitics is always definite (and specific, at least in most cases). And thirdly, by the fact that these clitics double definite DPs (see Anagnostopoulou 1999), i.e. DPs introduced by the definite article. 55 55· Note that according to a number of researchers (see e.g. Anagnostopoulou 1999) ditic doubling ls subject to definiteness. Nevertheless, there are also other researchers (see e.g. Kallulli 2001) who have argued against this analysis presenting data that show that doubling is possible with both definite and indefinite DPs, and that the crucial factor is topicality or [- focus].
Chapter 3. Internal structure of ditics and cliticlzation
All these facts show that clitics must be definite determiners both semantically and syntactically (see also Chapter 2). According to Dechaine & Wiltschko's theory determiners are R-expressions. Note, however, that we already know that elitics are (bound) variables. In other words, clitics can be both D-elements and variables. This is, however, an important contradiction that makes their theory look at least problematic.56
3·5 Summary
Summing up the above discussion, we have seen that a whole range of proposals have been put forward in the literature in order to account for the properties of pronouns and the subtypes thereof. Moreover, it was shown that these proposals can be subdivided into distinct subtypes, depending on whether they assume the same internal and/or external structure for distinct pronominal types. These distinct subtypes are summarized in the following tree: (60)
Pronouns DP/D 0 (Type A)
Intransitive D
XP//X0 (Type B)
Transitive D
Heads on Extended r-------.:rbal Projection (D or AGR)
All Clitics Distinct pronouns only internal structure
Same internal structure
Although I think that KallulU is eventually correct, Anagnostopoulou, on the other hand, has thrown light to an important generalization. namely that doubling in Greek is sensitive to the phonological/overt realization of the D head, either via the definite article or via the indefinite article or some other quantifier. This is evidenced by the fact that doubling of an indefinite quantifier gives a much better result than doubling of a noun headed by a null D, which is entirely out. It appears then that indefinites or other quantifiers can be doubled as long as they satisfy the -focus requirement of doubling and they realize the D head of the doubled DP, on a par with definite (see also Chapter 4). 56. Perhaps, one could claim that these are instances of the so-called expletive article (see Vergnaud & Zubl.zaretta 1992, Longobardi 1994), which is a q>. Even with such an assumption. doubling of non-generics and non-proper names would require aD-head on the doubled DP, and as a result on the ditic as well (assuming that the ditic and its doubled DP must agree in their features - see Suil.er 1988).
111
CHAPTER4
Derivation of proclisis
4.1 Introduction
In this chapter I will discuss the derivation of proclisis in Greek. I will argue that clitics A-move as XPs/Xs to the specifier of v*/T into which they incorporate as the result of being minimal non-phases. Incorporation has an effect on the outcome, by satisfying the PF and LF properties of clitics. I will also discuss auxiliary echo-constructions, indirect object constructions and clitic doubling constructions.
4.2 Properties of elitics 4.2.1 Clitics: Movement vs. base generation
Within the generative literature there have been two main analyses of cliticization, the movement approach and the base generation approach. According to the movement analysis, which was first proposed by Kayne (1975) for French clitics, clitics are morphemes that get bound onto the verbal host by moving from the postverbal object position to the preverbal position. This movement is called cliticization. The base generation analysis, on the other hand, claims that the clitic is directly merged on the verb, while an empty pro in the object position is associated with the clitic. Among the arguments in favor of the movement approach are the complementary distribution argument, the argument that the clitic-XP relation displays properties of movement, the Participle agreement argument, and the non-local character of cliticization argument. For example, in French and Italian an object clitic is in complementary distribution with an object DP, which has been used as evidence in favor of the fact that the elitic is generated in the object position of the clause whence it moved to its surface position next to the verb: Je le vois. I him see 'I see him~
French (France)
b. Je vois Jean. I see John. 'I see John:
French (France)
(1) a.
114
Cities in Greek: A Minimalist Account ofProdisls and Endisis
*Je le vois Jean. I him see John 'I see John:
French (France)
Furthermore, in French a clitic cannot appear in the main clause apparently moving across an embedded subject, a Specified Subject Condition (SSC) violation which provides evidence in favor of syntactic movement (example adjusted from Kayne 197 5:288 (his 42-43)): (2) a.
Cela fera [leur fils telephoner a ses parents.] French (France) That will make their son telephone to his parents 'That will make their son telephone to his parents: Deep structure
fera telephoner a ses parents. b. Cela le That him will make telephone to his parents 'That will make him telephone to his parents:
French (France)
*Cela leur fera telephoner leur fils. That them will make telephone their son That will make their son telephone to them:
French (France)
c.
Moreover, in languages like Italian or French an object may or must (depending on the dialect) trigger participle agreement when it precedes the participle, for example when it is relativized, questioned or cliticized. Assuming that agreement involves a spechead configuration and that the relevant specifier lies somewhere in between the object position and the verbal host position, it follows that the clitic must move through that particular specifier in order to get the agreement e1fect1: (3) a.
Jean a peint(*E) la porte. John has painted(agr) the door 'John painted the door.
French (France)
b. La porte que Jean a peint(E) t The door that John has painted(agr) 'The door that John painted:
French (France)
c.
Jean fa peint[E] t French (France) John ithas painted(agr) 'John painted if (example taken from Sportiche 1992/1998: 16 (his 22a-c))
Adam Ledgeway asks whether the same agreement fucts could be accounted for through a base generation approach of ditics. It is not dear to me how agreement between the base generated ditic on the auxiliary and the participle would be possible, given that in this case the goal would c-command the probe. Moreover, we would loose the parallelism between ditidzation and question formation and relativization: all three cases involve movement of a constituent through the specifier of a past participle. 1.
Chapter 4. Derivation of proclisis 115 Finally, the postulation of a clitic placement transformation by Kayne correctly predicts the possibility of non-local cliticization, namely the ability of clitics to appear on heads by which they have not been selected, as in the following examples from Greek:
Su ine pistos. Youel is.3sg faithful.masc 'He is faithful to you:
Greek
b. Ton ekana na fai. Him.cl made.lsg subj. eat3sg 'I made him eat:
Greek
(4) a.
The words in italics are the ones that theta-select for the clitic in each sentence, however the latter is found attached to the matrix verb ine and ekana respectively. A basegeneration approach would complicate selectional rules and restrictions considerably. In favor of the base generation approach to clitics are the non -complementarity of clitics and object DPs, as well as the presence of inherent clitics and of ethical datives (or more generally what is known as 'the source problem'). First of all. Kayne's argument from complementarity has been challenged by the existence of languages where there is no complementarity between an object eli tic and the corresponding object DP, such as Spanish, Greek, Romanian or Hebrew, the so-called clitic doubling languages. The following example is from Greek: (S) a.
Ton vlepo. Him.cl see.lsg 'I see him:
b. Vlepo ton Jani. See.lsg the John.acc 'I see John: c.
Ton vlepo ton ]ani. Him.cl see.lsg the John.acc 'I see John:
Greek
Greek
Greek
Here we see that ton !ani is not in complementary distribution with the clitic ton. Assuming that ton Ja.ni is in the object position, it is not clear where the clitic is moving from, hence the proposal that it is directly base generated on the verbal host Moreover, the only difference between (Sa) and (Sc) is that in (Sa) there is an empty pro base generated in the complement position, whereas in (Sc) there is a full DP. In such an analysis, SSC effects of the type illustrated in (2) would not constrain movement, but the relationship that is established between the clitic and the object position/empty category in V.
116
Cities in Greek: A Minimalist Account ofProdisis and Endisis
Similar concerns arise with regard to ethical datives. These may only be realized by clitics and not by full DPs, which suggests that they must be base generated as clitics on the verb. This is evidenced from the following example from Greek: (6) *(Mu) xtipisan emena to pedhi. (me.gen.cl) hit3pl me.gen.str the child.acc 'They hit the child on me:
Greek
The same applies to what Sportiche ( 1992/1998) calls inherent clitics. These are semantically empty clitics which are used in idioms and are very widespread at least in Greek but also in other languages with clitics (e.g. French) (for Greek see Bibis 2002): (7)
Tin
epesa gia m1s1 ora. Her.cl fell.1sg for half hour.acc 'I slept for half an hour:
Greek
The issue here is that these clitics do not seem to satisfY any theta-role connected to the main predicate, and hence they do not appear to be real objects to the verb from which they would move to their host 2 The standard arguments are not enough to let us decide in favor of or against one approach over the other. For this reason I will briefly discuss a number of further arguments which will allow me to follow Kayne's movement hypothesis and to argue that in Greek as well clitics move from the object position(s) of the clause, where they merge as syntactically independent words, to their surface position. This proposal is based on facts put forward in the literature, which show that clitics move in the narrow syntax. First of all, there is sufficient evidence that elitics A-move to the specifier of v*: (i) Movement is local: iksera oti i Maria apata. (8) .. Ton Him.cl knew.1sg that the Mary.nom cheats.3sg on 'I knew that Mary cheats on him:
Greek
(ii) It does not manifest weak cross over effects: (9)
To" parusiasa tis ghinekas tu. Him.cl presented.3sg the wife.3sg.gen his.gen.cl 'I presented him to his wife:
Greek
2. But see Sportiche 1992/1998 for an answer pointed out to him by R Kayne. Adam Ledgeway points out that ethical datives and inherent ditics do not necessarily argue against the movemement approach; they could be the exception, the real lexical/base-generated examples not associated with an argument position.
Chapter 4. Derivation of proclisis 117
(iii) It does not license parasitic gaps: (10) *To archiothetisa choris na dhiavaso. It.cl filed.lsg withoutsubj. read.lsg 'I filed it without reading it:
Greek
(iv) It licenses floating quantifiers: (11)
Tus echo dhi olus. Them.acc.cl have.lsg seen all.acc 'I have seen them all:
Greek
(v) It feeds binding, by obviating weak crossover violations or by inducing weak crossover violations. For example in (12b) movement of the clitic obviates the weak crossover violation in (12a): (12) a.
b.
?*Q Nikos epestrepse dhiorthomeni tu fititi The Nik.os.nom retumed.3sg corrected the student.gen tis tin kathe erghasia chtes to proi. her.cl the every essay.acc yesterday the morning 'Nick returned his student every essay yesterday morning:
Greek
0 Nik.os tin epestrepse dhiorthomeni tu fititis tis tin kathe erghasia chtes to proi. Greek
On the other hand, the reverse effect happens in (13a-b ): clitic doubling of the direct object leads to a weak crossover violation: 3 (13) a.
Sistisa tu kathe kathighiti ti ghramatea tu. Greek Introduced.lsg the every professor.gen the secretary.acc his.gen 'I introduced every professor to his secretary:
b. ?*Ti sistisa tu kathe kathighiti ti ghramatea tu.4 'I introduced his secretary to every professor:
Greek
Moreover, there is evidence that clitics move to v* and/or Tin order to check an EPP feature with their nominal D- feature. This is shown by Anagnostopoulou (2003a) who discusses a series of constructions involving passives, unaccusatives and raising 3· There ls also evidence that doubling of direct objects turns a stage level predicate into an individual level predicate. For instance the verb plno ('to drink') is interpreted as an individual level predicate (meaning 'like') if the object is doubled:
To pino to krasi. ltd drink.lsg the wine.acc 'I drink the wine' (meaning I like drinking the wine). This is related to the derived object position of the doubled DP. (i)
Greek
4· Elena Anagnostopoulou points out that the ungrammaticality of this example might not be related to the creation of a WCO effect but might have to do instead with topichood!spedficity.
us
Cities in Greek: A Minimalist Account ofProdisis and Endisis
constructions. In all of these doubling of the genitive clitic is obligatory, it involves spelling out offormal nominal features that undergo overt EPP-triggered movement to T, and it allows checking of the formal features of the intervening genitive phrase and hence subsequent checking/movement of the lower nominative theme: (14)
teliose tis Popis o kimas. (her.gen.cl) ended.3sg the Popi.gen the mince.nom 'Popi ran short of mince meat:
?.. (Tis)
Greek
Finally, clitic climbing gives further evidence for movement, as we find it in a number of constructions: past participles, adjectives, kana-constructions, among others. 5 (15) a.
To echo idhi teliosi. Itel have.l sg already finished 'I have already finished if
b. Su ine pisti? You.gen is.2sg faithful.fem? 'Is she faithful to you?'
Greek
Greek
Based on this evidence we can conclude that clitics in Greek move to their surface position. I have already presented evidence thatthey A-move, i.e. they move asXPs/ A-specifiers (cf. also participle agreement in Romance). However, the hypothesis that they move (at least partially) as heads (see e.g. Kayne 1991, Belletti 1999, Carver & Delfitto 1999) has always been part of the traditional movement analysis of clitics. So, the question that needs to be answered now is whether they move as heads, as XPs or as both X0 and XPs. 4.2.2
XP vs. X0
The standard assumption in the generative literature has been that clitics move both as XPs and as X0s. This was due to the conflicting evidence regarding their movement properties. For instance, the tact that clitics were subject to the Specified Subject Condition or that they triggered Participle agreement without violating the Head Movement Constraint (HMC) counted as evidence in favor of their XP status. On the other hand, the fact that clitics could not move across an embedded negation head or a complementizer to the matrix clause (see Kayne 1989b and Terzi 1997), apparently due to a violation of the HMC, showed that they must also move as heads. The same applied to all those analyses which proposed left-adjunction of the clitic to a functional head, such as e.g. INFL, since the landing site was a head. Moreover, if the functional head happened to move, the clitic would move along, as for example in the case ofV-to-C. 5· Adam Ledgeway asks whether ditic climbing is possible out ofPPs and NPs in Greek, as is possible in Romance. Greek does not have partitive clitics like Italian ne or prepositional clitics like French y, therefore it does not have clitic climbing out ofPPs or NPs.
Chapter 4. Derivation of proclisis 119
It is not however clear to me whether clitics move as X0s or simply land as X0s. What I mean by this is that the evidence for head movement is not conclusive that elitics move on their own as X0s. First of all, Sportiche (1996) has convincingly shown that what Kayne's (1989b) evidence really shows is that a negation head or a complementizer does not allow restructuring, which in turn does not allow clitic climbing or long object preposing. The latter are allowed as long as restructuring has taken place, as is evidenced by the fact that clitics may trigger participle agreement in the matrix clause in constructions in which restructuring has taken place: (16) Non li avrei saputi I *saputo a chi dare Not them I-would-have known-pl/*known-sg to whom give 'I would not have known to whom to give them: Italian (Italy) Now, V-to-C movement and left-adjunction to INFLdo not necessarily show that clitics move as X0s; they could equally show that clitics land as X0s.6 The truth is that typically landing entails moving. That is, if a constituent lands as an X0, it must have moved as an X0, and vice versa. However, with clitics things are different Recall that clitics are both XPs and X0s. This follows for free from Chomsky's Bare Phrase Structure Theory (see Chomsky 1994), given that clitics do not project at all (hence are minimal) but do not project any further (hence are maximal) at the same time. If that is true, we can assume that clitics generally move as X Ps/X0s, but as X0s only after they incorporate into the verbal functional projection. 7 In this way we can explain why elitics are the only 'heads' that may violate the HMC: for example, they may precede past participles or the main verb, or they may be preceded by gerunds or imperatives across a number of languages. 4.2.3 Clitic placement Adjunction or incorporation?
In this section I discuss the exact nature of clitic placement. in particular the way the clitic attaches onto its host. I show that current proposals for cliticization cannot capture the Greek facts, and propose my own solution to the problem. In the literature there are two main approaches to cliticization: (a) the syntactic approach and (b) the PF approach. (a) argues that clitics adjoin to their host (via syntactic head movement) within narrow syntax (ct~ e.g. Rizzi 1993; Belletti 1999), whereas (b) proposes that strictcliticization (i.e. attachment to the host) is a PF (or LF (= Logical Form) phenomenon. where the elitic head-moves (and adjoins) to an empty 6. Adam Ledgeway correctly points out another possibility: namely, that clitics land as XPs, their X0 status arising only after incorporation to the verbal host. In fact, this ls what I am arguing below, when I am saying that ditics move as XP/X0s but as X0s after they incorporate into the verbal functional projection. 7· Another possibility which was suggested to me by Ian Roberts (and see also Roberts 2009) is that ditics move as heads, and there ls no HM C to be violated.
120
Cities in Greek: A Minimalist Account ofProclisis and Enclisis functional head c-commanding the head that contains the host, the two merging after Spell-Out (cf. Kayne 1994). The PF approach does not make the right predictions with regard to Greek cliticization. First of all, it leaves room for interpolation (cf. Kayne 1991 ), which is nevertheless totally impossible in Greek: 8 (17) 0 Janis (~ sosta) ttl (* sosta) apantise. The John-nom (rightly) him-clgen (rightly) answered-3sg 'John did right to answer to him: In (17) a subject-oriented adverb cannot intervene between the clitic (which would be adjoined to the head right above T) and the verb (in T), contrary to what the analysis predicts. 9 Accordingly, the clitic and its host must already be a single (syntactidmorphological) unit before they actually move to T. 10 Note that one could not possibly Unavailability of interpolation does not mean that syntactic clitic!zation must be at work, as other factors may be involved (e.g. non-generation of the relevant adverb in that particular position in some language, or a morphological effect a la Bobaljlk 1995). However, it suggests it, since: (a) the particular adverb in (17) is independently available in that position; (b) clitic+host move as a single morpho-syntactic element already (at least) after the vP level (see main text below). Moreover, an adverb may intervene between a clitic-like element (the article) and its host in the nominal domain, where crucially D and N are on separate heads. 9· Following a suggestion by Adam Ledgeway, an alternative reason for the ungrammaticality of (17) could be thatthe clitic is attracted to v*, but because there is no v-to-T movement the clitic cannot appear to the left of scsto. Although this is a posslbility, there should be at least one case ofv-to-T movement (given that in Greek v-to-T movement satisfies the EPP feature on T - see Alexladou & Anagnostopoulou 1998) and hence of interpolation, contrary to fact. 8.
Note that by the same reasoning V + particle complexes must also form a unit before they move to T, since they also do not allow interpolation (Elena Anagnostopoulou p.c.). The question is whether this is plausible. One idea would be to analyze particles along the lines of the German particle zu, i.e. as morpho-syntactic features that are part of the verbal head from the start of the derivation. This would account for both the lack of interpolation as well as for the fact that they do not take wide scope over a conjunction, on a par with clitic pronouns. However, note that this would render V-movement to CModal V-related, and hence would predict encllsis for subjunctives and for other forms taking particles in CModal, contrary to fact. Therefore, a more adequate solution would be to assume that particles are independent heads that merge in the left periphery of the clause and that cliticize to/merge with the verb at PH Note that clitics must merge syntactically with their host so as not to allow interpolation, since we see them moving up the clause adjoined to the verb. On the other hand, we do not see the particle and the verb moving together to higher positions; only the particle may move, otherwise it blocks V-movement. This is an argument in favor ofPF cliticization in the case of particles. As a result, lack of interpolation in clitics and particles is due to distinct factors (morpho-syntactic vs. morphological/PF respectively). Finally, as far as the position of V+ particle complexes with regard to adverbs is concerned, these complexes may appear higher or lower in the clause (e.g. higher or lower than aspectual adverbs). This is possible because the PF may select a higher or a lower copy of the verb to spell out Presumably, the particle is also part of the lower copy,
10.
Chapter 4. Derivation of proclisis t:n
argue that (17) is out because clitics cannot take adverbs as their (phonological) host: there are examples of cliticization onto an adverb, namely in the nominal domain, where the definite article is a determiner clitic and PF-attaches to whatever is on its right (cf. (18) and see Alexiadou 2001 for evidence that there is no N-to-D movement in Greek). Therefore, further stipulations would be needed to account for the ungrammaticality of (17). (18) To tromera eksipno pedhi vs. to eksipno pedhi vs. to pedhi The very smart kid the smart kid the kid The asymmetry of the cliticization process in (17) and (18) is further supported by the fact that the definite article may take wide scope 11 over a conjunction, as opposed to verbal clitics: (19) a.
0 filos ke sinadhelfos. The friend-nom and colleague-nom 'The friend and colleague:
dhiavaza ke *(to) dhiorthona. b. To It-clacc read-imperf-lsg and (it-dace) corrected-imperf-1sg 'I was reading and correcting u:
Greek
Greek
(19a) trivially follows from the higher position of D with regard toN. (19b) on the other hand suggests that the clitic and its host must be located syntactically on the same head, therefore PF-attachment is not plausible without further stipulations. Kayne (1994) claims that cases like the one in (19b) cannot be analyzed as instances of V+V conjunction (which is independently banned by the Linear Correspondence Axiom), exactly because if V-coordination were available, nothing would rule out these constructions (c-command being available), indicating that head conjunction is in tact possible. Therefore, he argues that they are instances of Right Node Raising (RNR) and their ungrammaticality is due to semantic factors (the clitic cannot identify a null operator in the second conjunct). Moreover, sporadic grammatical cases of clitic coordination, such as those given for Romance by Beninca. & Cinque (1993), are (more) acceptable because in these cases the two verbs are semantically close, allowing reinterpretation of the operator as an empty eli tic bound by the overt clitic. Although the semantic restriction is quite intuitive (and possibly on the right track- see below since it has become part ofthe higher copy after PF cliticization, but also because the lower copy contains uninterpretable features which it checks against the particle head. n. See Miller (1992) and Takano (2004) for evidence that such cases are instances of V+V (or VP+VP) conjunction, and Kayne (1994) for a Right Node Raising approach as deletion under identity. Even if we postulate that the actual structure is TP+TP, as long as the clitic adjoins toT (or even to some higher head), narrow scope remains a problem. On the other hand. if the clitic adjoins to v-V and we have TP+ TP conjunction, then narrow scope trivially follows, since in that case the ditic will be embedded too low in the structure.
122
Cities in Greek: A Minimalist Account ofProdisis and Endisis
for a similar restriction in the nominal domain), it nevertheless cannot explain why parallel examples in Greek are ungrammatical (cf. 20a-b). Perhaps it could be said that this is an instance of cross-linguistic parameterization, in that Greek eli tics simply cannot reinterpret a null operator in the second conjunct However, this wouldn't explain why (preverbal) clitic conjunction is not available either (as opposed to French), where no such restriction holds and the empty category precedes its licenser (cf. 20c) 12: (20) a. *To dhiavaza ke ksanadhiavaza oli tin ora. Greek It-dace read-imperflsg and reread-imperf-2sg all the time-ace 'I was reading and rereading it all the time? b. *To anigha i eklina kathe pente lepta. It-dace opened-imperf-lsg or closed-imperflsg every five minutes 'I was opening or closing it every five minutes. Greek c. .. Sas ke tus ekana ti chari. You-clgen and them-clgen did-lsg the pleasure-ace 'I did you and them a pleasure?
Greek
Itis not clear why (20c) should not be acceptable, were it an instance ofRNR (and the same applies if it is an instance of head adjunction).B Note that in the nominal domain we find a rather similar semantic restriction in that conjunction is only possible if the two N s are semantically close enough and/ or compatible enough to be taken as properties of one and the same person (see Heycock & Zamparelli 2003): ke antras irthe na me di. (21) *To pedhi The child-nom and man-nom came-3sg subj me-clgen see-3sg 'The child and man came to see me.
Greek
Although (21) could be accounted for via RNR, this wouldn't explain in a straightforward manner why the nominal domain differs from the verbal domain in this respect, but more importantly it would miss the fact that this difference co-occurs with a (possible) structural difference, which might be an indication that these two may in Elena Anagnostopoulou has pointed out to me that she finds 20 (a-b) ok and 20 (c) ok with stress. Although (20a) appears to sound better than the following sentence* To diavaza ke petoksa sta skupidhia. (It.d was.readlng.1sg and threw.lsg to-the garbage), where two distinct verbs are being used, it is still worse than the corresponding sentence where the clltic is repeated in both conjuncts. As for (20b), I find it ungrammatical, especially if the verbal aspect is changed to perfective: * To aniksa I eklisa pri11 pente lepta (It.d opened.1sg or dosed.1sg before five minutes). Finally, I agree with her that stress on the clitics in (20c) renders the sentence more acceptable, however acceptance in this case seems to be related to a post-syntactic operation. Without stress the sentence is fully ungrammatical, and that is what interests me here. 12.
13. And in this respect, see Kayne (1994: 144, ft. 6). Note that disjunction does not seem to make any difference in Greek (cf (20a-20b)).
Chapter 4. Derivation of proclisis 123
fact be correlated. This suggests, then, that modulo semantic restrictions, wide scope is possible if syntactic structure allows it If wide scope is not allowed in V+V cases, even when semantic compatibility is possible, we might conclude that other (morphosyntactic) factors are at play. This is further supported by auxiliary conjunction: Ta echo dhiavasi k.e dhiorthosi. Them-cl have-lsg read-part and corrected-part 'I have read and corrected them: b. Ta echo dhiavasi k.e *(ta) echo dhiorthosi.
(22) a.
Greek
Greek
Note that in (22b) the ditic does not have wide scope. This follows from the structure of the sentence which involves TP+TP conjunction ( (22a) involves VP+VP conjunction). What is more interesting though is that we can have wide scope in auxiliary constructions with verbs that are not semantically compatible in any obvious way (i.e. they cannot form a (complex) single event): (23) a.
nse
iche dhiri ke aghapisi. You-el had-3sg beaten-part and loved-part 'He had beaten and loved you:
Greek
Although the sentence is semantically strange, it is not ungrammatical in the way the sentences in (20) are. This is confirmed by the fact that the sentence does not improve with full D Ps, which do not seem to fall under the same restriction as the one in (20) (cf. 23c ): b. nESENA iche dhiri k.e aghapisi. YOU -strong had-3sg beaten-part and loved-part
Greek
c.
Greek
EMENA idhe ke aghapise. ME-strong saw-3sg and loved-3sg
Kayne (1994) argues that the data in (22a-b) follow from the fact that in these cases clitic deletion is parasitic on auxiliary deletion (gapping). However, this leaves unexplained the fact that 'parasitic deletion' is obligatory with clitics but not with a full DP: (24) a.
ERGASIES echo dhiavasi ke echo dhiorthosi. Greek THE PAPERS-ace have-lsg read-part and have-lsg corrected-part TIS
b. TIS ERGASIES echo dhiavasi ke dhiorthosi. THE PAPERS-ace have-lsg read-part and corrected-part 'It is the papers that I have read and corrected:
Greek
All in all, it seems that clitics (but not full DPs) cannot have wide scope, which indicates that they cannot be PF-adjoined to their host 14
14. Unless of course conjunction is a PF-deletion process and the ditic PF-attaches to its host before this process takes place. However, such a derivation would not explain the nominal domain facts. Moreover, in aT-model, it is unclear in what way the output of a PF-process can be
124
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
Additional evidence against the Kaynean analysis is given by data involving adverb incorporation (see Rivero 1992 and Alexiadou 1997). The generalization is that incorporated adverbs (independently of whether incorporation is syntactic or lexical) behave like clitics in many respects. 15 If adverb incorporation involves head adjunction (either in syntax or in the Lexicon), a PF-approach to cliticization would fail to capture their common properties straightforwardly. Incorporated adverbs behave like clitic pronouns in several respects: a. They do not allow interpolation (by an adverb, a pronoun/D R etc.) masisa. Greek (25) a. *Kala to kotopulo Well the chicken-ace chewed-lsg b. Kalomasisa to kotopulo. Greek c. *To kotopulo kala/a- to masisa. (without intonation on the adverb)Greek 'I chewed the chicken carefully/well~ b.
They may only have narrow scope over a conjunction (26) a.
Kalomasisa ke efagha to komati to kotopulo. Well-chewed-lsg and ate-lsg the piece-ace the chicken-ace 'I chewed the piece of chicken carefully and then ate it:
Greek
b. Masisa ke ghrighorokatapja to komati to kotopulo. Chewed-lsg and quickly-swallowed-lsg the piece-ace the chicken-ace Greek 'I chewed and then quickly swallowed the piece of chicken~ c.
d.
Wide scope is only possible if the adverb has moved onto the auxiliary in a complex tense conjunction psilofai ke pji. (27) a. Echo Have-lsg little-eaten-part and drunk-part 'I have eaten a bit and I have drunk~
Greek
b. HPsiloecho fai ke pji. 'I have eaten a bit and drunk a bif
Greek
The adverb may 'adverb-climb: which is similar to what we find in clitic-auxiliary constructions. The main difference is that it does not need to climb all the way up: instead, it may stop at the participle (which is impossible for a clitic): fai ksana vs . ..J echo ksana-fai (28) a. ..J Echo Have.lsg eaten again vs. have.lsg again-eaten
ruled out for semantic reasons (and in fact Kayne 1994 assumes that deletion is a syntactic operation, for obvious reasons). 15. Note that incmporated adverbs are morphologically deficient in that they are unstressed and may take the morpheme -o-, as inkalo-troo from troo kala ('eat well').
Chapter 4. Derivation of proclisis 125
vs . ...J ksana-echo fai vs. again-have1sg. eaten 'I have eaten again'. b. *Echo
fai
to
vs. *echo
have.lsg eaten itcl vs. 'I have eaten it'
Greek
to-fai
vs. ...J to-
have.1sg itcl-eaten vs.
fai Greek itcl have.lsg eaten echo
Note that the adverb cannot stay in situ unless it is stressed and does not contain the a-morpheme (see Alexiadou 1997, and Cardinaletti & Starke 1999). If moved however, it must attach to its host. Finally, the verb may not leave the adverb behind moving higher within the clause (see Alexiadou 1997 for discussion): (29) a. *efagha kala- vs . ...J kalo-efagha ate.1sg well vs. well-ate.1sg 'I atewelf.
Greek
Greek
b. *efagha to- vs . ...J to-efagha ate.1sg it vs. it-ate.sg 'I ate if c. *to komati kotopulo to efaghe o Nikos the piece chicken itcl ate.3sg the Nick.nom vs . ...J to komati kotopulo to kalo-efaghe o the piece chicken it.cl well-ate.3sg the 'Nick pleasantly ate the piece of chicken:
kalawell Nikos Nick.nom
What (29c) shows is that the adverb+verb complex moves as a single unit in syntax, therefore it cannot be tanned in PF. Since these structures behave very much like clitic structures, it is simpler to assume that the latter also form a unit (at some point) in narrow syntax. 16 Further evidence is offered by the relative position of the 'clitic+host' complex with regards to adverbs (cf. Ledgeway & Lombardi 2005 for Italian and the dialects, following the Cinque 1999 approach to adverb placement). In particular, finite (lexical) verbs in Greek may appear in a range of positions above the vP, but crucially may stay at least as low as below aspectual adverbs like idhi (=already) or schedhon (=almost) (although higher than e.g. telios/entelos (= completely)/kala (=well)): (30) a.
Daniela ...J idhi/ ...J schedhon/*kala to" iksere. The Daniela-nom already/ almost/well him-dace knew-3sg
16. This symmetry, however, may break down (see Alexiadou 1997). Also, inN-V incorpora-
tion N attaches only to the partklple; this is probably because N-V incorporation is theta-role related (possibly also involving case checking - see Baker 1988), as opposed to ditic incorporation, which involves phi-feature checking, located in the aux head
12.6
Cities in Greek: A Minimalist Account ofProdisis and Endisis
b.
I Daniela to" iksere -.J idhi!.V schedhon!.V kala.
c. *I Daniela ton idhi!schedhonlkala iksere. 'Daniela (already/almost) knew him (well):
Greek
These data indicate that Greek clitics attach to their host in narrow syntax and in fact quite low in the clause (i.e. lower than T, but probably outside vP), since the complex may only move higher up as a unit (cf. 30c). Moreover, examples (30a-c) show that the T position is not targeted separately by the clitic and by the verb (leading to clitic-verb adjacency, as e.g. in French- see Ledgeway & Lombardi 2005), since the clitic always follows the verb, no matter where the latter lies in the structure. It seems therefore that there is some evidence against the PF-adjunction theory of cliticization, at least for Greek. The other possibility available is the syntactic approach, which views cliticization as clitic incorporation via head adjunction (in the sense of Baker 1988). However, this possibility is not without problems, either. In particular, assuming that the clitic is (presumably) the last head to adjoin to Tmax (cf. Raposo & Uriagereka 2005), it is not obvious why the clitic cannot have wide scope over a conjunction (but see ft. 11). Moreover, Rizzi & Roberts's (1989) suggestion that cliticization (and incorporation more generally) involves substitution into a subcategorized slot wouldn't solve the c-command problem, since the resulting structure would be identical to an adjunction structure 17 (see Borer 1998):
xo
(31)
~
yo
xo
lfY = clitic and X= T(P) + T(P), then Y should be able to c-command into X. 18 The same problem would also apply to approaches which put inflection in the syntax (cf. for example Pollock 1989; Belletti 1990). 19 These assume that clitics are agreement heads/affixes (or specificity heads- see Sportiche 1992/1998) merged above the vP periphery and picked up by the v-V head. One way to solve the problem would be by means of a post-Spell-Out morphological component that would 'restructure' the c-command relation between the clitic and the complex T-head. And this is exactly what Distributed Morphology has to offer, namely a PF morphological component that operates on syntactic structures, altering
17· Ian Roberts (p.c.) has pointed out to me that this is not really true given that in bare phrase structure the resulting structure would be set merge and not pair merge. Howeve~ he agrees that substitution movement would not be available in bare phrase structure in the first place. 18. For further problems (as well as for a criticism of Roberts's (1991b) attempt to solve some of these via his X- 1 proposal) see Borer (1998) and Julien (2002). 19. For additional problems of this approach with regard to Greek ditkization (as for example with Baker's Mirror Princtple) see Mavrogiorgos (2004).
Chapter 4. Derivation of proclisis 127
their structural properties.20 That. in combination with a multiple Spell-Out approach, as e.g. Chomsky (2001a), would be enough to account for the lack of wide scope for clitics in conjunction. For example, we could assume that by the end of the syntactic derivation a complex T-head has been formed (by head movement), part of which is the clitic, and that at Spell Out the features of this complex head are bundled together in such a way so that they become syntactically opaque. This restructuring process/ bundling would be a morphological operation taking place at Spell-Out and it would have the effect that the clitic wouldn't c-command the complex T-head.21 In what follows I am going to discuss in more detail the nature of this morphological restructuring/bundling process.
4·3 Clitics as non-phases
In this section I present my analysis regarding cliticization and I use proclitic environments to show how it works. 4·3·1 Derivation of proclisis
I follow Chomsky's Derivation by Phase model (Chomsky 2001a,b ), and especially the assumptions he makes with regards to phase heads in Chomsky (2005, 2006). I also adopt his hypothesis that feature inheritance applies in the lower (v""P) phase, triggering short object movement to the periphery ofVP. This allows me to capture the following property of Greek cliticization, namely that Greek clitics attach to their host low in the clause (see discussion in 4.2.3).22
10. Although for DM the ditic would probably be a bound morpheme moved to its host from the object position 11. Alternatively, it might be the case that the ditic head has to be realized on each conjunct separately for morpho-phonological reasons (if conjunction is PP-deletion). Either way, c-command (and wide scope) in head adjunction structures would be possible in syntax but invisible/ immaterial to PP-deletion, which only sees the whole complex head. 11. Attachment to the host at VP, or more generally to some head lower than v or T, is more theoretically and less empirically motivated: given that cliticization in Greek has A-properties (see Alexiadou & Anagnostopoulou 1997), but not X-properties (see Mavrogiorgos 2004), and that in Chomsky's (2005, 2006) system external specs of phase heads are strictly X-positions, it follows that the clitic must move to [Spec VP], the A-position. 0 n the other hand. the empirical evidence only shows that the ditic attaches lower than T, not necessarily at VP. In fact, I will assume later on that the ditic attaches at the v* level, i.e. basically at v*. This assumption is based on the fact that clitic pronouns contain object features which also appear as uninterpretable features on v*, as well as on adverb placement facts which show that the ditic +verb complex may follow aspectual adverbs which mark the vP periphery.
rr
128
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
Recall that interpolation in Greek is not allowed at all, which may be due to low cliticization (the idea being that syntactic cliticization binds the clitic to its host early within the derivation, making thus interpolation impossible - see Ledgeway & Lombardi 2005). Although adverbs cannot give us reliable evidence for vP-internal cliticization (as they are vP-external- see Cinque 1999), I believe there is some empirical evidence coming from the position of floating quantifiers. In particular, stranded object floating quantifiers may appear (also) within the vP. If we assume along with Boskovic (2004) that such quantifiers cannot adjoin to the object theta position, it follows that when they are vp-internal they must be somewhere between V and v: 23 (32) a.
Ta pedhia ta echi idhi dhi o Nikos ola. The kids-ace them-cl has-3sg already seen-part the Nick-nom all-ace 'Nick has already seen all the children: Greek
b.
Tis elise o Nikos ales me apistefti taxitita. Them-cl solved-3sg the Nick-nom all-ace with incredible velocity Greek 'Nick solved them all incredibly fast:
Interestingly enough, the sentences below, in which the quantifier is actually higher, at the so-called Object-Shift position (periphery ofvP - see Chomsky 2001a), are even less marked: (33) a. Ta pedhia ta exhi dhi ala o Nikos. b. Ta pedhia ta exhi ala dhi o Nikos. Tis elise ales o Nikos me apistefti taxitita.
Greek Greek Greek
This shows that there must be additional positions for the clitic/created by the clitic, namely at least one below the participle (presumably the OS-position; see Friedeman & Siloni 1997) and another one above the participle. The issue here is how to reconcile the sentences in (32) and (33). One possibility would be to assume that the quantifier may adjoin acyclically to each one of these positions (but crucially not to the thetaposition), with any differences in markedness linked to stylistic reasons. This follows from the nature of stranded quantifiers (they adjoin to phrasal constituents), while it also offers evidence for the fact that eli tics are found in further specifiers besides the lower specifier ofVP. But let us see how exactly the derivation proceeds. Following Kayne's movement hypothesis, the clitic is merged as a DPID in the complement position of the root. its phi-features being visible to appropriate probes higher up in the clause. V (or possibly some higher verbal v head), being a phase head by inheritance of phi-features from v*, attracts the clitic to its (external) specifier (A-chain). 24 At the same time, an optional 23. Post-verbal subjects are in [Spec, vP], complex manner adverbs are complements to VP (c£ Alexiadou 1997). Stranded quantifiers in sentence final position receive main stress.
24. Adam Ledgewayasks why we need feature inheritance. For theoretical reasons see Chomsky (2005), Richards (2007). For empirical reasons see main text above.
Chapter 4. Derivation of proclisis 129 EPP feature at v*, which is linked to a familiarity/- focus/old information D feature probes and forms a parallel chain with the clitic (A'-chain). The clitic moves and lands as an XP/X after which it incorporates into the edge ofv*, which is a minimal morphosyntactic phase transparent at the edge. 25 The incorporated clitic will move along with the v*-V complex to T. The derivation is illustrated below: (34)
TP
~
T' ~ DP/Dcl-V-v*-T
v*P
~ (DP/Dcl)
v*'
~~ (V-v*)
VP
~ (V)
(DP/Dcl)
What happens with full pronouns and other DPs? Why don't they also move to the periphery ofv*? One way out would be to speculate that EPP addition is blind to the D PI Cl distinction, and that the grammar filters out ungrammatical derivations. In that case, ungrammaticality would follow from independent interface properties as well as from economy considerations.26 Yet. another way would be to take the eli tic to be the realization of the optionaVtopic EPP in all environments. Recall that the clitic is in essence a topicalizer. To put it in other words, the EP PI clitic would be freely added to the numeration, independently of the type of DP complement available. Certain combinations would give rise to cliticization, others to clitic doubling, while others would be infelicitous. In those cases where no EPP is added we would get the non-doubling
25. In chapter 5 I discuss the possibility proposed by Gallego (2006) that in null subject languages v* reprojects at the T level, so that in essence we have a complex v*/T head. If we want to keep to this assumption, then we would have to assume that the ditic moves to the edge of the complexv*/T head. In any case the diticwill be found on T, as v*- V, into which the ditic incorporates, will move to Tin Greek (see Alexiadou & Anagnostopoulou 1998). 26. Elena Anagnostopoulou has pointed out to me that object DPs optionally move in VOS orders. Note, however, that it is not dear whether this is syntactic or PF movement In addition, if it is syntactic movement, it is not dear whether it is A- or A-bar movement. Moreover, there is also the possibility that VOS is derived via remnant movement of VO across S, and not by 0 movement across S. UntU these issues have been settled, one may not use VOS as a counterargument against the generalization that only ditics move to the periphery of v*P.
130
Cities in Greek: A Minimalist Account ofProdisis and Endisis
instances, which would only involve A-checking by V. So, we would get the following distribution: (35) (i) Optional EPP/clitic added //suitable- DP present: clitic doubling (ii) Optional EPP/clitic added// non-suitable DP present: non felicitous clitic doubling (iii) Optional EPP/clitic addedl/pro DP present: cliticization only (iv) No EPP/clitic addedi/DP present: no cliticization This would work for Greek and other languages which have clitic doubling. In the doubling constructions the clitic would move out of the double DP to the v*P periphery, while the double DP (i.e. the full DP or pronoun) would stay in situ. I will explain how this is done exactly in Section 4.6, where 1will be discussing doubling structures. In case (iv) the DP or pronoun would not move higher than the periphery ofVP, which is the typical object position, as there is no (optional) EPP feature and as a result no clitic. Going back to incorporation, I assume that it involves morpho-syntactic restructuring, which breaks down into two steps: The first step involves 'intrusion' of the clitic into the edge ofits host. This takes the form of morpho-syntactic adjunction, which 'glues' the clitic onto its host (i.e. the clitic becomes morpho-syntactically dependent onto it), ensuring that it will be carried along with the host toT. It is some form of super agreement between the clitic and the host that keeps the clitic within the specifier of its host. Another way of looking at intrusion would be as a super feature that keeps on attracting the clitic to the higher specs along the derivation. 27 However, intrusion does not restructure the clitic that is within the spec into a head, as yet. As a result, and due to the fact that the edge remains open until the next higher phase (here I assume the Phase-Impenetrability Condition (PIC) proposed in Chomsky 200la: 14, according to which "the domain of His not accessible to operations at ZP; only H and its edge are accessible to such operations."), the clitic is still visible, as an XP element, to syntax [presumably an EPP feature) (i.e. it is syntactically independent/active), which explains why it retains its XP-properties (projecting an A-spec) when it moves toT, even though morpho-syntactically it is part of a complex head. Morpho-syntactic restructuring is completed with the second step. which involves relocation/merger of the feature matrices of the clitic and (the phi-features of) v* (and see (36) for details). Technically, this is achieved via syntactic agreement and deletion of the phi-feature bundle of v*, and leads to closure of the edge, rendering the clitic (or any other sub-part of the complex head) inaccessible to further (syntactic)
27. Earlier on in this chapter I introduced Gallego's (2006) phase sliding approach, according to which~ re- projects when moving to T. Accordingly, we can reinterpret 'intrusion' as successive cyclic movement of the ditic XP to the specifier of v* IT through a number oflower specs.
Chapter 4. Derivation of proclisis 131
operations. 28 As a result, the clitic loses its syntactic independency, becoming a proper sub-part of the host and hence acquiring true head/X properties. This predicts that from this stage on the clitic counts as (part of) a head for syntax, a fact borne out by empirical data which show that the clitic becomes sensitive to the head movement constraint at the T level (see, for example, the blocking effect of negation on clitic climbing, as reported in Kayne 1989b).29 This analysis predicts that syntactic clitics will always target a phase head (C, T, v* or V, depending on parametric variation with regard to feature inheritance and the position of object agreement), which naturally captures the essence of special clitics, namely that their cliticization is syntactically motivated. Moreover, it helps us get rid of the extra head movement, which does not bode well with the XP/X status of clitics, and of the extra greedy/look-ahead features which drive this movement: the properties of clitics, then, do not follow from any special features they need to check but from the incorporation operation itself, which saturates them morphologically (see 4.3.3). Finally, this analysis accounts for the morphological properties of clitics (cf. narrow scope, etc.).30. 31 4.3.2 The Syntax-Morphology mapping: Words as m-phases and clitics
In this section I discuss in detail the notion of phase as I use it here. The general idea is taken from Marantz (2001), who proposes that words are (morpho-syntactic) phases or spell-out domains, closed-off by functional phase/categorizing heads, which render them opaque to further operations. Marantz wants to equate domains significant for morpho-phonology with domains significant for morpho-syntax (Elena Anagnostopoulou p.c.). Here I propose that certain categories of words are minimal phases (i.e. PJ:-Pilin), namely those that are a lexical phase (i.e. v, n, a, etc.) or those that 18. Note that syntactic agreement has already taken place between v* and the clitic when the latter was ln its base position. Howeve~ deletion of object features on v* and relocation of feature matrices may only take place at TRANSFER of the v* -cl complex, at the CP leveL
19. Apparently, this is a new way oflooklng at the XP/X nature of clitics, and moreover it gives
new content to the old intuition that the ditic moves to T for PP-reasons (the 'gluing' effect). Two further points are in order here: (a) if an EPP feature attracts the clitic before the second step, it is predicted that the ditic should be able to 'excorporate' out of the complex head. This is what we find ln (some) interpolation languages; (b) although the same should be possible for the host when a head feature probes the complex head, it actually does not strand the clitic. This means that for some reason the syntax can only see inside the edge of the host (Le. the label, which contains the adjoined clitic) but not any further down, excluding therefore the possibility of head excorporatioiL If empirical data prove that the latter exists after all, the system would predict that it should be possible until step 2, i.e. at the C-T level. However, we would need a principled way of predicting what moves (Le. the whole complex or only the host) and when/why. 30. Any morpho-phonological differences between ditics and other (lexical) agreement morphemes follow from the post-lexical nature of ditic incorporatioiL 31· AB far as endisis is concerned. see chapters 5 & 6.
132
Cities in Greek: A Minimalist Account ofProdisis and Endisis
contain a lexical phase.32 For instance, a root which has been selected by a phase/ categorizing head vis a minimal v-phase, and the same applies to other roots with regard to other phase heads. More functional heads may be added, but they will need to have their features valued in order to ensure valid phase-hood. What characterizes a (minimal) phase is that it has a (left) edge (as any other phase). The edge of the phase though (which is defined as the label) is minimally transparent, in the sense that it allows a non-phase to incorporate into it under certain conditions (i.e. sisterhood and proper subset-hood in terms of features) [as opposed to a non-phase, which is fully transparent, and therefore may fully incorporate into an appropriate host]. In this respect, incorporation is viewed as feature restructuring, meaning that the features of the 'incorporee: which are a proper subset of the features of the incorporator, are relocated/ 'sucked' into the label of the host. 33 Accordingly, the edge becomes the locus of the syntax-morphology interface: it contains lexical information that is relevant to syntax (e.g. phi-features, Tense features, theta-roles and categorial features, which determine the (internal and external) merge position of the word, etc., and hence must be saturated within the syntactic derivation), while on the other hand it is the place where morpho-syntactic restructuring is possible within the syntactic derivation. It follows that the edge will become opaque/non-accessible (both for morphological as well as for syntactic operations) as soon as all the features contained in it are checked off. This follows from the fact that spell out domains must be fully interpretable. The idea that I would like to propose here with regards to phases and clitics is that clitics are not phases and therefore they are able to incorporate into their verbal host, which is a minimal phase, hence minimally transparent at the edge. ln this I follow Cardinaletti & Starke (1999) who suggested thatclitics are deficient elements, accounting in this way for a number of their properties. The question of course is why elitics are not phases. Or to put it differently, how do we define non-phases? An answer could be that clitics are non-phases because they only contain a D-head, with no lexical root or any other functional projections to close-off. The problem with that is that a D- head counts as a phase-head and therefore itself as a phase. An alternative analysis would be to assume that although a clitic is a phase head, it has too few features, being a single head. Perhaps, Don its own is incomplete, without the lexical n/N phase (Ian Roberts, p.c.). 34 As a result, when it moves to the edge of a constituent it can be sucked into it if it is a proper subset of the features of that label. Then, the clitic will become a subpart
32. Ian Roberts (p.c.). 33· The idea that featural proper subsethood is a necessary condition for incorporation is taken from Roberts (2009) who argues that: "incorporation can take place only where the label of the incorporee is non-distinct from that of the incorporation host" (Roberts 2009: 75). 34· Or, perhaps, it is a defective D-head, similar to defective (passive/unaccusative) v (Adam
Ledgeway, p.c.).
Chapter 4. Derivation of prodisis 133
of the word/ phase it incorporates into, and will cease to have access to syntax as soon as the edge is closed off. Incorporation takes place in the way (36) shows: (36)
v*-V
~ {
cl
}
ld(lnJla.b.cJ
{
v*-V tense phi-features
} ld(aJed(lnJ{a,
b, c}} etc.
In (36) the clitic incorporates into the v*-V complex head. It carries phi-features as well as aD-feature that denotes definiteness (what I call here an internal D index). Unvalued tokens of the same features are found on the complex v*-V head with which the clitic agrees. Valuing of the phi-features on v*-V and incorporation allow the id(ex) index (i.e. the external D index, which denotes familiarity, old information, topichood) on the phi-feature bundle of v*-V to be copied on the corresponding bundle on the cl. In the second step of morpho-syntactic restructuring the v*-V-phi-bundle is deleted, and the cl-phi-bundle takes up its place (feature matrix substitution/relocation). This is possible only at TRANSFER in the CP phase, since itis only then that the v*-V-bundle (including the v*-V-phi-bundle) gets deleted.35 This feature relocation/substitution, which incorporation amounts to, was possible in this case due to the fact that the features of the clitic were a proper subset of the features of the host, namely the v*-V head. In other words, the features of the minimal phase v*-V were exhausted by the features of the clitic. It is in this respect that I want to call the elitic a non-phase in relation to a phase. It appears then that a non-phase may be a relational notion. As a result, a clitic is a non-phase to a verb but not for example to another clitic (provided these two particular clitics share the same features).36 A full DP, on the other hand, may only merge (including pair-merge), being a phase. This is so, because it has more features, and hence it cannot incorporate (i.e. it contains distinct features from another phase/incorporation host). Finally, lexical roots, i.e. bare L-heads, may incorporate into non-phases/ L-roots as well as into little x's (i.e. to phase heads selecting for L-roots), as the latter are the only heads which select for lexical heads/bundles and hence have the appropriate features that trigger incorporation. In other words, on the one hand lexical roots may incorporate with other 35· If for some reason the V-bundle is not deleted, the clitic should be further extractable, given the right conditions. This is an empirical issue for further research. A plausible example might be restructuring constructions, where the dl:tic is further extracted to the matrix clause. 36. A relevant example would involve constructions with two dative dl:tics, such as those that have been reported in Spanish (IDeam 1999) and Catalan (Bonet 1991). As for dl:tic dusters (e.g. direct- indirect object clitics) I assume that the two clitics do not incorporate into each other (given that they contain distinct features); rather, they may only syntactically adjoin with each other, with any further manipulations in their order and internal structure taking place in the post-syntactic morphological component (see Bonet 1991).
134 Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
lexical roots, given that both of them contain only lexicaVroot features. On the other hand, categorizing heads may act as hosts to lexical roots, since they contain lexicaV root features as complement features, i.e. the lexical root exhausts the features on the (selecting) categorizing head.
4·3·3 The effect on the outcome: LF and PF saturation of eli tics Since the clitic host is reinterpreted as an (external) morpho-syntactic phase head, and since phase-heads check off the features of their domains, it follows that the host should check the PF and LF properties/features of the clitic. My claim is that this is the effect of incorporation, which is an additional operation and therefore must have an effect on the outcome. 4·3·J.l PF valuation Clitics cannot be mapped into an independent Prosodic Word (being deficient functional words-cf. Selkirk 1995), therefore they need to be parsed into an adjacent PWd (cf. Principle of Full Interpretation (see Richards 2004)). Incorporation trivially satisfies this requirement. by providing a phonological host to the clitic. In this respect. incorporation has a similar effect to that of shifting in object shift constructions in Germanic languages: in those as well, object shifting provides the weak pronoun with a phonological host (although the movement itself takes place for purely syntactic reasons). ln addition. incorporation captures the difference between PF-clitics and syntactic clitics: the latter may only lean onto a specific host (which is a phase head), because they are syntactically attracted to it.J7 4·3·3-2
LF valuation
The clitic pronoun in Greek is typically understood as a topicalizer, i.e. a topic marker that marks its double as old or fiuniliar information. I propose that this interpretation is assigned to the doubled DP (overt or pro) via the optional EPP feature the clitic checks on the verbal host when it incorporates. In particular, it gets assigned an external D index, i.e. a familiarity D-head (the outer D head), that gives it the appropriate interpretation. In this sense, saturation by a (higher) D-category takes place in the syntax giving rise to a special interpretation.
4·4· Have - constructions In this section I will discuss the derivation of the so-called echo/icha constructions with clitics. Echo/icha constructions involve the periphrastic forms of the present and
37· In those cases where incorporation does not take place (e.g. in interpolation structures) the phonological host is provided via PF-cliticization.
Chapter 4. Derivation of proclisis 135
past perfect respectively.33 They are formed by the combination of the present or imperfect of the verb echo ('to have') and the active or passive form of the participle: (37) a. Echis fai vradhino? Have.2sg eaten dinner.acc 'Have you eaten dinner?' b. Icha dhi aftin tin tenia. Had.lsg seen this the movie.acc 'I had seen this movie:
Greek
Greek
What is very characteristic about these clauses is that their kernel appears to be the participle. For example, the theta roles and case( s) of the subject and the object participants are restricted by the participle, and not the auxiliary. as the following examples clearly show: dhosi tu Jiani ta portokalia. (38) a. Dhen echo Not have.lsg given the John.gen the oranges.acc 'I have not given John the oranges:
Greek
(*tu Jiani) ta portokalia. b. Dhen echo Not have.lsg (*the John.gen) the portokalia. 'I don't have the oranges:
Greek
c. Dhen edhosa tu Jiani ta portokalia. Not gave.lsg the Jiani.gen the portokalia.acc 'I didn't give John the oranges:
Greek
d. 'O agheras dhen iche chtipisi ti Meri epitidhes. The wind not had.3sg hit the Meri.acc deliberately 'The wind had not hit Mary deliberately:
Greek
Moreover, the participle contains aspectual and voice information: (39) a.
Dhidho echi maghirepsi simera. The Dhidho.nom has.3sg cooked-perf. today 'Dido has cooked today:
b. To grama echi stall edho k.e meres. The letter has.3sg been-sent here and days 'The letter has already been sent:
Greek
Greek
Therefore, it seems reasonable to assume that the participle clause is constructed of a transitive or passive voice head (depending on the diathesis of the verb) with an 38. The echo form is also used periphrastically with 8a to form the future perfect: ea to echo dhosi mechri tote. Fut itd have.sg given until then 'I will have given it by then.' I will confine myself to the present and past perfect for the purposes of the discussion. (i)
136 Cities in Greek: A Minimalist Account ofProdisis and Endisis
aspectual head on top of it. The voice head will take a VP as its complement, while it will be a camp lement to the auxiliary V. The iteration of vPs is illustrated below: (40)
VauxP ~ Vaux VaspP ~ Vasp VPrtP ~
DPsubj
VPrt' ~
VPrt
VP ~
V
DPobj
The VPrt is a proper v* which checks accusative case on the object DP and which assigns the external theta-role to the subject DP. The auxiliary verb is a light, raising verb that will move from Vaux to Tense. This explains why its presence is obligatory for the well formedness of the construction but also why on the other hand it does not contribute much to the general syntax or the semantics of the whole structure. What interests me here is what happens when the complement is a clitic. The question that arises is how the clitic ends up on the auxiliary. Normally, the direct object clitic will initially merge in the direct object position. It will be probed by the Prt v* as well as by the optional EPP feature, so it will move to the edge of Prt as an XP, where it will incorporate being a non-phase (a subset of the features on the participle). Next, the participle head will move higher to the aspectual head to pick up the aspectual morphology. Then, the empty auxiliary, which I assume has an unvalued v* containing phi features, will probe the complex participle head, and will copy the values of v and of the clitic that is contained in v on the complex participial head. In other words, there will be a chain formed between the empty auxiliary and the participial head, where the two heads will be sharing the same content. This must be related to the temporal chain that is formed between the empty auxiliary and the participle, which makes them two parts of one camp lex tense. In any case, the idea behind this feature sharing is that the complex form is a single form as far as clitic checking/climbing is concerned. 39 In particular, the clitic is checking its features only once against the 39· An alternative way of seeing this would be to assume that the AgrO head is on the Vaux, like Friedemann & S!loni (1997) have proposed. This would inlply that the participle is dependent
on the Vaux to case and theta mark its object(s) and subject. In that case the ditic would be directly probed byVaux.
Chapter 4. Derivation of proclisis 137
phi-features ofthe participle. Then, it is Vaux that is copying the v/phi features of the participial head, in which the clitic head is also contained due to incorporation. 40 4·4·1 Other clitic climbing constructions There are a few other clitic climbing constructions in Greek besides the present and past perfect ones, such as e.g. the following ones: (41)
To
echo ghrameno to ghrama. It.acc.cl have.lsg written.neut.sing.acc the letter.acc 'I have it written, the letter:
ine adhiafori. (42) tnu Me.gen.cl is.3sg uninterested.neutsg.nom 'I am not interested in her:
Greek
Greek
(41) has been analyzed by Iatridou (1996) as a passive small clause that is an object to
echo. The accusative is assigned by echo to the object/derived subject of the small clause (an ECM construction). We could assume that echo probes the clitic from the object position of the small clause: (43)
[v echo [sci ghrameno to]]~ [v to echo [sci (to) ghrameno (to)]]
Afl for (42), itinvolves the verb 'to be' along with an adjective and its object clitic mu ('me'). I assume that mu is introduced by an applicative head right above the adjective and thatitincorporates into ine since it only has a subset of the features of this higher probe. The subject agrees with both the adjective and the verb.
4·5· Indirect object constructions As far as indirect object constructions are concerned, I basically assume the analysis put forward by Anagnostopoulou (2003a). Since the main focus of this chapter is on direct object elitics, here I will only give a brief summary of her analysis; the reader is referred to her book for more details. According to Anagnostopoulou (2003a) Greek has a variety of indirect object constructions in which the indirect object is realized as a genitive DP, as an accusative DP or as a PP. These three options are illustrated below: (44) a.
Estila tu ]ani ena ghrama. letter.acc Sent.lsg the John.gen a 'I sent John a letter:
Greek
40. Note that it is not the whole participle that is copied/incorporates into Vaux, but only its v*-features, and as a result the object ditk that has incorporated into Vpart from the object position. This is so, since Vecho is an empty verbal shell containing v*- object features.
138 Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
b. Estila ena ghrama stan ]ani. Sent.lsg a letter.acc to-the John.acc 'I sent a letter to John:
Greek
c.
Greek
Dhidhaksa ton Jani mathimatika. Taughtlsg the John.acc mathematics.acc 'I taught John mathematics:
The alternation between the genitive D P and the PP shares a number of characteristics with the English dative shift construction: tor example, non-animate goals are largely not tolerated, while there are verb classes which allow the constructions and others which do not (see Anagnostopoulou 2003a for more details). The genitive DP may be optionally doubled. The genitive clitic is introduced in the specifier of an applicative head that is sandwiched between the higher v*- transitive head (which introduces the subject DP) and the lower VP/root head where the direct object clitic is merged. The v* -transitive head probes and finds the indirect object clitic first, since the latter is closer to v*-transitive. It checks the person feature of v*- transitive and moves to the external specifier position. The accusative is probed next, moving also to a (lower) external specifier (tucking in), where it checks the number teature of v*- transitive. Since the higher clitic (namely the indirect object one) always checks a person feature, only 3rd person direct object clitics may check the number features of v* -transitive in the second step, which is how the so-called Person-Case constraint is derived. I further assume that incorporation into v*- transitive takes place when the clitic moves to its specifier, since it only has a subset of the features contained in the v*- transitive head. The direct object clitic also incorporates into the verbal host after tucking in.
4.6 Clitic doubling In the final section of this chapter I will discuss clitic doubling. By the term elitic doubling I refer to the case where a clitic and a full D P are not in complementary distribution, but co-occur in the same syntactic domain forming a discontinuous constituent (see Anagnostopoulou 2006: 520). Here I give an example from Greek and (Peninsular) Spanish: (45) a.
Ton idha ton ]ani. Him.cl saw.lsg the Jani.acc 'I saw Janis:
b. Lo
vi a el Him saw.lsg a him.strong 'I saw him:
Greek
Peninsular Spanish
Chapter 4. Derivation of proclisis 139
In Greek both direct and indirect objects may be doubled. Note that no preposition is required, in fact doubling of a preposition is ungrammatical contrary to what happens in Romance languages (Kayne's generalization) 41 : (46) a.
To efagha to psari. Ilcl ate.1sg the fish.acc 'I ate the fish:
Greek
b. Tu dhanisa tu Taki xrimata. Him.cl lent.3sg the Taki.gen money.acc 'I gave Takis some money.'
Greek
c. *Ton idhe stan Taki. Him.cl saw.3sg to-the Taki.acc '(S)he saw Takis:
Greek
However, there seem to be restrictions on what can be doubled. For example, Anagnostopoulou (1999) argues that only definite direct object DPs may be doubled in Greek. However, this may be too strong. For instance, doubling of generic indefinites sounds to me as acceptable: (47) Mia kali mitera pantote to cherete ena skili te~u tipu. A good mother always it.cl enjoys.3sg a dog.acc this.gen type.gen 'A good mother is always fond of a dog of this kind: Greek Kallulli (2001) gives further examples of doubled indefinites which are not right dislocated: te~o na simvi. Greek (48) 0 Janis to perimeni kati The John it.cl expects.3sg something such subj happen.3sg 'John expects something like this to happen' (Kallulli 2001: 131 (her 7b))
A further example of a double indefinite is given below: tin klotsise mia aspri ghata (49) Mbamba, o Nikos Dad, the Nikos.nom it.cl kicked.3sg a white cat.acc pu pernuse ekso apo to spiti. that passed.3sg outside from thehouse.acc 'Dad, Nick kicked a white cat that was passing by the house?
Greek
41. Note that this restriction is not so straightforward for indirect objects where we find exam-
ples like the following one:
to ipe se mena. Me.d itd said.3sg to me 'She said it to me: It would be interesting to see why indirect objects differ in this respect from direct objects. (i)
?? Mu
140
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
(49)is acceptable provided a strong/specific/generic and/or partitive reading is available in the context. The same applies to the following example: (SO) Tus idha dhio fitites su na pernane apo edo. Them.cl saw.lsg two students.acc yours.cl subj. pass.3pl from here 'I saw two of your students passing by:
Greek
Finally, in the next example, a bare quantified expression may be doubled forcing a strong reading and hence destroying the ambiguity in (Sla): (51) a.
Machi dhen idhe polus astinomikus. The Machi.nom not saw.3sg many policemen.acc
Greek
neg> many//many>neg b.
Machi dhen tus idhe polus astinomikus. The Machi.nom not them.cl saw.3sg many policemen.acc
Greek
many>negl/*neg>many 'It is not the case that Machi saw many policemen: [neg> many] 'There are many of the policemen whom Machi did not see: [many > neg] Even the example that Anagnostopoulou (1999) takes to be ungrammatical (the example is adapted from Sufter 1988) does not sound that bad in the right context: (52) ?!Kathe mera, i mana Every day the mother tin akughe mia her was-listening-to.3sg stin piso avli. in-the backyard
mu, enoso katharize, my while was-cleaning.3sg ghineka pu traghudhuse rempetika a lady.acc that was-singing.3sg rempetika.acc Greek
Of course, non-referential quantifiers like kanis 'noone' or tipota 'nothing' cannot be doubled: idha kanena. (53) *Dhen ton Not him.cl saw.lsg no one.acc 'I didn't see anyone:
Greek
The same applies to bare nouns. These also cannot be doubled42 : (54) *To aghorasa meli. It.cl bought.lsg honey.acc. 'I bought some honey.'
Greek
42· Note that the examples with kanenas or with bare nouns are much worse than those with indefinite quantifiers. If we follow Longobardi (1994) and assume that these nouns have a D head, being arguments, it follows that not all DPs are allowed to be doubled. In other words, doubling ls not a questton of DPs vs. non- DPs, but rather a question of overt vs. covert Ds.
Chapter 4. Derivation of proclisis 141
A further characteristic of Greek doubling is that it is not restricted by any semantic features such as [±human] or[± animate], as opposed to Romance languages which are restricted by such features. 43 As a result. both animates and in animates may be freely doubled in Greek. (55)
Ton idha (ton pinak.a/ ton andhra). Him.cl saw.1sg (the painting.acc/the man.acc) 'I saw him (the painting/ the man):
Greek
Finally, note that not all definite DPs may be doubled. Those DPs which are doubled need to be old information, i.e. they need to be defocalized or [- focus], as Kallulli (2001) calls them. So, for example, doubling is not telicitous in the tollowing dialogue: (56) a.
b.
Ti aghorase o Nikos? Whatacc bought.3sg the Nikos.nom 'What did Nikos buy?' Nikos to aghorase to podilato. The Nikos.nom itcl bought3sg the bicycle.acc 'Nikos bought the bicycle:
1
0
Greek
Gree
Now, as far as indirect object doubling is concerned, it is much less restricted than direct object doubling. For example, doubling of a non-referential quantifier is allowed: (57) Dhen tu edhosa kanenos ti valitsa mu. Not him.cl gave.sg no one.gen the buggage.acc my.gen 'I didn't give anyone my luggage:
Greek
On the other hand bare nouns cannot be doubled even when they are indirect objects (and as opposed to other languages, e.g. Spanish): prosteran ftochon ikojenion tiri ke krasi. (58) *Tils Them.gen.cl. offered.3pl poor families.gen cheese.acc and wine.acc 'They offered cheese and wine to poor families: Greek (adapted from Sufier 1988: 395 (her (5b)) As opposed to direct objects, indirect [+ focused] items may be (marginally in some cases) doubled: (59) ??Pjanu tr4 edhoses faghito? Who.gen him.gen gave.2sg food.acc 'Who did you give food to?'
Greek
43· Note that doubling of[- animate] referents is commonplace in many (colloquial) Romance varieties.
142
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
(60) ??Tu NIKU tu edhosa faghito. THE NIKOs.gen him.gen gave.lsg food.acc 'It is to Nik.os that I gave food:
Greek
As a result of this difference between direct and indirect object clitic doubling, some researchers have proposed that indirect object clitic doubling is more akin to agreement as opposed to direct object clitic doubling which involves a proper direct object clitic. Although it is true that indirect object clitics look more like agreement in this respect, it does not necessarily mean that they are agreement heads (see also Chapter 3 ). First of all if indirect clitics were agreement heads we would expect them to show some of the properties of agreement For instance, they should define the properties of a particular syntactic construction, as e.g. the double object construction, however they don't because in most cases they are optional (see Anagnostopoulou 2006 for this argument). Moreover, in environments where doubling is obligatory for whatever reason, failure to double should lead to default agreement and not to ungrammaticality, although ungrammaticality is what we get: (61) *(Tus) teliose ton pedhion i zachari. *(them.gen.cl) ended.3sg the children.gen the sugar.nom 'The children were short of sugar:
Greek
In (61) absence of doubling does not lead to default agreement between the theme and the verb; rather, ungrammaticality arises due to the genitive ton pedhion that blocks agreement between the theme and the verb. All in all, if the genitive does not act like proper agreement, then there must be some other reason why it is more flexible in terms of doubling when compared with direct objects. How are we to analyze clitic doubling, then? I propose that elitic doubling involves movement of the higher D-head of the doubled DP to v*-transitive. That is, I assume that the doubled DP in Greek has two D-heads, a higher one (Dl) and a lower one (D2). The higher one encodes tamiliarity/prominence/topicality, while the lower one encodes definiteness (see also Alexiadou, Haegeman & Stavrou 2007 for a similar proposal). In doubling structures the two Ds appear together in the numeration, while in non-doubling structures only the lower D2 head appears. Dl, by being in the edge of the minimal phase, is extractable, i.e. it may be probed by an external probe, and this
Chapter 4. Derivation of proclisis 143
is exactly what happens with v*-transitive. D 1 moves to v*-transitive and incorporates, giving rise to a clitic doubling structure. This is illustrated below:
(62)
v*P
~ v*
VP ~~ V DPl
~ ._____ _ _ _ _ Dl
~
~ DP2
~ D2
~ Since part of the DP has moved to a higher position, it is as if the whole DP has moved to that position, even though the rest of the DP has stayed in situ. According to this proposal the clitic is analyzed as the highest D-head that may merge within the DP, which marks the DP as familiar and which connects it with the outside world both semantically/pragmatically and syntactically. As for the semantic restrictions on doubling, these must arise from the interaction of the two D-heads, however I will not go into this discussion here, as it would take me too far afield.44
4·7 Conclusions In this chapter I discussed the derivation of proclisis in Greek. I argued that clitics move as XPs/Xs to the specifier of v*/T into which they incorporate as the result of their being non-phases. (Non)-Phasehood is defined on the basis of the features in
44 To give an example, we could asswne that in direct object constructions the external determiner contains a set of uninterpretable unvalued definiteness features which it checks against
the interpretable valued definiteness features on the internal determiner. This agreement relationship would give us the definiteness restriction on direct object doubling. On the other hand, in indirect object constructions we could asswne that the external determiner does not contain any definiteness features, and hence is underspecified in this respect This would account for the lack of definiteness effects in indirect object constructions.
144 Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
the clitic in relation to the features contained in the minimal phase v*/T: the clitic is defined as a non-phase with regards to v"'IT as it only contains a subset of the features contained in v*/T. I also argued that incorporation of clitics has an effect on the outcome by satisfying their LF and PF properties. Finally, I discussed several issues regarding echo-constructions, indirect object constructions as well as clitic doubling constructions.
CHAPTER
5
Patterns of proclisis/enclisis and the role ofV-movement
5.1 Introduction
In this chapter I am going to discuss the proclisis-enclisis alternation as it appears in Greek and other languages that follow the so-called '(non- )finiteness' pattern. I will try to define which level of finiteness, if any, is relevant to the alternation. Also, I will discuss the V-movement theory (see Kayne 1991), and I will show that V-movement across the cliticization site is neither a necessary nor a sufficient condition for enclisis. I will claim that both movement to the cliticization site and resolution of proclisisenclisis on the basis of finiteness at the elitic incorporation site are necessary conditions for the derivation of proclisis or enclisis.
5.2 De-limiting the phenomenon A rather typical characteristic of (different types of) clitic pronouns is that they may be mobile, i.e. they may linearly precede or follow their (syntactic and/or phonological) host depending on properties of the structure containing them. To illustrate, in Romance languages clitics normally follow the positive imperative form of the verb, even in those languages where proclisis appears to be the default case, as in Standard French (see Rooryck 1992 for discussion and for some exceptions). On the other hand, in the same languages positive indicative clauses normally take proclitics: (1) a.
Faisle! Do-2sg.imp. it-cl '(You) do it!'
b. Je le fais. I it-cl do-1sg 'I do
French (France)
French (France)
it:
Clitic mobility, or alternatively the proclisis-enclisis alternation, has been analyzed in many different ways, usually depending on one's assumptions regarding the morphosyntactic status of clitics, as well as the more general morphological theory one is following. To illustrate, for those researchers who assume that clitics are lexical affixes,
•
146 Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
merged on their host in a pre-syntactic Lexical component, the alternation becomes a lexical matter, typically mediated via a diacritic feature (like finiteness, tense, agreement or mood) on the host (so that if this particular feature is on the host, the clitic is merged as a suffix), which marks the dependency of the alternation on properties of the host This is for example what Halpern (1995) suggests for enclitic constructions in a number of languages (and see also Monachesi 1999 for Romance languages). On the other hand, researchers following a syntactic approach to morphology typically assume that clitics are as independent as any other morpheme in the language, and end up on their host by means of syntactic movement Clitic mobility, then, derives from the interaction of the movement of the host and the movement of the clitic (see e.g. Kayne 1991, Rizzi 1993). In a way that parallels lexicalist approaches, the scope and target of host and/or clitic movement are related to independent properties of the clause (e.g. properties on T/INFL). As I have mentioned above, most analyses of the proclisis-enclisis alternation, independently of their assumptions on the morpho-syntactic properties of the clitic, as well as on the exact details of the process of pro-cliticization and en-cliticization, account for the alternation by referring either to properties of the host or to properties of the clause (which are indirectly related to those of the host, given a theory of extended projections). Some accounts also link the alternation to properties of the clitic itself, while others to properties that are not directly related to the host but rather to properties of the left periphery of the clause. This plurality of approaches stems from the fact that enclisis is not an empirically uniform phenomenon; on the contrary, there has been enough evidence in the literature to show that there are different types of enclisis, at least on the empirical level. In what follows I will discuss in more detail some of the basic and most widely discussed sub-types of enclisis pointed out in the literature, in order to point out their main similarities and differences, but also in order to put my discussion on Greek enclisis, which will be the main tocus of my research here, into a wider perspective.
5·3 Different types of enclisis: Syntax vs. PF
In this section I discuss three patterns of enclisis found in languages with pronominal clitics. Pattern A: Finiteness In a number oflanguages with pronominal clitic pronouns the proclisis-enclisis alternation depends on morpho-syntactic properties of the elitic host, such as the presence or absence of tense, mood, and/or subject agreement This is one of the most typical and widely discussed patterns in the literature, and is found among others in (some) Romance languages, as well as in Standard Greek, Macedonian and Afuanian. In these languages clitics typically (immediately) precede
Chapter 5. Patterns ofproclisis/endisis and the role ofV-movement 147
verbal forms which are marked for tense, and/or subject agreement, and/or mood, whereas they normally (immediately) follow forms that are not marked for tense, agreement and/ or mood. An alternative way to describe these basic facts would be to say that finite forms take proclitics while non-finite ones take enclitics. Given that finiteness is traditionally linked to subject agreement and/or tense (and not e.g. to voice or aspect), the two alternatives would have more or less the same empirical coverage. Although, as we shall see later on, finiteness appears to be problematic, both on the empirical and the theoretical side, 1 it is quite commonly used in the clitic literature, as it successfully captures the core cases of the proclisis-enclisis alternation in these languages, but also because finiteness has been independently used as a linguistic notion in other fields of the grammar (such as e.g. in the formulation of certain rules like nominative assignment or the Tensed-S condition and the Specified Subject Condition (see Chomsky 1973 for the last two, and Chomsky 1981 for nominative assignment)), showing that it is operative in language and that any problems arising with its problematic definition could be dealt with, possibly at a more abstract level, without having to totally dispense with it (see Joseph 1983 for a discussion on finiteness across languages). To illustrate, I will give examples from different languages which appear to follow the finiteness pattern. Starting with Greek. object pronominal clitics precede an indicative or a subjunctive verbal form, but follow a (positive) imperative or a gerund (see Joseph & Philippaki-Warburton 1987 and Holton, Mackridge & PhilippakiWarburton 1997) (Greek does not have infinitives; it uses controlled subjunctive clauses instead- see Terzi 1992; Philippaki-Warburton & Katsimali 1999; Roussou 2001; Landau 2004, among others).
Standard Greek (2) a. Me simpathuse i spitonikokira mas. Me.cl liked.3sg the landlady.nom us.cl 'Our landlady liked me: b. ea ithela na to pari o Nikos. Put wanted.1sg Subj itcl take.3sg the Nick.nom 'I'd like Nick to have
u:
c.
Pes mu tin alithia! Tell.2sg.imp me.cl the truth.acc 'Tell me the truth!'
d. Vlepondas ton i Maria kokinise. Seeing him.cl the Mary.nom blushed.3sg 'When Mary saw him, she blushed:
[indicative)
[subjunctive]
[imperative]
[gerund]
On the empirical side, cf e.g. that the Old Neapolitan Inflected Infinitive allowed both proclisis and enclisis (Adam Ledgeway, p.c.). This shows that finiteness on its own cannot account for the whole range of empirkal data. Something more needs to be said
1.
148 Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
The indicative and the subjunctive form are fully marked for agreement, and may also inflect for tense. as shown by the following examples:2 (3) a. Mas eksipireti I eksipiretuse. Us.cl serves.3sg.ind.imperf/ served.3sg.ind.imperf 'He is/was serving us? b. na to vrazo I evraza Subj itcl boil.1sglboiled.1sg 'To boil if On the other hand, gerunds/absolutives are not marked for subject agreement and tense, while (positive) imperatives are marked only for [+ imperative], 2nd person and for number, although not for tense. In other words, in terms of pure morphological marking they are less marked than the indicative and the subjunctive: (4) a.
Ksechnondas tin Forgetting.ger her.cl 'Forgetting about her:
tin! b. Ksechna Forget.imp.2sg. her.cl 'Forget about her!' According to traditional grammars, the indicative and the subjunctive are considered to be finite forms (since they manifest tense and agreement distinctions), as opposed to gerunds, which are taken to be non-finite (since they do not manifest any distinctions). The status of imperatives is more controversial, because although they manifest partial/restricted agreement, they do not have any tense. Accordingly, in the generative literature Greek imperatives have been treated either as finite or as non-finite forms by different researchers (see e. g. Philip paki-Warburton 1998 and Joseph 1983, among others). In any case, it seems that at least as far as morphological marking is concerned, imperatives lie somewhere in between the two extremes on the finiteness scale, namely indicatives/ subjunctives on the one hand and gerunds on the other. However, in terms of clitic placement (possibly a syntactic criterion, but surely a distributional one) they pattern on a par with gerunds, rather than with indicatives or subjunctives. In this respect, imperatives have a similar distribution with uncontroversial non-finite forms, a fact which could be argued to lend some support to the traditional description that clitics follow non-finite verbal forms. The finiteness pattern is not restricted to Greek, but it is also found in many IndoEuropean languages which have elitic pronouns, especially those coming from the Romance branch. In Standard Italian, for example, clitics precede indicative and
Mood is marked as [- imperative] on the indicative and subjunctive verbal form. The subjunctive is further marked by the particle na.
2.
Chapter 5. Patterns ofproclisis/endisis and the role ofV-movement 149
subjunctive forms but follow (positive and possibly negative) imperatives, infinitives, gerunds and past participles:
Standard Italian (5) a. Mi interessa il mio lavoro. Me-cl interests-3sg.ind. the mine work 'I'm interested in my work~ b. Spero que tu me lo dica presto. Hope-lsg that you me-cl it-cl say-3sg.subj. soon 'I hope you say it to me soon~
[indica.tive]
[subjunctive]
Dammelo! Give.2sg.imp.- me.cl-itcl 'Give it to me!'
[impera.tive]
d. Voglio telefonargli. Want.lsg call.inf. -him.cl 'I want to call him~
[infinitive]
e.
f.
Guardandoli bene... Looking-at-ger.-them.cl well 'Looking at them carefully..: Salutatala... Greeted-past.part.-her.cl 'Having greeted her.. ~
[gerund]
[past participle]
In Italian, as well, the most marked forms take proclitics: the indicative is fully marked for tense. subject agreement and mood, while the subjunctive displays reduced marking for agreement and arguably no marking for tense (aspect only). Gerunds, on the other hand, and possibly infinitives (modulo the infinitive ending) are not marked for any of these features, and take enclitics. Finally, both participles and (positive) imperatives lie somewhere between fully marked and fully non-marked forms, since the former are partially marked for (object) agreement (i.e. gender and number) while the latter are marked for 2nd person sg./pl as well as 1st person pl. (obligatory inclusive interpretation)-1 (see Graffi 1996; Zanuttini 1997 on Italian imperatives). Both of these forms, however, take enclitics. A similar situation is found in Castilian Spanish: clitics precede indicatives and subjunctives, which are marked for tense. mood, and subject agreement, whereas they follow infinitives (which are not marked for tense/ agreement), gerunds (not marked
3· Italian also exhibits 2nd person singular and plural polite imperatives, which employ forms of the 3rd person singular and plural respectively of the present subjunctive. In this case the clitic precedes the verbal form. Kayne (1992) and Graffi (1996) analyze these types of sentences as declarative clauses with imperative force, which would explain prodisis.
150
Cities in Greek: A Minimalist Account ofProdisls and Endisis
for tense/agreement) and (positive) imperatives (these are only marked for 2nd person sg./pl., 1st person pl (+ mood) (obligatory inclusive interpretation) and 3rd person sg./ pl. (+mood) (formaVpolite imperatives)). In this language, as well, imperatives, which are semi-marked forms, behave on a par with fully unmarked forms as far as clitic placement in concerned:
Castilian Spanish (6) a. Me lo has dicho t:U. Me.cl itcl have2-sg said you-nom 'You've told me that' b. No creo que te lo traiga Marfa. Not think.lsg that you.cl it.cl brings.3sg Maria.nom 'I do not think that Maria will bring it to you: c. De ganarlo ellos, los felicitaremos. Of win-inf.-itcl they-nom, them.cl congratulate.futlpl. 'If they win it, we will congratulate them: d. jDfmelo ahara! Tell-2sg.-me.cl-itcl now 'Say it to me now!' e. Estaba estudi.andolo. Was-3sg studying-it.cl 'He/she was studying it'
[indicative]
[subjunctive]
[infinitive]
[imperative]
[gerund]
Romanian is another Romance language that appears to follow the same pattern: indicative and subjunctive verbal forms are inflected for tense, subject agreement and mood, and take proclitics, while (positive) imperatives are only marked for 2nd person sg./pl., and take enclitics, and gerunds are not marked for tense or subject agreement, and also take enclitics·.4•5 Here, too, imperatives team up with fully unmarked forms with regard to cliticization:
4· In Romanian, the 3rd person singular feminine accusative clitic (i.e. the pronoun o) normally follows the non-finite part of a compound verbal form (namely, the participle, gerund or bare infinitive), i.e. it follows the whole verbal complex (although not e.g. the future indicative, formed by the auxiliary verb vol, vel, va, vom, vefi, vor, and the bare infinitival form). I will not discuss this rather particular case of enclisis here, given that this phenomenon has been argued to be PF-driven (see e.g. Dobrovie-Sorin 1994), but most importantly because o-endisls does not appear to be sensitive to the morpho-syntactic properties of the verbal host (e.g. tense or subject agreement), but on its simplex: vs. complex status. 5· Romanian still makes restrictive use of the infinitive form of the verb, as an independent form (with or without a), but mainly in the formation of compound tenses (where the bare form ls used). In compound tenses a ditic pronoun always precedes the auxiliary heading the verbal complex: (modulo the ditic o). In independent uses, ditics normally precede the infinitive
Chapter 5. Patterns ofproclisis/endisis and the role ofV-movement
Romanian (7) a. M4 asculta ~i im•-6 r.lspunde. Me.cl-acc hears- 3sg and me.cl-dat responds-3sg 'He/she is listening and answering to me: b. Vreau sa m4 asculte. Want-lsg subj me.cl-acc hears-3sg.subj. 'I want that he/ she listens to me: c.
e.
Dandu-mi -1 ... Giving me.cl it.cl 'Giving it to me' Cil.nta -1! Sing-2sg.imp. it-d 'Sing it!'
[indicative]
[subjunctive)
[gerund]
[imperative]
Standard French is another Romance language that patterns like the previously mentioned languages, albeit with a notable exception: in French infinitives and gerunds may only take proclitics, despite the fact that the latter are not marked for tense and/or subject agreement Indicatives and subjunctives are fully marked for these features (as well as for mood) and take proclitics, as expected, whereas (positive) imperatives are marked for 2nd person sg./pl. and for 1st person pl. (with an obligatory inclusive reading) but not tor tense, and take enclitics. That is, imperatives behave on a par with non-finite forms in other languages:
Standard French (8) a.
Jean Ia voit John her.cl sees-3sg.ind 'John sees her:
b. II n'est pas probable qu'il leur parle. It NEG-is not probable that he them.cl talks-3sg.subj 'It is unlikely that he will talk to them.
[indicative]
[subjunctive]
(and follow a, if the latter is present). I will defer discussion ofthe Romanian infinitive (and other forms from both Romanian and from other languages) for later on, where I will argue that morphological marking is only indirectly linked to the prodisis-endisis alternation. 6. Romanian has two series of unstressed pronouns, namely full and short unstressed pronouns. Some researchers have analyzed those as weak and ditic pronouns respectively, in the spirit of Cardinaletti & Starke (1999). In most cases, both full and short pronouns are possible in a 'ditk' position (e.g. immediately before or after the verb, not introduced by a preposition in the accusative, among others), although in some cases only the short ones are grammatkal (see Dobrovie-Sorin 1994 for further discussion).
151
152
Cities in Greek: A Minimalist Account ofProdlsis and Endlsls
Presente -le -moi! Present-2sg.imp - him.cl - me.cl 'Present him to me:
[impera.tive]
pas le faire serait pour nous incomprehensible. not it.cl do-infwould-be-3sg for us incomprehensible '(For the French government) not to do it would be incomprehensible for us:
d. Ne
NEG
[infinitive] French appears to pose a serious problem tor the finiteness pattern, at least on the descriptive level, since infinitival forms, which are the least marked (or one of the least marked) forms in the language take proclitics. I will discuss this issue in more detail later on in this and the following chapter. Here, it suffices to point out that although French infinitives do not fit the finiteness pattern, presumably for a number of reasons to be later identified and further analyzed, French, more generally, is more compatible with this than with any other pattern, given that all other possible clitic hosts follow the general pattern found in other Romance languages. Greek is not the only non- Romance language with pronominal clitics that follows the finiteness pattern. For example, Macedonian, a South- Slavic language (see Friedman 1993), behaves in a remarkably similar way: clitics precede forms which are fully marked for tense and subject agreement (e.g. indicatives and subjunctives), whereas they tallow non-marked (e.g. gerunds, which take an invariant suffix) or semi-marked forms (e.g. imperative forms, which are marked for 2nd person singular and plural) (most of the examples are taken from Tomic 1996)7:
Macedonian (9) a. Im gi dadov knigite. Them-gen.cl them-acc.cl gave-1sg.past books.the.pl 'I have given them the books: b. Saka da mu go dades. Wants subj him.dat.cl him.acc.cl give.2sg.perf '(S)he wants you to give it to him:
[indicative)
[subjunctive]
7· Note that the situation in Macedonlan ls quite more complex than is implied in the main text. For example the position of object pronominal ditics with regard to complex verbal forms (Le. forms composed of an auxiliary verb and a partidpial form) is complicated. since clitics may precede or follow certain auxiliaries depending on a number of fuctors (for example on whether they are have or be auxiliaries, on the person properties of be-auxiliaries), or they precede 1-partictples but follow present (Le. gerunds), past and passive particlples. Although some researchers have attempted to link these more complicated patterns to finiteness (see for example Joseph 1983), it is not clear that this is possible (see e.g. discussion in Legendre 1998 and Tomic 2001).
Chapter 5. Patterns ofproclisis/endisis and the role ofV-movement 153 Daj tni ja knigata! Give-2sg.imp me.cl it.cl book-the 'Give me it, the book!' d. Davaj:Ki mi ja knigata, me bakna. Giving me.cl it.cl book-the me.cl kissed-3sg 'Giving me the book, he/she kissed me:
[impera.tive]
[gerund]
All in all, the finiteness pattern appears to be quite widespread, as it can be found in a number of related or unrelated languages. Although the basic empirical observation described by the Finiteness Pattern will be questioned later on in this chapter, it still remains a fact that at least in some European languages object pronominal clitics precede forms which are fully marked for subject agreement, tense and/or mood (i.e. finite forms), while they follow forms which are unmarked for subject agreement, tense and/ or mood (i.e. non-finite forms). In other words, verbal forms which are at the two extreme ends of the morphological finiteness continuum tend to behave in the same way with regard to clitic positioning across a number of languages (as opposed to verbal forms lying anywhere else within this continuum, namely forms which are marked for some (but not all) features and which appear to be more 'unpredictable' as far as their position with regard to clitic pronouns is concerned); this very fact in itself is quite interesting, first of all because it suggests that finiteness of the host (at least in its morphological guise) may be an underlying factor in the proclisis-enclisis alternation, but also because it suggests that whatever role this finiteness factor is playing in the alternation, it is shared by many languages having adverbal cliticization, which means that investigation into the plausible proclisis-enclisis alternation - finiteness correlation could possibly reveal valuable information about the process of cliticization itself. Whether this turns out to be true or not will be investigated in this and the following chapter. Pattern B: Syntactic and PF factors In a number of languages with clitic pronouns the proclisis-enclisis alternation correlates with (a) morpho-syntactic properties of the host (such as e.g. finiteness); and/or (b) morpho-syntactic properties of the left periphery of the clause, such as focus and topic; and/or (c) PF properties of the pronoun (enclitic vs. proclitic) and/or of the pronoun host This is a complex pattern, as it is fairly obvious from the distinct fuctors mentioned in the above paragraph, which include both morphological/syntactic and/or phonological properties of the clitic pronouns and/ or their host The basic property shared by all languages within this group is that they ban first-position clitics. In other words, they do not allow clitic pronouns to appear in the first position of a sentence. However, non-first-position languages differ in other important respects. For example in some clitic pronouns must appear in a designated position within the clause, typically the
154
Cities in Greek: A Minimalist Account ofProdisis and Endisis second position (Second Position Clitic [2P] or Wackemagel3 languages), while in others they may not appear in first position but are free to appear in any other position within the clause, provided they are adjacent to the verb (Tobler-Mussafia languages9 ). Serbo-Croatian is a typical example of a 2P language, while European Portuguese and Old Romance, more generally, are Tobler-Mussafia languages. To illustrate the main differences between these two sub-types, for example one may find the following pattern"# XP XP CL V" in Tobler-Mussafia languages, but not in Wackernagel ones (where # =beginning of the clause). This is illustrated below with examples taken from Serbo-Croatian and Bulgarian (the examples are taken from Franks 1998: 5, and 49-50; clitics are in bold while the verb is in italics): (10) a.
b.
*Zoran stalno mi kupuje knjige. Zoran constantly me.dat buys books 'Zoran is constantly buying me books: Zoran mi stalno kupuje knjige. *Mi Zoran stalno kupuje knjige.
Serbo-Croatian
go dade vcera. d. Vera mi Vera me.dat itacc gave yesterday 'Vera gave me it yesterdaY. e.
VCera Vera mi go dade.
f.
*Mi go dade Vera vcera.
Bulgarian
Examples (lOa-c) show that Serbo-Croatian, a Wackemagellanguage, restricts clitics to the second position of the clause, while examples (lOd-f) show that Bulgarian, a 8. 'Wackernagel' or 'second position' languages is a term used in the linguistic literature in order to refer to those languages in which some grammatical constituent(s)/element(s) appear(s) in what is loosely defined as the second position of the clause (or the phrase, e.g. in NPs/DPs). Normally, these elements are phonologically weak/unstressed and enclitic, i.e. they need to attach to a host on their left (hence their inability to appear in clause initial position). Examples of Wackernagellanguages are Ancient Greek, Sanskrit, Serbo-Croatian, some other (West) Slavic languages, or Walpirl. As is widely known, the name is attributed to the Swiss philologist Jakob Wackernagel, who first pointed out this property in a number oflndo-Europeanlanguages such as Sanskrit, Latin and Ancient Greek (see Wackernagel1892). Within the generative literature, the term "Wackernagellanguage" has been mainly used with regard to Slavic languages, while the requirement itself is usually morpho-syntactically implemented via a syntactic node in the higher part of the clause (e.g. C, W, l:, or IP - see Dimitrova-Vulchanova 1999 for a detailed discussion on 2P Slavic languages). 9· Tobler-Mussafia languages do not allow clitics to appear in sentence initial position, however they do not require a ditic to appear in a certain position within the clause as long as the previous requirement is met and the ditic is adjacent to the verb. In other words, CL-V orders are disallowed in clausal initial position, whereas V-CL orders are always acceptable, independent of context (Le. independent of whether V is first in the clause or not).
Chapter 5. Patterns ofproclisis/endisis and the role ofV-movement 155 Tobler-Mussafia language, only bans them from clause initial position. Similar examples can easily be found in medieval Romance texts and in modern north-western lbero- Romance varieties as well as in Cypriot Greek and Medieval Greek: (11) a.
0 Joao ja o viu. The John already him saw. 'Peter already saw him~ European Portuguese example adapted from Barbosa 1996: 6 (her 24b))
b. E aqui ei Ia le:x:a. And here he her left.3sg ~nd here, he left her~ c.
Old Catalan; example taken from Fischer 2002: 175(her 233)
Poll anthropi panda kamnoun to sosta. Many people always do-3pl it correctly 'Many people always do it correctly? Cypriot Greek; Terzi 1999b:228 (her Sa)
d. Dhia tuton parakalo se. for this beg.1sg.pres you.oo.sG.WP Late Medieval Greek/ 'For this reason I beg you? Makhairas 83; example taken from Pappas 2004: 140 (his 26) Putting aside CL-V ordering, the examples in (lla-d) show that clitics do not have to be positioned in the second position in these languages. One of the immediate consequences of this basic difference is that Wackernagel clitics do not need to be adjacent to the verb and may encliticize to almost any constituent on their left hand side.
eisko. (12) a. Tudeidei min egoge daiphroni panta Son-of-T him I valiant in-all-ways liken 'I liken him in all ways to the valiant son ofTudeus~ Homeric Greek; example taken from Fontana 1996: 49 (his lOa) b. Da li nam Olga neSto dovikuje. Q us Olga something tells 'Is Olga telling us something?' Serbo-Croatian; example taken from Rivero 1997: 183 (her 29b) This behavior contrasts with that ofTobler-Mussafia languages, where clitics tend to be as adjacent as possible to the verbal host, resembling Pattern A languages in this respect. (13) a.
Petko tni go dade veera. Petko me it gave yesterday 'Petko gave me it yesterday?
b. *Petko mi go vcera dade.
Bulgarian
156
Cities in Greek: A Minimalist Account ofProdisis and Endisis
*en tal que no ouiesses tenido en to poder lo. in such that not you-had held in your power it Old Spanish; example taken from Wanner 1991: 346 (his 28) All in all, Tobler-Mussafia languages appear to lie somewhere between Wackernagel languages and Pattern A languages (see Wanner 1991).10 In this work I will not deal with Pattern B languages, since my main goal is to focus on the finiteness factor only (although see Shlonsky 2004 for an attempt to unite Pattern A with Pattern B languages). Pattern C: D P-ditics A third pattern, which however I am not going to deal with here, involves clitics in the nominal domain. As is well known, in many Balkan languages possessors within a DP may be expressed either by a full DP or by a pronoun. Possessive pronouns are typically special clitics, in the sense that they are phonologically weak elements that need to cliticize on a phonological host and their syntactic distribution is different from that of full DPs and/or pronouns (see Pancheva 2004 for an introduction to the properties of Balkan possessive clitics ). They may appear within or outside the D P to which they are thematically related (cf. e.g. Greek vs. Serbo-Croatian), and they may either occupy a specific/constant position within the DP/ clause or appear in a number of possible distinct positions. For example, in Serbo-Croatian possessive clitics appear outside the DP and must occupy the second position slot of the clause that is also used by any other clitic pronoun. opet pojavila slika. Serbo-Croatian (14) U novinama mi se In paper l.cl refl.cl again appeared picture 'My picture came out again in the paper: (example taken from Pancheva 2004: 181 (her S(a))) On the other hand, in Macedonian possessive clitics may appear both within and outside of the DP. (15) a.
fena mi Wife mine 'Mywite:
Macedonian (from Pancheva 2004: 180 (her 2a))
b. Mu gi zede pari-te Macedonian he.cl.dat. they.clacc took money-the 'Youl(s)he took his money: (from Pancheva 2004: 189 (her 21a))
Another characteristic of Tobler-Mussafia languages is that they have prodlsis in the case of focalization, while they have endlsis In the case of topicalization.
10.
Chapter 5. Patterns ofproclisis/endisis and the role ofV-movement 157 Finally, in Greek possessive clitics may appear either pre-nominally or post-nominally and typically within the DP, but not necessarily in the second position. To oreo mu spiti. The nice me.clgen. house 'My beautiful house:
Greek
b. To oreo spiti tnu. The nice house me.cl.gen. 'My beautiful house:
Greek
(16) a.
What characterizes all these possessive clitics is that (a) they are always enclitic, and therefore they cannot appear at the beginning of the D P or the clause, and (b) they are identical in form with indirect object clitics in the verbal domain. Although Balkan languages differ in terms of the restrictions they may impose on the placement of possessive clitics, they nevertheless treat them as Pattern B clitics in the sense of (a). This appears to be the case even in languages like Greek or Macedonian, which do not impose a clause initial ban on their verbal clitics. Even more interestingly, in both languages possessive clitics have identical forms to indirect object clitics (b). The reasons for this identity seem to be rather complicated, and many analyses have been proposed bringing also the historical development of these possessives into the picture (see Pancheva 2004 and references therein). However, this is not of my concern here. Nevertheless, it might be interesting to point out that according to various researchers Greek possessive clitics differ from indirect object clitics in several respects, even if they appear to have the same form on the surface. I will not discuss these differences here, but I will come back to them later on, when I will be discussing the relation between free clitic ordering in enclitics and the person-case constraint (see Part 6.3.3.3, Chapter 6). It suffices to merely point out the existence of obligatory enclitic possessives in Balkan DPs. Therefore, we can formulate the following descriptive generalization: Pattern C: In a number of Balkan languages possessive clitics are inherently enclitics and hence are not allowed to appear in DP-initial position. However, any further positional restrictions on these clitics appear to differ across these languages. Having discussed the three most common patterns of clitic distribution and having explicitly stated that only Pattern A languages will be thoroughly investigated in this chapter (with only a brief reference on Patterns Band C), I move on to the discussion of Pattern A and the factors that may be involved.
5·4 Finiteness: A rather problematic notion As we have already seen, morphological finiteness, namely the degree to which the verbal form is marked for such features as agreement, mood and tense, appears to play a crucial role in the proclisis-enclisis alternation. However, if we look more carefully at
158 Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
Pattern A languages, it appears that clitic distribution in terms of proclisis vs. enclisis does not always correlate with the (degree) of morphological finiteness on the verbal host. On the contrary, there are quite a few instances of "non-canonical" positioning, namely cases where proclisis obtains instead of the expected enclisis, and vice versa. These non-expected patterns are found both within a single language and across languages. For example, the French infinitive, which does not inflect for tense, subject agreement or mood, and which takes infinitival marking, i.e. a typical non-finite form, takes proclitics. This is exactly the opposite of what we find in other Pattern A languages with infinitives, where a form with similar markings to those of the French form takes enclitics. Similarly, some (positive) indicative forms, fully inflected for present tense, mood and agreement take enclitics in Standard Italian when used as imperatives (e.g. 1st pi, 2nd pi). This is not possible in Spanish, where the same forms may only take proclitics. Also, in Spanish some subjunctive forms may be used as imperatives, in which case they take enclitics even though they are fully marked for mood, agreement and present tense. Obviously, these counterexamples pose a challenge to the hypothesis that morphological finiteness is related to the proclisis-enclisis alternation in Pattern A languages. Besides morphological finiteness, there are also other types or levels of finiteness, such as syntactic and semantic finiteness, which might be related to the proclisis-enclisis alternation. The former refers to abstract correlates of finiteness in the clause, such as for example abstract features on functional heads of the clause (T and C mainly), with concomitant syntactic effects (such as for example overt nominative subjects, creation of syntactically opaque domains in the spirit of the Tensed Subject Condition and the Specified Subject Condition). As for the latter, a plausible interpretation, which is also related to Giv6n's (1990) functional approach on the semantic dependency between two events as well as to Timberlake's (1976) [±dependent] clause distinction, is that clauses may vary in terms of the degree to which they may be semantically (in)dependent For example, an embedded infinitive may be more semantically dependent than an embedded subjunctive, for example in terms of tense or subject agreement. But more generally, semantic finiteness is used in the sense of LF-related finiteness. In other words, it refers to the LF/semantic consequences or reflexes of the particular syntactic heads contained in the derivation. So, for example the presence or absence of a tense operator/tense features in the T head of the clause will give rise to a tensed or a tenseless clause in the LF representation. This approach is in accord with the overall design of the grammar as envisaged in the Minimalist Program according to which it is syntax that feeds PF and LF, with any operations within the interface components not being able to affect syntax itself. but which may create mismatches between the syntactic representation and the phonological and the semantic representation. Moreover, it may be compatible with a 'degree of semantic autonomy' approach to semantic finiteness, provided that the degree of clausal (in)dependency is calculated on the basis of syntactic structure (e.g. the presence or absence of clausal functional heads such as C or T). Whether the syntactic
Chapter 5. Patterns ofproclisis/endisis and the role ofV-movement 159 structure in turn depends on lexical properties of verbs taking clausal complements (as for example argued for in Ledgeway 2007) is an independent question that will not concern me here. 11 As a result, in most cases semantic finiteness will be correlated with (abstract) morpho-syntactic finiteness in a 1:1 relation, and therefore I will normally talk about syntactic finiteness with semantic correlates (unless LF operations change the syntactic representation). Such an approach can easily capture the basic generalization that the different levels of finiteness do not necessarily correlate with each other. Untortunately, if we look carefully at the data, it is fairly easy to conclude that neither of these two additional levels of finiteness is empirically adequate to make the correct predictions, a fact which creates problems for our initial hypothesis. For example, semantic and (abstract) morpho-syntactic finiteness would predict that controlled subjunctives in Greek should take enclitics, given that they are [-tense] and[- referential agreement], contrary to fact. Or, they would predict that imperatives should take proclitics, since they are [+referential agreement] (cf. that they take pro as a subject), although we know that they take enclitics. Concretely, both the G B framework and the Minimalist Program take syntactid semantic finiteness of clauses to be tied to the following manifestations: (a) the establishment of a syntactically and semantically opaque domain 12; (b) the presence of an overt (lexical), nominative subject (or pro, if the language allows for one) which is potentially referentially independent; (c) the definition of a separate temporal domain (alternatively, the projection of a tense operator) by the clause, which may or may not be dependent on the temporal domain of another clause (e.g. subjunctives vs. indicatives). These properties are taken as surface hallmarks of finiteness by most (if not all) syntactic analyses within the Chomskian tradition, even though there may be differences in the actual mechanisms/teatures/heads that are postulated in order to derive the above mentioned manifestations. However, two points are generally accepted by most current (minimalist) analyses, and hence appear to be less controversial: (a) the head(s) and/or feature(s) responsible for the presence or absence of the above surface manifestations of syntactid semantic finiteness are located in the C- T domain; (b) these features may be subject to parametric variation, depending on the language (e.g. Tense
u. Adam Ledgeway has pointed out to me that it is semantic finiteness that determines syntactic finiteness. For instance, according to him it ls control properties which determine whether nominative subject is present or not, whether T/AGR is present or not, etc. Presumably, semantic finiteness here reduces to lexical properties of verbs taking clausal complements, however it is not dear whether these properties are not also morpho-syntactic in nature. Note that what actually matters is to have an overt, referentially free subject. The exact case realized by the subject appears to be language dependent and possibly related to other properties ofthe clause [e.g. nominal properties] (c£ for example genitive subjects of finite gerunds in Turkish (see Kornfilt 2007), and Ancient Greek infinitives with accusative subjects (see Sevdali 2006) ).
12.
160
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
vs. AGR vs. Mood)Y As a result. finiteness is viewed as the correlate of features (both) on C and T (or INFL). 14 This captures the traditional observation that properties on T (such as the presence of Tense and/ or Agreement) typically correlate with properties on C (cf. e.g. selection of different types of INFL by different types of Cs, the absence or presence ofC, depending on the properties on INFL- see Rosenbaum 1967; Stowell 1982, or the choice of a particular mood depending on C, as is the case e.g. in Balkan languages (see Roussou 2001)). More recently, Rizzi ( 1997) has attempted to capture this correlation in syntactic terms, by postulating the projection of a lower C head, which he calls CFin, and which is argued to be the locus of clausal finiteness that is independent of purely INFL properties/heads (such as T/ AGR/mood), although it agrees with them, establishing as a result a syntactic bond with them. This is possible, because CFin has a number of INFL-lik.e syntactic features such as Tense and AGR, although in a rather rudimentary way. These features trigger C-T agreement. overt manifestations of which we may see in a number of languages (e.g. agreement or tense on complementizers - see Adger 2007 for a discussion). An alternative way of capturing this fact is to use Chomsky's (2005) inheritance operation, according to which C selects T, giving it its phi-features (arguably, tense is already on T, being an interpretable teature, however this does not necessarily exclude the presence of (uninterpretable) T feature on C, given that C appears to select tensed vs. untensed INFL-clauses). Additionally, other researchers have claimed that CFin, being the locus of clausal finiteness, also contains semantic properties, in the sense that it defines the semantic dependency of the clause. In this sense, they try to capture the functional intuition that finite clauses are semantically independent, as opposed to non-finite clauses which are semantically dependent (in the sense that their tense and agreement (syntactic and intepretational) properties are dependent on those of the matrix clause). This maybe implemented in different ways: for example Bianchi (2003) assumes that CFin contains a variable for the speech event that is identified deictically or anaphorically (i.e. within the clause or by being anchored by the speech time of the matrix clause), and which is syntactically realized as a ± finite feature. In this way, she captures the intuition that the temporal and logophoric centres of the clause (which are related to the speech event) may or may not be interpreted by means of those of another (the matrix) clause. Roussou (2001) also proposes something similar, assuming that the finite/ non-finite distinction is contextually defined, by means of a CFin head which provides the anchoring point to the speech time and which may be interpreted either
For example in Greek tense is the relevant feature for finiteness rather than agreement or mood (see Iat:ridou 1993), while In English it ls both tense and agreement that appear to be crucial. It seems then that languages are parameterized as to which of these features play a role In finiteness.
13.
14. So, finiteness ls not taken to be a primitive, after all Correlates of finiteness do not occur in
v*, even though diticization Involves incorporation into v*, since In Greek object agreement does not interact with the subject agreemenUtense/mood system In any way.
Chapter 5. Patterns ofproclisis/endisis and the role ofV-movement deictically or anap horically depending on the semantic ( selectional) properties of the selecting predicate. In this way, she can capture the fact that semantically speaking non-finite and finite clauses have similar properties cross-linguistically, even if they are realized morphologically in different ways. Based on the above discussion, we may define the syntactic/semantic properties of different types of clauses which have been traditionally labeled as finite or non-finite, and see whether the proclisis-enclisis alternation correlates with a particular feature, or a particular value of a feature, or a combination of more than one features and/or values. This is what I attempt to do, using a small number of clauses, in the table below. We have: (17) a.
Greek manner gerunds Irthe trechondas tin. Came.3sg running it.cl.acc. 'He came running:
Greek
These do not inflect for tense or agreement and they take a PRO subject They take enclitics. (see also Chapter 6) b. Greek temporal gerunds Eftase echondas tin treksi oli ti diadromi. Arrived.3sg having it.cl.acc run all the route.acc 'He arrived after having run throughout the whole route:
Greek
These have abstract tense and possibly abstract agreement, take a pro subject, but do not inflect for agreement (although they inflect for tense- cf. the complex tense with echo). They take enclitics. (see also Chapter 6) c.
Greek tensed subjunctives Thelo o Nikos na tus dhioksi avrio. Want.lsg the Nikos subj them.cl.acc. kick.out3sg. tomorrow 'I want Nikos to kick them out tomorrow:
Greek
They have both syntactic and morphological tense and agreement and they take a pro subject. They take proclitics. (see also Chapter 6) d. Greek tenseless subjunctives Ksero na tu apando sto tilefono. Know.lsg subj. him.cl.gen answer.lsg. to-the phone.acc 'I know how to answer him on the phone:
Greek
161
162
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
They have morphological agreement, take a PRO subject and they take proclitics. (see also Chapter 6) e.
French infinitives Ne pas le faire
serait
pour nous incomprehensible. French (France) NEG not it.cl do-inf. would-be-3sg for us incomprehensible '(For the French government) not to do it would be incomprehensible for us: French infinitives take no other marking than the infinitival ending, take a PRO subject, they have no tense and take proclitics. f.
Tensed Spanish infinitives De ganarlo ellos,
los
felicitaremos. Spanish (Spain) De win.inf.it.clacc they.nom, them.cl.acc will.congratulate.l pl 'If they win it, we will congratulate them:
They take overt nominative subjects (or pro), have syntactic/ semantic tense but no agreement and they take enclitics. g. Spanish subjunctives with imperative syntax jAmela Usted! = jque la arne Usted! Love-her you = subj her love you Love her!
Spanish (Spain)
They take overt nominative subjects (or pro), they do not inflect for tense, but they are marked with the 3rd person sg. and pl. subjunctive endings, and they take enclitics. h. Romanian F(ree)-subjunctives Vreau sa mii asculte. Want that me.cllistens.3sg 'I want that (s)he listens to me:
Romanian
Romanian Free - subjunctives are marked for agreement, they take a nominative subject (or pro), and take proclitics. (see also Landau 2004) i.
Romanian C(ontrolled)- subjunctives pedepseasca pe Mihai. Ion a incercat sa -1 John have-3sg try-PART sa him.cl.acc punish-3sg.subj. pe Mihai 'John tried to punish Mihai? (example taken from Dobrovie-Sorin 2001:45 (her l(a))
Romanian controlled subjunctives take PRO, have morphological agreement, but are not marked for tense, and they take proclitics.
Chapter 5. Patterns ofproclisis/endisis and the role ofV-movement j.
Obviated Romance subjunctives Quiero que ttl lo hagas. Want.lsg that you itdo.subj.2sg 'I want you to do it'
Spanish (Spain)
They take a nominative subject or a null pro, morphological subject agreement, as well as syntactic tense and agreement, and they take proclitics. k. Spanish/ Greek true imperatives jEsttldialo! Study.2sg.imp- it.clacc Dhiavase to! Read.2sg.imp it.cl.acc 'Study/read it!'
Spanish (Spain) Greek
True imperatives do not carry tense (morphological or syntactic) although they have agreement. take a nominative subject (or a pro) and take enclitics. (see also Chapter 6)
1.
Obviated Romanian and Albanian subjunctives e 21"1 sa mamnce. Ion1 vrea ca John wants COMP e PRT eats 'John wants (him) to eat'
Romanian (Terzi 1992: 106)
Obviated Romanian subjunctives have (semantic) tense and semantic/morphological agreement (see Landau 2004), take a nominative subject (pro), and take proclitics. m. Greek non-embedded na-clauses na min to stili subj not it.cl.acc send.3sg. (S)he should not send it.
Greek
Non-embedded na-clauses take nominative subjects (or pro), i.e. they have syntactic agreement (as well as morphological agreement), although they do not have syntactic or morphological tense, and they take proclitics. n. Greek raising constructions matheni ta italika. 0 Janis arxise na ta The Janis started subj them.cl.acc leam.3sg the Italian.acc 'John started learning Italian:
Greek
Greek raising constructions are C-subjunctives, i.e. they do not have syntactic or morphological tense, but they have syntactic and morphological agreement, a copy in their subject position, and they take proclitics.
163
164 Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
All the data presented in (17) above. are summarized in the table below.
(18) Table 1: Proclisis/Enclisis and Finiteness Syntactic Syntactic Subject Morpho- Morpho- C-properties Clitic (mainly logical positionTense Agreement logical C-Fin) Tense Agreement ing Greek manner gerunds Greek temporal gerunds Greek tensed subjunctives (Landau's 2004 F-subjunctives) Greek tenseless subjunctives (Landau's 2004 C-subjunctives) French infinitives (cornplernentsto control verbs) Tensed Spanish infinitives Spanish subjunctives with imperative
-I+
PRO
-ondas ending Enclisls (possibly)
+
+
pro
+
+
+
pro
+
-I+
PRO
-I+
PRO
-I+
Lexical subject
+
+
pro
+
pro
-ondas ending Enclisls (possibly)
+
Napartide
Prodisls
+
Napartide
Prodlsls
Particles llkea (but not de)
Proclisls
Endlsls
Enclisls
+
syntax Romanian
F-subjunctives
+
+
+
Sa particle
Procllsls
Chapter 5. Patterns ofproclisis/endisis and the role ofV-movement 165
Syntactic Syntactic Subject Morpho- Morpho- C-properties Clitic (mainly logical logical positionTense Agreement C-Fin) Tense Agreement ing Romanian C-subjunctives Obviated Romance subjunctives (French/
+
-I+
PRO
+
+
pro
+
+
pro
+
verb
Enclisis
+
pro
+
Sa particle// overt compl.
Proclisis
+
pro
+
Napartide
Proclisis
+
copy
+
Napartide
Prodisis
Sa particle
Proclisis
Proclisis
Italian/ Spanish) Spanish/ Greek true imperatives Obviated Rumanian Albanian subjunctive Greek non-embeddedna/ subjunctive clauses Greek raising
+
From this table we can infer that the proclisis-enclisis alternation does not correlate with any value (or combination of values) tor syntactic/ semantic tense or agreement For example. enclisis is found with[--/+],[++].[+-/+].[-+], whereas proclisis is found with[++], [- -/+], [+ -/+], [-+].In other words, the same combination of values maps onto both proclisis and enclisis . .M for [+ Tense] and [- Tense] and [+Agreement] and[- Agreement] (as single values), both are found with both proclisis and enclisis. Moreover, the alternation does not correlate with the type of subject, or with the properties of C-Fin (in the sense of what element appears in that head). Note that as far as PRO is concerned, I have put both± values, given that in the literature PRO has been analyzed in two distinct ways, namely as involving Agree of q>-features (see e.g. Landau 2004) or as involving an absent or inactive AgrS node in the embedded clause (see e.g. Roussou 2001). Either of these options would do, since the conclusion remains the same. Finally, note that the values for morphological Tense and Agreement are also included in the table. This was done for the convenience of the reader, as one may figure out quite easily that although morphological agreement
166
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
appears to correlate in most cases with proclisis (while lack thereof correlates with enclisis ), nevertheless there are cases where this does not hold, confirming the conclusion we have already reached, namely that morphological finiteness (in this case agreement) does not always make the right predictions. What is more interesting is the fact that when morphological agreement does not map onto the same value with abstract/ syntactic agreement, which happens only when morphological agreement is negative, the most reliable correlation (in the sense that it captures more structures) is morphological agreement rather than syntactic agreement (modulo the two cases which are not captured by morphological agreement). The same result follows if we apply Landau's (2004) system on the above data. Landau attempts to account for the appearance of pro/PRO in both finite and non-finite clauses in a number of languages (including Hebrew, English, and Balkan languages) by means of a calculus that involves Agreement, Tense and R(eferential) features on C, Taswell as PRO (as far as
Greek manner gerunds Greek temporal gerunds Greek tensed subjunctives (Landau's 2004 F-subjunctives) Spanish/Greek true imperatives Frenclt infinitives (complements to control verbs) Tensed Spanish infinitives Spanish subjunctives with imperative syntax
MorphoSyntactic Agreement
C-properties Clitic (mainly C-Fin) positioning
+
PRO pro pro/PRO
-T T +T +Agr
Endlsls Endlsls Prod isis
+
pro
Verb
Endlsls
PRO
-T
Prod isis
Lexical subject pro
T +Agr -T
Endlsls
+
+
Subject
+
+
End isis
Chapter 5. Patterns ofproclisis/endisis and the role ofV-movement 167
Romanian F-subjunctives Romanian C-subjunctives Obviated Romance subjunctives (French/ Italian/Spanish) Obviated Rumanian and Albanian subjunctive Greek non-embedded na/ subjunctive clauses Greek Raising
Syntactic Tense
MorphoSyntactic Agreement
Subject
+
+
pro
+
+ +
+
C-properties (mainly C-Fin)
Clitic positioning Prodisis
PRO pro
+T +Agr -T +T
+
pro
+T
Prodisis
+
pro
-T
Prodisis
+
Copy
-T
Prodisis
Prodisis Prodisis
This table shows that proclisis or enclisis cannot be correlated with any feature or combination of features either on T or on C. The only observation we can make is that [+ Agr] correlates with proclisis in most but three cases. Note that in this system a positive value for Agr implies both morphological and syntactic agreement, whereas a negative value refers only to abstract agreement. This means that proclisis in this system will correlate in most instances with both morphological and syntactic agreement However, as we have seen in the previous table, it seems that morphological agreement is a more relevant factor, because in enclisis [- morphological agreement] will always correlate with enclisis (modulo the three exceptions) even if there is abstract agreement on T. The conclusion from the above discussion is that morpho-syntactic agreement (and possibly morphological agreement) on T offers us the best correlation we can get as far as the proclisis-enclisis alternation is concerned, contrary to pure syntactic/ semantic tense or agreement, which do not appear to make any valid predictions. However, morphological agreement (and morphology more generally) does not come without problems: first of all, the correlation it offers us is only partial. Recall that this was the main reason we decided to look into other levels of finiteness in the first place. Second, even if we assume that the correlation holds, it is not clear why cliticization, a syntactic operation according to our analysis, would be sensitive to morphology. Although there might be a good reason for that, and although it is not a priori impossible, it still is suspicious, especially if we take into consideration the exceptions to the correlation. Now, if all levels of finiteness are inadequate to capture the empirical fact, the question that arises is how to proceed from here. I think there are two possible ways: one is to abandon finiteness altogether and look into other properties of these clauses which might correlate with the alternation. The second way is to try to stick to finiteness as a more general observational tendency. This could be achieved either by
168
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
trying to account for the exceptions in a principled way. so that they can be dismissed as irrelevant to the correlation, or by investigating the nature/components of cliticization itself in the clause, especially with regard to the clitic movement and head movement and their morpho-syntactic interaction in proclitic and enclitic structures. ln the rest of this chapter as well as in the following chapter, 1 will follow the second approach. I will argue that finiteness does play a central role in the alternation, albeit at the morpho-syntactic level only. and more specifically via the interaction of the C-T domain with (independent morpho-syntactic) properties of clitics and their hosts as well as with the operation that brings them together (i.e. cliticization). All known exceptions to the traditional observation will be given an independent syntactic or purely PF account, and therefore will cease to be a problem for the established correlation. Finally, an explanation will be given for why morphological agreement appears to be the best correlation we can get in the above tables, based on the interaction between syntax and morphology/PF. In this way, the original intuition will be kept alive, although reinterpreted according to current syntactic theory so as to account for all the available facts. Before that, however, I will first look into V-movement and its morpho-syntactic correlates, as it has been proposed as the alternative solution to the proclisis-enclisis alternation. Although V-moment is an indispensable part of the alternation, without which we cannot account for the facts, nevertheless it will be argued that it alone cannot explain them. That is, although it is a necessary condition, it is not a sufficient one.
5·5 V-movement Within the generative literature (i.e. GB and Minimalism) quite atew different types of analysis of the prodisis-enclisis alternation have been proposed. Some of these analyses try to incorporate the finiteness 'intuition' into their proposal, while others barely refer to it, giving it only a very marginal role, or alternatively they dispense with it altogether. One of the most influential approaches of the latter type has been what I will generally call 'the V-movement analysis~ The basic idea behind this approach, which involves a number of similar although not identical analyses (see e.g. Kayne 1991; Rizzi 1993; Kayne 1994; Rivero 1994; Philippaki-Warburton 1995; Rivero & Terzi 1995), is that the proclisis-enclisis alternation is directly related to the relevant position of the verb in the clause with regard to the clitic pronoun. More specifically, proclisis obtains when the verb moves to a functional head within the clause that is located in a lower position and to the right of it in relation to the functional head targeted independently by the ditic pronoun(s). On the other hand, enclisis obtains when the verb moves to functional head that is located in a higher position than the one targeted by the clitic(s). All this follows from the independent assumption that clitics and verbs target their position within the clause in an independent manner, and in this respect this is in essence a Kaynean approach (see Kayne 1975, 1991). As it is
Chapter 5. Patterns ofproclisis/endisis and the role ofV-movement 169
expected, different analyses differ in details (sometimes crucial), such as for example whether clitics target the same position across structures both within and across languages, orwhethertheymove to that position or are directly merged there, in the exact nature of the position where the clitic ends up (e.g. whether it is a head or a spedfier), or, finally, in the way the actual positioning of the verb in relation to the elitic( s) takes place as well as the trigger for this structural result However, all of them assume that in enclisis the verb is higher in the clause than the clitic pronoun(s). It was mentioned above that most of these analyses only give a marginal role to finiteness. The general idea is that the verb moves (overtly) to a functional head if it has morpho-syntactic features that are attracted (via Agree) by corresponding features on that head. Accordingly, a form, and more specifically, a non-finite form will end up higher than elitics (yielding enclisis ), if a relevant head with the appropriate features is merged above the clitic position(s). On the other hand, a finite form will end up lower than clitics (yielding proclisis), if there is no head above the clitic position(s) carrying appropriate features (and as a result the verb will only move to some head lower than clitics ). Whether this is true among Pattern A languages, and why this may be the case (i.e. why certain forms correlate with certain syntactic positions) is a matter ofinquiry. In order to illustrate how this type of analysis works, I will discuss Rivero (1994), which is one of the most representative analyses of this type, especially for Greek and other Balkan languages. Rivero looks into three types of non-finite movement which are found in Balkan languages and which involve V-movement to heads that lie higher than the AGR/T head, which is where clitics are located, yielding enclisis. These movements include Long Head movement (LHM) (i.e. movement of a non-finite lexical verb (typically an infinitive or a participle) across a finite auxiliary), found in Bulgarian and Romanian (as well as other conservative Romance languages- e.g. European Portugueseand see Lema & Rivero 1989), (positive) imperative movement, found in Romanian, Greek and partially in Albanian (also tound in other Romance varieties), and gerund movement, found in Romanian, Greek but not in Albanian (possibly in other Romance languages, as well- e.g. Spanish). Assuming a common basic clausal structure for Balkan languages, according to which Agr/T is c-commanded by M(ood), Neg and C (in that particular order), the verb is located in Agr/T in so-called finite clauses and clitic pronouns are directly merged in Spec Agr/T, as well as based on a number of syntactic criteria (such as the root/ embedded asymmetry, proclisis vs. enclisis, possibility vs. impossibility of negation) she analyzes these three structures as involving V-movement toM or C, both lying above Agr/T. Moreover, she follows a relativized minimality approach to the Head Movement Constraint, according to which heads are distinguished into A and A-bar, on a par with fullXPs (see Roberts 1993, 1994), and where only heads of the same type may count as potential interveners. In this way she can predict which head selectively blocks V movement to Cor toM in the above cases, based on the properties of the landing site and the properties of the potentially blocking head. Focusing on imperatives and gerunds, given that Greek does not have LHM, positive imperatives in Greek (and Romanian) involve V-to-C movement, triggered by a
170
Cities in Greek: A Minimalist Account ofProdlsis and Endlsls
special imperative affix on C with the corresponding illocutionary force. Since C is arguably an operator head (cf. for example the illocutionary force), it has A-bar properties. This fact in combination with the assumption that Neg is also an operator head correctly predicts that true imperatives cannot be negated. 15 Non-embeddedness is also accounted for, given that C is occupied by the moved verb, while enclisis simply follows from the fact that clitics lie lower in the clause. in Spec Agr/T. Presumably, Mood is either not projected, or if it is it must be empty (this is on the assumption that Mood particles are also A-bar heads, arguably because they have modal interpretation). As for surrogate imperatives (i.e. the use of forms other than imperative with an imperative use but a non-imperative syntax, such as for example negated subjunctives in Greek and Spanish, which are used as negative imperatives but which have a finite. non-special (subjunctive) syntax), she argues that these do not involve an affix inC; as a result. they remain in T/Agr, they take proclitics, they can be negated, and more generally they do not have a special syntax. A tree diagram of this analysis for true/ positive imperatives is given in (20) below (her (73) on page 106, slightly modified): CP
(20)
~ (MP)
C
I~ V
(M) (t)
Agr/TP
~ CL Agr/T' ~ Agr/T
VP
t
v t
Note that in the above tree diagram a negative head merged between Agr/T and C would induce a HMC violation according to this analysis. This is shown in the following simplified structure: (21) *(cPV ... [NEGP NEG [Agr/TP
CL
(V) ... [VP (V)]]]]
15. Note that an extra assumption is also necessary, namely that V cannot adjoin to Neg, taking it along to C, or alternatively that V cannot adjoin to Neg and afterwards excorporate to C. See also Iatridou (1994) for a critical view of the A vs. A-bar head analysis.
Chapter 5. Patterns ofproclisis/endisis and the role ofV-movement 171
(22a) shows the structure proposed by Rivero for surrogate (subjunctive) imperatives. Note that although the subjunctive particle is merged below negation, it adjoins to negation yielding the surface order Subj-Neg (for an alternative analysis of preverbal particles in Greek see Philippaki-Warburton 1998, and Roussou 2000 for an overview of all the main analyses that have been proposed on that issue). Note that the exact analysis of preverbal particles is not of concern here, given that what we are interested in is how proclisis/enclisis is derived. In most (if not all) analyses this is usually achieved via the intervention of a preverbal particle: (22) a.
[cP ... [NEGP NEG [MP na [Agr/TP CL V ... [VP (V)]])]]
b. Na min to dhis. Subj not it.cl see-2sg 'Don't watch it!'
Greek
A rather similar analysis is given for gerunds, the only difference being that this time the affix is located lower, in theM head (at least in Greek). The immediate consequence is that the gerund takes enclitics and can be negated, given that clitics are in Agr/T and Neg is merged above M. In Albanian, where theM particle duke is not affixal, the verb stays in Agr/T, yielding proclisis, and negation is also possible. The relevant structures are given below (exampled is taken from Rivero 1994: 115 (her 92c)): (23) a. b. c.
[NEGP NEG [MP V-ondas [Agr/TP CL- (V) ... [VP (V)]]]] [Greek gerunds] [NEGP NEG [MP duke [Agr/TP CL V ... [VP (V)]]]] [Albanian gerunds] Min gnorizondas to ... not knowing it.cl
mollen]. d. Jam [duke e hanger Be-pres-ls [PRT it eat-Prtcpl apple-the] 'I am eating the apple:
Greek Albanian
Whether this analysis is empirically and theoretically adequate is an issue that I will discuss in more detail when I will be presenting the structure I am assuming for these clauses in the following chapter. What we should keep in mind is that according to this (type of) analysis enclisis is derived by moving V to a head that is higher than Agr/T (the head clitics are located at), and that the trigger for this movement may either be morphological (as in true imperatives and gerunds) or related to independent phonological properties of clitics (in LHM, at least in Bulgarian, although not in Romanian). 16 The advantage of this approach, apart from the fact that it appears to be able to explain a gamut of structural properties of the above mentioned clauses, is that In subsequent papers (e.g. Rivero & Terzi 1995) the trigger for imperative V-to-C movement becomes more abstract, taking the form of abstract syntactic features with a special semantk interpretation (i.e. imperative logical mood) which are strong (in the sense of Chomsky 1995) and which are present both in imperative C heads and on Vs with true imperative morphology, inducing in this way overt V movement. This means that the verb does not move to
16.
172
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls it is flexible in two ways: first, it correctly predicts that a morphologically non-finite
form may take proclitics, or that a morphologically finite form may take enclitics, given that all that matters are the abstract morpho-syntactic features of the form and the exact position they will be eventually checked in with regard to the clitic position. Given that morphology only partially mirrors syntax, especially when a form is unmarked, such mismatches are expected. Second, although this approach links the proclisis-enclisis alternation to properties of the C-head, it allows us to keep the spirit of the traditional intuition that lies behind the correlation between finiteness and the proclisis-enclisis alternation, albeit in a more indirect and possibly more reliable way. Concretely, the alternation does depend on properties of the verbal host, nevertheless these are abstract, morpho-syntactic properties (not the morphological realization thereof) that are linked to the surface position of the verb (by agreeing with it and triggering overt V-movement), but are not necessarily agreement and/or tense properties (cf for example imperatives and gerunds), although they are still verbal properties/ features (since they are contained in heads that are part of the verbal extended projection in the sense of Grimshaw 1991). 17 Given the traditional empirical observation that C-features typically correlate with T-features (e.g. via selection - see Stowell1981 ), the prodisis-enclisis alternation simply reduces to the fact that different types of T-heads are selected by different types of C-heads. 18 Apparently, according to such an approach morphological finiteness would only partially mirror abstract finiteness (i.e. T/Agr features) which in tum would be linked to the (abstract) properties on C, that is to the actual 'controllers' of the alternation. In other words, morphological finiteness (i.e. marking) would be constrained to the indirect role of partially mirroring abstract properties on the C-T complex. Although it is not at all dear whether this link between finiteness and properties of the C-T domain can be maintained for all cases and types of clauses (cf. for example restructuring infinitives in Romance languages, which have been analyzed as bare VP complements- see for example Wurmbrand 2003 and references therein), the basic proposal of the analysis, namely that enclisis follows from V-movement across the cliticization site to a higher head, may be applicable in all cases, provided there is independent support for the trigger of this movement In other words, it wouldn't really matter which head triggers V-movement, as long as that head lies higher than the cliticization site. Of course, this would have concomitant effects on the validity of the correlation between V-movement, C-T properties and the proclisis-enclisis alternation, at least in those cases where C-T is not involved. However, it is also possible that the correlation could be interpreted in a more abstract sense, so as to include also these pick up an affix; rather it moves to check its own abstract features (as well as the corresponding features on the C head). This would help us account for the fact that the least finite forms are those that move the furthest, which seems to be counterintuitive in terms of finiteness (Adam Ledgewayp.c.).
17.
18. Here C and T are used as general labels for different types ofheads.
Chapter 5. Patterns ofproclisis/endisis and the role ofV-movement 173
cases, for example by arguing that the absence of certain features (and as a result the nature of a clause) may either involve lack of these features on a head or total absence/ truncation of the relevant head. This seems to be supported both by empirical evidence as well as by standard theoretical assumptions on the notion of finiteness (see discussion on finiteness in the previous section). Note that although Rivero's analysis does not analyze infinitival structures, given that Greek and more generally Balkan languages do not have infinitives (besides a few residual cases in some languages only - see also Joseph 1983), the analysis can readily extend to these structures, too. For example, Kayne (1991, 1994) analyzes infinitives with enclitics as involving V-movement across the cliticization site, as opposed to infinitives with proclitics which involve V-movement to a head that is lower than the cliticization site (cf. French infinitive). Similar analyses for proclisis or enclisis in different languages have been proposed in the literature (see for example Ledgeway & Lombardi 2005 for Southern ltaliandialects, and for Romance). Although these analyses may differ in other respects, all of them share V-movement as a common mechanism for deriving proclisis vs. enclisis. The additional advantage as suggested by all the above proposals is that the proclisis-enclisis alternation does not depend on the actual heads where clitics and verbs are located; to put it in other words, there is no requirement that clitics only target a specific head/ spec across structures both within a single language and across languages, and the same applies to verbs. What is crucial is merely that there is some feature triggering overt V-movement to a head that lies higher than the cliticization site, which may or may not differ from one clause to the other. As a result, both enclisis and proclisis may correspond to one, two or more structures. Whether this is empirically correct is a different issue I will discuss right below. To summarize what I have discussed so far, Rivero (1994) and other researchers have proposed that enclisis obtains when the verb moves to a higher functional head across the cliticization site, whereas in proclisis the verb does not move past the clitic. V-movement is related to abstract morpho-syntactic features, which in many non-finite clauses appear to be located in heads that are higher than the clitic position, capturing thus in a more indirect way the traditional correlation between finiteness and the proclisis-enclisis alternation. This type of analysis raises two main issues: (A) the relation of clitic pronouns to V-movement; (B) the empirical adequacy ofV-movement Starting with (A), recall that Rivero (1994) locates clitic pronouns in Balkan languages like Greek, Romanian, Bulgarian and Albanian as 'attached to Agr/TP, presumably in the Spec of this projection (Rivero 1994: 68), although in a footnote (see ft. 3, page 68) she explicitly states that she does not take any stand as to the XP status of Balkan clitics, or as to whether they adjoin to Agr/TP or simply move to its Spec. However, as I have shown in Chapter 4, there is enough evidence that Greek clitics are (also) heads, and similar proposals have also been made for Romanian, Albanian and Bulgarian (as well as Macedonian- see Dobrovie-Sorin 1994, Franks & King 2000). As a result, clitic pronouns should block head movement, on a par with particles in M and Neg (unless they are A heads- Elena Anagnostopoulou p.c.). In fact, Rivero & Terzi
174
Cities in Greek: A Minimalist Account ofProdisis and Endisis
(1995) point out this problem, as they assume that Greek clitics adjoin as heads to a functional head that is located right above T/Agr (i.e. the head independently targeted by the verb in finite clauses), following parallel proposals for Romance in Kayne (1994 ). Their solution is to adopt and adapt Roberts's (1994) analysis on head movemenl 19 In particular, they assume that heads are distinguished into A- and A-bar, on a par with XPs, as well as into L-and non-L-related (in the sense of Chomsky & Lasnik 1993), however they argue that these two categorizations should be kept apart, pace Roberts (1994), who reduces the A vs. A-bar distinction to the L- vs. non-Lrelated one (following in this the spirit of Chomsky & Lasnik 199 3, who also claim that the A vs. A-bar position distinction should be reinterpreted as L- vs. non- L related position). Based on this dichotomy, they further claim that the ECP interpretation of the HMC is sensitive only to the A vs. A-bar distinction. This is evident from imperative V-to-C movement, where Neg, C and the (F head occupied by the) clitic are all L-related to V, nevertheless only Neg blocks V-to-C movement, given that C and Neg are A-bar positions (due to their semantic/operator properties). The benefit of such a solution to the 'V (head)-movement across clitic' problem is that it appears to work, at least in the imperative case discussed above, without adding any non independently needed machinery to the theory. Recall that the most popular analysis by far among syntacticians of the ban on negated true imperatives in a number of languages is that negation blocks V-to-C movement This is even the case in the mimimalism era, during which the ECP is not being used any more, while the blocking effect of negation is simply interpreted in terms of do sest goal (see Chomsky 200 1a, Rizzi 2001). On the other hand, the main disadvantage of such an analysis is that it is not always clear why a certain head is specified as A or A-bar (or, alternatively as L- or non-L related). As it was pointed out above, not all researchers agree on the nature of the dichotomy, as some assume that the new Unon- L distinction should subsume the old N A-bar distinction based on morpho-syntactic properties of heads alone, while others assume that these two distinctions cannot be reduced to one, given that the old distinction (viz. N A-bar) (also) refers to semantic properties of heads, while the new distinction (viz. Unon- L) refers to morpho-syntactic properties of heads, and these
19. Note that for Roberts (1994) the ditic does not appear to cause any problems to long V movement in enclitic constructions, mainly because it is adjoined to a L- related head (Agr ), and
he assumes that V-to-C is a non-L-related movement chain. Of course, in a strict interpretation of Chomsky's & Lasnik's (1993) notion ofL-relatedness, V-to-C movement of true imperatives should probably be analyzed as L-related movement, which predicts that the head on which the ditic is adjoined should block this movement, contrary to fuct. This is exactly the reason why Rivero & Terzi (1995) maintain the AIA-bar distinction, which is based more on semantic properties, rather than purely morphological ones. This, again, shows the problematic nature of this type of analysis, which is not based on an independently defined typology of heads (and/or features), but rather on a pre-theoretical notion of a distinction that should hold for heads as it holds for maximal projections.
Chapter 5. Patterns ofproclisis/endisis and the role ofV-movement 175 two do not necessarily coincide with each other.20 It is obvious, though, that if there is no principled way to establish the type of each head, a system that allows or disallows particular head movements depending on head-types becomes ad hoc and circular, given that the hypothesized interpretation for the presence or absence of a blocking effect becomes the much wanted proof for the head/movement type that was only assumed but never independently argued for. That this might be a serious problem is revealed by contradictory assumptions and (hence) predictions made by these theories with regard to the status of heads and their blocking potentials. For example, Terzi (1992) argues that the Greek subjunctive particle na, an M head right above T/Agr, is an A head, although it does not assign case or a theta-role to its Spec (in this she follows Roberts 1991a21 ). Terzi claims that this is the case because it may host PRO in its spec (and pro if the verb is in C), according to her analysis of Greek subjunctive clauses. Besides the fact that it is not clear at all why PRO moves to Spec MP in the first place, if not to receive case or a theta-role (recall that according to the Control Theory she is following, PRO cannot take case, because it must be ungoverned, being+ anaphor; +pronominal- see Chomsky 1981), the claim itself is not uncontroversial given that it is not shared by all researchers on Greek (see e.g. Rivero 1994, Alexiadou & Anagnostopoulou 1998), but more importantly because it is based on a rather theory-internal interpretation of a number of facts. As a result. the postulation that Spec MP is an A position (and hence M is A as well) is not supported by independent evidence. In other words, it is a hypothesis, and therefore it cannot be used 20. Note that more subtle distinctions have been proposed even for a single criterion in the literature. For example according to Terzi (1992), Roberts (1991 a) claims that heads should be divided into A and A-bar, this dichotomy being based on the role morpho-syntactic properties of functional heads have for case and theta-role assignment ofXPs that appear in the Specs of the relevant heads. Chomsky & Lasnik (1993) (and Roberts 1994later on. where he revises his previous analysis), take the L/non-L distinction (viz. the older A/A-bar distinction) to refer only to (abstract) morpho-syntactic features on functional heads that check corresponding features on lexical items. Although, presumably, this change wlll not make any difference for many features, and hence for most heads, stlll there may be consequences for some heads. To give an example, if case is given by Agr and not T, the latter counts as an A-bar head in Roberts's system, while it counts as an AIL-head in Chomsky & Lasnik's system, as well as in Roberts's revised 1994 system. However, this difference would predict that in the former system Agr and T should have disjoint behavior with regard to V-movement, as opposed to the latter systems. Accordingly, while in the former T would block V-to-C movement in imperatives (given that C also does not license case or theta-roles), the same movement would be allowed by T in the latter systems. 21. Note that in essence such a classification simply constitutes the extension of the XP position classification onto the corresponding heads, the specifiers ofwhich are occupied by the relevant XPs. That is, according to this classification an A head counts as A because it may assign case or theta-role to an XP in its Spec, i.e. because its Spec may be occupied by an argument XP. Given the obvious circularity of this definition. it is not clear why we need to also label the relevant heads as A or A-bar, unless of course this pays off in the empirical domain. This is however far from dear, as latridou 1994 also points out.
176
Cities in Greek: A Minimalist Account ofProclisis and Enclisis
as an argument in favor of another hypothesis (namely the blocking effect of M). This is further corroborated by the fact that in other similar, though distinct systems the same head would count as an A-bar head. For example in Rivero's (1994) system, as well as in Rivero & Terzi (1995) the same head would have to be A-bar because if it were an A-head we would expect to find a true imperative along with a postverbal M particle and a clitic. This is so, because in these systems a head is not allowed to cross another head of the same type, while the opposite holds in Terzi's (1992) system. Note that according to Rivero (1994) and Rivero & Terzi (199 5) the arguable semantic properties of M, namely the tact that it is linked to modality (ct~ the indicative/subjunctive distinction in these languages, which is marked via the interaction of C, M and Neg), counts as independent evidence for it's A-bar status, and hence further supports their theoretical account for the blocking effect of M particles. Similar inconsistencies arise with regard to other heads, such as for example Neg, T, and C,2 z and they stem directly from the criteria used for the classification of heads as well as from the more theoretical assumptions regarding the exact mechanism of blocking (cf. e.g. Terzi 1992 vs. Rivero & Terzi 1995 or Roberts 1994). For these reasons I think that as far as the blocking potential of clitics is concerned, the best solution is to assume that the clitic counts as a head only after its incorporation into the extended verbal projection and the resolution of the proclisisenclisis alternation complex. This is possible since the clitic becomes subpart of the label of the extended verbal projection only via incorporation. As a result, one does not 11. Note that T and C involve further complications, in the sense that Tis not normally crossed over by a head. This is so, because at least in finite clauses (but also possibly in (some) non-finite ones) the verb has Tense features to check. Moreover, in tenseless clauses, especially in those languages where nominative depends on tense (as e.g. Greek), there is high chance that T is not projected at all, given that according to standard assumptions in Minimalism radically empty heads (i.e. heads with no features) do not exist (Adam Ledgeway points out to me that even radically empty heads do have features, such as categorical features or features like no Tense). As a result, even though dtlferent systems classify T as A or A-bar, it wouldn't make any difference if the verb has to move via T to a higher head when the latter is projected. As for C, only clitic movement appears to provide an example of a head moving over C; in all other cases C plays a role as a landing site for V. In any case, it is interesting to note that C is consldered to be A-bar by all analyses, albeit for dtlferent reasons: for Roberts (1994) because it is non-L related, for Rivero (1994) and Rivero & Teni (1995) because it contains an operator (note that according to their system C must be L-related, pace Roberts 1994 and Chomsky 1995). As I will discuss later on, their assumption is partially correct, as is Roberts's and Chomsky's assumption), and finally for Teni (1992) because its Spec is not an argument position. What is more interesting is to note that whereas Rivero, Rivero & Teni and Roberts would all predict that clitic climbing across C is not allowed, the opposite would be predicted by Terzi (1992). The irony of this is that some complementizers (like Italian di or Spanish a) allow ditic climbing, whlle others do not (e.g. si/se in Romance). What this shows is that these analyses may make contradictory predictions, depending on their assumptions, but also that in many cases additional assumptions are needed in order for all the data to be accommodated.
Chapter 5. Patterns ofproclisis/endisis and the role ofV-movement 177
need to assume that clitics block any other head since before incorporation they are both minimal and maximal projections. As long as the proclisis-enclisis alternation is resolved at incorporation, the clitic does not block movement of the verb. B) In this section I will show that the V-movement approach makes wrong empirical predictions, and as a result cannot explain without further stipulations the proclisis-enclisis alternation in Pattern A languages. As has already been shown, this analysis makes quite clear predictions with regard to CL-V distribution, namely it predicts that proclisis obtains when the verb moves no higher than the functional head occupied by the clitic, while enclisis tallows when the verb moves higher than the cliticization site. Put differently, V-movement below the cliticization site is a necessary and sufficient condition for proclisis, while V-movement across the cliticization site is a necessary and sufficient condition for enclisis. Unfortunately, this statement is too strong, since it makes wrong predictions. First of all, V-movement across the cliticization site is neither a necessary nor a sufficient condition for enclisis. In particular, it is not a sufficient condition, because there are structures where V-movement across the cliticization site has taken place but what we get is proclisis and not enclisis. Probably, the most striking examples to illustrate this tact are cases involving V-to-C movement. since imperatives are also analyzed as V-to-C structures, however they take enclitics. Such examples involve, among others: (a) Hypothetical se deletion constructions in hypothetical and counterfactual clauses in Standard Italian, including Aux-to-Comp constructions (see Rizzi 1982, Paletta 2000) (b) Subject Clitic Inversion constructions in root interrogatives in Northern Italian Dialects (NIDs) (see Paletta 2000); (c) Subject Clitic Inversion constructions in non-assertive clauses in North-Eastern Italian dialects; (d) V2 clauses in Rhaetoromance varieties (see Paletta 2000); (e) Complementizer deletion constructions in Standard Italian (see Paletta 2000, 2001); (f) possibly, Subject Clitic Inversion and Complex Inversion constructions in French (see Rizzi & Roberts 1989; Roberts 2007); (g) wh-questions in Standard Italian, Spanish or Greek (see Rizzi 1991, 1997, 2001, Tsimpli 1995); (h) non-embedded li questions in Macedonian (see Tomic 1996). Note that some of these constructions are controversial, in the sense that not everyone accepts a V-to-C analysis for them. For example, Subject Clitic Inversion and Complex Inversion in French have been analyzed as involving V-to-C only at LF, the verb targeting IIAgr-Tin the overt syntax (cf. e.g. Hulk 1993, Kayne 1994, Sportiche 1997).23 In a similar manner, wh-questions in many Romance languages (e.g. Spanish, Catalan, Romanian, E. Portuguese) as well as in Greek have been argued to be IP clauses with V-to-I movement (see Drachman & Klidi 1992; Sola 1992; Anagnostopoulou 1994; Barbosa 1995, 2001, 2006; Suiier 1994, 2002, Zubizarreta 1998; Goodall2001; Kotzoglou 2006; and references therein). Note, however, that the V-to-1/AGR-T analysis does not come without its own problems. As the analysis for these structures is not clear, I will not use examples from these constructions. Also note that I assume, following, in this, more or less 23. There is also a remnant movement analysis (see Pollock 2006 and references therein).
178 Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
standard assumptions, that excorporation of the verb or the clitic is not allowed once they have incorporated. This is important since excorporation would make possible an analysis that could account in principle for the proclisis/enclisis alternations. For example, in (24b) below, we could assume that the verb excorporates out of the clitic-verb complex in T and raises to C, followed by the clitic which also moves to C as a proclitic. Such an analysis could explain why C can move higher than the clitic in T and still give rise to proclisis, however it would do so in a rather complex and stipulative way. As a result, excorporation would open up so many possibilities that a principled analysis, which is what 1am looking for, would be hard to reach. For this reason I will not allow excorporation Below, I give examples from (a), (b) and (h): (24) a.
*(Se) Giorgio f
avesse comunicato al direttore, tutto If George it.cl. had communicated to-the director, everything sarebbe a posto. be-cond ok 'If Giorgio had told it (the change) to the director, everything would be ok:
b. (*Se) favesse Giorgio comunicato al direttare, tutta sarebbe a posto. c. (Se) favesse comunicato Giorgio al direttore, tutta sarebbe a posto. (example adapted from Paletta 2000: 116 (her 118)) Italian (Italy) (25) a.
Mi auguro (che) lui le abbia fornite tutte. Italian (Italy) hope (that) he them.cl. has (+sub) provided all
b. Mi auguro (*che) le abbia lui fornite tutte. Italian (Italy) 'I hope (that) he has provided them all? (example adapted from Rizzi 1982: 85 (his (25a)-(26a)) (26) a.
Co I fas-to pa? how it do-you interr.marker? 'How do you do it?'
Pera di Fassa
b. Dime co che tu I fas. Pera di Fassa Tell-me how that you it do 'Tell me how you do c. ..Dime co I fas-to (pa)? Pera di Fassa tell-me how it do-you interr. marker (example taken from Paletta 2000: 46 (her 7a-c))
it:
(27) a.
vide li? Him-ace saw Q 'Did you see him?' Go
b. Petko go vide vcera. Petko him-ace saw yesterday 'Petko saw him yesterdaY.
Macedonian
Macedonian
Chapter 5. Patterns ofproclisis/endisis and the role ofV-movement 179 Donesi mi go! Bring-imp me-dat it-ace 'Bring me it!'
Macedonian
Macedonian d. *Mi go donesi! (example (a-b) taken from Boskovic 2001: 199, 180-83; examples (c-d) taken from Franks 1998: 56)) (24a-c) is an example of an Aux-to-Comp construction with a subjunctive hypothetical clause in Standard Italian (typically used at a very special stylistic level according to Rizzi 1982). It shows that if subj-aux inversion has taken place, the complementizer must be deleted. On the other hand, when no inversion has taken place, the complementizer must be present unless the subject is postverbal (or a null pro), in which case it is optional. The presence of inversion, the ungrammaticality of the overt complementizer as well as the fact that similar structures involve subject-aux inversion in English (cf. Had George mentioned it to Jane, ... ), have been used as arguments in favor of a V-to-C analysis. Similar conclusions can be drawn from (25), which involves the same phenomenon with complement subjunctives (although this case is slightly simpler from that in (24) given that the complementizer is optional independent of the position and the nature of the subject). Now, the interesting part is that in both (24) and (25) the verb either stays lower than the subject, i.e. in T/Agr under standard assumptions, or it raises past it, to C, however in both cases the clitic is a proclitic. I.e. assuming that in (24a/25a) the cliticization site is in some functional head above T (where the verb lands when no inversion has taken place), V/T-to-C movement in (24b/25b) does not have any effect on the clitic position, contrary to what is predicted, as well as against the analysis of imperative clauses. In a similar manner, the Subject Clitic inversion example in (26) from the NID Pera di Fassa, has been analyzed by Paletta (2000) as involving V-to-C movement (as more generally Subject Clitic inversion in these dialects), given that the verb precedes pa, an interrogative marker that can be independently shown to be merged in the Spec of the lowest C head (assuming a split CP domain, in the sense of Rizzi 1997). Arguably, when the interrogative clause is embedded the overt complementizer che blocks V-to-C movement, and as a result subject clitic inversion is ungrammatical. Note that Paletta (2000), based on a number of syntactic and morphological arguments, analyzes subject clitic inversion as an interrogative form of the verb that appears only ifV has moved to C (independently of whether this form is assembled in the syntax or taken ready from the Lexicon). In any case, in the embedded clause the verb is assumed to occupy a head position in I, with the eli tic being merged in a higher I head (cf. proclisis). Again, we observe that V-to-C movement does not leave the clitic behind, yielding enclisis, contrary to the predictions of the V-movement hypothesis. Finally, in (27a) an example of a root interrogative from Macedonian is given. According to the standard and most widely accepted analysis of this construction (see Tomic 1996, Boskovic 2001), I has left-adjoined to interrogative C. which is
180
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
realized by the li-marker. In a declarative clause (see 27b) the verb is in T/Agr and takes a proclitic, the latter being presumably in some higher functional head. (27a-b) together show that V/1-to-C movement in root interrogatives yields proclisis, contrary to the predictions of the V-movement approach (see imperative clauses in 27c-d, which require enclisis, presumably the result ofV-to-C movement). What all the above examples show is that there are cases of V-to-C movement in which the verb moves across the cliticization site (given the CL-V distribution in V-to-T structures in the same language) but still takes proclitics. Now, have a look at the following Aux-to-Comp examples from Standard Italian, and compare them with (24) and (25): (28) a.
b.
Avendolo Gianni visto, ... Having-it John seen, .. . 'Having John seen it. .. :
Italian (Italy)
r: [
averlo lui affermato] non implica che sei fuori The [to-have-it he/him asserted] not implies that you-are out dai guai. of trouble 'His having confirmed it does not imply that you are out of trouble: (example (28b) adapted from Rizzi 1982:85 (his 28a)) Italian (Italy)
These examples show that a non-finite auxiliary (gerund or infinitive) may also move across a preverbal subject in formal styles, but in this case the clitic follows the auxiliary. Presumably, these structures also involve V-to-C movement, given the aux-subject inversion, however in this case we get enclisis (contrary to finite auxiliaries which take proclitics). What is even more interesting is that the same (non-finite) auxiliaries may be found in a lower position, i.e. in Tor even lower, still taking enclitics. This is shown below, where the gerund in (29) appears within Cinque's (1999) lower adverb domain (i.e. in some head between T/Agr and v), while the infinitival in (30) is located higher than lower adverbs, but lower than sentential/higher adverbs (i.e. somewhere around T/Agr- see Belletti 1990 and Cinque 1999 for details; examples (30a-b) are adapted from Belletti 1990:76 (her (107a-b) and (106a/108) respectively), while examples (30c-d) are taken from Belletti 1990: 76 (her (107a) and (106a) respectively) 24: (29) Gianni parla [solitamente leggendol~. John talks [usually reading-them] 'John talks usually by reading his notes: (30) a.
Italian (Italy)
Ho trovato qualcuno a cui (*? spesso) affidarlo (-v spesso). Have-l found someone to whom (often) to-assign-it (often) 'I found somebody to whom often to assign it: Italian (Italy)
24. Examples (30c-d) are used here in order to show that the adverbs in (30a-b) may be used in non-topic (i.e. right dislocated) positions.
Chapter 5. Patterns ofproclisis/endisis and the role ofV-movement
b. Ho trovato qualcuno a cui (-v probabilmente) Have- I found someone to whom (probably) to-assign-it affidarlo (*? probabilmente). (probably) 'I found somebody to whom probably to assign it' Italian (Italy) c.
Ho trovato qualcuno a cui affidar spesso questo tipo di Have- I found someone to whom to-assign often this kind of incarico. task 'I found somebody to whom often to assign this kind of task: Italian (Italy)
d. Ho trovato qualcuno a cui probabihnente affidar questa Have- I found someone to whom probably to-assign this tipo di incarico. kind of task 'I found somebody to whom probably to assign this kind of task.' Italian (Italy) (29) and (30) show that enclisis in (28a-b) cannot be due to V-to-C movement, because in that case we would expect to have proclisis in those examples where the gerund and the infinitival appear lower than C. In other words, gerunds and infinitivals (modulo restructuring verbs) always take enclitics, independent of their position in the clause (in T or C), a fact which cannot be accounted for by the V-movement approach without further assumptions and stipulations. This is in accord with what we saw happening in (finite) subjunctive clauses, which take proclitics independently of whether they appear in Tor in C. Note that the enclitic examples additionally show that V-movement across the cliticization site is not a necessary condition: this follows if clitics target T/ AGR in Italian, while gerunds and infinitivals may target lower heads in the clause. Moreover, the same conclusion is forced by the fact that infinitives may target the same head as finite verbs, as can be illustrated by means of adverb positioning, however the former take enclitics and the latter take proclitics. 25 In particular, Belletti (1990) has pointed out that both finite and non-finite verbs raise to Agr in Italian. For example, both precede VP-adjoined (i.e. lower v P) adverbs like spesso, but they follow Agr P-adjoined (i.e. higher) adverbs like apposta: (31) a.
Gianni apposta mi da spesso i suoi appunti. Italian (Italy) John deliberately me.cl gives often his notes 'Deliberately John often gives me his notes:
25. In French infinitives move lower than main verbs and yet both take proditics. This is expected. though, if the ditic moves to a position that is higher than the one occupied by both main verbs and infinitives (see Kayne 1991, Ledgeway & Lombardi 2005).
181
182
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
b. Ho trovato qualcuno a cui apposta affidarlo spesso. Have-l found someone to whom deliberately to-assign-it (often) 'I found somebody to whom deliberately often to assign it: Italian (Italy) In (47a-b) a finite (lexical) verb and a (lexical) infinitive precede a vP adverb and follow a speaker-oriented adverb. Belletti interprets these facts as evidence in favor of the hypothesis that both verbs move to the same clausal head (namely Agr). If we accept this analysis, along the lines of much of the literature on Italian head movement (see also Cinque 1999, for whom the relevant head will be an INFL head), and assuming that clitics target Agr/T (presumably a head just above Agr/T, given Kayne 1994 and the general spirit of the V-movement approach which treats clitics as targeting a separate functional head), we can observe that V-movement to the same head yields both proclisis and enclisis, depending on the finiteness of the form. Put in other words, enclisis apparently obtains even when the verb moves to a head lower than the clitic position. The same observation can be made for finite and non-finite auxiliaries, which may move across AgrP-adjoined (i.e. higher) adverbs, and in fact above the highest adverb possible, francamente ('frankly'). This is illustrated in ( 32a-b). Again, if we analyze the proclitic structure in (32a) as involving V-movement to a high INFL head, right below the head that has been independently targeted by the clitic, then we are forced to conclude that the same movement yields enclisis in (32b ), which involves an infinitival auxiliary. (32) a.
Ti ho francamente detto la verita. You.cl have- I frankly said the truth. 'I have frankly told you the truth:
Italian (Italy)
averLe francamente annunciate b. Mi mette in imbarazzo, announced Me.cl puts in embarrassment, to-have-it.cl frankly una tale notizia. a such news. 'I am sorry to frankly have announced you such news: Italian (Italy) Similar examples which challenge the V-movement approach can be constructed on the basis of Greek data. For example, Greek finite (lexical) verbs may appear in a range of positions above vP, and in fact, they may raise (at least) above tense adverbs like cht(h)es ('yesterday') or kapote ('once') (i.e. they move to Agr, in the old-style Pollock 1989 system- see Alexiadou 1997). However, like Italian lexical verbs, they may also appear within the 'lower-vP adverbs' area of Cinque (1999), since they may stay at least as low as aspectual adverbs like schedhon ('almost') or idhi ('already'). (33) a.
To edhose chthes o Jianis to vivlio stin Alexandra ltd Gave-3sg yesterday the John-nom the book-ace to-the Alexandra 'John yesterday gave the book to Alexandra: Greek
Chapter 5. Patterns ofproclisis/endisis and the role ofV-movement 183
b. Ta kerdhi tu chthes schedhon/idhi ta triplasiase The winnings-ace his.cl yesterday almost/already them.cl tripled-3sg o Jianis. the John-nom 'John yesterday almost tripled/already tripled his winnings: Greek c.
Ta kerdhi tu ta triplasiase chthes schedhon/idhi o Jianis.
Greek
The postverbal nominative subject is in [Spec,vP], following standard assumptions in the Greek literature (see Alexiadou & Anagnostopoulou 1998 and reterences therein). Chthes is in T (higher adverb area)26 while schedon/idhi are above v but below T (presumably assodated with some Aspectual head). We observe that V-movement does not appear to correlate with the position of the eli tic: in both the lower and the higher V-position the clitic appears as a proclitic. Assuming that in the lower position the clitic targets a functional head right above V, V-movement across this head to a higher position does not yield enclisis, contrary to what is expected: (34) a. *Ta kerdhi tu triplasiase ta chtes schedhon/idhi o Jianis. b. *Ta kerdhi tu triplasiase chtes schedhon/idhi ta o Jianis. In (34a-b) we can see that the verb cannot pick up the clitic as an enclitic on its way to T, or simply cross over it, leaving it in some lower aspectual head. One way of accounting for the lack of enclisis would be to assume that the clitic targets the Agr position, while the verb can stay in the low aspectual position. However, this would predict that interpolation would be possible, which is certainly not true in Greek: (35) *Ta kerdhi tu ta chtes schedon triplasiase o Jianis.
Greek
This shows that the clitic must already be targeting some lower head between T and v. As a result, the prediction of the V-movement approach (i.e. enclisis) is not met. Rather, it seems that, like the proclitic V-to-C cases, also here the clitic piggy-backs on the moving V (presumably via some kind of incorporation), without inducing any change in the established CL-V ordering. This would immediately account for the lack of enclisis despite V-movement across the cliticization site. A similar situation also arises with French (auxiliary) infinitives, which must move to an intermediate position outside vP, but which may optionally move across the negation adverb pas to a higher position (presumably T, or some low INFL head), taking the clitic along as a proclitic (see Pollock 1989, Kayne 1991, and Zanuttini 1997): (36) a.
ne pas f avoir mange. neg not it.cl have-inf. eaten
French (France)
26. Elena Anagnostopoulou asks whether chtes is truly so high. She points out that yesterday and tomorrow seem to be low In English, like adverblals such as tin Tetarti ('on Wednesday') In Greek. Even if chtes is not in T, the point here is that it lies higher than schedhon and idhi, which is enough for the purposes of the argument.
184 Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
French (France) French (France) French (France)
b. ne favoir pas mange. c. *n' avoir pas le mange. d. *ne le pas avoir mange.27
(36a-c) show that V-movement across the clitic and pas does not yield enclisis, as expected. On the other hand, (36d) clearly shows that the clitic cannot target the position above pas independently; as a result, one cannot argue that in (36b) the clitic has targeted this particular position. More examples can be constructed from other structures. For example Italian gerunds may stay low in the clause (i.e. below Agr/T but above v) or they may move to higher positions (e.g. in or around Agr/T), on a par with lexical finite verbs, although the former take enclitics in each possible position, while the latter take proclitics. Or, Italian finite auxiliary verbs may move to the highest INFL head in Cinque's (1999) hierarchy, exactly as absolute past partidples (see Belletti 1990) (which must move to this position) or infinitival verbs in subject and complement infinitives (see Cinque 1999 for details), however, while the finite auxiliaries take proclitics, both gerunds and infinitivals take enclitics, even though all of these forms target the same V-position. Of course, the proclitic vs. enclitic ordering of finite auxiliaries and infinitivals (which may also occupy lower positions) does not change, again confirming the observation that the linear ordering of the elitic with regard to its host typically remains constant within a language, no matter where it lies in the clause (although this might be too strong - see later on the discussion on surrogate morphological imperatives with imperative (i.e. non-surrogate) syntax). All in all, although the V-movementapproach presumably captures the difference between indicatives/ subjunctives and gerunds or imperatives, it cannot account equally straightforwardly for other constructions. There are two apparent possible ways to 'save this approach: either to try to show that V-movement across the cliticization site has not taken place in cases of unexpected proclisis, or that V-movement targets distinct heads in those cases where V-movement to the same head gives both proclisis and enclisis. This is illustrated below: (37) a. X ... CL...V
~ b.
vs.
X... CL...F. ..V
tJ
I
x ... cL...H ...v
tJ
vs.
X... G... CL...H ...V
~
I
Following (37a), one could assume that e.g. in those cases ofV-to-C movement with proclitics the verb does not really move to C, but to a lower head below the clitic position. This is in fact the standard explanation of the proponents of the V-movement approach (see e.g. Kayne 1994 on subject clitic inversion and (positive) imperatives in French). Although this might work for some structures, for which there is not enough 27. This kind of order (not really with pas) ls marginallliterary (see Pollock 2006) (Ian Roberts p.c.).
Chapter 5. Patterns ofproclisis/endisis and the role ofV-movement 185 evidence, or there is conflicting evidence, it wouldn't possibly work for those cases which do not have this problem. This would hold both for V-to-C movement structures as well as for V-to-F structures (where F stands for a lower functional head (e.g. T) that is located higher than the clitic position to which a verb may adjoin). As for (37b ), although such a solution would be plausible in principle, it would leave the adverb distribution facts unaccounted for (this would be the case both for a Cinque as well as for a Belletti 1990/Pollock 1989 type of approach to adverb placement, as cases with such high adverbs as francamente obviously show us), and it would also face problems with the tact that gerunds and infinitives may occupy many distinct positions in the clause: the crossing variation of (37b) in a low position would entail that the clitic should remain in that position every time the verb moves to a higher head. However, this type of interpolation does not appear to be possible in Romance languages (see e.g. Rizzi 1993, Rooryck 1992, Shlonsky 2004), a fact which would complicate this analysis considerably. The second way to deal with this issue is to assume multiple cliticization sites on a par with multiple V-sites. Accordingly, for the proclitic Greek cases mentioned in (33) one could argue that there are at least two cliticization sites in the clause, one in the lower aspectual area, and another one in the higher Agr/T area. An additional cliticization site in the C-domain, right above the moved interrogative V-head would also solve the problem of proclisis in root interrogatives. However, this approach faces a number of problems. First of all, it is not clear how the two cliticization sites are created. Assuming that a clitic moves to the lower cliticization site as the result of an Agree/Move operation, it is not straighttorward why the same clitic would synta.ctically move to the higher cliticization site, given that it can perfectly survive in the lower position (as is obvious from the facts). Second, the movement of the clitic to the lower or the higher head appears to be dependent on V-movement in some respects, although not in others, however it does not follow from the V-movement approach why this is the case. In particular, for Greek clitic movement to the higher head entails V-movement to (a head in) the same area, and vice versa. For those languages which allow interpolation, it appears that clitic movement to a higher head does not necessarily correlate with V-movement to the same area, although it is possible. Finally, no language appears to have V-movement to the higher head without clitic movement (cf. the lack of interpolation in enclisis ). These possibilities are illustrated below (where 1, 2 stand for different vaguely defined domains such as T or v/ Aspect): (38) a.
H 1a-CL... H 1b-V ... (H 28 -CL) ... (H2b-V) (Greek/languages allowing interpolation)
b.
H 1a ••• H 1b••• H 2a-CL ... H2b-V
(Greek/languages allowing interpolation)
c.
H 1a -CL... H 1b••• (H2a -CL) ... H2b-V
(languages allowing interpolation)
d. *H1a ••• H 1b-V... H2a-CL ... (H2b-V)
(unattested)
186
Cities in Greek: A Minimalist Account ofProdisis and Endisis The problem here is that it is not clear why (d) is (presumably) not found in any language.28 Put differently, it is not clear why H 1b cannot be activated unless H 18 is also activated, although the opposite is possible. 29 It seems that simple movement of V across CL- H cannot be the whole story, at least for those structures involving low and high enclisis. For these cases, the standard assumption in the literature is head adjunction/incorporation: the verb head adjoins to the clitic (or the head containing the clitic-depending on one's assumptions), picking it up on its way to a higher head. This is something the standard V-movement approach does not discuss very explicitly. Moreover, it is not clear why (c) is allowed only in some languages but not in others. A third problem is that it is not straightforward why proclisis is generally not possible with imperatives/gerunds, or enclisis with indicatives and subjunctives.30 I.e. once we assume the existence of multiple clitic sites, it is easy to imagine a situation where an imperative verb moves to C and a higher cliticization site gets activated at the same time. Given that all lower sites typically involve a higher clitic head (i.e. a proclitic), a similar head in the C-domain would also be possible.31 Alternatively, V-to-C movement in questions could give enclisis if no other cliticization site above the one in the TI Agr area were activated (here I put aside the issue of interpolation - presumably it is banned from enclitic constructions tor some reason, however this would not rule out the possibility of having enclisis in interrogatives, in the same way we do with imperatives). Possibly, we could argue that whenever a V-head is activated in some part of the clause, a corresponding clitic-head must also be activated, forcing clitic movement to this higher head. Although again it is not clear why this would be the case, or why all Pattern A languages appear to take the same option as far as this construction is concerned (c.f. the inexistence of enclitic object pronouns in interrogatives), at least it is in accord with the more general patterns given in (38) above.
18. In fact this may be too strong. As Adam Ledgeway has pointed out to me, Borgomanero is a north-western Itallan dialect that has been described and analysed in a number ofwell-known published studies by Christina Tortora (cf for example Tortora 2002) and which allows enclisis with interpolation. For this reason we should be talking about a rare pattern.
19. Note that another plausible way of viewing the patterns illustrated in (38) is to assume a spell-out copy approach, according to which both CL and V move to the higher heads, however PF may decide on independent grounds which combinations of copies may be pronounced (H-H, H-L, L-L, L-H). Putting further details aside, the questionstlll remains, although set in one of the interfaces (i.e. PF instead of syntax): why is the f\.- Lc1 combination not grammatical? Presumably, it ca.tft be due to PF properties of the ditics, given that they are not clause-initially, and they do not have to be adjacent to the verb, at least in those languages allowing for interpolation. 30. Prodisis is possible with French gerunds, while endisis is possible with Italian indicatives and Spanish subjunctives when used as imperatives. 31· One way out of this problem would be to assume that enclitics in imperatives and gerunds indeed involve clitic-heads in the C-dom.ain, however in all these cases V further moves to the highest possible C-head, yielding endisis.
Chapter 5. Patterns ofproclisis/endisis and the role ofV-movement 187 Although the multiple clitic site approach does not appear to solve all problems of the V-movement analysis, it nevertheless highlights a basic cross-linguistic fact that seeks for an independent explanation: namely, for the fact that clitics may attach higher or lower in the clause, both across languages and within the same language. This has already been pointed out by Kayne (1991), who analyzes French infinitivals as involving V- and clitic movement to a lower head, in comparison with languages such as Italian, which have V- and clitic movement to the (higher) T head. In a similar manner, Ledgeway & Lombardi (2005) offer a very interesting study of the interaction of clitic and V-movement in a number of Romance languages. They show how languages differ in the positions targeted by clitics and their verbal host, as well as how in some languages, namely those having interpolation, the clitic (and the verb) may target more than one position within some part of the clause. The latter (and more generally, the distribution of clitics with regard to their host) is defined by means of adverbs, in the spirit of Cinque (1999), where adverbs are argued to correlate with (semantically) corresponding syntactic heads, offering us in this way a full exposition of the clausal structure (the cartography approach). Following this line of analysis, they argue that in Standard Italian the clitic targets a low head position just above vP, as opposed to some Southern Italian Dialects (SIDs), where the clitic may attach to a number of distinct, though higher head positions. Furthermore, they analyze cases like the one illustrated in (33) for Greek (i.e. cases of constant CL-V ordering independent of the position of the CL-V cluster within the clause) as involving elitic incorporation (i.e. syntactic head adjunction) to the verbal host, accounting in this way for the constant ordering within the cluster as well as for the lack of interpolation. As a result, they do not assume distinct clitic sites/positions for these cases; rather, they assume that cliticization is syntactic, at least in these cases, and that the higher copy involves a morpho-syntactic cluster, rather than two distinct (V and CL) positions/heads. As for the interpolation cases, they analyze them in the following way: they show, based on their distribution with regards to adverbs, that clitics may target a number of positions within a designated portion of the clause, while verbs can also be located in a number of positions which are included in the particular portion, but which may also involve higher heads. In the latter case the clitic moves along with the verb to higher parts of the clause, and no interpolation is possible. This is illustrated below (assuming a Kaynean base-generation and movement approach): (39) a. [adv2 d adv 1 V [(V) (cl)]] (clitic moves to a position higher than V) b. *[cl [adv2 adv1 V [(V) (d)]]] (clitic moves to a position higher than a cer-
tain adverb) c. [adv2 d V adv 1 (V) [(V) (cl)]] (V follows the clitic to the higher position) d. *[d [adv2 (cl) V adv1 (V) [(V) (cl)]]] (the clitic strands Vat the higher position) e. .Y[cl V [adv2 (cl) (V) adv1 (V) [(V) (cl)])] (the clitic moves along with V across adv2.
188
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
f. [adv1 cl V [(V) (cl))] (cl and V stay lower) g. *[[adv2 V adv1 cl (V) [(V) (d)]]] (V moves to adv2 stranding the clitic in advl) The clitic may move in a number of positions up to the head hosting Adv2, above which it cannot appear on its own. If the verb stays lower, interpolation arises. If it moves to the same head position (Adv2), or if both CL and V stay in the lower Advl position, no interpolation arises. If the verb moves above Adv2 it takes the clitic along, presumably via syntactic incorporation. Finally, the verb cannot strand the clitic in the advl position. Summing up, according to this analysis multiple clitic sites are only present in languages which allow interpolation. Languages with low cliticization or languages with high cliticization (i.e. cliticization toT- e.g. French) only involve one additional elitic head besides the base clitic position (the latter not necessarily being a head). Moreover, cases like the one in (39d) are ungrammaticaL presumably because the clitic must always attach to the verb as long as they appear on the same head (cf. Adv2). On the other hand, it is not clear why (39g) is ungrammatical. Presumably, the verb cannot strand the clitic when it moves to a higher position, although apparently the clitic may do that (cf. 39f) (See Chapter 4 for a possible explanation). Of course, a number of questions arise, such as e.g. whether syntactic cliticization takes place in (f), and if not why not (cf. that Ledgeway & Lombardi argue that (f) involves PF cliticization only, in order to explain why the clitic may appear above Advl on its own), or why only the clitic may strand the verb, but not vice versa (cf. 39g). Moreover, given that Ledgeway & Lombardi assume that enclisis is a morphological rather than a syntactic phenomenon (i.e. enclitics are already attached on the verb from the Lexicon), it is obvious that their analysis does not have anything interesting to say with regard to enclitic structures, the role ofV-movement, finiteness or multiple clitic sites. Finally, no explanation is given as to why some languages have more than one potential clitic site, and what drives selective movement to these sites in each case (recall that this was also a problem for the V-movement approach). Whether these or other problems can be accounted for in an adequate manner is a separate question that is out of the scope of this discussion. What is more important is the following: (a) at least some languages have more than one cliticization site; (b) it seems that clitic pronouns indeed differ from particles, since they appear to be able to syntactically attach to another head (cf. the fact that CL + V must move as one unit above the position marked by Adv2).32 This can explain to some extent the observation that in those 32. Elena Anagnostopoulou points out to me that 8a/na + V can also occur before or after adverbs like pitha,w11, profanos, idhi, on a par with clitic pronouns. This suggests, again, that these particles might be part of the verbal head already from the root level Alternatively, and more in agreement with the analysis of enclisis in chapter 6, one could claim that particles merge in the left periphery of the clause and that verbal heads contain unintepretable features that they must check against the particles in the syntax. The PF chooses which copy of the verb +particlefeature complex to spell out, giving rise to cl-particle-adverb and adverb- cl-particle series.
Chapter 5. Patterns ofproclisis/endisis and the role ofV-movement 189 languages which do not allow interpolation, the CL-V/V-CL cluster retains its internal ordering independent of the scope of V-movement, a fact that brings us back to the importance of the particular verbal form (in the sense that a particular form takes proclitics or enclitics, independent of the position of the verb and the clitic in the clausal structure), and as a result (only indirectly) to the notion of finiteness. On the other hand, we still need to explain the possibility of multiple sites as well as the strong intuition that, at least for some structures, V-movement seems to play a big part (e.g. imperatives, gerunds, or more generally those cases where a low V-CL attachment approach would face more problems), although not for others (cf. V-to-C movement with proclitics). So, let us see the general picture that emerges from the discussion above with regard to the proclisis-enclisis alternation and the success of a V-movement approach to account for the alternation. (i) Verb movement across the cliticization site is neither a necessary nor a sufficient condition for enclisis, and as a result it cannot account for the proclisis-enclisis alternation in an adequate way, on a par with finiteness, without further stipulations and assumptions. However, in a quite striking manner, in many cases where V-movement makes the wrong predictions, finiteness makes the right predictions (as for example with gerunds/infinitivals vs. subjunctives in Aux-to-Comp constructions in Italian), while on the other hand there are also cases where finiteness makes the wrong predictions while V-movement makes the right ones (as for examplewith Spanish and Italian suppletive imperatives with imperative syntax). (ii) It appears that V-movement to the cliticization site is a necessary condition for enclisis (cf. Rizzi 1993). This is so, because there doesn't seem to be a case where a verb has stayed lower than the cliticization site but enclisis obtains. This is particularly clear in languages allowing interpolation. E.g. in Cosentino, a SID, in non-finite clauses the elitic always targets a position that is higher than the one targeted by the infinitive (i.e. they never end up in the same head position), and only proclisis is possible for all infinitival clauses (see Ledgeway & Lombardi 2005 for more details). (iii) Given (i) and (ii), as well as the interpolation/multiple clitic site facts, which suggest that at least proclitics attach syntactically to their host (via incorporation/ head adjunction) when they target the same head with the verb, not allowing further interpolation or changes in the internal ordering of the CL-V cluster, in combination with the well known observation that enclitics do not allow interpolation or changes in the internal V-CL ordering (even in cases where cliticization takes place in a head lower than T), it seems plausible to assume that the choice of a value for the alternation takes place locally, i.e. when the clitic and the V-host target the same head. This may happen earlier or later in the derivation, depending on which heads are independently targeted by the clitic and the verb (as evidenced from interpolation). It seems, though, that targeting of the same head by both the clitic and the verb does not always result in head adjunction and
190
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls subsequently in resolution of the alternation. However, there is always a particular head which is eventually targeted by both the clitic and the verb, and after which no further independent movement of the verb or the clitic is allowed. Assuming that this is an epiphenomenon of head adjunction or incorporation (see Chapter 4), it follows that the latter may only take place at a designated head, presumably the phase head, as we have argued for in Chapter 4. The data from interpolation suggest that this is always the highest possible clitic-related functional head from the row of possible clitic sites. This in turn suggests that syntactically speaking the head that triggers the operation is that particular highest head, given that incorporation normally involves the phase head (which, under standard assumptions, is the highest head in the phase), or the highest head containing the relevant features. Moreover, the resolution of the alternation may only take place at this particular head, i.e. at the moment of incorporation. This appears to be the case e.g. in literary French (see Kayne 1991), where the clitic and the verb do not attach syntactically if they target a head lower than bien, but they do attach when they target a higher head (lower than pas), yielding proclisis. Higher syntactic heads may only see the complex head (so that any further raising involves movement of the whole cluster). Although this is the layout of a more general approach, and although it faces a number of questions, similar to those of the V-movement approach or the multiple clitic site approach, 33 it links the proclisis-enclisis alternation partly to the morpho-syntactic operation of incorporation. As a result, it emerges that both V-movement and the properties of the morpho-syntactic relation between the clitic and its verbal host play a role in the alternation. One step further would be to assume that V-movement is responsible for cl-V adjacency, while incorporation, and subsequently the related morpho-syntactic properties of the incorporation site which regulate the operation, is responsible for the outcome of the alternation. Since at least in some cases finiteness appears to correlate with this outcome, a further hypothesis would be to assume that in fact finiteness is the factor regulating the outcome of incorporation (as proclisis vs. enclisis) locally (i.e. at the cliticization site), even in those cases which seem to pose a problem. This is in essence what researchers like Rizzi (1993) or Kayne (1991) have independently argued for. I will discuss their analyses later on, when I present my own theory on the role of finiteness in the alternation. Of course, we have to bear in mind that such a move, namely to consider finiteness as the factor affecting the alternation at the incorporation site, would also bring back the problems that are inherently related to the notion of finiteness, such as e.g. the level of finiteness that would be
33· For example, it ls not clear how the Intermediate cliticization sites can be catered for. I.e. one must argue whether these extra sites are the result of feature checking (e.g. EPP/person, following Chomsky 2005), or simply the result of successive-cyclic movement (see Bo8kovic 2007). In any case, I assume that successive-cyclic movement of ditics involves XP movement, and not X movement.
Chapter 5. Patterns ofproclisis/endisis and the role ofV-movement responsible for the alternation. or the exact properties of finiteness (e.g. abstract or morphological features, depending on the chosen level of finiteness). For example, one would have to ask questions like: 'Why should the placementofclitics (w.r.L their verbal host), which are object phi-feature morphemes, correlate with the presence-absence of features on the T/Agr note (following the standard assumption that finiteness is directly related to this node)'? Or: 'Which features are relevant for the alternation, and why is it only that certain types or sub-types of features are relevant? Are the same features relevant in all languages with a finiteness-sensitive alternation. and if yes/no why?'. Despite all these issues, it is fairly obvious that V-movement alone cannot account for the data, so we'll have to answer these questions in order to be able to use both finiteness and V-movement as regulating factors for the alternation. ln the following chapter I will look into the specification/definition of'finiteness: and how it interacts with V-movement in order to yield proclisis or enclisis. To do that, I will discuss a range of Greek constructions which appear to form a gradation of (both morphological and syntactic) finiteness. The main empirical observation will be that enclisis obtains when a person feature is not present on the subject agreement morpheme. Building on Kayne's (1991) intuition that non-finiteness creates an empty slot for the clitic, as well as on Rizzi's (1993) intuition that syntactic checking in terms of features is crucial for the alternation to take place, I will argue what is needed to capture the right results is a combination of a relativized minimality approach based on features with a non-1:1 mapping between syntax and morphology (as is e.g. the Late Insertion theory of Distributed Morphology). Accordingly, it will be shown that enclisis in Greek and other languages with a similar pattern is the result of the interaction of both syntactic and morphologicaVPF factors at the head level. This is strongly reminiscent of a similar interaction at the XP level in languages with Wackernagel or Tobler-Mussafia clitics.
5.6 Conclusions In this chapter I discussed the proclisis-enclisis alternation in Greek and other Pattern A languages and the role of finiteness and V-movement in the relevant patterns. It was argued that although finiteness appears to be a problematic notion it nevertheless seems to play an important role in the alternation, and in particular at the resolution site of the alternation, namely the cliticization site. This role needs to be further investigated, which will be the issue of the following chapter. Moreover, it was shown that movement across the cliticization site is neither a necessary nor a sufficient condition for enclisis; rather, it is movement to the cliticization site that is a necessary condition for enclisis.
191
CHAPTER6
Enclisis in Greek The role of CM in deriving enclisis
6.1 Introduction
In this chapter I will discuss the role of finiteness, and in particular, of CM(odal) (which corresponds to Rizzi's 1997 CFin) in the proclisis-enclisis alternation. More specifically, I will investigate the finiteness properties of a series of mood structures in Greek, and I will formulate the following generalization: "Proclisis correlates with non-restricted/full person agreement on T, while enclisis correlates with restricted person agreement on T. Non-restricted agreementis associated with the [-imperative] paradigm and a non-inflectional CM head (and, thus, with non-obligatory movement to C), while restricted agreement is associated with the[+ imperative] paradigm and an inflectional CM head (and, thus, with obligatory movement to C)". In other words, for enclisis to obtain, the verb must move to CM, a V-related site, across the initial cliticization site (v*/T), while the clitic incorporates into the edge of CM. At the end of the chapter I also discuss negated imperatives, surrogate imperatives as well as the socalled clitic switch in Greek imperatives/gerunds, namely the free ordering of dative and accusative clitics in enclitic orders.
6.2 The finiteness gradation: From indicatives to gerunds/absolutives
In this section I will investigate the finiteness properties of a number of Greek structures with distinct mood properties, especially in relation to the proclisis-enclisis alternation. Greek is a rather appropriate language in this respect for two reasons: (a) because it is a morphologically rich language, it allows us to test the relation between abstract (syntactic/semantic) and morphological finiteness, as well as their relevance to the proclisis-enclisis alternation; (b) because it has tense-less but fully marked for agreement subjunctive clauses (instead of infinitives), it gives us the opportunity to investigate the tense vs. agreement distinction, which it is not possible to test in Romance languages for independent reasons (modulo Romanian, and possibly Tensed Infinitivals in Spanish (see also Ledgeway 1998)). The main empirical observations will be: (i) that enclisis in Greek obtains when person is not marked on the subject agreement morpheme; (ii) that the final position of the verb in Greek correlates with the agreement and mood properties of the verb and the semantic/pragmatic properties on
194 Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
C (which are in turn related to syntactic EPP features on C-T). 1 It will be also shown that V-movement to the clitidzation site is a necessary condition for enclisis, confirming what was discussed above. Finally, I will very briefly discuss Greek interrogatives and the issue of whether they involve V-to-C movement and how the answer to this question affects our claim about the role ofV-to-C movement in the alternation. 6.2.1 Some preliminary remarks on verbal marking
in Greek
Here I give a brief overview of the morphological properties of the Greek verb, which will be useful in the following discussion. The facts presented here are based on Holton, Mackridge & Philippaki-Warburton (1997) and Roussou (2006). Greek verbs may be marked tor: (a) voice (± active); (b) aspect (± perfective); (c) mood(± imperative); (d) agreement(± person, number); and (e) tense(± past). Aspect is marked on the stem only in the perfective value. Agreement is marked on the ending, while tense is marked via different agreement endings (so that the latter mark both agreement and tense, although not in the +imperative paradigm), as well as by stress shift (or addition of a prefix e - (traditionally called 'augment>) that carries the stress in case ofbisyllabic forms) in the +past forms. Finally, mood on the actual verbal form only marks the basic paradigmatic opposition between + imperative and - imperative forms. The latter are taken to form the indicative paradigm. Traditional mood and other distinctions (such as for example subjunctive-indicative, realis-irrealis, future-non-future, etc.) are marked by the presence-absence of special particles added to the -imperative/indicative form (which appears in different featuralcombinations, i.e.+ past/-past, etc.). In purely morphological terms, traditional finiteness properties such as agreement and tense interact with mood, in that distinct mood types involve distinct agreement/ tense combinations and endings. In this sense, they appear to form a system on their own, where particular values in one dimension correlate with particular values in the other dimensions. On the other hand, voice and aspect are not part of this system, given that they do not correlate with any changes in the Agreement/Tense/Mood system (in fact, as gerunds show, they are totally independent from these properties). Thus, different moods may be organized into a preliminary finiteness hierarchy, based on marking of morphological distinctions. Accordingly, indicative forms, which inflect for (voice; aspect; mood; agreement; tense) are more marked than subjunctive forms, which inflect for (voice; aspect; mood; agreement; and possibly for tense), which in turn are more marked than (positive) imperatives, which inflect for (voice; aspect; mood; (partial) agreement, but not for tense). Finally the latter are more
Note that something should be said about the fuct that verbs appear to move optionally across certain types of adverbs, disregarding in this way the particular properties on the verb or on C. One way out would be to assume that optional.ity arises only at a level at which the agreement and mood properties of the verb as well as the semantic/pragmatic properties on C have already been satisfied
1.
Chapter 6. Enclisis in Greek marked than gerunds, which do not inflect for mood, tense or agreement (instead they take a default ending), but only for voice and aspect. but which take enclitics like imperatives.2 & proclisis shifts to enclisis at the transition from the subjunctive to the imperative, further investigation of this hierarchy could potentially assist us in the investigation of the role of finiteness in the proclisis-enclisis alternation. 6.2.2
Mood types and their finiteness properties
Indicative forms Indicative forms inflect for all verbal categories. Their marking remains constant for mood (always -imperative with no other distinctions), although they inflect for tense, agreement, voice and aspect. This is illustrated below with the verb trexo ('to run') (adapted from Roussou 2006) 6.2.2.1
(1) -imperative//± past of verb trexo a. trex-o, trex-is, trex-i, trex-ume, trex-ete, trex-un stem- tense/agr ending ( lsg,- past, etc.) (- imperative; - past; + number/ person; - perfective; + active) b. trek-s-o, trek-s-is, trek-s-i, trek-s-ume, trek-s-ete, trek-s-un stem- perfective marker3- tense/agr ending (lsg, -past, etc.) (- imperative; - past; +number/person; +perfective; +active) c. e-trex -a, e-trex-es, e- trex-e, (e)-trex-arne, (e)-trex -ate, e-trex -an augment (stress carrier) - stem- tense/agreement ending (lsg, +past, etc.) (- imperative; +past;+ number/person;- perfective; +active) d. e-trek-s-a, e-trek-s-es, e-trek-s-e, (e)-trek-s-ame, (e)-trek-s-ate, e-trek-s-an augment(stresscarrier)-stem-perfectivemarker-tense/agrending(lsg+ past, etc.) (- imperative; + past; + number/person; + pertective; + active) [- past + perfective] forms are traditionally called 'dependent forms: since they cannot appear on their own but must be preceded instead by one of the particles na, 8a4 and as. 5 Na is analyzed by most researchers as the subjunctive marker (e.g. Veloudis & 1. Adam Ledgeway asks how participles fit into the finiteness hlerarchy. These are marked for agreement, aspect and voice, however they do not accept (post-verbal) ditics (possibly, due to the fact that they are in the passive/middle voice), therefore they are not going to be discussed here. 3- Presumably, the outer aspect opposltioo ls marked via the pair zero(- perfective)/s (+ perfect.Lve). 4· 8a is the only word in this monograph where the sound /th/ will be presented with the Greek symbol theta 8. 5· Note that typically the two negation particles mi(n) and dhen cannot support the dependent form on their own - the presence of one of the other particles is required. There is only one
195
196
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
Philippaki- Warburton 1983) (cf. that Greek has lost the relevant marking on the verbal form itself- see Philippaki-Warburton & Spyropoulos 2004), on a par with Romanian sa (see e.g. Terzi 1992), although it also appears to express different types of modality. In this respect, na realizes syntactically the semantic import of subjunctive (see Giannakidou 2007). 8a is used along with the- imperative form of the verb in order to form the periphrastic future construction, although it is also used with a modal rather than a tense interpretation. As Roussou (2000) points out, this appears to be the case for all three preverbal particles, which yield a range of different modal readings (such as countertactual readings, epistemic modality readings) by combining with different values for tense and aspect of the subcategorized verbal form. Finally, as is the hortative marker, found only in root contexts. In the literature there have been several analyses of these preverbal particles and their position in the clause with regard to negation, clitics, the verb and complementizers of various sorts (for an overview see Giannakidou 1998, 2007; Iatridou 2000; Philippaki-Warburton 1998; Roussou 2000; 2006; Philippaki-Warburton & Spyropoulos 2004). Here, I will adopt the general clausal structure that is proposed in Roussou (2000) for two basic reasons. 6 First of all, because I think that it can account for the syntactic and semantic properties of the preverbal domain in Greek in an adequate manner, and in fact more adequately than other structures that have been proposed in the past Second, because one of its central points is that it makes a distinction between a modality position and a clause typing position, a distinction which will prove to be crucial in the proclisis-enclisis alternation. In particular, Roussou proposes, adopting and adapting Rizzi (1997), that Greek clauses have an extended CP domain, which can be split into three basic C positions: the highest C head hosts pure subordinators with no modal properties like pu. The middle Cis called COp (see Manzini & Savoia 1999). It is a clause typing head roughly corresponding to Rizzi's ( 1997) CForce head, and hosts proper complementizers, that is heads that encode operator teatures relating to clause typing (and which can determine the illocu tionary force of the clause - in this respect force is a pragmatic notion (see Roussou 2000, 2006)). Finally, the lowest C head, which is merged right above the highest INFL head is called CM (M stands for modal), and hosts modal particles that interact with properties realized in the INFL domain (such as tense/agreement and aspect) in order to give different modal readings. CM roughly corresponds to Rizzi's CFin head, which is argued to carry information
case where mi(n) can support the dependent form on its own, namely when used with a negative imperative meaning, but it must be restricted to 2nd person sg/pl: (i) min to kanis! vs. min to kanete! vs. *(na) min to kani! ('not itd do-2sgldep.form vs. not itcl do-2pl/dep.form vs. (subj) not it.cl do-3sg) 6. See also Alexiadou 1997, Giannakidou 1998, 2007 and Phillppaki-Warburton 1998 for alternative structures.
Chapter 6. Enclisis in Greek 197
regarding finiteness? Negation is merged between CM and COp, while Focus/Topic is realized right above COp and below the highest C. Finally, clitics and the V-form are in I, following all C heads. The overall structure is given below: (2)
[cpu [Topic /Focus
leap oti!an!na/as [ NegP dhenlmin [ CM 8a/tnalas[Icl+V... ]]])]] Roussou (2000: (19))
According to (2) the similarities between 8a and na (cf. their modal readings in combination with INFL, common morphological and syntactic properties) are accounted for by both particles being generated in CM. Na and as differ from 8a in that they move across the Neg head to COp,8 capturing in this way their common properties as well as their complementary distribution with complementizers proper like oti (indicative) and an (interrogative) (cf. for example the tact that they may introduce root clauses), and their differences with ea. Finally, oti/ an (but not na/as) may also (optionally) move to C, provided puis not realized, which captures why they cannot co-occur.9 The overall feature composition of preverbal heads, as specified in Roussou (2000) depending on their positional, semantic and syntactic properties, is as follows: (3) a. 8a: modal b. na., as: modal, clause-typing c. oti, an: clause-typing, (subordinator) d. pu: subordinator
(Roussou 2000: 79)
Phillipaki-Warburton (1992, 1998) and Giannakidou (1998) propose a slightly different clausal architecture: (4)
[ep oti/na [MoodP tn/zero (indica.tive)fimperative [NegP dhenfmin AgrP
[Now-TP
8a [T
cl-V [yp ... ]]]]]]
7· The main idea behind Rizzi's proposal is that finiteness appears to be associated with a nwnber ofinfiectional properties typically encoded on the INPL head (and marked on the verbal form), but which may be mirrored (though in a rough manner) by the CFin head. For some researchers (see e.g. Adger 2007) this relation between CFin and INFL can be encoded via a ± finite feature on the former.
According to Roussou (2006), when the negation particle mi(n) is present in the clause na is directly merged in COp (i.e. it does not move there). The same would apply to as. Mi(n) on the other hand first merges with CM and then moves to Neg, which according to her captures the fact that negation is not compatible with imperatives. The relevant structure is illustrated below:
8.
(i) lop na [Neg min [M \n1n [CL to lr dhis ...]]]]] 9· Note that Roussou's analysis does not come without problems. For example. the traditional intuition is that negation is selected by different modality/mood markers (for example. dhen is selected by indicative mood while min is selected by subjunctive). It is not dear how this is captured in her analysis, given that mood is merged lower than negation. The same problem may arise with 8a, which is also selected by indicative (see Giannakidou 1998, Phillppaki-Warburton 1998).
198
Cities in Greek: A Minimalist Account ofProdisis and Endisis
As with Roussou (2000), this analysis also distinguishes between a (higher) clause typing head and a lower mood/modal head, although Mood is taken to be an INFL head (cf. that it contains a presumably zero indicative marker and the specific imperative morphology), so thatna is viewed as amood/INFL head (cf. Rivero 1994). This structure has both positive and negative aspects, as is expected, and possibly fares better than Roussou's analysis in certain respects (see Giannakidou 2007 for a detailed comparison). However, for my purposes both structures stand on equal grounds, and for this reason I will continue to use Roussou's analysis. Later on in the analysis I will specifically point out any differences between the two analyses, if these are relevant to the discussion. All in all, based on the above, the actual structure I adopt for indicative clauses (both root and embedded) is as follows. Note that Csub is only optionally present, i.e. if the clause is embedded, while oti!an move to Csub only optionally (i.e. if they precede topic/focus). Oti!an are not present in root indicatives, which might suggest that COp is not projected in these cases, although topics/foci are possible (see also Roberts 200 5 for CForce in root indicatives): (5) [CP Sub[CPtopiclfo~us[CP Op (oti!an) [NEGP Neg dhen [CPM 8a [TP cl-T-v* V]]]]]]
Indicative clauses have full morphological and abstract agreement (marked and checked for all person/number distinctions), as well as full morphological and semantic/ abstract tense (independent tense). A lexical subject or pro is merged in [Spec, v P], although it is not clear whether this subject also moves to [Spec, TP], given the standard assumption that (all) preverbal subjects in Greek are topics (see Alexiadou & Anagnostopoulou 1998, but see also Roussou & Tsimpli 2006, Revithiadou & Spyropoulos 2007b for a different view). The verb may precede all lower vP adverbs as well as higher adverbs including deictic tense adverbials, and possibly epistemic, evidential or evaluative modality adverbs (auxiliaries are more felicitous in this case, though). Lexical verbs do not normally precede ilikrina ('frankly'), as opposed to auxiliary verbs. These facts suggest that both lexical and auxiliary verbs can raise at least as far as (deictic) T (and optionally to one of the higher modality heads - see Cinque 1999, and Alexiadou 1997 for the claim that the verb raises to Agr past tense adverbials10), which is consistent with Roussou's structure, according to which V moves no
Recall that verbs may also stay quite lower, between v and T, although they must precede such adverbs as completely or well. Adam Ledgeway asks how this squares with the idea that preverbal subjects are always topicalized-ifV doesn't rise toT, how is the EPP satisfied? Also, later on it is argued based on Alexiadou & Anagnostopoulou ( 1998) that V must always raise to INFL in Greek and similar languages for EPP reasons. How can this be reconciled with lower movement to some head in between v and T? One idea to solve this puz:lle would be to assume that V targets T in all cases, checking the EPP, however PF may choose to spell out a lower or a higher copy of the verb + d complex. 10.
Chapter 6. Enclisis in Greek 199 further than INFL (cf. the lower position of the verb with regard to 8a). Finally, clitics are always proclitics: (6) a.
(Ilikrina/eftixos/profanos/pithanon) dhen to pighe (Frankly/fortunately/apparently/probably) neg it-cl took-3sg (*?ilikrinal?eftixos/profimos/pithanon) xtes (o Janis) (.. ? frankly/?fortunately/apparently/probably) yesterday (the John) sti Flora. to-the Flora. 'Frankly/Fortunately/Apparently/Probably John didn't take it yesterday to Flora: Greek
Finally, let me point out that it is not clear what the situation is with interrogative clauses (root or embedded). According to some researchers (see e.g. Tsimpli 1995), they could possibly involve V-to-C movement (presumably to CM or CForce, depending on whether the clause is root or embedded), given the obligatory inversion found in these clauses (see also Rizzi 1991 for the same analysis in Romance). However, this view has been criticized by a number of researchers (see for example Drachman & Klidi 1992, Anagnostopoulou 1994, Kotzoglou 2006, and references therein), based on evidence such as the lack of the root/ embedded asymmetry (a typical argument in favor of a V-to-C analysis since den Besten 1983) (cf. 7), the presence of preverbal particles which typically block V movement (cf. imperatives) but which are compatible with negation (cf. 8), as well as the fact that not all wh-words trigger inversion (cf. 9) (cf. Anagnostopoulou's distinction between D-linked and nan-D-linked wh-words). (7) a.
Pjon (* i Eleni) idhe (Vi Eleni)? Who.acc (* the Eleni.nom) saw.3sg (V the Eleni.nom) 'Who did Eleni see?'
Greek
pjon (* i Eleni) idhe (Vi Eleni). Greek b. Me rotise Me.cl asked.3sg who (*the Eleni.nom) saw.3sg (V the Eleni.nom) '(S)he asked me who Helen saw: (8) Dhen ea aghorasis kerja? Neg. fut. buy.2sg candles.acc ~ren't you going to buy any candles?'
Greek
(9) a. Ti (* i Martha) rotise (Vi Martha)? What (*the Martha.nom) asked.3sg (V the Martha.nom) 'What did Martha ask?'
Greek
b. Pate (Vi Martha) rotise (Vi Martha) ghia ton When (the Martha.nom)asked.3sg (the Martha.nom)about the Niko? Niko.acc 'When did Martha ask about Nikos?'
Greek
200
Cities in Greek: A Minimalist Account ofProcllsls and Encllsls
Pjo
ap'
ala ta pedja
(..J i Martha)
dhen idhe
Which from all the children (the Martha.nom) not (...J i Martha)? (the Martha.nom) 'Which one of all the children did Martha not see?'
saw.3sg Greek
The strongest argument against a V-to-C movement approach to interrogatives is the preverbal particles argument, 11 since there is enough evidence that verbs cannot move across these heads (either by skipping them or by adjoining to them12). IfV-movement is not possible in other structures, the question arises why it should be possible in interrogative clauses. Presumably, in these structures the force/illocutionary type feature (or the wh-feature in case ofwh-questions) on the COp head will have to be checked. 13 Assuming that dhen or Oa do not have clause-type properties (cf. that they cannot introduce an embedded clause), it follows that the relevant feature is either checked via long distance agree with the verb in Agr/T, or via a null interrogative complementizer directly merged on the COp head. For na-clauses see Section 6.2.2.2. The question, of course, is what happens when no particle is present in the clause. Theoretically speaking, in that case the verb would be free to move to the clause typing head, at least in root contexts (recall that in embedded clauses oti/an occupy the COp head). The problem is that the evidence in favor or against such a movement is not conclusive. For example, there does not seem to be any difference between declaratives and interrogatives as far as V-placement in relation to adverbs is concerned. To illustrate, in both cases a verb (lexical or aux) cannot precede ilikrina. ('frankly'), the highest IP-adverb in Cinque's (1999) hierarchy, suggesting that in both clauses the verb stays in the IP domain: (10) a.
b.
(...J Ilikrina) echo (?* ilikrina) varethi. (Frankly) have.lsg (frankly) bored-participle 'Frankly, I am bored:
Greek
(..J Ilikrina) echis
(?* ilikrina) katalavi ti simveni? Greek (frankly) have.2sg (frankly) understood what is.going.on3sg 'Frankly, have you realized what's happening?'
On the other hand, auxiliaries, especially those participating in the past perfect periphrasis, appear to be acceptable (in varying degrees, depending on the speaker) when tollowed
u. Further arguments include the Aux + V > Subj argument, and the WH isos ('maybe') V argument (Elena Anagnostopoulou p.c.). Although it seems that particles may move across particles or adjoin to them (see Rivero 1994, Roussou 2000, although see also Roussou 2006).
12.
13. In fact, according to Rizzi (2001) there may be a special interrogative head carrying the relevant feature, which is separate from wh-/focus. This head ls presumably merged higher than
wh!focus, but lower than CForce in Rizzi's ( 1997) system. In Roussou's system interrogative can only be realized in COp.
Chapter 6. Enclisis in Greek by a lexical subject within a root yes/no or wh-interrogative (although this is much less accepted in the latter). Given that this is not possible in declarative clauses or in embedded yes/no interrogatives, one could interpret this distribution in terms ofV-to-C movement: (11) a.
(V 0 Janis) echi/iche (*o Janis) dhi (V o Janis) tin tenia (V o Janis).
Greek (The John-nom) has/had-3sg (the John) seen (the John) the movie-ace (the John) 'John has/had seen the movie: b. HEchi/ ?Iche o Janis dhi tin tenia? Has/ had-3sg the John-nom seen the movie 'Has/Had John seen the movie?'
Greek
c. *?Me rotuse an echi/iche o Janis dhi tin Me.cl was-asking-3sg if has/had-3sg the John-nom seen the tenia. Greek movie-ace 'He/she was asking me whether John has/had seen the movie: d. ?*Pjon echi/iche o Janis pirovolisi sto kefali? Whom has/had-3sg the John-nom shooted in-the head 'Who has/had John shoo ted in the head?'
Greek
e. ?*Me rotuse pjon echi/iche o Janis pirovolisi Me.cl was-asking-3sg whom has/had-3sg the John-nom shooted sto kefali. Greek in-the head 'He/she was asking me who John has/had shooted in the head.
At the moment, it is not clear to me why wh-interrogatives would differ from yes/no interrogatives. One idea might be that in Greek the wh-feature is preferably checked via longdistance Agree. 14 However, this is merely a hypothesis. On the other hand, the fact that a lexical subject may appear between an auxiliary verb and the participle only in root yes/no questions (and not in embedded ones) could possibly suggest V-to-C movement across the preverbal subject (this is merged either in Spec IP/TP or in a low topic position). Moreover, Rizzi (2001) argues that at least in Italian wh-interrogatives introduced by wh-adverbials like when or how exactly do not trigger inversion15 because they are merged in an Interrogative head above the focus/wh head(s), Int being intrinsically endowed with a wh-feature. His arguments can be replicated for Greek. First of all, these adverbials may co-occur with a preverbal focused element, although this is not 14. Adam Ledgeway suggests that o Janis could be the reason for the ungrammaticality, because If it were topicalized in situ. then pjon echl/ iche would not have any obviously positions to target (apart from Spec and head of CPSub).
15. Anagnostopoulou (1994) also talks about the adjunct/ non-adjunct distinction.
201
202
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
normally possible with other wh-words like e.g. complements- and see Tsimpli 1995 for the incompatibility of foci and wh-questions in root contexts in Greek16): (12) a.
?Giati AFTO na tu pume (ke oxi to ala)? Greek Why THIS.acc subj him.cl.gen say.1pl (and not the other-ace) 'Why should we tell him THIS (and not that)?'
KOSTAS pighe sto scholio? b. ?Pate 0 When THE KOSTAs-nom went-3sg to-the school -when did KOSTAS go to the school?' C.
Greek
*PJO DORO TU KOSTA edhose? Which present.acc the Kosta.gen gave.3sg *-which present did (s)he give to KOSTAs?'
Greek
Second, although declarative verbs may precede or follow higher IP adverbs such as idhi ('already'), interrogative ones must always precede them (on a par with imperatives and gerunds, which have been independently shown to occur in C - see discussion below). What is more interesting is that non-inversion triggering wh-words do not show this restriction (see Rizzi 2001 for these arguments applied in Italian):
(.V idhi) echi (.V idhi) dhi tin Alexandra. (13) 0 Janis The John-nom (already) has-3sg (already) seen the Alexandra-ace 'John has already seen Alexandra:
Greek
(14) Pjon (* idhi) echi (.V idhi) dhi? Whom (already) has-3sg (already) seen 'Whom has he/ she already seen?'
Greek
(15) a.
Jati
fili
su
(?idhi)
echun
(.V idhi)
fighi? Greek Why the friends-nom your-cl (already) have-3pl (already) left 'Why have your friends already left?'
b. Ti (*? idhi) echun (.V idhi) kani? What (already) have-3pl (already) done? 'What have they already done?'
Greek
The same applies to adverbs such as 'yesterday': they may also not intervene between the verb and a wh-word, but may do so with non-inversion triggering wh-elements (such as 'when: 'how come, 'where, among others). If this account goes through, then one might explain in a more straightforward way why inversion does not always take place with all wh-interrogatives. Note, however,
16. Elena Anagnostopoulou has pointed out to me that she finds the examples in (12a-b) un-
grammatical. However, I and other speakers I have asked find them acceptable.
Chapter 6. Enclisis in Greek 203
that these arguments are not very strong, and could possibly be accounted for in a different way. Moreover, the available evidence can be conflicting (see position of the verb with regard to frankly). ln any case, we should bear in mind that when no particles are present, V-to-C movement could potentially take place, at least in theory. Ifwe take the above arguments as evidence in favor of this theoretical possibility, then we could take at least some interrogatives to involve V-to-C movement Below is a table summarizing the main properties of clauses taking indicative forms: (16) Table 1: Properties of indicative clauses (main & embedded): Agreement Tense Mood CP periphery Subject Vposition Force Clitic position
6.2.2.2
Full Abstract & Morphological/[- anaphoric] (in the sense of Borer 1989) Full Abstract & Morphological! semantic (independent) Marked for mood (- imperative) +focus/+ wh/ +topics/+ interrogatives pro/ overt (nominative)
T. and possibly eM /C FOIICE in (some) interrogatives Dedaratives; interrogatives Proclitic (in all cases)
Subjunctive clauses
Subjunctive clauses come next in the finiteness list, mainly because they may not be marked for tense (although they are marked for agreement in all cases), and as a result minimally differ from indicative clauses in this respect. Subjunctives involve the indicative form of the verb preceded by the subjunctive particle na. They are negated by the deontic modality negation marker mi(n), which follows na. and immediately precedes the clitic+V complex. I assume the following structure for subjunctive clauses (taken from Roussou 2000): (17)
[CP Sub[CPtopklfucus[CP Op na [NEGP min [CPM (na) [TP cl- T-v*-V]]]]]]
Subjunctives are fully inflected for voice, aspect, agreement (person/number) and are constantly marked as -imperative (although na specifies them as subjunctive). They may be either root or embedded clauses. & embedded clauses they are complements to factive, perception, volitional, aspectual raising, causative and modal predicates, as well as verbs of saying and thinking (see Sitaridou 2002, Sevdali 2006 and Roussou 2006 ). In most cases the subjunctive clause corresponds to an infinitive in other languages. & main clauses, na-clauses may express different types of modal readings (imperative, optative, epistemic modality).
204
Cities in Greek: A Minimalist Account ofProdisis and Endisis
Their tense marking depends on whether they are free or controlled (complement) subjunctive clausesP Free subjunctives may be marked as [-/+ past] (although this is not necessary- see Varlokosta & Hornstein 1993) and have dependent tense (and a pro subject- hence referentiaV-anaphoric agreement), while controlled ones do not inflect for tense (they are constantly marked as -past) and have anaphoric/null tense 18 (and a PRO subject-hence non-referentiaVanaphoric agreement) (see Landau 2004 19 ). Note that it is not clear where exactly in the clausal structure the subject is merged. Some argue that it lies in [Spec, MoodP], Mood being realized by na (ct~ that an overt subject must precede na, or tallow the whole particle-V complex, but cannot appear between na and the V head) (see for example Philippaki-Warburton 1992, Terzi 1992, Hornstein & Varlokosta 1993). Others assume that it is in [Spec AgrP/TP] (see for example Alexiadou & Anagnostopoulou 1999, Krapova 2001, Landau 2004), while others argue that there is no subject projected in the structure (see for example Roussou 2001, 2006, who argues against the existence of pro or PRO). (18) a.
Ksero na ghrafo /*ghrafis /*eghrafa /*ghrapso. Greek Know-1sg subj write-1sgl write-2sgl write-1sg.imp.past/ write-1sg. perf. (controlled subjunctive) 'I know how to write:
b. 0 ]orgas chtes ithele na figho avrio. Greek The ]orgas yesterday wanted-3sg subj leave-1sg tomorrow 'Yesterday Jorgos wanted me to leave tomorroW. (free subjunctive/ abstract tense) c.
Elpizo [o Kostas na tighanise psaria]. Greek Hope-1sg [the Kostas-nom subj fried-3sg fish-pl] 'I hope Kostas fried fish? (free subjunctive/ morphological+ abstract tense) (example taken from Iatridou 1993: 177 (her 4'))
The verb moves to T obligatorily, although it does not seem to be able to move higher (cf. that it follows na obligatorily). This is supported by the fact that it is more felicitous
17· Main na-dauses may inflect for tense distinctions (cf na figho vs. na efo'l'gha (subj leave- I sg vs. subj. was-leaving-lsg), however in many cases there is no actual tense interpretation; rather they may have different modal readings (see Roussou 2000, 2006 for details). 18. This is not always the case. Unrealized future is also possible (see Landau 2004). 19. The distinction between pro and PRO in complement 11a-clauses is based on the presence
or absence of control, and on the semantic and syntactic properties of the null element (see Landau 2004 for more details). Phllippaki-Warburton & Katsimali (1999) have argued against a PRO analysis even for the controlled cases. On the contrary, they assume a pro subject across the board, with control being the result of the semantic properties of the matrix/selecting clause. For Landau (2004), controlled subjunctives may be ambiguous between a pro and a PRO structure (the former involves accidental coreference).
Chapter 6. Enclisis in Greek if it precedes higher tense adverbs (no topic interpretation intended in the following example 20): (19) Elpizo i Xristina (?*tora21 ) na milai (tara) Hope-lsg the Christina-nom (now) subj talk-3sg (now) sti Daniela. to-the Daniela 'I hope Christina is talking now to Daniela:
Greek
All types of na-clauses take proclitics. This means that tense (either in the morphological or in the semantic/abstract sense) does not correlate with the alternation. As finiteness on the Til node is typically related to Tense, Agreement and Mood, and given that between indicatives and subjunctives we have both tense and mood differences but no agreement differences, we may assume that agreement (syntantic or morphologica~ or both?) must be the relevant factor. This first conclusion will be confirmed by the investigation of imperatives and gerunds. Finally, a few comments on interrogative subjunctives. Again, we have the problem of deciding whether this structure could possibly involve V-to- C movement Based on the structure adopted here for subjunctives, it is quite clear that na (and negation, if present) would intervene between the COp head and the verb. This assumption is further supported by the fact that na has COp properties, presumably because it is located in COp at some point of the derivation. For example, it may introduce root clauses, it cannot co-occur with other complementizers (besides pu) (cf. Roussou 2000 on the Romanian and Albanian subjunctives, which may involve an overt complementizer in the COp head) and it may also give rise to different illocutionary force readings (e.g. imperatives, interrogatives, declaratives). These are all properties of a COp head. One possibility would be to assume that na is probed by (the features on) the COp head, forcing overt movement. This is in fact what Roussou (2000) suggests for na in all cases. An alternative would be to assume that na never moves from the CM position; rather, it is probed by COp by long-distance Agree. This is possible and we would still be able to block V-movement across na, however, given that negation is merged higher than na, we would not be able to account for the na-neg order. A third possibility is suggested in Roussou (2006), who assumes that na is directly merged on COp (checking 10. M far as higher IF-adverbs are concerned, judgements are not cleat: It seems that 'adv nad-V' orderings are more acceptable than 'na-d-V adv' orderings (cf. pithanos na min to tpe o Nikos ston Kosta vs. ?? na min to ipe pithanos o Nik.os ston Kosta [probably subj not it.d said3sg the Nick-nom to-the Kosta-acc]). In any case, given that we asswne that na ls in CM, the d-V complex may theoretically appear in the higher IP field. which means that the marginal grammaticality of the 'na-CL-V orderings' may be due to independent factors. I will not go into this matter here.
Elena Anagnostopoulou points out to me that she finds the example with tora in the preverbal position not bad. however I am not sure that this is not the topical reading of tora. Adam Ledgeway reports that also in SIDs V has to precede VP-adverbs in MODO-/CU-dauses.
21.
205
206 Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
its features) when negation is present. The latter then moves from CM to Neg. All these derivations are possible in prindple, and all of them have the same result: that the verb does not move to the C-domain. Presumably, if we assume that the verb may also have interrogative features (given that when no particle is present, it would be the only head available to check them on COp), it checks them by agreeing with CM, which agrees with COp (i.e. a dependency is formed between V and the higher preverbal heads, presumably an Agree correlate of Grimshaw's 1991 notion of Extended V projection). Below follow two tables with a summary of the main properties of subjunctives: (20) Table 2: Properties of free subjunctives Agreement
Full abstract/MorphologicaJ/- anaphoric
Tense
Present/past/perfect (no future)/± morphological/ semantic: dependent anaphoric tense (even when not marked)
Mood
Marked for mood(- imperative)+ na particle/+ semantic (irrealls) mood
CP periphery
+focus,+ topics (also+ wh with some predicates (e.g. verbs of saying)) Pro/ overt (nominative)
Subject
Vposition
Dedaratives/ interrogatives (wh-questions) with a few selecting predicates/ imperatives T (higher INFL domain)
Clitic position
Proclisis
Force
Table 3: Properties of controlled subjunctives Agreement
Full abstract/ MorphologicaJ/ + anaphoric
Tense
Only present/ no marking for tense/ + anaphoric tense
Mood
Marked for mood (- imperative) + na particle/ + semantic (irrealls) mood
CP periphery
+ focus, + topics PRO (nominative)
Subject Force Vposition
Declaratives T (higher INFL domain)
Clitic position
Proclisis
Common properties of na-clause types: full, morphological agreement, marked for modality (- imperative, + na), V is in T and proclisis. Difference from indicatives: morphological/semantic tense, semantic mood, PRO subject is possible. 6.2.2.3 Imperatives
Imperatives are divided into positive and negative ones. I will first discuss positive imperatives, and then proceed to negative ones.
Chapter 6. Enclisis in Greek 207
6.2.2.3.1 Positive imperatives Positive imperatives may inflect for voice and aspect, and are always marked as +imperative (marked by means of specific imperative agreement endings). According to Philippaki-Warburton (1998), Giannakidou (1998) and Roussou (2000) imperatives are the only inflectionally marked mood in Greek (assuming that indicative is marked by zero and that it never moves to the CM head, and that subjunctive/future, etc. are marked via particles in combination with the -imperative form). They do not inflect for tense 22 (in the sense that they may only take one pair of agreement endings which does not enter into any kind of opposition with another pair of endings 23 - recall that in Greek tense and agreement are marked by a single morpheme), and they only inflect for 2nd person singular and 2nd person plural. The former is a true form, in that it takes a special ending that is not used for the same person/number combination in any other paradigm (see Zanuttini 1997). This appears to be true for both voices. The latter is typically taken to be a true form as well. However, this is not entirely accurate. Holton, Mackridge & Philippaki-Warburton (1997) give -ete/-ate (depending on the conjugation) as the ending for the second plural imperative (active), which is identical to the -past/indicative (i.e. -imperative) active form. This is illustrated below:
(21) a.
plinete me! vs. na me wash.perf-2pl.imp me.cl subj me.cl Wash me!' vs. 'You should wash me:
plinete Greek wash.perf.-2pl.-imp
b. aghapate me! vs. (na) me aghapate! Love.perf.-2pl.imp. me.cl subj. me.cl love.perf.-2pl.-imp 'Love me!' vs. 'You should love me' I 'You love me'.
Greek
The same appears to be the case with non-active voice 2pl. imperative forms, which are homophonous with the corresponding [-past+ pertective,- imperative] forms (i.e. the so-called dependent forms). The relevant ending is -ite: (22) a. lith-ite! vs. untie.perf.- 2pl.imp 'Untie yourselves!'
na lith-ite! subj untie. perf- 2pl.-imp
Greek
On the other hand, in most cases where the first vowel syllable of the imperative ending is not stressed, it may be deleted (optionally, many cases depending on style-formal styles appear to prefer non-deletion of the vowel), or it must be deleted when the stem Note. however, that they permit the Perfect vs. Non-Perfect distinction: dhiavase to vs. eche to dhiavasi ('read..2sg.imp it.d vs. have.2sg.imp it.d readprt') (Elena Anagnostopoulou p.c.).
22.
23. In this sense. true imperative endings do not coincide with -past tense/agreement endings from the indicative(- imperative) paradigm. For those forms which are homophonous in the 2nd person plural with the corresponding indicative forms, the morphological marking would be -past (i.e. if we assume that these forms are not true forms).
2o8 Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
ends in particular consonants. I.e. vowel deletion may be optional or obligatory, depending on morpho-phonological factors. Interestingly enough, mono-syllabic, irregular imperatives delete both the lst.sg ending as well as the vowel of the 2nd.pl ending, which are typically homophonous (namely, -e): (23) a.
dhes-e me! vs. *desme! tie.perf.-2sgimp me.cl vs. * tie.perf.-2sg.imp me.cl 'Tie me!'
Greek
b. dhes-ete me! vs. des-te me! tie.perf.-2plimp me.cl vs. tie.perf.-2pl.imp me.cl 'Tie me!'
Greek
c.
pes to! vs. *pese tv! say.perf.-2sg..imp it.cl vs. *say.perf.-2sg.imp itcl 'Say it!'
d. pes-te say.pert~-2pl.imp
tv! vs. *pes-ete tv it.cl vs. * say.perf.-2pl.imp. itcl
Greek
Greek
'Say it!' As a result, it seems that -(e)te is the true imperative ending (for first conjugation verbs). Given that in certain cases the indicative/ subjunctive form cannot be homophonous with the imperative form (since -e deletion is obligatory), we could assume that also the 2nd plural imperative torm is a true imperative: (24) a.
graps-te tv! vs. *grap-s-ete tv! write.perf.-2plimp it.cl vs. *write.perf.2plimp itcl
b. na tv grap-s-ete vs. *na tv subj itcl write.perf.-2pl.-imp.-past vs. *subj itcl grap-s-te write.perf.-2pl.-imp.-past
Greek
Greek
However, the status of homophonous forms (in those cases where deletion is optional, subject to style) is not so clear. The reason I am referring to these details is that Rivero (1994) and Rivero & Terzi (1995) have argued that only (morphologically) true imperative forms can have imperative syntax based primarily on Greek imperatives. The above discussion, however, shows that in many cases imperative and indicative 2nd plural forms may also be identical in Greek, nevertheless they still have imperative syntax. This is in tact what we find in other European languages (e.g. Italian or Spanish), and constitutes a serious challenge for Rivero/Rivero & Terzi.
Chapter 6. Enclisis in Greek 209
Positive imperatives obligatorily take enclitics. Also, they must precede (higher) temporal adverbs (Tl ), and they also precede higher adverbs such as surely or certainljM: (25) (* sighura/ oposdhipote/ tara) fer -to (*surely/ certainly/now) bring -2sg.imp.perfit-cl (?? sighural ??oposdhipote/ ...Jtora)! (?? surely/?? certainly/ ...Jnow) 'Bring it surely/now!'
Greek
Note that imperatives are not compatible with evaluative, epistemic or root modal adverbs (fortunately, probably and intelligently), independently of the V-ADV order: (26) *(eftichos/pithanos/eksipna) kan- to (fortunately/probably/intelligently) do.perf.2sg.imp-itcl (eftichos/pithanos/eksipna)! (fortunately/probably/intelligently)! *'Fortunately/probably/intelligently do it!'
Greek
This is possibly due to independent interpretational factors. Similar facts have also been observed for Italian (see Zanuttini 1997). Note that speech-act mood adverbs such as frankly typically precede true imperatives in Greek: (27) (ilikrina) fer to (*ilikrina) tara! (frankly) bring.perf.-2sg.imp -itcl (*frankly) now 'Frankly, bring it now!'
Greek
Given that for Cinque (1999) the speech-act mood node typically encodes illocutionary force, it is natural to assume that it coincides more or less with Roussou's (2000) COp head (note that Cinque calls it mood, because illocutionary force may be encoded in the verbal form in a number oflanguages). Assuming that the corresponding adverb is in the Spec of this head, then the order frankly-imperative verb is exactly what we would expect, if imperatives moved to COp in order to check the relevant force features (see also analysis below). In any case, it is plausible to assume that the adverb facts show that positive imperatives move to the C-domain. Finally, positive imperatives cannot be embedded; a periphrastic imperative formed by na +-imperative form is used instead: (28) a.
pekse! play. perf.-2sg.imp 'play!'
Greek
24. These adverbs may precede the imperative verb if they are focalized or in a topic position. This is also possible when they follow the verb. In the example the non-focus/topk usage of these adverbs is used.
210
Cities in Greek: A Minimalist Account ofProdisis and Endisis b. mu zitise na pekso. Me.cl-gen asked-3sg subj come. perf.-3sg.-imp 'H/she asked me to plaY. I.e. a complement na-clause is used in subordination.
Greek
For positive imperatives I assume the following structure (taken from Roussou 2000). Note that the verb is inC (see also Rivero 199425 ): (29)
[CPTopic/Fo.:us [CPOp V imp [CPM (Vimp) [TP (Vimp) [vP subj. [VP (Vimp)]]]]]]
According to this structure, imperative verbs move through CM to Cop. This is based on the following facts: a. imperative forms cannot co-occur with particles that are located in the CM head, such as na or as: (30) pline! vs. *nalas pline! wash. perf-2sg.imp vs ... subj/hort wash. perf.-2 sg.imp 'Wash!'
Greek
Only the -imperative form is compatible with such particles, forming periphrastic positive imperatives (these may differ from true imperatives in pragmatic terms, as well- see Roussou 2000). This clearly shows that an imperative form must move to CM, as opposed to a -imperative one, supporting in this sense the intuition of Philippaki-Warburton ( 1998) that imperative in Greek is (also) a mood category (i.e. a modal category encoded on the verbal form- see Palmer 1986 on the distinction between mood and modality). b.
positive imperatives encode the imperative illocutionary force. Moreover, given that they are CM elements, they are expected to move to COp, on a par with na and as. This would explain why both na/as and imperatives are incompatible with complementizers proper. 26
15. Note that according to some researchers (see fur example Roberge & Pirvulescu 1999, Shlonsky 2004) imperatives are truncated clauses, either vPs or some aspectual phrase, but definitely not TPs or CPs. However, such an assumption would fail to predict the following empirical facts: (a) possibllity of foci/topics ln imperative clauses; (b) possibility of negated imperatives in a number of languages where this is allowed; (c) possibility ofwh-movement ln some imperative clauses (Reis & Rosengren 1992); (d) possibility ofembedded imperatives (e.g. Oassical Greek- see Rivero & Terzi 1995); (e) the presence of morpho-syntactic agreement features (according to standard assumptions, these are downloaded from C heads); (f) the presence of imperative force; (g) the presence of a modal C head with corresponding semantic propertles ln positive imperatives - see Portner & Zanuttlni (2003). More generally, by adopting this assumption we would fail to capture the interaction of imperative forms with preverbal maikers in languages like Greek. as well as the association between CM and the prodlsis-endlsis alternation, which I wlll formulate later on. As a result, for all these reasons I will keep on assuming that imperatives involve a CP domain. 26. Recall that according to Roussou (2000), declarative (oti) or interrogative (a11) complementizers may optionally bear subordination features, which force them to move to the highest C
Chapter 6. Enclisis in Greek
Now, the fact that imperatives move through two distinct C heads is an important characteristic ofRoussou's and Philippak:i-Warburtons analysis (see also Giannak:idou 1998 ), which sets them apart from standard analyses of imperative clauses. The latter typically treat the functions/features encoded by CM and Cop/Force as properties of one single C head (see for example Rivero 1994, who assumes that in imperative clauses Mood is not projected, and Rivero & Terzi 1995, who argue that imperative morphology encodes illocutionary force which is checked on a single head C). This partly stems from the theoretical assumptions of the particular stage of the theory on the internal structure of the CP domain at the time of the development of the relevant analyses, and partly on the assumption that in true (positive) imperatives morphology, syntax and semantics correlate with each other, so that imperative force can only be encoded by imperative morphology. However, we have evidence, both from imperative as well as from non-imperative clauses, that these two heads should be projected separately. In the analysis that I will propose for the proclisis-enclisis alternation later on, I will argue that this distinction is crucial for the alternation. In this respect, clitic positioning, as syntactic phenomenon, offers additional support to an independently established structural distinction. In what follows I will briefly discuss the particular morpho-syntactic properties of the tense/agreement ending of positive imperatives. I will argue that a closer examination of these properties will help us account for a number of properties that are inherent to imperatives, and which also play a role in the proclisis-enclisis alternation. In particular, note that although the imperative ending is marked for both person and number, it has been pointed out by some researchers that no true person opposition is encoded within the imperative paradigm. As Joseph (1983), Graffi (1996) and Nikolaeva (2007) among others have argued, imperatives truly inflect only for number. Joseph bases this assumption on Greek positive imperatives: given that the only available forms are 2nd person sg and pl, and that 2nd person is a pragmatically inherent property of the imperative force, he concludes that person is not really marked by the imperative endingsP head, where pu is also merged, when they are present in the structure. Roussou argues that in this way she can account for the otl/an -.focus/topic ordering (which is, apparently, optional), as well as the ungrammaticality of* pu oti/an. Na/as do not have the same optional subordinator feature; rather, they've got modality and clause typing features. This explains why na is possible with pu. Given that as is restricted to direct speech only, one could assume that a subordinator C-head, which is selected by a matrix predicate, never projects when as is present, thus accounting for the non-embeddedness of as-clauses. Presumably, one could assume something similar for imperatives, which also may not be embedded The use of na is the most readily available alternative structure, given that it maybe selected by a subordinator (allowing thus subordination), and it may check CM and COp features, on a par with true imperative forms. 27. The question arises what happens in languages like Ancient Greek which had a 2nd vs. 3rd person opposition in the imperative paradigm. Would that count as a person opposition or not? For Joseph that should probably count as a person opposition, although note that the pragmatics
211
212
Cities in Greek: A Minimalist Account ofProclisls and Enclisis
Graffi bases the same conclusion on Italian positive imperatives. He notices that in Italian the forms used to express imperative force may be distinguished into two groups, depending on their syntax: 2nd person sg and pl as well as 1st pl are (true) imperative forms, whereas 3rd sg and pl forms are surrogate forms which lack the properties of (true) imperative syntax. 28 This distinction does not necessarily correlate with true imperative marking, since in Italian surrogate forms (namely indicatives) may have true imperative syntax (see also Zanuttini 1997). Moreover, it should be kept distinct from the notional interpretation of what one calls 'imperative modality, since 3rd person imperative forms are notionally/pragmatically imperatives (they are used with an imperative force), but syntactically they are indicatives. He further points out that what distinguishes true (syntactic) imperatives from other forms is that they must appear in the CP domain, and that the reason they must move to C is that they have to check a special impoverished agreement feature on C. His claim is that this special, defective AgrC does not contain a person feature, but only a number feature. Finally, Goria (2004) (and see also Nikolaeva 2007) points out that in Northern Italian Dialects (NIDs) non-finite forms (including infinitives and gerunds) as well as imperatives do not take (person) subject clitics, in a way that is reminiscent of Semitic languages, which do not mark person in imperative forms (cf. also Swahili). The idea is again that person is not syntactically present, a property that is assumed to be a characteristic of imperative C. All the above views on person marking in positive imperatives have two main common properties: (a) they point out the traditional intuition that 2nd person is an inherent feature of the imperative situation/force, in the sense that the subject of an imperative verb necessarily refers to the addressee (an individual or a set of individuals). This is clearly evident from so-called 1st plural imperatives, which may only have an inclusive interpretation (i.e. their subject always refers to the addressee (second person)), as opposed to the 1st person plural torm in other moods, which may or may not have an inclusive interpretation (depending on the properties of the particular language). It is also evident from the fact that imperative subjects, which cross-linguistically may range over null subjects (i), overt pronominal subjects (ii), quantificational subjects (iii), and non quantificational, non pronominal subjects (iv) (see Zanuttini 2007), all have to refer to the addressee (or a group of addressees, or a group containing the addressee - see Mauck & Zanuttini 2005). Finally. it is also evident from Spanish polite imperatives: these are morphologically surrogate forms ('borrowed' from the subjunctive paradigm) which however have a true imperative syntax (contrary
of 3rd person imperative involve the expression of a wish/order that has to be realized by the present hearer. In this respect, it ls the 2nd person that is really expressed even by the 3rd person form, leading thus to the conclusion that person is not really marked by the imperative endings. 28. These properties include enclisis and the possibility ofhaving both neg-CL-V and neg-V-CL orders in negative imperatives: 3rd person forms take proclitics both with and without negation, as opposed to all other forms.
Chapter 6. Enclisis in Greek
to the corresponding 3rd person Italian 'imperatives'), and, more importantly, semantically refer to the hearer/addressee. This shows that AgrC, or more generally the imperative agreement (i.e. the agreement of forms with imperative syntax, which end up in the C domain) inherently refers to the addressee feature. The second (b) common characteristic of these approaches is that they interpret the traditional intuition about imperatives as lack of a person feature, either morphologically, or syntactically, or, most commonly, both ways. They typically use either morphological or syntactic evidence in favour of this claim. Note, however, that it is one thing to claim that imperatives inherently include the addressee property, and another to conclude from this that these forms do not have a syntactic person feature. So, the question we are facing is whether we want to interpret the notion 'inherent addressee' as 'lack of a syntactic person feature~ Following standard assumptions in the literature, this would not be possible; in fact, it should be wrong.z9 This is so, because interfaces are assumed to interpret syntactic structures. As a result, if semantically speaking an addressee feature (interpretation) is present at the LF representation, this can only be the case because a corresponding abstract syntactic feature was present (and manipulated) by the syntactic computation. This is essentially the same kind of idea that has been prevalent in the literature on the PF-syntax interface: morphological overtness entails syntactic features, but not vice versa. Additional evidence for this view comes from morphology: recall that Joseph (1983) based his claim that imperative forms do not mark person on the notion of 'morphological paradigm' for quite obvious reasons. However, on an individual level, all these forms mark both person and number. This is evident from a number of facts: (a) they typically borrow a corresponding ending from another paradigm (this appears to be the case especially for the plural forms; singular forms are more special, in that they may have their own specific ending, or they may totally lack an agreement encoding ending- see for example Zanuttini 1997 for this claim in Italian. See also Nikolaeva 2007 and Koch 1995). This ending marks both person and number; (b) according to the Subset Principle (see Halle 1997 for an exact formulation) vocabulary insertion (namely the insertion (under competition) of phonological exponents into terminal syntantic nodes, which morphologically realize a set of syntactic features present in that particular node) cannot take place if the vocabulary item contains features that are not also present in the syntactic node. As a result, in order for an imperative ending to mark person, the latter must be present in the relevant syntactic head that is realized by this particular ending. Although the Subset principle is mainly used by realizational theories of morphology, and in particular by such models as Distributed Morphology (see Halle & Marantz 1993, 1994), it appears to be a rule with cross-modular applicability (see for example Kiparsky 1973), and therefore its status is 29. But c£ the Benveniste tradition that 3rd person = no person. 1his would not pose a problem since semantically the absence of person could correlate with absence thereof at the syntactic level
213
214
Cities in Greek: A Minimalist Account ofProdisis and Endisis quite substantive, independent of and above the particular morphological theory used. In any case, that would also follow from minimalist assumptions. In this respect, note that in order to capture the inclusive interpretation of 1st person pl imperative forms, we would be forced to refer to person features ( 1 2 or simply partid pant- see Adger & Harbour 2007, and McGinnis 2004), both in the syntax and in the morphology. As a result. if we want to maintain the interpretationaVrealizational nature of PF, we need to assume that person is present as a syntactic feature in the imperative Agr node.30 Note that also from a purely syntactic point of view the postulation of a person (addressee/2nd) feature tor imperative clauses seems to be favoured. This is so, because imperative forms license a null (pro) subject, even in languages that do not otherwise license pro in finite clauses (such as e.g. English- see Potsdam 1996), which is syntactically active (e.g. it binds 2nd person anaphors). Independently of whether we assume that this subject is a full DP, or a nominal feature in the verbal ending, and independently of our assumptions regarding feature specification (e.g. how the value of this feature is assigned), in all cases a syntactic (person =addressee) feature is required in the syntactic representation in order to account for the syntactic effect of person. All in all, according to what was said above we must conclude that person must be part of the syntactic spedfications of imperative Agr, contra Graffi and Goria. However, it is equally true these approaches capture the cross-linguistic fact that imperative forms are akin to non-finite forms in two interrelated ways: (a) they are morphologically deficient, in that they either do not take any agreement ending( s), or they borrow them from other paradigms; (b) that they are also syntactically/ semantically restricted, in that they have to refer to an addressee.31 In this sense, one could say that they resemble controlled or anaphoric agreement. For this reason, in order to be able to express these generalizations but also to maintain the conclusion that person is present in imperative forms, I will tallow Jensen (2002) in assuming that imperative T contains an interpretable and valued (as an addressee = 2nd) person teature.32 A syntactically 30· Presumably, for those cases like the Semitic or Swahili imperatives, or even for NID imperatives, where the independent person morpheme is missing, we would need to assume a zero morpheme. Whether this assumption is independently supported by additional evldence is a separate issue that will not concern me here. 31· Note that controlled subjunctives as well are in essence restricted in their agreement features, given that they must be copies of the corresponding features on the controller. However, this type of restriction is different from that found in imperatives, given that the former are free to take on any number/person combinations (always depending on the properties of the controller), as opposed to the latter which must be 2nd person in all cases. This difference is also evident in the distinct interpretations that null subjects receive in imperatives and subjunctives: the former must refer to the addressee, while the latter either has a controlled or a generic interpretation. 32· Note that according to Jensen (2002) the addressee does not necessarily coinclde with the agent of the action/event/property that must be carried out This is achieved by splitting the notion of thematic subject into two parts, the thematic (or intended agent) part (see also Potsdam 1996) and the addressee part According to her, declarative subjects differ from imperative
Chapter 6. Enclisis in Greek 215
present though non-active syntactic person feature will ensure the special syntactic and semantic properties of imperatives. I will further assume that number is also syntactically present and active. Regarding the syntactic representation of tense in imperatives, I won't have much to say. Typically imperatives are assumed not to have any tense properties (see Zanuttini 1997), given that they do not inflect for tense, they do not take negation (presumably, negation requires the presence of a tense node - see Zanuttini 1996 - and cf. small clauses which lack T and cannot be negated), and more generally they do not appear to have any semantic/syntactic tense properties that would give evidence for the presence of a tense node/operator (see also Platzack & Rosengren 1998). Syntactically speaking, this could mean that the T node is present but empty (which normally would not be allowed in minimalism, given that projections project from features), or alternatively that the T node is absent altogether. The immediate consequences of the non-presence ofT in imperative clauses is that their agreement features (see previous discussion) must be parasitic on some other node, either Cor some lower (aspectual/ tense 2- cf. Giorgi & Pianesi 1997) node (unless, of course, pure Agr nodes are possible, or if Agr nodes may have semantic properties), and that clitics cannot attach to T. Apparently, cliticization will have to take place in some lower functional head (e.g. v*), which as we have already seen is an independently available strategy cross-linguistically, and therefore should not be a problem.33 On the other hand, one could argue, following Rupp (2003) and Jensen (2002) that imperative T contains an interpretable Tense feature anchored to speech time (according to Rupp, this is 'a non-finite unrealized future tense; following Boskovic 1997). In this sense, imperatives would be formally closer to controlled infinitives which also have an irrealis/future interpretation (see Stowell1982).
subjects in that the former are subjects of predication semantically, as opposed to the latter which are not (cf Platzack's & Rosengrerls 1998 statement that "the imperative N ... can be used only to talk TO the addressee not ABOUT him or her" (Platzack & Rosengren 1998: 177). Accordingly, in the default case agent and addressee are perceived as one unified subject, but it may also be the case that they are not related by identity. In that case they are related by CONTRO I.., in the sense that the addressee may control the intended agent Note in this respect that not all languages allow for such a split interpretation ofthe imperative subject (e.g. Greek, which only allows 2nd person pro or overt pronominal subjects (vocatives are also possible, but they do not seem to be real subjects)). See Zanuttini (2007) for an account of why languages may differ in what types of imperative subjects they allow. 33· Adam Ledgeway does not agree with the view that imperatives do not have tense. First of all, he argues that the daim that imperatives lack a T node because they show no tense inflection cannot be taken seriously, given that morphological distinctions do not mirror lack of syntactic distinctions/categories. In addition, it is not really true that lack ofT node implies lack of negation, as we find negated small clauses: d. It. lo aedevo [sc 11011 adatto]. Finally, he points out that imperatives have an inherent unrealized future (irrealis) temporal interpretation.
216
Clitics in Greek: A Minimalist Account ofProdlsls and Endlsls Nothing really hinges on the choice of either one of the two options, as tense does not seem to play a role in the proclisis-enclisis alternation: recall that the basic difference between subjunctives and imperatives is that the former are fully inflected for agreement (although not necessarily for tense), while the latter are restricted in which person agreement features they may realize. As a result, 1 will not take a stand as to whether aT node is present or not in positive imperatives. However, I will clearly continue to assume that a CFin head is projected in these clauses, pace Platzack & Rosengren (1998) and DiDomenico (2004), as we have evidence for the presence of this node in Greek (see discussion above), but also because absence of CFin entails absence of a logophoric/speech centre, making in this way the interpretation of deictic elements (such as e.g. expressions oflocationor time, but also the interpretation of the addressee feature itself) impossible. To return to the comparison between (positive) imperatives and subjunctives, it allows us to isolate one single feature out of the camp lex agreement feature bundle that appears to be relevant for the alternation. namely person. Note however that this difference in person appears to cluster with other differences, namely differences in (i) mood (-imperative vs. +imperative) and (ii) scope ofV-movement (C-head vs. T/ INFL-head). This is evident from the following table, which summarizes the main properties of positive imperatives: ( 31) Table 4: properties of positive imperatives
Agreement Tense Mood CP periphery Subject Force Vposition Clitic position
Restricted person agreement/morphological/-anaphoric No morphological/no semantic tense
Marked for mood(+ imperative)/ semantic mood + topics, + focus Pro/overt (nominative; restricted to second person only) Imperative CForce (not compatible with particles) Endisis (free ordering in clusters, as opposed to proclitics)
For the time being, the empirical generalization, as it emerges from the Greek data, could be formulated as follows:
Generalization A (to be revised) (32) Absence of a syntactically active (i.e. unvalued) person teature on the T/ Agr head (i.e. restriction on person agreement) correlates with enclisis, + imperative mood and V-to-CM movement (obligatory). Presence of full agreement on the T/Agr head correlates with proclisis,- imperative mood and V-to-T movement (obligatory).
Chapter 6. Enclisis in Greek 6.2.2.3.2 Negati11e imperatives Negative imperatives are suppletive forms formed by the negation particle min and the subjunctive form of the verb. They are typically root clauses, although they may also be embedded under a variety of verbs expressing commands. In their syntax they behave more or less like other subjunctive clauses: they always take proclitics, the verb moves to Tl Agr but not further, they are fully inflected for agreement (number and person), although not for tense. Despite all this, they also appear to have a cluster of properties that all together distinguish them from the rest of subjunctive clauses.34 For example, although they take a pro or an overt lexical subject, they cannot inflect for tense (in essence, they do not tolerate the± past distinction, being consistently (un)marked as -past). In this respect, they resemble positive imperatives, which have an irrealis meaning. On the other hand, they allow for more number/person combinations (1st. 2nd and 3rd) than positive imperatives, although the resulting possible interpretations of the subject are the same with those of positive imperatives in languages that allow 3rd person subjects with imperative verbs (e.g. English), depending of course on the properties of the subject DP as well (e.g. quantificational vs. non-quantificational, pronominal, overt lexical). Regarding illocutionary force, there are two possibilities. One assumes that these structures do not encode imperative force, since they are not true imperatives (see e.g. Rivero 1994, Rivero & Terzi 1995). As a result, they do not require V-to-C movement (checking). The other one assumes that negative imperatives (as well as surrogate positive imperatives) may also encode! realize such a force feature (see e.g. Zanuttini 1997 ). The main difference between these two analyses is that the former takes only V forms to be able to check force features, while the latter assumes that these may be checked by a range of elements such as verbs, complementizers, negation and modal particles (see also Isac & Jakab 2004). Given that Roussou's analysis, which I assume here, follows the latter approach, we can assume that na raises from CM to COp in order to check the imperative COp feature. Alternatively, following Roussou (2006), we can assume that na is directly merged on COp. Note that further empirical support for this approach is offered by the fact that in Greek negative imperatives the na particle may be dropped, leaving only the negation particle and the verbal torm. This is illustrated below:
(33) na min tv dhosis! vs. min to dhosis! subj not ilcl give-2sgvs. not it.cl give-2sg 'Do not give it!'
Greek
According to Isac & Jakab (2004), Zanuttini (1997) and Roussou (2000, 2006) in this case it is the negation head that checks the relevant feature on C. This follows from the fact that these sentences are also interpreted as commands, even though no overt 34· Note that most of these properties are also found with positive periphrastic imperatives, which are also subjunctive clauses. The uses of periphrastic imperatives include traditional notions such as 'jussive: which refers to 1st and 3rd person commands (see Palmer 1986).
217
218
Cities in Greek: A Minimalist Account ofProdisis and Endisis
modal particle is present. Note however that na-drop is only possible if the verb is in 2nd person; a 3rd or 1st person negative imperative cannot drop na: (34) na min to dhosi! vs. *min to dhosi! subj not itcl give-3sg not it.cl give-3sg
Greek
In this respect, na-less negative imperatives resemble true imperatives in one more respect: not only do they have to be -past, they must also be 2nd person. In other words, in terms of syntactic/semantic and morphological properties, they become more restricted as soon as na is dropped. Given that the 2nd person restriction is a hallmark of imperative modality (besides force), i.e. of the special modality that is associated with positive imperatives, I think that the best way to account for this restriction is to assume that in the absence of na. in the clause, the negation head must also check the imperative modality feature in CM, along with the negation feature on Neg and the imperative force feature on COp.35 This would force both an imperative reading (note in this respect that na-less negative imperatives are closer in meaning to true imperatives than negative imperatives with na, which receive a wider range of readings, always within the more general 'jussive/imperative' interpretation). The more interesting part, nevertheless, is that we have managed to find a clause that is almost as restricted as a true imperative clause in its morpho- syntactic/ semantic properties, but which may be negated and which also takes proclitics. If we assume that also in na-less clauses the 2nd person feature is inactive, an assumption that seems to be forced by the analysis so far, then the generalization put forward in (32) cannot be right: an inactive person feature is a necessary but not a sufficient condition for enclisis. It seems that the crucial difference between the two structures is in the way C-modality is checked, namely by a particle vs. a V-form. I will return to this point later. A final note should also be added in relation to the impossibility of true negated imperatives (i.e. morphologically true imperatives that cannot be negated by preverbal negative markers, namely negative heads that can negate a clause by themselves - see Zanuttini 1997 for details). This is a quite common phenomenon across languages, although it is definitely not a universal (see Han 1998, Zeijlstra 2006 for an extensive typology of negation and true imperatives). Moreover, it appears not to be entirely uniform among those languages that exhibit it: e.g. in Italian surrogate (true - syntactically) imperatives can be negated, even if they take enclitics. On the other hand, in Spanish surrogate (but syntactically true) imperatives can only be negated if they take proclitics, i.e. only if they are subjunctive clauses. The typical explanation for the incompatibility 35· Things might be more complicated, though: it seems that na cannot be dropped also with other uses/interpretations of periphrastic imperatives. For example, it cannot be dropped when the verb is +past (cf 11a min to edhine vs. * min to edhine- subj not it.cl gave.imper£3sg/+past = 'she/he should not have given it'), which has a counterfuctual reading. Similar restrictions apply to as, which can never be dropped in front of min. At the moment, it is not dear to me why this restriction should apply. It seems that min alone is only 'compatible' with imperative modality, but not with other types of modal readings. I will leave this issue here for further research.
Chapter 6. Enclisis in Greek between negation and true imperatives is that given in Rivero (1994): negation acts as an intervener between V in T/ Agr and C. As we have seen, this may be theoretically implemented either as a HMC/ECP violation or as a closest goal case. For approaches like that ofRoussou (2000) or Zanuttini (1997), which assume that the imperative force in CForce/ Op can be checked by a range of items, including negation, this explanation is not enough: if negation can satisfy the force feature, why can a true imperative not be negated? Zanuttini's answer is that negation that is selected by an imperative force head selects a mood head that must be checked by a strong (in terms of mood features), rich mood form. Based on the observation that true imperatives either involve the stem plus a thematic vowel, and/or some agreement ending, but no mood marking, she concludes that imperative forms have weak/meager mood features, and as a result cannot check the strong mood head. Although this approach may work for Italian, it is not clear how it would work for languages which have imperative mood morphology and which despite this morphology do not allow negated true imperatives, like e.g. Greek. One might argue that even these languages have mood features that count as weak for the purposes of imperative negation, however it is obvious that this would render the whole approach ad hoc and circular. This is confirmed by the fact that there are languages with morphologically true imperatives that can be negated by a preverbal negative head (e.g. many Slavic languages). One wonders under what exactly criteria these forms would not count as true in terms of negation and its selected strong mood. Roussou's answer is more complicated. She assumes that the imperative verb is like na!as in that it has to check features both in CM and COp. Given that negation intervenes between these two heads, and given that it may optionally have operator features, it constitutes a closer goal for COp, and hence blocks Agree between COp and the verb in CM. The problem that arises with this solution is, of course, what happens with na in negated imperatives: if na must also move to COp, why does negation not block this movement (assuming that neg may have COp features)? Roussou argues that in these cases na is directly merged in COp, and it is Neg (which may also have CM features) that checks CM. The problem with this solution is that it is rather theory dependent: it assumes that movement is reinterpreted as the spell-out of more than one features (of distinct positions), and as a result forms are directly merged in the positions where they surface. lf na were merged in CM, though, Neg would in principle be able to block CM-to-COp movement Equally. negated imperatives should be grammatical, given that CM is checked by the verb, while COp is checked by Neg. If imperatives have COp features as well, these could be checked via agreement with the neg head- this is in fact what was argued tor with regard to na in COp/CM. I will leave this issue here. I will return to it in the analysis section, where I will argue that Philippaki-Warburton (1998) was right in assuming that negated imperatives are out because neg blocks CM feature-checking. This will become obvious if we compare true with surrogate imperatives (in syntactic terms). Below I give a table with a summary of the main properties of negated imperatives:
219
220
Cities in Greek: A Minimalist Account ofProdisis and Endisis (35) Table 5: properties of negative (suppletive) imperatives
Agreement Tense Mood CP periphery Subject Force Vposition Oitic position
Full abstract// Morphological (ahhough further restrictions imposed if na is not present)//-anaphoric Only marked as -past// no morphological or semantic tense Marked for - imperative + the subjunctive particle// semantic mood + focus, + topics pro/overt (nominative) 'Imperative'- like/ jussive T Prodisis
6.2.2.4 Gerunds/Active participles as absolutives!free adjuncts Gerunds are the final form I will be looking into in this chapter. Gerunds are not a mood, however they also take clitics, and they are the only form that fits as close as pas sible to the traditional notion of non-finiteness. Given that they only accept enclitics, it will be interesting to see whether the correlations we found in true imperatives also apply in their case as well. More particularly. gerunds are active participles, formed from the [- perfective] verbal stem and the (modal) default ending -ondas,36 and syntactically they are (temporal or manner) adjuncts (see Rivero 1994; Tsimpli 2000 and Tantalou 200437 ). They are not marked for tense or agreement at all (and they are always + active, - perfective), however they may or may not have semantic/dependent tense, and± anaphoric agreement (i.e. pro or PRO) (depending on the function/interpretation of the gerund).38
36. According to Philippaki-Warburton (1995) the -ondas ending is an Agr ending (see Reuland 1983 for English -ing), not a M(ood)-head. For Drachman & Klldi (1992), Rivero (1994), Roussou (2000), Tsimpli (2000) and many others, -ondas lies in M. Note, however, there is little consensus among these researchers on whether this is an INFL-type head or a C-type head. Nothing hinges on this choice with regard to my purposes, given that what I need later on in the analysis is that this head has verbal/inflectional content, i.e. that it contains an affix/features that must be picked up/checked by the verb obligatorily. Since both sides agree on this point, I will asswne, following Roussou (2000) that the gerund ending lies in CM. 37· Different types of gerunds appear to adjoin to the main clause at distinct levels (e.g. VP vs. TP). Note that gerunds may also be used with a conditional or a modal interpretation (see 'Thoulas 1996, Tsimpli 2000. See also Iatridou 2000 on mood!modality and counterfactuality). As far as their syntactic structure is concerned. some researchers assume that all gerunds have the same structure, while others argue for a distinct structure for each type. Moreover, some assume that gerunds are CPs (e.g. Tsoulas 1996), while others argue against this (e.g. Tsimpli 2000). 38. Note that not everyone agrees on that point. For example, Tsimpli (2000) argues that any temporal or control readings of gerunds, which would suggest the presence of a temporal node
Chapter 6. Enclisis in Greek
Manner gerunds appear to have less structure than temporal gerunds (as has been shown by means of structural tests39). Moreover, manner gerunds can precede low manner adverbs (e.g. kala 'well'), but they cannot e.g. precede an imperfective aspectual adverb (e.g. panta 'always') (or other aspectual adverbs such as sinithos 'usuallY, sichna. 'often, or CP adverbs such as isos 'maybe' or pithanon 'possibly'): (36) a.
b.
Marina erchotan sto spiti mu The Marina-nom was-coming-3sg to-the house my.cl [(.V panta) traghudondas (*panta)]. [(always) singing (always)] 'Marina was coming to my house always singing:
Greek
Marina erchotan sto spiti mu The Marina-nom was-coming-3sg to-the house- my-el [(*ghrighora) trechondas (.V ghrighora)]. [(quickly) running (quickly)] 'Marina was coming to my house by running fast:
Greek
Moreover, manner gerunds are not compatible with tense adverbs (see Tantalou 2004 ). Temporal gerunds, on the other hand, appear to be able to raise to a higher position (hence, supporting the view that structurally they are richer in comparison to manner gerunds). In particular, temporal gerunds (both complex and non-complex) must precede deictic temporal (realis) adverbs (and they may precede even evaluative adverbs such as e.g. eftichos ('luckily'): or of a PRO subject, are not syntactically determined; rather, they are pragmatically conditioned, the only constraint imposed by the grammar being the situation type of the gerund Nevertheless, she assumes the presence of a functional projection, with aspectual and a ± perfect feature, where the verb or the auxiliary move to and which specifies the E/R relation. This node coincides in some respects with Giorgi & Pianesi's (1997) lower T2 head (cf. also Cinque's 1999 T-anterior head). On the other hand, Tsoulas (1996) argues that all types of gerunds have a temporal reading, while Tantalou (2004) claims that manner gerunds do not project a TP, as opposed to temporal gerunds which do so. Here, I will follow Tantalou (2004), since I believe there is enough evldence for both a temporal operator in temporal gerunds (and lack thereof in manner gerunds) as well as for a pro/PRO distinction. 39· Properties of manner gerunds: they cannot form complex tenses, i.e. no auxlliary is possible. The may only denote an event contemporaneous to the main clause event; they only allow for a PRO subject (cf. that their subject is obligatorily controlled by the matrix subject, it must be covert and it can only be in nominative); no negation, topics, focus or wh-questions are possible, and they must always follow the main verb (postverbal gerunds) (see Tantalou 2004 for discussion). Properties of temporal gerunds: they can form complex tenses (i.e. they can take auxiliaries), they have a pro subject in nominative case (cf. the covert-overt possibllity), although it typically corefers with a DP in the matrix dause, and it must be postverbal. Moreover, they can be negated (using the marked negation particle mi (which patterns with modal readings), and while focus and topics are possible, wh-questions are marginal (they seem to be more acceptable when an auxlliary is present). Finally, they may license semantic/dependent tense. (see Tantalou 2004 for discussion)
221
222
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
(37) (..,J Eftichos/* Simera) vlepondas ("' eftichos/...J simera) ti Maria, (Today/luckily) seeing (luckily/today) the Mary-ace, tha sinidhitopiisi oti tin aghapa. fut realize-3sg that her.clloves-3sg 'Possibly after seeing Maria today, he will realize that he loves her'
Greek
The structure I am adopting for each gerund type is given below (adapted from Tsimpli 200<J40):
Ma.nner geru.nds (38) a.
[CPM gerund [AspP (ger) [vp PRO (ger) [VP (ger) DPo]]]]
b.
Daliai pighe sto spiti [eirj perpatontas]. The Dalia-nom went-2sg to-the house [walking] 'Dalia went home on foot:
Greek
Temporal geru.nds (39) a.
[NEGP min [CM gerund [TP (ger) [PERFP/ASP/T2 (ger) [vP subject /pro (ger) [VP (ger) object]]]]]]
b. Echontas dhi ton Kosta
chtes,
simera apofasisa
na ... Greek
Having seenthe Kosta-acc yesterday today decided-lsg to... 'Having seen Kostas yesterday, today I decided to ... : Below I give two tables summarizing the properties of manner and temporal gerunds respectively: (40) Table 6: Properties of manner gerunds Agreement Tense Mood CP periphery Subject Vposition Clitic position
No agreement marking!/ abstract+ anaphoric agreement No tense marking! no semantic tense Mood marking/ modality readings - (and no negation) PRO (nominative-presumably default)
~ Enclisis
40. Elena Anagnostopoulou asks why it ls that gerunds check modality features. She points out that this is particularly unclear for manner gerunds which are smaller than temporal ones. I am not aware of any independent evidence in favor of this daim, however note that most researchers that have worked on Greek gerunds assume that they check mood/modality features (see e.g. Rivero 1994, Roussou 2001, Tsirnpli 2000). Some plausible evidence comes from the fact that (temporal) gerunds take the special negation marker min, which ls selected by non-indicative mood/modality. Unfortunately, this would not work for manner gerunds, which cannot be negated
Chapter 6. Enclisis in Greek 223 (41) Table 7: Properties of temporal gerunds
Agreement Tense Mood CP periphery Subject Vposition Clitic position
No agreement markingl/abstract - anaphoric agreement No tense marking/ semantic tense possible Mood marking/modality readings +topics, +focus,+ questions (marginal, although more acceptable with auxiliaries) pro (nominative), postverbal
eM Endisis
Note that clitics are always enclitics with gerunds. Comparison of the two types of gerunds reveals that their only other common property is their modal marking, the lack of any agreement or tense marking (although they may differ in terms of semantic (± anaphoric) agreement/tense), as well as the position of the gerund in the CM node, which bears the modal inflection. Further comparison of gerunds with subjunctives (which are always marked for agreement, although they may lack tense (both morphological and semantic) and have+ anaphoric agreement (i.e. PRO)) suggests that overt marking of agreement correlates with proclisis. In turn, full marking is associated with the position of the verb in distinct inflectional heads (modal vs. nonmodal). Now, if we compare gerunds with imperatives, two things become evident: (a) that in both cases enclisis appears to correlate with properties of the person agreement feature in particular, and in particular with restricted agreement marking (only 2nd person vs. no marking); (b) that this restricted person marking is also associated with a modal/mood property in CM which forces V-movement to that head; (c) that restricted person marking on T does not necessarily entail a syntactically non-active (i.e.+ interpretable;+ valued) feature; on the contrary, it may be associated with more than one syntactic structures on T (e.g. in gerunds it must involve a syntactically active (person) feature- cf. pro/PRO). This is quite straightforward, given that syntax does not map to morphology isomorphically. However, while syntactic activity/inactivity on T does not appear to correlate with person marking, person marking appears to correlate with the (obligatory) position of the verb, i.e. with the highest inflectional verbal head. To put it in other words, restrictive vs. non-restrictive person agreement on the verb is associated with distinct (higher) inflectional positions, namely CM and T respectively. We see that immediately, if we compare gerunds with subjunctives: although they both come in controlled and non-controlled variants, and therefore seem to involve similar (abstract) syntactic properties on T, one has special person agreement marking as opposed to the other, which doesn't. In gerunds, special agreement correlates with CM being inflectional whereas with subjunctives full!standard agreement correlates with T (but not CM) being inflectional. As a result, restricted vs. non-restricted
224 Cities in Greek: A Minimalist Account ofProdisis and Endisis
agreement should not be linked to the properties of syntactic features on T directly; rather, it should be related to the properties of CM, and only indirectly to those on T. As a result, the empirical generalization in (32) should be reformulated as:
Gene,.alization B (42) Proclisis correlates with non-restricted/full person agreement on T, while enclisis correlates with restricted person agreement on T. Non-restricted agreement is associated with the [-imperative] paradigm and a non-inflectional CM head (and, thus, with non-obligatory movement to en), while restricted agreement is associated with the [+ imperative] paradigm and an inflectional CM head (and, thus, with obligatory movement to C). What the correlation in (42) proposes is that both proclisis-enclisis and restricted/ non-restricted person agreement on T correlate with properties of CM. In this respect, the traditional intuition that mood/modality (i.e.± imperative for Greek] is related to finiteness and to the proclisis-enclisis alternation is successfully captured. The question of course is how the scope of obligatory V-movement fits into the picture, and why the inflectional nature of CM appears to be crucial to the alternation. 6.2.3 Summary of empirical points
In this section I will highlight the main empirical points stemming from the previous discussion, which are relevant for the proclisis-enclisis alternation, and which ask for explanation. They are given below: a. Tense (± morphological/abstract) does not play a role in the alternation. b. Person is the feature whose properties (restricted vs. non-restricted in the morphological sense) have an effect on the alternation. In particular, person is both a sufficient and a necessary condition for the alternation, as opposed to V-movement.42 41. Recall that full agreement forms (i.e. -imperative forms) may move further to C at least in interrogatives. The point here is that a full agreement form must stop at T in any case, independently of whether it moves to a higher head or not, whereas a non- full agreement form must stop at CM in any case, although it may move higher (cf imperatives). Adam Ledgeway points out that also Italian present subjunctives where agreement in the singular does not mark person, just singular number, could be associated with propertles of the CM head (cf. e.g. che deletion). Although association with CM cannot be ruled out and is an issue for further research, I am not sure that restricted agreement in Italian subjunctives is related to CM, given that subjunctive morphology in Italian as well as in other Romance languages is traditionally linked to a lower mood head (see also Harris 1995). 42. If we assume that indicative interrogatives may move to CForce also in Greek (see discussion above), and based on the observation that interrogatives may only take proclitics, we can conclude that V-movement across the clitic!zation site (assuming the latter is within the INFL domain, possibly T, but see also discussion below) triggers enclisis only if movement to C involves at least one functional head with V-related inflectional propertles.
Chapter 6. Enclisis in Greek 225
c.
Restricted vs. non-restricted person agreement is not associated with a particular (abstract) syntactic property on T/1 (i.e. activity/inactivity or presence/absence of a person feature). Rather, it appears to be associated with properties of the CM head, and in particular with the presence/absence of inflectional modal features that need to be checked by the verb (for Greek this would reduce to the [± imperative] distinction (modulo gerunds)). CM properties, in turn, are directly associated with the scope ofV-movement/V-position (as well as with the selecting Force/operator feature (at least for imperatives vs. indicatives/subjunctives43 )). Afl a result, also restricted vs. non-restricted person agreement correlates with the position of the verb in CM vs. T/ AG R.
We see then that finiteness indeed correlates with the alternation, at the morphological person agreement level. However, it does not seem to have any direct relation with what happens on T, and the same appears to be true for the alternation more generally. Rather, it seems that syntax/ abstract properties kick in at the CM level: it is CM that is directly related to V-position and to the proclisis-enclisis alternation. CM properties, though, leave morphological 'traces' on the lower T/AGR head. The question that arises is how exactly CM affects the proclisis-enclisis alternation.
6.3 EPP and head movement: Extending the V domain 6.3.1 Preliminary remarks Before moving on to the actual analysis I would like to briefly recapitulate what has been done so far, in order to have a more accurate view of the generalizations that this analysis will need to account for. We started our discussion by testing the empirical adequacy of the traditional observation that the finiteness of the verbal host/clause correlates with the outcome of the proclisis-enclisis alternation. It was argued that although some degree of correlation could possibly be justified, at least for the core cases, finiteness, both in its abstract/syntactic and its overt/morphological guise, appears to give wrong or contradictory results even at the descriptive level (and sometimes even for verbal forms that lie at one of the two extremes of the finiteness continuum), let alone at a more abstract leveL In a similar manner, scope ofV-movement was shown to also make wrong predictions, although it fares much better in theoretical strength. The general conclusion was that, descriptively speaking, both factors appear to play some role in the alternation, although neither of them alone could possibly account for all the existing cases. I then moved on to the detailed discussion of a number of (mood) constructions in Greek, in order to investigate how V-movement and finiteness (both morphological and abstract) affect the proclisis-enclisis alternation. I argued that the Greek data show us that proclisis vs. enclisis correlates with non-restricted (i.e. full) vs. 43· In the sense that force types appear to correlate with specific mood and agreement properties.
226 Cities in Greek: A Minimalist Account ofProdisls and Endisis
restricted (i.e. default, in the case of gerunds, and underspecified, in the case of imperatives) person agreement on the verb respectively, as well as (partially- cf. gerunds) with the [± imperative] dichotomy. I showed that this difference in person agreement marking should not be viewed as the reflex of distinct syntactic properties on T; rather, it should be associated with differences on the higher CM head, and in particular with presence vs.lack of obligatory V-movement to CM. As a result of this, proclisis vs. enclisis is linked to CM properties, i.e. to mood/modality features, rather than to T properties.44 In this sense, finiteness, at least at the morphological level, which appears to be the only level to be giving any correlations at all, becomes a PFreflex of mood/modality properties. This is not something new: in fact, it has quite often been pointed out that finiteness correlates with mood, tense and agreement on the inflected verb (see e.g. Rizzi 1997). However, what is new here is that of all these features mood/modality appears to be the one which is (more directly) associated with the proclisis-enclisis alternation.45 Now, if'finiteness' (i.e. person agreement marking) correlates with the mood/modal properties on CM, and if the latter correlate with the presence or absence of Vmovement to the C-domain, one could plausibly assume that these two are in essence the two sides of the same coin as far as the proclisis-enclisis alternation is concerned. In other words, one could simply speculate that enclisis obtains when there is V-movement to CM (under the assumption that clitics attach toT, which is more or less the standard assumption), which in turn is related to the finiteness properties of that particular head. This would reduce V-movement to finiteness, and vice-versa. However, as we have seen, there is evidence that V-to-C movement is not a sufficient condition tor enclisis: recall that in many languages V-to-C yields proclisis, and the same could possibly be the case for Greek particle-less interrogatives. What could be the reason for the different clitic positioning in each case? The natural assumption would be to associate this difference to CM. Recall that what unifies imperatives and gerunds in Greek is that they have to check a strong modality feature on CM and that this feature takes the form of mood, i.e. it is a property that shows up as part of the inflectional composition of the verb. On the other hand, subjunctives for example, which also involve an activated CM head but do not exhibit enclisis, do not have V-to-CM movement, since the non-inflectional particle na. (which, as Roussou 2006, convincingly shows has more nominal rather than verbal properties, making it in this respect a truly C element in the sense of Rizzi 1997) merges directly in the relevant head, preempting V-movement, and hence leaving
44 Of course, this doesn't mean that T properties are not related in any way to CM properties. It ls prodlsls vs. enclisis in particular that appears to be related to CM, not T (or other) properties more generally. 45· AB Adam Ledgeway has pointed out to me the primacy of mood in finiteness distinctions ls not a new idea and has been developed quite extensively by Vincent (1998). Note, however, that it ls quite new in the Greek literature.
Chapter 6. Enclisis in Greek 227
V unmarked for mood. Furthermore, we know that imperatives, contrary to gerunds, involve an additional step to CForce/Op, and that the same step is taken by na. This suggests that the second step is not related to enclisis. As a result, a reasonable hypothesis to make would be that in those cases where V-to-C movement yields proclisis, only the movement step to CForce/Op takes place,46 or alternatively that movement to CM takes place, however it is not a V-related movement (i.e. it does not take place in order to check a V-property, i.e. a feature that is part of the verbal morphology- see Chomsky 1993, Rizzi 1997). Accordingly, enclisis would reduce to V-related movement to CM, i.e. to V-movement that takes place only tor inflectionaVmood purposes. I will come back to this distinction later on. Here, it suffices to say that similar remarks have been made for Romance imperatives. For example, Paletta & Zanuttini (2003) have shown that positive imperatives in Badiotto, a Rhaetoromance variety spoken in north eastern Italy, obligatorily co-occur with special particles, which roughly speaking express point of view (and which are merged in the Spec of the relevant modal heads). They argue that imperatives in this language must check both a modal-like feature in a ModP merged above INFL, as well as a CForce head higher up in the clause. Putting details aside, the idea is that imperatives need to check a modal head before moving on to CForce, and their assumption is that the same thing must be happening in Standard Italian, although in that case no overt particles show up. Additional evidence for the postulation of this modal head comes from the semantic properties of imperatives more generally (see Portner & Zanuttini 2003), which include a modal-like operator. Following the standard assumption that semantic properties more or less mirror syntactic properties, they analyze imperative movement as V-to-M movement (they seem to dispense with CForce altogether, for independent semantic reasons, but this would not affect the argument here). Interestingly enough, this analysis is also confirmed by Greek imperatives and gerunds, which also involve V-movement to a modal-like element A further related point regards the actual trigger for V-to-CM movement. In Greek imperatives and gerunds, we have seen that this is typically attributed to modality feature checking. One could possibly argue, however, that this movement step takes place (also) for agreement checking, given that the actual movement correlates with special agreement. In this sense both modality and agreement features would be checked on CM. However, there are reasons to believe that even in these cases agreement features are checked on T, as in indicatives and subjunctives. First of all, in subjunctive clauses both agreement and modality features are checked, the former in T by the verb, the latter in CM by na. Moreover, na. moves on to CForce/Op, to check clausal type features. In imperatives we also have the same three checkings, namely agreement, modality and force. Given that na is in complementary distribution with 46. CForce/Op is used here in a rather general fushion for any clause type complementizer. Rizzi (2001) argues explicitly in favor of separate wh- and interrogative heads that are merged lower than CForce (really a subordinator in this case).
228 Cities in Greek: A Minimalist Account ofProclisls and Enclisis
imperatives, it is natural to assume that imperatives must also be checking the same (or similar) features in CM. For na these are modality features, so the same would expected to be the case for imperatives. 47 Similar arguments apply to gerunds, which however do not involve the additional movement step to CForce/Op. This hypothesis is supported by the existence of morphologically surrogate but syntactically true imperatives: these are -imperative forms, however their syntax is identical to that of + imperative forms. Given that they share the same syntax, they must also involve V-to-CM movement, however no effect in morphology arises. On the contrary, given the strong evidence that -imperative morphology typically correlates with agreement checking in T, we would have to conclude that agreement checking in surrogate imperatives also takes place in T, and that obligatory movement to CM is related to other properties, obviously to the extra modal properties of the verbal form. Presumably, if this were the case for surrogate forms, we would have to assume the same derivation for true forms, if we want to keep our syntax as simple as possible. Obviously, similar remarks would apply to Romance imperatives, which use surrogate forms across the board in the plural paradigm, borrowing them from the indicative or the subjunctive in most cases (see Zanuttini 1997 for details). In a rather opposite manner, we can also find special forms that are related to particular types of C heads (C flavors, cf. the term v-flavor proposed by Arad 1998 for v heads with different types offeatures, such as causative, agentive, etc.), which however do not involve V-to-C movement. A rather typical example is subject-clitic inversion in non-embedded interrogatives in French. This has been analyzed (although not without controversy- see Rizzi & Roberts 1989, Roberts 2007 for arguments in favor of overt movement to C in these cases, although this movement is related to a whfeature, not to agreement features) as involving V-movement toT, with further LF movement of I to C (or alternatively long distance agreement between the interrogative C and the wh-feature on I) (see Hulk 1993, Kayne 1994, Sportiche 1997 among others). In this case agreement features are presumably checked in T, while the whfeature is checked on CForce/Cinterrogative without overt movement. In this sense, the special interrogative inflection is not related to the C-movement step; rather it is a 47· Provided a non-scattering approach to features, i.e. assuming that features tend to appear in specific heads within the clausal architecture. Note that it ls not dear whether CM does not have any agreement features at all, given that it selects T (and according to Chomsky's 2005 inheritance theory it should be the head that is downloading phi-features to T). This may be true, however it does not mean that V-movement is triggered due to this agreement checking (if there is any at all). Rather, the latter seems to be related to selectional properties, i.e. it ensures that the right head combines with the right complement. Note in this respect that Badiotto, which has been discussed above, offers some additional evidence in favor of this analysis: there modality checking involves both V-movement and an overt particle at the spec of the head the verb has moved into. One could see this effect as the result of spec-head agreement (in the Chomsky 1995 sense), however what is important is that these particles show overtly what feature ls checked on the CM head, i.e. modality but not phi-features (recall that the particles are not inflected in any sense).
Chapter 6. Enclisis in Greek 229
morphological marking of the wh-inflection on I. or rather of the choice of a special type of C(force). It seems that in other Romance languages this same LF movement can be overt (in fact, this has also been proposed for French, based on the root/embedded asymmetry of subject clitic inversion in that language), which is expected under standard assumptions oflexical parametrization. This might be the case for example for subject clitic inversion cases in NIDs, which have been analyzed by Munaro (1997) and Paletta (2000) as involving V-movement to a C-Force head, presumably to check (among others) an interrogative operator feature (hence the interrogative reading). Although, as Paletta (2000) points out, nothing really prevents subject agreement from also being checked on the C head, nothing forces it, either. However, given (a) that the interrogative interpretation, which under standard assumptions involves checking of a wh-feature on I (not V- see Chomsky 1993), is uniquely related to verb movement to the interrogative C head; (b) that the same movement (and the same inflection) is triggered in the presence of other C- Force heads, such as hypothetical, counterfactual, etc., which in other languages (like e.g. English or Standard Italian) force V-to-CForce/Op movement (cf. deletion of complementizer); (c) the fact that the same movement obligatorily takes place in those NIDs which also have a true interrogative meaning but which do not involve a special inflection (rather, the enclitic subject clitic seems to be simply by-passed by Von its way to C-interrogative), suggest that V-movement to C might be better analyzed as force-related movement. As such, it would not be inflectional in these languages, a fact which may also be related to the observation that we do not seem to be able to find a true interrogative (or hypothetical, counterfactual, etc.) inflectional marking on any of these verbs. The only inflection we find is the typical mood inflection (indicative in most cases, but subjunctive is also possible in some Romance languages -see Aux-to-Comp constructions in Standard Italian), which is typically lower than T in these languages (see Laka 1990, Harris 1995 tor Spanish, Zanuttini 1997 tor Italian, Cinque 1999 for a number oflanguages including Romance languages), as opposed to Greek, as well as the subject clitic inflection, which is subject agreement related to/selected by a special C flavor (namely interrogative). In other words, if there is movement to C in these constructions, the latter would not be V-related. Although there is no absolute argument in favor of this approach, it nevertheless makes it possible to account for both interrogatives, and imperatives (and gerunds for the case of Greek) in a uniform way, accounting also for the fact that all these structures that involve force related C-movement take proclitics. According to this approach, if there is special subject agreement on the verbal form, this is not related directly to VII-to -C movement; rather, itis the morphological reflex ofT-selection by a special higher C head, with the actual agreement features being located (and checked by the verb) on T.48
48. Note that it is not dear to me whether the special form is related to the actual content of the agreement features downloaded toT by C (cf that in most cases the same agreement features with indicatives and subjunctives appear to be present on T, independent ofV-to-C movement
230 Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
All in all. the main point of the discussion above is that although V-to-C movement may co-occur with special subject agreement morphology on the verb. the (abstract) agreement features are checked on T, whereas C involves modality and force feature checking only. This appears to be rather straightforward for Greek imperatives, gerunds and possibly non-negated root interrogatives, as well as for Romance imperatives, but less so for Romance interrogatives involvingV-to-C movement Accordingly, special agreement is linked to special properties of the selecting C head (which in most cases appears to be CForce/Op (ct~ e.g. imperatives vs. interrogatives/declaratives), although it could be CM if no CForce is projected (cf. e.g. gerunds)). This could either involve different types of agreement features on T (courtesy of the higher C head), or the same types of features but with distinct morphological realization (presumably, due to the fact that a single vocabulary item in the PF component can be used to realize agreement, modality and CForce features at the same time, as opposed to the use of null morphemes for other features than agreement). The generalization that seems to emerge for the particular Greek and possibly Romance constructions is then as follows: enclisis obtains when V moves to C and this movementis V-related (this notion will be discussed in detail below), otherwise proclisis obtains, independent ofwhether V-movement to C has taken place or not The obvious question is whether C is an absolute value in this case, or whether it can refer to any head that is higher than/at the clitici:zation site, i.e. whether this is a more general constraint, possibly relevant to all proclitic-enclitic environments. Given the discussion in Section 2, C should only be a subcase of possible heads. As a result, the generalization could become stronger: enclisis obtains when V-movement across/to the cliticization site is V-related. For the time being I will concentrate on the C-head cases, however I will come back to the other cases (involving lower cliticization) later on. Although this generalization (in its less strict form) seems to hold for the Greek cases, as well as tor some Romance cases, it is still not dear to me whether it holds tor other cases as well. Nonetheless, it is a testable hypothesis, as it makes a number of predictions: e.g. it predicts that if V-movement to CForce/Op in some language is V-related, enclisis should obtain (unless clitics are merged higher than CForce/Op for independent, yet unknown, reasons). Since I am not aware of any such language with Romance type or Greek type object elitic pronouns, the testing of this prediction must await further research.4 9 According to this analysis V-related movement across the clitici:zation site appears to be crucial for enclisis to occur. ln this way one may capture the role of finiteness and to a spectalhead), or to selection by a higher special C head (which could involve Agree (although see Landau 2007 for arguments against unifying selection with Agree based on the EPP), which could be marked by special subject agreement. I will leave this issue here. 49· It is possible that this could be the case in Pattern B languages, like European Portuguese or Cypriot Greek, however further research ls needed before any definite conclusions are reached. Thanks to I. Roberts (p.c.) for suggesting this to me.
Chapter 6. Enclisis in Greek 231 the role ofV-movement in one go: V-movement plays a role only if it satisfies inflectional requirements of the verbal host. The position of the clitic seems to be relevant only as far as the V-related position is higher than (or possibly as high as) the clitic position (following standard assumptions of the V-movement approach). Inflection here does not refer to overt marking on the verb: as I have pointed out above, it seems that restricted vs. non-restricted marking is nothing more than an indirect reflex of what is going on in higher heads. Rather, it mirrors specific properties of the selecting C head on T (cf. the special properties of imperatives, which are in accord with the featuraL semantic and pragmatic properties of CForce-imperative more generally). This means that CM checking and whether it is V-related or not is not dependent on overt marking. Although in some languages marking could correlate with CM checking (see for example special modality particles), this is not necessary in any way. As a result, it is the overt obligatory movement of the verb to a V-related position that is crucial, and not the marking on the verb as the result of this movement Of course, one needs to better define when some movement is V- or not V-related. For most cases, the traditional distinction between INFL and C heads would suffice, however we have just seen that this cannot be true for all cases. Moreover, we need to establish how exactly V-related movement above/toT, in the structures discussed, affects clitic positioning. If clitics attach to a functional head in the clause, to which the verb (or the functional head the verb is attached to) may or may not attach as welL how is proclisis vs. enclisis derived? Finally, we need to say something about lower cliticization cases: in Section 2 it was pointed out that languages allowing interpolation offer us evidence in favor of a special head at the level of which incorporation seems to take place. lt was pointed out that sometimes an enclitic or a proclitic structure seem to be already formed at a level lower than T. If this is true, how can these structures fit in into the larger picture? In what follows I will discuss these and other questions, and I will attempt to offer an analysis that can capture the generalizations formed based on the Greek data. 6.3.2 The analysis 6.3-2.1 Assumptions In this section I will discuss the assumptions I follow in my analysis. These basically involve triggers for V-movement (EPP or non-EPP related), as well as the conclusions I have reached in chapters 2, 3 and 4 regarding the phrase structure, categorial and featural properties of clitics in Greek, as well as the derivation of proclisis, which I take to be the basic configuration for reasons that I will shortly explain (see also Kayne 1991, 1994). First of alL following Chomsky (2000, 2001a, 2005) I assume that the EPP is a generalized Occurrence (OCC) or Edge feature (EF) that forces the projection of a specifier at the edge of the relevant head. In this sense, it does not specifically reter to the traditional EPP (i.e. the requirement for an overt external subject at Spec TP), but it applies to any functional head (ct~ Core Functional Categories ( C, T, V'") of Chomsky 2000). Furthermore, I assume that this EPP/edge feature is added optionally to a phase
232 Cities in Greek: A Minimalist Account ofProdisis and Endisis
head (i.e. C or v* 50), triggering internal merge at the edge of the head/projection of a specifier, without requiring feature matching (although see Rizzi 2004 for a different approach which requires agree of criteria! features, even for intermediate positions). Internal merge/ movement to the EPP probe has an effect on the outcome, giving rise to semantic (LF) (for example scopal properties in the C domain- see Rizzi 2004, or specificity/ in lower v*- see Chomsky 2001a) and phonological/morphological (PF) effects (for example overt displacement/pied-piping of features, incorporation of deficient elements, prosodic restructuring). In addition, an EPP/edge feature (possibly in the guise of a person/D feature- see Chomsky 1995, 2001a) is also obligatorily present (maybe even cross-linguistically) on the/a lower phi-probe (typically on T/V), which requires phi-feature agree and which has a PF effect (triggering overt movement to T/V) and/or an LF effect (e.g. predication on T) (see Chomsky 2005 on EPP inheritance51 ). Note that both A- and A-bar checking are properties of the probe. This follows from Chomsky's (2005) assumption that phase heads carry phi-features and an edge feature, triggering parallel scanning of lower potential goals in the complement domain of the probe. Phi-feature (and possibly edge feature) inheritance toT and V respectively gives the traditional NA-bar distinction. Based on this distinction, I will informally call the EPP/edge teature on C/v* an 1\-bar EPP: as opposed to the EPP/ edge feature on TIV. which I will call the 1\-/phi-probe EPP'. I further assume, following Alexiadou & Anagnostopoulou (1998), that the EPP property on T (i.e. the traditional subject of predication requirement) is universal, but can be parametrized in terms of how it is satisfied: via merger (internal or external) of an XP constituent or via merger (internal or external) of an X0 constituent Null Subject languages (NSLs) appear to be able to satisfy this requirement by X0- movement of the verb, at least in most cases. This is so, because V presumably contains a nominal element (i.e. a [+ person/ D] element), namely the subject agreement phi-feature bundle, which is independently listed in the Lexicon (in their terms, it is strong) and which may check the D-/EPP feature on the T head. Note, in this respect, that it is not clear that all NSLs behave in the same way across the board, in the first place (see in this respect Biberauer et. al. 2009). The main reason for this is that there seems to be evidence that also NSLs project a specifier at T, either null (e.g. pro, full or expletive), or overt (e.g. quantifiers); ifV-to-T movement checked the EPP, then [Spec TP], if projected at all, would be an A-bar position and hence should not be able to host a subject (see Sheehan 2007 for an overview of these matters; see also Roussou & Tsimpli 2006,
so.
Here such labels as C, v", T, V are used ln the sense of Chomsky (1995, 2000, 2001 a, 2005, among others), i.e. as collective labels that define separate domains. Apparently, each head may be decomposed into a number ofheads, depending on the features it contains (the cartographic approach). Alternatively (but not equivalently), one could assume multiple Specs. In the general case I will be abstracting from this difference, although I will specifically make reference to it whenever it is crucial for the analysis. The alternative is simply to assume that phi-probe triggers overt movement, as ln Chomsky (2005). In that case, however, the EPP should be always present on the phase head
51.
Chapter 6. Enclisis in Greek 233
Revithiadou & Spyropoulos 2007b for a discussion on Greek). Moreover, some researchers have pointed out a number of theoretical problems for Alexiadou & Anagnostopoulou's (1998) account (for example the possible intervention of the subject DP) (see Gallego 2006, and references therein). Here I do not concern myself with the status of [Spec TP], nor with technical issues. 52 Therefore, I will keep on assuming that V-movement to T/Agr is triggered by a phi-probe EPP. However, I will assume that subject agreement is a PF-reflex of EPP-checking, not of case checking (see Alexiadou & Anagnostopoulou 2002, pace Chomsky 2000, 2001a, 2005). Finally, with regard to head movement more generally, I assume that it is syntactic (see Matushansky 2006, Roberts 2006, Lechner 2007, among others) and that it is driven either by means of a generalized V-feature (with the EPP property) or by specific features (such as T/vN) that need valuing (and also have the EPP property) (in the sense ofPesetsky &Torrego 2001). This does not involve V-to-Tin NSLs, which as I said appears to be a special case, involving D-/person feature I/agreement checking. As far as object clitic pronouns are concerned, I will follow the assumptions made in Chapter 4, namely that they are XPs/X0s (i.e. minimal and maximal) (cf. Chomsky 1995) and D/DPs, merged in the object theta positions and moving in the syntax (see also Kayne 1975, Anagnostopoulou 2003a, Matushansky 2006, Mavrogiorgos 2006, Roberts 2006, 2009). Structural deficiency allows (and in fact forces) clitics to incorporate as soon as this becomes structurally possible (see also Cardinaletti & Starke 1999 ). Movement of clitics is feature driven, and hence blind to any post-syntactic effects. This means that they cannot be moving for semantic (cf. e.g. Diesing & Jelinek 1995) or phonologicaVmorphological reasons (cf e.g. Kayne 1989). Rather, any PF/LF effects will follow from their structural position (see Richards 2004, and for the more general approach Chomsky 2001a). Finally, let me repeat here very briefly the analysis for proclisis that I have proposed in Chapter 4. According to the analysis, clitics are merged in the complement position (in essence they are introduced by V /v heads, on a par with full DPs ). A direct object/accusative clitic is probed by the phi and edge features on v* (presumably the former have been downloaded to a lower V head), checking v*'s phi-features andreceiving case.53 The clitic also moves to v*min (i.e. to the root), targeting the specifier of 52. Note, however, that ever since the proposal of Cardinaletti & Roberts (1991) that there may be more than one subject position in the clause (see also Cardinaletti 2007), these problems could easily be catered for (see also Barbosa 2001, 2006), without having to dispense with the X-movement approach altogether. 53· For genitive clitics see Chapter 4. However, let me here remind the reader that I adopt the analysis of Anagnostopoulou (2003a) according to which genitive clitics are merged in the spec of a v-applicative head right above V and right below v*. I assume that in ditransitive constructions both clitics are probed by v* moving to the periphery ofv*P. For unaccusative or nontransitive verbs (e.g. in cases of ethical datives) genitive ditics are again introduced by a v-applicative head, however they are not probed to the highest v; rather, they incorporate to the v-applicative, the latter moving along with V to higher heads.
234 Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
the relevant head. Furthermore, I have claimed that there is no separate head movement/ cliticization to the verbal host (or to a higher head that is also targeted by the verb). Rather, this is an epiphenomenon of incorporation (see also Matushansky 2006; Roberts 2006, 2009), which takes place between the clitic and v\ and which ensures that the clitic will piggy-back on v*, moving along with it toT (cf. Diesing & Jelinek 1995 on a similar in spirit analysis of Egyptian Arabic enclitic pronouns). Incorporation takes place, given that the elitic contains a subset of the features carried by v* and that the edge ofv* is still accessible/has not been sealed off yet Since in most languages with object clitics V-v* moves to T the surface position of the clitic tallows straightforwardly. 54 Lack of clitic movement to T basically follows from the fact that the object clitic cannot be checking an EPP/phi- feature on T (T's phi-features probe the subject agreement morpheme and trigger overt V-to-T movement) or on C (the clitic does not appear that high in the clause, and moreover it does not have semantic properties typical of the CP domain, at least in languages like Greek). i.e., only features within the vP phase seem to be relevant for the object clitic. Assuming that movement is syntactically triggered and that there is no principled reason why v* should wait until it moves toT in order to start acting as a probe (although see next paragraph), it follows that clitics must check their phi-features and move already as early as the vP phase. Such an analysis captures among others: (a) the similarities between object clitic movement in languages like Greek or in Romance languages, and object shift/ scrambling in Germanic languages (see Alexiadou & Anagnostopoulou 1997; apparently any differences would be related to the (un)availability of incorporation and piggy-backing in each language type respectively- see Cardinaletti & Starke 1999). In this respect, at least direct object clitics are not taken to be pure agreement markers (cf. that although they spell out the phi-features of v*, they also have semantic properties- see Sportiche 1992/1998); (b) the obligatory character of subject agreement as opposed to the optional character of clitics (although see Alexiadou & Anagnostopoulou 1997). This difference would follow from the obligatory nature of phi-probe related EPP (cf. that subject agreement is checking T's phi-features, moving obligatorily to T), as opposed to the optional EPP nature of phase heads (i.e. the object agreement elitic is added only if it has an effect on the outcome). As Anagnostopoulou (2003a) has convincingly shown, cliticization does not appear to be linked to specificity in all languages. E.g. they may also act as animacy or[+ human] markers. Note that specificity/ topic reading seems to differ from these latter readings: specificity appears to be related to a structural position that is outside the VP crosslinguistically (see Diesing 1992). On the other hand, animacy or [+ human] are independent lexical properties of D Ps which
54 In fact this should not be allowed, given the Earliness Principle, or some other rule that ensures cyclicity. Note that in the phases framework (Chomsky 2001a), a strong EPP feature must be checked as soon as possible (see also Pesetsky & Torrego 2001). As a result, if clitic movement to the vP periphery ls EPP-driven, the clitic must move to the edge of vP before the next head ls merged (or, alternatively, the next phase starts).
Chapter 6. Enclisis in Greek 235
may or may not trigger agreement, depending on the properties of v* (the available v* flavours, i.e.) in each language. If the elitic does not move out of the vP, then deviance occurs for two reasons: LF crash (no interpretation) and PF crash (no host). On the other hand, in languages where (direct) eli tics appear to be agreement markers (this would also apply for indirect clitics in many languages with non-agreement direct clitics- see Alexiadou & Anagnostopoulou 1997), like e.g. Albanian, cliticization may not involve an EPP feature on v* (only on V, or alternatively it would only involve phiprobing by V (see Chomsky 2001a)). Note that according to such an analysis, the presence of the clitic inTis contingent on verb movement (i.e. [V-v*] movement) to T. This predicts that whenever v* does not raise to T the clitic will also appear in a position lower than T, and that whenever it does move toT, the eli tic will also appear in T (at least in languages in which clitics incorporate into their host). This seems to be generally true, as can be seen e.g. from languages with low cliticization and low verb movement 55 Moreover, it predicts that clitic movement/licensing is contigent on the presence of a v*/v type head: if the latter is not present in a clause, then a clitic cannot possibly occur, unless it can be probed by a higher suitable head. An alternative idea which would give similar results, at least as tar as cliticization is concerned, would be Gallegds notion of phase sliding (see Gallego 2006), which is also contingent on V-movement toT in Null Subject Languages. In particular, Gallego argues that in Null Subject Languages obligatory V-v*-to- T movement yields reprojection ofv* in T (i.e. v*/T). As a result, v*/T becomes a phase head, in the sense that it inherits the (optional) edge feature ofv* (but not its phi-features, which stay in the VP domain). This would force the object clitic to check its phi-features inside VP but move to v*/T due to the edge feature. This would immediately explain why object clitics in languages like Greek or Italian, with v- to-T movement, end up in T. Given that 55· The question arises what happens in languages where the diticization head appears to be separated from the verbal head by a number of nodes (interpolation). It appears that in these cases the verb targets a position that is located lower than the one targeted by the ditic. Given, however, that the verb must at least have moved to a v*-type head (see Chomsky 2006) crosslinguistically, we would have to assume that the v* domain may also form an extended domain (on a par with C). In that case, the ditic would move to the periphery of the top-most head of this domain, as opposed to the verb which would stay lower. This is of course a hypothesis and needs to be tested on empirical grounds. On the other hand, a separate problem emerges with languages like French, which do not allow the (finite) verb + ditic complex to stay lower than T. This however is not surprising, given that at least in standard French the verb must always appear in T, as opposed to languages like Greek or Italian, in which the verb normally appears in T but may also appear lower. Accordingly, we could simply assume that in both French and Greek the ditk ditictzes on the verb at the edge ofvP, but that for independent reasons the ditic+verb complex must move to T in French, whereas it may move to T in Greek and Italian. Given that V-to-T movement appears to be syntactically motivated also in Greek/Italian (see Alexiadou & Anagnostopoulou 1998), we would have to assume that this effect is PF driven (i.e. it is related to the pronunciation of higher vs. lower copies of the d-V complex in each language).
236 Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
the creation ofv*/T is dependent on v-to-T movement in syntax, and that the edge feature is a property of v* (and not of the C-T phase), this analysis makes similar predictions to my analysis as far as clitic movement to T is concerned. Whether they give the same results for enclisis is something that I will discuss later on. 6.3-2.2 CM and enclisis
Many analyses on enclisis assume that either proclisis or enclisis is the basic construction, the idea being that the marked case appears only if the default construction leads to a crash. This is for example the idea put forward by Shlonsky (2004), who assumes that enclisis is the default construction, proclisis being formed only if enclisis would lead to a crash. Obviously, such an approach is reminiscent of the last resort operations that have been proposed in the literature, such as Long Head Movement, and V-to-C movement in case ofTobler-Mussafia or Wack.ernagellanguages, among others. Other analyses argue that the two structures differ from each other in that one is more deficient than the other. This idea tries to capture the intuition that finiteness is related to the proclisis-enclisis alternation, enclisis being primarily found with non-finite/deficient clauses. It seems, then, that each type of analysis looks at the relation of proclisisenclisis from a different point of view, depending on their assumptions as well as on their particular goals. From a purely derivational perspective, like the one advocated in the Minimalist Program, neither of the two constructions can be assumed to be more basic than the other; rather, the alternation itself must follow from the interaction of universal operations such as merge and Agree, and the featural content of the lexical items selected in each clause.56 In this sense, the idea that clausal finiteness could be related to the alternation gains support. But how and why should finiteness play a role in the alternation? First of all. in the previous sections we reached the conclusion that finiteness does not seem to be related to the alternation, at least directly: on the abstract syntactic/ semantic level no correlation appears to be possible, while on the morphological level there is a correlation between restricted vs. non-restricted person marking and the alternation. However, the latter distinction seems to be associated to properties of the CM head and to the presence vs. absence of verb movement to that particular head, rather than to any of the features on T. Given, though, that CM encodes mood/modality and that the latter has also been considered to be one of the components of finiteness, one could assume that this difference in CM properties is the part of finiteness that plays a role in the proclisis-enclisis alternation (presumably, other 'parts' of finiteness will play a role in other linguistic phenomena). But again, the question is why? I.e. why does CM play a role in the alternation? Given what we said above on proclisis, the null hypothesis to make would be that enclisis does not differ from proclisis in terms of phi-probe checking: i.e., also in 56. Of course, further non-syntactic constraints may apply that may change the outcome of the syntactic derivation, as fur example has been argued for languages llke Bulgarian or Serbo-Croatian.
Chapter 6. Enclisis in Greek 237
enclisis the object clitic will check its features against those contained in the phi-probe of the vP phase. This is further supported by the simple fact that in enclitic contexts object clitics are still case marked and interpreted as objects, and not as subjects (cf. nominative case or a subject of predication interpretation). Moreover, they do not appear to participate in any phi-feature related co-restrictions with the (abstract) subject agreement features, as the two object clitics do with regard to v*'s phi-features (cf. the Person-Case constraint- see Anagnostopoulou 2003a, 2005b), which suggests that the subject and the object do not check features against a single head (namely, T). In other words, enclitics appear to be as vP-related as proclitics are. If this is the case, one wonders how finiteness could possibly affect their positioning, if finiteness is typically related to properties of the C-T domain. If phi-features cannot be logically related to the alternation, then the EPP/edge feature could possibly be, given that it is the EPP that drives movement in this framework. Possible EPP/edge feature targets could be v*/V, or CIT. Again, the edge feature(s) on v*/V would not be expected to differ from proclisis, given that the enclitics are also optional and receive a specific/topic reading, which mark movement out of vP. Assuming that movement may only proceed through phase edges (see Chomsky 2001a, 2005), it is natural to conclude that the EPP teature(s) in the v*P phase have exactly the same properties in enclisis as in proclisis. The only case where this would not be true would be in structures where one of the two heads (logically v*) were missing from the structure altogether.57 This could plausibly be the case in restructuring infinitives in Italian and Spanish (see Wurmbrand 2003). We have seen above evidence that both imperatives and gerunds involve a rich clausal structure, and that their vP domains do not appear to have any differences in comparison to finite clauses proper. As a result, these structures could not be analyzed on a par with restructuring infinitives. Moreover, it is logical to assume that the enclitic is also not related to the edge property on C, given that there does not seem to be any related interaction between those two heads, at least in Pattern A languages (e.g. peripheral features/ semantics in the sense of Chomsky 2000). Another possibility is the EPP feature on T. This feature makes a good candidate, because it is directly related to the phi-probe properties ofT, and as a result with finiteness. Nevertheless, recall that this particular EPP is related to phi-probing, and we have already seen evidence that this cannot be the case for object enclitic pronouns. Imperatives and gerunds in Greek could give us a hint as to which EPP feature could possibly be related to enclisis. Recall that the traditional EPP maybe satisfied by verb movement toT (the IN FL head) in NSLs. What is characteristic of this movement is that it is obligatory - there must be raising to the INFL domain in Greek and other similar languages. This is exactly the property of phi-probe checking in Chomsky's (2005) system: phi-probes induce obligatory movement to the relevant head, V or T. 57· Or, alternatively, from v* already being lexicalized by infinitival marker or final-e, as Adam
Ledgeway points out to me.
238 Cities in Greek: A Minimalist Account ofProdisis and Endisis
The interesting part is that in imperatives and gerunds the same property is found slightly higher, in CM. I.e. CM takes on the phi-probe characteristics ofT, triggering obligatory verb movement 58 This is further supported by the properties of the CM head in these constructions. I.e. typically modality/mood is taken to be a property of C (cf e.g. Chomsky 2000). Rizzi (1997) on the other hand claims that mood is an IP/finiteness property (in the sense that only finite verbs normally mark mood distinctions) which is related to/ selected by an appropriate C. Moreover, he assumes that only V-related features/heads, i.e. features marked on the verbal form, are IP teatures/heads. As we have seen, Roussou (2000) argues that the modality head merged above T must be a C head, based on the observation that it typically hosts un-inflected modal particles. However, as we know, it may also host mood-inflected verbs, namely the [+imperative] forms as well as gerunds, which also have special modal marking. This has led for example Philippaki-Warburton (1998) and Giannakidou (1998) to argue that Mood must be an IP head, given that it is realized on the verb in imperatives and gerunds (for[- imperatives] they assume a zero morpheme in Mood, on a par with overt modal particles). As a result, CM has I properties in some structures but C properties in others. A fairly natural proposal would be that CM is a V-related head in imperatives/ gerunds (i.e. an IN FL head, following Rizzi 1997), and a non-V related head (i.e. a C head) in indicatives/ subjunctives. Now, if obligatory V-movement to Tis only found in indicatives/subjunctives, while obligatory V-to-CM movement is only found in imperatives/gerunds, it follows that the locus of phi-probe EPP in imperatives/ gerunds is distinct from that of indicatives/subjunctives (i.e. CM vs. T). This holds even if we assume that agreemenVtense (if present as abstract features) are always checked on T: mood/modality will count (also) as a 'phi'(= I) feature, but only in imperatives/gerunds.59 Following this line of 58. Elena Anagnostopoulou asks what we do with cases where the d- V complex follows adverbs llke ldhl, i.e. it stays low in the clause. As we have seen, this is indeed possible giving rise to optionality in the movement of the verb. 0 ne idea would be to assume that in all cases the verb moves to the highest posslble position, however it may be spelled out by PF at various positions within the clausal structure. According to this analysis, the EPP on Tor CM will always be checked by the verb, even if the latter appears lower in the structure. 59· In this respect, recall that imperatives only have deficient agreement marking (which nevertheless distinguishes them from-imperative forms: i.e. it is agreement that marks imperatives as +imperative, not an imperative marker), while gerunds have zero agreement marking and a default gerund ending. On the contrary, both indicatives and subjunctives have full agreement (and in many cases tense distinctions): here agreement marking is not used to mark mood Interestingly enough, V-position correlates with agreement/mood marking: agreement on a verb that moves to T is always full and doesn't mark mood, whereas agreement on a verb that moves to Mood is deficient/zero and does mark mood However, it doesn't necessarily follow from this correlation that agr features are on Mood in imperatives/gerunds: it could be the case that in both cases agr features are on T, their differences in use/marking stemming from distinct properties of C-domain heads (force and mood). Nevertheless, it is indicative of the phi-nature of mood in imperatives and gerunds.
Chapter 6. Enclisis in Greek 239
reasoning, we could further assume that full person agreement is a reflex of a phiprobe EPP checking on T. while restricted person agreement (as well as full mood agreement) is a reflex of a phi-probe EP P checking on CM. In this way, we would capture the correlation between marked vs. unmarked agreement, position of the verb and distinct properties of CM in each structure. So, using GB terminology, the EPP-related difference between proclitic and enclitic structures would be associated to the INFL (A) vs. C (A-bar) status of CM. Note, however, that this EPP feature is related to the verb, not to the clitic. This means that the clitic could not possibly change position on its own; rather, either the verb alone would change position, or both the verb and the clitic would, given that clitics attach to v*, a subpart of the verb. Clarification of these parts would probably give us some insight into what affects clitic positioning in these cases. If the EPP on CM is only related to verb movement, as seems to be the most plausible hypothesis, this amounts to saying that V-movement is the only relevant factor. This however is problematic for two reasons: first of all, because we have seen that only this particular EPP is linked to enclisis (recall that optional EPPs on CForce heads (optional in the sense that they are optionally included in the numeration and have an effect on the outcome, tor example giving rise to a question) do not induce enclisis (the V-movementapproach problem)). Second, because ofwhat I call the attachment problem. If clitics left-attach to v* (or to some v*P-phase related head), and assuming that excorporation is not a possibility (see Kayne 1994 60), V-movement alone could not possibly affect clitic positioning: we would still get proclisis. This is always a problem for those theories of enclisis that assume that both clitics and verbs attach onto the same head: in enclisis clitic movement must precede V-movement (to the same head), whereas in proclisis it is the other way round. This may not be that much of a problem tor theories assuming distinct heads (like the one by Kayne 1994, or Rivero & Terzi 1995), however they must find some way to allow head skipping, and we have already seen that such approaches may face problems. Kayne ( 1991) and Rizzi ( 1993) have proposed two of the most popular theories on the proclisis-enclisis alternation. Both of them use V-movement and finiteness in order to get the right order, although Kayne's theory makes use of the separate heads/ landing sites approach, while Rizzi's theory makes use of the same head approach. Rizzi argues that this is required, given that interpolation is never allowed with enclitics; assuming that in enclisis the verb simply crosses over the clitic would leave open the possibility for interpolation (see also Beninca & Cinque 1993, on whom he heavily draws). Although I am not convinced that this is necessarily a problem, let's grant this point and see where it leads us. Both of them assume that clitics (and verbs) do not necessarily involve the same landing site in all clauses and across languages. In this way they can account for cross-linguistic differences (e. g. French vs. Italian infinitives), as well as for differences within the same language. Kayne argues that in enclisis the 6o. See Roberts (1991b), Rooryck (1992).
240
Cities in Greek: A Minimalist Account ofProdisis and Endisis
elitic moves to the same head as in proclisis, namely I. If non-finite verbs do not have any features to check on I, given that they lack any subject/tense inflections (a proposal attributed to Esther Torrego), then they would not need to move there in the first place. If, moreover, they happened to move to a higher head than I, then enclisis would follow. Presumably, no intervention would arise by I, given that it is not relevant to the verb in these particular clauses. In this theory, finiteness directly affects V-movement, which alone suffices to render enclisis. For Rizzi the problem must be tackled in another way. Since he assumes that both the verb and the elitic target the same position, he needs a timing mechanism, i.e. some device to control which movement takes place first, V- or clitic movement. He argues that finiteness, in terms of overt morphological marking, is the relevant device. ln particular, he proposes that enclisis takes place only under two conditions: (a) V-movement at least to the cliticization site; (b) V must have fully checked its features by the cliticization site. I.e. the idea is that when the clitic and the verb manage to reach the same position (which is not always necessary- cf. interpolation), proclisis vs. enclisis is decided on finiteness; namely ifV has checked all its morphological features by then, enclisis obtains. If it hasn't, proclisis is the only option. He argues that this difference stems from the assumption that affixes and clitics involve distinct modes of incorporation, namely head substitution into an m-slot provided by an affixal head, and head adjunction respectively. Accordingly, mixed chains are not allowed (on a par with XP-movement), in the sense that once adjunction starts, substitution is no more possible. This constraint on head chains has as an immediate result the fact that proclisis is required whenever the verb has not checked all its features by the cliticization site. Assuming that the latter is typically AG Ro, it follows that finite verbs, which need to check features at the higher AGRs, will always require proclisis. On the other hand, enclisis obtains when the verb is non-finite and hence does not have to do any more checking already at the AG Ro level. Proclisis is not allowed in these cases because it is assumed that the clitic has features to check against AG Ro, and the verb would act as an intervener between those features and the clitic.61
61. Note that Shlonsky (2004) follows the same account, however he uses Relativized Minimal-
ity for all cases (i.e. both for prodisis and for enclisis), the idea being that enclisis obtains only when the verb does not have any more features to check already at the cliticization site, presumably because the clitic would act as an intervener. The problem with this account is similar to that of Rizzi's: it is not dear why the ditic would act as an intervener of verbal properties such as negation, focus, question or tense, assuming that Relativized Minimality can be formulated in terms of features (see Rizzi 2001). Moreover, it makes wrong predictions: for example it predicts that Greek gerunds should take proclitics, given that they are negated, however this is definitely wrong. Problems also arise with his analysis of positive imperatives (which, he assumes, do not contain any agreement features). More generally, the analysis appears to suffer from the same problems with Rizzi's and Kayne's analyses, Le. the assumption that there is a 1:1 relation between syntax and morphology.
Chapter 6. Enclisis in Greek 241 More generally, both analyses face problems, since they are spelled out in an older framework, many assumptions of which (e.g. existence of separate Agr heads, the distinction between substitution and adjunction for heads) are no longer used (or are even definable). However, besides these rather technical issues, these analyses face two important problems which have immediate empirical consequences: (a) they use finiteness with the assumption that morphology and syntax map onto each other isomorphically; (b) they are not selective enough in the finiteness features they assume may be relevant to the alternation (i.e. they assume that any verbal feature may be relevant to the alternation). Rizzi's analysis also has an additional problem, namely that it is not clear on what grounds the clitic or the verb moves first in each case. As a result, his timing mechanism is representational rather than derivational in character. The immediate outcome of these problems is that both analyses make wrong predictions, and as a result, they need to have recourse to additional stipulations in order to change the initial prediction. To illustrate, Rizzi's analysis predicts that imperatives must take proclitics, given that the cliticization site is AG Ro and imperatives have overt subject agreement. In order to account for enclisis, he assumes that in this particular case AGRo cannot be selected, because imperatives do not contain aT node. As a result, the clitic chooses a new cliticization site, an AgrC head, at the level of which the verb has already checked all of its features, and hence no problem arises. Although it is indeed possible that clitics could have a higher cliticization site in imperatives, the assumption that AGRo is not projected is a mere stipulation, given that enclitics are as accusative marked as proclitics. In addition, there are structures which have aT node (like some infinitives or gerunds) but nevertheless take enclitics. It is not clear why in this case AGRo selection would be affected, yielding enclisis. Similar remarks can be made for Kayne's analysis, although the latter is more compatible with a derivational system. As we saw, for Kayne enclisis is derived simply because the verb does not need to pass through I. Besides the issue with the lack of interpolation in enclisis, there is an additional issue, that of feature selectivity. In particular, it is not clear why a non-finite verb does not have to check features on I, given that even non-finite verbs may have a tense node and a tense operator. Presumably, this follows from the syntax = morphology equation, which was generally assumed in the '90's at least for inflectional morphology. Moreover, full-agreement controlled subjunctives in Greek do not have tense features, however they take proclitics, even though they should take enclitics according to Kayne. All in all, it seems that not any kind of morphological marking and/or syntactic feature is relevant to the alternation. This was also the result of our investigation, and here we can see that this problem forces both Kayne and Rizzi to make wrong predictions. Presumably, Rizzi's analysis is less restricted than Kayne's, the former taking any feature marked on the verb to be relevant, as opposed to the latter who only talks about features on I (i.e. tense, and possibly agreement). Based on minimality, as well as on the fact that clitics are feature bundles, we would expect only phi-features to be involved in the alternation. Given the traditional intuition/assumption that non-finite
242 Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
clauses do not have subject agreement features, one could hypothesize that the presence of such features forces proclisis, because otherwise the clitics would act as an intervener between the verb and those features. However, we have already seen that nonfinite forms may also have abstract agreement features, hence it is not clear why an enclitic would be an intervener in these cases as well. More generally, though, it is not clear how these analyses would capture the role of the phi-probe EPP carrying CM in enclitic constructions. For Rizzi, this would trigger proclisis in all cases (assuming the cliticization site is lower than CM- although judging from his treatment of imperatives, he might assume that the clitic site is higher, but again this would not capture the special role of CM, as the lower T could also do the job without problems), whereas for Kayne it would simply offer an explicit trigger for moving the verb higher than T (however, it wouldn't give a causal role to CM). Therefore, although these analyses seem to be intuitive in taking both V-movement and finiteness to play a role in the alternation, they suffer from several theoretical and empirical problems, and in addition they do not seem to be able to capture the generalizations put forward by the Greek data in an adequate way. Lefs see then how my analysis would work. The immediate problem is again the timing mechanism, i.e. the device which will torce V to move second, preceded by the clitic. Recall that I have said that proclitic and enclitic v*P phases do not seem to differ in any respect This would entail that in enclisis clitics also move to the periphery ofv*, incorporating into it However, if the verb moves first, followed by the clitic, on a par with proclisis, V-movement to higher heads, and finally to CM would not affect clitic positioning in any way, given that the latter is already incorporated into v* and excorporation is not allowed. Hence, neither enclisis would be derived, nor the role of CM would be accounted for. For this reason, we need to find a mechanism to force the verb to move second. 62 One idea put torward in Roberts (2006) is that imperatives (as well as infinitives in Italian and Spanish) check their morphology already inside VP (i.e. lower than v*). As soon as v* is merged, it attracts the clitic first, given that the verb does not have any more features to check. Further V-imp movement to the cl-v* complex (presumably triggered by a V-feature on v*) will result in enclisis. The basic problem with this account would be threefold: (a) it is not clear that imperative morphology lies inside the VP, first because we never see this morphology (only subject-agreement is overt on the verb), contrary to the infinitive morphology, which is visible, and second because there is compelling evidence that this morphology lies at 62. Chomsky (1995) argues that heads must generally raise before elements of their domain (Le. before XP constituents). This means that V-movement to v* would have to take place before ditic movement to the same head. This problem could be overcome by assuming that ditic movement ls an instance of head movement already at the v*P phase. Presumably, such an account would have to explain the A-properties of ditic movement, as opposed to other types of head movement, which do not have dear A-properties. Possibly, a theory which assumes that also head are A- or A-bar, or which takes head movement to be equal to XP movement (see e.g. Matushansky 2006, Roberts 2006) might do the trick
Chapter 6. Enclisis in Greek 243
a much higher head, namely CM (cf. discussion on imperatives above); (b) it is not clear that the imperative form does not have other features to check, such as subject agreement, modality, etc. To assume that imperatives are bare vPs would not work, given that imperatives can be negated, even when they are true (cf. e.g. Bulgarian, etc.). Similar problems would arise for gerunds, as well; (c) it would not capture the role of CM in the alternation. As a result, unless we can find a principled way to force clitic movement to v* first, the timing mechanism would not be more than a mere stipulation. Now, if such a mechanism cannot work, what other alternatives do we have? An idea would be to assume that clitics do not incorporate at the v* level, but at a higher level. This would allow them to excorporate. In fact I have suggested something like that in Chapter 4 (and see also Mavrogiorgos 2006), where I argued that clitics incorporate to their host (and hence become heads) only if the latter has checked off all its features. For finite verbs, this would beT (or some high INFL head, more generally). This idea was based on the observation that clitics may obviate locality effects at the T-level with unaccusatives and passives (see Anagnostopoulou 2003a), which suggests that they form an Aspec at that level. Presumably, clitics were assumed to attach to the v* host, although only loosely, so as to ensure alongside movement of the clitic with the verb to the higher T head as well as an XP status for the clitic. Besides all the technical and theoretical problems of such an analysis, it is not clear to me how it would help me with enclisis, given that it only allows the clitic to excorporate, but not the verb. However, under a derivation which moves the verb first and the clitic second, it would be the verb that should excoporate (see Rooryck 1992 for exactly the same problem). As a result, we would still need the timing device, and hence this analysis would be even more costly. An alternative would be to assume that the above proposal is correct in its basics, however the actual reason why this is so differs. More precisely, recall Gallego's claim that in NSLs v* reprojects in T, forming a hybrid v*/T head that has the optional EPP feature. Given that the clitic is attracted by this feature we could assume that it is the complex v*/T head that attracts the clitic. This would capture the fact that clitics may form an A-XP spec both at the T and at the v* level. as well as the common intuition that clitics adjoin toT in NSLs, even though it is v* that really attracts them. 63 On the other hand, if clitics incorporate into the v*/T head as soon as they move to its Spec, we would expect that they incorporate, so that v*/T movement to CM would yield again proclisis, contrary to fact So, what we need in this case is that the clitic does not incorporate to v*/T in those cases where the latter complex moves to CM, but not in
63. Elena Anagnostopoulou asks how the v* IT analysis Interacts with the observation that the complex d + V surfuces In positions below aspectual adverbs. The problem in this case is that the cl + V complex would be expected to surfuce always at the T level, which is higher than aspectual adverbs. One idea would be to assume that the d + V complex always targets the T level in the syntax. however PF chooses to spell out lower copies of the complex.
244 Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
those cases when the complex has finished its checking by v*/T (as in indicatives/ subjunctives). A way to achieve this result would be to capitalize on the nature of CM in enclisis, namely on the fact that CM is an INFL (V-related) head, i.e. the highest of a series of V-related heads on which a phi-probe EPP is checked. One could assume that the INFL domain may consist of more than one head (cf. for example Poletto 2000, who argues for the existence of separate person, number and other heads within the INFL domain ofNSLs (and cf. the cartographic approach more generally)), and that incorporation of the clitic takes place only after the phi-probe EPP has been checked. In the case of enclisis this would be the higher CM head (as opposed to proclisis in which the lower T/v* would be the relevant head), and as a result one may view the clitic and the verbal heads in T/v* and CM as being parts of the same phi-chain that may be reorganized by movement until the EPP on CM, the highest head, is checked. This hypothesis really says more or less the same thing that was proposed in Mavrogiorgos (2006)/ Chapter 4, although the closure of the clitic-V-complex is now triggered by the phiprobe EPP, and not by the checking of all V-related features on the label ofV (it is not clear to me whether these two could actually be the same, although it looks like a possibility). The derivation of enclisis would be as follows (here exemplified by positive imperatives and gerunds (irrelevant details are ommited): (43)
V+tmp
CM[EPP/INPL+IMP]
v*P/TP
v/*T
cl
CM[INFLl
v*/T
v*'/T'
/~ v*P
(v*/T)
~ Spec v*P ~ (cl) v*P
~ VP
(v*)
~
Chapter 6. Enclisis in Greek 245
(44)
CM[EPP/INPL+GER]
/"--... v/*T
CM[INFLJ
/"--... yger
v*/T
v*P/TP
/"--... cl
v*'IT'
/"--... (v*/T)
v*P
/"--... Spec
v*P
/"--... (d)
v*P /~ (v*) VP ~
In both (43) and (44) the ditic moves to the specifier ofv*/T, at which it does not incorporate, since v*/Tis not a phase head. Alternatively, we can say that it moves to the specifier ofv* and then to the specifier ofT, which becomes the specifier ofv*, as the latter reproj ects in N SLs. Then, the verb moves along to CM (the phase head), where it checks the phi-probe EPP. The ditic incorporates into CMmiu64 , exactly as in proclisis, i.e. when all the features on the extended INFL-chain have been checked off/when the EPP-probe has been checked (although in proclisis it takes place lower, at [Spec v... P/ TP65] and it involves a spec-head structural relation, as opposed to a head-spec relation in enclisis66 ). Incorporation in a head-spec fashion is strongly reminiscent of merger 64. NotethattheEPP on v*/T relatestov*s EPP, notto T's/CM's. The latter has not been checked and it is free to attract the verb. One problem with this analysis is that it is not dear whether we can avoid an intervention effect of a ditic. This is so, because ditics are nominal elements, and hence could be visible in principle for EPP checking. One possibility would be to relate the nonattraction of ditics in this case to the modality features in CM, which may only be checked by V: if the latter can check both these and the EPP, as opposed to the ditic which can only check the EPP, then principles of economy would force the verb to move, instead of the ditic. 65. Note that Tis used here as a generic label In those structures where T is not even projected we could assume that a lower INFL head does the job (e.g. the T2 position proposed by Tsimpli 2000 for Greek gerunds). 66. The complex imperative head will move to CForce, along with the clitic, although this is not illustrated in the tree diagram.
246 Cities in Greek: A Minimalist Account ofProdlsls and Endlsls under adjacency (see Marantz 1988). In (43) the verb also moves to CForce, although this is not shown in the structure. Note that lack of interpolation here would not be accounted for as the result of head movement through the clitic-head (see Rizzi 1993). Rather, it would follow from the fact that the complexT/CM head is sent to PF as soon as the lNFL EP P is checked (presumably the whole series ofiNFL heads will be sent off as one unit). Apparently, according to such an analysis sending to PF (i.e. spell out) cannot take place before the EPP of a related series ofheads (i.e. INFL in this case) has been checked, although it must take place before a new (strong) head (like C) is merged. This allows the clitic not to incorporate at the v*/T level (although it torces it to do so in proclisis, where the EPP (and hence head closure) takes place at the v*/T level); in fact, it makes CL- T/CM incorporation obligatory at the T/ CM level (i.e. when the verb has checked all its V-related features). In the following example the derivation of a subjunctive clause is given: (45)
CPM
---~
na
cl
v*'/T'
y*/T[EPPIINFL-IMP]
~ T
v*
~ vlmP
v*
v* p
~ v*P
Spec
~ (cl)
v*P
~ (v*)
VP
~ Incorporation at the v*P/TP level would yield: (46)
[CM na [v*P/TP d-v*/T]-._]].
Here. the clitic is probed by the v* head in v*/T. CM is not V-related/INFL because na merges in CM, satisfying its requirements. Therefore, it is v*/T that contains the phiprobe EPP, triggering obligatory verb movement to v*/T. The clitic moves to the specifier ofv*/T and incorporates into v*/T, yielding proclisis. Incorporation takes place
Chapter 6. Enclisis in Greek 247
since the elitic only contains a subset of the features contained in v*/T, but also because the phi-probe EPP has been checked (i.e. the edge is accessible, although after checking it is sealed off). Finally, indicative clauses in root interrogatives would involve the following structure (if we assume that they involve V-to-C movement) (details are omitted here/ CM is either not projected, or if it is projected it is realized by a zero morpheme (see Philippaki-Warburton 1998)): (47) [Cporcl cl-v*/T-CM [CM (cl-v*/T) [v*P/TP (cl) (v*/T [EPPIINFL-impl)
[ ••• ]]]]
In this case again, the phi-probe EPP is -imp on v*/T, and not +imp/+ger on CM, i.e. CM is not V-related. The clitic is probed by v""/T, to the specifier of which it moves and into which it incorporates, due to the tact that it only contains a subset of features contained in v*/T and moreover the edge is accessible. Further V-movement to higher heads is triggered by an optional EPP on the interrogative CForce head, which moves the elitic and the verb as one unit to CForce. Further manipulation of the cl-v*IT complex is not possible after incorporation (recall that the phi-probe EPP is in v*/T and it has been checked), which explains why in questions the proclitic order remains even though the verb moves 'across the cliticization site'. An issue that arises under this analysis is the Spell-Out issue. The problem is as follows. On the one hand, Chomsky (2001a, 2005) assumes that Spell-Out involves whole derivational chunks, which coincide with the complement domain of a strong phase. Spell-Out, in this sense, is viewed as transfer of already built syntactic structure to the PF and LF interfaces/components, and takes place at the next strong phase (i.e. Spell-Out ofVP, the complement of v* takes place when Cis merged in the structure). Gallego's (2006) phase sliding would shift the complement domain higher, to the v*P level. The next Spell-Out operation would take place at the CP level (presumably, involving the highest C-head, given that the derivation cannot know in advance whether a higher matrix verb will be merged, so that Spell-Out will take place when matrix v* is merged). The problem is that following such a Spell-Out approach, clitics would not have to incorporate before the end of CP. Although this might not be a problem for gerunds, which do not move higher than CM anyway, it would be a problem for imperatives, which as I have said move to CForce/Op as well. This would predict that interpolation would be possible in principle also with enclitics, which is exactly the argument Rizzi ( 1993) brings forward in order to support his claim that elitics must attach on the same functional head with the verb. Two possibilities seem to be available in this case: one is that this is what happens, but enclitics attach to the verb anyway, given that intervening specs do not block PF-cliticization (on a par with affixhopping heads in languages like English- see Bobaljik 1995). The problem with that is that it would make it much harder to explain why interpolation languages allow
248 Cities in Greek: A Minimalist Account ofProdisis and Endisis
interpolation in the first place. I think it is quite obvious that affix-hoping heads (i.e. a:ffixal heads) behave differently from clitic heads in this respect. 67 The second possibility is that the complex head picks up the clitic on its way to the higher CM head, ensuring that no interpolation is possible. This would follow from: (a) the fact that the clitic has not incorporated to the v*/T head as yet, although it is located at its edge; (b) from the fact that v*IT movement to the higher head must move through the edge of v*/'P11 ax (not the edge of v*/'P11i1168). Presumably, this would be possible due to the weak phrase structure status of clitics (maximal, minimal) which allows it to have an XP or an X role according to the demands of the structure. In this respect, encliticization (or the final step thereof) would be syntactic (involving some kind ofhead adjunction), and not PF. In the proclisis case such a possibility would never arise, given that the EPP forces incorporation to the min-head, so that further movement cannot leave the clitic behind (neither can the clitic excoporate to a higher head). Another approach to Spell-Out is that proposed in Matushansky (2006). According to her analysis, sending to PF takes place immediately before a higher head is merged (i.e. at each head). As far as I understand, for clitics this would entail that incorporation should take place at the v*/T case in both proclisis and enclisis. That is, her system would not allow EPP to play any role in the timing of Spell-Out. The problem again is that it is not at all clear how enclisis could possibly be derived by morphosyntactic means (i.e. movement, head adjunction, etc.) in this approach. I guess she would have to assume a purely morphological/PF approach. E.g. she could capitalize on the fact that non-finite/enclitic-taking forms more generally appear to have restricted agreement (recall the restricted person agreement on imperatives/gerunds), so that enclisis would involve some kind ofPF merger of the CL-v*/T head to the CM head. Nevertheless, it is far from clear how such an account could actually be implemented (cf. for example the fact that the whole complex head would invert, not only the clitic). Moreover, it is not the case that all enclitic-taking forms realize exactly the same set of morphological features, rendering thew hole approach extremely complex. Finally, it would not be possible to link the phi-probe EP P property of CM and enclisis, given that the latter would only be sensitive to PF properties. All in all, the analysis that I am proposing here seems to work, although the SpellOut issue and its presumed relation to EPP-checking is not so clear. Of course, it is not that counter-intuitive: it is quite well accepted that verbs form extended projections of 67. Unless, of course, this dltference would only apply in interpolation languages. I.e. in languages not allowing interpolation ditics would behave as atfixal heads, as opposed to languages which do allow interpolatioiL Although this might be possible, independent evidence is needed. So, I leave this issue here as it is. 68. By this I mean that the ditic is at the edge of v*JTDln, i.e. at the edge of the complex functional head, but not at the edge of the projection of this head. In essence one could visualize the ditic as standing between v*trmoz. (i.e the (XP) projection of the head) and v* tpnin (i.e. the actual head of the projection): then the min-head would have to move through the edge of the max-head, picking up the clitic on its way up.
Chapter 6. Enclisis in Greek 249
V-related heads which must be adjoined to each other in some way (typically via head adjunction/movement) and which also must form an undividable complex head (i.e. no excorporation is allowed (although PF can possibly change the internal ordering of heads, provided certain conditions are met- see Embick & Noyer 2001)). Clitics also take part in the construction of this extended INFL domain which surfaces as a verbal form, however they do not seem to have exactly the same properties as these a:ffixal INFL heads. Rather, they seem to be related to the whole complex in a looser sense. This is in fact the essence of Roberts' ( 1991) account that classified (V-related) heads as x-1, whereas clitics were taken to be X0 heads. This intuition cannot be captured in such terms any more, however the EPP-Spell-Out approach could possibly help us in that: one would simply have to assume that head movement (and especially V-related head movement) involves classic head adjunction, which does not allow any excorporation, under standard assumption, whereas the EPP-Spell- Out would ensure that further adjunction of weak elements (like clitics) that are contained in this complex head domain are also incorporated at the final stage, just before the whole complex seals off. Note that under such an approach clitic movement cannot be fully equated to head movement, at least in terms of their attachment properties. However, such a distinction seems to be independently required, given that v''"/T-to-CM movement is not considered to violate the HMC, exactly because clitics are taken to move as XPs/Xs to the spec of the head that attracts them (namely, v* /T). I do not want to argue here that this is a very elegant solution, however it may allow us to put the timing of EPP-satisfaction (which is linked to finiteness, namely to properties of CM, and to V-movement) in the picture, associating the proclisis-enclisis alternation with it In any case, and independently of the particular technical implementation proposed, this approach can capture the generalizations it was meant to capture, and also offers a new way of viewing the interaction of finiteness and V-movement in Pattern A languages. In 'typologicaf terms, it is a Kaynean analysis, although it takes the role of finiteness to be crucial in the CM head, and not in T. Now, it is important is to see whether it can work for other cases as well. This issue is dealt with in the following section.69 69. There is also a slightly different analysis, which I have not included in the main text, given that it is not dear to me whether it fares better than the analysis proposed, or whether they are variants of each other. Accordingly, one could assume that T is divlded into further sub-heads, like PersonP, Numbed~ etc (an assumption I have already accepted). PersonP would be the highest sub-head of the T head, following standard assumptions (see Shlonsky 1989). Given that enclitk structures typically involve non-finite clauses in Pattern A languages, and given that in Greek (and possibly in other languages) lack of finiteness seems to correlate with restricted person agreement, one could speculate that the Person head is not checked by the verb (in imperatives, presumably because the person feature is inactivated, and in gerunds because the subject agreement is not listed in the Lexicon (following Alexladou & Anagnostopoulou 1998), and hence cannot check it). The result would be that the verb would not have to pass through the relevant head (c£ Kayne 1991). The ditic would be attracted to the v*/T periphery, and would incorporate
250
Cities in Greek: A Minimalist Account ofProclisis and Enclisis
6.3.3 Further instances of the proclisis-enclisis alternation in Greek In the previous section I proposed an analysis for enclisis in Greek. In this section I will discuss a few related issues, as well as a number of proclitic and enclitic constructions in Greek. My goal will be to show that this analysis can capture a number of related phenomena in an adequate way. 6.3-3-1 Negated imperatives/gerunds in Greek
Negative (temporal) gerunds are negated via the deontic modality negative head min, and they can take enclitics, on a par with positive gerunds. This is illustrated below: (48) a.
Katalavenondas to. Understanding itcl 'Understanding i~
Greek
b. Min katalavenondas to. Not understanding it
Greek
According to the analysis of Roussou (2000) which I have adopted, negative gerunds are possible because they only need to move up to CM, whereas negation is merged right above this head. Since v*/T-to-CM movement is V-related (an EPP feature is checked on CM), enclisis obtains, according to the analysis. This is illustrated below (I omit details regarding lower perfective heads): (49) [NEGP min (CMP V-v*/T-CM [v*P/TP cl (V-v*/T) [... [vP subject/pro (V-v*) [VP (V) (d)]]]]]]
to the (empty) Person head. The verb in Nwnb would simply by-pass the head, although it would pick it on its way up, asswning that the top T-label has an uninterpretable V-feature. Presumably, no intervention effect would be created by the inactive/non-checked Person head. Controlled subjunctives, which also take a PRO subject, like temporal gerunds, would not take enclitics, because the verb would be able to check the person feature (c.f full agreement- strong ending), and the EPP would ensure closure of the CL-v" IT head. The main problem I see here is that restricted agreement correlates with lack of Person checking (and hence of movement of the verb to the person head, assuming that normally it is the verb that checks this feature). However, we have already argued that restricted agreement is only a refiex of a particular C-head, which is linked to V-related movement to a higher head, and not necessarily with the syntactic properties of the T head(s ). Hence, relying on overt agreement in order to decide whether the verb can check any of the features on Person is not an adequate approach (see also Alexiadou & Anagnostopoulou 1998 for the same argument regarding EPP checking on Agr in finite clauses). Moreover, it is not dear that it would work for imperatives, since the latter do not seem to have an active person feature, and as a result would not possibly contain an empty person head in T at all (cf that person would possibly stay in C, following Chomsky's 2005 feature inheritance theory). For these reasons I will continue to assume the proposal outlined in the main body text.
Chapter 6. Enclisis in Greek
On the other hand, negative true imperatives are not possible; rather, the subjunctive is used: (SO) a. *Min grapse to! Neg write.perf.-2sg.imp it.cl
Greek
b. *Min tv grapse! Neg itcl write.perf-2sg.imp
Greek
c.
Na min to grapsis! Subj neg it.cl write.perf.-2sg 'Do not write it!'
Greek
According to Roussou (2000) (see also Rivero 1994, Rivero & Terzi 1995) the examples in (SOa-b) are ungrammatical because neg blocks V-movement to CForce (for Roussou, due to an intervention effect; for Rivero and Rivero & Terzi, due to an ECP/HMC violation). On the other hand, Philippaki-Warburton (1998) and Giannakidou (1998) claim that these same examples are ungrammatical, because Neg blocks V-movement to a Mood head. This is the case, given that they assume that Neg is merged lower than Mood, contrary to Roussou (2000): (51) a
*l<1:orre Vimp-el [NegP Neg [~P (Vimp-d) [INFLP cl (Vim~ [...]]]]] (Roussou)
b. *[MoodP v;mp-cl [NegP Neg [INFLP cl (Vimp) [... ]]]] ( Philippaki-Warburton 1998) The problem for Philippaki-Warburton (1998)/Giannakidou (1998) is that in negative gerunds Mood does not seem to block V-to-Mood movement. In order to account for this, Philippaki-Warburton argues that gerunds do not involve movement to Mood; rather, they only move up to INFL. This assumption immediately creates a problem for elitics, given that they typically target v*IT in all other structures. Philippaki-Warburton claims that elitics remain in situ within the VP, however it is obvious that this would create a number of issues, which make such an analysis problematic (unless she assumes VP-internal merger, in the spirit of Raposo & Uriagereka 2005 ). The more general problem, though, is that it misses the common property between gerunds and imperatives, namely the fact that both involve V-related movement to CM, for which there is independent evidence (see Roussou 2000 for details, as well as Rivero 1994 for a more cross-linguistic argumentation). On the other hand, I believe that Philippaki-Warburton/Giannakidou are right in attributing the ungrammaticality of negated imperatives to negation. The reason is that there is evidence that modality/ mood checking rather than force checking is the crucial factor. First of all, recall that in Greek the verb is divided into two moods, [+)
251
252
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
and [-) imperative. The [-imperative] form stays in INFI.. while the [+imperative] form must move to CM. If it doesn't, ungrammaticality arises (see Roussou 2000): (52) a.
b.
*na grapse! subj write.perf.-2sg.imp 'Do not write!' grapse! write. perf-2sg.imp Write!'
c. *na grapsis subj write.perf-2sg.-imp Write!/ You should write!'
Greek
Greek
Greek
Given that na. is merged in CM, and that CM is the head where +imp mood is checked, in (lOla) the verb has stayed in INFL, and as a result ungrammaticality arises. Of course, someone could point out that this does not prove that CM is the factor for the ungrammaticality; it only shows that it is one possible factor, CForce being another one. However, different researchers have pointed out that modal and negation particles can also check CForce (as well as modal and negation) properties. For example, this is the standard assumption for (lOlc) (also supported by independent evidence from possible and impossible combinations of na with other complementizers and particles), and a similar analysis has been given for negative clauses (e.g. for Italian interrogatives - see Zanuttini 1997, for negative periphrastic imperatives in Greek without a subjunctive particle). If negation or na can check the features on CForce in other constructions, then why can't they do the same with imperatives? Someone could argue that this is due to the fact that imperatives also need to check a force feature on their head. However, Force more generally is not a V-related feature in these languages, as is evidenced by grammatical root interrogatives with preverbal particles/heads (such as negation or modal particles): in these clauses the verb does not need to raise to CForce, nevertheless no ungrammaticality arises. Moreover, Roussou (2000/2006) argues that in negated subjunctives, the na particle is directly merged in Force, while the negation particle first merges in CM and then moves to Neg. In this way she captures the optionality of the na particle (assuming that some element must check the modal head) as well as the fact that negation also has modal characteristics, given than at least in Greek it comes in two varieties, depending on the veridicality of the clause (see Giannakidou 1998 ). As a result, min is assumed to be a non-veridical negator, as opposed to dhen which is a veridical one. This structure is only allowed with the [- imperative] paradigm; the [+ imperative] one yields ungrammaticality: (53) a.
(na) min grapsis subj neg write.perf.-2sg.-imp
Greek
Chapter 6. Enclisis in Greek 253 b. *(na) min grapse! (subj) neg write.perf.-2sg.+imp 'Do not write!' (54) a. b.
Greek
[CForce na [Neg min [CM (min) [INFL V [... ]]]]] [CForce min [Neg (min) [CM (min) [INFL V [... ]]]]]
Given that in (54a) na checks Force, while min checks modality, negation and force (when na is not present), and given that [-imperative] is fine in IN FL. as opposed to [+imperative], we can conclude that[+ imperative] cannot 'survive' in INFL; it must raise higher to CM. This is in accord with the more general intuition that [+ imperative] is checked in CM, but also with the idea that true imperatives cannot survive in INFL in some languages (see below for further discussion).7° If we assume that [+ imperative] forms must move to CM, then the question arises why they cannot be negated, given that negation lies higher than CM and it could check the force teature on CForce. Here, I propose that this might be due to the fact that non-veridical min merges directly in CM, as we have seen in (54) above, given that it encodes modality features, too. Apparently, the same happens in periphrastic negated imperatives, however in that case the [- imperative] verb may stay low in IN FL. as it does not encode modal/mood features, and hence does not need to raise higher, contrary to true imperative forms. This account captures: (a) the requirement that [ + imperative] forms must move to CM; (b) the fact that min can check CM, Neg as well as Force features; (c) the fact that both imperatives and gerunds take enclitics, since both involve obligatory movement to CM. As for why gerunds are always negated (which means that neg cannot start off at CM, where the gerund must obligatorily move), it could be related to the fact no CForce head is projected in gerunds (recall that CForce is projected in all the imperative cases). However, it is not clear to me at the moment why this restriction should hold.71
70. Note that when na ls absent ln negated periphrastic imperatives, the verb can only be in the 2nd person singular and plural, i.e. it exhibits the same person restriction found with true imperatives, although the form is definitely[- imperative]. This fact could be accounted for either
by assuming that negation cannot really check force features in this particular case (so that the person restriction is required in order to get the imperative/command interpretation that is inherently characterized by the second person restriction), or alternatively by assuming that neg asswnes an imperative modal/mood function by realizing CM and CForce at the same time (which would render CM similar to a + imperative CM). 71. One possibility could be that negation can be interpreted as a modal element (and hence check the CM features) only if it is in the scope of an operator/CForce head. This appears to be more generally the case in Greek, as we can also find examples with subjunctive interrogatives where 11a cannot be deleted (na min to kano? vs. * min to kano? 'subj neg it.d do-lsg?' vs. 'neg it.d do-l sg?').
254 Cities in Greek: A Minimalist Account ofProdisis and Endisis
Surrogate imperatives Syntactically true imperative clauses may use verbal torms from other paradigms, typically in the 3rd person. or in the 2nd person plural. These forms are called surrogate forms, and they are very common in Romance languages (e.g. Spanish, Italian, Romanian). In Greek things are less clear, as some verbs in the 2nd person plural appear to be identical to the indicative paradigm, while on the other hand there are also forms which use a truncated version of the corresponding indicative ending. The issue that concerns me here is how exactly surrogate imperatives may have true imperative syntax. My proposal is that this is simply the effect of morphology, i.e. of the tact that languages tend to use less specific torms as an economy strategy. In other words, I propose that surrogate forms are hidden true imperatives, i.e. forms that must check modality and force in the C-domain. 72 The reason why they may surface as subjunctives or indicatives (depending on the language as well as on other properties) has to do with the fact that these forms are morphologically less specific than true imperative forms. & a result. due to the Subset Principle, they can phonologically realize the smaller set of morphological features encoded by true forms. Harris (1995) has convincingly shown that this is the case for Spanish surrogate forms with a true imperative syntax, and that a post-syntactic realizational morphological component can rather easily account tor these facts without needing to complicate imperative syntax in any 6.3-3.2
71. Note that there is independent semantic evidence that all clauses having a sentential imperative force must involve checking of some modal element (see Portner & Zanuttini 2003). Moreove~ we have seen overt evidence in certain Italian dialects, where imperatives co-occur with modal particles in the C domain. Note however that at least for Romance languages CM encodes modality and not mood, since mood is assumed to be merged lowe~ below T (see Laka 1990, Harris 1995 for Spanish, Zanuttini 1997 for Italian). This however would not affect our analysis since also in Greek there is no overt imperative modality morpheme on the true imperative form. Apparently, what is crucial is that the verb contains abstract modality features which can only be checked on CM. Finally, regarding surrogate forms with a non-imperative syntax (i.e. subjunctives and indicatives taking proditics), these are commonly assumed as involving movement toT. There is strong evidence in favour of this view (e.g. position of verbs w..t:t. adverbs, posslbility of negation, position of negation, etc.). Nevertheless, given that in Greek as well as in many languages modality/mood maybe checked in CM, one could wonder whether this is also the case in CL-subj/ind complexes. This is rather crucial, since if there is a CM head above T in these structures, and if this triggers obligatory V-movement, then enclisis should obtain according to my analysis. At this point I do not have much to say on this point. Unfortunately, these languages do not have preverbal markers, which would help us in locating the exact position of the verb, as in the Greek cases (obviously, clitics would not do in this case). On the other hand, there is independent evidence that the verb stays in T in indicatives and subjunctives in these languages, while CFin (i.e. CM) has been shown to host complementizers in many Romance varieties (see Rizzi 1997, Paoli 2007). These facts suggest that indicatives and subjunctives do not require V-movement to a head higher than T. Presumably, if modality features are present on CFin, they are checked either by some other head (e.g. negation or complementizer) or by a zero morpheme. This is a very tentative discussion and I will leave it here.
Chapter 6. Enclisis in Greek 255
way. Note that this analysis is further supported by the fact that in many languages the forms borrowed to the imperative (syntactic) paradigm typically exhibit tense or modality properties that are compatible with those of imperatives (i.e. -tense,+ future or +subjunctive, both of which involve irrealis modality, etc.). Moreover, it explains quite straightforwardly why surrogate forms can also appear in the INFL position, as opposed to imperative forms which cannot: presumably, subjunctive forms may realize both [+ imperative] and [- imperative) feature sets, as opposed to imperatives which have narrower specifications and which therefore may only realize [+ imperative] morphology in the C domain. Finally, such an approach allows us to keep the generalization that enclisis obtains when a form must check V-related properties in a head that is higher than the cliticization site?3 6.3-3.3
Clitic switch
Clitic switch refers to the phenomenon whereby both dat-acc and acc-dat (clitic) orders are possible. Clitic switch is only found in Greek, however it is more common in languages having weak pronouns (preverbal and/or postverbal) (e.g. Zurich German, French). The basic observation for Greek enclitic clusters is that their ordering is free (although it is subject to post-syntactic phonologicaVphonetic restrictions):
(55) a.
Dos(e) mu to! Give-2.sg.imp. me-dat it-ace
b. Dos(e) to mu! 'Give it to me!'
(56) a.
mu to! Pare Take-2.sg.imp me-dat it-ace 'Take it from me!'
Greek Greek Greek
b.
Pare to mu!
Greek
c.
*?Par-mu-to!
Greek
d.
Par-to-mu!
Greek
Free ordering is not found in other languages with special clitics (modulo French, which is somewhat more complicated and will be discussed separately below), at least in Romance. According to many researchers (cf. e.g. Morin 1979; Somesfalean 2005) in Romance the ordering within a clitic duster remains constant across
73· The fact that 1st plural forms may be true imperatives when preceding ditks is not a problem, since imperative force requires that the subject is interpreted as an addressee, not that is exclusively a 2nd person. For Spanish 3rd person imperatives, it is obvious given their obligatory addressee interpretation that they must involve some kind of morphologkal operation which links the feature combination 'addressee + formal' to morphological 3rd person.
256
Cities in Greek: A Minimalist Account ofProdlsis and Endlsls
proclisis and enclisis. In fact, according to Morin, it is only Standard French that does not follow this general trend:
(57) a. Lui ce lo dam.
IO>DO
Italian (Italy)
IO>DO
Italian (Italy)
He us it will-give 'He will give it to us: b. Spediscicelo! send us it 'Send it to us!'
(58) a. El mi-l da.
IO>DO
Romanian (Romania)
IO>DO
Romanian (Romania)
IO >DO
Lorrain dialect of Ranru pt
10 >DO
Lorrain dialect of Ranru pt
IO >DO
Normand dialect of Saire
IO >DO
Normand dialect ofSaire
He me it gives 'He gives it to me' b. Da-mi-1! give me it 'Give it to me!'
(59) a. II me le donne He me it gives 'He gives it to me: b. Donne- me-le! Give me it! Give it to me! (60) a.
Il lui le donne He him it gives 'He gives it to him:
b. Donne-lui-le! Give him it! 'Give it to him!'
Within Romance, we can find two cases that are somewhat similar to the Greek case: A. Standard FJtmch: preverbal ordering DAT > ACC (modulo 3rd person74 ) turns to ACC > DAT in postverbal position: (61) a.
Il me le donne. He me it gives 'He gives it to me:
IO>DO
St. French (France)
74· Alternatively we can say that it is Person > Non - Person with DO > I0 when 3rd person
(Elena Anagnostopoulou p.c.).
Chapter 6. Enclisis in Greek 257 b. Donne-le-moi! Give it me! 'Give it to me!'
DO>IO
St. French (France)
Note that in other French dialects and/or registers, the DAT > ACC order may appear in postverbal position: (62) a.
Donne-moi-le!
IO>DO
Quebec French
IO>DO
Paris RF
Give me it
b. Donne-moi-le! Give me it 'Give it to me!'
Note that in the third person the preverbal DO> 10 remains as it is in the postverbal position. while in dialects with preverbal 10 >DO we find the same order ( 10 >DO) in enclisis: (63) a.
Ille lui donne~ Donne-le-lui!
DO > 10
~DO
10 > DO
~
> 10
He it him gives ~ give it him 'He gives it to him' ~ 'Give it to him!' b.
Illui le donne ~ Donne- lui- le! He him it gives ~ give him it 'He gives it to him'~ Give it to him!'
10 > DO
So, more generally, in French in some persons the preverbal 10 >DO order may reverse to DO > 10 in enclisis, while in most dialects other than Standard French the preverbal order remains constant. B.
Ordonez (2002) reports Romance dialects which have free clitic ordering in postverbal position: (64) a. Lo me dusset pas veire. It me let not see 'You did not let me see it:
b. Daussa-m-lo!
Occitan
Occitan
Let me it c.
Daussa-lo-tne! Let it me 'Let me it!'
Occitan
However, (B) appears to be rare, at least in the Romance family (note in this respect that free ordering is not allowed also in Southern Balkan languages like Macedonian,
258 Cities in Greek: A Minimalist Account ofProdisis and Endisis
Bulgarian or Albanian). What is more common is to have the (A) pattern, or languages which have the DO > 10 in both preverbal and postverbal position: (65) a.
Ditz-lo-me!
Gascon
Tell-it-me 'Say it to me!' b. Lo me quies dar. it me want2sg give.inf 'You want to give it to me' c.
Aragones
Da-lo-me! give it me 'Give it to me!'
d. D6na-la-me! Give it me 'Give it to me!'
Mallorcan Catalan
Ordonez points out that in all these lo-me dialects the 10/dative clitic is obligatorily stressed. He links this fact to the well-known analysis of DO > 10 post-verbal orders of Standard French, according to which the 10 is a weak pronoun (as opposed to the DO, which is a clitic). The idea is that the 10 is stressed in all these dialects because it is a weak pronoun.75 Laenzlinger (1994, 1998) as well as Cardinaletti & Starke (1999) analyze the 10 object in post-verbal position in Standard French as a weak pronoun. Evidence in favor of this analysis comes from the fact that only the strong form of 1st and 2nd person sg. is allowed in this particular position, as well as the fact that the 10 pro-form receives lexical/word-level stress. Cf. in this respect that the reduced form of moi/toi (namely melte) is not allowed in final position, although i tis possible if the DO follows them (in some dialects; there are also dialects in which a strong form precedes a final DO. Laenzlinger analyzes such cases as clitics, whereas Ordonez (2002) argues that V-moi/me-le orders involve XP movement ofV-moi/me across le to a higher position. In any case, a final 10 is considered to be a weak pronoun): (66) a.
Donne-le moi! Give
Standard French(France)
it me
b. *Donne-le-me! Give it me
Standard French (France)
75· Adam Ledgeway points out to me that this is not exclusive to lOs, but is also found with DOs (e.g. schwa to front rounded vowel in -le infais-le). He argues that stress is not really an indicator of weak status and that French cannot do otherwise because of regular placement of word stress. He finally points to other Romance varieties with stressed enclitics (SIDs) for which there is no need to treat them as weak pronouns. However, note that in the case of mol/to I there are further arguments besides stress that they are weak pronouns (see main text).
Chapter 6. Enclisis in Greek 259
*Donne-tne-le!
Standard French (France)
Give me it 'Give it to me!' (67) a.
Donne-moi-le!
non-standard varieties
Give me it b. Donne-me-le! Give me it c.
Donne-lrli-le! Give him it 'Give it to me/him!'
[Note that *Donne-le-me is not allowed in any of the varieties mentioned in the literature] According to this type of analysis, lui in orders such as donne-le-lui is also a weak pronoun, even if it is homophonous with the preverbal clitic lui (cf. Ille lui donne). DO > 10 ordering involves the combination of a clitic and a weak pronoun. The standard analysis is that the DO cliticizes into T/INFL while the 10 stays in a spec position right below T/INFL (or, more generally, the functional head occupied by the host): (68)
[CP Donne- [AGRs le] [AspP/AgroP lui]] (Laenzlinger 1998: 171, 83b)
Another well known case of non-rigid order in pronoun clusters is that of Germanic languages. E.g. in Dutch weak pronouns/ clitics are less strictly ordered than e.g. German weak pronouns or Romance clitic clusters: (69) dat Jarlfrl Jarlrt gisteren gegeven heft. That Jan it her I Jan her it yesterday given has 'That Jan gave it to her yesterdaY:
Dutch
The same applies to West Flemish: (70) da Jarlt-zel Jan-ze't gisteren gegeven eel That Jan it her/ Jan her it yesterday given has. 'That Jan gave it to her yesterdaY:
West Flemish
In a similar way, weak pronouns in Swiss German are not strictly ordered: (71) a.
D' Maria zeigt et' mir. The Mary shows him me
b. D' Maria zeigt mir m. The Mary shows me him 'Mary shows him to me:
Swiss German Swiss German
26o
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
Similar phenomena are also found e.g. in Swedish and Norwegian. As Anagnostopoulou (2007) has pointed out, pronoun serialization in languages having weak pronouns is less strict than languages with clitic clusters. E.g. even in German, which normally only allows DO> 10, speakers accept some 10 >DO orders in the Wack.ernagel position. Now, French (and other lo-me Romance languages) does not have weak pronoun clusters, nevertheless the presence of one weak pronoun within the cluster correlates with less strict order. What about Greek? Could free ordering be related to the presence of a weak pronoun (instead of clitics), e.g. in DO> 10 orders, on a par with French? In French form, word stress, as well as the positional restrictions imposed on the reduced me/te forms, provide evidence in favor of this hypothesis as far as enclitic clusters are concerned. On the other hand, in Greek: a. There is no difference in form between proclitic and enclitic clusters as well as between 10 > DO and DO > 10 enclitic clusters: (72) a. Mu to edikse. Me-gen it-ace showed-3sg 'She showed it to me:
b. Dikse mrl to! Show-2.sg.imp. me-gen it-ace 'Show it to me!' c. b.
Dikse to mu!
Greek
Greek
Enclitics cannot take word stress for independent reasons (i.e. addition of clitics to the right hand side of the word/host only triggers rhythmic/ secondary stress), whereas on the other hand proclitics may take - under particular conditions - word stress: (73) a.
'Dikse 'mu to!
To 'ediksa 7 'Todiksa c. Tu 'edhosa 7 'Tudhosa b.
c.
Greek
Greek Greek Greek
There are no restrictions on the position of pronouns within the enclitic cluster, as opposed to French: (74) a.
..Donne-le-me!
French (France)
b.
Dose-ton-mu!
Greek
c.
Dose-mu-ton!
Greek
Of course, this might simply follow from the fact that in Greek the alleged weak pronoun is homophonous with the elitic pronoun, contrary to French (so that mu may appear without problems both in preverbal and postverbal position).
Chapter 6. Enclisis in Greek 261
So, is there any evidence that the IO in DO> IO enclitic clusters is a weak pronoun in Greek? Lefs take Cardinaletti & Starke's (1999) well-known criteria that distinguish between clitics and weak pronouns: a. Clitics are heads (cf their ungrarnmaticality in initial position in V2 structures, or the fact that they are carried along with the verb when the latter is moved to a higher functional position). On the contrary, weak pronouns are XPs (they may appear clause-initially in V2 structures, they may have wide scope, they do not need to be adjacent to the verb, among others). According to this criterion, Greek enclitics are clitics and not weak pronouns, given that they always follow the verb, they do not have wide scope76 and they must be adjacent to the verb. (75) a. *Parto k.e petak.se! Take-2sg.imp. it-ace and throw-2sg.imp 'Take it and throw it away'
Greek
b. *Dose -to grigora mu! Give-2sg.imp. it-ace quickly me-gen 'Give it quickly to it!'
Greek
Note, on the other hand, that this criterion would classify French enclitic IO as a clitic (rather than a weak pronoun), given that the IO does not allow interpolation and does not have wide scope (see Laenzlinger 1998 for details). As a result, it seems that even weak pronouns may have X-like properties in some languages. b.
Co-occurrence of clitics gives rise to special morpho-phonemic processes (e.g. vowel lowering), which are never found with weak pronouns.
Unfortunately, Greek does not have such special morpho-phonemic properties in clitic clusters. The only cases we find is prosodic restructuring (e.g. dose mu to~ dosmu to!), but this is characteristic of weak and eli tic pronouns more generally. No person-case constraints arise with weak pronouns. This is not entirely true, given that languages with weak pronouns are also sensitive to this constraint (see Bonet 1991 on Swiss German, and also the discussion in Anagnostopoulou 2003a and 2007 on German weak pronouns). Rather, whatis characteristic of weak pronouns is that they may re-shuffle in order to avoid Person Case
76. This is not entirely true. In DO > IO enclitic orders, the IO ditic (and only that) may have wide scope over a conjunction as the following example shows: D!kse to ke dhanise to show.2sg.imp. it.acc.d and lend2sg.imp it.acc.d 'Show it and lend it to me!'
mu me.gen.d
Greek
262
Cities in Greek: A Minimalist Account ofProdisls and Endisis
Constraint violations, as opposed to clitics which are normally strictly serialized and as a result cannot have recourse to such repair strategies. ln any case, in Greek one may notice a peculiar situation, which has also been mentioned very briefly by Bonet (1991) (due to Iatridou ): namely, in D 0 > I0 enclitic constructions the effects of the person-case constraint may be weakened. ln fact, with certain verbs, such as e.g. sistino ('to introduce'), such effects vanish altogether, at least for certain speakers. More importantly, the orders that are allowed involve violations of the Weak version of the Person Case Constraint (i.e. 3 1 orderings for imperatives). On the other hand, in the 10 > DO ordering, the effects arise again. This situation, as Bonet ( 1991) herself points out is strongly reminiscent of the repair strategy found for example in Swiss German. Note in this respect that the sensitivity of the judgements to the choice of the lexical verb is something that is not uncommon in person-case constraint judgements (see Ormazabal & Romero 2007 on Spanish and Romance more generally): *Sistise tu me! Introduce-2.sg.imp. him-gen me-ace
[10 > DOll 3 > 1]
b. rsistise me tu77! Introduce-2.sg.imp. me-ace him-gen
[DO > 10//1 > 3]
(76) a.
(77) a.
b.
*Danise tu me! Lend-2.sg.imp. him-gen me-ace 'Lend me to him!' ??Danise me tu!
Greek Greek
[10 >DOll 3 > 1] Greek [DO > 10// 1 > 3] Greek
Cf. with proclitic orders: (78) a. *Tu me sistise. (He/she introduced me to him)
Greek
b. *Me tu sistise.
Greek
c. *Tu me danise. (He/she lent me to him)
Greek
d. *Me ttl danise.
Greek
The insensitivity of the DO> 10 order to the Person Case Constraint may suggest that the IO is in fact a weak pronoun. This would agree with the criterion put forward by
n.
Other examples involve 1st and 2nd person plural ditics, which in Greek have syncretic forms for the dative/accusative, giving rise to ambiguity: (i)
Zoghrafise tus tnas. Paint.2sg.imp them.dacc/dat us.cldaUacc 'Paint them for us/us for them.'
(ii)
Zoghrafise tnas tus. Paint.2sg.imp us.ddat/acc them.clacc/dat 'Paint them for us/us for them.'
Chapter 6. Enclisis in Greek 263
Cardinaletti & Starke (1999), but most importantly with the cross-linguistic observation that languages may avoid violations of the Constraint by changing the order of the pronoun clusters (where typically the IO will be the last pronoun). In this respect, note that Cardinaletti (1999) points out that: "The combination of a 1st/2nd person accusative pronoun with a 3rd person dative pronoun is possible with pronouns belonging to different classes, but not when both pronouns are clitic or both are weak". In other words, when a weak pronoun combines with a elitic pronoun, violation of the Person Case Constraint may be avoided. This seems to be exactly what happens in Greek. 78 d.
Clitics can double, as opposed to weak pronouns.
As far as Greek is concerned, this would be a problem, given that all clitics, both proclitics and enclitics, may double a DP: Tu Jiorghu to vivlio tu to edhosa. The George-gen the book-ace him-gen it-ace gave-lsg 'I gave the book to George:
Greek
b.
Tu to edhosa tu Jiorghu to vivlio.
Greek
c.
Tu Jiorghu to vivlio dhos to tu The George-gen the book-ace give-2.sg.imp. it-ace him-gent /tu to! him-gen it-ace 'Give George the book!'
Greek
(79) a.
d. Dhos to tu/tu to tu Jorghu to vivlio!
Greek
However, Stanca Somesfalean (2005) has pointed out that this is not always the case. For example in Romanian weak pronouns can clitic-double (or left-dislocate) a DP, indicating that this difference may not be relevant cross-linguistically but only in certain languages (possibly related to the fact that in Greek the alleged weak pronoun is homophonous to the eli tic pronoun, not the strong pronoun). e.
Weak pronouns are found in distinct positions from both strong and clitic pronouns. Weak pronouns must occur in a case position at S-structure.
This appears to be true for example for Italian lora or tor Germanic pronouns, which appear in a number of distinct positions, not necessarily next to the verbal hosl This would follow, according to Cardinaletti & Starke from their XP status (as opposed to the
78. Note that interestingly enough in Northern varieties of Greek free ordering of double accusatives (recall that dative evolved to accusative in these varieties) is not possible, so that one may only have dikse me to! but not * dikse to me! ('show.2sg.imp. me.d itd'). Dikse to mu is, however, allowed One wonders whether this is a weak genitive pronoun instead of a dative ditic one taken from the standard variety.
264 Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
X status of clitics). Note, nevertheless that weak pronouns in French (postverbal moi/toi/ lui etc.) must also be next to the verb, and in fact appear in the same position as elitics: (80)
a. Regarde done la fleur! Look well the flower 'Well look at the flower!
French (France)
b. Regarde-moi done! Look me well 'Well look at me!'
French (France)
c. Parle-moi done! Talk me well 'Well talk to me!'
French (France)
d. Regarde-la done! Look her well 'Well look at her!
French (France)
e. *Regarde done moi! look well me 'Well look at me!' f.
(from Morin 1979)
French (France)
While both weak and eli tic pronouns may receive phrasal and contrastive stress, it is only weak pronouns that may have lexical/word stress. "In somewhat metaphorical terms, while both series can acquire accentuation, only one of the two has it from the start"
As 1 have already pointed out above, this criterion will classify Greek 10 pronouns in enclitic clusters as clitics, given that they do not have lexical/word stress: Jean me connait, moi. b. 0 Janis mu to edhose, * mu! ...J emena.
(81) a.
French (France) Greek
g. The vast majority of weak-strong pairs are homophonous. The Italian dative pair loro - a loro is not. The same appears to apply in French moi - amoi. In Greek the alleged weak pronouns are homophonous with the clitic, not the strong pronoun. Moreover, in Romanian weak pronouns are obviously structurally related to clitics rather than to weak pronouns (and they are not homophonous with strong pronouns). It appears then that strong-weak homophony is not a necessary criterion. From the above discussion, it becomes evident that the weakening of person case constraint effects in DO> 10 orders in enclisis as well as the wide scope of the 10 in coordination are strong indications for the weak status of the I0 pronoun. Additional evidence may be offered from the fact that in Greek NPs genitive pronouns are weak (in the sense that they have all the characteristics put forward by Cardinaletti & Starke with regard to the weak-clitic distinction- see also Alexiadou & Stavrou 2000 and
Chapter 6. Enclisis in Greek 2.65
Lechner & Anagnostopoulou 2005 for evidence that genitive pronouns in NPs are PFclitics). This means that there is independent evidence for the existence of a weak 10 pronoun series in Greek. which lends support to the proposal that the 10 pronoun in enclitic DO> 10 clusters is a weak pronoun and nota clitic. The ordering DO > 10 could be derived ala Laenzlinger (1998) and Ordonez (2002). One possible derivation would involve both pronouns moving to the periphery ofvP, where the DO clitic incorporates into its hostv, as opposed to the 10 cliticwhich does not incorporate (being an XP) and will remain at the vP periphery, attaching to its host only at PF (this could account for the coordination facts). A possible problem with this analysis (which is also found in the Laenzlinger & Shlonsky 1997 analysis of Germanic weak pronouns as LF clitics) is that one has to ensure that nothing intervenes between the IO pronoun and the V-DOcl cluster, even if the latter moves to C (e.g. in imperatives). Although this might be straightforward for an intervening subject (cf. the claim that in Null Subject Languages [Spec, TP) does not normally project), it is not clear whether the same would apply for possible intervening adverbs. Of course, if adverbs do not count as interveners between the weak pronoun and its host (see Bobaljik 1995), this would not be a problem. An alternative derivation would involve both clitics checking their features in a higher position (e.g. Asp or T), which will ensure adjacency between the V-DOd cluster and the 10 weak pronoun. Another issue has to do with the weakening of the Person Case Constraint Most current morpho-syntactic analyses assume that the Person Case Constraint violations involve multiple/parallel checking of two goals (10 and DO) against one probe (normally V-tr for these cases- see Anagnostopoulou 2003a, Nevins 2007 for discussion). The main idea is that Person Case Constraint violations are incurred only if the dative/1 0 moves/ agrees first, followed by the DO. Anagnostopoulou (2003a, 2007) builds on this basic premise of her analysis in order to account for the repair strategy found in languages like Swiss German: presumably, the latter allows two base orders, so that a person case constraint only arises when the 10 is merged higher than the DO, but not vice versa. Now, in Greek there is strong evidence that only the 10 >DO is base generated, whereas DO> 10 orders of full DPs/pronouns appear to be the result of A-bar movement of the DO across the 10 (see Anagnostopoulou 2003a for more details). I have applied the same diagnostics to gerunds and imperatives and it appears that the same appears to be true for these structures as well. (82) a.
Dhose tu kathe fititi tin erghasia tu. Give the every studenlgen the essay.acc his.cl.gen
Greek
b.
*1Dhose
tu fititi tis tin kathe erghasia. Give the student her.clgen the every essay.acc 'Give every student his essaY.
Greek
c.
Dhixnontas tu enos pedhiu ti mitera tu alu. Showing the one child.gen the mother.acc the other.gen.
Greek
266 Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
d. *Dhixnontas ti mitera tu alu to ena pedhi. Greek Showing the mother.acc the other.gen the one.acc child.acc 'Showing to the one child the mother of the other child~ In other words, both in gerunds and in imperatives the I 0 must be merged higher than the DO, because it asymmetrically c-commands it As a result, it is not clear how the DO may move/agree first with the v-Tr head, on a par with Swiss German. Interestingly enough, a similar problem arises in Standard German, where Person Case Constraint effects do not arise when the pronouns are not in the Wackernagel position. Recall that in German the order of pronouns is strictly DO> IO, even though the base order appears to be IO >DO (see Anagnostopoulou 2007, pace MUller 1995, 2001). Building on Cardinaletti's (1999) point that Person Case Constraint effects only arise with pronouns of the same type, Anagnostopoulou (2007) proposes that in this case one of the two pronouns is parsed as strong. The intuition behind this proposal is that strong pronouns do not need to agree with v-Tr, as opposed to weak ones. One could possibly extend this proposal to the Greek case, postulating that the weak IO pronoun does not need to agree with v-TR, and that this is only possible if we have a cluster of non-similar pronoun types (following Cardinaletti's original proposal). Given that the IO clitic must move at least up to vP, we would have to assume that in this case the weak pronoun moves to the vP periphery by means of an EPP feature, without it being necessary to agree with v-TR. This would parallel the strategy found in other languages like Basque or even Greek (with strong pronouns, at least). More generally such an analysis would make clitic syntax simpler, as clitics do not appear to allow both orders cross-linguistically. Moreover, most languages having free clitic ordering have been analyzed as having weak pronouns in post-verbal position (Ordo:iiez). Moreover, Terzi (1999a) assumes two possible cliticization patterns in non-finite clauses, based on an empty T and on Kayne's antisymmetry approach. In particular, she proposes that the Linear Correspondence Axiom of Kayne (1994) (LCA) makes two options available for the way in which clitics adjoin in the clause: they adjoin to distinct heads or the one to the other. When they adjoin to distinct heads (one of which is T) the accusative > dative order arises. On the other hand, when they adjoin to each other, the dative > accusative order arises. Terzi argues that Greek chooses both options as opposed to Romance which chooses only the second option. Although her analysis makes sense, it has no typological or predictive power. It does not explain why Romance languages differ in this respect, or why in those languages where clitic switch is possible, there is additional evidence that this may be a weak pronoun (e.g. lexical stress).
6.4 Conclusions In this chapter I discussed the role of finiteness, and in particular the role of CModal in the proclisis-enclisis alternation. More specifically, I investigated the finiteness
Chapter 6. Enclisis in Greek 267
properties of a series of mood structures in Greek. and ended up formulating the following generalization: "Proclisis correlates with non-restricted/full person agreement on T, while enclisis correlates with restricted person agreement on T. Non-restricted agreement is associated with the[- imperative] paradigm and a non-inflectional CM head (and, thus, with non-obligatory movement to C), while restricted agreement is associated with the[+ imperative] paradigm and an inflectional CM head (and, thus, with obligatory movement to C)". To put it in other words, for enclisis to obtain, the verb must move to CM, a V-related site, across the initialcliticization site (v*/T), while the clitic incorporates into the edge of CM. I also discussed negated imperatives, surrogate imperatives as well as clitic switch constructions in Greek enclitic orders. I argued that true negated imperatives are ungrammatical because the negative marker is merged in CM, the head through which the[+ imperative] verb must move in order for the sentence to be grammatical. Surrogate imperatives, on the other hand, involve forms which realize only a subset of the features of a [+ imperative] form, and hence are compatible with both the[+ imperative] (CM) and the [-imperative] (T) position within the clause. Finally, it was argued that in clitic switch constructions, in the accgen order the genitive pronoun is a weak pronoun and not a clitic.
CHAPTER 7
Concluding remarks
This dissertation aimed at providing a principled analysis for two interrelated phenomena in the morphosyntax of Greek clitic pronouns: proclisis (1) and enclisis (2), respectively: (1) When the verb is in the indicative or the subjunctive, the clitic pronoun precedes the verb, and nothing may intervene between it and the verb. (2) When the verb is in the imperative or the gerund, the clitic pronoun follows the verb, and nothing may intervene between it and the verb.
Starting with proclisis, it was argued that the elitic pronoun is a topicalizer, that is an optional definite head that is merged in the left periphery of the (direct or indirect) object DP of the clause and which marks it as a topic. The optional topicalizer can be associated with an optional EP P feature, since it also has an effect on the outcome. The clitic/topicalizer is probed by v*/T as well as by the EPP feature, and moves to v*/T. When it moves to the left edge of v*/T it incorporates into it, yielding proclisis. Proclisis takes place for two interrelated reasons. On the one hand, the clitic is not a phase, as opposed to the v*/T which is a minimal phase. This is so, since the clitic contains only a subset of features of those contained by v"'"/T, and according to a relevant law incorporation is allowed only if the features of the incorporee are not different from the features of the incorporator (Roberts 2009 ). 0 n the other hand, incorporation into v*/T is allowed because the edge of the minimal phase is still accessible, due to the fact that v*/T has not checked all its features yet. The above derivation gives an answer to the what, why and where questions of Rizzi (1993): clitics are topicalizers; clitics move because they agree with v*/T and the optional EPP feature on v*/T; clitics move to the phase head that has probed them. In enclisis, now, the generalization was that restricted person agreement on T is associated with the +imperative/+gerund paradigm and an inflectional CM(odal) head. This means that in enclitic environments the verb must target the higher CM head, so as to check the special verbal morphology. I argued that the clitic pronoun targets the v*/T head, as in proclisis, however it does not incorporate into it, because the edge of the phase is not accessible; the phase in this case includes the CM head, given that the latter is inflectional/verbal, part of the INFL domain of the clause. As a result, the edge is transferred to the CM head, and that is where the elitic is able to incorporate. Therefore, the following derivation was proposed: the clitic targets v*/T as in proclisis, however it does not incorporate, given that no edge is available. The verb moves higher to
270
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
the CM head, where an edge becomes available. and the clitic, which lies in the specifier of v*IT merges with the higher edge at CM, yielding enclisis. This answers to the how question of Rizzi, namely to the proclisis-enclisis alternation question. It also pinpoints the role of CM in the alternation, which is quite central. An issue for further research would be to show how the role of CM could be generalized over that particular head to other cases ofV-related heads that allow enclisis in higher or lower positions within the clause. For example, we could have a similar derivation in a lower position with infinitives, or in a higher position with an indicative or a subjunctive (as in Pattern B languages). Another issue for further research would be to try to apply this theory to Romance clitics. Romance clitici:zation has been extensively investigated over the years, it has many phenomena, and it surely is worthwhile to see whether my analysis works with the Romance data or what kind of changes are necessary to make it work. Finally, a further issue for research would be to test the borders of this analysis by taking the two general principles (i) subset of features and (ii) edge availability/accessibility and start 'playing with them, so as to see what kind of cross-linguistic predictions they would make. Such an investigation will no doubt be highly revealing for the theory of comparative syntax.
References
Abels, K. 2003. Successive Cyclicity, Anti-locality, and Adposition Stranding. PhD Dissertation, University of Connecticut Abney, S.P. 1987. The English Noun Phrase in its Sentential Aspect. PhD Dissertation, MIT. Adger, D. (2007) Three domains of finiteness: A minimalist perspective. In Finiteness. Theoretical and Empirical Foundations, I. Nikolaeva (ed), 23-58. Oxford: OUP. Adger, D. & Harbour, D. 2007. The syntax and syncretisms of the Person Case Constraint. Syn-
tax 10: 2-37. Alexiadou, A. 1997. Adverb Placement: A Case Study in Antlsymmetric Syntax [Linguistik AktuelULinguistics Today 18]. Amsterdam: John Benjamins. Alexiadou, A. 2001. Functional Structure in Nominals. Nominalization and Ergativity [Linguistik AktuelULinguistics Today 42]. Amsterdam: John Benjamins. Alexiadou, A. 2004. Inflection class, gender and DP internal structure. In Exploration in Nominal Inflection, G. Milller, L. Gunkel & G. Zifonun (eds.), 21-50. Berlin: Mouton de Gruyter. Alexiadou, A. 2005. Possessors and (in)definiteness. Lingua 115: 787-819. Alexiadou A. & Anagnostopoulou. E. 1997. Toward a uniform account of scrambling and clitic doubling. In German: Syntactic Problems-Problematic Syntax, W. Abraham & E. van Gelderen (eds.), 142-161. Tlibingen: Max Niemeyer. Alexiadou, A. & Anagnostopoulou, E. 1998. Parametrizing Agr: Word order, V-movement and EPP-checking. Natural Language and Linguistic Theory 16: 491-539. Alexiadou, A. & Anagnostopoulou, E. 1999. EPP without Spec IP. In Specifiers: Minimalist Approaches, D. Adger, S. Pintzu.k. B. Plunkett & G. Tsoulas (eds.), 93-109. Oxford: OUP. Alexiadou, A. & Anagnostopoulou, E. 2001. The subject in situ generalization and the role of Case in driving computations. Linguistic Inquiry 32: 193-231. Alexiadou, A. & Anagnostopoulou. E. 2004. Voice morphology in the causative-inchoative alternation: Evidence for a non- unified structural analysis of unaccusatives. In The Unaccusativity Puzzle. Explorations of the Syntax-Lexicon Interface, A. Alexiadou, E. Anagnostopoulou & M. Everaert (eds.), 114-136. Oxford: OUP. Alexiadou, A. & Stavrou. M. 2000. Adjective-ditic combinations in the Greek DP. In Clitics in Phonology, Morphology and Syntax [Linguistik Aktuell/Linguistics Today 36], B. Gerlach & J. Grijzenhout (eds.), 63-84. Amsterdam: John Benjamins. Alexiadou, A., Haegeman, L. & Stavrou, M. 2007. Noun Phrase In the Generative Perspective. Berlin: Mouton de Gruyter. Anagnostopoulou, E. 1994. Clitic Dependencies in Modern Greek PhD Dissertation, Salzburg University. Anagnostopoulou, E. 1997. Clitic left dislocation and contrastive left dislocation. In Materials on Left Dislocation [Linguistik AktuelULinguistics Today 14], E. Anagnostopoulou, H. van Riemsdijk& F. Zwarts (eds.), 151-192. Amsterdam: JohnBenjamins. Anagnostopoulou, E. 1999. Conditions on ditic doubling in Greek. In Clltics in the Languages of Europe, Henk van Riemsdijk (ed.), 761-798. Berlin: Mouton de Gruyte1:.
272 Cities in Greek: A Minimalist Account ofProdisis and Endisis
Anagnostopoulou, E. 2003a. The Syntax ofDitransltives. Evide11ce from Clitics. Berlin: Mouton de Gruyter. Anagnostopoulou, E. 2003b. Review of P. Panagiotidis, Pronouns, Clitics and Empty Nouns. Amsterdam: John Benjamins, 2002. Joun1al of Greek Linguistics 4( 1): 157-170. Anagnostopoulou, E. 2005a. Cross-linguistic and cross-categorial variation of datives. In Advances in Greek Ge11erative Syntax. b1 Honor of Dimitra Theopha11opoulou-l((mtou, M. Stavrou &A Terzi (eds.), 61-126. Amsterdam: John Benjamins. Anagnostopoulou, E. 2005b. Strong and weak person restrictions. In Clltic and Affix Combblatlons. Theoretical Perspectives [L!nguistlk Aktuell/Linguistics Today 74], L. Heggie & F. Ordonez (eds.), 195-235. Amsterdam: John Benjamins. Anagnostopoulou, E. 2006. Clitic doubling. In The Bladwell Companion of Sy11tax, M. Everaert & H. van Riemsdijk. (eds.), 519-581. Oxford: Blackwell Anagnostopoulou, E. 2007. Notes on the person case constraint in Germanic (with special reference to German). Ms,lingBuzz/000519. Anagnostopoulou, E. & Giannakidou, A 1995. Clitics and prominence, or why spedficity is not enough. CLS 31. Anderson, S.R 1992. A-Morphous Morphology. Cambridge: CUP. Anderson, S.R 2005. Aspects of the Theory of Clitics. Oxford: OUP. Baker, M.C. 1988. Incorporation: A Theory of Grammatical Function Changing. Chicago IL: University of Chicago Press. Barbosa, M. 1995. Null Subjects. PhD Dissertation, MIT. Barbosa, P. 1996. Clitic placement in European Portuguese and the position of subjects. In Approaching Second: Second Position Clltics and Related Phenomena, A. Halpern & A. Zwicky (eds.), 1-40. Stanford CA: CSLI. Barbosa, P. 2001. On inversion in Wh-questions in Romance. In Subject Inversion In Romance and the Theory of Universal Grammar, A Hulk& J.-Y. Pollock (eds.), 20-59. Oxford: OUP. Barbosa, P. 2006. Two kinds of subject pro. Ms. Bartos, H. 2001. Object agreement in Hungarian- a case for Minimalism. In The Minimalist Parameter [Current Issues in Linguistic Theory 192], G.M. Alexandrova & 0. Arnaudova (eds.), 311-324, Amsterdam: John Benjamins. Belletti, A 1990. Generalized Verb Movement: Aspects of Verbal Syntax. Thrin: Rosenberg and Sellie.t: Belletti, A 1994. Verb positions: Evidence from Italian. In Verb Movement, D. Lightfoot & N. Hornstein (eds.), 19-40. Cambridge: CUP. Belletti, A. 1999. Italian/Romance ditics: Structure and derivation. In Clltics In the Languages of Europe, Henk van Riemsdijk (ed.), 543-582. Berlin: Mouton de Gruyter. Belletti, A. 2005. Extended doubling and the VP periphery. Probus 17: 1-35. Benincil, P. & G. Cinque. 1993. Su alcune ditferenze tra endisi e prodisi. In Omaggio a Gianfranco Folena. Editoriale Programma, Padova, 2313-2326. Benveniste, E. 1971. Problems in General Linguistics. Coral Gables FL: University of Miami Press. Bianchi, V. 2003. On finiteness as logophoric anchoring. In Temps et Point de Vue/Tense and Point of View, J. Gueron & L. Tasmovski (eds.), 213-246. Paris: Universite Paris X Nanterre. Biberauer, T., Holmberg, A, Roberts, I. & Sheehan. M. (2009). Parametric Variation. Null Subjects in Minimalist Theory. Cambridge: CUP. Bibis, N. 2002. The Syntax ofClitks in Idiomatic and Other Fixed Expressions. PhD Dissertation, University of Toronto. Bleam, T. 1999. Lefsta Spanish and the Syntax of Oitic Doubling. PhD Dissertation, University of Delaware.
References 273 Bobaljlk. J.D. 1995. Morphosyntax. The Syntax of Verbal Inflection. PhD Dissertation, MIT. Bobaljlk. J.D. 2002. A-chains at the PF-interface: Copies and "covert" movement. Natural La11guage andLi11gulstic.1heory 20(2): 197-267. Bonet, E. 1991. Morphology after Syntax: Pronominal Clitics in Romance. PhD Dissertation, MI't Booij, G. 1996 Clitic!zation as prosodic integration: The case of Dutch. The Li11gulstlc Review 13: 219-242. Borer, H. 1984. Parametric Syntax. Dordrecht: Foris. Borer, H. 1986. Introduction. In The Sy11tax ofPronomi11al Clitics. Syntax a11d Sema11tics, Vol19, H. Borer (ed) 1-11. Orlando FL: Academic Press. Borer, H. 1998. Mmphology and syntax. In: The Handbook ofMorphology. A. Spencer and AM. Zwicky (eds.), 151-190. Oxford: Blackwell Borer, H. (2005) Structurb1gSense, Vol I: In Name 011ly. Oxford: OUP. Bo8k.ovit, Z. 1997. The Sy11tax ofNo,ifinite Compleme11tation: An Economy Approach. Cambridge MA: The MIT Press. Bo8k.ovit, Z. 2001. On the Nature of the Sy11tax-Phonology Interface: Cliticization and Related Phe,wmena. Amsterdam: Elsevier. Bo8k.ovit, Z. 2004. Be careful where you float your quantifiers. Natural La11guage and Linguistic Theory 22:681-742. Bo8k.ovi.C, Z. 2007. On the locality and motivation of Move and Agree: An even more minimal theory. Ms, University of Connecticut Final version appeared in Linguistic Inquiry 38: 589-644. Bresnan, J. & Mchombo, S.A 1987. Topic, pronoun and agreement in Chichewa. Language 63: 741-782. Cardtnaletti. A. 1994. On the internal structure of pronominal DPs. The Linguistic Review 11: 195-219. Cardtnaletti. A. 1999. Pronouns in Germanic and Romance languages: An overview. In Clitics In the Languages of Europe, H. van Riemsdijk (ed), 33-82. Berlin: Mouton de Gruyter. Cardtnaletti. A. 2007. Subjects and Wh-questions. Some new generalizations. In Romance Linguistics 2006: Selected Papers from the 36th Linguistic. Symposium on Romance Languages (LSRL), New Brunswick, N.J., March 31-Aprl12, 2006 [Current Issues in Linguistic Theory 287], J. Camacho, N. Flores Ferran, L. Sanchez, V. Deprez & M.J. Cabrera (eds.), 57-79. Amsterdam: John Benjamins. Cardtnaletti. A & Roberts, I. 1991. Clause structure and X second Ms. Cardtnaletti. A & Starke, M. 1996. Deficient pronouns: A view from Germanic. A study in the unified description of Germanic and Romance. In Studies in Comparative Germanic Syntax II, H. Thrainsson, S.D. Epstein & S.Peter (eds.), 21-65. Dordrecht: Kluwer. Cardtnaletti. A & Starke, M. 1999. The typology of structural deficiency: A case study of the three classes of pronouns. In CliNes in the Languages of Europe, H. van Riemsdijk (ed), 145-233. Berlin: Mouton de Gruyter. Cecchetto, C. 1999. A comparative analysis of left and right dislocation in Romance. Studia Llnguistica 53: 40-67. Cecchetto, C. 2000. Doubling structures and reconstruction. Probus 12: 93-126. Chomsky, N. 1973. Conditions on transformations. InA Fe.stschrift for Morris Halle, S. Anderson & P. Kiparsky (eds.), 232-286. New York NY: Holt, Rinehart and Wmston. Chomsky, N. 1981. Lectures 011 Government and Binding. Dordrecht: Foris. Chomsky, N. 1982. Some Concept5 and CmiSequences of the Theory of Government a11d Bi11di11g. Cambridge MA: The MIT Press.
274 Cities in Greek: A Minimalist Account ofProdisis and Endisis
Chomsky,. N. 1993. A minimalist program for linguistic theory. In The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger, K. Hale & S.J.Keyser (eds.), 1-52. Cambridge MA: The MIT Press. Chomsky,. N. 1994. Bare phrase structure. In Govemment and Binding Theory and the MblimimalistProgram: Principles and Parameters in Syntactic Theory, G. Webel.huth (ed..), 383-439. Oxford: Blackwell Chomsky,. N. 1995. The Mi11imalist Program. Cambridge MA: The MIT Press. Chomsky,. N. 2000. Minimalist inquiries: The framework. In Step by Step: Essays on Minimalist Syntax in Ho,wr of Howard Lasnik, R Martin. D. Michaels & J. Urlagereka (eds.), 89-155. Cambridge MA: The MIT Press. Chomsky,. N. 2001a Derivation by phase. In Ken Hale: A Life in Language, M. Kenstowicz (ed..), 1-52. Cambridge MA: The MIT Press. Chomsky,. N. 2001b. Beyond explanatory adequacy. Ms, MIT. Reprinted in Structures and Beyond, Vol. 3 of The Cartography of Syntactic Structures, ed. by A Belletti, 104-131. Oxford: OUP. 2004. Chomsky,. N. 2005. On phases. Ms, MIT. (Now published in R Freidin. C. Otero & M.-L. Zubizarreta (eds.), 2008, Fou11dational Issues in Linguistic Theory. Essays i11 Honor of Jea1!Roger Vergnaud, 133-66. Cambridge MA: The MIT Press). Chomsky,. N. 2006. Approaching UG from below. Ms, MIT. Published in U. Sauerland and H.-M. Gartner (eds.), 2007, Interfaces +Recursion= Language?: Chomsky's Minim.alism and the view from syntax-Semantics, 1-29. Berlin: Mouton de Gruyter. Chomsky,. N. & Lasnik. H. 1993. The theory of Principles and Parameters. In Syntax: An International Handbook of Contemporary Research, J. Jacobs, A. von Stechow, W. Sternefeld & T. Vennemann. Berlin: Walter de Gruyter. Christides, A- P. 1990. On the categorial status of particles: The case for 'Holophrasis: Lingua 82: 53-82. Cinque, G. 1990. Types ofA-Dependencies. Cambridge MA: The MIT Press. Cinque, G. 1999. Adverbs and Functional Heads. A Cross-Linguistic Perspective. Oxford: 0 UP. Cinque, G. 2004. Restructuring and functional structure. In Structures and Beyond. The Carto~ raphy of Syntactic Structures, Vol. 3,A. Bellettt (ed.), 132-191. Oxford: OUP. Condoravdi, C. 1990. Sandhi rules of Greek and prosodic theory. In The Phonology-Syntax Connection, S. Inkelas & D. Zec (eds.), 63-84. Chicago I L: University of Chicago Press. Condoravdi, C. & Kiparsky, P. 2001. Oitics and clause structure. Journal ofGreek Linguistics 2: 1-40. Corbett, G.G. 1991. Gender. Cambridge: CUP. Corbett, G. G. 2003. Agreement: The range of the phenomenon and the principles of the Surrey database of agreement. Transactions of the Philological Society 101(2): 155-202. Corbett, G. G. 2006. Agreement. Cambridge: CUP. Corver, N. & Delfitto, D. 1993/1999 On the nature of pronoun movement. In Clitlcs in the Languages of Europe, Henkvan Riemsdijk (ed.), 799-864. Berlin: Mouton de Gruyter. Cuervo, M.C. 2003. Datives at Large. PhD Dissertation, MIT. Cysouw, M. 2003. The Paradigmatic Structure ofPerson Marki'IK· Oxford: OUP. Danon, G. 2001. Syntactic definiteness in the grammar of Modern Hebrew. Linguistics 39(6): 1071-1116. de Hoop, H. 1992. Case Configuration and Noun Phrase Interpretation. PhD Dissertation, University of Groningen. de Hoop, H. 1996. Case Configuration and Noun Phrase Interpretatio11. New York NY: Garland.
References 275 de Hoop, H. 2000. Optional scrambling and interpretation. In Interface Strategies, H. Bennis, M. Everaert & E. Reuland (eds.), Amsterdam: KNAW. Dechaine, R-M. & Wiltscb.ko, M. 2002. Decomposing pronouns. Linguistic Inquiry 33: 409-442. Demonte, V. 1995. Dative alternation in Spanish. Probus 7: 5-30. den Besten, H. 1983. 0 n the interaction of root transformations and lexical deletive rules. In 011 the Formal Syntax of the Westgerma11la [Linguistik Aktuell/Linguistics Today 3], W. Abraham (ed.), 47-132. Amsterdam: John Benjamins. Deprez, V. 1994. Parameters of object movement In Studies on Scrambling, N. Corver & H. van Riernsdijk. (eds.), 101-152. Berlin: Mouton de Gruyter. Di Domenico, E. 2004. Placed, non-placed and anaphorically placed expressions. Ms, University of Siena. Diesing, M. 1992. Indefinites. Cambridge MA: The MIT Press. Diesing, M. 1996. NP types and conditions on interpretation. In OTS Yearbook 1995, J. Don, B. Schouten & W. Zonneveld (eds.), 1-12. Utrecht: Onderzoeksinstituut voor Taal en Spraak, Utrecht University. Diesing, M. 1997. Yiddish VP order and the typology of object movement in Germanic. Natural Language a11d Linguistic Theory 15: 369-427. Diesing, M. & Jelinek, E. 1995. Distributing arguments. Natural Language Semantics 3: 123-176. Dimitrova-Vulchanova, M. 1999. Clit:tcs in the Slavic languages. In Clltics In the Languages of Europe, H. vanRiernsdijk (ed.), 83-122. Berlin: Mouton de Gruyter. Dobrovie-Sorin, C. 1990. Clitic doubling, Wh-movement, and quantification in Romanian. Linguistic Inquiry 21(3): 351-397. Dobrovie-Sorin, C. 1994. The Syntax of Romanian. Comparative Studies In Romance. Berlin: Mouton de Gruyter. Dobrovie-Sorin, C. 2001. Head-to-head merge in Balkan subjunctives and locality. In Comparative Syntax ofBalkan Languages, M. L Rivero & A. Ralli (eds. ), 44-73. Oxford: OUP. Drachman, G. 1997. Some properties of clitics (with special reference to Modern Greek). In Studies on Universal Grammar and Typological Variation [Ltnguistik Aktuell/Ltnguistics Today 13], Alexiadou, A. & Hall, A. (eds.), 219-248. Amsterdam: John Benjamins. Drachman, G. & Klidi. S. 1992. The proper treatment of adverbial questions in Greek: The extended minimal structure hypothesis. Studies in Greek Linguistics 13:371-390. Drachman, G. & Malikout:t-Drachman. A. 1999. Greek word accent. In Word Prosodic Systems In the Languages ofEurope, H. van der Hulst (ed.), 897-945. Berlin: Mouton de Gruyter. Elbourne, P. 2005. Situations and Individuals. Cambridge MA: The MIT Press. Embick, D. & Noyer, R 2001. Movement operations after syntax.JJnguistic Inquiry 32: 555-595. Embick, D. & Noyer, R. 2005. Distributed morphology and the syntax/morphology interface. To appear in The Oxford Handbook of Linguistic Interfaces, G. Ramchand & C. Reiss (eds.). Oxford: OUP. Espa.ilol-Echevarria & Ralli. A. 2000. Case mismatches in Greek. Evidence for the autonomy of morphology. Acta Linguistlca Hungarlca 47: 179-203. Everaert, M. & van Riernsdijk, H. 2006. The Blackwell Companion to Syntax. Oxford: Blackwell. Fischer, S. 2002. The Catalan Clitic System: A Diachronic Perspective 011 its Syntax a11d Phonology. Berlin: Mouton de Gruyte.t Fontana, J. 1996. Phonology and syntax in the interpretation of the Tobler-Mussafia law. In Approachi11g Seco11d. Seco11d Position Clitics and Related Phtmomena, A. Halpern & A. Zwicky (eds.), 41-83. Stanford CA: CSLL Franks, S. 1998. Clitics in Slavic. Ms, Indiana University.
276 Cities in Greek: A Minimalist Account ofProdisis and Endisis
Franks, S. & King, T.H. 2000. A Handbook of Slavic Clitics. Oxford: OUP. Fried.emann. M.A. & S!loni, T. 1997. AGRobj is not AGRparticlple. Linguistic Rev lew 14: 69-96. Friedman. V. 1993. Macedonlan. In The Slavonic La11guages, B. Comrie & G.G. Corbett (eds.). London: Routledge. Gallego, A. 2006. Phase sliding. Ms, UAB & UMD. Georgopoulos, C. 1991. Sy11tactic Variables: Resumptive Pronouns a11d A' Binding i11 Palauan. Dordrecht: Kluwer. Gerlach, B. 2002. Clitics Between Syntax and Lexicon [LinguistikAktueWLinguistics Today 51]. Amsterdam: John Benjamins. Giannakidou, A. 1998. Polarity Sensitivity as (non) Veridical Depe11de11cy [Linguistik Aktuell/ Linguistics Today 23]. Amsterdam: John Benjamins. Giannakidou, A. 2007. Time for mood: the subjunctive revisited. Ms, University of Chicago. Giannakidou, A. & Merchant, J. 1997. On the intetpretationofnull indefinite objects in Greek. In Studies in Greek Linguistics. Proceedings of the 17th Annual Meeting of the Department of
Linguistics, Faculty of Philosophy. Aristotle University ofThessaloniki. Giannakidou, A. & Stavrou, M. 1999. Nominal!zation and ellipsis in the Greek DP. Linguistic Review 16: 295-331. Giorgi, A. & Planesi, F. 1997. Tense and Aspect: From Semantics to Morphosyntax. Oxford: OUP. Giusti, G. 1997. The categorial status of determiners. In The New Comparative Syntax, L Haegeman (ed), 95-124. London: Longman. Giv6n, T. 1976. Topic, pronoun, and grammatical agreement. In Subject and Topic, C. Li (ed), 149-188. New York NY: Academic Press. Giv6n, 't 1990. Syntax: A FuncNonal-'I}pologlcalintroduction, VoL 2. Amsterdam: John Benjamins. Goodall, G. 2001. The EPP in Spanish. In Objects and Other Subjects: Grammatical Functions, Functional Categories, and ConfiguraNonality, W.D. Davies & S. Dubinsky (eds.), 193-223. Dordrecht: Kluwer. Goria, C. 2004. Subject CliNes in the Northern Italian Dialects. Dordrecht: Kluwer. Graffi, G.1996. Alcune rifl.essioni sugli imperativi italianL In Italiano e DialetN nel Tempo: Studi di Grammatica per Giulio Lepschy. P. Beninca, G. Cinque, T. de Mauro & N. Vincent (eds.),l43-48. Roma: Bulzonl. Grimshaw, J. 1991. Extended projection. Ms, Brandeis University. Grimshaw, J. 1997. The best ditic: Constraint conflict in morphosyntax. In L Haegeman (ed), Elements of Grammar: Handbook in Generative Syntax, 169-196. Dordrecht: Kluwer. Grodzinsky, Y. & Reinhart, T. 1993. The innateness of binding and co reference. Linguistic Inquiry 24(1): 69-101. Grohmann, K. 1997. Pronouns and the left periphery of West Germanic embedded clauses. In Werner Abraham & Elly van Gelderen (eds.), German: Syntactic Problems - Problematic Syntax. 163-189. 'llibingen: Niemeyer. Haegeman, L 1993. The motphology and distribution of object dittcs in West Flemish. Studia Lingulstica 47: 57-94. Hale K. & Selkirk. E. 1987. Government and tonal phrasing in Papago. Phonology Yearbook 4: 151-183. Halle, M. 1997. Distributed motphology: Impoverishment and fission. MIT Worki11g Papers i11 Linguistics 30: 425-449. Halle, M. & Marantz. A. 1993. Distributed morphology and the pieces of inflection. In The View from Building 20. Essays i11 Linguistics in Honor ofSylvain Bromberger, K. Hale & S.J. Keyser (eds.),lll-176. Cambridge MA: The MIT Press.
References 277 Halle, M. & Marantz, A 1994. Some key features of Distributed Morphology. In MIT Working Papers in Linguistics 21, Papers on Phonology and Morphology, 275-288. Halpern, A 1995. 011 the Placeme11t and Morphology of Oitics. Stanford CA: CSLI. Han, Chung-hye. 1998. The Structure and Intetpretation of Imperatives: Mood and Force in Universal Grammar. PhD Dissertation, University of Pennsylvania. Harris, J. 1995. There is no imperative paradigm in Spanish. In Issues in the Phonology and Morphology ofthe Major Iberian Languages, F. Martinez-Gil &A. Morales-Front (eds.), 537-557. Washington DC: Georgetown University Press. Heim, I. 1982. The Semantics of Definite and Indefinite Noun Phrases. PhD Dissertation, University of Massachusetts at Amherst. Heim, I. & Kratzer, A 1998. Semantics in Ge11erat1ve Grammar. Oxford: Blackwell Heycock, C. & Zamparelli, R 2003. Friends and colleagues: Plurality, coordination and the structure ofDP. Ms, University of Edinburgh and UniversitadiBergamo. Higginbotham, J. 1985. On semantics. Li11guisticlnquiry 16(4): 547-594. Holmberg, A 1986. Word 0 rder and Syntactic Features in Scandinavian Languages and English. PhD Dissertation, University of Stockholm. Holmberg, A 2005. Is there a little pro? Evidence from Finish. Linguistic Inquiry 36(4): 533-564. Holton, D., Mackridge, P. & Phlllppaki-Warburton, I. 1997. Greek: A Comprehensive Grammar of the Modern Language. London: Routledge. Hor nsteln, N. & Varlokosta, S. 1993. Control in Modern Greek. Proceedings of NELS 23. Horrocks, J. 1997. Greek: A History of the Language and Its Speakers. London: Longman. Hulk. A 1993. Residual verb-second and the licensing of functional features. Probus 5: 127-154. Iatridou, S. 1991. Clitics and island effects. Ms, MIT. Iatridou, S. 1993. 0 n nominative case assignment and a few related things. MIT Working Papers In Linguistics 19: 175-196. Iatridou, S. 1994. Comments on the paper by Roberts. In Verb Movement, D. Lightfoot & N. Hornstein (eds.), 243-60. Cambridge: CUP. Iatridou, S. 1996. To have and have not: on the deconstruction approach. Proceedings ofWCCFL 14: 185-201. Iatridou, S. 2000.1he grammatical ingredients ofcounterfactuality. Linguistic Inquiry 31: 231-270. Iatridou, S. & Embick, D. 1997. APROPOS pro. Language 73(1): 58-78. Isac, D. & Jakab, E. 2004. Mood and force features in the languages of the Balkans. In Balkan Syntax and Semantics, O.M. Tomic (ed.), 315-339. Amsterdam: JohnBenjamins. Jaeggll, 0. 1982. Topics In Romance Syntax. Dordrecht: Foris. Jaeggll, 0. 1986. Three issues in the theory of ditics: Case, doubled NPs, and extraction. In Syntax and Semantics 19: The Synta.r: of Pronominal Clitlcs, H. Borer (ed.), 15-42. Orlando FL: Academic Press. Jensen, B. 2002. Variation in imperative subjects: Evidence from MSc and English. Paper Presented at the Grammar in Focus, Lund University. Joseph, B. & PhiUppaki-Warburton, I. 1987. Modern Greek. London: Croom Helm. Joseph, B. 1983. The Synchrony a11d Diachrony of the Balkan Infinitive. Cambridge: CUP. Joseph, B. 1988. Pronominal affixes in Modern Greek: The case against disis. In Papers from the 24th Regio11al Meeting, Chicago Linguistic Society (1988), D. Brentari, G. Larson & L. MacLeod (eds.), 203-215. Joseph, 8.1994. On weak subjects and pro-drop in Greek. In Themes in Greek Linguistics, I. PhlllppakiWarburton, K. Nkolaidis & M Sifianou (eds.), 21-32. Amsterdams: John Benjamins.
278 Cities in Greek: A Minimalist Account ofProcllsls and Encllsls Joseph, B.D. 2002. Defining 'Word' In Modern Greek: A response to I. Phlllppaki-Warburton & V. Spyropoulos 1999. Yearbook ofMorphology 2001: 87-114. Julien, M. 2002. Syntactic Heads and Word Formation. Oxford: OUP. Kallulli, D. 2001. Direct object clitic doubling in Albanian and Greek. In Comparative Syntax of Balka11 La11guages, M.-L Rivero & A. Ralli (eds.), 127-160. Oxford: OUP. Kayne, RS. 1975. French Syntax. 1he Tra11sj0rmational Cyde. Cambridge MA: The MIT Press. Kayne, RS. 1983. Chains, categories external to S, and French complex inversion. Natural La11guageandLi11gulstic Theory 1:107-139. Kayne, R 1989a. Facets of Romance past participle agreement. In Dialect Variation and the Theory of Grammar, P. Beninca (ed), 85-103. Dordrecht: Forls. Kayne, RS. 1989b. Null subjects and clitic climbing. In The Null Subject Parameter, 0. Jaeggli & K. J. Safir (eds.), 239-261. Dordrecht: Kluwer. Kayne, RS. 1991. Romance clitics, verb movement and PRO. Linguistic Inquiry 22: 647-686. Kayne, RS. 1994. 1he A11tisymmetry ofSyJ~tax. Cambridge MA: The MIT Press. Kayne, RS. 2000. Person morphemes and reflexives in Italian, French and related languages. In Parameters and U11iversals, 131-162. Oxford: OUP. Kiparsky, P. 1973. 'Elsewhere' In phonology. In A Fe.stschrtft for Morris Halle, P. Kiparsky & S. Anderson (eds.). New York NY: Holt, Rinehart, and Winston. Klavans, J. 1985. The independence of syntax and phonology In diticization. Language 61: 95--120. Koch, H. 1995. The creation of morphological zeroes. In Yearbook ofMorphology 1994, G. Booij &]. van Marle (eds.), 31-71. Dordrecht: Kluwer. Kornfilt, J. 2007. Verbal and nominalized finite clauses in Thrklsh. In Finiteness. Theoretical and Empirical Foundations, I.Nikolaeva (ed.), 305-332. Oxford: OUP. Kotzoglou, G. 2006. Subject-verb inversion In Greek: Implications for head movement and typology. Journal of Universal Language 7: 91-137. Krapova, I. 2001. Subjunctives in Bulgarian and Modern Greek. In Comparative Syntax of Balkan Languages, M.-L.Rivero & A. Ralli (eds.), 105-126. Oxford: OUP. Kratzer, A. 1998. More structural analogies between pronouns and tenses. In Proceedings of SALT VIII, D. Strolovttch & A. Lawson, 92-109. Ithaca NY: Cornell University. Kratzer, A. 2008. Making a pronoun. Fake indexicals as windows into the properties of pronouns. To appear In IJnguistic Inquiry. Published In Lingusttc Inquiry 40(2): 187-237. Laen7ltnger, C. 1994. Enclitic clustering: The case of French positive imperatives. Rivlsta di Grammatica Generatlva 19: 71-104. Laemlinger, C. 1998. Comparative Studies in Word Order Variations: Pronouns, Adverbs and German Clause Structure [Linguistik Aktuell!Linguistics Today 20]. Amsterdam: John Benjamlns. Laen7ltnger, C. & Shlonsky, U. 1997. Weak pronouns as LF-clitics: Qustering and adjacency effects in the pronominal systems of German and Hebrew. Studia Linguistica 51:154-186. Laka, I. 1990. Negation in Syntax. 0 n the Nature of Functional Categories and Projections. PhD Dissertation, MIT. Landau, I. 2004. The scale of finiteness and the calculus of control. Natural Language and Linguistic Theory 22: 811-877. Landau, I. 2007. EPP extensions. Linguistic Inquiry 38: 485-523. Lechner, W. 2007. Interpretive effects ofhead movement. Ms,lingBuz:z/000178. Lechner W. & Anagnostopoulou, E. 2005. Clitics and adjacency in Greek PPs. Ms. Ledgeway, A. 1998. Variation In the Romance infinitive: the case of the Southern Calabrian inflected infinitive. Transactio,IS of the Philological Society 96: 1-61.
References 279 Ledgeway, A. 2007. D!achrony and finiteness: Subordination in the dialects of Southern Italy. In Finitene.ss. Theoretical and Empirical Fou11dations, I. Nlkolaeva (ed.), 335-365.0xford: OUP. Ledgeway, A. & Lombardi, A. 2005. Verb movement, adverbs and ditic positions in Romance. Probus 17:79-113. Legendre, G. 1998. Second-position ditics in a V21anguage: Conflict resolution in Macedonian. In Proceedings of the 1997 ESCOL Meeting, 139-149. Ithaca NY: CLC Publications. Lema,J. & Rivero, M.-L 1989. Long head movement: ECPvs. HMC. In NELS, J. Carter, R.-M. Dechaine, B. Philip &T. Sherer (Eds.), 333-347. Amherst MA: GLSA. Leonetti, M. 2004. Specificity and differential object marking in Spanish. Catalan Journal of Linguistics 3:75-114. Liberm.an. M. 1975. The Intonational System of English. PhD Dissertation, MIT. Liberman, M. & Prince, A. 1977. On stress and linguistic rhythm. Linguistic. Inquiry 8: 249-336. Lobed{, A. 1991. Phrase structure of ellipsis in English. In Perspectives 011 Phrase Structure: Heads a11d Lice11si11g [Syntax and Semantics 25], S.D. Rothstein (ed.), 81-107. San Diego CA: Academic Press. Lobed{, A.C. 1995. Ellipsis, Fu11ctional Heads, Licensing and Ide11tijication. Oxford: OUP. Longobardi, G. 1994. Reference and proper names: A theory ofN-movement in syntax and Logical Form. Li11gulstlc Inquiry 25: 609-665. LujruJ, M. 1986. Stress and binding of pronoWJS. In Chicago Linguistics Society 22, A.M. Farley, P.T. Farley & K.E. McCullough (eds), 248-62. Chicago IL: CLS Lyons, C. 1999. Definiteness. Cambridge: CUP. Mahajan, A. 1994. Toward a unified theory of scrambling. In Studies on Scrambling, N. Corver & H. van Riemsdijk (eds.), 301-330. Berlin: Mouton de Gruyter. Malikouti-Drachman, A. & Drachman, G. 1988. Greek ditics and lexical phonology. Phonolo~ lea 1988: 157-206. Manzln!, M.R. & Savoia, L. 1999. The syntax of middle-reflexive and object ditics: A case of parametrization in Arberersh dialects. In Studi in Onore dl Luigi Marlekaj, M. Mandala (ed.), 328-383. Bart: Adr!atica Manzln!, M.R. & Savoia, L.M. 2005. I dialetti ltalianl e romancl. Morfosintassi generatlva, Vols. 1-3. Alessandria: Edizioni dell'Orso. Marantz. A. 1984. On the Nature of Grammatical Relations. Cambridge MA: The MIT Press. Marantz. A. 1988. Clit:ics, merger and the mapping to phonological structure. In Theoretical Morphology. Approaches in Modern Linguistics, M. Hammond & M. Noonan (eds.), 253-270. New York NY: Academic Press. Marantz. A. 1997. No escape from syntax: Don't try morphological analysis in the privacy of your own lexicon. In University ofPennsylvania Working Papers In Linguistics 4(2), A. Dimitriadis, L. Siegel, C. Surek-Clark & A. Williams (eds.), 201-225. Marantz. A. 2001. Words. Handout from WCCFL XX. Marantz. A. 2006. Phases and words. Ms, NYU. Matushansky, 0. 2006. Head movement in linguistic theory. Linguistic Inquiry 37: 69-109. Mauck, S. & Zanuttin!, R. 2005. The subjects of English imperatives. In GlT worki11g papers i11 theoretical linguistics, C. Brandstetter & D. Rus (eds.), 53-85. Washington DC: Georgetown University Department of Linguistics. Mavrogtorgos, M. 2004. Greek ditics in finite and non-finite clauses. Ms, University of Cambridge. Mavrogiorgos, M. 2006. The syntax-morphology interface is at the edge: Evidence from Greek ditics. Ms. University of Cambridge. Mavrogiorgos, M. 2009. Prodisis and Endisis in Greek. PhD Dissertation, University of Cambridge.
28o Cities in Greek: A Minimalist Account ofProdisls and Endisis
McGinnis, M. 2004. Implications of an asymmetry in the representation of person. Ms, University of Calgary. Merchant, J. 1999. The Syntax ofSUence: Sluicing, Islands and Identity in Ellipsis. PhD Dissertation, University of California, Santa Cruz. Miller, P. 1992. Clitics and Const1tue11ts in Phrase Structure Grammar. New York NY: Garland Monaches!, P. 1995. A Grammar of Italian Clitics. PhD dissertation, TUburg University. Monaches!, P. 1999. A Lexical Approach to Italia11 Cliticization. Stanford CA: CSLI. Monaches!, P. 20051he Verbal Complex i11 Romance: A Case Study i11 Grammatical Interfaces. Oxford: OUP. Montalbett!, M. 1984. After Binding: On the Interpretation of Pronouns. PhD Dissertation, MIT. Morin, Y.-C. 1979. More remarks on French clitic order. Linguistic Analysis 5: 293- 312. Moro, A 2000. Dy11amic Antisymmetry. Cambridge MA: The MIT Press. Miiller, G. 1995. A-bar Sy11tax. A Study in Moveme11t Tjpes. Berlin: Mouton de Gruyter. Miiller, G. 2001. Order preservation, parallel movement, and the emergence of the unmarked. In Optimality-Theoretic Syntax, G. Legendre, J. Grimshaw & S. Vikner (eds.), 279-313. Cambridge MA: The MIT Press. Munaro, N. 1997. Proprieta Strutturali e Distribuzionali dei Sintagmi Interrogativi in Alcuni Dialetti Italiani Settentrionali (Structural and Distributional Properties of Interrogative Phrases in some Northern Italian Dialects). PhD dissertation, University ofPadova. Muysken, P. 1982. Parametrizing the notion 'Head: Journal of Linguistic Research 2: 57-7 5. Nespor, M. & Vogel, I. 1986. Prosodic Phonology. Dordrecht: Foris. Nevins, A 2007. The representation of third person and its consequences for person-case effects. Natural Language and Linguistic Theory 25: 273-313. Nikolaeva, I. 2007. Constructional economy and non-finite independent clauses. In Finiteness. Theoretical and Empirical Foundations, I. Nikolaeva (ed. ), 138-182. Oxford: OUP. Ordonez, F. 2002. Some dit:tc combinations in the syntax of Romance. Catalan Journal of Linguistics 1: 201-224. Ormazabal J & Romero, J. 2007. The object agreement constraint. Natural Language and Linguistic 1heory25: 315-347. Ouhalla, J. 1991. Functional Categories and Parametric Variation. London: Routledge. Palmer, P.R. 1986. Mood and Modality. Cambridge: CUP. Panagiotidis, P. 2000. Demonstrative determiners and operators: The case of Greek. Lingua 110(10): 717-742. Panagiotidis, P. 2002. Pronouns, Clitics and Empty Nouns [Ltnguistik AktuelULinguistics Today 46]. Amsterdam: John Benjamins. Pancheva, R. 2004. Balkan possessive ditics. The problem of case and category. In Balkan Syntax and Semantics, O.M. Toml.C (ed.), 175-219. Amsterdam: John Benjamins. Paoli, S. 2007. The fine structure ofthe left periphery. COMPs and subjects. Evidence from Romance. Lingua 117:1057-1079. Pappas, PA. 2004. Medieval Greek weak object pronouns and analogical change: A response to Condoravdi & Kiparsky, 2001. Joun1al of Greek Linguistics 5: 127-183. Partee, B. 1989. Binding implicit variables in quantified contexts. In Papers from the 25th Regional Meeting of the Chicago Li11guistic Society, Part One: 1he general sessio11, C.WUtshire, R. Graczyk & B. Music (eds.), 342-365. Chicago IL: CLS. Peperkamp, S. 1996. 0 n the prosodic representations of clitics. In b1terjaces in Phonology [Studla Grammatica41], U. Kleinhenz (ed.), 102-127 Berlin: Akademie Verlag.
References Pesetsky D. & Torrego, E. 2001. T-to-C movement: Causes and consequences. In Ken Hale. A Life 111 La11guage, M. Kenstowicz (ed), 355-426. Cambridge MA: The MIT Press. Phillppaki-Warburton, I. 1992. On mood and complementizers in Modern Greek. Readi11g University Working Papers in Linguistics 1: 5-41. Phillppaki-Warburton, I. 1995. Metakinisi rimatos ke ikatanomi tonklitlkon antonimion. Studies in Greek Linguistics 15: 892-905. Phillppaki-Warburton, I. 1998. Functional categories and Modern Greek syntax. The Linguistic Review 15: 159-186.
Phillppaki-Warburton, I. & Katsimall, G. 1999. 0 n control in Greek. In Studies i11 Greek Sy11tax, A. Alexiadou. G. C. Horrocks & M. Stavrou (eds.), 153-167. Dordrecht: Kluwer. Phillppaki-Warburton, I. & Spyropoulos, V. 1999. 0 n the boundaries of inflection and syntax: Greek pronominal ditks and particles. The Yearbook ofMorphology 1998: 45-72. Phillppaki-Warburton I. & Spyropoulos, V. 2004. A change of mood: the development of the Greek mood system. Linguistics 42: 791-817. Phillppaki-Warburton, I., Spyropoulos, V., Georgiafentis, M. & Kotzoglou, G. 2004. Moving from theta-positions: Pronominal ditic doubling in Greek. Li11gua 114: 963-989. Platzack, C. & Rosengren, I. 1998. On the subject of imperatives: A minimalist account of the imperative clause. The Journal of Comparative Germanic Linguistics 1: 177-224. Poletto, C. 2000. The Higher Functional Field: Evidence from Northern Italian Dialects. Oxford OUP. Poletto, C. 2001. Complementizer deletion and verb movement in Standard Italian. In Current Studies In Italian Syntax: Essays Offered to Lorenzo Renzi, G. Cinque & G. Salvi (eds.), 265-286. Amsterdam: Elsevier. Poletto, C. & Zanuttini, R. 2003. Making imperatives: Evidence from Central Rh.aetoromance. In The Syntax ofItalian Dialects, C. Tortora (ed), 175-206, Oxford: OUP. Pollock, J.- Y. 1989. Verb movement, Universal Grammar, and the structure ofiP. Linguistic Inquiry 20: 365-424. Pollock, J- Y. 2006. Subject ditics and complex inversion. In The Blackwell Companion to Syntax, M. Everaert & H. van Riemsdijk (eds.), 601-659. Oxford: Blackwell Portner, P. & Zanuttini, R. 2003. Decomposing imperatives. Talk Presented in the IX Giornata di Dialettologia, June 26, 2003. Postal, P.M. 1969. 0 n so-called 'pronouns' in English. In Modern Studies in English, D. Reibel & S. Schane (eds.), 201-224. Englewood Cliffs NJ: Prentice-Hall. Potsdam, E. 1996. Syntactic Issues in the English Imperatives. Phd Dissertation, University of California at Santa Cruz. Preminger, 0. 2009. Breaking agreements: Distinguishing agreement and clitic doubling by their failures. Linguistic Inquiry 40(4): 619-666. Ralli. A. 2000. A feature-based analysis of Greek nominal inflection. Glossologla 11-12: 201-227. Ralli. A. 2002. The role of morphology in gender determination: Evidence from Modern Greek. IJnguistlcs 40(3): 519-551. Ralli. A. 2003. 0 kathorismos tou grammatikou genous sta ousiastika tis Neas Elinikis (Gender determination on nouns in Modern Greek). In To Genos, A. Anastasiadhi-Symeonid.hi, A. Ralli & Dh. Hila-Markopoulou (eds.), 57-99. Athens: Patakis. Raposo, E. & Uriagereka, J. 2005. Clitk placement in Western Iberian: A minimalist view. In 1he Oxford Handrook of Comparative Syntax, G. Cinque & R.S. Kayne (eels.), 639-697.0xford: OUP. Reinhart, T. 1983. Anaphora and Sema11tlc Interpretatio11. London: Croom Helm. Reis, M. & Rosengren, I. 1992. What do Wh-imperatives tell us about Wh-movement? Natural Language a11d Linguistic Theory 10: 79-118.
281
282
Cities in Greek: A Minimalist Account ofProdisls and Endisis Reuland, E. 1983. Governing -ing. Linguistic Inquiry 14: 101-136. Revithiad.ou, A. 2006. Prosodic filters on syntax: An interface account of second position ditics. Lingua 116: 79-111. Revithiad.ou, A. & Spyropoulos, V. 2007a. A typology of Greek ditics with special reference to their diachronic development Ms, University of the Aegean. Revithiad.ou, A. & Spyropoulos, V. 2007b. Subject chains in Greek and PF processing. Ms, University of the Aegean. Rezac, M. 2007. Escaping the person case constraint: Reference-set computation in the q>-system. Linguistic Variatio11 Yearbook 2006:97-138. Richards, M. 2004. Object Shift and Scrambling in North and West Germanic: A Case Study in Symmetrical Syntax. PhD Dissertation, University of Cambridge. Richards, M. 2007. On feature inheritance: An argument from the Phase Impenetrability Condition. Linguistic b1quiry 38: 563-572. van Riemsdijk, H. 1999a. Clitics i11 the La11guages ofEurope. Berlin: Mouton de Gruyter. van Riemsdijk, H. 1999b. Clitics: A state-of-the-art report In Clltics i11 the La11guages ofEurope, H. vanRiemsdijk (ed), 1-30. Berlin: Mouton de Gruyter. Rigau, G. 1986 Some remarks on the nature of strong pronouns in null-subject languages. In Generative Studie.s i11 Spanish Syntax, I. Bordelois, H. Contreras & K. Zagona (eds.), 143-163. Dordrecht: Foris. Ritter, E. 1993. Where's gender? Linguistic Inquiry 24: 795-803. Ritter, E. 1995. On the syntactic category of pronouns and agreement. Natural Language and Linguistic Theory 13: 355-404. Rivero, M. L 1992. Adverb incorporation and the syntax of adverbs in Modern Greek. Linguistics and Philosophy 15: 289-331. Rivero, M-L. 1994. Clause structure and V-movement in the languages of the Balkans. Natural Language and Linguistic Theory 12:63-120. Rivero, M.- L. 1997. 0 n two locations for complement ditic pronouns: Serbo-Croatian, Bulgarian and Old Spanish. In Parameters in Morphosyntactic Change, A. van Kemenade & N. Vincent (eds.), 170-206. Cambridge: CUP. Rivero, M.- L & Terzi. A. 1995. Imperatives, V-movement and logical mood Journal of Linguistics 31: 301-322. Rizzi. L 1982. Issues in Italian Syntax. Dordrecht: Forls. Rizzi. L 1986. Null objects in Italian and the theory ofpro. Linguistic Inquiry 17: 501-557. Rizzi. L. 1991. Residual verb second and the Wh-criterion. In Technical Reports In Formal and Computational Linguistics 2, Universite de Geneve. Rizzi. L 1993. Some notes on Romance dtticl.zation. Eurotyp Group on Clitics. Rizzi. L 1997. The fine structure of the left periphery. In Elements of Grammar: A Handbook In Generative Syntax L. Haegeman (ed). Dordrecht: Kluwer. Rizzi L. 2001. 0 n the position int(errogative) in the left periphery of the clause. In Current StudIes In Italian Syntax. Essays Offered to Lorenzo Renzi, G. Cinque & G. Salvi (eds.), 287-296. Amsterdam: Elsevier. Rizzi. L 2004. On the form of chains: Criterial positions and ECP effects. Ms. University of Siena Rizzi. L & Roberts, I. 1989. Complex inversion in French. Probus 1: 1-30. Roberge, Y. & Pirvulescu, M. 1999. Romanian, objects and imperative constructions. In Formal Perspectives on Romance Linguistics, J.-M. Authier, B.E. Bullock & LA. Reed (eds.). Amsterdam: John Benjamins.
References 2.83 Roberts, I. 1991a. Long head movement, Compand agreement in Romance. Paper Presented at the Verb Movement Workshop at the University of Maryland. Roberts, I. 1991b. Excorporation and minimality. Linguistic Inquiry 22: 209-218. Roberts, I. 1993. Verbs and Diachronic Sy11tax. Dordrecht: Kluwer. Roberts, I. 1994. Two types of head movement in Romance. In Verb Movement, D. Lighfoot & N. Hornstein (eds.), 207-242. Cambridge: CUP. Roberts, I. 2005. Principles a,ld Parameters in a VSO language: A Case Study in Welsh. Oxford OUP. Roberts, I. 2006. Clitics, head movement and incorporation. Ms, University of Cambridge. Roberts, I. 2007. A deletion analysis of null subjects: French as a case study. Ms, University of Cambridge. Roberts, I. 2009. Agreement and head movement: Clitics, incorporation and defective goals. Ms, University of Cambridge. Roberts, I. & Roussou, A 2003. Syntactic Cha11ge. A Minimalist Approach to Grammaticalizatlon. Cambridge: CUP. Roca, F. 1992. Object clitics in Spanish and Catalan. Catala11 Working Papers 111 Li11guistics 2: 245-280. Rohrbacher, B.W. 1994. The Germanic VO Languages and the Full Paradigm: A TheoryofVto I Raising. PhD. Dissertation. University of Massachusetts at Amherst. Rooryck. J. 1992. Romance enclitic ordering and Universal Grammar. Linguistic Review 9: 219-250. Rosenbaum, P. 1967. The Grammar of English Predicate Complement Constructions. Cambridge MA: The MIT Press. Roussou, A 2000. On the left periphery. Modal particles and complementlsers. Journal of Greek IJnguistics 1: 65-94. Roussou, A 2001. Control and raising in and out of subjunctive complements. In Comparative Syntax ofBalkan Languages, M.L. Rivero &A Ralli (eds.), 74-104. Oxford: OUP. Roussou, A 2006. In the mood for controL Ms, University of Patras. Roussou A & Tstmpli. I.M. 2006. 0 n Greek VSO again! Journal ofLinguistics 42: 317-354. Rullman. H. 2004. First and second person pronouns as bound variables. Linguistic Inquiry 35: 159-168. Rupp. L. 2003. The Syntax of Imperatives in English and Germanic. Word Order Variation in the Minimalist Framework. New York NY: Palgrave MacMillan. Seidl. A & Dlmitriadls, A. 1997. The discourse function of object marking in Swahili. Chicago IJnguistic Society 331: 373-387. Selkirk. E. 1995. The prosodic structure of function words. In Papers in Optimality Theory, J. Beckman. L. Dickey & S. Urbanczyk (eds.) 439-469. Amherst MA: GLSA Sevdali, C. 2006. Infinitival Clauses in Ancient Greek: Overt and Null Subjects. the Role of Case and Focus. PhD Dissertation. University of Cambridge. Sheehan, M. 2007. The EPP and Null Subjects in Romance. PhD Dissertation. Newcastle University. Shlonsky, U. 1989. The hierarchical representation of subject agreement. Ms. University of Haifa. Shlonsky, U. 2004. Endisis and prodisis. In The Cartography of Syntactic Structures, Vol. 2, L. Rizzi (eel), 329-353. Oxford: OUP. Sitaridou, I. 2002. Non-finiteness in Old Neapolitan and Modern Greek. Studie.s i11 Greek Li11guisNcs, 700-711. Smirniotopoulos, J. C. & Joseph. B.D. 1997. On so-called 'adverb-incorporation in Modern Greek. In Greek Linguistics '95. Proceedings of the 2nd Internatlo11al Conference 011 Greek IJnguistlcs, G. Drachman. A Malikouti-Drachman. J. Fykias & C. Klidi (eds.),117-128. Gnu: Neubauer Verlag.
284 Cities in Greek: A Minimalist Account ofProdisis and Endisis
Sola, J. 1992. Agreement and Subjects. PhD Dissertation, Universitat Autonoma de Barcelona. Somesfalean, S. 2005. On two issues related to the ditic dusters in Romance languages. In UG a11d Exten1al systems. LanguagE, Brain and Computation [Linguistik Aktuell/Linguistics Today 75], AM. Di Sciullo (ed.), 27-53. Amsterdam: John Benjamins. Speas, M. 1990. Phrase Structure in Natural La11guagE. Dordrecht: Kluwer. Sportiche, D. 1990. Movement, case and agreement. Ms, UCLA Sportiche, D. 1992/1998. Clitic constructions. Ms, Published in Partitions and Atoms of Clause Structure: Subjects, Agreeme11t, Case a11d Clltics.London: Routledge. Sportiche, D. 1996. Oitic constructions. In Phrase Structure and the Lexicon, J. Rooryck & L. Zaring (eds.), 213-276. Dordrecht: Kluwer. Sportiche, D. 1997. Subject ditics in French and Romance: Complex inversion and ditic doubling. In Studies in Comparative Syntax, K. Johnson & I. Roberts (eds.). Dordrecht: Kluwer. Spyropoulos, V. 2002. A note on arbitrary null subjects. Reading Working Papers in Linguistics 6:85-100. Stowell, T. 1981. Origins ofPhrase Structure. PhD Dissertation, MIT. Stowell, T. 1982. The tense of infinitives. Linguistic Inquiry 13: 561-570. Sufl.er, M. 1988. The role of agreement in ditic-doubled constructions. Natural LanguagE a11d Linguistic Theory 6: 391-434. Swler, M. 1994. V-movement and the licensing of argumental Wh-phrases in Spanish. Natural LanguagE and Linguistic Theory 12: 335-372. Swler, M. 2002. The lexical preverbal subject in a Romance null-subject language: Where art thou? In A Romance Perspective on Language KnowledgE and Use [Current Issues in Linguistic Theory 238], R. Nui1ez-Cedeil.o, L. Lopez & R. Cameron (eds). Amsterdam: John Benjamins. Szabolcsi, A 1994. The noun phrase. In The Syntactic Structure of Hungarian. Syntax and Semantics 27, F. Kiefer & K.E. Kiss (eds.),179-27 5. New York NY: Academic Press. Takano, J. 2004. Coordination of verbs and two types of verbal infiectioiL Linguistic Inquiry 35: 168-178. Tantalou, N. 2004. Gerunds in Modern Greek. Ms, Johns Hopkins University. Terzi, A. 1992. PRO in Finite Clauses. A Study of the Inflectional Heads of the Balkan Languages. PhD Dissertation, CUNY. Terzi, A 1997. PRO and null case in finite clauses. The Linguistic Review 14: 335-360. Terzi, A 1999a Clitic combinations, their hosts and their ordering. Natural Language and Linguistic Theory 17: 85-121. Terzi, A 1999b. Cypriot Greek ditics and their positioning restrictions. In Studies In Greek Syntax, A. Alexiadou, G. Horrocks & M. Stavrou (eds.), 227-240. Dordrecht: Kluwer. Theofanopoulou-Kontou, D. 2000. '-0/-me alternations in MGpatient oriented constructions: Anticausattves and passives: Proceedings of the 20th Annual Meeting of the Linguistics De-
partment of the University ofThessalonikl. Thr~nsson,
H. 2001. Object shift and scrambling. In The Handbook of Contemporary Syntactic Theory, M. Baltin & C. Collins (eds.), 148-202. Oxford: Blackwell.
Ticio, M.E. 2003. On the Structure ofDPs. PhD. Dissertation, University of Connecticut. Ticio, M.E. 2005. Locality and anti-locality in Spanish DPs. Syntax 8 (3): 229-286. Timberlake, A 1976. Nonfiniteness in Finnish. Papers from the Transatlantic Finnish Conference. Texas Linguistics Forum 5: 129-150. Tomic, 0. M. 1996. The Balkan slavic clausal clitics. Natural La11guage and Linguistic Theory 14: 811-872.
References 2.85 Tomic, 0. M. 2001. The Macedonian negation operator and diticization. Natural Language a11d Linguistic Theory 19: 647-682. Torrego, E. 1988. Pronouns and determiners: A DP analysis ofSpanish nominals. Ms, University of Massachusetts at Boston. Torrego, E. 1998. The Dependencies of Objects. Cambridge MA: The MIT Press. Tortora, C. 2002. Romance endisis, prepositions, and aspect. Natural Language and Linguistic Theory 20(4): 725-757. Tsakall, V. 2004. A different type of ditic doubling configurations. Ms, UCL/MIT. Tsakall, V. 2006. The Syntax and Acquisition ofPronominal Clitics: A Crosslinguistic Study with Special Reference to Modern Greek PhD Dissertation, UCL. Tsimpll, I.-M. 1995. Focusing in Modern Greek. In Discourse Configurational Language.s, K.:t Kiss (ed.), 176-206. Oxford: OUP. Tsimpll, I.-M. 1999. Null operators, ditics and identification: A comparison between Greek and English. In Studies in Greek Syntax, A Alexiadou, G. Horrocks & M. Stavrou (eds.), 241-262. Dordrecht: Kluwer. Tsimpll, I-M. 2000. Gerunds in Greek. Journal of Greek Lb1guistics 1: 133-169. Tsimpll, I.M. & Stavrakaki, S. 1999. The effects of a morphosyntactic deficit in the determiner system: The case of a Greek SLI child Lingua 108: 31-85. Tsoulas, G. 1996. Notes on temporal interpretation and control in Modern Greek gerunds. York Papers In Linguistics 17: 441-470. Tzak.osta. M. 2003. The endisis-prodisis asymmetry in the acquisition of ditics in Greek. Talk presented at the 6th International Conference of Greek Linguistics, University of Crete. 18-21 September. 2003. Uriagereka. J. 1995. Aspects of the syntax of ditic placement in Western Romance. Linguistic Inquiry 26: 79-124. van der Does J. & H. de Hoop. 1998. Type-shifting and scrambled definites. Journal of Semantics 15: 393-416. Varlokosta. S. & Hornstein, N. 1993. Control in Modern Greek. University ofMaryland Working Papers In Linguistics, 144-163. Veloudis, I. & Philippaki-Warburton, I. 1983. I ipotald:lki sta Nea Elinika (Subjunctive in Modern Greek). Studies in Greek Linguistics 4. Vergnaud, J.-R. & Zubizaretta, M.-L. 1992. The definite determiner and the inalienable construction in French and English. Linguistic Inquiry 23: 595-652. Vigario, M. 1999. On the prosodic status of stressless function words in European Portuguese. In Studies on the Phonological Word. T.A. Hall& U. Kleinhenz(eds.), 255-294. Amsterdam: John Benjamins. Vikner, S. 1995. Verb Movement and Expletive Subjects in the Germanic Languages. Oxford: OUP. Vincent, N. 1998. On the grammar of infiected non-finite forms (with special reference to Old Neapolitan). In Clause Combining and Text Structure, Copenhagen Studies in Language 22, I. Kor:zen&M. Herslund (eds.), 135-158. Vos, R. & Veselovska, L. 1999. Clitic questionnaire. In Clitics i11 the La,~guages ofEurope, H. van Riemsdijk. (ed), 891-1010. Berlin: Mouton de Gruyte.t Wackernagel, J. 1892. Ober ein Gesetz der Indogermanischen Worstellung.Indogermanische Forschu11gen 1: 333-436. Wanner, D. 1991. The Tobler-Mussafia law in Old Spanish. In Current Studies i11 Spanish Lingu~ tics, H Campos & F. Martinez-Gil, 313-378. Washington DC: Georgetown University Press.
286 Cities in Greek: A Minimalist Account ofProdisls and Endisis
Wurmbrand, S. 2003. bifinitives: Restructurb1g a11d Clause Structure. Berlin: Mouton de Gruyter. Zanutt!ni, R. 1996. On the relevance oftense for sentential negation. In Parameters a11d Functional Heads. Essays in Comparative Sy11tax, A. Belletti and L. Rizzi (eds.), 181-207. Oxford: OUP. Zanutt!ni, R. 1997. Negation and Clausal Structure. A Comparative Study ofRoma11ce Languages. Oxford: OUP. Zanutt!ni, R. 2007. Encoding the addressee in the syntaJC Evidence from English imperative subjects. To appear in Natural La11guage and Linguistic Theory. Zec, D. & Inkelas, S. 1991. The place of ditks in the prosodic hierarchy. In .Proceedings ofWCCFL 10, D. Bates (ed), 505-519. Stanford CA: SLA. Zeijlstra, H. 2006. Don't negate imperatives! Imperatives and the semantics of negative markers. In Proceedi11gs ofSinn und Bedeutung 10, C. Ebert & C. Endriss (eds.), 453-468. Berlin: ZAS. Zubizarreta, M.-L. 1998. Prosody, Focus, a11d Word Order. Cambridge MA: The MIT Press. Zwicky, A. 1977. 011 Clitics. Bloomington IN: Indiana University Linguistics Club. Zwicky, A. 1985. Clitks and particles. La11guage 61:283-305. Zwicky, A. & Pullum, G. 1983. Clitic!zation vs. inflection: English n't Language 59:502-513.
Index A A/A- bar distinction 232 A/Phi-probe EPP 232, 233, 238, 239· ::142, 244-2.48 A-bar chain 129 A-bar head 169 A-bar EPP 232 A-bar movement 265 Acc-dat 255, 266 Accessible 247, 269 Accusative clitic 101 Accusative head 101 A-chain 128 Addressee 212, 213 Adjectival constructions 118 Adverbal diticiza:l.ion 153 Adverbial incorporation 47, 12.4 Adverbs 128, 265 Affix hopping heads ::147, 2.48 Affixal ditic 35
Affixes 2.40 Aftos: 3 rd person strong pronoun in Greek 89 Agr 182,241 Agree 6o, 61, 75, 100, 169, 236 Agree/move 185 Agreement (morpheme) 98,142 AGRo 1oo A-head 169,173-176 Albanian 72, 1¢, 169, 171, 173, 258 Anagnostopoulou 137 [+ anaphor; + pronominal] 175 Anaphoric/null tense 204 Animacy/human 13, 61, 62, 67, 92, 98, 141, 234 Applica:t.ive construction 101 Applica:t.ive head 101, 138 Apposta 181 A-properties of chains 50 Arbitrary pro 31 As- horta:l.ive Greek particle 195-197, 210 A-spec 243
Aspect 194. 195, 203, 207 Aspectual adverbs 221 Auxiliary 5, 135-136, 180, 182-184 Auxiliary conjunction 123 Avoid Pronoun Principle (Chomsky 1981) 33
B Badioto 227 Balkan 156, 166,169,173 Bare DP analysis of pronouns 87 Bare nouns J.40, 141 Bare output conditions 62 Bare Phrase Structure 62, 102, 119 Basque 266 Belletti, A. 182 Benincl & Cinqllil 121 Bien 190 Binding Theory 25, 108 Bleam 95 Borer 91 Bound variable 27, 28, 30, 31, 34,108 Bulgarian 72,154,169.171, 173, 2.43,258
c c
160, 166, 167, 169, 170, 172, 179· 1% 21l, 228, 2JO, 246 C-flavor 228, 229 Cardinaletti, A. & Starke M. s.
n.
u-15,32--34. 37--39, 74, 104, 107, 132, 263 Case 6, 57, 59-61, 67, 77, 92, 97, 104,175 Case assignment in situ 105 Case marker a 91 Case marker pe 91 Catalan 177 Categorial features 132 Categorial sta:tus/fea:tures of pronouns 57, 76, 88, 108
Categorizing heads 134 C-command 48 Cerlaitily 209 CFin 3, 16o, 165, 193, 196, 216 CForce 196, 211, 227-231, 246, 247· 251-253
Che 179 Chichewa 5 Choice Principle 26, 27, 32,34 Chomsky & Lasnik 174 Chomsky 62, 75, 102, 12.7, 130, 175 Cht(h)es 182, 183 Cinque 187 Cinterrogative 228, 229 Clause typing head 198 Clitic climbing 118, 119,131, 137 Clitic coordination 121, 122 Semmrtic restriction on 12.1, 122 Clitic Criterion 93 Clitic doubling 91, 94, 103, 113, llj,l2.9,130, 138, 141,143,144
Obligatory of the genitive clitic in Greek 118 Of definite dired objed DPs in Greek 110, 141, 142 Ofgeneric indefinites in Greek 139 Of it1dired objed DPs in Greek 141, 142 Of nm1-rejerential quantifier in Greek 141 Clitic Group 35 Clitic Left Dislocation (CLLD) 16 Clitic movement 168, 2JJ, 234. 2J6,2.49 Clitic pronoun( object) 1, 2, 12, 13,15,17,19. 21-23,26,32.34. 37-40, 47> 49. 51, 58, 6J, 75. 76. 97. 98, 101, 104, 108, 113, 128, 132, 136, 137· 143· 145. 168, 173, 177-179, 182-184. 197, 233· 2.40· 2.47-2.49· l51, 258, 26o-265, 26!} and coindexation properties of 101
288 Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
and head properties of 41, 51, 63,131
and internal structure of 52 and lexicalist accounts of 42, 43.49· 50
and LP properties of 3, 113, 134> 144·153
Base Generation approach to 91, 96, 113,115 Movement approach to 113 PF approach to 119, 120, 124. 126,188
Syntactic approach to 119, 126,188
and morphological/sytltactic
CM(odal) 3, 193,196,205-207,
properties of 153 and morplro-synfactic status of 2,52 and movement properties of 64,101 and PF properties of J, 113,
210, 217-21!), 223-228, 231,
134.144,153
and plri-Jeafllres 2, 22, 59
See also nomiml iriflectior~al foat11res and plronological properties of 153 and syntactic accounts of 42, 43.49· 50
and XP properties of 63, 130, 173, 243. 263
as a bour1d variable 21, 22, 2J,31, 109
as a definite dete1111ir1er 111 as a dete1111iner 10, 95 as a fopicalizer 1, 129, 134. 269
as agreement marker 72, 91, 95· 96, 101, 102, 142, 234
as both maximal and minimal projection (before ir1cmporation) 62, 177, 233, 2'f8,249
as bllmlk ofD foat11res 93 as morpheme/lexical affix 42, 50, 52, 145
as syntactic head! morpheme 52, 91, 102 as syntactic word 42, so, 52 as pronominal S~~jJix head 95
as S~~bject agreement head 97
as XO head 249, 261, 264 PF- 134,265 Clitic switch 193, 255, 267 Clitic Voice 92 Cliticization 1, 62, 63, 78, 98, 113, 119,121, 126, 128-130,153,167, 168,215
2J8, 239· 24l-2j0, 252, 253· 266, 269, 270
and inflectionaW-related 2, J, 19J, 224.231, 2J8, 244. 250, 251, 267, 269
and non-inflectional/non- V-
related 193, 224. 238, 246, 247,267
Edgeof193 Colloquial Standard German 13, 16 Command 217 Complement features 134 Complement Matching Requirement 92 Complementary distribution 113 Complementizer 5, 196 Complementizer deletion constructions in Standard halian 177 Complex Inversion i French 177 Complex T head 127 Complex Tense conjunction 124 Computational system 62 Contrastive stress 264 Control Theory 175 Controlled subjunctive clauses in Greek: 147,159.204,241 Coordination 12,48,105 COp 196, 198, 200, 20 5, 206, 209-211, 218, 219, 227, 228, 230
Core Functional Categories (CFCs) C, T, v* 231 Cosentino 189 Covert N° head 90 CPadverbs CP domain 179, 185, 206, 212 C-position 196 Criteria by Cardinaletti & Starke 1999; Kayne 1975 12, 261 Csub 196, 198 C-T domain 159,168, 194,236, 237
CypriotGreek 155
D D- head 48, 64, 73, 132 Dative> accusative order 255, 266
Dative 72 Dative shift construction 138 Declarative clause 180, 20D-20l, 230
[def] 53 Default agreement 142 Default ending -ondas 220 Default phi-features 100 Defective AgrC 212, 213 Defective intervention 100 Deficiency hierarchy 104 Definite article 10, no, 121 Definiteness feature 62, 73, 77, 104. 105, 133, 142
Deictic temporal (realis) adverbs 221 Deictic tense adverbials 198 Demonstrative pronoun 6, 14, 15, 25, 26, J1, 7(>, 79· 82 Deontic modality negation marker mi(n) 203, 217, 250, 253
+1- dependent clause 158 Dependent forms in Greek verbal paradigm 195 Dependent tense 204 Derivation by Phase 127 Derivational affix 89 Determiner 66 D-feature 232 Direct Object 266 Direct object construction 101 Distributed Morphology 52, 66, 126,191
Distribution 12, 17 D-linked wh-words 199 Double obje...-r. construction 142 Doubled DP 91, 92, 96,102, 142
Doubled DP hypothesis 96 Doubling parameter 102 Doubly filled voice filter 93 DP hypothesis 65, 66 Duke 171 Dutch 70, 259 Dynamic view to pronoun construction 53
Index 289 E
F
Echolicha construction 113,
P-' head 64. 69, 73
13-•b 1.44 Edge feature (EF) 231, 232, l35-l36,237 Edge of a phase 2, 3, 129, 130, 1J2, lJJ, 142, 247· 269, 270 Eftichos 221 Egho:lst person strong pronoun in Greek 89 mHptical DP S9 Embedded yes/no interrogative 201 Empty Category Principle (ECP) 46, 174, 219, 251 Empty Pronominal Noun EN So, S1, S6-SS Bndisis as a morphological phenomenon 1SS Bnclisis 1--3,145, 1.46, 158, 165-173· 177,180-184.186, 189, 224. 227, 230, 236, 23!}-24l, 244,269 Enclitic 36, 1.46,149-151, 153, 157-159,161-163,172, 173, 18o, lSl, 184. 18!), 209, 21S, 220, 22J, 239. 242, 250' 260, 261, 263,267 Bncliticization 146 English 13, So, 166,229,247 English one 82 Entelos 125 Epistemic adverbs 209 BPP driven movement 103, 118 BPP feature 2, 75, 117, 129, 130, 134. 136, 194, 2Jl, 2J2, 237. 239. 243.246-250,266,269 BPP property on T 232, 234. 239 Esi: 2nd person strong pronoun in Greek 89 Ethical datives 115, 116 European Portuguese 49.154, 169,177 Evaluative adverbs 209, 221 Excorporation 17S, 239. 242, 249 Expletive 31, 70 Expletive uses 13 Extended INFL domain 249. 269 Extended Projection 65, 66, 87,248 Extended Verbal Projection 39, 49. 73. 91, 95. ¢, 102, 172, 176,206
FamiUarity/- focus/old information feature 129, 133, 134. 141,1.42 Familiarity D-head 134 Feature checking 6o, 61, 73, 74.92 Feature inheritance 127, 128, 131, 160,232 Feature matrix substitution/ relocation 133 Feature restructuring 132, 133 Finite forms J.4S, 153, 161, 166, 169, 172, 1S2, 18-'b 240 First position 153 Floating quantifier 117, 12S Focalization 12 Focus (syntactic) 153, 197, 198 Focus features 105 Foot J4 Force/illocutionarytype feature lOO, 230, 253, 254 Fragmentation view to pronoun construction 53, 94, 95 Pranmmente 182 Free clitic 35 Free subjunctive clauses in Greek 204 French 12, lJ, 16, 17,41, 49, 91, 103, 113,114,122,145, 151, 152, 183, 1S7, 229, 239· 255-257· 260, 261,264 French object ditics as proPs 108 Full Interpretation Principle 75 Full/tonic pronoun 2 See also strong pronoun Fusion 100
G Galician 64 Gallego, A. 235 GB case theory 91 GB framework 159,168 Gender 6, 53, So, S6-S9, 97, 105,149 Generalized Licensing Criterion 93 Genitive pronouns inside NPs in Greek 264. 265 as ditics 265 German weak pronouns 261
Germanic languages 5, 12, 13, 39. 134, 234. 263 Gerundlabsolutive 1, 17, 119. 1.47-152, 169,171, 172, 180, 181, 184-186, 189, l02, l12, 220, 227,228,230,237, lJ8, 241, 243· 244· 248, 250, 251, 265, 266,269 as active participle 220 Gerunds as adjuncts 220 Giv6n 158 Goal 100 Goria 212 Graffi 212 Greek J, 5, 10, 13, 16, 21, 32--35, 38-40, 46,49· so, 57· 61, 72· 77· 7S, SJ, S9, 95, 9S, 100, 102, 107, 113-116, 11S, 121, 122, 126-12S, lJO, 134,137,139,142,145, 147, 156,157.169-171,173-177. 1S5, 1S7, 191,193, 194,201,219, 227, 229, lJ4, 235· 237· 252, l54, 255· 260-26j, l66, 267, l69
See also Standard Modern Greek Greek aspectual adverbs 125, 1S2 Greek clitic pronouns as 109-110 Greek manner gerunds 161, 221 Greek non-embedded na clauses 163 Greek preverbal subjects as topics 19S Greek raising constructions 163 Greek temporal gerunds 161, 221 Greek tense adverbs 1S2, 209, 221 Greek tensed subjunctives 161 Greek tenseless subjunctives 161 Grimshaw 66, 172 Group feature 54
H Halpern 146 Head 39. 173, 174
See also X!' amstituent Head Adjunction 47, 51, 52, 126, 1S9, 190, 240, 241, 248' 249 in the Lexicon 124 in the syntax 124 Head conjunction 121 Head movement 131, 168, 173, 174. 233· 248, 249
290
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
Head Movement Constraint (HMC) 46,102,118,170,174, 219,251
a Relativized Minimality approach to the 169 Head skipping 239 Head substitution :140, 241 Head-Spec fashion 245 Hebrew 86--87,115,166 Hiatus resolution 36 High 188 Hungarian 73, 98, 99 Hypothetical se deletion constructions in hypothetical and counterfactual clauses in Standard Italian including Aux-to-Comp 177,179,180, 189,229
T.dhi 125, 182, 183, 202 Idiom 116 Idiomatic constructions 21, 22,31
llikrina 200 lU~iD~iUud
64
Imperative Agr 214 2nd person sg imperative ... true funn 207, 2.08 Imperative as mood category 210,216
+imperative/+gerund paradigm 2,3,193,194,207, 216, 224. 226, 247· 252, 253· 255. 267, 269
-imperative (indicative) paradigm 3. 193-196, 203, 206, 207, 216, 224, 226, 247· 252, 253. 267
Imperative C 212 Imperative furce 217 Imperative illocutionary force 210 Imperative modality 218 Imperative morphology 242. Imperative subjects 212. Imperative tense as nonfinite unrealized future tense 215 Imperative 1, 17, 36, 119, 145, 147-152,158, 159· 163, 169,172, 174.176, 177, 179· 180,184. 186, 189, 195· 202, 206, 209, 212, 216, 219,223,227,228,230,231, 237.238,241, 243. 244,248, 254. 255, 26j, 266, 269
and forrnt.ll!polite imperatives 150, 212 and negated imperatives 3, 170, 174.193, 217, 252, 253· 267
and surrogate imperatives 3, 170, 171, 184, 189, 193· 218, 22.8,267
See also + imperative/+ gerund paradigm as suppletive furms 217, 254, 255
Imperfective aspectual adverb 221 Impersonal 13, 31 Incorporated 46 Incorporation 2, 3, 113, 124, u6, 131-134. 137. 138, 144, 1]6, 177, 183, 186--190, 231, 233. 234. 244--248 Incorporator 132 Incorporee 132 Indicative 1, 17, 145,147-152,158, 159· 176,184,186, 195· 198,203, 207, 227--229, 238, 244, 247· 254,269 Indirect object 157, 266 as an accusative DP 137 as a PP 137 Indirect object as a genitive DP 137 Indirect object construction 113,
Interrogative li-marker 180 Interrogative subjunctives in Greek 205 Intervention effect 101, 251 Intransitive D0 hypothesis 76 Intrusion 130 IO >DO, DO> IO orders of full DPs/pronouns 2.65, 2.66 IO >DO and DO > IO enclitic dusters 26o--262 IO as a weak pronoun 258 Irrealis modality 255 Isos 221 halian 16, 39, 41, 49, 103, 113, 114, 125, 148, 149· 158, 180, 184. 187, 189, 227, 229, 235· 237· 239· 252,254 Italian dialects 125 Italian 1oro 2.63 Italian positive imperatives 212
Jaeggli 91 Japanese kare 8l Joseph 173 K Kala us, 220 Kano- constructions 118 Kilpote 182
Kayne, R. 5, 12, 13,39.41,49,102, 113, 114.145· 1¢, 168, 173, 187
137· 144
Infinitive morphology 242 Infinitives 147,149, 151, 158,162, 169, 173, 180-185, 187, 189, 193. 212,241 INFL 118, 184. 244, 2.51-253,
Kayne's generalization 91, 96, 97· 139 Kayne's movement hypothesis 128 Kratzer A. 22
255,259
Inflectional ending 89 Inflectional morphology 241 Inherent clitics 115, 116 Inherent stress 13 Internal clitic 35 Interpolation 120, 124, 128, 183, 185, 18r189, :131, 2.39-241, 2.47· :148,261 Interpretable features 100 Interpreted anaphorically 161 Interpreted as specific 65 Interpreted deictically 161 Interrogative clause 179, 185, 186, 199· 200, 202, 203, 230, 247,252 Interrogative head 2.01
L
Label 132 Landau, I. 166 Late Insertion 191, 213 Ledgeway; A. & A. Lombardi 125,173 Lexical 124 Lexical concept 76, So Lexical content 75 Lexical feature 73, 134 Lexical Insertion 52 Lexical Integrity Hypothesis 43, 44
Lexical N head 105 Lexical phase 2, 131, 132 Lexical properties of verbs 159
Index Lexical root 66, 89, 132-134 Lexical stem 88 Lexical/word level stress 258, 260,264 Lexicon 43, 146, 179 LF 61, 75, 8o, 81, 103, 119, 158, 159. 21J, 2.J2, 2JJ, 235. 247· 265 LF/covert movement 61, 62., 74 LF-related finiteness 158 Light verb 66, 88, 136 Linear Correspondence Axiom (LCA) 121, 266 Locality driven movement 103 Locality e1fe...-ts 101, 243 Logophoric centres of the clause 160, 216 Long Distance Agree 2 205,22.8 Long Head movement (LHM) 169, 171, 236 Long object preposing 119 Low 188 Low manner adverb 125, 220 Lower (vP) adverbs 18o, 182, 198 L-related head 174 Lui 259,264
M M- head 171, 173, 175, 176 Macedonian 146, 152, 157, 257 Mainland Scandinavian 13 Mapping Hypothesis 65 Marantz 131 Maximal projection 39
See also XP constituent Mi!/te 258, 260 Medieval Greek 155 Medieval Romance 155 Medieval Spanish 49 Merge 236 Merger 52, 245, 248 Metrical Stress Theory 34 Minimal non-phase 113 Minimal phase 2, 12.9, 1J1-1J3, 142,269
See also leximl phase Minimalism 168 Minimalist Inquiries (MI) 6o, 61 Minimalist Programme (MP) 61,158,159, 236 Minimallytransparent 132 Minimise Structure 33 Modal interpretation 196 Modal particle 2.18
Modality 176, 196, 206, 223, 227, 2JO, 2J6, 243, 251, 253, 254 Modes of 240 Modification 12, 17, 105 Moi/toi 258, 264 Monosyllabic irregular imperatives 208 Montalbetti 31 Mood 146, 147,149.150,153, 157, 158, 16o, 169,170, 194,206, 227, 2J9, 251 Mood/modal head 198, 227, 2J6,2J8 Morpheme concatenation 52 Morphological deficiency 13, 15 Morphological finiteness 157, 158, 166,172, 19J, 240 Morphological person agreement 225 Morphological subject agreement 162, 163, 165-168, 198, 205, 220, l2J, 240 Morphological tense 161, 163, 165, 198, 203, 205--207, 2:10, :123, 224, 240 Morphology 172, 241 Morpho-phonemic processes 261 Morpho-syntactic adjunction 130 Morpho-syntactic properties 172 Morpho-syntactic properties of heads 174 Movement 67, 73, 77, 187, 249 Multiple Cliticization sites 185188,190 Multiple Spell Out 127 Multiple V-sites 185
N n/N phase 132. Na-demonstrative 7 Na-subjunctive particle 41, 171, 175. 195. 197. 198, 203--205, 210, :117, :118, 219, 2.26--22.8, 246, 252,253 Na-drop 218 Neg 169, 170, 173, 174, 176 Negation 39, 45, 171, 174. 197, 200, 205, 206, 215, 218, 21!), 251, 252, 253 Negative 206 Neutral context 33
Nominal D feature 117 Nominal inflectional/phi- features of pronouns 57,59-61, 75· 77· 92, 96, 97· 100, 1% 105, 108, 128, 1J3 Nominalizer n 53, 88 Nominative 7 Non D-linked wh-words 199 Non- finite forms 148, 151,153, 158,161,166, 169,172,173,182, 189, 212, 214. 24Q, 242· ::148, 266 Non quantificational, non pronominal subjects 212 Non-embedded li-questions in Macedonian 177, 179.180 Non-L-related head 174 Non-phase 132 as a relatioml notion 133 (Non-) phasehood 143 Non-phonological reduction/ restructuring 105 Non-referential 20--22 Non- referential quantifier 140 Non-referential uses 13 Non-referential/anaphoric agreement 204. 2.14 Northern Italian Dialects (NIDs) 212, 229 North-western Ibero-Romance varieties 155 Norwegian 260 Null interrogative complementizer 200 Null N(P)-pro 66, 73, 74 Null Subject Language (NSL) 29, 232, 235.237, 243-245. 26j Null subjects 212 Number 6, 53, 61, 67, 87, 89, 97, 105, 1J8, 1.;8, 149. 211-21J, 244 Numeration 100
0 Object agreement 131, 149 Object Shift 128, 134, 234 Obligatory inclusive interpretation 149,150,212 Obviated Romance subjunctives 162 Obviated Romanian and Albanian subjunctives 163 Occurrence feature ( OCC) 231 Ostension 20 Oti/an 197, 198
291
292
Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
Overt movement 61 Overt Nll head 86 Overt nominative subject 158, 159· 16J, 217 Overt pronominal subjects 212. Overt Pronoun Constraint 33
p Pa: interrogative marker in Pera diFassa179 Pair merge 133 Panagiotidis, P. 76, 82, 87, 90 Pancheva.. R. 156 Panta 221 Parallel checking of two goals against one probe 265 Parasitic gaps so, 117 Participant features 54 Participle 135-137, 169 Participle agreement 113, 11.4. 118,119 Pas 183, 184, 190 Passive construction 117, 143 Passive small clause 137 Past participles 118, 119. 149, 184 Pattern A languages 155-158, 169,177, 186,237.249 Pattern B languages 156,157,270 Pattern C languages 157 Pera di Fassa 179 [-perfective] stem 220 Periphrastic future construction 196 Person 6, 61, 64, 65, 67, 68, 71, 73. 74· 77. 89, 97. 105, 1J8, 148, 191, 211-214, 216, 214, 232, 244 1st person 54, 6s, 71 2nd person 54. 65, 71 Jrd person 53, 64, 65, 68, 69, 86,89 Person agreement on T and non-restricted 3, 193, 214, 225,267 and restricted 2, 3, 193,223-225, 226,239.248.267,269 Personal pronoun 57, 82. Person-Case constraint 138,157, 237. 261-26j, 265,266 PF 81,100,119,125,126,158, 168, 191, 21J, 214, lJO, lJ2, 233, 235, 246-149 Phase head J, 88, 12.7, u8, 131, 132, 134· 190' 2J2, 235. 245. 269
Phase Impenetrability Condition 130 Phase sliding 235, 147 Phi probe on T 232 Phi/INFL-chain :1.44.145 Phi-features 132,160, 165, 166, 232 Phonological host 5, 134 Phrasal stress 264 Phrase structure 39
See also Bare Phrase Structure Phrase structure status of pronouns 57
Pitlranon
2.21
Polarity (assertion negation) features 105 Positive 206, 207, 209, 216, 217 Positive imperatives with an irrealis meaning 217 Possessive 156,157 Possessors 156 Post Spell Out morphological component 126, 254 Present tense 158 (pre-) Verbal particles 39, 44, 47· 171,188,195, 199 Principle B 26, 108, 109 Principle C 25, 26, so, 54,108 Principle of Full Interpretation 134 PRO 161, 162., 165, 166, 204. 206,220 Pro- DP 108
English 1st and 2nd person pronouns as 108 Greek strong pronouns as 109 Pro- cliP 108
English 3rd person pronouns as 108 Pro- NP 108 Probe 100, 12.8 Proclisis 1-3,36, 113, 145, 158, 165-169, 171,173, 177, 1]8, 180-182., 186, 189-191, 193, 194. 206, 22J, 214. 227, 231, 236, 140, 242, 245. 246, 256, 267,269 Proclitic 145,150, 151-153,158, 159· 161-16J, 1]l, 173,178-181, 18J-185, 18!}, 205, 217, 218, 141, 260,263 Procliticization 146
See also the proclisis-enclisis alternatiorr
Procrastination Principle 75 Pro-DP 108 Projection principle 106 Prominence 142 Pronominality 76, 8o, 81 Pronoun serialization 26o Pronouns as intransitive oo heads 6o,81 Pronouns as transitive D 0 heads 63,76 Properties of the pro be 232 pro-PP 6 Prosodic Deficiency 34 Prosodic Hierarchy 34 Prosodic Phrase 35 Prosodic Restructuring 13 Prosodic Word 34, 134 Pu(n) demonstrative 7 pu ... pure subordinator 196, 197,205
Q Quantificational subjects 212
R Raising construction117 Range 74 Recoverability 60-62, 73-75, 106 Referential 20, 21, 166 Referential properties 13 Referentia\J-anaphoric agreement 204 Referentiality 67, 70, 73, 77, 93, 104,105 Reprojection ofv* in T 235 Restructuring 119, 127, 129, 181 Restructuring infinitives 172, 237 R-e:xpression 26, 66, 108,111 Right Node Raising (RNR) 121, 122 Ritter, E. 86 Rivero,M.-L.&A.Terzi 176 Rivero, M.-L. 168,169,173,176 Rizzi, L~ 1, 58, 146, 168, 179, 193 Roberts, I. 174 Romance languages 5, 10, u, 13, so, 59· 61-65, 86, 91, 95· 121, 138,141, 145· 146, 1]2-174, 177, 185, 187, 22]-2JO, 234. 254, 257, 259,262.,266,270 Romanian 61,j'2,115,150,169, 171, 173, 177, 254, 264
Index 293 Roman ian C( ontro Ued)-subjunctives 162 Romanian F(ree)-subjunctives 162 Root 201 Root modal adverbs 209
s 8- VDicing 35 Sandhi rules 13, 35, 37, 44 schedhon 125, 182, 183 Second position cUtic languages 154
see also Tobler-Mussafia languages see also Wackernagellanguages Secondary/rhythmic stress 36. 260 Semantic content 75 Semantic dependency of the clause 160, 161 Semantic finiteness 158, 159, 193 Semantic properties of heads 174 Semitic languages 5, 212 Sentential/higher (speaker oriented) adverbs 180, 182, 185,202 Serbo-Croatian 154.156 Siclrna 221 Sinithos 221 Sistirw 262 Slavic languages 5 Sloppy identity reading 21, 22, 27,28 Southern Italian dialects 173, 187 Spanish 16, 39,49, 72, 78, 87, 92, 95· 98.107, 115, 138, 149· 158, 170, 177, 189,229, 237· 254,262 Spanish subjunctive with imperative syntax 162 Spec-head configuration 114.245 [Spec TP] 232,233,265 Special imperative affix on 170 Specificity effect 72 Specificity feature 61, 67, 73, 77, 92, 93· 96, 98, 234 Specificity head 95 Specified Subject Condition (SSC) 50,114, n8, 147,158 Speech event 16o Speech time 160,215
c
Spell Out 120, 127, 247-249
Spesso 181 Sportiche, D. 92, 102, 103 Standard Gennan 13, 17, 259, 260,266 Standard Modern Greek 5, 146 StowelL T. 172 Strict serialization of ditics 262 Strong features 59, 61 Strong pronoun 2, 5, 6, 7,12-15, 1]'-19,23, 26,32-35.37.39. 64. 67, 70, 71, ]6, 83, 86, 89. 91, 104, 107, 108, 129. 258, 26J, 266 Strong pronoun as a bound variable 107 Strong pronoun as transitive determiner 106 Strong subject pronoWJ 29-32 Strong/specific/generic/partitive reading 140 Structural deficiency 106, 233 Subject (nuU) pro 31--33, 64. 159, 161, 16J, 175· 179· 198, 204, 214, 217,220 Subject 29, 49, 201, 212 Subject agreement morphology 29, 95,100,146--153,157, 158, 16o, 161, 16J, 165-167,172, 191, 1% 195, 198, 203, 205, 206, 216, 217, 227, 228, 223, 234. 241--2.43 Subject Clitic Inversion constructions in North- Eastern Italian dialects 177 Subject Clitic Inversion constructions in root interrogatives in Northern Italian dialects (NIDs) 177.179 Subject Clitic Inversion in French 177, 228 Subject oriented adverb 120 Subject-aux.iliary inversion 179, 229 Subjunctive 1, 17, 147-152, 158, 159. 17fJ, 175.176,181,184.186. 189, 193. 195, 200' 203, 207, 216, 217, 223, 227--229, 238, 2f4, 251, 252, 254.269 Subset Principle 213, 254 Swier 92 Surely 209 Swahili 212 Swedish 70,260
Swiss German 259, 261, 262, 265,266 Syncretism 8, 9 Syntactic 124, 134. 161, 233 Syntactic finiteness 158,159, 172,193 Syntactically opaque domain 158 Syntax- morphology interface 132 Syntax 168, 172, 213, 241
T T 158, 16o, 166,167,169-17::1. 174. 178, 180, 183, 184, 187, 188, 191, 216, 217, 219, 224--232, 234--241, 243. 245. 246, 249. 259,266 Telios 125 Temporal centres of the clause 160 Temporal chain 136 Tense features 132, 146--153, 157-161, 16J, 16j,167, 172,1% 195· 198, 205, 206, 215-217, 224.241 Tense operator 158,159 Tensed Spanish infinitives 162 Tensed Subject condition 147, 158 Teni, A. 175 The prodisis-endisis alternation 3,145-147,153, 157,161, 165,167,168,172, 173· 176--1]8, 189, 190, 19J, 195· 196, 211, 216. 225, 239· 249· 266. 270 and finiteness 160, 167-169. 1]2, 173, 182, 189, 190, 191, 193,195.197,225, 226, 236--::139. 241, 249. 266 and the (non) finiteness pattern J, 145, 152, 153 and the fmiteness properties of the verbal host 17, 146,147 and the gradation offiniteness 191 and thelel•el ofjirliter1ess 3, 159.190,191 Theta role 116, 132, 136, 175 Timberlake, A. 158 Tirolese German 13 Tobler-Mussafia languages 49, 154-156,191,236 Topic (syntactic) 153,197,198 Topic chain 96
294 Cities in Greek: A Minimalist Account ofProdlsls and Endlsls
Topic marker 99, 269 Topicality feature 62, 77, 96, 1.42 Topicality head 95 TP + TP conjunction 123 Transitive 76 Transitive DOJ:rypothesis 76 Tme negated imperatives :u8, 251 Truncation (of a head) 173 Thcking in 138 Types of enclisis 146
u Unaccusative construction 117, 243 Uninterpretable features 6o Uriagereka 63-65, 69. 71-74
v v
2J2,237 v- flavor 228 v* 128, 130, 133, 136, 183, 184, 234--237· 239· 242, 243· 245·
246 EPP feat!lTe of 129 v* 236, 246--:149. 251, 269, 270 v*P phase 127, 247 Vl clauses in Rhaetoromance variaties 177 V2 structure 261 V-coordination 121 Verbal 157 V-feature 233 V-movement 168,172,173,177, 182, 184, 185, 187, 189-191, 205, 216, 224--226, 2J1, 235. 236, 238--l40o242o247o249 V-movement theory 3, 145.168, 177,179-181,183-188,190
rr
Voice 194, 195, 203, 207 VP + VP conjunction 123 vP phase 234, 237, 242 V-related 249 V-relatedheads 249,270 V-related movement 227, 230 V-to(-1-to)-Cmovement 169. 174> 177, 179· 180,181.184-186, 189,194,199,201,203,205, 227, 228, 2JO, 236, 238, 247, 267 V-to-CForce/Op movement 229 V-to-CM movement 227,228, 238,242 V+ V conjunction 121
w
Wh-questions in Greek 177 Wh-questions in Spanish 177 Wh-questions in Standard Italian 177 Words 131 as marpho-syntactic phases 131 as spell-out domains 131,132
X
XC constituent 39 Clitic pronmms as 39, 50 Clitic pronouns move as 118 XC vs. XP properties 13, 49 XP constituent 39 Clitics move as 118
Strong pronouns as 39
Wackernagellanguages 154,156, 191,236 Wackernagel/2nd position clitics 155 Weak Crossover Effects 50, 116,117 Weak 10 pronoun series in Greek 265 Weak version of the PersonCase constraint 262
Violatiotl of the constraim 263 Weak/atonic pronoun 2, 3, 5-7, 12.-14,17, 32,34> 70. 76, 78. 83, 89,104> 107, 108,134> 249. 255. 258-266 asat~XP 261 with X -like properties 261
See also ditic pronoun West Flemish 13, 259 Wh-feature 2oo,20J. 228 Wh-interrogative 201
y Yes/no interrogative 201
z Zero pronouns 22 Ziirich German 255
e ®-role 91 ®a-future particle 41,45, 195-197. 200
I :E,. a dummy support morpheme 105
Linguistik Aktuell/Linguistics Today A complete list of titles in this series can be found on the publishers' website. www.benjamins.wm 169 SANCHEZ, Liliana: The Morphology and Syntax of Topic and Focus. Minimalist inquiries in the Quechua
periphery. Expected November 2o1o 168 FELDHAUSEN, logo: Sentential FollD and Prosodic Structure of Catalan. xiii, 280 + index. Expected
October 2o1o 167 MERCADO, Raphael, Eric POTSDAM and Lisa deMena TRAVIS (eds.): Austronesian and Theoretical Linguistics. vii, 374 pp. + index. Expected October 201 o 166 BRANDT, Patrick and Marco GARdA GARdA (eds. ): Transitivity. Form, Meaning. Ac.quisition, and Proc.essing. vi, 300 pp. + index.ExpectedOctober2o1o 165 BREUL, Carsten and Edward GOBBEL (eds.): Comparative and Contrastive Studies of Information Structure. 2010. xii, 306 pp. 164 ZWART, Jan-Wonter and Mark de VRIES (eds.): Structure Preserved. Studies in syntax for Jan Koste& 201 o. xxiii, 395 pp. 16 3 KIZIAK, Tanja: Extraction Asymmetries. Experimental evidence from German. 2010. xvi, 273 pp. 162 BOTT, Olilel': The Processing of Events. xix. 379 pp. +index. Expected September2o1o 161 HAAN, Germen J. de: Studies in West Frisian Grammar. Edited by Jarich Hoekstra. Willem Visser and Golfe Jensma. 2010. x, 384 pp. 160 MAVROGIORGOS, Marios: Clitics in Greek A minimalist account of proclisis and enclisis. 2010.
x. 294 PP· 159 BREITBARTif, Anne, OJ.ristopber LUCAS, Sbella WATTS and David WILLIS (eds.): Continuity and Change in Grantmar. 2010. viii, 359 pp. 158 DUGUINE,Maia,SasanaHUIDOBRO and Nerea MADARIAGA (eds.): Argument Structure and Syntactic Relations. A cross-linguistic perspective. 2010. vi, 348 pp. 157 FISCHER, Sus ann: Word-Order Change as a Source of Grammaticalisation. 201 o. ix, 200 pp. 156 DI SCIULLO, Anna Maria and Virginia HILL (eds.): Edges, Heads, and Projections. Interface properties. 2010. vii, 265 pp. 155 SATO, Yosake: Minimalist Interfaces. Evidence from Indonesian and Javanese. 2010. xiii, 159 pp. 154 HORNSTEIN, No !:bert and Maria POLINSKY (eds.): Movement Theory of ControL 2010. vii, 330 pp. 153 CABREDO HOFHERR,Patricia and Ora MATUSHANSKY (eds.): Adjectives. Formal analyses in syntax and semantics. 2010. vii, 335 pp. 152 GALLEGO, Aogel J.: Phase Theory. 2010. xii, 365 pp. 151 SUDHOFF, Stefan: Focus Particles in German. Syntax, prosody, and information structure. 2010. xiii, 335 pp. 150 EVERAERT, Martia, Tom LENTZ, Hannah de MULDER, 0ystein NILSEN and Arjen ZONDERVAN (eds.): The Linguistics Enterprise. From knowledge of language to knowledge in linguistics. 2010. ix, 379 pp. 149 AELBRECHT,Lobke: The Syntactic Licensing of Ellipsis. 2010. xii, 230 pp. 148 HOGEWEG,Lotte,Hel.en de HOOP and Andrej MALCHUKOV (eds.): Cross-linguistic Semantics of Tense, Aspect. and Modality. 2009. vii, 406 pp. 147 GHOMESHI, Ji)a, Deana PAUL and Martina WILTSCHKO (eds.): Determiners. Universals and variation. 2009. vii, 247 pp. 146 GELDEREN,Flly van (ed.): Cyclical Change. 2009. viii, 329 pp. 145 WESTERGAARD, Marit: The Acquisition of Word Order. Micro-cues, information structure, and economy. 2009. xii, 245 pp. 144 PUTNAM, Michael. T. (eel.): Towards a Derivational Syntax. Survive-mininlalism. 2009. x, 269 pp. 143 ROTIIMAYR, Antonia: The Structure of Stative Verbs. 2009. xv, 216 pp. 142 NUNES, Jairo (eel.): Minimalist Essays on Brazilian Portuguese Syntax. 2009. vi, 243 pp. 141 ALEXIADOU, Artemis, Jorge HANKAMER, Thomas McFADDEN, Justin NUGER and Florian SCHAFER (eds.): Advances in Comparative GellDanic Syntax. 2009. xv, 395 pp. 140 ROEHRS, Dorian: Demonstratives and Definite Articles as Nominal Auxiliaries. 2009. xii, 196 pp. 139 HICKS, Glyn: The Derivation of Anaphoric Relations. 2009. xii, 309 pp. 138 SIDDIQI, Daniel: Syntax within the Word. Economy, allomorphy, and argument selection in Distributed Morphology. 2009. xii, 138 pp.
137 PFAU, Roland: Grammar as Processor. A Distributed Morphology account of spontaneous speech errors. 2009. xiii. 372 pp. 136 KANDYBOWICZ, Jason: The G1.'3111mar of Repetition. Nupe grammar at the syntax-phonology interface. 2oo8. xiii, 168 pp. 135 LEWIS, William D., Slmin KARIMI, Heidi HARLEY and Scott 0. FARRAR (eds.): Time and Again. Theoretical perspectives on formal linguistics. In honor of D. Terence Langendoen. 2009. xiv, 265 pp. 134 ARMON-LOTEM, Sharon, Gabi DANON and Suan D. ROTifSTEIN (eds. ): Current Issues in Generative Hebrew Linguistics. 2008. vii, 393 pp. 133 MACDONALD, Jonathan E.: The Syntactic Nature of Inner Aspect A minimalist perspective. 2008. XV, 241 pp. 132 WBERAUER, Theresa (ed.): The Limits of Syntactic Variation. 2008. vii, 521 pp. 131 DE CAT, Q!cile and Kalherine DEMUTH (eds.): The Bantu-Romance Connection. A comparative investigation of verbal agreement, DPs. and information structure. 2008. xix. 35 5 pp. 130 KALLUW, Dalina and Liliane TASMOWSKI (eds.): Clitic Doubling in the Balkan Languages. 2008. ix. 442 PP· 129 STURGEON, Anne: The Left Periphery. The interaction of syntax, pragmatics and prosody in Czech. 2008. xi. 143 PP· 128 TALEGHANI, AritalL: Modality, Aspect and Negation in Persian. 2008. ix, 183 pp. 127 DURRLEMAN -TAME, Stephanie: The Syntax of Jamaican Creole. A cartographic perspective. 2008. xii, 190 pp. 126 SCHAFER. Florian: The Syntax of (Anti-)Causatives. External m:guments in change-of-state contexts. 2008. xi, 324pp. 125 ROTifSTmN,Bjilm: The Perfect Time Span. On the present perfect in German. Swedish and English. 2008. xi, 171 pp. 124 ffiSANE, Tabea: The Layered DP. Form and meaning of French indefinites. 2008. ix, 260 pp. 123 STOYANOVA, Marina: Unique Focus. Languages without multiple wh-questions. 2008. xi, 184 pp. 122 OOSTERHOF, Albert M.: The Semantics of Generics in Dutch and Related Languages. 2008. xviii, 286 pp. 121 TUNGSETH, Mal Ellin: Verbal Prepositions and Argument Structure. Path. place and possession in Norwegian. 2008. ix, 187 pp. 120 ASBURY, Anna, Jakab DOTLACII.,Berit GEHRKE aad Rick NOUWEN (eds.): Syntax and Semantics of Spatial P. 2oo8. vi, 416 pp. 119 FORTUNY, Jordi: The Emergence of Order in Syntax. 2008. viii, 211 pp. 118 JAGER, Agnes: History of German Negation. 2008. ix, 350 pp. 117 HAUGEN, Juon D.: Morphology at the Interfaces. Reduplication and Noun Incorporation in Uto-Aztecan. 2008. XV, 257 PP· 116 ENDO, Yoshio: Locality and Information Structure. A cartographic approach to Japanese. 2007. x, 235 pp. 115 PUTNAM, Michael T.: Scrambling and the Survive Principle. 2007.x, 216 pp. 114 LEE-SCHOENFElD, Vera: Beyond Coherence. The syntax of opadty in German. 2007. viii, 206 pp. 113 EYTifORSSON, Thorhallar (ed.): Grammatical Change and Linguistic Theory. The Rosendal papers. 2008. vi, 441 PP· 112 AXEL, Katrin: Studies on Old High German Syntax. Left sentenc.e periphery, verb plac.ement and verbsecond. 2007. xii. 364 pp. 111 EGUREN, Luis and Olga FERNANDEZ-sORIANO (eds.): Coreference, Modality, and Focus. Studies on the syntax-semantics interface. 2007. xii, 239 pp. 110 ROTHSTEIN, Suan D. (ed.): Theoretical and Crosslinguistic Approaches to the Semantics of Aspect 2008. viii, 453 pp. 109 CHOCANO, Gema: Narrow Syntax and Phonological Form. Scrambling in the Germanic languages. 2007. x. 333 PP· 108 REULAND,Eric,lllnmoy BHATTACHARYA and Giorgos SPATHAS (eds.): Argument Structure. 2007. xviii, 243 pp. 107 CORVER, Norbert and Jaim NUNES (eds.): The Copy Theory of Movement 2007. vi, 388 pp. 106 DEllE, Nicole and Yordanb KAVALOVA (eds.): Parentheticals. 2007. xii. 314 pp. 105 HAUMANN,Dagmar: Adverb Licensing and Clause Structure in English. 2007. ix, 438 pp. 104 JEONG, Yonngmi: Applicatives. Structure and interpretation from a minimalist perspective. 2007. vii, 144PP·
103 WURFF, Wim nn der (ed.): Imperative Clauses in Generative Grammar. Studies in honour of Frits Beukema 2007. vii~ 352 pp. 102 BAYER, Josef, Tanmoy BHATTACHARYA and M. T. Hany BABU (eds.): Linguistic Theory and South Asian Languages. Essays in honour of I<. A. Jayaseelan. 2007. x. 282 pp. 101 KARIMI, Simin, Vida SAMHAN and WendyK. WILKINS (eds.): Phrasal and Clausal Architecture. Syntactic derivation and interpretation. In honor of Joseph E. Emonds. 2007. vi, 424 pp. 100 SCHWABE, Kentin and Susanne WINKLER (eds. ): On Information Structure, Meaning and Form. Generalizations across languages. 2007. vii, 570 pp. 99 MARTINEZ-GIL,Fernaado and Sonia COLINA (eds.): Optimality-Theoretic Studies in Spanish Phonology. 2007. viii, 564 pp. 98 PIRES, Acrisio: The Mininlalist Syntax of Defective Domains. Gerunds and infinitives. 2006. xiv, 188 pp. 97 HARTMANN, Jutta M. and Uszl6 MOLN ARFI (eds.): Comparative Studies in Germanic Syntax. From Afrikaans to Zurich German. 2006. vi, 332 pp. 96 LYNGFELT,Benjamin and Torgrim SOLSTAD (eds.): Demoting the Agent Passive, middle and other voice phenomena. 2006. x. 333 pp. 95 VOGELEER, Svedana and Llliane TASMOWSKI (eds.): Non-definiteness and Plurality. 2006. vi, 35 8 pp. 94 ARCHE. Marla J.: Individuals in Time. Tense, aspect and the individuaVstage distinction. 2006. xiv, 281 pp. 93 PROGOVAC, Ljiljana, Kate PAESANI, Eugenia CASIELLBS and Fllen BARTON (eds.): The Syntax of Nonsententials. Multidisciplinary perspectives. 2006 . .1, 372 pp. 92 BOECKX, Cedric (ed.): Agreement Systems. 2006. ix, 346 pp. 91 BOECKX, Cedric (ed.): Minimalist Essays. 2006. xvi, 399 pp. 90 DALMI, Grete: The Role of Agreement in Non-Finite Predication. 2005. xvi, 222 pp. 89 VEIDE, John R. te: Deriving Coordinate Symmetries. A phase-based approach integrating Select. Merge, Copy and Match. 2oo6. x. 385 pp. 88 MOHR, Sabine: Clausal Architecture and Subject Positions. Impersonal constructions in the Germanic languages. 200 5· viii, 207 pp. 87 JULIEN, Marit: Nominal Phrases from a Scandinavian Perspective. 200 5· xvi. 348 pp. 86 COSTA, Joio and Maria Cristina FIGUEIREDO SILVA (eds.): Studies on Agreement 2006. vi, 285 pp. 85 MIKKELSEN, Line: Copulae Clauses. Specification, predication and equation. 2005. viii. 210 pp. 84 PAFEL, Jllrgen: Quantifier Scope in German. 2006. xvi, 312 pp. 83 SCHWEIKERT, Walter: The Order of Prepositional Phrases in the Structure of the Clause. 2005. xii, 338 pp. 82 QUINN,Heidi: The Distribution of Pronoun Case Forms in English. 2005.xi~ 409 pp. 81 FUSS, Eric: The Rise of Agreement. A formal approach to the syntax and grammaticalization of verbal inflection. 2005. xii. 336 pp. 8o BURKHARDT SCHUMACHER, Petra: The Syntax-Disrourse Interface. Representing and interpreting dependency. 2005. xii, 259 pp. 79 SCHMID, Tanja: Infinitival Syntax. Infinitivus Pro Participio as a repair strategy. 2005. xiv, 251 pp. 78 DIKKEN, Marcel den and Christina TORTORA (eds.): The Function of Function Words and Functional Categories. 2005. vii, 292 pp. 77 OZTtl"RK, BalkJz: Case, Referentiality and Phrase Structure. 2005. x. 268 pp. 76 STAVROU, Melita and Amon to TERZI (eds.): Advances in Greek Generative Syntax. In honor of Dimitra Theophanopoulou-Kontou. 2005. viii, 366 pp. 75 DI SCIULLO, Anna Maria (ed.): UG and External Systems. Language, brain and computation. 2005. xviii, 398 pp. 74 HEGGIE,Lorie and Francisco ORD6:Rllz (eds.): Clitic and Affix Combinations. Theoretical perspectives. 2005. viii, 390 pp. 73 CARNIE,Andrew,HeidiHARLEY and Sheila Ann DOOLEY(eds.):Verb First On the syntax of verbinitial languages. 2005. xiv, 434 pp. 72 FUSS, Eric and Carola TRIPS (eds.): Diachronic Clues to Synchronic Gramm a~: 2004. viii, 228 pp. 71 GEIDEREN,lllly van: Grammaticalization as Economy. 2004. xvi, 320 pp. 70 AUSTIN, Jennifer R., Stefan ENGELBERG and Gisa RAUH (eds.): Adverbials. The interplay between meaning, context, and syntactic structure. 2004- x. 346 pp. 69 KISS,Katalin ~and Henkvan RIEMSDIJK (eds.): Verb Clusters. A study of Hungarian, German and Dutch. 2004. vi, 514 pp.
68 BREUL, Carsten: Focus Structure in Generative Grammar. An integrated syntactic, semantic and 67
66
65 64 63 62 61 6o 59 58 57 56 55 54 53
52 51 50 49 48 47
46 45 44 43 42 41 40 39
38 37
36 35 34
intonational approach. 2004. x. 432 pp. MI~SKA TOMIC, Olga (eel.): Balkan Syntax and Semantics. 2004 xvi, 499 pp. GROHMANN, Kleanthes K.: Prolific Domains. On the Anti-Locality of movement dependencies. 2003. xvi. 372 pp. MANNINEN, Satu Helena: Small Phrase Layers. A study of Finnish Manner Adverbials. 2003. .xii, 275 pp. BOECKX, Cedric and Kleanthes K. GROHMANN (eels.): Multiple Wh-Fronting. 2003.x, 292 pp. BOECKX, Cedric: Islands and Chains. Resumption as stranding. 2003. :xii, 224 pp. CARNIE, Andrew, Heidi HARLEY and MaryAnn WILLIE (eels.): Formal Approaches to Function in Grammm: In honor of Eloise Jelinek. 2003. xii, 378 pp. SCHWABE, Kentin and Susanne WINKLER (eds. ): The Interfaces. Deriving and interpreting omitted structures. 2003. vi, 40 3 pp. TRIPS, Carola: From OV to VO in Early Middle English. 2002. :xiv, 359 pp. DEiffi, Nicole: Particle Verbs in English. Syntax, information structure and intonation. 2002. xii, 305 pp. DI SCIULLO, Anna Maria (eel.): Asymmetry in Grammm: Volume 2: Morphology, phonology, acquisition. 2003. vi, 309 pp. DI SCIULLO, Anna Maria (eel.): Asymmetry in Grammm: Volume 1: Syntax and semantics. 2003. vi, 405 pp. COENE, Martine and Yves D11ULST (eds.): From NP to DP. Volume 2: The expression of possession in noun phrases. 2003.x. 295 pp. COENE, Martioe a ad Yves D11ULST (eds.): From NP to DP. Volume 1: The syntax and semantics of noun phrases. 2003. vi, 362 pp. BAPTISTA, Marlyse: The Syntax of Cape Verdean Creole. The Sotavento varieties. 2003. xxii, 294 pp. (incl CD-rom). ZWART, Jan-Woater and Werner ABRAHAM (eels.): Studies in Comparative Germanic Syntax. Proceedings from the 15th Workshop on Comparative Germanic Syntax ( Groningen, May 26-27, 2ooo). 2002. xiv, 407 pp. SIMON, Horst J. and HeJke WIESE (eels.): Pronouns - Grammar and Representation. 2002. xii, 294 pp. GERLACH, Birgit: Clitics between Syntax and Le:xicon 2002. :xii, 282 pp. STEINBACH, Malkas: Middle Voice. A comparative study in the syntax-semantics interface of German. 2002. .xii, 340 pp. ALEXIADOU,Artemis (eel.): Theoretical Approaches to Universals. 2002. vii~ 319 pp. ALEXIADOU, Artemis, Flena ANAGNOSTOPOULOU, Sjef BARBIERS and Hans-Martin GARTNER (eds. ): Dinlensions of Movement From features to remnants. 2002. vi, 345 pp. BARBIERS, Sjef, Frits BEUKEMA and Wim van der WURFF (eels.): Modality and its Interaction with the Verbal System. 2002. :x:. 290 pp. PANAGIOTIDIS, E. Phoevos: Pronouns, Clitics and Empty Nouns. 'Pronominality' and licensing in synt.ax. 2002. X. 214 pp. ABRAHAM, Werner and Jao-Wouter ZWART (eds.): Issues in Formal German(ic) Typology. 2002. xviii, 336 pp. TAYLAN, Eser Ergu-nnll (eeL): The Verb in Turkish. 2002. xviii. 267 pp. FEATIIERSTON, Sam: Empty Categories in Sentence Processing. 2001. xvi, 279 pp. ALEXIADOU, Artemis: Functional Structure in Nominals. Nominalization and ergativity. 2001. x, 233 pp. ZELLER, Jocheo: Particle Verbs and Local Domains. 2001. :xii, 325 pp. HOEKSEMA, Jack, HolzeRULLMANN, Victor SANCHEZ-VALENCIA aod Ton van derWOUDEN (eels.): Perspectives on Negation and Polarity Items. 2001. :xii, 368 pp. GELDEREN,Flly van: A History of English Reflexive Pronouns. Person. Self, and Interpretability. 2000. xiv, 279 pp. MEINUNGER, Andre: Syntactic Aspects of Topic and Comment 2ooo. .xii, 247 pp. LUTZ, Uli, Gereon MOLLER and Amim von STECHOW (eels.): Wh-Scope Marking. 2000. vi, 483 pp. GERLACH, Bbgit a ad Janet GRIJZENHOUT (eels.): Clitics in Phonology. Morphology and Syntax. 2001. xii, 441 pp. HR6ARSD6TTIR, lhorbjllrg: Word Order Change in Icelandic. From OV to VO. 2001. xiv, 385 pp. REULAND,Eric (eeL): Arguments and Case. E:xplaining Burzio's Generalization. 2ooo. xii, 25 5 pp.
33 PUSKAS, Genoveva: Word Order in Hungarian. The syntax of A-positions. 2000. xvi, 398 pp. 32 ALEXIADOU, Artemis, Paul LAW, Aodre MEINUNGER aod Chris WILDER (eds.): The Syntax of Relative Clauses. 2000. vi, 397 pp. 31 SVENONIUS,Peter (ed.): The Derivation ofVO and OV. 2000. vi, 372 pp. 30 BEUKEMA, Frits aod Mala!! deo DIKKEN (eds.): Clitic Phenomena in European Languages . .x. 324 PP· 29 MIYAMOTO, Tadao: The Light Verb Construction in Japanese. The role of the verbal nouiL 2ooo. xiv, 232 pp. 28 HERMANS,Beo aod Marc vao OOSTENDORP (eds.): The Derivational Residue in Phonological Optimality Theory. 2ooo. viii, 322 pp. 27 RUZICKA,Rudolfi Control in Grammar and Pragmatics. A cross-linguistic study. 1999· .x, 206 pp. 26 ACKBMA,Pelier: Issues in Morphosyntax. 1999· viii, 310 pp. 25 FELSER, Oaudia: Verbal Complement Clauses. A mininlalist study of direct peoc.eption c.onstructions. 1999. xiv, 278 pp. 24 REBUSCHI, Georges aod Laurice TULLER (eds.): The Grammar of Focus. 1999. vi, 366 pp. 23 GIANNAKIDOU, Anastasia: Polarity Sensitivity as (Non)Veridical Dependency. 1998. xvi, 282 pp. 22 ALEXIADOU, Artemis aod Chris WILDER (eds.): Possessors, Predicates and Movement in the Determiner Phrase. 1998. vi, 388 pp. 21 KLEIN,Heooy: Adverbs of Degree in Dutch and Related Languages. 1998. x, 232 pp. 20 LAENZLINGER, Christopher: Comparative Studies in Word Order Variation. Adverbs, pronouns, and clause structure in Romance and Germanic. 1998 . .1, 371 pp. 19 JOSEFSSON, Guolog: Minimal Words in a Minimal Syntax. Word formation in Swedish. 1998. ix. 199 pp. 18 ALEXIADOU, Artemis: Adverb Placement A case study in antisymmetric syntax. 1997 . .x, 256 pp. 17 BEERMANN,Dorotbee,David LEBLANC aod Henkvao RIEMSDIJK (eds.): Rightward Movement 1997. vi, 410 pp. 16 UU,Feog-bsi: Scope and Specificily. 1997. viii, 187 pp. 15 ROHRBACHER,BemhardWolfgllog: Morphology-Driven Syntax. A theory ofV to I raising and prodrop. 1999· viii, 296 pp. 14 ANAGNOSTOPOULOU,Eieoa,HeokvaoRIEMSDIJK aodFraos ZWARTS (eds.): Materials on Left Dislocation. 1997· viii, 349 pp. 13 ALEXIADOU,Artemis aod T.AlaoHALL (eds.): Studies on Universal Gran1mar and Typological VariatioiL 1997. viii, 252 pp. ABRAHAM, Weroer, Samuel David EPSTEIN,Hilskuldur 'fHR.AINSSON aod Jao-Wouter ZWART (eds.): Minimal Ideas. Syntactic studies in the mi imalistframework.1996. xii, 364 pp. u LUTZ, Uli aod Jtlrgeo PAFEL (eds.): On Extraction and I!xtraposition in German.1996. xii, 315 pp. 10 CINQUE, GusJ.ielmo aod Giuliaoa GIUSTI (eds.): Advances in Roumanian Linguistics. 1995. xi. 172 pp. 9 GELDEREN,EIIy vao: The Rise of Functional Categories. 1993. .x, 224 pp. 8 FANSELOW, Gisbert (ed.): The Parametri2al:ion of Universal GrammaL 1993. xvii, 232 pp. 7 AFARI1, Tor A.: The Syntax of Norwegian Passive Constructions. 1992. xii, 177 pp. 6 BHATT, Christa, Elisabeth LOBEL aod Oaudia Maria SCHMIDT (eds.): Syntactic Phrase Structure Phenomena in Noun Phrases and Sentences. 1989. ix. 187 pp. 5 GREWENDORF, Gilother aod Wolfgllog STERNEFELD (eds.): SCranlbling and Barriers. 1990. vi. 442 pp. 4 ABRAHAM, Weroer aod SjaakDe MEIJ (eds.): Topic, Focus and Configurationality. Papers from the 6th Groningen Graromar Talks, Groningen, 1984 1986. v, 349 pp. 3 ABRAHAM, Weroer (ed.): On the Formal Syntax of the Westgermania. Papers from the 3rd Groningen Gran1mar Talks (3e Groninger Gran1matikgesprlkhe), Groningen, January 1981. 1983. vi, 242 pp. EHLICH,Koorad aod Jtlrgeo REHBEIN: AugenkommunikatioiL Methodenreflexion und Beispielanalyse. 1982. viii.150 pp. With many photographic ills. KLAPPBNBACH,Ruth (1911-1977): Studien zur Modernen Deutschen Lexikographie. Auswahl aus den Lexikographischen Arbeiten von Ruth Klappenbach, erweitert urn drei Beitrage von Helene MaligeKiappenbach. (Written in German). 1980. xxiii. 313 pp.