Clitics between Syntax and Lexicon
Linguistik Aktuell/Linguistics Today Linguistik Aktuell/Linguistics Today (LA) p...
64 downloads
1350 Views
4MB Size
Report
This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form
AUTHOR ""
TITLE "Clitics between Syntax and Lexicon"
SUBJECT "Linguistik Aktuell/Linguistics Today, Volume 51"
KEYWORDS ""
SIZE HEIGHT "220"
WIDTH "150"
VOFFSET "4">
Clitics between Syntax and Lexicon
Linguistik Aktuell/Linguistics Today Linguistik Aktuell/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.
Series Editor Werner Abraham University of Vienna
Advisory Editorial Board Guglielmo Cinque (University of Venice) Günther Grewendorf (J.W. Goethe-University, Frankfurt) Liliane Haegeman (University of Lille, France) Hubert Haider (University of Salzburg) Christer Platzack (University of Lund) Ian Roberts (Cambridge University) Ken Safir (Rutgers University, New Brunswick NJ) Lisa deMena Travis (McGill University) Sten Vikner (University of Aarhus) C. Jan-Wouter Zwart (University of Groningen)
Volume 51 Clitics between Syntax and Lexicon by Birgit Gerlach
Clitics between Syntax and Lexicon
Birgit Gerlach Heinrich-Heine-University
John Benjamins Publishing Company Amsterdam/Philadelphia
8
TM
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 Gerlach, Birgit Clitics between syntax and lexicon / Birgit Gerlach. p. cm. (Linguistik Aktuell/Linguistics Today, issn 0166–0829 ; v. 51) Includes bibliographical references and index. 1. Grammar, Comparative and general--Clitics. 2. Romance languages--Clitics. I. Title. II. Linguistik aktuell ; Bd. 51. P288.G47 2002 415-dc21 isbn 9027227721 (Eur.) / 1588112454 (US) (Hb; alk. paper)
2002074771
© 2002 – 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
Contents
Contents Acknowledgements Abbreviations 1 Introduction 1.1 DeWnitions of clitics in linguistic research 2 1.2 Problematic clitic phenomena in Romance 5 1.2.1 Clitic paradigms 5 1.2.2 Clitic sequencing 7 1.2.3 Clitic placement and clitic doubling 9 1.3 Theoretical preliminaries 12 1.3.1 Minimalist Morphology 12 1.3.2 Lexical Decomposition Grammar 13 1.3.3 Correspondence Theory 15 2 The status of Romance clitics between words and aVixes 2.1 Clitic phonology 19 2.2 Clisis vs. inXectional aVixation 25 2.2.1 Pronominal clitics 30 2.2.2 Auxiliaries and other clitic-like material 53 2.3 Clisis vs. phrasal aVixation 61 2.4 Conclusions 65 3 Clitics in the lexicon 3.1 The place of morphology in grammar 74 3.2 Romance clitic inventories 83 3.2.1 Grimshaw’s Optimality-theoretical analysis of Italian clitics 86 3.2.2 European and Brazilian Portuguese clitics 89 3.2.3 European and Argentinian Spanish clitics 97 3.2.4 Standard Italian clitics 102 3.2.5 Piattino clitics 110 3.2.6 French clitics 112 3.2.7 Romanian clitics 117 3.3 Conclusions 121
ix xi 1
17
73
v
vi
Contents
4
Clitic sequences 127 4.1 Properties of clitic sequences in Romance 128 4.1.1 No separation 128 4.1.2 Strict order and combinatory restrictions 129 4.1.3 Opacity 133 4.2 Approaches to clitic sequences in recent research 136 4.3 Interaction of morphology and syntax in clitic sequences 140 4.3.1 The order of clitics 142 4.3.2 Clitic choice when combinations are prohibited 148 4.3.3 Deviations from the clitic order in Romanian 156 4.3.4 Markedness hierarchies and clitic order 159 4.4 Interaction of morphology and phonology in clitic sequences 163 4.4.1 Prosodic minimisation of clitic sequences 163 4.4.2 Dissimilation of clitics within sequences 170 4.5 Conclusions 183 5 Syntax at the interfaces: clitic placement and clitic doubling 189 5.1 Movement and base generation 190 5.2 Clitic placement 195 5.2.1 The domain of cliticisation 198 5.2.2 Clitic split and mesoclisis 207 5.2.3 Issues for further research 213 5.3 Clitic doubling 216 5.3.1 Simple cases: prohibited vs. obligatory clitic doubling 220 5.3.2 Person-dependent clitic doubling in Piattino and French 223 5.3.3 Semantic restrictions on clitic doubling 226 5.4 Conclusions 232 6 Summary 237 References 241 Appendix 255 I Paradigms 255 II Clitic sequence internal phenomena 265 III Clitic placement and clitic doubling 267 IV Clitic speciWc constraints 268
Contents vii
Language Index Name Index Subject Index
271 273 275
viii Contents
Contents
Acknowledgements Several people have assisted me in often surprising ways in the completion of this book and I would like to acknowledge their input here. First and foremost I would like to thank Dieter Wunderlich who was always willing to share his inexhaustible knowledge of the nature of language. His comments on earlier versions of the book have been extremely supportive. Barbara Stiebels read the pre-Wnal version and I thank her for her careful criticism and detailed comments. For other helpful comments on an earlier version of the book I thank Werner Abraham, James Kilbury and Ana Luis. I am grateful to Janet Grijzenhout, Sandra Joppen-Hellwig, Ingrid Kaufmann, Renate Lakämper, Albert Ortmann, Alexandra Popescu and Carsten Steins for many fruitful discussions of several topics of this book. Alexandra also provided most of the Romanian data. For assisting me with the French data I thank Anja and Pascal Latrouite and Tahar Guellil. I owe many Spanish data to Teresa Parodi’s previous work. For help with the Piattino and Italian data I thank Emanuela Canclini and her family and friends in Piatta. Claudia Kursawe and Stefano Bergonzoni in Bologna also helped me with the Italian data. The enterprise of writing this book would hardly have been possible without the help of my partner in life Manfred Lebbing and my parents. I dedicate the book to my son Julius, who has been very understanding and diligent during this last year.
ix
x
Contents
Contents
Abbreviations prn pss p c d acc dat
pronoun possessive pronoun preposition complementiser determiner accusative marker dative marker
m f sg pl 1 2 3 df idf
masculine feminine singular plural 1st person 2nd person 3rd person deWnite indeWnite
[-f] [+f] [-pl] [+pl] [+1] [+2] [-1,-2] [+df] [-df]
s o do io rfl ip ptv loc neg aux
subject clitic [ ] ambiguous object clitic [+hr] direct object clitic [+hr] indirect object clitic [+hr,+lr] reXexive clitic [+rX] impersonal subject clitic partitive clitic [+ptv] locative clitic [+loc] negation clitic [+neg] auxiliary clitic [ ]
fin inf fut pfv sbj cnd imp ger ptc
Wnite inWnite future tense perfective aspect subjunctive mood conditional mood imperative mood gerund participle
MM LDG OT CT HPSG
Minimalist Morphology Lexical Decomposition Grammar Optimality Theory Correspondence Theory Head Driven Phrase Structure Grammar
AS ES BP EP
Argentinian Spanish European Spanish Brazilian Portuguese European Portuguese
P F I R
Piattino French Italian Romanian
[+Wn] [-Wn] [+fut] [+pfv] [+sbj] [+cnd] [+imp]
xi
xii Abbreviations
Definitions of clitics in linguistic research
Chapter 1
Introduction
This book presents a lexically-based analysis of verbal pronominal clitics in Romance. It is argued that Romance clitics constitute a distinct morphological unit with special clitic-type properties. Therefore, the clitic phenomena are examined from a morphological point of view. Clitics play an increasingly prominent role in other Welds of linguistic research, too, and the status of clitics is a topic of much debate in phonology, morphology and syntax, as well as in the respective interfaces. The discussion on clitics has given rise to diverse and controversial analyses. I will look at the phonological, syntactic and prosodic aspects of clitic phenomena insofar as they are relevant to the suggested morphological analysis. From a morphological point of view, it is questionable whether a distinct morphological category of clitics is linguistically desirable beyond a purely descriptive means. In recent analyses, it has been proposed to accommodate clitics in one of the categories “word” or “aYx” (see, e.g., Barbu 1998; Crysmann 2000; Miller 1992; Monachesi 1995; Miller & Sag 1997). Other researchers propose creating a category “phrasal aYx” to refer to clitics (see, e.g., Anderson 1992; Halpern 1998; Harrison 1997; Legendre 1996, 2000a,b). These issues will be discussed in chapter 2, taking into account the criteria of Zwicky (1985) and Zwicky & Pullum (1983) as well as Nübling’s (1992) “bundles of scales” to distinguish between aYxes, clitics and words. In chapter 3, the relevant clitic paradigms are introduced and the clitics are supplied with lexical entries based on the early-insertion model of Minimalist Morphology (Wunderlich & Fabri 1996). In chapter 4, it is shown that the combinatory restrictions that underlie the occurrence of opaque clitics in clitic clusters pose a problem for phonology and syntax and seem to demand a morphological analysis. Chapter 4 provides such an analysis and also looks at the phonological principles that inXuence the combination of clitics. It will be shown that syntactically-orientated late-insertion models of morphology such as Distributed Morphology (Bonet 1995; Halle & Marantz 1994; Harris 1995) or Paradigm Function Morphology (Spencer 2000; Stump 1993) cannot satisfactorily
1
2
Introduction
explain the combinatory restrictions observed in connection with clitics. I will argue that the morphological component as proposed in Minimalist Morphology (Wunderlich & Fabri 1996) is able to capture the problems speciWc to clitics, especially since this framework allows for the interaction of phonological, morphological and syntactic constraints in an Optimality-theoretical way (see also e.g., Grimshaw 1997; Legendre 1996). Chapter 5 deals with clitic placement and clitic doubling in Romance languages. It is shown that in recent generative literature, the issue remains unsettled whether these phenomena are best accounted for in an entirely syntactic (Kayne 1975; Sportiche 1996; Uriagereka 1995), a prosodic (Halpern 1992), a prosodic-syntactic (CZ avar & Wilder 1994), or an Optimality-theoretical approach (Anderson 1996; Legendre 1996). The morphosyntactic status of clitics is still under debate. One of the main problems is whether clitics are arguments as proposed by Kayne (1975) and many others, or whether they are functional heads as proposed by, e.g., Sportiche (1996), Uriagereka (1995, 2000). The arguments for both positions are considered in chapter 5. The shortcomings of these syntactic analyses, especially when dealing with clitic doubling, will be demonstrated. I will show that Optimality-theoretical analyses are most successful with respect to clitic placement and, in my own analysis, I will adopt the linking model of Lexical Decomposition Grammar (Joppen & Wunderlich, 1995). Before going into the analysis, a deWnition of clitic-hood is given in section 1.1, and in section 1.2, a short overview of the particular clitic phenomena and clitic-speciWc problems in the examined languages, i.e. clitic sequencing, clitic placement and clitic doubling. Section 1.3 provides the reader with the basic assumptions of the theoretical framework chosen for the analysis.
1.1 DeWnitions of clitics in linguistic research A clitic is generally understood to be a word that cannot stand on its own and “leans” on a host word. The term “clitic” originates from the Ancient Greek verb klineîn ‘to lean on’ (see Kühner 1834/1966). The main characteristic of clitics is captured by this word: clitics are not able to bear stress or — as in the Ancient Greek case — accent by themselves and therefore have to “lean on” another word — called the “host” — in order to become integrated into a prosodic constituent. In Ancient Greek, as in many other languages, most
Definitions of clitics in linguistic research
clitics are enclitics, i.e. they have to lean on a host to their left. In contrast, proclitics lean on a host to their right. (1a) and (1b) show the Ancient Greek indeWnite enclitic pronoun tis as a single clitic, while (1c) shows the proclitic group ouχ hoos ho en, consisting of the negation ouk, the complementiser hoos, the relative pronoun ho and the preposition en. (1) Ancient Greek clitics a. phílos tis friend prn.idf
‘a friend’ or ‘any friend’
b. pólemós tis enemy prn.idf
‘an enemy’ or ‘any enemy’
c. ou χ hoos ho en téèi géèi neg c d.m in d.dat.f earth.dat.f ‘not like the one in the earth’ Kühner (1834/1966: §§ 87f) As can be seen in (1b), Ancient Greek enclitics may aVect the metric structure of their host, since pólemos usually bears the main stress on the antepenultimate. The enclitic tis causes additional secondary stress on the ultimate. On the basis of the Ancient Greek data, clitics were originally deWned by their phonological defectiveness. Therefore, clitics are a main topic of phonological research and their prosodic structure is under discussion. In addition to their typical lack of stress, clitics are prosodically deWcient in that they fail to meet prosodic minimality conditions. Unlike prosodic words, clitics need not consist of a full vowel. Moreover, clitics often exhibit diVerent phonological behaviour from other categories. A central issue in recent phonological theory is how clitics should be prosodiWed and whether we should distinguish a separate “clitic group” in the prosodic hierarchy, as originally proposed by Nespor & Vogel (1986) and Hayes (1989), or whether this category is superXuous. Under the latter view, which is supported in section 2.1, clitics are attached to — or integrated into — categories such as the prosodic word and the phonological phrase (see e.g., Booij 1996; Selkirk 1995; Zec & Inkelas 1991). With respect to the category, clitics are usually weak forms of functional elements such as pronouns, determiners, auxiliaries, negation particles and question particles. Zwicky (1977) has introduced the distinction between simple and special clitics. Special clitics diVer from simple clitics in that they are not necessarily derived from corresponding full forms in a transparent way and may be placed diVerently from the corresponding full form if there is one. Nübling (1992) argues that special clitics are at a transitional stage
3
4
Introduction
between simple clitics and aYxes. It will be shown that the Romance clitics examined in this study are special clitics. Considering the morphological and syntactic properties of clitics, one can observe that their behaviour in this respect shows some kind of defectiveness as well. A catalogue of criteria to distinguish clitics from aYxes on the one hand and to distinguish clitics from words on the other has been drawn up by Zwicky (1985) and Zwicky & Pullum (1983). Nübling (1992) has re-organised the so called Zwicky-criteria in a bundle of scales which also contain some additional criteria concerning the relation of clitics and full forms. This issue will be dealt with in chapter 2. A comparison of clitics and corresponding full forms is also made by Kayne (1975). He aims at a mainly syntactic description of French, which leads him to a deWnition of clitics based on their syntactic behaviour compared with full pronouns or NPs. These properties are shared by nearly all Romance pronominal clitics. For Italian and French, consider the sentences in (2a–d) and (2e) from Cardinaletti (1999). The Wrst sentence always contains the full pronoun, i.e. lui or voi respectively, and the second sentence contains the clitic, i.e. lo or vi respectively. (2)
Properties of clitics and full pronouns a.
i. ii.
Maria conosce solo lui. *Maria lo conosce solo.
‘Maria knows only him.’
b. i. ii.
Maria conosce lui e voi. *Maria lo e vi conosce.
‘Maria knows him and you.’
c.
Maria conosce LUI, non voi. *Maria LO conosce, non voi.
‘Maria knows HIM, not you.’
i. ii.
d. i. ii. e. i. ii.
Chi conosce, Maria? Lui. ‘Whom does Maria know? *Chi conosce, Maria? Lo. Him.’ Je ne connais/*achète que lui. ‘I don’t know/buy (other) than him.’ Je l’ai connu / acheté. ‘I have bought/known him/it.’
Contrary to full pronouns, clitics are syntactically deWcient in that they cannot be modiWed (2a), co-ordinated (2b) or contrastively stressed (2c). Furthermore, they can occur neither in isolation (2d) nor in the same positions as full pronouns or NPs. With respect to their referential properties, they diVer from
Problematic clitic phenomena in Romance
full pronouns insofar as they can refer to non-human entities, while full pronouns cannot (2e).
1.2 Problematic clitic phenomena in Romance Verbal pronominal clitics in the Romance languages are at the centre of this study. A morphological analysis of clitics has to face at least the following three problematic issues. First, it has to be explained that the clitic inventories of the various languages — though phonologically similar — diVer a lot with respect to their paradigm structure, i.e. which paradigm cells are Wlled by distinct elements and which are Wlled by syncretism. Second, it has to be decided how the distinctive ordering and combinatory restrictions that combinations of clitics with other clitics obey are accommodated in the grammar. Third, the unique conditions on the placement as well as on the occurrence of clitics need a careful re-examination. Analyses will be given for verbal clitics and verbal clitic phenomena in Standard Italian, Iberian and Rio de la Plata Spanish, Standard French, Iberian and Brazilian Portuguese and Standard Romanian. Furthermore, the northern Italian dialect Piattino spoken in Piatta in Lombardy is analysed. For Piattino, a completely new sample of data is introduced. Other dialects or strata of these languages will be taken into account where necessary or helpful.
1.2.1 Clitic paradigms The Wrst part of the book aims at a description of the inventories of the verbal clitics in the respective languages. As will be seen, all of the languages have pronominal object and reXexive clitics. In addition, we Wnd subject clitics in French and Piattino. Moreover, Italian and French have locative and partitive clitics, and Romanian has clitic auxiliaries. Though subject and object clitics are at the centre of my research, I will also provide lexical entries for the other verbal clitics in the respective languages. Two leading questions concerning the clitic paradigms are examined. – –
What are the lexical entries for the respective clitics? What kind of syncretism can be observed within the clitic paradigms?
The answers to both questions need a careful exploration of the occurrences of
5
6
Introduction
each clitic in the respective languages. Consider for example the Italian sentences in (3). The Italian clitic ci can be used as a 1st person direct object (3a), a 1st person indirect object (3b), a 1st person reXexive pronoun (3c), a prepositional object (3d), a locative pronoun (3e), or an expletive pronoun (3f). (3) Italian ci and its diVerent uses a.
direct object Arrivederci. see.again.us ‘Good bye’
b. indirect object Ci da questo libro. us give.3sg this book ‘He gives us this book.’
c.
reXexive Ci divertiamo. ourselves enjoy.1pl ‘We are happy.’
d. prepositional object È stata proprio una brutta storia, ma non ci penso più. is been really a ugly story but not about.it think yet ‘It has really been an ugly story, but I don’t think about it anymore.’ e.
locative Ci vado domani. Claudia non c’ é. there go.1sg tomorrow Claudia not here is ‘I go there tomorrow.’ ‘Claudia is not here.’
f.
expletive In Italia ci sono molte grandi città. In Italy there are many big cities ‘There are many big cities in Italy.’ Ce l’ha il bilieto? there it have.3sg the ticket ‘Do you (polite) have the ticket?’
Other clitics such as gli or lo are also found as determiners. This huge number of diVerent uses can be observed quite often in connection with clitics. With regard to underspeciWcation and syncretism, one needs to decide in each case which diVerent functions should be described by the same entries, i.e. which syncretism should be seen as systematic and which one coincidental. This will be provided by lexical entries in the framework of Minimalist Morphology in chapter 3.
Problematic clitic phenomena in Romance
1.2.2 Clitic sequencing In Romance languages, clitic combinations maintain a strict internal order, they resist separation, they are often conWned to at most two elements, and they exhibit unique morphophonological behaviour. The principles that underlie the building of clitic sequences are the topic of the fourth chapter of this thesis. In the Italian sentence in (4), the direct object must precede the indirect object if both objects consist of full NPs. (4) Order of object DPs in Italian Devo dare il libro a Stefano. must.1sg give the book to Stefano. ‘I must give the book to Stefano.’
If both objects are replaced by clitics, however, the indirect object clitic gli (or its allomorph glie, which is restricted to clitic sequences) always must precede the direct object lo, regardless of the proclitic (5a) or enclitic (5b) status of the clitic sequence. One of the most noticeable properties of clitic combinations is that the strict order of their elements is maintained with respect to each other and not — as is the case for aYxes — with respect to the host. Another property of clitic combinations is that they usually resist separation, even if there are diVerent positions in the sentence which may potentially accommodate clitics (5c,d). In Italian modal verb constructions, for instance, a clitic cluster can attach either to the embedded inWnitive as in (5a), or to the Wnite modal verb as in (5b). Adjacency seems essential and separation of two clitics is ungrammatical (5c,d). (5) Order of clitics in Italian a. Devo dar-glie-lo. must.1sg give-him-it ‘I must give it to him.’
b.
Glie-lo devo dare. him-it must.1sg give ‘I must give it to him.’
c. *Gli devo dar-lo.
d.
*Lo devo dar-gli.
Moreover, clitics in clitic sequences cannot combine freely. Clitic sequences in Romance languages, for example, are often restricted to two clitics, even if the clitic inventory consists of more than just a direct and an indirect object clitic. Standard Italian, for example, has locative and partitive clitics, and in Standard French we Wnd subject, locative and partitive clitics. Furthermore, re-
7
8
Introduction
strictions on possible combinations of clitics are based on case and person speciWcations. As mentioned above, an indirect object clitic always precedes a direct object clitic in Italian. In addition, 1st and 2nd person clitics can combine neither with each other nor with a 3rd person indirect object clitic. In these cases, one of the clitics is omitted and the respective argument is expressed by a full DP. Moreover, 3rd person clitics and reXexive clitics exhibit particular morphological behaviour in clitic sequences (see Bonet 1995:607, who refers to these cases as “opaque clitics”). These clitics are either deleted, or replaced by other clitics, or their surface form in combination with other clitics diVers from their form in isolation. The above-mentioned Italian clitic ci, for instance, is found as a substitute for the Wrst si in combinations of impersonal and reXexive si (see 6). (6) Restriction on the combination of impersonal and reXexive si in Italian a. I ragazzi si rallegrano delle vacanze estive. the teens oneself look.forward.3pl to.the holidays summer(adj) ‘The teens look forward to the summer holidays.’ b. Quando si é ragazzi ci si / *si si when one is teen ci oneself / one oneself rallegra delle vacanze estive. look.forward.3pl to.the holidays summer(adj) ‘When one is a teen one looks forward to the summer holidays.’
The following questions shall be examined in the analysis of clitic sequences. – – –
– –
Which clitic combinations are attested and which are not? Are there similar paradigms of clitic combinations in the Romance languages? Which are the underlying principles that determine the order of clitics within clitic combinations and the restrictions on possible clitic combinations? How do these principles interact? When do we Wnd deletion and when do we Wnd substitution within clitic sequences?
We will see that the answers to these questions can be found in the interaction of phonological and morphological constraints.
Problematic clitic phenomena in Romance
1.2.3 Clitic placement and clitic doubling The placement of clitics in the sentence is a purely syntactic issue. The following questions need to be answered. – – –
Why are the positions in which clitics surface diVerent from the positions in which full DPs surface? Why are these positions exclusive to clitics? How can the co-occurrence of both, clitic and DP, be explained?
To solve these questions, we not only need to consider to which syntactic position clitics are assigned on the surface, i.e. “where do they go”, we also need to ask in which syntactic position clitics are generated, i.e. “where do they come from”. Let us consider the Italian examples in (4) and (5) above again. As illustrated in (5), clitics are adjacent to a verb. Direct object DPs follow the Wnite verb (see 4), while direct object clitics precede it (see 5b). With regard to morphosyntax, the alternation between pre- and postverbal clitics with, e.g., Wnite verbs vs. inWnitives (see 5a vs. 5b) needs to be explained. Does the clitic-verb complex coalesce into a single morphosyntactic unit or do they both retain a certain independence? For both possibilities, it needs to be explained how the morphosyntactic features of the verb can inXuence the direction of clisis. A huge number of diVerent syntactic analyses is concerned with these questions for Romance as well as for Slavic languages. Clitic placement in Portuguese, where the morphology/syntax as well as the syntax/ semantics interface play a role, is particularly complex. The unmarked position for clitics in European Portuguese is behind the verb including inXectional morphology (see 7a), but in future and conditional tense, clitics occur directly behind the verb stem and in front of the inXectional endings (see 7b). Certain syntactic and semantic triggers (e.g. negation, question, emphasis), however, cause the clitic to occur in front of the verb in all tenses (see 7c). (7) Clitic Placement in European Portuguese a.
Eu vi-te. prn1sg see.pst-you ‘I saw you’
b.
c.
Não me lembrarei. not me remember.fut.1sg ‘I will not remember.’
Comprá-lo-ás buy-it-fut.2sg ‘You will buy it’
9
10
Introduction
In chapter 5, an overview will be given of current syntactic approaches to clitic placement including the two prevailing approaches, namely the “movement approach” (Kayne 1975, among others) and the “base-generation” approach (Jaeggli 1982, among others), and the most recent approaches, which treat clitics as functional D° heads (Sportiche 1996; Uriagereka 1995, 2000). These approaches are compared to recent Optimality-theoretical accounts of clitic placement in Balkan languages (Anderson 1996; Legendre 1996). It is argued that Optimality-theoretical approaches are capable of analysing clitic placement in a particularly elegant way since they allow constraints from diVerent modules to interact. For the Romance languages, an analysis is provided which allows for the interaction of syntactic, morphological and prosodic constraints. While in Standard French and Standard Italian a clitic and a DP cannot co-occur in the same sentence (see 8a,b), Spanish, Romanian and colloquial varieties and dialects of French and Italian allow clitic doubling, i.e. the cooccurrence of co-referential clitic and DP in the same sentence. Indirect object DPs in, e.g., Spanish are obligatorily doubled by clitics, and direct object DPs may also be doubled by clitics, for instance in Rio de la Plata Spanish and in Romanian (see 8c,d from Jaeggli 1986 and Sportiche 1996 respectively). (8) The co-occurrence of clitics and DPs in Romance a.
French Je la vois. Je vois Claudia. *Je la vois Claudia. I her see.1sg Claudia ‘I see Claudia’
c.
Rio de la Plata Spanish Lo veo a Juan. him see.1sg to Juan ‘I have seen Juan.’
b. Italian La vedo. Vedo Claudia. *La vedo Claudia. her see.1sg Claudia ‘I see Claudia.’
d. Romanian L-am va[ zut pe Popescu. him-aux.1sg see.ptc acc Popescu ‘I have seen Popescu.’
At Wrst glance, the co-occurrence of clitics and DPs seems to be licensed by the
Problematic clitic phenomena in Romance
additional occurrence of a case-marking element (a in Spanish and pe in Romanian) that precedes the DP. This observation is captured by “Kayne’s Generalisation”. This generalisation, however, does not withstand a closer examination of the data. In Spanish, for instance, the case marker is obligatory for human entities (9a), whereas the clitic is obligatory for speciWc entities (9b, where al stands for the case marker a plus the deWnite article el). (9) The case marker a and the clitic in Spanish
a. *La
veo a Veo a *La veo her see.1sg to ‘I see a woman.’
una una una d.idf
mujer. mujer. mujer. woman
b. *Lo veo al libro. * Veo al libro. Lo veo el libro. him see.1sg to.d.df d.df book ‘I see the book.’ This implies that the presence of the clitic and the presence of the case marker are required (or licensed) by diVerent semantic features of the object. Their co-occurrence in (8c,d) above is coincidental (i.e. dependent on the object having the semantic features [+human] and [+speciWc]) and not — as predicted by Kayne’s Generalization — necessary. Recent syntactic approaches are discussed in chapter 5. It is shown in detail that they all fail to explain certain data such as the relevance of person and animacy as a necessary condition for doubling structures. As an alternative, I will suggest that clitic doubling is a kind of agreement relation (in accordance with, e.g., Auger 1993; Suñer 1988; Torrego 1996), restricted by the lexical entries of the clitics on the one hand and (in accordance with Ortmann 2002) grammatical economy principles on the other hand. The requirement of agreement by clitics and the requirement of economy are thus in conXict, a problem which is in turn resolved by an Optimality-theoretical hierarchical ordering of these requirements. Moreover, it is shown that the morphological prohibition of the combination of, for instance, 1st and 2nd person object clitics (see section 1.2.2 above) may also conXict with the agreement requirement. In such cases, the clitics cannot be combined even if both
11
12
Introduction
clitics are required due to their speciWcity or animacy. Here, it is the requirement for morphological explicitness which governs the choice as to which of the arguments is realised as a clitic and which is realised as a pronoun.
1.3. Theoretical preliminaries In the following, I will brieXy outline the main assumptions I will be using in my study, taken from the integrated framework of Minimalist Morphology, Lexical Decomposition Grammar and Correspondence Theory (see also Gerlach 1998a,b; Stiebels 2000a,b; Wunderlich 2001 a,b).
1.3.1 Minimalist Morphology Minimalist Morphology (= MM, Wunderlich & Fabri 1996; Wunderlich 1996a,b,c) as an early-insertion model was originally developed for inXectional morphology. Morphemes are considered to be functors, represented by a lexical entry that consists of a phonological form, several morphosyntactic features, which are understood as output information, and a subcategorisation frame, which is understood as input information. The lexical entries for the single morphemes in the Italian verb form vedo ‘I see’, for instance, are represented as follows (see Canclini 1999 for a detailed study of Italian verbal inXection): (10) Lexical entries a. /-o/ : [+min], [+1]/[+V] λV ∃ x<+1> V(x) b. /ved/ : [ ], [+V], λy λx see(x,y) c. /ved-o/ : [+max], [+1, +V], λy ∃ x<+1> see(x,y)
The inXectional suYx with the phonological form /-o/ has the feature [+1] in the output and the subcategorisation restriction [+V] in the input (10a). Other output features such as [-pl], encoding singular, or [–hr] and [–lr], encoding the grammatical function “subject” (see section 1.3.2 below), are not part of the lexical entry itself but may be added by default. The output of the verb stem /ved/ (10b) consists of its categorical features, i.e. [+V], its semantic form, i.e. see(x,y), and its argument structure, i.e. λy λx. InXectional aYxes are characterised by the feature [+min], indicating that they are morphologically minimal. Words, on the other hand, are morphologically maximal, which is represented by the feature [+max]. Only elements with the latter
Theoretical preliminaries
feature may be projected into the syntax (see Stiebels & Wunderlich 1994). By combination of the stem and the aYx the inXected form vedo emerges, which can be projected into the syntax and thus bears the feature [+max]. In the verbal paradigm, this form occupies the cell characterised by the features [+1,+V]. The paradigm is considered to be the interface between syntax and morphology. The subject agreement information is assigned to the respective λ-abstractor, which is changed to the existential quantiWer due to the prodrop properties of Italian, similarly to co-indexing in HPSG (Pollard & Sag 1994). (Note that agreement information in non pro-drop languages such as German, for instance, does not aVect the abstractor in such a way.) The combination of morphemes, i.e. stems and aYxes, takes place in the lexicon and follows general principles that may be ordered. MM makes extensive use of underspeciWcation. Only those features necessary for building up the paradigm are speciWed in the lexical entry. This means that [+]-valued features are part of the lexical entry, while [-] values are added by default. In MM, the following principles are responsible for paradigm building: each cell of a paradigm is occupied (Completeness), more than one element per cell is prohibited (Uniqueness). A form with a more speciWc output is preferred over a more complex form (Output SpeciWcity), as well as to a more complex form with the same speciWcation (Simplicity). The structural features for argument roles as well as the linking mechanism are taken from Lexical Decomposition Grammar.
1.3.2 Lexical Decomposition Grammar The linking mechanism in Lexical Decomposition Grammar (= LDG, Joppen & Wunderlich 1995; Wunderlich 1997) essentially depends on relational features which encode the argument roles of the verb and, at the same time, the linkers, namely morphological case, structural position and agreement morphemes. These features go back to Kiparsky (1989, 1992). They are modiWed in LDG as follows: (11) +(-)hr: +(-)lr:
there is a (no) higher role there is a (no) lower role
In LDG, the Semantic Form (SF, see 12) determines the argument structure of the verb (TS = theta structure). Each position in the TS is characterised by features depending on whether a lower or higher position exists.
13
14
Introduction
(12) give SF: cause(x, become(poss(y, z)))(s) TS: λz λy λx λs
The TS of a (canonical) intransitive verb only contains one argument (in addition to the situation argument, which is ignored here). This means that there is neither a higher nor a lower role. Thus, this position is assigned the features [–hr,-lr] (13a). With (canonical) transitive verbs, the lowest role (the “direct object”) is assigned [–lr], since there is no lower role, and [+hr], because of the presence of a higher role (the “subject”). The highest role (the “subject”) is assigned the reverse feature values (13b). For the medial role of (canonical) ditransitive verbs (the “indirect object”), a higher role exists (the “subject”), hence the feature value [+hr], as well as a lower role (the “direct object”), hence the feature value [+lr] (13c). (13) Features of the TS positions of canonical verbs a. intransitive b. transitive c. ditransitive λx V(x) λy λx V(x,y) λz λy λx V(x,y,z) –hr +hr –hr +hr +hr –hr –lr –lr +lr –lr +lr +lr
The same features deWne the structural cases, with the diVerence that they are not fully speciWed, but only characterised by [+]-valued features. It is assumed that argument clitics are assigned the same features as structural cases. These case features are part of the clitics’ lexical entries. In nominative-accusative languages, dative case (dat) and indirect object clitics (io) are fully speciWed by [+hr,+lr], accusative case and direct object clitics (o) only bear the feature [+hr], and nominative case and subject clitics (s) remain unmarked [ ] and are thus least restricted: (14) case features: dat/io: [+hr,+lr]
acc/o: [+hr]
nom/s:[ ]
Structural linking consists in the choice of the optimal case pattern for a given TS. Ideally, this includes the choice of the most speciWc compatible linker for each argument role (SpeciWcity Principle) and the use of each linker only once (Uniqueness Constraint). As a result of these principles, the arguments of canonical verbs are linked to structural cases as shown in (15).
Theoretical preliminaries
(15) Linking for canonical (nominative-accusative) verbs a. λ x V(x) –hr –lr ↓ nom s
b. λ y +hr –lr ↓ acc o
λ x V(x,z)
–hr +lr ↓ nom s
c. λ z +hr –lr ↓ acc o
λy
λ x V(x,y,z)
+hr +lr ↓ dat io
–hr +lr ↓ nom s
The above-mentioned linking principles, however, are not always observed. Restrictions on clitic sequences (see section 1.2.2), for instance, may sometimes force substitution of clitics, so that it is not the most speciWc clitic that is linked to an argument role. This means that other principles or restrictions may compete with the linking principles and sometimes outrank them.
1.3.3 Correspondence Theory In clitic combinations, it is not only suppletion of clitics which is observed, but also deletion of clitics. Combinations of 1st and 2nd person object clitics, for instance, do not occur, even if both objects should cliticise due to clitic doubling requirements. Moreover, clitics often do not surface in their underlying phonological form. These cases, as will be argued in this study, are the result of the competition of constraints, which may belong to various grammatical modules. It depends on the relative ranking of the constraints whether they must be observed or not. Optimality Theory (= OT, McCarthy & Prince 1993; Prince & Smolensky 1993) and its more recent version Correspondence Theory (= CT, McCarthy & Prince 1995) provide a suitable mechanism for implementing this idea. In CT, grammar consists of a set of violable ranked constraints. While the constraint ranking is language-speciWc, the constraints themselves are universal. It is thus possible to account for the diVerences between languages with the same set of constraints but diVerent rankings. The set of output candidates to a given input is evaluated in view of the constraint ranking. The best (hence optimal) output candidate is the one with the fewest violations of highly ranked constraints. The correspondence relation of input and output candidates is checked by faithfulness constraints (in 16 quoted from McCarthy & Prince 1995).
15
16
Introduction
(16) Relation of S1 (input) and S2 (output) (Faithfulness): The max Constraint Family Every segment of S1 has a correspondent in S2. The dep Constraint Family Every segment of S2 has a correspondent in S1. The ident(F) Constraint Family Let α be a segment in S1, and β be any correspondent of α in S2. If α is [γ F], then β is [γ F].
max constraints restrict deletion, while dep constraints restrict epenthesis. If we generalise the notion of segment, the two constraint families together state that there is a one-to-one correspondence of elements regardless of which elements correspond to each other. ident makes sure that corresponding elements are identical with respect to their features. In chapter 3, the input and output and the model of grammar that is suggested here is introduced. For agreement relations, the concept of transitive correspondence is developed. For clitic sequences, a more elaborate formulation of the constraints is needed and, in addition, some morphological and phonological constraints. These constraints will be introduced step by step throughout the analysis in chapter 4. Additionally, alignment constraints are necessary in order to explain the placement of clitics. These constraints are developed in chapter 5. Finally, chapter 5 deals with the constraints restricting clitic doubling.
The status of romance clitics between words and aYxes
Chapter 2
The status of Romance clitics between words and aYxes
The morphological status of clitics has been under much discussion in recent research. The debate focuses on whether clitics constitute an autonomous morphological category, or whether they can be described as one of the independently motivated categories “aYx” or “word”. A threefold system that distinguishes words, aYxes and clitics is argued to be undesirable for reasons of economy. Therefore, many researchers would like to dispense with the term “clitic”, even if useful from a traditional descriptive point of view. The so-called Zwicky-criteria (Zwicky 1985; Zwicky & Pullum 1983, see below section 2.2) are often used to determine whether a linguistic element is a clitic, an aYx, or a word. Nevertheless, it is often diYcult to determine the status of these elements, since these criteria are gradual rather than absolute. Therefore, the Zwicky criteria have been supplemented by various other syntactic and morphological criteria (see for instance Kayne 1975; Cardinaletti & Starke 1999; Kaiser 1992; Nübling 1992). Sometimes, an accurate reexamination of the elements that are traditionally called clitics may lead to a re-categorisation of these elements but diVerent researchers may also come to diVerent results. The extraordinary properties of Romance clitics often lead to contradicting analyses. Let us consider in this respect the European Portuguese examples in (1). The pronominal clitics in European Portuguese occur after the verb including its inXectional endings in unmarked sentences (see 1a from Hundertmark-Santos Martins 1998). In future tense and conditional mood, however, the clitics occur before the inXectional endings (see 1b from Hundertmark-Santos Martins 1998). This seems to be clear evidence in favour of the aYxal status of these elements. There is, however, also evidence which strongly contradicts the categorisation as aYxes. In the presence of certain syntactic and semantic triggers (e.g. negation, question markers, certain adverbs, etc.) the clitics occur preverbally, even in future tense and conditional mood (see 1c).
17
18
The status of Romance clitics between words and aYxes
(1) European Portuguese clitic placement a. Ela tem uma casa linda e ontem prn have.3sg d.idf house nice and yesterday mostrou-no-la. show.pst.3sg-1o-3f.do ‘She has a nice house and she showed it to us yesterday.’ b. Ela tem uma casa muito linda. Mostrar-no-la-á? prn have.3sg d.idf house very nice. show-1o-3f.do-fut.3sg ‘She has a very nice house. Will she show it to us?’ c.
Não no-lo mostrar-á. not 1o-3f.do show-fut.3sg ‘She will not show it to us.’
European Portuguese clitics thus display very particular behaviour in that they — like aYxes — occur within inXected words but — like words — are aVected by their syntactic and semantic environment. I will re-examine the elements called clitics in the Romance languages below with respect to the Zwicky-criteria and show that at least pronominal clitics and auxiliary clitics behave diVerently from inXectional aYxes as well as from words. In section 2.2, I will then argue that the independent category “clitic” is reasonable for more than traditional descriptive grounds. Another approach which dispenses with the term clitic is Anderson’s theory of “phrasal aYxation”. Several researchers have adopted Anderson’s (1992) suggestion to analyse special clitics as a particular kind of aYx. Based on Klavans’ (1980, 1985) work on the positional properties of clitics, Anderson argues that clitics behave like aYxes and that the only diVerence is that clitics are adjoined to syntactic phrases, while aYxes are adjoined to words. For this reason, Anderson refers to clitics as “phrasal aYxes”. I will discuss this proposal in section 2.3. Whether clitics should be considered an independent category is not only an issue in morphological research but also in phonological research. As already mentioned in the introduction, the original deWnition of clitics is a phonological one, relying on their prosodic deWciency. Mostly, clitics are prosodically deWcient in that they fail to meet prosodic minimality conditions. Unlike prosodic words, clitics need not contain a full vowel. Moreover, clitics often exhibit diVerent phonological behaviour from other categories. In phonology, the prosodic structure of clitics is disputed. A central issue in recent phonological theory is how clitics should be prosodiWed and whether we
Clitic phonology
should distinguish a separate “clitic group” in the prosodic hierarchy, as originally proposed by Nespor & Vogel (1986) and Hayes (1989), or whether this category is superXuous. Under the latter view, clitics are attached to — or integrated into — categories such as the prosodic word and the phonological phrase (Booij 1996; Hall 1999; Kleinhenz 1998; Selkirk 1995; Zec & Inkelas 1991). Which view one chooses to adopt might be crucial also for the morphological status of clitics. If it can be shown that the phonology of clitics is entirely distinct from the phonology of aYxes and words, it is more likely that the morphology of clitics is also distinct from the morphology of aYxes and words. In section 2.1, I will brieXy discuss the question whether the distribution of clitics justiWes a category “clitic group” in the prosodic hierarchy or not and whether there are convincing arguments to assume that they are either part of prosodic words, or linked to categories higher than the prosodic word.
2.1. Clitic phonology Prosodic structure theory holds that syntax does not provide domains for phonological rules in a direct fashion. Instead, it is assumed that phonological rules can be explained in purely phonological terms, if their domains of application are expressed in relation to prosodic constituents such as the prosodic word or the phonological phrase. Prosodic constituents are deWned on the basis of the morphosyntactic structure of sentences, but they need not be isomorphic to morphosyntactic constituents. Kleinhenz (1998) illustrates this with the following example where the German article einen and the noun Fehler form one syntactic constituent (see 2a), whereas they are part of diVerent prosodic constituents (see 2b): (2) a.
ich muß [[ einen Fehler]NP gemacht haben]VP I must a mistake made have ‘I must have made a mistake’
b. ich [mu.z66 n]PW [fe:.l!! ]PW gemacht haben
In example (2b), the weak form of the indeWnite article einen, /6n/, needs a host word and forms one prosodic constituent (here: PW, i.e. prosodic word) with muß. For pronominal clitics in Romance, however, the phonological host mostly belongs to the same syntactic phrase as the clitic itself as it constitutes the head of the respective phrase.1 As originally proposed by Selkirk (1978),
19
20
The status of Romance clitics between words and aYxes
the prosodic structure of an utterance is composed of a speciWc hierarchy of prosodic units. The relevant part of this hierarchy is given in (3) with the extension of a so-called “clitic group” (see Nespor & Vogel 1982, 1986 and Hayes 1989): (3) utterance intonational phrase phonological phrase clitic group prosodic word
(U) (IPh) (PPh) (CG) (PW)
The phonological utterance is the largest constituent in the prosodic hierarchy. It consists of one or more intonational phrases, which are usually dominated by the same Xn (where Xn is the highest node of a syntactic tree). The intonational phrase groups together one or more phonological phrases and the phonological phrase is formed from one or more clitic groups. For evidence to assume the categories “utterance”, “intonational phrase”, “phonological phrase” and “prosodic word”, I refer to Nespor & Vogel (1982, 1986) and Selkirk (1978). The domain of the clitic group is deWned by Nespor & Vogel (1986:154) as follows: (4) The domain of the Clitic Group consists of a Prosodic Word (PW) containing an independent (i.e. nonclitic) word plus any adjacent PWs containing: a. a directional clitic2 b. a clitic such that there is no possible host with which it shares more category memberships.
According to the Strict Layer Hypothesis (Selkirk 1981, 1984, see below), the implication of this proposal is that each clitic may form a prosodic word of its own. This prediction does not stand up to close scrutiny, as becomes clear from the following reXection (for a more detailed refutation of the clitic group as a prosodic constituent see Zec 1988; Inkelas 1989; Zec & Inkelas 1991; Inkelas & Zeg 1995; Booij 1996; Hall 1999). Since the prosodic word is agreed to be the domain of stress assignment this means that clitics bear stress, which clearly contradicts the deWning property of clitics, namely to be stressless. In most Romance languages, clitics are either proclitic or enclitic, depending on the morphosyntactic features of the verb (see section 1.2.4 of the introduction and chapter 5). We never Wnd stress on proclitics, which I take as primary
Clitic phonology
evidence against the clitic group. Some enclitics, though, are able to aVect stress assignment within the constituent they cliticise to and may even bear stress themselves. Therefore, I will examine next whether the respective enclitics provide evidence in support of the clitic group. While Italian, Romanian, French and Spanish clitics do not aVect stress assignment, Neapolitan and Portuguese clitics do. Standard Italian (see 5a from Monachesi 1995) and Neapolitan (see 5b from BaWle 1993) both have lexical stress and both are subject to the “3-Silben-Gesetz” (Wackernagel 1892) a wellformedness condition which requires that the main stress lies on one of the three last syllables. Moreover, both varieties do not show stress readjustment with only one enclitic. In this case, violations of the “3-Silben-Gesetz” occur, since verb+ clitic units consisting of more than three syllables consequently bear stress to the left of the antepenult syllable, as shown for teléfonami and p7Üttinal6. If there are more clitics, however, stress readjustment takes place in Neapolitan but not in Italian. The Neapolitan forms dámmíll6 and p7Üttinal6 have two stressed positions — the Wrst one being identical with the lexical stress on the verb and the second one lying on the Wrst of the clitics. (5) a.
Standard Italian: no stress readjustment after clisis dá + mi + lo : dámelo give + 1sg.o + 3sg.do teléfona + mi call + 1sg.o
‘Give me it’
: teléfonami ‘Call me.’
péttina + me + li : péttinameli ‘Comb them to me.’ comb + 1sg.o + 3pl.do b. Neapolitan: stress readjustement after clisis: dá + m6 give + 1sg.o
: dámm6
p7Üttina + l6 comb + 3pl.do
: p7Üttinal6 ‘Comb them’
dá + m6 + l6 : dámmíll6 give + 1sg.o + 3pl.do
‘Give me’
‘Give me it’
p7Üttina + t6 + l6 : p7Üttinatíl6 ‘Comb them to comb + 2sg.o + 3pl.do yourself’
21
22
The status of Romance clitics between words and aYxes
Nespor & Vogel (1986) argue that clitics which do not aVect stress assignment are not integrated into the prosodic word which is formed by their host. Therefore, they must form a diVerent, higher constituent together with their host, namely the clitic group. As pointed out above, this implies that single clitics form a prosodic word of their own and consequently receive stress. Single clitics in Italian, however, never bear stress, so we can conclude that they never form prosodic words of their own. With single clitics, the constituent clitic group is thus not evident. Selkirk (1995) claims that there is no justiWcation for the prosodic constituent clitic group. She points out that the clitic group can only be defended if one can show that relevant phenomena exist which cannot be accounted for by structures based on the prosodic hierarchy without the clitic group. She redeWnes English and Serbo-Croatian clitics with respect to the category prosodic word and proposes that a lexical word (lex) may form a prosodic word of its own. A function word (fnc) may also form a prosodic word of its own (6a), or it may be prosodiWed as one of three types of prosodic clitics, viz. free clitics (6b), internal clitics (6c) and aYxal clitics (6d). (6) Prosodic structure of function words (fnc): a. prosodic word ( (fnc)Pw (lex) Pw ) Pph b. free clitic (fnc (lex) Pw ) Pph c. internal clitic ( (fnc lex) Pw ) Pph d. aYxal clitic ( (fnc (lex) Pw ) Pw ) Pph
Selkirk (1995) shows that with the diVerent prosodiWcations of function words, the clitic phenomena which lead Nespor & Vogel to the assumption of the clitic group can be explained without reference to this constituent. The only extension to the model of prosodic phonology is a revision of the Strict Layer Hypothesis which has been reinterpreted as a number of constraints (7). While Layerdness and Headedness are not violable, Exhaustivity and Non Recursivity are. (7) The Strict Layer Hypothesis (Selkirk 1995:443) (i) Layerdness
No Ci dominates a Cj, j>i e.g. ‘No σ dominates a F.’
(ii) Headedness:
Any Ci dominates a Ci–1 e.g. ‘A PW must dominate a F.’
Clitic phonology
(iii) Exhaustivity:
No Ci immediately dominates a Ck, k
(iv) Non Recursivity: No Ci dominates another Ci e.g. ‘No F dominates a F.’
As to proclitics, the free clitic option (6b) is most reasonable (see also Peperkamp 1997). Taking into account that proclitics never bear stress, they are directly incorporated into the phonological phrase in Standard Italian as well as in Neapolitan (see 8) and thus avoid the stress assigning constituent, namely the prosodic word. This structure causes a violation of Exhaustivity as can also be observed with clitics in other languages such as Dutch (see Booij 1996) and German (see Hall 1999). (8) Prosodic structure of proclitics a. Standard Italian b. Neapolitan PPh PPh
PW
PW
F σ ti
σ le
σ σ σ pet ti na
F σ ti
σ le
σ σ σ pεt ti na
Standard Italian enclitics behave the same as proclitics and are therefore assumed to have the same structure except for the clisis site, i.e. they are incorporated into the right of the phonological phrase (see 9a). Enclitics in Neapolitan, on the other hand, may inXuence stress assignment. Therefore, I suggest that enclitics are adjoined to the prosodic word violating NonRecursivity. This leads to the structure of aYxal clitics (6d). A single enclitic does not aVect stress assignment. It is not integrated into the internal prosodic word and thus cannot form a foot together with other material of the internal prosodic word, as for instance the syllable -na. Therefore, it is dominated directly by PW and cannot receive stress or inXuence stress assignment. Pairs of enclitics, however, are adjoined to the prosodic word at the same level and thus form a foot together, so that stress assignment takes place (9b).
23
24
The status of Romance clitics between words and aYxes
(9) Prosodic structure of enclitics a. Standard Italian b. Neapolitan PPh PPh PW PW
PW
F σ σ pe ti
σ na
σ σ ti le
F
F
σ σ σ pε ti na
σ ti
σ le
For the same data, Peperkamp (1997) suggests a slightly diVerent analysis. She argues that pro- and enclitics should have similar structures for reasons of consistency. Therefore, she proposes that proclitics in Neapolitan are treated as aYxal clitics and that they are thus adjoined to the prosodic word. This proposal does not seem convincing, since there is no reason why pro- and enclitics should not have diVerent prosodiWcations, in particular if one considers that preWxes and suYxes also often diVer in this respect. It is more reasonable to assume that proclitics resemble preWxes while enclitics resemble suYxes, rather than that pro- and enclitics resemble each other (see Booij 1996 for Dutch clitics and Rivera-Castillo 1992 for Spanish clitics). Moreover, Peperkamp’s otherwise careful analysis suVers from the parallel prosodic structure of pro- and enclitics, because it is not clear how pairs of proclitics are prevented from forming a foot and subsequently receiving stress. In addition to stress assignment, Nespor & Vogel (1986) and Nespor (1986) study internal and external sandhi processes (e.g. intervocalic s-voicing, truncation and radoppiamento sintattico). They argue that these processes can only be accounted for if one assumes the existence of a clitic group, since the phonological behaviour of clitics does not resemble the behaviour of other morphological categories such as words, derivational aYxes and inXectional aYxes with respect to these processes. Nespor & Vogel point out that clitics are thus neither directly dominated by prosodic words nor by phonological phrases so that an additional constituent in between both is needed. These arguments are thoroughly reconsidered by Monachesi (1995, 1996) and Peperkamp (1995, 1997). Monachesi (1995, 1996) shows that the behaviour of clitics is not as exceptional as Nespor & Vogel claim, since there are certain aYxes which trigger the same processes, although it is undesirable to assume
Clisis vs. inXectional aYxation
that these aYxes are also part of clitic groups instead of prosodic words. Peperkamp (1995, 1997) shows that with the theory of prosodic structure exempliWed above, the clitic-speciWc behaviour with regard to truncation and radoppiamento sintattico can be explained, too, so that the clitic group can be dispensed with. (For a more detailed discussion of these topics see Monachesi 1995 and Peperkamp 1997.) In conclusion, despite the distinct phonological behaviour of clitics, there is no need to assume the additional prosodic constituent clitic group. Clitic phenomena can be explained on the basis of the prosodic hierarchy and the diVerent processes of integration, adjunction and incorporation into prosodic constituents. It should be emphasised, however, that clitics still behave diVerently, so that the term “clitic” is nonetheless useful when referring to this particular class of linguistic elements. Let us now examine whether the term “clitic” is dispensable in morphology.
2.2 Clisis vs. inXectional aYxation In morphology, recent research has focussed on the question whether clitics constitute an autonomous morphological category, or whether they can be described as one of the independently motivated categories “aYx” or “word”. The so-called “Zwicky-criteria” (Zwicky 1985; Zwicky & Pullum 1983) are used to determine whether a linguistic element is a clitic, an aYx, or a word. Since these criteria are gradual rather than absolute, it is often diYcult to determine the status of these elements. According to the Zwicky-criteria, Romance pronominal clitics have been analysed as inXectional aYxes (e.g. Crysmann 2000; Barbu 1998; Kaiser 1992; Miller 1992; Miller & Sag 1997; Monachesi 1995, 2000) but also as clitics (e.g. Dobrovie-Sorin 1994; Gerlach 1998a,b, Popescu 2000, Vigário 1999). The status of other elements within the verbal complex is similarly controversial. Auxiliaries have been analysed as words (Monachesi 2000), as aYxes (Barbu 1998) and as clitics (Popescu 2000). In sections 2.2.1 and 2.2.2, I will examine these and other Romance elements which are traditionally called clitics. I will demonstrate that pronouns and auxiliaries are neither inXectional aYxes nor words but clitics. There are, however, elements which will be shown to be words (i.e. the negation element in Romanian) or aYxes (i.e. Romanian monosyllabic inten-
25
26
The status of Romance clitics between words and aYxes
siWers). Further cases will also be examined, for example pronominal clitics in European Portuguese (see Crysmann 2000), which exhibit aYx-like behaviour with respect to morphology but word-like behaviour with respect to their placement, i.e. syntax, and are therefore even more diYcult to categorise. Before examining the clitics in detail, the Zwicky criteria will be introduced together with Nübling’s bundle of scales to distinguish between aYxes, clitics and words. Zwicky’s and Zwicky & Pullum’s most important morphological and syntactic tests to distinguish clitics from aYxes and words are summed up in table (10).3 (10) The Zwicky Criteria (following Zwicky 1985; Zwicky & Pullum 1983) Criterion words I.
Clitics are word–like 1. arbitrary gaps in combination 2. morphophonological idiosyncrasies 3. semantic idiosyncrasies 4. combination with stems 5. combination with phrases
II. Clitics are aYx–like 1. bound element 2. strict ordering 3. restricted distribution 4. morphological complex 5. subject to syntactic rules 6. deletion under identity III. Clitics are neither aYxes nor words 1. high degree of selection (host/stem) 2. attach to material containing clitics
aYxes
clitics
– – – – +
+ + + + –
– – – – +
– – – + + +
+ + + – – –
+ + + – – –
+ –
– +
The criteria in table (10) are to be read as tendencies and not as absolute statements.4 Therefore, the Zwicky criteria leave some space for interpretation, so that even if one author Wnds that one of the criteria is met by the element under examination, another author might Wnd that the same criterion is not met by the same element. In addition to the Zwicky criteria, I use Nübling’s (1992) bundle of scales to determine the extent to which Romance clitics have developed to inXectional elements (see 11).
Clisis vs. inXectional aYxation
(11) Nübling’s (1992) bundle of scales determine the grade of clitic-hood simple clitics > special clitics > inXection > 2
1
>3
> 4
A: phonological fusion of host and clitic additive regular combination
contact phenomena
processes of fusion
irregular forms
B1: phonological relation of clitic and full form -
synchronic derivation
partly synchronic diachronic partial / total no full derivation derivation suppletive form
B2: interchangeability of clitic and full form everywhere partial impossible no change in no change in slight change heavy change — meaning meaning in meaning in meaning C: distribution and selectivity Syntax C=F
Syntax ∃ P=position C morphoC morphoC ⊂ F C in P &¬ F in P logically selected logically or idiosyncratic selected syntax of C
D1: clitic paradigm (forms no longer interchangeable with F) less than 1/4 1/4 to 1/2 1/2 to 3/4 complete paradigm D2: host paradigm combination with some hosts belonging to diVerent word classes
preference of one/few word classes as host; combination with less than the half of its members
combination with more than the half of the members belonging to one host word class
combination with only one host word class and with all of its members
27
28
The status of Romance clitics between words and aYxes
E1: scope of the clitic scope C ≈ scope C C and F may occur in the same minimal scope F becomes sentence -> C is no longer scope smaller complement but agreement marker E2: functionalisation of the clitic function C ≈ F function C ≈ C functionalised, C completely function function F, interchangeability functionalised F focus or with F restricted emphatic, C topic
The main diVerence between the Zwicky criteria and Nübling’s bundle of scales is that the scales take into account the relation between clitic (i.e. “C” in 11) and full form of the respective clitic (i.e. “F” in 11). Nübling argues that clitics start as independent (functional) words, which are reduced in fast speech. As they are reduced more regularly, they become simple clitics and subsequently special clitics. Finally, they end up as inXectional aYxes. Nübling contends that this development aVects the phonology, morphology and syntax as well as the semantics and pragmatics of clitics. The development is parallel in the diVerent modules. Scale A states that phonological processes between clitic+aYx and host+stem units are hardly attested with simple clitics. With special clitics, we Wnd processes of fusion, while irregular forms occur in inXection. The scales B1 and B2 are concerned with the phonological and semantic relation of clitic and full form. Phonological derivation from corresponding full forms is synchronic with simple clitics. As to special clitics, the phonological relation of clitic and full form is only partly synchronically derivable, or diachronically derivable or suppletive. InXectional aYxes lack corresponding full forms. This phonological digression of clitic and full form goes hand in hand with a change in meaning. While simple clitics have the same meaning as their corresponding full form and can always be replaced by them, special clitics diVer in meaning and can only partially be replaced by full forms, if at all. InXectional aYxes do not have corresponding full forms and can thus not be replaced. Interchangeability of clitics and full forms depends on distribution and selectivity as shown in scale C. While simple clitics have the same distribution as their full forms, special clitics develop a special syntax, the Wrst step being a reduction of the positions admissible for the full
Clisis vs. inXectional aYxation
form. The result then is that positions which are occupied by clitics are no longer admissible for full forms, so that clitic and full form underlie complementary distribution. Morphological selection is the next step on the way to inXection. Two further scales — D1 and D2 — compare the paradigms of clitics/aYxes and the paradigms of clitics/aYxes and hosts respectively. Nübling contends that both paradigms become more complete as clitics develop into inXectional aYxes. Finally, Nübling argues that clitics undergo a process of scope diminishing (scale E1) when they become aYxes, so that, Wnally, they may occur in the same sentence as corresponding full forms (i.e. as agreement markers) and the clitic tends to become more and more grammaticalised (scale E2). Nübling’s scales coincide partly with Zwicky’s criteria. The degree of phonological processes between host/stem and clitic/aYx (scale A), for instance, can easily be identiWed as Zwicky’s “morphophonological idiosyncrasies” (I.2). The clitic-host paradigm (scale D2) and the clitic paradigm (scale D1) resemble Zwicky’s “degree of selection” (III.1) and “arbitrary gaps” (I.1). Nübling’s scales B and E on the other hand, are concerned with the relation between full forms and clitics, which is not part of Zwicky’s criteria.5 One might recall from section 1.1 that the semantic-syntactic relation between full form and clitic in particular is also subject to Kayne’s (1975) deWnition of clitics. The special properties of clitics that Kayne observes (i.e. clitics cannot be modiWed, coordinated or contrastively stressed, they cannot occur in isolation or in the same positions as corresponding full forms) are reXected in Nübling’s scales B2, C, E1 and E2.6 Let us now look at the analyses of clitics in recent literature and examine whether they are clitics, aYxes or words regarding the above criteria. I will base the study on Monachesi’s careful examination of Italian pronominal clitics and compare them with pronominal clitics in other Romance languages. With respect to their aYxal status, I will also take into account further recent research (e.g. for Romanian: Barbu 1998; Legendre 2000b,c; Monachesi 1999, 2000; Popescu 2000; for French: Miller 1992; Miller & Sag 1997; for Spanish: Parodi 1998a; for Portuguese: de Carvalho 1989; Crysmann 2000; van der Leeuw 1995; Vigário 1999). The following section will show that Romance pronominal clitics are indeed not inXectional aYxes (contrary to Miller 1992; Miller & Sag 1997; Monachesi 1995, 2000 but in agreement with Legendre 2000b). The analysis will then be extended to other material whose status is under debate such as auxiliaries in Romanian, negation in French and
29
30
The status of Romance clitics between words and aYxes
Romanian, partitive and locative elements in Italian and French and intensiWers in Romanian.
2.2.1 Pronominal clitics Monachesi (1995) has carefully examined Italian weak pronouns. Assuming that clitics, like aYxes, are bound elements (criterion II.1) and morphologically simple (criterion II.4), Monachesi argues that for the following reasons the respective units actually are aYxes and not clitics. They only attach to verbs (criterion III.1.) and they have to be repeated in coordination structures (criterion II.6).7 As to combination with other clitics as well as with their hosts, they follow a rigid ordering (criterion II.2) and they show a lot of morphophonological idiosyncrasies (criterion I.2) as well as arbitrary gaps (criterion I.1). These arguments will be considered closely below and transferred to weak pronouns in other Romance languages, which show a quite similar behaviour. Let us Wrst consider Monachesi’s most convincing argument in favour of the aYxal status of clitics, i.e. the existence of arbitrary gaps (criterion I.1).
2.2.1.1 Arbitrary gaps Monachesi refers to two kinds of gaps. On the one hand, clitics do not combine freely with all verb forms and, on the other hand, pronominal clitics do not combine freely with each other. Monachesi shows that there are gaps in clitic-verb combinations, in particular if the verb is a present participle. While the present participle can be followed by indirect object clitics regardless of their person value (12a), only 1st and 2nd person direct object (12b) but not 3rd person direct object clitics (12c) are allowed to follow a present participle.8 (12) Arbitrary gaps in Italian clitic-host combinations a.
I compensi spettanti-ci /-mi /-ti /-vi d.df.pl compensation belong.ptc-1pl.o /-1o /-2o /-2pl.o /-gli /-le /-3io /-3f.io ‘The compensation belonging to us/ me/ you(pl)/ you(sg)/ him/ her’ b. Gli argomenti riguardanti-ci /-mi /-vi /-ti d.df.pl topics concern.ptc-1pl.o /-1o /-2pl.o /-2o ‘The topics concerning us/ me/ you(pl)/ you(sg)’
Clisis vs. inXectional aYxation
c. Gli argomenti riguardanti-*lo /*-la /*-le / ?-li d.df.pl topics concern.ptc-3do /-3f.do /-3f.pl.do /-3pl.do ‘The topics concerning him/ her/ them(f)/ them(m)’ (Monachesi 1995:42)
Similar restrictions are found in French. As Miller & Sag (1997) point out, the Wrst person subject clitic je cannot combine freely with verbs if the order of subject and verb is inverted in questions: (13) Arbitrary gaps in French clitic-host combinations a. Chantes-tu? b. *Chante-je?
‘Are you going to sing?’ ‘Am I going to sing?’
Gaps like these cannot be explained by any syntactic restrictions but demand a morphological analysis. However, this does not inevitably mean that the respective elements are inXectional aYxes. Combinatory restrictions in inXectional morphology do not allow much variation. With regard to the inversion of subject clitic and verb, however, Miller & Sag point out that it is speaker dependent which verb allows subject verb inversion and which verb does not. One can therefore argue that the restrictions observed with clitic-host combinations belong to a transitional state between syntax and inXectional morphology, namely clitic morphology. Regarding combinations of two clitics, combinatory restrictions are more frequent. In Italian, the relation of 3rd person indirect object and 1st person direct object, for instance, cannot be expressed if both objects are represented by clitics (see 14b). However, expressing the same relation by free pronouns is grammatical (see 14a). (14) Arbitrary gaps in Italian clitic-clitic combinations a.
Emanuela presenta me a lui. Emanuela introduce.3sg prn.1 p prn.3m.sg ‘Emanuela introduces me to him.’
b. *Emanuela gli mi / me gli presenta. Emanuela 3io 1o / 1o 3io introduce.3sg
Similarly, combinations of 2nd person and 3rd person indirect object and combinations of 1st person and 2nd person are ungrammatical. These kinds of gaps are quite unusual for words but often attested for object aYxes, particularly if 1st and 2nd person are involved.9 Clitic-clitic combinations in the
31
32
The status of Romance clitics between words and aYxes
other Romance languages are also restricted when 1st and 2nd person object are involved, though some variation can be observed. Romanian, for example, allows the 2nd person singular direct object clitic to combine with all 1st as well as 3rd person clitics (15a). 1st person clitics and 2nd person plural clitics, however, cannot represent the direct object if combined with other pronominal clitics (15b,c). (15) Arbitrary gaps in Romanian clitic-clitic combinations a.
Ni te recomanda[ . 1pl.o 2do recommend.3sg ‘He recommends you to us’
b. *Ni t6 i / it6 i ni recomanda[ . 1pl.o 2io/ 2io 1pl.o recommend.3sg intended: He recommends us to you. c.
*Ni va[ / va[ ni recomanda[ . 1pl.o 2pl.o / 2pl.o 1pl.o recommend.3sg intended: He recommends you(pl) to us.
Some dialects of Spanish allow combinations of 1st and 2nd person clitics and combinations of both with 3rd person indirect object clitics as well, in particular if one of both is reXexive and the other one represents an ethical dative (16a). Even here, the 1st person clitic is more restricted in that it cannot represent the (direct object) reXexive (16b).10 (16) Arbitrary gaps in Spanish clitic-clitic combinations a.
Te escapaste. ‘You escaped.’ Te le escapaste. ‘You escaped from him.’ Te me escapaste. ‘You escaped from me.’
b. Me escapé. Me le escapé. *Me te escapé.
‘I escaped.’ ‘I escaped from him.’ ‘I escaped from you.’
(Perlmutter 1970:194)
Restrictions for 1o are still more complex in Portuguese. As in the other Western Romance languages, 1st person clitics are unspeciWed with respect to case and reXexivity. That means that the 1o clitic me can be used as (reXexive) direct object (see 17a) and as (reXexive) indirect object (see 17b). In combinations with other clitics, however, the reXexive interpretation is blocked as Crysmann (2000) shows.
Clisis vs. inXectional aYxation
(17) Arbitrary gaps in Portuguese clitic-clitic combinations a. Eu ofereci-me um bolo. prn.1 oVer.1sg-1o d.idf biscuit ‘I oVered myself a biscuit.’ b. Ela ofereceu-mo. prn.3f oVer.3sg-1o.3do ‘She oVered it to me.’ c. *Eu ofereci-mo. prn.1 oVer.1sg-1o.3do d. Eu ofereci-o a mim. prn.1 oVer.1sg-3do p prn.1o ‘I oVered it to myself.’
(Crysmann 2000:123–124)
Gaps like these can neither be explained by syntactic nor by semantic means. This does not mean, however, that an analysis in terms of inXectional morphology is necessary. In chapter 4, I will show that there are morphological constraints at work which restrict the combination of clitics. The diVerence between inXectional and clitic morphology results from the domain of the constraints (see also Gerlach, 1998a,b; Legendre 1999, 2000b). It will be demonstrated below that the domain of the constraints restricting clitic combinations is the clitic sequence, that is, the clitics without their host. This domain diVers from that usually assumed for aYxal morphology, which includes the stem and all other aYxes. In the following sections, I will prove that there are more reasons to assume that clitic morphology diVers from aYxal morphology.
2.2.1.2 Morphophonological idiosyncrasies Monachesi lists a number of morphophonological processes in Italian clitic-host and clitic-clitic combinations, which she considers to be morphophonological idiosyncrasies that prove the aYxal status of clitics (criterion I.2). In the case of (pro)clitic-host combinations, vowel deletion takes place with 3do lo and 3f.do la, if the host begins in a vowel (see 18). In spoken language, vowel deletion also takes place with 3rfl, 1o and 2o clitics.
33
34
The status of Romance clitics between words and aYxes
(18) Vowel deletion in Italian clitic-host combinations Martina l’ accetta. Martina 3do accept.3sg ‘Martina accepts it.’
(Monachesi, 1995:43)
Before we can decide about the status of vowel deletion in Italian, we have to examine whether its domain of application is indeed (inXectional) morphology. That means we have to explore whether it is an internal or rather an external sandhi process. With derivational preWxes and with compounds vowel deletion is not attested: (19) No vowel deletion in Italian derivation and composition fotoamatore reattivo preavvisare
‘amateur photographer’ ‘reactive’ ‘to announce in advance’
Postverbal clitics are not subject to vowel deletion, since all Italian clitics have an onset and thus the phonological conditions are not met. InXectional morphology is suYxal and vowel deletion processes are diYcult to prove. One might argue that 1st person singular canto ‘I sing’ (20a), for instance, results from the stem canta- and the inXectional ending -o involving the deletion of the theme vowel. There are, however, also examples where the theme vowel and the onset vowel of the inXectional ending form a diphthong, as for instance in the perfective forms cantai (20b). (20) No vowel deletion in Italian inXectional morphology a. cant-o b. canta-i
‘I sing’ ‘I sang’
Deletion of the theme vowel and diphthongisation are both obligatory in the respective morphological contexts. With clitics, however, vowel deletion is optional, though strongly preferred, at least for la and lo. As I will also show with respect to a number of other processes below, optionality of morphophonological processes often distinguishes clitics from inXectional aYxes and from words. Further evidence against Monachesi’s claim that vowel deletion indicates aYxal morphology is that there are other contexts for vowel deletion in order to avoid hiatus, most of which are clearly not aYxal. As the Dizionari Garzanti (1994) lists, syllable-Wnal vowels are deleted in combinations of (singular) deWnite article+noun and preposition+deWnite article (21a). Italian
Clisis vs. inXectional aYxation
deWnite articles are clitics.11 One could thus argue that deWnite articles are aYxal as well and that the deletion of the vowels is a consequence of this status. Vowel deletion also occurs in quite diVerent cases, though. The indeWnite articles uno / una are not clitics, even though vowel deletion takes place (21b). If the conjunction come ‘how’ is combined with the non-clitic auxiliary essere ‘be’, the syllable-Wnal vowel is deleted (21c). Even between adjective and noun we Wnd vowel deletion, in particular with the adjectives quello ‘this’, bello ‘beautiful’ and santo ‘holy’.12 (21) Vowel deletion between Italian clitics and hosts and between words a.
l’uomo la vita dell’uomo b. un’ape un’oca c. com’eravamo d. quell’orto bell’uomo sant’Onofrio
‘the man’ ‘the man’s life’ ‘a bee’ ‘a goose’ ‘how we were’ ‘this garden’ ‘beautiful man’ ‘holy Onofrio’ (Dizionari Garzanti 1994:1428)
The morphophonology of Italian clitic-host combinations shows external rather than internal sandhi processes. Other Romance languages also show a distinct morphophonology of clitic-host combinations. Portuguese 3rd person object clitics o, a, os and as, for example, trigger neutralisation of the coda consonant of hosts ending in /s/, /z/ or /r/ to [l] (see 22).13 (22) Neutralisation in Portuguese clitic-host combinations a. b. c. d.
fazer + o ela faz + o tu fazes + o far + o + ia
→ → → →
fazê-lo ela fa-lo tu faze-lo fá-lo-ía
‘to do it’ ‘she does it’ ‘you do it’ ‘would do it’ (Crysmann 2000:124–125)
As Crysmann (2000) points out, neutralisation is restricted to clitic-host combinations and does not occur at word boundaries or between stems and inXectional aYxes with similar phonological or morphophonological environment.
35
36
The status of Romance clitics between words and aYxes
(23) No neutralisation in Portuguese inXectional morphology Ele foi preso / *pre[l]o há três dias. prn.3sgaux.pst.3sg arrest.ptc.m have.3sg three days ‘He was arrested three days ago.’ (Crysmann 2000:125)
The only other regular context in which neutralisation occurs is the combination of two clitics.14 However, neutralisation does not apply without exception in clitic-clitic combinations but only if the Wrst of the clitics is 1st or 2nd person (e.g. nos, vos), with the result of clitic sequences such as no-lo and vo-lo. With other clitic-clitic combinations in European Portuguese, vowel deletion can be observed (see 24). (24) Vowel deletion in European Portuguese clitic-clitic combinations a. b. c. d.
me + o / a / os / as te + o / a / os / as lhe + o / a / os / as lhes + o / a / os / as
→ → → →
mo / ma / mos / mas to / ta / tos / tas lho / lha / lhos / lhas lhos / lhas
Deletion of vowels is important in Portuguese sentential prosody.15 It takes place between words (25a) but neither between stems and inXectional aYxes (25b) nor between verbs and clitics (25c). (25) Vowel deletion in Portuguese a.
um grande amigo [u; .Ágr!n.d !.Ámi.gu]
b. rece + o → rece + amos → c.
vendi + a → vendeste + o →
recei-o/ *reco rece-amos / *recamos
‘a great friend’ ‘I fear’ ‘we fear’
vendi-a / *vend-a ‘you sell it/her’ vendeste-o / *vendest-o ‘you(pl) sell it/him’
Portuguese vowel contraction thus resembles Italian vowel deletion in that clitics and words group together but clitics and aYxes do not (for more details on the diVerent prosodic behaviour of European Portuguese clitics and aYxes see Vigário 1999; Mateus & Andrade 2000). Clitic-clitic combinations in the Romance languages show many idiosyncratic morphophonological processes. Monachesi lists a number of idiosyncrasies in the Italian clitic sequence, most of which are also referred to as “opaque clitics” (see Bonet 1995). She points out that the Wrst of two clitics undergoes the process of vowel lowering if the second diVers from locative ci or reXexive si (26a). In the same context, 3f.io is replaced by 3io (26b).
Clisis vs. inXectional aYxation
ReXexive and impersonal si do combine, but the Wrst si is replaced by ci (26c). Locative vi does not combine with any other object clitic but is replaced by ci (26d). (26) Italian: vowel lowering and clitic substitution in clitic-clitic combinations a. b. c. d.
vowel lowering: 3f.io substitution: 3rfl substitution: loc substitution:
*mi ne *le lo *si si *vi vi
→ → → →
me ne; mi ci / *me ci glie lo ci si vi ci; *mi vi → mi ci
Chapter 4 contends that vowel raising as well as substitution are due to OCPlike restrictions on clitic sequences, i.e. within clitic sequences alliteration (as in *le lo) and homophony (as in *si si, *ci ci and *vi vi) are prohibited. The domain of these constraints is again the clitic sequence, since none of these processes is typical for aYxes or for words. Furthermore, we do not Wnd similar processes in clitic-host combinations. Vowel raising does not take place before the host, if its onset syllable is phonologically identical to the clitics which trigger vowel raising, as for example negherá (see 27a), and clitics and onset syllables of the host may be phonologically identical as in si situa (see 27b). (27) Italian: neither vowel lowering nor substitution in clitic-host combinations a. Martina ti /*te
negherà
una
spiegazione.
Martina 2o deny.fut.3sg d.idf explanation ‘Martina will deny an explanation to you.’ (Monachesi 1995:44) b. Se si guarda dall’ alto, if ip look.3sg p.d.df height il duomo si situa a destra della Piazza. d.df dome rfl locate.3sg p right p.d.df piazza ‘If one looks up, the dome is accommodated on the right side of the Piazza.’
We can conclude that Italian clitic-clitic combinations show a unique morphophonological behaviour which demands clitic sequence-speciWc rules or
37
38
The status of Romance clitics between words and aYxes
constraints. Further evidence comes from other Romance languages. We Wnd strikingly similar morphophonological idiosyncrasies of clitic-host and cliticclitic combinations which are not attested elsewhere in the morphology of the respective language. In Spanish, for instance, combinations of alliterating and homophonous clitics are also prohibited. As in Italian, the former are subject to substitution by the reXexive clitic (see 28a) — a process well-known as the “spurious se rule” (going back to Perlmutter 1970). (28) Clitic substitution and deletion in Spanish clitic-clitic combinations a. 3io substitution: b. 3rfl deletion:
*le lo → *se se →
se lo se
Chapter 4 proposes that diVerences between Spanish and Italian in the resulting clitic sequences come from diVerent clitic inventories. Since Spanish does not have unspeciWed clitics like the Italian ci, combinations of impersonal and reXexive se are not possible and one of both is deleted (see 28b).16 For a last example, let us look at the morphophonology of Romanian object clitics. Romanian object clitics show a great amount of phonological (and orthographic) variation which Monachesi (1999, 2000) takes as morphophonological idiosyncrasies.17 Popescu (2000) shows that all seemingly idiosyncratic forms of the clitics can be analysed in terms of prosody.18 She suggests that an important condition for Romanian clitics and clitic sequences demands that they do not consist of more than one syllable (see also Legendre 2000b,c). Therefore, clitics are obligatorily reduced if combined with other clitics (see 29a). In Romanian orthography, full and reduced forms are represented diVerently, 1st person direct object for instance as ma[ or m. Cliticclitic combinations again diVer from clitic-host combinations in that reduction is only optional with the latter (see 29b). (29) Alternation of the prosodic representation of Romanian object clitics a.
ma[ + am + prezentat → *ma[ -am / m-am prezentat 1do + aux.pfv.1sg + present.ptc *[m6.am] / [mam] ‘I presented myself’
b. ma[ + ara[ t 1do + show.1sg ‘I show myself’
→
ma[ -ara[ t / m-ara[ t [m6] / [m] (Popescu 2000:779)
What Monachesi calls idiosyncratic forms of clitics can therefore be analysed
Clisis vs. inXectional aYxation
as regular phonological change if there are constraints which refer to the clitic sequence and restrict it. This will be illustrated in chapter 4. In conclusion, morphophonological idiosyncrasies of Romance clitics exist. However, they cannot be taken as evidence in favour of the inXectional aYx status of clitics, since they often do not resemble the morphophonological processes observed with inXectional morphology. Some processes as, for instance, vowel deletion in Italian and Portuguese, are external rather than internal sandhi phenomena. Other processes as for instance clitic deletion or substitution in Spanish and Italian, are clitic-speciWc and not attested elsewhere in the phonology or morphophonology of the language. As will be shown in chapter 4, morphophonological constraints whose domain of application is the clitic sequence can account for this behaviour of clitics.
2.2.1.3 Rigid ordering Monachesi suggests that the strict internal ordering (Criterion II.2) of clitic combinations is another argument in favour of the aYxal status of clitics. Italian clitics strictly obey the order of indirect object preceding direct object (30b); the reverse order is ungrammatical (30c). The clitic order is diVerent from the order of full object DPs (30a). (30) Rigid ordering of Italian object clitics a.
Martina spedisce la lettera a Vito. Martina send.3sg d.df letter p Vito ‘Martina sends the letter to Vito.’
b. Martina me lo spedisce. Martina 1o 3do send.3sg ‘Martina sends it to me.’ c.
*Martina lo mi spedisce.
(Monachesi, 1995:38)
The ordering criterion, however, cannot be taken into account if one wants to prove the aYxal status of clitics, since it distinguishes clitics and aYxes on the one hand and words on the other. This becomes clear from Zwicky’s statement: “Ordering. ..... an element that is strictly ordered with respect to adjacent morphemes is almost surely a clitic (or an aYx), while an element exhibiting free order with respect to adjacent words is certainly an independent word. ...” Zwicky (1985:288)
39
40
The status of Romance clitics between words and aYxes
Strict ordering of elements is therefore not to be taken as an argument against clitic-hood but only as an argument against word-hood. Moreover, Zwicky’s criterion not only demands strict ordering of two elements with respect to each other but also with respect to the stem/host they attach to.19 With stem/ hosts, however, a remarkable ordering diVerence exists between Italian pronominal clitics and Italian inXectional aYxes. While the latter are suYxes without exception and do not allow any alternation in the direction of attachment to the respective stems, the former alternate: they occur behind inWnitives, participles and imperatives (see 31b,c) but before other Wnite verb forms (see 31a). A similar alternation is also found in the other Romance languages.20 (31) Alternation of position with respect to the host in Italian a.
Martina lo legge. Martina 3do read.3sg ‘Martina reads it.’
b. Visto-lo, fu facile decidere. see.ptc-3do be.pfv.3sg easy decide.inf ‘Having seen it, it was easy to decide.’ c.
Leggi-lo! read.2sg-3do ‘Read it!’
(Monachesi, 1995:47)
Monachesi argues that, although this alternation is limited to clitics and never found with aYxes, it is not strong evidence against clitics being aYxes. She gives two reasons for this. The Wrst is that some Italian dialects eliminate this alternation though the clitics have otherwise similar properties and the other that there are languages which allow mobile aYxes.21 In Neapolitan, for instance, postverbal clitics are only found with imperatives, but not with inWnitives or negative imperatives while in some northern dialects preverbal clitics seem to have vanished. If one regards clisis as a transitional stage in the development of aYxes from formerly free elements, this is exactly what is to be expected. Clitics maintain the relatively free order with respect to the host they attach to in the more prescriptive standard variants, while in less prescriptive dialects clitics have lost this characteristic property. In Standard Italian, however, clitics still behave as clitics and not as aYxes with respect to this criterion. As to the argument that languages are attested in which aYxes alternate between pre- and suYxation, there is one phenomenon in Italian
Clisis vs. inXectional aYxation
(and in the other Romance languages) which is not attested with aYxes and which cannot be explained if clitics are analysed as aYxes. Monachesi notes that in all languages which show this alternation, there is only one aYx (namely the person aYx) which alternates, for either phonological or morphological reasons. A hypothetical morphological complex structure with an alternating pers aYx is outlined in (32), the brackets indicating optional additional aYxes (the stem and the pers aYx are in bold letters). (32) Alternation of person aYx a. preWxation: (af1)-pers-(af2)-(af3)-stem-(af4)-(af5) b. suYxation: (af1)-(af2)-(af3)-stem-(af4)-pers-(af5)
AYxes attach to stems and not to other aYxes. If more than one of the aYxes were mobile — say, for instance, pers and af3 — we would expect that the ordering of pers and af3 with respect to the stem would remain the same, i.e. that there would be some kind of mirror image of the aYxes.22 The only possible suYxal alternates to the preWxation structure in (33a) are therefore the morphological structures in (33b.i and 33b.ii) where af3 is nearer to the stem than pers. Any other order is highly atypical for aYxes. (33) Alternation of person aYx and af3 a. preWxation: b. suYxation: i. ii. iii. iv.
(af1)-pers-(af2)-af3-stem-(af4)-(af5) (af1)-(af2)-stem-(af4)-af3 -pers-(af5) (af1)-(af2)-stem-af3-(af4)-pers-(af5) *(af1)-(af2)-stem-(af4)-pers-(af5)-af3 *(af1)-(af2)-stem-(af4)-pers-af3-(af5)
Let us now look at the alternation of pre- and postverbal clisis in Italian. As mentioned above, indirect object clitics always precede direct object clitics, so that 3io glie precedes 3do lo, regardless of the preverbal (34a) or postverbal (34b) position of the clitic sequence. (34) Italian: alternation of clitic-sequence position with respect to the host a.
Devo dar-glie-lo. must.1sg give-3io-3do ‘I must give it to him.’
b. Glie-lo devo dare. 3io-3do must.1sg give ‘I must give it to him.’
*Devo dar-lo-gli.
*Lo-gli devo dare.
41
42
The status of Romance clitics between words and aYxes
One of the most noticeable properties of clitic combinations is therefore, that the strict order of clitics is sustained with respect to each other and not — as is the case for aYxes — with respect to the host. I see this behaviour as a striking argument against the aYxal status of clitics. Furthermore, this proves that the clitic sequence must be considered a unit which is formed independently of the host. In chapter 4, I will show that there are morphological alignment constraints whose domain is the clitic sequence. These constraints lead to the above described order of clitics. With respect to the ordering criterion, all Romance languages behave similarly in that the clitic combinations maintain a strict internal order but alternate with respect to their host. Except for Portuguese, this alternation is morphologically triggered: preverbal clisis takes place with Wnite verb forms and postverbal clisis takes place with inWnitives, gerunds and imperatives.23 Portuguese, however, diVers remarkably from these generalisations in some respects. This leads to diVering conclusions about the aYxal status of the elements in question. One of the most obvious diVerences to other Romance languages is that European Portuguese clitics occur postverbally in unmarked sentences (35a). As in other Romance languages, the clitics may occur either at the inXected auxiliary/modal verb or at the inWnitive in analytic tenses and in modal verb construction, even if the clitics occur postverbally (see 35b,c). (35) Postverbal clisis is unmarked in European Portuguese a.
Eu vi-te ontem. prn.1 see.pst-2o yesterday ‘I saw you yesterday’
b. Nós vamo-nos encontrar amanhã. prn.1pl go.1pl-1pl.o meet.inf tomorrow ‘We shall meet tomorrow.’ c.
Nós vamos encontrar-nos amanhã. prn.1pl go.1pl meet.inf-1pl.o tomorrow ‘We shall meet tomorrow.’ (de Carvalho 1989:407–408)
Colloquial Brazilian Portuguese on the other hand, strongly favours the preverbal position for clitics (36a), though post-positioning of clitics is also attested. As in European Portuguese, however, there is no alternation of postverbal clitics at inWnitives and preverbal clitics at Wnite forms. Instead, we Wnd preverbal clitics throughout (36b,c)
Clisis vs. inXectional aYxation
(36) Preverbal clisis is unmarked in Brazilian Portuguese a.
Eu te vi ontem. prn.1 2o see.pst yesterday ‘I saw you yesterday’
b. Nós nos vamos encontrar amanhã. prn.1pl 1pl.o go.1pl meet.inf tomorrow ‘We shall meet tomorrow.’ c.
Nós vamos nos encontrar amanhã. prn1.pl go.1pl 1pl.o meet.inf tomorrow ‘We shall meet tomorrow.’ (de Carvalho 1989:407–408)
Brazilian Portuguese clitics, therefore, are not problematic for an inXectional aYx analysis. They are in fact more like aYxes than other Romance clitics. Preverbal position of clitics is also attested in European Portuguese but triggered by the semantics and syntax of the sentence.24 Clitic pronouns must occur in front of the verb in subordinated, negated and quantiWed clauses, in questions, after certain adverbs, indeWnite pronouns and certain prepositions and with emphatic scrambling structures.25 Pre-positioning of clitics is obligatory if one of these conditions is given, as for example with negation in (37). (37) Preverbal clisis in European Portuguese Não me lembrarei. not 1o remember.fut.1sg ‘I will not remember.’
(Krenn & Mendes, 1987:244)
It is irrelevant whether the syntactic trigger of pre-positioning is placed directly in front of the pronoun-verb complex or whether there are other elements separating trigger and clitic-verb complex. Furthermore, it is irrelevant which grammatical relation is expressed by the trigger. Therefore, pre-positioning licensing cannot be captured by any morphological means. Instead, one has to admit that European Portuguese clitics are aVected by syntactic and semantic properties of the sentence which makes them unlikely to be inXectional aYxes. There is, however, another phenomenon attested in both varieties of Portuguese (but not in other Romance languages) which implies that pronominal clitics should be analysed as aYxes. With future tense and conditional mood verb forms, the clitics occur after the stem and before the inXectional tense/mood suYx as in (38a,b). This phenomenon is know as “mesoclisis”. 26
43
44
The status of Romance clitics between words and aYxes
(38) Mesoclisis in Portuguese a.
comprá-lo-ás buy-3do-fut.2sg
‘you will buy it’
b. tê-lo-ás comprado ‘you will have bought it’ aux-3do-fut.2sg buy.inf (Hundertmark-Santos Martins 1998:196)
Future tense and conditional mood origin from the Vulgar Latin construction inWnitive + present tense of habe¯re ‘have’ and inWnitive + past tense of habe¯re, respectively. The inXectional endings nowadays are short forms (i.e. the inXectional endings of present and past tense) of the Portuguese verb haver ‘have’. From this data, we can at least conclude that the mesoclitic elements cannot be words. The question then is, whether they must necessarily be categorised as inXectional aYxes (as Crysmann 2000 does). One possibility, of course, is that the inXectional future and conditional endings are clitics, too. This option, though in agreement with the clitic status of weak pronouns, seems unlikely. If future and conditional endings are not clitics but aYxes, however, this is clear evidence in favour of the aYxal status of Portuguese clitics. There are still arguments against this conclusion, however. First, in European Portuguese, pre-positioning of clitics is still obligatory in future tense and conditional mood if one of the triggering conditions is given (see e.g. the negated sentence in 37 above). The inXectional pattern of words, as the complex of stem, clitic and tense suYx, for instance, is syntactically integer and cannot be changed by any syntactic trigger. Second, the assumption that Portuguese pronouns have developed from free elements to clitics and on to aYxes would predict that the complex of verb, clitic and inXectional ending is stable even in younger varieties of the respective language. Interestingly, de Carvalho (1989: footnote 1) gives examples from colloquial Brazilian Portuguese with proclisis in future tense and conditional mood. As alternatives to the standard mesoclisis in ele poder-se-ia lavar ‘he could wash himself’ the sentences ele se poderia lavar and ele poderia (já) se lavar, with se as proclitic in both sentences, are also possible. In contrast to Brazilian Portuguese, some variants of European Portuguese show postverbal clisis in these cases (i.e. poderia-se (já) lavar and ele poderia lavar-se). Moreover, van der Leeuw (1995: footnote 2) observes that synthetic future and conditional is no longer the unmarked strategy but is replaced by other analytic forms which then no longer show mesoclisis. In agreement with van der Leeuw (1995), I therefore
Clisis vs. inXectional aYxation
suggest that the inXectional endings are aYxes but that their morphologisation is not yet completed. I will return to this analysis in section 2.2.2. Together with the data on the prosodiWcation of clitics and the syntactic source of preverbal clitic position discussed above, mesoclisis can be taken as evidence that Portuguese clitics are at a transitional stage between aYxes and words, i.e. that they are neither aYxes nor words but clitics. How to analyse these complex alternation between pre- and postposition and mesoclisis will be shown in chapter 5.
2.2.1.4 Coordination According to Monachesi (1995, 2000), obligatory repetition of clitics in coordination structures is another argument in favour of the aYxal status of these clitics. This criterion goes back to Miller (1992) who argues that object clitics in French are aYxes because they cannot have wide scope over a coordination of hosts/stems (this criterion resembles Zwicky’s criterion “Deletion under identity”). Monachesi shows that Italian clitics also have to be repeated if two inXected verbs are coordinated, regardless of whether the respective tense is synthetic (39a,b) or analytic (39c,d).27 Deletion of the clitic in the second part of the coordination structure is only possible in analytic tenses, if the Wnite auxiliary is also deleted.28 Postverbal clitics are never left out in coordination structures (39f). (39) Scope of Italian clitics in coordination structures a.
Martina lo comprerà e lo indosserà alla festa. Martina 3do buy.fut.3 and 3do wear.fut.3 p.d.df party ‘Martina will buy (it) and wear it at the party.’
b. *Martina lo comprerà e indosserà alla festa. Martina 3do buy.fut.3 and wear.fut.3 p.d.df party c.
Martina lo ha comprato Martina 3do aux.3sg buy.ptc e lo ha indossato alla festa. and do aux.3 wear.ptc p.d.df party ‘Martina has bought it and worn it at the party.’
d. *Martina lo ha comprato Martina 3do aux.3sg buy.ptc e ha indossato alla festa. and aux.3sg wear.ptc p.d.df party
45
46
The status of Romance clitics between words and aYxes
e.
Martina lo ha comprato e indossato alla festa. Martina 3do aux.3sg buy.ptc and wear.ptc p.d.df party ‘Martina has bought and worn it at the party.’
f.
*Per comprare e indossarlo. p buy and wear.3do ‘To buy and wear it.’
(Monachesi 1995:40)
In other Romance languages, pronominal clitics usually do not have wide scope over coordination either (see, e.g., Miller 1992 for French; Monachesi 2000 and Legendre 2000b for Romanian). Only Portuguese behaves somewhat exceptionally again. In Portuguese, preverbal clitics have wide scope not only over coordination of bare Vs (40a) but also over coordination of VPs (40b). Postverbal clitics, on the other hand, never have wide scope over coordination of synthetic verbs (40c). With analytic verbs, wide scope over coordination is observed with pre- as well as with postverbal clitics (40c, d). As in the Italian example above, the complex of clitic and auxiliary is shared in both conjuncts. Deletion of only the clitic is not possible as (40f) shows. (40) Scope of Portuguese clitics in coordination structures a.
que o João o comprou e leu that d.df João 3do buy.pst.3sg and read.pst.3sg ‘that John bought it and read it’
b. Não sei se ele se barbeou cuidosamente not know.1sg if he 3rfl shave.pst.3sgcarefully ou lavou lentamente. or wash.pst.3sg slowly ‘I do not know whether he shaved (himself) carefully or washed slowly.’ c.
*O João comprou-o e leu / *O João comprou e leu-o
d. Ele já o tinha lido atentamente he already 3do aux.3sg read.ptc carefully e comentado com os amigos. and comment.ptc with the friends ‘He had already read it carefully and commented on it to his friends.’ e.
Eles tinham-nas ouvido e contado aos Wlhos. they aux-3pl.do hear.ptc and tell.ptc p.d.df sons ‘They had heard them and told them to their sons.’
Clisis vs. inXectional aYxation
f.
*Ele já o tinha lido atentamente e tinha comentado com os amigos. (Crysmann 2000:131)
Therefore, Portuguese clitics behave inconsistently regarding the coordination criterion. While preverbal clitics behave like words, postverbal clitics do not. This is again evidence in favour of the transitional status of Portuguese clitics.29 In general, it is not clear why this test should distinguish between aYxes and clitics as Miller (1992) claims.30 Following Zwicky (1985:288), deletion under identity is a property which whole words may undergo but neither proper parts of words nor proper parts of [word+clitic] combinations. Therefore, this test only shows that pronominal clitics in Italian are not words, i.e. they are not subject to syntactic processes. Nothing can be said, however, about the aYxal or clitic status of these elements.
2.2.1.5 High selectivity with respect to the host category Monachesi goes on to show that high selectivity with respect to the category of the host (criterion III.1) is attested for Italian pronominal clitics. Attachment to categories other than verbs, as for instance to attributive adjectives, is ungrammatical, even if the corresponding full pronoun is allowed in the same construction as an argument of the respective adjective (41b). (41) Host selectivity of Italian clitics a.
Martina lo legge. Martina 3do read.3sg ‘Martina reads it.’
b. *Un articolo gli incomprensibile. d.idf article 3io incomprehensible Un articolo incomprensibile per lui d.idf article incomprehensible p him ‘An article incomprehensible to him.’ (Monachesi, 1995:38)
The clitics’ degree of selection with respect to the host, however, diVers from that of inXectional aYxes, which are much more restricted regarding the stems they attach to. The choice of Italian verbal subject agreement suYxes, for instance, depends on tense, mood and aspect information of the stem (see 42a). Furthermore, they diVer with respect to inXectional classes, as the comparison of the forms of the a-verb cantare ‘sing’, the e-verb cedere ‘give up’ and
47
48
The status of Romance clitics between words and aYxes
the i-verb Wnire ‘Wnish’ shows (data from Canclini 1999:19). We can therefore argue that each subject agreement aYx has a very high degree of selection. As far as clitics are concerned, on the other hand, a similar degree of selection is not attested. Although subject clitics in French and Piattino are agreement markers, as will be shown in chapter 5, they do not show a similar degree of selection. As to Italian object clitics, we can add one and the same clitic to each of the verbal forms (see 42b). Moreover, clitics can also be added to inWnitives and participles but aYxes cannot. (42) a. verbal subject agreement in Italian 1sg cant-o ced-o Wnisc-o 3sg canta cede Wnisce 1sg.pfv canta-i cede-i Wni-i 3sg.pfv cant-ò cedé Wnì 1sg.sbj cant-i ced-a Wnisc-a 3sg.sbj cant-i ced-a Wnisc-a b. verbal subject agreement forms with 3rd person direct object clitic 1sg lo canto lo cedo lo Wnisco 3sg lo canta lo cede lo Wnisce 1sg.pfv lo cantai lo cedei lo Wnii 3sg.pfv lo cantò lo cedé lo Wnì 1sg.sbj lo canti lo ceda lo Wnisca 3sg.sbj lo canti lo ceda lo Wnisca
Another argument which calls into doubt the high degree of selection with respect to the host is that the phonological and the morphological host of the clitics may diVer in Romanian and possibly also in European Portuguese.31 This is an important diVerence to other Romance languages such as Italian, where the morphological and the phonological host of the clitic are identical. In Romanian, clitics may occur in front of the verb or after it. Adjacency to the verb is necessary and the verb can thus be regarded as the morphological host, but the clitics may be prosodiWed together with the element to their left. That means that they are phonologically enclitic. Strictly speaking, the clitics are enclitic but alternate regarding pre- and post-positioning to their morphological host, the verb.
Clisis vs. inXectional aYxation
(43) Romanian: the phonological host diVers from the morphological host [ma.ma l.]Pw place mother 3do like.3sg ‘mother likes him’
(Popescu 2000:784)
The category of the phonological host in these instances is irrelevant; it may consist of, e.g., nouns, verbs, complementisers or negation adverbs. Monachesi (2000) argues on the basis of Miller (1992) that this behaviour does not contradict the aYxal status. As pointed out above, there are two kinds of aYx analyses (inXectional and phrasal, respectively) suggested for clitics. Monachesi claims that clitics are inXectional aYxes. For inXectional aYxes, however, this kind of promiscuity is absolutely atypical. Although inXectional aYxes sometimes choose the lexical category of the stem they attach to relatively freely, as e.g. deWnite markers may occur either with adjectives or with nouns in Basque, they usually attach morphologically and phonologically to the same element. Moreover, inXectional aYxes are usually restricted to natural classes of lexical categories and are not free to choose either a negation element, a verb, a complementiser or a noun. The diversity of phonological and morphological hosts in Romanian (and at least for some speakers in European Portuguese) clearly contradicts the aYxal status of pronominal clitics in these languages. The concept of phrasal aYxes, on the other hand, explicitly includes the possibility of promiscuous attachment. This issue will be discussed in section 2.3 below. Let us Wnally consider two more criteria which are not part of Zwicky’s and Zwicky & Pullum’s list but result from Nübling’s scales B,C and E, which relate clitics to their respective full forms.
2.2.1.6 The relation of clitic and full form Regarding the phonological relation of clitics and full forms (scale B1), I will conWne the examination to Standard Italian.32 For other Romance languages, the argument is quite similar, so that the limitation to just one of the languages is advisable not only for reasons of space.33 Standard Italian clitics and full forms are shown in (44a) and (44b), respectively.
49
50
The status of Romance clitics between words and aYxes
(44) a.
Standard Italian clitics sg
1 2 3m 3f
do
io
lo la
mi ti gli le
ci, vi : ‘locative’
pl rfl do
io
rfl
ne:
‘partitive’
ci vi li le
si
gli
si
b. Standard Italian object pronouns sg
pl
do
io
rfl
1 2
me te
a me (a) me noi a te (a) te voi
a noi (a) noi a voi (a) voi
3m 3f
lui lei
a lui a lei
a loro (a) sé
(a) sé (a) lei
do
loro
io
rfl
di + lui, lei, loro: ‘partitive’
As to their phonological relation, a synchronic derivation of clitics from full forms is not discernible. Diachronically, 1st person mi, 2nd person ti and 3rd person reXexive si can be derived from the full pronouns me, te and sé by vowel raising.34 These forms resemble the Latin direct object pronouns me¯, te¯ and se¯, which were often used in non-contrastive and non-emphatic contexts (see Hofmann & Szantyr 1972:173–174) so that the development to special clitics in Italian did not involve a major change in meaning.35 Latin indirect object pronouns (mih¿H /m¿H , tib¿H , sib¿H ) diVered from direct object pronouns. The former did not survive but were replaced by the latter. With 3rd person clitics, however, the derivation from full forms is more complicated. The 3rd person feminine direct object la could be a reduced form of the full 3rd person feminine pronoun ella and the 3rd person indirect object gli a reduced form of the full 3rd person pronoun egli. This relation, however, is not synchronic, since both, ella and egli are restricted to subjects. 3f.do la could also be derived from the feminine deWnite article la. Analogously, 3rd person direct object lo and 3rd person indirect object gli resemble the masculine deWnite article in some instances.36 DeWnite articles in Italian are clitics, too. Full pronouns,
Clisis vs. inXectional aYxation
pronominal clitics and deWnite articles originate in the Latin demonstrative ille ‘that one yonder’ which lost its emphatic deictic meaning (see Wanner 1987 for a detailed diachronic analysis). 1st person plural ci and the full pronoun noi and the partitive and locative clitics ne and ci, Wnally, are suppletive. With respect to scale B1, Italian clitics therefore should still be categorised as special clitics, although verging on inXectional aYxes (see 45). (45)
simple clitics
special clitics >2
1
inXection > 3
> 4
B1: phonological relation of clitic and full form synchronic derivation
partly synchronic derivation
diachronic derivation
partial / total suppletive
no full form
Nübling’s scales B2, C and E2 interact strongly, so they are best considered together. In all Romance languages, changeability of pronominal clitics and full pronouns is only partial and involves a change in meaning (scale B2). The reason for this lies in the fact that full pronouns are always emphatic or focused, while clitics are chosen for topic objects (scale E2). In contrast to full forms, clitics cannot be modiWed, coordinated or contrastively stressed and they cannot occur in isolation (see Kayne’s 1975 syntactic criteria in section 1.1). As already pointed out above (see also 30), clitics and full forms do not occur at the same positions in the sentence. With the exception of European Portuguese, Romance clitics are selected only morphologically, as they have to be adjacent to the verb and the position with respect to their host is morphologically determined (scale C). For European Portuguese clitics, one has to admit at least an idiosyncratic syntax. Even the referential properties of clitics and full forms diVer in that clitics can refer to non-human entities while full forms usually cannot. Where to place Romance clitics in Nübling’s scales is shown in (46).
51
52
The status of Romance clitics between words and aYxes
(46)
simple clitics 1
special clitics > 2
inXection >
3
>
4
B2: interchangeability of clitic and full form everywhere partial impossible no change in no change in slight change heavy change — meaning meaning in meaning in meaning C: distribution and selectivity Syntax Syntax C=F
C⊂F
∃ P=position C in P &¬ F in P
E1: scope of the clitic scope C scope C ≈ becomes scope F smaller
C morphologically selected or idiosyncratic syntax of C
C morphologically selected
C and F may occur in the same minimal sentence -> C is no longer complement but agreement marker scope
E2: functionalisation of the clitic function C ≈ function C functionalised, function C ≈ C completely F,F focus or interchangeability function F functionalised emphatic, C topic with F restricted
With respect to the scope properties of clitics and full forms, at least a scope diminishment can be observed as the coordination data above shows. Some Romance languages, however, show a further decrease of the scope of the clitic in that full forms and clitics occur in the same sentence as for example in the Spanish sentence in (47). (47) Clitic doubling in Spanish Le do il libro a Juan. 3io give.1sg d.df book p Juan ‘I give the book to Juan.’
These clitic doubling structures are attested in Spanish, in Romanian, in French and in Piattino. In Italian and Portuguese, clitic doubling is only
Clisis vs. inXectional aYxation
marginal or not allowed at all. Monachesi (2000) argues that clitic doubling in Romanian is important evidence in favour of the aYxal status of clitics. However, there is still a diVerence between inXectional agreement morphology and clitic doubling. While inXectional agreement aYxes are functionalised and usually not restricted by the semantics of the respective argument, clitic doubling is highly restricted in several ways. First, case is relevant, as indirect objects are doubled more often than direct objects. Second, speciWcity is relevant, as speciWc objects are doubled more often than unspeciWc ones. Third, animacy is relevant, as human entities are doubled more often than non-human ones. Finally, person is also relevant, in particular with subject clitics, though a general preference tendency cannot be observed. In chapter 5, I will show that the requirement for aYxal or clitic verbal agreement morphology and the restriction of morphological marking to salient entities compete with each other. This competition leads to the complex clitic doubling data. In agreement with, e.g., Auger (1993) and Miller & Sag (1997), I will argue that the aYxal status of clitics and their (potential) agreement marking properties are not necessarily the same.
2.2.2 Auxiliaries and other clitic-like material As to other clitic-like material, the morphological status is no less a matter of dispute. For Romanian auxiliary clitics, for example, Monachesi (2000) argues that they are words, while Barbu (1998) argues that they are aYxes and Popescu (2000) argues that they are clitics. Monachesi reasons that auxiliaries must be words, because they function as hosts of pronominal clitics and form a prosodic word together with them. Under the assumption that clitics do not form prosodic words of their own, it must be the auxiliary clitic which has word-like properties. This argument does not stand up to a closer look at the data. In fact, Romanian auxiliaries do not form prosodic words of their own, if there are no pronominal clitics. As Popescu (2000) points out, they do not behave as verbal aYxes either (contrary to Barbu 1998), since they attach phonologically to their left, i.e. they are enclitic, though they occur preverbally by default (see 48). (48) Phonological host may diVer from morphological host in Romanian [twi.neÛ ar]Pw face asta? who aux.cnd.3sg do.inf this ‘Who would do this?’
(Popescu 2000:784)
53
54
The status of Romance clitics between words and aYxes
Monachesi furthermore argues that Romanian auxiliaries resemble non-clitic auxiliaries in other Romance languages, because they occur in similar positions with respect to the verb and they trigger clitic climbing. There are a number of diVerences between Romanian and the other Romance languages. Romanian word order is less restricted than word order in other Romance languages. Romanian permits SVO, VSO and VOS (Monachesi 2000), but the subject does not occur between the clitic auxiliary, which is a reduced form of the full verb avea ‘to possess’, and the matrix verb (all forms of the auxiliary clitic are given in the appendix). (49) Word order in Romanian a. mama a faa[ cut o pra[ jitura[ . mum aux.3sg make.ptc d.idf cake ‘Mum has made a cake.’ b. a faa[ cut o pra[ jitura[ mama aux.3sg make.ptc d.idf cake mum c.
a fa[ cut mama o pra[ jitura[ aux.3sg make.ptc mum d.idf cake
d. *a mama fa[ cut o pra[ jitura[ . aux.3sg mum make.ptc d.idf cake
(Monachesi 2000:278)
Subject-auxiliary inversion, however, is attested with other auxiliaries as the modal a putea ‘can’ (see 50a,b), as Legendre (2000b) points out. This inversion is prohibited in other Romance languages. (50) Subject-auxiliary inversion in Romanian a. Cînd vine Ion? when come.3sg John ‘When is John coming?’ b. Cînd poate Ion veni mîine? when can.3sg John come.inf tomorrow ‘When can John come tomorrow?’ c.
Ce a spus Ion? what aux.3sg say.ptc John ‘What has John said?’
d. *Ce a Ion spus? what aux.3sg John say.ptc
(Legendre 2000b:221)
Clisis vs. inXectional aYxation
This diVerence between modal auxiliaries and other (clitic) auxiliaries is strong evidence against analysing both of them as words. If we assume that the auxiliaries are clitics, however, this diVerence comes out automatically. Romanian clitic auxiliaries can be separated from the verb by some monosyllabic intensiWers which are also often argued to be clitics (see DobrovieSorin 1994) or aYxes (see Monachesi 2000) but not words. I will support the latter view below. Finally, the coordination facts contradict Monachesi’s claim that auxiliaries are words. As we have seen above, words may be deleted under identity, that is in coordination structures, while aYxes and clitics tend to be repeated. Legendre (2000b) points out that Romanian auxiliaries — in contrast to auxiliaries in other Romance languages — have to be repeated in coordination structures as a comparison of the Romanian (51a) and French (51b) examples shows. (51) Auxiliary coordination in Romanian and French a. Am va[ zut-o s¸i i-am vorbit. aux.1sg see.ptc-3f.do and 3f.io-aux.1sg talk.ptc ‘I have seen her and talked to her.’ b. J’ai bu et mangé à midi. I aux.pfv.1sg drink.ptc and eat.ptc p noon ‘I drank and ate at noon.’ (Legendre 2000b:227)
As Legendre notices, distinct word-like auxiliaries may be conjoined, as for example in English (52a) and French (52b). (52) Conjunction of distinct auxiliaries in English and French a. I can and will go out tonight. b. Pierre est et Pierre aux.3sg and sera sans aucun doute réélu sénateur. aux.fut.3sg without any doubt reelect.ptc senator ‘Pierre is and will without any doubt be reelected senator.’ (Legendre 2000b:227–228)
Romanian clitic auxiliaries cannot be conjoined if they are distinct (53a), while lexical verbs (53b) as well as verbs in diVerent tenses can, even if one of them contains a clitic auxiliary as in the coordination of the compound past tenses with simple present tense in (53c).
55
56
The status of Romance clitics between words and aYxes
(53) Coordination of verb phrases in Romanian a. *Ion va sau ar veni. John aux.fut.3sg or aux.cnd.3sg come ‘John will or would come.’ b. Lupul vîneaza[ sau ignoreaza[ oile. wolf.d.df hunt.3sg or ignore.3sg sheep.d.df ‘The wolf hunts or ignores the sheep.’ c.
Lupul a urma[ rit s6 i acum ucide oile. wolf.d.df aux.pfv.3sg follow.ptc and now kill.3sg sheep.d.df ‘The wolf followed and is now killing the sheep.’ (Legendre 2000b:228)
As has been shown above, repetition in coordination is a criterion which proves that the elements under examination are not words. Regarding Monachesi’s claim that auxiliaries are words, these coordination facts are problematic, particularly because her analysis of the pronominal clitics as aYxes relies very much on the coordination criterion. She thus has to assume that auxiliaries are aYxes as well. As a last point of evidence against the word-status as well as the inXectional aYx-status of Romanian auxiliary clitics, it should be noted that they do not always occur preverbally. Older or literary variants, aVectionate exclamations and curses show a similar change in the direction of attachment as attested with the pronominal clitics above. (54) Change in direction of attachment in certain registers a. Mâncat-ai? b. Mânca l-ar mama! eat.ptc-aux.pfv.2sg eat.inf 3do-aux.cnd.3sg mama ‘Did you eat?’ ‘He is sweet!’ (lit.: ‘Mama would eat him’) (Popescu 2000:784)
The order of pronominal clitics and auxiliary clitics remains the same, regardless whether they occur pre- or postverbally. This is highly atypical for words as well as for aYxes. Legendre (2000b) argues that this seems to be a remnant of formerly second position clitic phenomena. This assumption is supported by examples like (48) which imply that auxiliaries are, phonologically, enclitics by default. A diVerent explanation for enclisis with auxiliaries is based on the development from Vulgar Latin and sets into the same context Romanian auxiliary enclisis and Portuguese pronoun mesoclisis. Van der Leeuw (1995)
Clisis vs. inXectional aYxation
lists a number of archaic Romanian verb forms with enclisis of auxiliarypronoun complexes: (55) Enclisis in Archaic Romanian a. lua-le-ar take-3pl.do-aux.cnd.3sg
‘he would take them’
b. bate-se-vor Wght-rfl-aux.fut.3pl
‘they will Wght’
c. arde-l-ar focul burn-3do-aux.cnd.3sg Wre
‘may the Wre burn him’
d. vazutu-l-ai see.ptc-3do-aux.pfv.2sg
‘you have seen him’
e. dusu-s-au go.ptc-rfl-aux.pfv.3pl
‘they have gone’
f. cerutu-i-le-ai ‘you have asked it to them’ ask.ptc-3pl.io-3do-aux.pfv.2sg (van der Leeuw 1995:37)
Van der Leeuw (1995) compares these archaic Romanian forms with Portuguese mesoclisis (see 38 above, repeated here as 56). (56) Mesoclisis in Portuguese a.
comprá-lo-ás buy-3do-fut.2sg
‘you will buy it’
b. tê-lo-ás comprado ‘you will have bought it’ aux-3do-fut.2sg buy.inf (Hundertmark-Santos Martins 1998:196)
There are a number of diVerences between Romanian and Portuguese, as van der Leeuw observes: (i) Romanian mesoclisis is not restricted to future tense and conditional mood, while Portuguese is, (ii) in Romanian, more than one clitic may occur between stem and auxiliary (see 55f), while in Portuguese this position is restricted to one clitic or to contracted clitics and (iii) Romanian mesoclitic forms have only one primary stress, while Portuguese have two.37 Therefore, as van der Leeuw convincingly argues, a diVerence exists between Romanian and Portuguese in the degree of morphologisation of the Vulgar Latin inWnitive + auxiliary complex.38 This is further supported by the fact that structures like these are rather marginal in modern Romanian as pointed out above (see 54). Modern Romanian prefers pre-positioning of the pronominal
57
58
The status of Romance clitics between words and aYxes
+ auxiliary complex which has been shown to have clitic status. Let us now turn to the morphological status of the future and conditional endings in Portuguese. Future tense marking has developed from the free postverbal auxiliary in Vulgar Latin (57a) to reduced forms of the auxiliary in Old Portuguese (57b) and Wnally to inXectional endings in modern Portuguese (57c).39 (57) Development of future tense marking in Portuguese a. Vulgar Latin: ama¯re habe¯ o¯ ‘I will love’ b. Old Portuguese: amare hei ‘I will love’ c. Modern Portuguese: amarei ‘I will love’
Structures like these are also attested in Old Spanish.40 The unmarked object position in Vulgar Latin was between inWnitive and auxiliary so that sentences like ama¯re illum habe¯ ¯o ‘I will love him’ can be assumed to be the origin of mesoclisis in Portuguese (see 56).41 Van der Leeuw shows that the synthetic future and conditional forms diVer from other inXectional forms not only with respect to mesoclisis of pronouns but also with respect to stress assignment. Portuguese stress assignment usually obeys the “3-SilbenGesetz” (Wackernagel 1892, see also section 2.1 above). Which of the last three syllables receives the stress is mostly lexically determined. As the examples in (58a) show, verbs have a stress on the stem or on the theme vowel (which traditionally belongs to the stem) and there is no stress-readjustment after enclisis. With future and conditional verb forms, however, stress is always on the inXectional ending as shown in (58b). Remarkably, an additional primary (i.e. equally strong) stress on the stem or the theme vowel appears after mesoclisis.42 (58) Stress assignment in Portuguese a. stress assignment in present and past tense dígo → dígo-lho ‘I say it to him’ fazém → fazém-no ‘they do it’ contámos → contámo-lo ‘we told/counted it’ b. stress assignment in future tense and conditional mood diréi → faríam → contaríamos →
dír-lho-éi ‘I will say it to him/her’ fá-lo-íam ‘they would do it’ contá-lo-íamos ‘we would tell/count it’ (van der Leeuw 1995:32)
Clisis vs. inXectional aYxation
According to van der Leeuw, Portuguese does not have secondary stress. There are three other structures though, which appear with two primary stresses, namely compounds, adverbs with the suYx -mente and diminutives/ augmentatives with the suYxes -zinho / -zarrão. Future and conditional forms that contain clitics are therefore considered compounds consisting of two prosodic words.43 For mesoclisis, van der Leeuw gives an analysis in terms of Optimality Theory which I will discuss in detail in chapter 5. For the time being, only the conclusion that future and conditional inXectional endings may still behave like second parts of compounds is relevant, since this is strong evidence in favour of van der Leeuw’s claim that the morphologisation of these elements is not yet complete. Furthermore, except from three irregular verbs (dizer ‘say’, fazer ‘do’, trazer ‘bring’) the future/conditional stem is always identical to the inWnitive. For these irregular verbs, a reduced form of the inWnitive is chosen (i.e. dir, far, trar, see Hundertmark-Santos Martins 1998:127–128). The phonological processes between mesoclitic pronouns and future/conditional stems are exactly the same as with inWnitives (see above in 22). There are still several reasons not to regard these elements as clitics, however. First, the stress bearing property of the respective elements contradicts the deWning property of clitics, that they thus do not to bear stress. Second, the position of these elements does not change. In contexts that trigger preverbal clisis, pronominal clitics appear in front of the verb but future and conditional endings stay in the usual suYx position (see 37 above). Such separation or split of clitics is neither attested elsewhere in the Romance languages nor in other languages which have clitics. If the respective elements were clitics, we would expect clitic clustering either pre- or postverbally.44 Finally, there are a number of other clitics in Romance which are neither auxiliaries nor canonical arguments. All Romance languages have reXexive clitics. Italian, Piattino and French have locative and partitive clitics. French and Romanian have an additional negation clitic. Romanian, moreover, has a number of monosyllabic intensiWers which are often referred to as clitics (see, e.g., Dobrovie-Sorin 1994). As to reXexive, locative and partitive clitics, they belong to the clitic sequence and the same processes are attested as with other pronominal clitics. They are strictly ordered with respect to other clitics but not with respect to the host. Alternation between pre- and postverbal position depends on the morphological speciWcation of the host. It is especially reXexive clitics that take part in morpho-phonologically induced substitution and deletion processes (see, e.g., the examples in 26 and 28 on p. 47–49 above).
59
60
The status of Romance clitics between words and aYxes
Therefore, I conclude that they are clitics as well. More diYcult is the case of negation clitics in French and Romanian. For French, it is reasonable to assume that the negation has the same status as the pronominal clitics discussed above, since it is positioned between subject and object clitics (see 59) and cannot occur in isolation. It is stressless and the vowel is omitted before verbs or other clitics (i.e. partitive en and locative y) which begin with a vowel. (59) French negation clitic Je ne l’ aime pas du tout. 1s neg 3do love neg p.d.df all ‘I don’t like him a bit.’
(Klein & Kleineidam 1979:103)
In Romanian, the negation element nu shows more word-like properties.45 It is never embedded into clitic clusters but occurs on the left of VPs including clitics (60a). Although this might be due to the fact that Romanian does not have subject clitics, the Romanian negation may occur in isolation (see 60b), which is clear evidence against the clitic status. That the negation may serve as a host for other clitics (see 60c) does not mean that it must be a clitic itself, since this is a well attested property also of elements which are clearly not clitics (compare the examples in 43 and 48 above). I therefore conclude that the Romanian negation is a word and not a clitic (in agreement with Monachesi 2000). (60) Romanian negation a. Nu l-am mai va[ zut. neg 3do-aux.pfv.1sg again see.ptc ‘I haven’t seen him again.’ b. Radu a venit dar Ion nu. Radu aux.pfv.3sg come.ptc but Ion neg ‘Radu has come but Ion hasn’t.’ c.
Nu-l da. neg-3do give.3sg ‘He doesn’t give it.’
(Monachesi 2000:287)
The Romanian monosyllabic intensiWers mai, cam, prea, ¸si, tot, on the other hand, are presumably derivational aYxes and not clitics (in agreement with Legendre 2000c). As to their clitic status, it has been argued that intensiWers are the only elements which may occur between clitic sequences and verbs (Dobrovie-Sorin 1994). However, there are a number of diVerences between
Clisis vs. phrasal aYxation
pronominal and auxiliary clitics on the one hand and intensiWers on the other. Other than auxiliary and pronominal clitics, their position with respect to the host is always Wxed, i.e. they always occur preverbally. In instances of obligatory postverbal position of clitics (i.e. with imperatives), intensiWers and clitics are therefore split up. The morphological and the phonological host are never distinct, though the intensiWers are not restricted to verbs only but may also attach to adverbs. IntensiWers do not behave like adverbs with word status either, since the latter are placed behind the verb. Another recent approach which dispenses with the term “clitic” is that of Anderson (1992). Going back to Klavans’ (1980, 1985) taxonomy of clitics, Anderson suggests that clitics should be treated as a kind of aYxes, namely as phrasal aYxes. In the following section, I will examine Anderson’s (1992) suggestion to analyse special clitics as a particular kind of aYx, namely “phrasal aYx”.
2.3. Clisis vs. phrasal aYxation Based on Klavans’ (1980, 1985) work on the positional properties of clitics, Anderson (1992) argues that clitics behave like aYxes and that the only diVerence is that clitics are adjoined to syntactic phrases, while aYxes are adjoined to words. Anderson argues that clitics are bundles of morphosyntactic features which are added to heads. Their surface form is the result of the application of morphological word formation rules. The concept of clitics as phrasal aYxes is adopted in the literature, for example in the framework of Paradigm Function Morphology (Spencer 2000) and in the framework of Optimality Theory (see Legendre 1996, 2000b,c).46 I will therefore outline Klavans’ taxonomy of clitics and Anderson’s adaptation together with his deWnition of phrasal aYxation before I discuss the beneWts and faults of the concept of phrasal aYxation with respect to clitics in Romance languages. Klavans (1985) observes that the placement of clitics can be categorised with respect to three parameters: (61) Clitic Placement Parameters (Klavans 1985) Parameter I (Dominance): initial/final Parameter II (Precedence): before/after Parameter III (Phonological Liaison): proclitic/enclitic
61
62
The status of Romance clitics between words and aYxes
The Wrst of these parameters expresses the possibility that a clitic attaches to the initial or Wnal constituent dominated by a speciWed syntactic phrase which is understood to be the domain of clisis. The second parameter speciWes whether a clitic occurs before or after the element chosen by parameter I. The third parameter gives the direction of phonological attachment, which is a property of the clitic itself. To clarify how these parameters work let us look at the well-known English possessive ’s. Before we can assign the parameter values we have to determine the domain of clisis. For English possessive ’s, the domain consists of an NP, as can be seen by comparing the examples in (62). (62) English possessive ’s a. the queen of England’s hat b. the boy who I saw’s mother c. the boy who I talked to’s sister
English possessive ’s chooses the Wnal element of its domain (parameter I), it occurs behind this element (parameter II) and it phonologically leans on the element to its left (parameter III). The parameter values are thus final/after/ enclitic. Seven additional cases result from parameter value combinations. Some of these cases are rather problematic and rarely (or not) attested in the languages of the world.47 Anderson (1992) adopts Klavans’ parameters and compares the placement of clitics to the placement of aYxes. He observes that aYxes can also be characterised by the domain of aYxation, which diVers from the domain of clisis in that it consists of a morphological and not a syntactic constituent. Within this constituent, the reference element must be determined. For aYxes, the reference element is the Wrst, the last or the stressed element within the domain. The choice of the reference element is equivalent to Klavans’ parameter I. Finally, the direction regarding the reference element (i.e. before or after) must be determined. This equals Klavans’ parameter II. Anderson argues that Klavans’ parameter III is unnecessary for clitics as well as for aYxes, since it results from the rules for stray adjunction in the respective language. He goes on to say that the only diVerence between clitics and aYxes is the kind of domain to which they attach. While aYxes attach to morphological constituents, clitics attach to syntactic constituents. Anderson thus calls clitics “phrasal aYxes”, an expression which emphasises the parallel behaviour of both kinds of linguistic material but also points to the main diVerence between them, namely the relevance of the syntax for clitics. The case of English
Clisis vs. phrasal aYxation
’s Wts nicely into the model, particularly, because the syntactic domain of clisis contains its phonological host. For aYxes this is always the case. For clitics, however, we can observe that they may be phonologically integrated into material which is not part of the clisis domain. In section 2.2.1.5, we saw that this is the case at least in Romanian (see 43, repeated as 63 for convenience). (63) Syntactic domain and phonological host are distinct in Romanian a. [ma.ma l.]Pw place mother 3do like.3sg ‘mother likes him’ b. [twi.neÛ ar]Pw face asta? who aux.cnd.3sg do.inf this ‘Who would do this?’
(Popescu 2000:784)
In Klavans’ analysis, this behaviour results from diVerent directions with regard to the parameters I/II on the one hand and III on the other hand. Let us assume in accordance with Klavans that the domain of clisis is V for Romanian.48 The clitics choose the initial element of their domain and are placed before it, but they are integrated into the phonological word on their left, which is not part of the clisis domain. The parameter setting is thus initial/ before/enclitic. In Klavans’ analysis, the phonological direction of clisis is part of the lexical information of the clitic, while Anderson argues that the language-speciWc rules with regard to stray adjunction are responsible for this behaviour. Anderson as well as Klavans are thus able to analyse the data above. The model of phrasal aYxation, however, is problematic for empirical as well as for theoretical reasons. The Wrst problem arises with respect to the domain of clisis in other cases of Romance clitic placement. Consider Wrst the alternation of preverbal and postverbal position in Portuguese. As explained above, it is a syntactic or a semantic trigger (e.g. negation, question marker, certain quantiWers, etc) that requires the clitics to occur preverbally (i.e. initial vs. final). It is unclear, how this condition can be integrated into the model. Secondly, we Wnd mesoclisis in future tense and conditional mood (see 38). (64) Mesoclisis in Portuguese Dar-me-ão eles um bom odenado? give.inf-1o-3pl.fut they d.idf good salary ‘Will they give me a good salary?’ (Hundertmark-Santos Martins 1998:128)
63
64
The status of Romance clitics between words and aYxes
It is not clear how the model can explain that the reference element in this case is no longer a whole word but instead an inXectional ending. As regards the general alternation in Romance between preverbal position with Wnite verbs, auxiliaries and modals and postverbal position with imperatives, inWnitives and participles, more problems arise. Mostly, clitics attach obligatorily to the auxiliary in analytic tenses (65a), but optionally either to the modal or to the inWnitive in modal verb constructions (65b). (65) a.
Analytic tenses in Italian L’ ho vista ieri. 3do.f aux.1sg see.ptc.f yesterday
‘I saw her yesterday.’
*Ho vista-la ieri. aux.1sg see.ptc.f-3do.f yesterday b. Modal verb constructions in Italian Devo dar-lo a Claudia. must.1sg give.inf-3do p Claudia Lo devo dare a Claudia. 3do must.1sg give.inf p Claudia
‘I must give it to Claudia.’ ‘I must give it to Claudia.’
The question is then, whether it is one and the same domain which applies to all these cases, or whether non-Wnite and Wnite verbs must be considered diVerent domains, i.e. V[-Wn] and V[+Wn] respectively (see Klavans 1980).49 With the latter option, diVerent domains of clisis require diVerent parameter settings. Since the domains consist of only one element, however, the setting of parameter I is irrelevant. For Wnite verbs, Klavans’ parameters are (initial,final)/before/proclitic, while for non-Wnite verbs the parameters are (initial,final)/after/enclitic. With the former option, two diVerent parameter settings would be necessary, namely (i) initial/before/proclitic and (ii) final/after/enclitic. While (i) must be obligatory for Wnite verbs and auxiliaries and (ii) must be obligatory for imperatives, participles and inWnitives in isolation, the choice between (i) and (ii) must be free for modal verb constructions. Neither Klavans’ (1980, 1985) nor Anderson’s (1992) model provides the means to explain these diVerences. Moreover, both options need parameter III with the values proclitic for preverbal clitics and enclitic for postverbal clitics, which guarantees that the clitics are phonologically part of their domain. This parameter, however, cannot be lexically determined, as Klavans argues, since the same clitic is either proclitic or
Conclusions
enclitic depending on the rest of the parameters — at least for Italian clitics. Anderson’s proposal to replace this parameter by the rules of stray adjunction must be rejected for the same reasons. Recently, however, Optimality-theoretical analyses have been developed which advance the Klavans/Anderson model by partly converting the parameters into violable alignment constraints. These clitic-speciWc alignment constraints compete with other alignment constraints for, e.g., Wnite verbs (see e.g. Anderson 1996; Legendre 1996, 2000b,c). Another strength of Optimality Theory in the placement of clitics is that constraints from other modules as, for instance, syntax and semantics, may take part in this competition and thus inXuence the placement of clitics. These analyses will be discussed in chapter 5.
2.4. Conclusions The Romance languages possess a number of elements which are traditionally called clitics. In the preceding chapter, the phonological and morphological properties of these elements have been examined in detail. As to the phonology of clitics, it has been shown that they diVer from aYxes as well as from words. A distinct prosodic category like the clitic group (Hayes 1989; Nespor & Vogel 1986), however, is not necessary to capture the behaviour of these elements. Clitic phenomena can be explained on the basis of the prosodic hierarchy and the diVerent processes of integration, adjunction and incorporation into prosodic constituents. The term “clitic” is nonetheless useful when referring to this class of linguistic elements, which exhibits a particular phonological behaviour. The main part of the preceding chapter was concerned with the morphological foundation of a distinct category “clitic”. The respective elements have been recategorised recently as inXectional aYxes or words (see e.g. Barbu 1998; Crysmann 2000; Miller 1992; Miller & Sag 1997; Monachesi 1995, 2000). A detailed reexamination according to the Zwicky criteria and Nübling’s bundle of scales, however, has lead to diVerent conclusions. I have demonstrated that Romance pronominal subject, object and reXexive clitics as well as locative and partitive clitics are not inXectional aYxes, yet, but special clitics. There are two main reasons for this conclusion. The Wrst one is that these elements, unlike aYxes, change the direction of attachment with respect to their host. The second and more important one is that the rigid internal order
65
66
The status of Romance clitics between words and aYxes
of clitic combinations is sustained regardless of whether they are pre- or postpositioned with respect to their host. There is only one test which contradicts the clitic status of these elements, and that is the existence of arbitrary gaps (criterion I.1) in clitic-host combinations (attested in Italian and French) as well as clitic-clitic combinations (attested in all Romance languages except Brazilian Portuguese). On the other hand, researchers who advocate the aYxal status of the respective elements ignore the fact that they do not combine with stems, as is expected from aYxes (criterion I.4), but with words, as is the case with clitics and words. (Note that Romance clitics do not combine with phrases, as would be predicted by criterion II.5, either.) Some of the other tests only show that clitics are not words (criteria II.1-6). The order and placement of clitics (criterion II.2) with respect to their host and with respect to each other particularly contradict the aYxal status of clitics. One might recall for instance, that pre-positioning of clitics in European Portuguese is syntactically or semantically triggered and that pre-positioning even prevails over mesoclisis in future tense and conditional mood. Moreover, combinations of clitics but not combinations of clitics and hosts maintain a strict internal order, which is highly untypical for aYxes. Instead, these facts suggest that clitics form a unit, the clitic sequence, which is subject to speciWc ordering requirements. Still other criteria show that clitics are not yet aYxes, since the morphophonological processes between clitics and hosts resemble the morphophonological processes on word boundaries and not the morphophonological processes between aYxes and stems (criterion I.2). Furthermore, the morphophonological processes between clitics are not attested elsewhere in the morphophonology of the respective languages, so that it seems reasonable to assume clitic sequence-speciWc constraints. Finally, clitics are not as selective as inXectional aYxes are regarding their host category, and their phonological host may diVer from their morphosyntactic host, which is not attested for inXectional aYxes. Auxiliaries have been considered in Romanian and Portuguese. While the former have been shown to be clitics for the same reasons as pronominal clitics, the latter have been identiWed as inXectional endings which are not yet completely morphologised. As to other elements, the picture is not as uniform. While the negation element in French is a clitic because of its placement between subject and object clitics, the negation element in Romanian turns out to be a word, because it occurs to the left of the clitic cluster and not within it and, moreover, it may occur in isolation.
Notes
Monosyllabic intensiWers in Romanian, Wnally, are derivational aYxes. The Wnal part of this chapter has been concerned with the concept of phrasal aYxation (Klavans 1980, 1985; Anderson 1992), which relies on two or three diVerent parameters to predict the placement of clitics. As far as Romance languages are concerned, the concept of phrasal aYxation has been shown to be too rigid. Recent analyses within the framework of Optimality Theory, however, convert these parameters into violable constraints. We will come back to these analyses in chapter 5. Before detailed analyses of clitic sequences, clitic placement and clitic doubling will be presented in chapters 4–5, the Romance clitic paradigms will be introduced and the clitics will be provided with lexical entries in the next chapter. Furthermore, the morphological model suggested here will be delineated and the place of morphology in grammar sketched.
Notes 1. In Romanian (see Popescu 2000) and in medieval varieties of Romance languages (see Fischer 2000) (and — according to de Carvalho 1989 — also in Portuguese) the phonological host might be distinct from the syntactic phrase which contains the clitic (see also section 2.2.1.5 below). 2. “Directional clitics” differ from other clitics in that the phonological dependence on an element to the left or right is an inherent property of directional clitics, but not of other clitics (which may find their host either to the left or to the right). 3. The numbers in the table are my own. 4. Most criteria are stated as tendencies. Criterion I.2, for instance, reads “Arbitrary gaps are more characteristic of affixed words than of clitic groups.” (Zwicky & Pullum 1983: 504). 5. Scale E1 subsumes Zwicky’s criterion II.6 “deletion under identity”, though. 6. See, however, Morin (1981) who shows the limits of this characterisation of French clitics. 7. A more elaborate version of criterion II.6 “deletion under identity” can be found in Miller (1992) referred to as the “coordination criterion”. 8. Note that there is a slight difference in grammaticality with respect to 3pl.do. While the combination with a plural participle is marginally accepted (12c), the combination with a singular participle is ungrammatical. *L’ argomento riguardante-li d.df topic concern.ptc.prs-3do.pl (Monachesi, 1995:42)
67
68
The status of Romance clitics between words and aYxes
9. Such restrictions are attested, e.g. for person marking affixes in Quechua (see Lakämper & Wunderlich 1998; Lakämper 2000) and in Bantu languages (see, e.g. Cocchi 2000). Haspelmath (2001) observes that similar restrictions also affect the combination of full pronouns in many languages. 10. See Perlmutter (1970) for a detailed description. 11. The definite articles il, lo, la, i, gli and le cliticise to nouns and to the prepositions a, da, di and in. The article-preposition combinations for, e.g., di + article, are del, dello, della, dei, degli and delle. 12. Note that Italian orthography distinguishes between ‘elision’ and ‘truncation’. Elision denotes deletion of a syllable-final vowel in order to avoid hiatus at word (or clitic-host) boundaries and is marked by an accent. Elision does not take place with inflectional endings (amori infelici /*amor’infelici ‘unhappy love affairs’; mangiava erba/ *mangiav’ erba ‘she ate herbs’). Truncation denotes mostly optional deletion of unstressed /e/ in word-final syllables which begin in /l/, /r/ or /n/ and is not marked at all (male/mal ‘bad’; andare/andar ‘go’; grande/gran ‘great’). Truncation is restricted to unmarked number and gender, i.e. singular (cuor contento ‘content heart’; *cuor / cuori contenti ‘content hearts’) and masculine (un bel ragazzo ‘a beautiful young man’; *una bel / una bella ragazza ‘a beautiful young woman’). In addition, there are phonological restrictions on truncation as well as some obligatory cases of truncation (see, e.g., Dizionari Garzanti 1994: 1429). As Nespor (1990) shows, truncation is optional between words but obligatory between hosts and enclitics (see also Peperkamp 1997:199–204). 13. In future tense and conditional mood, Portuguese clitics appear between stem and inflectional ending, a phenomenon called “mesoclisis”. I will return to this problem below. 14. Clitic sequences in Brazilian Portuguese are restricted to only one clitic. 15. While /e/ as a final vowel is usually deleted, other vowels are contracted (e.g. [ɑ] + [u] = [f]) or become diphthongs ([ɑ] + [u] = [ɑu]). 16. The lexical entries for the clitics are introduced in chapter 3. 17. See also Avram (1986) on sandhi phenomena in Romanian. 18. Popescu is in agreement with the conclusions of this paper in that she regards Romanian weak pronouns as clitics (in contrast to Barbu 1998 and Monachesi 1999, 2000, who argue that they are affixes). 19. In the remainder of the quote, Zwicky actually allows for a certain degree of freedom in the order of clitics: “...Again, some complexity exists, since clitics on occasion exhibit some freedom of order with respect to one another (this is true for Tagalog clitics ...) though not normally with respect to their hosts....” (Zwicky, 1985: 288). 20. See, e. g., Muller (1998) regarding French and Portuguese and Popescu (2000) regarding Romanian. 21. Monachesi refers to Arabic discussed by Fontana (1993), Huave discussed by Noyer (1994) and Afar discussed by Fulmer (1990).
Notes
22. This is only a hypothetical case, since I am not aware of a language with more than one mobile affix category. 23. In Romanian, the infinitive occurs with preverbal clitics as well. 24. See also Barbosa (1993, 1996) and Martins (1993). 25. There are certain quantifiers which do not trigger pre-positioning, e.g. muitos ‘many’ and alguns ‘some’, while poucos ‘few’ and nenhuns ‘no’ do. Crysmann (2000) argues that the inherent lexico-semantic property ‘downwards monotonicity’ is the crucial property which licenses proclisis. 26. Portuguese has different instances of infinitives: comprar indicates simultaneity and comprado indicates anteriority. 27. According to Monachesi, deletion is possible with verbs of the type V and re-V as in Lo leggo e rileggo. ‘I read it and reread it.’ The inflectional person suffix -o, of course, cannot be deleted in such contexts. Even with these verbs, enclitics cannot be deleted *Per leggere e rileggerlo. ‘To read and reread it.’ (Monachesi 1995:40-41) 28. My informant does not agree with Monachesi in the judgement of (39b) and (39e). According to him, both examples are equally good. 29. See also Kato (1993) on conditions which allow deletion of clitics in Brazilian Portuguese. 30. Monachesi (1995, 2000), Crysmann (2000), Legendre (2000b) adopt Miller’s coordination criterion and argue that wide scope over coordination is a sufficient criterion in favour of the affixal status of elements. See also Ortmann & Popescu (2000). 31. Regarding the choice of morphological and phonological host in European Portuguese, conflicting data are presented. While Vigário (1999) and Ana Luis (p.c.) assert that European Portuguese clitics are always prosodified together with their morphological host, the verb, de Carvalho (1989) argues that morphological and phonological host always differ if the clitics occur preverbally. According to de Carvalho, preverbal clitics are phonologically enclitic not to the verb but to the element to their left “whatever the preceding word may be from a grammatical point of view”. De Carvalho gives the following examples (clitics are bold). (i) The phonological host differs from the morphological host in EP ‘if he saw you...’ a. [se-ele-te]Pw visse ... ‘who sees me ?’ b. [quem-me]Pw vê ? ‘I did not see you.’ (de Carvalho 1989: 409) c. [não-te]Pw vi. In unmarked European Portuguese sentences the morphological and the phonological host do not differ, so we find post-positioning and enclisis to the verb (see 35a above). Brazilian Portuguese clitics, on the other hand, do not alternate between pre-postioning/ proclisis and post-positioning/enclitsis but are stable with respect to the direction of attachment. Their morphological and phonological host is identical. 32. Compare also Calabrese (1995) on the development of Italian clitic systems. 33. See, e.g., Aguado & Lehmann (1989) on Catalan, Fontana (1994) on Spanish.
69
70
The status of Romance clitics between words and aYxes
34. Vowel raising of syllable-final vowels is not attested synchronically in Standard Italian. Note, however, that vowel lowering takes place within clitic sequences (see p. 47). 35. As to a general scenario on the development from free pronouns to clitics and on to affixes in Uto-Atztecan languages see Steele (1977). Some problematic issues of the evolution of clitics are discussed in Jeffers & Zwicky (1980). 36. The phonological representation of the masculine d.df is dependent on the following word. In the singular, it is lo before any cluster consisting of /s/+consonant and before /z/, /ps/, /gn/, /x/, and l’ before a vowel (see also footnote 12). In plural, it is gli in these cases. Elsewhere, it is il in singular and i in plural. 37. Contrary to Hundertmark-Santos Martins (1998: 128) who lists oferecer-no-lo-ão ‘they will present it to us’ and Spencer (1991: 366) who lists mostrar-no-los-á ‘he will show it to us’, van der Leeuw’s informants did not recognise structures with two distinct clitics between stem and inflectional ending as correct. Instead, they preferred one of the objects to be expressed by a DP. According to Alexandra Popescu (p.c.), forms with two clitics have a similar, i.e. “grammatical but not in use”, status in Romanian. They still occur in liturgical texts. 38. Note, however, that the conclusion which van der Leeuw draws is somewhat problematic. Van der Leeuw argues that morphologisation of Romanian auxiliaries has proceeded further than that of Portuguese auxiliaries for two reasons. Firstly, there is only one primary stress in Romanian, and secondly, modern Romanian does not have mesoclisis anymore. This last fact, though, is due to the pre-positioning of the pronoun-auxiliary complex and not of the extra-positioning of pronouns alone which we would expect if the auxiliary had developed to an inflectional ending and which is what actually happens in modern colloquial Portuguese (see p. 56). 39. The complete paradigm of haver and of the respective inflectional endings is listed in the appendix. 40. Van der Leeuw (1995) notes that some Italian dialects as, for instance, Pugliese also still show similar structures and Rivera-Castillo (1997) discusses word-internal clitics in Caribbean Spanish. 41. As mentioned above, the synthetic mesoclitic verb forms are no longer the standard way to express the future tense. Instead, the future tense is formed analytically with the verb ir ‘to go’ or with the auxiliary haver in combination with the preposition de and the infinitive (see van der Leeuw 1995). 42. Note that stress marks on the vowels are taken from van der Leeuw and do not always coincide with Portuguese orthographic stress marks. 43. Van der Leeuw notes that there is a difference in stress assignment of compounds on the one hand and future and conditional forms on the other hand. While compounds always have one stronger and one weaker stress, future and conditional forms have two identical stresses when used in isolation. 44. See Crysmann (2000) for other arguments against the clitic status of future/conditional endings.
Notes
45. The negation element ne is obviously an affix (see Monachesi 2000 on this topic). 46. The respective analyses will be discussed in detail in chapters 4 (i.e. Clitic sequences) (Spencer 2000) and 5 (i.e. Clitic placement) (Legendre 2000b). 47. See Spencer (1991) for discussion and criticism on this topic. 48. I will return to the domain of clisis in Romance below and, in more detail, in chapter 5. 49. A problem remains with imperatives which behave like non-finite verbs.
71
72
Point of view and grammar
The place of morphology in grammar
Chapter 3
Clitics in the lexicon
During the last decade, the morphological component has received a great deal of attention. Not only derivational and compositional morphology but also inXectional morphology has been shown to pose severe problems to purely syntactic analyses. Irregular morphology, suppletion processes and morphophonological processes often cannot be explained by transformational models such as that of Pollock (1989). From the properties of clitics discussed in chapter 2, it is clear that they are particularly resistant to a transformational model. Their rigid order with regard to each other is predictable from neither their origin in syntactic argument roles nor from their origin in agreement phrases. Which clitics are allowed to combine and which are not is also not syntactically restricted. Their order with respect to the host is also diYcult to determine syntactically, especially when sequences of clitics are involved. Furthermore, clitic substitution and deletion is determined morphophonologically and cannot be solved in a model which denies the existence of a morphological component. Finally, the conditions of clitic doubling are semantically and pragmatically too complex to be captured by purely syntactic means. Clitics therefore demand a theory which regards morphology as an autonomous component, as for example the lexical theory of MM. Prior to the analysis of the complex clitic phenomena such as clitic sequencing, clitic placement and clitic doubling, the lexical inventories have to be identiWed. There are two issues to be examined in this respect. First, certain similarities can be observed across languages, such as the restriction of gender and case distinctions to 3rd person clitics. It will be shown that such restrictions reXect universal tendencies which result from general considerations on markedness. Second, the question has to be answered how the optimal output correspondent to each argument role is established. I will demonstrate that the work between constraint ranking and lexical entries is divided so that underspeciWed entries together with adequate constraint rankings lead to the optimal output candidates. Before introducing the Romance clitic inventories in section 3.2, I will
73
74
Clitics in the lexicon
locate the morphological component in the grammatical model suggested in section 3.1. As a Wrst example of how the clitic choice can take place if grammar consists of violable constraints, Grimshaw’s OT analysis of Romance clitics is discussed in section 3.2. The following section analyses the Romance clitic inventories and provides the clitics with lexical entries in the fashion of MM in sections 3.2.2 to 3.2.7. First, I will look at the languages with a rather small clitic inventory, i.e. Portuguese and Spanish. The basics of MM such as the speciWc notion of paradigms, the structure of lexical entries and the interaction with faithfulness and markedness constraints in determining the lexical inventory will be explained, along with the analysis of the Portuguese clitic inventory. Then I will move on to Italian, Piattino and French. Finally, the Romanian clitic inventory is analysed.
3.1. The place of morphology in grammar The grammar model suggested here combines the ideas of MM (see section 1.3.1) and of LDG (see section 1.3.2) with the mechanism of constraint ranking and the concept of constraint violability provided by OT and CT (see section 1.3.3).1 The lexicon is assumed to contain the language-speciWc inventory including all functional and non-functional elements. Clitics also play a major role in other recently developed morphological models, especially in Distributed Morphology (Halle & Marantz 1994), Paradigm Function Morphology (Stump 1993) and Anderson’s (1992) A-Morphous Morphology. The main diVerence to these other morphological models is that MM is an “earlyinsertion” model, while the other models are so-called “late-insertion” models. In MM, all linguistic elements are represented in the lexicon with their phonological form, their morphosyntactic features and their semantics. The meaning of the clause results from the composition of its elements, and linking follows from the lexical representation of the elements. A lexical entry consists of its phonological representation and its categorial information. The lexical entries of the stems in (1a) and (1b) also contain their semantic representation and their argument structure and following from it, the theta raster. Lexical entries of aYxes in addition consist of their subcategorisation frame (or input condition, following the back slash) and their morphological features (or output information, preceding the back slash). For clitics, similar entries will be proposed in the following.
The place of morphology in grammar
(1) Lexical entries a. see: /si:/; [+V]; λy λx b. dog: /dfg/; [+N]; λx c. -s: /s/; [–1,–2,-pl] / [+V]; λV ∃ x<–1,–2,-pl>
SEE(x,y) DOG(x)
V(x)
Late-insertion models, in contrast, assume that terminal syntactic nodes are complexes of semantic and syntactic features but systematically lack phonological features. They are organised into hierarchical structures by the principles and operations of the syntax proper. Phonological features are supplied after the syntax and only after vocabulary items have been inserted into the terminal nodes. This insertion process does not add anything to the semantic and syntactic features making up the terminal nodes. The morphological component applies to the morphological and phonological features in late insertion models in diVerent ways than in early insertion models.2 Clitic analyses within the former frameworks will be discussed in detail in chapter 4. In the framework suggested here, rules, derivations and constraints do not apply within the lexicon. Instead, the lexicon supplies the material for input and output candidates in the following way. Lexical elements are mapped to the conceptual structure including discourse-referential features. The input of the utterance, clause, phrase or word consists of the lexical representation of its elements, that is, for verb phrases including clitics: the lexical representation of the verb, together with its arguments and their discourse-semantic and morphosyntactic information. To this input, the function GEN generates a set of output candidates by arbitrary combination of the input material with other language-speciWc lexical material, as for example agreement markers such as clitics and aYxes. All lexical elements may be combined with each other, since GEN is unrestricted. Moreover, GEN provides each output candidate with a correspondence relation to the input by co-indexing output and input elements. Each of the output candidates is then evaluated by the evaluation function consisting of the language-speciWc constraint ranking. This correspondence relation is checked by faithfulness constraints, which require identity of input and output elements (see chapter 1.3.3 for the basic formulation of faithfulness constraints). There are faithfulness constraints which require that certain input information corresponds to clitics, namely the max constraints in (2b), (3b), (4b). Each of these constraints is a speciWc case of a more general constraint which requires the respective information to occur in the output, regardless whether as aYx, clitic or morphologically free element (2a), (3a), (4a).
75
76
Clitics in the lexicon
(2) Constraints on argument roles a. max(arg) An argument role has an overt correspondent in the output. b. max(arg)M An argument role has a morphological correspondent (i.e. a clitic or an aYx) in the output. (3) Constraints on tense information a. max(tense) Tense information is overtly realised in the output. b. max(tense)M Tense information is morphologically realised in the output. (4) Constraints on negation a. max(neg) A negation is overtly realised in the output. b. max(neg)M A negation is morphologically realised in the output.
The relation between these speciWc and general max constraints will be discussed with respect to the argument clitics in the following. Additionally, it will be explained how agreement relations are analysed within the framework. Pronominal subject and object clitics, and locative and partitive clitics are linked to argument roles, as we have seen above. This results from the constraint max(arg)M which requires that argument roles are realised morphologically (2b). “arg” is used as a variable which is replaced by more speciWc information about the argument in question in the concrete tableaux below. “s” stands for the feature speciWcation of the subject, i.e. [–hr], “io” for the feature speciWcation of indirect objects, i.e. [+hr,+lr], “do” for the feature speciWcation of direct objects, i.e. [+hr], “loc” for local arguments etc. The Romance languages provide the lexical material for morphological argument marking either by clitics (for objects, local and partitive arguments and for some subjects in French and Piattino) or by aYxes (for subjects). Clitics and aYxes are chosen as the morphological means to express argument roles so as to avoid violations of max(arg)M. The constraint max(arg) covers the case of morphological argument marking but is also obeyed if the argument is realised by morphological free material, e.g. by a DP. The output candidate set which obeys the more speciWc constraint in (2b) is therefore a subset of the output candidate set which obeys the more general constraint in (2a). In this respect, the representations of the subject in the Italian examples in (5b.i-iv) are considered to be diVerent output candidates to the input in (5a), which consists of the lexical representation of the suppletive verb andare ‘go’ with its
The place of morphology in grammar
argument structure and information about the referential properties of the subject, namely that it is 1st person (the correspondence relations of input and output elements are indicated by arrows).3 (5) Subject realisation in Italian a. Input:
λx
go(x), x = speaker
λx
go(x), x = speaker
b. Output candidates: >
>
(i) inXectional subject:
>
> vad-o go–1sg
go(x), x = speaker
>
>
λx
>
(ii) syntactic and inXectional subject:
>
λx >
go(x), x = speaker
>
>
>
(iii) syntactic subject:
>
> io vad-o 1prn go–1sg
λx
go(x), x = speaker
(iv) no subject:
>
io vad 1prn go
>
vad go
Of the set of output candidates, only the most promising ones are considered here. These are: (i) the verb stem and the subject information added by the subject agreement aYx, (ii) the verb stem and the subject information added by the subject agreement aYx and a full free pronoun in addition, (iii) the verb stem and the subject information added by a full free pronoun and (iv) only the verb stem and no overt realisation of the subject at all. Output candidates such as va-i ‘go–2sg’, tu va-i ‘2prn go–2sg’ etc. violate high ranked identity or unification constraints respectively and are neglected for the moment (but we will come back to these constraint violations shortly below). The evaluation of these output candidates regarding the constraints in (2) is demonstrated in the tableau in (6).4 In addition to the phonological representation of the output candidates in (5b), the eVect of the application of the subject aYx
77
78
Clitics in the lexicon
and/or pronoun on the representation is shown. (6) The relation of max(arg) and max(arg)M in Italian λx go(x), x = speaker vado ∃ x<+1> go(x) b. io vado go(io) c. io vad go(io) d. vad λx go(x)
max(s)
max(s)M
a.
* *
*
Candidate (6a) does not violate any of the constraints. The morphosyntactic features of the subject agreement aYx are assigned as an index to the abstractor of the respective argument.5 In Italian, as in other pro-drop languages, the λ-abstractor is changed into the existential quantiWer by adding the agreement information. On the one hand, this operation still allows the argument role to be accessible for DPs. On the other hand, the argument role needs not necessarily to be saturated by an additional DP, as is the case with non pro-drop languages, where the λ-abstractor remains unchanged. The latter option is exempliWed by candidate (6b), where the argument role is no longer available. This candidate does not violate the constraints in (2), either. Candidate (6c) violates max(arg)M but not max(arg), since the subject is not realised morphologically, but only as a DP. Here the argument role is no longer available due to application of the DP. Candidate (6d), Wnally, has no eVect on the representation of the verb but neglects the input information [x = speaker] (this kind of faithfulness violation will be discussed in chapter 5). Since the subject is not realised morphologically, max(S)M is violated, and since it is not realised at all, max(S) is also violated. This candidate is thus the worst of those above. We can assume that the speciWc constraint max(arg)M is relevant in languages that have morphological material (i.e. clitics or aYxes) which correspond to the respective argument roles. If the general constraint max(arg) is violated by any candidate, the speciWc constraint is as well. That means that the general constraint max(S) is not relevant in these languages, since there is always a better candidate which only violates max(S)M. This constraint may then decide the competition. In languages such as English the lexicon does not
The place of morphology in grammar
provide the respective morphological material. Here, the relevance of the constraints is changed. Let us consider in this respect the tableau in (7), in which the input of (5a) above corresponds to English output representations. (7) The relation of max(arg) and max(arg)M in English
λx go(x), x = speaker
max(S)
max(S)M
a. I go go(I) b. go λx go(x)
* *
*
Since there is no lexical material to encode the subject morphologically, max(S)M is always violated. This constraint is therefore irrelevant to the decision on the optimal output candidate, while max(arg) distinguishes between better and worse candidates. In neither of these cases can the ranking of the constraints be Wxed. Two further remarks are necessary regarding morphological argument marking. First, let us reconsider the Italian example in (5) and (6). There are two candidates which do not violate the constraints in (2), namely (i) vado and (ii) io vado. Though both structures are grammatical in Italian, there is not only a diVerence in meaning but also a formal diVerence.6 The correspondence relation in vado is a one-to-one correlation, i.e. in the output, only the subject aYx corresponds to the subject argument role of the input. In io vado, on the other hand, two output elements, i.e. the subject agreement aYx and the full pronoun correspond to the same argument role. Such multiple correspondence relations lead to violation of a constraint. The correspondence relation is therefore restricted by *mc (suggested originally by Lamontagne & Rice 1995, for vowel coalescence). (8) *multiple Each argument role in the input has correspondence (*mc) only one correspondent in the output.
As will be discussed in chapter 5, violations of *mc are common, if it is lowerranked than max(arg)M and additional higher-ranked constraints require the occurrence of full DPs. The conclusion from the argumentation above is that *mc is always violated in languages which have agreement morphology. The second problem which occurs with structures like io vado is that the subject pronoun io and the subject agreement marker -o are both part of the output. With the correspondence relation outlined above, their features can thus not be compared, since only input-output correspondences are deWned.7
79
Clitics in the lexicon
For agreement relations, however, diVerent output candidates which correspond to the same input position must be comparable. This is possible because of the concept of “Transitive Correspondence”. (9) Transitive Correspondence: For each O1, O2 ∈ Output and I ∈ Input: If O1 and O2 correspond to the same I, O1 and O2 correspond to each other. Regarding the output candidate io vado from (5b) and (6b), for instance, Transitive Correspondence instantiates a correspondence relation of the pronoun io (O1) and the inXectional suYx -o (O2), because both output elements correspond to same input element. (10) Input: Output:
λx go(x), x = speaker io vado I: >
>
O1: io
x = speaker > >
80
> > O2: -o
The features of corresponding DPs and clitics or aYxes, and the features of input arguments and their output correspondents can then be checked by identity or unification constraints as deWned for morphosyntactic features and referential properties in the following. (11) a.
unif(mf) Corresponding (output) elements are compatible with respect to their morphosyntactic features.
b. unif(ref) Corresponding elements are compatible with respect to their referential properties.
unification is a weaker form of the CT-constraint identity. While identity requires that features are identical and is thus also violated if underspeciWed output candidates correspond to fully speciWed input elements, unification requires compatibility and thus allows for underspeciWed correspondents.8 UnderspeciWed lexical elements such as, for instance, subject clitics correspond to fully speciWed argument roles in the theta raster without causing violations of unification but only of identity. It is unif(mf) — together with Transitive Correspondence — which guarantees that candidates with surface agreement violations such as tu vado (12d) or io vai (12c) are
The place of morphology in grammar
outranked, while unif(ref) causes the above-mentioned output candidates (tu) vai (see 12a,b) to fail. (12) unification in the evaluation of agreement and argument saturation λx go(x), x = speaker a.
tu vai go(tu) b. vai ∃ x<+2> go(x) c. io vai ∃ x<+2>go(io) d. tu vado ∃ x<+1> go(tu) e. vado ∃ x<+1> go(x)
unif(ref) unif(mf)
max(s)
max(s)M
** * *
*
*
*
The various chapters of this dissertation are concerned with diVerent levels of evaluations which all take place simultaneously. This is illustrated in the grammar model outlined in (13). (13) The grammar model Conceptual structure and discourse-referential features (e.g. z = topic, y = Proper Noun, x = speaker)
⇔ ⇓
The Lexicon: Functional and non-functional morphemes e.g.: Verb with its AS (following LDG) λz λy λx V(x,y,z) [+hr] [+hr] [–hr] [–lr] [+lr] [+lr] phonological representation morphological, syntactic and semantic features (following the lexical representations developed in MM and LDG).
81
82
Clitics in the lexicon
Input string of linguistic elements provided with conceptual structure, discourse-referential features
⇓ GEN Free combination of lexical elements provided with Correspondence Relations (by co-indexing) for:
⇓
a. conceptual structure — semantic form, morphological and syntactic realisation b. semantic form — output morphological and syntactic realisation c. lexically underlying morphological elements — output realisations d. lexically underlying phonological elements — output realisations
⇓
Output InWnite set of candidates Sentences and their parts: e.g. Verb with clitics, Verb without clitics
⇓ Evaluation phonological, morphological, syntactic and semantic constraints faithfulness, alignment, markedness
Optimal Output Candidate
Romance clitic inventories
The clitic which is most faithful to each morphosyntactic feature complex of the input is chosen out of the lexical inventory. This is the clitic which occurs in isolation (see section 3.2). In addition to the faithfulness constraints, however, constraints on optimal output structures also play a role. Morphological and phonological alignment and markedness constraints are particularly relevant when it comes to clitic sequences in chapter 4. We will see there that high ranking of these constraints may cause deviations from the clitic choice in isolation, such as deletion and substitution of clitics. Morphosyntactic and prosodic alignment constraints are responsible for the placement of the clitics in the sentence as will be shown in chapter 5. These constraints usually do not cause substitution or deletion of clitics. Deviations in the phonological representation of clitics in combination with their hosts result from other prosodic constraints. Finally, constraints on the realisation of arguments regulate the occurrence of clitics in that they determine the representation of discourse-referential information. These constraints either belong to the faithfulness family or to the markedness family, which also may suppress the faithful representation of arguments by clitics. Let us now turn to the Romance clitic inventories.
3.2. Romance clitic inventories The verbal clitic inventories in the Romance languages resemble each other in many ways.9 Direct, indirect object and reXexive clitics are attested in all Romance languages. These “minimal” clitic inventories are supplemented by other clitics depending on regional diVerences. We Wnd the most remarkable diVerentiation between the eastern Romance language Romanian and the other western Romance languages. Romanian is the only Romance language that has auxiliary clitics. This may result from the inXuence of the other Balkan languages, where auxiliary clitics are rather usual. In the group of the western Romance languages, a general tendency to enlarge the verbal clitic inventory can be observed from west to east. While Spanish and Portuguese only have object and reXexive clitics, we also Wnd partitive and locative clitics in Italian, Piattino and French. Only French and Piattino have subject clitics. The latter two languages thus have the most diVerentiated clitic inventories. French, Wnally, has a clitic negation.
83
Clitics in the lexicon
t
Italian + locative + partitive
t
Portuguese Spanish
t
(14) Romance clitic inventories t
Piattino + subject
French +negation
indirect object direct object reXexive t
84
+ auxiliaries Romanian
Not only the syntactic function of clitics is similar in the diVerent Romance languages but also their phonological representation. The direct and indirect object clitic inventories in (15) and (16) illustrate this. (15) Direct object clitics EP me you him her us you them(m) them(f)
BP
ES
AS me te lo
o a nos
le la nous vous les les
P
I
R
al
mi ti lo
ma[ te îl o ne va[ îi
me ve i li
ci vi li
F
P
le
lui
ie
les
nous me vous ve leur ie
I mi ti gli le ci vi gli
vos
los/las
os as
F
los las
le
(16) Indirect object clitics EP to me to you to him to her to us to you to them
BP
ES
AS me te
lhe nos vos lhes
les
R îmi ît6 i îi ne va[ îi
The 1st person singular direct object clitic is me in Piattino (P), French (F), European and Argentinian Spanish (ES and AS) and European and Brazilian
Romance clitic inventories
Portuguese (EP and BP). The respective clitics in Italian (I) and Romanian (R), i.e. mi and ma[ , only diVer in their vowel quality.10 As the Wgure in (14) correctly suggests, the phonological representations of the Romanian clitics are quite distinct from the western Romance clitics. We Wnd only one match, namely le (f.pl.do), which is the same in Italian. Of the western Romance languages, Italian shows the most distinctive clitic inventory, but we Wnd three matches with clitics from the western Romance group, i.e. (i) lo (3rd person, sg.m.do), which is the same in Spanish, (ii) la (sg.f.do), which is the same in Spanish, French and Piattino, and (iii) le (f.pl.io), which is also attested in Spanish. As expected, the European and Argentinian varieties of Spanish and Portuguese hardly diVer at all in the phonological representation of their clitics. The subject clitics in French and Piattino, on the other hand, are remarkably dissimilar, as the table in (17) shows. Apart from te or tu (2sg.s), the clitics obviously originate from diVerent sources, namely personal and demonstrative pronouns in French, and deWnite articles in Piattino. (17) Subject clitics
I you he she we you they(m) they(f)
Piattino
French
[a]l te [a]l la [a]n – i li
je tu il elle nous/on vous ils elles
In French, the additional clitic on also fulWls the impersonal subject function, which is — besides nous — also used as 1pl.s (see 17). As to the other clitics (see 18), we Wnd that the reXexive clitic is nearly identical in the western Romance languages, i.e. si in Italian and Piattino and se in French, Spanish and Portuguese. Romanian is the only Romance language which distinguishes direct and indirect object reXexive clitics, namely se and îs¸i The partitive clitic is ne in Italian and Piattino, and en in French. The locative clitic is ci or vi in Italian and Piattino, and y in French. The negation clitic is ne in French.11
85
86
Clitics in the lexicon
(18) Other clitics EP 3rfl io 3rfl do loc prt neg
BP
ES
AS
F
P
se – – –
– – –
– – –
I
R
–
îs6 i se – – –
si – – –
y en ne
ci ne –
That the clitic inventories are nearly identical does not mean that the lexical entries are nearly identical as well, since they are determined by syncretism and paradigmatic opposition within the language-speciWc clitic paradigm. Let us now turn to the choice of clitics in isolation, i.e. the most faithful realisation of arguments by clitics. In this respect, the analysis suggested here resembles Grimshaw’s (1997) Optimality-theoretical analysis of Romance clitics in many ways. In particular, the question of how clitic lexicons are represented and how the optimal clitic output candidate to a given input is chosen is answered in very similar ways. Therefore, Grimshaw’s analysis will be discussed before I present my own analysis in sections 3.2.2 to 3.2.7. I will conWne the discussion to her analysis of Italian clitics.
3.2.1 Grimshaw’s Optimality-theoretical analysis of Italian clitics Similar to the analysis suggested below but diVerent from the late-insertion models mentioned above, Grimshaw provides clitics with underspeciWed lexical entries. The feature speciWcations in her lexical entries are thus to a certain degree identical to the ones suggested below, but there are also some diVerences. Before we discuss the diVerences between the two analyses let us consider Grimshaw’s proposal in more detail. Grimshaw argues that clitics bear the features Cl[itic], R[eXexivity], P[erson], N[umber], G[ender], C[ase]. The feature Cl is part of all clitics but the other features may be underspeciWed, marked in Grimshaw’s notation by brackets: (R), (P), (N), (G), (C). UnderspeciWcation regarding the features R, G and C is assumed for 1st and 2nd person clitics, while 3rd person clitics only show underspeciWcation with respect to the feature (P). The feature R is marked with [-] values in 3rd person clitics, but unmarked in all other clitics though they can all be used in reXexive contexts. This seems to be somewhat peculiar, since even the reXexive clitic is
Romance clitic inventories
underspeciWed for R and in fact underspeciWed for all other features as well.12 The Italian clitic lexicon as suggested by Grimshaw is shown in (19). Grimshaw does not give an analysis of the 3rd person plural clitic gli. Furthermore, the Italian partitive clitic ne and the locative clitics ci and vi (which are homophonous to 1pl.o and 2pl.o) are not part of Grimshaw’s clitic lexicon. (19) The Italian clitic lexicon (Grimshaw 1997:172) si mi ti ci vi lo/la li/le gli/le
(R) (R) (R) (R) (R) –R –R –R
(P) 1 2 1 2 (P) (P) (P)
(N) sg sg pl pl sg pl sg
(G) (G) (G) (G) (G) m/f m/f m/f
(C) (C) (C) (C) (C) acc acc dat
self me, to me you, to you us, to us you, to you him, her, it them to him, her, it
Grimshaw suggests that faithfulness constraints like parse and fill exist for each of the features mentioned above.13 The ranking of these constraints is eventually responsible for the clitic choice. The input consists of morphosyntactic features, presumably given by the syntactic representation. Pairs of inputs and optimal output candidates together with the constraint rankings responsible for the clitic choice are shown in (20) (Grimshaw neglects gender information for the sake of simplicity here). The diVerence between input and output representations consists mainly in their feature speciWcation: input representations given by the syntax are fully speciWed. Therefore parse violations follow whenever an underspeciWed clitic is chosen. (20) The best clitic Input
optimal violated
a. [-R 3 pl acc] b. [R 2 pl acc]
li/le vi
c. [R 3 pl acc]
si
d. [3 pl]
si
parse p parse r, parse c parse r, parse p parse c parse p, parse n
crucial constraint rankings
2nd-best
no evidence for any ranking fill r >> parse c, parse g
lo/la li/le
fill p >> parse n
mi, ti, ci vi li, le ci, vi, li, le
fill r >> parse n,parse c fill p >>parse n
87
88
Clitics in the lexicon
The simplest case is an input representation consisting of non-reXexivity, 3rd person plural, accusative (20a). All competing candidates — that is, the clitics from the clitic lexicon shown in (19) — violate at least parse P: 3rd person clitics do so, because their representation does not contain any person speciWcation at all, and 1st and 2nd person clitics do so, because their representation contains the wrong person speciWcation, which means that fill P is also violated.14 Since all candidates other than li/le violate more constraints in addition, they are worse. For this reason, no constraint ranking can be derived here. Grimshaw’s second example has the input representation consisting of the features reXexivity, 2nd person plural and accusative (20b). In this case, all competing candidates violate parse R, and the optimal candidate vi also violates parse C since it is underspeciWed for case. Only the 3rd person accusative clitics do not violate parse C but violate fill R due to their [-R] feature speciWcation. The singular clitics are outranked because of additional parse N and fill N violations. Therefore, only the plural clitics remain in competition with the optimal candidate. In this case, the ranking fill R >> parse C is crucial for the choice of the optimal output. The next example that Grimshaw gives has the input consisting of the features reXexivity, 3rd person plural, accusative (20c). Even here, the constraints parse R and parse P are violated by all competing output candidates. The optimal candidate also violates parse C and parse N, which are also violated by all 1st and 2nd person and by all singular clitics, respectively. Singular clitics are outranked because of additional violation of fill N. Since the only diVerence between 1st and 2nd person plural clitics and the optimal candidate si is the violation of fill P in contrast to parse N, the constraint ranking must be fill P >> parse N in order to outrank the second-best candidates. This ranking, however, does not cause the 3rd person plural clitics to be as bad since these clitics violate neither fill P nor parse N. Instead, they violate fill R in addition to parse R and parse P. Therefore, this constraint must be ranked higher than parse N and parse C, which are violated by the optimal candidate. Grimshaw’s last example has the input for the impersonal si, namely the syntactic feature speciWcation 3rd person plural.15 All competing candidates violate parse P, and the optimal candidate violates parse N in addition. Singular clitics are outranked due to additional violation of fill N, so that only plural clitics remain which do not violate parse N but either fill P or fill R. The rankings must therefore be fill P >> parse N and fill R >> parse N. As to the concept of underspeciWcation, there are slight diVerences between MM and Grimshaw’s term. One might recall that in MM, only binary
Romance clitic inventories
features are allowed and only [+] valued features should be part of the lexical entry, while [-] values are added by default. Number speciWcation, for example, is speciWed by [+pl] for plural clitics. Singular clitics do not have a number speciWcation in their lexical entry but are interpreted as [pl] by default. The eVect on the analysis then is that there would be only a parse N but no fill N violation if a singular clitic replaces a plural clitic. Other agreement features are handled similarly: person is (as in Grimshaw’s model) marked by [±1] and [±2] and gender (unlike Grimshaw proposes) by [±f]. With respect to case, I will use the features developed in LDG as introduced in chapter 1.3.2. Therefore, [+hr] is part of the lexical representation of all object clitics, since the subject is a higher argument and [+lr] is only part of indirect object clitics, since the direct object is a lower argument. Both features may remain underspeciWed. Lexically underspeciWed forms like 1st and 2nd person clitics show syncretism, that is, the ability to Wll more than one cell within the paradigm, which is considered the interface of morphology and syntax. However, there are also other diVerences between Grimshaw’s and my own approach. In particular, the restrictions regarding the combination and linearisation of clitics and the substitution or deletion of clitics in clitic sequences (as far as discussed by Grimshaw) diVer in many respects. Clitic sequences will be discussed in detail in chapter 4.
3.2.2. European and Brazilian Portuguese clitics The clitic inventories of European and Brazilian Portuguese are identical, therefore both are listed together in (21). Portuguese has only object and reXexive clitics. (21) The Portuguese clitic inventory
1 sg 2 sg 3 sg.m 3 sg.f 1 pl 2 pl 3 pl.m 3 pl.f
direct object
indirect object
reXexive
me te o a nos vos os as
me te lhe lhe nos vos lhes lhes
me te se se nos vos se se
89
90
Clitics in the lexicon
A representation of the lexical inventory such as that given in (21), however, does not reXect the economy of lexical storage. 1st and 2nd person direct and indirect object and reXexive clitics, for instance, are identical, while 3rd person clitics show more diVerentiation: (i) indirect and direct object clitics diVer from each other; (ii) direct object clitics diVer regarding gender; and (iii) there is a special reXexive clitic. MM aims to account for such syncretism and contrast by means of underspeciWcation. In MM, a diVerence is made between systematic and non-systematic syncretism. The former is assumed to be the result of underspeciWcation in the lexicon (i.e. only one ambiguous lexical entry exists for diVerent functions), while the latter is not reXected in the lexical entries (i.e. there are diVerent lexical entries for diVerent functions). An MM paradigm reXects this diVerence in that there are diVerent cells for unsystematic syncretism but the same cell across diVerent dimensions for systematic syncretism. The paradigm and the lexical entries are directly related: lexical underspeciWcation of the features that build the dimensions of the paradigm leads to systematic syncretism. With respect to the Portuguese clitic inventory above, only systematic syncretism is observed. For 1st and 2nd person clitics, the dimensions number (with the values [±pl]) and person (with the values [+1] and [+2]) are relevant. Other agreement features as, for example, gender are not distinguished in 1st and 2nd person. Furthermore, 1st and 2nd person clitics never occur as subjects but only as direct or indirect objects, or as reXexives. Their phonological representation is not distinct with respect to the latter functions. Therefore, only the case feature [+hr] is relevant for them. 1st and 2nd person clitics are adequately described by the paradigm in (22), which consists of the dimensions person, number and case. For mnemonic reasons, the LDG-feature [+hr] is referred to as “o” and the agreement features [-pl], [+pl], [+1], [+2], [–1,–2] are referred to as “sg”, “pl”, “1”, “2”, “3” in the paradigms and the text.16 (22) Paradigm for 1st and 2nd person clitics o 1 2
sg
pl
me te
nos vos
For 3rd person clitics, the dimensions case and reXexivity are also distin-
Romance clitic inventories
guished. One can therefore argue that 3rd person clitics build up a more diVerentiated paradigm with the dimensions [±lr] and [±rX], which have to be integrated into the paradigm above. This subparadigm Wlls the cells 3o and 3pl.o, that is, the cells with the feature values [–1,–2,-pl] and [–1,–2,+pl]. In (23), the feature combinations [-rX,+hr,–lr], [–rX,+hr,+lr] and [+rX,+hr,-lr] are referred to as “do”, “io” and “rfl”. Moreover, 3do distinguishes gender, so that the cells 3do and 3pl.do (with the feature values [–1,–2,±pl,–rX,+hr]) consist of a subparadigm with the additional dimension gender (i.e. [±f] referred to as “m” and “f” in 23). (23) Integration of subparadigms for 3rd person clitics o sg me
1 2 3
do m o f a
pl nos
te io
rfl
lhe
se
do m os f as
vos io
rfl
lhes
se
The dimensions of the subparadigms can be integrated into the general paradigm with the result that clitics that are unspeciWed with respect to a certain dimension Wll more than one cell along this dimension as, for instance, 3io does with respect to gender or 1o with respect to case and reXexivity (see 24). (24) Portuguese object clitics sg
do 1 2 3m 3f
io
pl
rfl
do
me te o a
lhe
io
rfl
nos vos se
os as
lhes
se
The lexical entries in (25) reXect these syncretisms: ambiguous 1o and 2o, for instance, are unspeciWed with respect to indirect object case (i.e. [±lr]), reXexivity and gender (see 25a-d below), so that they Wt into the diVerent morphosyntactic contexts. 3rd person clitics, on the other hand, are speciWed
91
92
Clitics in the lexicon
with respect to these features. Regarding case, the indirect object clitics lhe and lhes bear the feature [+lr] in addition to the feature [+hr] (25i,j). Hence, they are not ambiguous but restricted to indirect object positions. As to 3do clitics, they are thus not ambiguous, either, but restricted to direct object positions (25e,f). Additionally, 3do clitics are speciWed with respect to gender. For wellknown reasons (i.e. masculine gender refers to mixed plurals etc.), I regard feminine as the marked gender, so that only the feature [+f] is part of the lexical entries. (25) Lexical entries for Portuguese clitics: +D; [-max,-min] [ a. me +1,+hr b. nos +1,+pl,+hr c. te +2,+hr d. vos +2,+pl,+hr e. o +hr f. os +pl,+hr g. a +f,+hr h. as +f,+pl,+hr i. lhe +hr,+lr j. lhes +pl,+hr,+lr k. se +rX
]/+V
It is worth noting that the clitic paradigms of all examined languages show a similar behaviour: we Wnd case syncretism in the 1st and the 2nd person and case distinction in the 3rd person (but see section 3.2.7 for Romanian). Gender distinction is found with the 3rd person direct object but neither with the 1st or the 2nd person nor with the 3rd person indirect object. This kind of restriction can be conceived as universal markedness constraints which prohibit that the features [+lr] and [+f] co-occur (i.e. *[+f]/[+lr]) and that they occur in the same feature structure as [+1], [+2] (e.g. *[+lr]/[+1]) respectively. (See Stiebels 2000a,b and Wunderlich 2001a,b for a detailed discussion of such markedness constraints.) For the time being, all clitics in (25) have the categorial feature [+D], which as a premise restricts their referential potential to [+deWnite] entities. (Details about this and other reference restricting features of clitics are discussed in chapter 5.) Furthermore, all clitics have the features [–max,–min]. These features were originally proposed by Stiebels & Wunderlich (1994) to account for the diVerence between morphologically free elements which can be projected into the syntax (i.e. words with the feature [+max]) and morphologically bound elements (i.e. stems and aYxes with the feature [+min]). Clitics, as has been shown in chapter 2, are neither words nor aYxes. Therefore, they are assigned negative values for both features. Finally, all clitics in (25) are subcategorised by the input condition [+V], which means that they
Romance clitic inventories
attach to verbs. (A thorough discussion and analysis of clitic placement and especially the alternation of pre- and postverbal clisis is the issue of chapter 5.) The lexical entry of the reXexive clitic se in (25) is peculiar: obviously, it has no case feature but only the feature [+rX]. This is due to the fact that the reXexive pronoun may occur in so-called “reXexive middle constructions”, which are characteristic for Romance languages. ReXexive middle constructions denote a group of diVerent constructions all including the reXexive clitic (for a detailed discussion see Kaufmann 2000; Kemmer 1993; Klaiman 1988 a.o.; on similar functions of reXexives particularly in West-Germanic see Abraham 1995). In Portuguese, the reXexive clitic se in combination with 3rd person active verbs indicates such a reXexive middle construction, translated into English as an impersonal passive construction as, for instance, in (26). (26) Contam-se muitas histórias sobre o velho castelo. tell.3pl-rfl many stories about d.df old castle ‘Many stories are told about the old castle.’ (Palma Caetano et al. 1986:50)
The capability of the reXexive clitic to bind the subject suggests that the reXexive clitic se is not assigned the feature [+hr] (25k). UnderspeciWcation with respect to this feature does not mean that the reXexive clitic cannot Wt into contexts which are assigned this feature. Therefore, it may also occupy the respective cells in the paradigms above. 1st or 2nd person clitics never occur as impersonal clitics, so there is no conXict for these clitics regarding the feature [+hr]. The lexical entries and the constraints are both relevant when it comes to the choice of clitics. On the one hand, faithfulness constraints like max(arg)M (see above, on p. 97) require that each input argument has a clitic correspondent in the output.17 Moreover, max(+feature) constraints require that each of the ([+valued]-)features in the input has a correspondent in the output. Markedness constraints are in conXict with these constraints for each of the [+valued] features (*(+feature)). Stiebels (2000a,b) discusses this conXict in detail and shows that the choice of optimal linkers takes place in a most elegant way. Presumably, the max constraints are higher-ranked than the markedness constraints. Let us now look at some examples which illustrate the interaction of faithfulness and markedness constraints as it comes to the evaluation of the lexical inventory. To begin with, the most underspeciWed entry, the 3do clitic, is chosen if the argument role in the input contains the feature [+hr] as the only [+]-valued feature (see 27).18 The only relevant max
93
94
Clitics in the lexicon
constraint is therefore max(+hr), which is violated merely by the reXexive clitic. All clitics which bear [+]-values of features that are [-]-valued in the input induce further constraint violations of the unification(feature) type. These constraints require that the features of the correspondents are compatible and most probably rank rather high. They are never violated in the optimal candidates if single clitics are evaluated.19 In (27), diVerent violations of this type are subsumed by the constraint unif(feature), which is on top of the ranking. Candidate (27a), for instance, violates unif(1), candidate (27f) violates unif(feminine) and unif(plural) and candidate (27h) violates unif(lr) and unif(plural). As regards the markedness constraints, all clitics have the feature [+hr]. They therefore violate *(+hr). Only the optimal 3do as the most unspeciWc clitic (i.e. candidate 27c) does not violate any other constraint. The number of markedness violations which the clitics induce increases with the number of additional features they bear. The violations of the markedness constraints remain the same to each input. In the following, these constraints are neglected as they do not contribute to the evaluation. (27) Input:3rd person direct object [–1,–2,–f,–pl, +hr,lr,-rX]
unif max * * * (F) (+hr) (+hr) (+lr) (+1)
a.
me [+1,+hr]
*!
*
*
b.
nos [+1,+pl,+hr]
**!
*
*
c.
o
[+hr]
* * * (+f) (+pl) (+rX) *
*
d.
os [+hr,+pl]
*!
*
e.
a
[+hr,+f]
*!
*
*
f.
as [+hr,+f,+pl]
**!
*
*
g.
lhe [+hr,+lr]
*!
*
*
h.
lhes [+hr,+lr,+pl]
**!
*
*
i.
se [+rX]
*!
*
* * *
*
The more speciWc the input is, the more relevant are the max constraints. If the direct object, for instance, has the additional features [+feminine] and [+plural], max(+f) and max(+pl) take part in the evaluation (see 28). Only 3f.do (candidate 28d) does not violate any of the max constraints and thus comes out as optimal.
Romance clitic inventories
(28) Input:3rd person feminine plural direct object [–1,–2,+f,+pl,–rX,+hr,–lr] unif(F) a. b. c. d. e. f. g.
o os a as lhe lhes se
[+hr] [+hr,+pl] [+hr,+f] [+hr,+f,+pl] [+hr,+lr] [+hr,+lr,+pl] [+rX]
max(+f) max(+pl) *! *!
max(+hr)
* *!
*! **! *!
* * *
* *
*
So far, the max constraints have not been ranked. A ranking is necessary if no fully speciWed clitic exists to a speciWc output and the optimal candidate is chosen out of a number of clitics which are underspeciWed with respect to diVerent features. If, for instance, the input is a 3rd person feminine plural indirect object (see 29), there is no clitic in the inventory which bears the feature [+lr] as well as the feature [+f]. Consequently, a decision must be made whether the feature [+lr] or the feature [+f] should be realised in the output. In Portuguese, the former option is chosen: 3pl.io (candidate 29f) is optimal, while 3f.pl.do (candidate 29d) is outranked. Therefore, the constraint max(+lr) must rank higher than the constraint max(+f). The rank of max(+pl) and max(+hr) cannot be established in this evaluation. This is reXected by the double line to the left of max(+pl). (29) Input:3rd person indirect object feminine [–1,–2,+f,+pl, -rX,+hr,+lr] a. o b. os c. a d. as e. lhe f. lhes g. se
[+hr] [+hr,+pl] [+hr,+f] [+hr,+f,+pl] [+hr,+lr] [+hr,+lr,+pl] [+rX]
unif(F)
*!
max (+lr)
max (+f)
max (+pl)
*! *! *! *!
* *
*
*
max (+hr)
* * * *
*! *
*
max(+lr) can also be violated. Consider in this respect the tableau in (30). Here, the input consists of the 3rd person feminine plural reXexive indirect
95
96
Clitics in the lexicon
object. The underspeciWed reXexive clitic is chosen in this context, even if the features [+hr], [+lr], [+f] and [+pl] are not part of the output and violations of the respective max constraint are admitted. (30) Input 3rd person feminine plural reXexive indirect object [–1,–2,+f,+pl, +rX,+hr,+lr] a. o b. os c. a d. as e. lhe f. lhes g. se
unif(F)
[+hr] [+hr,+pl] [+hr,+f] [+hr,+f,+pl] [+hr,+lr] [+hr,+lr,+pl] [+rX]
max (+rX) *! *! *! *! *! *!
max (+lr)
max (+f)
max (+pl)
* * * *
* *
*
max (+hr)
* * * *
*
* *
*
Other features, such as the person features [+1] and [+2] must not be underspeciWed in the output. As the tableaux in (31) and (32) show, max(+1) — and analogously max(+2) — rank not only higher than max(+f) and max(+lr), but also higher than max(rX). (31) Input:1st person indirect object feminine plural [–1,–2,+f,+pl, +rX,+hr,+lr] a. me b. nos c. o d. os e. a f. as g. lhe h. lhes i. se
[+1,+hr] [+1,+pl,+hr] [+hr] [+hr,+pl] [+hr,+f] [+hr,+f,+pl] [+hr,+lr] [+hr,+lr,+pl] [+rX]
unif(rX) max (+1)
*!
*! *! *! *! *! *! *!
max (+lr)
max (+f)
max (+pl)
* * * * * *
* * * *
*!
*
max (+hr)
* *
* * *
* *
*
Romance clitic inventories
(32) Input:1st person feminine plural reXexive indirect object [–1,–2,+f,+pl, +rX,+hr,+lr] a. me b. nos c. o d. os e. a f. as g. lhe h. lhes i. se
[+1,+hr] [+1,+pl,+hr] [+hr] [+hr,+pl] [+hr,+f] [+hr,+f,+pl] [+hr,+lr] [+hr,+lr,+pl] [+rX]
unif max (rX) (+1)
*! *! *! *! *! *! *!
max (+lr)
max (+1r)
max (+f)
* * * * * * * *
* * * * * *
* * * *
*
max max (+pl) (+hr) *! * *
* * *
* *
*
So far, the partial rankings in (33), together with the underspeciWed lexical entries in (24) are suYcient to predict the choice of argument clitics in Portuguese. In general, faithfulness constraints (max(feature)) rank higher than markedness constraints (*(feature)). For the latter, no ranking could be derived, while the former are ranked below unif(F), as shown in (33). As to max(+hr) and max(+pl), their position within the ranking could not be Wxed. (33) Portuguese: Ranking of morphological max(F) and unif(F) constraints unif(F) >> max(+1), max(+2) >> max(+rX) >> max(+lr) >> max(+f) .... max(+pl) .... max(+hr)
Regarding the clitic inventories the context-sensitive markedness constraints *[+f]/[+lr], *[+f]/[+1] v [+2] and *[+lr]/[+1] v [+2] are also relevant. These constraint restrict the inventory by prohibiting grammatical distinctions where they are not necessary. As mentioned above, the European and Brazilian varieties of Portuguese do not diVer in their clitic inventories. In contrast, we Wnd some diVerences between European and Argentinian Spanish, though both varieties show certain similarities to the Portuguese clitic lexicon discussed above.
3.2.3 European and Argentinian Spanish clitics In European Spanish, only 2pl.o and 3do have a diVerent phonological representation to the corresponding clitics in Portuguese (ES os instead of EP/AP
97
98
Clitics in the lexicon
vos and EP lo, la, los, las instead of o, a, os, as).20 3ios diVer in their orthographic but not in their phonological representation. Furthermore, we Wnd the same syncretism as in Portuguese: (34) European Spanish object clitics sg do 1 2 3m 3f
io
pl rfl
do
me te lo la
le
io
rfl
nos os se
los las
les
se
We can therefore conclude that the lexical entries have exactly the same properties regarding underspeciWcation in European Spanish as in Portuguese (see 35). (35) Lexical entries for European Spanish clitics: +D; [-max, -min] [ ]/+V a. c. e. g. i. k.
me te lo la le se
+1,+hr +2,+hr +hr +f,+hr +hr,+lr +rX
b. d. f. h. j.
nos os los las les
+1,+pl,+hr +2,+pl,+hr +pl,+hr +pl,+f,+hr +pl,+hr,+lr
Rio de la Plata Spanish, in contrast, cannot be analysed in the same way, since a further syncretism must be observed. In the plural paradigm, 2pl.o and 3pl.do/3pl.io are not diVerent anymore but are both represented by former 3pl.do and 3pl.io. This neutralisation regarding person leads to a diVerentiation with respect to case and gender in the 2nd person plural. Although syncretism along the person dimension is rather rare, it is unlikely that this syncretism is accidental, since only one dimension is concerned. Therefore, we can still correctly reXect this syncretism by merging adjacent cells in the paradigm. The neutralisation of person features is represented in (36) by means of merging the cells [+2,+pl] and [–1,–2,+pl] so that in the plural paradigm the remaining contrast with respect to person is [+1] and [–1]. The additional diVerentiation with respect to case is represented by dividing the resulting cell into the three distinctive cells do, io and rfl. The additional
Romance clitic inventories
gender distinction for 2/3.pl.do is integrated as a subparadigm. (36) Argentinean Spanish object clitics sg do
io
1 2 3
pl rfl
do
me te m f
lo la
io
rfl
nos le
se
m
los
f
las
les
se
For the lexical entries, the question arises how this person syncretism can be captured and how it inXuences the rest of the lexical inventory. Theoretically, there are two options for integrating this syncretism, either by disjunction of features within the lexical entries or by underspeciWcation of features. Let us Wrst consider three disjunctive lexical entries for 2/3pl.do as an example (see 37). (37) Disjunctive lexical entry for 2/3pl.do a. los : [+2,+pl,+hr] v [+pl,+hr] b. los : [(+2 v –1,–2),+pl,+hr] c. los : [(+2 v –2),+pl,+hr]
At Wrst glance, the lexical entry in (37a) is the least attractive one. Both structures share all features except [+2], so that the entry becomes more complex than it needs to be. Consequently, the disjunction is reduced in the entry in (37b) to exactly the disjunctive part of the structures, namely the person features. However, given the reasoning of the relation of paradigm structures and lexical entries in MM, the speciWcity principle ensures that 1pl.o nos with the feature [+1] has precedence over less speciWc clitics in any context compatible with [+1]. Therefore, the feature [–1] in the second conjunct of (37b) is superXuous. The entry in (37c) with only the features [+2] and [–2] in its disjunctive part lacks this fault. Yet, there are three reasons to reject this entry. First, it is questionable whether the feature [+2] has any distinctive function at all in the plural paradigm of Argentinian Spanish clitics, because there is no opposition of [+2] and [–2] but only of [+1] and [–1]. Second, there is no reason why [-] valued agreement features should not be avoided in disjunctive lexical entries as they are in non-disjunctive lexical
99
100 Clitics in the lexicon
entries. Third, the disjunction (+2 v –2) formally consists of a tautology, which means that the lexical entry is as informative with it as it is without it. Leaving the uninformative disjunction out leads to the underspeciWed lexical entry in (38). (38) UnderspeciWed lexical entry for 2/3pl.do: los :
+pl,+hr
Disjunctive lexical entries are questionable from a general point of view, too. If we have entries like the one in (37a), it is not clear whether we can really speak of one and the same entry or whether the disjunction is only a formal option to unite two diVerent but accidentally homophonous entries. Moreover, the complex structure of disjunction may be problematic regarding language processing, so that two simple entries may be more accessible than one complex disjunctive one (Stiebels 2000b integrates this fact as a markedness constraint which prohibits disjunctive entries: *[v]). Disjunctive lexical entries must therefore be rejected in an economy-driven grammar model if there are other ways of describing the respective phenomenon. To explain the coincidence of 2nd and 3rd person plural discussed above, the underspeciWed entry in (38) has been shown to be the most attractive solution. Analogously, the entries for 2/3f.pl.do, 2/3pl.io and the reXexive pronoun remain underspeciWed and thus do not diVer from the respective entries in European Spanish and Portuguese (see 18d-j). (39) Lexical entries for Argentinian Spanish clitics: +D; [-max, -min][ ]\+V a. c. d. f. h. j.
me te lo la le se
+1,+hr +2,+hr +hr +f,+hr +hr,+lr +rX
b.
nos
+1,+pl,+hr
e. g. i.
los las les
+pl,+hr +f,+pl,+hr +pl,+hr,+lr
The diVerence between the paradigms in (34) and (36), however, is not suYciently captured by simply not providing an entry for 2pl.o. If a dimension [F] is part of the singular paradigm, this dimension should also be part of the plural paradigm. The question that has to be answered then is how the cell with the features [+F,+pl] should be Wlled if no speciWc lexical entity bears these features. There are two ways to Wll the particular cell (i.e. [+2,+pl]) in the
Romance clitic inventories 101
paradigm, both of which rely on underspeciWcation. Either [pl] remains unspeciWed and the feature [+2] is speciWed (i.e. the cell is Wlled by te) or [2] remains unspeciWed and the feature [+pl] is speciWed (i.e. the option which is actually chosen).21 Again, it is the ranking of max(feature) constraints and in particular the high rank of max(+pl) which determines the clitic choice here.22 The tableau in (40) illustrates this for a 2nd person plural direct object in the input. Since max(+pl) ranks above max(+2), los (candidate 40d) is preferred to te (candidate 40a) as optimal output correspondent. Other plural clitics such as nos or les are worse because of their additional features [+1] and [+lr] respectively. These features produce violations of the unif(F) and *(feature) constraints as illustrated for nos (candidate 40b). (40) Input:2nd person plural direct object [–1,+2,+pl,+hr,-lr] a. b. c. d.
te nos lo los
[+2,+hr] [+1,+pl,+hr] [+hr] [+pl,+hr]
unif(1)
max(+pl)
max(+2) *(+1)
*! *!
*
* * *
*!
Other max constraints, such as max(+f) and max(+lr), retain the same ranking as demonstrated above for Portuguese (see 41 and 42). (41) Input:2nd person feminine plural direct object [–1,+2,+f,+pl,+hr,-lr] a. te b. las c. la d. los
max(+pl)
[+2,+hr] [+f,+pl,+hr] [+f,+hr] [+pl,+hr]
max(+2)
max(+f)
*!
* * * *
*!
*!
(42) Input:2nd person feminine plural indirect object [–1,+2,+f,+pl,+hr,+lr] a. te b. las c. les d. los
[+2,+hr] [+f,+pl,+hr] [+pl,+hr,+lr] [+pl,+hr]
max(+pl) max(+2) max(+lr) max(+f) *! * * *
* *! *!
* * *
102 Clitics in the lexicon
The Argentinian Spanish ranking of the morphological constraints which induce the clitic choice can then be summed up as shown in (43). (43) Argentinian Spanish: morphological max(feature) and *(feature) constraints a. max(+1), max(+pl) >> max(+2) >> max(+rX) >> max(+lr) >> max(+f) >> *(feature) b. max(+hr) >> *(feature)
In the following section, we will turn to the Italian clitic inventory which is much richer than the Spanish and Portuguese ones.
3.2.4 Standard Italian clitics The Italian object clitic paradigm (44) shows a number of diVerences in comparison to the Portuguese and Spanish clitic inventory (24, 34 and 36 above). (44) a.
Standard Italian object clitics sg do
1 2 3m 3f
lo la
io mi ti gli le
pl rfl
do
io
rfl
ci vi si
li le
gli
si
b. Additional clitics ci, vi : ‘locative’ ne: ‘partitive’
With regard to the phonological form of the clitics, we Wnd that the vowels are generally raised, so that me, te, se in Spanish and Portuguese correspond to mi, ti, si in Italian. Furthermore, Italian clitics follow the Italian default syllable structure CV, (cf. Italian vi and Spanish and Portuguese vos). 1pl.o ci has a complex onset consisting of the aVricate [tw ] instead of the unmarked nasal [n] in the Spanish and Portuguese counterpart nos. Therefore, there is no longer any obvious phonological similarity. The lexical entries are shown in (45).
Romance clitic inventories 103
(45) Lexical entries for Standard Italian object clitics: +D[-max, -min] [ ]/ +V a. c. e. g. i. k. l.
mi ti lo la gli1 le2 si
+1,+hr +2,+hr +hr +f,+hr +hr,+lr +f,+hr,+lr +rX
b. d. f. h. j.
ci vi li le1 gli2
+1,+pl,+hr +2,+pl,+hr +pl,+hr +f,+pl,+hr +pl,+hr,+lr
With regard to syncretism, the opposition of 3io gli [ˆi] and 3f.io le leads to gender diVerentiation within the cell [–1,–2,-pl,+hr,+lr], which is not attested in Spanish and Portuguese. In the singular, max(+f) is therefore no longer violated. On the other hand, gli also Wlls the cell 3pl.io (i.e. [–1,–2,+pl, +hr,+lr]) for both genders so that another syncretism arises. Neither that syncretism nor the occurrence of le as 3f.io (45h) and 3f.pl.do (45k) can be captured by underspeciWcation and must thus be considered unsystematic syncretism for the following reasons. If there was only one entry for le, this entry would have to be compatible with the feature combination [-pl,+lr] (so that le could occur as an indirect object in singular) and compatible with the feature combination [+pl,–lr] (so that le could occur as direct object in plural) in order to avoid unif(F) violations. The entry must therefore be unspeciWed with regard to the features [lr] and [pl], i.e. must only contain the features [+f,+hr]. Consequently, le would contain the same features as la and no decision on either candidate could be achieved. Therefore, le must contain the additional feature [+lr] to be able to Wll the respective feminine indirect object cell in singular. Provided with this feature, however, le can no longer Wll the feminine plural direct object cell with the feature combination [+pl,+hr,–lr], and therefore we must assume that both occurrences of le do not go back to a systematic syncretism with one underspeciWed lexical entry, but are coincidentally homophonous. The diVerent occurrences of gli cannot be captured by one underspeciWed lexical entry, either. This becomes clear from the tableaux in (46) and (47). Let us Wrst consider the unproblematic case in (46) where the input is a 3rd person indirect object. le1 and le2 violate unif(f) and unif(pl) and lo violates max(+lr). Therefore, these candidates are outranked, but gli does not violate any of these constraints and is thus optimal.
104 Clitics in the lexicon
(46) Input:3rd person indirect object [–1,–2,–f,–pl,+hr,+lr] unif(F) max(+hr) max(+lr) a. le1 b. le2 c. gli e. lo
[+f,+pl,+hr] [+f,+hr,+lr] [+hr,+lr] [+hr]
max(+f)
* *! *! *!
If the input consists of a 3rd person feminine plural indirect object as in (47), however, le2 incorrectly results as optimal. UnderspeciWed gli lacks the features [+pl] and [+f], so that any ranking of the max(+feature) constraints results in two violations for gli but only one violation for le1 and le2 respectively.23 (47) Input:3rd person feminine plural indirect object [–1,–2,+f,+pl,+hr,+lr] a. b. c. e.
le1 le2 gli li
unif(F) max (+hr)
[+f,+pl,+hr] [+f,+hr,+lr] [+hr,+lr] [+pl,+hr]
max (+lr) *
max (+pl) * *
max (+f)
*! *!
*
Therefore, we must assume that an additional lexical entry for gli exists, which contains the feature [+pl]. We can see then that, as in Portuguese and Spanish, max(+lr) also ranks above max(+f) (see 48). (48) Input:3rd person feminine plural indirect object [–1,–2,+f,+pl,+hr,+lr] a. b. c. d. e.
le1 le2 gli1 gli2 li
[+f,+pl,+hr] [+f,+hr,+lr] [+hr,+lr] [+pl,+hr,+lr] [+pl,+hr]
unif( )
max (+hr)
max (+lr) *!
max (+pl) *! *!
*!
max (+f)
* * *
Not only is the object clitic paradigm more diVerentiated but there are also
Romance clitic inventories 105
other functions encoded by verbal clitics. The clitic ne, for instance, refers to quantiWed direct objects (see 49a,b) and is therefore often called the “partitive” clitic. (49) Partitive ne a.
Quanti esami hai fatt-o Wno adesso? how.many exam.pl have.2sg made.ptc-m till now ‘How many exams have you made yet?’ – Ne ho fatt-i solo tre. ne have.1sg made.ptc-m.pl only three ‘I have made only three.’ – *Ne ho fatt-o solo tre. ne have.1sg made.ptc-m only three
b.
Vuoi ancora un po’ di vino? want.2sg still a bit of wine – No, grazie, non ne voglio più. no thanks not ne want.1sg more ‘Would you like some more wine?’ – ‘No, thanks, I don’t want any more.’ (Da Forno & Manzini, 1995:164)
In the synthetic perfective tense (i.e. “passato prossimo”), the participle usually does not agree with full DP arguments of transitive verbs but shows the default agreement morphology (i.e. masculine singular: -o, see 49a). If the noun phrase is replaced by partitive ne, on the other hand, the participle obligatorily shows the gender and number agreement features of the replaced noun phrase (see 49b). For non-quantiWed direct objects, this is the case as well.24 This is illustrated by the sentences in (50) where the participle agrees with 3f.do la and 3pl.do li. Participle agreement only occurs with clitics but not with full, quantiWed NPs. (50) Participle agreement a. Hai chiamat-o Maria? have.2sg call.ptc-m Maria – Si, l’ ho chiamat-a ieri. yes, 3f.do have.1sg call.ptc-f yesterday ‘Did you call Maria?’ – ‘Yes, I called her yesterday.’
106 Clitics in the lexicon
b.
Avete incontrat-o i Bergonzoni? have.2pl meet.ptc-m d.pl.m B. – Li abbiamoincontrat-i ieri. 3pl.do have.1pl meet.ptc-pl.m yesterday ‘Have you(pl) met the Bergonzonis?’ – ‘We met them yesterday.’
These facts show that ne is in complementary distribution with direct object clitics and DPs and must be compatible with the feature combination [+hr, –lr]. In addition, replacement of non-quantiWed direct objects by ne must be excluded. Since ne also replaces non-partitive prepositional objects with the preposition di, which occur with a number of verbs (see the examples in 51a-b), it does not seem reasonable to attribute a feature as, for instance, [+quantiWed] directly to the semantics of ne. Replacement of non-quantiWed prepositional objects diVers from replacement of quantiWed direct objects in that with the former, no participle agreement takes place (see 51c according to Calabrese, 1991:555). (51) ne replaces prepositional objects with di a.
discutere di interessarsi di pensare di avere il tempo di essere contento di avere voglia di avere bisogno di
‘to discuss about’ ‘to take care of’ ‘to think (sth) of (sth or so)’ ‘to have time to’ ‘to be content with’ ‘to be in the mood to’ ‘to need’
b.
Che ne pensi?
‘What do you think of it?’
c.
i. (Della storia,) (about.df.f story(f)) Carlo/Carla ne ha discuss-o con Mario/Maria. Carlo/Carla ne has dicuss.ptc-mwith Mario/Maria ‘(With regard to this story,) Carlo/Carla has discussed it with Mario/Maria.’ ii. * (Della storia,) (about.df.f story(f)) Carlo/Carla ne ha discuss-a con Mario/Maria. Carlo/Carla ne has discuss.ptc-f with Mario/Maria
Romance clitic inventories 107
The preposition di also denotes partitive relations as in un po’ di vino (49b). Therefore, it is much more economical if ne remains unspeciWed with respect to the features “partitive” or “quantiWed” but restricted to objects which are assigned the preposition di.25 The semantic restriction to partitive objects is then attributed to the preposition and not to the clitic. The entry in (52) captures this restriction and allows for lexically required di as in (51) as well as for structurally required di as in (49). (52) ne: +D; [-max, -min] [+hr, P: di] / +V The “locative” clitic ci is much less restricted (see also chapter 1.2.1).26 ci replaces locative adverbial PPs with a number of diVerent local prepositions (see 53a). Locative use is also observed for vi, which also occurs as 2pl.o. Locative vi is mainly used in literary Italian and is not as common as locative ci. (53) locative ci a. Quando vai all’ università? – Ci vado domani. when go.2sg to.d.df university ci go.1sg tomorrow ‘When do you go to the university?’ ‘I go there tomorrow.’ b.
Claudia non c’ é. Claudia neg ci is ‘Claudia is not there.’
In addition, ci is used as an expletive clitic. In combination with the auxiliary essere ‘to be’, ci is similar to the locative adverb ‘there’ in English ‘there is’ / ‘there are’ or the neuter personal pronoun es in German es gibt (see 54a). Moreover, combinations of ci and vuole ‘he/she/it wants’ encode an impersonal need (see 54b) and ci is inserted in combinations with 3o clitics and avere ‘to have’ with possessive meaning (see 54c). (54) expletive ci a. In Italia ci sono molte grandi città. in Italy ci be.3pl many big cities ‘There are many big cities in Italy.’ b. Per girare a Venezia ci vuole il vaporetto. to move.around in Venice ci want.3sg d.df vaporetto ‘To move around Venice, one needs the vaporetto.’
108 Clitics in the lexicon
c.
Ma lei ce l’ ha o non ce l’ ha il biglieto? but prn.3f ci 3do have.3sg or neg ci 3 do have.3sg d.df ticket ‘But do you have the ticket or don’t you?’
In contrast to German es, however, expletive ci does not fulWl the grammatical function of impersonal subject but instead has an unspeciWc local meaning like English ‘there’.27 This is clear from number agreement marking as shown in (54a) and from the overt subject being diVerent from ci as, e.g., lei in (54c). ci replaces prepositional objects, just like ne, and occurs with verbs which are subcategorised for the preposition a (see 55a). ci cannot replace indirect objects which are also marked with the preposition a (56). (55) ci replaces prepositional objects with a a. riuscire a ‘to reach at’ dedicarsi a ‘to dedicate oneself to’ cominciare a ‘to begin to’ pensare a ‘to think of sth. or so.’, ‘to look after’, to provide’ b.
(56)
È stata proprio una brutta storia, ma non ci penso più. is been really d.idf ugly story but neg ci think yet ‘It has really been an ugly story, but I don’t think of it anymore.’ (Da Forno & Manzini, 1995:164) Hai dato il libro a Maria? aux.2sg give.ptc d.df book p M. ‘Did you give the book to Maria?’
–
Si, gliel’ ho dato ieri. yes, 3io.3do aux.1sg give.ptc yesterday vs. *Si, cil’ ho dato ieri. yes, ci.3do aux.1sg give.ptc yesterday
–
‘Yes, I gave it to her yesterday.’
At least three options can capture these data lexically. The Wrst is to have only one unspeciWed entry as, for instance, (57a) covering all diVerent occurrences of ci and thus subsuming 1pl.o. The second option is that two diVerent but homophonous ci exist as in (57b), one of them being 1pl.o and the other one subsuming the rest of the functions. The last option to be discussed is that there are three diVerent entries for ci, i.e. 1pl.o, local adverb and prepositional object as in (57c). The entries in (57) have the case feature [+hr] in common, since ci does not occur as a subject.
Romance clitic inventories 109
(57) Possible lexical entries for ci: +D, [-max, -min] [ a. ci b. ci1 c. ci1
+hr +1,+pl,+hr; +1,+pl,+hr;
ci2 +hr, +loc ci2 +hr, +loc;
]/V
ci3 +hr, P: a
Regarding vi, only option (57a) and (57b) — if the person feature [+1] is replaced with [+2] — are relevant here, because vi does not replace prepositional objects. For both clitics, we can directly reject unspeciWed entries like the one in (57a) for the following reasons. First, if the clitics were unspeciWed with regard to person, they would be interchangeable with regard to the object clitic paradigm, but they are not. Second, if they were speciWed for person but not for number features, they could be exchanged with 1o and 2o clitics in the singular paradigm, but they can not. To prevent this, the singular clitics mi and ti would need negative number values ([-pl]), which is unwanted in MM for reasons of economy. The latter is also the case if only one of them is speciWed with respect to number and the other is not, so that this option has to be rejected as well. Furthermore, none of the marked values [+1]/[+2] or [+pl], which might contribute to the entry in (57a), is compatible with the locative use of the respective clitics. As a last argument against the entry in (57a), it must be mentioned that without any speciWcation with respect to the possible adverbial function(s) which might be fulWlled by the clitic, no restriction to locatives is derivable and there is no reason why the clitics should not replace temporal or other adverbial expressions, either. We can therefore conclude that there are at least two homophonous entries for each of the clitics, one of them containing person and number values and the other a feature like [+loc] as sketched in (57b). For vi, these entries are suYcient to capture all its functions. For ci, however, there is still the replacement of prepositional objects with a which needs to be considered. Therefore, ci has an additional entry that is analogous to that of ne but restricted to the preposition a. The entries for vi and ci that result from these facts are thus: (58) Lexical entries for vi and ci a. vi1 +2,+pl,+hr vi2 +hr,+loc b. ci1 +1,+pl,+hr ci2 +hr, +loc
ci3
+hr; P: a
Having analysed the Standard Italian clitic inventory, I will now turn to a northern Italian dialect which has a still more complicated clitic system.
110 Clitics in the lexicon
3.2.5 Piattino clitics Piattino is an Italian dialect spoken in the Lombardian town Piatta. Like many other northern Italian dialects, Piattino not only has the full Italian clitic inventory but subject clitics as well. The subject and object clitic paradigm is given in (59). (59) a. Piattino subject and object clitics sg s 1 2 3m 3f
do
a (a)l la
pl io
rfl
me te ie
si
s (a)n – i li
do
io
rfl
me ve ie
si
b. Additional clitics ci, vi: ‘locative’ ne: ‘partitive’
The additional subject clitics make the organisation of the Piattino clitic inventory more complex, particularly because of diVerent strategies in locating the subject clitics in the paradigm. The lexical entries are given in (60). (60) Lexical entries for Piattino clitics: +D; [-max,-min] [ ]/ +V a. a +1 b. (a)n +1,+pl c. me +1,+hr d. te +2 e. ve +2,+pl,+hr f. (a)l [person ] g. i +pl h. la +f i. li +f,+pl j. ie +hr,+lr k. si +rX l. ne +hr; P: di m. vi1 +2,+pl,+hr n. vi2 +hr,+loc o. ci1 +1,+pl,+hr p. ci2 +hr,+loc q. ci3 +hr; P: a
As to the 1st person, 1s a and 1pl.s an are introduced. These clitics are unspeciWed with respect to the case features [hr] and [lr] and have only person
Romance clitic inventories
and number features (see 60a,b). Moreover, 1o me occurs in the singular and in the plural paradigm. Therefore, another syncretism has to be represented which diVers from the syncretism above in that it is the number feature which has to be underspeciWed. Within the framework of MM, this syncretism automatically results from the fact that no (number-speciWed) 1pl.o clitic exists in the lexicon. Thus, the (underspeciWed) 1o clitic can Wll both cells (see 60c). The choice between 1pl.s an and 1o me is determined by the SpeciWcity Principle, because case information is higher in the Hierarchy of Functional Categories and thus assumed to be more speciWc. Regarding the max(feature) constraints, we can derive the ranking max(+hr) >> max(+pl) from this data. Let us consider 3rd person clitics next. Here, quite a diVerent strategy in establishing subject clitics can be observed. Ambiguous subject and direct object clitics occur in each number and gender, while the indirect object clitic is unambiguous regarding case but ambiguous regarding gender and number. The subject/direct object clitics al, la, i and li are thus unspeciWed with respect to case features and bear only gender and number features (see 60f-i), while the indirect object clitic ie bears only case features but neither number nor gender features (see 60j). 2nd person clitics, Wnally, not only show diVerent behaviour from 1st and 3rd person clitics but also diVerent behaviour in singular and plural. In the singular paradigm, only one clitic covers subject, direct and indirect object. We can therefore assume that this clitic is unspeciWed with respect to case features (see 60d). The diVerence to the 1st person is due to the absence of a (case-speciWed) 2o clitic. Therefore, no competitor for the object cells exists and the underspeciWed clitic is free to Wll these cells as well. The plural paradigm, on the other hand, shows no 2nd person subject clitic at all. Independently of the theoretical framework, such a gap in the paradigm is diYcult to obtain. In MM, a paradigm cell that is not Wlled by a fully speciWed element is usually Wlled by a less speciWc element that is compatible with the feature speciWcation of the respective cell. The result then, is syncretism as shown above. For the feature speciWcation in question here, however, this mechanism must be impeded. How can this be done? Let us Wrst look at the unproblematic cases of 1s, 1o and 3io clitics. The 2nd plural subject cell has the feature speciWcation [+2,+pl,-hr,-lr], which is compatible neither with 1s and 1o because of their person feature [+1], nor with 2pl.o ve or 3io ie because of the case features [+hr] and [+hr,+lr], respectively (see 60e,j). unif(f) violations therefore prevent these clitics from Wlling the cell. The 2nd person singular clitic te and the 3rd person plural clitics i and li, however, do not bear
111
112
Clitics in the lexicon
positively speciWed feature values which are incompatible with the respective cell. Wunderlich (2001a,b) attributes similar gaps in the Yimas and Dalabon inventory to sociolinguistic taboos. He integrates such taboos into the analysis in terms of negative faithfulness constraints which prohibit certain input features from having a correspondent in the output in speciWc contexts. For Piattino, the constraint *max(+2)/[+pl,-hr] suYces to block the 2nd person plural subject cell.28 Let us now turn to the French clitic inventory which resembles the Piattino clitic inventory insofar as it also contains subject clitics.
3.2.6 French clitics As regards French, it is worth noticing that there are diVerences between the Standard and the Colloquial varieties. Although Colloquial French is not considered a distinct dialect, some of the clitics in particular are remarkably diVerent in both varieties. This might lead to general insights concerning language change and the development of clitics. I will therefore compare both varieties in this section. Let us Wrst consider the Standard French clitic inventory in (61). In addition to subject, direct and indirect object clitics, French also exhibits locative and partitive clitics akin to Piattino and Italian. Moreover, French has the impersonal clitic on and the negative clitic ne, both occurring in the verbal clitic sequence. (61) a. Standard French subject and object clitics sg s
do
je tu il elle
1 2 3m 3f
pl io
s
me te le la
lui
b. Additional clitics y: en: on : ne:
rfl
‘locative’ ‘partitive’ impersonal subject negation
do
io
rfl
nous vous se
ils elles
les
leur se
Romance clitic inventories
While in the singular paradigm, all subject clitics are unambiguous with respect to case, in the plural paradigm unique subject clitics only exist for the 3rd person. Therefore, I suggest that the subject clitics 1 je and 2 tu only bear person information (see 62a,d) and that it is due to the existence of the more speciWc 1o me and 2o te (62c,f) that je and tu are restricted to subjects. As to 1pl nous and 2pl vous, I also suggest that they do not have any case features (see 62b,e). Since there are no more speciWc object clitics in the plural paradigm, they are not restricted to subject positions. With 3rd person clitics, no case syncretism is observed. In addition, we Wnd gender distinctions with 3s il and 3f.s elle in the singular, with 3pl.s ils and 3f.pl.s elles in the plural and with 3do le and 3f.do la in singular (see 62g-j,k,m). 3io lui and 3pl.io leur, as well as 3pl.do les, show gender syncretism (see 62l,n,o). (62) Lexical entries for French clitics: +D; [-max,-min] [ ]/ +V a. c. d. f. g. i. k. m. n. p.
je me tu te il elle le la lui se
+1 +1,+hr +2 +2,+hr [person ] +f +hr +f,+hr +hr,+lr +rX
b.
nous +1,+pl
e.
vous +2,+pl
h. j. l.
ils elles les
+pl +f ,+pl +pl,+hr
o.
leur
+pl,+hr,+lr
The locative and partitive clitics y and en are quite similar to their Italian counterparts. en is restricted to the replacement of prepositional phrases with de which occur as local adjuncts determining the source with verbs of movement (63a), as partitive objects and as prepositional objects, the latter being restricted to [-human] entities in Standard French. (63) Locative and partitive clitic in French a.
Il en est rentré hier soir. ‘He came from there yesterday night.’
b. Vous avez acheté des oranges? – Qui, j’en acheté (un kilo). ‘Have you bought oranges ?’ – ‘Yes, I have bought some (one kilo).’
113
114
Clitics in the lexicon
c.
On en parle beaucoup.
‘One talks about it/*him/*her a lot.’ (Klein & Kleineidam 1979:31f)
However, there is participle agreement with direct object clitics but not with en: (64)
Participle agreement in French Tu as fait les sandwiches? ‘Have you made the sandwiches?’ – Oui, je les ai fait-s yes 1 3pl.do have.1sg make.ptc-pl
‘Yes, I have made them.’
– Oui, j’en ai fait yes, 1 en have.1sg make.ptc
‘Yes, I made some.’
Similar to Italian ci, y is less restricted as it may replace all other local prepositional phrases, i.e. phrases with à, dans, en and sur (see 65a). Furthermore, it replaces prepositional (see 65b) but not indirect objects with à (see 65c). Unlike en, y is not restricted to non-human prepositional objects. Finally, y occurs in the expletive construction il y a (lit.: ‘he/it there has’) which resembles the Italian construction c’é (see 65d), except for the additional subject. (65) DiVerent uses of y in French a. Il y retourne demain. b. On y pense beaucoup.
‘He returns there tomorrow.’ ‘One thinks of it/him/her often.’ c. Je lui/*y téléphone. ‘I call her.’ d. Qu’est-ce qu’il y a dans la chambre? ‘What is there in the room?’
In addition to the impersonal reXexive construction, there is the impersonal subject pronoun on which agrees with the 3rd person singular verb inXection and is thus similar to English one and German man.29 (66) Impersonal subject in French En France, le café se boit sans lait / on boit le café sans lait. ‘In France, one drinks the coVee without milk.’
The clitic negation ne, Wnally, occurs together with other, non-clitic negative elements as, for instance, personne ‘nobody’ (see 67a) or rien ‘nothing’ (see 67b). In negative statements, the negation pas has to occur in addition to ne (see 67c)
Romance clitic inventories
(67) The negation clitic ne in French a.
Personne ne m’ a vu. nobody ne 1o has.3sg see.ptc
b. Il ne m’ a rien dit. 3s ne 1o has.3sg nothing say.ptc c.
Je ne le crois pas. 1 ne 3do believe neg
‘Nobody has seen me.’ ‘He has told me nothing.’ ‘I don’t believe it.’ (Klein & Kleineidam 1979:103)
The lexical entries for these four additional French clitics are given in (68). The entry for en is similar to the entry of Italian ne with the additional restriction to non-human entities (see 68a).30 Regarding y, the lexical entries are again similar to Italian ci but since y is not used as 1pl.o, there is one entry less than in Italian (see 68b,c). The impersonal subject clitic on only has the feature [+generic], while the features [–pl], [–1], [–2], [–hr] and [–lr] are added by default as discussed earlier. The negation clitic ne, Wnally, only has the feature [+neg]. (68) Lexical entries for Standard French clitics: +D; [-max,-min] [ ]/ +V a. en +hr,-human; P: de b. y1 +hr,+loc c. y2 +hr; P: á d. on +generic e. ne +neg
Let us now come to the diVerences between Standard and Colloquial French and consider the changes in the lexical entries that have to be made in order to capture these diVerences. As to the subject and object clitics, the Standard French impersonal clitic on also functions as 1pl.s in colloquial French (see 69). In contrast to nous, on still agrees with 3rd person singular verb inXection. Furthermore, on cannot be used as 1o. (69) The use of nous in Standard and on in Colloquial French a.
Standard French nous portons tu nous attends
b. Colloquial French on porte *tu on attends
‘we carry’ ‘you wait for us’
In the case of the 1st person plural, we can therefore observe an assimilation to the singular paradigm with the opposition subject vs. object clitics. With 3rd person subject clitics, on the other hand, neutralisations with respect to gen-
115
116
Clitics in the lexicon
der and number can be observed. The substitution of elle by il and of elles by ils is widespread in Colloquial French. Subsequently, il sometimes replaces ils. This last step is not implemented into the paradigm of Colloquial French subject and object clitics in (70). (70) Colloquial French subject and object clitics sg s 1 2 3m 3f
do
je tu il
pl io
rfl
me te le la
s on
lui
se
ils
do
io
rfl
nous vous les
leur
se
In addition to the subject and object clitics, other clitics also show a diVerent behaviour from Standard French. The partitive clitic en, for instance, is not restricted to [-human] entities, so that (71) is correct in Colloquial French. (71) Je ne veux plus en entendre parler, de ce type. ‘I don’t want to hear anything again of this type.’ (Klein & Kleineidam 1979:31)
The locative clitic y is not diVerent in both varieties of French, but the negation clitic ne is often omitted in Colloquial French (see 72). (72) C’est pas la même chose. ‘It isn’t the same thing.’
(Klein & Kleineidam 1979:103)
These diVerences are captured with only a few changes in the lexical entries (see 73). First, with regard to 1pl.s on, a lexical entry must be added.31 This has an eVect on the former 1pl nous in that it now occurs only as object and is thus more speciWc, bearing the feature [+hr] which turns nous into 1pl.o in Colloquial French. Furthermore, 3f.s elle and 3f.pl.s elles are no longer part of the clitic inventory. Less speciWc 3s il and 3pl.s ils consequently Wll the cells. Partitive en diVers from Standard French only in lacking the feature [-human], so that it is unrestricted with respect to animacy in Colloquial French. The negation clitic ne, Wnally, is no longer part of the clitic inventory in Colloquial French.
Romance clitic inventories
(73) Lexical entries for Colloquial French clitics: +D; [-max,-min] [ ]/ +V a. je +1 b. on +1,+pl c. me +1,+hr d. nous +1,+hr,+pl e. tu +2 f. vous +2,+pl f. te +2,+hr g. il [person ] h. ils +pl i. le +hr j. les +pl,+hr k. la +f,+hr l. lui +hr,+lr m. leur +pl,+hr,+lr o. se +rX p. en +hr; P: de q. y1 +hr,+loc r. y2 +hr; P: á s. on +generic
In the last sections, we have been dealing with an increasing number of verbal clitics. We have seen that additional clitics may result from adding speciWc functions to clitics which are already part of the lexicon as shown for the development of ci in Italian and on in Colloquial French. Let us now look at the eastern Romance language Romanian, which is more inXuenced by the other Balkan languages than by neighbouring western Romance languages and has consequently developed a quite diVerent clitic inventory.
3.2.7 Romanian clitics The most obvious diVerence between Romanian and the western Romance languages is that Romanian has auxiliary clitics (see below in 77).32 Furthermore, there are distinct clitics for 3do.rfl and 3io.rfl. In (74), the pronominal clitics are listed by their default form in Romanian orthography.33 Although it is sometimes argued that Romance clitics show a number of diVerent idiosyncratic forms (see e.g., Monachesi 1999, 2000 and chapter 2), Popescu (2000) shows that all surface forms of the clitics can be derived from the forms in (75) by general phonological and prosodic constraints. I will come back to Popescu’s analysis in chapter 4.
117
118
Clitics in the lexicon
(74) Romanian object clitics sg
pl
do
io
1 1rfl
ma[
îmi
ne
2 2rfl 3m 3f 3rfl
te
ît6 i
va[
îl o se
îi îs6 i
do
îi se
io
le îs6 i
Let us Wrst look at the pronominal clitics in Romanian. Romanian — in contrast to Piattino and French but like Portuguese, Spanish and Italian — does not have any subject clitics, which might be the result of Romanian being a subject pro-drop language. A closer look at the paradigm reveals that some features of the object clitic paradigm are distinct from all western Romance languages. Particularly the relation between full speciWcation and ambiguity in the paradigm is diVerent. Regarding the 1st and 2nd person singular clitics, we Wnd fully speciWed 1do ma[ and 2do te and fully speciWed 1io îmi and 2io ît6 i respectively (75a-c). In western Romance languages, these cells are ambiguously Wlled with 1o and 2o clitics. In the plural paradigm, on the other hand, 1pl.o ne and 2pl.o va[ show the same syncretism as the other Romance languages (75g,h). With the 3rd person singular we Wnd the generally attested gender distinction 3do îl and 3f.do o (see 75d,f) and the case syncretism with 3io îi (see 75e). 3rd person plural clitics, however, show a diVerent behaviour. 3pl.do îi (see 75i) stands in opposition to 3f.pl.do le (see 75k) on the one hand, and to 3pl.io le (see 75j) on the other. (75) Lexical entries for Romanian object clitics: +D; [-max,-min] [ ]/ +V a. ma[ +1,+hr b. îmi +1,+hr,+lr b. te +2,+hr c. ît6 i +2,+hr,+lr d. îl +hr e. îi1 +hr,+lr f. o +f,+hr g. ne +1,+pl,+hr h. va[ +2,+pl,+hr i. îi2 +pl,+hr j. le1 +pl,+hr,+lr k. le2 +f,+pl,+hr
Romance clitic inventories
l. m.
se îs6 i
+hr,+rX +hr,+lr,+rX
In the lexical entries a decision has again been made about homophonous clitics and systematic and unsystematic syncretism. First, homophonous 3io and 3do.m îi are analysed as unsystematic syncretism and thus represented by two distinct lexical entries (see 75e,i). In this case, it is obvious that systematic syncretism must be ruled out, since the feature speciWcation in the paradigm cell of 3io [-pl,+hr,+lr] is not compatible with the feature complex [+pl,+hr, –lr] of 3pl.do îi, and underspeciWcation of both the conXicting features [pl] and [lr] would be necessary. The feature [lr], however, can only be assigned by paradigmatic contrast. If îi is not assigned the feature [+lr] in singular, this would thus lead to a lack of the feature [-lr] for îl, so that îl and îi would have the same feature speciWcation 3o. The undesired result would then be that no choice between the two clitics could be made. Analogously, îi could not be chosen as a plural clitic if it did not have the feature [+pl] because it could not be distinguished from 3do îl. The other 3rd person object cells in the plural paradigm, i.e. [+f,+pl,+hr,-lr], [+f,+pl,+hr,+lr] and [+f,+pl,+hr,+lr] are Wlled by only one clitic, namely le. Again, the question must be answered whether this syncretism is systematic, partly systematic or unsystematic, that is, whether we have one, two or three diVerent lexical entries for le. These options are outlined in (76). (76) Systematic, partly systematic and unsystematic syncretism of le a. le1: [+pl,+hr] = [+pl,+f,+hr,-lr] v [+pl,-f,+hr,+lr] v [+pl,+f,+hr,+lr] b. le1: [+pl,+f,+hr,-lr] =[+pl,+f,+hr,-lr] le2: [+pl,+hr,+lr] = [+pl,-f,+hr,+lr] v [+pl,+f,+hr,+lr] c. le1: [+pl,+f,+hr,-lr] le2: [+pl,-f,+hr,+lr] le3: [+pl,+f,+hr,+lr]
Systematic syncretism for all three cells needs extensive underspeciWcation as shown in the entry in (76a). To subsume the contradicting feature speciWcations [+f] vs. [-f] and [+lr] vs. [–lr], these features must be unspeciWed with the result of 3pl.o. Former 3pl.do îi would lack the feature [-lr], since no paradigmatic contrast would provide it. Consequently, it would also come out as 3pl.o so that no decision between the two clitics could be made. Therefore,
119
120 Clitics in the lexicon
at least the contrast between 3pl.do and 3pl.io must be marked by the features [+lr] and [–lr]. This contrast is contributed by the entry of le2 in (76b), which is assigned the feature [+lr]. This feature speciWcation, however, excludes the paradigm cell [+f,+pl,+hr,–lr], so that we must provide another entry, namely le1 in (76b), which is able to Wll this cell.34 The last option (76c) also Wts the diVerent paradigm cells perfectly. For reasons of economy, however, it is abandoned here and the option in (76b) is chosen for the lexical entries in (75j,k) instead. Let us Wnally come to Romanian auxiliary clitics (77). These clitics are somewhat problematic, since the syncretism they show is obviously systematic but nevertheless does not belong to natural classes which can be captured by underspeciWcation. (77) Romanian auxiliary clitics perfective 1 2 3 1pl 2pl 3pl
conditional
am
future
as6
oi
ar
o om ot6 i or
ai a am at6 i au
ar
The Wrst diYculty is that 2nd person singular and 1st and 2nd person plural perfective and conditional auxiliary clitics are identical, while the 1st person singular and 3rd person singular and plural are diVerent. The second diYculty is that 1st and 2nd person singular future auxiliary clitics are identical, while the other future auxiliary clitics diVer from each other. Disjunctive entries are not necessary in order to account for this syncretism (see 78). (78) Lexical entries for auxiliary clitics: [-min,-max],[asp, tmp, per, num]/ V a. am +1,+pfv b. as6 +1,+cnd d. ai +2 e. a +pfv f. ar +cnd h. am +1,+pl j. at6 i +2,+pl l. au +pl,+pfv
c. oi
(+1 v +2), +fut
g. i. k. m.
+fut +1,+pl,+fut +2,+pl,+fut +pl,+fut
o om ot6 i or
Conclusions
Since auxiliaries may induce the features [+fut], [+cnd] or [+pfv], the forms which are underspeciWed for the aspect and tense features (i.e. 78d,h,j) compete with otherwise equally speciWed forms which bear the feature [+fut] (i.e. 78c,g,i,k.m). They are therefore not [+fut] but either [+pfv] or [+cnd].
3.3. Conclusions In this chapter the Romance clitic inventories have been introduced and the morphological component has been located within the grammar. It has been shown that underspeciWed lexical entries in the manner of MM, together with a language-speciWc ranking of faithfulness and markedness constraints, can predict the clitic choice in the Romance languages. Particularly the similarities between the languages regarding case and gender syncretism vs. case and gender distinction in 1st and 2nd person vs. 3rd person clitics have been attributed to certain markedness constraints which prohibit the features [+lr] and [+f] from occurring in the same feature structure as the person features [+1] and [+2]. An exception to this generalisation is only documented in the Romanian singular clitic paradigm, where 1st and 2nd person clitics, too, show a distinction with respect to direct and indirect object. As regards the ranking of the faithfulness constraints, the following tendencies have been observed: (79) General ranking tendencies
a. b. c.
max(+feature) >> *(feature) max(+1),max(+2) >> max(+rX) >> max(+lr) >> max(+f) max(+pl) (>> max(+1),max(+2) ) >> max(+f)
Violations of unification constraints never occur in optimal output candidates and unification must therefore rank highly (see 79a). The ranking of max(+1) and max(+2) above max(+lr) and max(+f) accomplishes the representation of the features [+1] and [+2] rather than [+lr] or [+f] by the optimal output clitic candidates. Regarding the feature [+rX], diVerent ranks of the constraints max(+1), max(+2) on the one hand and max(+hr), max(+lr), max(+f) on the other have been observed. 3rd person reXexive arguments are always represented by a clitic which contains the feature [+rX], even if the features [+f], [+hr] and [+lr] are not part of its feature structure. 1st and 2nd person arguments, on the other hand, are represented by clitics which contain
121
122 Clitics in the lexicon
the features [1] or [2], even if they do not contain the feature [+rX] (see 79b). The ranking of the constraints max(+pl) and max(+f) has not been Wxed in all languages. However, the Italian data suggest that max(+pl) ranks higher than max(+f) and the Argentinian Spanish data suggest that max(+pl) even ranks higher than max(+2) (see 79c).
Notes 1. As to other approaches which also employ OT-like constraints rankings in the morphological analysis see Müller (1999). 2. See also Wunderlich & Fabri (1996), Wunderlich (1996a,b,c). 3. The referential information is provided by the discourse. In the theta-raster of the input, the information about the case features of the argument is encoded in the fashion of LDG. 4. In OT, the evaluation of diVerent output candidates is illustrated by so-called “tableaux”, in which the lines represent the output candidates and the columns the constraints. The ranking of the constraints is mirrored by their position: constraints on the left are higher-ranked than constraints on the right (notationally, “x >> y” means that “x is ranked higher than y”). Equally highly ranked constraints are divided by a dotted line instead of a continuous line. Constraint violations are marked by “*” in the respective cell, fatal constraint violations with “*!”. The best (hence optimal) output candidate is marked with “”. 5. The representation of agreement information as an index on the respective argument resembles the co-indexing mechanisms which is chosen in HPSG (Pollard & Sag 1994). 6. Moreover, there are discourse-semantic differences between such structures. These will be discussed in chapter 5. 7. Constraints on output-output correspondences are deWned by Benua (1995). This correspondence relation does not concern different outputs to the same input as it is the case with agreement relations but different outputs to different inputs which are organised in paradigms. 8. Instead of inventing the new constraint unification, one could of course also reinterpret the (genuinely phonological) identity constraints according to the needs of morphology. I believe, however, that both concepts are useful also in morphology (which will be shown in chapter 4). 9. Most Romance languages also have clitic determiners, which are often phonologically identical to pronominal clitics. 10. In Romanian orthography, a[ encodes schwa, so that the clitic ma[ is phonologically represented as [m6], î encodes the central high vowel [G], t6 encodes [ts] and s6 encodes [w ].
Notes 123
i in the end is often phonologically represented not by a full vowel but by palatalisation of the preceding consonant. 11. As argued in chapter 2, the Romanian negation ne is an afWx and nu is a word. 12. This proposal goes back to Bonet (1995). 13. The OT constraints parse and fill roughly equal the CT constraints max and dep. 14. In the correspondence variant of Optimality Theory below, it is assumed that there are different constraints at work, namely max(person), which is violated with underspeciWed clitics, and ident(person), which is violated if the person feature is parsed with the wrong feature. 15. The plural speciWcation comes from the agreement with plural verbs as in Si mangiano pasta in Italia. rfl eat.pl pasta in Italy ‘One eats pasta in Italy.’ 16. Recall from section 1.3.1 that [–] values are usually not part of the lexical entries but added by default. They deWne, however, the cells of the paradigm by dividing the space of the dimensions into [+] and [–]valued cells. On the differences between radical and principled underspeciWcation compare Inkelas (1994). 17. dep(arg)M constraints, on the other hand, prohibit that clitics occur in the output if there are no correspondents in the input. These constraints are not violated in the candidates considered here. See also the discussion in Gerlach (1998a) on this topic. 18. In the following tableaux, 2nd person clitics are neglected. They behave analogously to 1st person clitics. 19. These constraints may also be violated. Violations are relevant as it comes to clitic substitution and morphological feature transfer in clitic sequences, a topic which will be discussed in chapter 4. 20. Recall from chapter 2, however, that 3rd person clitics in Portuguese may trigger phonological neutralisation of coda consonants so that they phonologically resemble their Spanish counterparts. 21. In earlier versions of MM, conXicts like this were solved by the hierarchy of functional categories (see i) together with the SpeciWcity Principle which requires that of two potential afWxes the most speciWc one is preferred. The hierarchy of functional categories reflects the degree of speciWcity in that categories that are higher (i.e. more on the right) on this hierarchy are considered more speciWc. (i) Relevant part of the Hierarchy of Functional Categories verb/noun <... < gender < number < person < case In that case, however, 2o is more speciWc than 3pl.o and should thus be preferred. Since this option is not theoretically excluded — and actually should not be excluded, as will be clear by the Piattino data in section 3.2.5 — the lexical entries must be changed so that the correct choice of clitics is determined. Therefore, 2o te would need the additional feature [-pl] which prevents it from being chosen in this context.
124 Clitics in the lexicon
Note moreover, that according to the hierarchy of functional categories max(case) should rank higher than max(person). However, the ranking of these constraints is reverse in all languages examined. 22. If one assumes that the Argentianian Spanish clitic paradigm is derived from Iberian Spanish clitic paradigm by re-ranking of max(+pl) it is reasonable that this re-ranking takes place step by step. max(+pl) is therefore ranked above max(+2) but not above max(+1). Otherwise, one would expect that also the 1pl.o clitic gets lost. 23. Note that, as above, max(+pl) must rank higher than max(+f), because a non-feminine plural clitic is preferred to a feminine singular clitic as the optimal correspondent to the input 3rd person feminine plural direct object. 24. Moreover, participle agreement is obligatory for subjects of intransitive verbs which form the passato prossimo with the auxiliary essere. Mio padre è venut-o qui ‘nel 66, my father is come.ptc-m here in ’66, due anni dopo è arrivat-a mia madre, .... two years later is arrive.ptc-f my mother ‘My father came here in ’66, two years later, my mother arrived, ...’ (Da Forno & Manzini, 1995: 80) 25. In LDG, verbs which obligatorily combine with prepositional objects are lexically marked. The respective argument role bears the information about the preposition as shown for pensare: pensare: λy λx think(x,y) [+hr] [-hr] [-lr] [+lr] PP: di 26. On the various uses of ci, see also Calabrese (1995). 27. As in other Romance languages, impersonal subjects are expressed by the reflexive clitic si. 28. To block the unwanted Wll-up of this cell, one could also assume that the lexical entries are fully speciWed and thus incompatible with the respective cell. te could easily Wll the empty subject cell in the plural paradigm in the fashion the 1o clitic Wlls the object cells in singular and plural. Therefore, te must be negatively speciWed by the number feature [pl]. With the 3rd person plural clitics, blocking can be achieved by adding the negative person feature [-2] to the lexical entries. Consequently, the 3rd person singular clitics are least speciWed and therefore able to Wll the empty cell. To avoid this, they have to be assigned the feature [-2], too. 29. Furthermore, French has the construction il faut to express impersonal need as in Pour faire une vinaigrette il faut de la moutarde. ‘To make a vinaigrette one needs mustard.’ Note that this construction resembles the Italian ci vuole (see 54b). 30. We will come back to restrictions with respect to animacy and speciWcity in chapter 5. 31. The feature [+pl] in the lexical entry of on, however, is not unproblematic, since on agrees with [-pl] verb forms. On conflicting agreement features, see Urbas (1993).
Notes
32. Recall from chapter 2 that other elements which are sometimes referred to as clitics have been shown to be words (i.e. the negation nu) or afWxes (i.e. the negation ne and monosyllabic intensiWers). 33. For the phonological representation of Romanian clitics, see note 10. 34. Note that another decision has implicitly been made by this choice of entries. Theoretically, one could also think of [lr]-underspeciWcation in the entry of le1 in (76b), so that le1 as well as le2 would be able to Wll the cell [+hr,+lr,+pl,+f].
125
126 Clitic sequences
Properties of clitic sequences in Romance 127
Chapter 4
Clitic sequences
This chapter is concerned with the combinatory restrictions and linearisation properties of verbal clitics. When clitics combine with each other, their behaviour diVers from that in isolation. With clitic combinations we can observe three problematic phenomena: deviations in phonological form, substitution of clitics, and the unexpected non-occurrence of particular clitic combinations. We can further observe that the phonological and morphological contexts in which these restrictions apply are strikingly similar in all the languages examined. I will suggest that an Optimality-theoretical approach provides the most promising means to analyse these data, since such an approach can handle interactions of concurring requirements of diVerent linguistic modules as well as explain the variation between the languages. I will show that clitic sequences form a morphological unit with a Wxed order of clitics. For each case, the optimal sequence results from interacting constraints on the syntax, morphology and phonology of clitic sequences within an Optimalitytheoretical account. The chapter is organised as follows. First, the clitic sequence-speciWc morphological and morphophonological properties which in part have already been discussed in chapter 2 are assembled in section 4.1. In section 4.2, a number of recent analyses for clitic sequences will be discussed, in particular that of Bonet (1995) and that of Harris (1995), both placed in the framework of Distributed Morphology (Halle & Marantz 1994), and that of Spencer (2000), placed in the framework of Paradigm Function Morphology (Stump 1993). It will be argued that all of them suVer from certain shortcomings when it comes to the theoretical motivation of clitic sequence-speciWc phenomena, in particular when interactions of morphological and phonological or morphological and syntactic requirements inXuence the respective sequence. In sections 4.3 and 4.4, I therefore suggest an Optimality-theoretical analysis which is able to capture such interactions.
128 Clitic sequences
4.1. Properties of clitic sequences in Romance This section introduces morphological properties of clitic combinations. In Romance languages, clitic combinations resist separation, they maintain a strict internal order, they are often conWned to at most two elements, and they exhibit unique morphophonological behaviour.1
4.1.1 No separation One of the most noticeable properties of clitic combinations is that they usually resist separation, even if there are diVerent positions in the sentence which may potentially accommodate clitics. In Italian modal verb constructions, for instance, a clitic cluster can either attach to the embedded inWnitive as in (1a), or to the Wnite modal verb as in (1b). Adjacency seems essential and separation of two clitics is ungrammatical, as is shown in (1c) and (1d). (1) Clitics resist separation (Italian) a. Devo dir-glie-lo. b. must.1sg say-3io-3do ‘I must tell it to him.’ c.
*Gli devo dir-lo.
d.
Glie-lo devo dire. 3io-3do must.1sg say ‘I must tell it to him.’ Lo devo dir-gli.
Only Romanian diverges from this generalisation in a single case. If the 3f.do clitic o is used in analytic tenses in combination with the clitic auxiliary, it has to occur postverbally even if the conditions for preverbal position are given. Consider in this respect the examples in (2). While 3f.do o occurs preverbally and clusters with potential other clitics in present indicative in (2a), it occurs postverbally and does not cluster with the clitic auxiliary and potential other clitics, which occur in the default preverbal position (2b). With other clitics, such behaviour is not observed (2c). (2) Clitic split in Romanian a. no clitic split in synthetic tenses (j-)o cumpa[ r / *(i) cumpa[ r o / *cumpa[ r (j-)o 3io-3f.do buy ‘I buy it (for him).’
Properties of clitic sequences in Romance 129
b. clitic split in analytic tenses with 3f.do (le-)am cumpa[ rat o / *cumpa[ rat (le-)o am/ 3pl.io-aux.1sg buy.ptc 3f.do *(le-)o am cumpa[ rat
‘I bought it.’ c. no clitic split in analytic tenses with other clitics l-am cumpa[ rat / *am cumpa[ rat îl 3pl.do-aux.pfv.1sg buy.ptc ‘I bought them.’ (compare Popescu 2000:778) As Popescu (2000) shows, there are particular prosodic and phonological restrictions on clitic combinations which are responsible for this exceptional behaviour. In chapter 5, I will analyse in detail the positioning of Romance clitics with respect to their host and also brieXy sketch Popescu’s analysis. I will argue that certain alignment constraints determine the position of the clitics with respect to their host as well as the adjacency of clitics. For the time being, let us take the inseparability of clitics as a Wrst argument in favour of analysing the clitic sequence as a morphological unit.
4.1.2 Strict order and combinatory restrictions Another property of clitic combinations is that the clitics are strictly ordered with respect to each other and not — as is the case for aYxes — with respect to the host. The Italian examples above show that 3io gli has to precede 3do lo, regardless of the pre- or postverbal status of the clitic sequence. This order is the reverse of the order of full object DPs (see 3). (3) Order of DPs in Italian Devo dare il libro a Stefano. vs. *Devo dare a Stefano il libro. must.1sg give d.df book p Stefano ‘I must give the book to Stefano.’
Similarly, io clitics precede do clitics in the other Romance languages with only the exception of combinations of two 3rd person clitics in French, where this order is reversed.
130 Clitic sequences
(4) Order of clitics in French a. Je le lui ai déjà montré. 1s 3do 3io aux.1sg already show.ptc ‘I have already shown it to him.’ b. Je te l’ ai déjà montré. 1s 2o 3do aux.1sg already show.ptc ‘I have already shown it to you.’
Not only case is responsible for the order of clitics, but also the person features of the respective clitics. 1st and 2nd person clitics always precede 3rd person clitics. (5) Relevance of person features for the clitic order
Devo dar-te-lo. or must.1sg give-2o-3do ‘I must give it to you.’
Te-lo devo dare.
The facts that clitic combinations resist separation and that the inherent order is Wxed lead several researchers to the conclusion that clitic combinations need to be analysed as one morphological unit (Gerlach 1998a,b; Harris 1995; Spencer 2000) which may be called “clitic sequence”. This assumption is further supported by data from Wrst language acquisition. Children acquire clitic sequences very early and do not make mistakes in the ordering of clitics (see Monachesi 1995:76). One might recall moreover from chapter 2 that the unique morphophonological behaviour of clitic combinations is a further argument in favour of considering them a speciWc morphophonological unit. Clitics in clitic sequences cannot combine freely. Sequences of clitics which encode internal arguments of the verb (i.e. object, reXexive, locative and partitive) in Romance languages, for example, are often restricted to two clitics, even if the clitic inventory consists of more than just a direct and an indirect object clitic. In Spanish, clitic sequences containing more than two clitics are not attested, and although Standard Italian has locative and partitive clitics, clitic sequences consisting of more than two clitics are hardly attested. A templatic representation of clitic sequences, as for instance shown in (6), is therefore often misleading because it suggests that any number of clitics can combine as long as the clitics belong to diVerent slots.
Properties of clitic sequences in Romance
(6) Clitic cluster templates
Spanish: se 2 1 3 Perlmutter (1970:45) French: s neg 1/2/se 3do 3io y en Perlmutter (1970:57) Italian: 1/2/3io ciloc sirX 3do siimp ne Monachesi (1995:122) Restrictions on possible combinations of object clitics are based on case and person speciWcations. As mentioned above, an io clitic always precedes a do in Italian. In addition, 1o and 2o can neither combine with each other (see 7b), nor with 3io (see 7d). (7) Restrictions on possible clitic combinations in Italian a. Emanuela presenta me a te. Emanuela introduce.3sg prn.1sg p prn.2sg ‘Emanuela introduces me to you.’ b. *Emanuela me ti / te mi presenta. Emanuela 1sg.o 2sg.o / 2sg.o 1sg.o introduce.3sg c.
Emanuela presenta me a lui. Emanuela introduce.3sg prn.1sg.o p prn.3sg ‘Emanuela introduces me to him.’
d. *Emanuela glie mi / mi glie presenta Emanuela 3sg.io 1sg.o / 1sg.o 3sg.io introduce.3sg
Even if clitic-doubling requirements demand the realisation of a 1st and a 2nd person clitic, or a 1st/2nd and a 3rd person clitic, only one of the two actually occurs as a clitic (see Gerlach 1998a,b and the discussion in chapter 5 below). With regard to this person-dependent prohibition of the combination of clitics, Romanian is again exceptional. In Romanian, the second person singular direct object clitic can combine with 1io as well as with 3io. With regard to Spanish, some variation can be observed. While Parodi (1998a,b) excludes combinations of 1st and 2nd person clitics, Perlmutter (1970) gives examples for such combinations. The speakers of Spanish who accept such combinations clearly accept only the order 2 < 1 and not the reverse order. Ordering conditions and combinatory restrictions for clitics in clitic sequences are assembled in (8) (“x < y” should be read as “x precedes y”).
131
132 Clitic sequences
(8) Order and combinatory restrictions of object clitic sequences general tendency exception a. io < do French: 3do < 3io b. 1/2 < 3 Romanian: 3io < 2do c. *1/2 1/2 Romanian: 1io < 2sg.do, 3io < 2sg.do (Spanish: 2 < 1)
In addition, French and Italian have a number of other clitics, namely locative and partitive clitics, which combine with direct and indirect object clitics. Although these clitics obtain their own speciWc slots in the templates above (see 6), the total number of object, reXexive, partitive and locative clitics within one and the same clitic sequence is still restricted to two. While in French all object clitics precede loc, in Italian only 3io and 1o and 2o precede loc which in turn precedes 3do (see 9a). In both languages, all object clitics precede prt (see 9b). In combinations of object and reXexive clitics (see 9c), rfl precedes all object clitics in Spanish and all 3o in French, but in Italian, 3do, 1o and 2o precede rfl. In French, combinations of 1o and 2o with rfl are prohibited. As discussed above, rfl is used not only as a 3rd person reXexive clitic but also as an impersonal clitic in all Romance language. In Italian, the position of rfl with respect to 3do depends on its function: in reXexive use, it precedes 3do, while it follows 3do in impersonal use. loc precedes rfl in Italian but follows it in French (see 9d). In both languages, rfl and loc precede prt (see 9e-f). (9) Combinations of clitics in the internal clitic sequence Italian io < loc loc < 3do 1/2 < loc
French o < loc
Spanish —
b. object + partitive clitic
o < prt
o < prt
—
c.
rfl < 3do rfl < o io < rfl 1/2 < rfl *1/2 rfl 3do < rflimp
rfl < o
loc < rfl
rfl < loc
—
e.
partitive + reXexive clitic rfl < prt
rfl < prt
—
f.
locative + partitive clitic
loc < prt
—
a.
object + locative clitic
object + reXexive clitic
d. locative + reXexive clitic
loc < prt
Properties of clitic sequences in Romance
Romanian and French depart from the limitation to two clitics to some extent. In both languages, the clitic inventories provide clitics which are not internal arguments of the verb. These clitics may occur in addition to the internal object cluster. In Romanian analytic tenses, auxiliary clitics occur in clitic sequences in addition to the object clitics. Except from the above-mentioned case with 3f.do o (see 2), the auxiliary follows both object clitics (10). (10) Romanian: io < do < aux
exception: io < aux V-3f.do
In French, subject and negation do not belong to the internal arguments of the verb and may therefore occur in addition to a group of maximally two other clitics. In such instances, the subject clitic precedes the negation, which in turn precedes any other clitics (see 11). (11) French:
s < neg < {o / loc / prt}
4.1.3 Opacity 3rd person clitics and reXexive clitics exhibit particular morphological behaviour in clitic sequences (see Bonet 1995:607 who refers to such cases as “opaque clitics”). These clitics are either deleted or replaced by other clitics, or their surface form diVers from their form in isolation in combination with other clitics. Deletion is found for example in Spanish where reXexive se and impersonal se cannot both occur in the same sequence (12a). Substitution is found in the same context in Italian, where the Wrst si is replaced by the Wrst person plural or locative clitic ci (see 12b).2 Substitution within clitic sequences is also found in Spanish, e.g. in a combination of the two object clitics le and lo, the Wrst one is replaced by the reXexive clitic se (see 12c). In Italian, the same phonological context is created by combination of the feminine indirect object clitic le and another 3rd person clitic. Here, it is not the reXexive clitic, which replaces the Wrst one, but the indirect object clitic gli which is unmarked with respect to gender (see 12d). (12) Opaque clitics I: deletion and substitution a. Spanish: clitic deletion b. Italian: clitic substitution *se se → se *si si → ci si rfl rfl rfl rfl rfl 1pl/loc rfl c.
Spanish: clitic substitution *le lo → se lo 3io 3do rfl 3do
d.
Italian: clitic substitution *le lo → glie-lo 3f.io 3do 3io-3do
133
134 Clitic sequences
In Mexican and Uruguayan dialects of Spanish, a 3do clitic referring to a nonfeminine and/or singular entity may exhibit feminine and/or plural morphology if the number and gender of the 3io clitic in the same sequence is no longer visible due to clitic substitution (following from 12c: le →se). As a result, the sequence se las, for instance, has many diVerent readings in these dialects (i.e. in 13: 3io-3f.pl.do, 3f.io-3f.pl.do, 3pl.io-3f.do, 3pl.io-3f.pl.do, 3f.pl.io.-3f.pl.do, 3f.pl.io-3do, 3f.io-3pl.do, see Bonet 1995:635). These clitic combinations are thus no longer transparent, which is additional evidence in favour of the independent morphological category clitic sequence. (13) Opaque clitics II: morphological feature transfer a. *le las b. *les la c. *les las → se las d. *les lo e. *le los
Clitics and clitic combinations are also attested in the Slavic languages and in Greek. Here, the internal order of clitics obeys restrictions similar to those mentioned above for the Romance languages.3 As already discussed in chapter 2, clitic sequences show special phonological processes which are not attested elsewhere in the morphophonology of the respective languages. In Italian, the vowel in the Wrst of two clitics in clitic combinations is lowered if the second one begins in /l/ or /n/ (see 14a). In Romanian clitic combinations, on the other hand, [e] and [6] in the Wrst of two clitics are raised to [i] (see 14b). In Portuguese clitic sequences, the coda consonant /s/ of 1pl.o nos and 2pl.o vos is replaced by /l/ before a 3rd person clitic (14b). This process also takes place if 3rd person clitics encliticise to verb forms which end in /s/, /z/ or /r/ and is known as “neutralisation”. (14) Opaque clitics III: clitic sequence internal phonological processes a.
Italian vowel lowering mi → me ti → te ci → ce / _ {lo,ne} vi → ve gli → glie
Properties of clitic sequences in Romance
b. Romanian vowel raising ne → ni va[ → vi / _ {te, îl, îi, le} le → li c.
Portuguese coda neutralisation {nos, vos} →{nol, vol} /_{o, a, os, as}
Other Portuguese clitic sequences result in only one syllable, because the Wnal vowel of the Wrst clitic is either deleted (as with me, te and lhe) or replaced (as with lhes) (15a). Romanian also shows deletion and reduction of vowels which is obligatory within clitic sequences but optional in clitic host combinations. (15) Opaque clitics IV: clitic sequence internal prosodic processes a.
Portuguese vowel deletion me → m te → t / _ {o, a, os, as} lhe → lh lhes + {o, a, os, as}→ {lhos, lhas}
b. Romanian obligatory vowel reduction ne → [neÛ] le → [leÛ] îmi → [mj] /_ V ît6 i → [tsj] îi → [j] c.
Romanian obligatory vowel deletion îl → [l] îi → [j] /_ V ma[ → [m] / V_ va[ → [v]
We have seen in chapter 2 that the morphophonological behaviour of clitics and clitic sequences is also unique with regard to their hosts and does not resemble processes which are attested in aYxal morphology. Thus, clitic combinations have to be regarded as a separate issue in linguistic theory. It is worth noticing that the properties of clitic combinations not only pose a problem, but seriously challenge the explanatory power of a framework which
135
136 Clitic sequences
denies the independence of a morphological component. This is especially true for any theory which attempts to assign all morphological processes to transformational syntax.
4.2. Approaches to clitic sequences in recent research One of the Wrst attempts to account for the phenomena observed with clitic combinations in generative grammar, Perlmutter (1970), has a long-standing tradition. Perlmutter suggests that clitic combinations obey so-called “Surface Structure Constraints” which demand a templatic order of clitics within the sequence. The clitic sequence template consists of diVerent slots, each of which can be occupied only once (see the examples in 6 on p. 131 above). Combinatory restrictions like those mentioned above result from the fact that both clitics belong to the same templatic slot, so that they cannot co-occur within the same sequence.4 In addition, the substitution of the 3rd person indirect object clitic by the reXexive clitic se in Spanish is captured by the socalled “spurious se rule” (Perlmutter 1970:191) which states that the indirect object clitic should be replaced by se. Surface Structure Constraints are thus a type of formalised description of the properties of clitic sequences but they do not go beyond that as far as their explanatory power is concerned (compare also Hetzron 1976 for similar criticism). Furthermore, a detailed examination of clitic sequences shows that the Surface Structure Constraints do not suYce to correctly exclude all prohibited combinations nor to predict all grammatical combinations (see Emonds 1975; Herschensohn 1980 for French; Hetzron 1976 for Spanish). It is also diYcult to extend the Surface Structure Constraints to other languages (see Wanner’s 1977 detailed examination of the clitic order in Italian). Until recently, Perlmutter’s Surface Structure Constraints were adopted to account for the order of clitics (see, among many others, Avgustinova 1997; Monachesi 1995; Sportiche 1996; Sylla 1979).5 In the last decade, however, the view that morphology is an independent grammatical module with particular types of rules, advocated e.g., in Minimalist Morphology (Wunderlich & Fabri 1996), Distributed Morphology (Halle & Marantz 1994) and Paradigm Function Morphology (Stump 1993, 1997), has re-emerged. In addition, Optimality Theory (McCarthy & Prince 1993, 1994, 1995; Prince & Smolensky 1993) has introduced the concept of violable constraints and the mechanism of con-
Approaches to clitic sequences in recent research 137
straint ranking, which allows phonological and morphological requirements to interact. Thus, recent theoretical developments oVer new possibilities in dealing with problematic issues and clitics are currently even more an object of linguistic research than ever before. Bonet (1995) and Harris (1995) both propose an analysis of Romance pronominal clitics within the framework of Distributed Morphology. For them, clitics constitute morphological feature structures to which morphological impoverishment and readjustment rules apply. They furthermore assume that the linearisation of morphological elements also takes place in the morphological component. Bonet suggests that insertion and delinking of morphosyntactic features lead to the opaque forms of clitics. For example, the opaque Italian clitic sequence ci si results from the insertion of the person agreement feature [1] – for the Wrst person – into the combination of impersonal si and reXexive si, which are represented by the underlying morphosyntactic feature structures arg-[pl] and arg respectively: (16) Italian: substitution of impersonal si cl cl cl cl arg arg → arg arg insertion [pl] [pl] 1 si si ci si
In the example above, Bonet assumes that the underlying feature structure for the impersonal reXexive clitic contains the number feature [pl]. If this feature structure is combined with the reXexive clitic, a morphological rule applies which inserts the feature [1] and links it to the impersonal clitic. The resulting feature structure, then, is spelled out as ci, which is the Wrst person plural clitic. In the Spanish example below, Bonet implements Perlmutter’s spurious se rule by the concept of delinking, i.e. the node containing the person agreement feature 3rd and its daughter obl are delinked. Bonet assumes that, in contrast to Italian, bare [arg] nodes are incompatible with agreement features in Spanish. Subsequently, ([pl]) is deleted because of structure preservation.
138 Clitic sequences
(17) Spanish: substitution of 3io le cl cl cl cl cl cl delinking stray arg 3rd arg 3rd → arg arg 3rd → arg arg 3rd erasure ([pl]) obl ([pl]) le lo lo se lo
To account for the order of clitics within a clitic cluster, Bonet (1995) assumes templates which supply diVerent slots for diVerent clitics or feature structures respectively. Harris (1995), who conWnes himself to the analysis of Spanish clitics, replaces these templates by Wlters. In both analyses, the surface forms of clitic sequences result from morphophonological “spell out” rules. The problem with this approach is that it remains unclear why a particular feature is inserted or delinked rather than another feature and that the theory predicts insertion or delinking processes that are unattested. In Distributed Morphology, the only purpose of the morphological component seems to be to re-organise the feature structures provided by the syntax so that the correct surface form emerges. It is not clear how the morphological component can be restricted. Another model for morphology is that of Paradigm Function Morphology (Stump 1993, 1997). Spencer (2000) argues that clitic combinations are best analysed within this framework. He shows that in Bulgarian, clitic clusters and inXectional aYxes form paradigms and adds the notion of the “clitic cluster” to the model of Paradigm Function Morphology. He then derives the internal order of clitic clusters and the sound-meaning correspondences by the mechanisms provided by the theory. Rules of exponence, called “realisation rules”, describe the relation between morphosyntactic features and their phonological representation. For clitics, these rules are deWned over the VP. In the Bulgarian clitic cluster, 1st and 2nd person auxiliary clitics are initial, then follow dative and accusative clitics and Wnally 3rd person singular auxiliary clitics (see 18a). For each of the clitic types, the realisation rules in (18b) generate a speciWc slot which is preWxed to a sequence X, where X is also allowed to be empty.
Approaches to clitic sequences in recent research 139
(18) Bulgarian: Realisation rules for the clitic cluster a. aux1,2 dat acc aux.3sg b. RI[CASE:Acc, P/N:1sg...](X) ⇒ [me]/[te]/[go] ... X (acc clitics) RII[CASE:Dat, P/N:1sg...](X) ⇒ [mi]/[ti]/[mu] ... X (dat clitics) RIII[TNS:Perf, P/N:1sg ...](X) ⇒ [s6m]/[si]/[e] ... X (aux clitics) (Spencer 2000:368)
The internal sequence of clitics in clusters is derived from the Paradigm Function Schema (see 19), which maintains the order in which the diVerent realisation rules must apply. For Bulgarian, the Paradigm Function Schema splits up into the Functional Composition Default (19a) and an additional, more speciWc function which is restricted to 3rd person perfective auxiliary clitics (19b). The Functional Composition Default states that RI takes place before RII, which in turn takes place before RIII so that accusative clitics are nearer to the verb than dative clitics, which in turn are nearer to the verb than auxiliary clitics. As regards 3rd person auxiliary clitics, the presence of such clitics causes a change in the order of the realisation rules according to the function in (19b), so that the 3rd person auxiliary clitic occurs on the right edge of the cluster. (19) Paradigm Function Schema a. Functional Composition Default: PF(σ) = def RIII(RII(RI(σ))) b. Extra function for 3rd person auxiliary clitics: For σ = [TNS:Perf, PERS:3sg] PF(σ) = def RII(RI(RIII (σ))) (Spencer 2000:369-370)
The interrogative clitic li, however, cannot be analysed within this framework since its position is not morphologically, but rather prosodically determined. Spencer proposes that such instances of conXicting requirements of diVerent grammatical modules can be best accounted for by the mechanisms of Optimality Theory as suggested by Legendre (1996, 1999, 2000a,b,c).6 Paradigm Function Morphology suVers from similar faults as Distributed Morphology. Although the mechanisms of rule ordering and the realisational rules are rather complex, it is not clear what they contribute to the understanding of clitic sequences beyond the straightforward description of what they look like. Moreover, the failure to integrate prosodic conditions on the placement of certain clitics into the framework and the additional use of Optimality Theory
140 Clitic sequences
in this respect raises the question why an Optimality Theory analysis (as e.g. Legendre’s analysis for the same data) should no be used throughout. For the opaque clitics in Romance discussed above, an analysis within Paradigm Function Morphology seems to be impossible. Below, an Optimality-theoretical solution is proposed for clitic sequences. It has become clear from the discussion so far that the position of clitics in a sentence and the occurrence of clitic clusters depends on phonological, morphological, and syntactic properties. An advantage of Optimality Theory regarding clitic phenomena is that constraints from diVerent grammatical modules may be considered simultaneously. The analysis is much in the line with Gerlach (1998a,b), Grimshaw (1997, 2001) and Legendre (1999, 2000b).
4.3. Interaction of morphology and syntax in clitic sequences As pointed out in section 3.1, the evaluation of clitic sequences is considered parallel to the evaluation of the clitic most faithful to the morphosyntactic information of each of the argument roles (the “best” clitic in Grimshaw’s terms). The phonological form of this most faithful clitic is subject to both phonological and morphological constraints on optimal outputs, which may cause deviations from the underlying phonological structure. Phonological faithfulness constraints compare the phonological representation of the most faithful clitic to other possible phonological output representations so as to limit the deviations to a minimum. A premise for the comparison of clitic sequences which results from the combination of io and do clitics is that the syntactic and semantic conditions which require that such clitics occur are met. In all Romance languages, clitics obligatorily denote previously mentioned discourse participants. Clitics then develop into agreement markers and occur in speciWc syntactic and semantic environments together with coreferential pronouns or DPs. In section 5, I will show the nature of these syntactic and semantic conditions and provide an analysis. With regard to the analysis of the sequences here, I generally assume that these conditions are met. Moreover, the evaluation of the morphosyntactically most faithful clitic as explained in chapter 3 is presupposed and its phonological representation as well as the schematic lexical entry for a verb is considered a part of the input. The relevant set of output candidates, i.e. the set of possible clitic sequences, is formed from elements of the input. A schematic input and
Interaction of morphology and syntax in clitic sequences
output representation for the evaluation of a object clitic sequence is shown in (20a) and (20b), respectively. (20) a.
Input of the morphosyntactic evaluation:
Ca1, Cb2, V3 with Ca, Cb ∈ C C = {/ϕ/; [(+hr), (βlr)], ([+rX]), ([+loc]), ([+1]), ([+2]), ([+pl])} and λz λy λx V (x,y,z), Ca= y, Cb = z b. Output set for the morphosyntactic evaluation CS = {[Ca1-Cb2]-V3; [Ca1-∅2]-V3; [∅1-Cb2]-V3; [∅1-Ca2]-V3; [Ca1-Cd2]-V3; ...} with Ca, Cb, Cd ∈ C C = {/ϕ/; [(αhr), (βlr)], ([+rX]), ([+loc]), ([+1]), ([+2]), ([+pl])} and ∃ z ∃ y λx V (x,y,z) In (20a), C is the set of clitics in the respective language, Ca and Cb are the most faithful clitics for direct object and indirect object. Each clitic in C is identiWed by a phonological representation /ϕ/, which encodes its phonological form, including a marker indicating it is prosodically defective. The lexical entry contains some additional case and agreement features. The output set (20b) consists of all possible combinations of the input elements, but also includes every other — less faithful — combination of the respective verb with (other) clitics or clitic sequences, as for instance Cd or ∅. In (20), the correspondence relation is marked by super-indices. The tableaux below display only relatively good candidates, already evaluated by highly ranked constraints. Romance clitic sequences are often limited to two clitics, in particular if no auxiliary and/or subject clitics belong to the inventory. Regarding this limitation to two internal argument clitics, not only the morphology-syntax interface has to be considered but also constraints from all other linguistic components, which will mainly be discussed in chapter 5. Firstly, morphosyntactic clitic placement constraints require that clitics cluster together, either to the left or to the right of their morphosyntactic host. These constraints secondly interact with the prosodic requirement that clitic sequences group together with their phonological host, which is not necessarily identical to their morphosyntactic host. Thirdly, integration into a prosodic word is the more promising the less phonological material the clitic sequence contains. Therefore, prosodic constraints require that clitic sequences are as minimal as
141
142 Clitic sequences
necessary, i.e. preferably at most one syllable. Finally, the information structure of the sentence also comes into play. Since pronominal clitics represent topics and hardly more than two internal arguments are topical, sequences of more than two internal clitics are not to be expected. Of these issues only the third, i.e. the constraints on prosodic minimality of clitic sequences, will be discussed in this chapter (see section 4.4.1). Before we come to the requirements of prosodic minimality, the order of clitics within clitic sequences (see 8 on p. 132) will be explained by morphological alignment constraints in section 4.3.1. The combinatory restrictions will be shown to result from the higher order of these constraints with respect to the syntactic max(arg)M constraints which require the occurrence of clitics (see chapter 3). In section 4.4, it will be pointed out that deviations from the order of clitics and opaque clitics (see 12 to 15 on pp. 133–135) occur when phonological constraints — mostly similar to the OCP — inXuence the clitic sequence. Variation between the languages results from diVerent constraint rankings on the one hand and diVerent clitic inventories on the other.
4.3.1 The order of clitics I will start the analysis with the “core of the clitic sequence”, that is, clitic sequences which consists of two clitics, marking direct and indirect objects (in reXexive and non-reXexive use). After that, the additional clitics which are attested in the individual languages will be integrated into the analysis. Let us look at the Italian clitic sequence paradigm in (21) as a Wrst example. As pointed out above, the Romance languages are very similar with respect to the ordering of clitics and the prohibition of certain combinations (see 8 on p. 168). As can be seen in (21), Italian io clitics always precede do clitics and 1o and 2o clitics neither combine with each other nor with 3io clitics so that in this case all of the predictions above are met. The empty cells in the paradigm mean that a combination of clitics is not possible. If such combinations are required syntactically, one of the elements is realised by a full pronoun and the other one as a clitic (see below).
Interaction of morphology and syntax in clitic sequences 143
(21) The Italian object clitic sequence paradigm io do 1sg 2sg 3sg.m 3sg.f 1pl 2pl 3pl.m 3pl.f
mi ti lo la ci vi li le
1sg mi
2sg ti
me lo me la
te lo te la
me li me le
te li te le
3sgm gli
3sgf le
1pl ci
2pl vi
3pl gli
glielo gliela
ce lo ce la
ve lo glielo ve la gliela
glieli gliele
ce li ce le
ve li ve le
glieli gliele
We can observe that the order of clitics within the clitic sequence mainly depends on the markedness relation between the two clitics. The clitics have to be marked diVerently with respect to case and person, and, in general, the more marked clitic, i.e. the clitic which is high either in the animacy hierarchy (see 22) or in the argument hierarchy (see 23 below), occurs on the left edge of a clitic sequence.7 (22) Animacy hierarchy (according to Siewierska, 1988) 1 > 2 > 3 human > higher > other > anorganic > abstracts animals organisms matter (23) Argument hierarchy: s > io > do
As regards the argument hierarchy, morphological markedness does not automatically reXect semantic markedness. Subjects, although linked to the morphological nominative, refer to the highest role according to the semantic form of the verb and are thus considered higher in the argument hierarchy than objects. The relation of indirect and direct objects with respect to the argument hierarchy reXects their morphological markedness; we can say that elements which bear the feature [+lr] (i.e. indirect objects) are on the left edge of a clitic sequence.8 As to the animacy hierarchy, only the part which refers to person is relevant to the order of clitics in clitic sequences. Both 1st and 2nd person are considered to be marked, and alignment constraints exist for each of the features. The order of clitics within the clitic sequence (‘cs’ in the constraints) is evaluated entirely by the constraints in (24).
144 Clitic sequences
(24) Morphological alignment constraints reXect morphological markedness align-l(+1,cs): 1st person clitics are initial in a clitic sequence. align-l(+2,cs): 2nd person clitics are initial in a clitic sequence. align-l(+lr,cs): Indirect object clitics are initial in a clitic sequence.
These constraints imply a conXict between marked person values and the feature [+lr] because only one of them can be initial in a clitic sequence. It is exactly this conXict that induces gaps in the clitic sequence paradigm, as will be shown below. The following predictions concerning optimal forms result from the interaction of these morphological constraints and the syntactic constraints explained in chapter 3. If all alignment constraints are ranked above max(io)M and max(do)M, it is predicted that 1o and 2o cannot combine with each other and that they cannot combine with 3io. The order of clitics within the cluster is io < do. The constraints refer to the clitic sequence as the domain of alignment and presuppose that it is a morphological unit. Morphologically more marked elements, i.e. elements whose lexical entries contain [+] valued case and person features, are positioned to the left in this domain. A remark is necessary regarding the nature and interpretation of the alignment constraints. For each of the elements to be aligned, a parameter determines the alignment site. In the constraints above, these parameters are invariantly l[eft] for both elements, the respective feature and the clitic sequence; their extended versions are therefore of the type align(x,l,y,l) (i.e. “the left edge of x is aligned with the left edge of y”). Two distinct interpretations of alignment constraints prevail. (25) Two intrepretations of alignment a. directional alignment [[ ]x ]y / [ [ ]x]y ‘x is the initial / Wnal element of a constituent y’ ~ x is edgemost within the domain y b. mutual alignment (i) [[ ]x ]y / [ [ ]x]y ‘the left edge of x is aligned with the left edge of y’ / ‘the right edge of x is aligned with the right edge of y’ ~ edgemost if y is regarded a domain
Interaction of morphology and syntax in clitic sequences 145
(ii) [ ]x [ ]y / [ ]y [ ]x ‘the right edge of x is aligned with the left of y’ / ‘the right edge of y is aligned with the left edge of x’ ~ x and y are adjacent constitutents / elements
First, the constraints are interpreted as determining some kind of directional alignment of an element within a speciWc domain as: “x is the initial/Wnal element of a constituent y”. The constraints in (24) above are of that kind. They are violated if an element “x” (e.g. [+lr]) exists in the output that is not initial/Wnal in the respective domain “y” (i.e. cs). If the constituent “y” (i.e. cs) is part of the output but “x” (e.g. [+lr]) is not, the constraint is not considered to be violated. Legendre (e.g. 1996, 2000b) therefore more precisely calls this kind of alignment constraints edgemost with respect to a speciWc domain (see 25a). Other alignment constraints select diVerent directional parameters for the elements to be aligned, for instance align(x,r,y,l) (i.e. “the right edge of x is aligned with the left edge of y”) interpreted as “x is adjacent to y and x precedes y” and align(x,l,y,r) (i.e. “the left edge of x is aligned with the right edge of y” interpreted as “x is adjacent to y and x follows y”, see 25b). The alignment is considered to be mutual. Consequently, it is not reasonable to speak of the “domain of alignment” with these constraints. This type of alignment constraint will be brieXy discussed in section 4.4.2 and in detail in chapter 5. For the individual evaluations, let us consider Wrst the combination of two 3rd person clitics in (26). High ranking of align-l(+lr,cs) requires that 3io precedes 3do. Therefore, candidate (26b) with the reverse order imposes a fatal violation. Leaving out one of the clitics, as in (26c) and (26d), causes a fatal violation of max(arg)M.9 Candidate (26a) violates none of the constraints and is thus optimal.10 (26) Italian: combination of two 3rd person clitics
lexical entries of the clitics: gli: +hr,+lr lo: +hr λz λy λx V(x,y,z) a. glie lo b. lo gli c. gli d. lo
align-l(+lr,cs)
max(arg)m
*! *! *!
146 Clitic sequences
Similarly, align-l(+1,cs) and align-l(+2,cs) require that both 1o and 2o are leftmost in combinations with 3do. This is illustrated in the tableau in (27) for the combination of 1pl.o and 3do. (27) Italian: combination of 1pl.o and 3do
lexical entries of the clitics: ci: +1,+pl,+hr lo: +hr align-l(+1,cs) λz λy<+1,+pl> λx V(x,y,z) a. ce lo b. lo ci *! c. ci d. lo
max(arg)m
*! *!
So far, no ranking of the morphological alignment constraints and the syntactic requirement for clitics postulated by max(arg)M is derivable. However, such a ranking is necessary regarding the prohibited combinations. One might recall that neither 1o nor 2o clitics can combine with 3io. If such a combination is intended, the optimal candidate is not a clitic sequence, but rather a single clitic. The same applies to mutual combinations of 1o and 2o. Consider in this respect the evaluations of the combinations of 1pl.o ci and 3io gli in (28) and the combination of 1pl.o ci and 2o ti in (29). (28) Italian: combination of 1pl.o and 3io lexical entries of the clitics: ci: +1,+pl,+hr gli: +hr, +lr align-l align-l λz<+1,+pl> λy λx V(x,y,z) (+1,cs) (+lr,cs) a. ci gli *! b. gli ci *! c. gli d. ci (29) Italian: combination of 1pl.o and 2o lexical entries of the clitics: ci: +1,+pl,+hr ti: +2, +hr λz<+1,+pl> λy<+2> λx V(x,y,z) align-l align-l λz<+2>λy<+1,+pl> λx V(x,y,z) (+1,cs) (+2,cs) a. ci ti *! b. ti ci *! c. ci d. ti
max(arg)m
* *
max (arg)m
* *
Interaction of morphology and syntax in clitic sequences 147
In both tableaux, the morphological alignment constraints are ranked above max(arg)M. The requirement that both clitics should appear on the left edge of the clitic sequence ranks higher than the requirement that the arguments are realised by clitics, but is impossible to satisfy. Hence, the combination fails. Basically, the Romance languages obey the same order and combinatory restrictions for clitics. Deviations from these restrictions, as observed in Romanian and French, result from minimal re-ranking and a limited number of additional constraints. We can therefore conclude that the same ranking of morphological alignment and morphosyntactic faithfulness constraints underlies all Romance clitic sequences (see 30). (30) Basic ranking of morphological constraints on Romance object clitic sequences align-l(+1,cs), align-l(+2,cs), align-l(+lr,cs) >> max(arg)M Let us now turn to the other internal argument clitics, which are also part of the clitic sequence. The French and Italian inventory, in particular, have a number of non-object clitics (see 9 on p. 132 above). These clitics are also positioned in the clitic sequence according to alignment constraints (see 31a and 32a) whose ranking leads to the order of the clitics (31b and 32b). (31) Italian clitic order in the entire clitic sequence a. align-l(+lr,cs) align-l(+1,cs) align-l(+2,cs) >>max(arg)M >>align-l(loc,cs) >>align-l(rX,cs) >>align-l(+hr,cs)>>align-l(imp,cs) b. 1/2/3io > loc > rfl > 3do > imp > prt mi/ci/ti/vi/gli ci/vi si lo/la/li/le si ne
148 Clitic sequences
(32) French clitic order in the entire clitic sequence a. align-l(+1,cs) align-l(+2,cs) align-l(rX,cs) >>max(arg)M >>align-l(+lr,cs)>>align-l(loc,cs)>>align-l(+hr,cs) b. 1/2/rfl > 3io > loc > 3do > prt me/nous/te/vous/se lui/leur y le/la/les en If the combination of clitics is prevented because of morphological restrictions, only one of the clitics appears instead of the clitic sequence, and the other argument is realised by a full pronoun. In the next section it will be shown how morphological constraints inXuence the choice between clitic and full pronoun.
4.3.2 Clitic choice when combinations are prohibited In Italian, as indicated by the two optimal candidates resulting from the evaluation in the tableaux in (28) and (29), any of the objects may cliticise (see 33, clitics and their English pronominal counterpart are in bold letters); the choice which one actually does depends on discourse-semantic constraints which are discussed in chapter 5. (33) Italian combination of 1st person direct object and 3rd person indirect object a.
Claudia presenta noi a Stefano.
b. *Glielo presenta. c. Gli presenta noi. d. Ci presenta a lui.
‘Claudia presents us to Stefano.’ ‘She presents us to him.’ ‘She presents us to him.’
The same is the case for the clitic choice in Portuguese and Piattino. In the other Romance languages, however, the choice of the clitic is not free if the combination is prohibited. Instead, the direct object is represented by a clitic, while the indirect object is represented by a full pronoun or DP which is either inXected for dative case or occurs together with a dative marking preposition, as a comparison of the examples in (34) to (36) from French, Romanian and Spanish shows.11
Interaction of morphology and syntax in clitic sequences 149
(34) French a.
*[1o 3io]cs: [1o]cs (dat) 3prn *Tu me lui présenteras. Tu me présenteras à lui. *Tu lui présenteras moi. ‘You introduce me to him.’
c.
*[1o 2o]cs: [2o]cs (dat) 1prn *Elle me te recommande. Elle te recommande a moi. *Elle me recommande toi. ‘She recommends you to me.’
b. *[1o 2o]cs: [1o]cs (dat) 2prn *Il me vous recommande. Il me recommande à vous. *Il vous recommande moi. ‘He recommends me to you.’
(35) Romanian a. *[2o 3io]cs: [2o]cs 3prn.dat b. *[1o 2o]cs: [1o]cs 2prn.dat *I va[ recomand. *Ni va[ recomanda[ . Va[ recomand lui. Ne recomanda[ voua[ . *Îi recomand pe voi. *Va[ recomanda[ pe noi. ‘I recommend you (pl) to him’ ‘S/he recommends us to you (pl).’ c. *[2o 1o]cs: [2o]cs 1prn.dat *Ni va[ recomanda[ . Va[ recomanda[ noua[ . *Ne recomanda[ pe voi. ‘S/he recommends you to us.’ (36) Spanish a.
*[2do 3io]cs: [2do]cs (dat) 3prn *Te le recomiendo. Te recomiendo a ella/la empresa. *Le recomiendo a ti. ‘I recommend you to her/the company’
b. *[1do 2io]cs: [1do]cs (dat) 2prn *Te me recomienda. ?Me recomienda a ti. *Te recomienda a mi. ‘S/he recommends me to you.’ c.
*[1io 2do]cs: [2do]cs (dat) 2prn *Me te recomienda. ?Te recomienda a mi. *Me recomienda a ti. ‘S/he recommends you to me.’
150 Clitic sequences
In the examples above, the choice of the indirect object clitic instead of the prohibited clitic sequence makes the form ungrammatical. In French and Romanian, the direct object clitic is always chosen as the grammatical output. In Spanish, combinations of 1st and 2nd person are only marginally accepted, which might result from the diVerent functions of the preposition a. The Spanish preposition a is not only a case marker, like the preposition á in the French examples above, but also marks direct objects with animate reference. Its lexical entry therefore contains the disjunction of [+hr,+lr] v [+hr,+animate]. However, the diVerence in grammaticality to the clitic sequence or the indirect object clitic instead of the direct object is still signiWcant; both are not accepted at all. The choice of the direct object in these cases cannot be explained by the ranking of max(io)M above max(do)M, since indirect objects are more likely to be represented by clitics than direct objects in contexts which are not morphologically restricted. For Spanish, this is shown in (37).12 (37) European Spanish a. [1o 3f.do]cs Me la recomienda. ‘S/he recommends her to me.’ *La recomienda a mi. (intended: ‘S/he recommends her to me.’) Me recomienda a la mujer. ‘S/he recommends the woman to me.’ b. [3io 3do]cs Se la recomienda (al uomo). ‘S/he recommends her to him (to the man).’ *La recomienda al’ uomo. (intended: ‘S/he recommends her to the man.’) Le recomienda a la mujer. ‘S/he recommends the woman to him.’
If the clitic sequence is prevented, the direct object may be realised either as a clitic within the clitic sequence or as a full DP; the indirect object must be realised as a clitic, which also may occur in addition to a full DP. There is thus a choice between the clitic sequence and the indirect object clitic (which is eventually determined by referential properties of the arguments) but no choice between the clitic sequence and the direct object clitic. A ranking like max(do)M >> max(io)M would predict the contrary. Therefore, such a ranking cannot account for the choice of the direct object clitic if the clitic sequence is prohibited, as in the examples in (34) to (36). Instead, I suggest that
Interaction of morphology and syntax in clitic sequences
the markedness of the clitics again plays the crucial role: 1st and 2nd person arguments are more likely to be morphologically encoded by agreement markers than 3rd person arguments. This generalisation is captured by the constraint max(person)M (see 38).13 (38) max(person)M Person features from the input are encoded morphologically in the output.
Ranking of max(person)M higher than or as high as max(arg)M leads to 1o (or 2o respectively) as the optimal candidate in combinations of 1o (2o) and 3io. This is shown for Spanish in (39). The diVerence between the candidates in (39c) and (39d) is that the former realises the feature [+lr] morphologically by the clitic and the feature [+1] syntactically by the full pronoun, while the latter realises the feature [+1] morphologically and the feature [+lr] syntactically. Obviously, the latter is optimal and max(person)M must rank high to exclude the former. That the DP a mi in (39c) in contrast to the DP a la mujer in (39d) lacks the feature [+lr] results from the disjunctive lexical entry of the preposistion a. As explained above, a bears the features [+hr,+animate] if it combines with a(n animate) direct object as in (39c) or the features [+hr,+lr] if it combines with an indirect object as in (39d). (39) Spanish: combination of 1o and 3io lexical entries of the clitics: me: +1,+hr features of the DPs: a mi: +1,+hr λz<+1> λy ∃ x recommend (x,y,z) a. me le recomienda b. le me recomienda c. le recomienda a mi d. me recomienda a la mujer
le: a la mujer:
+hr,+lr +hr,+lr
align-l align-l max max m (+1,cs) (+lr,cs) (person) (arg)m *! *! *! * *
That max(person)M ranks below the alignment constraints results from the evaluation of combinations of 1st and 2nd person objects (see 40). Again, it is the direct object which is realised as a clitic, while the indirect object is realised as a full pronoun. max(person)M is violated by both potentially optimal candidates. It is thus not decisive in the evaluation of the optimal candidate, but it must rank below the morphological alignment constraints.
151
152 Clitic sequences
(40) Spanish: combination of 1o as direct object and 2o as indirect object lexical entries of the clitics: me: +1,+hr features of the DPs: a mi: +1,+hr <+1> <+2> λz λy ∃x align-l recommend(x,y,z) (+1,cs) a. me te recomienda b. te me recomienda *! c. ? me recomienda a ti d. ? te recomienda a mi
te: +2, +hr a ti: +2, +hr,+lr align-l max max (+2,cs) (person)m (arg)m *! * *
* *
The decision in such combinations is provided by a diVerent context-speciWc constraint: max(+lr/+hr).14 This constraint requires that the input feature [+lr] is realised in the output if it belongs to an indirect object, i.e. elements which correspond to indirect objects should not be underspeciWed in the output. In contrast to max(person)M, max(+lr/+hr) does not require morphological realisation of the feature [+lr]. (41) max(+lr/+hr) The feature [+lr] of an input element I has a correspondent in the output, if I also bears the feature [+hr].
The tableaux in (42) and (43) show how the Spanish examples in (36b,c) are evaluated if this constraint takes part in the evaluation.15 The clitics are identical in both examples and only the relation to the argument roles diVers: while 1o refers to the direct object and 2o to the indirect object in (42), 1o refers to the indirect object and 2o to the direct object in (43). In both tableaux, the clitic sequences (i.e. candidates 42a,b and 43a,b) are outranked because of align-l violations. Of the remaining candidates, (42c) and (43c) are better than (42d) and (43d) insofar as the feature [+lr] is part of the output element that corresponds to the indirect object, which is a pronominal PP. In this case, the lexical inventory does not provide the morphological means (i.e. fully speciWed 1io or 2io clitics) to mark the feature, but the PP consisting of the dative marking preposition a together with the full pronouns meets this requirement. Although a is also part of the output candidates (42d) and (43d), it does not mark dative here, but the features [+hr,+animate].
Interaction of morphology and syntax in clitic sequences
(42) Spanish: combination of 1o as direct object and 2o as indirect object lexical entries of the clitics: me: +1,+hr te: +2,+hr features of the DPs: a mi: +1,+hr a ti: +2,+hr,+lr λz<+1> λy<+2> ∃ x max max max align-l recommend (x,y,z) (per)M (arg)M (+lr/+hr) * a. me te recomienda *! b. te me recomienda *! * c. me recomienda a ti * * d. te recomienda a mi * * *! (43) Spanish: combination of 1o as indirect object and 2o as direct object lexical entries of the clitics: me: +1,+hr te: +2,+hr features of the DPs: a mi: +1,+hr,+lr a ti: +2,+hr
a. b. c. d.
λz<+2> λy<+1> ∃ x align-l recommend (x,y,z) me te recomienda *! te me recomienda *! te recomienda a mi me recomienda a ti
max (per)m
max (arg)m
* *
* *
max (+lr/+hr) * * *!
Let us now consider how these constraints inXuence the argument realisation if the clitic sequence is not prevented as exempliWed by (37a,b) above. In (37a), we saw that in combinations of 1st person indirect object and 3rd person direct object, either both objects are realised as clitics or the indirect object is realised as a clitic and the direct object as a full DP. As mentioned above, it is the referential properties of the object which determine whether it is realised as a DP. While topical arguments, for instance, are preferably not realised as full DPs, focused elements must occur as full DPs. Moreover, morphological economy constraints prohibit noun referents with low salience from being realised morphologically, in Spanish particularly for direct objects. These constraints (e.g. max(foc), drop topic, *m-structure see chapter 5) are merged in the tableaux below as ‘is’ (short for ‘information structure’). In Spanish, these constraints only aVect the realisation of the direct object but not the realisation of the indirect object. is is therefore ranked below max(io)M and above max(do)M. Both objects are realised as clitics and the clitic sequence is optimal, as (44) shows, if only the morphological constraints which are introduced above are involved. This is the case, for instance, if the
153
154 Clitic sequences
direct object is topical and has a speciWc reference. Although the optimal candidate (44a) violates max(+lr/+hr), the other candidates are worse. If the direct object is realised as a PP (44c), not only max(+lr/+hr) is violated, but also max(do)M. Even though max(+lr/+hr) is not violated if the indirect object is realised as a PP (44d), this candidate is worse than the sequence because it violates high-ranked max(per)M and, in addition, max(io)M. If the referential properties of the direct object, on the other hand, are in conXict with its realisation as a clitic (e.g. if it is focused and non-speciWc, see 45), the formerly optimal clitic sequence violates the constraint is and is therefore outranked. Candidate (45c) with the direct object as a full DP and the indirect object as a clitic is then the optimal output representation. (44) Spanish: combination of 1o as indirect object and 3do as direct object lexical entries of the clitics: me: +1,+hr lo: +hr features of the DPs: a mi: +1,+hr,+lr al uomo: +hr λz<+topic,+speciWc>λy<+1>∃ x align maxm maxm is maxm max recommend (x,y,z) (per) (io) (do) (+lr/+hr) a. me lo recomienda * b. lo me recomienda *! * c. me recomienda al uomo *! * * d. lo recomienda a mi *! * (45) Spanish: combination of 1o as indirect object and 3do as direct object lexical entries of the clitics: me: +1,+hr lo: +hr features of the DPs: a mi: +1,+hr,+lr al uomo: +hr λz<-speciWc,+focus>λy<+1>∃ x max max max max align is recommend (x,y,z) (per)m (io)m (do)m (+lr/+hr) a. me lo recomienda *! * b. lo me recomienda *! * * c. me recomienda al uomo * * d. lo recomienda a mi *! *
Combinations of two 3rd person clitics are still more complex, because in addition to the constraints on the realisation of morphological and referential features above, a purely phonologically induced substitution of the 3io clitic le by the rfl clitic se comes into play. For the sake of illustration of the clitic choice in such instances, let us assume that this substitution is indicated by the
Interaction of morphology and syntax in clitic sequences
constraint phon in (46) and (47) (the respective morphophonological evaluation will be discussed in detail in section 4.4.2.1). The correct results are achieved as long as phon ranks above max(+lr/+hr). As can be seen in (46), the clitic sequence is optimal if the referential features of the direct object do not present any conXict to is. However, this time it is not max(person)M that causes the representation of indirect object clitic and direct object PP (candidate 46c) to be ungrammatical but max(do)M. This shows that max(do)M ranks above max(+lr/+hr). The same structure wins the competition, on the other hand, if the referential features of the direct object are in conXict with the is constraints. In this case, the clitic sequence loses the competition because of is violations (see 47). (46) Spanish: combination of 3io as indirect object and 3do as direct object lexical entries of the clitics: le: +hr,+lr lo: +hr se: +rX features of the DPs: al uomo: +hr,+lr a la mujer: +hr max max max λz<+topic,+speciWc>λy∃ x recommend (x,y,z) align (io)m is (do)m phon (+lr/+hr) a. se la recomienda b. la se recomienda c. le recomienda a la mujer d. la recomienda al uomo e. le la recomienda
* *
*! *! *!
* *!
(47) Spanish: combination of 3io as indirect object and 3do as direct object lexical entries of the clitics: le: +hr,+lr lo: +hr se: +rX features of the DPs: al uomo: +hr,+lr a la mujer: +hr λz<+focus,-speciWc>λy∃ x max max max align (io)m is (do)m phon (+lr/+hr) recommend (x,y,z) a. se la recomienda *! * b. la se recomienda *! * * c. le recomienda a la mujer * d. la recomienda al uomo *! e. le la recomienda *! *
We can now add the constraints max(person)M and max(+lr/+hr) to the ranking of morphological and syntactic constraints which determine the order and combinatory restrictions on clitic sequences (48).
155
156 Clitic sequences
(48) Spanish: Determining the clitic choice if clitic sequences are prohibited align-l(+1,cs) align-l(+2,cs) >> max(person)M align-l(+lr,cs) >>max(io)M >> is >> max(do)M >> max(+lr/+hr)
French and Romanian show certain exceptions to the general Romance pattern. These exceptions result from the downgrading of one of the alignment constraints which reveals the eVect of a delimited number of additional constraints. While in French, it is an entirely phonological constraint which interacts with alignment (see section 4.4.2.1), only re-ranking of the morphological constraints determines diVerent order and combinatory restrictions in Romanian.
4.3.3 Deviations from the clitic order in Romanian Romanian is exceptional insofar as combinations of 2sg.do with 3io and 1io are attested (highlighted by bold letters in the table in 49). In such combinations, 2sg.do te follows the indirect object clitic. Romanian is somewhat more problematic than the varieties of Spanish which also allow combinations of 1st and 2nd person, because, in Spanish, these combinations are independent of number. (49) The Romanian object clitic sequence paradigm io 1sg 2sg 3sg 1pl do îmi ît6 i îi ne 1sg ma[ 2sg te mi te i te ni te 3sg.m îl mi-l t6 i-l i-l ni-l 3sg.f o mi-o t6 i-o i-o neÛ-o 1pl ne 2pl va[ 3pl.m îi mi-i t6 i-i i-i ni-i 3pl.f le mi le t6 i le i le ni le
2pl va[
3pl le
vi-l vi-o
li te li-l le-o
vi-i vi le
li-i li le
If alignl(+2,cs) is dominated by align-l(+1,cs), align-l(+lr,cs) and max(arg)M, the combination of 2sg.do te and 3io îi (see 50) results in the clitic sequence îi te as the optimal candidate.16
Interaction of morphology and syntax in clitic sequences
(50) Romanian: combination of 2sg.do te and 3io îi lexical entries of the clitics: îi: +hr,+lr te: +2,+hr λz<+2> λy λx V(x,y,z) align-l max(arg)M (+lr,cs) a. îi te b. te i *! c. te *! d. îi *!
align-l (+2,cs) *
The combination of 2sg.do te and 1pl.o ne also results in a clitic sequence because of this ranking (see 51). (51) Romanian: combination of 2sg.do te and 1pl.o ne lexical entries of the clitics: ne: +1,+pl,+hr te: +2,+hr λz<+2> λy<+1,+pl> λx align-l max V(x,y,z) (+1,cs) (arg)M a. ni te b. te ne *! c. ne *! d. te *!
align-l (+2,cs) *
The combination of 2pl.o va[ and 3do îl results in a clitic sequence with the order 2 < 3 just as in the other Romance languages (see 52). (52) Romanian: combination of 2pl.o va[ and 3do îl lexical entries of the clitics: îl: +hr va[ : +2,+pl,+hr λz λy<+2, +pl> λx V(x,y,z) max(arg)M align-l(+2,cs) a. îl va[ *! b. vi-l c. va[ *! d. îl *!
There are, however, a number of problems which must be considered. First, only 2sg.do but not 2pl.o can combine with indirect objects. Second, 2sg.io can only combine with 3do but not with 1do. Both cases can be accounted for if we assume that a more speciWc alignment constraint for 2nd person plural clitics exists which ranks as high as the alignment constraints for 1st person and indirect object clitics. Let us Wrst consider combinations of 2pl.o va[ and
157
158 Clitic sequences
1pl.o ne which — like in the other Romance languages — never result in clitic sequences but in the clitic which encodes the direct object (see 35b,c above). As shown for Spanish, the constraint max(+lr/+hr) is responsible for the choice between the clitics (53c and 54d). In addition, align-l(+2,+pl) and align-l(+1,cs) outrank the clitic sequence (53a,b and 54a,b). In contrast to Spanish, Romanian morphologically distinguishes indirect object personal pronouns (in noua[ and voua[ in 53c and 54d). Direct object pronominal DPs, however, consist of the unmarked (subject) pronoun and the additional case marker pe. (53) Romanian: combination of 2pl.o va[ and 1pl.o ne lexical entries of the clitics: ne: +1,+pl,+hr va[ : +2,+pl,+hr features of the pronoun DPs: pe noi: +1,+pl,+hr voua[ : +2,+pl,+hr,+lr λz<+1,+pl> λy<+2,+pl> ∃ x recommend (x,y,z) a. b. c. d.
ne va[ recomanda[ va[ ne recomanda[ va[ recomanda[ pe noi ne recomanda[ voua[
align align max max align max (+1) (+2,+pl) (per)M (arg)M (+2) (+lr/+hr) *!
*
*! * *
* *
* * *!
(54) Romanian: combination of 2pl.o va[ and 1pl.o ne lexical entries of the clitics: ne: +1,+pl,+hr va[ : +2,+pl,+hr features of the pronoun DPs: noua[ : +1,+pl,+hr,+lr pe voi: +2,+pl,+hr λz<+2,+pl> λy<+1,+pl> ∃ x recommend (x,y,z) a. ne va[ recomanda[ b. va[ ne recomanda[ c. va[ recomanda[ noua[ d. ne recomanda[ pe voi
align align max max align max (+1) (+2,+pl) (per)M (arg)M (+2) (+lr/+hr) *! * * *! * * * * * *!
Combinations of 2pl.o va[ and 3io are also still ungrammatical. The direct object is realised as a clitic instead of the clitic sequence (see 55c) because of the high ranking of max(person)M.
Interaction of morphology and syntax in clitic sequences 159
(55) Romanian: combination of 2pl.o va[ and 3io îi lexical entries of the clitics: îi: +hr,+lr va[ : +2,+pl,+hr features of the pronoun DPs: lui: +hr,+lr pe voi: +2,+pl,+hr λz<+2,+pl> λy ∃ x recommend (x,y,z) a. îi va[ recomanda[ b. va[ i recomanda[ c. va[ recomanda lui[ d. îi recomanda[ pe voi
align-l align-l (+lr,cs) (+2,+pl,cs)
max max (per)M (arg)M
*! *! *!
* *
For Romanian, the following partial constraint rankings are relevant.17 (56) Romanian: Determining clitic order and clitic choice align-l(+1,cs), align-l(+2,+pl,cs) align-l(+lr,cs)
align-l(+2,cs) M>>max(arg)M>> >>max(per) max(+lr/+hr)
The question arises whether combinatory restrictions can be analysed alternatively by diVerences in markedness. One might recall that 1st and 2nd person clitics in Romance are usually unmarked for the case feature [lr]. In Romanian, underspeciWcation of [lr] is conWned to the plural clitics. These unspeciWed clitics still cannot combine. 1st and 2nd person singular clitics, on the other hand, are distinct with respect to indirect and direct object. Of these clitics, it is only 2sg.do which is found in combinations with 3io or 1io clitics. One could therefore propose that combinatory restrictions of 1st and 2nd person ambiguous clitics are integrated into the analysis by additional morphological markedness constraints.
4.3.4 Markedness hierarchies and clitic order The markedness constraints which are needed do not have to relate the markedness of one of the clitics to a position within the clitic sequence (as the alignment constraints above do). Instead, they must prevent 1st and 2nd person clitics from occurring in combinations with 3io clitics if their interpretation is not distinctively either direct or indirect object. That means that clitics with diVerent ranks in the person hierarchy should also display a diVerent rank in the argument hierarchy if they combine. SpeciWcally, that means that in combinations of a 1st (2nd) person and a 3rd person clitic, the
160 Clitic sequences
1st (2nd) person clitic should encode an argument which is higher in the argument hierarchy than the 3rd person clitic (see also Parodi 1998a who argues for that view)18. This is achieved above by the requirement that elements which are high in either of the hierarchies occur on the left edge of a clitic sequence.19 Another account of the mutual relation of diVerent hierarchies has been put forward in recent Optimality-theoretical analyses (see Aissen 1999, 2000; Stiebels 2000a,b) which employ the concept of harmonic alignment.20 As to combinatory restrictions on clitic sequences, Artstein (1998) shows that harmonic alignment is indeed a useful means for Sesotho data, and Grimshaw (2001) suggests that the Romance data can be analysed in a similar way. Let us therefore consider whether harmonic alignment of the hierarchies above also provides an analysis for the combinatory restrictions in Romance clitic sequences. The relevant parts of the person and argument hierarchy are repeated in (57). As suggested by Artstein (1998), 1st and 2nd person are merged into one position in (57a), which seems reasonable considering that 1st and 2nd person clitics behave similarly. (57) Relevant hierarchies for clitic sequences a. person: 1, 2 > 3 b. argument hierarchy: s > io > do
Harmonic alignment of these hierarchies leads to the new preference hierarchies in (58). Elements which are high on the person hierarchy are aligned with elements which are high on the argument hierarchy. As a result, 1s, 2s and 3do are most harmonic, while 1do, 2do and 3s are least harmonic. These hierarchies are then translated into rankings of markedness constraints (58b). The constraint ranking for 2nd person, for instance, is interpreted in a way that a 2do is more marked than a 2io which in turn is more marked than a 2s. As to 1st and 2nd person, there are two identical scales since they share one position in the markedness hierarchy in (57a). (58) Harmonic alignment of person and argument hierarchy a. harmonic alignment b. constraint ranking 1s ⊃ 1io ⊃ 1do *1do>> *1io >> *1s 2s ⊃ 2io ⊃ 2do *2do>> *2io >> *2s 3do ⊃ 3io ⊃ 3s *3s >> *3io >> *3do
By local conjunction (see Smolensky 1995) it is then possible to refer to combinations of elements which occur, for instance, in clitic sequences. As to
Interaction of morphology and syntax in clitic sequences
the constraint rankings which reXect markedness hierarchies, Aissen (1999) suggests that local conjunction of the constraints preserves the ranking of the subhierarchies.21 Local conjunction of the hierarchies above leads to the constraint rankings in (59) (subjects are neglected), which state that a 3io in combination with a 1do (2do) is worse than a 3do in combination with a 1io (2io). (59) Local conjunction of markedness rankings a. *3io & *1do >> *3do & *1io b. *3io & *2do >> *3do & *2io
High ranking of *3io & *2do then causes the combination of 2pl.o and 3io to be ungrammatical just as required (compare 60 to 55). (60) Romanian: combination of 2pl.o va[ and 3io îi lexical entries of the clitics: îi: +hr,+lr va[ : +2,+pl,+hr features of the pronoun DPs: lui: +hr,+lr pe voi: +2,+pl,+hr λz<+2,+pl> λy ∃ x align-l max max *3io & *2do M recommend(x,y,z) (+lr,cs) (per) (arg)M a. îi va recomanda[ *! b. va[ i recomanda[ *! * c. va[ recomanda[ lui * d. îi recomanda[ pe voi *! *
The problem with this constraint ranking, however, is that the grammatical combination of 2sg.do and 3io is also prevented (compare 61 to 50 above). (61) Romanian: combination of 2sg.do te and 3io îi lexical entries of the clitics: îi: +hr,+lr te: +2,+hr features of the pronoun DPs: lui: +hr,+lr pe tine: +2,+hr λz<+2> λy ∃ x align max max align *3io & 2do recommend(x,y,z) (+lr,cs) (per)M (arg)M (+2,cs) a. îi terecomanda[ *! * b. te i recomanda[ *! * c. te recomanda[ lui * d. îi recomanda[ pe tine *! *
Therefore, the analysis in terms of harmonic alignment and local conjunction of markedness hierarchies must be rejected for Romanian clitic sequences. However, one could still argue that at least the markedness constraints in
161
162 Clitic sequences
(58), which result from harmonic alignment of the person and argument hierarchy, could replace other constraints which are suggested above, as for instance constraints of the alignment family or max(person)M and max(+lr/+hr). A closer look at the data reveals that the markedness constraints, though often compatible with the facts, sometimes make wrong predictions about the clitic choice. To analyse the realisation of both objects in sentences where clitic combination are prohibited, we need two further constraints with regard to the realisation of objects: *NP and *cl (see also Artstein 1998). I assume that these constraints are ranked, so that the syntactic structure, i.e. the NP, is more marked than the morphological structure, i.e. the clitic (see 62a). Harmonic alignment to the person hierarchy, then, results in the constraint rankings in (62b). By local conjunction of the rankings (i) and (iv), combinations of 1st person indirect objects and 2nd person direct object are evaluated (62c) and, by local conjunction of the rankings (ii) and (iii), combinations of 1st person direct objects and 2nd person indirect objects (62d). (62) a.
Harmony scales cl ⊃ NP 1io ⊃ 1do 2io ⊃ 2do
b. Harmonic alignment 1io cl ⊃ 1io NP 1do NP ⊃ 1do cl 2io cl ⊃ 2io NP 2do NP ⊃ 2do cl
Constraint ranking *NP >> *cl *1do >> *1io *2do >> *2io
*1io NP >> *1io cl *1do cl >> *1do NP *2io NP >> *2io cl *2do cl >> *2do NP
Local conjunction (1) / (4) *1io NP & *2do NP *2do cl & *1io NP >> *2do cl & *1io cl
(i) (ii) (iii) (iv)
c.
>> *1io cl & *2do NP
d. Local conjunction (2) / (3) *1do NP & *2io NP *1do cl & *2io NP >>
>> *1do NP & *2io cl *2io cl & *1do cl
The representation of the direct object by an NP and of the indirect object by a clitic is least marked. This prediction is met exactly if clitic sequences are not morphologically prohibited and information structure prevents the direct
Interaction of morphology and phonology in clitic sequences 163
object being realised as a clitic (see pp. 150 above). If clitic sequences are prohibited, however, it has been shown that direct objects are realised as clitics and indirect objects are not. This can only be guaranteed by additional constraints such as max(+lr/+hr) which require that indirect objects are realised unambiguously. As to the alignment constraints, the local conjunction of markedness constraints in (59) also leads to the prohibition of unwanted clitic sequences, but the alignment constraints are still needed in order to explain the ordering of clitics. The alignment constraints are able to do all the work without the markedness constraints and the latter are therefore dispensable. Moreover, high ranking of the markedness constraints makes an additional kind of constraint necessary which allows that, under speciWc morphological conditions, otherwise prohibited combinations are possible, as for instance in the Romanian singular clitic sequence paradigm. Not only ungrammaticality of certain marked combinations must be accounted for, but also the fact that the very same combination becomes grammatical as soon as certain morphological features are available. So far, we have only concerned ourselves with morphological and morphosyntactic constraints on clitics sequences. There are, however, also phonological restrictions which inXuence them. Below, it will be argued that the deviation of the clitic order in French results from such phonological restrictions.
4.4. Interaction of morphology and phonology in clitic sequences Two kinds of phonological constraints inXuence the combination of clitics. First, prosodic minimality constraints shape clitic sequences. These constraints hardly interact with morphological and morphosyntactic constraints but mainly with general phonological constraints. Second, OCP-like phonological constraints require that clitics within clitic sequences are dissimilar. These constraints interact with the above-mentioned alignment constraints as well as with morphological and phonological faithfulness constraints with the eVect of metathesis, substitution and deletion of entire clitics or their phonological constituents.
4.4.1 Prosodic minimisation of clitic sequences Clitic sequences have to obey certain prosodic minimality conditions, in particular if they are integrated into the prosodic word which is formed by
164 Clitic sequences
their host. From the Romance languages, especially Romanian and Portuguese possess internal clitics (in Selkirk’s 1995 terms, see section 4.1.2 above). The (mutually interpreted alignment) constraints align-l(Pw,stem) and align-r(Pw,stem) can capture the minimality restriction (for similar constraints in clitic combinations see also Legendre 1996, 1999, 2000a,b,c for Balkan languages and Popescu 2000 for Romanian). (63) align-l(Pw,stem) The left edge of a prosodic word is aligned with the left edge of a stem. align-r(Pw,stem) The right edge of a prosodic word is aligned with the right edge of a stem.
Since we are not concerned with the placement of clitics here, these constraints are merged into align(Pw,stem) in the following. Each syllable of a clitic sequence counts as a violation of this constraint. For Romanian, Popescu (2000) shows that it is mainly this constraint which is responsible for obligatory reduction and deletion of vowels within clitic sequences. She demonstrates that the diVerent phonological representations of Romanian clitics are not idiosyncratic but can be derived by interaction of align(Pw,stem) with other phonological and prosodic constraints. Since Romanian has auxiliary clitics in addition to object clitics, there are three diVerent kinds of clitic sequences to be considered. First, combination of indirect and direct object clitics (see 64), second, combinations of direct object and auxiliary clitics (see 65), and third, combinations of indirect object and auxiliary clitics (66). The cells in the tables show the resulting phonological form of the clitic sequences. (64) Romanian: prosodic shape of io-do clitic sequences io 1sg 2sg 3sg 1pl do îmi ît6 i îi ne 1sg ma[ 2sg te [mi.te.] [i.te.] [ni.te.] 3sg.m îl [mil.] [tsil.] [il.] [nil.] 3sg.f o [mjo.] [tsjo.] [jo.] [neÛo.] 1pl ne 2pl va[ 3pl.m îi [mij.] [tsij.] [ij.] [nij.] 3pl.f le [mi.le.] [tsi.le.] [i.le.] [ni.le.]
2pl va[
3pl le
[vil.] [vjo.]
[li.te.] [lil.] [leÛo.]
[vij.] [lij.] [vi.le.] [li.le.]
Interaction of morphology and phonology in clitic sequences 165
(65) Romanian: prosodic shape of do-aux clitic sequences aux do ma[ te îl o ne va[ îi le
1sg 2sg 3m.sg 3f.sg 1pl 2pl 3m.pl 3f.pl
1sg am
2sg ai
3sg a
1pl am
2pl ît6 i
3pl au
[mam.] [teÛam.] [lam.]
[maj.] [teÛaj.] [laj.]
[ma.] [teÛa.] [la.]
[mam.] [matsj.] [mauÛ .] [teÛam.] [teÛatsj.] [teÛauÛ .] [lam.] [latsj.] [lauÛ.]
[neÛam.] [vam.] [jam.] [leÛam.]
[neÛaj.] [vaj.] [jaj.] [leÛaj.]
[neÛa.] [va.] [ja.] [leÛa.]
[neÛam.] [neÛatsj.] [neÛauÛ .] [vam.] [vatsj.] [vauÛ.] [jam.] [jatsj.] [jauÛ .] [leÛam.] [leÛatsj.] [leÛauÛ .]
(66) Romanian: prosodic shape of io-aux clitic sequences aux io 1sg 2sg 3sg 1pl 2pl 3pl
îmi îtL i îi ne va[ le
1sg am [mjam.] [tsjam.] [jam.] [neÛam.] [vam.] [leÛam.]
2sg ai [mjaj.] [tsjaj.] [jaj.] [neÛaj.] [vaj.] [leÛaj.]
3sg a [mja.] [tsja.] [ja.] [neÛa.] [va.] [leÛa.]
1pl am [mjam.] [tsjam.] [jam.] [neÛam.] [vam.] [leÛam.]
2pl 3pl ît6 i au [mjatsj.] [mjauÛ.] [tsjatsj.] [tsjauÛ .] [jatsj.] [jauÛ .] [neÛatsj.] [neÛauÛ .] [vatsj.] [vauÛ.] [leÛatsj.] [leÛauÛ .]
Let us Wrst look at combinations in which the left clitic is vowel Wnal and the right one is vowel initial. These contexts are created in io-do clitic sequences which contain 3f.sg.do o or 3m.pl.do îi and in all o-aux sequences except for combinations with 3m.sg.do îl. Let us start with Popescu’s (2000) evaluation of 2sg.io ît6 i and 3sg.f.do o (see 67). For reasons of clarity, the host verb cumpa[ ra[ ‘s/he buys’ is also shown in the output. In addition to align(Pw,stem), the familiar phonological constraints no hiatus and onset take part in the evaluation. max(ph) requires that phonemes of the input are represented in the output as well.
166 Clitic sequences
(67) Romanian: obligatory deletion of cs initial [G] and reduction of Wnal [i]: 2sg.do ît6 i and 3sg.f.do o /Gtsi/ + /o/ + /cump6r6/ a. [tsjo.] cumpa[ ra[ b. [tsi.o.] cumpa[ ra[ c. [tso.] cumpa[ ra[ d. [Gtsj.o.] cumpa[ ra[
align-l no hiatus onset max(ph) (Pw, stem) * ** * **
*
* * **
* *
Candidate (67a) has the lowest number of violations and is thus optimal under each possible ranking. Therefore, it is not the violation of one particular constraint which is fatal, but the total number of constraint violations. To avoid confusion, the exclamation mark is not used to indicate which violation is fatal. Candidates (67b,d) violate align(Pw,stem) twice and onset as well. Candidate (67b), moreover, violates no hiatus. Candidates (67a,c), on the other hand, violate align(Pw,stem) only once and neither of them violate onset nor no hiatus. Both candidates diVer only in the number of violations of max(ph). While in (67c) two of the input elements are not represented in the output, in (67a) only one of the input elements is deleted. The latter candidate is therefore optimal. In combinations of the singular indirect object clitics îmi, ît6 i, îi and the direct object clitics îl and îi with other object and auxiliary clitics, clitic sequence initial [G] is deleted for similar reasons. [G] is also deleted if it is internal to the clitic sequence but initial to the second clitic of the sequence, as for example in combinations of 1sg.io îmi and 3sg.do îl (see 68). (68) Romanian: obligatory deletion of cs internal [G]:1sg.io îmi and 3sg.do îl no align-l no /Gmi/+/Gl/+/cump6r6/ hiatus (Pw, stem) onset coda max(ph) a. [mi.Gl.] cumpa[ ra[ * ** * * * b. [mil.] cumpa[ ra[ * * ** c. [G.mil] cumpa[ ra[ ** * * *
Interaction of morphology and phonology in clitic sequences 167
Moreover, we Wnd obligatory deletion of clitic Wnal [6] in combinations of 1sg.do ma[ and 2pl.o va[ with auxiliary clitics, which also results from align(Pw,stem), no hiatus and onset violations. Reduction of clitic Wnal vowels as shown in (67) is also observed with 1pl.o ne and 2pl.o va[ . In these cases, [e] is reduced to [eÛ ] and [6] to [j] before vowel initial clitics. Similarly, Portuguese clitic sequences are restricted by prosodic minimality requirements. In contrast to Romanian, no reduction takes place if vowel Wnal and vowel initial clitics are combined whereas Wnal vowels, and sometimes also codas, are deleted. Consider the clitic sequence paradigm in (69). (69) The Portuguese clitic sequences paradigm io do 1sg 2sg 3sg.m 3sg.f 1pl 2pl 3pl.m 2pl.f
me te o a nos vos os as
1sg me
2sg te
3sg lhe
1pl nos
2pl vos
3pl lhes
mo ma
to ta
lho lha
no-lo no-la
vo-lo vo-la
lho lha
mos mas
tos tas
lhos lhas
no-los no-las
vo-los vo-las
lhos lhas
In the Wrst clitic of a sequence, only [e] is deleted (or replaced) but not [o] (compare lhes + o →lhos and nos + o →no-lo). I take this as evidence that the underlying representation of the Portuguese clitics me, te, lhe and lhes contains only an underspeciWed vowel position but no features indicating the vowel quality. This position is then Wlled by the default vowel [e], if it occurs at the end of a prosodic word. All other Portuguese clitics, i.e. o, a, os, as, nos, and vos, have fully speciWed vowels. Let us Wrst consider combinations with 3io lhe and 3pl.io lhes. In (70), 3io lhe is combined with 3m.do o. The optimal candidate is lho (70c), which consists of only one syllable, i.e. only one violation of align(Pw, stem) occurs. The underspeciWed vowel position /V/ of the input does not surface, which leads to a violation of max(V). In the only other monosyllabic candidate (70b), the fully speciWed vowel /o/ does not surface instead of the underspeciWed vowel. Since 3m.do only consists of this vowel, an entire argument is not realised if this vowel does not surface. Therefore,
168 Clitic sequences
(70b) also violates max(arg)M. In this evaluation, the proposed ranking is not necessary. In agreement with the facts below, however, it is presupposed for the time being. (70) Portuguese: combination of 3io lhe and 3m.do o /lV/ + /o/ max(arg)M a. lhe.o. b. lhe. c. lho.
*!
align onset no hiatus (Pw,stem) **! * *
*
max(V)
* * *
Combinations of 3io lhe with 3m.pl.do os are analysed similarly (see 71). no coda violations and max(V) violations are accepted in the optimal candidate (71c) in order to parse the morphological number information as a comparison with (71d) shows. (71) Portuguese: combination of 3io lhe and 3m.pl.do os /lV/ + /os/ a. lhe.os. b. lhes. c. lhos. d. lho.
max align max no no max (arg)M (Pw,stem) (pl) coda hiatus (V) **! * * *! * * * * * * * *! *
max (C)
*
When both objects are plural, number is only marked once in the optimal candidate (see 72). Therefore, max(arg)M and align(Pw, stem) rank higher than max(pl). (72) Portuguese: combination of 3pl.io lhes and 3pl.do os /lV/ + /os/ max align max no onset max (arg)M (Pw, stem) (pl) coda (V) a. lhes.os. **! ** * b. lhes. *! * * * * c. lhos. * * * * d. lhe.sos. **! *
max (C) * *
If the indirect object is plural but the direct object singular, the optimal candidate also violates max(pl). In this combination, another constraint be-
Interaction of morphology and phonology in clitic sequences 169
comes important, namely morpheme integrity (see also Anderson 1996 for other integrity constraints). morpheme integrity prohibits the discontinuous representation of morphemes in the output. In this case, candidate (73e) is worse than the optimal candidate (73c), because the information of the indirect object is marked before and after the direct object. (73) Portuguese: combination of 3pl.io lhes and 3m.do o /lVs/ + /o/ a. lhes.o. b. lhe.so. c. lho. d. lhes. e. lhos.
max morpheme align max(pl) no coda onset (arg)M integrity (Pw,stem)
*! *!
**! **! * * *
*
*
* *
* *
Let us now examine combinations of 1o and 2o with 3do. As explained above, in these combinations, the Wnal [s] of the Wrst clitic, i.e. 1o nos or 2o vos, is dropped and the onset of the second clitic becomes [l]. In traditional grammars, this process is called “neutralisation”. It also occurs if 3do encliticises to hosts which end in /s/, /z/ or /r/. Neutralisation results from another often attested (mutual) phonological alignment constraint, which requires that the edges of morphemes, in particular clitics, are identical to the edges of syllables (see 74). (74) align l(m,σ) The left edge of a morpheme is aligned with the left edge of a syllable. alignr(m,σ) The right edge of a morpheme is aligned with the right edge of a syllable.
In the tableaux above, these constraints are often violated in the optimal candidates, but their integration into the tableaux, particularly if they rank low, does not change the evaluation, as one can easily see. Instead of 3m.pl.do os, 1pl.o nos is combined with 3f.pl.do as to show that the merging of both clitics into one syllable (candidate 75d) is impossible even if the resulting form diVers from other clitics in the inventory (i.e. the merged form nas but not nos diVers from 1pl.o nos). High ranking of the additional phonological constraint max(V-f) (i.e. ‘speciWed vowel features from the input must be present in the output’) excludes such merged forms. The candidate with l-epenthesis (75e) instead of the candidate with resyllabiWcation (75b) is optimal, because the latter violates align(m,σ) which is worse than violations of dep(ph). 22
170 Clitic sequences
(75) Portuguese: combination of 1pl.o nos /nos/ and 3m.pl.do as /as/
a. b. c. d. e.
max align align-r align-l dep no coda onset (V-f) (Pw, stem) (m,σ) (m,σ) (ph) nos.as. ** **! * no.sas. ** * *! * nos.las. ** **! * nas. *! * * * * no.las ** * *
In addition to the constraints on the prosodic shape of clitic sequences, there are also a number of phonological constraints which are the source of deviations in the phonological form, substitution and deletion of clitics.
4.4.2 Dissimilation of clitics within sequences In the introduction to this chapter, we have seen that a number of opaque clitics occur if clitics are combined with other clitics (see section 4.1.3). These deviations from the underlying phonological form of the clitics result from phonological constraints which all originate in an OCP-like requirement of distinct elements within clitic sequences.23 (76) Phonological constraints: the elements of a clitic sequence are dissimilar *sonority Adjacency The sonority levels of adjacent elements in a clitic sequence are not adjacent in the sonority hierarchy. *alliteration *[[X1...]c1 [X1...]c2]CS ‘Clitics in a clitic sequence do not have identical onsets.’ *rhyme ‘Clitics in a clitic sequence do not have identical nuclei.’ antihomophony *[ ... χ χ ...]Pw χ: some stretch of sound, (Golston,1995) ↓↓ A,B: some stretch of AB meaning. (‘Homophonous clitics are avoided’.)
In addition to opacity, these phonological constraints sometimes inXuence the order of clitics, as for instance in French, or make certain clitic combinations fail completely, as for instance in Spanish. In the following section, I will
Interaction of morphology and phonology in clitic sequences
discuss how diVerent rankings for diVerent languages are able to explain these data.
4.4.2.1 Deviations from the French clitic order Let us start with clitic order deviation in French. French clitic sequences obey the order of io < do discussed above, except that in combinations of two 3rd person clitics 3do precedes 3io (see 77). (77) The French object clitic sequence paradigm io 1sg 2sg 3sg do me te lui 1sg me 2sg te 3sg.m le me le te le le lui 3sg.f la me la te la la lui 1pl nous 2pl vous 3pl les me les te les les lui
1pl nous
2pl vous
nous le vous le nous la vous la
3pl leur
le leur la leur
nous les vous les les leur
While the exceptional order of 2nd person in Romanian depends solely on the re-ranking of morphological constraints (see section 4.3.3), the exceptional order of 3rd person clitics in French is caused by the phonological constraint *sonority-adjacency. *sonority-adjacency excludes clitic sequences in which elements like /r/ (in leur) or /i/ (in lui) and /l/ are adjacent. Since the sonority level of [i] is the lowest for vowels and the sonority level of [l] or [r] is high for consonants in the sonority hierarchy, [i] and [l] as well as [l] and [r] may not be adjacent within the clitic sequence. As a consequence, the order of clitics is changed. align-l(+lr,cs) ranks below max(arg)M in French. The evaluation of the combination of 3do le and 3io lui is illustrated in the tableau in (78). (78) French : combination 3io lui and 3do le lexical entries of the clitics: le: +hr lui: +hr,+lr λz λy λx V(x,y,z) max(arg)M *son-adj a. le lui b. lui le *! c. lui *! d. le *!
align-l(+lr,cs) *
171
172 Clitic sequences
Despite the re-ranking of align-l(+lr,cs), combinations of 1o and 2o and combinations of 1o and 2o are still excluded, because align-l(+1,cs) and alignl(+2,cs) still rank highly (see 79 and 81). max(+lr/+hr) and max(person)M guarantee that it is the direct object which is chosen instead of the prohibited clitic sequence as has already been shown above for Spanish (see section 4.3.2).24 (79) French: combination of 1o me and 2o te lexical entries of the clitics: me: +1,+hr te: +2,+hr λz<+1>λy<+2>λx align align max max align- max V(x,y,z) (+1,cs) (+2,cs) (per)M (arg)M (+lr,cs) (+lr/+hr) a. b. c. d.
me te te me te me
*!
*
*! * *
* *
* * *!
In French, just like in Romanian, demotion of the alignment constraint reveals that markedness also plays a role. For the same reasons as argued above (see section 4.3.4), however, speciWc markedness constraints are avoided. Instead, other alignment constraints come into play which determine that clitics bearing the feature [+lr] are on the left of clitics bearing the feature [+1] or [+2]. These constraints are also in agreement with Romanian, where 3io precede 2do, if such sequences are allowed at all (see section 4.3.3). (80) align(+lr,r,+1,l) The right edge of a clitic with the feature [+lr] is aligned with the left edge of a clitic with the feature [+1].
align(+lr,r,+2,l) The right edge of a clitic with the feature [+lr] is aligned with the left edge of a clitic with the feature [+2]. In French, these constraints conXict with the alignment constraints that require 1o and 2o clitics to be left in the sequence which results in ungrammaticality of combinations of 1o (2o) and 3io (see 81).
Interaction of morphology and phonology in clitic sequences 173
(81) French: combination of 1o me and 3io lui lexical entries of the clitics: me: +1,+hr lui: +hr,+lr λz<+1> λy λx align-l align max max align-l V(x,y,z) (+1,cs) (+lr,r,+1,l) (per)M (arg)M (+lr,cs) a. me lui *! * b. lui me *! c. lui *! * d. me *
Combinations of 1o/2o and 3do, on the other hand, still result in clitic sequences, the order of io>do being achieved by high ranking of align-l(+1,cs) and align-l(+2,cs) (see 82). Violations of align(+lr,r,+1,l) do not occur, since neither of the arguments bear the feature [+lr]. (82) French: combination of 1o me and 3do le lexical entries of the clitics: me: +1,+hr le: +hr λz λy<+1> λx align-l align max V(x,y,z) (+1,cs) (+lr,r,+1,l) (per)M a. me le b. le me *! c. me d. le *!
max (arg)M
*! *
The constraint ranking for French is thus quite similar to the ranking we have stated above for Spanish with the addition of the phonological constraint *sonority-adjacency and demotion of align(+lr,cs) which results in revealing the eVect of other alignment constraints (i.e. align(+lr,r,+1,l) and align(+lr,r,+2,l)). (83) French: constraints that determine the clitic order and clitic choice align-l(+1,cs) max(person)M >> max(arg)M align-l(+2,cs) >> >> max(+lr/+hr) align(+lr,r,+1,l) sonority adjacency align-l(+lr,cs) align(+lr,r,+2,l)
*sonority-adjacency is also relevant in Italian clitic sequences but with the result of vowel lowering instead of re-ordering of clitics.
174 Clitic sequences
4.4.2.2 Changes in the vowel quality of Italian and Romanian clitics In Italian, io clitics end in a high fronted vowel whenever they occur in isolation (mi, ti, ci, vi, gli), but the [i] is lowered to [e] in combinations with do clitics. I assume that this results from *sonority-adjacency. This assumption is supported by the comparison of combinations of 3io gli with 3do lo or prt ne on the one hand (84a), and of rfl si or loc ci on the other (84b). The former have sonorant onsets and consequently trigger lowering, while the latter have obstruent onsets and do not trigger lowering.25 (84) Italian vowel lowering
a. *gli lo → glielo *gli ne → gliene
vs. b.
gli si (*glie si) gli ci (*glie ci)
The tableau in (85) shows the evaluation of these clitic sequences (the phonological form of the clitics is given in square brackets). In addition to *sonority adjacency, we need max(V(owel)), max(onset), dep(V(owel)), and ident(high); these constraints are relevant for the phonological evaluation in general and not restricted to clitic sequences.26 The optimal output candidate glielo results from the high ranking of *sonority adjacency, max(F), max(onset) and dep(F) in relation to ident(high). Optimal input-output correspondence as in [ˆilo] (candidate 85a) violates *sonority adjacency and is therefore ruled out. Inserting a vowel as in [ˆielo] (candidate 85b) violates dep(V), and deleting either the vowel as in [ˆ:o] (candidate 85d) or the consonant as in [ˆio] (candidate 85e) violates max(V) or max(onset) respectively. Since any violation of these constraints is fatal, no ranking can be derived by the competing output candidates. In the optimal form [ˆelo] (candidate 85c) the vowel feature [+high] of the input is not part of the corresponding element in the output. This candidate thus violates ident(high) which is ranked below the aforementioned constraints. *sonority-adjacency has no eVect on the choice of the clitics so that syntactic constraints are not involved in this evaluation.
Interaction of morphology and phonology in clitic sequences
(85) Italian vowel lowering lexical entries of the clitics: gli [ˆi]: +hr,+lr lo [lo]: +hr λz λy λx V(x,y,z) max max dep *son-adj max(V) (arg)M (onset) (V) a. gli lo [ˆi.lo.] *! b. gli-e-lo [ˆie.lo.] *! c. glielo [ˆe.lo.] d. gl lo [ˆ:o.] *! e. gli-o [ˆi.o] *! f. gli [ˆi.] *! g. lo [lo] *!
To the morphological constraints in (30) we can therefore add *sonorityadjacency as relevant to the Italian clitic sequence. (86) Italian: Interaction of morphological and phonological constraints I a. alignment >> max(arg)M b. *sonority-adjacency
Another purely phonological change takes place in the Romanian clitic cluster. As shown in the table above (in 49), 1pl.o ne, 2pl.o va[ and 3pl.do le surface as ni, vi and li in combinations with consonant initial 2sg.do te and 3f.pl le. (87) Romanian vowel raising *ne te → ni te *va[ te → vi te
This change results from a constraint which seems to be irrelevant to the other Romance languages, namely *rhyme (see 76). (88) Romanian vowel raising lexical entries of the clitics: ne [ne]: +1,+pl,+hr te [te]: +2,+hr λz λy λx V(x,y,z) *rhyme ident(high) a. ne te *! b. ni te *
Phonological constraints do not only cause deviations from the morphologically conditioned clitic order and from the most faithful phonological output representation but also from the most faithful morphosyntactic output representation.
175
176 Clitic sequences
4.4.2.1 Clitic substitution in Italian Combining 3f.io le with any 3do does not result in the optimal input-output correspondent form le-lo nor in a purely phonological change. Instead, 3f.io le is replaced by 3io gli. Violation of *alliteration is the reason for this substitution. The evaluation of this combination is shown in the tableau in (89). (89) Italian 3f.io substition by 3io gli vs. phonological deviation lexical entries of the clitics: lo: +hr le: +f,+hr,+lr; additional elements of C in the output candidates: gli: +hr,+lr λz λy λx V(x,y,z) a. le lo b. glie lo c. e lo d. le o e. Ce lo f. le Co
*alliteration max(arg)M ident(gender) *! * *! *! *! *!
The most faithful correspondent to the input, le lo, is bad because it violates *alliteration. The question arises why the optimal candidate does not result from a purely phonological change as is the case with violations of *sonorityadjacency. There are instead two main reasons for the morphological replacement of one of the clitics. First, clitics must be identiWable as clitics. For Italian, this means that clitics must have the structure CV. Consequently, leaving out the onset (as in the output candidates 89c and 89d) causes the corresponding elements to be no longer identiWable as a clitic, and a max(arg)M violation emerges. Second, the morphosyntactic information of the clitics must be identiWable. If we compare the form of the Italian clitics in isolation, we see that the onset is the distinctive morphophonological information when person/number oppositions are determined. The vowels only determine the gender/number distinction for 3o clitics; all other clitics have the same vowel features. Hence, identifying clitics as belonging to the clitic paradigm also depends on the phonological features of their onsets. Replacement of the onset (as in the output candidates 89e and 89f) also causes a fatal violation of max(arg)M.27 Since the ranking of max(arg)M is at least as high as *alliteration, only morphosyntactic substitution remains for the optimal output candidate. As we can see, the 3io clitic gli is the best choice for this
Interaction of morphology and phonology in clitic sequences 177
replacement. The morphosyntactic features of gli diVer from those of le only minimally, in that [+f(eminine)] is absent. Therefore, ident(gender) must be ranked lower than *alliteration and max(arg)M.28 There are several other possible clitics in Italian which could also serve as suppletive forms. First, let us consider the reXexive clitic si, which is indeed chosen in the same morphophonological context in Spanish (see 97 below). si does not add any conXicting agreement feature, since it is unmarked for person, just like the object clitic that is to be replaced. Neither does it add any conXicting case feature — instead, it is less speciWed than the object clitics, because it also occurs as an impersonal subject marker. Consider the tableau in (90). (90) Italian 3f.io substitution:3io gli vs. rfl si lexical entries of the clitics: lo: +hr, additional elements: gli: +hr,+lr λz λy λx *alliteration V(x,y,z) a. b. c. d.
le lo glie lo se lo le si
le: +f,+hr,+lr; si: +rX ident ident (case) (rX)
ident (gender)
*! *! *!
* *
* * *
Replacement of one of the clitics by si would cause the case features [+hr] (if si replaces the 3do clitic as in candidate 90d) and [+hr,+lr] (if si replaces the 3f.io clitic as in candidate 90c) to be not identical in the corresponding elements. Furthermore, si bears at least the feature [+rX]. This feature is not part of the input representation. The replacement of an object clitic by si therefore causes a violation of ident(rX). Therefore, replacement of le by gli leads to the optimal candidate, as can be seen in the tableau (90). A ranking of ident(case) and ident(rX) can be derived neither with respect to *alliteration nor with respect to ident(gender). The correct outcome of the evaluation is solely due to the ranking *alliteration >> ident(gender). Similarly, replacement of one of the clitics by the locative clitic ci (which is distinct from the 1pl.o clitic, as shown above) causes an ident(case) violation, since the locative clitic is unspeciWed for case features, and, in addition, an ident(loc) violation. In this case, too, glielo remains the optimal output candidate (see 91).
178 Clitic sequences
(91) Italian 3f.io substitution:3io gli vs. loc ci lexical entries of the clitics: lo: +hr le: +f,+hr,+lr; additional elements: gli: +hr,+lr ci: +loc λz λy λx V *alliteration ident ident (x,y,z) (loc) (case) a. le lo *! b. glie lo c. ce lo *! * d. le ci *! *
ident (gender) * * *
We can now add *alliteration to the Italian constraint ranking. (92) Italian: Interaction of morphological and phonological constraints II a. alignment >> max(arg)M b. *sonority-adjacency c. *alliteration >> ident(gender)
There is one further morphophonological context in Italian which triggers substitution. (see the sentences in 93). The impersonal use of si is demonstrated in (93a) and the reXexive use of si in (93b). Combinations of impersonal and reXexive si do not result in the most faithful sequence *si si as in (93d), but in the sequence ci si as in (93c), where the Wrst of the reXexive pronouns is replaced by locative ci. (93) Italian: impersonal and reXexive si a.
Alessio ha due anni. Lo si lava ogni giorno Alessio aux.3sg two years. do imp.s wash.3sg every day alle otto. p.d eight ‘Alessio is two years old. They wash him every day at eight.’
b. Quando avrá tre anni, when aux.fut.3sg threeyears si laverá probalbilmente da solo. rfl wash.fut.3sg probably P alone ‘When he will be three years, he will probably wash himself alone.’ c. Ci si lava ogni giorno alle otto. imp.s rfl wash.3sg every day p.d eight ‘One washes oneself every day at eight.’ d. *Si si lava ogni giorno alle otto.
Interaction of morphology and phonology in clitic sequences 179
The combination of identical clitics is prohibited due to antihomophony.29 In this case, too, the Italian clitic inventory provides a less speciWc suppletion form, namely locative ci. As argued above, ci is analysed not as 1pl.o but as locative. It is also often assumed that 1pl.o ci, rather than locative ci, replaces the reXexive (see e.g. Monachesi 1995; Grimshaw 1997). This assumption is supported by dialectal data from Florence, where the 1pl pronoun noi is still used in this context additionally to ci (Paola Monachesi, p.c.). This shows that the impersonal subject si rather than the reXexive si is replaced. In chapter 3, however, it has been shown that ci has a number of functions in contemporary Standard Italian which are typical for locative pronouns — in particular expletive — but not compatible with the feature speciWcation [+1,+pl]. Therefore, locative ci has to be analysed as the replacing clitic although 1pl.o ci may be the historical source. As to the evaluation of the combination of reXexive and impersonal si, no ranking of antihomophony and max(arg)M can be derived, as is shown in the tableau (94). (94) Italian : combination of reXexive and impersonal si lexical entries of the clitics: si : +rX ci: +loc λy λx V(x,y) antihomophony max(arg)M a. si si *! b. si *! c. ci si
dep(loc)
*
The question arises why such substitutions of clitics only occur when phonological constraints prevent the combination of the clitics but not when morphological constraints prevent it. In section 4.3, it has been shown that combinations are prohibited whenever 1o or 2o are interpreted as direct objects. Therefore, the sentence in (95a), which contains such a clitic sequence, is ungrammatical. Instead, one of the objects is realised as a clitic and the other as a full pronoun (95b,c). Italian is the only Romance language that has a free choice which of the arguments is realised as a clitic and which one as a pronoun. If less speciWc forms such as locative ci or lo replace 3io, such combinations are either ungrammatical (95d) or grammatical but have a distinct meaning (95e) which would result in fatal violations as regards uniWcation of input and output speciWcations as exempliWed in chapter 3.
180 Clitic sequences
(95) Italian: combination of 1o and 3io a.
*Glie-mi / *mi gli presenta. 3io–1o 1o 3io introduce.3sg
b. Mi presenta a lui. 1o introduce.3sg p prn3sg ‘S/he introduces me to him.’
c. Gli presenta a mi. 3io introduce.3sg p prn.1sg
d. *Ci gli / *glie-ci presenta. e. #Me lo presenta. loc 3io 3io–1o introduce.3sg 1o 3do introduce.3sg (meaning: ‘S/he introduces him to me.’)
Moreover, it is worth noticing, that si — in contrast to object clitics — does not have a full pronominal counterpart which could support the intended meaning of the sentence if a clitic sequence is avoided. We can therefore conclude that substitution in cases of combinations of 1o and 2o with 3io is worse than deletion of one of the clitics in the clitic sequence. The interaction of morphological and phonological constraints in Italian is summed up in (96). (96) Italian: Interaction of morphological and phonological constraints III a. alignment >> max(arg)M b. *sonority adjacency c. *alliteration >> ident(gender) d. ident(case), ident(loc), ident(rX)
Let us now turn to Spanish. As pointed out above, *alliteration aVects the combination of clitics in nearly the same context as it does in Italian but with a diVerent result.
4.4.2.4 Clitic substitution and deletion in Spanish The phonological representation of the prohibited combination of 3io and 3do is identically le lo in both Italian and Spanish. In Spanish, however, the reXexive clitic is chosen as a suppletion form (see 97), instead of a less speciWc 3io as in Italian. (97) Spanish:
le lo → se lo
The observation that Italian and Spanish have similar phonologically conditioned suppletion, but only diVer in the choice of suppletive forms, contradicts the approaches to the Spanish data which assume that there is only one slot (or functional projection) for the 3rd person and that, therefore, two 3rd
Interaction of morphology and phonology in clitic sequences
person clitics may not combine. Such approaches suggest that the reXexive clitic originates from another slot and may thus be chosen instead (see, e.g. Harris 1995 and Manzini 1998). Grimshaw (2001) analyses these opaque clitic combinations in a similar way and diverges from her previous (1997) proposal that the phonological constraint *XX is responsible for the ungrammaticality of combinations like le lo. Her recent analysis cannot be transferred to the Italian data and the remarkable similarities cannot be analysed on a common theoretical base. In agreement with her prior proposal, I assume that *alliteration is highly ranked in both languages. The result is that opaque forms and diVerent repair strategies result from diVerent clitic inventories. The diVerent solutions to the violation of *alliteration result from diVerent sets of clitics in the Spanish respective Italian clitic lexicon. In Spanish, all 3o clitics have the onset [l]. Thus, every replacement of one of the clitics within the object clitic paradigm results in an *alliteration violation. Moreover, there is no underspeciWed clitic other than se in the Spanish clitic inventory which can serve to replace one of the object clitics. In Spanish, we can therefore derive a constraint ranking for *alliteration, ident(rX), and ident(case), namely *alliteration >> ident(rX), ident(case), which gives us se lo as the optimal output candidate. In addition to these constraints, a morphological alignment constraint for the feature [+rX] is at work which guarantees that the indirect object is replaced. (98) Spanish:3io substitution lexical entries of the clitics: lo: +hr le: +hr,+lr additional clitics : si : +rX λz λy λx align-l max ident *alliteration V(x,y,z) (rX,cs) (arg)M (case) a. le lo *! b. e lo *! * c. se lo * d. le se *! * e. le o *!
ident (rX)
* *
The question why se does not replace clitics in otherwise prohibited combinations can easily be accounted for by high ranking of align-l(rX,cs). In Iberian as well as in South American Spanish, se replaces 3io le in combinations with 3do lo. The resulting sequences, however, diVer in the range of interpretation. While in Iberian Spanish number and gender features
181
182 Clitic sequences
of the indirect object are lost after substitution by se, in some Uruguayan dialects these features are retained at the direct object clitics (see the examples in 13 on p. 134 above). I assume that the morphological merger of these clitics within the sequence is much more advanced, so that the whole sequence is regarded as a unit to which gender and number features are added. Formally, the diVerence between Uruguayan and Iberian Spanish results from diVerent rankings of max(number)M or max(gender)M and ident(number) or ident(gender) respectively. (99) Restricting morphological feature transfer in Spanish a. Iberian Spanish: ident(number) >> max(number) ident(gender) >> max(gender) b. Uruguayan Spanish: max(number) max(gender)
>> ident(number) >> ident(gender)
The diVerent evaluation of the combination of 3pl.io les and 3do lo is exempliWed in the tableaux in (100) and (101). (100) Uruguayan Spanish: combination of 3pl.io les and 3do lo lexical entries of the clitics: lo: +hr les: +pl,+hr,+lr additional elements se: +rX los: +pl,+hr λz λy<+pl> λx V(x,y,z) max(number)M ident(number) a. se lo *! b. se los * (101) Iberian Spanish: combination of 3pl.io les and 3do lo lexical entries of the clitics: lo: +hr les: +pl,+hr,+lr additional elements: se: +rX los: +pl,+hr <+pl> λz λy λx V(x,y,z) ident(number) a. se lo b. se los *!
max(number)M *
As we have seen, phonological constraints may cause deviations from the most faithful output candidate. Two kinds of violations have been identiWed: (i) purely phonological changes resulting from the omission of phonological features, thereby violating ident(phonological feature) (i.e. ident(high) in the Italian and in the Romanian examples in 86 above); (ii) substitution by other — usually less speciWc — clitics in which morphosyntactic features are omitted (or inserted), thereby violating ident(phonological feature) and
Conclusions 183
ident(morphological feature) (i.e. ident(gender), ident(rX) and ident(case) in the Italian and Spanish examples above). So far, omission of one of the clitics, thereby violating max(arg)M, has not been found among the optimal output candidates. In Spanish, however, the violation of antihomophony that results from the combination of impersonal and reXexive se is avoided by exactly this kind of strategy (see 102). (102) Se lava. (*Se se lava.) rfl wash.3sg ‘S/he washes her/himself.’ or ‘One washes oneself.’
Only one of the two reXexives is realised in the optimal output candidate. (103) Spanish: combination of impersonal and reXexive se clitics : a. b.
se : +rX λy λx V(x,y) se se se
antihomophony *!
max(arg)M *
The optimal candidate results from ranking antihomophony higher than max(arg)M. In this case, it is not possible to decide which of the two clitics is left out.
4.5 Conclusions From this chapter, it is evident that cliticisation is an interface phenomenon. I have shown that the order and combinatory restrictions on clitic sequences in Romance may be accounted for by the interaction of morphological and morphosyntactic constraints. Morphological alignment constraints determine the position of clitics in the clitic sequence, i.e. the order of clitics, according to their position in the person and argument hierarchy. The relevant alignment constraints require that the left edge of a clitic marked for case and person corresponds to the left edge of a clitic sequence. I have suggested that indirect object clitics are more marked than direct object clitics and that the 1st and 2nd person are more marked than the 3rd person. If these three alignment constraints rank high, a marked clitic always precedes an unmarked one. In sequences which contain two marked clitics (an indirect object clitic and a 1st or 2nd person clitic), alignment constraints are in
184 Clitic sequences
conXict and the combination fails. Moreover, clitics can only combine if their positions with respect to the markedness hierarchies diVer in a harmonic way. Otherwise the combination of the clitics fails, too. In such cases, only one of the arguments cliticises, while the other is realised as a full pronoun. At Wrst glance, the choice of clitic in the case where the cs is morphologically outranked contradicts what one expects from the data of clitic doubling (which will be discussed in detail in the next chapter). I have shown that this choice results from additional constraints that regulate the speciWcity of feature marking. Ultimately, the observed conXict can be traced back to the particular lexical entries of clitics. As to the morphology-phonology interface, certain prosodic constraints determine that clitic sequences are prosodically as minimal as possible. These constraints are particularly important in languages in which the whole clitic sequence is integrated into the same prosodic word as their host (Romanian and Portuguese). Moreover, we have seen that opaque clitics are best analysed if we take into account phonological OCP-like constraints. In such cases, we Wnd phonological deviations from the most faithful output candidate, clitic substitution or clitic deletion. It depends on the lexical inventory of the language whether one of the respective clitics is omitted or replaced by another clitic that does not violate the highly ranked phonological constraints. Presumably, substitution is preferred to deletion. However, substitution is only possible with elements which belong to the clitic paradigm of the language, and which are less speciWc than the clitic which has to be replaced. In the next chapter, I will show how the issue of clitic host combination can be integrated into the analysis.
Notes 1. In Slavic languages, some of these properties are found as well. 2. In section 4.4 I will argue that it is the loc clitic and not the 1o clitic which is chosen here (contrary to Bonet 1995; Grimshaw 1997; Monachesi 1995). 3. As in Italian, the indirect object clitic vi in Bulgarian, for instance, has to precede the direct object clitic ja. Ne sâm li vi ja pokasval? neg aux.1sg q 2pl.io 3do.sg.fem show.pcp.sg.m ‘Haven’t I shown her to you?’ (Avgustinova 1997)
Notes 185
In addition to direct and indirect object clitics, the Bulgarian clitic inventory contains an interrogative clitic, a negation clitic (also attested in French) and clitic auxiliaries (also attested in Romanian). Furthermore, the number of clitics which may be combined in a single cluster is not restricted to two. This is unusual for most of the Romance languages, except for Romanian and French, but it is also observed in other Slavic languages like Macedonian (see, e.g., Spencer 2000 and Tomic™ 2000), Serbian and Croatian (see, e.g. Spencer 1991 and Anderson 1996). In most Slavic languages the position of clitics is independent of the morphosyntactic category of their host. They always occur in the second position of the sentence. The placement of these “Wackernagel” or “P2” clitics has to be analysed in terms of their prosodic or syntactic properties (see chapter 5). 4. Another templatic analysis of clitic clusters can be found in Simpson & Withgott (1986). 5. Laenzlinger (1993), however, proposes a syntactic analysis of Romance pronominal clitic sequences. 6. The subject of Legendre’s (1996, 1999, 2000a,c) analysis of Balkan clitic clusters is not only the order of clitics in a cluster, but also — and mainly — the placement of the whole clitic cluster with respect to the host. The crucial assumption is that constraints that belong to the alignment family (edge in Legendre’s terminology) are in conflict with constraints which prohibit that certain features are initial in a specific domain (non initial) (see also Anderson’s 1996 analysis of Serbian and Croatian clitic placement). I will therefore focus on her analysis in chapter 5. 7. Both hierarchies are not only relevant with respect to the combination of clitics but also with respect to clitic doubling. Artstein (1998) and Grimshaw (2001) also relate markedness hierarchies to clitic phenomena, applying the mechanism of harmonic alignment with the result of particular markedness hierarchies. This kind of analysis with respect to Romance clitic phenomena will be discussed in section 4.3.4 below. 8. Subjects are still further to the left than indirect objects. This does not arise from the alignment constraints which refer to the clitic sequence as their domain, but from other alignment constraints whose domain is V’ or VP presumably. 9. max(arg)M covers max(io)M and max(do)M in this tableau. 10. In (26) and (27), the phonological change from gli to glie and from ci to ce is presupposed. How this change results is shown in section 4.4.2 below. 11. In French and Romanian this dative marking preposition is à and a respectively. 12. Due to phonological restrictions which will be discussed in section 4.4 below, the 3rd person indirect object clitic in (37b) is replaced if combined with another 3rd person clitic. 13. max(person)M is equivalent to spec(person) in Gerlach (1998a,b). 14. In Gerlach (1998a,b) the constraint ident(lr) was employed to guarantee that indirect objects do not remain underspecified. 15. The French and Romanian examples in (34) and (35) are evaluated analogously. 16. Note that, in Romanian, information structure plays a very similar role to that in
186 Clitic sequences
Spanish. While indirect object clitics are obligatory, regardless of the referential properties of the argument, direct objects are only realised as clitics if their referents are specific and human (see chapter 5). For the morphological evaluations here, it is presupposed that the direct objects bear these features and that therefore the clitics should occur. 17. One problem remains with respect to the Romanian data. Consider the sentences in (i) and (ii) with combinations of 1sg.do and 2io and 1pl.o and 2io respectively. (i) a. b. c.
*Ma[ t6 i / *t6 i ma( recomanda[ . Ma[ recomanda[ t6 ie. *Ît6 i recomanda[ pe mine. ‘He recommends me to you.’
(ii)
a. *Ne t6 i /* T6 i ne recomanda[ . b. Ne recomanda[ t6 ie. c. *Ît6 i recomanda[ pe noi. ‘He recommends us to you.’
The clitic combinations are prevented in these cases because of align(+1,cs) and align(+lr,cs). Instead of the clitic sequence, the direct object cliticises, while the indirect object has to be expressed by a full pronoun. The problem which arises now is how to determine the clitic choice in these cases. In the tableaux above (see, e.g., 53 and 54), the constraint max(+lr/+hr) renders unspecified 1o or 2o worse when they refer to indirect objects than when they refer to direct objects. In these cases, however, 2io is fully specified with respect to the feature [lr] and thus no violation of max(+lr/+hr) occurs. 18. Compare also Haspelmath (2001) who suggests a usage-based explanation for the above described combinatory restrictions. Haspelmath states that “Grammars are likely to put restrictions on Recipient-Theme combinations to the extent that the Recipient argument is not higher on the topicality scales than the Theme argument” (Haspelmath 2001: 31). He makes the interesting proposal that the frequency of combinations of full pronous is restricted by this constraint determines which combinations may be grammaticalised (as affixes or clitics) and which may not. 19. A constraint like ‘A 1st person clitic cannot correspond to a direct object position, if a 3rd person clitic refers to an indirect object position’ could describe this relation. The reasoning of a constraint like this would be in agreement with the underlying idea of the alignment constraints above, i.e. the whole clitic sequence should reflect the ranking of both the person and argument hierarchy. A markedness constraint of this kind, however, on the one hand refers to the input-output correspondence relation in that it prohibits a certain interpretation of 1st person clitics and on the other hand refers to the output representation as the restricting context for this prohibition at the same time. Such constraints are far too complex to be considered part of the morphological component. Therefore, constraints like these are avoided here, even if they could prevent a splitting of the alignment constraints into, e.g., align-l(+2,cs) and align(+2,+pl,cs) with different ranks. 20. Harmonic alignment has been proposed by Prince and Smolensky (1993) to account for differences in the markedness of syllables by aligning the scales of positional prominence and sonority prominence. 21. A problem with Aissen’s concept of subject- and objecthood is that there is no clear distinction between logical and grammatical subjects (and objects). In the following, I will refer with the term ‘subject’ or ‘s’ to the element which corresponds to the highest
Notes 187
argument role in the input argument structure, i.e. the features of subject, indirect and direct object of a ditransitive verb are s = [–hr,+lr], do = [+hr,-lr], io = [+hr,+lr]. 22. In Gerlach (2001b) it is argued that the occurrence of /l/ in the onset of the cliticcontaining syllable results from two facts: First, Portuguese 3rd person clitics have an underspecified onset position which is filled if resyllabification takes place after encliticisation. Second, the feature specification of this onset follows from the consonantal feature geometries as developed in Grijzenhout (2001) with the result of /l/. 23. Grimshaw (1997) expresses the prohibition of identical elements as *XX. Similar constraints are suggested by different authors. Compare, for instance, Plag (1998), who assumes different OCP-like constraints which together exclude the adjacency of identical morphemes. In Plag’s model, *alliteration could be translated as ocp(onset) and *rhyme could be translated into ocp(nucleus). Both could then collaborate by local conjunction (see, e.g. Lubowicz 1997) to replace antihomophony. 24. Recall from the discussion above that in cases where the direct object is realised by a clitic, the indirect object is realised by a full pronoun. The pronominal DP is then specified for the feature [+lr]. 25. Kaisse (1985: 99–100) also suggests that clitic vowel lowering is triggered by a following sonorant. Based on Grandgent (1933), she also provides an appealing diachronic explanation on this phenomenon. 26. Max(V) is only relevant for clitic sequences with the clitics ti, ci, vi as indirect object, since leaving out the vowel in mi or gli is already ruled out by fatal violation of *sonority adjacency. 27. To capture this morphological motivation, one might argue for a constraint which requires morphological identifiability instead of using the more general phonological constraints max(onset) and ident(onset). 28. unif(gender) is not violated here. 29. Phonologically identical elements may be adjacent in Italian, as for example se (conditional) and se (clitic), if they do not belong to the same prosodic word (see chapter 2). Antihomophony also causes the prohibition of *ci ci (1pl.o and locative) und *vi vi (2pl.o and locative).
188 Clitic sequences
Movement and base generation 189
Chapter 5
Syntax at the interfaces: Clitic placement and clitic doubling
The placement of clitics in the sentence is a purely syntactic issue. The question that needs to be answered is why the positions in which clitics surface are diVerent from the positions in which full DPs surface. Moreover, the question arises why these positions are exclusive to clitics. To solve these questions, we not only need to consider to which syntactic position clitics are assigned on the surface, i.e. “where do they go”, but we also need to ask in which syntactic position clitics are generated, i.e. “where do they come from”. A huge number of diVerent syntactic analyses is concerned with these questions for Romance as well as for Slavic languages. For reasons of space, I can only present a selection of the most recent proposals.1 As illustrated in chapter 2 and chapter 4, clitics in Romance are adjacent to a verb. In Italian, for instance, direct object DPs follow the Wnite verb (1a), while direct object clitics precede it (1b). Any other order with Wnite verbs is ungrammatical: see (1c) and (1d). (1) Italian: Placement of object DPs and object clitics a. Vedo Claudia. ‘I see Claudia.’ b. La vedo. ‘I see her.’ c. *Claudia vedo. d. Vedo la.
With regard to the origin of clitics, there are two standpoints in the literature, namely the “movement approach” and the “base-generation” approach. These approaches will be discussed in the next section. After that, an analysis for clitic placement in terms of Optimality Theory will be proposed in section 5.2. It will be argued that the placement of clitics can be explained by the interaction of syntactic and phonological constraints. Finally, I will show in section 5.3 how the interaction of morphological and syntactic constraints determines whether an argument is realised as a clitic, as a DP or as a clitic doubling structure, i.e. by co-referential clitic and DP which occur in the same sentence.
190 Syntax at the interfaces
5.1. Movement and base generation The movement approach goes back to Kayne’s (1975) work on French syntax. He observes that in Standard French the object is realised either as a full DP or as a clitic. In Standard French and Italian, a clitic and a DP cannot co-occur in the same sentence: (2) No clitic doubling in Standard French and Italian a. *Je la vois Claudia. b. *La vedo Claudia.
Kayne suggests a transformational analysis of clitics. He argues that clitics are base generated as arguments (XPs) and as such move out of their XP-position to join the verb by left adjunction. For French, Kayne’s analysis is supported by the fact that clitics obey standard constraints on movement. In Italian, however, clitics behave diVerently. In constructions with modal verbs (see chapter 4) or other so-called “restructuring” verbs like cominciare ‘begin’ cliticisation to the Wnite verb is also possible. In (3) (from Rizzi 1978), for instance, the partitive clitic ne cliticises to the Wnite matrix verb but denotes a partitive argument of the embedded inWnitive. (3) Italian: Restructuring verbs Ne ho cominciato a discutere con Mario da Gianni. prt have begun p discuss with Mario at Gianni ‘I have begun at Gianni’s house to discuss (of) it with Mario.’
Rizzi (1982) attributes this behaviour to a lexical speciWcation of the matrix verb, i.e. the property of triggering restructuring of the construction, so that the movement approach may still be adopted. In Kayne’s later work (1989, 1991), these properties are captured without referring to lexical properties of the matrix verbs. He suggests that clitics may move to I° even if the embedded verb does not. This also accounts for Italian participle agreement with clitics: Kayne suggests that the clitic moves through SpecAgrO and the verb remains in AgrO°, where it receives the respective agreement morphology. A problem for this kind of analysis are clitic sequences (see chapter 4), as a comparison of the position of the object DPs in (4a) to the position of the object clitics in (4b) and (4c) shows.
Movement and base generation
(4) Italian positions of object DPs and object clitics a. Devo dare il libro a Stefano. must.1sg give d.df book p Stefano ‘I must give the book to Stefano.’ b. Glie-lo devo dare. 3io-3do must.1sg give ‘I must give it to him.’
c. Devo dar-glie-lo. must.1sg give-3io-3do ‘I must give it to him.’
As discussed in the preceding chapter, the position of the object clitics is determined by constraints on the order of clitics within clitic sequences, but not with respect to the verb. Their order with respect to each other diVers from the order of object DPs. If clitics originate in argument positions of the verbs and are moved to their surface positions, speciWc clitic transformations have to be maintained to guarantee that the order of object clitics with respect to each other is the reverse of the order of object DPs with respect to each other (see e.g. Emonds 1975: “Le La Les Rule”, “Pronominal Clitic Placement” and “Clitic Interchange” or Warburton 1977: “Clitic Permutation Rule”). Moreover, the alternation of preverbal and postverbal position requires a speciWc mechanism to derive the reverse order of the clitics with respect to the host. Other problems concerned with clitic sequences, such as combinatory restrictions and opaque clitics cannot be explained within a transformational analysis, either.2 A more serious problem for Kayne’s approach results from the prohibition of clitic doubling structures like the one in (2). The fact that the case assigned by V is absorbed either by the clitic or by the DP predicts that only one of them may occur in the same sentence. This prediction does not apply to languages like Spanish and Romanian or colloquial varieties and dialects of French and Italian. The indirect object DP in the Spanish example in (5), for instance, is obligatorily doubled by a clitic. (5) Indirect object clitic doubling in Spanish a. Le doy un libro a ella/ a un mujer. b. *Doy un libro a ella/ a un mujer. 3io give.1sg a book p prn.f.sg / p a woman ‘I give a book to her / to a woman.’
Direct object DPs may also be doubled by clitics, for instance in Rio de la Plata Spanish (6a) and in Romanian (6b).
191
192 Syntax at the interfaces
(6) Direct object clitic doubling in Rio de la Plata Spanish and Romanian a. Rio de la Plata Spanish b. Romanian Lo veo a Juan. L-am va[ zut pe Popescu. 3do see.1sg p Juan 3do-aux seen acc Popescu. ‘I have seen Juan.’ ‘I have seen Popescu.’ (Jaeggli 1986) (Sportiche 1996)
At Wrst glance, the co-occurrence of clitics and DPs seems to be licensed by the additional occurrence of a case-marking element (a in Spanish and pe in Romanian) which precedes the DP. This observation is captured by a generalisation known as “Kayne’s Generalisation” which requires that “... doubled elements must appear aYxed by a morpheme (dative-looking in Spanish, the object marker pe in Romanian) that does not appear in the absence of the clitic” (via Sportiche 1996:222). This generalisation does not stand up to a closer examination of the data. In Rio de la Plata Spanish, for instance, the case marker is obligatory with human entities (7a), whereas the clitic is obligatory with speciWc entities (see 7b, where al stands for the case marker a plus the deWnite article el). (7) Restrictions on clitic doubling in Rio de la Plata Spanish a. *La veo a una mujer. b. *Lo veo al libro. Veo a una mujer. *Veo al libro. *La veo una mujer. Lo veo el libro. ‘I see a woman.’ ‘I see the book.’
This implies that the presence of the clitic and the case marker are required — or licensed — by particular semantic features of the object. The co-occurrence of both in (6a) above is coincidental (i.e. depending on the object having the semantic features [+human] and [+speciWc]) and not — as predicted by Kayne’s Generalisation — necessary. Data such as these suggest that the movement approach to cliticisation must be abandoned, at least for languages which allow clitic doubling (see also Jaeggli’s 1986 critical assessment of Kayne’s Generalisation). Alternative “base generation approaches” to cliticisation (e.g. Jaeggli 1982) attempt to account for the possible co-occurrence of clitics and DPs. Base-generation approaches assume that clitics are generated in the position in which they surface. However, it is not quite clear how the relation with regard to case and agreement features between clitic and doubled DP is
Movement and base generation 193
captured and how semantic restrictions on their co-occurrence are to be integrated. Suñer (1988) makes the interesting proposal to account for cooccurrence restrictions in terms of agreement. She suggests that clitics are generated as heads of a VP and neither absorb case nor a θ-role. In her view, clitics are co-indexed with the argument position by chain co-indexing. In Argentinian Spanish, for instance, direct object clitics are lexically speciWed as [+speciWc]. Suñer’s “Matching Principle” requires that clitics and NPs in the same chain must have identical agreement features, and this therefore restricts their co-occurrence. We will return to her analysis in section 5.3.3. Recent syntactic theories suggest that clitics are functional D°s which are in a Spec-Head-Relation with the doubled DP at some stage of the derivation. Below, two prevailing analyses will be examined, namely that of Uriagereka (1995) and that of Sportiche (1996). Uriagereka (1995) concentrates on 3rd person accusative clitics in Spanish, which he considers “weak clitics” in contrast to 1st and 2nd person clitics, which he considers “strong clitics”. Weak clitics are analysed as heads, while strong clitics are heads and phrases at the same time, i.e., weak clitics are base generated in the head position of a DP and the doubled DP occurs in its speciWer position. Because of this Spec-head-relation, agreement features may be checked. Strong clitics, on the other hand, project full DPs and doubled DPs may therefore only be analysed as adjuncts. Furthermore, weak and strong clitics diVer in their speciWcation with respect to person: weak clitics are defective with respect to person features, whereas strong clitics are not. All clitics have in common that they are inherently speciWc and referential. The whole DP is generated in an argument position of the verb and thus receives case. Uriagereka assumes that 3rd person accusative clitics move either to AgrO (French) or to the functional head F above I (Spanish, Galician) which, according to him, contains features like speciWcity. Clitics which stay in AgrO can double unspeciWc DPs, but clitics which move to F cannot. io-clitics only move to AgrIO, hence doubling of unspeciWc indirect objects is more likely to be observed than doubling of unspeciWc direct objects. Following this assumption, indirect object clitic doubling should even be expected to be obligatory. For strong clitics, Uriagereka assumes that they move Wrst as phrases to SpecAgrO and then as heads to F (if present).3 Uriagereka (2000) suggests that the semantics of clitic doubling resembles the semantics of inalienable possession. Sportiche (1996) proposes that clitics are base-generated as heads of so-
194 Syntax at the interfaces
called “clitic voice” projections which are adjoined to the highest verbal element of the clause. Double DPs originate in a lower position, namely in a thematic position of that verb. To guarantee that features of DPs and clitics can be checked, the DP has to move into the speciWer position of the clitic voice projection. The diVerence between dative and accusative clitics results from the diVerent nature of dative clitic voice (DatCliticVoice) and accusative clitic voice (AccCliticVoice) respectively. According to Sportiche, accusative clitics have interpretative import and their associated argument must be speciWc. Therefore, AccCliticVoice diVers from AgrO, which is mainly the position for case assignment and participle agreement. Dative clitics do not have interpretative import and DatCliticVoice is considered to be the dative equivalent of AgrO or AgrS. Since speciWcity is licensed in AccCliticVoice, all speciWc direct object DPs have to raise to this projection at least at LF. Sportiche derives the preference of doubling with pronouns from this speciWcity condition arguing that “Since pronouns are quintessentially speciWc DPs, we expect that if anything is doubled (i.e. singled out as speciWc by a clitic marking speciWcity) it will at least include the archetype of speciWc DPs, i.e. pronouns” (Sportiche 1996:264). Projections are assumed to be universal and the occurrence of clitics and clitic doubling constructions is parameterised. DiVerent parameter settings may then result in languages with undoubled clitics, doubled clitics, or object agreement. Cocchi (2000) extends Sportiche’s analysis to Bantu agreement aYxes and compares them with Romance clitics. For Uriagereka’s and Sportiche’s approaches, clitic doubling in particular is a problem. First, as brieXy mentioned above, 1st and 2nd person clitics often diVer from 3rd person clitics with respect to their doubling properties. However, it is not the case that 1st and 2nd person clitics always prefer to be doubled, whereas 3rd person clitics do not, as one would expect from the generalisations above. In French, for example, 1st and 2nd person singular subject clitics are obligatory and 3rd person and plural clitics are optional. In the northern Italian dialect Piattino, in contrast, 1st and 2nd person singular clitics are optional, but 3rd person singular and 1st person plural clitics are obligatory. The diVerences in doubling can therefore not be attributed to a general diVerence between 1st/2nd person clitics on the one hand and 3rd person clitics on the other. For subject clitics, at least, the diVerences in the doubling properties seem to result from a defective verbal agreement paradigm which is re-completed by clitics, as will be argued in section 5.3.2 (see also Auger 1993; Kaiser 1992; Miller & Sag 1997; Müller & Riemer 1998).
Clitic placement 195
Second, with 3rd person clitics speciWcity is not the only condition which restricts doubling. Romanian direct objects, for example, also require animacy as a necessary condition for doubling structures. It is not clear how features like [+/-animate] can be integrated into the purely syntactic analyses.4 The issue of clitic doubling will be analysed by means of conXicting OTlike constraints in section 5.3. Based on the common assumption that clitics in clitic doubling languages are agreement markers, I will argue that clitics are basically required for each argument. Grammatical economy constraints which prohibit morphological marking for e.g. unspeciWc or non-human nominal referents are in conXict with this requirement (see also Ortmann 2000). The phenomenon of clitic doubling is thus not tied to the clitic placement problem. The placement of clitics has also been the subject of recent OT analyses. In the following section, I will discuss recent OT-analyses for clitic placement in Balkan languages (Serbian and Croatian: Anderson 1996; Macedonian, Bulgarian, Romanian: Legendre 1996, 1999, 2000b,c) and afterwards provide an analysis of the phenomenon in other Romance languages.
5.2. Clitic placement The placement of clitics has been the subject of several recent OT analyses. Clitic placement in Slavic languages in particular is problematic for purely syntactic analyses, since clitics do not attach to a host of a particular category like [+V]. Instead, they occur in the second position in the sentence.5 Syntactic analyses for second-position clitics (e.g., CZ avar & Wilder 1994) assume that clitics are moved to a position on the left of the sentence and that a process takes place which moves another syntactic element in front of them. As Anderson (1996) argues, these second-position clitics are problematic for syntactic approaches in several respects. The element to the left-hand side of the clitic is not always a whole constituent as in Serbian (8a), but even twopart proper names may be interrupted by clitics as in Croatian (8b). Under a syntactic analysis it is unclear how syntactic constituents like two-part proper names can be split up in order to host a clitic. (8) Clitic placement in Serbian and in Croatian a. Lav Tolstoi je veliki ruski pisak. Leo Tolstoi aux great Russian writer ‘Leo Tolstoi is a great Russian writer.’
196 Syntax at the interfaces
b. Lav je Tolstoi veliki ruski pisak. Leo aux Tolstoi great Russian writer. ‘Leo Tolstoi is a great Russian writer.’(Anderson 1996:174)
Second, Anderson convincingly argues that mechanisms like prosodic inversion, i.e. the PF movement of a constituent to the left of the clitics (Halpern 1992), though able to describe the facts, are not restricted enough and do not account for the diVerences between (8a) and (8b). Anderson suggests that second-position clitics may be accounted for in a similar way as in Legendre’s analysis (1996) of the Bulgarian verbal clitics. Legendre’s (1996, 1999, 2000b, c) analyses of Balkan clitic clusters are not only concerned with the placement of the whole clitic cluster with respect to the host, but also with the order of clitics in a cluster. The crucial assumption is that constraints that belong to the alignment family (edgemost in Legendre’s terminology; see chapter 4, p. 144) are in conXict with constraints which prohibit certain features being initial in a speciWc domain (non initial). Legendre proposes that this domain should be V’. An element which bears feature F may occur in initial position (9a) if align-l(F,V’) outranks non initial(F,V’). If non initial(F,V’) outranks align-l(F,V’), the element in question never occurs initially, but obligatorily in 2nd position as, for example, verbal pronominal clitics in Bulgarian (9b).6 (9) Interaction of align-l and non initial a. align-l(cl,V’) >> non initial(cl,V’): cl > V b. non initial(cl,V’) >> align-l(cl,V’): V > cl
Legendre (1999, 2000b,c) compares the properties of Bulgarian clitics with other Balkan clitics, namely those of Macedonian and Romanian. She shows that in Macedonian, where the clitic inventory is nearly identical to that of Bulgarian, the order of clitics can be predicted by the same constraints and constraint rankings. The order of the clitics with respect to their verbal host, though, diVers from Bulgarian: clitics follow imperatives, but precede other Wnite verbs. In order to account for this behaviour, Legendre assumes that — all other things being equal — the morphological constraint align-l(imperative) is higher-ranked than the other align-l constraints for clitic features and thus determines the correct clitic-host order. In addition to Legendre’s constraints, Anderson (1996) suggests the constraints integrity(word) and integrity(XP), which state that words and
Clitic placement 197
syntactic phrases respectively may not be interrupted by other material.7 With the constraint rankings in (10), clitics have to occur in the second position, i.e. for Serbian after the Wrst phrase (10a) and for Croatian after the Wrst word (10b). (10) a.
integrity(word), integrity(XP) >> non initial(cl) >> align(cl)
b. integrity(word) >> non initial(cl) >> align(cl) >> integrity(XP)
In the following, we will examine whether Legendre’s analysis of Macedonian can be transferred to Romance languages where the position of the clitics alternates depending on the morphological information of the verb as she proposes, e.g., for Romanian. In Italian, for instance, clitics occur behind inWnitives, participles and imperatives (see 11b,c) but before other Wnite verb forms (see 11a repeated from chapter 2). 8 (11) Alternation of position with respect to the host in Italian a.
Martina lo legge. Martina 3do read.3sg ‘Martina reads it.’
b. Visto-lo, fu facile decidere. see.ptc.pfv-3do be.pfv.3sg easy decide.inf ‘Having seen it, it was easy to decide.’ c.
Leggi-lo! read.2sg-3do ‘Read it!’
A similar alternation is also found in other Romance languages (see 12). (12) The position of clitics with respect to the verb Wnite Piattino Italian Spanish Romanian French
inWnitive
participle, gerund
preverbal preverbal preverbal preverbal
imperative
postverbal postverbal postverbal postverbal preverbal
postverbal
198 Syntax at the interfaces
In Piattino and Spanish, clitics are preverbal with Wnite verb forms but postverbal with inWnitives, participles, gerunds and imperatives just as in Italian. In Romanian, the clitics are preverbal with inWnitives, but postverbal with gerunds, participles and imperatives. In French, clitics are only postverbal with imperatives and preverbal with all other verb forms. Legendre’s ranking for Romanian will be discussed in section 5.2.1 together with the question which domain one has to assume for the constraints that are responsible for the placement of the clitics. In this respect, Grimshaw’s (2001) proposal will also be considered. In section 5.2.2, the interaction of morphological and prosodic constraints on clitic placement will be examined and, in particular, the problems of clitic split in Romanian and mesoclisis in Portuguese. Finally, I will discuss some remaining problems such as the more complex cases of optional clitic climbing in section 5.2.3.
5.2.1 The domain of cliticisation The domain of cliticisation has been controversially discussed in recent literature. As shown in chapter 2, Klavans (1985), for instance, assumes that the domain of cliticisation is V in the Romance languages. This means that constraints which determine the placement of clitics must refer to the verb stem as their point of reference. This assumption is problematic if one considers analytic tenses, where clitics attach obligatorily to the auxiliary. In the Italian example in (13), the clitic must move up to the (non-clitic) auxiliary, although clitics are in principle allowed to attach to participles as a comparison with (11b) shows. (13) Italian: analytic tenses L’ ho vista ieri. 3do.f aux.1sg see.ptc.pfv.f yesterday vs. *Ho vista-la ieri. aux.1sg see.ptc.pfv.f-3do.f yesterday ‘I saw her yesterday.’
Considering the sentences in (13), it seems to be reasonable to assume that clitics must attach to Wnite elements. Much in line with recent syntactic analyses, as e.g. in Kayne’s (1991) and Uriagereka’s (1995) work, Grimshaw (2001) assumes that clitics are left-adjoined to a functional head and proposes the following structure for preverbal cliticisation in Romance.
Clitic placement 199
(14) Structure of IP (Grimshaw 2001) IP Spec
I’ I1
CLL
VP I2
CLR
I3
Grimshaw suggests that the domain of cliticisation is either I’, or IP, so the clitics are thus still visible to the syntax, i.e. their placement is determined by rules which belong to the syntax, and they do not form a morphosyntactic unit.9 Apart from the problem that the morphophonological relations between clitics in clitic combinations are unique and require a unique solution (as argued in chapter 4), the order of postverbal clitics can only be explained if one adds a number of additional assumptions (see also chapter 4). Moreover, there are a number of structures in Italian and Spanish where the clitic can attach either to the Wnite matrix verb or to an inWnite embedded verb. With modal verbs such as volere ‘want’, dovere ‘must’, and sapere and potere ‘can’ (see 15a), aspectual verbs such as cominciare ‘begin’, Wnire ‘Wnish’, continuare ‘continue’, stare ‘stay’ (see 15b), and verbs of motion such as andare ‘go’, venire ‘come’ and tornare ‘turn’ (see 15c), the clitics may occur either before the Wnite verb or after the inWnitive. Left alignment to I’ does not explain this optionality. (15) Italian: clitic climbing
a. Mario vuole far-lo da solo. M. want.3sg make-3do p alone or Mario lo vuole fare da solo. M. 3do want.3sg make p alone ‘Mario wants to do it alone.’ b. La mamma comincerà a cucir-la domani. d.df.f mum begin.fut.3sg p sew-3f.do tomorrow or La mamma la comincerà a cucire domani. d.df.f mum 3f.do begin.fut.3sg p sew tomorrow ‘Mother will begin to sew it tomorrow.’
200 Syntax at the interfaces
c.
Piero andò a chiamar-lo. P. go.fut.3sg p call-3do
or Piero lo ando a chiamare. P. 3do go.fut.3sg p call ‘Piero will go to call him.’
(Calabrese 1991:573)
As pointed out above, Legendre (2000b,c) proposes that the domain of cliticisation is V’ across Balkan languages, including Romanian.10 According to her, the placement of clitics is a morphological issue and clitics are not assumed to be visible to the syntax. Objections to the use of V’ thus do not apply. For Wniteness, Legendre also adopts a morphological analysis rather than a syntactic one and combines this with the VP-internal subject hypothesis (Koopmann & Sportiche, 1991). Based on these assumptions, she analyses preverbal clitics, postverbal clitics and clitic-climbing structures in the following way. (16) Romanian a. Preverbal clitics with Wnite verb: [V’ mi-o da[ ] ‘he gives it to me’ b. Postverbal clitics with gerund: [V’ dîndu-i-l] ‘giving it to him/her’ c. Clitic climbing with modal verb: [V’ îl pot [V’ vedea]] ‘I can see him’
In Romanian, unlike the other Romance languages, clitic climbing is obligatory with modal verbs. Romanian moreover diVers from Italian in that the clitics are preverbal with inWnitives. Preverbal placement of clitics with Wnite verbs then results from a ranking of alignment constraints as in (17i), postverbal placement of clitics with gerunds results from (17ii) and obligatory clitic climbing from (17iii).11 (17) Legendre’s rankings for Romanian clitic placement12 i.
align-l(io,V’)>>align-l(do,V’)>>align-l(+Wn,V’) îmi + o + da[ align-l(io,V’) align-l(do,V’) align-l(Wn,V’)
a. mi-o da[ . b. Da[ mi-o.
*!
* **
**
Clitic placement 201
ii. align-l(ger,V’)>>align-l(io,V’)>>align-l(do,V’) îi + îl + dînd a. b.
dîndu-i-l îl dînd
align-l(ger,V’)
align-l(do,V’) *
*!
iii. alignVs>>align-l(io,V’)>>align-l(do,V’)>>align-l(+Wn,V’) îl + pot + vedea alignVs align-l(do,V’) align-l(Wn,V’) a. b. c.
îl pot vedea. pot îl vedea pot vedea-l
* *!
* **!
Clitics in Romanian are also postverbal with pure participles and imperatives. In these cases, align-l(ptc,V’) and align-l(imp,V’) are ranked in the same position as align-l(ger,V’) in (17ii). Legendre proposes that a constraint alignVs captures a clause union/restructuring analysis in that it requires two verbs to be adjacent to each other. This constraint ranks high and prohibits material such as clitics occurring between modal verb and inWnitive in (17iii). In analytic tenses, the object clitics also occur before the Wnite auxiliary, regardless of whether it is a clitic or not. The analysis is presumably parallel to that of modal verb constructions. The rankings in (17i) and (17ii) in particular imply that align-l(ger,V’) ranks above align-l(Wn,V’), with the result that gerunds precede Wnite verbs if they occur in the same analytic tense. There are, however, also other analytic forms in Romanian verbal inXectional morphology which contradict this ranking. Let us consider in this respect the Romanian presumptive present tense in (18). The present presumptive consists of a Wnite form of the nonclitic auxiliary a vrea ‘want’, the nonclitic inWnitive W ‘be’ and the gerund. Clitics obligatorily precede the verbal complex. (18) Romanian presumptive present îl voi W chemând ‘I am probably calling him’ 3do want be.inf call.ger
If we apply the ranking align-l(ger,V’) >> align-l(+Wn,V’), which results from the evaluations of analytic tenses containing object clitics to these presumptive present forms, the wrong order of verbs evolves as optimal as the tableau in (19) shows.
202 Syntax at the interfaces
(19) Evaluation of Romanian presumptive present without object clitics I voi + W + chemând alignVs align-l(ger,V’) align-l(Wn,V’) a. voi W chemând b. voi chemând W c. W voi chemând d W chemând voi e. chemând voi W f. chemând W voi
** ! *! ** ! *!
* ** * **!
According to the ranking of align-l(ger,V’) >> align-l(Wn,V’), the candidates in which the auxiliary precedes the gerund (19a-d) are worse than the candidates with the reverse order (19e,f). From the latter, the candidate with the inWnitive in Wnal position is better than the candidate with the inWnitive in second position. These predictions are not met by the data. I therefore suggest the reverse order for the constraints align-l(ger,V’) and align-l(Wn,V’). In addition, the constraint align-l(inf,V’) is ranked between them. The presumptive present as well as all other analytic forms are then captured by this ranking (see 20). (20) Evaluation of Romanian presumptive present without object clitics II voi + W +chemând a. voi W chemând b. voi chemând W c. W voi chemând d W chemând voi e. chemând voi W f. chemând W voi
align-l (Wn,V’)
*! **! *! ** !
align-l (inf,V’)
align-l (ger,V’)
* ** !
** * ** *
** *
As to the placement of clitics within the verbal complex, Legendre’s constraints edge(do,V’) and edge(io,V’) (i.e., in 20: align-l(do,V’) and align-l(io,V’)) cannot be integrated into this ranking in any way so that the clitics come out in the correct order. Therefore, I suggest that the placement of clitics with respect to their hosts is supplied by morphological alignment constraints which refer to the morphological speciWcation of verbs on the one hand and to clitic sequences on the other. In a way, this means that I return to Klavans’ (1985) suggestion that the domain of Romance clitics is V: clitics are aligned with verbs and the direction of alignment depends on the features of
Clitic placement 203
the verbs. In contrast to Legendre’s constraints, these constraints do not refer to the features which are encoded by the clitics but to clitic sequences (which may contain one or more clitics). (21) Morphological alignment of clitic sequences and their host align(cs,l,V[Fα],r) The left edge of a clitic sequence is aligned with the right edge of a verb with the feature speciWcation Fα. ‘cs attaches to the right of V’ align(cs,r,V[Fβ],l) The right edge of a clitic sequence is aligned with the left edge of a verb with the feature speciWcation Fβ. ‘cs attaches to the left of V’ Romanian: Fα = gerund, imperative, participle Fβ = inWnitive, Wnite verbs
These alignment constraints require that clitic sequences always attach to verbs. Depending on the feature speciWcation of the verbs, the clitic sequences attach to their right or to their left edge, i.e. the right edge of the clitic sequence meets the left edge of the verb and vice versa. More precisely, there is one speciWc constraint for each of the verbal feature speciWcations in the respective language. For Romanian, for instance, Wve diVerent alignment constraints are relevant. These diVerent constraints are necessary because there are diVerent rankings within one group of verbal speciWcations if analytic verb forms consist of, e.g., inWnitives and Wnite verbs. Let us now return to the examples in (16) and (18). Clitics precede Wnite verbs because of high ranking of align(cs,r,VWn,l) as shown in (22). (22) Romanian: Placement of clitic sequences with Wnite verbs îmi + o + da[ a. b. c.
mi-o da[ . da[ mi-o. mi da[ o
align(cs,r,VWn,l)
align-l(VWn,V’) **
*! *!
*
With gerunds, on the other hand, align(cs,l,Vger,r) guarantees that the clitics follow the verb (see 23). With respect to align-l(ger,V’), no ranking can be derived.
204 Syntax at the interfaces
(23) Romanian: Placement of clitic sequences with gerunds
îi + îl + dind a. b.
align (cs,l,Vger,r)
align-l (Vger,V’)
*!
*
dîndu-i-l îl dînd
In the evaluation of analytic tenses and modal verb complexes, more than one of the alignment constraints are involved (see 24). Legendre’s constraint alignVs, however, is dispensable in this evaluation, since high ranking of align(cs,r,VWn,l) suYces to exclude candidates where the clitic occurs between the two verbs (i.e. candidate 24b). (24) Romanian: Placement of clitic sequences with modal verb complexes îl + pot + vedea
align align-l align align-l (cs,r,VWn,l) (Wn,V’) (cs,r,Vinf,l) (inf,V’)
a. îl pot vedea. b. pot îl vedea c. pot vedea-l
*
* * *
*! *!
** ** *
In the analytic presumptive present tense, there are still more of the alignment constraints to be considered. High ranking of align(cs,r,VWn,l), however, again provides the correct candidate (see 25). (25) Romanian: Placement of clitic sequences with presumptive present tense îl + voi + W + chemând a. b. c. d.
align align align align-l align-l alig-nl (cs,r, (cs,r, (cs,l, (Wn,V’) (inf,V’) (ger,V’) VWn,l) Vinf,l) Vger,r)
îl voi W chemând voi îl W chemând *! voi W îl chemând **! voi W chemând îl ***!
*
*
**
* **
** ** * *
*** ** * **
*** *** ***
As to the ranking of align(cs,r,Vinf,l) and align(cs,l,Vger,r) with respect to each other and with respect to the other alignment constraints, no order can be derived from this evaluation. There is, however, another analytic tense which requires a constraint ranking (for a detailed examination of Romanian
Clitic placement 205
analytic tenses, see Barbu 1998). The perfect subjunctive (see 26) consists of the clitic subjunctive particle sa( (see chapter 2), the non-clitic inWnitive W ‘be’ and the perfective participle. Object clitics and subjunctive particle precede the inWnite forms. The perfect subjunctive is invariable with respect to subject inXection. (26) Romanian perfect subjunctive sa[ -l W chemat sbj-3do be.inf call.ptc.pfv
‘I / you / he would have called him’
In (27), the constraint align(cs,r,Vinf,l) ranks above align-l(inf,V’) and both rank above align(cs,l,Vptc,r) and align-l(ptc,V’) to determine the correct candidate as optimal. (27) Romanian: Placement of clitic sequences with perfect subjunctive mood sa[ + îl + W + chemat a. sa[ -l W chemat b. sa[ -l chemat W c. W chemat sa[ -l d W sa[ -l chemat e. chemat sa[ -l W f. chemat W sa[ -l g. sa[ W-l chemat h. sa[ W chemat-îl
align alignl align (cs,r,Vinf,l) (inf,V’) (cs,l,Vptc,r) *! *** ! ** ! ** ! *! ** !
** ***
*** ! * * *
** * ** * * * *
align-l (ptc,V’) *** ** * ***
*** **
Let us now come back to the other Romance languages. According to the table in (12), the constraints in (21) are speciWed in the following way. (28) Alignment of clitic sequences and their host align(cs,l,V[Fα],r) The left edge of a clitic sequence is aligned with the right edge of a verb with the feature speciWcation Fα. align(cs,r,V[Fβ],l) The right edge of a clitic sequence is aligned with the left edge of a verb with the feature speciWcation Fβ.
206 Syntax at the interfaces
Romanian:
Fα = gerund, imperative, participle Fβ = inWnitive, Wnite verb
French:
Fα = imperative Fβ = participle, gerund, inWnitive, Wnite verbs
Italian, Piattino, Spanish: Fα = gerund, imperative, participle, inWnitive Fβ = Wnite verb
If we transfer the analysis of Romanian clitic placement to the Italian examples in (11), we see that the same constraints and similar rankings can explain the data. (29) Italian: clitic placement with Wnite verbs lo + vedo a. b.
align(cs,r,VWn,l)
lo vedo. vedo-lo.
align-l(Wn,V’) *
*!
(30) Italian: clitic placement with participles a. b.
lo + visto
align(cs,l,Vptc,r)
lo visto. visto-lo.
*!
align-l(ptc,V’) *
(31) Italian: clitic placement in analytic tenses la + ho + vista a. l’ ho vista b. ho vista-la c. ho la-vista d. vista l’ho e. vista ho la
align align-l align align-l (cs,r,VWn,l) (Wn,V’) (cs,l,Vptc,r) (ptc,V’) * *! *! *!
* *
**! *
** * **
*
The placement of clitics is not only inXuenced by morphosyntactic constraints but also by prosodic constraints. As we have seen in chapter 4, there are a number of prosodic constraints which aVect clitic sequences, especially in Romanian and Portuguese. In the next section, I will show that these and similar constraints are also relevant to the placement of clitic sequences in the sentence.
Clitic placement 207
5.2.2 Clitic split and mesoclisis In chapter 4 (p. 206), the constraints align-l(Pw,stem) and align-r(Pw, stem) in particular have been shown to be essential in restricting the number of syllables and consequently the number of clitics in a clitic sequence. In the following, I will show how these constraints also aVect the placement of clitic sequences with respect to their hosts and how they take part in clitic split and mesoclitic structures, respectively. As to clitic placement in Romanian, it has been shown in chapter 2 that there is a single case where clitics do not cluster together but split up: if the 3f.do clitic o is used in analytic tenses in combination with the clitic auxiliary. While 3f.do o occurs preverbally and clusters with other potential clitics in present indicative in (32a), it occurs postverbally and does not cluster with the clitic auxiliary and other potential clitics which occur in the default preverbal position (32b). With other clitics, such behaviour is not observed (32c) . (32) Clitic split in Romanian
a. no clitic split in synthetic tenses / *(i) cumpa[ r o (j-)o cumpa[ r 3io-3f.do buy ‘I buy it (for him).’
/ * cumpa[ r (j-)o
b. clitic split in analytic tenses with 3f.do (j-)am cumpa[ rat o / *(j-)o am cumpa[ rat /*cumpa[ rat 3io-aux.pfv.1sg buy.ptc 3f.do (j-)o am ‘I bought it (for him).’ c. no clitic split in analytic tenses with other clitics l-am cumpa[ rat / *am cumpa[ rat îl 3pl.do-aux.pfv.1sg buy.ptc ‘I bought them.’ (Popescu 2000:778) As Popescu (2000) shows, there are particular prosodic and phonological restrictions on clitic combinations which are responsible for this exceptional behaviour. As one might recall from chapter 4, prosodic minimality is essential in Romanian clitic sequences (see pp. 163-167). However, not only reduction and deletion of vowels in clitic sequences results from this minimality requirement, but clitic split, too (see 33).
208 Syntax at the interfaces
(33) Clitic Split in Romanian i. no clitic split in synthetic tenses (îi +) o + cumpa[ r no hiatus a. (j-)o cumpa[ r b. (i) cumpa[ r o c. cumpa[ r (j-)o
align align onset (Pw, stem) (cs,r,VWn,l) * **! *
* *!
*
ii. clitic split in analytic tenses with 3f.do (îi +) am + no align align align onset cumpa[ rat hiatus (Pw,stem)(cs,r,VWn,l) (cs,l,Vprt,r) a. (j-)am cumpa[ rat o b. (j-)o am cumpa[ rat c. cumpa[ rat (j-)o am
*! *!
** ** **
*
* *
*
*
iii. no clitic split in analytic tenses with other clitics îl + am + no align align align onset cumpa[ rat hiatus (Pw,stem)(cs,r,VWn,l) (cs,l,Vprt,r) a. l-am cumpa[ rat b. am cumpa[ rat îl
* **!
* *
*
*
In synthetic tenses (33i), clitics do not split up, because the preverbal clitic sequence violates align(Pw,stem) only once, while the clitic split structure violates it twice. In analytic tenses (33ii), on the other hand, each of the candidates contains at least two syllables. The clitics must split up, since there is no reduced form of 3f.do o. With regard to align(Pw,stem), the clitic split structure is thus no longer worse than structures with single clitic sequences. Moreover, a single clitic sequence violates the constraint no hiatus which makes it worse than the clitic split structure. All other object clitics are reduced in front of the auxiliary clitic (33iii). Therefore, they can form only one syllable together with the auxiliary and no violation of no hiatus occurs. The clitic split structure is again worse than the structures with only one sequence. (For more details on this analysis, see Popescu 2000.) Another phonologically-induced clitic phenomenon is the attachment to a prosodic host which diVers from the morphological host of the clitic, i.e. the verb. As explained in chapter 2, distinct morphological and phonological hosts are attested in Romanian (and possibly also in European Portuguese). The category of the phonological host in these instances is irrelevant, they may
Clitic placement 209
consist of, e.g., nouns, verbs, complementisers or negation adverbs. Taking the preverbal position of the clitics for granted for the moment, the evaluation of preverbal enclitic structures is then achieved by high ranking of align-r(cs,Pw). (34) align-r(cs,Pw) The right edge of a clitic sequence is aligned with the right edge of a prosodic word.
This constraint guarantees that the clitics are prosodically enclitic (van der Leeuw 1995 proposes a similar constraint for Portuguese: align(clitic,r,w,r)). A more challenging problem of particular Portuguese clitic placement is the case of mesoclisis in future tense and conditional mood.13 Here, the clitic occurs after the stem and before the inXectional tense/mood suYx as in (35a,b). (35) Mesoclisis in Portuguese a.
comprá-lo-ás buy-3do-fut.2sg
‘you will buy it’
b. tê-lo-ás comprado ‘you will have bought it’ aux-3do-fut.2sg buy.inf (Hundertmark-Santos Martins 1998:196)
As already pointed out in chapter 2, future tense and conditional mood originate from the Vulgar Latin construction inWnitive + present tense of habe¯re ‘have’ and inWnitive + past tense of habe¯re, respectively. The inXectional endings nowadays are short forms of the Portuguese verb haver ‘have’ in present and past tense (see the full paradigm in the appendix). In (36), the relevant data from chapter 2 is repeated. As van der Leeuw shows, the synthetic future and conditional forms diVer from other inXectional forms not only with respect to mesoclisis but also with respect to stress assignment. Verbs bear a stress on the stem or on the theme vowel (which traditionally belongs to the stem) and there is no stress-readjustment after enclisis (36a). With future and conditional verb forms, however, stress is always on the inXectional ending (36b). Remarkably, an additional primary stress on the stem or the theme vowel appears in mesoclitic forms.14
210 Syntax at the interfaces
(36) Stress assignment in Portuguese verbal morphology a.
present and past tense dígo → dígo-lho fazém → fazém-no contámos → contámo-lo b. future tense and conditional mood diréi → dír-lho-éi faríam → fá-lo-íam contaríamos → contá-lo-íamos
‘I say it to him’ ‘they do it’ ‘we told/counted it’ ‘I will say it to him/her’ ‘they would do it’ ‘we would tell/count it’ (van der Leeuw 1995:32)
According to van der Leeuw, Portuguese does not possess secondary stress. Future and conditional forms that contain clitics therefore resemble compounds which consist of two prosodic words. However, van der Leeuw notes that there is a diVerence in stress assignment of compounds on the one hand and future and conditional forms on the other. While compounds always bear one stronger and one weaker stress, future and conditional forms bear two identical stresses when used in isolation. For the mesoclitic structures, van der Leeuw assumes that the constraint align-r(cs,Pw) introduced above is essential for mesoclitic forms as well. Moreover, general constraints such as morpheme integrity (morph int, morphemes are not interrupted by other morphemes see also chapter 4), stem dominance (stm dom, primary stress falls on the verb stem), prosodic word stress (pws, a prosodic word bears stress), *structure (*struc, avoid prosodic structure) and foot binarity (ft bin) are relevant for van der Leeuw’s analysis of Portuguese clitic placement.15 There are, however, some problems concerning his analysis. Van der Leeuw does not assemble all the constraints in a single tableau, but the ranking which results if one does is given in (37). The Wrst problem that arises is that van der Leeuw ranks foot binarity relatively high, i.e. above align-r(cs,Pw). While this ranking permits mesoclitic forms like the correct output in (37i.b), it excludes enclitic forms such as the past tense form contávamolo (37ii.c) in favour of (37ii.b) or forms without clitics such as the future tense form contaríamos (37iii.d) in favour of (37iii.a). Neglecting foot binarity (indicated by the light grey column in 37), on the other hand, does not lead to the correct results either. The enclitic form would achieve an optimal result both incorrectly in future tense (37i.e) and correctly in past tense (37ii.c). With future tense forms, stress on the stem would incorrectly be better than stress on the ending.
Clitic placement
(37) Future tense in Portuguese in line with van der Leeuw (1995) i. Mesoclisis in future tense ft align-r morph stm pws *struc bin (cs,Pw) int dom a. [contá][lo-íamos] * *! * b. [contá-lo][íamos] * *! c. [contá-lo][iamos] * *! *! d. [contá][íamo-lo] * *! e. [contáriamo-lo] *** f. [contaríamo-lo] *** *! g. [conta-lo-íamo] *** **! * h. [co-lo-ntaríamo] *** ***! * * ii. Enclisis in past tense
a. b. c. d. e.
ft align-r morph stm pws *struc bin (cs,Pw) int dom [contá][lo-vámos] * *! * [contá-lo][vámos] * *! [contávamo-lo] *** [conta-lo-vámo] *** *! [co-lo-ntavámo] *** **! * *
iii. Future tense without clitics
a. b. c. d.
ft align-r morph stm pws *struc bin (cs,Pw) int dom [contár][íamos] * * [contar][íamos] * * * * [contáriamos] *** [contaríamos] *** *
Beside the fact that the predictions of van der Leeuw’s analysis are incorrect, it suVers from not making clear which stress is part of the input and which is not. As he points out, stress is almost always lexically determined in Portuguese. I therefore assume that, in addition to stem dominance, faithfulness constraints (i.e. max(stress) and dep(stress)) evaluate whether stress from the input occurs in the output or not. As to future and conditional tense aYxes, their stress already belongs to the input, and max(stress) is violated if they are not stressed in the output. Stress on the stem, on the other hand, still
211
212 Syntax at the interfaces
results from stem dominance. This constraint, however, ranks lower than max(stress) (but higher than dep(stress), presumably) and also lower than another additional constraint, namely no stress clash. This ranking is responsible for the diVerent stress patterns of future tense forms with and without clitics respectively (compare 38i and iii). Other inXectional endings do not have stress in the input. With these forms, it is *structure (besides dep(stress)) which leads to the correct output candidate (38ii). (38)
Future tense in Portuguese i. Mesoclisis in future tense: /contar/ morph +/íamos/ + /o/ int a. [contá][lo-íamos] b. [contá-lo][íamos] c. [contá-lo][iamos] d. [contá][íamo-lo] e. [contáriamo-lo] f. [contaríamo-lo] g. [conta-lo-íamo] h. [co-lo-ntaríamo] *!
align-r max no stm *struc (cs,Pw) (stress) clash dom *! * * *! *! *! *! *! * * *
ii. Enclisis in past tense /contar/ + /vamos/ + /o/ [contá][lo-vámos] a. b. [contá-lo][vámos] c. [contávamo-lo] d. [conta-lo-vámo] e. [co-lo-ntavámo]
morph int
*!
align-r max no stm *struc (cs,Pw) (stress) clash dom *! * *! *! *
* *
iii. Future tense without clitics
a. b. c. d.
/contar/ + +/íamos/ [contár][íamos] [contar][íamos] [contáriamos] [contaríamos]
morph int
align-r max no stm *struc (cs,Pw) (stress) clash dom *! * * *! *! *
Clitic placement 213
The complex mesoclitic forms in Portuguese can therefore be explained by referring to the interaction of phonological constraints. A remaining problem is the analysis of the phonological change of the mesoclitic pronouns from o, a, os, as to lo, la, los, las after future/conditional stems. These changes can also be observed in enclitic forms following inWnitives. It has been shown in chapter 4 that quite similar phonological changes can be observed within clitic sequences (nos + o →no-lo) which can be explained by the eVect of phonological constraints on the structure of syllables.16 The analysis of the mesoclitic forms is assumed to work in the same way as proposed in chapter 4. In the subsequent section, some ideas are outlined concerning issues in clitic placement which need further research: (optional) clitic climbing and syntactically triggered preverbal clitics in European Portuguese.
5.2.3 Issues for further research Clitic placement in complex sentences is one of the most complicated issues in the Romance languages. The very brief overview of the Italian clitic placement data below can illustrate the phenomenon only very roughly (see Calabrese 1991 and Monachesi 1995 for a detailed description). In Italian, clitics obligatorily climb to the matrix verb in constructions with perceptive and causative verbs. The subject of the embedded clause is then realised as a do clitic in the matrix clause (raising to object). (39) Italian: obligatory clitic climbing with perception verbs a. Mario ha visto Giovanni uscire. or Mario ha visto uscire Giovanni. ‘Mario saw Giovanni leave’ b. Mario lo ha visto uscire. ‘Mario has seen him leave.’
(Calabrese 1991:585)
(40) Italian: obligatory clitic climbing with causative verbs a. Carlo ha fatto andare via Mario. vs. *Carlo ha fatto Mario andare via. ‘Carlo has made Mario leave.’ b. Carlo lo ha fatto andare via. vs. *Carlo ha fatto andar-lo via. ‘Carlo has made him leave.’
(Calabrese 1991:585)
214 Syntax at the interfaces
Clitics are not allowed to climb if the embedded verb is Wnite, nor if the subject of the embedded inWnite verb is diVerent from the subject of the matrix verb (41). (41) Italian: no clitic climbing over speciWed subjects
a. Carlo vuole che lo compri. vs. *Carlo lo vuole che compri. ‘Carlo wants you to buy it.’ (lit.: ‘Carlo wants that you buy it.’) b. Mario persuase Giovanni a comprar-lo. vs. *Mario lo persuase Giovanni a comprare. ‘Mario persuades Giovanni to buy it.’ (Calabrese 1991:574) Conative verbs such as tentare ‘try’, cercare ‘search’, provare ‘try’ only marginally allow clitic climbing (42a), and other verbs such as decidere ‘decide’, dire ‘say’, promettere ‘promise’ do not allow clitic climbing at all (see 42b). (42) Italian: no clitic climbing with speciWc verb classes
a. Marco tenta di studiar-lo. vs. ?Marco lo tenta di studiare.
‘Marco tries to study it.’
b. Aldo ha deciso di legger-lo. vs. *Aldo lo ha deciso di leggere.
‘Aldo has decided to read it.’ (Calabrese 1991:573-74)
Spanish and Italian and, to a certain degree, also French show optionality in the placement of clitics with the above-mentioned ‘restructuring’ verbs and in modal verb constructions.17 In the Italian modal verb complexes in (15) (see p. 199-200), the clitic sequence climbs optionally and not — as shown in (16c) (see p. 200) for Romanian — obligatorily to the Wnite matrix verb. The clitic sequence may be placed either in front of the matrix verb or behind the inWnitive, but the clitics cannot be separated.18 Clitics can climb up more than one phrase, if an embedded inWnitive consists of one of the classes which allow clitic climbing (see 43). In such cases, however, the clitics cannot stay with the low-est verb. (43) Italian: clitic climbing over more than one phrase a. ?Carlo continua a saper far-lo. vs. Carlo continua a saper-lo fare. or Carlo lo continua a saper fare. ‘Carlo continues to know (how) to make it.’
Clitic placement
b. *Carlo ha detto di saper far-lo. (lit.: ‘Carlo has said to know to make it.’) vs. Carlo ha detto di saper-lo fare. ‘Carlo said he knew how to make it.’
(Calabrese 1991: 574)
I do not aim at an analysis of the complex clitic placement data here but will leave this issue for further research. See, however, Monachesi (1995), who provides a lexical analysis in the framework of HPSG. Monachesi’s suggestion that the diVerence between obligatory and prohibited clitic climbing lies in the lexical properties of the matrix verbs seems to be quite reasonable. With respect to the treatment of optional clitic climbing in the framework of OT, I suspect that it is not constraint ranking which causes optional clitic placement but that two diVerent structures underlie the diVerent options. An argument in favour of this hypothesis is, that in the Italian perfective aspect, a mutual inXuence of clitic climbing and auxiliary choice can be observed. In the analytic passato prossimo, the auxiliary choice is entirely semantic but roughly follows the general rule that intransitive verbs take on forms of the auxiliary essere ‘be’, while transitive verbs take on forms of avere ‘have’.19 With modal verbs, avere is preferred (see Schwarze 1995:201) but essere is also possible if the embedded verb selects essere in the passato prossimo (see 44a without clitics). If the structure contains a clitic, the position of the clitic alternates with respect to the auxiliary choice. The clitic is realised before the modal verb if essere is chosen (i.e. with intransitive verbs), but the clitic stands behind the embedded inWnitive if avere is chosen (44b, clitics in bold letters). (44) Italian: clitic climbing and auxiliary choice a. Ieri non sono / ho potuto venire a casa tua. ‘Yesterday, I couldn’t come to your house.’ b. Ieri non ci sono / *ho potuto venire. or Ieri non *sono / ho potuto venir-ci. ‘Yesterday, I couldn’t come there.’
(Calabrese 1991:575)
As to the underlying structures to (44), I suggest that the clitic in situ structure results from subordination of the inWnitive (45a). The internal place argument may either already be saturated (45a.ii) or belong to the discourse and thus be presupposed (45a.i). In the latter case, the local argument is not inherited by the matrix verb and the clitic can thus not be assigned to any argument role. With the clitic climbing structure (see 45b), modal verb and inWnitive un-
215
216 Syntax at the interfaces
dergo clause union. All arguments of the embedded verb are transferred to the matrix verb (45b.i), so that the clitic can be assigned to the local argument role (45b.ii). (45) Auxiliary choice and clitic placement in Italian modal verb constructions a. subordination structure i. [V’ [V’ ho potuto] [V’ [V° venire]]] ∃ u<+1>λs can(u,[move(u) & bec(loc(u,place))]) (s) place ∈ discourse ii. [V’ [V’ ho potuto] [V’ [V° venir] ci]] ∃ u<+1>λs can(u, [move(u) & bec(loc(u,there)])(s) b. verb union structure i. [V’ sono [V’[V° potuto venire]]] (λl) ∃ u<+1>λs can(u,[move & bec(loc(x,l/place))]) (s) ii. [V’ ci [V’ sono [V’ [V° potuto venire]]]] ∃ u<+1>λs can(u, [move(u) & bec(loc(u,there)])(s)
Another open issue in the placement of Romance clitics are European Portuguese preverbal clitics. The main problem is that their placement is not morphologically triggered but depends entirely on the syntax and semantics of the sentence. In contrast to the other Romance languages, the alternation of preverbal and postverbal position in European Portuguese is triggered syntactically or semantically by, e.g., negation, question marker, certain quantiWers etc. Such a trigger demands that the clitics occur preverbally; in all other cases, the clitics occur postverbally (see chapter 2 for data). In contrast, the preverbal position for clitics is strongly favoured in colloquial Brazilian Portuguese and no alternation between postverbal clisis with inWnitives and preverbal clisis with Wnite forms takes place. Brazilian Portuguese clitics, therefore, are not diYcult to integrate into an analysis like the one outlined above.
5.3. Clitic doubling As pointed out in section 5.1, clitic doubling, i.e. the co-occurrence of clitics and co-referent DPs, constitutes a serious problem for purely syntactic analyses. The main diYculty with clitic doubling is that not only is the grammatical relation of the respective argument relevant to the occurrence of co-referent
Clitic doubling 217
clitics and DPs in the same sentence but so are semantic properties such as speciWcity and animacy (see the examples 5 to 7 on pp. 191–192 above), discourse-referential properties such as old vs. new referent and other grammatical properties such as pronominality and person. The table in (46) gives a (simpliWed) overview of clitic doubling conditions in Romance (‘+’ means obligatory; ‘(+)’ means optional; ‘*’ means not allowed). (46) Conditions on clitic doubling in Romance20 subject indirect object direct object +pron –pron +pron –pron +pron –pron +spec –spec +spec –spec Piattino (+)1 + + + + + (+)3 LA Spanish + + + + + E Spanish + + + + * Romanian + + * + +/*2 E Portuguese + * * + * LA Portuguese + * * + * Standard Italian * * * * * Standard French (+)1 (+) * * * * * 1 2 3 depends on person; [-focus,-human]; +: old /-: new discourse referent
* * * * * * * *
As to the conditions person and pronominality, let us consider the Piattino example in (47) and the European Spanish example in (48). In Piattino, the clitic must occur in addition to a 1st person plural subject DP (47a), while with a 1st person singular subject the clitic is optional (47b). In European Spanish, do clitics must occur in addition to pronominal direct objects (48a) but are prohibited in addition to any other direct object DPs (48b). (47) Piattino: 1 plural vs. 1 singular a. Noaltri an guarda. b. *Noaltri guarda. 1pl.prn 1pl watch ‘We watch.’ (48) European Spanish a. La veo a ella. *Veo a ella. 3do.f see.1sg p 3f.prn ‘I see her.’
Mi a guardi. Mi guardi. 1prn 1 watch ‘I watch.’
b. *La
veo a la mujer. Veo a la mujer. 3do.f see.1sg p d.df woman ‘I see the woman.’
218 Syntax at the interfaces
This distribution of clitic doubling cannot be solved by any syntactic means nor by any generalisation about a particular person which generally tends to be doubled and another person which generally does not (see e.g. Uriagereka 1995; Sportiche 1996). French subject clitics, for instance, are obligatory with other persons than Piattino subject clitics (see the detailed discussion in section 5.3.2). In addition to the conditions listed in (46), clitic doubling is also morphologically restricted. As shown in chapter 4, clitics cannot combine freely, but have to obey speciWc combinatory restrictions. 1st and 2nd person clitics, for instance, are usually not allowed to occur within the same clitic sequence. Even if both objects should be represented by clitics, only one of them is, while the other one is represented by a full pronoun. In the Romanian example in (49), the semantic and grammatical conditions for obligatory clitics listed above are met: following (46), indirect objects are obligatorily represented by clitics and the direct object is animate and speciWc and should therefore also be represented by a clitic. Since 1do cannot combine with 2io, however, the direct object is realised as a clitic, while the indirect object is realised as a full pronoun instead of the clitic sequence. (49) Morphological restrictions on clitic doubling a. *Ma[ t6 i / *tt6 i ma[ recomanda[ . 1do 2io/ 2io 1do recommend.3sg b. Ma[ recomanda[ t6 ie. 1do recommend.3sg 2prn.io c. *Îtt6 i recomanda[ pe mine. 2io recommend.3sg acc 1prn ‘He recommends me to you.’
This clitic choice contradicts the general conditions on clitic doubling listed in (46). In the following, I will examine these facts on clitic doubling. In particular, I will argue that the relation between the argument role, the clitic and the coreferential DP is an agreement relation entirely (see also Suñer 1988, 1991, 1992; Franco 1993; Torrego 1996). Clitics are required to saturate argument positions because of the max constraints that demand morphological argument realisation. One might recall from chapter 3 that there are two distinct constraints regarding the realisation of arguments (repeated in 50), one of them requiring that an argument role has a correspondent in the Wrst place (50a) and the other one requiring that this correspondent be morphological
Clitic doubling 219
(50b). The candidates which satisfy (50b) are a subset of the candidates which satisfy (50a). (50) Constraints on argument realisation:
a. max(arg):
An argument role has an overt correspondent in the output. M b. max(arg) : An argument role has a morphological correspondent (i.e. a clitic or an aYx) in the output.
For the morphosyntactic evaluation of doubling structures, the lexical representation of the verb together with its argument roles and its arguments is regarded as input information. The arguments are represented by DPs which are assigned the discourse-relevant features as, e.g., [±topic], [±focus], [±speciWc]. The input to the sentence in (51a) consists of the representation in (51b). (51) Structure of the input a. Am va[ zut-o pe Popescu. aux.1sg see.ptc-3f.doacc Popescu b. λy
‘I see Popescu (fem.).’
∃ x<+1> am va[ zut (x,y), y = [Popescu] <+speciWc>
+hr –hr –lr +lr
The output consists of a set of candidates (clauses) which combine the input element with other, language-speciWc material, as, e.g. clitics (see 52). (52) {am va[ zut; am va[ zut-o; am va[ zut pe Popescu; am va[ zut-o pe Popescu; am va[ zut Popescu; ...}
The constraint *multiple correspondence (53) restricts the number of output correspondences to the input argument role. 21 (53) *multiple An argument role in the input has only one correspondence (*mc) correspondent in the output.
With respect to the constraints introduced so far, co-occurrence of co-referent clitics and DPs is only possible if max(arg)M ranks above *multiple correspondence. Below, it is argued that exactly this ranking allows for clitic doubling structures. Agreement relations between clitics and aYxes on the one hand and DPs on the other hand are mediated by argument roles. This is attained by the concept of Transitive Correspondence. Transitive Correspon-
220 Syntax at the interfaces
dence means that — if there is more than one output correspondent to an input element — there is a correspondence relation between these output elements. (54) Transitive Correspondence: For each O1, O2 ∈ Output and I ∈ Input: If O1 and O2 correspond to the same I, O1 and O2 correspond to each other.
Transitive Correspondence is only relevant if *multiple correspondence is violated. Because of Transitive Correspondence, identity and unification constraints (see chapter 3) can also apply to DPs and their corresponding agreement markers. The restrictions on clitic doubling come from diVerent sources. Clitic doubling is generally prohibited by high ranking of *multiple correspondence. If clitic doubling is obligatory for a particular argument ‘a’, max(a)M ranks above *multiple correspondence. This will be shown for Italian and Spanish in section 5.3.1. Person-dependent clitic doubling only occurs with subjects in the Romance languages. In section 5.3.2, I will show that it is the morphological requirement for completeness in agreement paradigms, i.e. max(s)M, which causes certain clitics to be obligatory in Piattino and French. In section 5.3.3, the semantic restrictions on clitic doubling are discussed. The question will be examined whether these restrictions must be encoded in the lexical entries of the clitics (as, for instance, Suñer 1988 proposes) or whether they reXect grammatical economy constraints.
5.3.1 Simple cases: prohibited vs. obligatory clitic doubling In Standard Italian, object doubling is prohibited. Objects are represented either by DPs or by clitics. However, the choice between object clitic and DP is not free, as the small pieces of discourse in (55) show. If the object is already part of the previous discourse, as, e.g., in the questions in (55a), the following answer does not contain the DP but the clitic. If, on the other hand, the object is not part of the previous discourse, the object is represented by the DP and not by the clitic (55b).
Clitic doubling 221
(55) No object doubling in Standard Italian a. Vedi la tavola / Claudia / lei?– *La vedo la tavola / Claudia / lei. La vedo. # Vedo la tavola / Claudia / lei. ‘Do you see the table/ Claudia / her?’– b. Che vedi?
‘I see it / her.’
– *La vedo la tavola /Claudia / lei. # La vedo. Vedo la tavola /Claudia /lei. – ‘I see the table / Claudia / her.’
‘What do you see?’
This generalisation is captured by two additional constraints on the mapping of information structure to grammar. The Wrst constraint requires that DPs whose referents are relevant for the information structure are overtly realised, i.e. DPs which bear the feature [+focus] cannot be dropped. 22 (56) max(foc) Arguments which are relevant for the information structure (i.e. bear the feature [+focus]) are realised by DPs. The other constraint requires that DPs which are already part of the discourse and which are not emphasised, i.e. DPs which bear the feature [+topic], are dropped. (57) drop topic Arguments which are co-referent with the topic are not realised by DPs. (following Grimshaw & Samek-Lodovici 1995)
The Italian sentences above are then evaluated as shown in the tableaux in (58) and (59) respectively. (58) Italian: the direct object is topic λyλx V(x,y);
y = Claudia<+topic> a. Vedo Claudia. b. La vedo. c. La vedo Claudia. d. Vedo.
*mc
drop topic max(do)M max(do) *!
*!
*
* *!
*
If the direct object is topical, only candidate (58b) does not violate any of the constraints. All other candidates violate two constraints. Candidate (58b) is
222 Syntax at the interfaces
thus optimal and no ranking for the constraints can be derived. Output candidates which contain the direct object DP violate drop topic and are thus always worse than candidates that do not. If the clitic occurs in addition, as in (58c), *multiple correspondence is violated. The clitic doubling structure is thus worse than structures which contain only the clitic. If the clitic does not occur at all, but only the DP, as in (58a), max(arg)M is violated in addition to drop topic. This candidate is again worse than the candidate which only contains the clitic. We can conclude that, with topic arguments, the structure with the clitic is always optimal, regardless how the constraints are ranked. Therefore, topic arguments are neglected in the following. The relation between max(arg) and max(arg)M has already been discussed in chapter 3: candidates which violate max(arg) (i.e. max(do)) are always worse than candidates which violate only max(do)M (i.e. max(do)M). In the following, the constraint max(arg) is therefore neglected. With focus objects, the evaluation is diVerent. Here, we need a ranking of the constraints *multiple correspondence, max(foc) and max(do)M (see 59). (59) Italian: the direct object is focus λyλxV(x,y); y = Claudia<+focus> *m c max(foc) a. Vedo Claudia. b. La vedo. *! c. La vedo Claudia. *!
max(do)M *
Structures with only the clitic (59b) violate max(foc) and are thus suboptimal. Candidates with the direct object DP do not violate max(foc), and the decision between the doubling structure (59c) and the non doubling structure (59a) is obtained by the ranking of *mc above max(do)M. We can conclude that in all languages which prohibit clitic doubling a similar ranking is at work. In languages which have obligatory clitic doubling, on the other hand, the ranking of these constraints is the reverse. As mentioned above, Spanish is a case of obligatory clitic doubling, insofar as indirect objects are concerned. Indirect object clitics are always obligatory and indirect object DPs only occur in addition if their referents are emphasised in discourse (see 60).
Clitic doubling 223
(60) Obligatory indirect object doubling in Spanish a. Le doy un libro (a ella / a Maria / a una mujer / a nadie). ‘I give her / Maria / a woman / nobody a book.’ b. *Doy un libro a ella / a Maria / a una mujer / a nadie.
The crucial diVerence between Spanish obligatory clitic doubling and Italian prohibited clitic doubling is the ranking of *multiple correspondence and max(arg)M. In Spanish, the ranking max(io)M and max(foc) >> *multiple correspondence causes the doubling structure to be optimal (see 61). (61) Spanish: the indirect object is focus λyλx V(x,y); y = a nadie<+focus> max(foc) a. Le doy un libro a nadie. b. Doy un libro a nadie. c. Le doy un libro.
max(io)M
*m c *
*! *!
Person-dependent clitic doubling results from the same ranking, as will be argued in the next section.
5.3.2 Person-dependent clitic doubling in Piattino and French Person-dependent clitic doubling is attested for subjects in Piattino and French.23 The most surprising fact about subject clitic doubling is that, across the languages, it is not the same persons that require clitics. In Piattino clitics occur obligatorily with 3rd person subjects and 1st and 2nd person plural subjects (62), while 1st and 2nd person singular clitics are optional (63) (see also the examples in 47 on p. 273 above). (62) Obligatory clitic with 1st person plural and 3rd person subjects in Piattino a. (Noaltri) an guarda. Noaltri *(an) guarda. 1pl.prn 1pl watch ‘We are watching.’ b. (Nigun / un omen / l’ omen / Alessio / lu) alguarda. Nigun / un omen / l’ omen / Alessio / lu *(al) guarda. nobody / d.idf man / d.df man / Alessio / 3m.prn 3m watch ‘(Nobody / a man / the man / Alessio / he) he is watching.’
224 Syntax at the interfaces
(63) Optional clitic with 1st person singular subject in Piattino Mi (a) guardi. (Mi) a guardi. 1prn 1 watch.1sg ‘I am watching.’
In Colloquial French, on the other hand, clitics are obligatory in the 1st and 2nd person singular and with the 1st person plural clitic on (64), while all other clitics are optional (65). (64) Obligatory clitic with 1st person in French a. Moi je porte la table. Je porte la table. *Moi porte la table. 1prn 1s carry d.f.df table ‘I carry the table.’ b. Nous
on On
porte porte *Nous porte 1pl.prn 1pl.s carry ‘We carry the table.’
la table. la table. la table. d.f.df table
vs. Alexandra et
moi on porte la table. On porte la table. *Alexandra et moi porte la table. A. and 1prn 1pl.s carry d.f.df table ‘Alexandra and I (we) carry the table.’
(65) Optional clitic with 1st person plural and 3rd person in French a.
Jules il porte la table. Il porte la table. Jules porte la table. J. 3s carry d.f.df table ‘He/Jules carries the table.’
b. Alexandra et moi
nous portons la table. Nous portons la table. Alexandra et moi portons la table. A. and 1prn 1pl.s carry.1pl d.f.df table ‘Alexandra and I (we) carry the table.’
Clitic doubling 225
A syntactic or semantic analysis which attributes the requirements of clitics to particular persons cannot analyse these facts. I will show in the following that person-dependent doubling is instead morphologically induced. Let us Wrst consider Piattino subject clitics. A closer look at the subject agreement paradigm (66) shows that subject clitics may not be left out whenever the verbal forms lack subject person agreement features (obligatory clitics and subject agreement aYxes are in bold letters). A particular feature of the Piattino subject paradigm is that the 3rd person may not remain unspeciWed. Consequently, the clitic is obligatory for 3rd person and 1st person plural.24 (66) Subject agreement in the verbal paradigm 1sg (a) guardi 1pl an guarda 2sg (te) guardesc 2pl – guardé 3sg al/la guarda 3pl li guarden
In line with the analysis above, high ranking of max(s)M provides this result. The diVerence between obligatory and optional clitics lies in the input, as can be seen in (67). The verb forms without distinctive subject agreement aYxes but with only the theme vowel violate max(s)M, as long as no subject clitic occurs (67i). The verb forms with distinctive subject agreement aYxes (i.e. -i, -é, -esc), on the other hand, do not violate max(s)M, so they may also occur without clitics (67). (67) Obligatory subject clitics result from high ranking of Max(agrS) i. λx V(x) max(s)M ii. λx V-agrS(x) max(s)M a. Noaltri guarda. *! a. ? Mi guardi. b. ? an guarda b. ? a guardi. c. ? Noaltri an guarda. c. ? Mi a guardi.
The choice between these diVerent optimal candidates is again induced by the discourse-referential constraints drop topic and max(foc). Pure clitics (67i.b and 67ii.b) result from high ranking of drop topic, while clitic and DP result from high ranking of max(foc) and max(s)M. A problem remains regarding the optionality of 1st and 2nd person singular clitics together with focused DPs, i.e. the choice between (67ii.a) and (67ii.c). For this problem, the data seems to imply that a speaker-dependent re-analysis of verbal morphology takes place. While one (mostly older) group of speakers prefers the structure without clitics, another (mostly younger) group prefers the clitic doubling structure.25 Therefore, clitics seem to become the relevant subject agreement
226 Syntax at the interfaces
markers, though this development is not complete yet. For French subject clitics, the analysis is similar. As to the subject agreement paradigm in (68), most of the verbal agreement aYxes are no longer represented by any phonological material (see the phonological form of the verbs in square brackets).26 (68) Subject agreement in the verbal paradigm 1sg je porte [pfrt] 1pl on porte [pfrt] (nous) portons [pfrÁtf«] 2sg tu portes [pfrt] 2pl (vous) portez [pfrÁte:] 3sg (il/elle) porte [pfrt] 3pl (ils/elles) portent [pfrt]
In contrast to Piattino, the 3rd person may remain underspeciWed. Otherwise, clitics are obligatory if the verb forms are not distinctive. Therefore, 1st and 2nd person singular clitics and the clitic on together with the unmarked verb are obligatory. Nous and vous as clitics may be left out with co-referential DPs (i.e. Alexandra et moi) and the agreement-marked verb form. Let us now examine the semantic restrictions on clitic doubling.
5.3.3 Semantic restrictions on clitic doubling As to the semantic restrictions on clitic doubling, there are two diVerent ways of encoding them. I will concentrate on Romanian clitic doubling to illustrate the analysis. Let us Wrst consider the option that the lexical entries of the clitics are responsible for the restrictions on clitic doubling, a proposal that has been made by Suñer (1988, 1991, 1992). One has to assume then that clitics have semantic and discourse-referential features which must be compatible with the features of the respective DPs. The constraint unif(ref) is able to check such correspondence relations. This constraint resembles Suñer’s matching principle.27 (69) unif(ref) Corresponding elements are compatible with regard to their referential features.
In Romanian, focused indirect object DPs are only realised together with
Clitic doubling 227
clitics if they refer to speciWc entities. Otherwise, they occur without clitics. This is exempliWed by (70). (70) (Le) cauta[ leac unor boli incurabile. 3pl.io search.3sg remedy some.io.pl disease incurable ‘He searches remedies for incurable diseases.’ (Popescu 1997:54)
The sentence is grammatical either with or without a clitic. The interpretation, however, diVers. With the clitic doubling structure, well-known speciWc diseases are meant, while with the pure DP, it is unknown which diseases are meant. Therefore, it seems reasonable to assume that io clitics bear the feature [+speciWc]. If unif(ref) ranks above max(io)M, clitic doubling structures with unspeciWc DPs are excluded. (71) a.
The lexical entry of Romanian io clitics contains the feature [+speciWc] ([+hr, +lr] ⇒[+speciWc]). b. Ranking: max(foc), unif(ref) >> max(io)M
Let us now examine whether a similar analysis can also be adopted for Romanian do clitics. Focused do DPs occur together with clitics only if they refer to speciWc and human entities (see 72). Otherwise, they occur alone (see 73 and 74). (72) do DP: [+human, +speciWc] L– am va[ zut pe doctor. * Am va[ zut pe doctor. # L– am va[ zut. 3do-aux.1sg see.ptc acc doctor ‘I have seen the doctor.’ (73) do DP: [+human, -speciWc] *L– am va[ zut un copil. Am va[ zut un copil. # L– am va[ zut. 3do-aux1.sg see.ptc d.idf child ‘I have seen a child.’
228 Syntax at the interfaces
(74) do DP: [–human, +speciWc] *L– am va[ zut Wlmul. Am va[ zut Wlmul. # L– am va[ zut. 3do-aux.1sg see.ptc Wlm.df.m ‘I have seen the Wlm.’
In keeping with the discussion above, one must assume that the lexical entries of Romanian do clitics contain the features [+human] and [+speciWc]. The constraints unif(speciWc) and unif(human) must then both rank above max(do)M (see 75). (75) a.
The lexical entry of Romanian do clitics contains the features [+human] and [+speciWc]. ([+hr] ⇒ [+speciWc], [+hr,-lr] ⇒ [+human]). b. Ranking: max(foc), unif(ref) >> max(do)M
This kind of analysis runs into problems, however, if one considers other kinds of Romanian constructions. Focused direct objects may also occur on the left of the verb. Clitics are then obligatory with all speciWc direct object DPs, regardless whether human or not (see 76–78). (76) Pe doctor l-am va[ zut. *Pe doctor am va[ zut. ‘I have seen the doctor.’ (77) *Un copil l-am va[ zut. Un copil am va[ zut. ‘I have seen a child.’
(78)
Filmul l-am va[ zut. *Filmul am va[ zut. ‘I have seen the Wlm.’
There is no intonational break or any other indication that structures like this are left dislocation structures. Instead, the object is a proper part of the sentence. Nevertheless, the position left of the verb is the result of some kind of movement which is reXected in the grammar by a constraint violation. The direct object Wlmul in (78), for instance, is assumed to be moved to a focus position to the left of the VP resulting in the structure [FP Wlmul [VP l-am va[ zut]]. The constraint which is violated if the object does not occur in its basic position is domain-oVP, which requires that objects are realised in the VP (see 79).
Clitic doubling 229
(79)
domain-oVP
For each object role of a verb V (i.e. an argument role bearing the feature [+hr]), at least one correspondent (i.e. a direct object DP or a direc object clitic) must be realised within the VP.
The constraint unif(ref) must then split up into a constraint which requires that the DP and the clitic are compatible with respect to the feature [+human] and another one which requires that they are compatible with respect to the feature [+speciWc], i.e. unif(human) and unif(speciWc) respectively. domain-oVP must be ranked diVerently with regard to unif(speciWc) and unif(human) so that unspeciWc human DPs still remain undoubled (80) but speciWc non-human DPs are doubled obligatorily (81). (80) Un copil am va[ zut.: max(foc), unif(speciWc) >> domain-oVP >> max(dO)M (81) Filmul l-am va[ zut.: max(foc) >> domain-oVP >> unif(human) >> max(dO)M
A ranking like (81) allows a feature clash with regard to referential features, since the two elements which refer to the direct object are conversely assigned [+human] and [–human]. The question arises therefore how such elements can be interpreted. Still more awkward is the interpretation of the direct object if it is topical. As discussed above, topic do-DPs are dropped and the arguments are realised by clitics. In such sentences, the clitic refers to entities which are speciWc, regardless of whether they are human or not. In the answer to the question in (82), the clitic refers to the speciWc but non-human noun Wlmul. If the clitic had the feature [+human] in its lexical entry, this would mean that the only element in the sentence which refers to the non-human object would bear the feature [+human], an assumption which does not make any sense and is incompatible with what is known about semantics. (82) Filmul?
L-am va[ zut. *L-am va[ zut Wlmul. # Am va[ zut Wlmul. ‘The Wlm?’ – ‘I have seen it.’
Therefore, the assumption that the feature [+human] is part of the lexical entry of the clitics must be rejected. In general, violations of unif(ref) are not
230 Syntax at the interfaces
in accord with the concept of a compositional semantic component in grammar and cannot be accepted. The second option in the analysis of the restrictions on clitic doubling is that it is not the lexical entries but the grammar which conWnes the occurrence of clitics. The clitics do not have any semantic or discourse-referential features but only agreement and case features as introduced in chapter 3. Grammatical economy constraints prohibit that noun referents with low salience are marked morphologically as aYxes or clitics, an idea that goes back to Ortmann (2000).28 An economy constraint that conveys such a prohibition is avoidagr (83). For the speciWc clitic contexts, avoidagr refers to diVerent features such as [human] and [speciWc] in the diVerent instantiations in (83).29 (83) avoidagr(-hum) avoidagr (-spec)
Non-human noun referents are not marked morphologically. Non-speciWc noun referents are not marked morphologically.
avoidagr(-hum) and avoidagr(-spec) can rank diVerently — just like the
unification constraints discussed above. This is shown in the tableaux in (84) and (85). (84) Romanian: focused, human, unspeciWc direct object DP left from the verb max AvoidAgr domain- max λy λxV(x,y); y = un copil<-speciWc,+human,+focus> (foc) (-spec) oVP (do)M a. Un copil am va[ zut. b. L-am va[ zut. c. Un copil l-am va[ zut.
* *!
*
* *!
(85) Romanian: focused, non human, speciWc direct object left from the verb λy λx V(x,y); y =Wlmul<+speciWc,-human,+focus> a. Filmul am va[ zut. b. L-am va[ zut. c. Filmul l-am va[ zut.
max domain- AvoidAgr max (foc) oVP (-hum) (do)M *! *!
* * *
Clitic doubling 231
Although the *m-structure constraints are ranked at the same positions as the unification constraints in (80) and (81) above, there is a huge diVerence between both analyses. While violability of the unification constraints poses a severe problem with regard to interpretation, such a problem does not arise with violability of the *m-structure constraints. The latter do not rely on any lexical features of the clitics. UniWcation is not violated since the clitics are assumed to be underspeciWed with regard to the features [human] and [speciWc].30 Let us Wnally examine how these insights are related to the analysis of the clitic choice in the morphologically restricted combinations presented in chapter 4. As pointed out there, 1o and 2o cannot combine with each other nor with 3io in the same sequence (see 86, 87 and 88). Instead of a clitic sequence, the direct object cliticises, while the indirect object is realised as a full pronoun or DP. (86) 1pl.do / 2pl.io a.
*Ni va[ / * Vi ne recomanda[ voua[ pe noi. 1pl.o 2pl.o /2pl.o 1pl.orecommend prn.2pl.io acc prn.1pl
b. *Va[ recomanda voua[ 2pl.o recommend prn.2pl.io
(pe noi). (acc prn.1pl)
c.
pe noi. acc prn.1pl.
Ne recomanda[ (voua[ ) 1pl.o recommend (prn.2pl.io) ‘He recommends us to you.’
(87) 1pl.io / 2pl.do a.
*Ni va[ / * Vi ne recomanda[ noua[ pe voi. 1pl.o 2pl.o/ 2pl.o 1pl.o recommend prn.1pl.io acc prn.1pl
b. *Ne recomanda[ (noua[ ) 1pl.o recommend (prn.1pl.io)
pe voi. acc prn.1pl
c.
(pe voi). (acc prn.1pl)
Va[ recomanda[ noua[ 2pl.o recommend prn.1pl.io ‘He recommends you to us.’
(88) 2pl.do / 3io a.
*I va[ / * va[ i recomand pe voi 3io 2pl.o/2pl.o 3io recommend acc prn.2pl lui / profesorului. prn.3io professor.dat
232 Syntax at the interfaces
b. *Îi recomand (lui / profesorului) pe voi. 3io recommend (prn.3io / professor.dat) acc prn.2pl c.
Va[ recomand lui / profesorului (pe voi). 2pl.o recommend prn.3io / professor.dat (acc prn.1pl) ‘I recommend you to the professor.’
As discussed above, clitic doubling is obligatory for speciWc indirect objects and for direct objects which are speciWc and human. In unrestricted circumstances, the 3io clitic in (88) is therefore obligatory as well as the 1o and 2o clitics in (86, 87 and 88), since 1st and 2nd person imply the features [+speciWc,+human]. In chapter 4, it has been argued that equally highly ranked alignment constraints for the person features [+1] and [+2] and for the case feature [+lr] are responsible for the restrictions on the combination of these clitics. In these cases it is always the direct object which is realised as a clitic, while the indirect object is realised as a full pronoun. This is also the case if the indirect object has speciWc and human reference. The question that arises now is how this contradictory clitic choice in morphologically restricted and morphologically unrestricted circumstances can be explained. While it is the direct object which is realised as a clitic in the former case, in the latter case it is the indirect object. The answer to this question comes from the kind of morphological constraints which have been argued to be responsible for the clitic choice in section 4. Analogously to economy constraints for referents with low salience, there are certain constraints which require an utterance to be as explicit as possible, in particular when it comes to particularly salient referents. In chapter 4, it was argued that max(person)M requires 1st and 2nd person referents to be morphologically marked. Moreover, max(+lr/+hr) requires that the grammatical relation of the arguments which are marked at the verb is unambiguously identiWable. If both economy and explicitness requirements are in conXict, as is the case if combinations of clitics are morphologically prohibited, the requirement of deWniteness of arguments is obviously more important than the restrictions on economical argument marking.
5.4 Conclusions In the present chapter, the shortcomings of purely syntactic analyses of clitic placement and clitic doubling have been demonstrated. Moreover, an alternative analysis in terms of Optimality Theory has been proposed.
Notes 233
The placement of clitics within the sentence has been attributed to three kinds of alignment constraints. First, syntactic alignment constraints determine the order of clitics and other elements within their domain V’. Second, morphological alignment constraints determine whether the clitics occur before or after the verb. These alignment constraints refer to certain morphological features of the verbs. Finally, prosodic alignment constraints determine whether the clitics are integrated into the phonological word on their right or on their left. Mesoclitic forms in Portuguese as well as clitic split structures in Romanian have been shown to result from the interaction of these constraints. However, there are still open questions. The complex data on clitic climbing has been exempliWed in section 5.2.3. For these data, an analysis has only been hinted at. Moreover, Portuguese syntactically triggered proclisis needs a more thorough examination. Clitic doubling and clitic placement interact with regard to morphological restrictions on the combination of clitics, but clitic doubling is not licensed by particular positions in the sentence. The distribution of clitics and DPs to argument roles has been explained in the following way. Clitics are agreement markers and are able to saturate argument roles. Whether co-referential DPs occur in addition depends on the discourse-referential properties of their referents. Clitics occur in addition to DPs — i.e. the typical clitic doubling structure — if the requirement for morphological argument realisation ranks higher than the prohibition of multiple correspondence relations and if the morphosyntactic features of clitic and DP are compatible. avoidagr restricts clitic doubling to salient noun referents. So far, person-dependent clitic doubling as well as clitic doubling depending on speciWcity and animacy have been analysed. A problem still open to further research is the restriction of direct object clitic doubling to pronouns in European Spanish.
Notes 1. For overviews on syntactic approaches on cliticisation see also Manzini (1998), Sportiche (1996), Uriagereka (1995). 2. On criticism of transformational approaches to the order of clitics, see also Morin (1979). 3. It remains unclear how the metamorphosis of strong clitics from phrases to heads fits in the common assumptions about generative syntax. 4. Another syntactic approach to Romance clitics, in particular on Romanian clitics is
234 Syntax at the interfaces
that of Dobrovie-Sorin (1990, 1994, 1995). On the problems with her analysis see Legendre (2000b) and Monachesi (2000). 5. On the relevance of the 2nd position to clitic placement in Romance, see among others Fontana (1996). 6. For the Bulgarian negation clitic ne and the question clitic li, prosody is relevant as well. As observed by Hauge (1976: 17,16), stress is assigned to the syllable following ne, and li is placed immediately to the right of the first stressed element within the verb constituent. For these clitics, Legendre assumes additional prosodic constraints which interact with the morphological constraints as well as with other prosodic constraints (e.g. those which require clitics to be stressless). One constraint requires that the right edge of ne be aligned with the left edge of the head of a prosodic word and another one says that the left edge of li must be aligned with the right edge of a prosodic word. 7. See also the constraint morpheme integrity in chapter 4 (p. 169) . 8. On the historical development of clitic placement, see, e.g., Fischer (2000); Fontana (1996); Jacobs (1993). 9. In the latter case, an additional — higher-ranked — alignment constraint requires that subjects are still further left in IPs, according to the following ranking: align-l(specifier,XP) >> align-l(case/person, IP). 10. Legendre shows that, based on this assumption, 2nd position clitics and verbal clitics are not different with respect to the domain of the alignment constraints but with respect to their ranking. 11. Note that gerunds end in [u] only if clitics follow. 12. The align-l constraints in (17) correspond to Legendre’s edgemost constraints, see the discussion on the different interpretation of alignment constraints in chapter 4 (pp. 143–145). 13. Recall that mesoclisis is also attested to some extent in Archaic Romanian and other medieval varieties of Romance languages (see also chapter 2). Other languages also show word internal clitics, see e.g. Harris’ (2000) analysis of Udi clitics. 14. Note that stress marks on the vowels are taken from van der Leeuw and do not always coincide with Portuguese orthographic stress marks. 15. Van der Leeuw call this constraint lexical integrity. 16. In Gerlach (2001b) it is argued that /l/ in the onset of the clitic results from (i) a lexically underlying but unspecified onset in the syllable structure of 3rd person clitics and (ii) of spreading of the features of the preceding coda. According to a feature geometry as suggested by Grijzenhout (2001) /l/ is exactly what is expected. 17. On clitic climbing in Spanish see also Franco & Landa (1995). 18. In French modal verb constructions, negation and subject clitic occur in front of the matrix verb, but the internal clitics occur in front of the infinitive. 19. See, e.g., Schwarze (1995: 197–202) for a more elaborate description. 20. Data from Canclini (1997); Farkas (1978); Farkas & Kazazis (1980); Gierling (1997); Kaiser (1992); Parodi (1998a); Popescu (1997). 21. See also Burzio (1998) on multiple correspondence relations in morphology. 22. On the discourse-referential properties of full DPs in Spanish, see also Bentivoglio (1993).
Notes 235
23. Subject clitic doubling is also attested in many other French and northern Italian dialects, see e.g. Auger (1993, 1994); Auger & Miller (1995); Brandi & Cordin (1989); Poletto (1999); Rizzi (1986); Roberge (1997); Suñer (1992). 24. The 3rd person plural is presumably only marked for number and not for person. 25. This data, however, is somewhat vague and needs further examination. The basis is grammaticality judgements of about 10 speakers who live in Piatta. 26. See also Auger (1993, 1994) who also analyses French clitics as agreement affixes. 27. In Suñer’s analysis, S-clitics in northern Italian dialects and dO-clitics in Argentinian Spanish are lexically specified as [+specific] (i.e. ‘identifiable’). The Matching Principle requires that clitics and NPs in the same chain must have identical features and thus restricts their co-occurrence. 28. Ortmann refers to Aissen (1999, 2000) and Stiebels (2000) for a more elaborated version of economy constraints which restrict morphological marking. 29. Note that this kind of restriction on clitics is exactly what is expected as regards the harmonic alignment of markedness hierarchies such as +clitic>–clitic and +spec>-spec (see Stiebels 2000a,b). 30. As an alternative, one could also think of a kind of job-sharing between lexical entries and economy constraints. Since *m-structure(-specific) is never violated, the feature [+specific] could be lexical. This is not true for the feature [+human].
236 Syntax at the interfaces
Summary 237
Chapter 6
Summary
In this thesis, the properties of Romance clitics have been examined in detail. The Romance languages possess a number of elements which are traditionally called clitics. It has been shown in chapter 2 that the phonological and morphological properties of pronominal clitics are unique and that they resemble neither aYxes nor words, so that it is justiWed to speak of a distinct morphological category “clitic”. A distinct prosodic category like the clitic group (Hayes 1989; Nespor & Vogel 1986), however, is not necessary in order to capture the behaviour of these elements since clitic phenomena can be explained on the basis of the prosodic hierarchy and the diVerent processes of integration, adjunction and incorporation into prosodic constituents. Morphologically, the respective elements have recently been re-categorised as inXectional aYxes or words (see e.g. Barbu 1998; Crysmann 2000; Miller 1992; Miller & Sag 1997; Monachesi 1995, 2000). A detailed re-examination according to the Zwicky criteria and Nübling’s bundle of scales has lead to the conclusion that Romance pronominal subject, object and reXexive clitics as well as locative and partitive clitics are not yet inXectional aYxes, but special clitics. Clitics do not combine with stems as is expected from aYxes, but with words. They are not as selective with respect to their host category as inXectional aYxes are, and their phonological host may diVer from their morphosyntactic host, a behaviour which is not attested for inXectional aYxes. Unlike aYxes, clitics change the direction of attachment with respect to their host, and their rigid order holds with respect to each other and not with respect to the host. Moreover, pre-positioning of clitics in European Portuguese is syntactically or semantically triggered and even prevails over mesoclisis in future tense and conditional mood. The morphophonological processes between clitics and hosts often resemble the morphophonological processes on word boundaries rather than those between aYxes and stems. The strict order of clitics with respect to each other suggests that they form a unit, the clitic sequence, which is subject to speciWc ordering requirements. The morphophonological processes between clitics in such clitic sequences
238 Summary
are not attested elsewhere in the morphophonology of the respective languages, so that it seems reasonable to assume constraints speciWc to clitic sequences. As to non-pronominal elements, the picture is not as uniform. Auxiliaries in Romanian have been shown to be clitics for the same reasons as pronominal clitics. Similar elements in Portuguese have been analysed as inXectional endings which are not yet completely morphologised. While the negation element in French has been shown to be a clitic because it is positioned between subject and object clitics, the negation element in Romanian is best considered a word because it occurs to the left of the clitic cluster and not within it and may also occur in isolation. Monosyllabic intensiWers in Romanian, Wnally, have been argued to be derivational aYxes. Moreover, the concept of phrasal aYxation (Klavans 1980, 1985; Anderson 1992) has been shown to be too rigid for the Romance clitic phenomena. Following the discussion on the morphological status of Romance clitics, the clitic paradigms were introduced in chapter 3, before I delineated the morphological model and outlined the place of morphology in grammar. As a result of a high ranked FAITHFULNESS constraint which requires that arguments are realised morphologically, clitics have to occur in the output to realise particular argument roles in the input. In competition with this constraint, the prohibition of multiple correspondence relations restricts the number of output correspondents to a single argument role. The agreement relation between co-occurring clitics and DPs is mediated by the concept of Transitive Correspondence which requires the morphosyntactic features of clitics and DPs to be checked. The clitics have then been provided with lexical entries in the manner of Minimalist Morphology (Wunderlich & Fabri 1996). In comparison to a similar analysis by Grimshaw (1997), it has been shown how the interaction of violable FAITHFULNESS and MARKEDNESS constraints may determine the choice of single clitics. It has been shown that the clitic inventories of diVerent Romance languages — though superWcially similar — are distinct when it comes to their lexical speciWcation. Universal tendencies, however, have been observed as regards which features are speciWed and which features remain underspeciWed. Romance clitic sequences were analysed in chapter 4. The order and combinatory restrictions on clitic sequences were accounted for by the interaction of morphological and morphosyntactic constraints whose domain was assumed to be the clitic sequence. Morphological ALIGNMENT constraints require that the left edge of a sequence of clitics should correspond to the left
Summary 239
edge of elements marked for case and person. Indirect object clitics are more marked than direct object clitics and Wrst and second persons are more marked than third persons. If the ALIGNMENT constraints are high-ranked, a marked clitic always precedes an unmarked one. In sequences which contain two marked clitics (an indirect object clitic and a 1st or 2nd person clitic), the combination fails due to conXicting ALIGNMENT constraints. Clitics can only combine if their positions with respect to the markedness hierarchies diVer in harmonic ways. Otherwise the combination of clitics fails. In such cases, only one of the arguments cliticises, while the other one is realised as a full pronoun. Which of the arguments is realised as a clitic results from additional constraints concerning the speciWcity of feature marking. As to the morphology-phonology interface, certain prosodic constraints determine that clitic sequences are prosodically as minimal as possible. These constraints are particularly important in Romanian and Portuguese, where the whole clitic sequence is integrated into the same prosodic word as their host. It has been shown that opaque clitics are best analysed if phonological OCP-like constraints are taken into account. In such cases, the optimal clitic sequence either deviates phonologically from the most faithful output candidate, or clitic substitution or clitic deletion takes place. Which of these repair mechanisms is chosen depends on the lexical inventory of the language. In the Wnal chapter, the shortcomings of purely syntactic analyses of clitic placement and clitic doubling have been demonstrated. Moreover, an alternative analysis in terms of Optimality Theory has been proposed. The placement of clitics within the sentence has been attributed to three kinds of ALIGNMENT constraints, namely syntactic, morphological and prosodic ALIGNMENT. The syntactic constraints align clitics with the domain of V’, determining the order of clitics and other elements within this domain. The morphological constraints refer to the verb as their domain. They determine whether the clitics occur before or after the verb, depending on the morphological features of the verb. The prosodic constraints, Wnally, determine whether clitics are integrated into the phonological word on their right or into the phonological word on their left respectively. Mesoclitic forms in Portuguese as well as clitic split structures in Romanian have been shown to result from the interaction of these constraints. Clitic doubling and clitic placement interact with regard to morphological restrictions on the combination of clitics, but clitic doubling is not licensed by particular positions in the sentence. The distribution of clitics and DPs to argument roles has been explained in the following way. Clitics are
240 Summary
the unmarked argument realisation. In Italian and other languages where clitic doubling is prohibited, the requirement for morphological argument realisation ranks lower than the ban on multiple correspondence relations. The decision whether the argument role is realised by a clitic or by a DP depends on its relevance for the information structure. As a tendency, topical arguments are realised by clitics, while focused arguments are realised by DPs. If clitic doubling is allowed in a language, on the other hand, the ranking of the constraints is the reverse. Clitics may occur in addition to DPs — i.e. the typical clitic doubling structure — if the requirement for morphological argument realisation ranks higher than the prohibition of multiple correspondence relations and if the morphosyntactic features of clitic and DP are compatible. Only focused co-referential DPs occur in addition to clitics. Other than regular agreement markers, however, the occurrence of clitics is further restricted by grammatical economy constraints which restrict clitic doubling to salient noun referents.
References 241
References
Abraham, Werner. 1995. “Diathesis: the middle, particularly in West-Germanic. What does reXexivation have to do with valency reduction?” In Discourse grammar and typology, W. Abraham, T. Givon and S. Thompson (eds), 3–47. Amsterdam: John Benjamins. Aguado, Miquel and Christian Lehmann. 1989. “Zur Grammatikalisierung der Klitika im Katalanischen.” In Romanistik, Sprachtypologie und Universalienforschung [Beiträge zum Freiburger Romanistentag 1987], Wolfgang Raible (ed), 151–161. Tübingen: Gunter Narr. Aguado, Miquel and Grzegorz Dogil. 1989. “Clitics in lexical phonology: alleged counterevidence?” Linguistische Berichte 120:99–106. Aissen, Judith. 1999. “Markedness and subject choice in Optimality Theory.” NLLT 17(4):673–711. Aissen, Judith. 2000. DiVerential object marking: iconicity vs. economy. Ms. UC Santa Cruz. Anagnostopoulou, Elena and Anastasia Giannakidou. 1995. “Clitics and prominence, or why speciWcity is not enough.” CLS 31, Clitics Parasession Volume:1–49. Anagnostopoulou, Elena. 1999. “Conditions on clitic doubling in Greek.” In Clitics in the Languages of Europe, Henk van Riemsdijk (ed), 761–798. Berlin/New York: Mouton de Gruyter. Anderson, Stephen R. 1992. A-Morphous Morphology [Cambridge Studies in Linguistics 62]. Cambridge: CUP. Anderson, Stephen R. 1993. “Wackernagel’s revenge: clitics, morphology and the syntax of second position.” Language 69:68–98. Anderson, Stephen R. 1996. “How to put your clitics in their place or why the best account of second-position phenomena may be something like the optimal one.” The Linguistic Review 13:165–191. Anttila Arto and Y.Y. Cho. 1998. “Variation and change in Optimality Theory.” In Lingua special issue on Optimality Theory, Dicky Gilbers and Helen de Hoop (eds), 31–56. Amsterdam: Elsevier Science. Artstein, Ron. 1998. Hierarchies. Ms. Rutgers University. Auger, Julie. 1993. “More evidence for verbal agreement marking in colloquial French.” In Linguistic perspectives on Romance languages, W. Ashby, M. Mithun, G. Perissinotto and E. Raposo (eds), 177–198. Amsterdam/Philadelphia: John Benjamins. Auger, Julie. 1994. Pronominal clitics in Québec Colloquial French: a morphological analysis. Phd diss. Philadelphia: University of Philadelphia. Auger, Julie and Philip Miller. 1995. Les aYxes pronominaux du français: à l’interface entre syntaxe, morphologie et phonologie. Ms. McGill University, Montréal and Université Lille 3.
242 References
Avgustinova, Tania. 1997. Word order and clitics in Bulgarian. Phd diss. University of the Saarland. Avram, Andrei. 1986. “Sandhi phenomena in Romanian.” In Sandhi Phenomena in the Languages of Europe, Henning Andersen (ed), 551–574. Berlin/New York/Amsterdam: Mouton de Gruyter. BaWle, Laura. 1993. Fonologia prosodica e teoria metrica: accento, clitizzazione e innalzamento vocalico in napoletano. Phd diss. Universitá di Firenze. BaWle, Laura. 1994. “La riassegnazione postlessicale dell’ accento nel napoletano.” Quaderni del dispartimento di linguistica — Università di Firenze:1–23. Barbosa, Pilar. 1993. “On clitic placement in Old Romance and European Portuguese.” CLS 29,1:33–59. Barbosa, Pilar. 1996. “Clitic placement in European Portuguese and the position of subjects.” In Approaching Second: Second Position Clitics and Related Phenomena, Aaron L. Halpern and Arnold M. Zwicky (eds), 1–40. Stanford: CSLI publications. Barbu, Ana Maria. 1998. Complexul Verbal. Ms. Institutul de Cervceta™ ri în Informatica™ . Bucuresti. Belletti, Adriana. 1999. “Italian/Romance clitics: structure and derivation.” In Clitics in the Languages of Europe, Henk van Riemsdijk (ed), 543–579. Berlin/New York: Mouton de Gruyter. Bentivoglio, Paola. 1993. “Full NPs in spoken Spanish: a discourse proWle.” In Linguistic perspectives on Romance Languages, W. Ashby, M. Mithun, G. Perissinotto and E. Raposo (eds), 211–224. Amsterdam/Philadelphia: John Benjamins. Benua, Laura. 1995. “Identity eVects in morphological truncation.” In Papers in Optimality Theory [University of Massachusetts Occasional Papers 18], Jill N. Beckman, Laura Walsh Dickey and Suzanne Urbanczyk (eds), 77–136. Cambridge MA: MIT Press. Berendsen, Egon. 1986. The phonology of cliticization. Dordrecht: Foris. Bonet, Eulalia. 1995. “Feature structure of Romance clitics.” NLLT 13:607–647. Booij, Geert. 1996. “Cliticization as prosodic integration: The case of Dutch.” The Linguistic Review 13:219–242. Brandi, Luciana and Patrizia Cordin. 1989. “Two Italian dialects and the null subject parameter.” In The Null Subject Parameter, Osvaldo Jaeggli and Kenneth SaWr (eds), 111–142. Dordrecht: Reidel. Bresnan, Joan. 2000. “Optimal syntax.” In Optimality Theory: syntax, phonology, and acquisition, Joost Dekkers, Frank van der Leeuw, and Jeroen van de Weijer (eds), 334– 380. Oxford: OUP. Burzio, Luigi. 1998. “Multiple correspondence.” In Lingua special issue on Optimality Theory, Dicky Gilbers and Helen de Hoop (eds), 79–110. Amsterdam: Elsevier Science. Calabrese, Andrea. 1991. “I pronomi personali.” In Grande grammatica italiana di consultazione [volume I] Salvi Renzi and Anna Cardinaletti (eds), 535–641. Bologna: Il Mulino. Calabrese, Andrea. 1995. “Syncretism phenomena in the clitic systems of Italian and Sardinian dialects and the notion of morphological change.” NELS:151–173.
References 243
Canclini, Emanuela. 1999. Die italienische VerbXexion: eine morphophonologische Analyse. Masters thesis. Düsseldorf: Heinrich-Heine-University. Cardinaletti Anna. 1999. “Pronouns in Germanic and Romance languages: an overview.” In Clitics in the languages of Europe, Henk van Riemsdijk (ed), 33–82. Berlin/New York: Mouton de Gruyter. Cardinaletti, Anna and Michal Starke. 1999. “The typology of structural deWciency: on the three grammatical classes.” In Clitics in the languages of Europe, Henk van Riemsdijk (ed), 145–233. Berlin/New York: Mouton de Gruyter. Carvalho, Joaquim Brandâo de. 1989. “Phonological conditions on clitic placement: on syntactic evidence for stress and rhythmical patterns.” Linguistics 27:405–436. CZ avar, Damir and Chris Wilder. 1994. “‘Clitic third’ in Croatian.” ESF-Eurotyp Working Papers, Grammatical Models Section 6:19–61. Cocchi, Gloria. 2000. “Free clitics and bound aYxes: towards a unitary analysis.” In Clitics in phonology, morphology and syntax, Birgit Gerlach and Janet Grijzenhout (eds), 85– 120. Amsterdam/Philadelphia: John Benjamins. Crysmann, Berthold. 2000. “Clitics and coordination in linear structure.” In Clitics in phonology, morphology and syntax, Birgit Gerlach and Janet Grijzenhout (eds), 121– 160. Amsterdam/Philadelphia: John Benjamins. Dimitrova-Vulcharnova, Mila and Lars Hellan. 1999. “Clitics and Bulgarian clause structure.” In Clitics in the languages of Europe, Henk van Riemsdijk (ed), 469–540. Berlin/ New York: Mouton de Gruyter. Dizionari Garzanti. 1994. Bologna: Il Mulino. Dobrovie-Sorin, Carmen. 1990. “Clitic doubling, wh-movement and quantiWcation in Romanian.” Linguistic Inquiry 21:351–397. Dobrovie-Sorin, Carmen. 1994. The syntax of Romanian: comparative studies in Romance. Berlin/New York/Amsterdam: Mouton de Gruyter. Dobrovie-Sorin, Carmen. 1995. “Clitic clusters in Romanian: deriving linear order from hierarchical structure.” In Advances in Romanian linguistics, Guglielmo Cinque and G. Giusti (eds), 55–82. Amsterdam/Philadelphia: John Benjamins. Emonds, Joseph. 1975. “A transformational analysis of French clitics without positive output constraints.” Linguistic Analysis 1:3–24. Enc¸, Murvet. 1991. “The semantics of speciWcity.” Linguistic Inquiry 22:1–25. Farkas, Donka. 1978. “Direct and indirect object reduplication in Rumanian.” CLS 14:88– 97. Farkas, Donka and Kostas Kazazis. 1980. “Clitic pronouns and topicality in Rumanian.” CLS 16:75–82. Fischer, Susann. 2000. Reopening the case of Old Romance clitic placement. Data from Old Catalan. Ms. University of Potsdam. Fontana Josep M. 1993. Phrase structure and the syntax of clitics in the history of Spanish. Phd diss. University of Pennsylvania. Fontana, Josep M. 1994. “A variationist account of the development of the Spanish clitic system.” CLS 30(2):87–100. Fontana, Josep M. 1996. “Phonology and syntax in the interpretation of the ToblerMussaWa Law.” In Approaching Second: second position clitics and related phenomena,
244 References
Aaron L. Halpern and Arnold M. Zwicky (eds), 41–83. Stanford: CSLI publications. Franco, Jon. 1993. “Conditions on clitic doubling: the agreement hypothesis.” International Journal of Basque Linguistics and Philology 27.1:285–298. Franco, Jon and Alazne Landa. 1995. “Conditions after SPELL OUT: the case of Spanish clitic climbing.” CLS 31, Clitics Parasession Volume:105–117. Fulmer, S. 1990. “Dual position aYxes in Afar: an argument for phonologically-driven morphology.” In Proceedings of the Ninth West Coast Conference on Formal Linguistics, Aaron Halpern (ed), 189–203. Stanford. Gerlach, Birgit. 1998a. “Optimale Klitiksequenzen.” Zeitschrift für Sprachwissenschaft 17:35–91. Gerlach, Birgit. 1998b. Restrictions on clitic sequences and conditions on the occurrence of clitics in Romance. Arbeiten des Sonderforschungsbereichs 282: Theorie des Lexikons, No. 105. Düsseldorf: Heinrich-Heine-University. Available as ROA 277-0898. Gerlach, Birgit. 2001a. “Optimality aspects of clitic doubling in Romance.” In Clitiques et cliticisation. Actes du Colloque de Bordeaux, octobre 1998, Claude Muller (ed), 343– 363. Geneve: Edition Slatkine. Gerlach, Birgit. 2001b. Neutralisation in European Portuguese. Ms. Düsseldorf: HeinrichHeine-University. Gierling, Diana. 1997. “Clitic doubling, speciWcity, and focus in Romanian.” In Clitics, pronouns, and movement, J.R. Black and V. Motapanyane (eds), 63–85. Amsterdam/ Philadelphia: John Benjamins. Givón, Talmy. 1976. “Topic, pronoun, and grammatical agreement.” In Subject and topic, Charles Li and Sandra Thompson (eds), 146–429. Austin: University of Texas Press. Golston, Chris. 1995. “Syntax outranks phonology: evidence from Ancient Greek.” Phonology 12:343–368. Grandgent C. 1933. From Latin to Italian. Cambridge, MA: Harvard University Press. Grijzenhout, Janet. 2001. West-Germanic sound structure. A contrastive study. Habilitationsschrift. Düsseldorf: Heinrich-Heine-University. Grijzenhout, Janet and Martin Krämer. 1998. Final devoicing and voicing assimilation in dutch derivation and cliticization. Arbeiten des Sonderforschungsbereichs 282: Theorie des Lexikons. Nr. 106. Düsseldorf: Heinrich-Heine-University. Grimshaw, Jane. 1997. “The best clitic: constraint conXict in morphosyntax.” In Elements of grammar, Liliane Haegemann (ed), 169–196. Dordrecht: Kluwer. Grimshaw, Jane. 2001. “Optimal clitic positions and the lexicon in Romance clitic systems.” In Optimality-theoretic syntax, G. Legendre, J. Grimshaw and S. Vikner (eds). Cambridge, MA: MIT Press. Grimshaw, Jane and Vieri Samek-Lodovici. 1995. “Optimal Subjects.” In Papers in Optimality Theory [University of Massachusetts Occasional Papers 18], Jill N. Beckmann, Laura Walsh Dickey and Suzanne Urbanczyk (eds), 589–605. Cambridge, MA: MIT Press. Grimshaw, Jane and Vieri Samek-Lodovici. 1998. “Optimal subjects and subject universals.” In Is the best good enough? Optimality and competition in syntax, Pilar Barbosa, D. Fox, P. Hagstrom, M. McGinnis and D. Pesetsky (eds) 193–219. Cambridge, MA: MIT Press.
References 245
Hall, T.A. 1999. “Phonotactics and the prosodic structure of German function words.” In Studies on the Phonological Word [Current Studies in Linguistic Theory 174], T.A. Hall and Ursula Kleinhenz (eds), 99–132. Amsterdam/Philadelphia: John Benjamins. Halle, Morris and Alec Marantz. 1993. “Distributed Morphology and the pieces of inXection.” In The view from building 20: essays in linguistics in honor of Sylvain Bromberger [Current Studies in Linguistics 24], K. Hale and S.J. Keyser (eds), 111–176. Cambridge, MA: MIT Press. Halle, Morris and Alec Marantz. 1994. “Some key features of Distributed Morphology.” In Papers on phonology and morphology [The MIT Working Papers in Linguistics 21], Andrew Carnie and Heidi Harley (eds), 275–288. Cambridge, MA: MIT Press. Halpern, Aaron L. 1992. Topics in the placement and morphology of clitics. Phd diss. Stanford University. Halpern, Aaron L. 1998. “Clitics.” In The handbook of morphology, Andrew Spencer and Arnold M. Zwicky (eds). Oxford: Blackwell. Harris, Alice. 2000. “Where in the word is the Udi clitic?” Language 73(3):593–616. Harris, James. 1994. “The syntax-phonology mapping in Catalan and Spanish clitics.” In Papers on phonology and morphology [The MIT Working Papers in Linguistics 21], Andrew Carnie and Heidi Harley (eds), 321–353. Cambridge, MA: MIT Press. Harris, James. 1995. “The morphology of Spanish clitics.” In Evolution and revolution in linguistic theory, Hector Campos (ed), 168–197. Washington: Georgetown University Press. Harrison, David K. 1997. “The morphology of special NP clitics: the deWnite article in Albanian.” In Yale Amorphous Linguistic Essays. Studies in the morphosyntax of clitics, Lizanne Kaiser (ed), 45–73. New Haven: Yale University. Haspelmath, Martin. 2001. Explaining the Ditransitive Person-Role Constraint: A usagebased approach. Ms. Leipzig: Max-Planck-Institut für evolutionäre Anthropologie. Hauge, Kjertil R. 1976. The word order of predicate clitics in Bulgarian. Oslo: Universitetet i Oslo, Slavisk-Baltisk Institutt. Hayes, Bruce. 1989. “The prosodic hierarchy in meter.” In Rhythm and meter, Paul Kiparsky and G. Youmans (eds), 201–260. New York/Orlando: Academic Press. Heim, Irene. 1987. “Where does the deWniteness restriction apply? Evidence from the deWniteness of variables.” In The representation of (in)deWniteness, Eric Reuland and Alice ter Meulen (eds), 21–42. Cambridge, MA: MIT Press. Herschensohn, Julia. 1980. “On clitic placement in French.” Linguistic Analysis 6:187– 219. Hetzron, Robert. 1975. “Where the grammar fails.” Language 51:859–72. Hetzron, Robert. 1976. “Clitic pronouns and their linear representation.” Forum Linguisticum 1:189–215. Hofmann, J.B. and Anton Szantyr. 1972. Lateinische Syntax und Stilistik. Handbuch der Altertumswissenschaft 2. Munich: Beck. Hundertmark-Santos Martins, Maria Teresa. 1998. Portugiesische Grammatik. 2. verbesserte AuXage. Tübingen: Niemeyer. Inkelas, Sharon. 1989. Prosodic constituency in the lexicon. Phd diss. Stanford University.
246 References
Inkelas, Sharon. 1994. “The consequences of optimization for underspeciWcation.” NELS 25:287–302. Inkelas, Sharon and Draga Zec. 1995. “Syntax-phonology interface.” In The handbook of phonological theory, John A. Goldsmith (ed), 535–549. Oxford: Blackwell. Jacobs, Haike. 1993. “The phonology of enclisis and proclisis in Gallo-Romance and Old French.” In Linguistic perspectives on Romance languages, W. Ashby, M. Mithun, G. Perissinotto and E. Raposo (eds), 149–163. Amsterdam/Philadelphia: John Benjamins. Jaeggli, Osvaldo A. 1982. Topics in Romance syntax. Dordrecht: Foris. Jaeggli, Osvaldo A. 1986. “Three issues in the theory of clitics: case, doubled NPs, and extraction.” In The syntax of pronominal clitics [Syntax and Semantics Vol. 19], Hagit Borer (ed), 15–42. New York/Orlando: Academic Press. JeVers, Robert J. and Arnold M. Zwicky. 1980. “The evolution of clitics.” In Papers from the 4th International Conference on Historical Linguistics, Elisabeth Traugott, Rebecca Labrum and Susan Shepherd (eds), 221–231. Amsterdam: John Benjamins. Jespersen, Otto. 1912. Elementarbuch der Phonetik. Leipzig/Berlin: Teubner. Joppen, Sandra and Dieter Wunderlich. 1995. “Argument linking in Basque.” Lingua 97:123–169. Joseph, Brian D. 1990. “The beneWts of morphological classiWcation: on some apparently problematic clitics in Modern Greek.” In Contemporary morphology, W. Dressler et al. (eds), 221–231. The Hague: Mouton de Gruyter. Kaiser, Georg. 1992. Die klitischen Personalpronomina im Französischen und Portugiesischen. Frankfurt:Vervuert. Kaiser, Georg. 1994. “More about INFL-ection: the acquisition of personal pronouns in French.” In Bilingual Wrst language acquisition. French and German grammatical development, J.M. Meisel (ed), 131–159. Amsterdam/Philadelphia: John Benjamins. Kaisse, Ellen M. 1985. Connected speech. The interaction of syntax and phonology. New York/Orlando: Academic Press. Kato, Mary Aizawa. 1993. “The distribution of pronouns and null elements in object position in Brazilian Portuguese.” In Linguistic perspectives on Romance languages, W. Ashby, M. Mithun, G. Perissinotto and E. Raposo (eds), 225–235. Amsterdam/Philadelphia: John Benjamins. Kaufmann, Ingrid. 2000. Medium Voice. Arbeiten des Sonderforschungsbereichs 282: Theorie des Lexikons, No 117. Düsseldorf: Heinrich-Heine-University. Kayne, Richard S. 1975. French syntax: the transformational cycle. Cambridge, MA: MIT Press. Kayne, Richard S. 1989. “Facets of Romance past participle agreement.” In Dialect variation and the theory of grammar, Paola Benincà (ed), 85–103. Dordrecht: Foris. Kayne, Richard S. 1991. “Romance clitics, verb movement, and PRO.” Linguistic Inquiry, 22(4):647–686. Kayne, Richard S. 1994. The antisymmetry of syntax. Cambridge, MA: MIT Press. Keenan, Edward. 1987. “A semantic deWnition of ‘IndeWnite NP’.” In The representation of (in)deWniteness, Eric Reuland and Alice ter Meulen (eds), 286–317. Cambridge, MA: MIT Press.
References 247
Keenan, Edward. 1995. “The semantics of determiners.” In Handbook of contemporary semantic theory, Shalom Lappin (ed), 42–64. Oxford: Blackwell. Kemmer, Suzanne. 1993. The middle voice. Amsterdam: John Benjamins. Kenstowicz, Michael. 1993. Phonology in generative grammar. Cambridge: Blackwell. Kiparsky, Paul. 1989. Agreement and linking theory. Ms. Stanford Universitiy. Kiparsky, Paul. 1992. Structural case. Ms. Center for Advanced Studies. Berlin. Klaiman, M.H. 1988. “AVectedness and control: a typology of voice systems.” In Passive and voice [Typological Studies in Language 16], Masayoshi Shibatani (ed), 25–83. Amsterdam/Philadelphia: John Benjamins. Klavans, Judith L. 1980. Some problems in a theory of clitics. Bloomington: Indiana University Linguistics Club. Klavans, Judith L. 1985. “The independence of syntax and phonology in cliticization.” Language, 61(1):95–120. Klein, Hans-Wilhelm and Hartmut Kleineidam. 1979. Französische Grundgrammatik für Schule und Weiterbildung. Stuttgart: Klett. Kleinhenz, Ursula. 1998. On words and phrases in phonology: a comparative study with focus on German, Phd diss. Universität Tübingen. Koopmann H. and D. Sportiche. 1991. “The position of subjects.” Lingua 85:211–258. Krenn, Herwig and M. Ardélia Soares de Carvalho Mendes. 1987. Modernes Portuguesisch. Grammatik und Lehrbuch. Tübingen: Niemeyer. Kühner, Raphael. 1834/1966. Grammatik der griechischen Sprache I. Wissenschaftliche Buchgesellschaft: Darmstadt. Laenzlinger, Christopher. 1993. “A syntactic view of Romance pronominal sequences.” Probus 5:241–270. Lakämper, Renate. 2000. Plural- und Objektmarkierung in Quechua. Phd diss. Düsseldorf: Heinrich-Heine-University. Lakämper, Renate and Dieter Wunderlich. 1998. “Person marking in Quechua — A constraint-based minimalist analysis.” Lingua 105:113–148. Lamontagne, Greg and Keren Rice. 1995. “A correspondence account of coalescence.” In Papers in Optimality Theory. [University of Massachusetts Occasional Papers 18], Jill N. Beckmann, Laura Walsh Dickey and Suzanne Urbanczyk (eds), 211–223. Cambridge, MA: MIT Press. Leeuw, Frank van der. 1995. “Cliticization, stress and phonological words in European Portuguese: An Optimal(ity) approach.” Probus 7:31–68. Legendre, Geraldine. 1996. Clitics, verb (non-) movement, and Optimality in Bulgarian. Technical Report JHU-CogSci–96–5. Baltimore: Johns Hopkins University. Legendre, Géraldine. 1998. “Second position clitics in a verb-second language: conXict resolution in Macedonian.” In Proceedings of ESCOL 1997, S. Avrutin and D. Jonas (eds), 139–149. Cornell University: GLC Publications. Legendre, Géraldine. 1999. “Morphological and prosodic alignment at work: the case of South-Slavic clitics.” In Proceedings of WCCFL XVII, S.J. Blake, E.-S. Kim and K.N. Shahin (eds), 436–450. Stanford: CSLI Publications. Legendre, Géraldine. 2000a. “Morphological and prosodic alignment of Bulgarian clitics.” In Optimality Theory: syntax, phonology, and acquisition, Joost Dekkers, Frank van
248 References
der Leeuw, and Jeroen van de Weijer (eds), 423–462. Oxford: OUP. Legendre, Géraldine. 2000b. “Positioning Romanian verbal clitics at PF: an Optimaliytheoretic analysis.” In Clitics in phonology, morphology and syntax, Birgit Gerlach and Janet Grijzenhout (eds), 219–254. Amsterdam/Philadelphia: John Benjamins. Legendre, Géraldine. 2000c. “Optimal Romanian clitics: a crosslinguistic perspective.” In Comparative studies in Romanian syntax, V. Motapanyane (ed). Amsterdam: Elsevier. Lubowicz, Anna. 1997. Derived environment eVects in Optimality Theory. Ms., Amherst: University of Massachusetts. Maaßen, Christine. 1994. “Spanische klitische Pronomina: ‚Es war einmal ein Wackernagelpronomen ...’.” In Was determiniert Wortstellungsvariation?, Brigitte Haftka (ed), 345–361 Westdeutscher Verlag: Opladen. McCarthy John J. and Alan S. Prince. 1993. Prosodic Morphology I: constraint interaction and satisfaction. To appear, MIT Press. Technical Report #3. Rutgers University Center for Cognitive Science. McCarthy, John J. and Alan Prince. 1994. “Generalized alignment.” In Yearbook of Morphology 1993, Geert Booij and Jaap van Marle (eds), 79–153. Dordrecht: Kluwer. McCarthy, John J. and Alan Prince. 1995. “Faithfulness and reduplicative identity.” In Papers in Optimality Theory [University of Massachusetts Occasional Papers 18], Jill N. Beckman, Laura Walsh Dickey and Suzanne Urbanczyk (eds), 249–384. Cambridge, MA: MIT Press. Manzini, Rita. 1998. “Syntactic approaches to cliticization.” Glot International 3(3):3–7. Marantz, Alec. 1988. “Clitics, morphological merger and the mapping to phonological structure.” In Theoretical morphology: approaches in modern linguistics, Michael Hammond and Michael Noonan (eds), 253–270. London: Academic Press. Martins, Ana Maria. 1993. “Clitic placement from Old to Modern Portuguese.” In Historical Linguistics [Current Issues in Linguistic Theory 124], Henning Andersen (ed), 295–307. Amsterdam: John Benjamins. Mateus, Maria Helena and Ernesto d’Andrade. 2000. “The phonology of Portuguese.” Oxford: OUP. Miller, Philip. 1992. Clitics and constituent in Phrase Structure Grammar. New York: Garland. Miller, Philip and Ivan Sag. 1997. “French clitic movement without clitics or movement.” NLLT 15(3):573–639. Monachesi, Paola. 1994. “Towards a typology of Italian clitics.” CLS 30:266–281. Monachesi, Paola. 1995. A grammar of Italian clitics. Phd diss. Tilburg University. Monachesi, Paola. 1996. “On the representation of Italian clitics.” In Interfaces in phonology [studia grammatica 41] Ursula Kleinhenz (ed), 83–101. Berlin: Akademie Verlag. Monachesi, Paola. 1999. “The syntactic structure of Romanian auxiliary (and modal) verbs.” In Constraints and resources in natural language syntax and semantics, G. Bouma, G. KruiV and R. Oehrle (eds), 99–115. Stanford: CSLI Publications. Monachesi, Paola. 2000. “Linearization properties of Romanian clitics.” In Clitics in phonology, morphology and syntax, Birgit Gerlach and Janet Grijzenhout (eds), 255– 294. Amsterdam/Philadelphia: John Benjamins.
References 249
Morin, Yves-Charles. 1979. “More remarks on French clitic order.” Linguistic Analysis 5:293–312. Morin, Yves-Charles. 1981. “Some myths about pronominal clitics in French.” Linguistic Analysis 8:95–109. Muller, Claude. 1998. “On the position of Romance clitics.” Hermes 20:115–138. Müller, Gereon. 2000. Das Pronominaladverb als Reparaturphänomen. In Linguistische Berichte 182. Forschung, Information, Diskussion:39–178. Müller, Natascha and Beate Riemer. 1998. Generative Syntax der romanischen Sprachen. Tübingen: StauVenburg. Nespor, Marina. 1990. “Vowel deletion in Italian: the organization of the phonological component.” The Linguistic Review 7:375–398. Nespor, Marina. 1986. “The phonological word in Greek and Italian.” In Sandhi phenomena in the languages of Europe, Henning Andersen (ed) 65–74. Berlin/New York/ Amsterdam: Mouton de Gruyter. Nespor, Marina and Irene Vogel. 1982. “Prosodic domains of external sandhi rules.” In The structure of phonological representations 1, Harry van der Hulst and Norval Smith (eds), 222–255. Dordrecht: Foris. Nespor, Marina and Irene Vogel. 1986. Prosodic Phonology. Dordrecht: Foris. Noyer, Rolf. 1994. “Mobile aYxes in Huave: optimality and morphological wellformedness.” In Proceedings of the Twelfth West Coast Conference on Formal Linguistics, Erin Duncan, Donka Farkas and Philip Saelti (eds), 67–82. Santa Cruz: University of California. Nübling, Damaris. 1992. Klitika im Deutschen. Schriftsprache, Umgangssprache, alemannische Dialekte. Tübingen: Gunter Narr. Ortmann, Albert. 2002. Kategorien des Nomens: Schnittstellen und Ökonomie. Tübingen: Niemeyer. Ortmann, Albert and Alexandra Popescu. 2000. “Romanian deWnite articles are not clitics.” In Clitics in phonology, morphology and syntax, Birgit Gerlach and Janet Grijzenhout (eds), 295–324. Amsterdam/Philadelphia: John Benjamins. Palma Caetano, A. José, Johannes J. Mayr, Renate Plachy, Franz Ptacek. 1986. Grammatik Potugiesisch. Munich: Hueber. Parodi, Teresa. 1998a. “Aspects of clitic doubling and clitic clusters in Spanish.” In Models of inXection, Ray Fabri, Albert Ortmann and Teresa Parodi (eds), 85–102. Tübingen: Niemeyer. Parodi, Teresa. 1998b. Clitic doubling en español como lengua nativa y como segunda lengua. Ms., Cambridge: Cambridge University. Peperkamp, Sharon. 1995. “Enclitic stress in Romance.” CLS 31, Clitics Parasession Volume:234–249. Peperkamp, Sharon. 1997. Prosodic Words. Phd diss. University of Amsterdam. Perlmutter, David M. 1970. “Surface structure constraints in syntax.” Linguistic Inquiry 1:187–255. Philippaki-Warburton Irene. 1987. “The theory of empty categories and the pro-drop parameter in Modern Greek.” Journal of Linguistics 23:289–318. Plag, Ingo. 1998. “Morphological haplology in a constraint-based morpho-phonology.” In
250 References
Phonology and morphology of the Germanic languages, Wolfgang Kehrein and Richard Wiese (eds), 199–215. Tübingen: Niemeyer. Poletto, Cecilia. 1999. “The internal structure of AgrS and subject clitics.” In Clitics in the languages of Europe, Henk van Riemsdijk (ed), 581–637. Berlin/New York: Mouton de Gruyter. Pollard, Carl and Ivan Sag. 1994. Head-Driven Phrase Structure Grammar. Ms. Stanford, CA: CSLI and University of Chicago Press. Pollock, Jean Yves. 1989. “Verb movement, universal grammar and the structure of IP.” Linguistic Inquiry 20:365–424. Popescu, Alexandra. 1997. Objektklitika und Argumentlinking im Rumänischen. Masters thesis. Düsseldorf: Heinrich-Heine-University. Popescu, Alexandra. 2000. “The morphophonology of the Romanian clitic sequence.” Lingua 110:773–799. Prince, Alan S. and Paul Smolensky. 1993. Optimality Theory: constraint interaction in generative grammar. To appear, MIT Press. TR–2, Rutgers University Cognitive Science Center. Rivera-Castillo, Yolanda. 1992. “Enclitic pronouns in Caribbean Spanish.” BLS 18:425– 434. Rivera-Castillo, Yolanda. 1997. “Proclitic groups and word order in Carribean Spanish.” In Clitics, Pronouns and Movement, James R. Black and Virginia Motapanyane (eds), 237–271. Amsterdam/Philadelphia: John Benjamins. Rivero, Maria-Luisa. 1986. “Parameters in the typology of clitics in Romance and Old Spanish.” Language 62:774–807. Rizzi, Luigi. 1978. “A restructuring rule in Italian syntax.” In Recent transformational studies in European languages, J. Keyser (ed), 113–158. Cambridge, MA: MIT Press. Rizzi, Luigi. 1982. Issues in Italian syntax. Dordrecht: Foris. Rizzi, Luigi. 1986. “On the status of subject clitics in Romance.” In Studies in Romance linguistics, Osvaldo Jaeggli and Carmen Silva-Corvalán (eds), 391–419. Dordrecht: Foris. Roberge, Alain. 1997. “Subject doubling, free inversion and null argument languages.” The Canadian Journal of Linguistics 31(1):55–79. Samek-Lodovici, Vieri. 1996. Constraints on subjects. An Optimality Theoretic analysis. Phd diss. Rutgers University. Schwarze, Christoph. 1995. Grammatik der italienischen Sprache. 2. verbesserte AuXage. Tübingen: Niemeyer. Selkirk, Elisabeth O. 1978. “On prosodic structure and its relation to syntactic structure.” Nordic Prosody 2:111–120. Selkirk, Elisabeth O. 1981. “On the nature of phonological representation.” In The cognitive representation of speech, J. Anderson, J. Laver and T. Meyers (eds). Amsterdam: North Holland. Selkirk, Elisabeth O. 1984. Phonology and syntax: the relation between sound and structure. Cambridge, MA: MIT Press. Selkirk, Elisabeth O. 1995. “The prosodic structure of function words.” In Papers in Optimality Theory [University of Massachusetts Occasional Papers 18] Jill N. Beck-
References 251
mann, Laura Walsh Dickey and Suzanne Urbanczyk (eds), 439–469. Cambridge MA: MIT Press. Siewierska, Anna. 1988. Word order rules. London: Croom Helm. Simpson Jane and M. Withgott. 1986. “Pronominal clitic clusters and templates.” In The Syntax of Pronominal Clitics [Syntax and Semantics 19], Hagit Borer (ed), 149–174. New York/Orlando: Academic Press. Smolensky, Paul. 1995. On the internal structure of the constraint component Con of UG. Handout of talk given at UCLA, April 7. Available as ROA–86. Spencer, Andrew. 1991. Morphological theory. An introduction to word structure in Generative Grammar. Oxford: Blackwell. Spencer, Andrew. 1998. Bulgarian clitics and Optimality Theory. Essex research reports in linguistics. University of Essex. Spencer, Andrew. 2000. “Verbal clitics in Bulgarian: a Paradigm Function approach.” In Clitics in phonology, morphology and syntax, Birgit Gerlach and Janet Grijzenhout (eds), 355–386. Amsterdam/Philadelphia: John Benjamins. Sportiche, Dominique. 1996. “Clitic constructions.” In Phrase structure and the lexicon, Johan Rooryck and Laurie Zaring (eds), 213–276. Dordrecht: Kluwer. Steele, Susan. 1977. “Clisis and diachrony.” In Mechanisms of syntactic change. Li, Charles N. (ed), 539–579. Austin: University of Texas. Stiebels, Barbara and Dieter Wunderlich. 1994. “Morphology feeds syntax. The case of particle verbs.” Linguistics 32:913–968. Stiebels, Barbara. 2000a. “Linker inventories, linking splits and lexical economy.” In Lexicon in focus, Barbara Stiebels and Dieter Wunderlich (eds), 211–245. Berlin: Akademie Verlag. Stiebels, Barbara. 2000b. Typologie des Argumentlinkings: Ökonomie und Expressivität. Habilitationsschrift. Düsseldorf: Heinrich-Heine-University. Stump, Gregory T. 1980. “An inXectional approach to French clitics.” Ohio State University Working Papers in Linguistics 24:1–54. Stump, Gregory T. 1993. “Position classes and morphological theory.” In Yearbook of Morphology 1992, Geert Booij and Jaap van Marle (eds), 129–179. Dordrecht: Kluwer. Stump, Gregory T. 1997. “Template morphology and inXectional morphology.” In Yearbook of Morphology 1996, Geert Booij and Jaap van Marle (eds), 217–241. Dordrecht: Kluwer. Suñer, Margarita. 1988. “The role of agreement in clitic doubled constructions.” NLLT 6:391–434. Suñer, Margarita. 1991. “Two properties of clitic doubled constructions.” In Logical structure and linguistic structure: crosslinguistic perspectives, James Huang and Robert May (eds), 233–251. Dordrecht: Kluwer. Suñer, Margarita. 1992. “Subject clitics in the northern Italian vernaculars and the matching hypothesis.” NLLT 10:641–672. Sylla, Yero:1979. Grammatical relations and Fula Syntax. Phd diss. Los Angeles: University of California. Tomic´, Olga Miseška. 2000. “Operator clitics.” In Clitics in Phonology, Morphology and
252 References
Syntax, Birgit Gerlach and Janet Grijzenhout (eds), 387–404. Amsterdam/ Philadelphia: John Benjamins. Torrego, Esther. 1996. “On the nature of clitic doubling.” In Evolution and revolution in linguistic theory, Hector Campos (ed), 399–418. Washington: Georgetown University Press. Urbas, Martina. 1993. Numeruskongruenz und NumeruskonXikte in Kopulasätzen. In Arbeiten des Sonderforschungsbereiches 282 ‘Theorie des Lexikons’ No. 38. Düsseldorf: Heinrich-Heine-University. Uriagereka, Juan. 1995. “Aspects of the syntax of clitic placement in western Romance.” Linguistic Inquiry, 26(1):79–123. Uriagereka, Juan. 2000. “On doubling.” In Clitics in phonology, morphology and syntax, Birgit Gerlach and Janet Grijzenhout (eds), 405–431. Amsterdam/Philadelphia: John Benjamins. Vigário, Marina. 1999. “On the prosodic status of stressless function words in European Portuguese.” In Studies in the phonological word, [Current Studies in Linguistic Theory 174], T.A. Hall and Ursula Kleinhenz (eds), 255–294. Amsterdam: John Benjamins. Wackernagel, Jacob. 1892. “Über ein Gesetz der indogermanischen Wortstellung.” Indogermanische Forschungen 1:333–436. Wanner, Dieter. 1974. “The evolution of Romance clitic order.” In Linguistic studies in Romance languages: Proceedings of the Third Linguistic Symposium on Romance Languages, R.J. Campbell et. al. (eds), 158–177. Washington: Georgetown University Press. Wanner, Dieter. 1977. “On the order of clitics in Italian.” Lingua 43:101–128. Wanner, Dieter. 1987. The development of Romance clitic pronouns. From Latin to Old Romance. Berlin/New York/Amsterdam: Mouton de Gruyter. Wanner, Dieter. 1993. “Multiple clitic linearization principles.” In Linguistic perspectives on Romance languages, W. Ashby, M. Mithun, G. Perissinotto and E. Raposo (eds), 281–302. Amsterdam/Philadelphia: John Benjamins. Wanner, Dieter. 1996. “Second position clitics in Medieval Romance.” In Approaching Second: second position clitics and related phenomena, Aaron L. Halpern and Arnold M. Zwicky (eds), 537–579. Stanford: CSLI. Warburton, Irene P. 1977. “Modern Greek clitic pronouns and the ‘surface structure constraints’ hypothesis.” Journal of Linguistics 13:259–281. Weinrich, Harald. 1982. Textgrammatik der französischen Sprache. Stuttgart: Klett. Wheeler, Max W. 1986. “Catalan sandhi phenomena.” In Sandhi phenomena in the languages of Europe, Henning Andersen (ed) 475–488. Berlin/New York/Amsterdam: Mouton de Gruyter. Wunderlich, Dieter. 1996a. “Minimalist Morphology: the role of paradigms.” In Yearbook of Morphology 1995, Geert Booij and Jaap van Marle (eds), 17–41. Dordrecht: Kluwer. Wunderlich, Dieter. 1996b. “A minimalist model of inXectional morphology.” In The Role of Economoy Principles in Linguistic Theory, Chris Wilder, Manfred Bierwisch and Hans-Martin Gärtner (eds), 267–298. Berlin: Akademie Verlag. Wunderlich, Dieter. 1996c. “Lexical categories.” Theoretical Linguistics 22:1–48.
References 253
Wunderlich, Dieter. 1997. “Cause and the structure of verbs.” Linguistic Inquiry 28:27–68. Wunderlich, Dieter. 2001a. “How gaps and substitutions can become optimal: the pronominal aYx paradigms of Yimas.” In Transactions of the Philological Society 99, Special number on morphological paradigms, James P. Blevins (ed). 315–366. Wunderlich, Dieter. 2001b. “A correspondence-theoretic analysis of Dalabon transitive paradigms.” In Yearbook of Morphology 2000, Geert Booij and Jaap van Marle (eds), 233–252. Kluwer Academic Publishers. Wunderlich, Dieter and Ray Fabri. 1996. “Minimalist Morphology: an approach to inXection.” Zeitschrift für Sprachwissenschaft 14:236–294. Zec, Draga. 1988. Sonority Constraints on Prosodic Structure. Phd diss. Stanford University. Zec, Draga and Sharon Inkelas. 1991. “The place of clitics in the Prosodic Hierarchy.” In Proceedings of the 10th West Coast Conference on Formal Linguistics, D. Bates (ed), 505–519. Stanford: SLA. Zwicky, Arnold M. 1977. “On clitics.” Indiana University Linguistics Club. Again in Phonologica 1977. Akten der dritten internationalen Phonologie-Tagung, Wien, 1.–4. Sept, 1976, Wolfgang U. Dressler and Oscar E. PfeiVer (ed), 29–39. Innsbruck: Institut für Sprachwissenschaft der Universtät Innsbruck. Zwicky, Arnold M. 1985. “Clitics and particles.” Language, 61(2):283–305. Zwicky, Arnold M. and GeoVrey K. Pullum. 1983. “Cliticization vs. inXection: English n’t.” Language, 59(3):502–512.
Appendix
i. Paradigms i.i Portuguese 1. Object clitics
1 2 3m 3f
sg pl do io rfl do io rfl me te nos vos o os lhe se lhes se a as
2. Lexical entries: +D; [-max,-min][ ]/+V a. me +1,+hr b. nos +1,+pl,+hr d. vos +2,+pl,+hr e. o +hr g. a +f,+hr h. as +f,+pl,+hr j. lhes +pl,+hr,+lr k. se +rX 3. Clitic sequences io do 1sg 2sg 3sg.m 3sg.f 1pl 2pl 3pl.m 3pl.f
me te o a nos vos os as
1sg me
2sg 3sg 1pl te lhe nos
mo ma
to ta
lho no-lo lha no-la
2pl vos
3pl lhes
vo-lo lho vo-la lha
mos tos lhos no-los vo-los lhos mas tas lhas no-las vo-las lhas
c. te +2,+hr f. os +pl,+hr i. lhe +hr,+lr
256 Appendix
4. Personal pronouns
1 2 3m 3f
sg s do/io pp eu mim tu ti de/para ele ele ela ela
pl s do/io pp nós nós (vós) vocês de/para eles eles elas elas
5. The forms of haver and the inXectional endings of future and conditional sg 1 2 3 haver: present tense hei hás há [+fut] -ei -ás -á haver: past tense havia havias havia [+cnd] -ia -ias -ia
pl 1 havemos -emos havíamos -íamos
2 haveis -eis havíeis -íeis
i.ii European Spanish 1. Object clitics sg pl do io rfl do io rfl 1 me nos 2 te os 3m lo los le se les se 3f la las 2. Lexical entries: +D; [-max,-min][ ]/+V a. d. g. j.
me os la les
+1,+hr +2,+pl,+hr +f,+hr +pl,+hr,+lr
b. e. h. k.
nos lo las se
+1,+pl,+hr +hr +pl,+f,+hr +rX
c. te +2,+hr f. los +pl,+hr i. le +hr,+lr
3 hão -ão haviam -iam
Paradigms 257
3. Object clitic sequences io do 1sg 2sg 3sgm 3sgf 1pl 2pl 3plm 3pl.f
me te lo la nos os los las
1sg me
2sg te
3sg le
1pl nos
me lo me la
te lo te la
se lo nos lo se la nos la
2pl os
os lo se lo os la se la
me los te los se los nos los os los se los me las te las se las nos las os las se las
4. Personal pronouns pl s do/io pp 1m nosostros nosostros yo a mi / conmigo 1f nosostras nosostras 2m vosostros a / con vosostros tu a ti / contigo 2f vosostras vosostras 3m él él ellos ellos a / con 3f ella ella ellas ellas s
sg do/io pp
i.iii Argentinian Spanish 1. Object clitics
1 2 3
3pl les
sg pl do io rfl do io me nos te m los m lo le se les f las f la
rfl
se
258 Appendix
2. Lexical entries: +D; [-max,-min] [ ]\+V a. me d. lo g. las j. se
+1,+hr +hr +f,+pl,+hr +rX
b. nos +1,+pl,+hr e. los +pl,+hr h. le +hr,+lr
c. te f. la i. les
+2,-pl,+hr +f,+hr +pl,+hr,+lr
i.iv Standard Italian 1. Object clitics sg do io rfl do 1 mi 2 ti 3m lo gli si li 3f la le le
pl io ci vi gli
rfl
si
Additional clitics: ci, vi : ‘locative’
ne: ‘partitive’
2. Lexical entries: +D, [-max,-min][ ]/ +V a. mi +1,+hr b. ci +1,+pl,+hr d. vi +2,+pl,+hr e. lo +hr g. la +f,+hr h. le +f,+pl,+hr j. gli2 +pl,+hr,+lr k. le2 +f,+hr,+lr m. ne +hr, P: di n. vi1 +2,+pl,+hr p. vi2 +hr,+loc q. ci2 +hr,+loc
c. f. i. l. o. r.
ti li gli1 si ci1 ci3
+2,+hr +pl,+hr +hr,+lr +rX +1,+pl,+hr +hr, P: a
3. Object clitic sequences io do 1sg 2sg 3sg.m 3sg.f 1pl 2pl 3pl.m 3pl.f
mi ti lo la ci vi li le
1sg mi
2sg 3sgm 3sgf 1pl 2pl ti gli le ci vi
3pl gli
me lo te lo me la te la
glielo gliela
ce lo ve lo glielo ce la ve la gliela
me li me le
glieli gliele
ce li ve li glieli ce le ve le gliele
te li te le
Paradigms 259
4. Personal pronouns
1 2 3m 3f
sg s do io pp io me a me tu te a te lui a lui lei a lei
pl s do noi voi
io pp a noi a voi
loro
a loro
i.v Piattino 1. Subject and object Clitics
1 2 3m 3f
sg pl s do io rfl s do io a me [a]n me te ve [a]l i ie si ie la li
Additional clitics: ci, vi: ‘locative’
rfl
si
ne: ‘partitive’
2. Lexical entries: +D;[-max,-min][]/+V a. a +1 b. [a]n +1,+pl d. te +2 e. ve +2,+pl,+hr g. i +pl h. la +f j. ie +hr,+lr k. si +rX m. vi1 +2,+pl,+hr n. vi2: +hr,+loc p. ci2: +hr,+loc q. ci3 +hr; P: a
c. f. i. l. o.
me [a]l li ne ci1
+1,+hr [person ] +f,+pl +hr; P: di +1,+pl,+hr
260 Appendix
3. Object clitic sequences io do 1sg 2sg 3sg.m 3sg.f 1pl 2pl 3pl.m 3pl.f
me te l’ la me ve i li
1sg me
2sg 3sg 1pl te ie me
2pl 3pl ve ie
mel’ tel iel mel’ vel iel mela tela iela mela vela iela
mei meli
tei iei mei teli ieli meli
vei iei veli ieli
i.vi Standard French 1. Subject and object clitics
1 2 3m 3f
sg pl s do io rfl s do io rfl je me nous tu te vous il le ils lui se les leur se elles elle la
Additional clitics: y: ‘locative’ en: ‘partitive’ on: impersonal subject ne: negation
2. Lexical entries: +D;[-max,-min][ ]/+V a. je +1 b. nous +1,+pl d. tu +2 e. vous +2,+pl g. il [person ] h. ils +pl j. elles +f,+pl k. le +hr m. la +f,+hr n. lui +hr,+lr p. se +rX q. en +hr,-human; P: di s. y2 +hr ; P: a t. on +generic
c. f. i. l. o. r. u.
me te elle les leur y1 ne
+1,+hr +2,+hr +f +pl,+hr +pl,+hr,+lr +hr,+loc +neg
Paradigms 261
3. Object clitic sequences io do 1sg 2sg 3sgm 3sgf 1pl 2pl 3pl
1sg me
2sg te
3sg lui
1pl nous
2pl vous
3pl leur
me te le me le te le le lui nous le vous le le leur la me la te la la lui nous la vous la la leur nous vous les me les te les les lui nous les vous les les leur
4. Personal pronouns sg 1 2 3m 3f
s do moi toi lui elle
pl io pp à moi à toi à lui à elle
s do nous vous eux elles
io pp à nous à vous à eux à elles
i.vii Colloquial French 1. Subject and object clitics
1 2 3m 3f
s je tu il
sg pl do io rfl s do io rfl me on nous te vous le lui se ils les leur se la
2. Lexical entries: +D;[-max,-min][ ]/ +V a. d. f. i.
je nous te le
+1 +1,+hr,+pl +2,+hr +hr
b. e. g. j.
on tu il les
+1,+pl +2 [person ] +pl,+hr
c. f. h. k.
me vous ils la
+1,+hr +2,+pl +pl +f,+hr
262 Appendix
l. lui p. en s. on:
+hr,+lr +hr ; P: di +generic
m. leur q. y1
+pl,+hr,+lr +hr,+loc
o. se r.y2 :
+rX +hr; P: a
b. e. h. k.
+2,+hr +hr,+lr +2,+pl,+hr +f,+pl,+hr
i.viii Romanian 1. Object clitics
1/1rfl 2/2rfl 3m 3f 3rfl
sg pl do io do io ma[ îmi ne te îti va[ îl îi îi le o se îs6 i se îs6 i
2. Lexical entries: +D;[-max,-min][ ]/ +V a. c. f. i. l.
ma[ ît6 i o îi se
+1,+hr +2,+hr,+lr +f,+hr +pl,+hr +hr,+rX
b. d. g. j. m.
îmi îl ne le îs6 i
+1,+hr,+lr +hr +1,+pl,+hr +pl,+hr,+lr +hr,+lr,+rX
3.1. Object clitic sequences io do 1sg 2sg 3sg.m 3sg.f 1pl 2pl 3pl.m 3pl.f
ma[ te îl o ne va[ îi le
1sg îmi
2sg ît6 i
3sg îi
1pl ne
2pl va[
3pl le
mi te mi-l t6 i-l mi-o t6 i-o
i te i-l i-o
ni te ni-l vi-l neÛ-o vi-o
li te li-l le-o
mi-i t6 i-i mi le t6 i le
i-i i le
ni-i ni le
li-i li le
vi-i vi le
te îi va[ le
Paradigms 263
3.2. The phonological representation of io-DO sequences io do
1sg 2sg 3sg.m 3sg.f 1pl 2pl 3pl.m 3pl.f
ma[ te îl o ne va[ îi le
1sg îmi [Gmj],[mi], [mj],[mj]
[m6], [m] [te], [teÛ] [Gl], [l] [o] [ne], [neÛ] [v6], [v] [Gj], [ij], [j] [le], [leÛ]
2sg 3sg 1pl 2pl 3pl îtt6 ii îi ne va[ le [Gtsj], [Gj] [ne], [v6],[v], [le] [tsi] [neÛ] [vj],[vi] [tsj],[tsj] [ni],[nj]
mi.te. mil. mjo.
tsil. tsjo.
i.te. ni.te. il. nil. jo neÛo.
vil. vjo.
li.te. lil. leÛo.
mij mi.le.
tsij. tsi.le.
ij. nij. i.le. ni.le.
vij. vi.le.
lij. li.le.
4. Auxiliary Clitics pfv cnd 1 am asi 2 ai 3 a ar 1pl am 2pl ati 3pl au ar
fut oi o om oti or
5. Lexical entries: [-min,-max],[asp, tmp, per, num]/ V a. d. g. j. m.
am ai o ati or
+1,+pfv +2 +fut +2,+pl +pl,+fut
b. e. h. k.
asi a am oti
+1,+cond +pfv +1,+pl +2,+pl,+fut
c. f. i. l.
oi ar o au
(+1 v +2), +fut +cnd +1,+pl,+fut +pl,+pfv
264 Appendix
6. AUX — DO sequences aux do 1sg 2sg 3m.sg 3f.sg 1pl 2pl 3m.pl 3f.pl
ma[ te îl o ne va[ îi le
1sg 2sg am ai mam. maj. teÛam. teÛaj. lam. laj.
3sg a ma. teÛa. la.
1pl am mam. teÛam. lam.
2pl ît6 i matsj. teÛatsj. latsj.
3pl au mauÛ teÛauÛ . lauÛ .
neÛam. vam. jam. leÛam.
neÛaj. vaj. jaj. leÛaj.
neÛa. va. ja. leÛa.
neÛam. vam. jam. leÛam.
neÛatsj. vatsj. jatsj. leÛatsj.
neÛauÛ. vauÛ jauÛ leÛauÛ
2sg ai mjaj. tsjaj. jaj. neÛaj. vaj. leÛaj.
3sg a mja. tsja. ja. neÛa. va. leÛa.
1pl am mjam. tsjam. jam. neÛam. vam. leÛam.
2pl ît6 i mjatsj. ?tsjatsj. jatsj. neÛatsj. vatsj. leÛatsj.
7. AUX — IO sequences aux io 1sg 2sg 3sg 1pl 2pl 3pl
îmi ît6 i îi ne va[ le
1sg am mjam. tsjam. jam. neÛam. vam. leÛam.
8. Personal pronouns sg s do io 1 mine pe mine mie 2 tine pe tine t ie 3m el pe el lui 3f ea pe ea ei
pl s do noi pe noi voi pe voi ei pe ei
io noua[ voua[ lor
3pl au mjauÛ tsjauÛ jauÛ neÛauÛ. vauÛ leÛauÛ
Clitic sequence internal phenomena 265
II. Clitic sequence internal phenomena ii.i Clitic order 1. Templates French: Spanish: Italian : Romanian:
s neg 1/2/se 3do 3io y en Perlmutter (1971:57) se 2 1 3 Perlmutter (1971:45) 1/2/3io ciloc sirX 3do siimp ne Monachesi (1995:122) io < do < aux exception: io < aux V-3f.do
2. Order and combinatory restrictions of object clitic sequences general tendency a. io < do exception b. 1/2 < 3 exception c. *1/2 1/2 exception
French: Romanian: Romanian:
3do < 3io 3io < 2do 1io < 2sg.do, 3io < 2sg.do (Spanish: 2 < 1)
3. Combinations of clitics in the internal clitic sequence Italian io < loc loc < 3do 1/2 < loc
French o < loc
Spanish —
b. object + partitive clitic
o < part
o > part
—
c. object + reXexive clitic
rfl < 3do rfl < o io < rfl 1/2 < rfl *1/2 rfl 3do < rflimp
a. object + locative clitic
rfl < o
d. locative + reXexive clitic loc < rfl
rfl < loc
—
e. partitive + reXexiveclitic rfl < part
rfl < part
—
f. locative + partitive clitic loc < part
loc < part
—
ii.ii Opaque clitics 1. Deletion and substitution a. Spanish: clitic deletion *se se → se rfl rfl rfl
b. Italian: clitic substitution *si si → ci si rfl rfl 1pl/loc rfl
266 Appendix
c. Spanish: clitic substitution d. Italian: clitic substitution *le lo → se lo *le lo → glie-lo 3io 3do rfl-3do 3f.io 3do 3io-3do 2. Spanish: morphological feature transfer a. b. c. d. e.
*le las *les la *les las *les lo *le los
→ se las
3. Clitic sequence internal phonological processes a. Italian vowel lowering mi → me ti → te / _ {lo,ne} ci → ce vi → ve gli → glie
b. Romanian vowel raising ne → ni va[ → vi / _ {te, îl, îi, le} le → li c. Portuguese coda neutralisation {nos, vos} → {nol, vol} /_{o, a, os, as} 4. Clitic sequence internal prosodic processes a. Portuguese vowel deletion me → m te → t / _ {o, a, os, as} lhe → lh lhes + {o, a, os, as}→ {lhos, lhas} b. Romanian obligatory vowel reduction ne → [neÛ] le → [leÛ] /_ V îmi → [mj] îtsi → [tsj] îi → [j] va[ → [vj]
Clitic placement and clitic doubling 267
c. Romanian obligatory vowel deletion îl îi ma[ va[
[l] [j] /_ V [m] /V_ [v]
→ → → →
III. Clitic placement and clitic doubling 1. The position of clitics with respect to the verb Wnite
inWnitive participle, gerund imperative
Piattino preverbal postverbal Italian preverbal postverbal Spanish preverbal postverbal Romanian preverbal postverbal French preverbal postverbal
2. Conditions on clitic doubling in Romance subject +pron -pron Piattino LA Spanish E Spanish Romanian E Portuguese LA Portuguese Standard Italian Standard French
(+)1
+
(+)1
(+)
indirect object direct object +pron -pron +pron -pron +spec -spec +spec -spec + + + + + + * *
+ + + + * * * *
+ + + * * * * *
+ + + + + + * *
(+)3 + * +/*2 * * * *
(‘+’ means obligatory; ‘(+)’ means optional; ‘*’ means not allowed ) 1depends on person; 2[-focus,-human]; 3+: old / -: new discourse referent (Data from Kaiser, 1992; Parodi 1998; Popescu 1997)
* * * * * * * *
268 Appendix
IV. Clitic speciWc constraints 1. Clitic sequence internal morphological alignment align-l(+1,cs)
1st person clitics are initial in a clitic sequence.
align-l(+2,cs)
2nd person clitics are initial in a clitic sequence.
align-l(+lr,cs)
Indirect object clitics are initial a clitic sequence.
align(+lr,r,+1,l)
The right edge of a clitic with the feature [+lr] is aligned with the left edge of a clitic with the feature [+1].
align(+lr,r,+2,l)
The right edge of a clitic with the feature [+lr] is aligned with the left edge of a clitic with the feature [+2].
2. Morphological alignment of clitic sequences and their host align(cs,l,V[Fα],r) The left egde of a clitic sequence is aligned with the right edge of verb with the feature speciWcation Fα. align(cs,r,V[Fβ],l) The right edge of a clitic sequence is aligned with the left edge of a verb with the feature speciWcation Fβ. Romanian:
Fα = gerund, imperative, participle Fβ = inWnitive, Wnite verb
French:
Fα = imperative Fβ = participle, gerund, inWnitive, Wnite verbs
Italian, Piattino, Spanish: Fα = gerund, imperative, participle, inWnitive Fβ = Wnite verb 3. Prosodic alignment constraints align-r(cs, ω)
The right edge of a clitic sequence is aligned with the right edge of a prosodic word.
4. Phonological constraints on clitic combinations *sonority adjacency
The sonority levels of adjacent elements in a clitic sequence are not adjacent on the sonority hierarchy.
*alliteration
*[[X1...]c1 [X1...]c2]CS Clitics in a clitic sequence do not have identical onsets.
*rhyme
Clitics in a clitic sequence do not have identical nuclei.
Clitic specific constraints 269
antihomophony (Golston,1995)
*[ ... χ χ ...]ω ↓↓ A B
χ: some stretch of sound, A,B: some stretch of meaning, ω : phonological word
5. Constraints on morphological explicitness and economy max(person)M
Person features from the input are encoded morphologically in the output.
max(+lr/+hr)
The feature [+lr] of an input element I has a correspondent in the output, if I also bears the feature [+hr].
*m-structure *(x)
Noun referents with low salience are not marked morphologically.
*(x<-human>)M
Non-human noun referents are not marked morphologically.
*(x<-speciWc>)M
Non-speciWc noun referents are not marked morphologically.
6. Constraints on argument roles max(arg):
An argument role has an overt correspondent in the output.
max(arg)M:
An argument role has a morphological correspondent (i.e. a clitic or an aYx) in the output.
*multiple correspondence
Each argument role in the input has only one correspondent in the output.
unif(mf)
Corresponding elements are compatible with respect to their morphosyntactic features.
unif(ref)
Corresponding elements are compatible with respect to their referential properties.
270 Appendix
7. Transitive Correspondence: For each O1, O2 ∈ Output and I ∈ Input: If O1 and O2 correspond to the same I, O1 and O2 correspond to each other. 8. Constraints on the realisation of DPs max(foc)
Arguments which are relevant for the information structure (i.e. bear the feature [+focus]) are realised by DPs.
drop topic
Arguments which are coreferent with the topic are not realised by DPs (Xwg. Grimshaw & Samek-Lodovici, 1995).
9. Syntactic object position domain-oVP
For each object role of a verb V (i.e. an argument role bearing the feature [+hr]) at least one correspondent (i.e. a direct object DP or a direct object clitic) must be realised within the VP.
Paradigms 271
Language index Balkan 10, 83, 117, 164, 185, 195-196, 200 Bantu 68, 194 Bulgarian 138-139, 184-185, 195-196, 234 Catalan 69 Croatian 22, 185, 195, 197 Dalabon 112 Dutch 23-24 English 22, 55, 62, 78-79, 93, 107-108, 114, 148 French 4-5, 10, 21, 29-31, 45-46, 48, 52, 55, 59-60, 66-68, 74, 76, 83-85, 112-117, 118, 124, 129-133, 136, 147-150, 156, 163, 170-173, 185, 190- 191, 193-194, 197-198, 214, 218, 220, 223-224, 226, 234-235, 238, 265, 267 Colloquial 112, 224 clitic inventory and lexical entries 115-117, 261-262 Standard 5, 7, 10, 112-116, 190, 217, 260-261, 267 clitic inventory and lexical entries 112-115, 260 clitic sequence paradigm 171, 261 German 13, 19, 23, 107-108, 114 Greek 134 Ancient 2-3 Italian 4-10, 12-13, 21-22, 29-31, 33-41, 45-48, 50-52, 59, 64-66, 68-70, 74, 76-79, 83-87, 102, 107, 110, 112-115, 117-118, 122, 124, 128-129, 131-134, 136-137, 142-143, 145-148, 173-180, 182-184, 187, 189-191, 194, 197-200, 206, 213216, 220-223, 235, 240, 265-267 Standard 5, 7, 10, 21, 23-24, 40, 49-50, 70, 130, 179, 217, 220-221, 258-259, 267 clitic inventory and lexical entries 102-109, 258 clitic sequence paradigm 143, 258
Neapolitan 21, 23-24, 40 Piattino 5, 48, 59, 74, 76, 83-85, 118, 123, 148, 194, 197-198, 206, 217, 220, 223-226, 267 clitic inventory and lexical entries 110-112, 259 clitic sequences 260 Pugliese 70 Latin 50 Vulgar 44, 56-58, 209 Macedonian 185, 195-197 Portuguese 9, 21, 29, 32-33, 35-36, 39, 4247, 52, 56-59, 63, 66-70, 74, 83-85, 8993, 95, 97-98, 100-104, 118, 123, 134135, 148, 164, 167-170, 184, 187, 198, 206, 209-213, 233-234, 238-239, 266 Brazilian 5, 42-44, 66, 68-69, 84, 216217, 267 clitic inventory and lexical entries 89-97, 255 clitic seqeunces 255 European 5, 9, 17-18, 26, 36, 42-44, 4849, 51, 66, 69, 84, 208, 213, 216-217, 237, 267 inventory and lexical entries 89-97, 255 clitic sequences 255 Modern vs. Old 58 Quechua 68 Romanian 5, 10, 21, 25, 29-30, 32, 38, 46, 48-49, 52-57, 59-60, 63, 66-70, 74, 83-85, 92, 122-123, 125, 128, 131135, 147-150, 156-159, 161, 163167, 171-172, 175, 182, 184-186, 191192, 195-198, 200-208, 214, 217218, 226-228, 230, 233-234, 238-239, 262-264, 265-267 clitic inventory and lexical entries 117-121, 262-263
272 Language index
clitic seqeunce paradigms 164-165, 262-264 Serbian 22, 185, 195, 197 Sesotho 160 Slavic 9, 134, 184-185, 189, 195 Spanish 10-11, 21, 23, 29, 32, 38-39, 52, 69, 74, 83-85, 102-104, 118, 123, 130134, 136-138, 148-156, 158, 170, 172173, 177, 180-181, 183, 186, 191-193, 197-199, 206, 214, 220, 222-223, 234, 265-267 Argentinian 84-85, 97-102, 122, 124, 193, 235, 257-258
clitic inventory and lexical entries 100-102, 257-258 Carribean 70 European 5, 84-85, 97-102, 124, 150, 181-182, 217, 233, 256-257, 267 clitic inventory and lexical entries 97-100, 256 clitic sequences 257 Old 58 Rio de la Plata 5, 10, 98, 191-192 South American 181, 217, 267 Uruguayan 182 Yimas 112
Language index 273
Name index Abraham, W. 93 Aguado, M. 69 Aissen, J. 160, 161, 186, 235 Anderson, S.R. 1, 2, 18, 61, 62, 63, 64, 65, 74, 169, 185, 195, 196, 238 Artstein, R. 160, 162, 185 Auger, J. 11, 53, 194, 235 Avgustinova, T. 136, 184 Avram, A. 68 Bafile, L. 21 Barbosa, P. 69 Barbu, A.M. 1, 25, 29, 53, 68, 205, 237 Bentivoglio, P. 234 Benua, L. 122 Bonet, E. 1, 8, 36, 123, 127, 133, 134, 137, 138, 184 Booij, G. 3, 19, 20, 23, 24 Brandi, L. 235 Burzio, L. 234 Calabrese, A. 69, 106, 124, 200, 213, 214, 215 Canclini, E. 12, 48 Cardinaletti, A. 4, 17 Carvalho, J.B. de 29, 42, 43, 44, 67, 69 CZ avar, D. 2, 195 Cocchi, G. 68, 194 Cordin, P. 235 Crysmann, B. 1, 25, 26, 29, 32, 33, 35, 36, 44, 47, 65, 69, 70, 237 d’Andrade, E. 36 Dobrovie-Sorin, C. 25, 55, 59, 60, 234 Emonds, J. 136, 191 Fabri, R. 1, 2, 12, 122, 136, 238 Farkas, D. 234 Fischer, S. 67, 234 Fontana, J.M. 68, 69, 234 Franco, J. 218, 234 Fulmer, S. 68 Gerlach, B. 12, 25, 33, 123, 130, 131, 140, 185, 187, 234
Gierling, D. 234 Golston, C. 170 Grandgent, C. 187 Grijzenhout, J. 187, 234 Grimshaw, J. 2, 74, 86, 87, 88, 89, 140, 160, 179, 181, 184, 185, 187, 198, 199, 221, 238, 270 Hall, T.A. 19, 20, 23 Halle, M. 1, 74, 127, 136 Halpern, A.L. 1, 2, 196 Harris, A. 234 Harris, J. 1, 127, 130, 137, 138, 181 Harrison, D.K. 1 Haspelmath, M. 68, 186 Hauge, K.R. 234 Hayes, B. 3, 19, 20, 65, 237 Herschensohn, J. 136 Hetzron, R. 136 Hofmann, J.B. 50 Hundertmark-Santos Martins, M.T. 17, 44, 57, 59, 63, 70, 209 Inkelas, S. 3, 19, 20, 123 Jacobs, H. 234 Jaeggli, O.A. 10, 192 Jeffers, R.J. 70 Joppen, S. 2, 12 Kaiser, G. 17, 25, 194, 234, 267 Kaisse, E.M. 187 Kato, M.A. 69 Kaufmann, I. 93 Kayne, R.S. 2, 4, 10, 11, 17, 29, 51, 190, 191, 192, 198 Kazazis, K. 234 Kemmer, S. 93 Kiparsky, P. 13 Klaiman, M.H. 93 Klavans, J.L. 61, 62, 63, 64, 65, 67, 198, 238 Klein, H-W. 60, 114, 116
274 Name index
Kleineidam, H. 60, 114, 116 Kleinhenz, U. 19 Koopmann, H. 200 Kühner, R. 2, 3 Laenzlinger, C. 185 Lakämper, R. 68 Lamontagne, G. 79 Landa, A. 234 Leeuw, F. van der 29, 44, 56, 57, 58, 59, 70, 209, 210, 211 Legendre, G. 1, 2, 29, 33, 38, 46, 54, 55, 56, 60, 61, 65, 69, 71, 140, 145, 164, 185, 195, 196, 197, 198, 200, 201, 202, 203, 204, 234 Lehmann, C. 69 Lubowicz, A. 187 Manzini, R. 181, 233 Marantz, A. 1, 74, 127, 136 Martins, A.M. 69 Mateus, M.H. 36 McCarthy J.J. 15, 136 Miller, P. 1, 25, 29, 31, 45, 46, 47, 49, 53, 65, 67, 69, 194, 235, 237 Monachesi, P. 1, 24, 25, 29, 30, 31, 33, 34, 37, 38, 39, 40, 41, 45, 46, 47, 49, 53, 54, 55, 56, 60, 65, 67, 68, 69, 71, 118, 130, 131, 136, 179, 184, 213, 215, 234, 237, 265 Morin, Y-C. 67, 233 Muller, C. 68 Müller, G. 122 Müller, N. 194 Nespor, M. 3, 19, 20, 24, 65, 68, 237 Noyer, R. 68 Nübling, D. 1, 3, 4, 17, 26, 27, 28, 29, 49, 51, 65 Ortmann, A. 11, 69, 195 Palma Caetano, A.J. 93 Parodi, T. 29, 160, 234, 267 Peperkamp, S. 23, 24, 25, 68 Perlmutter, D.M. 32, 38, 68, 131, 136, 137, 265 Poletto, C. 235 Pollard, C. 13, 122 Pollock, J.Y. 74
Popescu, A. 25, 29, 38, 49, 53, 56, 63, 67, 68, 69, 70, 118, 129, 164, 165, 207, 208, 227, 234, 267 Prince, A.S. 15, 136, 186 Pullum, G. 1, 4, 17, 25, 26, 49, 67 Rice, K. 79 Riemer, B. 194 Rivera-Castillo, Y. 24, 70 Rizzi, L. 190, 235 Roberge, A. 235 Sag, I. 1, 13, 25, 29, 31, 53, 65, 122, 194, 237 Samek-Lodovici, V. 221, 270 Schwarze, C. 215 Selkirk, E.O. 3, 19, 20, 22, 164 Siewierska, A. 143 Simpson, J. 185 Smolensky, P. 15, 136, 160, 186 Spencer, A. 1, 61, 70, 71, 127, 130, 138, 139, 185 Sportiche, D. 2, 10, 136, 192, 193, 194, 200, 218, 233 Starke, M. 17 Steele, S. 70 Stiebels, B. 12, 92, 93, 100, 160, 235 Stump, G.T. 1, 74, 127, 136, 138 Suñer, M. 11, 193, 218, 220, 226, 235 Sylla, Y. 136 Szantyr, A. 50 Tomic´, O. 185 Torrego, E. 11, 218 Urbas, M. 125 Uriagereka, J. 2, 193, 194, 198, 218, 233 Vigário, M. 25, 29, 36, 69 Vogel, I. 3, 19, 20, 22, 24, 65, 237 Wackernagel, J. 21, 58, 185 Wanner, D. 51, 136 Warburton, I.P. 191 Wilder, C. 2, 195 Withgott, M. 185 Wunderlich, D. 1, 2, 12, 13, 68, 92, 112, 122, 136, 238 Zec, D. 3, 19, 20 Zwicky, A.M. 1, 3, 4, 17, 25, 26, 28, 29, 39, 40, 45, 47, 49, 65, 67, 68, 70
Name index 275
Subject index 3-Silben-Gesetz 21, 58 affixal clitic 22-24 agreement 11-13, 16, 28-29, 52-53, 75-81, 90, 99, 105, 108, 122-123, 137, 140-141, 151, 177, 190-195, 218, 225-226, 230240 affix 48, 77-79, 194 phrase 73, 190, 193-194 subject 13, 48, 77-81 alignment constraints 16, 65, 83, 156, 183, 232234 directional 144 domain of 144-145, 234, 239 interpretation of 144-145, 234 morphological 42, 142-148, 151-152, 156-159, 163, 171-173, 181, 185-186, 196, 200-206, 233, 238-239 mutual 145, 169 phonological 164-169, 207-210, 239 harmonic 160-162, 185, 235 *alliteration 37, 170, 176-178, 180-181, 187 A-Morphous Morphology 74 see also phrasal affix analytic verb form 42, 45-47, 64, 128-129, 133, 198-204, 207-208 animacy 11, 53, 116, 125, 195, 217-218, 233 hierarchy 143 antihomophony 170, 179, 183, 187 see also homophonous clitics arbitrary gaps 26, 29-33, 66 argument hierarchy 143, 159-160, 183, 186 realisation see constraints on argument realisation role 13-15, 73, 76, 93, 140, 152, 187, 215, 218-219, 233, 238-239
structure 12-15, 74-81, 187 article definite 11, 34, 68, 85, 192 indefinite 19 auxiliary clitic 18, 53-59, 117, 120, 128, 133, 138-141, 164, 166-167, 207-208 base generation 189, 192-195 case features 14, 111, 113, 122, 177, 230 marker 11, 150, 157, 192 structural 14 syncretism 92 causative verb 213 chain co-indexing 193 clitic choice 74, 83, 87, 101-102, 148-156, 158-159, 162, 174, 186, 218, 231-232 climbing 54, 198-200, 213-216, 233234 cluster see clitic sequence cluster template 131, 136, 138 doubling 2, 10-12, 15-16, 52-53, 73, 184, 189-195, 216-232, 235, 239 semantic restrictions on 193, 217218, 226-232 group 3, 19-25, 65, 67, 237 inventory 7, 74, 83-121, 130, 132, 142, 179, 181, 185, 196 paradigm 5, 13, 29, 86, 89, 99, 102, 104, 109-112, 116, 118, 120-121, 176, 181, 184, 238 construction of 90-91, 100-101, 119 placement 2, 9-10, 18, 63, 73, 93, 189216, 232, 234, 239 see also position of clitic with respect to the host parameter 61 sequence 1, 7-8, 15-16, 31-33, 36-39, 59-60, 66, 70, 73, 89, 123, 127-185,
276 Subject index
199-203, 206, 208, 213-214, 218, 231, 233, 238 internal order 5, 7, 39, 42, 59, 65-66, 127, 129-134, 136, 138, 142-163, 170, 175, 183, 185, 191, 196, 233, 237, 239 see also ordering criterion deviation from 156-159, 171-174 paradigm 142, 156, 163-165, 167, 171 prohibited 148-156, 162, 172, 239 separation of 7, 59, 128-130 see also clitic split combinatory restrictions 1, 5, 31-32, 127, 129-133, 142, 146, 159-160, 183, 186, 191, 218, 238 other restrictions 37, 129, 207 split 59, 128-129, 198, 207-208, 233 see also separation of clitic sequence strong 193, 233 voice 194 weak 193 clitic-clitic linearisation see clitic sequence, internal order clitic-host linearisation see position of clitic with respect to the host Completeness 13 constraints on argument realisation 75-81, 153, 218-219, 233, 238 see also faithfulness constraints morphological 8, 10, 16, 33, 97, 102, 140, 147, 153, 163, 171, 175, 179180, 198, 232, 234, 238 see also alignment constraints phonological 16, 140, 142, 156, 163165, 167, 170-171, 174-175, 178-180, 182, 184, 187, 189, 213 see also alignment constraints prosodic 10, 83, 118, 141, 164, 184, 198, 206, 234, 239 syntactic 189, 238 coordination criterion 30, 45-47, 55-56, 67, 69 see also deletion under identity
correspondence relation 15, 75, 79, 93, 122, 141, 174, 186, 220, 226 Correspondence Theory 15-16, 74, see also Optimality Theory deletion of clitic 8, 15-16, 38-39, 59, 69, 73, 83, 89, 133, 163, 170, 180, 184, 239 under identity 26, 46-47, 67 dep see faithfulness constraints discourse-referential features 75, 217, 230, 233-234 see also topic and focus dissimilation 170 see also OCP Distributed Morphology 1, 74, 127, 136139 domain of clisis 62-63, 71, 198-206, 233 phonological 20 syntactic 63 early vs. late insertion 75 economy 17, 100, 109 constraints 11, 153, 195, 220, 230, 232, 235, 240 edgemost constraints see interpretation of alignment constraints elision 68 see also vowel deletion enclitic 3, 7, 20, 48, 53, 61, 69, 187, 209210, 213, prosodic structure of 23-24 evolution of clitics 70 faithfulness constraints 15, 74-82, 87-89, 93, 95, 97, 101, 103-104, 111-112, 121, 123, 140, 142, 145-146, 150-152, 154, 156, 162-163, 186, 218-221, 227-229, 232, 238 morphosyntactic 171-172, 175-177, 179, 181-183, 185 phonological 165-167, 169, 174-175, 211 fill see faithfulness constraints focus 28, 153, 219-223 position 228 foot binarity 210 free clitic 22-23 GEN 75 Head-driven Phrase Structure Grammar
Subject index 277
(HPSG) 13 hierarchy of functional categories 123 homophonous clitics 37-38, 119 see also *antihomophony host 2-3, 27 category 47 distinct morphological and phonological 48-49, 53, 61, 69, 208-209, 237 morphological 141 phonological 19-20, 22, 63, 67, 141, 239 human reference 5, 11, 51, 53, 113-115, 186, 192, 195, 217, 227-232, 235 ident see faithfulness constraints impersonal passive construction 93 inflectional affixation 25 information structure 141, 153, 162, 185, 221, 240 input 15-16, 75-76, 79-80, 82, 86-88, 93-94, 166, 219 information 12, 74-75, 78, 92 internal clitic 22 interrogative clitic see question clitic intervocalic s-voicing 24 Kayne’s Generalization 11 see also movement approach Lexical Decomposition Grammar (LDG) 2, 13-15 lexical entry 12-13, 74-75, 89-93, 98, 100, 103-104, 109-110, 113, 115, 117-118, 120, 141, 227-229 disjunctive 99, 151 underspecified 100, 103, 121 see also underspecification linking 2, 13-15, 74 principles 14-15 local conjunction 160-163, 187 locative clitic 5, 7, 51, 59, 65, 76, 83, 85, 87, 107-110, 113-114, 116, 130, 132, 177, 237 markedness constraints 74, 83, 92-93, 97, 100, 121, 159-163, 172, 186, 238 hierarchy 143, 159-163, 184-185, 235, 239
morphological 143, 151, 159, 232 rankings 160-161 Matching Principle 226 max see faithfulness constraints mesoclisis 43-45, 56-59, 63, 66, 68, 198, 209-213, 233-234, 237, 239 Minimalist Morphology 1, 6, 12-13, 136, 238 minimality restriction see prosodic minimality modal verb construction 7, 42, 64, 128, 190, 199-201, 204, 215-216, 234 see also clitic climbing morpheme integrity 169, 210, 234 morphological feature transfer 123, 134, 182 morphophonological idiosyncrasies 26, 29-30, 33-39 movement approach 10, 189-192 *multiple correspondence 79, 219-223, 234, 238, 240 *m-structure 153 negation 25, 29, 59-60, 66, 71, 76, 123, 238 clitic 59-60, 83, 85, 115-116, 133, 185, 234, 238 neutralisation 35-36, 123, 134, 169 no hiatus 165-167, 208 no stress clash 212 OCP 37, 142, 163, 170, 184, 187, 239 onset 165-167 opaque clitic 1, 8, 36, 133-135, 137, 142, 170, 181, 184, 191, 239 Optimality Theory (OT) 10, 15, 59, 61, 67, 74, 86-89, 122, 127, 136, 139, 160, 189, 232, 239 see also Correspondence Theory order of clitics see clitic sequence internal order ordering criterion 26, 39-45, 68 output 15-16, 79, 82-83, 140, 152, 166 candidate 75-77, 80, 86, 140-141, 151152 information 12, 74
278 Subject index
Output Specificity 13 output-output correspondence 122 Paradigm Function Morphology 1, 61, 74, 127, 136, 138-140 parse see faithfulness constraints participle agreement 105-106, 114, 124, 190, 194 partitive clitic 5, 7, 51, 59, 65, 76, 83, 85, 105-107, 113-114, 116, 130, 132, 190, 237 perception verbs 213 person hierarchy 159-160, 162 phonological phrase 3, 19-20, 23-24 phonological word see prosodic word phrasal affix 1, 18, 49, 61-65, 67, 238 position of clitic (sequence) with respect to the host 40, 56, 59, 61-65, 189, 191-216, 237 see also ordering criterion alternation of 40-42, 63-64, 93, 129, 191, 196-197, 199-201, 216 postverbal 9, 40-42, 44-47, 49, 61, 63, 69, 199-200 preverbal 9, 40-47, 49, 57, 63, 66, 69, 128, 198-200, 207-209, 213 prepositional object 6, 106, 108-109, 113114 proclitic 3, 7, 20, 61 prosodic structure of 23 pronoun full 4-5, 12, 47, 50-51, 68, 77, 140, 148, 151-152, 158, 161, 179, 184, 186-187, 218, 231-232, 239 weak 30, 44, 68 prosodic hierarch 2-3, 18-22, 25, 65, 237 Prosodic Inversion 196 prosodic minimality 3, 18, 142, 163-164, 167, 207 prosodic word 3, 18-24, 53, 59, 63, 163, 167, 184, 187, 209-210, 233- 234, 239 prosodic word stress 210 quantified NP 105 question clitic 139, 234 radoppiamento sintattico 24-25 referential features 154, 226, 229 see also
human reference, specific reference and discourse-referential features reflexive clitic 5-6, 8, 32, 36-38, 50, 59, 65, 85-86, 89, 93-94, 96-97, 100, 121, 132133, 136-138, 177-183, 237 impersonal 112-115, 178-179, 183 reflexive middle construction 93 restructuring verbs 190 resyllabification 169, 187 *rhyme 170, 175 scope of clitic 45-46, 52-53, 69 second position clitic 56 selectivity (with respect to host category) 27, 47-49 simple clitic 3, 27-28, 51-52 Simplicity 13 sonority hierarchy 170-171, 186 *sonority adjacency 170-171, 173-176, 180, 187 special clitic 3, 18, 27-28, 50-52, 61, 65, 237 specific reference 11, 53, 125, 154, 186, 192-195, 217-219, 227-233, 235 Specificity Principle 14, 99, 113, 117, 123, 179, 184 spurious se 38, 136-137 stem dominance 210-212 stress 2-3, 23, 211, 234 assignment 20-24, 58-59, 70, 209-210 readjustment 21 primary 3, 57-59, 70, 209-210 secondary 3, 59, 210 Strict Layer Hypothesis 21-22 subcategorisation frame see input information subject 12, 14, 77-81, 89, 93, 108, 143, 200, 213-214, 217, 237 agreement see agreement, subject – auxiliary inversion 54 clitic 5, 7, 14, 31, 48, 60, 65, 80, 83, 85, 110-113, 118, 124, 133, 141, 194, 218, 223-226, 234 impersonal 85, 93, 107-108, 112-115, 124, 132, 137, 177-179 substitution of clitic 8, 15, 37-39, 59, 73,
Subject index 279
83, 89, 116, 123, 127, 133-134, 136, 138, 154-155, 170, 176-182, 184, 239 syncretism 5-6, 86, 89, 98-99, 103, 111, 113, 118, 120-121 coincidental vs. systematic 90, 103, 119120 synthetic verb form 128 theta structure (TS) 74, 80, 122 topical argument 28, 51, 153-154, 219, 221-222, 229 transformational approach 73, 136, 190191, 233 Transitive Correspondence 16, 80, 220, 238 truncation 24-25, 68 underspecification 6, 73, 80, 86-89, 93, 96-
99, 101, 103, 111, 119-120, 125, 159, 167, 181, 185, 187, 226, 231, 238 principled vs. radical 123 Uniqueness Constraint 13-14 vowel contraction 36 deletion 33-36, 135, 164, 166-167, 207 lowering 36-37, 70, 134-135, 174-175, 187 raising 37, 70, 135, 175 reduction 135, 166 Wackernagel clitic see second position clitic word formation rule 61 word status 61 Zwicky criteria 17-18, 26, 28-29, 237
In the series LINGUISTIK AKTUELL/LINGUISTICS TODAY (LA) the following titles have been published thus far, or are scheduled for publication: 1. KLAPPENBACH, Ruth (1911-1977): Studien zur Modernen Deutschen Lexikographie. Auswahl aus den Lexikographischen Arbeiten von Ruth Klappenbach, erweitert um drei Beiträge von Helene Malige-Klappenbach. 1980. 2. EHLICH, Konrad & Jochen REHBEIN: Augenkommunikation. Methodenreflexion und Beispielanalyse. 1982. 3. ABRAHAM, Werner (ed.): On the Formal Syntax of the Westgermania. Papers from the 3rd Groningen Grammar Talks (3e Groninger Grammatikgespräche), Groningen, January 1981. 1983. 4. ABRAHAM, Werner & Sjaak De MEIJ (eds): Topic, Focus and Configurationality. Papers from the 6th Groningen Grammar Talks, Groningen, 1984. 1986. 5. GREWENDORF, Günther and Wolfgang STERNEFELD (eds): Scrambling and Barriers. 1990. 6. BHATT, Christa, Elisabeth LÖBEL and Claudia SCHMIDT (eds): Syntactic Phrase Structure Phenomena in Noun Phrases and Sentences. 1989. 7. ÅFARLI, Tor A.: The Syntax of Norwegian Passive Constructions. 1992. 8. FANSELOW, Gisbert (ed.): The Parametrization of Universal Grammar. 1993. 9. GELDEREN, Elly van: The Rise of Functional Categories. 1993. 10. CINQUE, Guglielmo and Guiliana GIUSTI (eds): Advances in Roumanian Linguistics. 1995. 11. LUTZ, Uli and Jürgen PAFEL (eds): On Extraction and Extraposition in German. 1995. 12. ABRAHAM, W., S. EPSTEIN, H. THRÁINSSON and C.J.W. ZWART (eds): Minimal Ideas. Linguistic studies in the minimalist framework. 1996. 13. ALEXIADOU Artemis and T. Alan HALL (eds): Studies on Universal Grammar and Typological Variation. 1997. 14. ANAGNOSTOPOULOU, Elena, Henk VAN RIEMSDIJK and Frans ZWARTS (eds): Materials on Left Dislocation. 1997. 15. ROHRBACHER, Bernhard Wolfgang: Morphology-Driven Syntax. A theory of V to I raising and pro-drop. 1999. 16. LIU, FENG-HSI: Scope and Specificity. 1997. 17. BEERMAN, Dorothee, David LEBLANC and Henk van RIEMSDIJK (eds): Rightward Movement. 1997. 18. ALEXIADOU, Artemis: Adverb Placement. A case study in antisymmetric syntax. 1997. 19. JOSEFSSON, Gunlög: Minimal Words in a Minimal Syntax. Word formation in Swedish. 1998. 20. LAENZLINGER, Christopher: Comparative Studies in Word Order Variation. Adverbs, pronouns, and clause structure in Romance and Germanic. 1998. 21. KLEIN, Henny: Adverbs of Degree in Dutch and Related Languages. 1998. 22. ALEXIADOU, Artemis and Chris WILDER (eds): Possessors, Predicates and Movement in the Determiner Phrase. 1998. 23. GIANNAKIDOU, Anastasia: Polarity Sensitivity as (Non)Veridical Dependency. 1998. 24. REBUSCHI, Georges and Laurice TULLER (eds): The Grammar of Focus. 1999. 25. FELSER, Claudia: Verbal Complement Clauses. A minimalist study of direct perception constructions. 1999.
26. ACKEMA, Peter: Issues in Morphosyntax. 1999. ° 27. RUZICKA, Rudolf: Control in Grammar and Pragmatics. A cross-linguistic study. 1999. 28. HERMANS, Ben and Marc van OOSTENDORP (eds): The Derivational Residue in Phonological Optimality Theory. 1999. 29. MIYAMOTO, Tadao: The Light Verb Construction in Japanese. The role of the verbal noun. 1999. 30. BEUKEMA, Frits and Marcel den DIKKEN (eds): Clitic Phenomena in European Languages. 2000. 31. SVENONIUS, Peter (ed.): The Derivation of VO and OV. 2000. 32. ALEXIADOU, Artemis, Paul LAW, André MEINUNGER and Chris WILDER (eds): The Syntax of Relative Clauses. 2000. 33. PUSKÁS, Genoveva: Word Order in Hungarian. The syntax of È-positions. 2000. 34. REULAND, Eric (ed.): Arguments and Case. Explaining Burzio’s Generalization. 2000. 35. HRÓARSDÓTTIR, Thorbjörg. Word Order Change in Icelandic. From OV to VO. 2000. 36. GERLACH, Birgit and Janet GRIJZENHOUT (eds): Clitics in Phonology, Morphology and Syntax. 2000. 37. LUTZ, Uli, Gereon MÜLLER and Arnim von STECHOW (eds): Wh-Scope Marking. 2000. 38. MEINUNGER, André: Syntactic Aspects of Topic and Comment. 2000. 39. GELDEREN, Elly van: A History of English Reflexive Pronouns. Person, ‘‘Self’’, and Interpretability. 2000. 40. HOEKSEMA, Jack, Hotze RULLMANN, Victor SANCHEZ-VALENCIA and Ton van der WOUDEN (eds): Perspectives on Negation and Polarity Items. 2001. 41. ZELLER, Jochen : Particle Verbs and Local Domains. 2001. 42. ALEXIADOU, Artemis : Functional Structure in Nominals. Nominalization and ergativity. 2001. 43. FEATHERSTON, Sam: Empty Categories in Sentence Processing. 2001. 44. TAYLAN, Eser E. (ed.): The Verb in Turkish. 2002. 45. ABRAHAM, Werner and C. Jan-Wouter ZWART (eds): Issues in Formal German(ic) Typology. 2002. 46. PANAGIOTIDIS, Phoevos: Pronouns, Clitics and Empty Nouns. ‘Pronominality’ and licensing in syntax. 2002. 47. BARBIERS, Sjef, Frits BEUKEMA and Wim van der WURFF (eds): Modality and its Interaction with the Verbal System. 2002. 48. ALEXIADOU, Artemis, Elena ANAGNOSTOPOULOU, Sjef BARBIERS and HansMartin GAERTNER (eds): Dimensions of Movement. From features to remnants. 2002 49. ALEXIADOU, Artemis (ed.): Theoretical Approaches to Universals. 2002. 50. STEINBACH, Markus: Middle Voice. A comparative study in the syntax-semantics interface of German. 2002. 51. GERLACH, Birgit: Clitics between Syntax and Lexicon. 2002. 52. SIMON, Horst J. and Heike WIESE (eds): Pronouns – Grammar and Representation. 2002.
53. ZWART, C. Jan-Wouter and Werner ABRAHAM (eds): Studies in Comparative Germanic Syntax. Proceedings from the 15th Workshop on Comparative Germanic Syntax (Groningen, May 26-27, 2000)(Workshop). n.y.p. 54. BAPTISTA, Marlyse: The Syntax of Cape Verdean Creole. The Sotavento varieties. n.y.p. 55. COENE, M. and Yves D'HULST (eds): From NP to DP. Volume 1: The syntax and semantics of noun phrases. n.y.p. 56. COENE, M. and Yves D'HULST (eds.): From NP to DP. Volume 2: The expression of possession in noun phrases. n.y.p. 57. DI SCIULLO, Anna-Maria (ed.): Asymmetry in Grammar. Volume 1: Syntax and semantics. n.y.p. 58. DI SCIULLO, Anna-Maria (ed.): Asymmetry in Grammar. Volume 2: Morphology, phonology, acquisition. n.y.p. 59. DEHÉ, Nicole: Particle Verbs in English. Syntax, information structure and intonation. n.y.p. 60. TRIPS, Carola: From OV to VO in Early Middle English. n.y.p. 61. SCHWABE, Kerstin and Susanne WINKLER (eds.): The Interfaces. Deriving and interpreting omitted structures. n.y.p. 62. CARNIE, Andrew, Heidi HARLEY and Mary WILLIE (eds.): Formal Approaches to Function in Grammar. In honor of Eloise Jelinek. n.y.p.
