The Verbal Complex in Romance: A Case Study in Grammatical Interfaces

The Verbal Complex in Romance OX FO RD ST UDI E S I N T H E O R ET I C A L LI NGUISTIC S GENERAL EDITORS David Adger...

Author: Paola Monachesi

14 downloads 899 Views 1MB 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

DOWNLOAD PDF

The Verbal Complex in Romance

OX FO RD ST UDI E S I N T H E O R ET I C A L LI NGUISTIC S GENERAL EDITORS

David Adger, Queen Mary College London; Hagit Borer, University of Southern California ADVISORY EDITO R S

Stephen Anderson, Yale University; Gennaro Chierchia, University of Milan; Rose-Marie Dechaine, University of British Columbia; Elan Dresher, University of Toronto; James Higginbotham, University of Southern California; Pat Keating, University of California, Los Angeles; Ruth Kempson, King’s College, University of London; James McCloskey, University of California, Santa Cruz; Gillian Ramchand, University of Tromsφ; Maria-Luisa Zubizarreta, University of Southern California PUB LISH ED

The Syntax of Silence: Sluicing, Islands, and the Theory of Ellipsis by Jason Merchant Questions and Answers in Embedded Contexts by Utpal Lahiri Phonetics, Phonology, and Cognition edited by Jacques Durand and Bernard Laks At the Syntax–Pragmatics Interface: Concept Formation and Verbal Underspecification in Dynamic Syntax by Lutz Marten The Unaccusativity Puzzle: Explorations of the Syntax–Lexicon Interface edited by Artemis Alexiadou, Elena Anagnostopoulou, and Martin Everaert Beyond Morphology: Interface Conditions on Word Formation by Peter Ackema and Ad Neeleman The Logic of Conventional Implicatures by Christopher Potts Paradigms of Phonological Theory edited by Laura Downing, T. Alan Hall, and Renate Raffelsiefen The Verbal Complex in Romance: a Case Study in Grammatical Interfaces by Paola Monachesi The Syntax of Aspect edited by Nomi Erteschik-Shir and Tova Rapoport IN PREPARATION

Aspects of the Theory of Clitics by Stephen Anderson Stratal Optimality Theory: Synchronic and Diachronic Applications by Ricardo Bermúdez Otero Tense, Mood, and Aspect edited by Alessandra Giorgi, James Higginbotham, and Fabio Pianesi The Ecology of English Noun–Noun Compounding by Ray Jackendoff A Natural History of Infixation by Alan Chi Lun Yu The Oxford Handbook of Linguistic Interfaces edited by Gillian Ramchand and Charles Reiss [published in association with the series]

The Verbal Complex in Romance A Case Study in Grammatical Interfaces PAOLA MONACHESI

1

3

Great Clarendon Street, Oxford OX2 6DP Oxford University Press is a department of the University of Oxford. It furthers the University’s objective of excellence in research, scholarship, and education by publishing worldwide in Oxford New York Auckland Cape Town Dar es Salaam Hong Kong Karachi Kuala Lumpur Madrid Melbourne Mexico City Nairobi New Delhi Shanghai Taipei Toronto With offices in Argentina Austria Brazil Chile Czech Republic France Greece Guatemala Hungary Italy Japan Poland Portugal Singapore South Korea Switzerland Thailand Turkey Ukraine Vietnam Oxford is a registered trade mark of Oxford University Press in the UK and in certain other countries Published in the United States by Oxford University Press Inc., New York © Paola Monachesi 2005 The moral rights of the author have been asserted Database right Oxford University Press (maker) First published 2005 All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, without the prior permission in writing of Oxford University Press, or as expressly permitted by law, or under terms agreed with the appropriate reprographics rights organization. Enquiries concerning reproduction outside the scope of the above should be sent to the Rights Department, Oxford University Press, at the address above You must not circulate this book in any other binding or cover and you must impose this same condition on any acquirer British Library Cataloguing in Publication Data Data available Library of Congress Cataloguing in Publication Data Data available Typeset by SPI Publisher Services, Pondicherry, India Printed in Great Britain on acid-free paper by Biddles Ltd., www.biddles.co.uk ISBN 0–19–927475–4 10 9 8 7 6 5 4 3 2 1

978–0–19–927475–8

Contents General Preface Dedication Abbreviations Acknowledgements

ix xi xiii xvi

Introduction 1.1 Outline . . . . . . . . . . . . . . . . . . . . . . . .

1 7

2 The Framework 2.1 Basic assumptions . . . . . . . . . . . . . . . . . . . 2.2 The principles . . . . . . . . . . . . . . . . . . . . . 2.3 Phrases and constituent structure . . . . . . . . . . . 2.4 Lexical generalizations . . . . . . . . . . . . . . . . . 2.5 The organization of the grammar and the implications for the interfaces: A comparison with other theories . .

9 9 12 15 17

1

3

Morphology and its Interfaces: The Case of Romance Clitics 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . 3.2 A constrained-based view of morphology . . . . . . . 3.2.1 The word–syntax approach . . . . . . . . . . . 3.2.2 Realizational approaches . . . . . . . . . . . . . 3.3 Romance clitics as a test case for grammatical interfaces 3.3.1 General properties of Romance clitics . . . . . . 3.3.2 The affixal status of Romance clitics . . . . . . . 3.3.3 Phonology . . . . . . . . . . . . . . . . . . . . 3.3.4 Conclusions . . . . . . . . . . . . . . . . . . . 3.4 A lexical analysis of Romance clitics . . . . . . . . . . 3.4.1 Clitic doubling . . . . . . . . . . . . . . . . . . 3.4.2 Subject clitics . . . . . . . . . . . . . . . . . . . 3.5 Morphophonological properties of Romance clitics . . 3.5.1 Syntactic and morphological approaches to clitic distribution . . . . . . . . . . . . . . . . . . . 3.5.2 The clitic cluster . . . . . . . . . . . . . . . . . 3.5.3 The morphophonological features . . . . . . . . 3.5.4 The spelling out of Romance pronominal clitics .

20 25 25 26 26 31 34 36 42 65 75 76 81 85 87 87 89 96 100

vi

Contents

3.6 3.7 3.8 3.9 3.10 3.11

3.5.5 An alternative realizational approach . . . . Germanic clitics . . . . . . . . . . . . . . . . . Slavic clitics . . . . . . . . . . . . . . . . . . . . A brief review . . . . . . . . . . . . . . . . . . . Comparing frameworks: Distributed Morphology Comparing frameworks: Optimality Theory . . . Conclusions . . . . . . . . . . . . . . . . . . .

. . . . . . .

. . . . . . .

107 115 117 121 123 126 129

4 Syntax and its Interfaces: The Case of Romance Auxiliary Verbs 131 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . 131 4.2 Basic properties of Romanian tense auxiliaries . . . . 133 4.3 The syntactic structure of tense auxiliaries . . . . . . 139 4.3.1 The syntactic structure of French auxiliary verbs 140 4.3.2 The syntactic structure of Italian auxiliary verbs 144 4.3.3 The syntactic structure of Romanian tense auxiliaries . . . . . . . . . . . . . . . . . . . . . . 146 4.4 Mesoclisis in European Portuguese . . . . . . . . . . 152 4.5 Romanian tense auxiliaries and the distribution of pronominal clitics . . . . . . . . . . . . . . . . . . 158 4.5.1 The position of pronominal clitics in the Romanian and Balkan clitic cluster . . . . . . . . . . 159 4.5.2 The ordering of Romanian clitics . . . . . . . . 161 4.5.3 The analysis of Romanian clitic climbing . . . . 164 4.6 Romanian tense auxiliaries and monosyllabic intensifiers 175 4.7 Romanian tense auxiliaries and negation . . . . . . . 181 4.8 Romanian future auxiliaries . . . . . . . . . . . . . 185 4.9 The perfect auxiliary fi . . . . . . . . . . . . . . . . 192 4.10 Romanian auxiliary-like constructions . . . . . . . . 197 4.10.1 Raising verbs . . . . . . . . . . . . . . . . . 197 4.10.2 The modal verb a putea . . . . . . . . . . . . 205 4.11 A brief review . . . . . . . . . . . . . . . . . . . . . 210 4.12 Comparison with other accounts . . . . . . . . . . . 211 4.13 The role of argument structure: A comparison . . . . 214 4.14 What is the role of syntactic structure? . . . . . . . . 217 4.15 Conclusions . . . . . . . . . . . . . . . . . . . . . 219 5

Phonology and its Interfaces: The Case of Italian Restructuring Verbs 221 5.1 Introduction . . . . . . . . . . . . . . . . . . . . . 221 5.2 The data . . . . . . . . . . . . . . . . . . . . . . . 222

Contents

vii

5.3 The syntactic structure of Italian restructuring verbs . 5.4 Restructuring verbs and clitic climbing . . . . . . . . 5.5 The Prosodic Representation of Restructuring Verbs . 5.5.1 Evidence against Prosodic Word status . . . . . 5.5.2 Evidence in favour of Phonological Phrase status 5.6 The syntax–phonology interface . . . . . . . . . . . . 5.6.1 Ghini (1993) . . . . . . . . . . . . . . . . . . . 5.6.2 Selkirk (1986, 1995) and Truckenbrodt (1999) . . 5.6.3 Nespor and Vogel (1986) . . . . . . . . . . . . . 5.7 A templatic view on restructuring verbs: Cinque (2000) 5.8 Conclusions . . . . . . . . . . . . . . . . . . . . . .

223 228 234 235 237 240 240 246 250 259 265

References Index

267 282

This page intentionally left blank

General Preface The theoretical focus of this series is on the interfaces between subcomponents of the human grammatical system and the closely related area of the interfaces between the different subdisciplines of linguistics. The notion of ‘interface’ has become central in grammatical theory (for instance, in Chomsky’s recent Minimalist Program) and in linguistic practice: work on the interfaces between syntax and semantics, syntax and morphology, phonology and phonetics, etc. has led to a deeper understanding of particular linguistic phenomena and of the architecture of the linguistic component of the mind/brain. The series will cover interfaces between core components of grammar, including syntax/morphology, syntax/semantics, syntax/phonology, syntax/pragmatics, morphology/phonology, phonology/phonetics, phonetics/speech processing, semantics/pragmatics, intonation/discourse structure as well as issues in the way that the systems of grammar involving these interface areas are acquired and deployed in use (including language acquisition, language dysfunction, and language processing). It will demonstrate, we hope, that proper understanding of particular linguistic phenomena, languages, language groups, or inter-language variations all require reference to interfaces. The series is open to work by linguists of all theoretical persuasions and schools of thought. A main requirement is that authors should write so as to be understood by colleagues in related subfields of linguistics and by scholars in cognate disciplines. In our ninth volume Paola Monachesi explores the way that grammatical interfaces can be conceptualized in Head Driven Phrase Structure Grammar, taking as her empirical domain the Romance Verbal Complex. She argues that a deeper understanding of this domain follows from integrating syntax, semantics, and prosody into a single uniﬁed perspective. This proposal can be elegantly implemented in a uniﬁcation based framework like HPSG by under-specifying the properties of lexical items. July 2004 David Adger Hagit Borer

This page intentionally left blank

Dedication To the memory of my father and to our common love for the country house in Montanello without which this book would have never been written.

This page intentionally left blank

Abbreviations Abbreviations in glosses: 1 2 3 acc aux cl conj dat f fin imp imper ind inf int loc m neg nom nonfin part pas pl prep pres part psp refl sg sub

first person second person third person accusative auxiliary clitic conjunction dative feminine finite impersonal imperative indicative infinitive intensifier locative masculine negation nominative nonfinite partitive passive plural preposition present participle past participle reflexive singular subjunctive

Languages: bg fr it mac

Bulgarian French Italian Macedonian

xiv

Abbreviations

na pt ro sl sp tor tr

Napoletano Portuguese Romanian Slovenian Spanish Torinese Trentino

HPSG

abbreviations:

acl-ss cl-ss mark-ss nom-obj noncl-ss npro p-ss prcl-ss psoa pw quant syll w-ss arg-st aux cat/c clts comps cons cont h hd-dtr loc/l lp nonhd-dtrs phon reln spr ss subj val vform vow

anaphoric synsem clitic synsem marker clitic synsem nominal object nonclitic synsem nonpronoun phrase synsem pronominal clitic synsem parametrized state of affairs prosodic word quantifier syllable word synsem argument structure auxiliary category clitics complements consonants content head head daughter local linear precedence nonhead daughters phonology relation specifier synsem subject valence verbal form vowels

Abbreviations

Symbols: ⊕ {} ∨ ¬ → →

append relation list set disjunction negation relates input and output in lexical rules used in implicational constraints

xv

Acknowledgements This book has grown out of my interest in Italian cliticization that dates back to my dissertation (Monachesi 1996a) and its revision (Monachesi 1999a). A three-year research grant from the Netherlands Organization for Scientiﬁc Research (NWO), which I gratefully acknowledge, has allowed me to pursue this interest further by taking into consideration cliticization in Romance languages, as well as complex predicates, and to focus on the interface dimension of these phenomena. During the years I spent writing this book, I greatly beneﬁted from opportunities to discuss my ideas with other people. Preliminary versions of the book were presented at several conferences, seminars, and summer schools. For their comments and suggestions, I am particularly grateful to the students and the audience of the EUROLAN Summer School in Tusnad (1997), the ESSLLI summer school in Helsinki (2001), and especially the seminars at Paris 7 on morphology and the interfaces (2002) and that on weak elements (2003). Special thanks to Anne Abeillé and Danièle Godard for inviting me to Paris and for creating a very stimulating working environment. Several people have given me helpful comments on this work. Among them, I am especially indebted to Anne Abeillé, Artemis Alexiadou, Laura Bafile, Miriam Butt, Carmen Dobrovie-Sorin, Birgit Gerlach, Mirco Ghini, Danièle Godard, Aditi Lahiri, Detmar Meurers, Philip Miller, Michael Moortgat, Marina Nespor, Alexandra Popescu, Frank Richter, Ivan Sag, and Manfred Sailer. For help with the judgements and the data, I would like to thank Ana Maria Barbu, Agnes Bende Farkas, Emil Ionescu, Manuela Lehau, Alexandra Popescu, Amalia Todiras¸cu, Cristina Vertan, and Diana Zaiu, (Romanian), Josep Quer (Spanish), Antonio Branco (Portuguese), Anne Abeillé, Danièle Godard, and Philip Miller (French). Publishing a book with Oxford University Press has been a very pleasant experience. I am particularly grateful to the series editor, David Adger, for suggestions and advice, John Davey and Stuart Fowkes for editorial assistance, and the anonymous reviewers for taking their task so seriously. The book has beneﬁted immensely from their comments, criticism, and suggestions. A stimulating and lively academic environment is necessary to carry out good research. However, writing up the results in a book is a lonely experience that requires dedication, time, discipline, and, in my case, a beautiful environment.

Acknowledgements

xvii

Without our country house in Montanello, I would have never written this book. The beautiful landscape of Marche, the olive trees, and especially the hammock under the pergola have helped my concentration immensely. I am also indebted to the peaceful house on the top of the dunes in Bergen aan Zee, to the sunny terrace near Piazza del Popolo in Rome, to the beautiful garden in via della Pietra in the Firenze countryside, and ﬁnally to the houseboat with the ﬂoating terrace on the Amstel: without them, writing this book would have never been such a wonderful enterprise. I regret that writing this book has taken far too much time and that my father couldn’t see it in its published version. I dedicate this book to him.

This page intentionally left blank

1 Introduction The theoretical domain of investigation of this volume is the nature of the interface between syntax and the other components of the grammar, in particular phonology, morphology, and argument structure. The empirical domain of analysis is concerned with the properties of the verbal complex in various Romance languages. Cliticization is taken into consideration as well as complex predicates, that is auxiliary verbs as well as modal, aspectual, and motion verbs in Romance. I show that these phenomena constitute ideal test cases which can help shed new light on the organization of the grammar. The idiosyncrasies exhibited by cliticization as well as the variation attested within complex predicates offer new insights into the mechanisms of the grammar. They provide relevant information about the structure of language and constitute a proving ground for controversial claims. Various aspects of these linguistic phenomena have already received considerable attention in the literature; however, their relevance for the study of the interface has not always been highlighted in previous works. The aim of this volume is to show that a comprehensive analysis of their properties can be achieved only in terms of an interaction of the various components of the grammar. The study of these phenomena thus plays a key role in attaining a better understanding of the organization of the grammar. The data under investigation come from the ‘core’ Romance languages: Italian, French, and Spanish; however, Portuguese is also considered in the analysis of cliticization and in the treatment of auxiliary verbs. Special attention, however, is dedicated to Romanian since it occupies a position that makes it of particular importance to linguists with comparative interests, given that it allows for several types of comparison to be made, as claimed by Joseph (1999). Comparisons of a genetic nature are possible by looking at Romanian in the context of those languages most closely related to it, that is the Romance languages. Comparisons of a geographical nature can also be made by taking into consideration its linguistic neighbours, that is the other Balkan languages. Furthermore, typological comparison can be carried out in order to pursue linguistic universals. Looking at Romanian from these perspectives is particularly relevant since in many ways it does not pattern with the other Romance languages and the reason for this is to be found in the fact that it patterns

2

Introduction

with the other major Balkan languages: Bulgarian, Macedonian, Greek, and Albanian. As already mentioned, one of the phenomena addressed in this volume is cliticization. A number of issues are analysed in depth including the representation and placement of clitics and the nature of clitic clustering in various Romance languages. An attempt is made to tease apart the roles that morphology, phonology, and syntax play in the analysis of cliticization. Romance clitics occur in a special position within the sentence, that is on the left of the finite verb (1a), while complements usually occur on the right (1b), as can be seen from the following Italian examples: (1) a. Martina glielo spedisce. [it] Martina CL.(DAT) CL.(ACC) sends ‘Martina sends it to me.’ b. Martina spedisce il libro a Vito. [it] Martina sends the book to Vito ‘Martina sends the book to Vito.’ Example (1a) shows that clitics cluster together in a fixed order which is different from that of the corresponding full phrases (1b). Furthermore, clitics cannot occur alone and they need a host to attach to, which is the verb in Romance languages. Nothing can intervene between the clitic and the verb, including adverbs, as shown by the following example: (2) *Martina glielo rapidamente spedisce. [it] Martina CL.(DAT) CL.(ACC) quickly sends ‘Martina sends it to me quickly.’ In certain Romance languages such as Italian (3a) and French, pronominal clitics are in complementary distribution with the correspondent full phrase, while in others such as Spanish and Romanian (3b), they can combine (under certain conditions) with the related complement: (3) a. *Martina gli dá un libro al presidente. [it] Martina CL.DAT gives a book to.the president d˘a pres¸edintelui o carte. [ro] b. Martina îi Martina CL.DAT gives president.the a book ‘Martina gives the president a book.’ From this brief overview, it is immediately evident that a comprehensive analysis of the properties of Romance clitics requires an interaction of the different modules of the grammar: they attach phonologically to the verb, they affect

Introduction

3

its subcategorization requirements, and they exhibit a fixed order which is typical of morphological elements. The challenge that clitics pose for any formal treatment is that their morphophonological properties must be reconciled with their syntactic and semantic characteristics. I provide evidence that Romance clitics share many properties with affixes, extending thus a line of research which I have initiated in Monachesi (1996a) and which I have developed further in Monachesi (1999a) (cf. also Miller 1992a; Miller and Sag 1997). While in these works I have focused on Italian clitics, in this volume I show that the motivations presented there can be extended to the other Romance clitics. Furthermore, I spell out in more detail the role that clitics play in achieving a better understanding of the grammatical interfaces. Clitics are not considered lexical items which are located in a speciﬁc position by the rules of syntax, but featural information which is provided in the lexicon and employed in morphophonology for the realization of the cliticized verb form. I do not claim that this analysis should be extended to cliticization in Germanic and Slavic languages, where it can be argued that clitics have a different status. Among the Slavic languages, however, Bulgarian and Macedonian represent an exception since they do not have second position clitics but verbal clitics patterning thus like the Romance languages, as the following example from Macedonian shows: (4) Mi

go

dade Petko. [mac] Petko ‘Petko gave it to me.’ CL.ACC CL.DAT gave

On the other hand, among the Romance languages, Romanian is rather exceptional in that it exhibits a complex clitic system which is similar to that of South Slavic languages such as Bulgarian and Macedonian. It includes not only pronominal clitics, as in the other Romance languages, but arguably also auxiliaries (fi), intensiﬁers (mai), complementizers (s˘a), and negation (nu), which all cluster together: mai fi v˘azut. [ro] (5) s˘a nu o that not CL.ACC again be seen ‘That I should not have seen her again.’ These elements have all been considered clitics in the literature (cf. DobrovieSorin 1994); however, I show that even though this term can be adopted as a general label, the elements in (5) do not constitute a uniform class. Instead of assuming the existence of a special class clitics, the elements of which exhibit different behaviour, I decompose their properties and I assimilate them to those of other well established categories. In particular, I claim

4

Introduction

that object pronouns and intensiﬁers exhibit affix-like properties. On the other hand, auxiliaries, complementizers, and negation share word-like behaviour. Under this proposal, the linearization properties of the clitics in the cluster in (5) can be derived rather than stipulated by means of a template or by an appropriate grid of functional projections. The order of the elements in the verbal complex is not accidental, but is the expected one given the different status of the clitics and the general architecture of the grammar. In particular, I assume that words enter syntax fully inﬂected. These data are particularly revealing since they clearly show that an appropriate division of labour between morphology, phonology, and syntax can deal with the differences and similarities exhibited by these elements. Furthermore, it is through the interaction of these different modules that it is possible to account for the linearization properties attested in the cluster in (5). Ultimately, characterizing these elements as clitics, afﬁxes, or words will always remain a controversial matter and probably a terminological one. The crucial issue which needs to be addressed, however, is how these labels translate in an appropriate analysis and, in particular, which module of the grammar is responsible for their analysis. The study of cliticization shows that Romanian differs from the other Romance languages in several respects and patterns more like Bulgarian and Macedonian which are also exceptional within the Slavic languages. This behaviour can be observed not only in the existence of a complex clitic system, but also in the presence of clitic doubling, in the use of ﬁnite complementation (i.e. subjunctives) instead of inﬁnitives, and in the properties of tense auxiliaries. These are features that characterize Romanian as a member of the Balkan Sprachbund and that differentiate it from the other Romance languages. In this volume, I discuss briefly clitic doubling and finite complementation in Romanian but I focus especially on tense auxiliaries in order to point out the similarities and differences with respect to the other Romance languages as well as with respect to the Balkan ones. In Romanian, nothing can intervene between the tense auxiliary and the lexical verb. As can be seen in the examples below, adverbs (6a), quantifiers (6b), subjects (6c), and complements (6d) cannot occur in this position: (6) a. *Am adesea v˘azut filme bune. [ro] have often seen films good ‘I have often seen good films.’ b. *Elevii t˘ai au tot¸i citit un poem de Verlaine. students yours have all read a poem by Verlaine ‘Your students have all read a poem by Verlaine.’

Introduction

5

c. *A mama f a˘ cut o pr˘ajitur˘a. has mum made a cake ‘Mum has made a cake.’ d. *Am filme bune v˘azut. have films good seen ‘I have seen good films.’ The situation is different if the other Romance languages are taken into consideration since there is variation with respect to which elements can occur between the two verbs. On the other hand, similarities are attested with South Slavic languages, such as Bulgarian and Macedonian, which exhibit clitic auxiliaries that pattern like the Romanian ones with respect to adjacency. I claim with Abeillé and Godard (2003) that tense auxiliaries represent an instance of complex predicate formation within Romance languages, together with modal, aspectual, and motion verbs as well as attributive, causative, and perceptual verbs. These verbs form a class semantically heterogeneous; however, they all show similar properties of monoclausality. More generally, Romance complex predicates can be recognized by two correlated properties: they subcategorize for a nonﬁnite complement and they trigger clitic climbing. In this construction, the clitic, which is an argument of the lexical verb, surfaces attached to the auxiliary verb: (7) a. Martina l-a citit. [ro] Martina CL.ACC has read ‘Martina has read it.’ letto. [it] b. Martina l’ha Martina CL.ACC has read ‘Martina has read it.’ lu. [fr] c. Martine l’a Martina CL.ACC has read ‘Martina has read it.’ a leído. [sp] d. Martina lo Martina CL.ACC has read ‘Martina has read it.’ As can be seen from the examples above, this is a property which Romanian auxiliary verbs share with their Romance counterparts and which they do not have in common with South Slavic languages, as can be seen in the sentences below in which the pronominal clitics follow the tense auxiliary (Franks and King 2000; Legendre 1998; 2000a):

6

Introduction

(8) a. Az sum mu go dal. [bg] I AUX CL.DAT CL.ACC give ‘I have given it to him.’ go dal. [mac] b. Sum ti AUX CL.DAT CL.ACC give ‘I have given it to you.’ ih juˇce(r) dala. [sc] c. Ti si joj you AUX CL.DAT CL.ACC yesterday gave ‘You gave her to them yesterday.’ jo. [sl] d. Prinesel sem ji bring AUX CL.DAT CL.ACC ‘I brought it/her to her.’ These data show that complex predicates, similarly to cliticization, are a perfect test case to investigate the organization of the grammar. The variation in adjacency requirements exhibited by Romance tense auxiliaries can be reconciled with the uniform behaviour with respect to monoclausality only through an appropriate interaction between syntax and argument structure. I suggest that tense auxiliaries in Romance languages share their arguments with the nonﬁnite verb through the lexical operation of argument composition (Hinrichs and Nakazawa 1994). The monoclausality aspect of complex predicates is thus accounted for by means of the extended valence of the tense auxiliary, which provides the tool for dealing with clitic climbing and with all the other properties which are driven by argument structure. The argument sharing mechanism, however, doesn’t determine the constituent structure of the complex predicates. This is an area in which variation is attested within Romance auxiliary verbs given that there are differences with respect to the structural properties of these verbs. I have already mentioned that while in Romanian no elements can intervene between the tense auxiliary and the lexical verb on a par with the Balkan tense auxiliaries, this is not the case in other Romance languages in which there is variation with respect to which elements can intervene. I claim that these differences with respect to the adjacency requirements of the two verbs should be dealt with in terms of different syntactic structures and I argue against a phonological treatment of adjacency which would not be able to deal with the variation in a principled way. In this volume, I provide evidence for a prosodic phonology (Nespor and Vogel 1986) view which assumes that prosodic constituency is distinct from syntactic constituency. In order to defend this view, I take as test case another instance of complex predicates, that is modal, aspectual, and motion verbs in Italian (i.e. restructuring verbs (Rizzi 1986)). Two different syntactic structures

Introduction

7

are associated with these verbs which, however, correspond to a single prosodic configuration. Restructuring verbs are thus another example of the lack of isomorphism between syntactic structure and prosodic structure. I claim that syntax exists as an autonomous module of the grammar, which creates the basis for a mapping into prosodic structure, by means of an appropriate algorithm which is based on prosodic principles. Restructuring verbs exhibit certain properties among which is (optional) clitic climbing, which can be viewed as a manifestation of their ability to share all the syntactic arguments of the nonfinite verb with which they combine. The analysis in terms of argument composition proposed for tense auxiliaries in Romance can also be adopted in this case (Monachesi 1993a, 1999a). However, in the same way as tense auxiliaries, the monoclausality of these verbs does not determine their syntactic structure.

1.1. Outline Chapter 2 provides a brief overview of the basic assumptions underlying the formalism adopted to analyse the various phenomena presented in this volume. Furthermore, it discusses the organization of the grammar with its implications for the interfaces. Chapter 3 focuses on morphology and investigates its interaction with phonology and syntax by taking Romance cliticization as a test case. The phonological, morphological, and syntactic properties of Romance cliticization are presented in detail in order to show the complexity of the phenomenon. A comprehensive analysis of cliticization has to take these various dimensions into consideration and it thus requires a deep understanding of the nature of the interfaces between morphology, phonology, and syntax. I propose a lexical analysis which views Romance clitics as morphological elements. They affect the subcategorization requirements of the verb as a result of a lexical operation which provides the necessary information for the spelling out and placement of the clitics. Different approaches to morphology are considered, including realizational and word–syntax approaches; however, the analysis of cliticization proposed is inspired by the realizational view. The role of morphology within a constrained-based framework is discussed, including the issue of whether this formalism contributes to the development of a new morphological theory with its use of lexical multiple-inheritance hierarchies. A comparison is carried out to investigate the position that morphology occupies in other frameworks such as Distributed Morphology (DM) (Halle and Marantz 1993; Harley and Noyer 1999) and Optimality Theory (OT) (McCarthy and Prince 1993b; Prince and Smolensky 1993).

8

Introduction

The interface between syntax and argument structure is the topic of Chapter 4 which considers a case study of a particular type of complex predicate, that is Romance auxiliaries. The focus, however, is mainly on Romanian auxiliaries, whose properties are compared with those of Italian, French, and Portuguese auxiliaries, on the one hand, and with those of Bulgarian and Macedonian auxiliaries, on the other hand. I show that Romanian tense auxiliaries pattern like the Balkan auxiliaries (and unlike the Romance ones) with respect to adjacency requirements. They behave, however, like the Romance auxiliaries (and unlike the Balkan ones) with respect to the distribution of pronominal clitics (i.e. clitic climbing). I suggest that an appropriate division of labour between syntactic structure and argument structure can deal with the idiosyncratic behaviour of Romanian tense auxiliaries, which I argue have the status of simple clitics. Different syntactic structures are proposed to deal with the variation attested within Romance languages with respect to which elements can occur between the tense auxiliary and the nonﬁnite verbs. Furthermore, other Romanian auxiliary(-like) constructions are taken into consideration to extend the typology and to provide additional evidence for the claims made. The role of argument structure is taken into consideration in dealing with clitic climbing and a comparison is carried out with the function that this level of description has in other frameworks such as Lexical Functional Grammar (LFG) (Kaplan and Bresnan 1982; Dalrymple 2001; Bresnan 2001; Falk 2001). I also address the issue of whether syntax exists as an autonomous module of the grammar or whether it acts merely as an interface between form and meaning. In Chapter 5, I defend the former view and I claim that syntax creates the basis for a mapping into prosodic structure, by means of an appropriate algorithm which is based on prosodic principles. As a test case, I take another instance of complex predicates into consideration, that is Italian restructuring verbs, and I investigate the problems that they raise for the interface between phonology and syntax. The constituency tests provide evidence that two different syntactic structures should be associated with restructuring verbs, while at the phonological level the various rules argue in favour of a single conﬁguration. Different algorithms which have been proposed in the literature are discussed to account for this non-isomorphism between syntactic structure and phonological structure. I show that those which are driven by syntactic principles do not make the right predictions in accounting for phonological constituency. This is not the case for the algorithm suggested in Ghini (1993), which is assumed in order to map syntactic structure to prosodic structure in the case of restructuring verbs. The advantage of this algorithm is that phonological constituency is driven mainly by prosodic principles.

2 The Framework In this chapter, I give a brief overview of the basic assumptions behind Headdriven Phrase Structure Grammar (HPSG), which is the framework I have adopted to formalize the analyses presented in this work. More speciﬁc concepts related to the analysis of linguistic phenomena are introduced, as needed, in the course of the book. I refer, however, to Pollard and Sag (1987, 1994) for a description of those aspects of the theory which are not of immediate relevance to this volume. HPSG emphasizes the need for integration of partial information from various linguistic sources in an order-independent way: the grammar is formulated as a declarative system of constraints. Linguistic information is organized into signs, which integrate several levels of description: phonological, syntactic, and semantic. In HPSG, the bulk of linguistic information is encoded in the lexicon, thus leading to a richly developed theory of lexical organization. Lexical information is not simply listed, in a redundant way, in each lexical entry. On the contrary, lexical principles and lexical rules allow cross-cutting generalizations about words to be expressed in a compact and efficient way. HPSG is a uniﬁcation grammar, that is uniﬁcation is the operation which is employed to verify the compatibility of the information coming from different sources (i.e. lexicon, syntax, semantics). Other frameworks which are also based on uniﬁcation are Lexical Functional Grammar (LFG) (Kaplan and Bresnan 1982), Generalized Phrase Structure Grammar (GPSG) (Gazdar et al. 1985) and Tree Adjoining Grammar (TAG) (Joshi et al. 1975). These theories are characterized by certain common properties, such as the fact that they represent an adequate compromise between linguistic expressivity and the possibility of implementation. Furthermore, they are based on logical models which are well studied and for which programming techniques have been developed.

2.1. Basic assumptions In HPSG linguistic objects are modelled by sorted feature structures which encode linguistic information in terms of attributes and their values.1 1 The terms attribute and feature are used interchangeably throughout this volume, as well as the terms sort and type.

10

The Framework

Sorts are ordered in terms of a sort hierarchy. For example, the type sign, which is a fundamental object in HPSG, has two subtypes which are word and phrase: (9)

sign

HH

phrase

word

A given type may thus have more specific subtypes and each one can introduce features that are appropriate for it. For example, the type sign has the features PHON and SYNSEM which are appropriate for it. The values of PHON and SYNSEM are list of phoneme strings and synsem, respectively: (10) sign PHON

list(phon) synsem

SYNSEM

These attributes are also appropriate for the subtypes of sign, which are word and phrase. A type thus inherits the feature declarations of its supertype: any feature which is appropriate for a given type is also appropriate for all of its subtypes. A sign has complex internal structure and it encodes, at least, the following information:2 (11)

⎤ sign ⎥ ⎢ ⎥ ⎢PHON list phon ⎥ ⎢ ⎥ ⎢ ⎤ ⎡ ⎥ ⎢ synsem ⎥ ⎢ ⎡ ⎤⎥⎥ ⎢ ⎢ ⎢ ⎥⎥ ⎢ local ⎢ ⎥⎥ ⎢ ⎡ ⎤ ⎢ ⎥ ⎥ ⎢ ⎢ ⎢ ⎥⎥ ⎢ ⎥⎥ ⎢ category ⎢ ⎥ ⎥ ⎢ ⎢ ⎢ ⎥⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎢ ⎥ HEAD head ⎢ ⎥ ⎥ ⎢ ⎢ ⎢ ⎢ ⎥⎥ ⎡ ⎢ ⎤⎥ ⎥⎥ ⎢ ⎢ ⎥ ⎢ ⎥ ⎥ ⎢ ⎢ ⎢ ⎢ ⎥⎥ SUBJ list synsem ⎥⎥ ⎢ ⎥⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎢SYNSEM ⎢ ⎢ ⎥⎥⎥⎥⎥ ⎢ ⎢ ⎢ ⎢LOCAL ⎢ CATEGORY ⎢ ⎥⎥⎥⎥ ⎢ ⎢ ⎥ ⎢ ⎢ list synsem ⎥ ⎢VALENCE ⎢SPR ⎢ ⎥⎥⎥⎥ ⎢ ⎥⎥ ⎢ ⎢ ⎥ ⎢ ⎥ ⎣ ⎦ ⎥ ⎢ ⎢ ⎢ ⎥⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ synsem COMPS list ⎢ ⎥ ⎢ ⎥ ⎥ ⎢ ⎢ ⎢ ⎥⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎣ ⎦ ⎢ ⎥ ⎥ ⎢ ⎢ ⎢ ⎥⎥ ⎢ ⎥⎥ ⎢ ARG-ST list synsem ⎢ ⎥ ⎥ ⎢ ⎢ ⎣ ⎦⎥ ⎢ ⎦⎥ ⎣ ⎦ ⎣ CONTENT psoa ∨ nom-obj ∨ quant ⎡

2 Only those features which are relevant for the analyses developed in this volume are mentioned. I refer to Pollard and Sag (1994) for a complete description.

The Framework

11

Sorted feature structures are described in terms of attribute value matrices (AVMs), such as the one above. In order to interpret the matrix in (11), one should bear in mind that:

• the items in italics in the upper left corner represent the type (or supertype) of the object described; • the items in small capitals are attribute names and what follows them on the right hand side is the type (or supertype) of their value; • list(σ) stands for a list of objects of type σ. List descriptions are abbreviated by angle brackets (e.g. σ1 , σ2 ); • set(δ) stands for a set of objects of type δ. Unlike list descriptions, the order of elements in a set description is not significant. Set descriptions are abbreviated by curly brackets (e.g. {δ1 , δ2 }); • the notation ∨ stands for disjunction.3 As already mentioned, a sign has at least the two attributes PHON and SYNSEM. The former encodes phonological information, while the latter includes both syntactic and semantic information. In Pollard and Sag (1994) it is assumed that the value of PHON is a list of phoneme strings. However, in Chapter 3, I suggest that it should have more structure in order to distinguish segmental and prosodic information. Furthermore, I propose certain modifications to the architecture of the sign to enable the encoding of morphological information. From the description presented in (11), it is evident that information from different linguistic levels is encoded in one structured object. The SYNSEM value is an object of type synsem and it includes the attribute LOCAL, with value local, which specifies the inherent syntactic properties of a sign. Among its appropriate attributes, the features CONTENT and CATEGORY can be distinguished. The CONTENT value constitutes the sign’s contribution to semantic interpretation. It is within this attribute that agreement information is handled and also information regarding semantic roles. As for CATEGORY, it encodes information related to the syntactic category of a word (or phrase) and also the grammatical arguments it requires. The CATEGORY value is an object of type category and it has the attributes HEAD, VALENCE, and ARG-ST. The attribute ARG-ST specifies the arguments which are selected by a lexical head. Furthermore, it encodes information about the obliqueness hierarchy. It thus plays a crucial role for binding theory.4 It should be mentioned that this feature was not introduced in Pollard and Sag (1994), but is the result of more recent developments within the theory (Manning and Sag 1998). In Chapter 4, I dedicate an extensive discussion to the introduction

3

Negation can also be expressed and the notation ‘¬’ is used. I refer to chapter 6 of Pollard and Sag (1994) for an exhaustive presentation of HPSG binding theory. 4

12

The Framework

of this feature and its current status within HPSG and I compare it with similar notions in other theories. As for the feature HEAD, it encodes information about grammatical properties such as part of speech, verbal form, or case. The type head has the following subtypes: head

(12)

H HH

HH

substantive

P @PPP @ P

noun

verb

adj

prep

functional HH det marker

Some of these types might have certain attributes which are appropriate for them, as in the case of verb: (13) verb vform boolean

VFORM AUX

The feature VFORM encodes verb inﬂections. Some of its possible values are fin (ﬁnite or tensed forms) and psp (past participle), while AUX is used to distinguish auxiliary verbs. As for nouns, they have the feature CASE as appropriate for them.

2.2. The principles Generalizations within HPSG can be expressed by means of principles. For example, in the case of head features, their transmission from the head to the mother node is regulated by the Head Feature Principle, which says: (14) A head daughter’s HEAD value is identical to that of its mother. This principle is very similar to the Head Feature Convention of Generalized Phrase Structure Grammar (Gazdar et al. 1985). It guarantees that headed phrases are projections of their heads. The subcategorization requirements of a given word are regulated by means of the Valence Principle and the different valence features (i.e. SUBJ, SPR, COMPS). SUBJ is an attribute for the selection of the subject, SPR for the selection of specifiers, while COMPS encodes the complements selected. For example, a verb like the Italian ama ‘loves’ subcategorizes for a subject and a direct object:

The Framework (15)

13

⎤ word ⎥ ⎢ ⎥ ⎢PHON ama ⎢ ⎡ ⎤⎥ ⎤ ⎡ ⎥ ⎢ ⎥ ⎢ verb ⎥ ⎢ ⎢ ⎥ ⎥ ⎢ ⎢ ⎢HEAD ⎣VFORM fin⎦ ⎥⎥ ⎥ ⎢ ⎢ ⎥ ⎢ ⎢ ⎥⎥ ⎥ ⎢ AUX − ⎢ ⎥ ⎢ ⎢ ⎡ ⎥⎥ ⎥ ⎢

⎤ ⎢ ⎥ ⎢

⎥⎥ ⎥ ⎢SS|LOC|CAT ⎢ SUBJ 3 NP nom ⎢ ⎥ ⎢ 1 3sg ⎥⎥⎥ ⎢ ⎢ ⎥ ⎢ ⎢ ⎥ ⎢ ⎥ ⎢

⎦⎥⎥ ⎢VAL ⎣ ⎥ ⎢ ⎢ ⎥ COMPS 4 NP acc 2 ⎢ ⎢ ⎥⎥ ⎥ ⎢ ⎣ ⎦ ⎥ ⎢ ⎥ ⎢ 3 4 ARG-ST , ⎥ ⎢ ⎥ ⎢ ⎡ ⎤ ⎥ ⎢ ⎥ ⎢ psoa ⎥ ⎢ ⎢ ⎥ ⎥ ⎢ ⎢ ⎥ RELN love ⎥ ⎢CONTENT ⎢ ⎥ ⎥ ⎢ ⎢ ⎥ 1 ARG1 ⎥ ⎢ ⎣ ⎦ ⎦ ⎣ ⎡

ARG2

2

The description above states that ama ‘loves’, is a finite verb and it is not an auxiliary: this information is encoded within the HEAD feature. As already mentioned, it subcategorizes for a (third person singular) subject which is the value of the SUBJ feature as well as for an object encoded as the value of the feature COMPS. The CONTENT feature is the locus for role assignment: the verb ama has two arguments, the first of which is linked to the subject while the second is associated with the object. The linking is formalized by means of structure sharing, that is a token identity of values which is indicated by multiple occurrences of boxed numbers called ‘tags’, for example 2 . In the AVM above, the notation NP [nom] 3 [3sg] abbreviates a saturated nominal sign whose case is [nom] and whose index is 1 , where 1 is specified as [Per 3] and [Num sg]. Notice that the feature SPR is omitted here (and in the entries in the rest of the book) since in the case of verbs its value is always the empty list (elist). Lexical heads select for the synsem objects of their complements, subjects or specifiers. Given the particular information which is included in objects of this type, it follows that category selection, head agreement, case agreement, and role assignment are all strictly local. An exception to this might be seen in the argument composition mechanism, introduced in Chapter 4, according to which an auxiliary verb inherits the arguments of the lexical verb.

14

The Framework

Furthermore, as can be seen in the description above, role assignment occurs within the lexical entry of the verb. As already mentioned, it is obtained by the structure sharing between the indices of the NPs and those of the roles in the CONTENT of the verb. It should also be noted that it is not the complements themselves which are assigned roles, but their indices. Agreement is accounted for by having the verb select a certain kind of subject, in this case a third person singular subject.5 As for case assignment, it also occurs within the lexical entry. Recall that subcategorization is regulated by the Valence Principle which says: (16) In a headed phrase, for each valence feature F (COMPS, SUBJ, or SPR) the F value of the head daughter is the concatenation of the phrase’s F value with the list of SYNSEM values of the F-DTRS value. The principle guarantees that the valence on the mother is that of the head with any complements taken off the valence list. It is thus similar to the cancellation of arguments in Categorial Grammar (CG) (Morrill 1994; Moortgat 1997; Steedman 1996). Consider an Italian sentence like the one below: (17) Martina ama Vito. [it] Martina loves Vito ‘Martina loves Vito.’ The following is an example of how the two principles discussed above, the Head Feature Principle and the Valence Principle, operate. Recall that the boxed numbers are used to indicate identity of values:6 As already discussed, the verb ama ‘loves’ subcategorizes for a subject, Martina, and for the object, Vito. The SYNSEM value of the object is shared with that of the single element in the COMPS list of the verb and similarly for the subject. Once the object is realized, the Valence Principle ensures that it is removed from the COMPS list; therefore the VP has an empty COMPS list while the SUBJ list still contains the information related to the subject since this element has not yet been realized. Once the subject is realized, the result is an S, that is a verbal sign with an empty COMPS list and an empty SUBJ list. 5

HPSG distinguishes different types of agreement: syntactic concord, anaphoric agreement, and pragmatic agreement. For more details, cf. ch. 2 of Pollard and Sag (1994). 6 It should be noted that trees are often used in this work since they represent a familiar way to display phrasal structure; however they are not part of the theory. Furthermore, in the tree in (18) (and in trees in the rest of the book) the AVMs are highly simplified since only the relevant attributes are mentioned.

The Framework

(18)

15

1

HEAD S SUBJ COMPS

2

HEAD

Martina

VP SUBJ

1 2

COMPS

HEAD

1 verb fin

SUBJ

2 NP[3sg]

COMPS

3 NP

3 Vito

ama It should also be noted that the HEAD value of the verb indicated by the tag 1 is shared with that of the VP, and similarly the head value of the VP is identical to that of the S. This is the effect of the Head Feature Principle which ensures that head features are transmitted from the head daughter to the mother node.

2.3. Phrases and constituent structure The typology of phrases has undergone some revisions since Pollard and Sag (1994). In this section, I assume the division proposed in Sag (1997), which eliminates the DTRS attribute suggested in Pollard and Sag (1994) and assumes instead the two attributes HD-DTR and NONHD-DTR. In HPSG, different kinds of phrases can be distinguished: (19)

phrase

X HX X HX HHXXXX XXX HH X hd-comp-ph hd-subj-ph hd-adj-ph hd-fill-ph …

Phrases are classified as head complement phrase, head subject phrase, head adjunct phrase, head filler phrase, head specifier phrase, and head marker phrase. In addition to inherited attributes, the type phrase has HEAD-DAUGHTER and

16

The Framework

NONHEAD-DAUGHTERS

(20)

HD-DTR

as appropriate features:7

sign

NONHD-DTRS

list(phrases)

The feature structure description of the tree in (18) given below shows the structure of Martina ama Vito ‘Martina loves Vito’, in which the different phrases are distinguished. The labels S, VP, and NP should be seen as abbreviations: ⎡ ⎤ (21) hd-subj-ph ⎢ ⎥ ⎢PHON Martina ama Vito ⎥ ⎢ ⎥ ⎢SYNSEM S ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ PHON Martina ⎢NONHD-DTRS ⎥ ⎢ ⎥ SYNSEM NP ⎢ ⎥ ⎢ ⎥ ⎡ ⎤ ⎢ ⎥ ⎢ ⎥ hd-comp-ph ⎢ ⎢ ⎥⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎢PHON mangia spaghetti ⎥⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎢SYNSEM VP ⎥⎥ ⎢ ⎥

⎢ ⎥ ⎢HD-DTR ⎢ ⎥⎥ ⎢ ⎥ HD-DTR PHON mangia ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎢ PHON spaghetti ⎥⎥ ⎢ ⎣ ⎦ NONHD-DTR ⎥ ⎣ ⎦ SYNSEM NP The structure described above is licensed by immediate dominance schemata, which play a role similar to the X-bar schemata in GOVERNMENT BINDING theory (Chomsky 1981) and are responsible for grouping constituents together. In particular, two schemata play a role: SCHEMA1 licenses phrases consisting of a subject daughter and a phrasal head daughter, while SCHEMA2 licenses phrases where there is a lexical head daughter and any number of complement daughters. It should be noticed that in Pollard and Sag (1994) six schemata are introduced. Besides the two already mentioned, there are the head-subjectcomplement schema, the head-marker schema, the head-adjunct schema and the filler-head schema. The schemata do not make reference to the surface order in which constituents are realized. HPSG follows the proposal advanced in GPSG that matters of word order should be taken into account by means of Linear Precedence (LP) constraints (Pollard and Sag 1987). In particular, they suggest speciﬁc constraints that play a role in the linearization of English, but they can also be 7 Since these features are assumed, the formulation of the Valence Principle should be changed slightly. I refer to Miller and Sag (1997) for a possible reformulation.

The Framework

17

adopted in other languages. In the case of the Italian example in (17), the order ama Vito is due to an LP constraint like the following, which accounts for the fact that the verb precedes its complement: (22) Linear Precedence Constraint 1 (LP1) In a phrase, a lexical head has to precede any of its sisters. On the other hand, if more than one complement is present, they should be ordered according to the obliqueness hierarchy expressed in ARG-ST by the following constraint: (23) Linear Precedence Constraint 2 (LP2) Less oblique complements must precede more oblique ones. In the course of this work, additional LP constraints will be formulated in order to deal with the ordering of particular constituents, as for example in Chapter 4 where Romanian auxiliaries are taken into consideration.8

2.4. Lexical generalizations In HPSG, most of the linguistic information is encoded within lexical entries. The lexicon provides the basis for expressing generalizations about phenomena which include not only inflection and derivation, but also constructions which were traditionally considered outside its scope, such as passive and control. Therefore, within HPSG, much research has been devoted to the crucial issue of how to structure linguistic information in the lexicon. Lexical representations must be able to capture not only common properties, but also exceptional ones without undue redundancy. According to Flickinger (1987), two kinds of generalizations can be distinguished within the lexicon: vertical and horizontal. Vertical generalizations indicate that certain properties are common to the words of a given class, while horizontal generalizations express a systematic relationship holding between two classes of words. It has often been assumed that vertical generalizations can be captured by means of macros or templates that allow the abbreviation of lexical speciﬁcations. However, Meurers (1997) points out that these devices do not actually capture theoretical generalizations and have no formal status within the theory. He suggests instead that the same implicational constraints which are used to express the grammatical principles in HPSG can also be employed to express 8 Within HPSG, some research has been devoted to the problem of linearization. Cf. Reape (1994), Kathol (1995), and Richter and Sailer (1995) for German.

18

The Framework

lexical generalizations. The following is a simple example of a constraint which says that ﬁnite verbs must have a nominative subject:

(24)

⎡

word ⎢ ⎢ ⎣SS|L|C|H verb

VFORM

⎤

⎥ ⎥→ SS|L|C|VAL|SUBJ NP nom ⎦

fin

Implicational constraints are interpreted as stating that every object described by the antecedent will also have to satisfy the consequent in order to be grammatical. In the example above, the antecedent is a complex one, but it is also possible to have constraints in which the antecedent is a simple type. In HPSG, constraints are interpreted as well-formedness conditions on linguistic objects. They state which linguistic object is grammatical and which one is not. Constraint violation determines ungrammaticality: there is thus a crucial difference from Optimality Theory in which constraints can be violated. More speciﬁcally, in OT, the best output candidate is the one with the fewest violations of highly ranked constraints. Horizontal generalizations can be expressed by means of lexical rules which are introduced in Pollard and Sag (1987) to account for inﬂection. In this use, the input of the rule makes reference to a certain base form while the output produces the inﬂected form. Pollard and Sag also employ lexical rules to account for phenomena which were traditionally handled by syntactic mechanisms, such as passives and unbounded dependency constructions. In current research within HPSG, lexical rules are used to account for different linguistic phenomena in several languages. However, a certain line of research is exploring the possibility of doing away with lexical rules in accounting for inﬂection and derivation, as in the work of Erjavec (1994), Kathol (1994), Kim (1994), Krieger and Nerbonne (1992), and Riehemann (1993). Another attempt in this direction, but with respect to the analysis of relative clauses, is Sag (1997). There are two possible ways in which lexical rules can be interpreted: either as meta-descriptions relating lexical entries (Calcagno 1995) or as descriptions relating word objects (Meurers 1995; Meurers and Minen 1997). In this work I assume the latter formalization. Under this formalization, a lexical rule establishes a relation between two sets of words:9

9 It should be noted that → is used to relate input and output descriptions in lexical rules, while → is used for implicational constraints throughout the book.

The Framework (25) Passive Lexical Rule (PALR) ⎤ ⎡ word ⎥ ⎢ ⎥ ⎢ ⎥ ⎢HEAD verb ⎥ ⎢ VFORM psp ⎥ ⎢ ⎢ ⎡ ⎤⎥ ⎥ ⎢ ⎥ ⎢ ⎢ ⎣SUBJ NP 1 ⎦⎥ ⎦ ⎣VAL COMPS NP 2 ⊕ 3 ⎡

verb

19

→

⎤

⎥ ⎢HEAD ⎥ ⎢ VFORM pas ⎥ ⎢ ⎢ ⎡ ⎤⎥ ⎥ ⎢ SUBJ NP 2 ⎥ ⎢ ⎢ ⎢ ⎥ ⎦⎥ ⎥ ⎢VAL ⎣ ⎦ ⎣ COMPS 3 ⊕ PP da 1

The rule is a modified version of that proposed by Pollard and Sag (1987) for the analysis of the English passive and it accounts for passive constructions in Italian. It relates words, in particular past participles, with passive verbs. A coindexation is established between the object in the input and the subject in the output, as well as between the subject in the input and the optional by-phrase. It should be noticed that, as traditionally assumed within HPSG, all specifications of the input are preserved in the rule output except for those which are explicitly mentioned on the right hand side. To summarize: two methods have been presented to express lexical generalizations within HPSG; they are lexical rules and implicational constraints. Both devices are employed in the analyses developed in the following chapters. It is not always obvious how to determine which generalizations are better expressed by means of implicational constraints and which ones by lexical rules. Generally, it seems that the latter are better suited to express a systematic relationship between two classes of words while the former are more appropriate to indicate that certain properties are common to the words of a given class. The crucial issue then is whether it is always possible to reduce one to the other with no reduction in expressivity or in elegance: this is a topic to which more research should be devoted within the framework. As already mentioned, implicational constraints and lexical rules are employed to provide an analysis of the various phenomena addressed in this volume. A relevant issue in this respect is whether a theory can be provided of the possible rules which can be formulated. It is often argued that the rules introduced in HPSG are descriptively adequate but they do not provide enough

20

The Framework

explanatory power. I believe that this is a feature shared by the various grammatical frameworks and that it is not typical of HPSG. The introduction of a new constraint in Optimality Theory or the postulation of a new functional projection or a new feature in the Chomskyan framework can also be seen as very stipulative moves. In particular, a distinction should be made between the formalism adopted and the grammatical theory related to it. The formalisms available are all very powerful: it is the task of grammatical theory to constrain the formalism by identifying the possible rules of the language and this is an empirical question. Deﬁning the possible rules of the language is thus a collective enterprise within HPSG as in any other framework: this volume should be considered as a step forward in identifying the basic rules in the area of cliticization and complex predicates.

2.5. The organization of the grammar and the implications for the interfaces: A comparison with other theories The crucial property of the HPSG sign is that it represents a structured complex of phonological, syntactic, semantic, discourse, and phrase-structural information. A distinction is posited by Ladusaw (1988) between level and stratum (Perlmutter 1983), to differentiate among ways in which a theory of grammar can employ multiple representations of structure. Ladusaw suggests that the term level should be reserved for the description of linguistic structure by multiple representations, each of which is drawn from a different vocabulary. On the other hand, the term stratum should be employed for the description of a particular type of structure in terms of multiple representations drawn from the same vocabulary. Under this view, the HPSG model could be characterized as multilevel but monostratal (Pollard and Sag, 1994). It is multilevel since more than one level of representation is posited, for example phonological information is encoded by means of the PHON feature, while syntactic information is expressed by means of the CATEGORY feature. However, it is monostratal because there is only one representation for each level. As shown by the lexical entry for ama ‘loves’ in (15), each word carries information about the various levels of description which allows for form and meaning to be built up in parallel. For example, phonological structure can be built up in parallel with syntactic structure (Asudeh and Mikkelsen 2000) as in recent proposals by Steedman (2000), who claims that information structure and surface structure coincide. In his view, intonation coincides with surface structure in the sense that all intonational boundaries coincide with syntactic boundaries.

The Framework

21

The approach that builds several levels of description in parallel differs from the architecture advocated in Optimality Theory, where each level of grammatical description is built up independently and then related through constraint interaction. It seems, however, that in more recent developments of the HPSG theory, such as Sag (1997) and subsequent work, there is a tendency to move towards a similar architecture of the grammar. The cross-classification framework suggested in Sag (1997) offers the possibility of deﬁning constraints independently for each grammatical level, for example, prosodic constraints can be deﬁned independently of syntactic constructions (Klein 2000) yet they can be related by means of multiple inheritance in a way that is indeed reminiscent of Optimality Theory. As discussed in Legendre (2000a), the OT candidates that undergo optimization are global structures given that these representations combine all traditional levels of representations (syntactic, morphological, and prosodic) into one single level. Furthermore, constraints may refer to any sublevel of representation: some refer to syntax, others to morphology, yet others to prosody. Legendre shows, however, that constraint interaction is modular and that ranking consists of blocks of constraints referring to the same sublevel of representation. Constraints belonging to different modules of the grammar may not intermix. However, the architecture of standard OT permits reference to any level of representation since alignment constraints may relate structures from one module to another directly: these are thus interface conditions. This is also the case in HPSG, in which constraints concerning various levels of description can all interact to produce the desired output. It is thus possible to deal with mismatches between levels of representation by simply using the standard combinatorics. This might be a desirable result given that a whole body of research has provided ample evidence about the existence of such mismatches and this volume adds to the existing literature. The HPSG model shares with LFG the property that it is multilevel according to the deﬁnition in Ladusaw (1988). In LFG functional structure, which contains a mixture of syntactic and semantic information, is distinct from both constituent structure and semantic representation. Ladusaw argues that (at least) two levels should be distinguished, that is c-structure and f-structure, since well-formedness is deﬁned by a series of conditions some of which are stated at one level and some at the other. There are, however, differences with respect to the organization of the grammar between HPSG and LFG. LFG is based on a model in which a given level feeds another, for example syntactic trees are mapped into prosodic representations in Butt and King (1998). In particular, they posit a p(honological)-structure

22

The Framework

which feeds into a further phonological component in the same way that s(emantic)-structure feeds into a more elaborate semantics. The p-structure thus represents the interface. They propose, however, a mutually constraining model in which principles govern not only the wellformedness of the syntax and of the phonology, but also of the possible mappings between them. There are thus similarities with the traditional T-model conception of the grammar which is based on four levels of representation: D-structure, Phonetic Form, Logical Form, and S-structure which connects the three previously mentioned levels. This model is based on the assumption that the grammar consists of subcomponent parts governed by their own rules and principles. Interactions among these autonomous subcomponents are restricted to the interfaces, given that the output of one subcomponent serves as the input for the next. This model has undergone revisions under Chomsky’s minimalist programme, which has abolished D-structure. As noted by Morrill (1994), this change yields the Montague monostratal architecture (in which a derivational structure can be found instead of S-structure), which is the basis of Categorial Grammar. However, this model has been simpliﬁed in the logical tradition of Categorial Grammar (CG), that is there is no syntactic representation deﬁning the theory of formation since this is deﬁned model-theoretically and not syntactically. In CG, the language model is projected from the lexicon, which consists of category formulas, that is sets of pairings of prosodic and semantic objects. Under this view, the grammar is simply the lexicon: this pure lexicalism has been adopted in various forms by virtually all the theories discussed in this section. The categorial model shares similarities with the further simpliﬁed model resulting from Chomsky’s minimalist programme, which has abolished S-structure. All properties are thus projected from the lexicon and Move-α is regulated by universal principles of computation as pointed out by Morrill (1994). It is this model which has been argued for in a recent work on the syntax– phonology interface by Boˇskovi´c (2001). He defends a derivational model in which syntax feeds phonology, with the operation Move applying only in the syntax and not in the phonological component. On the other hand, this model has been questioned by the literature on South Slavic cliticization which has postulated PF movement. Furthermore, it has argued for the necessity of look-ahead from the syntax to the phonology as well as for a copresence, bi-directional model in which phonology can feed information to the syntax. As noticed by Grohmann (2001), PF movement would be an unwelcome conceptual result, as one would have to posit parallel interfaces of the syntax,

The Framework

23

interpretation, and phonology. They would be hard to integrate even in a minimalist version of the T-model, since it does not make available a bidirectional connection between the syntax and the phonology, but assumes that syntax feeds phonology. Boˇskovi´c shows, however, that the relevant facts about South Slavic cliticization do not support non-standard claims about the syntax–phonology interface, as discussed in more detail in Section 3.7. He argues that there is no empirical basis for the possibility of movement in phonology, there is no need for a look-ahead from the syntax to the phonology, and there is no evidence for a co-presence, bi-directional model of the grammar. However, a disadvantage of the model proposed in Boˇskovi´c (2001) is that syntax overgenerates and the inappropriate structures are ﬁltered out in phonology. In a model such as HPSG, in which constraints concerning various levels of description can all interact to produce the desired result, overgeneration would be avoided. Syntactic information and phonological information would both interact to produce the appropriate representation, which seems to be a desirable result. This is a conclusion also reached by Legendre (2000a) with respect to OT. The architecture of this framework permits reference to any level of representation, that is alignment constraints may directly relate structures across different modules. They act in this way as interface conditions, which find a natural implementation in standard OT because candidate structures are global structures and optimization proceeds in parallel rather than serially. She also argues against the prevailing approach in which the phonology repairs what the syntax produces since, from the perspective of the grammar, there is in principle no reason why phonology should undo what the syntax does. Even though the T-model has been the leading one in grammatical theory, I show in this book that the organization of the grammar in a sign-based framework such as HPSG is better suited for a formal treatment of interface issues. This is due to the fact that it simultaneously captures generalizations in the various levels of descriptions as well as interactions between them. Research on grammatical interfaces, however, has not always received the necessary attention within HPSG and it has not always dealt with the various interfaces in equal terms. The interface between morphology and syntax has been the topic of various studies, as discussed at length in Chapter 3. This is also the case for the interface between syntax and argument structure, as shown in Chapter 4. The interface between phonology and syntax is still not well understood; I believe, however, that this is a more general problem that goes beyond the HPSG framework, as will be clear from the presentation in Chapter 5.

24

The Framework

This book will take pains to show that the HPSG organization of the grammar can deal successfully with the interfaces. In order to achieve this goal, I take into consideration phenomena such as cliticization and complex predicates, which have always been a challenge for any grammatical theory. They therefore appear to be ideal test cases to evaluate the various models of grammar.

3 Morphology and its Interfaces: The Case of Romance Clitics 3.1. Introduction The aim of this chapter is to analyse the role that morphology plays within grammatical theory and in particular its interaction with phonology and syntax. The focus is mainly on Romance clitics, which are a perfect test case as their morphophonological properties must be reconciled with their syntactic characteristics, that is the fact that pronominal clitics satisfy the subcategorization requirements of verbs. Clitics represent a topic of crucial importance for the understanding of the interaction of the different modules of the grammar and the way they are organized. It is from this perspective that they are addressed. In particular, the analysis of cliticization defended here is one in which clitics are assimilated to affixes. A morphological treatment within HPSG is proposed that is inspired by a realizational approach to morphology. The question of how morphology interacts with phonology and syntax is an area still under debate within HPSG. Both realizational and word–syntax approaches to morphology have been defended: an overview of both types of analyses is provided. Special attention is dedicated to the issue of whether HPSG merely provides a framework to formalize specific morphological analyses or if it contributes to the development of morphological theory. In particular, the use of lexical multiple-inheritance hierarchies is discussed. They provide structure to the lexicon by allowing for an explicit representation of linguistically relevant subclasses of words. This view implies that the best way to classify words is in terms of hierarchies. It is shown that HPSG is particularly well suited for expressing these generalizations because the cross-classification of the relations among linguistic types is one of the main concerns of the framework. However, it is the HPSG organization of the grammar that mostly inﬂuences the way morphology interfaces with the other components, as will become evident from the analysis of Romance clitics.

26

Morphology and its Interfaces

A comparison with other frameworks, including Distributed Morphology and Optimality Theory, is carried out in order to show that HPSG allows for a more explicit organization of the interfaces, which in turn leads to a better understanding of the phenomena analysed.

3.2. A constrained-based view of morphology The question of how morphology interacts with syntax and phonology is an area still under debate within HPSG. Different proposals have been made on how morphological information should be formalized. In particular, two of the leading morphological approaches, that is the word–syntax (Selkirk 1982) and the realizational approach (Matthews 1972; Anderson 1992; Stump 1992, 2001), have been explored and considered from an HPSG perspective. In the word–syntax approach, morphemes have the status of signs and they combine with the stem by means of morphological principles that are similar to those that operate in syntax. An X-bar model is often assumed; a complex word has a head, which has the same category as the mother. Under this view, a complex German word such as les-bar ‘read-able’ has the following representation: (26) [[les] V [bar] Aaf ] A In the first part, the verb lesen ‘to read’ can be distinguished. It combines by means of a phrase structure rule with the derivational suffix -bar resulting in an adjective. In this case, the category is determined by the suffix. In contrast, under the realizational approach to morphology, morphemes do not exist as signs. They simply represent the phonological realization of certain morphosyntactic properties of the host they combine with. The same example with the morpheme -bar receives the following representation: (27) [[les] V bar] A In this case, the suffix is introduced by a lexical redundancy rule and it is not itself marked for category. In the following sections, I discuss these approaches in more detail and, in particular, I focus on how they have been formalized within HPSG. 3.2.1. The word–syntax approach

There have been several formalizations of word–syntax approaches within HPSG, from the seminal work of Krieger and Nerbonne (1992) and Krieger (1994) to the most recent work of Reinhard (2001). In the following section,

Morphology and its Interfaces

27

I take the work of Krieger and Nerbonne (1992) and of Krieger (1994) into consideration, as representative of this line of research within HPSG. 3.2.1.1. Krieger and Nerbonne (1992) and Krieger (1994) Krieger and Nerbonne (1992) and Krieger (1994) adopt a phrase structure approach to morphology. They assume that derivational afﬁxes have category marked lexical entries. Objects of type afﬁx have the following attributes associated with them: (28)

⎡

⎤ affix ⎢ ⎥ ⎢MORPH affix-morphology⎥ ⎢ ⎥ ⎢SYN affix-syntax ⎥ ⎣ ⎦ SEM affix-semantics

In other words, the attributes present in the description above allow for a speciﬁcation of the morphological, syntactic, and semantic properties of afﬁxes, as will become clear by the speciﬁc lexical entry for the afﬁx -bar introduced in (34). In particular, the attribute MORPH has an object of type affix-morphology as its value with its appropriate features that allow for a detailed representation of the morphological contribution of afﬁxes. More speciﬁcally, the type affix-morphology has the following structure: ⎤ (29) ⎡affix-morphology ⎢ ⎥ ⎢FORM string ⎥ ⎢ ⎥ ⎢POS left ∨ right ⎥ ⎢ ⎥ ⎢ ⎥ ⎣ACAT prefix ∨ suffix ⎦ SUBCAT part-of-speech

The feature SUBCAT ensures that the afﬁx combines with the appropriate argument it subcategorizes for, ACAT states whether the afﬁx is classiﬁed as a preﬁx or a sufﬁx, and POS determines the position of the afﬁx with respect to the word it combines with.1 Notice that objects of type word also have MORPH as an appropriate attribute, but the value in this case is word-morph. In addition, they have the attribute CORD that reflects the order of the parts of a complex word and the attribute DTRS that determines the constituents of a word:

1 The POS feature seems redundant given that the information about the position could be derived from the type of affix, that is whether it is a prefix or a suffix.

28

Morphology and its Interfaces

(30)

⎡

⎤ word ⎢ ⎥ ⎢MORPH word-morphology ⎥ ⎢ ⎥ ⎢SYN word–syntax ⎥ ⎢ ⎥ ⎢ ⎥ ⎢SEM word-semantics ⎥ ⎢ ⎥ ⎢CORD list(part-of-speech ∨ affix)⎥ ⎣ ⎦ (DTRS affix-word-struct)

The DTRS attribute has affix-word-structure as its value. Objects of this type have the following appropriate attributes: (31) ⎡affix-word-struct ⎤ ⎢ ⎥ ⎣AFFIX affix ⎦ WORD part-of-speech This setup allows for a formalization of the word–syntax approach to morphology. Afﬁxes combine with the relevant stems and are subject to various morphological principles which resemble those playing a role in syntax. The following principles are envisaged: 1. 2. 3. 4. 5. 6.

Morphological Affix-Word Schema; Morphological Constituent Order Principle; Morphological Surface Realization Principle; Morphological Head Feature Principle; Morphological Semantics Principle; Morphological Subcategorization Principle.

The Affix-Word Schema is responsible for licensing complex words and it formally expresses a context-free rewrite rule such as the following, which combines an afﬁx (A) with a word (W) to produce a lexical element: (32)

M [LEX

+] → A, W

Matters of linear order are accounted for by the Constituent Order Principle, which determines in which order words and afﬁxes occur at the surface level. This sequence is then stored under the attribute CORD mentioned above. The Surface Realization Principle deals with morphophonemics: for example phenomena like umlaut. The Head Feature Principle is responsible for deducing the category of a new word from the category of the head daughter, while the Semantics Principle accounts for the fact that the semantics of the mother is equal to the semantics of the head daughter. As for subcategorization, a distinction is made between morphological and syntactic subcategorization. The morphological subcategorization ensures that the afﬁx combines with the appropriate argument. However, the subcategorization information of the new

Morphology and its Interfaces

29

complex word comes directly from the syntactic subcategorization of the afﬁx by means of structure sharing, which is indicated by the presence of the tag 1 : (33)

⎤ complex ⎥ ⎢ ⎥ ⎢ LOC | SUBCAT 1 ⎥ ⎢SYN | ⎡ ⎤⎥ ⎢ ⎡ ⎤ complex ⎥ ⎢ 1 →⎢ ⎢AFFIX ⎣SYN | LOC | SUBCAT ⎥⎥ ⎢ ⎦ DTRS affix-word-struct ⎥⎥ ⎥ ⎢DTRS ⎢ ⎢ ⎢ MORPH | SUBCAT part-of-speech ⎥⎥ ⎣ ⎦⎦ ⎣ ⎡

WORD

This formal apparatus is employed in Krieger (1994) to deal with derivational morphology and, in particular, he adopts it for the analysis of -bar sufﬁxation. Traditionally, it is assumed that the sufﬁx -bar can be attached to the stem of any transitive verb, as in the case of lesen ‘to read’ which becomes lesbar ‘readable’. The following lexical entry is associated with the afﬁx -bar: (34)

⎤ bar-suff ⎢ ⎤⎥ ⎡ ⎥ ⎢ aff-morph ⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢POS right ⎢ ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢FORM bar ⎢ ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ACAT suffix ⎢ ⎤⎥⎥ ⎡ ⎢ ⎢MORPH ⎢ ⎥⎥ bar-v ⎥ ⎢ ⎢ ⎡ ⎤⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎢ subcat-info ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎢ ⎥ ⎥ ⎢ ⎥ ⎢ ⎢SUBCAT ⎥⎥⎥ ⎢ 3 SYN | LOC | SUBCAT ⎢ ⎥ ⎢ ⎢ ⎢OBJ 1 ⎥⎥⎥ ⎢ ⎢ ⎥ ⎢ ⎣ ⎦⎦⎦⎥ ⎥ ⎢ ⎣ ⎣ ⎥ ⎢ COMPS 2 ⎥ ⎢ ⎥ ⎢ ⎤ ⎡ ⎥ ⎢ ⎥ ⎢ local ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢HEAD | MAJ A ⎥ ⎢ ⎢ ⎡ ⎤⎥ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ subcat-info ⎥ ⎥ ⎢SYN | LOC ⎢ ⎢ ⎥⎥ ⎢ ⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎥ ⎢ 1 ⎥⎥ ⎢SUBCAT ⎢SUBJ ⎥ ⎢ ⎣ ⎦⎦ ⎣ ⎥ ⎢ ⎥ ⎢ 2 COMPS ⎥ ⎢ ⎥ ⎢ ⎡ ⎤ ⎥ ⎢ ⎥ ⎢ affix-sem ⎥ ⎢ ⎢ ⎥ ⎥ ⎢ OPERATOR ♦⎥ ⎥ ⎢SEM ⎢ ⎣ ⎦ ⎦ ⎣ ⎡

SCOPE 3

30

Morphology and its Interfaces

It states that -bar is a sufﬁx which attaches to the right of the stem and subcategorizes for a particular class of verbs, that is -bar verbs. As in passives, the presence of a -bar afﬁx turns the object of the verb into the subject of the adjective. The dropped subject normally cannot be expressed as a by-phrase. This is reﬂected in the various subcategorization lists by means of structure sharing: the object of the verb indicated by the tag 1 in the morphological SUBCAT list is structure shared with the subject of the resulting adjective in the syntactic SUBCAT list. The afﬁx -bar functions as the head of the complex word, thus the Morphological Head Feature Principle ensures that the mother obtains the category of the head daughter, which is a desirable result since lesbar is an adjective. Semantically, a notion of possibility is added, as can be seen by the presence of the ♦ operator. On the other hand, the lexical entry for the verb lesen is the following: ⎤ (35) ⎡bar-v ⎢ ⎥ ⎢MORPH | STEM les ⎥ ⎢ ⎡ ⎤⎥ ⎢ ⎥ LEX + ⎢ ⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎢HEAD | MAJ V ⎥⎥ ⎢ ⎢ ⎡ ⎤⎥⎥ ⎢ ⎥ ⎢ subcat-info ⎥ ⎢SYN | LOC ⎢ ⎥⎥ ⎢ ⎢ ⎥⎥⎥ ⎢ ⎢ ⎢ ⎥⎥⎥ ⎢ ⎢ ⎥⎥⎥ ⎢SUBCAT ⎢OBJ 1 ⎢ ⎣ ⎦⎦⎥ ⎣ ⎢ ⎥ ⎢ ⎥ COMPS 2 ⎢ ⎥ ⎡ ⎤ ⎢ ⎥ ⎢ ⎥ RELN ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢read ⎥ ⎢ ⎥ ⎥ ⎢SEM ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ 1 ⎢SOURCE ⎥ ⎢ ⎥ ⎣ ⎦ ⎣ ⎦ THEME 2

It encodes the information that lesen belongs to the bar-V class of verbs as well as information on its subcategorization, its argument structure and the fact that we are dealing with the lexeme. The entries for lesen and -bar together with the various morphological principles presented above license the adjective lesbar: ⎤ (36) ⎡bar-comp-A ⎡ ⎤⎥ ⎢ ⎢ affix-word-str ⎥ ⎢ ⎥⎥ ⎢DTRS ⎢ ⎣AFFIX bar-suff ⎦⎥ ⎦ ⎣ WORD bar-v

Morphology and its Interfaces

31

It is usually assumed that -bar sufﬁxation is a productive process. However, as discussed in Riehemann (1998), corpus analysis has revealed that there are some counterexamples. Transitive verbs which do not allow -bar sufﬁxation and non-transitive verbs that do allow it are attested, as well as lexicalized exceptions and new formations. Riehemann notices that these data are a problem for most traditional approaches to morphology, namely those ones which assume a simple X-bar approach and those ones which posit one lexical rule for forming -bar adjectives. These approaches have no mechanisms for dealing with the counterexamples. Riehemann’s answer is an alternative treatment which is inspired by realizational approaches to morphology. Not only can it deal with the counterexamples, but it also avoids many of the shortcomings that are attested in the analysis proposed by Krieger and Nerbonne. An obvious problem with their treatment is that syntactic principles are duplicated within morphology. Furthermore, the information is stated both in terms of principles and in terms of inheritance leading to redundancy. For example, the category of -bar adjectives is multiply determined: by inheritance, since bar-comp-adjectives are a subtype of adjectives, and by the Morphological Head Feature Principle, which states that the category of the complex word is the same as that of the afﬁx. Also, the information as to which verbs the afﬁx can combine with is given twice: in the class deﬁnition, and also in the subcategorization requirements within the lexical entry of the afﬁx. The analysis is thus a mere implementation of word–syntax approaches within HPSG and does not exploit the devices that are available within the framework to avoid the redundancy. In the next section, I show how Riehemann’s analysis is a step forward in this direction. 3.2.2. Realizational approaches

As already mentioned, a certain line of research within HPSG has been inspired by realizational approaches to morphology. Analyses in this direction are those of Riehemann (1993, 1998), Kathol (1999), Erjavec (1994), Miller and Sag (1997), Monachesi (1996a, 1999a), and Bonami and Boyé (2001). In the following section, I discuss Riehemann’s treatment of -bar adjectives and I point out the advantages of her analysis. The approach presented in Monachesi (1996a, 1999a) is compared with that proposed by Miller and Sag (1997) in the context of the discussion about Romance clitics. It is shown that these analyses are not a mere implementation of realizational approaches within HPSG, but that the devices offered by the framework allow for extensions of these approaches and thus influence the development of morphological theory. 3.2.2.1. Riehemann (1993, 1998) The main challenge that an analysis of -bar adjectives poses is the fact that its productive character has to be reconciled

32

Morphology and its Interfaces

with the subregularities and exceptions which should be dealt with in a principled way. Riehemann’s claim is that a traditional approach formulated in terms of word syntax or lexical rules is not sufficient. Her proposal is based instead on a hierarchical model which allows for a fine-grained distinction of word classes which is needed to deal with the relevant data. As already mentioned, she points out that having a classificatory approach to morphology and a wordsyntactic treatment (as in Krieger and Nerbonne’s analysis) leads to several redundancies. She suggests instead that the hierarchical approach alone, if properly extended, should be able to capture the facts. Her proposal is based on a hierarchy of -bar adjectives, as illustrated in Fig. 3.1, in which stems are cross-classified according to various properties. The specific stems at the bottom of the hierarchy are the lexicalized ones that need to be listed because of irregular properties. However, only their idiosyncratic properties need to be expressed as the more general patterns can be specified at the higher levels. The type reg-bar-adj is introduced at the lowest level and it functions as the schema for productive word formation. It says that there is a class of adjectives ending in -bar which have transitive verb stems as their morphological basis. In other words, the value of the MORPH-B feature is an object of type trans-verb, as shown in (37). It also states how the syntax and semantics of the verb relates to that of the adjective. For example, the semantics of the verb indicated by the tag 4 is put into the scope of the possibility operator in the semantics of the adjective.

stem

complex

externalized

&

adjective

...

&

. . . & possibility . . .

verb . . .

trans-verb . . .

derived

compound

compositional

simple

affixed . . .

prefixed . . .

eb- . . . suffixed

bar-adj . . .

poss-bar-adj

trans-bar-adj

reg-bar-adj

ebbar . . .

dative-bar-adj

prep-bar-adj

unentrinnbar . . .

verfügbar

FIGURE 3.1 The -bar adjectives hierarchy

...

intr-bar-adj

brennbar . . .

fruchtbar . . .

Morphology and its Interfaces

33

At the syntactic level, the schema specifies that the verb’s accusative object is the same as the subject of the adjective, as indicated by the structure sharing represented by the tag 2 . Phonologically, the affix is added to the phonology of the verbal stem: ⎤ (37) ⎡reg-bar-adj ⎢ ⎥ ⎢PHON 1 ⊕ bar ⎥ ⎢ ⎥ ⎤ ⎥ ⎡ ⎢ ⎢ ⎥ trans-verb ⎢ ⎥ ⎥ ⎢ ⎢ ⎥ ⎥ ⎢PHON 1 ⎢ ⎥ ⎥ ⎢ ⎢ ⎤⎥⎥ ⎡ ⎢ ⎢

⎥ ⎥ ⎢ ⎢ ⎥ 2 ⊕ 3 ⎥⎥ ⎥ ⎢MORPH-B ⎢ CAT | VAL | COMPS acc NP : ⎢ ⎥ ⎢ ⎢ ⎥⎥ ⎥ ⎢ ⎢ ⎢ ⎡ ⎤ ⎥⎥ ⎥ ⎢SS | L ⎢ ⎢ ⎥⎥ ⎥ ⎢ ⎢ ⎢ ACT ⎥⎥ ⎥ ⎢ ⎢ ⎢ 4 ⎦ ⎦⎦ ⎥ ⎣CONT | NUC ⎣ ⎣ ⎢ ⎥ 2 ⎢ ⎥ UND ⎢ ⎥ ⎢ ⎥ ⎤ ⎡ ⎢ ⎥ ⎤ ⎡ ⎢ ⎥ HEAD adj ⎢ ⎥ ⎢ ⎤⎥⎥ ⎡ ⎢ ⎥ ⎢ ⎥ ⎢ ⎢ ⎥ ⎥ ⎢ ⎥ ⎢CAT ⎢ 2 ⎢ ⎥ ⎥ SUBJ NP: ⎢ ⎢ ⎥⎥⎥ ⎢ ⎥ ⎢ ⎢ VAL ⎣ ⎦⎦⎥ ⎢ ⎥ ⎣ ⎥ ⎢ ⎢ ⎥ 3 ⎥ ⎢ ⎢ ⎥ COMPS ⎥ ⎢SS | L ⎢ ⎥ ⎢ ⎡ ⎤⎥ ⎢ ⎥ ⎥ ⎢ ⎢ ⎥ RELN ♦ ⎥ ⎢ ⎢ ⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎥ 2 ⎢CONT | NUCL ⎢ARG 1 ⎥ ⎥ ⎢ ⎥ ⎣ ⎣ ⎦⎦ ⎣ ⎦ ARG2 4

Notice that part of the information expressed in the -bar schema can be isolated so that certain generalizations can be made. For example, the following constraint states that objects of type sufﬁxed are characterized by the fact that the phonology of the morphological base precedes the sufﬁx: ⎤ ⎡ (38) suffixed ⎥ ⎢ ⎢PHON 1 ⊕ suffix ⎥ ⎥ ⎢ ⎥ ⎢ ⎦ ⎣ 1 MORPH-B

PHON

This property is also shared by objects of type reg-bar-adj, which inherit it from the type suffixed: the information thus needs to be expressed only once at a more general level. Similarly, for what concerns the following constraint, which expresses generalizations with respect to the possibility aspect: (39) possibility SS | L | CONT | NUCL | RELN ♦

34

Morphology and its Interfaces

It should be noted that this is a property that -bar adjectives share with derived words ending in -lich. This treatment shares, with realizational approaches to morphology, the same assumptions with respect to the internal structure for derived words, which is as in (27), repeated below: (27) [[les] V bar] A In Riehemann’s analysis, however, -bar does not represent phonological material that is added by a lexical rule, but should be seen in terms of a schema, arising as a generalization about existing -bar adjectives in the lexicon. The main contribution of this proposal can be found in an interesting use of lexical multiple-inheritance hierarchies within morphological theory (cf. also Flickinger 1987). As Riehemann points out, they play a crucial role in providing structure to the lexicon, by representing linguistically relevant subclasses of words explicitly. They reduce redundancy in the representations for idiosyncratic or exceptional words since they relate them to the rule, rather than just listing them. The result is an approach that can easily handle exceptions and productive rules, but also partially regular words and new subregular cases representing a step forward with respect to previous analyses of this phenomenon within HPSG. A shortcoming of this approach, however, is that a signature is missing, that is a definition of the relevant types and the appropriate features. It is not always obvious how to interpret the features and the types introduced as well as the schema for -bar adjectives whose formal status is unclear. In other words, there is no clear proposal for an architecture within which morphological analysis can be performed. More generally, it is not clear how the treatment proposed here can be extended to deal with other phenomena. In the rest of this chapter, I discuss other approaches that, while making use of the multiple-inheritance mechanism, are more rigorous from a formal point of view.

3.3. Romance clitics as a test case for grammatical interfaces In the previous sections, I have provided an overview of the way morphology is dealt with in HPSG. Formalizations of word–syntax and realizational approaches have been presented together with their extensions within the framework. The discussion, however, has mainly focused on the way morphological theory has developed within HPSG and has not been concerned with the relation between morphology and the other modules of the grammar. It is to this topic that the rest of the chapter is dedicated. In particular, Romance clitics are taken into consideration as their analysis constitutes a challenge for any grammatical

Morphology and its Interfaces

35

theory due to their properties, which can be analysed only through an appropriate interaction of the different modules of the grammar. Clitics are an interesting subject of research not only because their study can shed new light on how the grammar is organized but also because of their problematic status: their behaviour is intermediate between that of independent words and that of afﬁxes. Even though they seem to be more autonomous than afﬁxes, they attach phonologically to a host, in contrast to words. Within early work in generative grammar, such as that of Kayne (1975), the assumption that clitics are syntactically independent elements was not questioned. More generally, the problematic status of clitics with respect to the interaction of syntax, morphology, and phonology was, to a large extent, neglected. It was only with the appearance of Zwicky (1977) that clitics began to be considered from a broader perspective and that a classiﬁcation of clitic types, which takes into account their various syntactic, morphological, and phonological properties, was proposed. In his typology, Zwicky distinguishes two classes of clitics: simple clitics, which are syntactically normal elements that are phonologically dependent on an adjacent word, and special clitics, which are elements whose placement cannot be accounted for by the normal processes of syntax and for which speciﬁc rules must be stipulated. Romance clitics are clearly special clitics according to this distinction since they do not have the same distribution as the corresponding full forms. It is their peculiar distribution that has captured the interest of many linguists and has assigned them a central position within several generative studies. A crucial question in the study of clitics is whether the elements that are grouped under this label constitute a separate class or, on the contrary, whether there is no uniﬁed category of clitics and they should be analysed as independent words or as afﬁxes according to the situation. The view defended here is one in which Romance clitics are assimilated to afﬁxes: motivations are provided to support this position. Given their afﬁxal status, a morphological analysis is proposed to deal with their characteristics. It is shown that a realizational approach is the most appropriate to handle not only their morphological properties, but also to deal with their interface character. Even though the issue of whether Romance clitics should be considered afﬁxes might be controversial, I believe that the evidence I provide should convince the reader that the best way to deal with their idiosyncratic behaviour is within morphology. It is well known that the main challenge that clitics pose is that their morphological properties must be reconciled with their syntactic properties; for example they satisfy the subcategorization requirements of the verb, as well as

36

Morphology and its Interfaces

their phonological properties, and that is the special position they occupy in sentence structure. The rest of the chapter tries to provide an answer to the questions of how morphology interacts with syntax and phonology and which is the best way to represent this interaction within HPSG. Two approaches are compared: the one proposed in Monachesi (1999a) and that of Miller and Sag (1997). Furthermore, the HPSG organization of the grammar and its consequences for the interfaces is contrasted with that of other frameworks. 3.3.1. General properties of Romance clitics

Clitics have been the object of much discussion in the linguistics literature; however their deﬁnition is still a topic of much debate. Since a clear deﬁnition of what clitics are is still lacking, they are usually identiﬁed on the basis of certain properties that they share. The following are some of the criteria suggested in Kayne (1975) (cf. also Cardinaletti and Starke (1999)). Clitics occur in a special position within a sentence and it is different from the one in which full phrases occur. The following examples from the various Romance languages illustrate this: (40) a. Martina lo legge. [it] Martina CL.ACC reads ‘Martina reads it.’ b. Martina lo lee. [sp] Martina CL.ACC reads ‘Martina reads it.’ lê. [pt] c. Martina não o Martina not CL.ACC reads ‘Martina doesn’t read it.’ d. Martine le lit. [fr] Martina CL.ACC reads ‘Martina reads it.’ cites¸te. [ro] e. Martina o Martina CL.ACC reads ‘Martina reads it.’ In these cases, the accusative clitics occur on the left of the finite verb, while direct objects usually occur on the right: (41) a. Martina legge il libro. [it] Martina reads the book

Morphology and its Interfaces

37

b. Martina lee el libro. [sp] Martina reads the book c. Martina lê o livro. [pt] Martina reads the book d. Martine lit le livre. [fr] Martina reads the book e. Martina cites¸te cartea. [ro] Martina reads book.the ‘Martina reads the book.’ It should be noticed that in the examples above the clitics combine prosodically with the finite verb, preceding it: they are proclitics whereas, if they follow the verb, as in the presence of imperatives, they are enclitics. The details on the conditions of their placement in the various Romance languages are discussed at length in Section 3.3.2. Example (40) shows that Romance clitics must be adjacent to the verb which constitutes their host. Nothing can intervene between the clitic and the verb as shown by the following ungrammatical examples where an adverb is present in the relevant position:2 (42) a. *Martina lo rapidamente legge. [it] Martina CL.ACC quickly reads ‘Martina reads it quickly.’ rápidamente lee. [sp] b. *Martina lo reads Martina CL.ACC quickly ‘Martina reads it quickly.’ rapidamente lee. [pt] c. *Martina não o reads Martina not CL.ACC quickly ‘Martina doesn’t read it quickly.’ rapidement lit. [fr] d. *Martina le reads Martina CL.ACC quickly ‘Martina reads it quickly.’ Clitics occur in a fixed order and this order is usually different from that of the corresponding full phrases:

2 This kind of example is not relevant in Romanian since adverbs generally follow the verb in this language.

38

Morphology and its Interfaces

(43) a. Martina me lo spedisce. [it] Martina CL.DAT CL.ACC sends ‘Martina sends it to me.’ b. Martina me lo envía. [sp] Martina CL.DAT CL.ACC sends ‘Martina sends it to me.’ c. Martina não mo envia. [pt] Martina not CL.DAT CL.ACC sends ‘Martina doesn’t send it to me.’ d. Martine me l’envoie. [fr] Martina CL.DAT CL.ACC sends ‘Martina sends it to me.’ e. Martina mi-l trimite. [ro] Martina CL.DAT.CL.ACC sends ‘Martina sends it to me.’ In all the Romance languages considered in the examples above, the direct object usually precedes the indirect object if they are both represented by full phrases: this is the unmarked order. However, matters are different when clitics cluster together. It is usually the case that they occur in a different order from that of full phrases. The example in (43) shows that, when Romance clitics are present, the opposite order is grammatical. It should be noted that this is the only order which is allowed with this combination of clitics; the sequence in which an accusative clitic precedes a dative one is ungrammatical: (44) a. *Martina lo mi spedisce. [it] Martina CL.ACC CL.DAT sends ‘Martina sends it to me.’ me envía. [sp] b. *Martina lo Martina CL.ACC CL.DAT sends ‘Martina sends it to me.’ c. *Martina não o me envia. [pt] Martina not CL.ACC CL.DAT sends ‘Martina doesn’t send it to me.’ d. *Martine le m’envie. [fr] Martina CL.ACC CL.DAT sends ‘Martina sends it to me.’

Morphology and its Interfaces

39

e. *Martina îl-mi trimite. [ro] Martina CL.ACC.CL.DAT sends ‘Martina sends it to me.’ In certain Romance languages such as standard French and Italian, clitics are in complementary distribution with the correspondent full phrase: (45) a. *Martina gli parla a Paolo. [it] Martina CL.DAT speaks to Paul b. *Martine lui parle à Paul. [fr] Martina CL.DAT speaks to Paul ‘Martina speaks to him, to Paul.’ The only exception to this is represented by those cases of left or right dislocation, exempliﬁed here: (46) a. Martina gli parla, a Paolo. [it] Martina CL.DAT speaks, to Paul b. Martine lui parle, à Paul. [fr] Martina CL.DAT speaks, to Paul ‘Martina speaks to him, to Paul.’ On the other hand, in Romanian and in Spanish, clitics can, or in certain cases must, combine with the correspondent full phrase. In these cases, the clitics behave almost as agreement markers and one speaks of clitic doubling:3 (47) a. Martina îi d˘a pres¸edintelui o carte. [ro] Martina CL.DAT gives president.the a book ‘Martina gives the president a book.’ b. Martina le habla a Pablo. [sp] Martina CL.DAT speaks to Paul ‘Martina speaks to him, to Paul.’ Another characteristic usually associated with clitics is that they cannot be conjoined: (48) a. *Martina mi e lo conosce. [it] Martina CL.(ACC) and CL.ACC knows ‘Martina knows me and him.’ 3

For a possible analysis of clitic doubling in different frameworks, cf. Farkas (1978), Steriade (1980), Jaeggli (1982), Borer (1984), Su˜ner (1988), Kallulli (1995), Uriagereka (1995), Sportiche (1996), Torrego (1996), Anagnostopoulou (1999), Bleam (1999), and Gerlach (2001) among others. In Section 3.4.1, I sketch a possible analysis of clitic doubling within the present framework.

40

Morphology and its Interfaces b. *Martina me y lo conoce. [sp] Martina CL.(ACC) and CL.ACC knows ‘Martina knows me and him.’ e o conhece. [pt] c. *Martina não me Martina not CL.(ACC) and CL.ACC knows ‘Martina doesn’t know me and him.’ et le connaît. [fr] d. *Martine me Martina CL.(ACC) and CL.ACC knows ‘Martina knows me and him.’ s¸i îl cunoas¸te. [ro] e. *Martina m˘a Martina CL.(ACC) and CL.ACC knows ‘Martina knows me and him.’

In certain configurations, clitic pronouns do not appear on the verb that subcategorizes for them, as illustrated by the ungrammatical examples below, but on a higher verb. This is the case of clitic climbing. Auxiliary verbs trigger this phenomenon in all the Romance languages under consideration: (49) a. *Martina ha lo letto. [it] Martina has CL.ACC read ‘Martina has read it.’ letto. [it] b. Martina l’ha Martina CL.ACC has read ‘Martina has read it.’ leído. [sp] c. *Martina a lo Martina has CL.ACC read ‘Martina has read it.’ a leído. [sp] d. Martina lo Martina CL.ACC has read ‘Martina has read it.’ lido. [pt] e. *Martina não tem-o Martina not has CL.ACC read ‘Martina hasn’t read it.’ tem lido. [pt] f. Martina não o Martina not CL.ACC has read ‘Martina hasn’t read it.’ lu. [fr] g. *Martine a le Martina has CL.ACC read ‘Martina has read it.’

Morphology and its Interfaces

41

h. Martine l’a lu. [fr] Martina CL.ACC has read ‘Martina has read it.’ i. *Martina a îl citit. [ro] Martina has CL.ACC read ‘Martina has read it.’ j. Martina l-a citit. [ro] Martina CL.ACC has read ‘Martina has read it.’ In certain Romance languages, such as Italian and Spanish, there are other verbs which allow clitic climbing, that is modal, aspectual, and motion verbs or restructuring verbs after Rizzi (1982).4 Finally, clitics do not constitute an independent domain for what concerns stress assignment and they cannot carry contrastive or focal stress: (50) a. *Martina LO legge. [it] Martina CL.ACC reads b. *Martina LO lee. [sp] Martina CL.ACC reads c. *Martina LE lit. [fr] Martina CL.ACC reads ‘Martina reads it.’ d. *Martina O cites¸te. [ro] Martina CL.ACC reads ‘Martina reads it.’ It should be noticed that the properties mentioned in this section are not always exhibited by all those elements which are usually called clitics in the various Romance languages and dialects. For example, in Italian, there is the third person plural dative clitic loro ‘to them’ which, like the monosyllabic clitics, occurs in a special position within the sentence and appears adjacent to the verb. However, certain adverbs can separate it from its host and it is stressed. In Napoletano, a southern Italian dialect, clitics have properties similar to those of standard Italian: they occur in a special position, they attach to a verb, they follow a rigid order, but, differently from standard Italian, in this dialect the combination of two clitics receives stress, as discussed at length in Monachesi (1999a). 4 Analyses of clitic climbing from different theoretical perspectives are those of Rizzi (1982), Aissen and Perlmutter (1983), Picallo (1990), Moore (1991), Rosen (1989), Roberts (1997), Kayne (1991), Burzio (1986), Bleam (1994), Kulick (1997), Nishida (1991), and Monachesi (1993a, 1999c), among others. Clitic climbing in the presence of restructuring verbs is discussed in Chapter 5, while clitic climbing which is triggered by auxiliary verbs is the topic of Chapter 4.

42

Morphology and its Interfaces

As pointed out in van Riemsdijk (1999), a crucial question is whether it is possible to reach a coherent and unified concept of clitics, thus a clear definition of what clitics are, or whether there is no unified category of clitics. Even though it is not possible to provide a general answer, I argue that in the case of Romance clitics there is ample evidence that they do not constitute a uniform class. Therefore, instead of assuming the existence of a special class clitics, the elements of which exhibit different properties, I show that it is possible to assimilate them to other well established categories: in particular, I claim that they behave as inflectional affixes. 3.3.2. The affixal status of Romance clitics

In this section, I provide evidence for analysing Romance pronominal clitics as inflectional affixes which combine with the host as a result of morphological processes, as also argued by Miller (1992a) for French; Monachesi (1996a, 1999a) for Italian; Monachesi (2000) for Romanian; and Crysmann (2000b) for Portuguese. There are similarities with Bonet (1991), who also argues in favour of a morphological analysis of pronominal clitics. She shows that a morphological treatment can properly deal with synthetic clitic clusters and co-occurrence restrictions in clitic combinations which did not receive an appropriate analysis under syntactic accounts. It should be noticed, however, that even though Bonet (1991) argues in favour of a morphological analysis for pronominal clitics, she is not committed to the view that considers them affixal elements. In other words, a morphological analysis is also compatible with the assumption that clitics constitute a special class and cannot be reduced to inflectional affixes. A similar view is also defended by Gerlach (2001). However, given the non-uniform behaviour among the elements of this class, it seems more economical to tease apart their various properties and to investigate whether they can be assimilated to other categories. In this section, I look in detail at the arguments in favour of the affixal status of Romance clitics which are mainly based on Zwicky (1977, 1985), and Pullum (1983), in which tests are proposed for distinguishing affixes from nonaffixes (or words from nonwords).5 3.3.2.1. Degree of selection with respect to the host Similarly to affixes, Romance clitics are quite selective with respect to the host with which they combine. This is always the verb, as can be seen in the following examples: 5 It could be assumed that the set of nonwords includes both clitics and affixes and that they are both handled within morphology. It would be more economical, however, if it were possible to assimilate at least certain clitics to affixes. The aim of this section is to investigate to what extent this is possible.

Morphology and its Interfaces (51) a. L’articolo gli sembra incomprensibile. [it] the article CL.DAT seems incomprehensible ‘The article seems incomprehensible to him.’ parece incomprensible. [sp] b. El artículo le the article CL.DAT seems incomprehensible ‘The article seems incomprehensible to him.’ c. O artigo não lhe parece incompreensível. [pt] the article not CL.DAT seems incomprehensible ‘The article doesn’t seem incomprehensible to him.’ semble incompréhensible. [fr] d. L’article lui the article CL.DAT seems incomprehensible ‘The article seems incomprehensible to him.’ pare de neînt¸eles. [ro] e. Articolul îi article.the CL.DAT seems incomprehensible ‘The article seems incomprehensible to him.’ incomprensibile. [it] (52) a. *Un articolo gli an article CL.DAT incomprehensible incomprensible. [sp] b. *Un artículo le an article CL.DAT incomprehensible c. *Um artigo lhe incompreensível. [pt] an article CL.DAT incomprehensible incompréhensible. [fr] d. *Un article lui an article CL.DAT incomprehensible e. *Un articol îi de neînt¸eles. [ro] an article CL.DAT incomprehensible ‘An article incomprehensible to him.’ (53) a. Un articolo incomprensibile per lui. [it] an article incomprehensible for him b. Un artículo incomprensible para él. [sp] an article incomprehensible for him c. Um artigo incompreensível para ele. [pt] an article incomprehensible for him d. Un article incompréhensible pour lui. [fr] an article incomprehensible for him e. Un articol de neînt¸eles pentru el. [ro] an article incomprehensible for him ‘An article incomprehensible for him.’

43

44

Morphology and its Interfaces

In (51), the examples show that the clitics attach to the verb, while in (52) the sentences are ungrammatical because the clitic combines with an adjective. In fact, if the host is not a verb a strong pronoun should be used instead, as shown in (53). The examples above show that Romance clitics are highly selective in their combinatory possibilities, a property they share with afﬁxes. Furthermore, similarly to inﬂectional afﬁxes, they do not affect the lexical category of the host. Romanian clitics, however, constitute an exception with respect to the other Romance clitics since they can optionally attach to elements that are different from their morphosyntactic host (i.e. the verb). They can combine with negation (54a), complementizers (54b), nouns (54c), or wh-elements (54d) (Dobrovie-Sorin 1994): (54) a. Mihai nu-l as¸teapt˘a. [ro] Michael not.CL.ACC.3MSG waits for ‘Michael doesn’t wait for him.’ vede. b. Cred c˘a-l believe that.CL.ACC.3MSG sees ‘I believe that he sees him.’ c. Maria-i scrie des. Maria.CL.DAT.3SG writes frequently ‘Maria writes to him frequently.’ d. Unde-mi lasa fratele tau cheia. Where CL.DAT.1SG leaves brother yours key.the ‘Where does your brother leave me the key.’ The examples in (54) represent cases of phonological attachment in which the enclitic form of the clitic combines with the preceding word. At first sight, these data might argue against the affixal view of clitics defended in this chapter. However, I have shown in Monachesi (1998b) that they can be reconciled with a (phrasal) affix analysis along the lines of that proposed by Miller (1992a) (cf. Section 4.7 for more details). 3.3.2.2. Rigid order The most crucial evidence for the afﬁxal nature of Romance clitics comes from their ordering. I have already mentioned that their position within the clitic cluster is fixed, for example only the order in (55) is correct, not that in (56): (55) a. Martina glielo spedisce. [it] Martina CL.DAT.CL.ACC sends ‘Martina sends it to him.’

Morphology and its Interfaces

45

b. Martina se lo envía. [sp] Martina CL.DAT CL.ACC sends ‘Martina sends it to him.’ c. Martina não lho envia. [pt] Martina not CL.DAT.CL.ACC sends ‘Martina doesn’t send it to him.’ d. Martine le lui envoie. [fr] Martina CL.ACC CL.DAT sends ‘Martina sends it to him.’ e. Martina i-l trimite. [ro] Martina CL.DAT.CL.ACC sends ‘Martina sends it to him.’ gli spedisce. [it] (56) a. *Martina lo Martina CL.ACC CL.DAT sends ‘Martina sends it to him.’ b. *Martina lo se/le envía. [sp] Martina CL.ACC CL.DAT sends ‘Martina sends it to him.’ c. *Martina não o lhe envía. [pt] Martina not CL.ACC CL.DAT sends ‘Martina doesn’t send it to him.’ d. *Martine lui l’envoie. [fr] Martina CL.DAT CL.ACC.sends ‘Martina sends it to him.’ e. *Martina l-i trimite. [ro] Martina CL.ACC.CL.DAT sends ‘Martina sends it to him.’ It is also interesting to notice that the order varies from one language to another without any obvious motivation. As the examples in (55) show, the third person dative pronoun must follow the accusative one in French, while it must precede it in Italian, Portuguese, and Romanian (Miller and Monachesi 2003). It does not seem to be possible to correlate this difference to some other more general opposition between French and Italian/Romanian/Portuguese syntax. It should be noticed that in example (43), discussed in the previous section and repeated here, the order is identical in all the languages if a dative first person clitic and a third person accusative clitic are present at the same time:

46

Morphology and its Interfaces

(43) a. Martina me lo spedisce. [it] Martina CL.DAT CL.ACC sends ‘Martina sends it to me.’ b. Martina me lo envía. [sp] Martina CL.DAT CL.ACC sends ‘Martina sends it to me.’ c. Martina não mo envia. [pt] Martina not CL.DAT CL.ACC sends ‘Martina doesn’t send it to me.’ d. Martine me l’envoie. [fr] Martina CL.DAT CL.ACC sends ‘Martina sends it to me.’ trimite. [ro] e. Martina mi-l Martina CL.DAT.CL.ACC sends ‘Martina sends it to me.’ As pointed out by Anderson (1992), there seems to be a similarity between clitics and affixes, which also exhibit a rigid ordering. In languages such as Swahili (Stump 1992) or Nimboran (Inkelas 1993), the idiosyncratic ordering of affixes has been accounted for by assigning them to different position classes. If this approach is also adopted in the case of Italian, clitics should be ordered according to the following template: (57) Template for the Italian clitic cluster

Position:

I

II

III

IV

V

VI

mi ti gli le ci vi

ci

si (refl.)

lo la li le (acc)

si (imp.)

ne

Similarly for French, the following template proposed in Perlmutter (1971) has been adopted by Miller (1992a):

Morphology and its Interfaces

47

(58) Template for the French clitic cluster

Position:

I

NOM

II

III

IV

V

VI

VII

ne

me te nous vous se

3 ACC

3 DAT

y

en

The situation is apparently less complex in Romanian, since only dative and accusative clitics are attested. However, it could be argued that the Romanian clitic inventory is richer and includes not only pronominal clitics but other elements such as auxiliaries, intensifiers, complementizers, and negation. I postpone the discussion on whether this is indeed the case to Chapter 4. Therefore, if Romance clitics are considered affixes, their ordering could be accounted for in morphology by those mechanisms usually employed in the analysis of languages with rich morphology. On the other hand, if it is assumed that they have word status, their order should be dealt with in syntax, but I have shown that there is no clear correspondence between the order of clitics and that of the related full complements. Therefore, it would not be possible to derive the sequencing of clitics from general principles of syntax, which is an undesirable result.6 3.3.2.3. Coordination Further evidence for the fact that Romance clitics behave like afﬁxes comes from coordination. The crucial test in this respect is whether clitics can have wide scope over a coordination of hosts or whether they have to be repeated on each conjunct. If clitics have word status, one would expect them to have wide scope over a coordination since this is the case for syntactic words. On the other hand, if they must be repeated on each conjunct they must have affixal status. These observations are formulated in the Coordination Criteria in Miller (1992b). 6 It could be argued that Germanic weak pronouns also show templatic effects since they cluster together in a fixed order (Zwart 1993). However, since they are not part of the verb they cannot be analysed as affixes and this might undermine the claim that templatic effects support a morphological approach to Romance clitics. However, the fact that they can be split up quite easily suggests that they are not morphological elements and that their order should receive a different explanation. Furthermore, the variation in judgements with respect to their order calls into question the validity of the templatic effects.

48

Morphology and its Interfaces

It can be seen that, if two verbs are coordinated that share the same clitic, the clitic has to be repeated within each conjunct: (59) a. Giovanni la desidera e la cerca. [it] John CL.ACC desires and CL.ACC looks for b. Giovanni la desea y la busca. [sp] John CL.ACC desires and CL.ACC looks for c. Jean la désire et la cherche. [fr] John CL.ACC desires and CL.ACC looks for d. Ion o dorea s¸i o c˘auta. [ro] John CL.ACC desires and CL.ACC. looks for ‘John desires her and looks for her.’ The clitic cannot have wide scope over coordination, as can be seen in the following examples: (60) a. *Giovanni la desidera e cerca. [it] John CL.ACC desires and looks for b. *Giovanni la desea y busca. [sp] John CL.ACC desires and looks for c. *Jean la désire et cherche. [fr] John CL.ACC desires and looks for d. *Ion o dorea s¸i c˘auta. [ro] John CL.ACC desires and looks for ‘John desires her and looks for her.’ In Italian, there is an exception to the fact that clitics must be repeated in front of each verb, as the following example from Benincà and Cinque (1993) shows: (61) Lo

leggo e rileggo. [it] and reread ‘I read and reread it.’ CL.ACC read

Similarly for French, as noted by Benviniste (1974) and Kayne (1975), and Spanish: (62) a. Il le lit et relis. [fr] he CL.ACC read and reread b. El lo lee y relee. [sp] he CL.ACC read and reread ‘He reads and rereads it.’

Morphology and its Interfaces

49

It is possible for a clitic to have wide scope over a coordination only if the two verbs are closely related, namely if they are of the X and re-X type. One could think that in this case the two verbs form a lexicalized expression. However, this does not seem to be true because if the clitic is enclitic, wide scope is not possible, as shown by the Italian example below: [it] (63) *Per leggere e rileggerlo. to read and reread CL.(ACC) ‘To read and reread it.’ Although it is not clear why this should be the case, given the marginality of these sentences, they are not considered further. The behaviour of Portuguese clitics is quite interesting in this respect since, unlike clitics in the other Romance languages under consideration, they can have wide scope over a coordination. The example below from Crysmann (2000a) shows that the object clitic o can be shared across two conjoined verbs: comprou e leu. [pt] (64) Que o João o that the John CL.ACC bought and read ‘that John bought and read it.’ However, this is the case only with proclitics. Enclitics can never have wide scope over coordination: (65) a. *O João comprou-o e leu. [pt] the John bought.CL.ACC and read b. *O João comprou e leu-o. the John bought and read.CL.ACC ‘John bought and read it.’ Wide scope over coordination is not restricted to ordinary pronominal clitics: reflexives and reciprocals, as well as the non-argument clitic se (in its use as a marker of verb valency change), can be shared across two conjoined verbs. Furthermore, also in this case, wide scope over coordination is restricted to proclisis. To summarize: Portuguese clitics exhibit different behaviour with respect to the possibility of having wide scope over coordination compared to other Romance clitics. However, it is only proclitics that can be shared across two conjoined verbs and not enclitics. It is interesting to notice that this asymmetry is not only attested in Portuguese but can also be observed in French and Italian in those few cases in which wide scope over coordination is marginally acceptable. It seems that the coordination data do not support the affixal status of Portuguese clitics. However, it should be noted that the obligatory repetition

50

Morphology and its Interfaces

is a sufficient criterion and not a necessary one for affixal status as argued for in Miller (1992b). Thus, the coordination criterion is not a conclusive test in the case of Portuguese clitics. 3.3.2.4. Arbitrary gaps Zwicky and Pullum (1983) point out that arbitrary gaps can occasionally occur in inflectional paradigms; this is the case in several Romance languages where some verbs lack certain verbal forms. For example, in Italian, the verb involvere ‘to wrap’ lacks the forms related to the past tense. Similarly, solere ‘to be used to’ and sapere ‘to know’ do not have a form for the present participle. Verbs such as incombere ‘to impend’, splendere ‘to shine’, prudere ‘to itch’, urgere ‘to be necessary’, vertere ‘to concern’, and vigere ‘to be in use’ lack the past participle and therefore all the forms which are made in combination with it. Arbitrary gaps are also present in the case of clitics. They occur both when clitics combine with each other and when they cliticize to the verb. An instance of the former case which is attested in most of the Romance languages under consideration can be seen in the impossibility of combining a first or second person accusative clitic with a dative one: (66) a. *Martina gli mi presenta. [it] Martina CL.DAT CL.ACC introduces b. Martina le me presenta. [sp] Martina CL.DAT CL.ACC introduces c. *Martine me lui presénte. [fr] Martina CL.ACC CL.DAT introduces ‘Martina introduces me to him.’ The sentences above become acceptable if one of the clitics is replaced by a strong pronoun: (67) a. Martina mi presenta a lui. [it] Martina CL.ACC introduces to him b. Martina me presenta a él. [sp] Martina CL.ACC introduces to him c. Martine me présente à lui. [fr] Martina CL.ACC introduces to him ‘Martina introduces me to him.’ The combination of a first and second person clitic does not produce acceptable results: (68) a. *Mi

ti

raccomanda. [it]

CL.ACC CL.DAT recommends

Morphology and its Interfaces b. *Me

te

51

recomienda. [sp]

CL.ACC CL.DAT recommends

c. *Il me te recommande. [fr] he CL.ACC CL.DAT recommends ‘He recommends you to me/me to you.’ Similarly, in Portuguese an interesting case of arbitrary gap is attested. Crysmann (2000a) mentions that in the set of conceivable combinations which could be found with a ditransitive verb, there are some gaps that do not follow from any general syntactic or semantic consideration. In particular, if the clitic me combines with another object clitic, it is possible to observe some arbitrary restrictions in the allowed combinations. The example below shows that the direct object clitic me is compatible both with a pronominal and a reflexive interpretation when it combines with another clitic: (69) a. Ela apresentou-me-lhe como sendo o representante da she introduced.CL.ACC.CL.DAT as being the representative of Microsoft. [pt] Microsoft ‘She introduced me to him as Microsoft’s representative.’ b. Eu apresentei-me-lhe como sendo o representante da I introduced.CL.ACC.CL.DAT as being the representative of Microsoft. Microsoft ‘I introduced myself to him as Microsoft’s representative.’ On the other hand, indirect object me can only be interpreted as a pronominal when it combines with a direct object clitic: (70) a. Ela ofreceu-mo. [pt] she offered.CL.DAT.CL.ACC ‘She offered it to me.’ b. *Eu ofreci-mo. I offered.CL.DAT.CL.ACC ‘I offered it to myself.’ If the indirect object is realized as a full NP, a reflexive interpretation of me is possible, showing that the restriction observed above cannot be captured on the basis of syntactic and semantic properties alone, but should be seen as a case of arbitrary gaps:

52

Morphology and its Interfaces

(71) Eu ofreci-o a mim. [pt] I offered.CL.ACC to me ‘I offered it to myself.’ Romanian presents similar gaps in the combination of clitics. Farkas and Kazazis (1980) mention that not all dative-accusative person and number combinations are grammatical. In particular, it is not possible to have the first person accusative clitic m˘a together with a dative one: (72) a. *t¸i

m-au

CL.DAT CL.ACC have

dat de nevast˘a numai pentru c˘a ai insistat. given as wife only because have insisted

[ro] ‘They gave me in marriage to you only because you have insisted.’ b. *am auzit c˘a p˘arint¸i mei vor s˘a i m˘a dea de have heard that parents mine want that CL.DAT CL.ACC give as nevast˘a. wife ‘I have heard that my parents want to give me in marriage to him.’ Similarly, a first person singular or plural clitic together with a second person plural one does not yield a grammatical result: (73) a. *vor s˘a mi v˘a omoare. [ro] want that CL.DAT CL.ACC kill ‘They want to kill you on me.’ v˘a omoare. b. *vor s˘a ni want that CL.DAT CL.ACC kill ‘They want to kill you on us.’ Farkas and Kazazis (1980) make an attempt to explain the ungrammaticality of the sentences in (72) in terms of topicality. They assume two topicality hierarchies: (74) a. Ethical > Goal > Theme b. I person > II person > III person They claim that the ungrammatical sentences in (72) are the consequence of a conﬂict in the two hierarchies in (74) that arises when the clitic sequence involves the extremities of the hierarchies.7 However, the topicality hierar7 It should be noted that no independent motivation is given for the order in (74), which needs to be modified in order to account for the fact that certain additional sequences are ungrammatical in languages such as French and Modern Greek.

Morphology and its Interfaces

53

chies cannot explain the ungrammaticality of the sentences in (73). Farkas and Kazazis admit that they should postulate an additional arbitrary constraint in order to rule them out. This fact suggests that it is not possible to provide a principled account of the ungrammatical clitic sequences, suggesting that they should be considered cases of arbitrary gaps. This is evidence in favour of the affixal status of these clitics and it supports an analysis of these data in morphology which is the appropriate locus for dealing with this kind of arbitrariness. Arbitrary gaps are attested not only in possible clitic combinations but also when they attach to verbs. For example, in Italian, Benincà and Cinque (1991) point out that present participles are limited in their combinations. In particular, they cannot combine with third person accusative clitics, as example (75a) illustrates: (75) a. Gli argomenti riguardanti*-lo/*-la/*-le/?-li. [it] the topics concerning CL.ACC b. Les arguments le/la/les concernant. [fr] the topics CL.ACC concerning ‘The topics concerning him/her/them.’ In contrast, the equivalent French example is fully grammatical as shown by (75b). In Portuguese, past participles can generally combine with clitics. However, this is not the case if the past participle occurs in an absolute participial clause, as noted by Crysmann (2000a): crianc¸as, vamos passear. [pt] (76) a. *Lidos-os às walk read.CL.ACC to.the children go b. *Lidos-lhes os livros, vamos passear. read.CL.DAT the books go walk ‘Having read the books to the children, we go out for a walk.’ In the examples above, neither the direct object clitic nor the indirect object clitic can combine with the past participle. Similar gaps are attested in French, as observed by Miller (1992a), where the first person singular subject clitic je cannot occur in an inverted configuration with the majority of verbs, as shown by (77), while this is not the case with other person clitics (78): (77) a. *Sors-je? [fr] go out I ‘Do I go out?’

54

Morphology and its Interfaces b. *Chante-je? sing I ‘Do I sing?’

(78) a. Sors-tu? [fr] go out you ‘Do you go out?’ b. Chante-tu? sing you ‘Do you sing?’ It should be noticed that the ungrammaticality of the examples in (77) cannot be due to pragmatic reasons since the following questions which make use of the locution est-ce que are fully grammatical: (79) a. Est-ce que je sors? [fr] is this that I go out ‘Do I go out?’ b. Est-ce que je chante? is this that I sing ‘Do I sing?’ To conclude: these arbitrary gaps which occur both when clitics combine with each other and when they combine with verbs constitute evidence in favour of their affixal status. Similar gaps do not occur if pronouns or other independent words are present. 3.3.2.5. Morphophonological idiosyncrasies The phonological shape of afﬁxes is often affected by the phonology of the stem or of other afﬁxes with which they combine. Zwicky and Pullum (1983) note that morphophonological idiosyncrasies are rather common for inﬂectional formations. Their examples include forms like oxen or dice for the plural afﬁx, slept or went for the past tense afﬁx, best or worst for the superlative. It is well known that morphophonological idiosyncrasies also occur in cliticization. In Italian, if a clitic like lo or la attaches as a proclitic to a verb that begins with a vowel, a rule of vowel deletion applies, so that the clitic loses its vowel: (80) Martina l’accetta. [it] Martina CL.ACC.accepts ‘Martina accepts it.’

Morphology and its Interfaces

55

This applies also to mi, ti, ci, vi, and si, although mainly in spoken language. In Italian, the phonological shape of clitics can also be affected by the presence of other clitics. For example, clitics ending in -i have their ending changed into -e if they are followed by another clitic which begins with l- or n-: spedirà. [it] (81) a. Martina te/*ti lo Martina CL.DAT CL.ACC will send ‘Martina will send it to you.’ b. Martina me/*mi ne spedirà molti. Martina CL.DAT CL.PART will send many ‘Martina will send me many of them.’ It should be noted that this alternation is triggered only by adjacent clitics. Hosts with the required phonological form do not induce it, as can be seen in the following example: (82) Martina ti/*te negherà una spiegazione. [it] Martina CL.DAT will deny an explanation ‘Martina will deny an explanation to you.’ A change similar to the one discussed above is triggered by the third person dative feminine le; if it precedes a clitic beginning with l- or n-, then the masculine dative form gli is used instead: (83) *Le/glie-le

ho date. [it] have given ‘I have given them to her/him.’ CL.DAT.CL.ACC

Another type of modification occurs in the presence of two identical clitics where one of the two changes its phonological shape. In Italian, if there is a sequence of two occurrences of si, the first one surfaces as ci: (84) Ci/*si si veste pesanti. [it] CL. CL. dress heavy ‘One wears heavy cloths.’ This ban against the combination of two identical clitics can be found in different languages and, in general, phonological rules have been advocated to obtain the desired result. However, it is interesting to see that there is some variation with respect to the possibility of having two adjacent homophonous morphemes. Guasti and Nespor (1999) discuss the following case, where in a similar configuration such a sequence is possible, arguing against the existence of a phonological rule:

56

Morphology and its Interfaces

(85) Se se la sente. [it] if CL. CL.ACC feels ‘If he feels like it.’ On the basis of this evidence, they suggest that the ban against adjacent homophonous morphemes should be stated as a lexical constraint and not as a phonological one.8 A change, which is apparently similar to the one discussed above, occurs in the sequence of two clitics which both surface as vi. In this case, the second vi is realized as ci: ci/*vi porta. [it]

(86) Vi

CL.ACC CL.LOC take

‘He takes you there.’ In Italian, the locative clitic can be expressed either as ci or as vi, with vi used mainly in literary language. However, it is the clitic ci which is used if the accusative clitic vi is already present. Even though this change is usually attributed to the ban against two adjacent homophonous clitics, a closer look at the data shows that this is not the case. More generally, it seems that it is not possible to have vi in combination with an accusative clitic, as can be seen from the following examples: (87) a. Mi

ci/*vi porta. [it]

CL.ACC CL.LOC take

‘He takes me there.’ b. Ti

ci/*vi porta.

CL.ACC CL.LOC take

‘He takes you there.’ In (87a), locative vi combines with the first person singular accusative clitic, while in (87b) it combines with the second person accusative clitic, and in both 8 Boˇ skovic´ (2001) mentions that in Serbo-Croatian it is not possible to have a sequence of homophonous wh-phrases. The solution in this case is to front one phrase and to leave the other in situ, instead of fronting both. In Serbo-Croatian the ban against the combination of two homophonous elements is also attested with clitics: the accusative feminine clitic je is replaced by ju when adjacent to the third person singular auxiliary je. There is thus a difference between sequences of homophonous wh-words and homophonous clitics in which one of them changes its phonological shape. It seems that a different strategy is at stake in the latter case. One wonders if this might be due to the fact that, in the case of clitics, morphophonology plays a role, while in the case of wh-words only syntax is involved. If this were the case, these data might provide additional evidence against a syntactic treatment of cliticization and it might be problematic for the analysis in Boˇskovic´ (2001) based on a Copy Theory of movement (Chomsky 1993) which I discuss in more detail in Section 3.7.

Morphology and its Interfaces

57

cases the result is ungrammatical. The sequence is acceptable if locative ci is used instead. Similar phenomena can be found in French, as argued in Miller (1992a). For example, there are cases of idiosyncrasies in the realization of a verbal root and a clitic. In spoken French, the sequences je suis ‘I am’ and je sais ‘I know’ are usually realized as [Úyi] and [Úe], which can be written as chuis and chais. This phenomenon is limited to these two verbs; it should be noticed that the homonymous form je suis ‘I follow’, which is derived from the verb suivre ‘to follow’, cannot be realized as chuis. Similarly, the sequence y ir- ‘will go there’, where ir- is the allomorph of the future of the verb aller ‘to go’, is obligatorily realized with the reduction of the sequence of two [i] to only one, as (88c) shows. Example (88b) illustrates that the sequence of two [i] is ungrammatical: (88) a. Pierre ira à Paris. [fr] Peter will go to Paris ‘Peter will go to Paris.’ b. *Pierre y ira. Peter CL.LOC will go c. Pierre ira. Peter CL.LOC will go ‘Peter will go there.’ The fact that the idiomatic expression y aller de ‘to be at stake’ also shows obligatory [i] reduction, as can be seen in (89b), provides evidence that (88c) should not necessarily be interpreted as a case of ellipsis. Example (89a) where there is no [i] reduction is ungrammatical: irait de tout mon bien. [fr] (89) a. *Quand il y when he CL.LOC will go of all my good de tout mon bien. b. Quand il irait when he CL.LOC.will go of all my good ‘Even if all of my possessions were at stake.’ The example below, which illustrates that [i] reduction is impossible, suggests that this phenomenon cannot be reduced to a productive rule of French phonology:

58

Morphology and its Interfaces

(90) Il y illustra sa valeur. [fr] he CL.LOC will show his skills ‘He will show his value.’ Several cases of idiosyncrasies are also attested in the realization of sequences of French clitic pronouns which cannot be explained in terms of productive phonological rules, as discussed by Morin (1979). This is the case of the sequence je lui which is reduced to jui, as well as the sequence le/la/les lui reduced to lui and le/la/les leur reduced to leur. Furthermore, there is deletion of the vowels [a] and [y] in the case of la and tu before a host that begins with a vowel. Object clitics also give rise to a number of morphophonological idiosyncrasies in Portuguese. They occur both at the juncture between clitic and host as well as inside the clitic cluster. An instance of the former is represented by the deletion of a consonant preceding a clitic, discussed in Vigário (1999b). She mentions that consonant loss occurs regularly in verb final position before accusative clitics, and not before clitics marked for a different case: (91) a. Dá-lo. [pt] give.CL.ACC ‘to give it.’ b. Dar-nos. give.CL.DAT ‘to give to us.’ The examples above show that the consonant [r] is deleted in (91a) before an accusative clitic, while it is present in (91b) because the clitic is a dative one. Consonant deletion is also triggered by specific combinations of verb forms and clitic forms. In particular, it occurs with verbs marked for first person plural in combination with a first or second person plural dative clitic, as shown below, where the [s] is deleted in the verb damos ‘we give’: (92) a. Damo-nos. [pt] give.PRES.1PL-CL.1PL.DAT ‘We give to us.’ b. Damo-vos. give.PRES.1PL-CL.2PL.DAT ‘We give to you.’ However, the process does not occur if the verb is in the first, second, or third person singular, even if the clitic is the appropriate one, as shown by the

Morphology and its Interfaces

59

following examples where the [s] is not deleted: [pt] (93) a. Pus-nos. put.PRES.1SG-CL.1PL.DAT ‘I put to us.’ b. Dás-nos. give.PRES.2SG-CL.1PL.DAT ‘You give to us.’ c. Faz-nos. give.PRES.3SG-CL.1PL.DAT ‘He does to us.’ Similarly, consonant deletion is not triggered if the verb is the appropriate one, but the clitic is a second or third person singular clitic or a third person plural: (94) a. Damos-te. [pt] give.PRES.1PL-CL.2SG.DAT ‘We give to you.’ b. Damos-lhe. give.PRES.1PL-CL.3SG.DAT ‘We give to her/him.’ c. Damos-lhes. give.PRES.1PL-CL.3PL.DAT ‘We give to them.’ Consonant deletion can also be found within the clitic cluster if a third person accusative clitic is preceded by a first or second person dative clitic nos/vos. The result is a form such as no-lo or vo-lo. Vigário concludes that this process is an idiosyncratic one. She notes that the sequences of consonants in which deletion occurs are attested in the language. Thus consonant deletion cannot be due to an ill-formed combination. Vigário (1999b) mentions another interesting case of morphophonological idiosyncrasy triggered by clitics. In Portuguese, there is an alternation in the form of the accusative pronoun, which may be lo, no, or o. The clitic appears as lo when the preceding element ends in a consonant. As already mentioned, the consonant is deleted when the clitic is added. The form no appears when the preceding verb ends in a nasal diphthong: (95) Comem-no. [pt] eat.CL.ACC ‘They eat it.’

60

Morphology and its Interfaces

The form o appears in all other cases. This alternation, however, occurs only when the clitic is preceded by a verb or by another clitic, but not by other elements, as shown by the following examples: dou. [pt] (96) a. Apenas o only CL.ACC give ‘I only give it.’ b. Não o dou. not CL.ACC give ‘I don’t give it.’ Vigário argues convincingly that there is no phonological process that could relate these three forms and she concludes that these facts can only be explained as instances of allomorphic variation. Another case of morphophonological idiosyncrasies is provided by Crysmann (2000a) who notices that in Portuguese certain cases of fusion are attested in the combination of a third person accusative clitic o(s)/a(s) with the third person dative plural lhes. The two forms contract to lho(s) or lha(s) resulting in the neutralization of the number distinction for the indirect object. It should be noted that deletion of ﬁnal /s/ does not constitute a general phonological process in Portuguese, thus providing evidence for the idiosyncrasy of afﬁx fusion. Morphophonological idiosyncrasies are also attested in the case of Romanian cliticization, as discussed in Monachesi (2000). Optional vowel deletion occurs if a clitic ending in a˘ precedes a verb beginning with unstressed a or o: (97) a. M˘a

as¸teapt˘a. [ro]

CL.ACC waits

‘He waits for me.’ b. M-as¸teapt˘a. CL.ACC waits However, if the verb is an auxiliary, vowel deletion is obligatory: (98) a. M-a

invitat. [ro] has invited ‘He has invited me.’ b. *M˘a a invitat. CL.ACC has invited CL.ACC

Similarly to clitics ending in a˘ , the clitic se undergoes optional vowel deletion in front of a verb which begins with unstressed a or o:

Morphology and its Interfaces (99) a. Se

61

as¸eaz˘a. [ro]

CL.ACC sits

‘He sits.’ b. S-as¸eaz˘a. CL.ACC sits On the other hand, there is no vowel deletion if the clitics te, ne, or le occur in a similar context: (100) a. Te

as¸teapt˘a. [ro]

CL.ACC waits

‘He waits for you.’ b. *t-as¸teapt˘a. CL.ACC waits The morphophonological idiosyncrasies that I discuss above for Romanian suggest thus that different allomorphs should be postulated in the case of pronominal clitics.9 Three different classes can be distinguished:10 (101) Different classes of clitics Person 1 2 3 Case D A D A D A m. f. Class 1 îmi m˘a ît¸i te îi îl o Class 2 mi- m- t¸i- te- i- l- oClass 3 -mi -m˘a -t¸i -te -i -l -o

4 5 DA DA D

6

A m. f. ne v˘a le îi le ne- v- le- i- le-ne -v˘a -le -i -le

3,6 D A refl. îs¸i se s¸i- s-s¸i -se

Class 1 includes those clitics that occur in front of any verbs (proclitics), while in Class 2 are those clitics that can occur only before verbs that begin with a or o. Those pronominal clitics that follow the verb (enclitics) are grouped in Class 3. It seems, therefore, that the morphophonological changes that occur in the case of the Romance clitics under consideration can be naturally interpreted if clitics are considered as affixes, while they would be unexpected if they were considered as independent words. 9

However, see Popescu (2000) for a different view. The dash which is present in the case of Class 2 and Class 3 clitics illustrates an orthographic convention that requires these forms to be written attached to the verb. However, in certain cases, its presence reflects a different pronunciation (cf. Avram 1986). 10

62

Morphology and its Interfaces

3.3.2.6. Verb left-detachment Further evidence for the afﬁxal status of Italian clitics comes from a construction discussed by Benincà (1988). In spoken Italian, it is possible for the inﬁnitive of a verb to be left-detached. The inﬁnitive is then doubled by the indicative of the same verb, as shown in the following sentence: (102) Dormire, dormo benissimo. [it] to sleep.INF, sleep.IND very well ‘As for sleeping, I sleep very well.’ Bare infinitives or cliticized infinitives can undergo left-detachment: vedo ogni giorno. [it] la (103) Vedere/vederla, to see/to see.CL.ACC, CL.ACC see every day ‘As for seeing her, I see her every day.’ It would not be possible to have this kind of left dislocation if the infinitival is followed by complements: (104) *Vedere Martina, la vedo ogni giorno. [it] to see Martina, CL.ACC see every day ‘As for seeing Martina, I see her every day.’ The fact that bare and cliticized infinitives have the same status seems to offer an argument in favour of the view that Italian clitics are affixes. On the basis of similar evidence, this view has also been advocated by Vallduví (2001) for Catalan clitics. 3.3.2.7. Mesoclisis A potential argument in favour of the afﬁxal status of European Portuguese clitics comes from the possibility of pronominal clitics preceding inﬂection. The example in (105b) shows the phenomenon of mesoclisis according to which pronominal clitics occur before the future (or conditional) marker. It should be observed that the opposite order illustrated in (105c), in which the clitic follows the inflection, is not grammatical: (105) a. Falaremos. [pt] speak ‘We will speak.’ b. Falar-lhe-emos. speak.CL.DAT.FUT.1PL ‘We will speak to him/her.’

Morphology and its Interfaces

63

c. *Falaremos-lhe. speak.FUT.1PL.CL.DAT ‘We will speak to him/her.’ Mesoclisis has been used as evidence by Zwicky (1987) and Halpern (1995) in favour of the affixal nature of Portuguese clitics. They assume that the future/conditional markers have the status of inflectional affixes; since object clitics are attested between the lexical verb and these markers it follows that they must be inflectional elements. This argument depends, however, on the analysis of mesoclisis. I show in the next chapter, in section 4.4, that the future/conditional markers should not be considered inﬂectional afﬁxes; rather I argue that they are standard auxiliaries subject to syntactic rules (cf. also Vigário 1999b). It should be noticed, however, that the analysis I propose is still compatible with the afﬁxal status of object clitics. It only shows that mesoclisis is not a conclusive test. 3.3.2.8. The position of the clitics A possible argument against the afﬁxal analysis of Romance clitics comes from their position with respect to the verb. For example, in Italian, clitics can appear both as proclitics and as enclitics. They are proclitic if the verb is finite and enclitic if the verb is nonfinite or imperative: (106) a. Martina lo legge. [it] Martina CL.ACC reads ‘Martina reads it.’ b. leggilo! read.CL.ACC ‘Read it!’ Therefore, it is the morphosyntactic properties of the verb that trigger proclisis and enclisis in the various Romance languages. While Spanish patterns like Italian in this respect, in French enclisis is only triggered by imperatives. Portuguese, however, constitutes an exception, since proclisis is triggered by the presence of negation (107a), interrogative words (107b), prepositions (107c), conjunctions (107d), adverbs (107e), and relative clauses (107f) (Rouveret 1999; Crysmann 2000b): (107) a. Não me chamo Brown. [pt] not CL.ACC call Brown ‘My name is not Brown.’

64

Morphology and its Interfaces b. Porque nos passa o sal? why CL.DAT pass the salt ‘Why do you pass us the salt?’ c. Antes de lhe falar. before of CL.DAT speak ‘Before speaking to him.’ d. Embora você o contrc¸a. Although you CL.ACC know ‘Although you know him.’ e. Sempre lhe digo a verdade. Always CL.DAT tell the truth ‘I always tell him the truth.’ f. Pec¸o que me diga quanto custa. want that CL.DAT tell how much costs ‘I want you to tell me how much it costs.’

On the other hand, enclisis represents the default configuration if clitics are present: (108)

um livro. [pt] Ele deu-me he gave.CL.DAT a book ‘He gave me a book.’

It could be argued that such alternation between proclisis and enclisis is not typical of affixes and in particular that inflectional affixes are usually enclitic in Romance languages. It should be noticed, however, that there are languages in which such alternation can be found with afﬁxes as discussed in Monachesi (1996a). This is the case of the person marker in the imperfect/perfect in Arabic (Fontana 1993) and of the mobile afﬁxes in Huave (Noyer 1994) or in Afar (Fulmer 1990). Given the existence of languages with mobile position afﬁxes, it seems that the alternation exhibited by proclitics and enclitics in Romance does not constitute strong evidence against their afﬁxal status. Furthermore, it should be mentioned that this alternation is often neutralized in certain dialects. An example is Napoletano (Bafile 1993), where clitics occur almost exclusively before the verb, while in a subset of Piedmontese dialects spoken between the Sesia and Ticino enclisis is present in all environments (Tuttle 1992). In these dialects, clitics exhibit otherwise similar properties to those of the Romance languages.

Morphology and its Interfaces

65

3.3.3. Phonology

In the previous section, I have shown that Romance clitics share the behaviour of affixes from a morphosyntactic perspective; the next issue is whether they also behave as affixes from a phonological point of view. In this section, I look at the phonological properties of Romance clitics in order to assess to what extent they provide evidence in favour of their affixal status. 3.3.3.1. Italian Previous literature (Nespor and Vogel 1986; Scalise 1986) has shown that the phonological properties of Italian clitics are different from those of affixes. Motivations are based on the fact that clitics behave differently from affixes with respect to the phonological rules of Intervocalic s-Voicing (ISV) and Raddoppiamento Sintattico (RS) and also with respect to stress (the cliticized verb form violates the constraint that stress must fall on one of the last three syllables). The different behaviour of clitics with respect to the rules mentioned above has been considered evidence for the introduction of the Clitic Group (Hayes 1989; Nespor and Vogel 1986). In Monachesi (1996b, 1999a) I have argued against the existence of this prosodic category and I have shown that the behaviour of clitics is no different from that of certain afﬁxes, provided that a distinction is made between cases where one clitic is present and cases where two or more clitics are attached to the verb. Such a distinction follows as a natural consequence of Prosodic Minimality (McCarthy and Prince 1990), according to which a minimal word consists either of two syllables or two moras.11 If a monosyllabic clitic is present, it cannot constitute a prosodic word, and it therefore adjoins to the host to form a prosodic word with it (Booji 1995). Similarly, if two or more clitics are present, I have provided evidence that they merge together into a unit which adjoins to the verb. It could be argued that the combination of two clitics forms a prosodic word since it satisfies the requirements imposed by Prosodic Minimality. However, I have shown that this cannot be the correct conclusion for Italian, while it seems to be appropriate for Napoletano, a southern Italian dialect. In the rest of this section, I review the arguments I have provided in order to show that Italian clitics also behave as affixes from a phonological perspective. As already mentioned, a crucial difference between the phonology of clitics and that of affixes is related to stress. While affixes usually trigger stress shift, so that stress always falls within the last three syllables, this is not the case with Italian clitics. If they are present, it is generally assumed that the stress can fall further back than the third to last syllable. However, I have shown in 11 Cf. Peperkamp (1996) for additional evidence against the existence of the Clitic Group and for a slightly different analysis of the data presented here.

66

Morphology and its Interfaces

Monachesi (1996b) and Monachesi (1999a) that the facts are not so simple in the case of clitics and that a distinction is in order. It should be noted that if one clitic attaches to the verb, the stress generally falls within the last three syllables: (109) Mán-gia-lo. [it] eat.CL.ACC ‘Eat it.’ An exception is constituted by a restricted class of verbs which have antepenultimate stress. In this case, if a clitic is added, the stress would fall on the fourth syllable: (110)

Te-lé-fo-na-mi. [it] call.CL.ACC ‘Call me.’

A similar result is obtained if the inflectional suffix corresponding to the third person plural of the present indicative is added to the same type of verb: (111)

Te-lé-fo-na-no. [it] call.3PL-IND

As noted in Nespor and Vogel (1986), the existence of this verbal form does not affect the general conditions on stress (i.e. that stress must fall on one of the last three syllables in Italian), since it represents a restricted and predictable case. A similar conclusion can be reached for those cases where one clitic is added to verbal forms with antepenultimate stress, since they exhibit the same behaviour. In these cases, the clitic or the inflectional affix is adjoined to the prosodic word: (112)

PW

H H PW σ

telefona mi PW

H H PW σ

telefona no This representation also makes the right predictions with respect to ISV. This rule, which applies in northern varieties of Italian, istriggered in the contexts

Morphology and its Interfaces

67

illustrated in (113a), but not in those in (113b): (113)

Intervocalic s-Voicing

a. i[z]ola cau[z]-ava ca[z]-ina di[z]-uguale b. ri-[s]alare porta-[s]igarette lavando-[s]i la [s]era bella [s]era

‘island’ ‘he/she was causing’ ‘little house’ ‘unequal’

(morpheme internal) (before inflection) (before deriv. suffix) (end of prefix)

‘resalt’ ‘cigarette case’ ‘washing oneself ’ ‘the evening’ ‘beautiful evening’

(after prefix) (compound) (before enclitic) (after proclitic) (in phrase)

Nespor and Vogel assume that a stem forms a prosodic word with all suffixes and with those prefixes that end in a consonant since in those cases ISV applies, while prefixes that end in a vowel form an independent prosodic word since they do not trigger ISV. As for clitics and for prefixes that end in a vowel, they are outside of the domain of the prosodic word, since in their analysis they constitute independent prosodic words which form the Clitic Group. However, given the representation suggested above, the rule is not triggered because the clitic is outside of the domain of the prosodic word, which is the desired result: (114)

[s]apevo. [it] CL.ACC knew ‘I knew it.’ Lo

Scalise (1986) claims that ISV is violated in cliticization, but not in affixation, but this is not the case since, as we have seen in (113b), there are certain prefixes, such as ri-, a-, which do not trigger this rule. Thus, these prefixes behave exactly like clitics and they are also adjoined to the prosodic word. The rule of Raddoppiamento Sintattico has been used as evidence for the word-like status of clitics by Nespor and Vogel (1986) and Scalise (1986). This is a rule which applies in central and southern varieties of Italian. Given a sequence of two words, the rule lengthens the initial consonant of the second word, if the first word ends in a stressed vowel: (115) a. Martina ha letto metà [l:]ibro. [it] Martina has read half book ‘Martina has read half a book.’

68

Morphology and its Interfaces b. Martina ha letto due libri. ‘Martina has read two books.’

As can be seen in (115a), the word metà ends in a stressed vowel and RS is triggered; this is not the case in (115b) where the word due does not end in a stressed vowel. Nespor and Vogel (1986) consider the lengthening of the consonant which occurs when a clitic attaches to a verb ending in a stressed vowel to be an instance of RS: (116) Da[m:]i. [it] give.CL.ACC ‘Give me.’ However, a similar lengthening also occurs with the inflectional suffix -no or with the suffix responsible for past participle formation -to: (117) a. Da[n:]o. [it] ‘They give.’ b. Fa[t:]o. ‘Done.’ Therefore, these seem to be cases of lengthening of the consonant that occur word internally. It should be noted that a similar lengthening occurs in Napoletano (a Southern Italian dialect) where it could not be considered a case of RS, since in Napoletano RS is a morpho-lexical process which is triggered by the notion of neuter or plural feminine carried by determiners and clitics and not by word-final stress. The situation is different if two or more clitics attach to the verb. In this case, the violation of the stress constraint occurs with almost every verb: (118) Méscola-glie-la. [it] load.CL.DAT.ACC ‘Stir it for him/her.’ In Section 3.5.2.1, I provide evidence that object clitics combine together into a morphological unit. If two or more clitics are present, they merge together into a new unit which adjoins to the verb. There is thus a parallel with the inflectional affixes which, as Scalise (1986) observes, function as a unit even though they can be decomposed. He provides the example of -rebbero and suggests that this termination as a whole carries the information: conditional, third person, and plural. Similarly, in the case of clitics, the unit gliela as a whole carries the information of dative, accusative and third person. As for the structure which is assigned in those cases, it is similar to the one proposed in the case of one clitic:

Morphology and its Interfaces (119)

69

PW

PW

H H σ

mescola gli(e)

H σ

la

Under the assumption that it is the embedded prosodic word that constitutes the domain of stress assignment, it is possible to account for the fact that the combination of two (or more) clitics does not trigger stress shift. Similarly for the fact that Intervocalic s-Voicing does not apply. Given that the combination of two clitics forms a bisyllabic unit, it could be argued that it constitutes a prosodic word. This is indeed the position which I defended in Monachesi (1996b). However, I have abandoned it in Monachesi (1999a) in the case of Standard Italian. The fact that the clitic cluster does not receive stress shows that this cannot be the case. However, it seems to be the correct conclusion for Napoletano, as discussed in the following section. 3.3.3.2. Napoletano Clitics in Napoletano share the same behaviour as those of Standard Italian. They are verbal clitics whose position is determined by the morphosyntactic features of the verb. As in Italian, they are proclitic with ﬁnite verbs. They follow the verb only in the case of imperatives, since gerunds and past participles are hardly used in the spoken language. The possible combinations of clitics in a cluster resemble those of Standard Italian even though sequences of more than two clitics are usually not attested in Napoletano, as noted by Bafile (1993). The most crucial difference between Italian clitics and those of Napoletano is with respect to stress. Napoletano is similar to Italian in having stress which may fall on one of the last three syllables of a word. However, if two enclitics are present, the stress falls on the first of the two enclitics: (120) P’ɔrtatíllə. [na] bring.CL.DAT.CL.ACC ‘Bring it with you.’ It is possible to account for this difference if in Napoletano, unlike in Standard Italian, the combination of two clitics forms a prosodic word which combines with the verb as in compounding. This is in accordance with the requirements imposed by prosodic minimality, since the two clitics form a bisyllabic entity. In Monachesi (1999a), I have provided detailed motivations to show that the combination of two clitics forms a fixed unit in Napoletano. Under this view example (120) receives the following representation:

70 (121)

Morphology and its Interfaces PW

H H

PW

PW

pɔrte tíllə In this case, two prosodic words can be distinguished, each with its own stress. It could be argued that another representation is possible, one in which the clitics incorporate into the prosodic word, which constitutes the domain of stress. Rules of stress assignment operate in order to move the main stress from the verb to the first clitic. Evidence against this proposal is given by the rule of Vowel Raising. A closer look at (120) reveals that in this example the rule has not applied, showing that it cannot be a case of stress shift. The rule does not apply within compounds and, as observed by Bafile (1993), it does not apply when the two enclitics attach to the verb. In fact, in example (120), there is an [◦] in p◦rta. This fact indicates that the primary stress of the first word is not reduced, suggesting that the host and the clitics constitute two different domains with respect to stress assignment. Furthermore, it shows that compounds and encliticization pattern alike with respect to Vowel Raising. Bafile also notices that when two enclitics attach to a monosyllabic verb the stress falls on the clitics: [na] (122) Dammíllə. give.CL.DAT.CL.ACC ‘Give it to me.’ This case cannot be interpreted as a situation in which the stress shifts to obey the third last well-formedness condition; in fact if the stress were to fall on the verbal root, as in Italian, the conditions on stress would not be violated. However, this example can be explained if a structure like that in (121) is assumed. Since two prosodic words are present, each maintains its own stress. 3.3.3.3. French In French, clitics behave as afﬁxes with respect to stress assignment, as discussed in Miller and Monachesi (2003). The stress moves to the last syllable which does not contain a schwa. The following example shows a case in which the stress falls on the penultimate syllable, but the presence of the clitic triggers stress to shift to the last one: [fr] (123) a. Regárde. look.2SG.IMP ‘Look!’

Morphology and its Interfaces

71

b. Regarde-lés. look.IMP.CL.3PL ‘Look at them!’ It should be noticed that this behaviour is similar to that triggered by inﬂectional sufﬁxes, as shown by the example below (Delais-Roussarie 1999): (124) Regar-déz. [fr] look.IMP.PL ‘Look!’ As for proclitics, the situation is less clear. Delais-Roussarie (1999) suggests that they behave differently from prefixes in terms of the possibility of having an optional stress at the beginning of the word. According to Delais-Roussarie it is possible to have the stress falling on the prefix re-, but it is not possible to have stress falling on the clitic le in Paul le reverra demain ‘Paul will see him tomorrow’. However, as noted by Miller and Monachesi (2003), the facts are not so straightforward, since it might be possible to have contrastive stress on the clitic me in il me le donnera ‘He will give it to me’, given an appropriate context. It thus seems possible to conclude that the phonological behaviour of French clitics is rather similar to that of affixes. 3.3.3.4. Portuguese The situation appears to be different in the case of Portuguese clitics, according to the evidence provided by Vigário (1999b). She offers several arguments to show that Portuguese clitics do not behave as afﬁxes from a phonological point of view. As in the case of Italian, clitics do not affect the stress pattern of the verb to which they are attached, since the requirement that stress should fall no further back than the third to last syllable is not obeyed: (125) a. Dizíamos. [pt] tell ‘We told.’ b. Dizíamo-no-lo. tell.CL.DAT.CL.ACC ‘We told it to us.’ Vigário assumes that pronominal clitics are not yet present in the string when word stress applies.12 12 Van der Leeuw (1997) claims, however, that stress assignment is not a phonological process, but a morphological one in Portuguese. This view might have consequences for the affixal status of the clitics. Cf. also de Carvalho (1989).

72

Morphology and its Interfaces

Additional phonological evidence against the affixal status of Portuguese clitics comes from word-final nasal gliding. Vigário shows that the insertion of clitics does not block this process, as shown in (126a). Nasal gliding can only be found in word-final position (126b) and never inside a word (126c): (126) a. Bat[ ɐ˜˜j]-te. [pt] hit you ‘They hit you.’ b. Bat[ ɐ˜˜j]. hit ‘They hit.’ c. Bat[ ~e]te. door-knocker ‘Door-knocker.’ The examples show that clitics are not incorporated into the verb, otherwise nasal gliding would not occur. Portuguese clitics do not trigger a rule of dissimilation of non-back vowels followed by a palatal segment, as illustrated by the example in (127a). The rule centralizes a stressed non-back vowel when followed by a palatal consonant, as shown by (127b): (127) a. D[e]-lha. [pt] give.CL.DAT.CL.ACC ‘Give it to him/her.’ b. T[ɐ]-lha. tile ‘Tile.’ The non-application of this process in the presence of clitics, even though the phonological conditions would allow it, shows that clitics are not attached to their host when the rule applies. Finally, as example (128a) illustrates, clitics do not trigger the process of glide insertion to break a hiatus, which comprises the insertion of a non-back glide between two vowels, when the first is a stressed [e]: (128) a. Vê-o. [pt] see.CL.ACC ‘I see it.’ b. Rece[j]o. fear ‘I fear.’

Morphology and its Interfaces

73

The non-application of this process is evidence that the clitics do not form a phonological word with the verb. On the other hand, affixes trigger this process, as illustrated in (128b). Additional evidence that clitics do not behave as afﬁxes from a phonological point of view comes from the possibility of clitics undergoing reduction contrary to lexical afﬁxes. Example (129a) shows the relevant case of reduction occuring with a clitic. It consists of the deletion of a stressless non-back vowel, [e] in both cases, that can also surface as a glide if followed by a word starting with a vowel: (129) a. Não te ouvi. not CL.ACC hear

[j]/0 [pt]

‘I didn’t hear you.’ b. Readaptac¸ão. readaptation

[j]/*0

‘Readaptation.’ As (129b) shows, this reduction process does not occur if an affix is present in a similar context. Additional evidence of the different behaviour of clitics and affixes comes from the application of certain morphophonological rules discussed in Vigário (1999b). She mentions that verbs belonging to the third conjugation centralize to schwa the theme vowel (/i/) in stressless final position (130a). This process of centralization also occurs when the vowel is followed by a pronominal clitic (130b), but it is not triggered before inflected affixes (130c): (130) a. Part[ə]. [pt] breaks ‘He breaks.’ b. Part[ə]-me. breaks CL.ACC ‘He breaks me.’ c. Part[i]remos. will break ‘We will break.’ Similarly, a process of vowel deletion is triggered if the vowel is followed by an inflectional affix starting with a vowel. This is shown in (131a) where the vowel [e] disappears in the presence of the affix [o]. However, vowel deletion does not apply when the vowel is followed by a pronominal clitic (131b):

74

Morphology and its Interfaces

(131) a. Como. eat.PRES.1SG

(come + o) [pt]

‘I eat.’ b. Come-o. eats CL.ACC ‘He eats it.’ An additional piece of evidence mentioned by Vigário, which argues in favour of the fact that clitics and affixes show different phonological behaviour, is that while there are portmanteau forms involving either clitics or affixes, there are no portmanteau forms involving both. As for the representation of clitics, Vigário (1999a) argues that pronominal clitics are prosodized in the lexicon as syllables but they cannot be combined into a foot or a prosodic word. The lack of stress is the most obvious piece of evidence against prosodic word status. Vigário follows the literature which argues against the existence of the Clitic Group and she suggests a treatment of Portuguese clitics which does not rely on this constituent. In particular, she proposes a distinction between enclitics and proclitics: the former are incorporated into the preceding prosodic word while proclitics, as well as prefixes, are adjoined to the following prosodic word. Evidence for the incorporation of the enclitics comes from various rules such as Non-Back Vowel Deletion which is a prosodic word domain limit rule. It is interesting to note that the rule applies in (132a) since the [-back] vowel [e] of the clitic is deleted, but not in (132b) where the vowel [e] is realized as a glide: (132) a. Pec¸o-te azeitonas. ask.CL.DAT olives

0 [pt]

‘I ask you for olives.’ b. Pede-o. asks.CL.ACC

[j]

‘He asks for it.’ The non-application of the rule in (132b) shows that no prosodic word boundary occurs between the host and the clitic. In contrast, the rule applies in (132a) because the clitic is in ﬁnal position within the prosodic word. Additional rules that support the incorporation analysis are the semivocalization of high vowels

Morphology and its Interfaces

75

when preceded by another vowel and the rule of Back Vowel Deletion, both discussed at length in Vigário (1999a), to which I refer.13 As for proclitics, evidence that they are adjoined to the following prosodic word comes from‘emphatic’ stress which seems to cue the prosodic word initial boundary as argued in Vigário (1999a). She also considers the impossibility of the semivocalization of a high vowel beginning the host when it is preceded by a clitic ending in any vowel as additional evidence for prosodic adjunction. To conclude: on the one hand, the evidence provided in Vigário (1999b) seems to argue against the assumption that Portuguese clitics behave as afﬁxes. On the other hand, the representation proposed in Vigário (1999a) for enclitics suggests the opposite conclusion since they are incorporated within the prosodic word, as is usually the case for afﬁxes. In other words, Vigário (1999a) takes pains to demonstrate the lack of prosodic word boundary between clitics and host, a property which is typical of afﬁxes, while Vigário (1999b) provides evidence, on the basis of several rules, of the presence of a boundary between the clitic and the verb and of the different behaviour of afﬁxes in this respect. There is thus contrasting evidence and it seems that additional research would be necessary in order to settle the issue. 3.3.4. Conclusions

In this section, the syntactic, morphological, and phonological properties of Romance clitics have been taken into consideration. I have shown that their behaviour is often not uniform, both within one language and across languages. As discussed in Miller and Monachesi (2003), the complexity of the data and the variety of intermediate situations that are attested make it difficult to provide a conclusive, simple, and at the same time convincing answer with respect to the status of Romance clitics. However, it seems appropriate to conclude that, while from a (morpho)syntactic point of view Romance clitics show a status which seems intermediate between that of an afﬁx and that of a word, from a morphological perspective they do behave as afﬁxes. As for their phonological properties, it is still quite unclear whether they can provide evidence for their afﬁxal status. This seems to be an area in which variation is the most striking: it is left to future research to shed more light in this direction. More generally, it is desirable that the complexity of the data and the non-uniform behaviour of Romance clitics will continue to trigger research which will allow us to advance in our knowledge and in our theories on the interaction of the modules of the grammar. 13 Stress constitutes a potential problem for this analysis which is solved by stipulating that word stress assignment is turned off postlexically. The underlying assumption is that cliticization is a postlexical phenomenon.

76

Morphology and its Interfaces

In the next section, I show that a lexical analysis cast within the HPSG framework can provide an appropriate treatment of the afﬁxal properties of Romance clitics and, at the same time, can deal with their ‘word-like’ properties. The analysis proposed has been shown to be appropriate for Italian clitics (Monachesi 1999a) and Romanian clitics (Monachesi 2000), as well as for French clitics (Miller and Sag 1997), and could easily be adapted to deal with Spanish clitics. It is not clear, however, whether it could be extended to Portuguese clitics, given their peculiar status. It should be noticed that Crysmann (2000b) provides a lexical analysis of Portuguese clitics, which he considers morphological elements. The mechanisms adopted to analyse them are different from those employed in the next section, which might be evidence for the fact that Portuguese clitics are somehow different from the other Romance clitics.

3.4. A lexical analysis of Romance clitics The challenge that object clitics pose for any treatment is that their morphophonological properties must be reconciled with their syntactic characteristics. A comprehensive analysis must be able to deal with the afﬁxal character of Romance clitics, that is the fact that they are attached to the verb and that they follow a rigid order: these properties are best treated within morphology. In addition, Romance clitics can precede or follow the verb and they are subject to various idiosyncratic phonological changes which are usually morphologically triggered: these characteristics should be dealt with in (morpho)phonology. Furthermore, they satisfy the subcategorization requirements of the verb they attach to, behaving as full complements: a property which should be dealt with in syntax. It is evident that clitics are a topic of major importance for any theory of grammar since they play a role at all the various levels. Within the generative literature, two kinds of approaches have been competing in accounting for the special status of Romance clitics: a movement approach and a base generation approach. The first to propose a movement analysis (for French) was Kayne (1975). In his treatment, the clitic is generated in an argument position and it moves from there to adjoin to the verb. The source position is analysed as a trace. One motivation behind such a movement analysis is the fact already discussed in Section 3.3.1 that, in French and in Italian, clitics and the related complements are in complementary distribution. However, a potential problem for an analysis such as the one proposed by Kayne is posed by languages which exhibit clitic doubling, as in Spanish

Morphology and its Interfaces

77

or Romanian. The existence of this phenomenon has led to the formulation of base generation analyses put forward by Rivas (1977), Strozer (1976), Jaeggli (1982), and Borer (1984). Under this view, a clitic is base generated in its surface position, while the argument position is filled by pro. Alternatively, in the case of clitic doubling, this position can be filled by the relevant NP. Recently, mixed approaches have emerged such as that of Uriagereka (1995) and Sportiche (1996) that try to reconcile the movement and the base generation approaches. Uriagereka (1995) claims that pronominal clitics are not arguments but rather represent the functional part of an argument, that is the determiner head, that undergoes movement from within the argument to the functional domain of the clause. On the other hand, Sportiche (1996) assumes that clitics are base generated in pre-existing slots, namely they are X0s which head their own projections, and that clitic constructions may also involve movement. In particular, he suggests that this process should be decomposed into a first step which has properties of XP movement, while the second step should be considered Head movement. Sportiche argues for this type of analysis on the basis of motivations related to blocking effects of intervening subjects on clitic placement, past participle agreement and the similarity of long NP movement and clitic climbing in restructuring environments. Under this view, in clitic-doubling languages, the XP that moves is overt, while in non-clitic-doubling languages it is covert. Sportiche claims that his analysis can capture the advantages of both the movement and the base generation approaches, while providing a uniform treatment of cliticization. I propose an alternative approach to cliticization that shares some insights with the syntactic accounts discussed above, in particular with the base generation analyses and with the proposal of Sportiche (1996). Clitics are not to be considered lexical items, signs thus, which are located in a speciﬁc position by rules of syntax, but featural information which is provided in the lexicon and employed in morphophonology for the realization of the cliticized verb form. I suggest that cliticization is a lexical operation which has both a syntactic/semantic effect and a morphophonological one (cf. Monachesi 1996a, 1999a; Miller and Sag 1997). The former is reﬂected in the fact that clitics satisfy the subcategorization requirements of the verb they are an argument of. The lexical analysis is thus able to deal with the challenge posed by the syntactic properties of clitics. Cliticized verb forms are present in the lexicon as the result of an operation that derives cliticized verbs from simple ones. The cliticized forms differ from the simple ones in that the former have the subcategorization list reduced. This does not imply that object clitics reduce

78

Morphology and its Interfaces

the number of semantic arguments of the verb, it only means that the verb no longer needs a lexical NP to occur as complement since that position is filled by the clitic which gets assigned the appropriate semantic role. I have already mentioned that in Italian and in French, clitics are in complementary distribution with the full complement: dà un libro a Paolo. [it] (133) a. *Martina gli Martina CL.DAT gives a book to Paul donne un livre à Paul. [fr] b. *Martine lui Martina CL.DAT gives a book to Paul ‘Martina gives him a book to Paul.’ It is not possible to have the clitic and the related complement present at the same time, as the ungrammatical examples above show.14 It seems, therefore, that the effect of cliticization in these languages is to reduce the subcategorization requirements of the verb. Within HPSG, a lexical rule can be proposed to achieve this effect; it relates two sets of words: (134) Complement Cliticization Lexical Rule (CCLR) ⎤ word ⎡ ⎢ ⎥ ⎢HEAD ⎥ verb VAL | COMPS ⎢ ⎥ →⎣ ⎢ ⎥ ⎢VAL | COMPS 1 2 ⎥ CLTS ⎣ ⎦ CLTS elist ⎡

1 2 list (cl-ss)

⎤ ⎦

The rule relates verbs which subcategorize for certain complements to other ones with the same properties except that their subcategorization list is reduced. In other words, the effect of the rule is that those complements which are in the COMPS list are removed and added as members of the CLTS list. This list contains syntactic and semantic information about those complements which are realized as clitics. This is illustrated by the presence of the tag 2 as the value of the CLTS list. This is a feature which is appropriate for objects of type category. A certain convention is employed in the formulation of the lexical rules, that is only those features that are affected are mentioned in the 14 I have already mentioned that if the full complement is left (or right) dislocated, its cooccurrence with a clitic pronoun is possible. For an analysis of this type of construction, in Italian, which is compatible with the treatment of cliticization proposed here, I refer to Sanfilippo (1998). Additional studies that investigate the role of left/right dislocation with respect to information structure are those of Vallduví (1996) and Hendriks (1994).

Morphology and its Interfaces

79

output. Therefore, the feature HEAD and the type (i.e. word) are not repeated in the output because they remain the same. This rule has a relevant side-effect: since elements that are removed from the COMPS list (to be added as value of CLTS) are also members of the ARG-ST list, the corresponding element in ARG-ST is a synsem object of type clitic. Different types of synsem should be distinguished (cf. also Miller and Sag (1997)), such as phrase synsem (p-ss), word synsem (w-ss), and clitic synsem (cl-ss): (135)

synsem H H H

noncl-ss H p-ss w-ss

H

cl-ss

H H

HH acl-ss prcl-ss mark-ss

While p-ss and w-ss are relevant to distinguish phrases from words, cl-ss plays a relevant role in the analysis of past participle agreement.15 Synsem objects of type cl-ss are further partitioned in order to distinguish reflexive clitics (acl-ss), pronominal clitics (prcl-ss), and marker clitics (markss). Acl-ss contains information relevant for binding theory, which relies on the obliqueness hierarchy encoded in ARG-ST. As for mark-ss, it is used to distinguish certain uses of the clitics si as discussed in detail in Monachesi (1999a). All the other clitics are associated with prcl-ss. The lexical rule presented above plays a key role in the analysis of cases like the ones represented by sentences in simple tenses such as the ones shown here: (136) a. Martina gli dà un libro. [it] Martina CL.DAT gives a book b. Martine lui donne un livre. [fr] Martina CL.DAT gives a book ‘Martina gives him a book.’ In particular, the lexical rule applies to words which are licensed by the lexical entry for da/donne, which subcategorizes for two complements, the relevant part of which is shown below: (137)

COMPS

NP

acc , NP dat

15 I refer to Monachesi (1999a) for an analysis of past participle agreement which is triggered by object clitics.

80

Morphology and its Interfaces

The effect of the rule is that the indirect object is removed from the COMPS list and it is added as a member of the CLTS list: ⎡ ⎤ (138) ⎢VAL | COMPS NP acc ⎥ ⎢

⎥ ⎥ ⎢ ⎦ ⎣ CLTS NP dat [ 3M sg ] The necessary featural information is thus available for the spell out of the cliticized verb form, in a way which is made clear in Section 3.5.4. Furthermore, it is possible to account for the complementary distribution of the clitic and the related phrasal complement in sentences like the ones in (133), repeated below: (133) a. *Martina gli dà un libro a Paolo. [it] Martina CL.DAT gives a book to Paul donne un livre à Paul. [fr] b. *Martine lui Martina CL.DAT gives a book to Paul ‘Martina gives him a book to Paul.’ As can be seen in (138), the COMPS list of the verb dá/donne does not contain the information about the indirect object which is encoded instead in CLTS. Therefore it is not possible for a phrasal argument to be present, if the corresponding clitic is also present. The ungrammaticality of sentences like (133) above can be accounted for in this way. The lexical rule can also account for Romanian sentences such as the one below, in which the clitic and the related complement are in complementary distribution: (139) Martina îi d˘a o carte. [ro] Martina CL.DAT gives a book ‘Martina gives him a book.’ However, I have already mentioned that Romanian (as well as Spanish) also allows for the possibility of clitic doubling: d˘a pres¸edintelui o carte. [ro] (140) Martina îi Martina CL.DAT gives president.the a book ‘Martina gives the president a book.’ In the sentence above, the dative clitic îi is present in combination with the indirect object pres¸edintelui. A possible account of clitic doubling is sketched in the next section.

Morphology and its Interfaces

81

3.4.1. Clitic doubling

Romanian and Spanish differ from the other Romance languages, such as Italian and French, in that they trigger clitic doubling. There is a parallel with Bulgarian and Macedonian, which are also exceptional within the Slavic languages in demonstrating this phenomenon.16 As noted in Franks and King (2000), this accords with the generalization that true clitic doubling, as opposed to clitics occuring with dislocated constituents, occurs only in languages with verb-adjacent clitics and not in those with second-position clitics (Halpern and Fontana 1994). The crucial issue in dealing with clitic doubling is to single out its triggers, that is what semantic and discourse factors license it. This is not an easy task. On the one hand, the data are rather complex and are subject, in each language, to variation depending on the dialects and on the registers. On the other hand, the possibility of doubling is related to pragmatic and discourse factors for which it is difficult to provide a clear definition. Given the complexity of the phenomenon and the variation involved, it is not possible to address it in an exhaustive manner within this chapter. I limit myself to a brief overview of its properties taking into consideration both Romance languages, such as Spanish and Romanian, and Slavic languages, such as Bulgarian and Macedonian. Furthermore, I sketch how the lexical analysis of Romance clitics presented in the previous section can be extended to deal with clitic doubling in Romanian. In Spanish, clitic doubling is obligatory in the presence of a pronominal element which has the function of either a direct or indirect object (Jaeggli 1982; Su˜ner 1988; Soriano 1999; Franco 2000). This is the case in the various dialects: (141) a. Lo

veo a él. [sp] to him

CL.ACC see

‘I see him.’ b. *Veo a él. see to him ‘I see him.’ c. Le

hablo a él. to him

CL.DAT talk

‘I talk to him.’ 16 It should be mentioned that there are also close similarities with doubling attested in Balkan languages such as Greek (Anagnostopoulou 1994, 1999) and Albanian (Kallulli 1995).

82

Morphology and its Interfaces d. *Hablo a él. talk to him ‘I talk to him.’

If a non-pronominal element is present, the situation differs depending on syntactic, semantic, and pragmatic factors and according to the various dialects. In standard Spanish, doubling is possible with indirect objects (142a), (142b), which are located after the verb in an unmarked position, while it is not possible with either animate direct objects (142c) or inanimate direct objects (142e): (142) a. Le

hablo a Juan. [sp] to John ‘I talk to John.’ Hablo a Juan. talk to John ‘I talk to John.’ *Lo veo a Juan. CL.ACC see to John ‘I see John.’ Veo a Juan. see to John ‘I see John.’ *La veo la mesa. CL.ACC see the table ‘I see the table.’ Veo la mesa. see the table ‘I see the table.’ CL.DAT talk

b.

c.

d.

e.

f.

In certain Southern American dialects (i.e. Rioplatense Spanish (Su˜ner 1988)), variation is attested since doubling of indirect objects is the norm and there is a clear preference for doubling direct objects that are animate and specific. In these dialects, the variant with clitics shown in (142a) and (142c) is strongly preferred. In Romanian, the situation is similar. As Farkas (1978) notes, if the direct object is a deﬁnite pronoun and it follows the verb, doubling is obligatory for those pronouns which are marked [+ human], it is optional for pronouns which are marked [+/−human] if they refer to non-human entities, and impossible for those pronouns which are marked [−human]. In the case of an

Morphology and its Interfaces

83

indirect object represented by indeﬁnite pronouns, doubling is impossible except in the presence of speciﬁc partitive referents; in this case it is optional. As for nonpronominal elements, doubling with direct objects is optional if they are deﬁnite and [+human]: v˘azut pe Ion. [ro] have seen ACC John ‘I have seen John.’

(143) L-am

CL.ACC

As for indirect objects, they can always be doubled by a clitic and they must be doubled if they are definite pronouns. Slavic languages such as Bulgarian and Macedonian differ crucially from the Romance languages taken into consideration in that clitic doubling is not associated with a preposition (i.e. the marker pe in Romanian or the preposition a in Spanish). Furthermore, doubling has become grammaticalized as an obligatory marker of speciﬁcity in Macedonian (Franks and King 2000). Specificity (and not definiteness) is a necessary requirement in triggering clitic doubling in both Bulgarian (Rudin 1997) and Macedonian (Berent 1980). In particular, clitic doubling occurs in Bulgarian when the relevant object is both specific and topicalized. Topicality is thus an additional requirement for clitic doubling in Bulgarian: (144) Deacata ja obi˘cat neja. [bg] children.DEF CL.ACC love her ‘The children love her.’ Only in a few cases is doubling of direct and indirect object obligatory; in most cases it is optional (Franks and King 2000). In Macedonian, clitic doubling has become grammaticalized and it is obligatory for speciﬁc direct objects (145a) and unacceptable for nonspeciﬁc ones (145b) (Berent 1980): (145) a. Marija *(go) poznava (uˇcenikot/Vlado/nego). [mac] Mary CL.ACC knows pupil.sc def/Vlado/him ‘Mary knows the pupil/Vlado/him.’ b. Marija (*go) poznava eden uˇcenik. Mary CL.ACC knows one pupil ‘Mary knows a pupil.’ As reported in Franks and King (2000), failure to double a definite topicalized direct object results in ungrammaticality, whereas a comparable indirect object is marginal.

84

Morphology and its Interfaces

From this brief overview of the properties of clitic doubling across different languages, it appears that specificity is a crucial condition for triggering it (Su˜ner 1988). However, there might be additional factors which can affect it: as can be seen in Bulgarian, topicality also plays a role, as well as animacy, in languages such as Romanian. Another crucial issue in the analysis of clitic doubling is to establish the role of the clitic and its relationship with the NP double. There are various possibilities which have been proposed in the literature. In one alternative clitics are considered pronouns with special surface positions and special prosodic requirements. Another alternative views clitics as verbal arguments while NPs are adjuncts which corefer to the pronominal argument. Finally, pronominal clitics are object agreement and they are thus not pronominal. Franks and King (2000) examine these three possibilities, which are proposed by Rudin (1997) for Bulgarian clitics. They conclude with Rudin that the clitics are indeed object agreement in this language and they show that her argumentation extends easily to Macedonian. Under this view, object clitics do not receive a theta-role from the verb but instead agree with the verb’s relevant argument. A similar conclusion is also reached by Borer (1984), (Su˜ner 1988), and Sportiche (1996) for clitic doubling based on Romance languages. This is essentially the view I adopt here in order to account for Romanian clitic doubling: (140) Martina îi d˘a pres¸edintelui o carte. [ro] Martina CL.DAT gives president.the a book ‘Martina gives the president a book.’ Notice that in this conﬁguration, the clitic behaves almost as an agreement marker, since it cooccurs with the indirect object pres¸edintelui. It is straightforward to extend the analysis of cliticization presented in this chapter, which is based on the assumption that pronominal clitics are morphological elements. A variant of the cliticization lexical rule in (134) can be proposed. In particular, the relevant complements remain on the COMPS list and structure sharing with the SYNSEM information encoded in the CLTS list ensures agreement with the pronominal clitic as well as the appropriate role assignment: (146) Clitic Doubling Lexical Rule (CDLR) ⎤ word ⎡ ⎢ ⎥ ⎢HEAD ⎥ verb ⎢VAL | COMPS ⎢ ⎥ ⎢ ⎥→ ⎣ 1 2 ⎢VAL | COMPS NP dat ⎥ ⎣ ⎦ CLTS CLTS elist ⎡

⎤

1 2 dat ⎥ 2

⎦

Morphology and its Interfaces

85

In Romanian, however, the clitic cannot simply cooccur with any direct or indirect object but combines with objects that satisfy certain semantic and/or pragmatic conditions. It is only with those complements that the presence of the clitic is allowed or in certain cases is compulsory. The crucial issue is how to identify those complements. Different solutions could be employed: one possibility would be to use features, that is [+/−specific] or [+/−human], to mark the relevant NP. The lexical rules which regulate the cooccurrence of the clitics on the verb should be sensitive to the presence of complements marked with the appropriate features. Alternatively, a specific typing system can be envisaged which classifies the SYNSEM of the NP on the basis of the specific semantic and pragmatic conditions underlying clitic doubling. In this section, I have provided an overview of the conditions which trigger clitic doubling, taking Romance and Slavic languages into consideration. Furthermore, I have sketched how the analysis of cliticization presented in this chapter could be extended to deal with clitic doubling, the basic assumption being that pronominal clitics should be considered agreement markers. As discussed in Miller and Monachesi (2003), the variation attested in clitic doubling is particularly interesting for typological studies on the origins of agreement markers. On the basis of work by Moravcsik (1974) and Givón (1976), it seems possible to assume that verbal agreement is the result of a morphological process which originates on pronoun doubling arguments. Givón proposes that agreement develops with those arguments which are more easily topics. Furthermore, it develops more rapidly with pronouns than with NPs, with deﬁnite NPs than with indeﬁnite ones, with dative NPs than with accusative ones. It is evident that the data which are available for Romance languages with respect to clitic doubling are widely compatible with the hypothesis formulated by Givón. 3.4.2. Subject clitics

The analysis of Romance clitics presented in this chapter has mainly focused on object clitics. In French and in certain Northern Italian dialects, however, subject clitics are attested (Rizzi 1986). A relevant question is, therefore, whether the analysis proposed could be extended to account for them. While in Italian (and in Spanish) the subject does not need to be expressed, being a pro-drop language, and the relevant information is present on the inflection on the verb (147a), in French and in certain Northern Italian dialects such as Trentino, a subject clitic appears (147b, 147d) and it cannot be omitted (147c, 147e): (147) a. Mangia. [it] eats

86

Morphology and its Interfaces b. Il

mange. [fr]

CL.SUBJ eats

c. *Mange. [fr] eats magna. [tr] d. El CL.SUBJ eats e. *Magna. [tr] eats ‘He eats.’ Kayne (1975) has proposed that the analysis of French subject clitics should be similar to that of object clitics since they are in complementary distribution with the NP. However, as discussed in Miller and Monachesi (2003), this is not the case in spoken French (148a) (Zribi-Hertz 1994), in Quebec French (148b) (Auger 1994), and in the Northern Italian dialects (example from Torinese) (148c) (Rizzi 1986), in which the subject clitics should be considered agreement markers since they cooccur with the NP subject: (148) a. Ben oui mais alors personne il a une table exhaustive. [fr] ok yes but then nobody CL.SUBJ has a table exhaustive ‘Ok, but then nobody has an exhaustive table.’ b. En campagne, quand quelqun’il dansait. [fr] in country when somebody CL.SUBJ danced ‘In the country, when somebody danced.’ dit gnent. [tor] c. Gnun l’a nobody CL.SUBJ has said anything ‘Nobody has said anything.’ The analysis proposed for object clitics could easily be extended to account for subject clitics in Standard French, in which they are in complementary distribution with the full NP. A lexical rule such as the following can be proposed: (149) Subject Cliticization Lexical Rule (SCLR) ⎤ ⎡ ⎡ word

⎤⎤ ⎤⎥ ⎡ ⎢ SUBJ elist ⎥ ⎢ ⎢ ⎥⎥ ⎢VAL HEAD verb ⎢ ⎥⎥ ⎢SS | L | C⎢ ⎥⎥ ⎢ → ⎥ ⎢ ⎦⎦ ⎣ ⎣

⎥ ⎢ ⎥ ⎢SS | L | C 1 CLTS ⎣VAL SUBJ 1 NP nom ⎦⎦ ⎣ ⎡

The effect of the rule is that the element which is in the SUBJ list is removed and added as a member of the CLTS list. This list contains syntactic and semantic information about those complements which are realized as clitics, thus

Morphology and its Interfaces

87

including subject clitics. Since the information about the subject is removed from the SUBJ list, it is possible to account for the complementary distribution between subject clitics and full NPs. In those Italian dialects in which subject clitics behave like agreement markers, as well as in spoken French and in Quebec French, an analysis similar to that proposed for clitic doubling can be proposed. The information about the subject is not removed from the SUBJ list, while structure sharing with the element in the CLTS list ensures agreement between the subject NP and the subject clitic. As noted by Miller and Monachesi (2003), the morphophonological idiosyncrasies and the arbitrary gaps which are attested in the case of French subject clitics provide evidence of the fact that the subject clitics in French have both the status of agreement markers and of lexically attached afﬁxes.

3.5. Morphophonological properties of Romance clitics The lexical analysis of Romance clitics, which I have discussed in the previous sections, has mainly taken into consideration the syntactic properties of cliticization, that is the fact that the presence of clitics affects the subcategorization requirements of the verb they combine with. However, I have provided ample evidence that Romance clitics also play a relevant role at the morphological level: any comprehensive analysis must be able to deal with their morphophonological properties. Crucial issues in this respect are how the fixed order in the clitic cluster can be derived and how it is possible to account for the morphophonological idiosyncrasies discussed at length in Section 3.3.2. Both syntactic and morphological approaches have been proposed in the literature to deal with these issues. In the following section, I review some of these syntactic analyses in order to point out their limitations. I show that morphological approaches are much more successful in dealing with the relevant questions. 3.5.1. Syntactic and morphological approaches to clitic distribution

The issue of clitic ordering was the topic of much discussion in the 1970s, following the seminal work of Perlmutter (1971). However, as discussed in more detail in Monachesi (1999a), these purely syntactic treatments cannot provide the solution to the formation of the clitic cluster. They all seem to suggest that syntactic transformations cannot appropriately account for clitic ordering. Therefore, other mechanisms have been introduced, such as templates (Perlmutter 1971; Wanner 1977) or numerical devices (Seuren 1976) in order to obtain the correct result.17 17 Emonds (1975) has proposed an alternative approach which is also based on the notion of templates while using transformations. See, however, Morin (1981) for a discussion of the shortcomings of this analysis.

88

Morphology and its Interfaces

More recent syntactic analyses which address the issue of clitic ordering, such as Sportiche (1996), still rely on the idea of a template. In fact, Sportiche postulates the presence of different clitic projections which are ordered according to a chart similar to the one proposed by Perlmutter for French. Bonet (1991) provides crucial evidence against a syntactic approach to clitic clusters. In particular, she notices that dialects can show a certain variation with respect to order, while this does not seem to correlate with signiﬁcant differences in syntax. Furthermore, it is often the case that certain clitics occupy a speciﬁc position within the cluster, regardless of their syntactic function. However, the most obvious criticism that syntactic approaches to the formation of the clitic cluster are confronted with is that they have generally ignored the problem of synthetic clusters and the cooccurrence restrictions in clitic combinations. If these issues are taken into consideration, it becomes even more evident that a syntactic approach to the distribution of clitics is inadequate. More recent approaches to the formation of the clitic cluster have not only addressed the problem of clitic ordering, but have also dealt with issues concerning synthetic clusters, that is non-transparent combinations of clitics, and cooccurrence restrictions. In particular, they have shown that in order to address these problems in an adequate way, it is necessary to shift towards a morphological treatment of the formation of the clitic cluster. Analyses in this direction are those of Bonet (1991), Inkelas (1993), Anderson (1992), and Simpson and Withgott (1986). For example, Bonet (1991) views clitics as a hierarchical structure of morphological features that is obtained through a mapping from S-structure to the morphological component. This component is characterized by speciﬁc morphological rules that can operate on the basic structure which is assigned to each clitic, by changing its features (cf. also Marantz 1993). It is in this way that she can account for synthetic clusters, as in the case of the Spurious se rule. Even though the notion of template introduced by Perlmutter (1971) still plays a relevant role in morphological treatments of clitic ordering, they represent a step forward with respect to syntactic approaches. In particular, Simpson and Withgott (1986) introduce template morphology, which involves the construction of a flat morphological template. Template morphology creates a structure which comprises slots that are filled by speciﬁc morphemes. This notion of template differs crucially from that proposed by Perlmutter (1971), in that the former creates a morphological structure while the latter acts as a ﬁlter on syntactic structure. A distinctive feature of template morphology is the notion of position classes. Template morphology is characterized by being a word formation process which does not make reference to the notion

Morphology and its Interfaces

89

of head and where the absence of information (zero morphemes) can play a meaningful role. Simpson and Withgott (1986) argue that pronominal clitic clusters constitute a clear case of template morphology. This is evident from the fact that they form a flat structure in which elements belong to different slots. Furthermore, the phonological form of clitics might depend on other clitics that appear in the cluster and it is not evident which element should be considered the head. The notion of position class, which is crucial in an account of clitics in terms of template morphology, also plays a relevant role in Inkelas (1993), which provides an analysis of the complex verb morphology of Nimboran. However, her approach is different from the proposal made in Simpson and Withgott (1986). The latter considers the position class template as a flat structure where linear precedence is encoded directly, while Inkelas proposes a template with hierarchical levels. Ordering relations are thus obtained by combining two sources of information: there is a precedence dimension which is the result of dividing morphemes into preﬁxes and sufﬁxes and a dominance dimension which follows as a consequence of the possible levels of attachment. In this system, afﬁxes are speciﬁed for a subcategorization frame which encodes information on the direction and on the level of attachment. In this way, the notion of position class also plays a relevant role in the analysis of languages where affixes are rigidly ordered, making the correlation between affixes and clitics more evident. To conclude: morphological analyses can offer a solution to the presence of synthetic clusters and to the cooccurrence restrictions exhibited by certain clitics. Therefore, they can provide a more comprehensive perspective on cliticization since they take into account those morphophonological properties of clitics that were traditionally neglected in syntactic accounts. 3.5.2. The clitic cluster

The previous section has shown that morphological analyses are rather successful in accounting for the clitic cluster as well as for the cooccurrence restrictions exhibited by certain clitics. I will also adopt a morphological analysis for these properties of Romance clitics; however, my treatment differs in certain respects if compared to the approaches discussed in the previous section. In particular, instead of assuming a templatic approach to clitic ordering, I suggest that the combination of two or more clitics is conceived as a new unit, which does not result from merging two individual forms. Evidence is provided to support this position which is mainly based on the Italian and Romanian clitic cluster (Monachesi 1999a, 2000).

90

Morphology and its Interfaces

3.5.2.1. The Italian pronominal clitic cluster Corpus research on the Italian Reference Corpus has revealed that almost all combinations of clitics that were found in the corpus consist of two elements. Cases with three clitics were very rare and in all of them the reflexive clitic si was present (Monachesi 1999a).18 The table in (150) reports the attested clusters which have been extracted from the corpus: (150)

ce la ce le ce li ce lo ce ne ci si

gliela gliele glieli glielo gliene gli ci gli si

la si le si li si lo si le ci

me la me le me li me lo me ne mi ci mi si

se la se le se li se lo se ne

te la te le te li te lo te ne ti ci ti si

ve la ve le ve li ve lo ve ne vi ci vi si

The table shows that Italian clitics seem to be ordered in a way that does not have a straightforward connection with syntactic, semantic, or phonological information. Their order appears completely idiosyncratic. While for Spanish the notion of grammatical person and case plays a role in the linearization of clitics, this information does not seem to be crucial for the ordering of Italian clitics as discussed in Monachesi (1999a). The possibility that clitics are ordered according to their phonological shape is rejected by Wanner (1977). Similarly, as already mentioned, the order of clitics is not related to that of full complements: in fact dative clitics precede accusative clitics, while the opposite holds in the case of full complements. It could be assumed that the ordering of afﬁxes on a stem is predicted to reﬂect the order in which the stem picks up afﬁxes via head to head movement according to the Mirror Principle (Baker 1988). This principle regulates the relation between the hierarchy of functional projections and the ordering of the corresponding morphemes in morphological structure. It speciﬁes that morphological derivations must directly reﬂect syntactic derivations (and vice versa). However, as Miller and Sag (1997) have argued, the constraints enforced by the Mirror Principle cannot be assumed in the case of Romance pronominals, especially as one has to deal not only with the ordering of clitics but also with their compatibility. They notice that the tight link between syntactic and morphological structure enforced by the Mirror Principle has been abandoned 18 There were around 14,000 examples of combinations of clitics which I examined; only six contained combinations with three clitics.

Morphology and its Interfaces

91

precisely by those who were dealing with the morphological and phonological properties of Romance clitics such as Bonet (1991) and Cummins and Roberge (1994). The conclusion, therefore, seems to be that the most appropriate way to characterize the distribution of clitics in Italian is by making reference to the descriptive notion of position class, which could be expressed by means of template morphology. There is some evidence which supports the possibility of considering the Italian clitic cluster as a case of template morphology. Clitics belong to different slots and it is clear that within the cluster, the notion of head plays no role. Furthermore, the phonological form of clitics might depend on which other clitics appear in the cluster and there are arbitrary gaps in the combinations of clitics. If an approach in terms of template morphology is adopted, clitics are ordered according to the following template, which holds for Standard Italian: (151) Template for the Italian clitic cluster Position: I

II III

IV

V

VI

mi ci si lo si ne ti (refl.) la (imp.) gli li le le ci (acc.) vi Clitics are assigned to different position classes and they appear in the order prescribed by the template. Since no more than one clitic per position can be present, it is possible to rule out ungrammatical sequences such as *gli mi or *gli ti simply because both gli and mi/ti belong to the same position class.19 However, a templatic approach does not represent an immediate solution to the analysis of the morphophonological idiosyncrasies described in Section 3.3.2. The template introduced above provides a straightforward account of the idiosyncratic order of clitics and of the fact that certain combinations are not possible. On the other hand, it does not deal easily with those cases of allomorphy, deletion, and sequencing of similar clitics already discussed. 19 It is obvious that some ungrammatical orders could be ruled out for syntactic reasons given that a sentence cannot have, for example, two direct objects. However, there would still be many combinations that cannot be excluded on those grounds.

92

Morphology and its Interfaces

Furthermore, a shortcoming of a templatic approach is that it predicts that it should be possible to have a sequence of six clitics (one for each position class) attached to the verb. The investigation of the Italian reference corpus has shown that in the general case, only two clitics are present in the cluster. An appropriate treatment should therefore take this fact into consideration. It seems more appropriate to suggest that Italian clitics merge together into a morphological unit which combines with the verb. In Monachesi (1999a), I have suggested that such a unit is not the result of a word formation process such as template morphology, but represents the phonological spelling out of certain morphosyntactic features of the verb. It is thus possible to account for the ordering of the clitics in the cluster. Under this view, the issue of their ordering does not even arise, since the combination of two or more clitics is conceived as a new unit. Several arguments support the assumption that Italian clitics merge together into a morphological unit which combines with the verb. For example, the same order is found in the clitic cluster, both when the clitics are proclitic and when they are enclitic: (152) a. Me

lo

spedisce. [it]

CL.DAT CL.ACC sends

‘He sends it to me.’ b. Spedisci-melo. send.CL.DAT.CL.ACC ‘Send it to me.’ There does not seem to be evidence that clitics are added one at a time, producing a hierarchical structure; in that case the mirror image would be expected.20 Additional evidence for the proposal that the combination of two clitics forms a unit can be seen in the fact that if a clitic climbing trigger verb is present, the clitics which originate as complements of the lower verb (153a) must climb together (153b); they cannot be separated as in (153c): (153) a. Martina deve spedirmelo. [it] Martina must send.CL.DAT.CL.ACC b. Martina me lo deve spedire. [it] Martina CL.DAT CL.ACC must send deve spedirlo. c. *Martina mi Martina CL.DAT must send.CL.ACC ‘Martina must send it to me.’ 20 French seems to constitute an exception to this since, in certain cases, the mirror image is obtained. However, it should be noted that when this is the case, a different clitic surfaces as enclitic, namely il me le donne ‘he gives it to me’ vs. donne le moi ‘give it to me’.

Morphology and its Interfaces

93

It seems, therefore, that this combination forms a unit which cannot be divided. Further support is provided by data from acquisition which show that children use combinations of clitics at a very early stage. An investigation of the CHILDES database has revealed that combinations of two clitics are used starting from the beginning of the second year and no mistakes are found in their ordering.21 Diachronic data also seem to support the hypothesis that the combination of clitics forms a unit. In Old Fiorentino, from which Italian derives, there was an evolution in the position of object clitics from an order of the type mii.o. lod.o. to an order of the type lod.o. mii.o. , which is found up to the end of the thirteenth century, and back to the order mei.o. lod.o. , which is the usual order by the first decades of the fifteenth century (Lombard 1934). On the other hand, the combination of two third person clitics such as gliele (originally lile) did not undergo any change. This form was considered a single morphological unit that marked the presence of two third person pronouns. Santangelo and Vennemann (1976) argue that it is for this reason that the two clitics were not affected by the change in ordering. It should also be noted that in current Italian, the dative gli has the function both of masculine and feminine when it is in combination with an accusative third person clitic, which can be considered a fixed expression. I would like to suggest that this possibility for two clitics to form a unit, which is evident in the combination of two third person clitics, has also extended to the other combinations of clitics. Given the motivations presented above, I assume that if two or more clitics are present, they merge together into a new unit which adjoins to the verb. It is clear that previous accounts have failed in providing a principled explanation of clitic ordering: a templatic approach is simply a mere description of the facts. It might therefore be more reasonable to assume that the different combinations of clitics are new units, which are ‘stored’ as such in the lexicon. An additional advantage of the approach I propose is that it can provide a straightforward account of the morphophonological idiosyncrasies in which Italian clitics are involved. 3.5.2.2. The Romanian pronominal clitic cluster Additional motivations in favour of the assumption that the combination of two pronominal clitics constitute a unit come from Romanian. Recall that in Romanian only accusative and dative clitics are attested, therefore making an approach to pronominal clitic ordering in terms of template morphology rather unmotivated. As already mentioned, it could be argued that the Romanian clitic system is much richer and includes not only pronominal clitics, but also auxiliaries, 21

I refer to Monachesi (1996a) for an exhaustive list.

94

Morphology and its Interfaces

intensifiers, complementizers, and negation. It is therefore more like the clitic inventory of Balkan and Slavic languages than that of Romance languages. I postpone the discussion on whether this is indeed the case to Chapter 4. I claim that while the term clitic has been used to refer to all those elements in Romanian, matters are more complex since they do not all have the same status. I argue that an analysis of the ordering of the clitic cluster in terms of templates is not necessary and that an appropriate interaction of the various modules of the grammar provides better results. In the rest of this section, I therefore focus only on the order of pronominal clitics. It is interesting to observe that, as in other Romance languages, in Romanian as well certain changes occur in the combination of two pronominal clitics. This fact seems to suggest that pronominal clitics are not realized one at a time, but cluster together forming a unit. If clitics were realized one at a time, one would expect forms such as ne le or ne îl, le le where the dative form is the one that occurs either before a vowel or a consonant. However, this is not the case, as can be seen in the table below, which gives the possible combinations of clitics: (154) Possible combinations of two clitics

1 Dat Sg 2 Dat Sg 3 Dat Sg 1 Dat Pl 2 Dat Pl 3 Dat Pl Refl.

3 Acc Sg Ms mi-l t¸i-l i-l ni-l vi-l li-l s¸i-l

3 Acc Sg F mi-o t¸i-o i-o ne-o v-o le-o s¸i-o

3 Acc Pl Ms mi-i t¸i-i i-i ni-i vi-i li-i s¸i-i

3 Acc Pl F mi le t¸i le i le ni le vi le li le s¸i-le

It can be seen that there is a special form of the dative clitic ni, vi, li which surfaces only when it is in combination with another pronominal clitic. The regular form (i.e. ne, le) would not be possible in this case. The forms ending in i cannot be explained by means of appropriate phonological rules; they surface only in the presence of other object clitics. These data seem to support the hypothesis that the combination of two clitics constitutes a new unit and does not result from the composition of two single forms. A closer look at the table above reveals, however, that the third person feminine accusative clitic o constitutes an exception. The combination with another pronominal clitic results in the composition of the two forms: it is the regular form of the clitic which surfaces (i.e. ne) and not the special one (i.e. ni) which generally occurs in combination with another clitic.

Morphology and its Interfaces

95

It is interesting to note that the clitic o also exhibits idiosyncratic behaviour in the presence of auxiliary verbs. In this configuration, a combination of two pronominal clitics is forced to climb, attaching to the tense auxiliary (155a). However, if o is present, it does not cluster with the other object clitics, but attaches to the nonfinite verb (155b). The result is therefore an unusual configuration with split clitics: (155) a. Mi

l-a

dat. [ro] CL.DAT CL.ACC.has given ‘He has given it to me.’ b. Le-am dat-o. CL.DAT.have given.CL.ACC ‘I have given it to them.’

Despite this exception, there is additional positive evidence in favour of the idea that pronominal clitics form a unit, as shown by the data below. Romanian clitics can phonologically cliticize to a host different from the verb. The example below shows that the clitic can be enclitic on negation: (156) Nu-l

d˘a. [ro] gives ‘He does not give it.’ NEG.CL.ACC

In this case, the enclitic form -l is used and not the proclitic form îl. However, if negation is present, as well as two pronominal clitics, the clitics cluster together and not with negation: (157) a. *Nu-ne

îl

d˘a. [ro] gives b. Nu ni-l d˘a. NEG. CL.DAT.CL.ACC gives ‘He does not give it to us.’ NEG.CL.DAT CL.ACC

It should be noted that the dative form that surfaces is the one which occurs in combination with another pronominal clitic, that is ni, and not the one that occurs as enclitic (i.e. ne). The pronominal clitics also form a unit in a configuration in which they occur together with negation and an auxiliary verb: (158) a. *Nu-ne

l-a

dat. [ro] given b. Nu ni l-a dat. NEG. CL.DAT CL.ACC.has given ‘He hasn’t given it to us.’ NEG.CL.DAT CL.ACC.has

96

Morphology and its Interfaces

Evidence that the two pronominal clitics cluster together can be found in the fact that the dative form which surfaces is the one that occurs in combination with another clitic. It is not the case that one clitic is enclitic on negation and the other is proclitic on the verb. The most crucial evidence in favour of the hypothesis that pronominal clitics cluster together is provided by the combination of two clitics which are enclitic on the verb: (159) a. *D˘a-ne-l. [ro] give.CL.DAT.CL.ACC b. D˘a-ni-l. give.CL.DAT.CL.ACC ‘Give it to us !’ The same combination of clitics which surfaces before the verb also occurs after the verb and not the composition of the enclitic forms. However, the latter is what one would expect if clitics were realized one at a time. The evidence presented in this section indicates that Romanian clitics are not realized one at a time, but that they cluster together forming a new unit. The only exception might be represented by the clitic o. However, I show in Section 3.5.4 that the morphophonological analysis proposed can deal with its properties, while the next chapter contains an analysis of its interaction with auxiliary verbs. 3.5.2.3. The Portuguese pronominal clitic cluster Vigário (1999b) provides evidence that sequences of clitics in Portuguese should be considered as lexically formed clusters. She notes that clitics behave differently in similar phonological environments. As already discussed in Section 3.3.2.5, the combination of nos/vos with (l)o yields the forms no-lo and vo-lo, but the combination of lhes with (l)o yields the form lhos. Furthermore, in a sequence of clitics, the first of two adjacent vowels is always absent, as in the combination of the clitic me with o which yields the form mo. Similarly for the sequence te with o which yields to and the sequence lhe with o which yields lho. Vigário argues that these changes cannot be accounted for in terms of phonological rules. As additional evidence in favour of the fact that the clitic sequence acts as a unit, she mentions that the order of clitics is identical in proclitic and in enclitic position. 3.5.3. The morphophonological features

The conclusion I have reached in the previous section is that the combination of two clitics forms a unit. In Monachesi (1999a) I have claimed that this unit

Morphology and its Interfaces

97

is the spelling out of the featural information present on verbs. In particular, I propose that appropriate constraints relate the featural information present on verbs to the actual phonological realization of the clitic. They are sensitive to the morphosyntactic form of the verb. Clitics are therefore the spelling out of certain morphosyntactic features of the verb. However, before presenting the relevant constraints, I introduce the general architecture I have assumed in Monachesi (1999a) in order to deal with morphology and its interfaces. In particular, I introduce the relevant types and the appropriate features, the signature therefore, which shares some similarities with that proposed in Bird and Klein (1994). Following Bird and Klein (1994) and previous work in HPSG morphology, I suggest that words have more structure than is proposed in Pollard and Sag (1994). Therefore, the type word has MORPH as additional appropriate attribute, with value morph: (160) word MORPH

morph

The introduction of this attribute creates the necessary level to deal with morphological analysis. It will be shown that the existence of several levels of description within the word allows for an appropriate treatment of interface phenomena such as cliticization. The new type morph has two subtypes, which are complex-morph and basicmorph: (161)

morph HH H complex-morph basic-morph

The former plays a role in the analysis of morphologically complex words, as in the case of inflection and cliticization. The latter is employed in the analysis of morphologically simple words such as roots and uninflected words. There are certain features which are appropriate for the types introduced above. In particular, complex-morph has the attribute AFFIX as appropriate for it, while morph has the attribute STEM as appropriate, which is inherited by both of its subtypes (i.e. complex-morph and basic-morph). The type complex-morph thus has the features AFFIX and STEM as appropriate for it: (162) complex-morph AFFIX affix STEM stemorword While STEM is the only feature appropriate for basic-morph:

98

Morphology and its Interfaces

(163) basic-morph STEM stemorword STEM

has stemorword as value, which is a subtype of sign:

(164)

sign

H H

stemorword phrase H H stem word The subtypes of sign I propose are thus slightly different from those assumed in Pollard and Sag (1994), which I introduced in Chapter 2. In the case of inflection, the attribute STEM has stem as value, while in the case of cliticization its value is word. The same apparatus can therefore be employed to account both for inflection and for cliticization. This is a desirable result given the view which I defended in Section 3.3.2 that Romance clitics should be treated on a par with inflectional affixes. As for the type affix, it should be further partitioned in order to distinguish prefixes from suffixes:22 (165)

affix

H H

prefix suffix It should be noticed that the only appropriate attribute for affix is PHON: (166) affix PHON

phon

It is thus evident that, in this approach, morphemes do not exist as lexical entries, but represent the phonological realization of certain morphosyntactic properties of the host they combine with. There is therefore a similarity with realizational approaches to morphology such as those of Matthews (1972), Anderson (1992), and Stump (1992, 2001) in which morphemes simply represent the phonological spelling out of certain morphosyntactic properties of the verb. However, in this case, a feature is employed to spell out the phonological information and not a function. In this way, it is also possible to encode the prosodic properties of clitics and afﬁxes. The treatment presented here is slightly different from the one discussed in Section 3.2.2.1 by Riehemann (1993, 1998). Recall that in her analysis of -bar 22

This partition could be further extended to include also infixes and circumfixes, if necessary.

Morphology and its Interfaces

99

adjectives, -bar is neither a sufﬁx with its own lexical entry and subcategorization information, nor is it just phonological material added by lexical rule or introduced via a constraint as in my approach. Instead it is seen in terms of a schema, arising as a generalization about existing -bar adjectives in the lexicon. By contrast, in my approach, afﬁxes do exist in the lexicon but simply as phonological material. As already mentioned, the feature PHON allows for an appropriate representation of the various phonological properties of the relevant morphemes. However, contrary to what is assumed in Pollard and Sag (1994), this attribute should have a different value from a simple list of phonemes, since this is not enough to formalize phonological information. Thus, I follow Bird and Klein (1994) in assuming that the value of PHON is phon, which has certain appropriate features which are necessary to distinguish the segmental structure: (167) phon SKEL list of segments CONS list of consonants VOW list of vowels In particular, the feature CONS identifies the consonants while VOW identifies the vowels within a word. The feature SKEL has for value the list of segments present in a word. It should be noted that coindexation is employed to establish a link between the skeletal level and the vowels/consonants, as can be seen from the representation of a Romanian word such as copil ‘child’: (168)

⎡

⎤ phon ⎢ ⎥ ⎢ ⎥ ⎢CONS 1 k 3 p 5 l ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢VOW 2 o 4 i ⎥ ⎢ ⎥ ⎢ ⎥ ⎣ ⎦ SKEL 1 2 3 4 5

Bird and Klein (1994) introduce these features and the use of coindexation to show that an autosegmental analysis of Sierra Miwok data (Goldsmith 1990), in which there is an intercalation of vowels in a consonantal verbal stem, can be successfully modelled in HPSG. However, it will become clear in the following section that these features are also appropriate for dealing with certain morphophonological idiosyncrasies attested in Romanian cliticization. In order to encode prosodic information, it is necessary to distinguish (at least) the following constituents: syllable, foot, prosodic word, and phonological phrase. I represent them as subtypes of the type phon:

100

Morphology and its Interfaces

(169) phon

P P @ PPP @ P

syll

foot

p-word p-phrase

The partition above encodes a Prosodic Hierarchy similar to that proposed by Nespor and Vogel (1986), which I discuss in more detail in Chapter 5. In this section, I have sketched a general architecture for morphological analysis. Even though I have only focused on those aspects which are relevant for the analysis of cliticization, I believe that the framework proposed here can be easily extended to deal with inﬂection. I leave it to future research to investigate whether an approach along these lines might also be appropriate for derivation. The architecture sketched in this section is meant to be the beginning of a theory of morphology which can be compatible with a constraint-based grammar framework such as HPSG. It is clear that the approach presented here is strongly inﬂuenced by realizational approaches to morphology; however it diverges from them in some respects: I refer to Section 3.8 for a detailed comparison. At the beginning of this chapter, I presented several formalizations of morphology within HPSG. However, a shortcoming of those proposals is that they have not shown how morphology interacts with the other modules of the grammar. They have limited themselves to dealing only with the morphological aspects of a given phenomenon. I would like the theory introduced here to go beyond that and make claims about the way morphology interfaces with phonology and syntax within the HPSG framework. In the next section, I show that the hierarchy of types proposed, as well as their appropriate features, are well suited to deal with the morphological properties of Romance clitics. Furthermore, I provide evidence that the morphological theory proposed integrates successfully with both phonology and syntax, as shown by the comprehensive analysis of Romance cliticization presented in this chapter. 3.5.4. The spelling out of Romance pronominal clitics

In Section 3.4, I suggested that verbs which have undergone the lexical operation of cliticization are enriched with the relevant featural information. The crucial issue is how this information can be used in order to spell out the cliticized verb form in phonology. On the basis of the signature presented above, speciﬁc constraints can be proposed to relate the information contained in the CLTS list to the actual phonological realization of the clitic. In the rest of this section, I take Romanian as a test case in order to show the constraints that are responsible for the realization of the cliticized verb form. However, the approach presented here can be easily extended to other Romance languages. I refer to Monachesi (1999a) for an analysis of the mor-

Morphology and its Interfaces

101

phophonological properties of Italian clitics which is based on the account sketched here.23 I consider a sentence like the following in which the indirect object is cliticized: d˘a un buchet. [ro] (170) Maria îi Maria CL.DAT gives a bouquet ‘Maria gives him a bouquet.’ Recall that the Complement Cliticization Lexical Rule presented in Section 3.4 accounts for the fact that clitics satisfy the subcategorization requirements of the verb they are an argument of. Therefore, in the case of the sentence above, the verb d˘a subcategorizes for two complements, as can be seen in the following lexical entry (only the relevant part is shown): (171) COMPS NP acc , NP dat The effect of the rule is that the indirect object is removed from the COMPS list and is added as a member of the CLTS list: ⎡ ⎤ (172) ⎢VAL | COMPS NP acc ⎥ ⎢

⎥ ⎥ ⎢ ⎦ ⎣ CLTS NP dat [ 3M sg ] The necessary featural information is therefore available for the spelling out of the cliticized verb form in phonology. An implicational constraint like the following can be formulated: (173) Realization of the dative, third person masculine singular clitic ⎤ ⎡ ⎤⎤ ⎡ complex-morph prefix ⎥ ⎣ ⎢ ⎦⎦ ⎦ → AFFIX ⎣ ⎣ STEM | SS | [L | C | CLTS NP dat 3Sg PHON îi ⎡

The constraint applies to morphologically complex words and it states that if this word contains the information about a dative, third singular masculine element in the CLTS list, then the clitic îi must also be present in the structure. Certain generalizations with respect to the linearization of the clitics can be easily expressed within this system. For instance, appropriate constraints can account for the fact that Romanian clitics precede finite verbs while they 23 Some modifications would be necessary in order to adopt the analysis presented here to account for Portuguese clitics, given their special status. On the other hand, I expect that it should not be problematic to extend it to deal with Spanish and French cliticization. I refer to Section 3.5.5 for an alternative approach to French clitics.

102

Morphology and its Interfaces

follow gerunds and imperatives. In particular, a constraint such as (174) deals with proclitics. The phonology of the clitic indicated by tag 1 should precede that of the verb, indicated by tag 2 :24 ⎤ ⎡ (174) word ⎤⎥ ⎡ ⎢ ⎥ ⎢ complex-morph ⎢ ⎤⎥⎥ ⎡ ⎢ ⎢ → ⎥⎥ ⎢MORPH ⎢ verb ⎥⎥ ⎢ ⎥ ⎢ ⎦ ⎣ ⎦⎦ ⎣STEM word SS | L | C | H ⎣ VFORM fin ⎤ ⎡ 1 ⊕ 2 PHON ⎢ ⎤⎥ ⎡ ⎥ ⎢ ⎥ ⎢ complex-morph ⎥ ⎢ ⎡ ⎤ ⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎥ ⎢ word ⎥ ⎢ ⎢ ⎦⎥⎥ ⎢STEM ⎣ ⎥ ⎢ 2 ⎥ ⎢ ⎢MORPH ⎢ PHON ⎥⎥ ⎥ ⎢ ⎡ ⎤ ⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎥ ⎢ prefix ⎥ ⎢ ⎢ ⎦ ⎦⎥ ⎣AFFIX⎣ ⎦ ⎣ 1 PHON

A similar constraint can be formulated to deal with enclitics: (175)

⎡

word ⎡ ⎢ ⎢ complex-morph ⎢ ⎡ ⎢ ⎢ ⎢ ⎢ word ⎢ ⎢ ⎡ ⎢ ⎢MORPH ⎢ ⎢ ⎢STEM ⎢ verb ⎢ ⎢ ⎢ ⎣SS | L | C | H ⎣ ⎣ ⎣

⎤

⎤⎥ ⎥ ⎤⎥⎥ ⎥⎥ ⎥ ⎥→ ⎤⎥⎥ ⎥ ⎥⎥ ⎥ ⎥ ⎥

⎦⎥ ⎦⎦⎥ ⎦ VFORM nonfin ∨ imper

⎡

⎤

2 1 ⎢PHON ⊕

⎡ ⎤⎥ ⎢ ⎥ ⎢ complex-morph ⎥ ⎢ ⎡ ⎤ ⎢ ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎥ word ⎢ ⎢ ⎦⎥ ⎢STEM ⎣ ⎥⎥ ⎢ ⎥ ⎢MORPH ⎢ PHON 2 ⎥ ⎢ ⎥⎥ ⎢ ⎡ ⎤ ⎥⎥ ⎢ ⎢ ⎢ ⎥⎥ ⎢ suffix ⎢ ⎥⎥ ⎢ ⎦ ⎦⎥ ⎣AFFIX⎣ ⎣ ⎦ 1 PHON

24 The system allows for a more detailed formalization of the phonological properties of clitics. I refer to Monachesi (1999a) for an analysis of the phonological properties of Italian clitics discussed in Section 3.3.3.1.

Morphology and its Interfaces

103

The constraint above states that if the verb is nonﬁnite (thus also a past participle) or imperative the clitic follows it. This is indicated by the fact that the phonology of the clitic identiﬁed by the tag 1 follows that of the verb, indicated by the tag 2 . In this way it is possible to deal with the placement of Romanian clitics which is accounted for in the phonological component. On the other hand, their syntactic properties, that is the fact that they satisfy the subcategorization requirements of the verb, are handled by the lexical rule which affects the value of the syntactic features COMPS and CLTS. The morphophonological idiosyncrasies are dealt with by the spelling out mechanism, which is formalized in terms of implicational constraints that operate at the morphological and phonological level. The architecture of the sign interacts with the general devices of the framework to allow for an appropriate analysis of the various aspects of Romance special clitics. Going back to the Romanian cliticized form îi d˘a, it is licensed through the interaction of the lexical rule and the constraints presented above: (176) Realization of the cliticized verb form îi d˘a ‘he gives him’ ⎡ ⎤ word ⎢ ⎥ ⎢PHON îi d˘a ⎥ ⎢ ⎥ ⎢ ⎤ ⎡ ⎤⎥ ⎡ ⎢ ⎥ ⎢ ⎥ word ⎢ ⎥ ⎢ ⎥⎥ ⎢ ⎢ ⎥ ⎥ ⎢ ⎥ ⎢ ⎢ ⎥ d˘ a PHON ⎢ ⎥⎥ ⎢ ⎢ ⎥ ⎥ ⎢ ⎤ ⎡ ⎥ ⎢

⎢ ⎥ ⎢ ⎥⎥ ⎢ ⎢ ⎥ ⎥ ⎢ ⎥ ⎢ +fin HEAD verb ⎢ ⎥ ⎢STEM ⎢ ⎥⎥⎥ ⎢ ⎢ ⎥ ⎥ ⎢ ⎥ ⎢

⎥ ⎢ ⎢ ⎥ ⎢ ⎥⎥⎥ ⎢ ⎢1 ⎢ ⎥ ⎢ acc VAL | COMPS NP ⎥ ⎢ ⎥⎥⎥ ⎢ SS | L | C ⎢MORPH ⎢ ⎥ ⎥⎥ ⎢ ⎢ ⎢ ⎢ ⎥⎥ ⎢ ⎥⎥ ⎢ ⎢ ⎥ ⎢ ⎦⎦⎥ ⎣ ⎣ ⎢ ⎥⎥ ⎢ dat CLTS NP ⎢ 3Sg ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎢ ⎤ ⎡ ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎢ prefix ⎥⎥ ⎢ ⎢ ⎦ ⎣ ⎦⎥ ⎣ AFFIX ⎢ ⎥ PHON îi ⎢ ⎥ ⎢ ⎥ ⎣ ⎦ SYNSEM 1

The description above states that there is a stem, which is of type word. More specifically, this word is a verb which has one element in the COMPS list. This implies that it can still combine with a direct object. On the other hand, the information about the indirect object is contained in the CLTS list, as a result of the application of the lexical rule. The constraint in (173) triggers its spelling out as the clitic îi. Since the verb d˘a is a finite one, the clitic precedes it,

104

Morphology and its Interfaces

according to the constraint in (174). The approach which I have sketched can easily account for the different classes of clitics shown in the table in (101), repeated below: (101) Different classes of clitics 1

Person Case D

2 A

D A

3

D

Class 1 îmi m˘a ît¸i te îi Class 2 mi- m- t¸i- te- iClass 3 -mi -m˘a -t¸i -te -i

4 5 DA DA D

6

3,6 A A D A m. f. m. f. refl. îl o ne v˘a le îi le îs¸i se l- o- ne- v- le- i- le- s¸i- s-l -o -ne -v˘a -le -i -le -s¸i -se

In particular, if the host starts with a vowel and it is not an auxiliary, both the clitics from class 1 and those from class 2 can precede it, as can be seen from the examples in (97) repeated below: (97) a. M˘a

as¸teapt˘a. [ro]

CL. ACC.1SG waits

‘He waits for me.’ b. M-as¸teapt˘a. CL.ACC.1SG.waits A constraint like the following can account for this fact: ⎤ ⎡ complex-morph ⎤⎥ ⎡ ⎢ ⎥ ⎢ word ⎢ ⎡ ⎤ ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎥ ⎢ 2 ⎢ 1 ⎢ ⎢VOW a ∨ o ⊕ ⎥⎥ ⎥⎥ ⎢ ⎥ ⎢ ⎢ ⎥ ⎢PHON ⎢ ⎢ ⎥⎥ ⎥⎥ ⎢ ⎢ ⎣ ⎦ ⎥⎥ ⎢ → (177) ⎢ SKEL 1 ⊕ 3 ⎥⎥ ⎢STEM ⎢ ⎥⎥ ⎢ ⎥ ⎢ ⎡ ⎢ ⎢

⎤ ⎥ ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ HEAD verb AUX − ⎥ ⎢ ⎥ ⎢ ⎢ ⎢ ⎥ ⎥ ⎥⎥ ⎢SS | L | C⎢ ⎢ ⎦ ⎦⎥ ⎣ ⎣ ⎦ ⎣ CLTS NP acc 1Sg

⎡

⎡

prefix ⎣AFFIX ⎣ PHON

⎤⎤ ⎦⎦ m˘a ∨ m

It says that if there is a lexical verb which begins with either the vowel a or o and it has one element in the CLTS list which is the first singular accusative NP, this can be realized either as the clitic m˘ a or as m. The tag 1 identifies the

Morphology and its Interfaces

105

vowel as the first segment of the word while 3 identifies the list of remaining segments. The tag 2 stands for a list of vowels. A similar constraint can be formulated to account for the fact that if the verb begins with a consonant or a vowel which is different from a or o, only the clitic m˘a is possible, as shown by the following example: (178) a. M˘a

vede. [ro]

CL.ACC.1SG sees

‘He sees me.’ b. *M-vede. CL.ACC.1SG

sees

It should be noted that the combination of two pronominal clitics can be expressed by means of the same mechanism. The following is an example of the realization of the dative, first person plural clitic and the third person, accusative, masculine, plural clitic:25 ⎤ complex-morph

⎥ ⎢

(179) ⎣ ⎦→ STEM | SS | L | C | CLTS NP dat 1P l , NP acc 3M P l ⎤⎤ ⎡ ⎡ affix ⎦⎦ ⎣AFFIX ⎣ PHON ni-i ⎡

Recall that in this case it is not the form ne that surfaces, but the form ni. Given the present approach, it is quite straightforward to account for this fact. The information about the two elements is available at the same time and the combination of two clitics is treated just like a new form.26 It is also possible to deal with those cases in which the clitic o combines with another clitic. Recall that in this configuration the regular form of the dative clitic surfaces and not the one ending in i. However, since the information about both clitics is available, it is possible to express the fact that the dative, first person plural clitic and the third person, accusative, feminine, accusative clitic are realized as ne o: 25 Recall that implicational constraints are interpreted as stating that every object described by the antecedent will also have to satisfy the consequent in order to be grammatical. If the CLTS list contains two elements it will never satisfy the description of an antecedent which contains only one element. It is thus not possible to realize the clitics one at a time. 26 It is evident that an approach that realizes the clitics one at a time would run into a series of problems. For example, it should justify why forms such as ni, vi and li cannot occur alone and must be realized in combination with another clitic.

106

Morphology and its Interfaces

⎤ complex-morph

⎥ ⎢

(180) ⎣ ⎦→ STEM | SS | L | C | CLTS NP dat 1P l , NP acc 3F Sg ⎤⎤ ⎡ ⎡ affix ⎦⎦ ⎣AFFIX ⎣ PHON ne-o ⎡

To summarize: the approach presented here can easily deal with the various morphophonological idiosyncrasies in which combinations of clitics are involved as well as with the ordering of the clitics both among each other and with respect to the host. It might be argued that the analysis proposed can merely describe the facts without providing a deeper understanding of the underlying mechanisms. I believe that the data I have discussed cannot be explained in a principled way. Previous accounts have failed to provide an explanatory analysis of clitic ordering: a templatic approach is simply a mere description of the facts. This also seems to be the case for accounts cast in the Chomskyan tradition, where appropriately ordered functional projections are employed, as well as for analyses within Optimality Theory, where alignment constraints of the EDGEMOST family simulate templates. This might suggest that clitic ordering cannot be explained in a principled way. It seems more reasonable, therefore, to assume that the different combinations of clitics are new units, which are ‘stored’ as such in the lexicon. The spelling out approach I have proposed is superior to OT analyses of the same data such as those of Grimshaw (1997) and Anderson (1996), which are described in Section 3.10. This is because the spelling out approach has a broader empirical coverage and it can deal not only with clitic ordering, but also with the morphophonological idiosyncrasies and the arbitrary gaps in which pronominal clitics are involved. Incidentally, the data considered in this section have not been addressed by most analyses of Romanian pronominal clitics which have not even reached descriptive adequacy. It may be the case that, from a linguistic point of view, it might be much more appealing to assume that the clitic cluster consists of individual clitics and to try to explain the irregularities. However, a serious attempt in this direction is bound to assume ad hoc morphophonological rules which apply only in the case of clitics and which do not extend to other languages. This approach would ultimately be no more explanatory than one which assumes that the combinations of clitics are new units. I believe that the fact that children do not make mistakes in producing clitic clusters while second language learners have serious difficulties in this respect might suggest that indeed there are no rules underlying the combinations of clitics and they should be learned as a whole.

Morphology and its Interfaces

107

Before concluding this section, I would like to consider a possible objection to this analysis. It could be argued that an approach in terms of inheritance hierarchies might constitute an alternative solution to the problem represented by the formation of the clitic cluster. After all, I have claimed that inheritance hierarchies could be viewed as one of the most interesting contributions within HPSG morphology. At first sight, a certain regularity can be observed in the combinations of clitics and it is exactly this regularity that is lost in an approach expressed in terms of implicational constraints. In contrast, an analysis in terms of inheritance hierarchies might indeed be able to express the general properties of the individual clitics which would then be inherited when they combine with each other. I believe, however, that the common properties are only apparent and that any attempt to express generalizations in the combinations of clitics would eventually fail given the additional idiosyncrasies that these combinations are subject to. 3.5.5. An alternative realizational approach

The analysis of Romance cliticization I have proposed in this chapter shares some insights with the approach that Miller and Sag (1997) have proposed for French clitics. In both proposals, clitics are considered lexically attached inﬂectional afﬁxes that satisfy the subcategorization requirements of the verb they are an argument of. In both approaches, the morpho(phono)logical aspect of cliticization is taken into consideration and plays a relevant role. The two approaches, however, differ significantly in the way these basic insights are formalized. In particular, there are differences in the hierarchies of the types proposed and in the way word formation operates as well as in the way morphology interacts with the other grammatical levels. In the following section, I briefly sketch the analysis of French clitics proposed by Miller and Sag (1997) and I highlight the differences and similarities to the analysis I have presented in this chapter. 3.5.5.1. The analysis of French clitics Miller and Sag (1997) propose a lexicalist analysis of French clitics which is based on the assumption, defended at length in this chapter, that French clitics are inflectional affixes. In particular, they suggest that verb forms bearing pronominal affixes should be formed in the lexicon and not in syntax. They claim that each inflected word must belong simultaneously to three compatible types, that is a (CLITIC)-REALIZATION (REALZN) type, an INFLECTIONAL (INFLN) type, and a LEXEME type, as can be seen in Fig. 3.2. The REALZN type is divided into various subtypes which distinguish cliticized words (and their subtypes subject clitic word and non-subject clitic word) from plain words.

108

Morphology and its Interfaces word REALZN cl-wd

pl-wd

su-cl-wd

ns-cl-wd

INFLN fin-vb indic-vb

sbjnctv-vb

LEXEME ... ...

...

... LAVER

VOULOIR

FIGURE 3.2 The word hierarchy

The

type specifies an inflectional form for a given lexeme, while the type specifies the morphological stem, part of speech, and meaning which is common to a family of inflected forms. Certain constraints are associated with each type. In particular, objects of type plain word require each element of a verb’s ARG-ST list to correspond to an overt phrase that combines with the verb syntactically. The following constraint is associated with it: ⎤ ⎡ ⎡ ⎤ 0 FORM ⎥ ⎢ ⎦ ⎥ ⎢MORPH ⎣ ⎥ ⎢ 0 I-FORM ⎥ ⎢ ⎥ ⎢ ⎡ ⎡ ⎤ ⎤ ⎡ ⎤ ⎥ ⎢ ⎥ 2 (181) pl-wd ⇒ ⎢ ⎢ ⎢ SUBJ ⎦⎥⎥⎥ ⎢ ⎣ ⎥ ⎢ ⎢ VAL ⎢ ⎥ ⎥ ⎢ ⎢ 3 ⎥⎥⎥ ⎥ ⎢SS⎢L | C ⎢ COMPS ⎢ ⎥ ⎥ ⎢ ⎣ ⎣ ⎦⎦⎥ ⎦ ⎣ ARG-ST 2 ⊕ 3 INFLN

LEXEME

It states that plain words have an ARG-ST list corresponding to the concatenation of the values of the Valence Features. On the other hand, cliticized words are subject to the following constraint: ⎤ ⎡ ⎡ ⎤ 0 FP RAF ( ,...)⎦ ⎥ ⎢MORPH ⎣FORM ⎥ ⎢ 0 ⎥ ⎢ I-FORM ⎥ ⎢ ⎤⎥ ⎢ ⎡ ⎡ ⎤ ⎥ ⎢ HEAD verb ⎥ ⎢ ⎢ ⎥ ⎢ ⎡ ⎤ ⎢ ⎥⎥⎥ (182) cl-wd ⇒ ⎢ ⎢ ⎥ ⎢ ⎥ 2 ⎥ ⎢ ⎢ SUBJ ⎢ ⎥⎥ ⎥⎥ ⎢ ⎢ ⎣ ⎦ ⎢ ⎥ VAL ⎥ ⎥ ⎢SS ⎢L | C ⎢ 3 list(non-aff) ⎥ ⎥ ⎢ ⎢ COMPS ⎢ ⎥⎥ ⎥ ⎢ ⎢ ⎣ ⎦⎦⎥ ⎦ ⎣ ⎣ ARG-ST ( 2 ⊕ 3 ) nelist(aff) It requires that cliticized words have at least one argument that is realized affixally rather than syntactically. This is accounted for by having one or more elements of type aff as members of the ARG-ST list. These elements must be

Morphology and its Interfaces

109

absent from the Valence Feature lists and they are realized as pronominal affix by means of the function FPRAF . It should be noted that the type aff is a subtype of synsem which, according to Miller and Sag (1997), is classified into various subtypes, as can be seen in Fig. 3.3. The type canon(ical-synsem) is associated with all signs, while noncan(onical synsem) corresponds to an ARG-ST position that is not realized as a local syntactic dependent of the head. The latter is further divided into aff(ixal-synsem) and gap(-synsem). The presence of elements of type aff on the ARG-ST of a verb triggers the morphological realization of the relevant clitics. The additional partition distinguishes afﬁxes in overt anaphors (a-aff) and personal pronouns (p-aff). The interaction of the various constraints gives rise to the following description for the plain word donne ‘gives’: ⎤ pl-wd & DONNER & 3sg-pres-indic-vb ⎥ ⎢ ⎥ ⎢ FORM donne ⎥ ⎢MORPH ⎥ ⎢ I-FORM donne ⎥ ⎢ ⎥ ⎢ ⎤ ⎡ ⎥ ⎢ ⎥ ⎢ verb ⎥⎥ ⎢HEAD ⎢ ⎥ ⎢ ⎥ ⎢ VFORM indic (183) ⎢ ⎥⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎡ ⎤ ⎥⎥ ⎢ ⎢ 1 ⎥ ⎢ ⎥ ⎢ SUBJ ⎥⎥ ⎢SS | L | C ⎢VAL ⎣ ⎦ ⎥ ⎢ ⎥ ⎢ ⎥⎥ ⎢ ⎢ COMPS 2 , 3 ⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎥ ⎢ ⎣ ⎣ 1 NP[3sg] , 2 NP[acc] , 3 NP[à ] ⎦⎦ ARG-ST 1 ⎡

synsem CANONICITY

AFFIXALITY

noncan canon

non-aff gap

aff a-aff

FIGURE 3.3 The synsem hierarchy

p-aff

110

Morphology and its Interfaces

It states that donne is a plain word and that it must satisfy the constraints that are associated with the lexeme donner, as well as those related to the subtype 3sg-pres-indic-vb. In particular, it has three elements in the ARG-ST list which correspond to those present in the valence lists. In this way, it is possible to express generalizations at the level of the higher types and let the specific entry inherit the relevant properties. As for a cliticized verb form such as lui donnera, it has the COMPS list reduced while the ARG-ST list contains an aff element which is realized as the clitic lui: ⎤ ⎡ ns-cl-wd & DONNER & 3sg-fut-indic-vb ⎥ ⎢ ⎥ ⎢ FORM lui-donnera ⎥ ⎢MORPH ⎥ ⎢ I-FORM donnera ⎥ ⎢ ⎥ ⎢ ⎤ ⎡ ⎥ ⎢ ⎥ ⎢ verb ⎥⎥ ⎢HEAD ⎢ ⎥ ⎢ ⎥ ⎢ VFORM indic (184) ⎢ ⎥⎥ ⎢ ⎥⎥ ⎢ ⎢ ⎡ ⎤ ⎥⎥ ⎢ ⎢ 1 ⎥⎥ ⎢ ⎢ SUBJ ⎥⎥ ⎢SS | L | C ⎢VAL ⎣ ⎦ ⎥⎥ ⎢ ⎢ 2 ⎥⎥ ⎢ ⎢ COMPS ⎢ ⎥ ⎢ ⎥ ⎥⎥ ⎢ ⎢ ⎦ ⎣ ⎣ 1 NP[3sg] , 2 NP[acc] , NP[p-aff,à ,3sg] ⎦ ARG-ST 1 This proposal shares, with the other morphological analyses discussed in this chapter, the property that hierarchies and multiple inheritance are employed to structure the lexicon: the hierarchy introduced allows for a representation of linguistically relevant subclasses of words. As in my approach, and unlike in Riehemann’s analysis, a general architecture for morphological analysis is suggested. As for the analysis of cliticization, there is a clear difference between the treatment I have presented in section 3.4 and the proposal of Miller and Sag: in the former a lexical rule is employed to deal with the syntactic aspects of cliticization, while in the latter constraints on types are used. In my approach, the Complement Cliticization Lexical Rule is employed to establish a relation between cliticized and non-cliticized verb forms. On the other hand, in Miller and Sag’s proposal the same insights are expressed by using the types pl-wd and cl-wd: they are subject to specific constraints that correspond roughly to the IN and OUT of the lexical rule.27 27

Lexical rules are not completely abandoned in the analysis of cliticization proposed by Miller and Sag (1997) since they are maintained to deal with the properties of the clitic en and for the treatment of the quantifier tous. Miller and Sag give no motivation for why constraints on types are used in one case while lexical rules are used for the others.

Morphology and its Interfaces

111

However, this approach requires a distinction of (morphosyntactic) words into plain and cliticized, which is not a natural one, at least for French. This is because only verbs can belong to the class of cliticized words, while any category can be a member of the plain word class. In the case of cliticization, the crucial opposition is between cliticized verbs and non-cliticized ones, but this distinction is not reﬂected in the proposal by Miller and Sag.28 In addition cliticized verbs and non-cliticized ones differ only in the way their complements are realized, but otherwise they share exactly the same properties. On the other hand, the lexical rule approach I have proposed allows a relation between classes of words to be established, such as those represented by cliticized and non-cliticized verbs: this is exactly what one wants to capture in the case of cliticization. The morphological realization of clitics is accounted for in the proposal of Miller and Sag by a three argument function FPRAF that constrains the FORM values of words: ⎡ ⎡ ⎤⎤ 0 , 1 , 2) FORM F ( P RAF ⎢MORPH ⎣ ⎦⎥ ⎥ ⎢ 0 ⎥ ⎢ I-FORM ⎥ ⎢ ⎤ ⎡ ⎢ ⎡ ⎤ ⎥ (185) cl-wd ⇒ ⎢ ⎥ 1 ⎥ ⎢ HEAD ⎥ ⎢ ⎢ ⎦⎦ ⎥ ⎣L | C ⎣ ⎦ ⎣SS 2 aff-list ARG-ST The first argument of the function is the value provided by the inflectional type, the second is the value of the HEAD feature of the verb, and the third is the value of the ARG-ST. This approach is clearly inspired by the Paradigm Function Theory presented in Stump (1992). Both the analysis of Miller and Sag and the treatment presented in Section 3.5.4 are thus inspired by a realizational approach to morphology in that clitics are not considered signs, but simply the spelling out of certain morphosyntactic features of the verb. However, in the proposal by Miller and Sag, the relevant information which plays a role in the realization of the clitic is encoded in ARG-ST, while in my approach, the feature CLTS is introduced. At first sight this solution might not be very economical, but this is not the case since the CLTS feature serves different purposes. Besides encoding information that is relevant for the spelling out of the cliticized verb form, it 28 It should be mentioned that this distinction has been abandoned in subsequent work such as that of Abeillé and Godard (2002) and it has been replaced by a distinction between plain verbal words (without clitics) and cliticized verbal words.

112

Morphology and its Interfaces

is used in the analysis of clitic climbing as will become clear in the following chapter. It is employed to identify those verbs that have already combined with clitics and those that have not. In the proposal by Miller and Sag, new types are introduced to achieve the same result: basic verbs to which non-cliticized verb forms belong and reduced verbs which include cliticized verbs. In their approach, the information that we have a cliticized verb form is encoded twice: in the subtyping of word and in that of verb. A certain amount of redundancy is thus introduced. Additional motivation in favour of the use of the CLTS feature is provided by Romanian clitics and their property to attach phonologically to categories which are different from the verb, as discussed in section 3.3.2. Given the analysis of cliticization presented in Section 3.4, it is possible to percolate the information related to the clitics, which is contained in CLTS, on to the relevant host. Appropriate constraints are responsible for the spell out of the cliticized form. However, this solution is not available if the feature ARG-ST is used instead; it is not desirable to have the various phonological hosts of the clitic to inherit the ARG-ST of the verb. It should be noted that the fact that Romanian pronominal clitics can combine with categories other than the verb does not argue in favour of the division proposed by Miller and Sag in terms of pl-wd and cl-wd. Their distinction is of a morphosyntactic nature since each inflected word must belong to one of the two types: according to them, objects of type cl-wd are verbs that have at least one argument that is realized affixally, rather than syntactically. On the other hand, the attachment of Romanian clitics to different categories is a phonological matter, as can be seen in the examples (54), repeated below: (54) a. Mihai nu-l as¸teapt˘a. [ro] Michael not.CL.ACC.3MSG waits for ‘Michael doesn’t wait for him.’ b. Cred c˘a-l vede. believe that.CL.ACC.3MSG sees ‘I believe that he sees him.’ c. Maria-i scrie des. Maria.CL.DAT.3SG writes frequently ‘Maria writes to him frequently.’ d. Unde-mi lasa fratele tau cheia. Where CL.DAT.1SG leaves brother yours key the ‘Where does your brother leave me the key.’ They seem to be cases in which the clitic attaches as enclitic to the word which precedes the verb and they are thus a remnant of the possibility that clitics

Morphology and its Interfaces

113

had in Old Romance to follow as enclitic the first word of the Phonological Phrase. As already mentioned, the order of clitics in French can be expressed, in the general case, in terms of a template like the following: (186)

Position: I

II

III IV V VI

me le lui te la leur NOM nous les y en vous se Miller and Sag deal with the French clitic ordering by assuming that the FORM values are objects of the type cl-fm. This has two subtypes: procl-fm and enclfm. Appropriate features are BASE, whose value is an inflected form, and seven slot features, whose values are pronominal clitics:29 ⎤ ⎡ (187) cl-fm ⎥ ⎢ ⎥ ⎢BASE infl-form ⎢ ⎥ ⎥ ⎢ ⎢SL-1 je, tu, il, elle, on, nous, vous, ils, elles, [ ] ⎥ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢SL-2 me, te, nous, vous, se, [ ] ⎥ ⎢ ⎥ ⎢ ⎥ ⎢SL-3 le, la, les, [ ] ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ lui, leur, [ ] ⎥ ⎢SL-4 ⎥ ⎢ ⎥ ⎢ ⎥ ⎢SL-5 moi, toi, nous, vous, [ ] ⎥ ⎢ ⎥ ⎢ ⎥ ⎢SL-6 y, zy, [ ] ⎥ ⎢ ⎦ ⎣ en, zen, [ ] SL-7 It is thus possible to provide a formalization of the template presented above. The approach that Miller and Sag adopt with respect to clitic ordering shares insights with template morphology since clitics belong to different slots. However, a shortcoming of this approach is that it predicts several clitics to occur attached to the verb while usually, in French, a clitic cluster is made up of two clitics. This is one of the reasons for not adopting the templatic approach in the analysis I have developed in this chapter. 29 Subject clitics and object clitics receive similar analysis in Miller and Sag (1997) (cf. also Kayne (1975)). They are both considered pronominal clitics and they are both included in the template which deals with the ordering of French clitics. Perlmutter (1971) also includes subject clitics in his template as well as the negative clitic ne.

114

Morphology and its Interfaces

Another problem is that a templatic approach does not offer any insights on how to deal with the kind of morphophonological idiosyncrasies that are exhibited by Romance clitics. Miller and Sag do not provide details on how their approach could be extended to account for the type of allomorphy, idiosyncratic vowel deletion, and sequencing of identical clitics which I have discussed in Section 3.3.2. As already mentioned, it is the function FPRAF that accounts for the morphological realization of the clitics. Miller and Sag deﬁne it in the following way, where X is an inﬂected form, Y is a HEAD value and Z is an argument structure list: (

(188) FPRAF (1)

(2) =

(3) =

=X encl-fm

X

BASE X

procl-fm

, if Y =

, if Y =

BASE X

,

Y,

VFORM

VFORM NEG

Z

)

=

W,

where

W

past-p , imp , –

, otherwise.

The function accounts for the fact that no clitics can combine with a past participle, as well as for the fact that enclitics combine with imperatives and proclitics with the other verbal forms. Furthermore it interacts with a number of further constraints on W, among which are the constraints that determine the form of subject clitics in SL-1 and those that are relevant for the realization of non-subject clitics, which are shown below: (189)

W’s

value for the feature F is v, just in case Z contains α and W satisfies

ω, where:

F C2a: SL-2 C2b: SL-2 C2c: SL-2 C2d: SL-2 C2e: SL-2 C3a: SL-3 C3b: SL-3 C3c: SL-3

C4a: SL-4 C4b: SL-4

v me te nous vous se la le les lui leur

α

[aff, 1sg, obj] [aff, 2sg, obj] [aff, 1pl, obj] [aff, 2pl, obj] [a-aff, 3, obj] [p-aff, 3sgf, acc] [p-aff, 3sgm, acc] ∨ [p-aff, +PRED] [p-aff, 3pl, acc] [p-aff, 3sg, à1 ] [p-aff, 3pl, à1 ]

ω

[procl-fm] [procl-fm] [procl-fm] [procl-fm] [procl-fm]

Morphology and its Interfaces F C5a: SL-5 C5b: SL-5 C5c: SL-5 C5d: SL-5

C6a: SL-6 C6b: SL-6 C7a SL-7 C7b SL-7

v moi toi nous vous y zy en zen

α

[aff, 1sg, obj] [aff, 2sg, obj] [aff, 1pl, obj] [aff, 2pl, obj] [p-aff, à2 ] [p-aff, à2 ] [p-aff, de ] [p-aff, de ]

115

ω

[encl-fm] [encl-fm] [encl-fm] [encl-fm] [procl-fm] [encl-fm] [procl-fm] [encl-fm]

A shortcoming of the approach to cliticization proposed by Miller and Sag is that they do not spell out the relation between morphology and phonology. They say that the phonology of clitic words should be determined by FPRAF ; however, it is obvious that this function determines the morphological form of the cliticized verbs and not their phonological form. Similarly, matters of enclisis and proclisis are treated as a morphological effect, while usually this is considered a phonological property of clitics. It seems, therefore, that while they claim that their approach is inspired by realizational morphology, it goes against its spirit. Recall that in realizational morphology, morphemes represent the phonological realization of certain morphosyntactic properties of the host they combine with. This is not the case in the approach proposed by Miller and Sag since phonology plays no role at all. By contrast, in the approach presented in Section 3.5.4, the implicational constraints establish the appropriate relation between the morphosyntactic features present on the verb and the phonological realization of the clitics.

3.6. Germanic clitics In this chapter, I have proposed an analysis of Romance clitics which is based on the assumption that they are affixal elements and I have argued for a lexical analysis to deal with their properties. One might wonder whether this analysis could be extended to deal also with Germanic clitics, given that in a series of studies these elements have been assimilated to their Romance counterparts, both being considered as heads (Cardinaletti and Roberts 1993), (Corver and Delfitto 1999), (Haegeman 1992), (Haegeman 1993), (Penner 1991), (Zwart 1993). Cardinaletti (1992) argues that Germanic weak pronouns pattern like Romance clitics in having a special distribution and a fixed position in the clause. It should be mentioned, however, that this position is different in Romance (i.e. verb) and in Germanic (preceding sentence adverbials and following the subject), as can be seen in the German (190a) and Dutch (190b) examples

116

Morphology and its Interfaces

below from Haverkort (1994):30 (190) a. Was hat die Mutter’m gegeben? [de] what has the mother him given b. Wat heeft moeder’m gegeven? [nl] what has mother him given ‘What has mother given to him?’ Furthermore, Cardinaletti argues that pronoun movement in Germanic cannot be accounted for in terms of scrambling or XP-movement and that an analysis in terms of head movement seems more plausible. She shows that, indeed, Germanic weak pronouns have the same syntactic properties as Romance clitics. In particular, they cannot be focused, modiﬁed, or conjoined and they form a cluster with a fixed order. If Germanic weak pronouns are assimilated to the Romance clitics, a natural question is why pronoun movement is obligatory. Cardinaletti suggests that this is the case because weak pronouns must be ‘licensed’ by a functional head: in Romance the clitics adjoin to the head which contains the inﬂectional morphemes, while Germanic weak pronouns are cliticized on an empty head (cf. also Zwart (1991) for Dutch, Cardinaletti and Roberts (1993) for German and Deprez (1990) for Scandinavian languages). She stipulates thus the introduction of the projection AGR1P, which is meant to host weak pronouns. As argued by Haverkort (1994), no independent motivation is given for the introduction of AGR1P except that it acts as a host for clitics. Furthermore, if West-Flemish is taken into consideration, in which the clitics can appear in several positions, it is necessary to assume unrestrictive AGR-recursion, as in Haegeman (1992). Haverkort argues that a head movement analysis of Germanic clitics is untenable since they behave very differently from their Romance counterparts in several respects. In particular, they can appear independently of the verb, they can license parasitic gaps, and, in addition, clitic clusters can be split up. Furthermore, they can interact with scrambled NPs, they can appear in positions in coordinated structures that are inaccessible to heads, and they can move along with Verb Projection Raising in certain dialects. He claims that these properties are easier to account for if Germanic clitics are considered maximal projections. He adopts the system of syntactic projections proposed in Muysken (1982), which is based on features such as [+/− projection] and [+/− maximal], to deal with the hybrid character of Germanic clitics and to provide an explanation of why they move when they are maximal projections. 30 German differs from the other Germanic languages in that it allows weak pronouns to follow the complementizer and to precede the subject.

Morphology and its Interfaces

117

Haverkort (1994) proposes that Germanic clitics should be characterized by the feature [+ maximal] which explains that these elements can adjoin to maximal projections and that they cannot have complements. On the other hand, the feature [− projection] allows clitics to be listed in the lexicon and to have a number of head-like features, similar to their Romance counterparts which have the features [− maximal], [− projection]. I share with Haverkort the view that Germanic clitics form a hybrid category which unites properties of heads and maximal projections. Therefore, I do not claim that the analysis proposed for Romance clitics in terms of affixes should be extended to them. At first sight, it seems that these clitics share some of the properties of the Italian bisyllabic clitic loro (Cardinaletti 1991). In Monachesi (1994) and Monachesi (1999a), I have argued that loro does not have affixal status, contrary to the other Italian (monosyllabic) clitics, and should be considered a maximal projection marked with the feature [+ CL]: its placement is thus dealt with in syntax. Furthermore, the interaction of morphology and syntax allows for a proper treatment of the similarities and differences between loro and the Italian monosyllabic clitics. I believe that a similar difference characterizes Romance and Germanic clitics: while the former are afﬁxal elements and their properties are accounted for through the interaction of morphology with the other modules of the grammar, this is not the case for Germanic clitics. They are maximal projections which are marked in a special way and should thus be dealt with in syntax.

3.7. Slavic clitics In this section, I briefly address Slavic clitics to show that they exhibit different properties from the Romance clitics discussed in this chapter. Unlike Romance clitics, they occur in second position in the clause and therefore the analysis presented here does not carry over to them naturally.31 The only exceptions are the verbal clitics attested in Bulgarian and Macedonian, which share several properties with Romanian and with Romance clitics, as shown in more detail in the next chapter. However, a brief overview of the properties of Slavic clitics is relevant in this context because they provide additional evidence that a comprehensive analysis of cliticization can be achieved only through an appropriate interaction of the various modules of the grammar. In particular, I take into consideration the analysis proposed in Boˇskovi´c (2001) as representative of this view: in the speciﬁc case of Slavic clitics, the locus of the analysis is not the interface between morphology and the other modules of the grammar as for Romance 31

I refer to Franks and King (2000) for a general survey of clitic systems in Slavic languages.

118

Morphology and its Interfaces

clitics, but that between syntax and phonology. In this section, I anticipate the discussion on the interface between syntax and phonology which is the topic of Chapter 5.32 Boˇskovi´c takes cliticization in Serbo-Croatian as a case study to investigate the second position properties of Slavic clitics: (191) Mi smo mu je predstavili juˇce? [sc] we CL.AUX CL.DAT CL.ACC introduced yesterday ‘We introduced her to him yesterday.’ In the example above there is a sequence of three clitics, that is the auxiliary clitic smo and the two object clitics mu and je, which occur in the second position in the clause, after the subject mi. Locating the clitics in other positions or splitting the clitic cluster would lead to ungrammaticality. Boˇskovi´c discusses four approaches to deal with the placement of these clitics which have been proposed in the literature, that is the strong syntax approach, the strong phonology approach, the weak syntax approach, and the weak phonology approach. The strong syntax approach assumes that syntax is completely responsible for second position cliticization in Serbo-Croatian and clitics do not move at PF. Works that represent this view include: Dimitrova-Vulchanova (1995), Franks (1997, 1999), Franks and Progovac (1994), Progovac (1996), Rivero (1997), Tomi´c ´ (1996), and Wilder and Cavar (1994). By contrast, in the strong phonology approach, it is phonology that is fully responsible for second position cliticization and the clitics move to this position at PF (Radanovi´c-Koci´c 1988). According to the weak syntax approach proposed by Halpern (1992, 1995), it is claimed that most movements of the clitics take place in the syntax, even though clitics are allowed to undergo phonological movement in certain well-defined conﬁgurations (cf. also King (1996) and Schütze (1994)). After discussing the advantages and disadvantages of these approaches, Boˇskovi´c provides evidence that the second position effect is phonological in nature and thus argues in favour of the weak phonology approach. Boˇskovi´c modiﬁes the analysis of Klavans (1985) and proposes that syntax is responsible for the placement of the clitics while phonology acts as a ﬁlter inducing the second position effect. Furthermore, following Stjepanovi´c (1998b, 1998a), he assumes that each pronominal clitic is located in the head of separate AGR projections which are hierarchically arranged. Boˇskovi´c argues against a templatic approach to clitic ordering and claims that his analysis is more principled than a templatic one. However, it seems to 32 In this section, Slavic cliticization is taken as a case study, while in Chapter 5 I address the relevant issues taking Italian restructuring verbs into consideration.

Morphology and its Interfaces

119

me that an approach which postulates slots in a template is equivalent to one which assumes an ordered hierarchy of functional projections. Clitics are ordered in syntax even though they do not have to cluster together in the syntax in the same head position, as in most syntactic accounts of clitic placement in Serbo-Croatian. Boˇskovi´c proposes that clitics are located in the second position of the Intonational Phrase (Nespor and Vogel 1986) in which the syntax places them, implementing in this way the second position requirement. Furthermore, a modified version of the notion of Morphological Merger (Marantz 1988, 1989) is assumed which, however, in this analysis takes place in PF under PF adjacency. The two adjacent elements merge and they take over the requirements of both, thus forming a single word. Clitics must cluster in the phonology. Boˇskovi´c does not, however, specify what is the phonological nature of this cluster, that is whether these elements form a prosodic word or not. The analysis relies heavily on the Copy Theory of movement (Chomsky 1993), which is the theoretical mechanism that allows Boˇskovi´c to capture phonological effects on word order without having to postulate movement. Under this view, movement leaves behind a copy of the moved element rather than a trace. While at LF there is a choice in deciding where deletion should take place in a non-trivial chain, it is standardly assumed that this choice is not available in PF. However, several authors have argued that in PF there is also a choice concerning which member of a chain survives deletion, Boˇskovi´c follows Franks (1998) who proposes that there is a preference to delete the elements in the tail position unless this would lead to a violation. Thus, PF affects word order by determining which copy of a non-trivial chain is pronounced and through a ﬁltering effect on the output of the syntax. Boˇskovi´c shows that this mechanism provides an elegant account of the properties of Serbo-Croatian clitics and it can explain the differences in clitic placement in Polish and Slovenian. Furthermore, he shows that the mechanism of pronunciation of the lower copies can be employed to deal with the relevant facts in the analysis of cliticization in Bulgarian and Macedonian which have provided evidence for PF movement in the literature. In Bulgarian, clitics occur as enclitics (192a), (192b), but they do not surface in second position (192c): go Petko vc¸era. [bg] (192) a. Dade mi gave CL.DAT CL.ACC Petko yesterday b. *Mi

go

dade Petko vc¸era. Petko yesterday

CL.DAT CL.ACC gave

‘Petko gave it to me yesterday.’

120

Morphology and its Interfaces c. sˇe Petko mi go dade vc¸era. that Petko CL.DAT CL.ACC gave yesterday ‘that Petko gave it to me yesterday.’

On the other hand, Macedonian clitics are proclitics and they can occur in sentence-initial position. Boˇskovi´c proposes an analysis in which a copy of pronominal clitics is present both above and below the verb: (193) a. X clitic V clitic b. clitic V clitic In Bulgarian, if there is lexical material preceding the clitic, it is the head of the chain of clitic movement which is pronounced. However, if nothing is present before the verb, it is the tail of the chain which is pronounced, since in this way one avoids a PF violation. In Macedonian, on the other hand, it is always the head of the chain which is pronounced, since the clitics are proclitics in this language. Even though the Copy Theory of movement combined with the deletion of the appropriate copy properly accounts for the data, one might wonder why there is a requirement to move at all in the syntax if, in the end, phonology prefers the version in which the element has not moved. Boˇskovi´c concludes that he can account for South Slavic cliticization in a derivational model in which syntax feeds phonology, as opposed to Zec and Inkelas (1990) who argue for a non-derivational, co-presence model in which syntax and phonology can feed each other. In addition, he claims that there is no need to look ahead from the syntax to the phonology and that PF movement is not required. His analysis gives support to his view that South Slavic cliticization does not provide evidence for non-standard claims about the nature of the syntax–phonology interface, while this has often been the case in the literature. I believe that a disadvantage of the model proposed in Boˇskovi´c (2001) is that syntax overgenerates and the inappropriate structures are ﬁltered out in phonology. He does not say, however, how ﬁlters should be formally interpreted in his theory. By contrast, in HPSG there is no need to repair in phonology what syntax has produced, since the various constraints all interact to produce the appropriate representation. To conclude: second position cliticization has been the topic of much discussion in the linguistic literature. This is not surprising as the solution to the problem can shed light on relevant theoretical issues such as the nature of the syntax–phonology interface. The analysis of Slavic clitics can help us to understand whether the interface is derivational or co-present, non-derivational or with bidirectional interaction between the syntax and phonology. Ultimately,

Morphology and its Interfaces

121

it will lead to a better understanding of the way the grammar is organized as we have also seen through the study of Romance cliticization, to which I return in the following section.

3.8. A brief review Research within HPSG has initially focused on syntax and semantics largely to the exclusion of morphology and especially phonology. Interest in morphology is a relatively recent development: in this chapter, I have provided an overview of various morphological approaches which have been proposed within the framework. A natural question to ask, at this point, is whether there is an HPSG theory of morphology or whether HPSG only provides a framework for the formalization of existing morphological analyses. More generally, it may be asked what contribution HPSG makes to morphological theory, if any. In principle, one could assume that HPSG doesn’t need to formulate its own theory of morphology and it might suffice to develop an interface to already existing theories which can be integrated into the framework. Both word–syntax and realizational approaches have been implemented in the HPSG framework. Certain analyses such as that of Krieger (1994) do not seem to go beyond a mere formalization of a word–syntax approach within HPSG. Others, such as those of Riehemann (1993, 1998) as well as Monachesi (1999a) and Miller and Sag (1997) improve on existing realizational analyses by exploiting the tools which are available within the framework. They make an interesting use of (multiple) inheritance hierarchies, which are at the basis of the HPSG framework, to establish a hierarchy of words, which is needed to give structure to the lexicon. By assuming a hierarchy of types, it is possible to express regular patterns at a general level, thus avoiding unnecessary redundancy. Those generalizations can express common properties of different classes of words that can thus be inherited: the advantage of this approach is obvious, as the relevant information is stated only once. At the same time, idiosyncrasies can also be expressed and associated only with those words exhibiting them. Notice that the approaches mentioned above diverge from standard realizational morphology approaches in certain interesting respects. For example, Riehemann (1998) does not consider the -bar sufﬁx simply as phonological material added by a lexical rule, as would be the case in realizational analyses such as those of Anderson (1992), but as a generalization about a speciﬁc class of words, that is the class of -bar adjectives. In Monachesi (1999a), phonological information is not realized by means of a function as in standard realizational approaches such as Stump (1992), but a feature has been introduced for this

122

Morphology and its Interfaces

purpose. The advantage provided by this modiﬁcation is a more straightforward account of the interface between morphology and phonology allowing for an appropriate representation of prosodic phenomena. As for Miller and Sag (1997), they use a function as in standard realizational analyses. However, unlike in realizational analyses, the function is employed to express the morphological form instead of the phonological form and it is not obvious what the advantages of such a move are supposed to be. A disadvantage is that the relation between morphology and phonology remains unclear. The overview presented shows that there is a growing interest, within HPSG, in morphology, which goes beyond the simple development of an interface between syntax and existing morphological theory. After an initial period in which works were mainly formalizations of existing ideas, there has been a shift of direction. In particular, the technical and linguistic innovations that are available within this constraint-based grammar are introduced in morphological analysis, producing some interesting results which improve sensibly on existing accounts. An example of this kind is Bonami and Boyé (2001) in which the formal approach based on types and inheritance drives a novel look at the data considered. They provide an analysis of suppletion in inﬂectional morphology which captures the regularities by means of online type construction and default constraints on the phonology of dependent signs. The advantage of this approach is that it directly captures their deﬁnition of suppletion which considers a sign as suppletive if and only if its phonology is unpredictable. The formal basis which underlies the HPSG framework allows for a strict formalization of the morphological analyses presented. In Monachesi (1999a), a signature is proposed which deﬁnes a possible architecture to deal with morphology: the relevant types and features are introduced and the interaction of morphology with the other modules of the grammar is made explicit. A natural question at this point is whether the HPSG organization of the grammar inﬂuences the treatment of morphology with respect to its interfaces. It seems that the notion of sign that is fundamental in HPSG provides, with its various levels of representation, a straightforward way to deal with the interfaces. The analysis of cliticization that is presented in this chapter offers evidence in this respect. It is, by now, obvious that cliticization phenomena directly inﬂuence the nature of the interface between syntax, morphology, and phonology. The analysis of Romance clitics proposed shows that the multilevel HPSG architecture is well suited to deal exactly with this type of phenomena. It is therefore interesting to investigate how HPSG compares to other frameworks with respect to the treatment of morphology and the interface between the other components of the grammar. It is to this issue that the following sections are dedicated.

Morphology and its Interfaces

123

3.9. Comparing frameworks: Distributed Morphology One of the prevailing approaches to morphological theory is represented by Distributed Morphology (DM) (Halle and Marantz 1993). It is thus relevant to compare it with the developments witnessed in morphology within HPSG, especially with respect to a test case such as that of cliticization. Even though the two theories differ in some very crucial respects, I show that the analysis of Romance cliticization argued for in this chapter shares some interesting insights with similar proposals within DM. The basic elements of DM are collections of morphosyntactic features which are manipulated by the syntax to perfom various kinds of operations. Syntax does not affect lexical items, but generates structures by combining morphosyntactic features by means of operations such as Move or Merge. These features are then realized when morphophonological forms, taken from a set called the Vocabulary, are inserted at syntactic terminal nodes. Vocabulary Insertion therefore involves the association of phonological pieces with abstract morphemes. Therefore, in this system, morphemes are viewed as atoms of morphosyntactic representation which do not include the phonological expression: this is provided as part of a Vocabulary item. If no particular speciﬁc form is consistent with the features in context, an elsewhere form is inserted. In addition to the Vocabulary and the collection of morphosyntactic features, an Encyclopedia links Vocabulary items with their meanings (Harley and Noyer 1999). Distributed Morphology adopts a view of the grammar in which syntax manipulates the morphemes into a hierarchical structure that is the input to computations concerned with morphological and phonological expression (PF) as well as with semantic interpretation (LF). The operations that take a representation from the PF/LF branching to its phonological form comprise the Spelling out of that representation: the module of grammar responsible for Spelling out is morphology. Morphology is thus a cover term for operations that occur on the PF branch following the point at which the syntactic derivation splits between PF and LF (Embick and Noyer 2001). A peculiar feature of this model of the grammar is that there is no lexicon: word formation takes place either in syntax or through post-syntactic operations during morphology. Marantz (1997) provides various arguments against the Lexicalist Hypothesis: the tasks assigned to the lexicon in other theories are distributed in DM through various other components. As a consequence of this view, the distinction between ‘lexical’ and ‘postlexical’ phonology is abandoned since phonology occurs in a single post-syntactic module. There is no need for an architecture of the grammar which divides phonology into a pre-syntactic and a post-syntactic module. It is obvious that the absence of a lexicon represents a crucial difference between DM and other theories of the grammar, such

124

Morphology and its Interfaces

as for example HPSG (but also LFG or even the Minimalist Program (Chomsky 1995)), in which the lexicon plays an important role. As opposed to DM, HPSG is a lexicalist theory which incorporates the principle of Strong Lexicalism (Scalise 1986) according to which the principles of word structure are independent from those governing syntactic structure and syntactic operations do not affect the internal structure of words. Strong lexicalism therefore precludes any analysis where lexical afﬁxes undergo syntactic rules. The absence of the lexicon has crucial consequences for the theory of clitics developed in DM. As I have amply discussed in this chapter, in a lexicalist framework such as HPSG, afﬁxes, as well as Romance clitics which are assimilated to them, are placed on their hosts in the lexicon while clitics such as the Slavic or the Germanic ones are dealt with in syntax. In Distributed Morphology, the distinction between clitics and afﬁxes loses its relevance, since there is no lexical mode of derivation. At the syntactic level, clitics and afﬁxes are treated in the same way, as bundles of features without phonological substance. In DM, the important factors are the provenance of the morpheme, its distribution, and the means by which it comes to be attached to its host. Embick and Noyer (2001) claim, on the basis of their account of the Bulgarian deﬁnite article, that the syntactic/lexical split underlying the clitic/affix distinction is incorrect. They suggest an analysis of the deﬁnite article in terms of ‘Lowering’, that is a movement occurring in morphology that unites syntactic terminals that are phonologically spelled together but not joined in overt syntax.33 They show that the analysis can be extended to deal with the distribution of the possessive marker, which is identical to that of the definite article. However, while the determiner is usually considered an affix, the possessive is generally analysed as a clitic. In a lexicalist framework, this leads to a paradox since the two elements have the same distribution which should be dealt with in one case lexically and in the other case syntactically. Embick and Noyer (2001) claim that this contradiction does not arise in their DM analysis. It should be noted, however, that their analysis largely neglects all those morphophonological idiosyncrasies that motivate, in a lexicalist framework, the treatment of the determiner as afﬁx. The DM treatment is only concerned with the distribution of the Bulgarian deﬁnite article and of the possessive, but only when it is shown that this analysis can also deal with their morphophonological properties can one believe in the merits of a non-lexicalist framework. 33 It should be noted that some of the morphological operations within DM such as ‘Lowering’, (but see also ‘Local dislocation’) share some crucial insights with related operations within Multimodal Categorial Grammar (Moortgat 1997).

Morphology and its Interfaces

125

Embick and Noyer (2001) claim that the distinction between clitics and afﬁxes obscures rather than clariﬁes basic aspects of the syntax–morphology interface. While this might be true, they have completely neglected the implications that this distinction has for the morphology–phonology interface. Only when they can demonstrate how they can deal with it, can one dismiss the distinction. More generally, while in DM there is an intimate connection between syntax and morphology and the interface between these two modules is clearly deﬁned, the situation seems different for the interface with phonology. It is not clear how the division of labour between morphology and phonology (PF) is organized. It is not obvious why certain operations such as Vocabulary Insertion, which associates a phonological expression with an abstract morpheme, should occur in morphology and not in phonology. One wonders which is the role of phonology within DM and more generally which theory of phonology should interface with DM. Harley and Noyer (1999) recognize that Lexical Phonology and Morphology have produced many important insights. However, they conclude that DM denies that these results require an architecture of grammar which divides phonology into a pre-syntactic and a post-syntactic module: it remains to be shown how they can be achieved otherwise. Although from this brief comparison it might seem that there are crucial differences between DM and HPSG, I will show in the rest of this section that there are some unexpected similarities which emerge in the analysis of Romance clitics proposed in this chapter. In particular, in both approaches the basic elements of the computation are features that are spelled out. Recall that, in my analysis, the CLTS list contains information about those elements which are realized as clitics, that is a bundle of features carrying morphological, syntactic, and semantic information, but crucially no phonological information. In both analyses, therefore, morphemes are viewed as a bundle of features without phonological substance: while in my approach this is the case only for afﬁxal elements (f-morphemes in DM terms), in Distributed Morphology all morphemes would have this representation, including lexical ones. It is only through the implicational constraints that the information contained in the CLTS list is related to the actual phonological realization of the clitic at the interface between morphology and phonology. In DM, the link is established in morphology via Vocabulary Insertion. It is interesting to notice that while in DM a special principle, such as the Subset Principle, needs to be stipulated in order to regulate insertion, within HPSG, nothing special needs to be said since implicational constraints apply under uniﬁcation, which is at the basis of the framework. There is, however, a crucial difference between the two approaches that relates to the locus of the operations: while in my analysis feature ‘manipulation’ occurs in the lexicon, in DM it occurs mainly in syntax.

126

Morphology and its Interfaces

I have already mentioned that in DM the distinction between clitics and afﬁxes is not relevant, since both are dealt with in syntax and are viewed as a bundle of features: what is crucial is their distribution and how they get attached to the host. I agree that the distinction between clitics and afﬁxes might be somewhat artiﬁcial and I have argued that clitics do not constitute a uniform class (cf. also Monachesi (1999a)). More generally, when dealing with functional elements, it is usually difﬁcult to characterize their analysis as lexical, syntactic, or morphological since it is often the result of the interaction of these various modules. This is what I have shown with the treatment of cliticization defended here and which I have analysed in its various aspects: it is for this reason that I believe that the analysis presented here is superior to DM accounts, which usually only focus on one of them.

3.10. Comparing frameworks: Optimality Theory The representation of dependent elements, among which are clitics, has also been the topic of several works within the framework of Optimality Theory (OT) (McCarthy and Prince 1993b); (Prince and Smolensky 1993) and its more recent version Correspondence Theory (CT) (McCarthy and Prince 1995). In this section, I take these analyses into consideration and evaluate them with respect to the approach to cliticization; proposed in this chapter, especially the aspects that concern their relevance to the interfaces. However, they often address only one aspect of cliticization, perhaps the only exception in this respect is represented by Legendre (2000a) and by Gerlach (2001), the latter being an attempt to provide a comprehensive analysis of cliticization. In OT, a grammar consists of a set of universal constraints which are ranked according to language-specific criteria. While in HPSG constraint violation leads to ungrammaticality, this is not the case in OT where constraints can be violated. The set of output candidates to a given input is evaluated on the basis of the constraint ranking. The best output candidate is the one with the fewest violations of highly ranked constraints. In CT, the correspondence relation of input and output candidates is checked by faithfulness constraints. Faithfulness constraints (together with markedness constraints) are one of the most crucial devices in the analysis of Italian and Spanish clitics developed by Grimshaw (1997). She addresses the issue of how clitics should be represented in the lexicon in order to obtain the appropriate clitic choice as well as the problem of clitic combinations. She argues that the unexpected clitic forms which occur in clitic combinations are the best available ones given the need to avoid sequences of identical clitics. A shortcoming of this approach is that not all clitics are taken into account and only a limited set of idiosyn-

Morphology and its Interfaces

127

crasies, that are attested in the case of clitics, are considered. In addition, no independent motivation is given for the peculiar lexical entries assumed in the case of clitics. As already mentioned, Grimshaw’s proposal addresses only one aspect of cliticization, that is its morphological character, and leaves the other ones unaccounted for. Even an important issue such as that of the ordering of the clitics is left for future research. A possible solution to this problem is constituted by the proposal put forward in Anderson (1996), which sketches a way to deal with clitic ordering in Romance languages. Anderson assumes that all affixes are present in the word at the same time. Their ordering is determined by the fact that they all compete for a certain position, but since in OT constraints are ranked relative to one another, one will outweigh the others. The analysis is developed for second position clitics, but Anderson extends it to also account for the distribution of Romance clitics. Clitics are subject to constraints of the form EDGEMOST (CL,L): which means that a given clitic should be as close to the left edge as possible. These constraints should be ranked in the appropriate way so that the observed order is produced. The analysis is thus similar to that proposed by Legendre (1996, 2000a) for the treatment of the Bulgarian clitic cluster. She assumes that the EDGEMOST constraints are in conflict with those which prohibit certain features being initial in a specific domain, which she suggests should be V’. A problem with Anderson’s approach is that it is not clear how the complementary distribution which holds for certain clitics can be derived. Furthermore, the insights underlying the ordering of clitics in both Anderson and Legendre’s analyses are not new: the EDGEMOST constraints are a way to formalize in OT the template approach. In this respect, these treatments do not really contribute to the understanding of clitic ordering, or perhaps they could be viewed as additional evidence that the order of the clitics in the cluster is indeed idiosyncratic and cannot be explained in terms of general principles. Some of the insights put forward in the analysis of cliticization presented in this chapter can also be found in Legendre (2000a). In particular, she defends a view that considers clitics not as syntactic atoms generated and moved around in the syntax but as functional features which are realized morphologically as phrasal afﬁxes (cf. also Anderson (1992)). They are subject to alignment constraints which regulate their realization at the edge of some domain: second position clitics, for example, result from the interaction of an alignment constraint whose domain is V’ and one whose domain is the Intonational Phrase. Her analysis can be extended to Bulgarian clitics even though they are not subject to second-position requirements (except for the question particle li) and are considered verbal clitics. More generally clitics positioning results from

128

Morphology and its Interfaces

one mechanism, which is constraint ranking. Similarly for Romanian clitics (Legendre 2000b) which are also analysed as phrasal afﬁxes. Furthermore, Legendre shows that a proper analysis of Bulgarian cliticization can only be obtained through the interaction of syntactic, prosodic, and morphological constraints which raise the issue of the modularity of the grammar. Since constraints may refer to any level of representation and they interact, a natural question is whether the rankings of constraints freely intermix or whether rankings consist of blocks of constraints referring to the same sub-level of representation. On the basis of the Bulgarian data, she argues that constraints belonging to different modules of the grammar may not intermix. An interesting proposal which addresses the understudied area of the phonology of Romanian clitics is Popescu (2000). She argues that the morphophonological phenomena of Romanian pronominal clitics, discussed in Section 3.3.2.5, should not be considered as idiosyncrasies. While her analysis has some interesting insights, I believe that it does not account for the full range of idiosyncrasies I have presented: for example it does not account for the fact that clitic deletion occurs only with verbs with unstressed a or o or for the special behaviour of the reﬂexive clitics. Popescu proposes an analysis which assumes that the various forms attested in the case of Romanian clitics should not be seen as real allomorphs, but as context-dependent realizations of the same morpheme. She claims that clitic clusters tend to have a minimal number of syllables, and it is for this reason that some of the clitics are reduced. However, such an analysis does not really deny the idiosyncratic character of these forms. This would be the case if she were to show that the phonological processes involved are generally attested in the language, which she fails to do. In general, the changes attested are triggered by the presence of other clitics and they do not occur or they become optional if there is a lexical verbal host. Furthermore, the OT analysis does not offer any insights into why in certain cases reduction occurs, while in other cases deletion is attested. The intuition that reduction occurs if there is the possibility of the vowel reducing to a glide, while deletion occurs if this possibility is not attested, is not immediately reﬂected in the mathematical calculation of the number of violated constraints. As already mentioned, the analyses discussed above deal mainly with one aspect of cliticization. A different approach in this respect is Gerlach (2001): she integrates the treatments summarized in this section into a coherent account providing a comprehensive analysis of cliticization in Romance. Gerlach assumes a high-ranked FAITHFULNESS constraint which requires that arguments are realized morphologically in order to account for the fact that clitics realize in the output particular argument roles in the input. As for the choice of the various clitics, her analysis is similar to that of Grimshaw (1997)

Morphology and its Interfaces

129

and is based on the interaction of violable FAITHFULNESS and MARKEDNESS constraints. On the other hand, the order and combinatory restrictions on clitic sequences are accounted for by morphological ALIGNMENT constraints while opaque clitics are dealt with by means of OCP-like constraints. The placement of clitics is analysed through the interaction of three kinds of ALIGNMENT constraints: syntactic ones, which align clitics with the domain of V’, morphological ones, which determine whether clitics occur before or after the verb, and prosodic ones, which establish whether clitics are integrated into the phonological word on their right or on their left. The merit of this treatment is to show that a comprehensive analysis of cliticization is only possible through an appropriate interaction of phonology, morphology, and syntax, a view I have also advocated in this chapter. Therefore, in Gerlach (2001), the best clitic sequence results from the interaction of constraints which might belong to various grammatical modules. For example, morphological constraints could be ranked higher than phonological constraints in order to deal with a certain phenomenon. This seems to me the equivalent of saying that a certain module of the grammar is more relevant than another, which doesn’t seem to be an adequate conclusion. In OT, therefore, the evaluation of the various candidates does not take place in several steps, but all candidates are evaluated on the same level. There is thus a similarity with HPSG, where the various constraints interact to state which linguistic object is grammatical and which one is not. Recall, however, that in HPSG constraints are interpreted as well-formedness conditions on linguistic objects. Constraint violation determines ungrammaticality: there is therefore a crucial difference with the way constraints are interpreted in OT. The OT constraints evaluate a given input with respect to various outputs. The input is stated as an unordered bundle of what one might interpret as phonological, morphological, syntactic, and semantic features. Unlike DM, in which features are ordered according to a certain hierarchy, or HPSG, where a feature theory has been developed, in OT features all have the same status, which might not be very economical. The analysis proposed by Gerlach (2001) provides a comprehensive treatment of cliticization which is obtained through the interaction of various constraints. The consequence, however, is an explosion of constraints that makes the analysis difﬁcult to evaluate.

3.11. Conclusions Romance pronominal clitics constitute a well studied topic in the linguistic literature. However, it is only recently that their key role in achieving a better understanding of the interfaces of the grammar has been highlighted.

130

Morphology and its Interfaces

Previous studies of cliticization have often focused on one aspect of this phenomenon, usually the syntactic one, neglecting the other ones. In this chapter, I have tried to provide an overview of the complexity of the data by addressing the various phonological, morphological, and syntactic properties of Romance clitics. If clitics are considered from a broader perspective, it is immediately obvious that they do not constitute a uniform class and that the term clitic is a label often used to address elements which exhibit very different behaviour both within one language and across languages. I have argued that Romance clitics behave as inﬂectional afﬁxes and I have suggested a morphological analysis of their properties. In particular, I have shown that HPSG constitutes an appropriate model to deal with the interface character of Romance clitics. The multilevel architecture and the use of implicational constraints which can relate different levels of description can reconcile their morphophonological properties with their syntactic characteristics. The morphological analysis of Romance clitics has allowed for a detailed discussion of the role of morphology within the theory of grammar and its relation with the other components. Furthermore, a comparison between HPSG morphology with its extended use of inheritance hierarchies and the view defended in other theories such as Distributed Morphology and Optimality Theory has been carried out.

4 Syntax and its Interfaces: The Case of Romance Auxiliary Verbs 4.1. Introduction The aim of this chapter is to analyse the interface between syntax and the other components of the grammar. The relation between syntactic structure and argument structure is investigated by taking into consideration a particular instance of complex predicate formation, that is Romance auxiliary verbs. However, the focus is especially on Romanian tense auxiliaries, which despite their relevance have received little attention in the literature, the only exceptions being Dobrovie-Sorin (1994), Rivero (1994), Legendre (1999, 2000b), and Monachesi (1999c, 2000). As already mentioned, Romanian is particularly interesting since it allows for several sorts of comparisons: genetic (with the other Romance languages), geographical (with the other Balkan languages), and typological (Joseph 1999). I show that these perspectives are particularly important when one investigates the various properties of Romanian tense auxiliaries. Only a close examination will allow an understanding of their idiosyncratic behaviour. Romanian tense auxiliaries pattern like the Balkan auxiliaries (and unlike the Romance ones) with respect to adjacency requirements. They behave, however, like the Romance auxiliaries (and unlike the Balkan ones) with respect to the distribution of pronominal clitics. I argue that they are morphosyntactic words, which are phonologically dependent on an adjacent word. They are simple clitics according to the terminology of Zwicky (1977). I claim that an appropriate interaction of syntax with argument structure can account for the idiosyncratic properties of the Romanian tense auxiliaries as well as for the similarities and differences to their Balkan and Romance counterparts. A comparison is carried out between Romanian tense auxiliaries and auxiliaries in French and in Italian; it is suggested that the relevant differences with respect to adjacency can be accounted for in terms of different syntactic structures. While in French and Italian the auxiliary verb, the nonﬁnite element, and the complements combine in a flat structure (Abeill´e and Godard 1994,

132

Syntax and its Interfaces

2002; Monachesi 1996a), in Romanian, the tense auxiliary and the lexical verb form a verbal complex (Monachesi 1999c, 2000; Abeill´e and Godard 2003). If a verbal complex is assumed, it is possible to account for the properties of Romanian tense auxiliaries and for the fact that they can be placed after the verbal complement, in certain registers. This characteristic of Romanian tense auxiliaries to occur in an inverted paradigm is compared with the phenomenon of mesoclisis in European Portuguese. It is suggested that the Portuguese tense/aspect markers for the conditional and future have the same status as tense auxiliaries in Romanian. There are some crucial similarities across Romance languages with respect to the distribution of the clitics in the presence of auxiliary verbs. If an auxiliary verb is present, the pronominal clitic, which is an argument of the embedded verb, must precede the auxiliary and cannot attach to the lower verb. On the other hand, pronominal clitics tend to follow the auxiliary clitics in Balkan and Slavic languages (Franks and King 2000). I show that the distribution of object clitics is an area in which Romanian differs from the Balkan (and Slavic) languages and patterns like the other Romance languages. I claim that Romanian exhibits clitic climbing and I suggest an analysis in terms of argument composition (Hinrichs and Nakazawa 1994) to deal with this phenomenon. This is a lexical mechanism according to which the auxiliary inherits the arguments of the embedded verb, including those ones which might be realized as clitics (Monachesi 1993a, 1998a). Romanian exhibits a rich typology of auxiliary(-like) constructions which are characterized by the presence (or absence) of clitic climbing and by the possibility (or impossibility) for certain elements to intervene between the auxiliary and the nonfinite verb. In addition to tense auxiliaries, the colloquial future auxiliary, the modal verb a putea ‘can’, the aspectual auxiliary fi, and raising verbs are taken into consideration in order to illustrate this typology. I show that these data provide additional support for the assumption that an appropriate division of labour between syntactic structure and argument structure can deal with the variation exhibited by these constructions with respect to adjacency and to the distribution of the clitics. Special attention is dedicated to the role of argument structure within HPSG and to its function, which is almost exclusively syntactic. A comparison is carried out with the role that argument structure has in other frameworks such as Lexical Functional Grammar. In LFG, a-structure operates as an interface between the semantics and syntax of predicators. An alternative analysis of the data presented in this chapter could be foreseen. It could be suggested that syntactic structure does not play a role in accounting for the adjacency of the auxiliary and the lexical verb: what is

Syntax and its Interfaces

133

crucial is the way elements merge in the form dimension. Under this view, syntax loses its role as an autonomous module of the grammar and it acts merely as an interface between form and meaning, as in current versions of Categorial Grammar (Moortgat 1997). However, if this view is accepted, it is not clear how constituency tests should be interpreted. Furthermore, prosodic constituency in the form dimension is determined on the basis of syntactic criteria instead of prosodic ones, which is an undesirable result. As will be discussed at length in the following chapter, I follow Nespor and Vogel (1986) in assuming that syntax exists as an autonomous module of the grammar, which creates the basis for a mapping into prosodic structure, by means of an appropriate algorithm which is based on prosodic principles.

4.2. Basic properties of Romanian tense auxiliaries In this section, I give a general overview of Romanian tense auxiliaries and I discuss the adjacency requirements they are subject to. Romanian tense auxiliaries can be found in the present perfect, the conditional, and the future. In the case of the present perfect, the tense auxiliary combines with a past participle, while in the other two cases it combines with a bare inﬁnitive, that is an inﬁnitive without the particle a. The following table summarizes the different forms of tense auxiliaries: (194) Different forms of the tense auxiliaries

Aux Psp. Aux Cond. Aux Fut.

1 Sg am as¸ voi

2 Sg ai ai vei

3 Sg a ar va

1 Pl am am vom

2 Pl at¸i at¸i vet¸i

3 Pl au ar vor

past participle bare infinitive bare infinitive

As Dobrovie-Sorin (1994) notices, these auxiliaries exhibit different forms from the related lexical verbs, thus the paradigm of the future auxiliary is distinct from that of a vrea ‘want’ or a voi ‘want’ (which is an archaic variant of the former).1 Similarly, the paradigm of the tense auxiliary in the present perfect and in the conditional is different from that of the verb a avea ‘to have’, which is given below:

1 There is similarity with Serbo-Croatian that forms the future with an auxiliary clitic, which represents a reduced form of the verb ht(j)eti ‘to want’ in combination with an infinitive (Franks and King 2000).

134

Syntax and its Interfaces

(195) The verb a avea ‘to have’ 1 Sg am

2 Sg ai

3 Sg are

1 Pl avem

2 Pl avet¸i

3 Pl au

It is relevant to notice that the tense auxiliaries are a monosyllabic, reduced variant of the verb a avea ‘to have’. Romanian tense auxiliaries exhibit certain properties that differentiate them from other Romance counterparts, most notably they are strictly adjacent to the lexical verb. In Romanian, adverbs cannot separate the tense auxiliary from the verbal complement, as illustrated by example (196a), in which adesea ‘often’ occurs between the two verbs, resulting in an ungrammatical sentence.2 The adverb usually follows the complex verb (Mallinson 1986), as shown by (196b): (196) a. *Am adesea v˘azut filme bune. [ro] have often seen films good b. Am v˘azut adesea filme bune. have seen often films good ‘I have often seen good films.’ The situation is different in French and Italian where adverbs can generally occur between the auxiliary and the verbal complement. In the French example below, the adverb évidemment ‘clearly’ separates the auxiliary a from the past participle lu: (197) Paul a évidemment lu ce texte. [fr] Paul has clearly read this text ‘Paul has clearly read this text.’ Similarly, in Italian, the adverb sicuramente ‘surely’ can intervene between the auxiliary ha and the verbal complement visto: (198) Ha sicuramente visto il nuovo film di Almodovar. [it] has surely seen the new film by Almodovar ‘He has surely seen the new film by Almodovar.’ It is interesting to note that Romanian behaves like other Balkan languages in this respect. As observed by Legendre (2000a), in Bulgarian, adverbs cannot occur between the clitic auxiliary and the lexical verb (cf. also Macedonian): 2 Abeill e´ and Godard (2003) mention that the situation is similar in Spanish and Portuguese, where only a limited class of adverbs can occur in this position. However, there is variation among speakers with respect to which adverbs belong to this class.

Syntax and its Interfaces

135

(199) *Ivana e naburzo proˇcela knigite. [bg] Ivana be-3 quickly read books-the ‘Ivana has quickly read the books.’ On the other hand, adverbs can occur between the lexical auxiliary beše and the past participle, as shown by the example below: (200) Ivana beˇse naburzo proˇcela knigite. [bg] Ivana was-3 quickly read books-the ‘Ivana had quickly read the books.’ In Romanian, there is an exception to the strict adjacency of tense auxiliaries and the lexical verb since a group of five monosyllabic intensifiers can separate them. They are mai, ‘again’, cam ‘little’, prea ‘very’, ¸si ‘also’, tot ‘still’ (DobrovieSorin 1994): (201) As¸ mai mînca. [ro] would again eat ‘I would still eat.’ In Section 4.6, I argue that these elements have affixal status and it is for this reason that they can occur between the two verbs.3 Romanian shares this property with other Balkan languages, such as Bulgarian. As noted in Krapova (1997, 1999); (cf. also Boˇskovi´c (2001); Avgustinova and Oliva (1991); Franks and King (2000)), certain short adverbs, such as veˇce ‘already’ and ošte ‘still’, can intervene between the clitic auxiliary and the lexical verb. They are unstressed when they occur in this position: (202) a. Ivana e veˇce proˇcela knigite. [bg] Ivana be-3 already read books-the ‘Ivana has already read the books.’ b. Ti ne si go oˇste napisala. you not be-2 CL.ACC still written ‘You have not yet written it.’ Boˇskovi´c (2001) notices, however, that not all speakers accept the sentences above as grammatical. A similar phenomenon is also attested in Macedonian, as mentioned in Franks and King (2000). 3 There is some similarity between these elements and certain ‘short adverbials’ attested in Czech which occupy various positions within the clitic cluster (Franks and King 2000; Avgustinova and Oliva 1995).

136

Syntax and its Interfaces

Similarly to adverbs, Romanian quantiﬁers such as tot¸i ‘all’ cannot occur between the tense auxiliary and the past participle, as shown by the ungrammatical example (203a). The quantiﬁer should follow the verbal complex (203b), as shown by the following examples from Dobrovie-Sorin (1994): (203) a. *Elevii t˘ai au tot¸i citit un poem de Verlaine. [ro] students yours have all read a poem by Verlaine b. Elevii t˘ai au citit tot¸i un poem de Verlaine. students yours have read all a poem by Verlaine ‘Your students have all read a poem by Verlaine.’ On the other hand, in French quantifiers can intervene between the auxiliary and the verb, as exemplified by (204) where the quantifier tous separates the auxiliary a and the past participle vus: a tous vus. [fr] (204) Marie les Marie CL.ACC has all seen ‘Mary has seen them all.’ The situation is more complex in Italian where there is variation with respect to acceptability:4 (205) %Elena li ha tutti mangiati. [it] Elena CL.ACC has all eaten ‘Elena has eaten them all.’ In the example above, the quantifier tutti intervenes between the auxiliary ha and the verbal complement mangiati. However, the sentence is only marginally acceptable; the standard version would have the quantifier following the verbal complex: ha mangiati tutti. [it] (206) Elena li all Elena CL.ACC has eaten ‘Elena has eaten them all.’ Once more Romanian patterns like Bulgarian in this respect. As noted by Legendre (2000a), quantifiers cannot occur between the clitic auxiliary and the past participle: vsiˇcki proc¸eli knigite. [bg] (207) *Studentite sa read books-the students-the be.3.PL all ‘The students have all read the books.’ 4

The symbol ‘%’ is used to indicate variation in acceptability.

Syntax and its Interfaces

137

On the other hand, they can occur between the lexical auxiliary bjaxa and the verb (cf. also Macedonian): (208) Studentite bjaxa vsiˇcki proc¸eli knigite. [bg] students-the were-3 all read books-the ‘The students had all read the books.’ In Romanian, subjects can be found in different positions. They can occur before the verb (209a), between the verb and the direct object (209b), and after the direct object (209c): (209) a. Mama a f a˘ cut o pr˘ajitur˘a. [ro] mum has made a cake b. A f a˘ cut mama o pr˘ajitur˘a. has made mum a cake c. A f a˘ cut o pr˘ajitur˘a mama. has made a cake mum ‘Mum has made a cake.’ However, the subject cannot intervene between the tense auxiliary and the lexical verb: (210) *A mama f a˘ cut o pr˘ajitur˘a. [ro] has mum made a cake ‘Mum has made a cake.’ Other Romance languages behave like Romanian with respect to the example above; however, they present at least one environment in which the subject can intervene between the two verbs. For example, in French, subject clitics can occur between the auxiliary and the lexical verb, as shown by tu in the sentence below: mangé? [fr] have you eaten ‘Have you eaten it?’

(211) L’as-tu CL.ACC

Notice, however, that this is the only configuration in which a subject can intervene in French. Similarly, in Italian, the subject can intervene between the auxiliary and the lexical verb, in certain constructions: (212) Avendo Martina deciso di partire, possiamo usare la sua stanza. [it] having Martina decided to leave, can use the her room ‘Martina having decided to leave, we can use her room .’

138

Syntax and its Interfaces

As Dobrovie-Sorin (1994) notes, unlike other Romance languages, Romanian does not present any construction in which the subject NP intervenes between the tense auxiliary and the verbal complement. Similarly in Bulgarian, as discussed in Legendre (2000a), subjects cannot intervene between the clitic auxiliary and the lexical verb: (213) *Kakvo e Ivan proˇcel? [bg] what be-3 Ivan read ‘What has Ivan read?’ Furthermore, in Romanian (as well as in French and in Italian), complements cannot occur between the two verbs: (214) *Am filme bune v˘azut. [ro] have films good seen ‘I have seen good films.’ In the example above, the direct object filme bune cannot separate the tense auxiliary am from the past participle v˘azut. The data presented in this section suggest that in Romanian, the tense auxiliary and the lexical verb form a unit which cannot be separated by other elements. This is not the case in French and Italian where (clitic) subjects, adverbs, and quantiﬁers can intervene between the two verbs. Furthermore, the data presented show a crucial similarity with respect to the adjacency requirements between Romanian tense auxiliaries and the clitic auxiliaries found in Balkan languages such as Bulgarian and Macedonian. On the basis of the evidence presented, I conclude that they are morphosyntactic words, which are phonologically dependent on an adjacent word. They are simple clitics according to the terminology of Zwicky (1977). The fact that they represent a reduced form of the verb a avea ‘have’ provides additional support for this hypothesis. In the analysis provided in this chapter, I argue that their dependency is accounted for in syntax in terms of the verbal complex structure. I show that a syntactic treatment of adjacency is the most adequate to deal with the variation exhibited by the Romance auxiliary verbs discussed in this section. The dependency is also reflected at the phonological level, given that the tense auxiliaries form a prosodic word with the lexical verb since they are monosyllabic and unstressed elements (Popescu 2000). It could be argued that there is an alternative analysis of the data presented in this section which accounts for the treatment of adjacency exclusively within phonology. I postpone the discussion of this possibility to Section 4.14, in which I provide evidence that this is not the case and that adjacency should indeed be considered a syntactic property.

Syntax and its Interfaces

139

4.3. The syntactic structure of tense auxiliaries In this section, I show that a syntactic treatment of adjacency, that is a treatment in terms of syntactic structure, can adequately account for the variation attested in the Romance auxiliary system. A hierarchical structure in which the auxiliary takes a VP complement is the one usually adopted to represent Romance tense auxiliaries (and more generally complex predicates) in the linguistic literature (Kayne 1975; Moore 1991; Pollock 1989; Zagona 1988; Burzio 1986; Cinque 2000; Miller 1992a; DobrovieSorin 1994; Picallo 1990; Alsina 1996; Manning and Andrews 1999): (215) Hierarchical structure VP

H H H

V [AUX

+]

VP

H H

V

COMPS

This structure makes it possible for elements to intervene between the finite and the nonfinite verb, it allows a coordination of sequences of nonfinite verbs in combination with their complements without having to repeat the finite verb, and it predicts that the combination of the nonfinite verb with its complements cannot be extracted or cliticized. However, Abeill´e and Godard (2002) have challenged this tradition with respect to French auxiliaries. They have provided crucial evidence against a hierarchical structure and in favour of a flat one, in which the nonfinite verb and its complements are sisters of the auxiliary verb (cf. also Legar e` and Rollin (1976)): (216) Flat structure VP

V [AUX

+]

H H V

H

H

COMPS

A flat structure has also been adopted in Monachesi (1996a) for Italian auxiliaries as well as by Legendre (1987), Miller (1992a), and Abeill´e et al. (1998)for the French causatives, by Aissen and Perlmutter (1983) for the Spanish restructuring verbs, by Monachesi (1993b, 1999a) for Italian restructuring verbs and by Monachesi (1999c, 2000) for the Romanian a putea. This structure predicts that the sequence made by a nonﬁnite verb with its complements cannot be extracted or cliticized. It allows for elements to intervene between the ﬁnite and the nonﬁnite verb and it permits a coordination of

140

Syntax and its Interfaces

sequences of nonﬁnite verbs in combination with their complements without having to repeat the ﬁnite verb. In the following section, I summarize the evidence in favour of a flat structure for French auxiliaries proposed by Abeill´e and Godard (2002). I show that their arguments carry over to Italian auxiliaries as well (Monachesi 1996a). Given the flat structure, it is possible to account for the fact that, differently from Romanian, (clitic) subjects and adverbs can occur between the auxiliary and the nonﬁnite verb, both in French and in Italian. I suggest a different structure in the case of Romanian tense auxiliaries (Monachesi, 1999c, 2000; Abeill´e and Godard 2003). I assume that the auxiliary and the nonfinite verb constitute a verbal complex, which seems the most appropriate structure to capture the intuition that the auxiliary and the lexical verb form a unit which cannot be separated by other elements: (217) Verbal complex VP

H HH

V

HH

V [AUX

+]

COMPS

V

If the flat structure is adopted, it would be necessary to formulate special linearization constraints in order to account for the fact that it is not possible for subjects, adverbs, quantiﬁers, and complements to intervene between the two verbs. Furthermore, the verbal complex structure predicts that the sequence made up of the ﬁnite verb and its complements cannot be extracted or cliticized and that sequences of nonﬁnite verbs with their complements cannot be coordinated. I also show that a hierarchical structure does not seem plausible in view of the fact that Romanian tense auxiliaries can be optionally placed after the verbal complement. This phenomenon of auxiliary inversion in Romanian is compared with that of mesoclisis in European Portuguese: I argue that there are strong similarities between the two. 4.3.1. The syntactic structure of French auxiliary verbs

Abeill´e and Godard (2002) provide several arguments in favour of a flat structure for French tense auxiliaries and against a hierarchical one. In particular, they note that the standard constituency tests do not support the view that the nonﬁnite verb and its complements form a constituent. There is thus no evidence for a hierarchical structure in which the auxiliary verb subcategorizes for a VP.

Syntax and its Interfaces

141

Motivations come from the fact that the sequence which comprises the nonfinite verb and its complements cannot be cliticized (218a), unlike the complement of a control verb (218b) or that of a passive participle (218c) for which a hierarchical structure is usually assumed: (218) a. *Jean croyait avoir compris, mais il ne l’avait pas. [fr] John believed have understood, but he CL.NEG CL.ACC had not ‘John believed to have understood, but he hadn’t.’ le veut pas. b. Jean peut venir, mais il ne John can come, but he CL.NEG CL.ACC want not ‘John can come, but he does not want to.’ furent pas épargnés par les c. Bien des bâtiments publics ne many buildings public CL.NEG were not spared by the bombes, mais, heureusement, la cathédrale le fut. bombs, but, fortunately, the cathedral CL.ACC was ‘Many public buildings were not spared by the bombs, but fortunately, the cathedral was.’ Similarly, it is not possible to question the complement of an auxiliary verb (219a), while this is not the case for a control verb (219b) or for a passive verb: (219) a. *Qu’a-t-elle? – Vendu ses livres. [fr] what has she? – sold her books ‘What has she? She has sold her books.’ b. Que veut-elle? – Partir. what wants she? – leave ‘What does she want? She wants to leave.’ Furthermore, the sequence composed of the nonﬁnite verb with its complements cannot undergo deletion (220a), preposing (220b), or clefting (220c), while this is not the case for the complement of a control verb or in the case of a passive auxiliary: (220) a. *Jean a fini son travail, mais Marie n’a pas. [fr] John has finished his work, but Mary CL.NEG has not ‘John has finished his work, but Mary hasn’t.’ b. *Lu cet article, non, je n’/ne l’ai pas. read this article, no, I CL.NEG CL.ACC have not ‘Read this article, no, I haven’t.’

142

Syntax and its Interfaces c. *Ce que Jean a, c’est bu trop de vin. what John has, is drank too much of wine ‘What John has, is to drink too much wine.’

It could be argued that the fact that a nonfinite verb and its complement cannot be questioned or cliticized when embedded under an auxiliary verb might be because auxiliaries select a VP, while clitics and question words are typically nominal in character. Auxiliaries differ from control verbs such as vouloir ‘want’, which can take nominal complements. The ungrammatical examples might be due to a violation of c-selectional restrictions. As Abeill´e and Godard (2002) note, this reasoning could be valid in the case of pronominalization, but extraction is not limited to NP constituents, therefore this explanation can only be partial. Furthermore, passive auxiliaries behave as control verbs in that their complements can undergo deletion, preposing, or clefting; however, unlike control verbs they cannot take nominal complements. It seems, therefore, that the ungrammaticality of the sentences above cannot be due to c-selection properties. On the other hand, the contrast can be easily explained if it is assumed that tense auxiliaries are associated with a flat structure, while control verbs and passive auxiliaries are associated with a hierarchical structure. Additional evidence against a hierarchical structure for French auxiliaries comes from the distribution of manner adverbs. Abeill´e and Godard (2002) argue that if such a structure is assumed, three possible attachments for adverbs could be foreseen. However, they are not able to reflect the appropriate scope, as well as the correct position with respect to the verb and to time adverbs. The behaviour of auxiliaries in bounded dependency constructions such as tough constructions, inﬁnitival relative clauses, and in the presence of causative verbs provides additional arguments against a hierarchical structure. Abeill´e and Godard (2002) therefore conclude that there is no positive evidence in favour of a hierarchical structure; even coordination does not seem to support it. Abeill´e and Godard (2003) consider as evidence cases of double coordination. They show that the conjunction is located in front of the participle (221a) but this does not constitute a structural indication. In the case of non-constituent coordination, the conjunction is located in front of the sequence of complements (221b), which clearly do not form a constituent: (221) a. Paul aura (ou) trouvé une solution ou demandé de l’aide. [fr] Paul will have or found a solution or asked of the help ‘Either Paul will have found a solution or he will have asked for help.’

Syntax and its Interfaces

143

b. Paul aura acheté (ou) un livre a` Jeanne ou un disque a` Paul will have bought or a book to Jeanne or a record to Sophie. Sophie ‘Either Paul will have bought a book for Jeanne or a record for Sophie.’ More generally, it seems that the conjunction is placed in front of the first word of each sequence of complements and not in front of the first word of each constituent. Abeillé and Godard conclude that coordination does not provide conclusive evidence in support of a given structure. They propose that the tense auxiliary does not subcategorize for a VP. Under this view, the coordination of the nonfinite verb with its complements, as in the example below, should be assimilated to cases of non-constituent coordination, which are also attested in French: (222) Paul a parlé avec Marie et compris son erreur. [fr] Paul has talked with Mary and understood his mistake ‘Paul has talked with Mary and understood his mistake.’ Abeillé and Godard (2002) show that a structure in which the auxiliaries and the past participle form a verbal complex is also inadequate for French. Such a structure runs into problems with examples of coordination like the one in (222). For many French speakers there is a clear contrast between (223a) and (223b), which are cases of non-constituent coordination: (223) a. Jean donnera [le livre de Proust] [`a Marie] et [le livre de John will give the book of Proust to Mary and the book of Camus] [`a Paul]. [fr] Camus to Paul ‘John will give the book by Proust to Mary and the book by Camus to Paul.’ b. ??Jean donnera [le livre de Proust] [`a Marie] et [de Camus] [`a John will give the book of Proust to Mary and of Camus to Paul]. Paul ‘Jean will give the book by Proust to Mary and by Camus to Paul.’ The judgements show that only sequences of sister constituents can be conjoined in French. This is not the case in the verbal complex structure, which would incorrectly predict that it is not possible to conjoin sequences of participles with their complements.

144

Syntax and its Interfaces

The second problem involves the possibility of inserting elements between the two relevant verbs such as quantifiers and adverbs. It would not be possible if a verbal complex structure were adopted. These problems would not arise if a flat structure were assumed: Abeill´e and Godard (2002) conclude, therefore, that the flat structure is the most appropriate to deal with French auxiliaries.5 Its interaction with the lexical mechanism of argument composition and with the relevant linearization principles provides an adequate account of the properties of French auxiliaries discussed in Section 4.2, as shown in Abeill´e and Godard (1994, 2002), to which I refer for further details. 4.3.2. The syntactic structure of Italian auxiliary verbs

The evidence provided by Abeill´e and Godard (1994, 2002) against a hierarchical structure for French auxiliaries carries over to Italian auxiliaries. The sequence of the nonﬁnite verb with its complements cannot be cliticized (224a) or questioned (224b) in Italian: (224) a. *Vito credeva aver capito, ma non l’aveva. [it] Vito believed have understood, but not CL.ACC had ‘Vito believed to have understood, but he hadn’t.’ b. *Che ha? – Venduto i suoi libri. what has? – sold the his books ‘What has he? He has sold his books.’ As shown in the previous section for French, the nonfinite complement cannot be deleted (225a), preposed (225b), or clefted (225c): (225) a. *Michele ha finito il suo lavoro, ma Elena non ha. [it] Michele has finished the his work, but Elena not has ‘Michele has finished his work, but Elena hasn’t.’ b. *Letto questo articolo, no, non (l’) ho. read this article, no, not CL.ACC have ‘Read this article, no, I haven’t.’ c. *Quello che Vito ha, e’ bevuto troppo vino. what that Vito has, is to drink too much wine ‘What Vito has, is drank too much wine.’ 5

Abeille´ and Godard (2000) conclude that passive auxiliaries and copular verbs are characterized by a hierarchical structure. These verbs behave like the tense auxiliary with respect to coordination but they differ with respect to the constituency tests. The different structures assumed, hierarchical in the case of passive auxiliaries, and flat in the case of the tense auxiliaries, are able to capture this contrast.

Syntax and its Interfaces

145

As for coordination, the data are similar to French since coordination of nonﬁnite lexical verbs is always accepted as well as coordination of sequences of nonﬁnite verbs together with their complements Abeill´e and Godard (2003): (226) a. Giovanni ha comprato e letto questo libro. [it] John has bought and read this book ‘John has bought and read this book.’ b. Giovanni ha sbattuto gli albumi e fatto la crema. John has mixed the egg whites and made the custard ‘John has mixed the egg whites and made the custard.’ In Italian, as in French, it is possible to conjoin sequences of complements in a non-constituent coordination. As in the case of French, acceptability is degraded if the complements are not sister constituents: (227) a. Giovanni regalerá [un libro di Proust] [a Maria] e [un libro di John will give a book of Proust to Mary and a book of Camus] [a Paul]. [it] Camus to Paul ‘John will give a book by Proust to Mary and a book by Camus to Paul.’ b. ??Giovanni regalerá [un libro [di Proust]] [a Maria] e [di John will give a book of Proust to Mary and of Camus] [a Paul]. Camus to Paul ‘John will give a book by Proust to Mary and by Camus to Paul.’ These coordination data argue against a verbal complex structure in Italian and it could be claimed that they provide support for a hierarchical structure. However, the existence of non-constituent coordination casts doubts on this argument, since it is possible to analyse the conjunction of participles and complements as an instance of non-constituent coordination which can be accounted for by a flat structure. In addition, the evidence provided in this section shows that the sequence consisting of the past participle and its complements fails all classical constituency tests (except for coordination). I conclude that a hierarchical structure where the auxiliary subcategorizes for a VP complement is unmotivated.6 Similarly for what concerns a structure in which the auxiliary and the past participle form a verbal complex: the line of reasoning developed for French 6 A hierarchical structure could be assumed instead for passive auxiliaries which behave like the tense auxiliaries with respect to the coordination data, but differ from those with respect to extraction and pronominalization. The different structures can account for this contrast.

146

Syntax and its Interfaces

holds also for Italian. Also in this case, the coordination data and the possibility of inserting material between the two verbs argue against this structure. The flat structure is therefore the most appropriate representation for Italian auxiliaries. It should be mentioned that a flat structure has also been proposed by Napoli (1981). Her evidence is based on four tests presented in Rizzi (1982) to argue that an infinitive and the following material do not form a constituent in a clitic climbing sentence. Napoli shows that the tests can also be used to argue in favour of the fact that a past participle and the following material do not form a constituent in the case of auxiliary verbs. The tests are based on pied-piping, clefting, right node raising, and complex NP shift. The flat structure proposed, in combination with the argument composition mechanism and the appropriate linearization constraints, can deal with the properties of Italian auxiliaries discussed in Section 4.2. I refer to Monachesi (1996a) for additional details on the analysis. 4.3.3. The syntactic structure of Romanian tense auxiliaries

In Romanian, the tense auxiliary and the lexical verb form a unit which cannot be separated by other elements. I have suggested that this property indicates that they are morphosyntactic words, which are phonologically dependent on an adjacent word, that is simple clitics. This is, however, only a theoretical classiﬁcation. The real issue is how simple clitics should be dealt with in the grammar. I suggest that an appropriate treatment of their status can be achieved through an interaction of syntax and phonology. I claim that their dependent status is best accounted for in syntax, that is the auxiliary and the nonfinite verb form a verbal complex. I show that this is the most appropriate way to deal with the adjacency requirements discussed in Section 4.2. This syntactic dependency is also reflected at the phonological level in which the tense auxiliary and the verb form a prosodic word. The verbal complex structure assumed in the case of Romanian tense auxiliaries requires an appropriate schema to license it. The underlying assumption is that this kind of ‘compounding’ occurs in syntax. As noted in Abeill´e and Godard (2003), it is generally assumed that complex predicates form a syntactic unit. Furthermore, they show that the fact that coordination of nonﬁnite verbs is always accepted provides additional evidence in favour of this assumption, otherwise coordination should be included as a regular morphological operation, which is not very economical: (228) Ion a cump˘arat s¸i citit aceast˘a carte. [ro] John has bought and read this book ‘John has bought and read this book.’

Syntax and its Interfaces

147

The schema suggested for the German verbal complex in Hinrichs and Nakazawa (1994) could also be adopted for this case:

(229) SS | L | C | HEAD verb → H word , C [SS | L | C | HEAD verb ] The schema combines a head which must be a word (the auxiliary verb in this case) with a verbal complement, and the result is a verb. It licenses the ‘verbal complex’ structure, repeated below: (217) Verbal complex VP

H HH

V

HH

V [AUX

+]

COMPS

V

The schema ensures that the auxiliary and the lexical verb combine together before non-verbal complements are added. That is, the verbal complex combines with its complements by means of SCHEMA 2 discussed in Chapter 2, Section 2.3, which licenses phrases where there is a lexical head daughter and any number of complement daughters. The structure above accounts for the fact that no element can intervene between the two verbs, which is the desired result. More generally, verbal complexes have been proposed in the literature to represent syntactic combinations of two verbs in order to exclude not only complements but also adverbs. It should be noted that if the flat structure is adopted, as in French and in Italian, it would be necessary to postulate special linearization constraints to account for the fact that it is not possible for subjects, adverbs, quantiﬁers, and complements to intervene between the two verbs. Furthermore, the structure predicts that in Romanian it is not possible to coordinate participles with their complements, given that in that configuration they are not sisters: (230) *Ion a cump˘arat aceast˘a carte s¸i citit primul capitol. [ro] John has bought this book and read first chapter ‘John has bought this book and read the first chapter.’ The underlying assumption is that it is not possible to coordinate sequences of complements belonging to different levels, as already confirmed by the French and Italian data (Abeill´e and Godard 2003). As for the hierarchical structure, it does not seem to be a plausible alternative in the case of Romanian auxiliaries. The standard constituency tests do not support the viewthat the nonﬁnite verb and its complements form a con-

148

Syntax and its Interfaces

stituent. In particular, this sequence cannot be cliticized (231a) or questioned (231b): (231) a. *Dragos¸ a crezut c˘a a int¸eles, dar nu l-a. [ro] Dragos¸ has believed that has understood, but not CL.ACC has ‘Dragos¸ believed to have understood, but he hadn’t.’ b. *Ce-a? – Vîndut c˘art¸ile lui. what has? – sold books.the his ‘What has he? He has sold his books.’ As in French and in Italian, the nonfinite complement cannot be deleted (232a), preposed (232b), or clefted (232c): (232) a. *Florin a terminat serviciul, dar Doina nu a. [ro] Florin has finished work.the, but Doina not has ‘Florin has finished his work, but Doina hasn’t.’ b. *Citit articolul acesta, nu, nu l-a. read article this, no, not CL.ACC have ‘Read this article, no, he hasn’t.’ c. *Ce Marius a, baut prea mult vin. what Marius has, drank too much wine ‘What Marius has, is to drink too much wine.’ The data above do not provide any evidence in favour of the claim that the nonﬁnite verb forms a constituent with its complements. Therefore, a hierarchical structure where the tense auxiliary subcategorizes for a VP complement is clearly unmotivated. Additional arguments against a hierarchical structure come from the fact that Romanian tense auxiliaries can be optionally placed after the verbal complement. This peculiar construction is the topic of the next section. 4.3.3.1. Romanian inverted auxiliaries Romanian allows for the possibility, in certain registers, to optionally place the tense auxiliary after the nonfinite verb: (233) a. V˘azut-am. [ro] seen have ‘I have seen.’ b. Mira-m-as¸. wonder-CL.ACC would ‘I would wonder it.’ c. Pleca-voi. leave shall ‘I shall leave.’

Syntax and its Interfaces

149

In example (233a), the auxiliary of the present perfect follows the past participle, in (233b) there is a conditional form in which the auxiliary follows the bare inﬁnitive, while in (233c) the future auxiliary comes after the bare inﬁnitive.7 Dobrovie-Sorin (1994) discusses these data at length, as well as Rivero (1994) who argues that they are triggered by illocutionary factors. However, this does not seem to be the case since native speakers have confirmed that this kind of inversion is only possible in some dialects, in poetry, and in curse expressions: they are thus limited to certain registers. If the two verbs belong to the same domain, as in the verbal complex conﬁguration, one can account for the inversion data by means of simple linearization principles. This is not feasible if a hierarchical structure is assumed since the two verbs would belong to different domains. In order to account for the inversion data, an approach similar to the one proposed by Pollard and Sag (1994) for English can be assumed. It relies on the use of the feature [INV boolean], which marks auxiliaries that can or cannot invert. Given the simplicity of the Romanian phenomenon, there is no need to assume a linearization approach as in Kathol (1995), which is necessary to deal with the more complex cases of German word order. Auxiliaries which cannot occur after the verb, such as the passive auxiliary, are specified as [INV −], while auxiliaries which optionally invert have [INV boolean], that is [INV +] or [INV −], as their value. This applies to the future and the present perfect auxiliaries. The following LP constraints account for the relevant orders:

• auxiliaries which are marked [INV −] must precede the lexical verb; • auxiliaries which are marked [INV +] must follow the lexical verb. It is thus possible to deal with the optionality of the phenomenon. In example (234a), the auxiliary is marked [INV −] and it is for this reason that it must precede the verb. On the other hand, in example (234b), the auxiliary is marked [INV +] and it must follow the verb according to the second linearization principle presented above: (234) a. Am v˘azut. [ro] have seen ‘I have seen.’ b. V˘azut-am. seen have ‘I have seen.’ 7 It should be noted that, in the case of the conditional, inversion is allowed only if object clitics are present, as discussed in more detail later in this section.

150

Syntax and its Interfaces

This account can easily deal with a peculiar fact of Romanian auxiliary inversion which has not received an analysis in the literature, but is reported in traditional grammars. Conditionals, unlike the other auxiliaries, allow for inversion only if object clitics are present, as shown in (233b), repeated below: [ro] (233b) Mira-m-as¸. wonder-CL.ACC would ‘I would wonder it.’ The auxiliary cannot occur after the verb if there are no object clitics: (235) *Mînca as¸. [ro] eat would ‘I would eat.’ There are various ways to interpret the way object clitics combine with the tense auxiliaries in the examples above. Given that the Romanian tense auxiliaries have clitic status, it could be argued that they combine with the object clitics in a cluster which is formed in morphology. In Section 4.5, however, I argue that this is not the case and that auxiliary verbs act as hosts for clitics, that is object clitics are attached to the auxiliary as a result of clitic climbing. Under this view, the example in (233b) represents a case in which the combination of the clitic with the auxiliary undergoes inversion. The auxiliary clitics have the status of morphosyntactic words which are (phonologically) dependent on a given host. Recall that in the analysis suggested in this chapter, this dependency is reflected at the syntactic level in terms of the verbal complex structure, while at the phonological level the tense auxiliary and the lexical verb form a prosodic word. The property that conditional auxiliaries allow for inversion only in the presence of object clitics can be easily accounted for if they are marked [INV boolean]. The lexical entry below contains partial information about the conditional auxiliary as¸: ⎤ ⎡ (236) PHON |SKEL as¸ ⎢ ⎤⎥ ⎡ ⎡ ⎤ ⎥ ⎢ verb ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎢ ⎥ ⎢ ⎥⎥ ⎢HEAD ⎣AUX + ⎦ ⎢ ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ INV boolean ⎢ ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎡ ⎤ ⎤⎥⎥ ⎢SS|L|C⎢ ⎡ ⎢ ⎥⎥ ⎢ w-ss ⎥ ⎢ ⎢ ⎢ ⎥ ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎢ ⎥⎥ ⎥ ⎢ ⎥ ⎢VAL ⎢COMPS ⎢ verb ⎣HEAD ⎦⎥ ⎥ ⎢ ⎢ ⎣ ⎦⎥ ⎥ ⎢ ⎦ ⎣ VFORM bse ⎦ ⎣

Syntax and its Interfaces

151

The sentence in (235) is ruled out by means of an implicational constraint like the following: (237)

⎡

verb ⎢HEAD ⎢ INV + ⎢ ⎢ ⎡ ⎡ ⎢ ⎢ w-ss ⎢ ⎢ ⎢ ⎢ ⎢ verb ⎢VAL ⎢COMPS ⎢ ⎣ ⎢ ⎣ HEAD ⎣

⎤

⎥ ⎥ ⎥ ⎤

⎤ ⎥ ⎥ ⎥→ CLTS nelist ⎥ ⎥ ⎥⎥ ⎥⎥ ⎥ ⎦⎥ ⎦⎥ ⎦ VFORM bse

It states that if a conditional verb is marked [INV +], its CLTS list must contain some members. As already discussed in the previous chapter, this feature encodes the information about those complements which are realized as clitics. The sentence in (235) does not comply with the constraint since in that case the CLTS list is empty: it is therefore ungrammatical. This is not the case for (233b) in which clitics are present. The CLTS list is not empty, thus the constraint in (237) is satisﬁed and the sentence in which the auxiliary occurs in an inverted conﬁguration is considered grammatical. Since no additional constraints are imposed, the auxiliary is also able to precede the nonfinite verb if there are clitics, as in the following sentence: (238) M-as¸ CL.ACC-would

mira. [ro] wonder

‘I would wonder it.’ It should be noticed that even if clitics are lacking, there is still a possibility of a configuration with inverted auxiliaries. However, in this case a different infinitive form, the so called ‘long infinitive’, must be present, which can only occur in the absence of clitics:8 (239) Mîncare-as¸. [ro] eat-would ‘I would eat.’ The sentence above, in which the auxiliary as¸ attaches to the long infinitive mˆıncare, is accounted for by the constraint below: 8 This -re form is much more nominal than clausal in its structure and function. It is called ‘long infinitive’ because of its morphological identity with those forms which were subsequently replaced by the short infinitive and because of the similarity with the Latin infinitive forms. Since constructions with the long infinitive are partly nominal, they are found frequently as objects, governed by prepositions, or as subjects (Mallinson 1986).

152

(240)

Syntax and its Interfaces ⎡

verb ⎢HEAD ⎢ INV + ⎢ ⎢ ⎡ ⎡ ⎢ ⎢ w-ss ⎢ ⎢ ⎢ ⎢ ⎢ ⎢VAL ⎢COMPS ⎢ ⎣HEAD verb ⎢ ⎣ ⎣

⎤

⎥ ⎥ ⎥

⎤ ⎤⎥ ⎥ ⎥→ CLTS elist ⎥ ⎥⎥ ⎥ ⎥⎥ ⎥ ⎦⎥ ⎦⎥ ⎦ VFORM long-inf

It states that if a conditional subcategorizes for a long infinitive, its CLTS list must be empty.

4.4. Mesoclisis in European Portuguese The inverted conjugations discussed in the previous section were also attested in older stages of Romance languages, in particular old Spanish. They still occur not only in Romanian, but also in modern European Portuguese (EP) in the future and conditional forms. However, these forms belong mainly to the literary language and they are disappearing from the spoken language, being replaced by the present and the imperfect indicative, respectively. In this section, I show that there are some crucial similarities between the Romanian and the EP auxiliaries which occur in these inverted paradigms. It is interesting to note that in both languages these constructions are rather marginal and limited to certain registers. Notice that the similarities between these two auxiliary systems are attested only when clitics are present. In European Portuguese, auxiliaries usually precede the main verb, as can be seen in the example below which illustrates a case of present perfect: (241) Tenho comprado muitas coisa. [pt] have bought many things ‘I have bought many things.’ Present perfect in EP is composed by the auxiliary ter (in the present tense) in combination with the past participle. The future and conditional, however, constitute an exception to this rule since a different form of the auxiliary is present and, in addition, the auxiliary follows thelexical verb, as can be seen

Syntax and its Interfaces

153

in the table below: (242) Future and conditional forms in EP

1 Sg 2 Sg 3 Sg 1 Pl 2 Pl 3 Pl

European Portuguese Future Conditional falar-ei falar-ia falar-ás falar-ias falar-á falar-ia falar-emos falar-íamos falar-eis falar-í falar-ão falar-iam

These paradigms have developed from an analytic construction involving an inﬁnitive together with the relevant form (present/imperfect) of the verb habere ‘to have’. Similarly for other Romance languages which have developed comparable paradigms; the table below illustrates Italian: (243) Future and conditional forms in Italian

1 Sg 2 Sg 3 Sg 1 Pl 2 Pl 3 Pl

Italian Future parler-ò parler-ai parler-à parler-emo parler-ete parler-anno

Conditional parler-ei parler-esti parler-ebbe parler-emmo parler-este parler-ebbero

French has a similar conjugation in the case of the future and the conditional: (244) Future and conditional forms in French

1 Sg 2 Sg 3 Sg 1 Pl 2 Pl 3 Pl

French Future manger-ai manger-as manger-a manger-ons manger-ez manger-ont

Conditional manger-ais manger-ais manger-ait manger-ions manger-iez manger-aient

154

Syntax and its Interfaces

The paradigms presented above show that in French and Italian the old auxiliary habere has been reanalysed into an inﬂectional afﬁx which marks the tense/aspect of the verb. A similar conclusion can be reached for EP where the conditional/future markers also seem to exhibit afﬁx status, as argued by Vig´ario (1999b). She shows that the verb stem and the afﬁxes are treated as one phonological word by the relevant phonological rules. In particular, there is one single primary stress and vowel reduction applies to the vowel that would bear stress if the afﬁxes were not added. The example below shows that the vowel [a] is reduced: this is the vowel that would bear main stress if the conditional marker were not present: (245) D[ɐ]r[í]amos. [pt] give.COND.PL ‘We would give.’ To summarize: the conditional and future markers have been reanalysed into inflectional affixes in those structures in which pronominal clitics do not intervene, either because they do not occur in the sentence or because they occur as proclitics. As already mentioned in the previous chapter, in EP proclisis is triggered by the presence of negation, interrogative words, prepositions, conjunctions, adverbs, and relative clauses, while enclisis represents the default configuration: um livro. [pt] (246) Ele deu-me he gave.CL.DAT a book ‘He gave me a book.’ An interesting peculiarity of EP is mesoclisis, a phenomenon according to which pronominal clitics intervene between the verb stem and the conditional/future markers: (247) a. Falar-lhe-emos. [pt] speak.CL.DAT.FUT ‘We will speak to him.’ b. *Falar-emos-lhe. speak.FUT.CL.DAT It is well known that the situation is different in French and in Italian where clitics cannot occur in this position with conditional and future forms: (248) a. *Parler-lui-a. [fr] speak.CL.DAT.FUT ‘We will speak to him.’

Syntax and its Interfaces

155

b. *Parler-gli-á. [it] speak.CL.DAT.FUT The conﬁguration attested in EP enclitic environments is similar to the one found in Romanian inverted conjugations. Also in that case, the object clitics can intervene between the lexical verb and the auxiliary which has not been reanalysed as an inﬂectional afﬁx, but is a simple clitic. In EP, Vig´ario (1999b) provides phonological evidence that, in the mesoclisis conﬁguration, the future/conditional markers behave as words and not as afﬁxes since they carry primary stress. In particular, she shows that these markers are not ignored by word stress rules, as is the case for object clitics: (249) *Fal[á]r-te-emos. [pt] Speak.CL.DAT.FUT ‘We will speak to you.’ The example above shows that stress cannot remain in its original place if the future/conditional marker is present. However, it is not the case that stress shift is triggered, as with inflectional affixes: (250) *Falar-te-[é]mos. [pt] Speak.CL.DAT.FUT ‘We will speak to you.’ Instead the marker carries its own primary stress, behaving as a word, both from a phonological and a morphosyntactic perspective, and not as an inﬂectional morpheme: (251) Fal[á]r-te-[é]mos. [pt] Speak.CL.DAT.FUT ‘We will speak to you.’ Additional evidence that the future/conditional marker does not behave as an inflectional affix comes from the fact that in inflectional environments, the first of two adjacent vowels deletes, as shown in (252a) where the vowel [a] is no longer present. This is not the case in the combination of an object clitic with a future/conditional marker, as illustrated by (252b). Example (252c) shows that vowel deletion results in an ungrammatical form: (252) a. Falo (< fala+o). [pt] speak ‘I speak.’

156

Syntax and its Interfaces b. Falar-t[j]-emos. speak CL.DAT.FUT ‘We will speak to you.’ c. *Falar-temos. speak CL.DAT.FUT ‘We will speak to you.’

The examples above provide evidence in favour of the word status, both from a morphosyntactic and from a phonological point of view, of these markers in EP, which behave as lexical auxiliaries. Two forms are thus attested: a synthetic one used in the absence of intervening clitics and an analytic one which is attested in the presence of object clitics. It should be mentioned that there are other similarities between the EP auxiliaries in the mesoclisis conﬁguration and the Romanian ones. In particular, adverbs cannot intervene between the verb and the future/conditional marker, as shown in (253a) where the adverb já ‘already’ separates the marker á from the verbal stem, resulting in an ungrammatical sentence. Example (253b) shows that the adverb should follow the verb or alternatively precede it (253c):

(253) a. *O João falar-lhe já á nisso. [pt] the John tell.CL.ACC already FUT it ‘John will already tell him about it.’ b. O João falar-lhe-á já nisso. the John tell.CL.ACC.FUT already it ‘John will already tell him about it.’ nisso. c. O João já falar-lhe-á the John already tell.CL.ACC.FUT it ‘John will already tell him about it.’

On the other hand, it should be noted that adverbs can separate lexical auxiliaries from the verb, as illustrated by the example below where the adverb já occurs between the auxiliary tinha and the nonﬁnite verb falado:

Syntax and its Interfaces

157

(254) O João tinha-lhe já falado nisso. The John had CL.ACC already told it ‘John had already told him about it.’ Given the similarities between Romanian and EP auxiliaries in the mesoclisis conﬁguration, I suggest that the syntactic structure proposed to deal with the properties of Romanian tense auxiliaries should be adopted to also account for EP. The future/conditional marker in EP therefore combines with the inﬁnitival in a verbal complex. The fact that nothing can intervene between the lexical verb and the marker argues in favour of this conﬁguration, which captures the intuition that the two elements form a unit. The data discussed in this section provide additional evidence in favour of a syntactic treatment of adjacency, which seems the most adequate to deal with the variety of possibilities attested within the Romance verbal complex. In the case of EP, the future/conditional markers do not have clitic status given that they can carry stress. They behave like words both from a morphosyntactic and from a phonological point of view, yet they pattern like the Romanian tense auxiliaries (which have clitic status) with respect to adjacency. The analysis in terms of syntactic structure allows for a common treatment of what synchronically appears as the same phenomenon (i.e. the same adjacency requirements). It should be noted that if one assumes a treatment of the adjacency properties of Romanian tense auxiliaries in the form dimension, it could not be extended to the EP future/conditional markers given their differences at this level. That is, while in EP the markers form a prosodic word of their own, this is not the case in Romanian where the clitic auxiliaries form a prosodic word with the verb. Given the verbal complex configuration, it is possible to account for the fact that nothing can intervene between the lexical verb and the marker. The analysis is, however, simpler than in Romanian as in EP the future/conditional markers always follow the lexical verb, while in Romanian inversion is an optional phenomenon. There is no need to employ the feature [INV] as in Romanian; a linearization principle orders the future/conditional marker after the infinitival:

• auxiliaries which are marked [VFORM future ∨ conditional] must follow the verb. The only elements that can occur between the lexical verb and the future/conditional marker are object clitics. Under the assumption that they are inﬂectional afﬁxes (Crysmann 2000b) on a par with French clitics (Miller and Sag 1997), Italian clitics (Monachesi 1999a), and Romanian clitics (Monachesi

158

Syntax and its Interfaces

2000), it is straightforward to deal with this fact. The clitics are enclitic to the inﬁnitival and are attached to it. The situation is, however, different in Romanian, where in the equivalent configuration, object clitics are attached to the auxiliary as a result of clitic climbing.9 The example (233b), repeated below, shows that the object clitic that surfaces is m. As already mentioned in Chapter 3, this is the form that occurs in front of verbs that begin with a or o, as in the case of the conditional auxiliary. It is not m˘a, the form that occurs in an enclitic configuration: (233b) Mira-m-as¸. [ro] wonder-CL.ACC would ‘I would wonder it.’ The form of the object clitic thus provides evidence that it is the combination of the clitic with the auxiliary that undergoes inversion in Romanian. The result is a configuration where the clitic intervenes between the two verbs. To summarize: the Portuguese phenomenon of mesoclisis exhibits some crucial similarities with that of inverted conjugations in Romanian and I have shown that it is possible to assume a common analysis for both based on linearization constraints and the assumption that the future/conditional marker forms a verbal complex with the lexical verb. In the next section, I turn back to Romanian tense auxiliaries and I take into consideration the way they combine with object clitics. I argue that Romanian auxiliaries share, with other Romance counterparts, the property of triggering clitic climbing.

4.5. Romanian tense auxiliaries and the distribution of pronominal clitics In the previous sections, I have shown that the different properties of Romanian tense auxiliaries and of French/Italian auxiliaries with respect to adjacency can be accounted for in terms of different syntactic structures. I have suggested that while in French and in Italian the auxiliary verb, the nonfinite verb, and the complements combine in a flat structure, in Romanian the auxiliary and the lexical verb form a verbal complex, which captures the intuition that the two verbs form a unit. I have also proposed that this structure can deal with EP auxiliaries in the mesoclisis configuration. Furthermore, I have claimed that the differences between Romanian and the other Romance languages with respect to adjacency are the result of Romanian patterning with the Balkan languages. 9 I refer to Section 4.5 for evidence that in Romanian the object clitics and the tense auxiliaries do not form a morphological cluster which has the lexical verb as host and for an analysis of their attachment in terms of clitic climbing.

Syntax and its Interfaces

159

In the following sections, I take into consideration the distribution of object clitics in the presence of tense auxiliaries. I show that this is an area in which Romanian differs from the Balkan (and Slavic) languages and patterns like the other Romance languages. I claim that Romanian exhibits clitic climbing and I suggest an analysis in terms of argument composition to deal with this phenomenon. 4.5.1. The position of pronominal clitics in the Romanian and Balkan clitic cluster

Romanian has a rich clitic inventory which differs from that of the other Romance languages, being more similar to that of Balkan (and Slavic) languages. It includes not only pronominal clitics, as in the other Romance languages, but also auxiliaries (as argued in Section 4.2), intensiﬁers, complementizers, and negation. These elements cluster together, as can be seen in the following examples from Dobrovie-Sorin (1994), which illustrate two maximal clitic strings in Romanian: (255) a. S˘a nu o that not CL.ACC

mai fi v˘azut. [ro] again be seen

‘That I should not have seen her again.’ ar mai fi întrebat. b. Nu l not CL.ACC would again be asked ‘He wouldn’t have asked him again.’ Schematically, the two sequences could be represented in the following way for (255a) and (255b), respectively: (256) Ordering templates for Romanian clitics SEQ. 255a SEQ. 255b

comp neg

neg pron

pron aux1

int int

aux2 aux2

The elements in (255) all have clitic status in Dobrovie-Sorin (1994); however, I show in the rest of this chapter that matters are more complex. More generally, ‘clitics’ has been used as a cover term for elements whose status is intermediate between that of an afﬁx and that of a word. I claim that the elements above do not all have the same status: while some exhibit affix-like properties, others share word-like behaviour. However, regardless of the terminology used, I believe that the most crucial issue is which module of the grammar can deal with their analysis. I claim that pronominal clitics and intensiﬁers are accounted for in morphology while the other elements in the clitic cluster in (256) are dealt with in syntax. More generally, an appropriate analysis of the properties

160

Syntax and its Interfaces

of these elements and the way they relate to each other requires an interaction of the different modules of the grammar. As already shown in (255b), if a tense auxiliary verb is present, the pronominal clitic, which is an argument of the embedded verb, must precede the auxiliary and cannot attach to the lower verb: (257) Le-am

v˘azut. [ro] have seen ‘I have seen them.’ CL.ACC

In the example above, the clitic le, which represents an argument of the lexical verb v˘azut, must precede the tense auxiliary and does not combine with the embedded verb that subcategorizes for it. Pronominal clitics, however, tend to follow the auxiliary clitics in Balkan and Slavic languages, as observed by Franks and King (2000). The position of object clitics in the cluster is therefore different from that attested in Romanian, as illustrated by the following examples from Bulgarian (Legendre 2000a), Macedonian (Legendre 1998), Serbian/Croatian (Franks and King 2000), and Slovenian (Franks and King 2000): (258) a. Az sum mu go dal. [bg] I AUX CL.DAT CL.ACC give ‘I have given it to him.’ b. Sum ti go dal. [mac] AUX CL.DAT CL.ACC give ‘I have given it to you.’ c. Ti si joj ih juˇce(r) dala. [sc] you AUX CL.DAT CL.ACC yesterday gave ‘You gave her to them yesterday.’ d. Prinesel sem ji jo. [sl] bring AUX CL.DAT CL.ACC ‘I brought it/her to her.’ The table below, which represents a modified version of that provided in Franks and King (2000), gives an overview of the clitic cluster in Serbian/Croatian (SC), Slovenian (SL), Macedonian (M), and Bulgarian (B):10 (259) Ordering templates for Slavic clitics 10 In Serbian/Croatian complementizers and negation also have clitic status but they are verbal proclitics unlike the other clitics which occur in second position. It is for these reasons that they are not included in the template by Franks and King (2000).

Syntax and its Interfaces SC SL M B

li (inter) naj (opt) da (comp) sˇte (aux)

aux bi/aux neg aux

pron pron aux pron

je (aux) je/fut aux pron e (aux)

161

e (aux)

It can be observed that the general tendency is for the pronominal cluster to follow the auxiliary clitics in these languages.11 Furthermore, the table shows that these languages exhibit a rich clitic system comparable to that of Romanian. It is well known that in some of these languages, such as Serbian/Croatian and Slovenian, clitics occur in second position in the clause while in others, such as Bulgarian and Macedonian, they attach to the verb (Franks and King 2000). However, in all these languages the clitics follow a rigid order and they form a cluster, which has received different analyses in the literature. 4.5.2. The ordering of Romanian clitics

It is often assumed that clitic ordering in the Slavic and Balkan languages is best expressed in the morphological component and this is typically done in templatic terms, as in Halpern (1995). Alternatively, a syntactic view of the clitic order could be adopted, as in Franks and King (2000). They argue that the clitic cluster is formed in syntax by successive head-adjunction and that this process entails most internal ordering properties of the cluster. Given the comparable structure of the Romanian and of the Balkan (and Slavic) cluster, similar approaches could also be adopted in the case of Romanian to deal with the ordering of the clitics. Under this view, the fact that the object clitic precedes the tense auxiliary might be due to the specific position that object clitics fill in the template. In particular, Popescu (2000) and Gerlach (1998, 2001) propose that Romanian object clitics cluster together with the tense auxiliaries in a clitic sequence which is formed in morphology. They assume that these elements all have the same status and they cluster before the lexical verb which acts as a host, as illustrated by the structure below: (260) [[ CL

CL AUX]

VERB]

In the rest of this section, I provide evidence that this is not the case in Romanian. 11 The only exception is represented by the third person singular form of certain tense auxiliaries which follow the object clitics, as well as by the future auxiliaries in Slovenian (Franks and King 2000). Notice that Bo sˇkovic´ (2001) mentions that the special position of the clitic je with respect to the other clitic auxiliaries might be due to the fact that it is in the process of losing its clitichood as discussed in Schütze (1994) and Tomic´ (1996). An alternative could also be that it has never fully gained its clitic properties. Notice that the position of je is similar to that of the tense auxiliaries in Romanian.

162

Syntax and its Interfaces

I have already claimed in the previous chapter that pronominal clitics are affixal elements that cluster together forming a unit in morphology. As for auxiliaries, they have the status of morphosyntactic words and are accounted for in syntax. Under this view, the position of the object clitics preceding the tense auxiliaries should be viewed as a case of clitic climbing. It is not the case that the object clitics and the tense auxiliaries combine in a morphological cluster which has the lexical verb as host. On the contrary, the tense auxiliary acts as host for the pronominal clitics and combines with them through a morphophonological spelling out process, as discussed in detail in the previous chapter. On the other hand, the auxiliary combines with the lexical verb in syntax through the schema introduced in Section 4.3.3. The following structure illustrates the way these elements cluster together: (261) [ [ [

CL CL] AUX] VERB]

In the rest of this section, I provide evidence that motivates the structure proposed. Recall that in Chapter 3, Section 3.2.2.2, it was shown that there are special forms of the dative clitic, that is ni, v˘a, li, which surface only when it is in combination with another pronominal clitic, as shown in (262a). On the other hand, the regular forms ne, ve, and le occur in isolation, as can be seen in (262b). I have argued that the forms ending in -i cannot be explained by means of appropriate phonological rules; they surface only in the presence of other object clitics (Monachesi 2000; Popescu 2000). The example in (262c) from Popescu (2000) gives additional evidence that the alternation between regular and special forms is not phonological since the form ne precedes a verb beginning with l like the clitic in (262a): (262) a. Ni-l

aduce. [ro] CL.DAT CL.ACC brings

‘He brings it to us.’ b. Ne

aduce.

CL.DAT brings

‘He brings us.’ c. Ne

leag˘a o sfoar˘a de picior. a rope from leg

CL.DAT bind

‘They bind us a rope to our legs.’ It is important to notice that if an auxiliary clitic is present, it is the regular form of the object clitic which is attested, as can be seen in (263):

Syntax and its Interfaces

163

(263) Ne-a

adus. [ro] has brought ‘He has brought us.’ CL.DAT

The regular form of the object clitic occurs both before auxiliary verbs which start with a vowel (263), as well as before those which begin with a consonant. The example below shows that the form ne precedes the future tense auxiliary vom: vom aminti. [ro] remember ‘We will remember.’

(264) Ne

CL.REFL will

The data above suggest that the auxiliary clitics behave differently from the object clitics when they cluster and I claim that this difference reveals a different status. Object clitics are afﬁxal elements and they cluster together in morphology: according to the view defended in the previous chapter, the combination of the pronominal clitics forms a new unit. On the other hand, auxiliary clitics do not have afﬁxal status; they are morphosyntactic words which are phonologically dependent on the verbal host. Tense auxiliaries are equivalent to lexical verbs from a morphosyntactic point of view. They differ, however, from a prosodic point of view since tense auxiliaries are dependent elements, while lexical verbs form independent prosodic words. Additional evidence for the differences between object clitics and tense auxiliaries comes from their different behaviour when they cluster with negation. As already mentioned in the previous chapter, negation can function as host for object clitics which can occur in a reduced form. The following examples show that the clitic îl occurs as -l in combination with nu: (265) a. Nu îl v˘ad. [ro] NEG. CL.ACC see v˘ad. b. Nu-l NEG. CL.ACC see ‘I don’t see him.’ On the other hand, tense auxiliaries can act as host for negation when they combine with it. In this case, it is nu which occurs in a reduced form and not the tense auxiliary, as can be seen in (266b): (266) a. Nu am v˘azut-o. [ro] neg. have seen CL.ACC b. N-am v˘azut-o. neg. have seen CL.ACC ‘I haven’t seen her.’

164

Syntax and its Interfaces

A similar behaviour can be observed with the lexical verb a avea ‘to have’, which can also act as host for negation. In this case, nu occurs in its reduced form, as can be seen in (267b): (267) a. Nu am reviste de mod˘a. [ro] NEG. have magazines of fashion b. N-am reviste de mod˘a. NEG.have magazines of fashion ‘I don’t have fashion magazines.’ These data show that auxiliaries behave differently from object clitics when they cluster with negation and that they pattern with the lexical verb a avea. I claim that this is due to their different status: object clitics are afﬁxes and they combine together in a morphological unit, while auxiliaries are morphosyntactic words on a par with lexical verbs. It should be mentioned, however, that the view defended in this section does not preclude the possibility that the object clitics cluster with the auxiliary clitics (and the verb) in a prosodic unit, as suggested in Popescu (2003). 4.5.3. The analysis of Romanian clitic climbing

In the previous section, it was shown that the distribution of the object clitics in the presence of tense auxiliaries differs from that attested in the Balkan (and Slavic) languages. I claim that Romanian behaves like the other Romance languages in this respect.12 Recall that in Romanian, the clitic precedes the tense auxiliary and does not combine with the verb that subcategorizes for it: (257) Le-am

v˘azut. [ro] have seen ‘I have seen them.’ CL.ACC

In French and Italian, the object clitic occurs in the same position, as can be seen in the examples below: (268) a. Gianni l’ha letto. [it] Gianni CL.ACC.has read ‘Gianni has read it.’ 12 There is another aspect in which Romanian tense auxiliaries pattern more like the Romance ones than the Balkan auxiliaries. Even though the formation of this compound tense is similar in terms of its form to that of the other Balkan languages, it is apparently different with respect to its meaning. Joseph (1999) claims that it has the meaning of a simple past and not that of a perfect; it is thus similar to the French passé composé. He concludes that (Daco)-Romanian seems more like a prototypical member of the Romance language group than a member of the Balkan Sprachbund in this respect, since it is taking part in the Romance drift concerning the value of tense formations.

Syntax and its Interfaces

165

b. Jean l’a lu. [fr] Jean CL.ACC.has read ‘Jean has read it.’ This phenomenon, attested in the various Romance languages, has been traditionally called clitic climbing, due to the analysis provided in the transformational grammar in which the clitic moves from its argument position to attach to the auxiliary verb. Clitic climbing is considered to be one of the syntactic diagnostics revealing complex predicate formation, the other being that the second predicate does not carry inﬂection (Abeill´e and Godard 2003). Complex predicates are characterized by the fact that a series of verbs behave as a single one with respect to certain syntactic or semantic properties. Romanian tense auxiliaries, as well as auxiliaries in other Romance languages including restructuring, causative, and attributive verbs, constitute complex predicates and they exhibit common properties of monoclausality. The Romanian data above therefore argue in favour of a monoclausal conﬁguration and against the biclausal one proposed by Dobrovie-Sorin (1994).13 I do not provide an analysis in terms of movement of clitic climbing, but I suggest instead a treatment in terms of argument composition (Hinrichs and Nakazawa 1994).14 Argument composition is a lexical operation according to which the auxiliary inherits the arguments of the embedded verb, including those which might be realized as clitics. It has already been adopted in the analysis of Italian clitic climbing due to auxiliary verbs (Monachesi 1996a) and restructuring verbs (Monachesi 1993a, 1993b, 1998a), as well as in the case of clitic climbing triggered by French auxiliaries (Abeill´e and Godard 1994, 2002) and Miller and Sag (1997) and French causatives Abeill´e et al. (1998). This analysis therefore allows for a common treatment for what appears to be the same phenomenon in Romanian and in the other Romance languages. In example (257), the tense auxiliary am has the following lexical entry associated with it: 13 Abeille´ and Godard (2003) mention other properties which correlate to clitic climbing in determining complex predicate formation. These are the possibility of triggering medio-passive forms, the occurrence of complex predicates in tough constructions, and the lack of scope on the nonﬁnite verb in the case of negation. I refer to Rizzi (1982), Burzio (1986), Monachesi (1999a), and Chapter 5 in this book for an analysis of Italian complex predicates, Aissen and Perlmutter (1983), Zagona (1988), and Moore (1991) for Spanish, Quicoli (1976) and Gonçalvès (1999) for Portuguese, and Kayne (1975), Legendre (1987), Rouveret and Vergnaud (1980), and Abeillé et al. (1998) for French. 14 I refer to Monachesi (1993c, 1996a) for a discussion of the problems related to an analysis of clitic climbing in terms of non-local features, which is the equivalent of movement within HPSG.

166 (269)

Syntax and its Interfaces ⎡ PHON

am

⎤

⎢ ⎡ ⎤⎥ ⎥ ⎢ verb ⎥ ⎢ ⎢ ⎢HEAD ⎥⎥ ⎥ ⎢ ⎢ ⎥ AUX + ⎢ ⎢ ⎥⎥ ⎢ ⎢ ⎡ ⎥ ⎤⎥ ⎢ ⎢ ⎥⎥ SUBJ 1 NP ⎢ ⎢ ⎥⎥ ⎡ ⎤ ⎢ ⎢ ⎢ ⎥⎥⎥ ⎢ ⎢ ⎢ ⎥⎥ ⎥ w-ss ⎢ ⎢ ⎢ ⎥⎥⎥ ⎢ ⎥ ⎢SS|L|C⎢ ⎢ ⎥⎥ ⎥ ⎢HEAD verb ⎥ ⎢ ⎢VAL ⎢ ⎥⎥⎥ ⎥ ⎥ ⎢ ⎢ ⎢COMPS 2 ⎢ ⎥ ⎢ ⎥ ⊕ 3 ⎥⎥ ⎢ ⎢ ⎢ ⎥⎥ ⎢ SUBJ 1 NP ⎥ ⎥ ⎢ ⎢ ⎢ ⎥ ⎥ ⎣VAL ⎦ ⎢ ⎢ ⎣ ⎦⎥⎥ ⎢ ⎢ ⎥⎥ COMPS 3 ⎢ ⎢ ⎥⎥ ⎢ ⎣ ⎦⎥ ⎦ ⎣ 1 2 3 ARG-ST , ⊕

In the description above the abbreviation w-ss stands for word synsem and it is employed to distinguish words from phrases. Alternatively the attribute [LEX + /−] could be used. Auxiliary verbs are raising verbs, as is indicated by the presence of the tag 1 both in the SUBJ list of the verb and in that of the verbal complement. The presence of the same tag indicates structure sharing; in other words, the syntactic and semantic information of the subject of the VP complement is shared with that of the subject of the auxiliary verb. Structure sharing also occurs in the case of the complements. The complements of the auxiliary verb are identiﬁed with those of the embedded verb, through the operation of argument composition. This is indicated by the presence of the tag 3 both in the COMPS list of the auxiliary verb and in that of the embedded verb. The effect of argument composition is even more visible if the feature ARG-ST is considered. Recall that the value of ARG-ST is a list of synsem objects, which are the syntactic arguments (including the subject) subcategorized for by the lexical item. In this case, the arguments of the embedded verb appear in the ARG-ST of the auxiliary. Argument composition is an operation which is related to a monoclausal structure and therefore associated with those complex predicates that trigger this condition. A natural question is whether argument composition could apply in biclausal structures. In principle, nothing in the formalism would prevent the application of this mechanism in such a configuration. However, it is the task of the theory to determine the possible ‘rules’ of the language which are postulated on the basis of empirical evidence. In the particular case, given the lexical character of the operation, it is possible to restrict it to those trigger verbs and lexical verbs that are clause-mates. The analysis of Romanian tense auxiliaries presented above interacts in the appropriate way with the analysis of Romanian clitics proposed in the previous

Syntax and its Interfaces

167

chapter. It is thus possible to provide an account of the distribution of the clitics exemplified by (257), repeated below: v˘azut. [ro] have seen ‘I have seen them.’

(257) Le-am CL.ACC

In this case, the auxiliary inherits from the past participle the direct object, which is eventually realized as clitic: ⎡ ⎤ (270) ⎣COMPS V COMPS 1 , 1 NP [acc][3fpl] ⎦ The Complement Cliticization Lexical Rule which I have introduced in the previous chapter can then apply to license cliticized verbs: (271) Complement Cliticization Lexical Rule (CCLR) ⎡ ⎤ word ⎡ ⎤ ⎢ ⎥ 1 verb ⎢HEAD ⎥ VAL | COMPS ⎢ ⎥→ ⎣ ⎦ ⎢ ⎥ 2 list (cl-ss) ⎣VAL | COMPS 1 2 ⎦ CLTS CLTS elist Recall that the rule relates verbs which subcategorize for certain complements to other ones with the same properties except that their subcategorization list is reduced. Its effect is that the relevant complement is removed from the COMPS list and added as value of the CLTS list: ⎡ ⎤ (272) ⎥ ⎢VAL | COMPS V COMPS 1 ⎥ ⎢ ⎥ ⎢ ⎦ ⎣ 1 CLTS NP[acc][3fpl] A constraint like the following is responsible for the spelling out of the cliticized verb form: (273) Realization of the accusative, third person feminine plural clitic ⎤ ⎡ ⎡ ⎤⎤ ⎡ complex-morph prefix ⎥ ⎣ ⎢ ⎦⎦ ⎦ → AFFIX ⎣ ⎣ STEM | SS | L | C | CLTS NP acc [3fpl] PHON le Its effect is that if the verb has an NP accusative, third person feminine in its CLTS list, it is realized as the clitic le. In addition, the following constraint, which

168

Syntax and its Interfaces

I have already introduced in the previous chapter, accounts for the fact that the clitic is proclitic to the auxiliary: (274) ⎡

word

⎤

⎡

PHON 1

⊕

⎤

2

⎡ ⎤⎥ ⎢ complex-morph ⎥ ⎤⎥ ⎢ ⎢ ⎡ ⎤⎥⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎥ ⎥ ⎢ word ⎤⎥⎥ ⎢ ⎢ ⎥ ⎥ ⎣ ⎦ ⎥⎥ → ⎢ ⎢STEM ⎥⎥ 2 ⎢ ⎥ ⎥ ⎥ ⎢ ⎥⎥ PHON ⎥⎥ ⎥ ⎢MORPH ⎢ ⎥⎥ ⎡ ⎤⎥⎥ ⎢ ⎥ ⎢ ⎦⎥ ⎦⎦ ⎢ ⎢ ⎥⎥ prefix ⎢ ⎥⎥ ⎢ VFORM fin ⎦⎦⎦ ⎣AFFIX⎣ ⎣

⎡ ⎢ ⎢ complex-morph ⎢ ⎡ ⎢ ⎢ ⎢ ⎢ word ⎢ ⎢MORPH ⎢ ⎢ ⎢ ⎢STEM ⎢ verb ⎢ ⎣ ⎣ ⎣ SS | L | C | H

PHON 1

The result is a cliticized verb form like the following:15 (275) ⎡ ⎤ word ⎢ ⎥ ⎢PHON le-am ⎥ ⎢ ⎥ ⎢ ⎤⎥ ⎡ ⎢ ⎥ complex-morph ⎢ ⎤⎥⎥ ⎡ ⎢ ⎢ ⎥ ⎥⎥ ⎢ ⎢ word ⎥ ⎢ ⎢ ⎥⎥⎥ ⎢ ⎢ ⎢ ⎥⎥⎥ ⎢PHON am ⎢ ⎢ ⎥⎥⎥ ⎢ ⎢ ⎢ ⎥⎥⎥ ⎢ ⎢ ⎡ ⎤ ⎢ ⎥⎥⎥ ⎢ ⎢ ⎢ ⎥⎥⎥ ⎢ verb ⎢ ⎢ ⎥⎥⎥ ⎢ ⎢ ⎥ ⎢ HEAD ⎢ ⎥⎥⎥ ⎢ ⎢ ⎥ ⎢ ⎢ AUX + ⎥⎥⎥ ⎢ ⎢ ⎥ ⎢ ⎢ ⎥⎥⎥ ⎢ ⎢ ⎥ ⎢

⎢ ⎥⎥⎥ ⎢ ⎥ ⎢STEM ⎢ ⎢ 2 NP acc ⎥⎥⎥ ⎢ CLTS ⎢ ⎥ ⎢ ⎢ [ 3fpl ] ⎥⎥⎥ ⎢1 ⎢ ⎥ ⎢MORPH ⎢ ⎥⎥⎥ ⎢ SS | L | C⎢ ⎥ ⎢ ⎢ ⎡ ⎤ ⎥⎥⎥ ⎢ ⎢ ⎥ ⎢ ⎢ ⎥⎥⎥ ⎢ w-ss ⎢ ⎥ ⎢ ⎢ ⎥ ⎢ ⎢ ⎥ ⎢ ⎢ ⎥ ⎢ ⎥ ⎥⎥ ⎥⎥ ⎢ ⎢ ⎥ ⎢ ⎢ ⎥ HEAD verb ⎢ ⎥ ⎥ ⎢ VAL | COMPS ⎣ ⎢ ⎥ ⎥ ⎢ ⎢ ⎦ ⎦⎥⎥⎥ ⎢ ⎢ ⎣ ⎢ ⎦⎥⎥ ⎣ ⎢ ⎢ ⎥ VAL | COMPS 2 ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎢ ⎤ ⎡ ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎢ prefix ⎥⎥ ⎢ ⎢ ⎦ ⎢ ⎦⎥ ⎣AFFIX ⎣ ⎢ ⎥ le PHON ⎢ ⎥ ⎣ ⎦ SYNSEM 1 15 The analysis in terms of argument composition, which I have proposed to deal with clitic climbing, also produces the correct output in the absence of pronominal clitics. In this case, the Complement Cliticization Lexical Rule does not apply and the arguments are realized as full NPs and subject to the standard Linear Precedence constraints. The situation is slightly different in the case of restructuring verbs which I discuss in Chapter 5, Section 5.4.

Syntax and its Interfaces

169

As already mentioned, Romanian clitics usually precede auxiliary verbs. However, there is an exception to this generalization, since the third person feminine clitic o must attach to the past participle, as in (276a), and cannot precede the auxiliary (276b) (Dobrovie-Sorin 1994): (276) a. Am v˘azut-o. [ro] have seen.CL.ACC ‘I have seen her.’ b. *O-am v˘azut. CL.ACC.have seen This is also the case for the conditional. The clitic must attach to the bare infinitive, as in (277a), and not to the auxiliary (277b): (277) a. As¸ vedea-o. [ro] would see.CL.ACC ‘I would see her.’ b. *O-as¸ vedea. CL.ACC.would see I suggest that the reason for the ill-formedness of the examples in (276b) and (277b) is of a phonological nature. The clitic o cannot precede an auxiliary which begins with a vowel. In the case of the future paradigm, the auxiliary begins with a consonant and the clitic o can occur either after the inﬁnitive (278a) or in front of the auxiliary (278b): (278) a. Voi vedea-o. [ro] will see.CL.ACC ‘I will see her.’ b. O voi vedea. CL.ACC will see The construction is similar, in one of the few other contexts in which clitic climbing is triggered in Romanian, namely in the presence of the modal verb a putea ‘can’: (279) O

pot vedea. [ro] see ‘I can see her.’ CL.ACC can

In the case above, the modal begins with a consonant and the pronominal clitic must attach to it, like all the other object clitics.16 16

A detailed analysis of Romanian modal verbs is postponed until Section 4.10.2.

170

Syntax and its Interfaces

It should be noted that the clitic o can occur before a main verb beginning with a vowel, as the following example from Boˇskovi´c (2001) shows: (280) O

am. [ro]

CL.ACC have

‘I have her.’ In order to account for these peculiar facts, past participles (and bare inﬁnitives) should be excluded as possible hosts for pronominal clitics. They can be allowed as possible hosts only in the case of the clitic o. Given the constraints proposed for the spelling out of the cliticized verb form, it is quite straightforward to add additional conditions which produce the desired results. Therefore, the constraint in (273), presented above, should be modiﬁed in the following way: (281) Realization of the accusative, third person feminine plural clitic ⎤ complex-morph ⎤ ⎡ ⎡ ⎢ ⎤⎤ ⎡

⎥ ⎥ ⎢ prefix ⎥ ⎢ ¬ (psp ∨ bare-inf) HEAD VFORM ⎥⎥→ ⎣AFFIX ⎣ ⎢ ⎦⎦ ⎢ ⎥⎥ ⎢STEM | SS | L | C⎢ PHON le ⎥⎥ ⎢ ⎢ ⎦⎦ ⎣ ⎣ CLTS NP acc [3fpl] ⎡

The additional condition states that the VFORM of the host must be different from the past participle or from the bare infinitive. However, the condition is not present in the constraint that is responsible for the realization of the clitic o. In this way, it is possible to account for examples (276a) and (277a), but it is also necessary to exclude those cases in which the clitic o attaches to the auxiliary, exemplified by (276b) and (277b). This can be achieved by adding further conditions to the constraint that accounts for the realization of the clitic o: ⎤

⎡

word ⎢ (282) ⎣

MORPH | ST | SS | L | C | CLTS

NP

acc [3fsg]

⎥ ⎦

→

Syntax and its Interfaces

171

⎤⎤ complex-morph ⎢ ⎤⎥⎥ ⎡ ⎢ ⎢ ⎥⎥ ⎢ word ⎢ ⎥⎥ ⎢ ⎡ ⎤ ⎥ ⎢ ⎥ ⎢ ⎢ ⎢ ⎥⎥ ⎢ 2 1 ⎥⎥ ⎢ VOW ⊕ ⎢ ⎥ ⎢ ⎢ ⎥⎥⎥⎥ ⎥ ⎢ ⎢PHON ⎢ ⎢ ⎣ ⎦⎥⎥ ⎥ ⎢ ⎢ ⎢ 3 ⎥ ⎢ ⎥⎥ ⎢ SKEL 1 ⊕ ⎥ ⎢ STEM ¬ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢MORPH⎢ ⎡ ⎢ ⎥⎥ ⎢ ⎤ ⎥ ⎥ ⎢ ⎥ ⎢ ⎢ ⎢ ⎥⎥ ⎢ verb ⎥⎥ ⎢ ⎢ ⎣ ⎦ ⎦ ⎣ SS | L | C HEAD ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ AUX + ⎢ ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎤ ⎡ ⎢ ⎥⎥ ⎢ ⎢ affix ⎥⎥ ⎢ ⎢ ⎦⎥ ⎣ ⎦ ⎣ AFFIX ⎦ ⎣ PHON | SKEL o ⎡

⎡

The constraint states that if a STEM contains an accusative, third person singular feminine element in its CLTS list, the clitic o must also be present in the structure, and the verb is not an auxiliary that begins with a vowel. Recall that the clitic o can combine with auxiliaries that begin with a consonant, as shown by example (278b). Furthermore, the clitic can combine with a main verb beginning with a vowel, as can be seen in (280). As already mentioned, if two object clitics are present, one of which is the clitic o, as well as an auxiliary verb, the two object clitics do not cluster together: (283) T¸i-am [ro] dat-o. CL.DAT.have given.CL.ACC ‘I have given it to you.’ Split clitics are very unusual both in Romance and in Balkan languages. Other examples can be found in French causative constructions in which a reﬂexive clitic remains attached to the inﬁnitival verb, while object clitics, which correspond to complements of the inﬁnitival, attach to the causative verb (Kayne 1975). Kayne (1991) mentions that split clitics are also attested in FrancoProvenc¸al. In constructions where there is an auxiliary and a past participle it is possible to have one clitic attached to the former and another one attached to the latter. The case of Romanian split clitics can be easily dealt with under the present analysis. The Complement Cliticization Lexical Rule in (271) accounts for the fact that the direct object is realized as a clitic. In particular, the relevant information is encoded as a value of the CLTS list and spelled out by the constraint (282), as the clitic o. The result is a cliticized verb form like the following:

172

Syntax and its Interfaces

(284) ⎡

⎤ word ⎢ ⎥ ⎢PHON dat-o ⎥ ⎢ ⎥ ⎢ ⎤ ⎡ ⎤⎥ ⎡ ⎢ ⎥ word ⎢ ⎥ ⎥⎥ ⎢ ⎢ ⎥ ⎢ ⎥ ⎢ ⎢ ⎥⎥⎥ ⎢PHON dat ⎢ ⎢ ⎥⎥ ⎢ ⎢ ⎢ ⎥ ⎡ ⎤⎥⎥ ⎢ ⎡ ⎤ ⎢ ⎢ ⎥ ⎥⎥ ⎢ ⎥ ⎢ ⎢ ⎥⎥⎥ ⎢ verb ⎢ ⎢ ⎥ ⎢ ⎥ ⎥ ⎢ ⎥ ⎢ ⎢ ⎥ ⎢HEAD ⎢ ⎥⎥⎥ ⎢ ⎣AUX − ⎦ ⎥⎥ ⎢ ⎢ ⎢ ⎥ ⎥ ⎢ ⎥ ⎢STEM ⎢ ⎢ ⎢ ⎥⎥⎥⎥ ⎢ ⎢ ⎢ ⎥⎥⎥⎥ VFORM psp ⎢ ⎢ ⎢ ⎥⎥ ⎢ 1 SS | L | C⎢ ⎥ ⎢MORPH ⎢ ⎥ ⎢ ⎥⎥⎥ ⎢ ⎢ ⎢ ⎥ ⎢VAL | COMPS NP dat ⎥⎥⎥ ⎢ ⎥⎥ ⎢ ⎢ ⎢ ⎥ ⎥ ⎢ ⎥ ⎢ ⎢ ⎢ ⎥⎥⎥⎥ ⎢ ⎢ ⎢ ⎢ ⎥⎥⎥⎥ ⎢ ⎢ ⎢ ⎣ ⎦⎦⎥⎥ ⎣ ⎢ ⎢ ⎥ CLTS NP acc [3fsg] ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎢ ⎤ ⎡ ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎢ suffix ⎥⎥ ⎢ ⎢ ⎦ ⎢ ⎦⎥ ⎣AFFIX ⎣ ⎢ ⎥ PHON o ⎢ ⎥ ⎣ ⎦ SYNSEM 1

It should be noted that the clitic o is enclitic to the past participle. This is accounted for by the following constraint that deals with enclitics: ⎤ word ⎤⎥ ⎡ ⎢ ⎥ ⎢ complex-morph ⎢ ⎤⎥⎥ ⎡ ⎢ ⎢ ⎥⎥ ⎢ ⎢ word ⎥⎥ ⎢ ⎥→ ⎢ ⎤ ⎡ ⎥ ⎢ ⎢MORPH ⎢ ⎥⎥ ⎥⎥ ⎢STEM ⎢ verb ⎥ ⎢ ⎥ ⎢ ⎢

⎦⎦⎥ ⎢ ⎣SS | L | C | H ⎣ ⎦⎥ ⎣ ⎦ ⎣ VFORM nonfin ∨ imper ⎡

(285)

⎡

PHON 2

⊕

⎤

1

⎡ ⎤⎥ ⎢ ⎢ complex-morph ⎥ ⎢ ⎡ ⎤⎥⎥ ⎢ ⎢ ⎥ ⎢ ⎥⎥ ⎢ word ⎢ ⎥ ⎢ ⎦⎥⎥ ⎢STEM ⎣ ⎢ ⎥ 2 ⎢ ⎥⎥ ⎢ PHON ⎥⎥ ⎢MORPH ⎢ ⎡ ⎤⎥⎥ ⎢ ⎢ ⎢ ⎥⎥ ⎢ suffix ⎢ ⎥⎥ ⎢ ⎦⎦⎦ ⎣AFFIX⎣ ⎣ PHON 1

Syntax and its Interfaces

173

The constraint above states that, if the verb is nonfinite (thus also a past participle) or imperative, the clitic follows it. This is indicated by the fact that the phonology of the clitic identified by the tag 1 follows that of the verb, indicated by the tag 2 . Going back to the split clitic case exemplified by (283), note that while the constraint in (282) allows for the clitic o to combine with the past participle, this is not the case for the clitic ¸ti. Recall that the spell out constraints do not allow pronominal clitics, different from o, to combine with past participles and bare infinitives. The clitic ¸ti is thus licensed next to the auxiliary by the mechanism of argument composition. The cliticized verb form ¸ti-am is thus obtained: (286)⎡ ⎤ word ⎢ ⎥ ⎢PHON ¸ti-am ⎥ ⎢ ⎥ ⎢ ⎤⎥ ⎡ ⎢ ⎥ complex-morph ⎢ ⎥ ⎢ ⎤ ⎡ ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ word ⎢ ⎥⎥ ⎢ ⎢ ⎥ ⎢ ⎥⎥ ⎢ ⎢ ⎥ ⎢ ⎥⎥ ⎢ PHON am ⎢ ⎥ ⎢ ⎥⎥ ⎢ ⎡ ⎤ ⎢ ⎥ ⎢ ⎥⎥ ⎢ ⎢ ⎥ ⎢ ⎥⎥ ⎢ verb ⎢ ⎥ ⎢ ⎥⎥ ⎢ ⎢ ⎥ ⎢ ⎥ ⎢ HEAD ⎥⎥ ⎢ ⎢ ⎥ ⎢ ⎥ ⎢ AUX + ⎥⎥ ⎢ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎥ ⎢ ⎢ ⎥ ⎢ ⎥ ⎢

⎥⎥ ⎢ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎥ ⎢ 2 NP dat ⎢ ⎥ ⎢ ⎥ ⎢ CLTS ⎥⎥ ⎢ [2sg] ⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎥ ⎢STEM ⎢ ⎢ ⎥ ⎢ ⎥ ⎡ ⎤ ⎥⎥ ⎢ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎥ ⎢ ⎥ ⎢MORPH ⎢ ⎥ ⎢ w-ss 1 ⎥⎥ ⎢ SS | L | C ⎢ ⎥ ⎢ ⎥ ⎢ ⎢ ⎡ ⎤⎥ ⎥⎥⎥⎥ ⎢ ⎢ ⎢ ⎥ ⎢ ⎢ ⎥ ⎢ verb ⎢ ⎢ ⎢ ⎥⎥ ⎢ ⎥⎥ ⎥⎥ ⎢ ⎢ ⎥ ⎢ ⎥ ⎥ ⎢ ⎢ ⎥ ⎢ ⎥ ⎢ ⎢ ⎢ ⎥⎥ ⎢ HEAD ⎣AUX − ⎦⎥ ⎥ ⎥⎥ ⎢ VAL | COMPS ⎢ ⎢ ⎥ ⎢ ⎥ ⎥ ⎢ ⎢ ⎥ ⎢ ⎢ ⎥⎥⎥ ⎢ ⎢ ⎢ ⎥⎥ ⎢ VFORM psp ⎥ ⎢ ⎥ ⎢ ⎥ ⎥ ⎢ ⎢ ⎥ ⎢ ⎢ ⎢ ⎢ ⎦ ⎥⎥⎥ ⎣ ⎥⎥ ⎢ ⎣ ⎦ ⎢ ⎥ ⎦ ⎣ ⎥ ⎢ VAL | COMPS 2 ⎢ ⎥⎥ ⎢ ⎢ ⎥ ⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎤ ⎡ ⎢ ⎥ ⎥ ⎢ ⎢ prefix ⎥⎥ ⎢ ⎢ ⎦⎥ ⎣AFFIX ⎣ ⎦ ⎢ ⎥ ⎢ ⎥ PHON ¸ti ⎢ ⎥ ⎣ ⎦ SYNSEM 1

The description above states that the auxiliary am has one element in the CLTS list as a result of the application of the CCLR in (271). The realizational constraint below spells it out as the clitic ¸ti:

174

Syntax and its Interfaces

⎤ complex-morph ⎤ ⎡ ⎢

⎥ ⎥ ⎢ ⎢ ⎢HEAD VFORM ¬ (psp ∨ bare-inf) ⎥⎥ → (287) ⎢ ⎥⎥ ⎢STEM | SS | L | C⎢ ⎥⎥ ⎢ ⎥ ⎢ ⎦⎦ ⎣ ⎣ CLTS NP dat [2sg] ⎤⎤ ⎡ ⎡ prefix ⎦⎦ ⎣AFFIX ⎣ PHON ¸ti ⎡

It should be noted that the constraint in (274) previously presented is responsible for the linearization of the clitic as proclitic. These data have received a different analysis in Popescu (2000) within the OT framework. As already discussed in Section 3.10, she proposes that clitic clusters tend to have a minimal number of syllables, and it is for this reason that some of the clitics are reduced. However, this reduction is not attested in the case of the clitic o, even though it could be reduced to a glide. Popescu claims that this clitic consists only of this vowel and cannot be reduced when it is adjacent to another vowel to form a diphthong. It is for this reason that the clitic encliticizes in spite of the proclitic context. She formulates an OT analysis that tries to capture this intuition. In particular, the constraint CONSTITUENT, which says that morphemes must project at least one syllable-constituent (i.e. they cannot consist of a glide only), plays a relevant role, being ranked very high. However, she does not say what happens in the case of examples such as (280), repeated below, in which the clitic occurs before a main verb beginning with a vowel: (280) O

am. [ro]

CL.ACC have

‘I have her.’ Furthermore, Boˇskovi´c mentions that in the example below, the clitic o contracts with the auxiliary, resulting in a glide: moart˘a! [ro] (288) Vedea-o-as¸ see.CL.ACC.would dead ‘May I see her dead!’ It is, therefore, questionable whether the explanation given by Popescu is indeed the correct one. Boˇskovi´c (2001) suggests a different analysis with respect to the position of the clitic o. As discussed in Chapter 3, Section 3.7, he assumes that movement leaves behind a copy of the moved element. He follows several authors who

Syntax and its Interfaces

175

have recently argued that the choice concerning which members of the chain survive deletion is attested not only at LF but also at PF. He adopts this assumption in his analysis of Slavic clitics and extends it also to the Romanian facts. Boˇskovi´c suggests that in the case of Romanian, the clitic o undergoes clitic climbing, which he considers a syntactic process involving clitic movement. Thus, in example (276a), repeated below, it is the lower copy of the clitic which is pronounced: (276a) Am v˘azut-o. [ro] have seen.CL.ACC ‘I have seen her.’ If the higher one were pronounced, it would not lead to a legitimate PF representation, since the clitic should undergo weakening, but o cannot take part in this process. It is difficult to evaluate the validity of this analysis which is only concerned with the problem of the placement of clitic o, since it is not clear how it could be extended to deal with the more general issues of Romanian cliticization.

4.6. Romanian tense auxiliaries and monosyllabic intensifiers In Section 4.2, it was shown that adverbs cannot intervene between the tense auxiliary and the lexical verb in Romanian. However, a small class of monosyllabic intensiﬁers can occur in this position. They are: mai ‘again’, cam ‘little’, prea ‘very/too much’, ¸si ‘also/already’, and tot ‘still’. In this section, I discuss the properties of these elements and I show that contrary to what has been assumed in the literature, they do not exhibit uniform behaviour. Although they share many common characteristics, they also differ from each other in some crucial respects. It is suggested that their properties argue in favour of an afﬁxal status on a par with pronominal clitics. Furthermore, a lexical analysis is proposed.17 Romanian intensifiers are the only elements which can occur between the tense auxiliary and the nonfinite verb:18 (289) Am mai/s¸i/cam/tot cîntat. [ro] have INT. sung ‘I have again/already/a little/still sung.’ 17 This section is based on data collection carried out by Manuela Lehau. I would like to thank her for sharing her data with me. 18 The aspectual marker fi can also occur in this position. I refer to Section 4.9 for a detailed discussion of the properties of this element.

176

Syntax and its Interfaces

The example above shows that they behave as a class with respect to this property with the exception of prea which can occur between the two verbs only if negation is also present (290a). Example (290b) where negation is absent is thus ungrammatical: (290) a. El n-a prea venit pe la noi. [ro] he NEG.has much come by to us ‘He hasn’t come by much.’ b. *El a prea venit pe la noi. he has much come by to us It should be noted that prea (291a), as well as cam (291b), can occur before the tense auxiliary, a position which is generally banned for most intensifiers, as example (291c) shows: (291) a. Prea ai venit pe la mine. [ro] much have come by ‘You have come by often.’ b. ?El cam ar veni. he little would come ‘He would come rarely.’ c. *El s¸i a venit. he also has come ‘He has also come.’ It should be mentioned that examples such as (291b) are usually attested in spoken language. The modal verb a putea differs from the tense auxiliaries since intensifiers cannot separate it from the infinitival (292a). They usually precede it (292b): (292) a. *(Nu) pot mai/cam/prea/tot veni. [ro] come NEG. can INT. ‘I can(not) come again/a little/too much/still.’ b. (Nu) mai/cam/prea/tot pot veni. can come NEG. INT. ‘I can(not) come again/a little/too much/still.’ This behaviour is rather surprising since elements such as adverbs and subjects can usually occur between the two verbs. It should be noticed, however, that some speakers consider the following example grammatical:

Syntax and its Interfaces

177

(293) El poate s¸i lucra. [ro] he can also work ‘He can also work.’ In the sentence above, ¸si intervenes between the modal and the infinitival, while an example such as the following in which ¸si precedes the modal is considered ungrammatical: (294) *El s¸i poate lucra. [ro] he also can work ‘He can also work.’ These data suggest that in this case as well, the intensifiers do not really behave as a class.19 The data above suggest that Romanian intensiﬁers have a constrained distribution. It should be added that they must precede the verb and nothing can separate them from it: vˇad. [ro] (295) a. Mai/cam/nu prea/tot vreau s˘a te want that CL.ACC see INT. ‘I want to see you again/a little/much/still.’ b. *Vreau s˘a te vˇad mai/cam/prea/tot. want that CL.ACC see INT. ‘I want to see you again/a little/much/still.’ The only exception is the aspect marker fi, since this element, which is discussed in detail in Section 4.9, can occur between the intensifiers and the verb: (296) As¸ mai/cam/prea/tot fi stat. [ro] have stayed would INT. ‘I would have stayed more/a little/much/still.’ The example below shows that the intensifiers cannot occur between fi and the past participle:20 (297) *As¸ fi mai/s¸i/cam/prea/tot stat. [ro] would have INT. stayed ‘I would have stayed more/also/a little/much/still.’ It is interesting to note that the intensifiers usually occur between pronominal clitics and the verb: 19 20

I postpone the discussion of these examples and of the modal a putea to Section 4.10.2. According to some speakers tot can occur between fi and the past participle.

178

Syntax and its Interfaces

(298) Îl

mai v˘ad. [ro] see

CL.ACC again

‘I see him again.’ However, mai or tot can also precede the object clitics; there is thus a certain degree of variation with respect to this property: (299) a. ?Mai îl

v˘ad. [ro]

INT. CL.ACC see

‘I see him again.’ b. Tot îl

v˘ad.

INT. CL.ACC see

‘I see him too often.’ To summarize: Romanian intensifiers exhibit uniform behaviour with respect to certain properties such as the fact that they must precede the verb in simple tenses and they should precede the aspect marker fi. They usually cannot intervene between the modal a putea and the infinitival with the exception of ¸si, while they are the only elements which can occur between the tense auxiliaries and the lexical verb. However, the intensifier prea can appear in this position only if negated. As for the position before the tense auxiliary and with respect to pronominal clitics, a certain degree of variation is attested. Given the properties of Romanian intensifiers, it can be argued that they exhibit affix-like behaviour on a par with object clitics. In particular, the fact that they can occur between a pronominal clitic and the verb, as shown by example (298) repeated below, provides motivation in favour of this position: (298) Îl

mai v˘ad. [ro] CL.ACC again see

‘I see him again.’ If it is accepted that pronominal clitics have affixal status, one must conclude that the intensifiers are also affixes since they can occur between the object clitics and the verb. Note that other types of adverbs cannot occur in this position: (300) *Nu (î)l-adesea v˘ad. [ro] NEG. CL.ACC.often see ‘I do not see him often.’ The prosodic weight of the adverb does not play a role: even light adverbs cannot intervene between the clitic and the verb.

Syntax and its Interfaces

179

Additional evidence in favour of the affixal status of these intensifiers is provided by the fact that they can also intervene between the negative element ne and the gerund (or participle): (301) Nemaiavînd bani, Radu a plecat. [ro] NEG.INT.having money, Radu has left ‘No longer having money, Radu left.’ The prefix status of the negative element ne is rather uncontroversial.21 It is used to negate adjectives, participles, and gerunds and it is comparable to the English negative prefix un: (302) Sînt nefericit. [ro] am unhappy ‘I am unhappy.’ The affixal status of the intensifiers is also claimed by Legendre (1999). She suggests that the intensifier mai should be considered a prefix as, for example, the French re- in relire ‘read again’. She also notes that mai does not show a high degree of selection with respect to the host since it can also modify adjectives and adverbs.22 However, a similar observation can be made for French re-, which is uncontroversially a prefix. Thus, she concludes that a high degree of host selection is a weak criterion to determine affixal status, contrary to the claim made by Zwicky and Pullum (1983). Even though pronominal clitics and the intensifiers share affixal behaviour, there are some properties that differentiate them. While pronominal clitics are related to an argument position, this is not the case for the intensifiers. I have already discussed how auxiliary verbs inherit the arguments of the nonfinite verb through the lexical operation of argument composition. In this way, it is possible to account for the fact that pronominal clitics precede the auxiliary. On the other hand, since the monosyllabic intensifiers do not relate to an argument position, they cannot be inherited by the auxiliary as a result of argument composition. Therefore, they cannot occur before an auxiliary, which is a desirable result, given that most intensifiers are banned from this position. An additional difference between pronominal clitics and the monosyllabic intensifiers is that, as already mentioned, the latter cannot occur after the verb. 21 It should be noted that this negative element is different from nu which is discussed in the next section. 22 This property holds for most intensifiers which can usually modify other categories than the verb.

180

Syntax and its Interfaces

I suggest that monosyllabic intensifiers attach to verbs as a result of a lexical process; it is thus possible to deal with the degree of variation attested. The following rule accounts for the behaviour of mai: (303)

MAI

⎡

Lexical Rule

⎤ word ⎢ ⎥ ⎢PHON 1 ⎥ ⎡ ⎤ ⎢ ⎥ ⎡ ⎡ ⎤⎤⎥ ⎢ 1 ⎢ ⎥ ⎣PHON mai ⊕ ⎦ → ⎢ ⎢CAT ⎣HEAD verb ⎦⎥⎥ 2 ) ⎢ ⎥ SS | L | CONT mai ( ⎢ ⎥ ⎢SS | L⎢ AUX − ⎥⎥ ⎢ ⎥ ⎣ ⎦⎦ ⎣ CONT 2

The lexical rule relates verbs with other ones, which differ in the phonology and in the semantics. In particular, mai is attached before the verb, thus dealing with the general property of intensifiers which are always proclitic. The condition [AUX −] accounts for the fact that mai cannot attach to auxiliary verbs, thus it cannot precede them: (304) *Ion mai i-a scris profesorului. [ro] John again CL.ACC.has written teacher.the ‘John has written to the teacher again.’ On the other hand, this condition is absent in the lexical rule that deals with cam: (305)

CAM

Lexical Rule

⎡

⎤ word ⎢ ⎥ ⎢PHON 1 ⎥ ⎡ ⎤ ⎢ ⎥ ⎡ ⎢ 1

⎤⎥ PHON cam ⊕ ⎢ ⎥ ⎦ CAT HEAD verb ⎥→ ⎣ ⎢ ⎦⎥ ⎢SS | L⎣ SS | L | CONT cam ( 2 ) ⎢ ⎥ CLTS elist ⎢ ⎥ ⎣ ⎦ CONT 2

Recall that this intensifier can occur before the tense auxiliary. The condition CLTS elist, which was absent in the previous lexical rule, allows for cam to combine with verbs only if no pronominal clitics are attached to them. In this

Syntax and its Interfaces

181

way, it is possible to account for the fact that this intensifier is closer to the verb than pronominal clitics.23

4.7. Romanian tense auxiliaries and negation In the previous section, it was shown that the prefix status of the element ne is rather uncontroversial. Recall that ne is used to negate adjectives, participles, and gerunds and it is comparable to the English negative prefix un: (306) Sînt nefericit. [ro] am unhappy ‘I am unhappy.’ In Romanian, however, there are two types of negation: in addition to the prefix ne, there is also the element nu which is used to negate constituents. Nu precedes all the elements in the clitic cluster. It comes before object clitics, tense auxiliaries, and intensifiers, as shown by the example below: (307) Nu l-am

mai v˘azut. [ro] NEG. CL.ACC.have again seen

‘I haven’t seen him again.’ Unlike the monosyllabic intensifiers, it cannot intervene between an auxiliary and the nonfinite verb, as in the following example from Dobrovie-Sorin (1994): (308) *Ar nu vrea s˘a te supere. [ro] would NEG. want that you bother ‘He wouldn’t want to bother you.’ Under the assumption that nu has word status, it is straightforward to deal with the examples above. Evidence in favour of this hypothesis comes from the fact that nu is stressed and can function as a phonological host for pronominal clitics, acting thus as a word: (309) Nu-l NEG.CL.DAT

da. [ro] gives

‘He does not give it.’ 23

A more explanatory analysis of these intensifiers could be in terms of their semantics. However, it is not clear how such an account could explain the difference in position with respect to tense auxiliaries and modal verbs, as well as the variation attested among intensifiers when they combine with auxiliary verbs.

182

Syntax and its Interfaces

Similarly, it can be the host for the reduced forms of the verb a fi ‘be’ (Mallinson 1986): (< nu + e) [ro]

bun. (310) a. Nu-i NEG.be.3s good ‘He isn’t good.’

(< nu + sînt)

bun. b. Nu-s NEG.be.1s good ‘I am not good.’

Furthermore, it can be used in isolation, which is not the case for affixal elements: (311) Radu a venit dar Ion nu. [ro] Radu has come but Ion NEG. ‘Radu has come but Ion hasn’t.’ Therefore, I consider nu a negative adverb which modifies a phrasal head, as can be seen in the entry below: ⎤ ⎡ (312) PHON nu ⎥ ⎢ ⎡ ⎢ ⎡ ⎤⎤⎥ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢CAT | HEAD ⎣adv ⎢ ⎦⎥⎥ ⎢ ⎢ 1 ⎥⎥ ⎢ MOD VP: ⎢ ⎥⎥ ⎢SS | L⎢ ⎡ ⎤ ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎢ not-rel ⎥⎥ ⎢ ⎢ ⎦ ⎦⎥ ⎣CONTENT ⎣ ⎦ ⎣ 1 ARG

Pollard and Sag (1994) propose that adjuncts select the head they modify. To this purpose, they introduce the feature MOD(ified), whose value indicates the syntactic and semantic information of the sign that is modified. In the case of negation, it is a VP which is selected through the MOD feature while its semantic contribution is reflected in the CONTENT attribute. Since nu modifies a VP, it follows correctly that it precedes all the elements of the clitic cluster, accounting thus for example (307), repeated here: (307) Nu l-am

mai vˇazut. [ro] NEG. CL.ACC.have again seen ‘I haven’t seen him again.’

The sentence above is associated with the following structure:

Syntax and its Interfaces

183

(313) VP

H HH H

ADV

nu

VP V

HH H

V [AUX

+ ]

l-am

V

mai v˘azut

The tense auxiliary forms a verbal complex with the nonfinite verb which projects into a VP that is modified by the negative element nu. The ungrammatical sentence in (308), where negation intervenes between the tense auxiliary and the nonfinite verb, is ruled out in a straightforward way: (308) *Ar nu vrea s˘a te supere. [ro] would NEG. want that you bother ‘He wouldn’t want to bother you.’ Given the verbal complex, it follows that negation cannot intervene between the two verbs. This is because negation modifies a VP, but the nonfinite verb is a V that cannot project into a VP because the auxiliary does not subcategorize for a maximal projection. Notice also that if negation intervenes between the lexical verb and a cliticized inverted auxiliary, the result is an ungrammatical sentence: (314) *Mira-nu-m-as¸. [ro] wonder.NEG.CL.ACC.would ‘I wouldn’t wonder it.’ The example above is ruled out for the same reasons that a sentence such as (308) is ruled out: negation modifies a VP, but the combination of the tense auxiliary with a clitic is a V that cannot project, given the verbal complex structure. In Section 4.5.2, I discussed a peculiar property of negation in combination with tense auxiliaries. Recall that nu can be reduced to n, thus obtaining the following combinations: (315) a. N-am trimis scrisoarea. [ro] NEG.have sent letter.the ‘I didn’t send the letter.’

184

Syntax and its Interfaces b. N-ai mînca? NEG.would eat ‘Wouldn’t you eat?’

The process is optional since the full form would be possible as well: (316) a. Nu am trimis scrisoarea. [ro] NEG. have sent letter.the ‘I didn’t send the letter.’ mînca? b. Nu ai NEG. would eat ‘Wouldn’t you eat?’ I would suggest that in (315) it is the clitic form of negation that surfaces: n- is thus a phonological clitic. It represents the reduced form of nu, which surfaces in combination with tense auxiliaries that begin with a vowel. Before concluding this section on negation, I would like to go back to the example (309), since it can be problematic in certain respects: (309) Nu-l NEG.CL.DAT.

da. [ro] gives

‘He does not give it.’ As already mentioned, in this case, negation functions as a phonological host for the object clitics. More generally, in Romanian, pronominal clitics can optionally attach to elements which are different from their morphosyntactic host (i.e. the verb), unlike in other Romance languages. They can combine not only with negation but also with complementizers, nouns, or wh-elements.24 Given the view adopted here, according to which object clitics should not be considered signs, but featural information, it seems possible to provide an analysis for the example above. The Complement Cliticization Lexical Rule in (271) accounts for the fact that even though the clitic attaches phonologically to a host different from the verb, it still satisﬁes its subcategorization requirements. The information about those complements which should be realized as clitics is thus encoded in the CLTS list. A speciﬁc principle can be formulated to account for the percolation of the relevant information on the appropriate 24 This fact might be used as evidence against the affixal view of pronominal clitics defended in the previous chapter. According to one of the tests proposed by Zwicky and Pullum (1983), affixes should be selective with respect to the host they attach to. However, as discussed in Miller (1992a), the relevance of this test is questionable under the view that elements which show ‘promiscuous attachment’ such as possessive ’s in English should be considered (phrasal) affixes (cf. Zwicky 1987). Similar evidence is shown by the affix re in French discussed in the previous section.

Syntax and its Interfaces

185

host. The morphophonological constraints spell out the correct clitic form. It should be noted that in the example above, it is the enclitic form of the clitic that surfaces. Therefore, it cannot be analysed in terms of phonological restructuring since, in that case, the proclitic form would be expected. I refer to Monachesi (1998b) for a more detailed discussion of these data. The optional enclitic placement of the Romanian object clitics to the element preceding the verb seems a remnant of the Tobler-Mussafia Law (Wanner 1987), according to which clitics could not appear in first position. Before concluding this section, I would like to mention another interesting property of negation which Romanian shares with the Balkan languages. As discussed in Lombard (1934), a word which is introduced by negation is stressed on the first syllable even if without negation it would not have been stressed. On the other hand, a secondary stress falls on the syllable which would have carried primary stress if negation were not present. More generally, the nform makes stress fall on the tense auxiliary, which would be stressless when it combines with it: (317) N-á v˘azùt. [ro] NEG.has seen ‘He hasn’t seen.’ A similar behaviour can be seen in Bulgarian, in which the negative marker (which is a clitic) gives stress to the following lexical item. As mentioned in Franks and King (2000), if the following lexical item is a finite verb there is no change in stress since the verb has lexical stress. If the following item is a clitic auxiliary (or a clitic pronominal), that clitic will carry stress. The situation is slightly different in Macedonian where, unlike in Bulgarian, ne itself can be stressed.

4.8. Future auxiliaries In Romanian, different forms are employed to convey the idea of future. The difference among the various forms is related to frequency and register. In Section 4.2, I have already introduced a form for the future which is composed of an auxiliary together with the bare infinitive. This construction is not very frequent and it is used mainly in written texts: I refer to it as the literary future. There are two other possible ways to express the future in Romanian, both being normally used in the spoken language. One form employs the present of the verb a avea ‘to have’ together with the subjunctive while the other makes use of the invariable auxiliary o with the subjunctive. I refer to these two constructions as the colloquial future and they are the subject of this section. I mainly address the one formed by the verb a avea with the subjunctive, but the

186

Syntax and its Interfaces

conclusions reached for this construction can be extended to the other form introduced by the auxiliary o. The three different forms are summarized in the table below: (318) Different forms of the Romanian future literary fut. colloquial fut. colloquial fut.

1 Sg voi am

2 Sg vei ai

3 Sg va are

1 Pl vom avem o

2 Pl vet¸i avet¸i

3 Pl vor au

complement bare infinitive subjunctive subjunctive

In the colloquial future the subjunctive is employed. It should be noted that in Romanian, the subjunctive is formed by the conjunction s˘a together with the present indicative of the verb, except in the third person where a different form occurs. Dobrovie-Sorin (1994) claims that there are crucial similarities between the colloquial future forms and the tense auxiliary constructions discussed in the previous sections. In particular, she notes that, as in the case of the tense auxiliaries (196), adverbs cannot intervene between the main verb and the subjunctive. This is shown in (319a), where adesea ‘often’ cannot occur between the two verbs. As illustrated by (319b), the adverb follows the verbal complex: (319) a. *Ion are adesea s˘a cînte la pian. [ro] John has often CONJ. play to piano ‘John will often play the piano.’ cînte adesea la pian. b. Ion are s˘a John has CONJ. play often to piano ‘John will often play the piano.’ Similarly for quantifiers: they cannot occur between the two verbs (cf. (203) for the equivalent cases with tense auxiliaries). Example (320a) shows that if tot¸i ‘all’ occurs between the auxiliary au and the subjunctive, the result is an ungrammatical sentence. The quantifier should follow the verbal complex, as illustrated in (320b): (320) a. *Copiii au tot¸i s˘a cînte la pian. [ro] children.the have all CONJ. play to piano ‘The children will all play the piano.’ b. Copiii au s˘a cînte tot¸i la pian. children.the have CONJ. play all to piano ‘The children will all play the piano.’

Syntax and its Interfaces

187

Recall that subjects cannot intervene between the tense auxiliary and the lexical verb (210). This is also the case for the colloquial future since subjects cannot precede the subjunctive, as can be seen in (321a). The preferred order is the one in which the subject Ion follows the verbal complex, as shown in (321b): cînte la pian? [ro] (321) a. *Are Ion s˘a has John CONJ. play to piano ‘Will John play the piano?’ b. Are s˘a cînte Ion la pian? has CONJ. play John to piano ‘Will John play the piano?’ These common properties have lead Dobrovie-Sorin (1994) to argue in favour of a unitary analysis for tense auxiliaries and the colloquial future. However, Dobrovie-Sorin does not mention that the colloquial future and the tense auxiliaries also exhibit a crucial difference, which is related to the distribution of pronominal clitics. Contrary to tense auxiliaries, the colloquial future does not trigger clitic climbing, as illustrated by the example in (322a), while (322b) shows that the object clitic se must attach to the embedded verb plimbe: (322) a. *S-are s˘a plimbe. [ro] CL.REFL.has CONJ. goes for a walk ‘He will go for a walk.’ plimbe. b. Are s˘a se has CONJ. CL.REFL goes for a walk ‘He will go for a walk.’ Romanian does not exhibit many environments in which clitic climbing is allowed: this might be due to the fact that the use of bare inﬁnitives is disappearing in this language and being replaced by the subjunctive. Additional differences can be seen in the fact that the forms of the auxiliary in the colloquial future are different from those which occur in the present perfect or in the conditional. As already mentioned, in the colloquial future the paradigm is constructed with the conjugated form of the verb a avea. On the other hand, in the present perfect and in the conditional, the auxiliaries no longer bear any relation to a avea even though they are historically related to it. The table below compares the relevant forms:

188

Syntax and its Interfaces

(323) Different forms of the auxiliaries 1 Sg am as¸ am

Aux Psp. Aux Cond. Aux Coll. Fut.

2 Sg ai ai ai

3 Sg a ar are

1Pl am am avem

2 Pl at¸i at¸i avet¸i

3 Pl au ar au

past participle bare infinitive subjunctive

To summarize: these data show that tense auxiliaries and the colloquial future share the same adjacency requirements while they differ with respect to the distribution of pronominal clitics. The best way to deal with these properties is to assume that the similarities between the two constructions (i.e. the impossibility of intervening elements) can be explained in terms of the same syntactic structure. It is suggested that both in the case of tense auxiliaries and of the colloquial future, the two relevant verbs form a verbal complex: (324) Verbal complex VP

H HH

V

HH

V [AUX

+ ]

COMPS

V

In this way, it is possible to deal with the fact that adverbs, quantifiers, and subjects cannot occur between the colloquial future and the subjunctive. The result is a uniform analysis of tense auxiliaries and of the colloquial future with respect to the common adjacency requirements. These data thus seem to argue in favour of a syntactic treatment of adjacency, if one were to assume a phonological treatment it would not be straightforward to extend it to deal with the future auxiliaries. On the other hand, the lack of clitic climbing in the presence of the colloquial future provides evidence that the lexical mechanism of argument composition is not triggered in this case. Recall that argument composition is the explanatory device that accounts for clitic climbing since it makes the complements of the embedded verb become arguments of the main verb. If argument composition is not triggered, clitic climbing cannot occur, which is the desired result. As a consequence of this analysis, different lexical entries are assumed for the tense auxiliaries and for those of the colloquial future. In particular, a future auxiliary such as are is associated with the following description:

(325)

⎡

Syntax and its Interfaces ⎤

189

are ⎢ ⎡ ⎤⎥ ⎢ ⎥ ⎢ ⎥ verb ⎢ ⎥ ⎢ ⎥ HEAD ⎢ ⎢ ⎥⎥ ⎢ ⎥ AUX + ⎢ ⎥ ⎢ ⎢ ⎡ ⎤⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎢ ⎥⎥ SUBJ 1 NP ⎢ ⎥ ⎢ ⎢ ⎥ ⎢ ⎤ ⎥⎥ ⎡ ⎢ ⎢ ⎥⎥ ⎢ ⎢ ⎢ ⎥⎥⎥ ⎢ w-ss ⎢ ⎢ ⎥⎥ ⎢ ⎥ ⎥ ⎥ ⎢ ⎢ ⎢ ⎥ ⎢ ⎥ ⎥ ⎥⎥ ⎢ ⎥ ⎢ verb ⎢SS | L | C⎢ ⎥⎥⎥⎥ ⎢HEAD ⎢ ⎢ ⎢ ⎥ ⎥⎥⎥ ⎢ ⎢VAL ⎢ ⎢ VFORM subjunct ⎥ ⎥⎥⎥ ⎢ ⎢COMPS 2 ⎢ ⎢ ⎢ ⎡ ⎢ ⎢ ⎥⎥ ⎢ ⎤ ⎥ ⎥ ⎥ ⎢ ⎢ ⎢ ⎥⎥ ⎥ ⎢ ⎥ ⎥ ⎢ 1 NP ⎢ ⎢ ⎥ SUBJ ⎢ ⎥ ⎥ ⎥⎥ ⎢ ⎣ ⎢ ⎢ ⎥ ⎦ ⎢ VAL ⎦ ⎦⎥⎥ ⎣ ⎢ ⎣ ⎢ ⎢ ⎥⎥ ⎢ COMPS 3 ⎢ ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎣ ⎦⎥ ⎣ ⎦ ARG-ST 1 , 2 PHON

The auxiliary subcategorizes for a subjunctive verbal complement, while the lexical mechanism of argument composition is not triggered in this case. It should be noted that, the ARG-ST list does not include the arguments of the embedded verb, indicated by the tag 3 . The underlying assumption behind this analysis is that the notion of argument composition is orthogonal to that of syntactic structure. In fact, argument sharing is not sufﬁcient to determine the syntactic structure since it is compatible both with the verbal complex structure and with a flat one, as will be shown in Section 4.10.2. This is a conclusion also reached by Abeill´e and Godard (2002), who show that in French argument composition is compatible with a flat structure in the case of tense auxiliaries and a VP complement for passive auxiliaries. The position of negation plays a crucial role in assessing the analysis proposed. In the case of the colloquial future, negation precedes the auxiliary: (326) Nu am s˘a plec. [ro] NEG. have CONJ. leave ‘I will not leave.’ The following sentences in which negation occurs between the two verbs (327a) or before the subjunctive (327b) are ungrammatical: (327) a. *Am nu s˘a plec. [ro] have NEG. CONJ. leave ‘I will not leave.’

190

Syntax and its Interfaces b. *Am s˘a nu plec. have CONJ. NEG. leave ‘I will not leave.’

It therefore seems that with respect to negation, the colloquial future behaves like tense auxiliaries in that negation should precede the auxiliary (328a) but it cannot intervene between the two verbs (328b): (328) a. N-am v˘azut. [ro] NEG.have seen ‘I haven’t seen.’ b. *Am nu v˘azut. have NEG. seen ‘I haven’t seen.’ Given these data, the crucial question to be answered is why it is not possible to negate the complement of the colloquial future and why negation must precede the auxiliary. Dobrovie-Sorin (1994) suggests that the answer must be found in the nature of the auxiliary which is not allowed to subcategorize for a negated complement. Dobrovie-Sorin has to stipulate that the colloquial future auxiliaries cannot subcategorize for a negated complement in order to deal with the position of negation. On the contrary, in the analysis I have proposed, the position of negation follows as a consequence of the syntactic structure assumed. Given the verbal complex, it is straightforward to deal with the fact that negation cannot intervene between the auxiliary and the subjunctive. Since negation modiﬁes a VP, but the embedded verb is a V that cannot project, it follows that negation cannot intervene between the two verbs, as in the case of the tense auxiliaries. Thus, the only possible position is the one in which negation precedes the verbal complex.25 The position of the intensiﬁers in the case of the colloquial future provides additional evidence for the analysis proposed. Intensiﬁers can only occur attached to the embedded verb, as illustrated by mai in the sentence below:

25 It could be argued that the fact that negation cannot occur between the two verbs might have to do with the semantic scope of negation. The two verbs form a complex predicate and we do not have two separate clauses. Abeille´ and Godard (2003) consider the impossibility of negating the second predicate as a property of complex predicates, on a par with clitic climbing. However, the fact that clitic climbing is triggered only by tense auxiliaries and not by the colloquial future seems to argue against this possibility. Furthermore, it constitutes a counterexample to the generalization that the lack of scope on the nonfinite verb in the case of negation and clitic climbing go together in determining monoclausality.

Syntax and its Interfaces

191

(329) Am s˘a mai plec. [ro] have CONJ. INT. leave ‘I will not leave.’ Intensifiers cannot precede the future auxiliary (330a) and they cannot intervene between the two verbs (330b): (330) a. *Mai am s˘a plec. [ro] INT. have CONJ. leave ‘I will not leave.’ b. *Am mai s˘a plec. have INT. CONJ. leave ‘I will not leave.’ This distribution is the expected one given the lexical rules assumed for intensiﬁers. Recall that intensiﬁers such as mai are accounted for by means of lexical rules like the following: (331) MAI Lexical Rule ⎡ ⎤ word ⎢ ⎥ ⎢PHON 1 ⎥ ⎡ ⎤ ⎢ ⎥ ⎡ ⎡ ⎤⎤⎥ ⎢ 1 PHON mai ⊕ ⎢ ⎥ ⎣ ⎦ → ⎢ ⎢CAT ⎣HEAD verb ⎦⎥⎥ 2 ) ⎢ ⎥ SS | L | CONT mai ( ⎢ ⎥ ⎢SS | L⎢ AUX − ⎥⎥ ⎢ ⎥ ⎣ ⎦⎦ ⎣ CONT 2

The lexical rule relates verbs with other ones that differ in the phonology and in the semantics. In particular, mai is attached before the verb, thus dealing with (329) where the intensifier is proclitic to the lexical verb. The condition [AUX −] accounts for the fact that mai cannot attach to auxiliary verbs, thus it cannot precede them, correctly accounting for the ungrammaticality of example (330a). To conclude this section: Dobrovie-Sorin (1994) has argued in favour of a uniform analysis for tense auxiliaries and for the colloquial future based on the fact that they share certain common properties. In particular, subjects, adverbs, and quantifiers cannot occur between the auxiliary and the lexical verb. She has suggested that both the tense auxiliary and the colloquial future must share the same structure: a biclausal one. Dobrovie-Sorin has also argued that it can capture the common properties of these constructions and has the advantage of providing a unitary analysis. However, this account is not able to

192

Syntax and its Interfaces

deal with a crucial difference, that is the fact that clitic climbing is triggered by tense auxiliaries, but is not allowed by the colloquial future. In the analysis I have proposed, a clear distinction is drawn between the adjacency requirements that are shared by both the tense auxiliaries and the colloquial future on the one hand and the presence or absence of clitic climbing on the other hand. I have suggested that the impossibility of certain elements occurring between the two relevant verbs is accounted for by the same syntactic structure in the case of tense auxiliaries and of the colloquial future. The lack of clitic climbing with the colloquial future and its presence with tense auxiliaries is a consequence of the lexical mechanism of argument composition. It is triggered in the latter case, but not in the former. Thus, the similarities and the differences between the colloquial future and the tense auxiliaries are accounted for by an appropriate division of labour between syntactic structure and argument structure: the former is in charge of the adjacency conditions while the latter deals with clitic climbing.

4.9. The perfect auxiliary fi The discussion of Romanian auxiliaries should also include the element fi which is addressed in this section. The auxiliary fi is an invariable marker of perfect aspect which represents the bare infinitive of the verb a fi ‘to be’. The following examples illustrate its use: (332) a. Va fi terminat lect¸ia. [ro] will be finished lesson.the ‘She will have finished the lesson.’ b. Nu cred s˘a fi spus Ion as¸a ceva. NEG. believe CONJ. be said John such thing ‘I don’t believe John to have said such a thing.’ Example (332a) shows a case of anterior future, which is composed by the auxiliary together with the marker fi and the past participle. Example (332b) represents an example of past subjunctive, which comprises the element s˘a together with the marker fi and the past participle. Nothing can separate the marker from the past participle; even object clitics and intensifiers must precede this element: (333) S˘a o mai fi v˘azut. [ro] that CL.ACC INT. BE seen ‘That I should have seen her again.’ The marker fi is the most internal element of the clitic cluster. Dobrovie-Sorin (1994) assumes that it adjoins to INFL after the intensifiers and that it forms a

Syntax and its Interfaces

193

discontinuous morpheme with the past participle. In particular, she comes to the conclusion that fi is not an auxiliary but simply a bound morpheme. A possibility would be to follow Dobrovie-Sorin in her analysis and to suggest that fi is a prefix which attaches to the past participle. Given that it is a marker of perfect aspect, it can be proposed that it represents the spelling out of a feature [ASP perf]. Assuming an approach to morphology such as the one presented in the previous chapter for the spelling out of object clitics, a specific implicational constraint can be formulated which is responsible for its phonological realization: ⎤ ⎡ (334) word ⎢ ⎤⎥ ⎡ ⎥ ⎢ complex-morph ⎥ ⎢ ⎢ ⎤⎥⎥ ⎡ ⎢ ⎢ ⎥⎥ ⎢ word ⎢ ⎥⎥ ⎢ ⎡ ⎤ ⎥ ⎢ ⎥ ⎢ ⎡ ⎤ ⎢ ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ verb ⎥ ⎢ ⎥→ ⎢MORPH ⎢ ⎢ ⎥⎥ ⎢ ⎥⎥ ⎢ ⎥⎥ ⎢STEM ⎢ ⎢ ⎥ VFORM psp HEAD ⎥ ⎢ ⎥ ⎢ ⎣ ⎦ ⎢ ⎥⎥⎥ ⎢SS | L | C⎢ ⎢ ⎥⎥ ⎢ ⎢ ⎥ ⎥ ⎢ ⎥ ⎢ ⎢ ASP perf ⎣ ⎦⎦⎥ ⎢ ⎣ ⎦⎥ ⎣ ⎦ ⎣ CLTS elist ⎡ ⎤ ⎤ ⎡ complex-morph ⎢ ⎤⎥ ⎥ ⎡ ⎢ ⎢ ⎥ ⎥ prefix ⎥ ⎢MORPH ⎢ ⎢ ⎥⎥

⎢ ⎢ ⎥ ⎥ ⎢ ⎢ AFFIX ⎦ ⎣ ⎦ ⎥ ⎣ ⎥ ⎢ PHON SKEL fi ⎥ ⎢ ⎢ ⎤⎥ ⎡ ⎥ ⎢ ⎥ ⎢ verb ⎢

⎦⎥ ⎦ ⎣SS | L | C | HEAD ⎣ VFORM comp psp The constraint states that if there is a past participle with the feature ASP equal to perf, this is realized as the prefix fi and the VFORM of the resulting word is of type composed past participle. The condition CLTS elist ensures that object clitics and intensifiers occur before fi. The position of the clitic o in the presence of these past forms is quite surprising. The clitic can attach to the past participle (335a) or it can precede the composed form (335b): (335) a. Nu cred s˘a fi spus-o Ion. [ro] NEG. believe CONJ. be say.CL.ACC. John ‘I don’t believe John to have said it.’ b. Nu cred s-o fi spus Ion. NEG. believe CONJ.CL.ACC. be say John ‘I don’t believe John to have said it.’

194

Syntax and its Interfaces

The position in (335a) is the expected one for the clitic o: as already seen in the case of the tense auxiliary constructions, this clitic usually attaches to the past participle. Examples such as (335a) are thus licensed by the constraint (282), repeated below, together with the one responsible for enclisis: ⎤

⎡

word ⎢ (282) ⎣

MORPH | ST | SS | L | C | CLTS

NP

⎥ ⎦ acc [3fsg]

→

⎤⎤ complex-morph ⎢ ⎤⎥⎥ ⎡ ⎢ ⎢ ⎥⎥ ⎢ word ⎢ ⎥⎥ ⎢ ⎡ ⎤ ⎢ ⎥ ⎢ ⎥⎥ ⎢ ⎢ ⎥ ⎢ 2 1 ⎥⎥ ⎢ VOW ⊕ ⎢ ⎥ ⎢ ⎢ ⎢ ⎥⎥⎥⎥ ⎥ ⎢ ⎢ ⎢ ⎦⎥⎥ ⎢ ⎢PHON ⎣ ⎥⎥ ⎢ 3 1 ⎥ ⎢ ⎥ ⎢ SKEL ⊕ ⎢STEM ¬⎢ ⎢ ⎥⎥ ⎥⎥ ⎢MORPH⎢ ⎡ ⎤ ⎥ ⎢ ⎢ ⎥⎥⎥ ⎥ ⎢ ⎢ ⎢ ⎢ ⎥⎥ ⎢ verb ⎥⎥ ⎢ ⎢ ⎣ ⎦ ⎦ ⎣ SS | L | C HEAD ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ AUX + ⎢ ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎤ ⎡ ⎢ ⎥⎥ ⎢ ⎢ affix ⎥⎥ ⎢ ⎢ ⎦ ⎦⎥ ⎣AFFIX ⎣ ⎦ ⎣ PHON | SKEL o ⎡

⎡

The order in (335b) is the most peculiar one, even though pronominal clitics also occur in this position: fi spus Ion. [ro] (336) Nu cred s˘a-l NEG. believe CONJ.CL.ACC. be say John ‘I don’t believe John to have said it.’ These cases can be dealt with if it is assumed that the composed past participle triggers proclisis. The constraint which accounts for proclisis should therefore be modified to include this form among the ones that allow for preverbal clitics. The distribution of the clitic o might suggest another alternative analysis in which fi has the same status as tense auxiliaries. Also in this case, it can be assumed that the perfect auxiliary and the past participle form a verbal complex: it is thus possible to capture the intuition that the two verbs form a unit which cannot be separated by other elements. Under this view, the position of the clitic o could be easily explained as a case of optional clitic climbing. The following lexical entry should be associated with the aspectual marker:

Syntax and its Interfaces (337)

⎤

⎡

word ⎢ ⎢PHON fi ⎢ ⎡ ⎢ ⎡ ⎤ ⎢ verb ⎢ ⎢ ⎢ ⎥ ⎢HEAD ⎢ ⎢ ⎣AUX + ⎦ ⎢ ⎢ ⎢ ⎢ VFORM comp psp ⎢ ⎢ ⎢ ⎡ ⎢ ⎢ ⎢ ⎢ 1 NP ⎢ SUBJ ⎢ ⎢ ⎢ ⎢ ⎢ ⎤ ⎡ ⎢ ⎢ ⎢ w-ss ⎢ ⎢ ⎢ ⎢ ⎢ ⎥ ⎢ ⎢ ⎢ ⎢SS | L | C⎢ ⎢ ⎥ ⎢ verb ⎢ ⎢ ⎢ ⎥ ⎢ HEAD ⎢VAL⎢ ⎢ ⎥ ⎢ ⎢ ⎢ VFORM psp ⎢ ⎥ ⎢ ⎢ ⎢COMPS 2 ⎢ ⎡ ⎢ ⎥ ⊕ ⎤ ⎢ ⎢ ⎢ ⎥ ⎢ ⎢ ⎢ ⎢ 1 NP ⎥ ⎢ SUBJ ⎢ ⎢ ⎢ ⎥ ⎢ ⎢ ⎢ ⎢ ⎢ ⎦⎥ ⎦ ⎣VAL⎣ ⎢ ⎣ ⎢ ⎢ COMPS 3 ⎢ ⎢ ⎢ ⎢ ⎢ ⎣ ⎣ ARG-ST 1 , 2 ⊕ 3

195

3

⎥ ⎥ ⎥ ⎤⎥ ⎥ ⎥ ⎥⎥ ⎥⎥ ⎥⎥ ⎥⎥ ⎥⎥ ⎥ ⎤⎥ ⎥⎥ ⎥⎥ ⎥ ⎥⎥ ⎥ ⎥⎥ ⎥ ⎥⎥ ⎥ ⎥⎥ ⎥ ⎥⎥ ⎥ ⎥⎥ ⎥ ⎥⎥ ⎥ ⎥⎥ ⎥ ⎥⎥ ⎥ ⎥⎥ ⎥ ⎥⎥ ⎥ ⎥⎥ ⎥ ⎥⎥ ⎥ ⎥⎥ ⎥ ⎦⎥ ⎥⎥ ⎥⎥ ⎥⎥ ⎦⎦

The perfect auxiliary subcategorizes for the past participle and its complements through the operation of argument composition, which identiﬁes the complements of the auxiliary verb with those of the embedded verb. This is indicated by the presence of the tag 3 both in the COMPS list of the auxiliary verb and in that of the embedded verb, as well as by the fact that these complements are included in the ARG-ST list of the auxiliary verb. Example (335b), repeated below, is thus assimilated to a case of clitic climbing: (335b) Nu cred s-o fi spus Ion. [ro] NEG. believe CONJ.CL.ACC. be say John ‘I don’t believe John to have said it.’ The auxiliary fi subcategorizes for the verbal complement and its complements which in this case are represented by the direct object which is realized as the clitic o. On the other hand, example (335a) is considered a case in which the clitic does not climb and accounted for by means of the lexical analysis of cliticization presented in the previous chapter. In particular, the Complement Cliticization Lexical Rule, together with the relevant realizational constraints, is responsible for the spelling out of the cliticized verb form spuso:

196

Syntax and its Interfaces

(335a) Nu cred s˘a fi spus-o Ion. [ro] NEG. believe CONJ. be say.CL.ACC. John ‘I don’t believe John to have said it.’ The position of the clitic o in the presence of the perfect auxiliary fi is thus similar to that attested in the case of the literary future in which the clitic occurs either after the infinitive or before the auxiliary, as illustrated by example (278) repated below: (278) a. Voi vedea-o. [ro] will see.CL.ACC ‘I will see her.’ b. O

voi vedea. see

CL.ACC will

I have mentioned that intensifiers generally precede the aspectual marker, but in spoken language as well as in the literature of the beginning of the century, they can intervene between fi and the past participle: (338) a. ??Eu as¸ fi mai stat. [ro] I would have INT. stayed ‘I would have stayed more.’ b. Eu as¸ fi tot stat. I would have INT. stayed ‘I would have still stayed.’ This position is similar to that found in the case of tense auxiliaries and it would receive the same explanation under the alternative analysis proposed that assimilates the aspectual marker fi to tense auxiliaries. In particular, the lexical rules proposed in Section 4.6 would account for the position of the intensifier attached to the lexical verb. The data presented in this section with respect to the aspectual marker fi might suggest that this element originally had the same status as tense auxiliaries, but it is now in the process of becoming a preﬁx. The position of the clitic o and that of the intensiﬁers seem to point towards this explanation. In synchronic grammar this might mean that the two analyses proposed should not be considered as possible alternatives, but they might both be available for Romanian speakers given that this element is undergoing a process of historical change.

Syntax and its Interfaces

197

4.10. Auxiliary-like constructions The data discussed in this chapter suggest that Romanian exhibits a rich typology of auxiliary constructions that are characterized by the presence (or absence) of clitic climbing and by the possibility (or impossibility) of certain elements intervening between the auxiliary and the nonﬁnite verb. However, these properties are also attested to various degrees by other verbal classes that are usually not considered auxiliary verbs. This is the case of raising verbs and modal verbs, which are discussed in the following sections. It is interesting to look at them in detail because of the variation they exhibit, suggesting the existence of a more complex typology. 4.10.1. Raising verbs

Romanian raising verbs such as a urmea ‘going to’ are characterized by the fact that they subcategorize for a verbal complement headed by a subjunctive instead of an infinitive, as in the following example from Dobrovie-Sorin (1994): (339) Ion urmeaz˘a s˘a cînte la pian. [ro] John follows CONJ. play to piano ‘John is going to play the piano.’ Romanian differs in this from the other Romance languages: the use of bare inﬁnitives is disappearing from the language, usually being replaced by subjunctives.26 Recall that in Romanian, the subjunctive is formed by the conjunction s˘a together with the present indicative of the verb, except in the third person where a different form occurs. The status of this conjunction is rather controversial, and I try to clarify it in the following section. Raising verbs share many of the properties already discussed in the case of tense auxiliaries. In particular, as in the case of tense auxiliaries (196), adverbs cannot occur between the relevant verbs, that is the raising verb and the embedded one (340a) (Dobrovie-Sorin 1994): (340) a. *Ion urmeaz˘a adesea s˘a cînte la pian. [ro] John follows often CONJ. play to piano ‘John is going to often play the piano.’ b. Ion urmeaz˘a s˘a cînte adesea la pian. John follows CONJ. play often to piano ‘John is going to often play the piano.’ 26 I refer to Quer (1995) for an overview of the use and the properties of the subjunctive in Romance and Balkan languages.

198

Syntax and its Interfaces

In example (340a), the adverb adesea ‘often’ occurs between the two verbs, resulting in an ungrammatical sentence. The adverb should follow the subjunctive as illustrated by the grammatical example in (340b). Similarly for what concerns quantifiers; they cannot occur between the two verbs (341a), but must follow the subjunctive (341b) (cf. (203) for the equivalent examples with tense auxiliaries) (Dobrovie-Sorin 1994): (341) a. *Copiii urmeaz˘a tot¸i s˘a cînte la pian. [ro] children.the follow all CONJ. play to piano ‘The children are all going to play the piano.’ b. Copiii urmeaz˘a s˘a cînte tot¸i la pian. children.the follow CONJ. play all to piano ‘The children are all going to play the piano.’ Raising constructions differ from both tense auxiliaries (210) and the colloquial future (321) in that subjects can precede the subjunctive (Dobrovie-Sorin 1994): (342) Crezi c˘a urmeaz˘a s¸i Ion s˘a cînte? [ro] believe that follows also John CONJ. play ‘Do you think that also John is going to play?’ In the example above, the subject Ion is allowed to surface between the two verbs. On the other hand, they pattern like the colloquial future with respect to the distribution of the clitics. Also in this case, clitic climbing is not triggered: (343) a. *Copiii l-urmau [ro] s˘a conduc la gar˘a. children.the CL.ACC.believed CONJ. take to station.the ‘The children were going to take him to the station.’ b. Copiii urmau s˘a-l conduc la gar˘a. children.the believed CONJ.CL.ACC take to station.the ‘The children were going to take him to the station.’ The example in (343a) illustrates the fact that the pronominal clitic cannot occur attached to the raising verb, since the result is an ungrammatical sentence. Instead it must attach to the verb in the subjunctive, as shown in (343b). I have already mentioned that Dobrovie-Sorin (1994) focuses on the common properties of tense auxiliaries and the colloquial future in order to motivate a unitary analysis of the two constructions. On the other hand, she ignores them in the case of raising constructions. The fact that adverbs and quantiﬁers cannot intervene between the relevant verbs receives a common explanation in the case of tense auxiliaries and the colloquial future, but a different one in the case of raising constructions. Furthermore, the impossibility of clitic

Syntax and its Interfaces

199

climbing with the colloquial future and with raising constructions is entirely ignored in her analysis. I suggest that the impossibility of clitic climbing in raising constructions receives the same explanation as the colloquial future. In both conﬁgurations, the operation of argument composition is not triggered. On the other hand, raising verbs differ slightly from the auxiliaries already discussed with respect to which elements can intervene between the relevant verbs: adverbs and quantiﬁers are not allowed, while subjects are. I suggest that these adjacency requirements should be dealt with by means of a hierarchical structure which allows for a subject to intervene between the raising verb and the subjunctive:27 (344) Hierarchical structure VP

H HH

V [AUX

−]

S

HH

NP

VP

H H

V

COMPS

The raising verb subcategorizes for a sentential complement; as will be discussed later, this allows for the subject to intervene between the raising verb and the subjunctive. A verb such as a urmea is associated with the following lexical entry: ⎤ ⎡ (345) a PHON urmeaz˘ ⎢ ⎤⎥ ⎡ ⎥ ⎢ ⎥ ⎢ verb ⎢ ⎥⎥ ⎢HEAD ⎢ ⎥⎥ ⎢ ⎥ ⎢ ⎢ ⎡ AUX ⎢ ⎤⎥ ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎢SUBJ 1 NP ⎢ ⎥ ⎥⎥ ⎢ ⎢ ⎢ ⎡ ⎤ ⎥ ⎥⎥ ⎢ ⎢ ⎥ ⎢ ⎥ ⎢ ⎢ p-ss ⎢ ⎥⎥ ⎥⎥ ⎢ ⎢ ⎢ ⎥ ⎢ ⎥ ⎥⎥ ⎢ ⎢ ⎢ ⎢ ⎥⎥ verb ⎥⎥ ⎢SS | L | C⎢ ⎢ ⎥ ⎢ ⎥ ⎥⎥ ⎢VAL ⎢ HEAD ⎢ ⎢ ⎥ ⎥ ⎥⎥ ⎢ ⎢ ⎥ ⎢ VFORM subjunct ⎥ 2 ⎢ ⎥ ⎢ ⎢COMPS ⎢ ⎥⎥ ⎢ ⎡ ⎥⎥ ⎢ ⎢ ⎤ ⎥ ⎢ ⎥ ⎢ ⎥ ⎥⎥ ⎢ ⎢ ⎢ ⎢ ⎥ ⎥ SUBJ 1 NP ⎥⎥ ⎢ ⎢ ⎥ ⎢ ⎥ ⎣ ⎦ ⎣ ⎦ ⎢ ⎣ VAL ⎢ ⎦⎥ ⎥⎥ ⎢ ⎢ COMPS elist ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎢ ⎦⎥ ⎣ ⎦ ⎣ 1 2 ARG-ST , 27 Since clitic climbing is not triggered this construction does not represent an instance of complex predicate formation.

200

Syntax and its Interfaces

The auxiliary subcategorizes for a sentential complement. The lexical mechanism of argument composition is not triggered in this case, as can be seen from the absence of the arguments of the embedded verb in the ARG-ST list of the raising verb. Given these assumptions, it is a straightforward matter to deal with the fact that both adverbs and quantifiers cannot intervene between the two relevant verbs. Since argument composition is not triggered, they must remain within the lower clause while an appropriate linearization principle orders them in their appropriate position. Subjects, however, are allowed to occur before the subjunctive as shown in (342), repeated below: (342) Crezi c˘a urmeaz˘a s¸i Ion s˘a cînte? [ro] believe that follows also John CONJ. play ‘Do you think that also John is going to play?’ Given the syntactic structure assumed, however, it is possible to deal with the fact that the subject can intervene between the raising verb and the subjunctive in (342). Another obvious consequence of the fact that argument composition is not triggered is the impossibility of clitic climbing with raising verbs. The lack of this property receives the same explanation in the various cases, as already mentioned. As for intensifiers, they can attach to the raising verb, as shown by example (346a) in which mai combines with the raising verb urmeaz˘a. They can also attach to the embedded verb as can be seen in (346c) where mai combines with the bare infinitive cînte. However, they cannot occur between the two verbs, as illustrated in (346b): (346) a. Ion mai urmeaz˘a John INT. follow

cînte la pian. [ro] CONJ. play to piano s˘a

‘John is going to play the piano again.’ b. *Ion urmeaz˘a John follows

mai

s˘a

INT.

CONJ. play

cînte la pian. to piano

‘John is going to play the piano again.’ c. Ion urmeaz˘a s˘a mai cînte la pian. John follow CONJ. INT. play to piano ‘John is going to play the piano again.’ This distribution is the expected one given the analysis of intensifiers according to which they can only attach to a V. The lexical rule proposed for mai in

Syntax and its Interfaces

201

Section 4.6 accounts for examples (346a) and (346c) in which the intensifier combines with the lexical verb. Similarly, example (346b) is ruled out since mai precedes a sentential complement. The position of negation is quite interesting in the case of raising verbs, since it can occur before the raising verb, as illustrated by (347a), as well as before the embedded verb, as shown by (347c). However, it cannot surface between the two verbs, as can be seen in (347b) which represents an ungrammatical example: (347) a. Ion nu urmeaz˘a s˘a cînte la pian. [ro] John NEG. follow CONJ. play to piano ‘John is not going to play the piano.’ b. *Ion urmeaz˘a nu s˘a cînte la pian. NEG. CONJ. play to piano John follows ‘John is not going to play the piano.’ c. ?Ion urmeaz˘a s˘a nu cînte la pian. CONJ. NEG. play to piano John follow ‘John is not going to play the piano.’ These data can be accounted for on the basis of the syntactic structure assumed. If raising verbs subcategorize for a sentential complement, it follows that negation cannot intervene between the two verbs since it modiﬁes a VP. An example such as (347b) is therefore ruled out. On the other hand, a sentence such as (347c) is accounted for because negation modiﬁes a VP which is represented by the sequence cînte la pian. Similarly for example (347a) which receives the same analysis: negation modiﬁes the VP represented by the raising verb with its complement. The position of negation might suggest that the element s˘a has the status of a complementizer. As can be seen in the examples below, negation follows the element c˘a which is uncontroversially a complementizer in Romanian: (348) S¸tiu c˘a n-ai timp. [ro] know COMP. NEG.have time ‘I know that you haven’t time.’ Given this distribution of negation, it is quite peculiar that in the case of the colloquial future, the element nu cannot occur after s˘a, as shown by (327b) repeated below: (327b) *Am s˘a nu plec. [ro] have CONJ. NEG. leave ‘I will not leave.’

202

Syntax and its Interfaces

In this case, the distribution is similar to that attested with tense auxiliaries in which negation must precede the verbal complex: (328a) N-am v˘azut. [ro] NEG.have seen ‘I haven’t seen.’ It therefore seems that the position of negation cannot be assumed as conclusive evidence to assess the status of the element s˘a, which is rather controversial. The controversy is based on whether it should be considered an independent word, that is a conjunction or a particle or simply a morphological element marking the verb as subjunctive. In the following section, I show that a typological perspective might shed some light on the status of this element since it shares many of its peculiar properties with other Balkan counterparts such as Bulgarian da and Greek na. 4.10.1.1. The status of the element s a˘ The controversial status of s˘a is recognized by Dobrovie-Sorin (1994). She shows that s˘a has properties of both complementizers and inﬂectional elements: the analysis she proposes tries to reconcile this double behaviour without having to argue in favour of either possibility. Dobrovie-Sorin proposes that s˘a is generated under COMP, which is indistinguishable from INFL since they are adjacent. In particular, she assumes that s˘a governs an IP constituent whose SPEC position is empty. Given this conﬁguration, the operations of Functional Coindexation and of Restructuring Incorporation can apply and the result is a complex head which is labeled C/I/V and whose projection is a CP/IP/VP ‘matching projection’. Thus, the INFLlike behaviour of s˘a, due to incorporation and its COMP-like nature, due to the position in which it is generated, are reconciled in this analysis. The result, however, is the classiﬁcation of s˘a into a rather unusual category and a speculative analysis which does not shed new light with respect to its status. In the rest of this section, I consider s˘a from a typological perspective and I compare it to its Balkan counterparts: Greek na and Bulgarian da. The Greek subjunctive marking na is similar to s˘a in exhibiting ambiguous behaviour between an inflectional element and a complementizer. As discussed in Quer (1995), na shares with complementizers the property that it is sentence initial, it precedes negation and clitics, and can head embedded clauses in the absence of any other complementizer. On the other hand, it behaves as an inflectional element in that it must be adjacent to the verb cluster, it must be repeated under conjunction, and it can cooccur with wh-phrases. Arguments for both positions have been provided in the literature: Philippaki-Warburton and Veloudis (1984), Philippaki-Warburton (1987), and Rivero (1994) analyse

Syntax and its Interfaces

203

na as an inflectional element while Agouraki (1991), Dobrovie-Sorin (1994), and Tsoulas (1995) consider it a complementizer.28 The comparison with Greek na and the way it has been accounted for in the literature does not settle the issue about the status of Romanian s˘a. I show that it might be more promising to look at Bulgarian da and at the evidence that Rudin (1985) has provided with respect to the status of this element. I show that it also applies to Romanian s˘a, for which a similar analysis could be proposed. Rudin (1985) notes that no definitive answer has been given in the literature with respect to the status of Bulgarian da. She fills this gap by arguing against the evidence which suggests that it should be considered a complementizer and showing that it should be analysed as an auxiliary similar to English to (Pullum 1982). Rudin notes that, at first sight, the element da seems to behave like a complementizer. It introduces sentential subjects (349a) and objects (349b): (349) a. Da preseˇce napravo beˇae nemislimo. [bg] straight was unthinkable DA cut ‘To cut straight across was unthinkable.’ b. Tja predpoˇcita da ostane tuk. here DA stay she prefers ‘She prefers to stay here.’ Romanian s˘a shares similar behaviour, as can be seen in these examples from Mallinson (1986): iei f a˘ r˘a plat˘a este s˘a furi. [ro] CONJ take without payment be CONJ steal

(350) a. S˘a

‘To take without paying is to steal.’ b. Vreau s˘a plec. want CONJ leave ‘I want to leave.’ Furthermore, both da and s˘a appear in complement clauses and are selected by a verb, noun, or adjective which subcategorizes for this clause. In many cases, Bulgarian da alternates with cˇe. However, it seems that while cˇe tends to occur with ‘intellectual’ verbs such as to think and to tell and in factive complements, da appears with ‘emotional’ verbs such as want and like and in nonfactive 28 As noted by Quer (1995), another possibility would be in terms of an analysis featuring I-to-C movement (either overt or covert), which could reconcile both points of view.

204

Syntax and its Interfaces

complements. A similar distinction can also be seen in Romanian, where the element s˘a often alternates with c˘a with verbs such as a ¸sti ‘to know’ (but there is a change of meaning) or a spera ‘to hope’.29 However, as noticed by Mallinson (1986), factivity seems to be associated with c˘a while s˘a is predictable with wish-type verbs. Despite these apparent similarities, da differs from complementizers such as ce¸ and from wh-words in many respects, as argued in Rudin (1985). The same evidence seems to hold in the case of Romanian s˘a. Rudin shows that focused constituents cannot follow da, as can be seen in (351b) in which the focused element decata cannot occur after da, while this is not the case in (351a) where it occurs after ce¸: (351) a. Kazvat cˇ e decata pejat. [bg] say COMP. children sing ‘They say that the children are singing.’ b. *Iskam da decata pejat. want DA children sing ‘I want the children to sing.’ The situation is similar in Romanian since the subject cannot occur after s˘a (352b), but it can occur in that position if c˘a is present (352a), as can be seen in these examples from Mallinson (1986): Ion mai vine ast˘azi. [ro] (352) a. Nu cred c˘a NEG. believe COMP. John more come today ‘I don’t believe John is coming again today.’ b. *Nu cred s˘a Ion mai vine ast˘azi. NEG. believe CONJ. John more come today ‘I don’t believe John is coming again today.’ More generally, nothing can intervene between da and a verbal complex that consists of pronominal clitics and negation; even adverbs cannot occur in this position. The situation is similar in Romanian since the only elements which can separate s˘a from the verb are clitics, that is pronominal elements, monosyllabic intensifiers, and (in certain cases) negation, as can be seen in the following example from Dobrovie-Sorin (1994): (353) Vreau s˘a nu-l mai ˇıntˇılnes¸ti. [ro] want CONJ. NEG.CL.ACC. INT. meet ‘I want that you don’t meet him again.’ 29

It should be noted that while s a˘ takes a subjunctive, c a˘ is followed by an indicative.

Syntax and its Interfaces

205

Furthermore, in Bulgarian da can cooccur with complementizers and with whelements. It should be noted that in the Chomskyan tradition, wh-elements are assumed to fill the same structural position as complementizers. Since in Bulgarian it is not possible to have more than one element in this position (i.e. doubly filled COMP), one can conclude that da cannot be a complementizer. The examples below show that da can occur with the complementizer dali ‘whether’ (354a) as well as with the wh-element k ude ˘ ‘where’ (354b): (354) a. Ne znaja dali da otida. [bg] NEG. know COMP. DA go ‘I don’t know whether to go.’ b. K˘ude da otidem? where DA go ‘Where should we go?’ Also in this case, one notices a behaviour similar to Romanian s˘a, since it can also cooccur with complementizers such as ca ‘for’ (355a) and with the wh-element unde ‘where’ (355b): (355) a. Vreau ca Ion s˘a plece. [ro] want COMP. John CONJ. leave ‘I want John to leave.’ b. Unde where

s˘a

l˘as˘am

CONJ. leave

mas¸ina? car

‘Where should we leave the car?’ The evidence discussed in this section leads Rudin to conclude that da cannot be a complementizer in Bulgarian and to propose an analysis in which it is considered an auxiliary on a par with English to, which Pullum (1982) analyses as a nonﬁnite auxiliary verb. Rudin notices that da shares its peculiar position with a group of invariable verbs which are arguably also auxiliaries. Furthermore, it clearly expresses modality. A similar conclusion could thus be reached also for Romanian s˘a which shares with da all these properties. 4.10.2. The modal verb a putea

A discussion of Romanian auxiliary-like constructions should also include the modal verb a putea since it shares some common properties with tense auxiliaries, the most relevant of which is clitic climbing. As already mentioned, Romanian does not exhibit many environments in which clitic climbing is allowed. This might be due to the fact that the use of

206

Syntax and its Interfaces

bare inﬁnitives is disappearing in this language. As already noted in the previous section, raising (and control) verbs are generally followed by a subjunctive. However, the clitic distribution shown with tense auxiliaries discussed in Section 4.5.3 can also occur if the modal verb a putea ‘can/may’ is present, as can be seen from example (279), repeated here, in which the modal verb is followed by a bare inﬁnitive: (279) O CL.ACC

pot vedea. [ro] can see

‘I can see her.’

In order to deal with these cases of clitic climbing, an analysis in terms of argument composition can be assumed, as for tense auxiliaries. The following lexical entry is associated with the modal verb: (356)

⎡

verb ⎢ ⎡ ⎢ ⎢ ⎢SUBJ 1 NP ⎢ ⎢ ⎡ ⎢ ⎢ ⎢ ⎢ w-ss ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢HEAD verb ⎢VAL ⎢ ⎢CLTS elist ⎢ ⎢ ⎢ ⎢COMPS 2 ⎢ ⎢ ⎡ ⎢ ⎢ ⎢ ⎢ ⎢ 1 SUBJ ⎢ ⎢ ⎢ ⎣VAL ⎣ ⎢ ⎣ ⎢ COMPS 3 ⎢ ⎢ ⎣ ARG-ST 1 , 2 ⊕ 3

⎤

HEAD

⎤ ⎥ ⎥ ⎥ ⎥ ⎥ ⎤⎥ ⊕ ⎥ NP ⎥ ⎦⎦

3

⎤⎥ ⎥ ⎥⎥ ⎥⎥ ⎥⎥ ⎥⎥ ⎥⎥ ⎥⎥ ⎥⎥ ⎥⎥ ⎥⎥ ⎥⎥ ⎥⎥ ⎥⎥ ⎥⎥ ⎦⎥ ⎥ ⎥ ⎥ ⎦

The description above states that a putea inherits the complements of the infinitival through the operation of argument composition. This is indicated by the tag 3 , which is present both in the COMPS list of the modal verb and in that of the embedded verb, as well as by the presence of the arguments of the embedded verb in the ARG-ST list of the auxiliary verb. It is therefore possible to deal with (279) where the clitic, which represents a complement of the embedded verb, does not attach to the infinitival, but to the modal. Recall that this example constitutes further support for the phonological explanation of the impossibility of (276b), repeated here:

Syntax and its Interfaces

207

v˘azut. [ro] (276b) *O-am CL.ACC.have seen ‘I have seen her.’ The modal begins with a consonant, so that o can attach to it, while this is not the case in (276b) where the auxiliary begins with a vowel. Furthermore, the condition CLTS elist excludes the possibility of the clitic o being attached to the bare infinitive: (357) *Pot vedea-o. [ro] can see-CL.ACC ‘I can see her.’ Clitic climbing is obligatory with the verb a putea it is not possible to have any type of clitic attached to the infinitival: (358) *Pot vedea le. [ro] can see CL.ACC ‘I can see them.’ These data provide further support for the condition imposed on the spelling out constraints which excludes bare infinitives as possible hosts for clitics. The constraint in (281), repeated below, accounts for the sentence above which is correctly ruled out: (281) Realization of the accusative, third person feminine plural clitic ⎤ complex-morph ⎤ ⎡ ⎡ ⎢ ⎤⎤ ⎡

⎥ ⎥ ⎢ prefix ⎥ ⎢ ¬ (psp ∨ bare-inf) HEAD VFORM ⎥⎥→ ⎣AFFIX ⎣ ⎢ ⎦⎦ ⎢ ⎥⎥ ⎢STEM | SS | L | C⎢ PHON le ⎥⎥ ⎢ ⎢ ⎦⎦ ⎣ ⎣ CLTS NP acc [3fpl] ⎡

The analysis of the modal a putea interacts in the appropriate way with that of the tense auxiliaries presented in Section 4.5.3. If both the modal and a tense auxiliary verb are present, the clitic attaches to the former and not to the latter: (359) a. Am putut-o vedea. [ro] have can-CL.ACC see ‘I have been able to see her.’ putut vedea. b. *O-am CL.ACC.have can see

208

Syntax and its Interfaces

These data constitute additional evidence for the constraint (282), which regulates the realization of the clitic o, repeated here: ⎤ ⎡ word ⎥ ⎢ → (282) ⎣ ⎦ MORPH | ST | SS | L | C | CLTS NP acc [3fsg] ⎡ ⎤⎤ ⎡ complex-morph ⎢ ⎤⎥⎥ ⎡ ⎢ ⎢ ⎥⎥ ⎢ word ⎢ ⎥⎥ ⎢ ⎡ ⎤ ⎥ ⎢ ⎥ ⎢ ⎢ ⎢ ⎥⎥ ⎢ 2 1 ⎥⎥ ⎢ VOW ⊕ ⎢ ⎥ ⎢ ⎢ ⎥⎥⎥⎥ ⎥ ⎢ ⎢PHON ⎢ ⎢ ⎣ ⎦⎥⎥ ⎥ ⎢ ⎢ ⎥ ⎢ 3 1 ⎢ ⎥ ⎢ SKEL ⊕ ⎥⎥ ⎢STEM ¬⎢ ⎢ ⎥ ⎥⎥ ⎢MORPH⎢ ⎡ ⎢ ⎥⎥ ⎢ ⎤ ⎥ ⎢ ⎥ ⎢ ⎥⎥ ⎢ ⎢ ⎥ ⎢ verb ⎥⎥ ⎢ ⎢ ⎣ ⎦ ⎦ ⎣ SS | L | C HEAD ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ AUX + ⎢ ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎤ ⎡ ⎢ ⎥⎥ ⎢ ⎢ affix ⎥⎥ ⎢ ⎢ ⎦⎥ ⎣AFFIX ⎣ ⎦ ⎦ ⎣ PHON | SKEL o In this constraint, the condition VFORM ¬psp is not present, allowing for sentence (359a) in which the clitic o combines with a past participle. On the other hand, the constraint states that the clitic cannot combine with an auxiliary which begins with a vowel, thus accounting for (359b). Despite the common property of clitic climbing, a putea differs from tense auxiliaries and from the other constructions discussed in the previous section in several respects. In particular, adverbs can separate the modal from the infinitival, unlike in the case of tense auxiliaries (196): (360) Se

putea uneori pierde. [ro] can sometimes lose ‘He could sometimes lose himself.’ CL.REFL

Among the adverbs which can occur in this position are adverbs such as probabil ‘probably’ and other ones such as iar ‘again’ or inca ‘again’:30 (361) a. Tu poti iar veni? [ro] you can again come ‘Can you come again?’ b. Eu pot înc˘a veni la tine. I can again come by you ‘I can come again to your place.’ 30 As for quantiﬁers intervening between the modal and the inﬁnitival, there is variation in judgements.

Syntax and its Interfaces

209

Similarly, a subject can separate a putea from the embedded verb, as can be seen in (362a) with a pronominal subject and in (362b) from Legendre (2000b), which represents a case of subject–aux inversion: (362) a. Ne-putem noi închipui. [ro] CL.ACC.can we imagine ‘We can imagine.’ b. Cînd poate Ion veni mîine? when can John come tomorrow ‘Can John come tomorrow.’ This is a property that modals share with raising constructions (342). Recall, however, that subjects cannot intervene between the relevant verbs in the case of tense auxiliaries (210) and in the colloquial future (321). These differences can be accounted for if the modal verb is associated with a flat structure since this configuration allows elements to intervene between the two verbs: (363) Flat structure VP

V [AUX

+]

H H V

HH

COMPS

Additional support in favour of the flat structure comes from the position of negation. As with tense auxiliaries (308) and the colloquial future (327), nu cannot intervene between the modal verb and the embedded one: (364) *El poate nu veni. [ro] he can NEG. come ‘He cannot come.’ This distribution is the expected one given that negation modifies a VP, but the nonfinite verb is a V that cannot project. On the other hand, intensifiers constitute a problem. Recall that they usually cannot occur between the modal verb and the infinitival (292a), but normally precede the verb a putea (292b):31 (292) a. *(Nu) pot mai/cam/prea/tot veni. [ro] come NEG. can INT. ‘I can(not) come again/a little/too much/still.’ 31 The only exception to this is constituted by the intensifier ¸si which can cooccur between the two verbs for some speakers.

210

Syntax and its Interfaces b. (Nu) mai/cam/prea/tot pot veni. can come NEG. INT. ‘I can(not) come again/a little/too much/still.’

Since intensifiers attach to the lexical verb, one would expect them to be able to combine with the infinitival, contrary to facts. It seems, therefore, that intensifiers behave as pronominal clitics in terms of their distribution in the presence of the modal verb: it is thus desirable to have a unitary analysis. If intensifiers are contained in the CLTS list, (292a) is ruled out for the same reason (357), repeated below, is considered ungrammatical: (357) *Pot vedea-o. [ro] can see-CL.ACC ‘I can see her.’ The condition that the CLTS list of the embedded verb must be empty, as stated in the lexical entry for a putea, excludes the possibility that pronominal affixes or intensifiers are attached to the bare infinitive. Intensifiers will be added to the CLTS list by the lexical rules proposed in Section 4.6, which should be thus modified. It should be noted that intensifiers do not cluster with the object clitics: for example, it might be possible to have enclitics which cooccur with intensifiers. Therefore, it does not seem adequate to have realizational constraints to spell them out. It is more appropriate to deal with their realization by means of the lexical rules. To conclude: the proposed analysis accounts for the differences between the modal a putea and the other auxiliary(-like) constructions by assuming a slightly different constituent structure while the similarities, such as the distribution of the clitics, follow as a consequence of the lexical operation of argument composition. It should be noted that the modal has the same structure as the Italian (Monachesi 1996a) and French auxiliaries (Abeillé and Godard 2002), with which it shares several properties. This is a conclusion also reached in Dobrovie-Sorin (1994) even though she assumes a different analysis from the one proposed here.

4.11. A brief review In this chapter, I have taken several Romanian auxiliary(-like) constructions into consideration and I have analysed their properties with respect to adjacency requirements and with respect to the presence or absence of clitic climbing. The following table summarizes these constructions:

Syntax and its Interfaces

211

(358) Properties of Romanian auxiliary(-like) constructions

tense aux. colloq. fut. perf. aux. raising verb modal

quant. no no no no ??

intervening elements adv. subj. neg. int. no no no yes no no no no no no no no/yes no yes no no yes yes no no

cl.cl. yes no yes no yes

I have claimed that the common properties with respect to adjacency should be accounted for in terms of syntactic structure which is the most appropriate way to deal with the variation attested. On the other hand, the presence/absence of clitic climbing has been dealt with in terms of the mechanism of argument composition which is lexically triggered. Furthermore, the analysis proposed can account for the common monoclausality effects which are exhibited by tense auxiliaries and by the modal verb a putea regardless of their structural diversity. This is because the notion of argument composition is orthogonal to that of syntactic structure.

4.12. Comparison with other accounts The Romanian verbal complex has not been the object of wide-ranging discussion in the linguistic literature; however, there are a small number of studies devoted to it. In this section, I would like to consider these works in more detail and to compare them with the analysis proposed here. The most influential and complete work on Romanian syntax is without doubt Dobrovie-Sorin (1994) which discusses the Romanian verbal complex, among other topics. I have already addressed it in this chapter; however, in this section I take a more critical view of it. Dobrovie-Sorin devotes several sections of her book to auxiliary constructions, which are analysed in great detail. She argues for a biclausal structure, in which Romanian auxiliaries are sisters of CP/IP constituents. It should be noted that the biclausal structure she proposes is not standard in that both the INFL node and the NP subject position of the main clause are lacking. She suggests that both the tense and the colloquial future auxiliaries should be analysed as adjoined to a CP/IP category, but it is not clear in which sense this is supposed to be a biclausal conﬁguration. Dobrovie-Sorin compares her ‘biclausal’ analysis to one in which the auxiliaries and the nonﬁnite verb form a verbal complex as in Emonds (1978) for English and to an analysis where the relevant verbs combine in a flat

212

Syntax and its Interfaces

structure.32 She says that the former might capture the intuition that there is a strong coherence between the auxiliaries and the lexical verb. This is an observation that I share and that I have emphasized in my treatment of tense auxiliaries. Dobrovie-Sorin concludes, however, that her analysis is superior to one which assumes a verbal complex for several reasons. In particular, she claims that it can deal with the affix-like behaviour of Romanian auxiliaries, which is shared by the tense auxiliaries and the colloquial future for which she proposes the same analysis. However, it is not entirely correct to assume a common treatment for tense auxiliaries and for the colloquial future. The fact that the auxiliaries which occur in the two configurations exhibit different forms and that the distribution of the clitics is not the same reveals that there are crucial differences between these constructions which need to be taken into account. It should be noted that raising verbs also exhibit affix-like behaviour, even though Dobrovie-Sorin has not argued for a unitary analysis in this case and has provided a different explanation for the affix-like properties of this construction. It is for this reason that I have drawn a distinction, in my account, between the adjacency requirements and the possibility of clitic climbing in the case of these constructions. In particular, the patterns of impossibility of elements intervening between the two relevant verbs are identical in the case of tense auxiliaries and the colloquial future and they are accounted for in terms of the same syntactic structure under my proposal. In contrast, the different behaviour with respect to clitic climbing is dealt with by assuming that the lexical operation of argument composition is triggered in one case but not in the other. Dobrovie-Sorin claims that another advantage in favour of her ‘biclausal’ analysis is that it can correctly account for the distribution of the clitic o and that of the intensifiers. However, I have shown that this is also the case within a ‘verbal complex’ analysis. Furthermore, she mentions that it can deal with the existence of inverted conjugations. However, in the analysis presented here, inverted conjugations are used as evidence for the appropriateness of a verbal complex. The analysis of clitic climbing proposed in Dobrovie-Sorin (1994) is open to several criticisms. We have seen that with tense auxiliaries the clitic o does not climb and attaches to the embedded verb. She proposes that this exceptional case represents the regular S-structure clitic position. On the other hand, the

32 The verbal complex analysis is similar to the one proposed in this chapter for tense auxiliaries, even though in Emonds (1978) a slightly different structure is assumed.

Syntax and its Interfaces

213

pre-auxiliary position which is occupied by all the other clitics is derived in Phonological Form by phonological rules. This approach is at least controversial especially as no insights are offered on the nature of such rules and the kind of phonological processes involved. It should be noted that there are similarities behind this treatment and the lower-copy analysis proposed by Boˇskovi´c (2001) in the case of the clitic o, which I have discussed in Section 4.5.3. They both assume that the clitic is originally located next to the lexical verb from which it moves: in syntax in Boˇskovi´c (2001) or in PF according to Dobrovie-Sorin (1994). It should be noted, however, that phonological rules do not seem to play a role in the analysis of clitic climbing proposed for the modal verb a putea by Dobrovie-Sorin: the same phenomenon therefore receives a different explanation. She suggests that the modal subcategorizes for a VP complement; it thus has a monoclausal structure, unlike the tense auxiliaries. Since she assumes that pronominal clitics adjoin to IP, they do not have a node to attach to in the lower clause.33 Therefore, the clitic must adjoin to the first available INFL, which is in the main clause. Dobrovie-Sorin therefore presents two different explanations for the same phenomenon of clitic climbing: a phonological one in the case of tense auxiliaries and a syntactic one in the case of the modal verb. Furthermore, she does not mention how such an approach could account for similar cases of clitic climbing in other Romance languages. In contrast, the analysis of clitic climbing in terms of argument composition proposed in this work has proved successful in accounting for a wide range of data both in Italian (Monachesi 1993a, 1998a, 1999a) and in French (Abeill´e and Godard 1994, 2002; Miller and Sag 1997). The analysis in terms of argument composition shares some insights with that of Rivero (1994), which assumes that Romanian auxiliaries are light verbs in the sense of Grimshaw and Mester (1988). Under this view, the main verb transfers its argument structure to the auxiliary; however Rivero does not mention how this idea should be formalized within the framework she adopts. Furthermore, she proposes that auxiliaries take a VP as complement, but she has to stipulate why adverbs and quantiﬁers cannot intervene between the two verbs. In contrast, under the approach suggested here, this property follows from the fact that the auxiliary and the lexical verb form a verbal complex. The Romanian verbal complex has been the topic of a recent paper by Legendre (1999) which shares some similarities with the analysis proposed here. 33

It should be noted that the adjunction of heads to maximal projections was not legitimate within the framework Dobrovie-Sorin assumed for her analysis. She stipulates, however, that this might be possible due to the defective nature of IP. In fact, she stipulates further that clitics adjoin to a SPECless IP.

214

Syntax and its Interfaces

In particular, in both approaches pronominal clitics and intensiﬁers do not have the status of autonomous syntactic elements, but represent the morphophonological realization of certain features. Legendre, however, extends her analysis to auxiliary verbs and negation. Furthermore, she assumes that clitics are phrasal afﬁxes which are subject to alignment constraints of the EDGEMOST family. Under her approach, their order is stipulated since the EDGEMOST constraints are ranked in order to obtain the appropriate linearization. On the other hand, an advantage of the analysis advocated in this chapter is that the order of the clitics follows as a consequence of their different status and of the general architecture of the grammar. That is, object pronouns and intensiﬁers exhibit affix-like properties; they combine with the verb as a result of lexical processes. On the other hand, tense auxiliaries and negation nu share word-like behaviour; they are thus items which combine with the verb in syntax.

4.13. The role of argument structure: A comparison The analysis of clitic climbing presented in this chapter is based on the lexical mechanism of argument composition, according to which the trigger verb inherits the complements of the embedded verb. It has been assumed that argument composition affects both the COMPS list and the ARG-ST list of the auxiliary verb. In other words complements of the embedded verb are merged with those of the trigger verb at both levels. However, this is not the only possibility, as shown in Abeill´e et al. (1998). They suggest that there might be two different kinds of argument composition which reflect two different ways in which predicates share their arguments. Before discussing these two different possibilities, I summarize why the feature ARG-ST has been introduced in HPSG and what its role is supposed to be. In Pollard and Sag (1987), and in the first eight chapters of Pollard and Sag (1994), the subcategorized arguments of a head are stored in the SUBCAT list. In the last chapter of Pollard and Sag (1994), following suggestions of Borsley (1987), the valence features SUBJ, COMPS, and SPR are introduced. This move allows for a distinction between external and internal arguments, since subjects are identified and distinguished from other complements (as in LFG), and it solves certain technical problems, as discussed in Pollard and Sag (1994), to which I refer for further details. However, the SUBCAT list, which is the append of the SUBJ, COMPS, and SPR list, is not dispensed with. It is kept in order to deal with Binding Theory, which otherwise would have had to be redefined over the new valence lists. At first sight, there seems to be a certain redundancy between the SUBCAT list and the VALENCE lists. However, the existence of these two lists is empirically

Syntax and its Interfaces

215

motivated, as argued at length in Manning and Sag (1998). For example, it allows for an analysis of pro-drop without empty elements by assuming that the relevant argument that appears on the SUBCAT list is not present on the VALENCE list. The SUBCAT list has, therefore, become similar to certain notions of argument structure and it has been renamed ARG-ST. However, the ARG-ST list should be viewed as a syntactic representation and not as a partial semantic representation. Given the existence of this list and of the VALENCE list, it is possible to dissociate argument structure from valence and have two independent syntactic notions of these concepts. This architecture not only allows for an analysis of pro-drop, but also makes it possible to deal with binding patterns in Austronesian and ergative languages, as well as the binding patterns attested in ‘subject-oriented’ reflexives when they occur with passive or causative verbs (Manning and Sag 1998). An interesting case of dissociation between valency and argument structure, which further motivates the existence of these various lists, is attested in the presence of clitic climbing in French. French auxiliary verbs and causative verbs exhibit different behaviour with respect to clitic climbing. Even though certain clitics can stay on the infinitival verb after causative faire ‘make’, no clitics can be realized on the past participle. Abeill´e et al. (1998) have proposed an analysis of French clitic climbing in terms of argument composition, but they have suggested a distinction between a-composition and c-composition. The former requires that part of the argument structure of the trigger verb is identical to that of the embedded verb, while the latter requires that part of the argument structure of the trigger verb is identical to the COMPS list of the embedded verb. A-composition involves the lexical verbal argument’s ARG-ST list, while c-composition only involves the COMPS list. On the basis of this distinction, it is possible to account for the different behaviour of the French causative and auxiliary verbs. It should be mentioned that there does not seem to be evidence for this distinction in Italian (Monachesi 1999a) and in Romanian. It is evident that the notion of argument structure associated with the ARGST list in HPSG is rather different from that encoded in the a-structure in LFG (Bresnan 2001; Dalrymple 2001). In LFG, argument structure is viewed as an interface between the semantics and syntax of predicators. On the semantic side, it represents the core participants in events designated by a single predicator; on the syntactic side it encodes the minimal information needed to characterize the syntactic dependents of an argument-taking head. In other words, argument structure captures lexical information about the number of arguments, their syntactic type, and their hierarchical organization necessary for the mapping to syntac-

216

Syntax and its Interfaces

tic structure. It is obvious that the HPSG representation of argument structure takes into consideration almost exclusively its syntactic side. The general conception of argument structure assumed within LFG is rather similar to that proposed by Rappaport, Hovav and Levin (1998a, 1998b), even though there is one crucial difference which is related to the role of syntactic structure. As noted by Bresnan (2001), Rappaport Hovav and Levin interpret syntactic structure as an underlying syntactic tree prior to movement. However, within the framework of LFG this structure is redundant. The redundancy attested in Rappaport Hovav and Levin has been eliminated in Hale and Keyser (1997). They have proposed a representation in which the lexical level of argument structure has been dispensed with. Hale and Keyser (1997) suggest that the generalizations about possible meanings of verbs reflect syntactic constraints on movement. In other words, their approach denies the lexical aspect of argument structure; what is crucial is the initial syntactic structure. In LFG, however, the opposite position is adopted: it is the syntactic structure which is eliminated, that is syntactic trees do not play a role in establishing a relation between lexical semantics and syntax. This is an obvious move given the lexical character of the theory. The representation of argument structure within LFG has undergone an evolution, as noted by Dalrymple (2001). In Kaplan and Bresnan (1982), it is the semantic form value of the PRED attribute which encodes the relation between semantic roles and syntactic functions. As for more recent research on the topic, it stems essentially from two sources: the Proto-Role argument classification of Dowty (1991) is adopted and modified by Zaenen (1989), Alsina (1993, 1996), and Ackerman (1992) among others, while the Conceptual Semantics framework of Jackendoff (1983, 1990) is assumed in Broadwell (1999) and Butt (1995), though Butt (1998) moves away from it. A substantial body of research within LFG has focused on the regularities in the mapping between argument structure roles and grammatical functions, as well as on the development of appropriate linking rules. An interesting work in this respect is Bresnan and Zaenen (1990), who have proposed a comprehensive theory of mapping. They suggest that the various syntactic functions are decomposable into the feature speciﬁcations + /− R(estricted) and + /− O(bjective). Furthermore, they propose that the relation between argument roles and grammatical functions is expressed by associating these features to thematic roles, given a theory of intrinsic and default classiﬁcation of grammatical functions which depends on the particular roles involved. Mapping theory, as originally formulated, was assumed to apply within the lexicon. However, Alsina (1996) and Butt (1995) show the need to extend it to

Syntax and its Interfaces

217

instances of complex predicates in which two independent heads contribute to the predication. Butt assumes the ‘standard’ linking principle but, since argument structure must be combined at the level of the c-structure, a revision of the architectural model is required. Further work on (Romance) complex predicates that addresses architectural and linking issues is represented by Manning and Andrews (1999). They argue that complex predicates create problems for the LFG architecture which assumes various levels related via correspondence functions, since complex predicates are multiclausal at the semantic structure level, but monoclausal at the functional structure level.34 It is for this reason that they suggest abandoning the idea of having separate projections related by correspondence functions and adopting instead a uniﬁed feature structure, as in HPSG, together with the notion of information spreading. It is clear, even from this brief overview, that argument structure and Lexical Mapping Theory have become one of the most intensively investigated areas of research within LFG. Recent developments have included insights from Optimality Theory that have prompted a new look at some of the basic assumptions and mechanisms, as witnessed by the work of Butt et al. (1997), Asudeh (2001), Lødrup (1999), and Morimoto (1999) among others. The HPSG view on argument structure is far less developed, as already mentioned; the ARG-ST level is mainly syntactic in nature. As for linking theory, very little research has been carried out, perhaps the only exceptions being Wechsler (1995), Davis (2001), and Kordoni (2001), to which I refer for further details.

4.14. What is the role of syntactic structure? The data presented in this chapter dealing with the various auxiliary(-like) constructions in Romanian could lead to a different analysis than the one argued for here. In particular, the adjacency requirements between the auxiliary and the lexical verb could be dealt with in phonology and not in syntax. In accordance with assumptions in (Multimodal) Categorial Grammar, more speciﬁcally those put forward in Moortgat (1997), it could be suggested that adjacency is the result of the way elements merge in the form dimension. Moortgat and Oehrle (1996) assume that there are different modes in which linguistic material can be put together. There can be at least a phrasal com-

34

See Butt et al. (1997) for an alternative architecture which would avoid these problems.

218

Syntax and its Interfaces

position mode and a clitic attachment mode. The former combines phrasal constituents while the latter plays a relevant role in the analysis of cliticization since it puts together two lexical heads. It should be noted that each mode is associated with speciﬁc rules that deﬁne their behaviour. They regulate the way in which the different modes can interact as well as which mode takes scope over the other. In addition to these modes, a third mode of composition is proposed, that is the clause union mode, which has an intermediate status between the other two. It combines two lexical elements which have an order feature associated with them; it is this feature in combination with the appropriate distribution rules that allows for the appropriate interaction with cliticization. Taking a similar direction, it could be argued that the strict adjacency exhibited by Romanian tense auxiliaries could be accounted for by assuming that the two words merge together (in the form component) to form a new one, as shown informally below: (366) [ Word Word] Word The following constraint enforces this composition mode: ⎡ ⎡ ⎤⎤ ⎤ ⎡ (367) word ⎢ ⎢ ⎥⎥ ⎢ ⎢PHON ⎣ 1 ⎦ ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎢ ⎥ ⎤⎥ ⎡ ⎢ ⎢ ⎥⎥ ⎢ ⎢ ⎥⎥ verb ⎢ ⎢ ⎥ ⎥⎥ ⎢H ⎢HD ⎢ ⎥ ⎥⎥ ⎢ AUX + ⎢ ⎢ ⎥ ⎡ ⎡ ⎥⎥⎥ ⎢ ⎢ ⎢ ⎥⎥⎥ ⎢ ⎢SS | L | C ⎢ ⎥ word ⎥⎥⎥ ⎣ ⎢ ⎢ ⎢ w-ss ⎥⎥⎥ → PHON ⎣ ⎢ ⎢ ⎢ 3 1 ⊕ ⎦⎦⎥ ⎣VAL | COMPS ⎢ ⎣ ⎢ ⎥ HEAD verb ⎢ ⎥ ⎢ ⎥ ⎡ ⎤⎤ ⎡ ⎢ ⎥ ⎢ ⎥ word ⎢ ⎥ ⎢PHON ⎣ ⎦⎥ ⎢ ⎥ ⎢ ⎥ 2 ⎢N-HD ⎢ ⎥ ⎥ ⎢ ⎥ ⎣ ⎦ ⎣ ⎦

⎤⎤ 2

SS 3

It states that the phonology of the auxiliary verb is concatenated with that of its verbal complement to form a word. Raising verbs, however, allow for subjects to intervene between the relevant verbs exhibiting adjacency requirements which are less strict. In this case, the raising verb combines with a phrasal complement to form a phrase, as shown informally below: (368) [ Word Phrase] Phrase

⎦⎦

Syntax and its Interfaces

219

A modified version of the constraint in (367) enforces this composition mode. As for modal verbs, their behaviour justifies the postulation of an additional composition mode, in which the two words merge together to form a phrase, as shown informally below: (369) [ Word Word] Phrase In the analysis sketched above, the form dimension is associated with the PHON level. However, it is questionable whether this is the appropriate interpretation of the ‘categorial’ concept of form. The various references are not explicit with respect to the definition of this term. In Moortgat and Oehrle (1996), it seems that the form dimension could be equated with the phonological level. It is in this dimension that the various modes that compose linguistic material are active, thereby accounting for prosodic constituency. Under the ‘categorial’ view, syntax loses its role as an autonomous module of the grammar and can be seen simply as a trigger for the appropriate mapping between form and meaning. If this is the case, we obtain the undesirable result that syntactic criteria, which are the trigger for the composition modes, determine prosodic constituency. This is indeed what happens in the analysis sketched in this section, which should therefore be rejected since the constituency at the PHON level is determined on a syntactic basis, instead of a prosodic basis. As will become more obvious in the following chapter, I argue in favour of a prosodic phonology (Nespor and Vogel 1986) view which assumes that prosodic constituency is distinct from syntactic constituency. Even though there is a clear relation between the two, there is no isomorphism. In particular, prosodic constituency should be determined on the basis of prosodic rules. Therefore, syntax exists as an autonomous module of the grammar, which creates the basis for a mapping into prosodic structure, by means of an appropriate algorithm that is based on prosodic principles. Notice, however, that in more recent versions of (Multimodal) Categorial Grammar (Moortgat 2002), the form dimension is associated with syntax. This is particularly evident in the discussion of the mapping from syntactic types (introduced in the form dimension) into semantic types. Under this view, form should be understood as surface form. The remaining question is, then, how the structural representation proposed is spelled out into a phonological form.

4.15. Conclusions A rich typology of auxiliary(-like) constructions is attested within the Romance languages. They are characterized by the presence (or absence) of clitic climbing and by the possibility (or impossibility) of certain elements intervening between the auxiliary and the nonﬁnite verb. The data discussed in

220

Syntax and its Interfaces

this chapter provide new insights on the relation between syntactic structure, argument structure, and prosodic structure. I have argued that an appropriate division of labour between syntactic structure and argument structure, accounts for the similarities among tense auxiliaries, that is the common distribution of pronominal clitics, and for the variation attested in the adjacency requirements between tense auxiliary and lexical verb. The lexical operation of argument composition accounts for the extended valence of tense auxiliaries, providing in this way an explanation for the apparent climbing of pronominal clitics. However, the sharing of arguments does not determine the structural properties of these verbs. This is because the notion of argument composition is orthogonal to that of syntactic structure. Argument composition is compatible with different syntactic structures: the verbal complex one in the case of Romanian tense auxiliaries, the flat one as in Romanian a putea, and even the hierarchical one attested in the case of French passive auxiliaries. These different syntactic structures allow for a proper treatment of the adjacency requirements exhibited in the Romance auxiliary system. I have argued against the possibility of dealing with adjacency at the phonological level and provided both theoretical and empirical motivations against an analysis in this direction. The data show that nothing can intervene between the auxiliary and the lexical verb, both when the auxiliary has prosodic word status as in European Portuguese and when it is a phonologically dependent element as in Romanian. The analysis of Romance auxiliary(-like) constructions argues in favour of a model such as that of HPSG in which the representation of valence information can be distinguished from that of syntactic structure. The data presented in this chapter, as well as that which is discussed in the next, show that a common monoclausal representation (with the properties related to it) can be associated with different syntactic structures (cf. also Abeill´e and Godard (2003)). While the HPSG model does not have a problem in dealing with it, a framework which adopts only one type of representation would face problems in dealing with the relevant data.

5 Phonology and its Interfaces: The Case of Italian Restructuring Verbs 5.1. Introduction Complex predicates present grammatical frameworks with interesting interface problems, as the discussion of auxiliary verbs in the previous chapter has already shown. While the issues they raise with regard to the syntax–semantics interface have received considerable attention in the literature, this is not the case for the syntax–phonology interface. In this respect, a particular kind of complex predicate is addressed, which occurs in the presence of Italian restructuring verbs (Rizzi 1982). Following Monachesi (1993b, 1998a, 1999a), I argue that an analysis of these verbs in terms of argument composition, along the lines of that suggested for Romanian auxiliary verbs, can provide a uniform account of clitic climbing as well as of their other syntactic properties that reflect the monoclausal character of complex predicates. It is the extended valence of the restructuring verbs that can explain the apparent movement of the clitics. In this respect, the complex predicates form a unit from the point of view of the arguments for which they subcategorize. I assume with Rizzi (1982) that two different syntactic structures should be associated with Italian restructuring verbs. The situation is, however, different at the prosodic level, where there is motivation to assume only one structure. Italian restructuring verbs constitute additional evidence in favour of the non-isomorphism between prosodic and syntactic constituents, given that two different syntactic structures correspond to a single prosodic conﬁguration (Monachesi 1999b). It therefore becomes of crucial relevance to determine the constraints which are responsible for the syntax–prosody mapping. I discuss different algorithms which have been proposed in the literature. In particular, I show that those suggested by Nespor and Vogel (1986), Selkirk (1986), and Truckenbrodt (1999) make the wrong predictions with respect to the prosodic structure of restructuring verbs. More generally, a shortcoming of these approaches is that prosodic constituency is driven to a large extent by syntactic principles. This is not the case for the algorithm suggested in Ghini (1993), which is assumed in order to map syntactic structure to prosodic

222

Phonology and its Interfaces

structure in the case of restructuring verbs. It is shown that the algorithm can properly deal with the relevant data, not only when two restructuring verbs are present, but also when there is a more complex configuration due to the presence of additional restructuring verbs or intervening adverbs.1 The approach presented here is compared with a recent analysis proposed in Cinque (2000), which is based on a fixed universal hierarchy of functional projections that host the restructuring verbs. The assumptions presented in Cinque (2000) are questioned and the role of functional elements such as restructuring verbs in the syntax–phonology mapping is taken into consideration.

5.2. The data Rizzi (1982) identifies three classes of restructuring verbs in Italian: (370) Restructuring verbs

• modal verbs (e.g. potere ‘can’, dovere ‘must’, volere ‘want’) • aspectual verbs (e.g. cominciare ‘to begin’, finire ‘to finish’, continuare ‘to continue’) • motion verbs (e.g. venire ‘to come’, andare ‘to go’, tornare ‘to come back’) The verbs in (370) are usually referred to as restructuring verbs, after Rizzi (1982) who proposed a restructuring rule in order to account for their behaviour.2 These verbs act as a class with respect to certain phenomena such as clitic climbing, long NP-movement, tough constructions, and auxiliary selection. In this chapter I focus on clitic climbing, while I refer to Monachesi (1999a) for a detailed analysis of the other properties of restructuring verbs, which is based on the approach presented here. Clitic climbing is one of the most salient diagnostics, revealing complex predicate formation: (371) a. Martina lo vuole leggere. [it] Martina CL.ACC wants to read ‘Martina wants to read it.’ 1 The analysis presented in this chapter has benefited from the comments and the suggestions of Mirco Ghini. I believe that his mapping algorithm, introduced in this chapter, has contributed very interesting insights to the research on the syntax–phonology interface. I regret that his premature death has not allowed him to develop his ideas further and has put an end to a very fruitful and challenging collaboration we had in this area. 2 Alternative analyses of restructuring verbs are those of Aissen and Perlmutter (1983), Picallo (1990), Moore (1991), Rosen (1989), Roberts (1997), Kayne (1991), Burzio (1986), Bleam (1994), Kulick (1997), and Nishida (1991) among others.

Phonology and its Interfaces

223

b. Martina vuole leggerlo. Martina wants to read.CL.ACC ‘Martina wants to read it.’ In this construction, a clitic which originates as dependent of a complement verb can climb and attach to the trigger verb as shown in (371a). The two verbs thus act as a unit with respect to clitic placement and the result is a complex verb. In standard Italian, clitic climbing is optional with restructuring verbs, as shown by example (371b), where the clitic attaches to the embedded verb.

5.3. The syntactic structure of Italian restructuring verbs The different behaviour of restructuring verbs with respect to clitic climbing seems to suggest that two different structures should be associated with these verbs—A flat one (372), which is related to the clitic climbing conﬁguration, and a hierarchical one (373), which represents those cases where the clitic remains attached to the lower verb: (373) Hierarchical structure (372) Flat structure VP

VP

HH H

CL-V

V

COMPS

HH

V

VP

HH

V-CL

COMPS

Rizzi (1982) argues in favour of this hypothesis and he suggests that while there is motivation to let the embedded verb and its complements form a constituent if no clitic climbing occurs, this is not the case if clitic climbing is triggered. He proposes four tests based on Pied-Piping, Clefting, Right Node Raising, and Complex NP Shift, which provide support for two different structures. They are summarized here:

• Pied-piping applies to an infinitive with following material, but not in a sentence where clitic climbing has occurred: (374) a. Questi argomenti, a parlarti dei quali verrò these topics, to talk.CL.DAT of which will come al più presto, mi sembrano interessanti. [it] very soon, CL.DAT seem interesting verrò b. *Questi argomenti, a parlare dei quali ti these topics, to talk of which CL.DAT will come al più presto... very soon... ‘These topics, to talk to you about which I will come as soon as possible, seem very interesting to me.’

224

Phonology and its Interfaces

• Clefting can apply to an infinitive with following material, but not in a sentence where clitic climbing has occurred: i soldi che sto (375) a. È proprio a riportargli is just, to bring back.CL.DAT the money that am andando. [it] going sto b. *È proprio a riportare i soldi che gli is just, to bring back the money that CL.DAT am andando. going ‘It is just to bring him back the money that I am going.’ • Right Node Raising can apply to an infinitive with following material, but not in a sentence where clitic climbing has occurred:3 (376) a. Mario vorrebbe, ma a mio parere non potrà mai, Mario would like, but to my opinion not can never, pagargli il debito. [it] pay.CL.DAT the debt vorrebbe, ma a mio parere non potrà b. *Mario gli Mario CL.DAT would like, but to my opinion not can mai pagare il debito. never, pay the debt ‘Mario would like, but in my opinion will never be able, to pay his debt to him.’

• Complex NP Shift can postpose an infinitive with following material, but not in a sentence where clitic climbing has occurred: (377) a. Fra qualche giorno verrò a Firenze ad esporti in few days will come to Florence to submit.CL.DAT la mia idea. [it] the my idea verrò a Firenze ad esporre b. *Fra qualche giorno ti in few days CL.DAT. will come to Florence to submit la mia idea. the my idea ‘In a few days, I will come to Florence to submit my idea to you.’ 3 While the second example is considered ungrammatical in Rizzi (1982), it seems acceptable to me. On the other hand, it is not clear if Right Node Raising can be considered a reliable test for constituency. See Abbott (1976) for English.

Phonology and its Interfaces

225

The constituency tests proposed by Rizzi are based on the idea that constituents can move. The ungrammatical examples where clitic climbing has occurred show that the embedded verb and its complements do not form a constituent since they cannot move. Additional motivations for the structures proposed come from the constituency tests discussed in the previous chapter. Similarly to auxiliary verbs, in the presence of restructuring verbs the sequence which comprises the nonﬁnite verb and its complements cannot be cliticized (378a). However, this is the case only if clitic climbing occurs; if it does not occur, the embedded verb and its complements can be cliticized (378b): (378) a. *Martina gli deve dare il libro, ma non lo vuole. Martina CL.DAT must give the book, but not DAT.CL.ACC. want [it] il libro, ma non lo vuole. b. Martina deve dargli Martina must give.CL.DAT. the book, but not CL.ACC. want ‘Martina must give the book to him, but she doesn’t want to.’ In (378a), the sequence dare il libro cannot be replaced by the clitic lo, while this is possible in (378b) suggesting that it forms a constituent. It should be noted that, in the former example, clitic climbing is triggered while this is not the case in the latter. Similarly, it is not possible to question the complement of a restructuring verb (i.e. dare il libro) which has undergone clitic climbing (379a), while this is not the case if clitic climbing does not occur (379b): (379) a. *Che gli vuole Martina? – Dare il libro. [it] what CL.DAT. wants Martina? – Give the book il libro. b. Che vuole Martina? – Dargli what wants Martina? – give.CL.DAT. the book ‘What does Martina want? To give the book to him.’ Furthermore, the sequence composed by the nonfinite verb together with its complements (i.e. dare il libro) cannot undergo deletion (380a) if clitic climbing occurs, while this is not the case if clitic climbing is not triggered (380b): (380) a. *Martina gli vuole dare il libro, ma Vito non gli Martina CL.DAT. wants to give the book, but Vito not CL.DAT vuole. [it] wants

226

Phonology and its Interfaces b. Martina vuole dargli il libro, ma Vito non vuole. Martina wants to give.CL.DAT. the book, but Vito not wants ‘Martina wants to give the book to him, but Vito doesn’t want to.’

Similar behaviour can be observed for preposing; if the sequence dare il libro is preposed, the result is an ungrammatical sentence in (381a), while the sequence dargli il libro in (381b) can be preposed since clitic climbing is not triggered: (381) a. *Dare il libro, no, Martina non gli vuole. [it] to give the book, not, Martina not CL.DAT wants il libro, no, Martina non vuole. b. Dargli to give.CL.DAT. the book, not, Martina not wants ‘Give the book to him, Martina doesn’t want to.’ The data presented above clearly show that the behaviour of restructuring verbs differs depending on whether clitic climbing has occurred or not.4 If clitic climbing is not triggered, the embedded verb and its complements form a constituent, suggesting that a hierarchical structure is the most adequate. This is not the case in a clitic climbing configuration, since the lexical verb and the complements do not form a constituent; a flat structure seems more appropriate. Before concluding this section, it should be noted that a structure in which the two relevant verbs combine in a verbal complex, as in the case of Romanian tense auxiliaries, might also be appropriate for Italian restructuring verbs which have undergone clitic climbing: (382) Verbal complex VP

HH

COMPS

V V

H

V

As already pointed out by Rizzi (1982), the choice between a structure in which the two verbs form a verbal complex and the flat structure assumed here is quite difficult. However, it should be observed that lexical material can intervene between the two verbs, arguing against a verbal complex. In particular, it is possible to find constructions in which the subject can occur between the restructuring verb and the infinitival: 4 It should be mentioned that the validity of these tests has been questioned in a recent paper by Cinque (2000). I refer to Section 5.7 for a critical discussion of the evidence presented there.

Phonology and its Interfaces

227

(383) Volendolo Anna comprare usato, non avrebbe potuto contare wanted.GER.CL.ACC Anna buy used, not had could count sulla garanzia. [it] on the guarantee ‘Since Anna wanted to buy it used, she couldn’t count on the guarantee.’ Similarly, adverbs can occur between the two verbs, as shown by the following examples: (384) a. Martina ha dovuto spesso viaggiare per affari. [it] Martina has must often travel for business ‘Martina has had to travel for business often.’ b. Anna lo ha voluto immediatamente comprare. Anna CL.ACC has wanted to immediately buy ‘Anna wanted to buy it immediately.’ c. Martina non ha voluto più lavorare. Martina not has wanted more to work ‘Martina has not wanted to work anymore.’ It should be noted, however, that not all adverbs can occur in this position. Furthermore, it does not seem to be the case that their distribution is determined by semantic criteria, since there is variation in the behaviour of adverbs which belong to the same class. For example, while certain time adverbs can occur between the two relevant verbs, as shown in (384a), others such as ieri ‘yesterday’, oggi ‘today’, ora/adesso ‘now’, dopo ‘after’, presto ‘soon’, tardi ‘late’ are banned from this position. Similarly in the case of manner adverbs, while those that are formed with the addition of the sufﬁx -mente can occur between the two verbs (384b), this is not the case for manner adverbs such as bene ‘well’, male ‘badly’, volentieri ‘with pleasure’, and piano ‘slowly’. The variation exhibited by adverbs with respect to their position is extremely interesting; however, it is outside of the scope of this chapter to investigate it further. What is relevant for present purposes is the fact that certain adverbs can indeed intervene between the restructuring and the lexical verb, as well as subjects, showing that a flat structure is more appropriate to deal with the data than one in which the two verbs form a verbal complex. This is because the flat structure allows for elements to intervene between the two verbs, as discussed in the previous chapter. Restructuring verbs that trigger clitic climbing thus share several properties with French and Italian auxiliaries, as well as with the Romanian modal verb a putea ‘can’. It seems appropriate to assume a flat structure in all these

228

Phonology and its Interfaces

cases. On the other hand, a hierarchical structure is adopted if clitic climbing does not occur, since the embedded verb and its complements form a constituent.

5.4. Restructuring verbs and clitic climbing In the previous chapter, it was shown that the explanatory device used to deal with clitic climbing in the case of Romanian auxiliary verbs is the lexical operation of argument composition. As already discussed in Monachesi (1993b, 1998a, 1999a), argument composition can also provide an adequate account of clitic climbing in the case of restructuring verbs. The trigger verb inherits the complements of the embedded verb, including those which might be realized as clitics. In the previous section, I suggested that two different structures should be associated with restructuring verbs: there are (at least) two lexical entries that describe a restructuring verb such as volere ‘want’. One is a structure in which it subcategorizes for a verbal complement and its argument is illustrated here (for the second structure, see example (392)): (385)

⎡

⎤ word ⎢ ⎥ ⎢PHON | SKEL volere ⎥ ⎢ ⎥ ⎡ ⎤⎥ ⎢ ⎢ ⎥ HEAD verb ⎢ ⎥ ⎢ ⎡ ⎥ ⎢ ⎥ ⎤ ⎢ ⎥⎥ ⎢ 1 NP ⎢ ⎥ ⎢ ⎥ ⎢ ⎢SUBJ ⎥⎥⎥ ⎢ 4 ⎢ ⎢ ⎥⎥⎥ ⎢ ⎢ ⎢ ⎥⎥⎥ ⎢ ⎡ ⎤ ⎢ ⎢ ⎥⎥⎥ ⎢ w-ss ⎢ ⎢ ⎥⎥⎥ ⎢ ⎢ ⎢ ⎥⎥⎥ ⎢ ⎢ ⎥ ⎢ ⎢ ⎥⎥⎥ ⎢ HEAD verb ⎢ ⎥ ⎢ ⎢ ⎥⎥⎥ ⎢ ⎢ ⎡ ⎤⎥ ⎥⎥⎥ ⎢SS | L | C⎢VAL ⎢ ⎢ ⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎥ NP 3⎥ ⎢ ⎢COMPS 2 ⎢ SUBJ ⎥⎥⎥ ⎢ ⊕ 4 ⎢ ⎥ ⎢ ⎥ ⎢ ⎢ ⎥⎥ ⎥ ⎢ ⎢VAL ⎣ ⎦⎥ ⎢ ⎢ ⎥⎥⎥ ⎢ ⎢ ⎥ ⎢ ⎢ ⎥⎥ ⎥ ⎢ ⎢ ⎥ ⎢ ⎢ COMPS 3 ⎥⎥⎥ ⎢ ⎣ ⎦ ⎢ ⎣ ⎢ ⎦⎥⎥ ⎢ ⎥⎥ ⎢ CLTS elist ⎢ ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎣ ⎦⎦ ⎣ 3 ARG-ST 1 , 2 ⊕

This entry plays a role in those cases in which clitic climbing occurs. Volere is a subject control verb; there is, therefore, a coindexation between the subject of the inﬁnitival complement and that of the control verb, which is indicated by the tag 4 . The notation ⊕ stands for the append relation, while the

Phonology and its Interfaces

229

tag 3 indicates structure sharing between the elements in the COMPS list of the inﬁnitival and that of the restructuring verb. This is the mechanism of argument composition that makes the complements of the embedded verb become complements of the restructuring verb. The condition CLTS elist ensures that argument composition occurs only if the embedded verb does not constitute a cliticized verb form. The argument composition mechanism plays a crucial role in the analysis of clitic climbing exemplified by (371a), repeated here: (371a) Martina lo vuole leggere. [it] Martina CL.ACC wants to read ‘Martina wants to read it.’ The verb vuole subcategorizes for the verbal complement and for the arguments of the latter: (386)

⎡

⎤ ⎣COMPS V COMPS 1 , 1 NP[acc] ⎦

This analysis interacts with that of cliticization presented in Monachesi (1999a), which has been extended to deal with Romanian clitics, as illustrated in Chapter 4. A lexical approach to cliticization is assumed which treats Italian clitics as affixes.5 The Complement Cliticization Lexical Rule, introduced in Chapter 3, deals with the syntactic/semantic aspect of cliticization, that is the fact that Italian clitics satisfy the subcategorization requirements of the verb of which they are an argument: (387) Italian Complement Cliticization Lexical Rule (ICCLR) ⎤ ⎡ word ⎤ ⎥ ⎡ ⎢ 1 ⎥ ⎢HEAD verb VAL | COMPS ⎥ ⎣ ⎢ ⎦ ⎥→ ⎢ ⎢VAL | COMPS 1 2 ⎥ 2 list (cl-ss) CLTS ⎦ ⎣ CLTS elist Recall that the effect of the rule is that the complements of the verb are removed from the COMPS list and are added as members of the CLTS list. The rule interacts with the lexical mechanism of argument composition and it accounts for the 5

I refer to Monachesi (1999a) for further details about the analysis.

230

Phonology and its Interfaces

cliticized verb form lo vuole: ⎡ ⎤ (388) ⎢VAL | COMPS V COMPS 1 ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎣ ⎦ CLTS 1 NP[acc][3sgm] In particular, the element corresponding to the direct object is moved from the COMPS list to become a member of the CLTS list. Verbs that have undergone this operation are thus enriched with the relevant featural information, which is used in morphophonology for the realization of the cliticized verb form. The constraint below is responsible for the phonological realization of the information contained in CLTS as the clitic lo: (389) ⎤ ⎡ ⎡ ⎤⎤ ⎡ complex-morph affix

⎥ ⎣ ⎢ ⎦⎦ ⎦→ AFFIX ⎣ ⎣ STEM | SS | L | C | CLTS NP acc [3sgm] PHON | SKEL lo It states that if there is a STEM with a CLTS list with one element that is a third person, singular, masculine accusative NP, it must be realized as a clitic whose phonological form is lo. The cliticized verb form lo vuole is thus obtained: ⎤ (390) ⎡word ⎢ ⎥ ⎢PHON | SKEL lo vuole ⎥ ⎢ ⎥ ⎤⎥ ⎡ ⎢ ⎢ ⎥ complex-morph ⎢ ⎥ ⎢ ⎡ ⎤⎥⎥ ⎢ ⎥ ⎢ ⎢ word ⎥⎥ ⎢ ⎢ ⎥ ⎢ ⎢ ⎥⎥⎥ ⎢ ⎢ ⎢PHON | SKEL vuole ⎥⎥⎥ ⎢ ⎢ ⎢ ⎥⎥⎥ ⎢ ⎢ ⎡ ⎤⎥⎥⎥ ⎢ ⎢ ⎢ ⎢ ⎥⎥⎥ ⎢ HEAD verb ⎢ ⎢ ⎥⎥ ⎢ ⎢ ⎥⎥⎥⎥ ⎢ ⎢

⎢ ⎥ ⎢ ⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎢CLTS 2 NP acc [3sgm] ⎥⎥⎥⎥ ⎢ ⎢ ⎢ ⎢ ⎥⎥⎥⎥ ⎢STEM ⎢ ⎢ ⎢ ⎥⎥⎥⎥ ⎢ ⎢MORPH ⎢ ⎡ ⎤ ⎥⎥⎥⎥ ⎢ ⎢ 1 SS | L | C⎢ ⎢ w-ss ⎢ ⎥⎥⎥⎥ ⎢ ⎢ ⎢ ⎥ ⎢ ⎢ ⎢ ⎥⎥ ⎢ ⎢ ⎥ ⎥ ⎢ ⎥⎥⎥⎥ ⎢ ⎢ ⎢ ⎥ ⎢ ⎥ HEAD verb ⎢VAL | COMPS ⎢ ⎢ ⎢ ⎥⎥ ⎢ ⎥ ⎥ ⎢ ⎥⎥⎥⎥ ⎢ ⎢ ⎢ ⎥ ⎣ ⎦ ⎢ ⎣ ⎦⎦⎥⎥ ⎣ ⎢ VAL | COMPS 2 ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎢ ⎤ ⎡ ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎢ prefix ⎥⎥ ⎢ ⎢ ⎦ ⎦⎥ ⎣AFFIX ⎣ ⎢ ⎢ ⎥ PHON | SKEL lo ⎢ ⎥ ⎣ ⎦ SYNSEM 1

Phonology and its Interfaces

231

A flat structure is associated with example (371a), as illustrated here: (391)

VP COMPS

H HH H ⎤ H ⎡ 2 V ⎥ ⎢COMPS 2 V COMPS 1 ⎥ ⎢ ⎥ leggere V⎢ ⎥ ⎢

⎦ ⎣ CLTS 1 NP acc, 3sgm

lo vuole However, there is also an alternative lexical entry that is associated with the verb volere ‘want’, one in which the verb subcategorizes for a saturated VP (i.e. a VP with an empty COMPS list): ⎡ ⎤ (392) word ⎢ ⎥ ⎢PHON | SKEL volere ⎥ ⎢ ⎥ ⎢ ⎡ ⎤⎥ ⎢ ⎥ HEAD verb ⎢ ⎥ ⎢ ⎡ ⎤⎥⎥ ⎢ ⎢ ⎢ ⎥⎥ ⎢ ⎢ ⎢SUBJ 1 NP ⎥⎥ ⎥ ⎢ ⎥ 3 ⎢ ⎢ ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎥ ⎢ ⎡ ⎤ ⎥⎥⎥ ⎢ ⎢ ⎢ ⎥ ⎢ ⎢ p-ss ⎥⎥ ⎢ ⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎢HEAD verb ⎥ ⎥⎥⎥ VAL ⎢ ⎢SS | L | C⎢ ⎢ ⎥ ⎡ ⎤ ⎢ ⎥ ⎥ ⎥⎥⎥ ⎢ ⎥ 2⎢ ⎢ ⎢ ⎢COMPS ⎢ ⎢ ⎥ ⎥ ⎢ ⎥⎥ NP ⎢ ⎥ SUBJ ⎢ ⎢ ⎢ 3 ⎥⎥ ⎥⎥⎥ ⎢ ⎢ ⎢ ⎦⎦ ⎦⎥⎥ ⎣VAL ⎣ ⎢ ⎣ ⎢ ⎢ ⎥⎥ ⎢ COMPS elist ⎢ ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎣ ⎦⎥ ⎣ ⎦ 1 2 ARG-ST , This entry plays a crucial role in the analysis of sentences such as (371b), repeated below, in which the clitic does not climb: (371b) Martina vuole leggerlo. [it] Martina wants to read.CL.ACC ‘Martina wants to read it.’ In this case, the Complement Cliticization Lexical Rule applies to words that are licensed by the lexical entry for leggere, and its effect is that the direct object is removed from the COMPS list and it is added as a member of the CLTS list:

232 (393)

Phonology and its Interfaces ⎡

VAL | COMPS

⎢ ⎣

CLTS

elist

1 NP[acc] [3sgm]

⎤ ⎥ ⎦

The cliticized verb form leggerlo is eventually spelled out by means of the constraint presented in (386): (394)

⎡

⎤ word ⎢ ⎥ ⎢PHON | SKEL leggerlo ⎥ ⎢ ⎥ ⎤⎥ ⎡ ⎢ ⎤ ⎡ ⎢ ⎥ ⎢ word ⎥⎥ ⎢ ⎢ ⎥ ⎥ ⎢ ⎢ ⎢ ⎥⎥ ⎢PHON | SKEL legger ⎥⎥ ⎢ ⎢ ⎥ ⎢ ⎥⎥ ⎢ ⎢ ⎢ ⎡ ⎤⎥ ⎥⎥ ⎢ ⎢ ⎥ ⎢ ⎥⎥ ⎢ ⎢ ⎥ ⎥ ⎢ verb ⎢ ⎢ ⎥ ⎢ ⎥⎥⎥ ⎢HEAD ⎥⎥ ⎢STEM ⎢ ⎢ ⎥ ⎢ VFORM inf ⎥⎥ ⎢ ⎢ ⎢ ⎥⎥ ⎢ ⎥⎥ ⎢ ⎢ ⎥ ⎥ ⎢ 1 ⎥ ⎢ SS | L | C⎢VAL | COMPS elist ⎢MORPH ⎢ ⎥ ⎥ ⎢ ⎥⎥⎥ ⎥ ⎢ ⎢ ⎢ ⎥ ⎢

⎥⎥⎥ ⎢ ⎢ ⎢ ⎦⎦⎥⎥ ⎣ ⎢ ⎢ ⎣ ⎥⎥ ⎢ CLTS NP acc [3sgm] ⎢ ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎢ ⎤ ⎡ ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎢ suffix ⎥⎥ ⎢ ⎢ ⎥ ⎦ ⎣ ⎦⎥ ⎣AFFIX ⎢ ⎢ ⎥ PHON | SKEL lo ⎢ ⎥ ⎣ ⎦ SYNSEM 1

It should be noted that in the case of Italian, additional constraints are responsible for proclisis and enclisis, thus accounting for the different positions of clitic lo in (371a) and (371b). They are similar to those presented in chapter 4 for Romanian. The restructuring verb subcategorizes for the VP headed by the cliticized verb form given in (394). In this case, there is a hierarchical structure like the following:6

6 Given the two lexical entries associated with restructuring verbs, the analysis overgenerates if the subcategorization slots are filled by complements instead of clitics. There is a flat representation and a hierarchical one where the lower verb and its complements form a constituent. However, data concerning long NP-movement, tough constructions, and auxiliary selection show that they are both necessary when clitics are not present. Furthermore, Skytte et al. (1991) notice that there are regional differences in the use of the two configurations. In the dialects of central Italy and Toscana, both structures are present, while in the northern Italian dialects only the hierarchical structure is attested and in the southern Italian dialects only the flat structure appears.

Phonology and its Interfaces

(395)

233

VP COMPS

HHH HH

V COMPS VP

vuole

VP

⎡ ⎢

V⎢ ⎣ CLTS

⎤

COMPS

NP

⎥

⎥ ⎦

acc, 3sgm

leggerlo An advantage of the argument composition approach is that it is lexically constrained, therefore the mechanism is triggered only by restructuring verbs. In addition, it can easily deal with the fact that several restructuring verbs can cooccur and the clitics can attach to any of these verbs: (396) a. Martina lo vuole poter leggere. [it] Martina CL.ACC wants to can read leggere. b. Martina vuole poterlo Martina wants to can.CL.ACC read c. Martina vuole poter leggerlo. Martina wants to can read.CL.ACC ‘Martina wants to be able to read it.’ A recursive application of the argument composition mechanism can account for (396a) where the clitic is attached to the higher verb. Similarly for (396b) where the clitic combines with the middle verb. It should be mentioned that the analysis of cliticization proposed correctly predicts that the clitic is always a suffix on the lower verbs and a prefix on the top one. This is because of the constraints responsible for proclisis and enclisis proposed in Section 3.5.4, which are sensitive to the morphosyntactic features of the verb. Note that, unlike Romanian tense auxiliaries, Italian restructuring verbs do not allow split clitics. It is therefore not possible to have one clitic attached to the embedded verb while the other occurs next to the restructuring verb (397a); the two clitics must climb together (397b): (397) a. *Vito lo voleva spedirgli. Vito CL.ACC wanted to send.CL.DAT b. Vito glielo voleva spedire. Vito CL.DAT.CL.ACC wanted to send ‘Vito wanted to send it to him.’

234

Phonology and its Interfaces

The condition that the CLTS list of the embedded verb must be empty, which is present in the lexical entry in (392), enforces the two clitics to climb together and rules out sentence (397a). I refer to Monachesi (1999a) for additional details on the data presented here and for an extended discussion of the various properties of restructuring verbs such as long NP-movement, tough constructions, climbing of loro, and auxiliary selection, which are outside of the scope of this chapter.

5.5. The Prosodic Representation of Restructuring Verbs In the previous section, I suggested that two syntactic structures should be associated with restructuring verbs: a flat structure if clitic climbing occurs and a hierarchical structure if the clitic combines with the infinitival verb. A crucial question is whether this double structure is attested at the prosodic level. Furthermore, it is relevant to determine what kind of prosodic constituent a restructuring verb and an infinitival verb form. I follow Nespor and Vogel (1986) in assuming a prosodic hierarchy, which includes the following prosodic constituents (from small to large): (398) Prosodic hierarchy Syllable Foot Prosodic Word Phonological Phrase Intonational Phrase Phonological Utterance Each of these prosodic categories may serve as a domain for the application of prosodic rules. The hierarchy proposed in Nespor and Vogel (1986) includes an additional constituent: the Clitic Group. This is a constituent between the Prosodic Word and the Phonological Phrase, which groups together a lexical head and one or more adjacent clitics. Even if the clitics are not present, a Prosodic Word should always project into a Clitic Group, given the Strict Layer Hypothesis (Selkirk 1984). According to the Strict Layer Hypothesis, a prosodic constituent of a certain level immediately dominates only constituents of the next level down in the prosodic hierarchy: no level can be skipped. However, in Monachesi (1996b, 1999a) I have provided evidence against the existence of this constituent in Italian (cf. also Inkelas 1989; Inkelas and Zec 1995; Booij 1996; Peperkamp 1996;

Phonology and its Interfaces

235

Zec and Inkelas 1991; Selkirk 1995). It is for this reason that it is not mentioned in the prosodic hierarchy presented above. Turning back to the question of what kind of prosodic constituent a restructuring verb forms with the inﬁnitival, a possibility would be that they merge together in a Prosodic Word (PW). This seems a reasonable assumption, especially in those cases in which clitic climbing occurs, given that the two verbs form a complex one. In the following section, I consider the various prosodic rules that have the Prosodic Word as domain of application in order to show that the restructuring verb and its complements do not form a Prosodic Word. 5.5.1. Evidence against Prosodic Word status

The Prosodic Word represents the interaction between the phonological and the morphological components of the grammar. Appropriate mapping rules put together the morphological elements in such a way that the resulting phonological unit is not necessarily isomorphic to the morphological constituent (Nespor and Vogel 1986). Nespor and Vogel (1986) argue at length that the Prosodic Word has certain rules as its domain of application, such as Intervocalic s-Voicing (ISV) and Vowel Raising (VR). As discussed in Chapter 3, ISV applies in northern varieties of Italian and it is triggered in the contexts illustrated in (399a), but not in those in (399b): (399) Intervocalic s-Voicing a.

i[z]ola cau[z]-ava ca[z]-ina di[z]-uguale

‘island’ ‘he/she was causing’ ‘little house’ ‘unequal’

(morpheme internal) (before inflection) (before deriv. suffix) (end of prefix)

b.

ri-[s]alare porta-[s]igarette lavando-[s]i la [s]era bella [s]era

‘re-salt’ ‘cigarette case’ ‘washing oneself ’ ‘the evening’ ‘beautiful evening’

(after prefix) (compound) (before enclitic) (after proclitic) (in phrase)

Nespor and Vogel assume that a stem forms a prosodic word with all suffixes and with those prefixes that end in a consonant since in those cases ISV applies, while prefixes that end in a vowel form an independent prosodic word since they do not trigger ISV. As for clitics and for prefixes that end in a vowel, they are outside of the domain of the prosodic word and it is for this reason that ISV cannot apply.

236

Phonology and its Interfaces

ISV is not triggered when a restructuring verb combines with an infinitival verb, thus providing evidence that the two verbs constitute two distinct PWs:

(400) a. Le

voleva

[s]alire. to climb b. Voleva [s]alirle. wanted to climb.CL.ACC ‘He wanted to climb them.’ CL.ACC wanted

Vowel Raising is another rule that has the PW as domain of application. In Italian, [ɔ] and [ε] may occur only in syllables that bear primary stress, as can be seen in the following example: (401) a. l ε´ ggo b. pɔ´co

‘I read’ ‘little’

If stress shift occurs due to derivation or inflection, Vowel Raising (VR) applies to change [ɔ] and [ε] into [o] and [e]: (402) a. l[e]ggévo b. p[o]chíno

‘I was reading’ ‘very little’

In the case of the combination of a restructuring verb with an infinitival, there are two main stresses and no Vowel Raising occurs, suggesting that there are two distinct PWs: vɔglio]PW [lεggere]PW . to read b. [Vɔglio]PW [lεggerlo]PW . want to read.CL.ACC ‘I want to read it.’

(403) a. [Lo

CL.ACC want

The data above show that the behaviour with respect to the rules of ISV and VR is the same whether clitic climbing has occurred or not. Furthermore, they provide evidence that the restructuring verb and its complements do not form a Prosodic Word. However, the most convincing motivation against such a proposal comes from the fact that adverbs can intervene between the two verbs. Recall that this is a possibility whether clitic climbing occurs (404a) or not (404b): (404) a. Lo

voglio sicuramente leggere. [it] to surely read b. Voglio sicuramente leggerlo. want to surely read.CL.ACC ‘I surely want to read it.’ CL.ACC want

Phonology and its Interfaces

237

An alternative hypothesis would be to assume that the two verbs combine in a Phonological Phrase (Φ). In the next section, I provide evidence in favour of this proposal and I consider in detail the rules that have this constituent as their domain of application. 5.5.2. Evidence in favour of Phonological Phrase status

Nespor and Vogel (1986) have argued that the rules of Raddoppiamento Sintattico, Stress Retraction, and Final Lengthening have the Phonological Phrase as domain of application. In this section, I introduce these rules in detail in order to show that the restructuring verb and its complements form a Phonological Phrase. As discussed in Chapter 3, Raddoppiamento Sintattico (RS) is a rule which applies in central and southern varieties of Italian. Given a sequence of two words, the rule lengthens the initial consonant of the second word, if the first word ends in a stressed vowel: (405) a. Martina ha letto metà [l:]ibro. Martina has read half book ‘Martina has read half a book.’ b. Martina ha letto due libri. ‘Martina has read two books’ As can be seen in (405a), the word metà ends in a stressed vowel and RS is triggered; this is not the case in (405b) where the word due does not end in a stressed vowel. A further condition for the application of RS is that the two words must belong to the same Phonological Phrase:7 (406) a. (Avrà [l:]etto)Φ (il libro)Φ . ‘He will have read the book.’ b. (Visita)Φ (le città)Φ // (molto vecchie)Φ ‘He visits very old cities.’ In both of the examples above, we have two words ending in a stressed vowel, that is avr`a and citt`a. However, RS applies only in example (406a) because the two relevant words avr`a and letto belong to the same Φ. This is not the case in (406b) where citt`a and molto belong to different Phonological Phrases. In this configuration, RS does not apply since only part of the triggering conditions are met. The double slashes indicate the failure of RS. It should be noted that RS occurs in the combination of a restructuring verb with an infinitival: 7

In this and in the following examples, prosodic structure is represented by parenthesis.

238

Phonology and its Interfaces

(407) a. Lo

potrò [l:]eggere. read ‘I will be able to read it.’ b. Potrò [l:]eggerlo. can read.CL.ACC CL.ACC can

In the examples above, the word potr`o ends in a stressed vowel and RS is triggered; this fact provides evidence that the two verbs belong to the same Phonological Phrase. Stress Retraction (SR) is a rule of northern Italian. Given a sequence of two words, if the first word ends in a primary stressed vowel and the second word has its primary stress on the first syllable, then the final stress of the first word is moved leftward to avoid stress clash. Also in this case, the rule applies if the two words belong to the same Phonological Phrase:8 (408) a. (Martina)Φ (ha mangiato)Φ (metá tórta)Φ → (méta tórta) ‘Martina has eaten half a cake.’ b. (La veritá)Φ (salta fuori)Φ quasi sempre → (*vérita) ‘The truth almost always comes out.’ In (408a), the word torta is stressed on the first syllable. In order to avoid stress clash, the stress of the preceding word moves to the left. It should be noted that SR can apply because the two words belong to the same Φ. This is not the case in (408b). The relevant words do not belong to the same Φ and SR does not apply, even though the triggering conditions are met. In the combination of a restructuring verb with an infinitival, SR applies as shown in the example below: potrò leggere → pòtro read ‘I will be able to read it.’ → pòtro b. Potrò leggerlo can read.CL.ACC

(409) a. Lo

CL.ACC can

In the examples above, the word legger(lo) is stressed on the first syllable. In order to avoid stress clash, the stress of the preceding word moves to the left; the result is thus p`otro instead of potr`o, suggesting that the two words belong to the same Phonological Phrase. Final Lengthening (FL) is a phonological rule that lengthens the vowel bearing main stress in the phonological word which is ﬁnal in a Phonological Phrase: 8 Unfortunately, it is not possible to test how RS and SR interact since the former rule is attested only in central and southern Italian dialects, while the latter is a rule of northern Italian.

Phonology and its Interfaces

239

(410) a. Ho mangiato (dei pasticcini ripieni)Φ . have eaten some doughnuts filled ‘I have eaten some filled doughnuts.’ b. Ho mangiato (dei pasticc[i:]ni)Φ (ripieni)Φ (di cioccolata)Φ . filled with chocolate have eaten some doughnuts ‘I have eaten some doughnuts filled with chocolate.’ In example (410b) the word pasticcini is final in a Φ and thus the vowel that the bears main stress is lengthened. This is not the case in (410a) where the word is not final within a Φ. FL is triggered in the combination of a restructuring verb with an infinitival, showing that the two words belong to the same Φ: poteva l[e:]ggere. read CL.ACC can ‘I was able to read it.’ b. Poteva l[e:]ggerlo. can read.CL.ACC

(411) a. Lo

The data presented in this section provide convincing evidence that the restructuring verb and its verbal complement form a Phonological Phrase. Furthermore, the behaviour with respect to the prosodic rules considered is the same whether clitic climbing has applied or not. It is thus evident that there is no correspondence between syntactic structure and prosodic structure, since two different syntactic conﬁgurations correspond to a single prosodic one: (412) Mapping of the flat structure VP

H H

V

V

(

)Φ

(413) Mapping of the hierarchical structure VP

H H

V

VP

(

)Φ

As argued at length in Nespor and Vogel (1986), the domain of application of phonological rules cannot be deﬁned in terms of syntactic constituents. It is thus necessary to assume appropriate phonological constituents and a way to relate syntactic structure to prosodic structure.

240

Phonology and its Interfaces

5.6. The syntax–phonology interface As observed by Inkelas and Zec (1995), it is now generally accepted that syntax does not provide domains for phonological rules in a direct fashion. Despite works such as Kaisse (1985), which constitutes an attempt to show that phonology can access the syntactic component directly, most of the current literature presents algorithms that map from syntactic structure into prosodic structure. I adopt a prosodic phonology view which assumes that prosodic constituency is distinct from syntactic constituency (Nespor and Vogel 1986). Even though there is a clear relation between the two, there is no isomorphism (Monachesi 1999b). Different mapping algorithms have been proposed in the literature to establish a link between syntax and phonology. In the following section, I introduce the algorithm proposed by Ghini (1993), which is adopted in order to map syntactic structure to prosodic structure in the case of restructuring verbs. I claim that this algorithm is superior to other ones discussed in the literature, such as the relation-based mapping algorithm in Nespor and Vogel (1986), the end-based mapping system proposed by Selkirk (1986), and that argued for in Truckenbrodt (1999), which represents an extended and updated version of the former, cast within Optimality Theory. I show that these algorithms face crucial problems when applied to the mapping of restructuring verbs, while this is not the case for the system proposed by Ghini (1993). I argue that this is due to the fact that in Ghini’s algorithm Φ-constituency is driven by phonological principles and not by syntactic principles as in most of the other systems. 5.6.1. Ghini (1993)

The innovative character of the algorithm proposed by Ghini (1993) is that syntax determines only the domain of Φ-formation which is delimited by the right edge of an XP. This first step therefore incorporates the end-based mapping system proposed by Selkirk (1986, 1995), which I discuss in more detail in section 5.6.2. However, additional prosodic principles such as those of uniformity and average weight, symmetry, and increasing units determine the formation of Phonological Phrases within this domain. Φ-constituency is sensitive to notions of phonological weight, balance, and symmetry: 1. Φ domain formation The domain of Φ formation is delimited by right-edge Xmax boundaries. 2. Φ formation Phonological words included in a string delimited by Φ domain formation are distributed according to the principles of:

Phonology and its Interfaces

241

A. uniformity and average weight A string is ideally parsed into the same length of Φs; the average weight of the Φ’s depends on tempo: at an average rate of speech (moderato), a Φ contains two Prosodic Words (PW); the number of PWs within a Φ increases or decreases by one speeding up or slowing down the rate of speech. B. symmetry Strings are symmetrically parsed. C. increasing units If strings with an odd number of primitive Φs are not symmetrically parsed according to (b), Φs on the recursive side are heavier than Φs on the nonrecursive side. It should be noted that languages should be distinguished between those that are left recursive and those that are right recursive. Among the languages which belong to the former group can be included Turkish, Bengali, and Japanese, while in the latter group there are English, French, and Italian. In Italian, the nonrecursive side is on the left of a head: in other words, it is that part of a constituent that does not expand that is rich in phonological rules. Given the algorithm presented above, it is relevant to consider what predictions it makes.9 In the case of a sentence with two Prosodic Words, they are grouped toge