CLITIC PHENOMENA IN EUROPEAN LANGUAGES

LINGUISTIK AKTUELL
Linguistik Aktuell/Linguistics Today (LA) provides a platform
for original monograph studies into synchronic and
diachronic linguistics. Studies in LA confront empirical
and theoretical problems as these are currently discussed
in syntax, semantics, morphology, phonology, and
systematic pragmatics with the aim to establish robust
empirical generalizations within a universalistic perspective.

Series Editor

Werner Abraham
Germanistisch Instituut
Rijksuniversiteit Groningen
Oude Kijk in ’t Jatstraat 26
9712 EK Groningen
The Netherlands
E-mail: [email protected]

Advisory Editorial Board

Guglielmo Cinque (University of Venice)
Günther Grewendorf (J.W. Goethe-University, Frankfurt)
Liliane Haegeman (University of Lille, France)
Hubert Haider (University of Salzburg)
Christer Platzack (University of Lund)
Ian Roberts (University of Stuttgart)
Ken Safir (Rutgers University, New Brunswick NJ)
Höskuldur Thráinsson (University of Iceland, Reykjavik)
Lisa deMena Travis (McGill University)
Sten Vikner (University of Stuttgart)
C. Jan-Wouter Zwart (University of Groningen)

Volume 30
Frits Beukema and Marcel den Dikken (eds.)

Clitic Phenomena in European Languages

CLITIC PHENOMENA IN
EUROPEAN LANGUAGES

Edited by

FRITS BEUKEMA
University of Leiden

MARCEL DEN DIKKEN

The Graduate School and University Center of
The City University of New York

JOHN BENJAMINS PUBLISHING COMPANY

AMSTERDAM / PHILADELPHIA
 TM The paper used in this publication meets the minimum requirements of
8

American National Standard for Information Sciences — Permanence of Paper

for Printed Library Materials, ANSI Z39.48-1984.

Library of Congress Cataloging-in-Publication Data

Clitic phenomena in European languages / [edited by] Frits Beukema, Marcel den Dikken.
p. cm. -- (Linguistik aktuell / Linguistics today, ISSN 0166-0829; v. 30)
Rev. papers first presented at the Workshop on “Clitic Phenomena in English and Other European
Languages,” which was held during the Fourth Conference of the European Society for the Study of
English, Sept. 5–9, 1997, at Lajos Kossuth University in Debrecen, Hungary; with one additional
contribution.
Papers in English, with examples in various languages.
Includes bibliographical references and indexes.
1. Europe--Languages--Clitics--Congresses. I. Beukema, F.H. (Frits H.). II. Dikken, Marcel den.,
1965– III. Workshop on “Clitic Phenomena in English and Other European Languages” (1997 :
Lajos Kossuth University) IV. Linguistik aktuell ; Bd. 30.
P380.C49 1999
415--dc21 99-054884
isbn 90 272 2751 9 (eur) / 1 55619 914 7 (us) (Hb; alk. paper) CIP
© 2000 – John Benjamins B.V.
No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any other
means, without written permission from the publisher.
John Benjamins Publishing Co. · P.O.Box 75577 · 1070 an amsterdam · The Netherlands
John Benjamins North America · P.O.Box 27519 · Philadelphia pa 19118-0519 · usa
 Table of Contents

List of Contributors vii

Preface ix
Clitics at the Interface: An Introduction to Clitic Phenomena in
European Languages 1
Steven Franks
Asymmetries in the Distribution of Clitics: The Case of Greek
Restrictive Relatives 47
Artemis Alexiadou and Elena Anagnostopoulou
Second Position Cliticisation: Syntax and/or Phonology? 71
Željko Bošković
Possessive Constructions and Possessive Clitics in the English and
Bulgarian DP 121
Mila Dimitrova-Vulchanova
Agreement as a Continuum: The Case of Spanish Pronominal Clitics 147
Jon Franco
Slovene Pronominal Clitics 191
Marija Golden and Milena Milojević Sheppard
Direct Object Clitic Doubling in Albanian and Greek 209
Dalina Kallulli
Where do Clitics Cluster? 249
Ljiljana Progovac
Clitic Doubling Constructions in Balkan-Slavic Languages 259
Ivanka Petkova Schick
vi TABLE OF CONTENTS

On Clitic Sites 293

Olga Mišeska Tomić
Subject Index 317
 List of Contributors

Artemis Alexiadou Jon Franco

Zentrum für Allgemeine Universidad de Deusto
Sprachwissenschaft Aptdo 1
Jägerstrasse 10/11 Bilbao 48080
D-10117 Berlin Spain
[email protected] [email protected]
Elena Anagnostopoulou Steven L. Franks
Linguistics Section Linguistics Department
School of Philology 1021 E. 3rd Street
University of Crete Memorial Hall East, Room 322
74100 Rethymno Indiana University
Greece Bloomington IN 47405–7005
[email protected] U.S.A.
[email protected]
Željko Bošković
Department of Linguistics, U-145 Marija Golden
University of Connecticut Kamniska 4a
341 Mansfield Road 2000 Maribor
Storrs, CT 06269 Slovenia
U.S.A. [email protected]
[email protected]
Dalina Kallulli
Mila Dimitrova-Vulchanova University of Durham
Lingvistik Institutt Elvet Riverside
University of Trondheim Durham DH1 3JT
7055 Dragvoll, Trondheim United Kingdom
Norway [email protected]
[email protected]
viii LIST OF CONTRIBUTORS

Ljiljana Progovac Milena Milojević Sheppard

English Department Department of English and
Wayne State University American Studies
51 W. Warren Faculty of Arts
Detroit, MI 48202 University of Ljubljana
U.S.A. Aškerćeva 2
[email protected] 61000 Ljubljana
Slovenia
Ivanka Petkova Schick
[email protected]
Institut für Slavistik der
Universität Potsdam Olga Mišeska Tomić
Am Neuen Palais 10 English Department
Postfach 601553 University of Novi Sad
D-14415 Potsdam Stevana Musića 24
[email protected] 21000 Novi Sad
Yugoslavia
[email protected]
 Preface

The papers included in this volume were first presented at the workshop on
“Clitic Phenomena in English and Other European Languages”, which was held
during the fourth conference of the European Society for the Study of English at
Lajos Kossuth University in Debrecen (Hungary), 5–9 September 1997. The
authors who responded to our call for papers and to our various deadlines for
handing in versions of their papers deserve our warmest thanks. The contribu-
tions were subsequently refereed and revised in the light of reviewers’ comments.
One contribution, by Ljiljana Progovac, was invited by the editors.
We are especially grateful to Steven Franks for writing the overview paper
Clitics at the Interface, which opens the volume.
Our debt to the linguists who assisted us in the reviewing is enormous.
Their help was freely and generously given; without it the volume could not have
come about.
A final word of thanks must go to Klaske van Leyden, who helped us in the
final stages of preparing the manuscript for the press. Without her wizardry on
the computer and her phenomenal accuracy, the book would have been far longer
in the making.

Frits Beukema Marcel den Dikken

Department of English The Graduate School and Univer-
University of Leiden sity Center of The City University
P.O. Box 9515 of New York
NL-2300 RA Leiden 365 Fifth Avenue
[email protected] New York, NY 10016–4309
USA
[email protected]
 Clitics at the Interface
An Introduction to
Clitic Phenomena in European Languages*

Steven Franks

Abstract

Surveying the contributions in this volume, with a special emphasis on

problems in South Slavic special clitic placement, this paper addresses the
interaction of syntax with morphology and phonology, and proposes a variety
of carefully detailed specific solutions. Clitics are treated as functional heads,
consisting exclusively of formal features, these being pied-piped along with the
verb’s formal features under feature-checking verb movement. The paper
proposes an interface approach, having principles of Optimality Theory
mediate the mapping between components, and supporting a division of labour
between OT and rule-based systems.

* I have been investigating Slavic clitics since 1994, and my thinking about them has evolved
considerably over the years. I took a strictly syntactic approach in Franks and Progovac (1994) and
tried to adhere to it also in Franks (1997). More recently I have come to the perspective adumbrated
in the present paper, that an Optimality Theoretic approach can be exploited to resolve syntactically
licit constructions in ways more acceptable to the phonology and morphology. Other works which
express this perspective within a larger context, although at earlier stages of conceptualization, are
Franks (1998) and Franks and King (in press). The specific mechanisms proposed here for imple-
menting the second position phenomenon, however, are not described elsewhere, although they have
been partially reported in a series of recent talks. My approach has been largely inspired by Progovac
(1998), has profited from numerous discussions with Ljiljana Progovac, and was in one instantiation
presented as Franks and Progovac (1999). Although I have also benefited greatly from the advice of
many others, Željko Bošković needs especially to be singled out as a continual source of stimulating
interaction. As should be obvious from what follows, I have not always heeded this advice.
2 STEVEN FRANKS

1. Overview

This volume offers a selection of contributions dealing with the properties of

clitics found in various languages spoken in Europe. The paper by Franco
examines Spanish (although it brings to bear data from a host of other languag-
es), Alexiadou and Anagnostopoulou write about Greek, Kallulli treats Greek and
Albanian, and fully six of the papers, those by Bošković, Dimitrova-Vulchanova,
Golden and Sheppard, Progovac, Schick, and Tomić, are primarily devoted to
clitic phenomena in the South Slavic languages. The particular problems
investigated by these authors also range considerably, as do the scope (and
length) of the papers. Yet within this seeming diversity there is a common
thread, a central puzzle which confronts most work on the nature of ‘special’
clitics.1 This is the issue of how best to formalise the idiosyncratic properties of
clitics and, in particular, what clitics are and where in the Grammar their
properties ought to be implemented.
As has been evident since the earliest work on clitics in the transformat-
ional-generative paradigm, such as Perlmutter (1971) or Kayne (1975), clitics
defy easy categorisation and as such pose unusual problems for strictly modular
theories of grammar. The reason is that they seem to have characteristic proper-
ties at multiple levels of representation — in addition to their morphological
quirks they display special syntactic and prosodic properties that differentiate
them from other words.2 This suggests that any definition of ‘clitic’ may have
to encompass more than one subsystem of the grammar, a situation which has
never been readily accommodated in Chomskyan models, including the most
recent ‘Minimalist’ program of Chomsky (1993, 1995, 1998).3 As Progovac
(this volume) writes, ‘So far, the literature has been successful in recognising
phonological aspects of cliticisation and syntactic aspects of cliticisation. The
challenge remains, however, of putting the two together into a coherent and

1. These are opposed to ‘simple’ clitics, which lack syntactic idiosyncrasy; these terms were intro-
duced in Zwicky (1977). Since, the papers of this volume deal only with ‘special’ pronominal and (to
a lesser extent) auxiliary clitics, in this introduction I similarly restrict the discussion to such items.
2. The by no means exhaustive list in Sadock (1991: 52) includes phonology, morphology, syntax,
semantics and the lexicon (although this last is technically speaking just the repository of all
idiosyncratic information from the first four categories).
3. If anything, in this newest approach modularity is brought to its logical extreme, the ‘syntax’
being nothing more than a computational system for interfacing between the phonology and the
semantics (and the lexicon, although this point is rarely highlighted).
 CLITICS AT THE INTERFACE 3

empirical picture; as far as I can tell, this hasn’t happened yet.’ Below, in
Section 3, I will describe my own view of how to meet Progovac’s challenge.
A central theoretical dilemma raised by clitics in general is whether they are
positioned syntactically, by the normal methods available in most theories of
syntax, or prosodically, by some special phonological reordering mechanism. In
this context, as discussed in detail by Bošković (this volume), the vagaries of
Serbo-Croatian (SC) clitic placement are particularly troublesome. Since the
seminal description of SC clitics in Browne (1974), a paper which circulated as
far back as 1966, these two factors have been standardly regarded as in competi-
tion, with the result that either may prevail. The typical explanation of the
variation in (1) is thus that the clitic cluster is free to fall either after the first
prosodic word (1a) or after the first syntactic phrase (1b):
(1) a. Zeleno mi je auto kupio
green me. .3 car bought
‘He bought me a green car.’
b. Zeleno auto mi je kupio
This explanation has become very widely accepted, so much so that it is given
as a paradigm example of the interplay between phonological and syntactic
criteria in such standard textbooks as Comrie (1989). Halpern (1995) character-
ises phonologically determined placement, as in (1a), as ‘second word’ or 2W,
and syntactically determined placement, as in (1b), as ‘second daughter’ or 2D.
Together they can be regarded as variants of ‘second position’ or 2; I shall
adopt these notations here for ease of reference. The kind of variation illustrated
by (1) is recognised as problematic for any uniform account of clitic placement,
since a purely syntactic analysis will have difficulty accommodating (1a) and a
purely phonological analysis will have difficulty accommodating (1b).
There are several conceptually distinct ways of approaching the dilemma of
how clitics are positioned. As Bošković (this volume) notes, the two most
restrictive are to attempt to reduce their idiosyncrasies in both domains either to
purely phonological or to purely syntactic requirements. It is, I believe, not likely
that either of these approaches will turn out to be entirely workable, since there
always seems to be a residue of incontrovertibly syntactic or phonological facts.
An important line of research, stemming from Zwicky (1977) and best exempli-
fied by the detailed studies of Klavans (1982) and Halpern (1995), has therefore
been to pursue a mixed system, whereby reference may be made to both types
of criteria in anchoring the clitics. This third alternative, to account for clitic
placement in terms of some combination of phonological and syntactic require-
ments, is less desirable on theoretical grounds although perhaps inescapable; the
4 STEVEN FRANKS

question really boils down to which properties result from requirements in which
module, and how the two domains interact.4
Regardless of how clitics ultimately become situated, most approaches
regard them as at some level of representation distinct syntactic entities.5 They
can in principle be heads, phrases or ambiguous between the two. The evidence
is I believe overwhelming that clitics are introduced as functional heads and that,
in Slavic at least, they move as such.6 Most of the papers in this volume take a
similar stance. Although verbal auxiliary clitics are invariably heads, pronominal
clitics are typically introduced in the canonical phrasal positions occupied by
theta-role bearing arguments of the verb. The analysis of auxiliary clitics as
necessarily heads is forced by traditional phrase structure, and implies that they
originate outside VP.7 For pronominal clitics, on the other hand, there is some
variation in how they should be treated. Bošković (this volume) and Golden and
Sheppard (this volume) clearly show that pronominal clitics are arguments in
Serbo-Croatian and Slovenian, respectively. Progovac (this volume) takes
advantage of the fact that pronominal clitics in SC differ from auxiliary clitics
in precisely this regard to explain some very curious ellipsis facts discussed by
Bošković (this volume), facts which otherwise fly in the face of purely syntactic
approaches to clitic placement.8 However, the status of pronominal clitics as
heads which are exhaustively phrases opens the door to their being treated as
non-arguments and introduced as functional heads external to VP in the first
place. This is, it turns out, indeed what has happened in Spanish, as argued by
Franco (this volume) largely on the basis of the phenomenon of clitic doubling,

4. A fourth way of accommodating clitics that so far as I know remains virtually unexplored is to
derive all phonological and syntactic properties of clitics as epiphenomena resulting from a defining
factor based in some other module, perhaps morphology, discourse or semantics. Schick (this
volume), which adopts a discourse-analytical perspective, is one promising attempt in this direction.
5. One notable exception to this generalization is work by Stephen Anderson and his students.
Anderson (1993, 1995, 1996) argues that clitics are a kind of phrasal inflection, introduced by the
morphology but on phrase-sized units. Similar stances have been taken in Distributed Morphology
accounts, as well as by Legendre (1999), who, working within an OT framework, analyzes clitics as
‘PF realizations of functional features attached to verbal nodes in the syntax.’
6. As demonstrated in Franks (1998) and Franks and King (in press), this is not true of pronominal
clitics in Polish, which move as phrases and have the status of simple clitics.
7. This fact is problematic for the account in Halpern and Fontana (1994), which capitalizes in some
very interesting ways on the distinction between head and phrasal clitics to create a typology of clitic
properties. In their system, 2 clitics are phrasal, hence adjoin to IP. A phrasal analysis, while credible
(although wrong) for SC pronominal clitics, is, however, almost unimaginable for verbal auxiliaries.
8. In Franks (1998) I develop an account of partial clitic cluster ellipsis that is consistent with purely
syntactic clitic clustering.
 CLITICS AT THE INTERFACE 5

as well as in a number of the Balkan languages. Literally half of the papers in

this volume deal with Balkan languages — Alexiadou and Anagnostopoulou
(Greek), Dimitrova-Vulchanova (Bulgarian), Kallulli (Greek and Albanian),
Tomić (Macedonian), and Schick (Bulgarian) — and all either assume or argue
for an analysis in which the clitic is a non-argument which doubles an argument
DP.9 Interestingly, these papers also show that the factors which condition
doubling vary across these languages, in ways exactly like one might expect an
agreement marker to behave; Franco’s typology of argument vs. agreement
characteristics particularly clearly highlights the kinds of variation one might
encounter. Thus, while there is consensus that clitics are bundles of functional
features, the issues of what features they instantiate and how they are eventually
realised as morphological entities are complex and not well understood. In what
follows I will outline some generalisations about what facts a successful account
should accommodate and sketch some of the properties of a possible solution.

2. Some Emerging Truths

In this section I survey some general conclusions about South Slavic clitics,
based in part on the papers in this volume and augmented considerably by my
own investigations. These conclusions will serve as the skeleton for the analysis
in Section 3.

2.1 Splitting is essentially syntactic

Historically, as noted above in conjunction with (1), an important issue of debate

has concerned the status of ostensible ‘splitting’ of phrases by clitics, found
especially in SC. It has been argued by Halpern (1995), Percus (1993) and
Schütze (1994, 1996), among others, that splitting requires phonological clitic
placement. Halpern’s mechanism of ‘prosodic inversion’ (PI) was intended to
resolve ‘mismatches between syntactic and surface structure’ by inverting the
clitic and adjacent material. He suggested that clitics in SC are positioned to the
left of IP, as follows:10

9. Rudin (1997) systematically reviews approaches to clitic doubling and similarly concludes that the
clitic is a non-argument and generated outside VP in functional head positions such as Agr0. Her
arguments about Bulgarian (as well as Macedonian and Albanian) closely parallel those of Franco
(this volume) for Spanish.
10. Halpern (1995: Chapter 3) slightly revises this so that clitics are adjoined to CleftP, a phrase he
posits between CP and IP.
6 STEVEN FRANKS

(2) CP

[SpecCP] C

C0 IP

cl IP

Taking the output of the syntax to be a sentential phrase structure essentially as

in (2), with the clitic cluster adjoined to IP, Halpern then proposed that PI places
the clitic cluster after the first prosodic word to its right. In this way, 2W as in
(1a) is obtained. 2D as in (1b) could be handled similarly, with PI targeting a
syntactic phrase instead of a prosodic unit, but Halpern reasonably argued against
this variant, which a number of previous PI analyses had espoused. Instead, he
concluded that for (1b) a phrase moves syntactically to SpecCP, with the clitic
cluster remaining in situ. Prosodic inversion is thus a special mechanism which
operates, in the phonology, to shift an enclitic which the syntax happens to have
kept initial in its prosodic domain one prosodic word to its right.
This kind of mixed system has the primary advantage that, by factoring
prosodic effects out of the syntax proper, it avoids the ‘look ahead’ problem
faced by purely syntactic accounts. There is, in other words, no teleology to the
movement of the verbal participle kupio in (1c) to before the clitics mi je; rather
the clitics are prosodically stranded hence undergo PI over kupio.
(1) c. Kupio mi je zeleno auto
bought me. .3 green car
‘He bought me a green car.’
PI extends to a variety of constructions in which clitics are arguably left as initial
in their intonational phrases by the syntax. An intonational break, represented by
‘#’, can precede otherwise licit clitics, as in the following SC example, discussed
by Bošković (this volume):
(3) a. #Tvome prijatelju# #prodali su knjigu#
#your friend #sold .3 book
‘To your friend, they sold the book.’
b. #Tvome prijatelju su prodali knjigu.#
 CLITICS AT THE INTERFACE 7

The pair in (3) shows that if the topicalised constituent is treated as a separate
intonational phrase, then PI of su is forced, and if it is not a separate intonational
phrase, PI is impossible. To handle such ‘clitic third’ facts, Progovac (this
volume) argues that phrases with ‘comma intonation’ are ‘external to their CP
domain’, so that clitic positioning can be entirely syntactic. This may be correct
in (3a), and almost definitely is for examples where comma intonation is
obligatory, as in her (4):
(4) #Svojim ro acima po majci# #šta će Jovan prodati?#
#self’s relatives by mother #what .3 Jovan sell
‘To his maternal relatives, what will Jovan sell?’
Elsewhere, however, the effects of prosodic conditioning seem inescapable. In
Bošković’s (5), the enclitics sam ti cannot immediately follow the subject NP ja
tvoja mama even though this is the initial/highest phrase, since the appositive
tvoja mama is necessarily flanked by intonational phrase boundaries:
(5) #Ja#, #tvoja mama#, #obećala sam ti igračku#
#I #your mother #promised .1 you. toy
‘I, your mother, promised you a toy.’
Clearly, (3a), (4), and (5) should enjoy similar prosodic motivation, whether or
not their syntactic structures are parallel. Evidence that the word order in SC (5)
involves a strategy for recovering from a prosodically invalid output of the
syntax can be found in its Slovenian (Slvn) version in (6), cited by Golden and
Sheppard (this volume):
(6) Jaz#, #tvoja mama#, sem ti obljubila igračko
The recovery strategy that comes into play in SC (5) is not necessary in Slvn (6),
since in this language the clitics sem ti are prosodically neutral, hence can find
support either to their left, as enclitics, or to their right, as proclitics.
It is, however, unlikely that this recovery strategy is Halpern’s PI. Although
the structure in (2) seems appropriate for (4), taking šta to be in SpecCP, it is
not applicable in (5), assuming ja tvoja mama to be in the canonical SpecIP
subject position. The assumption that clitic movement targets a uniform position
is problematic here, and the data suggest that a more flexible mechanism may be
called for in order to implement the effects of prosody on where clitics are
pronounced. Further issues arise with respect to deriving the ‘verb + clitics’
order. Given a sentence as in (7a), in order to prevent both XP movement and V
movement, as in hypothetical (7b), we must assume that (in declaratives at least)
V never moves past the clitics, so that the ‘verb + clitics’ order in (1c) or (7c)
8 STEVEN FRANKS

is always derived through PI:11

(7) a. Taj pesnik mi piše zanimljivu knjigu
that poet me. writes interesting book
‘That poet is writing an interesting book for me.’
b. *[CP [taj pesnik]i [C′ [C0 pišej [IP mi [IP ti tj zanimljivu knjigu]]]]]
c. Piše mi zanimljivu knjigu
‘She/he is writing an interesting book for me.’
Similarly, we must prevent superfluous PI from deriving ungrammatical (8b)
from (8a):12
(8) a. [CP [zanimljivu knjigu] [C′ C0 [IP mi [IP taj pesnik piše]]]]
b. *Zanimljivu knjigu taj mi pesnik piše
While blocking (8b) is straightforward if PI is a ‘last resort’ mechanism,
blocking (7b) requires the auxiliary assumption that PI is necessarily invoked in
the derivation of SC (7c). However, since (7c) reflects the normal word order for
Slvn as well (although as we have seen special clitics are prosodically neutral in
that language), this assumption cannot be correct.
For reasons such as these, the need for PI has been much challenged in the
literature; see Ćavar (1996), Franks (1998), or Bošković (this volume) for
additional arguments against PI. A common epistemological objection is that a
mechanism which enriches the power of PF to literally move syntactic entities is
conceptually undesirable. Moreover, the conclusion that special clitic placement
is essentially syntactic in nature is inescapable, since it displays so many obvious
syntactic effects, such as obeying locality restrictions. As surveyed by Bošković
(this volume), the syntactic properties of clitic placement have been abundantly
presented by many linguists; see, for example, Franks and Progovac (1994),
Franks (1997), or Franks and King (in press). The fundamentally syntactic nature
of clitic movement is also readily admitted by proponents of PI.
The rejection of PI raises of course the problem of how to accommodate
splitting in SC. The fact is, however, that splitting too displays clear syntactic
effects, as discussed at length by Bošković (this volume) and Franks (1998). I
therefore assume the kind of ‘remnant topicalisation’ analysis of splitting
advocated by Ćavar and Wilder (1994) or Franks and Progovac (1994), where
apparent splitting is the consequence of independently available syntactic

11. Halpern (1995) does not actually discuss how the ‘V + clitic’ order comes about.
12. Examples (7b) and (8b) are of course acceptable with an intonational pause after the initial XP.
 CLITICS AT THE INTERFACE 9

manipulations. That this is correct is verified by the correlation between the

possibility of ‘splitting’ by clitics and the ability of non-clitics also to intervene,
as well as non-constituents. This correlation holds for a single language, such as
SC, across its various constructions, as well as for various speakers of that
language. Thus, according to Željko Bošković (personal communication), for
example, the status of (9a) as slightly marginal or completely unacceptable co-
varies for different speakers with their judgements about the status of (9b):
(9) a. ?/*Roditelji su se uspešnih studenata razišli
parents .3  successful. students. dispersed
‘The parents of the successful students dispersed.’
b. ?/*Roditelji dolaze uspešnih studenata
parents arrive.3 successful. students.
‘The parents of the successful students are arriving.’
The same correlation holds across languages, so that SC-like splitting is marginal
to unacceptable in Czech and Slvn, the latter as mentioned by Golden and
Sheppard (this volume). In general, and so far as special clitics are concerned,
the 2W phenomenon seems to be parasitic on 2D.
The rejection of PI also raises the problem of how to accommodate the
kinds of prosodic effects noted in conjunction with (3a), (4), and (5). Clitics
move in the syntax, yet are also sensitive to PF factors. There are three ways to
deal with this paradox: (i) allow clitics also to move in PF; (ii) allow syntactic
movement to satisfy phonological deficiencies; (iii) overgenerate in the syntax
and only subsequently ‘filter out’ undesirable representations. PI is an instance
of the first approach, and the second approach suffers from the look ahead
problem, as well as a potential incompatibility with the minimalist desideratum
that all syntactic movement be motivated by feature checking. I therefore take
the stance that an ‘overgenerate and filter’ kind of approach is ultimately going
to be successful. Bošković (1995) first alluded to this possibility and clarifies it
in his contribution to the present volume. I develop variants of it in Franks
(1998) and Franks and King (in press), and in Section 3 below I elaborate a
filtering approach based on Franks and Progovac (1999).

2.2 ‘PI’ may still be a fact of life

I have argued that clitic placement is basically syntactic, and that any phonologi-
cal residue can be treated as options made available by the syntax but selected
as optimal on the basis of prosodic criteria. Even so, it is worth asking one more
question about PI before turning to other items in my survey of ‘emerging
10 STEVEN FRANKS

truths’. Can a filtering approach handle all prosodic effects? Although the vast
majority of clitic placement is syntactic, is it possible to assimilate everything to
independently motivated syntactic operations? Contrary to my claims in Franks
and Progovac (1994) or Franks (1997), and as far as I can tell contrary to the
views of the authors in the present volume, I now believe that this is not
possible. Some kind of strategy for realising an enclitic which has been stranded
in initial position at the right edge of the prosodic word to its right is ultimately
going to be required. The postulation of such a mechanism is unavoidable in
Universal Grammar (UG), given that the simple clitic li, which marks Yes/No
interrogatives in various Slavic languages, including Bulgarian (Bg) and Russian
(Ru), clearly seeks the right boundary of the prosodic word to its right. This
sensitivity to prosodic structure can be seen in the fact that Bg li is placed in the
verbal complex after the first prosodic word, which can vary depending on
whether the clause is affirmative, as in (10a) or negative, as in (10b).13
(10) a. DAde li ti go?
gave  you. it.
‘Did she/he give it to you?’
b. Ne TI li go DAde?
 you.  it. gave
‘Didn’t she/he give it to you?’
The reason that enclitic ti is stressed in (10b) is because the morpheme ne in Bg
is post-stressing, hence ne + ti form a prosodic word which li can be placed
after, superficially interrupting the verbal complex ne ti go dade.14 Rivero
(1993) proposed that li lowers to I0 in the syntax (in Bg, but not SC), and as a
reaction to this a number of papers, including King (1994), Izvorski, King, and
Rudin (1997), and Rudin, Kramer, Billings and Baerman (in press), have pointed
out that Rivero disregarded examples such as (10b). Li placement is determined
by the simple fact that it is introduced higher than any other element in the tree,
in C0, but it is phonologically enclitic. This conspires to have the occasional
result that li ends up sandwiched between other clitics. Rudin, Kramer, Billings

13. The stressed syllable is capitalized and li is underlined for ease of reference.
14. This is not true of Mac ne. Instead the entire unit receives regular antepenultimate stress, and Bg
(10b) comes out as (i) in Mac:
(i) Ne ti GO dade li?
 you. it. gave 
‘Didn’t she/he give it to you?’
See Rudin, Kramer, Billings and Baerman (in press) for details.
 CLITICS AT THE INTERFACE 11

and Baerman (in press) also note that, in addition to li placement requiring
reference to stress in Bg, li never counts in locating the antepenult in Macedo-
nian (Mac). It thus seems inevitable that li placement is subsequent to stress
assignment (or at least to the determination of prosodic domains) and hence must
be phonological.15 I suggest derivations roughly as in (11) for the items in (10):
(11) a. li [[dade]w ti go]w → [[DAde]w li ti go]w
b. li [[ne ti]w [go dade]w]w → [[ne TI]w li [go DAde]w]w →
[ne TI li go]w [DAde]w
Even more telling is the behaviour of li in Russian, a language which has lost all
special clitics. Here we find obligatory splitting, as shown in (12):16
(12) a. [Na ètom]w li zavode on rabotaet?
[in this  factory he works
‘Is it in THIS FACTORY that he works?’
‘Is it in THIS factory that he works?’
‘Is it in this FACTORY that he works?’
(*Na ètom zavode li on rabotaet?)
b. [Doroguju]w li knigu on čitaet?
[expensive  book he reads
‘Is it an EXPENSIVE BOOK that he is reading?’
‘Is it an EXPENSIVE book that he is reading?’
‘Is it an expensive BOOK that he is reading?’
(*Doroguju knigu li on čitaet?)
In Ru, li necessarily comes after the first prosodic word, although it could be this
word, the entire phrase, or the head noun (i.e., some other part of the phrase) that
is actually focused, with interpretational differences rendered through intonation.
Ru no longer has any special clitics, so that any idiosyncratic placement
properties of li have to be prosodic. If the focused constituent were in SpecCP,
then Ru should be like Bg.17 Since it is not, I conclude that the syntax really
leaves the focused constituents na ètom zavode ‘in this factory’ and doroguju

15. In Section 4.4 I will return to this problem and argue that the facts of li placement in Bg indicate
that the mapping to PF must proceed in a cyclic fashion.
16. Some speakers very marginally accept placement after the first syntactic constituent. This
discussion is drawn from Franks (1998) and the examples are slightly modified from King (1994);
see also Brown and Franks (1995) and Rudnitskaya (1999).
17. And if this were 2 clitic placement, as in SC, we would only expect the grammatical forms
in (12) as a variant of the ungrammatical ones.
12 STEVEN FRANKS

knigu ‘(an) expensive book’ in the specifier of whatever phrase is immediately

to the right of C0. If so, a prosodic readjustment just as in Bg (11) applies to
produce the Ru forms in (12). Interestingly, Stepanov (1998) adduces a number
of to my mind compelling arguments that wh-phrases in Ru actually front to a
position to the right of C0. Stepanov shows that Ru fails the superiority tests
applied to SC in Bošković (1998). I interpret this to mean that there is in fact no
movement to SpecCP in Ru and that focused phrases are just like wh-phrases (in
either moving to the specifier of a Focus Phrase located between CP and IP or
in being simply adjoined to IP). By PF, li ends up to the left of the only prosodic
word which can support it, as in (13):
(13) a. [li [na ètom]w] zavode on rabotaet
b. [li [doroguju]w] knigu on čitaet
At this point something like PI is necessary to produce the correct order.
However, in the same spirit as Bošković (this volume), I do not take this literally
to be an operation of PF, but rather to be a linearisation effect: an enclitic must
be pronounced at the right edge of its prosodic host. It then comes as no surprise
that li in Ru interrupts a focused phrase precisely after the first prosodic word
within that phrase.
Given that this phenomenon exists in UG (and purely prosodic positioning
of simple clitics is well documented outside Slavic), I see no reason not to
entertain it as an option even in SC. Indeed, there is one place where it seems to
me that a mechanism with the effect of PI may be called for. Progovac (1996)
provides the following judgements for examples of constituents that do not seem
independently splittable (‘fortresses’, in Halpern’s terminology) but which can
nonetheless be broken up by clitics, drawn from Browne (1975):18
(14) a. ??Sestraće i njen muž doći u utorak
sister .3 and her husband come on Tuesday
‘My sister and her husband will come on Tuesday.’
b. ??Lav je Tolstoj veliki ruski pisac
Leo .3 Tolstoi great Russian writer
‘Leo Tolstoi is a great Russian writer.’
c. ??Prijatelji su moje sestre upravo stigli
friends .3 my sister just arrived
‘Friends of my sister’s have just arrived.’

18. Speakers differ considerably on the status of such splitting. See Franks (1997, 1998) or
Bošković (this volume) for arguments that for some speakers splitting of declinable names is actually
sensitive to syntactic factors.
 CLITICS AT THE INTERFACE 13

I claim that these are marginal because, for Progovac, they are not actually
splittable in the syntax. Thus, the addition of more clitics makes them completely
unacceptable:
(15) a. *Sestra će mi ga i njen muž pokloniti
sister .3 me. it. and her husband give
‘My sister and her husband will give it to me.’
b. *Lav mi ga je Tolstoj poklonio
Leo me. it. .3 Tolstoi gave
‘Leo Tolstoi gave it to me.’
c. *Prijatelji su mi ga moje sestre poklonili
friends .3 me. it. my sister gave
‘Friends of my sister’s gave it to me.’
Consistently, a single clitic is marginally able to split ‘fortresses’ which cannot
be penetrated by larger amalgamations of clitics. This is a curious effect, one
which to my mind clearly calls for a phonological account. Clearly, remnant
topicalisation should not be able to discriminate the phonological weight of what
interrupts the syntactic phrase. Nor, however, should Halpern’s version of PI,
which is based on syntactic structure. These SC facts reveal a restriction that
only a single syllable can ‘undergo PI’, which is corroborated by the behaviour
of li in Bg and Ru. Taking li as canonical, I suggest that a simple enclitic which
happens to be stranded at the left edge of its potential prosodic host can be
pronounced to that host’s right. This is only available for single syllables, hence
the contrast in (14) vs. (15). If simple clitics are PF entities, whereas the
dependence of special clitics is syntactically defined, it becomes clear why we
hardly see any evidence for prosodic clitic placement in SC: this is a language
with a robust special clitic system. Apparently, however, lone special clitics can
marginally slip by and be treated as simple clitics in PF. That is, if a prosodic-
ally unacceptable output of the syntax (and morphology) is allowed to enter PF,
rather than be filtered out, it can be salvaged by realising a monosyllabic enclitic
to the left of the prosodic word at that syllable’s right.
Further evidence that this is the right kind of approach can be found in the
following paradigms; judgements are again from Ljiljana Progovac (personal
communication):
(16) a. ?Vesna će i njen muž doći sutra
Vesna .3 and her husband come tomorrow
‘Vesna and her husband will come tomorrow.’
14 STEVEN FRANKS

b. ?*Vesna će mu ga i njen muž kupiti

Vesna .3 him. it. and her husband buy
sutra
tomorrow
‘Vesna and her husband will buy him it tomorrow.’
Example (16) illustrates what we have already seen in (15), that multiple clitics
cannot undergo the necessary PF adjustment. However, (17) shows that even a
single monosyllabic clitic cannot split a fortress by appearing after the first
syntactic phrase, although of course this is perfect otherwise, as in (18).
(17) a. ?Mojaće sestra i njen muž doći sutra
my .3 sister and her husband come tomorrow
‘My sister and her husband will come tomorrow.’
b. ?*Moja sestra će i njen muž doći sutra
(18) a. Moja će sestra doći sutra
my .3 sister come tomorrow.
‘My sister will come tomorrow.’
b. Moja sestra će doći sutra
The contrast between (17a) and (17b) is comparable to what happens with Ru li,
as indicated in (12). In general, whenever there is forced splitting, we can
assume that PF factors are likely to be at work.19

2.3 2PL is the highest functional head position

One possible way of achieving 2 effects by exploiting independent syntactic

properties seems to have been arrived at independently by a number of people at
around the same time.20 The idea was that if clitics all move as heads to C0, and if
either an XP moves to SpecCP or the V moves to C0, then the basic properties of the
2 effect result automatically. This kind of approach is diagrammed in (19).

19. Bošković (this volume) discusses a possible exception, the well-known type in (i), from Browne
(1975):
(i) Jako su dosadni
very .3 boring
‘They are very boring.’
(*Jako dosadni su)
20. Relevant papers dealing with SC include Ćavar and Wilder (1994), Franks and Progovac (1994),
Progovac (1996), and Tomić (1996). Veselovská (1995) applies the ‘2 = C0’ approach to Czech, and
Golden and Sheppard (this volume) also advocate it for Slvn.
 CLITICS AT THE INTERFACE 15

(19) CP

XP C

C0 IP

V clitics

This structure raises several important issues. One is why V moves to C0 only
when nothing else has moved to SpecCP — as with the problem of superfluous
PI, something must prevent both movements from occurring. Motivating V-to-C
in (19) phonologically, as in often assumed, requires looking ahead to PF to
force a last resort syntactic movement. One could imagine V-to-C here as an
overt instantiation of a movement that ordinarily occurs at LF; in other words,
the verb is pronounced higher to satisfy the clitics’ prosodic requirements and
thereby save the structure. Another issue is what causes the clitics to move to C0
in the first place; perhaps, like V, the auxiliary clitic needs to check tense
features there, but why should this implicate the pronominal clitics as well?
These are the sorts of issues that appear repeatedly in the 2 literature.
There are also a variety of conceptual and empirical questions that can be
posed with respect to (19), many of which are raised by Bošković (this volume)
and the references therein. An analysis that targets C0 requires (i) that all clauses
be CPs and (ii) that 2 clitics are always in the same position. There is, I
believe, good reason to doubt both of these assumptions. Bošković presents two
reasons why clitics cannot always be in C0 in his Section 3.2.2, one having to do
with the scope of the adverb pravilno ‘correctly’ and the other demonstrating that
SC participles never climb as high as C0 although they can precede clitics.
Consider the following minimal pair:
(20) a. Kupuje li mi zeleno auto?
buys  me. green car
‘Is he buying me a green car?’
b. *Kupio li mi je zeleno auto?
Bought  me. .3 green car
‘Did he buy me a green car?’
Apparently, participles in SC can never cross the one clitic which we indepen-
dently surmise to be in C0, that is, the Yes/No interrogative li. Nonetheless, they
can precede clitics, as we saw in (1c), repeated as (20c):
16 STEVEN FRANKS

(20) c. Kupio mi je zeleno auto

bought me. .3 green car
‘He bought me a green car.’
This means that clitics may be lower than C0, contra (19).21
Others have produced a variety of arguments against the ‘2 = C0’
hypothesis. There is now so much evidence that, as far as I am aware, virtually
all the initial advocates of this account of 2 no longer adhere to it. Progovac
(in press), for example, argues that ‘eventive to’ is in a fixed position between
CP and IP. However, as shown in (21), it sometimes follows and sometimes
precedes the clitics:
(21) a. To mi je Novak kupio knjigu
that me. .3 Novak bought book
‘What you see is that Novak has bought me a book.’
b. Šta li si mu to kupio?
what  .2 him. that bought
‘What is it that you bought him?’
Progovac thus reverses her earlier stance that 2 in SC can be derived from
obligatory movement to C0.
The alternative which seems correct to me, as I have argued in a number of
works, including Franks (1997, 1998) and Franks and King (in press), is that 2
clitics move to the highest functional head position in their extended projection,
whatever that happens to be. Here it is interesting to note a similar evolution in
analyses of the verb second (V2) phenomenon in Germanic; cf., e.g., Travis
(1991) or Zwart (1997), who argue that SVO order is best accommodated by
having the subject in SpecIP and the verb in I0, not C0. I will have much more
to say about the connection to V2 below. In any event, it is clear that 2 in SC
cannot always involve C since it persists in domains that are not CPs, such as
adjectival (22a) or gerundive (22b) phrases:
(22) a. suvremeni mu pisci
contemporary him. writers
b. dajući joj ružu, …
giving her. rose
On the basis of (22b), Caink (1997) also rejects the ‘2 = C0’ hypothesis,
proposing instead that clitics simply move to the highest head position that they

21. Similarly, they are not always adjoined to IP, as in (2).

 CLITICS AT THE INTERFACE 17

can. While this argument clearly applies to phrases that are most probably
smaller than full-fledged clauses, in the sense that they lack certain functional
material, the same is probably also true of complete sentences. It is now
frequently argued from diverse theoretical perspectives that only phrase structure
which is independently required is projected; cf., e.g., Bošković (1997: 37–39) for
references and discussion. A sentence has all and only those functional projec-
tions that it needs (because of its V). In a non-interrogative main clause, for
example, there is no reason to project a CP, since selection of C is not required
in the absence of [+wh] features to be checked. An immediate and obvious
consequence of this conclusion, however, is that not all 2 clitics can be in C0.
Instead, one must retreat to the position that they are in the highest functional
head in the given structure. There is currently, I believe, a consensus in the field
that this is the most promising way to salvage syntactic approaches to 2 clitic
placement. Of course, at least two serious problems still remain: Why do clitics
move up the tree in this way? and How should deviations from strict 2 be
handled? I offer my suggestions about these problems in Section 3.

2.4 2PL and V2 are intimately related

Another ‘emerging truth’ that should, I think, be essential to any successful

analysis of clitics is that moving to second position must be connected to the fact
that this is something verbs also do. In other words, 2 and V2 are part and
parcel of the same Wackernagel phenomenon. This is by no means a new
insight, it is just that recently a variety of credible mechanisms for implementing
it have come to the fore. One way of expressing the relation between verbs and
clausal clitics is provided by Anderson (1996), who assimilates V2 to 2, both
as ultimately morphological; cf. also note 5. It seems to me, however, that this
sort of approach is backwards. Verb movement is traditionally regarded as a
syntactic phenomenon and, as much of the literature including the papers in this
volume demonstrates, clitic movement should also be. For example, as pointed
out by Franks and Progovac (1994) and Progovac (1996), and as reiterated in
many places, including Section 4.1 of Bošković (this volume), clitic climbing in
SC is like other syntactic movement rules in respecting the dichotomy between
‘indicative-like’ complement clauses, as in (23), and ‘subjunctive-like’ ones, as
in (24). Thus, (24b) is much better than (23b).
(23) a. Milan kaže da ga vidi
Milan says that him. sees
‘Milan says that he can see him.’
18 STEVEN FRANKS

b. *Milan ga kaže da vidi

Milan him. says that sees
(24) a. Milan želi da ga vidi
Milan wishes that him. sees
‘Milan wishes to see him.’
b. ?Milan ga želi da vidi
Milan him. wishes that sees
‘Milan wishes to see him.’
Patently, there can no phonological explanation of domain extension in subjunc-
tive clauses, hence clitic placement, like verb movement, is subject to strict
syntactic locality conditions; see Franks (1998) and Franks and King (in press)
for arguments that clitic movement obeys the Head Movement Constraint
(HMC). We thus want 2 to be in some sense parasitic on V2, rather than the
other way around (or for both to derive from non-syntactic principles, as in the
morphological approach of Anderson (1993, 1995, 1996) or Legendre’s (1999)
OT version of it).
Another obvious reason why 2 clitic placement should be connected to
the syntax of verbs is that the other option for clitic placement in Slavic,
instantiated by Bg and Mac, is in a position adjacent to the verb. This suggests
that the difference between 2 and verb-adjacent clitics is relatively superficial,
and that they reflect two closely related strategies for clitic placement. Corrobo-
ration for this idea can be found in the fact that, in the history of Slavic, there is
clear migration from one option to the other; the same has been documented for
other language families, such as Romance, as noted by Franco (this volume). It
thus seems to me that the correct analysis of clitic placement will be able to
connect 2, V2, and verb-adjacency; see Progovac (in press) for further claims
to that effect.
It also seems inescapable to me, as articulated in Franks and Progovac
(1999), that the right approach to 2 will derive its effects in some way from
the V2 phenomenon. In the previous subsection I concluded that both verbs (in
V2 systems) and clitics (in 2 systems) move to the highest functional head in
the extended projection of V. Given this, it is simply more credible that the verb
should care about its extended projection than that the clitics should. On the one
hand, there is presumably good reason for V2: V2 is motivated by the need to
check features of V against those of functional projections associated with V.22

22. As Marcel den Dikken (personal communication) reminds me, there are also significant problems
with feature-checking accounts of V2, such that Chomsky (1995) rejects the syntactic approach to
 CLITICS AT THE INTERFACE 19

On the other hand, there is no obvious reason why the clitics should, since they
are an assortment of apparently disparate categories. In fact, what relates them
may well be the fact that they are all either generated in or move to functional
heads in the extended projection of V anyway. If so, not only can the verb be
expected to pass through positions the clitics occupy, such movement could not
easily be avoided, assuming standard HMC effects, whether formalised as
‘Shortest Move’, as in Chomsky (1993), ‘Minimal Links’, as in Chomsky (1995),
or some other way. This line of reasoning strongly indicates that we need to look
for a way to derive clitic placement from verb placement such that 2 is a
direct consequence of V2, and that all other systems be treated as in some way
deviations from these.

2.5 Clitic properties are cross-modular

By now it should be clear that clitics can display special properties in various
components. Some problems we have examined pertain to phonology, such as
the clitic third facts in (3)–(5), and some to syntax, such as the clitic climbing
facts in (23) and (24). There are also unquestionably morphological facts, such
as the idiosyncratic morphological manipulations of adjacent clitics often evident
in Romance. For example, in standard Spanish the sequence le lo becomes se lo
and in standard Italian si si becomes ci si. Combinations of third person dative +
accusative clitics similarly result in the dative clitic being replaced either by a
reflexive or locative clitic in various Italian dialects. Interestingly, as noted by
Bonet (1995: 607), such ‘opaque’ clitic forms ‘always coincide with clitics which
exist independently in the language.’ These patterns suggest post-syntactic
adjustments of morphological feature matrices in the spirit of Halle and Marantz
(1993). Clitic behaviour is thus the result of the complex interaction of (at least)
prosodic, syntactic and morphological requirements of clitics. The challenge is to
express those requirements and implement that interaction in a way that does not
invoke derivational globality.
One could, of course, simply reject the derivational paradigm altogether, as
in OT approaches to clitics such as Legendre (1999) or Grimshaw (1997a), but
these systems in my opinion merely restate the clitics’ properties and make no
attempt to explain recurring patterns or restrictions. Instead, it is necessary to
take as a point of departure the idea that clitics originate as syntactic entities.
The puzzle is how then to capture their non-syntactic effects. Derivational

V2. Here I advocate a system as in Zwart (1997), as discussed below.

20 STEVEN FRANKS

accounts of clitic phenomena tend to pass on to the (morphology and) phonology

clitic structures which require restructuring in these components. PI is the classic
example of this. However, it seems to me that here too there is a potential lack
of explanatory adequacy. Why do factors relevant to distinct components interact
as they do? What is needed is a model of how different components of grammar
‘talk’ to one another, the things they can talk about, and the kinds of compromis-
es they can or cannot entertain.

3. Clitics as Interface Phenomena

In this section I outline a programmatic account of a derivationally based

solution to the general problem of clitic placement. The account exploits the
insight that 2 is parasitic on V2 and relies on a specific filtering approach to
the interaction between syntax and phonology. It also makes explicit claims
about the nature of feature movement and pronunciation.

3.1 Verbs as hosts

Verbs are canonical hosts for special clitics. The leading idea I wish to take
advantage of here is that the verb can be the syntactic ‘host’ of the clitics even
when the clitics are not actually pronounced adjacent to the verb, as in the SC
2 system. In essence, we want the verb to move to second position and to
take the clitics along with it, but then the clitics are realised in that position
while the verb is realised lower down. At this point I review some related
considerations.
First, consider that there is a compelling reason for the verb to move up
through its extended projection: the formal features of V are checked off against
those of associated functional heads. It thus moves to the top of the projection in
order to license its formal features. Moreover, it is the features of V rather than
V itself which need to move. Various ways of instantiating this are imaginable:
features of both V and the associated functional heads could be fully specified,
and (i) the latter could ‘Attract’ the former, as in Chomsky (1995) or (ii) the
former could ‘Move’ out of Greed considerations, adapting Chomsky (1993); (iii)
features of V could be left with their values unspecified, as in Franks (1995) and
be attracted to their fully specified functional counterparts, or (iv) features of V
could be fully specified but move to fill in unspecified values for those features
in the functional heads, as in Zwart (1997). The point is that, under any of these
scenarios, there is feature movement from V to the top of its extended projection.
 CLITICS AT THE INTERFACE 21

In other words, the claim is that all languages are abstractly verb second. This is
a matter of UG, and a direct consequence of checking theory. Grammars differ,
however, in where the verb is actually spelled out. There are a number of
mechanisms which have been proposed in the literature for implementing this
idea, which I review in the next subsection.
Second, consider that the clitics are all morphosyntactic feature matrices
which can reasonably be situated in functional heads associated with the verb,
and through which the verb must therefore pass on its way up the tree. Verbal
auxiliary clitics in various languages are presumably introduced above VP in
nodes such as T0, AgrS0, Aux0, Mod0, Pred0, or Asp0, voice markers probably
head a VoiceP or ReflP, and complementiser elements are introduced in C0.
Pronominal clitics, on the other hand, are most naturally analysed as introduced
in the same VP-internal positions as their phrasal counterparts — at least when
they function as arguments, as in SC and Slvn. However, they still display the
morphosyntactic properties of functional heads. In the Slavic languages, except
for Bg and Mac, nominal phrases are maximally KPs rather than DPs, as they
are in Romance.23 Consequently, the pronominal clitics have the morphology
of pure case endings, and so I will treat them as K0 elements, as opposed to the
D0 clitics typically found in Romance. Now notice that the fact that they are
introduced inside VP does not obviate the need for them to move to a functional
head above VP. As arguments, they still must move to a case-checking position,
just as a full KP or DP would. Depending on the details of one’s theory of
phrase structure, these could be any one of various functional heads, including
AgrO, AgrIO, Asp, Trans, or v, against which case-checking occurs. Crucially,
they are all positions through which V must pass. Apparent differences between
a K0 such as SC accusative/genitive ga ‘him’ and the KP uspešnoga studenta
‘(the) successful student’ are that (i) the former is a head (which is exhaustively
a phrase), while the latter is necessarily a phrase, and (ii) the former must move
to its case checking position (and further) overtly, while the latter typically does
not. In Balkan languages such as Mac and Bg, as well as in Southern Cone and
other varieties of Spanish, according to Franco (this volume), the pronominal
clitics are not arguments. Rather they double full DPs (which may be pro) and
are thus generated as feature complexes directly in the higher functional head.
Regardless of how the details are fleshed out, I therefore think one can make a
convincing case for analysing clitics as morphosyntactic feature matrices

23. They can also be QPs, following the analysis in Franks (1995: Ch. 5) of East Slavic numerically
quantified phrases.
22 STEVEN FRANKS

occupying head positions which the V is expected to pass through on its way to
the top of its extended projection.

3.2 Formal features vs. semantic features

In light of these considerations, we now ask what exactly differentiates the

clitics from the verb, such that (i) the verb must pied-pipe the clitics as it moves
through positions they occupy and (ii) the clitics can be pronounced higher than
the verb, as well as what differentiates a pronominal clitic from a full KP/DP,
again such that the clitic is pronounced in a higher position.
The key to the puzzle lies in moving the verb overtly in some way that does
not require it to be pronounced in its target position. In fact, a number of
proposals exist in the recent literature for accomplishing just this. The first such
proposal, so far as I know, appeared in Groat and O’Neil (1996), and others are
embodied in Zwart (1997), Richards (1997), and Roberts (1998). They all hinge
on Richards’s insight that ‘strong’ is in some sense an instruction to PF to
pronounce an element in that position. For Richards, it is nothing more than that,
simply a diacritic. Others try to identify ‘strong’ with the pied-piping of addition-
al features, beyond those particular formal features (F-features) which are being
checked, that eventually cause the element to be pronounced. Groat and O’Neil
(1996: 124) postulate that ‘strong features may be checked only in a checking
relation with a node specified for phonological features’, so that the (more costly)
movement of phonological PF-features only takes place when strong formal
features are being checked. Thus, as required, F-features move overtly but pied-
piping of the PF-features prerequisite to pronunciation only occurs selectively.
While Groat and O’Neil literally equate locus of PF-features with copy
pronounced,24 others opt for a more mediated approach. Zwart (1997) distin-
guishes F-features from ‘Lexico-Categorial’ features. The former have purely
grammatical content whereas the latter add meaning. I will make the same kind
of distinction, but call semantic features ‘S-features’, putting aside the debate
over how best to treat categorial features. For Zwart, overt movement involves
both F-features and S-features, whereas movement traditionally analysed as
‘covert’ only involves F-features. He writes: ‘… the separation of F-movement

24. Whenever explicit, Groat and O’Neil refer to phonological feature ‘copying’ rather than
movement. This, however, raises the problem of having phonological features in multiple places,
which entails pronunciation in the highest position possible. While they do not address this issue, it
seems to me the solution is similar to what I will propose, some kind of constraint or principle that
forces PF realization of the higher copy when given a choice.
 CLITICS AT THE INTERFACE 23

from LC-movement makes it possible to revise the notions ‘strong’ and ‘weak’.
Instead of saying that functional heads are specified for strength, we can say that
F-features are strong or weak. When F-features are strong, they must be spelled
out in the morphosyntactic complex which they have been adjoined to. When
F-features are weak, they must be spelled out in the position of their trace, i.e.,
in the position of the lexical head. Only in the former case is movement of
LC-features triggered. Thus, the strong-weak distinction refers to movement of the
LC-features [my emphasis], not to movement of the F-features …’ Zwart
(1997: 188, footnote 37). This makes much more sense to me than literal
movement of PF-features, which contra Chomsky (1993, 1995) do not belong in
a syntactic representation. Instead, as in Franks (1995), I opt for a system of
‘two-stage lexical insertion’, whereby lexical items as bundles of F-features and
S-features are drawn from the lexicon and eventually merged to form a phrase
structure, with the syntax licensing F-features through checking (or assignment,
either to functional heads if items in the numeration have specified F-feature
values, or to the lexical items, if these are drawn from the lexicon with open
F-feature values). Subsequently, the morphology may manipulate some features
in a local fashion. Eventually, then, the mapping to PF must entail a second
access to the lexicon and replacement of each morphosyntactic feature matrix by
a corresponding phonological one. Zwart, who calls this general kind of approach
‘Postlexicalism’, makes a similar move: ‘Morphology takes a morphosyntactic X0
object a and replaces it with a form from the Lexicon that optimally matches the
features of a’ (Zwart 1997: 182). The Distributed Morphology system of Halle
and Marantz (1993) is one version of Postlexicalism and, whether or not all the
particular machinery they espouse is motivated, I concur that some such model
is required.
We are now faced with our original question of why clitics display special
syntactic behaviour. The answer is revealed in Roberts’ (1998) account of why
English auxiliaries raise to T0 whereas main verbs do not. He casts the problem
in terms of pied-piping. As in Chomsky (1995), movement to check one
F-feature always pied-pipes the entire bundle of F-features. However, unlike in
Chomsky’s system, where overt F-feature movement leads to category move-
ment, for Roberts only strong feature checking causes ‘generalised pied-piping’
from Move F to Move a. Thus, as in the other systems that divorce target
position from phonologically realised position, for Roberts (1998: 124) ‘Move F
is always and only overt’. Strong features in addition cause the entire category
to be pied-piped. If V-features in English are weak, then as expected lexical
verbs are pronounced in situ. Auxiliaries, however, only have F-features, so that
when ‘Move F moves all features of the element it moves … checking the weak
24 STEVEN FRANKS

feature of the V node causes the entire auxiliary to move’ (Roberts 1998: 119).
This insight is crucial and, as observed at the end of the article, ‘another obvious
place to look for the effects of ‘overt Move F’ is the area of clitics and cliticis-
ation’ (Roberts 1998: 124). The implication is that if clitics are taken to be pure
F-feature bundles, and in this respect differ from other DPs/KPs, then clitics
should be pronounced higher than full DPs/KPs when weak case features are
being checked. Not only can this idea explain the distribution of clitics vs. full
phrases, it also solves our problem of how the verb can be used to drive 2
clitic placement.

3.3 Getting clitics to 2PL

Following in the spirit of this work, I assume that F-features always move
overtly and that movement to check one F-feature necessarily pied-pipes the
entire bundle of F-features. Movement to check strong features pied-pipes
S-features as well, whereas movement to check weak features does not. I also
assume that PF-features are not present in the syntactic part of the derivation.
Finally, I assume that special clitics are purely grammatical, hence consist
exclusively of F-features.
How do these assumptions give rise to 2 clitics? The F-features of the
verb move overtly through every functional head position above VP. The verb
contains both F-features and S-features, schematically ‘[[F-features], [S-fea-
tures]]verb’, but since these are weak F-features only they move.25 They move
through successive functional heads occupied by F-feature bundles which corres-
pond to the various clitics, schematically ‘[F-features]clitic ‘. When the F-features
of the verb move up the tree through its extended projection, these incorporate
the [F-features]clitic in the head positions through which they pass; here I am
assuming that the clitic F-features must be pied-piped as part of the larger package
of features. This has the empirical consequence that the clitic features are forced
to be ‘free riders’ on the verb features as they continue up the tree in the checking
process. The result is that the complex [[F-features]verb + [F-features]]clitics is

25. There may be ‘short’ verb movement to check strong F-features, but at some point only weak
features are involved. The verb will then be pronounced in its highest strong position. In my system,
contra Groat and O’Neil (1996), once F-features are copied to a weak position (i.e. not pied-piping
S-features), no further strong feature checking is possible: the S-features cannot be copied from an
intermediate weak position, since they are not there.
 CLITICS AT THE INTERFACE 25

situated in the highest head position.26 In lower (weak) positions, on the other
hand, a complex with the verb’s semantic features is retained, schematically
[[[F-features], [S-features]]verb + [F-features]]clitics, as in (25):
(25) [[F-features]verb + [F-features]]clitics … [[[F-features], [S-features]]verb
+ [F-features]]clitics
Clitics are then pronounced in the target position, but the verb is not.
There are a few technical details of this system which need to be made
explicit. Most importantly, recall Chomsky’s (1993) ‘copy and delete’ theory of
category movement, whereby movement actually consists of two independent
operations, syntactic copying and PF deletion (of all but the highest copy).
Adapting this approach to Chomsky’s 1995 feature movement theory, we should
also view F-feature ‘movement’ as copying. And, as in Chomsky (1993), my
system also requires that the highest copy possible to pronounce be the one
pronounced. Since pronunciation works by replacing a morphosyntactic object
with a phonological one, and in order to find the right item in the lexicon access
to appropriate features is necessary, the verb can only be pronounced where both

26. While I subscribe to the traditional view that the Slavic special clitics form a syntactic cluster,
Bošković (this volume, in progress) presents a viable prosodically oriented alternative. Since, for an
account such as mine, which exploits V2, to work, the clitics must form a syntactic constituent, it is
worth reproducing here a new and compelling argument from Bošković (in progress) that SC clitics
remain in separate syntactic heads. Bošković notes that adverbs such as pravilno ‘correctly’ can either
have a subject-oriented (sentential) or manner (VP) reading, as in (i).
(i) Oni su pravilno odgovorili Mileni
they .3 correctly answered Milena
‘They did the right thing in answering Milena.’
‘They gave Milena a correct answer.’
The adverb pravilno can thus be adjoined above or below the clitic su. He further claims, however,
that only the manner reading is possible if there is a pronominal clitic, as in (ii):
(ii) Oni su joj pravilno odgovorili
they .3 her. correctly answered
*‘They did the right thing in answering her.’
‘They gave her a correct answer.’
If su and joj were in the same place, this effect would be inexplicable. If, on the other hand, su and
joj occupy separate head positions, but must be phonologically (or morphologically) adjacent, then
Bošković’s judgements follow. While this argument seems to me to be virtually unassailable, I should
at least point out that other SC speakers consulted do not share Bošković’s judgements, finding that
the subject-oriented reading is possible in both (i) and (ii). For Ljiljana Progovac (personal
communication), in both cases this reading requires the adverb to be emphasized and set off by
comma intonation. The potential significance of such examples is great enough to warrant closer
investigation of whether or not there is indeed a contrast.
26 STEVEN FRANKS

its F-features and S-features are.27 The failure to copy the verb’s S-features
thus forces the verb to be pronounced in the lower position in (25). Clitics, on
the other hand, are purely F-features, hence given a choice between higher or
lower position it is the higher position in (25) that is phonologically realised,
crucially assuming that the highest copy prevails (everything else being equal).
In this way, discontinuous pronunciation of the verb and the clitics is achieved,
despite the fact that a single copying operation is implicated. In sum, even
though weak feature checking means only F-features are copied, if a word is
exclusively F-features, then it can be pronounced in the higher position. That is
precisely what a clitic is, but not what a verb is, hence SC and Slvn 2 can be
understood as parasitic on abstract V2. The significant difference between this
and an overt V2 language, such as German, is simply that the V features being
checked are strong, so that S-features are pied-piped as well, allowing the verb
to be pronounced in the target position.28

3.4 The role of optimality theory

I have argued that, given a sentence such as (26a), or an embedded variant (26b),
the clitic and verb F-features are copied to the highest head position, but the verb
predstavlja ‘introduces’ S-features are not.
(26) a. Jovan mu ga svaki dan predstavlja
Jovan him. him. each day introduces
‘Every day Jovan introduces him to him.’
b. Znam da mu ga Jovan svaki dan predstavlja
know that him. him. Jovan each day introduces
‘I know that Jovan introduces him to him every day.’
Example (26a) has a structure roughly as in (27):29

27. Zwart (1997: 190) also addresses this issue, noting that ‘without LC-features, a head adjunction
structure cannot be interpreted by Morphology’. Any workable solution to the problem of determining
pronunciation site must therefore have all features required for lexical access in pronounced position.
28. There are of course other differences hinging on the fact that whereas clitics have prosodic
dependencies which also must be met, verbs do not. The striking parallelisms between SC 2 and
German V2 are highlighted in Franks and Progovac (1999).
29. Many extraneous decisions about phrase structure have been made in (27), such as the labels and
relative heights of functional categories. These details are irrelevant to the central claims made in this
introduction, and are provided merely for the sake of explicitness of representation.
 CLITICS AT THE INTERFACE 27

(27) a. [AgrSP Jovan predstavlja

————— mu ga [AgrIOP predstavlja
————— mu —— ga
— [vP
svaki dan [vP predstavlja ga
— [AgrOP predstavlja
————— ga — [VP mu
——
————— ga
predstavlja —]]]]]]
An element which is ‘struck through’ is not pronounced. The clitics are pro-
nounced in the highest position, head of AgrS (= IP), and the verb is pronounced
lower down, since none of the higher copies actually have the relevant S-feat-
ures. In (26b) the complementiser da appears, with the structure in (27b):
(27) b. … [CP da predstavlja
————— mu ga [AgrSP Jovan predstavlja
————— mu —— ga
—
[AgrIOP predstavlja
————— mu —— ga
— [vP svaki dan [vP predstavlja ga
— [AgrOP
————— ga
predstavlja — [VP mu—— predstavlja
————— ga —]]]]]]]
I suspect that da is inserted in C0 in order to support the clitics. Since this is a
matter of prosody, da-support in fact takes place in the mapping from morpholo-
gy to PF.30 Crucially, da-support is not available in (26a) since, as argued
above in Section 2.3, main clause declaratives are not CPs. Main clause interrog-
atives are, however, CPs, with li as we have seen generated in C0. As expected,
here too da-support is possible, as can be seen in (28):
(28) Da li mu ga Jovan svaki dan predstavlja?
that  him. him. Jovan each day introduces
‘Does Jovan introduce him to him every day?’
How should we formalise the fact that the highest copy is pronounced? In
particular, why aren’t lower copies of the clitics pronounced, even though they
are eligible for replacement by PF-features? It seems to me that, as in
Chomsky’s original copy and delete model, this is a not a matter of the syntax
per se. Instead, I would argue that which copy to pronounce is determined in the
mapping to the phonology.31 I further propose that this mapping is mediated by
Optimality Theoretic (OT) principles; in general, I believe that the proper place
of OT in the grammar is at the interfaces between components. Information that
is irrelevant for one component is left vague or unspecified, with the choice

30. PF lexical insertion is possible for semantically vacuous elements, since they have no LF
features that would disrupt PF. Similarly, as argued in a series of papers by Bošković on wh-move-
ment, such as Bošković (1998), semantically active features such as [+wh] can be merged in LF if
they are phonologically null.
31. Although technically the relevant mapping is from a level of Morphological Structure to PF, in
most of the discussion I ignore this intermediate stage since there are few ramifications for Slavic
clitic data and since morphological representations are minimally modified syntactic phrase structure
representations.
28 STEVEN FRANKS

among open options determined only at the interface, in the mapping from one
component to the next.32 Thus, here I would simply assume a constraint P-
 H, that states that the higher of identical copies is the one
replaced by PF-features, everything else being equal:
(29) P H: Lower identical copies are silent.
In other words, the decision about which copies to strike through in (27) is made
in the mapping to PF, since the syntax does not care that there are multiple
copies. This system retains the insights both of traditional derivational models of
syntax as well as those of recent representational work in OT. The syntax is still
generative, but OT considerations reign at the interface to mediate selection of
what actually serves as the input to PF. In short, minimalist syntax provides the
candidates (‘’, in OT terms), but OT principles select the winner (‘’, in
OT terms). Syntax ‘composes’, and (the mapping to) phonology ‘disposes’.
With this in mind, let us return to the ‘delayed clitic placement’ problem
discussed in Section 2.1 in connection with SC examples such as (5), repro-
duced below:
(5) #Ja#, #tvoja mama#, #obećala sam ti igračku#
#I #your mother #promised .1 you. toy
‘I, your mother, promised you a toy.’
This example involves pronunciation of clitics lower than expected on the basis
of the syntax alone. Since the clitics immediately follow an intonational break,
they appear in a position lower than 2. The role of prosodic factors is clear in
(5). In earlier models, they suggested prosodic ‘movement’. However, we now
have an appropriate mechanism to express the fact that prosody can be expected
to police the mapping from syntax to PF, yet also retains the standard insights of
syntactic clitic placement.
Let us see how delayed clitic placement might work in the ‘OT at the
interface’ approach. The reason clitics are not realised in the highest syntactically

32. Other credible examples of this are (i) the interface between the lexicon and phonology, where
the lexicon can tolerate underspecification but the phonology requires feature values to be filled in,
and (ii) the interface between syntax and LF, where the syntax can tolerate ambiguity, e.g. about
which copy of a reflexive is interpreted or the scope of a quantifier, whereas LF cannot. One
‘minimalist’ principle for regulating the mapping to LF that has a distinctly OT flavor is the
‘Preference Principle’ of Chomsky (1993). The Preference Principle minimizes the restriction in the
operator position to the extent possible, so that the desire to minimize can be overriden by other
interpretive exigencies. While there is much work on the former kind of problem within OT, it seems
to me the latter kind of problem can also be profitably approached from the perspective on OT
advanced in this chapter.
 CLITICS AT THE INTERFACE 29

viable position is because pronunciation in second position would violate

prosodic requirements. I assume a very highly ranked constraint P
S which disallows prosodically stranded elements:
(30) P S: Everything pronounced must be prosodically
parsed.
Thus, (5) can be represented as in (31):
(31) #Ja#, #tvoja mama#, #sam
——— ti obećala sam ti igračku#
A lower copy of the clitics is pronounced, in violation of P H,
in order to satisfy P S. Note that this is unnecessary in Slvn (6),
since the clitics in that language are prosodically neutral, hence are able to meet
both constraints.

4. Some problems and extensions

We have seen that the central theoretical dilemma raised by Slavic clitics is
whether they are positioned syntactically, by the normal methods available in
most theories of syntax, or prosodically, by some special repair operation. I have
proposed that a general strategy for expressing the possibility of dependence on
prosody is to let the syntax leave open the matter of which copy is pronounced,
and to resolve this question by selecting the prosodically optimal candidate at the
interface with PF. In this section, I briefly discuss some extensions to other
phenomena, mostly from Bg and Mac, and associated problems that arise.

4.1 Verb-adjacent Slavic systems

In taking 2 as contingent on V2, an immediate question that arises is how this
approach can accommodate verb-adjacent clitic systems. In Bg and Mac, most
clitics immediately precede the verb, as shown in (32).
(32) a. Vera včera ti go dade
Vera yesterday you. it. gave
‘Vera gave it to you yesterday.’
b. Vera ti go dade včera
In Mac, this is true even if nothing else would be to the left of the clitics, as in
(33a). In Bg, however, we find that otherwise initial clitics must follow the verb,
as in (33b).
30 STEVEN FRANKS

(33) a. Ti go dade Vera včera [MacØ/Bg*]

b. Dade ti go Vera včera [BgØ/Mac*]
The so-called ‘Tobler-Mussafia’ effect arises with Bg clitics because, as in SC,
they are enclitic; in Mac, on the other hand, they are prosodically neutral, as in
Slvn. And, as with SC vs. Slvn, I would claim that the relevant differences
between the two languages are purely prosodic.
These facts require considerable elaboration within the framework put
forward in this introduction. First, I interpret them to mean that the clitics prefer
to be syntactically preverbal if possible. Thus, given the lexical difference in
prosodic properties of the clitics, example (33a) does not violate P
S in Mac but does in Bg. The problem here is how to express the idea
‘preverbal if possible’ without derivational globality. Within the system outlined
above, the obvious way to avoid look ahead in dealing with the Tobler-Mussafia
effect is to leave determination of linear order to the mapping to PF. In other
words, just like selection of which copy to pronounce, linearisation is part of the
PF mapping process, such that the verb will precede or follow the clitics based
on OT considerations. Let us therefore abstract linear order out of the syntactic
head-adjunction of the verb to the clitics (more precisely, of one matrix of
features to another) and see what effect this has on our model. I would like to
suggest the following as a guiding principle:
(34) L = H (LH): Features of the head are morphologically
realised to the left.33
This constraint means that, everything else being equal, a head (more precisely,
its features) moved from below will eventually be realised to the right of the
element that was higher than it, so that linearisation comes out as right-adjunc-
tion. Like P H, LH is ranked lower than P S,
so that candidate (33b) wins over (33a) in Bg, but not in Mac (since LH can
also be satisfied in Mac).
Notice that LH has a number of interesting consequences deserving of
further exploration. For one thing, it implies that the verb moves to the clitics in

33. In earlier work, such as Franks (1998), I describe this idea as stipulating that ‘head-adjunction is
to the right’. As Marcel den Dikken (personal communication) rightly points out, ‘adjunction’ is the
wrong term here, since I am explicitly denying that syntactic adjunction, treated as copying of features
from one head to another, is directional. Linearisation, which I will argue applies cyclically, is an
interface phenomenon, in the spirit of Chomsky (1995) and Epstein et al. (1998), rather than Kayne
(1994). LH states that the features of the head of the phrase prefer to be instantiated on the morpho-
logical item to the left, with the result that the raised head appears to be realized as ‘right-adjunction’.
 CLITICS AT THE INTERFACE 31

Bg and Mac, rather than the other way around. This, however, follows from the
fairly standard conclusion that the clitics are indeed introduced in functional
heads above VP in these languages. For SC and Slvn, on the other hand, I
contend that the verb must move first to the case-checking positions of the
clitics, before the K0 clitics, generated within VP as its arguments, themselves
move. One reason for this is simply that the verb’s F-features may need to
combine with those of Agr to impart specific case-checking requirements, as
when quirky case is assigned. Consequently, LH has the effect of establishing
‘verb + clitics’ as the neutral order in Slvn and SC, exactly the opposite of what
happens in Mac and Bg.34
Another conceptual problem raised by LH is that it cannot be applied to
completed tree structures, otherwise the derivational history that would reveal
which element had moved first would be obscured (without special appeal to
heavily enriched enchains). The solution, it seems to me, is to carry out the
mapping from syntax to PF cyclically. I take this as a very important conse-
quence of the model, one that is in conformity with a number of recent models
such as Epstein et al. (1998), Chomsky (1998), Castillo, Drury and Grohmann
(1999), and Uriagereka (in press). An empirical argument for cyclic application
of LH will be given in Subsection 4.4.
The most significant implication of the 2 analysis for verb-adjacent
systems, however, is that we are now forced to the conclusion that Bg and Mac
are in fact overt V2 languages. That is, since the clitics are pronounced on the
verb, the verb as well as the clitics must be at the top of its extended projection.
Mac and Bg are therefore like German in that strong F-features are being
checked, so that S-features are pied-piped as well. While this conclusion runs
contrary to most traditional analyses, including my own earlier claims in e.g.
Franks (1997, 1998) about the difference between SC/Slvn and Bg/Mac, there
are I believe advantages to treating Mac and Bg as surface V2. Of course, this
approach requires massive topicalisation in order to accommodate examples in
which the verb appears lower down. One clear example of the need for topical-
isation is given in (35):
(35) Ivan piše li pismoto?
Ivan writes  letter.
‘Is Ivan writing the letter?’

34. One complication with this approach to word order is that the auxiliary is presumably generated
above VP in all systems. Possibly, a solution in terms of when Aux moves to T0 and AgrS0 might
be successful.
32 STEVEN FRANKS

Since li is in C0, we have good reason to suppose that the subject Ivan has been
adjoined to CP.35 See also Krapova and Petrov (1999) for the claim that mate-
rial preceding the ‘clitics + V’ complex in Bg has been topicalised.

4.2 Atypical Macedonian hosts

In her contribution to the present volume, Tomić discusses some curious Mac
facts which reveal that the behaviour of the clitics depends on properties of the
syntactic host. Thus, although in principle these clitics are prosodically neutral,
hence as shown in (33) can precede the finite verb even when initial, she
observes the following paradigms:36
(36) a. Si mi go dal
.2 me. it. gave
‘You have (allegedly) given it to me.’
b. *Dal si mi go
(37) a. ?Mu e rečeno da bide točen povek’e pati
him. .3 told to be punctual more times
‘He was told to be punctual more than once.’
b. Rečeno mu e da bide točen povek’e pati
c. Na Petreta mu e povek’e pati rečeno da bide
to Peter him. .3 more times told to be
točen
punctual
‘Peter was told to be punctual more than once.’
(38) a. ??Si mu mil
.2 him. dear
‘He likes you.’
b. Mil si mu
c. Ti si mu mil
you .2 him. dear
‘He likes you.’

35. While this also happens in SC, as in example (4) in the text, discussed by Progovac (this
volume), it does not force an intonational break in Bg and Mac as it does in SC.
36. For the sake of consistency, I gloss forms of sum ‘be’ as ‘Aux’ even when they have a copular
function. Of course, the likelihood of verb-adjacency directly corresponds to the extent to which sum
is serving as an auxiliary, rather than a copula.
 CLITICS AT THE INTERFACE 33

(39) a. *Si mu tatko

.2 him. father
‘You are his father.’
b. Tatko si mu.
c. Ti si mu tatko
you .2 him. father
‘You are his father.’
What is to be made of this complex array of data provided by Tomić?
In earlier work, such as Tomić (1997), she had argued that lexical properties
of the clitics actually depended on their hosts. Quite reasonably, in the current
paper Tomić no longer thinks that a prosodic deficiency of a lexical item (i.e.
proclitic or enclitic) can vary according to stipulated syntactic criteria. Instead,
we want to maintain that these Mac clitics are always prosodically neutral, in the
sense that they can attach to a prosodic word either to their left or their right.
This is simply a lexical fact. In this volume, Tomić argues instead that the
variation lies in whether 2 or verb-adjacent behaviour is being displayed. I
believe that this new approach is probably on the right track, although it raises
the important question of why in a verb-adjacent system proclisis is preferred,
whereas in a 2 system enclisis is preferred. Since technically these Mac clitics
are prosodically neutral, the data in (36)–(39) strongly suggest to me that
considerable mileage may be achieved by exploiting an OT model: one order or
the other wins, although both are in principle possible.
Briefly, here is a way to think of Tomić’s facts. When the ‘clitics + host’
order obtains, the host raises. This in turn implies that S-features have been pied-
piped, hence strong F-features are being checked, which means that LH,
P H and P S can all be satisfied. Surveying the
Mac data, we can conclude that finite verbs check strong features, so that (33a)
beats (33b). Similarly, the l-participle also checks strong features, with the result
that (36a) beats (36b); note, as demonstrated by (20b) above, that this is not true
of SC.37 Next, the judgements in (37) indicate that the -n/-t participle checks
features which are only optionally strong. Strong features give rise to (37a),
weak produce (37b, c). A predicate adjective marginally checks strong features,

37. While Bošković (this volume) shows that SC l-participles never move to C0, he adduces further
facts in e.g. Bošković (1997) that indicate that the participle cannot even raise out of Aux (for
example, to T or AgrS), Aux being the head where verbal auxiliaries such as sam ‘am’ are
introduced. Thus, given my conclusion in note 25 that there is no movement from weak to strong
positions, raising of the participle to C0 in SC is blocked because S-features are unavailable for pied-
piping, even though finite verbs do raise, as in (20a).
34 STEVEN FRANKS

to derive (38a). Finally, the facts in (39) reveal that, as expected, a predicate
noun never moves to check features at all, so that (39a) is strongly ungrammati-
cal and some other phrase must move to the specifier of the highest functional
category, headed by the clitics. This approach nonetheless raises an interesting
residual problem: Since these Mac clitics are prosodically neutral, why is
anything ungrammatical? In particular, where does (37b) come from? In (37c) na
Petreta ‘to Peter’ has moved to the specifier of the phrase headed by the clitics
mu e, but in (37b) the S-features of rečeno ‘told’ fail to raise but the result is
nonetheless ‘rečeno + clitics’. Thus, following claims in Tomić (this volume),
the paradigm implies re-emergence of a 2 system, although there are some
divergences from strict 2, as in the following variant of (37), verified by Olga
Tomić (personal communication):
(40) Včera na Petreta mu e povek’e pati rečeno…
yesterday to Peter him. .3 more times told
‘Yesterday Peter was told more than once …’
It is thus unclear what is going on in (37b), although the solution is presumably
syntactic, not phonological, since there is nothing prosodically wrong with
starting with a clitic here. This conclusion is even more inevitable when (39a) is
taken into consideration. Apparently, if nothing supports the clitics syntactically,
then the result is ill-formed. One idea about how to handle this problem is that
there exists a morphological locality constraint, which we can call L, that
requires that special clitics must adjoin to something in their own maximal
projection.
(41) L: special clitics require morphological support within their
own maximal projections.38
L is generally met, i.e. whenever V raises or a phrase moves to the speci-
fier of the phrase headed by the clitics. A phrase indeed moves in (39b, c), but
L is violated in (39a) since tatko ‘father’ is in situ, even though the sen-
tence is prosodically well-formed. I suspect that this might be exploited to handle
(37b) as well, if we assume that the ‘rečeno + clitics’ order wins because in
some sense which needs to be made precise it is most compatible with L.

38. Marcel den Dikken (personal communication) raises the question of whether we want morpholo-
gy to make reference to syntactic structure. I suggest that, since the L constraint polices the
mapping from syntax to morphology, it is indeed expected to have access to phrase structure
information. LH is similar, hence crucially takes into account whether or not a set of features serves
as head of its phrase.
 CLITICS AT THE INTERFACE 35

Intuitively, the same approach can be extended to SC and Slvn ‘verb + clitic’
orders. Note that some explanation outside of phonology is going to need to be
invoked, since Slvn behaves just like SC in this regard, although it has proso-
dically neutral clitics. Consequently, the initial ‘V + clitics’ order in that
language cannot be exclusively prosodically driven: prosodic factors distinguish
SC and Bg from Slvn and Mac, but syntactic factors distinguish SC and Slvn
from Bg and Mac.

4.3 The ‘last resort’ pied-piping puzzle

These facts draw attention to a potential problem with the model I have devel-
oped for pied-piping of S-features. We have seen that although it is optimal to
pronounce the highest copy of the clitics, a lower copy may sometimes need to
be pronounced in order to meet prosodic requirements. The same might be
expected for verbs, so that (in SC and Slvn at least), while it is optimal to
pronounce a lower copy of the verb, a higher copy may sometimes need to be
pronounced in order to meet prosodic or morphological requirements. However,
we have explicitly ruled this out, since pronunciation of a higher copy would
crucially involve identification of a particular lexical item, which is impossible
if S-features have not been pied-piped. Whether the system needs to be adjusted
to allow ‘last resort’ pied-piping of S-features is empirically unclear, as is the
best mechanism for accomplishing this if it turns out to be necessary.
Recall the SC example (28):
(28) Da li mu ga Jovan svaki dan predstavlja?
that  him. him. Jovan each day introduces
‘Does Jovan introduce him to him every day?’
As an alternative to da-support, the following is also possible; cf. also (20a):
(42) Predstavlja li mu ga svaki dan?
introduces  him. him. each day
‘Does she/he introduce him to him every day?’
This shows clearly that a finite verb can move to C0 when C is [+wh]. Presum-
ably, this is an instance of AgrS/T-to-C raising; since the l-participle never gets
as high as AgrS/T in SC, it also cannot raise to C. Since, however, the finite
verb can raise, we might assume an optional strong feature in C0. Pied-piping of
S-features and pronunciation of the verb in C is thus not a technical problem.
Whether the same holds for a declarative example such as (43) is not as clear.
36 STEVEN FRANKS

(43) a. Predstavlja mu ga svaki dan

introduces him. him. each day
‘She/he introduces him to him every day.’
Is predstavlja being pronounced higher than expected here or not? If it is, then
we will need a mechanism for getting S-features to pied-pipe even when not
syntactically motivated. While technical fixes exist, they are unwieldy. Briefly,
one would need to pied-pipe S-features always, but then ignore them unless they
are in a strong position. ‘Strong’ then reverts to its diacritic sense ‘pronounce
me’, as in Richards (1997). Pronouncing a higher copy thus amounts to changing
a weak feature to a strong one, which can perhaps be taken as a violation of
‘faithfulness’.39 A much simpler alternative therefore might be to conclude that
in (43) we are really just pronouncing a lower copy of the ‘V + clitics’, as in
(43b), so that there is never any need to pied-pipe S-features when attracting
weak F-features.
(43) b. [mu
—— ga
— [predstavlja mu ga [… svaki dan…]]]
Interestingly, this leads us back to the puzzle of the previous section, of why the
‘verb + clitics’ order obtains. Once again, although prosody may seem to be the
culprit for SC, the facts that this order is also most natural in Slvn, as well as
that it reappears in Mac (37b), suggest that more may be involved. Possibly, here
we find a morphological constraint such as L at work, which has the effect
of ranking ‘host + clitics’ higher than ‘clitics + host’.40

39. I am using this notion somewhat differently than in standard OT. The idea here is that each
component provides (presumably generates) a single output, and the list of candidates is simply every
variant of that output, with unspecified information (such as which copy to pronounce, direction of
head-adjunction) filled in. I regard any divergence from these candidates, i.e. new candidates derived
by manipulating the ones with which the input component provided the interface, as a violation of
faithfulness. In a sense, we can take F as a literal constraint that says ‘pick one of the
candidates provided by the input component’. In this instance, the value of a feature (on some
F-feature!) is changed from weak to strong in the mapping from syntax to morphology. Note that PI,
as conceptualized in Section 2.2, involves making a comparable adjustment in the mapping from
morphology to PF.
40. Note that, as shown by Golden and Sheppard (this volume), initial clitics in Slvn need not be
adjacent to the verb. They note that ‘traditional Slovene grammar assumes that sentence-initial clitics
result from deleting the material preceding the clitics’. I believe that this is correct, and that the
deletion is a PF process, one which follows satisfaction of the morphological constraint L.
 CLITICS AT THE INTERFACE 37

4.4 An argument for cyclicity from Bg

In this section I review some curious facts from Bg that suggest that, however all
these technical problems of determining directionality are resolved, the solution
will most likely involve cyclic mapping to PF.
First, note that even i ‘and’ can support the clitics in Bg; compare (44) with
the ungrammatical (33a), which surfaces as (33b) instead:
(44) I ti go dade Vera včera
and you. it. gave Vera yesterday
‘And Vera gave it to you yesterday.’
(33) a. *Ti go dade Vera včera
b. Dade ti go Vera včera
Next, recall from Section 2.2 that the Yes-No clitic li is positioned by Prosodic
Inversion, which I now take to be a minimal violation of F (in the
sense of note 39) in the mapping from the morphology to PF, necessitated by the
need for li to satisfy the higher ranked ‘well-formedness’ constraint P
S. The result was an example like (10a), repeated below, which arose as
indicated in (11a):
(10) a. DAde li ti go?
gave  you. it.
‘Did she/he give it to you?’
(11) a. li [[dade]w ti go]w → [[DAde]w li ti go]w
We now understand (11) not as a PF operation, but rather as the creation of a
new candidate that minimally differs from the one which is provided to the
interface by the morphology.
Given all this, consider the following Bg sentence:
(45) I dade li ti go?
and gave  you. it.
‘And did she/he give it to you?’
Strikingly, it resembles (10a) rather than (44) in that the order which surfaces is
‘verb + clitics’ rather than ‘clitics + verb’. Apparently, the order dade ti go in
(45) is determined without consideration of the fact that i will ultimately be
available as a prosodic host for ti go, even though this fact does play a crucial
role in (44). Assuming a structure whereby li heads CP and i heads a higher
38 STEVEN FRANKS

(Boolean) phrase,41 the correct conclusion it seems to me is that linearisation

only makes use of very local information. In particular, I propose that the
interaction of li with linearisation shows that linearisation applies cyclically, with
linearising a lower syntactic concatenation required when the element defining
the next phrase up is merged. Thus, both for Bg (10) and (45), when ti go is
combined with dade, there are two candidates to be compared:
(46) a. li [ti go dade] or b. li [dade ti go]
The linearisation in (46b) wins, since enclitic li does not provide valid support
for ti go; it is thus the same as when li is not present, as in (33b). In (44), on the
other hand, (47a) wins, since i can support the enclitics ti go.
(47) a. i [ti go dade] or b. i [dade ti go]
These facts clearly reveal the effect of cyclic linearisation: the highest ranked
candidate is selected when the next element up is merged, so that initial i is
counted in (44) but must be ignored in (45), since it is protected from consider-
ation by the presence of li. The CP headed by li in (45)/(46) is, in the sense of
Chomsky (1998), a ‘phase’; (44)/(47) shows that IP is not.

5. Summary and Conclusions

In this introduction, I have mostly surveyed problems associated with South

Slavic special clitic placement. The overall approach I have advocated entails a
number of assumptions about the ways in which syntax interacts with morpholo-
gy and phonology, as well as a variety of specific solutions that required
considerable fleshing out. The analysis involves treating clitics as functional
heads, so that they consist exclusively of formal features, and then having those
features be pied-piped when F-features of the verb raise for checking purposes.
Only checking of strong F-features requires pied-piping of semantic features,
however, so that under 2 clitic placement the verb is pronounced lower, where
its S-features are.
In order to accommodate discrepancies between target and pronounced
position, as well as complexities of direction of adjunction, it was suggested that
OT principles mediate the mapping between components, which are themselves
generative. While the ideas expressed here are meant to be programmatic, I

41. For the notion of a ‘Boolean Phrase’ see Munn (1993) or Kayne (1994), among others.
 CLITICS AT THE INTERFACE 39

believe that the general structure of the model has much to recommend it. Within
the OT literature there is already some work that supports the division of labour
between OT and rule based systems which I propose. In his examination of the
extent to which syntax has an ‘OT character’, Pesetsky (1997) drew attention to
the fact that Optimality Theoretic ways of understanding syntactic problems are
most appropriate when those problems ‘lie at the boundary between syntax and
phonology’. My interface approach can be taken as a proposal to formalise
Pesetsky’s conclusions. Elsewhere, Golston (1995) argued that ‘syntax outranks
phonology’, in the sense that the syntactic component (which he takes to be a
system of constraints, operating as a block) outranks the phonological component
(also regarded as a block of constraints).42 His basic conclusions are completely
in accord with the approach to clitics envisioned here, since they confirm that
‘prosody chooses between structures which are equally well-formed syntactic-
ally.’ Tranel (1998) makes similar arguments that ‘syntax and phonology can
remain as separate modules, with syntax outranking phonology.’ Harford and
Demuth (in press), however, put forward a putative counterexample, claiming
that in various Bantu languages choice of relativisation strategy crucially depends
on phonological factors. Here again, however, it is the presence of a clitic which
forces I-to-C movement of the verb. They point out that the difference between
Sesotho (48a) and Chishona (48b) is simply that the verb must raise to the
relative morpheme in the latter:43
(48) a. dikobo tseo basadi ba- -di- -rekileng kajeno
blankets  women SM OM bought today
‘the blankets which the women bought today’
b. mbatya dza- -v- -aka- -sona vakadzi
blankets  SM TAM sew women
‘the clothes which the women sewed’
The verbal form vakasona raises to dza in (48b), but badirekileng does not raise
to tseo in (48a). Chishona dza- must be a clitic because its pronunciation in isolation
would violate M, a constraint against monosyllabic words (which is irrelevant
for bisyllabic Sesotho tseo). Raising the verb where prosodically necessary, in
violation of Grimshaw’s (1997b) constraint S that disprefers traces (hence
movement), thus implies that the syntactic constraint S is ranked lower.

42. Golston, however, also argues that prosodic constraints can force morphological ill-formedness,
implying that phonology outranks morphology.
43. ‘SM’ and ‘OM’ are subject and object markers, respectively, and ‘TAM’ is a Tense/Aspect/
Modality marker.
40 STEVEN FRANKS

While this is a persuasive argument, it does not hold in the current system,
which cannot even countenance a constraint like S. First of all, since there is
no OT syntax, the issue of how a ‘syntactic’ constraint such as S is ranked
is nonsensical. Secondly, in both Sesotho and Chishona, the F-features of the
verb are copied up to C0 overtly.44 Thus, the issue boils down to which copy
is pronounced, a decision that is made at the interface with PF. Harford and
Demuth’s data, just like with the Slavic clitic facts we have discussed, indicate
that P S is ranked very high, high enough to outrank whatever
normally forces the copy of V in I0 to be the one phonologically realised.45
Notice, now, that the Bantu situation is exactly comparable to the Slavic treated
in Section 4.3 under the rubric of ‘last resort pied-piping’. However, here we are
much more clearly forced to allow for realisation of a higher copy of the verb
even when the F-features that drive the raising are normally weak. This would
apply also in SC (43), and to get both (28) and (42), presumably violation of the
two ‘faithfulness’ constraints (cf. note 39), the one against da-support and the
other against pronouncing weak features, is equally onerous. While how this all
is best technically accomplished remains an open matter, I think Harford and
Demuth’s facts do not provide an argument that phonological constraints can
outrank syntactic ones, but rather serve as support for my general model in
which OT-like considerations (based on the input component, for which they are
relevant) discriminate among unresolved options (left open by the output compo-
nent, for which they are irrelevant).
While I have in this introduction sketched out a general framework for concept-
ualising clitics, there are countless details which remain as very hard problems of
practical analysis. Let me conclude with some general observations about where I
believe future work ought to take us. Most importantly, I still see accounting for
variation in clitic systems as a central puzzle. Even taking clitics to be functional
categories, the number of options that exist in their behaviour is complex. While
I believe these options derive from constraint interactions at various interfaces,
delimiting just what can occur is a formidable task. Even when considering a
relatively narrow phenomenon, such as clitic doubling, we encounter diverse
approaches. This is amply embodied in the three contributions to this volume
which concentrate on clitic doubling, those by Franco, Kallulli, and Schick, as
well as in the papers by Alexiadou and Anagnostopoulou, Dimitrova-Vulchanova,

44. Thus, the only way to understand Grimshaw’s S or Chomsky’s Procrastinate is that
F-features are weak in the default case.
45. The morphological constraint L also could be at work here, since this is violated in (48b)
unless the verb is in C.
 CLITICS AT THE INTERFACE 41

and Tomić, which treat doubling in less detail. Each to some extent deals with
its problematics in relative isolation. Alexiadou and Anagnostopoulou, for
example, thus do not offer any connection between their account of Greek and
conceptually similar (although empirically very different) Balkan Slavic prob-
lems. Franco’s paper particularly highlights the fact of variation, and the extreme
likelihood that different grammars will require different analyses of clitic
doubling. If so, UG must provide some way for the language learner to converge
on the appropriate analysis for her grammar, and a fundamental next step is to
come up with a theoretically sound typology of clitic systems.46
There are many other comparable matters of variation, some of which I
have touched upon in this introduction: Why is some bundle of features a clitic
in one grammar and not in another? (e.g., the difference between SC clitic ga
and English him, or SC tonic form njega for that matter); What determines
whether clitics are arguments or agreement markers? (e.g., Spanish vs. Catalan);
Why do indirect and direct object clitics sometimes behave differently? (e.g.,
with respect to clitic doubling); Why do reflexive and pronominal clitics
sometimes behave differently? (e.g., as in Czech or Slovene);47 What is the
difference between a V2 and verb-adjacent clitic system? (e.g., is it simply
where one pronounces the verb); What causes an element to move as a head or
as a phrase? (e.g., the role of F-feature copying);48 What is the relationship
between particular clitics and hierarchical phrase structure (e.g., how much does
the morphology reflect the syntax)?49 While the list of open questions about

46. This has been an obvious goal for some time, as evidenced by the seminal work of Klavans
(1982) and Halpern (1995). What I am proposing is a renewed version of this task, within more
current minimalist and/or OT models.
47. As Golden and Sheppard (this volume) note for Slvn, regardless of the origin of ‘reflexive’ se
(in diverse functions), it appears initial in the pronominal clitic string: ‘irrespective of its morphosyn-
tactic category and function, the morpheme se has its fixed and unique position in the clitic cluster.’
48. Within the system I have outlined, where all syntactic movement is feature movement, this
distinction should boil down to whether the features are copied from an XP (hence compositional)
or an X0.
49. Like the papers in this volume, I have not been explicit about the basic issue of the relative
ordering of the clitics. It seems to me that the fact that the same patterns are repeated in diverse
languages, with a  marker first, verbal markers such as subject agreement/tense/mood before
pronominal clitics, reflexive clitic peripheral in the pronominal cluster, and dative before accusative,
strongly indicates that general principles rather than an idiosyncratic morphological template are at
work. (The flip side of this is that there is good evidence that deviations from these patterns are
idiosyncratic.) Whether these are syntactic principles, as claimed for example in Franks and Progovac
(1994), Progovac (1996), or Tomić (this volume), or can be handled using interface linearisation
requirements, as I have argued here, is still unclear, however. One interesting fact that needs to be
42 STEVEN FRANKS

clitics is virtually endless, one thing is certain: the questions raised in this
volume serve to define a research program for innumerable future studies
devoted to the description and analysis of clitics.50

References

Alexiadou, A. and E. Anagnostopoulou. This volume. ‘Asymmetries in the distribution of

clitics: the case of Greek restrictive relatives.’
Anderson, S. 1993. ‘Wackernagel’s revenge: Clitics, morphology, and the syntax of
second position.’ Language 69: 68–98.
Anderson, S. 1995. ‘Rules and constraints in describing the morphology of phrases.’ In:
Dainora, A., R. Hemphill, B. Luka, B. Need and S. Pargman (eds.) Papers from
Chicago Linguistic Society 31: Parasession on Clitics, 15–31.
Anderson, S. 1996. ‘How to put clitics in their place or why the best account of second-
position phenomena may be something like the optimal one.’ The Linguistic Review
13: 165–191.
Bonet, E. 1995. ‘Feature structure of Romance clitics.’ Natural Language and Linguistic
Theory 13: 607–647.
Bošković, Ž. 1995. ‘Participle movement and second position cliticisation in Serbo-
Croatian.’ Lingua 96: 245–266.
Bošković, Ž. 1997. The Syntax of Nonfinite Complementation: an Economy Approach.
Cambridge, Mass.: MIT Press.
Bošković, Ž. 1998. ‘Wh-phrases and wh-movement in Slavic.’ Position paper for the
Comparative Slavic Morphosyntax Workshop, held at McCormick’s Creek State Park,
5–7 June. To appear in Proceedings.
Bošković, Ž. This volume. ‘Second position cliticisation: Syntax and/or phonology?’

explained is why the internal order of the core complex (of verbal auxiliary and pronominal clitics)
is preserved whether that complex precedes or follows its host, both in a single grammar and across
grammars; see Franks and King (in press) for discussion. This suggests to me that clitic clustering
takes place independently of and prior to determination of direction of support. Perhaps this can be
done with an Agr∀P, as in Richards (1997), or with a carefully worked out system of which
functional category attracts what. I have avoided specifics because of complexities endemic to the
espoused system in which the verb’s F-features pied-pipe those of the clitics up through the tree, and
hope to address this problem in future work.
50. Even as I write this, new projects are coming to the fore. I have just learned that the publisher
of this volume, John Benjamins Publishing Company, plans to produce a new edited volume on
clitics, with papers by Stephen Anderson, Jane Grimshaw, Geraldine Legendre, Andrew Spencer, and
Juan Uriagereka, among others. The volume has the preliminary title Clitics from Different Perspec-
tives, and is to be edited by Birgit Gerlach and Janet Grijzenhout.
 CLITICS AT THE INTERFACE 43

Bošković, Ž. In progress. ‘On the syntax-prosody interface: A view from the syntax.’ Ms.,
University of Connecticut.
Brown, S. and S. Franks 1995. ‘Asymmetries in the scope of Russian negation.’ Journal
of Slavic Linguistics 3: 239–287.
Browne, W. 1974. ‘On the problem of enclitic placement in Serbo-Croatian.’ In: Brecht,
R. and C. Chvany (eds.) Slavic Transformational Syntax. Ann Arbor: Department of
Slavic Languages and Literatures, University of Michigan, 36–52.
Browne, W. 1975. ‘Serbo-Croatian enclitics for English-speaking learners.’ In: Filipović,
R. (ed.) Contrastive Analysis of English and Serbo-Croatian, Vol.1. Zagreb: Institute
of Linguistics, Faculty of Philosophy, University of Zagreb, 105–134.
Caink, A. 1997. ‘Extended projections in South Slavic.’ Paper presented at the Second
Conference on Formal Approaches to South Slavic Linguistics, Sofia.
Castillo, J., K. Drury and K. Grohmann 1999. ‘The status of the merge over move
preference.’ Paper presented at the 75th Annual Meeting of the Linguistic Society
of America, Los Angeles.
Ćavar, D. 1996. ‘On cliticisation in Croatian: syntax or prosody?’ ZAS Papers in Linguis-
tics 6: 51–65.
Ćavar, D. and C. Wilder 1994. ‘Long head movement? Verb movement and cliticisation
in Croatian.’ Lingua 93: 1–58.
Chomsky, N. 1993. ‘A minimalist program for linguistic theory.’ In: The Minimalist
Program. Cambridge, Mass.: MIT Press, 167–217.
Chomsky, N. 1995. ‘Categories and transformations.’ In: The Minimalist Program.
Cambridge, Mass.: MIT Press, 219–394.
Chomsky, N. 1998. ‘Minimalist inquiries: the framework.’ MIT Occasional Papers in
Linguistics 15, Cambridge, Mass.: MIT Dept. of Linguistics and Philosophy.
Comrie, B. 1989. Language Universals and Linguistic Typology: Syntax and Morphology
(2nd edn.). Chicago: University of Chicago Press.
Dimitrova-Vulchanova, M. This volume. ‘Possessive constructions and possessive clitics
in the English and Bulgarian DP.’
Epstein, S., E. Groat, R. Kawashima and H. Kitahara 1998. A Derivational Approach to
Syntactic Relations. New York: Oxford University Press.
Franco, J. This volume. ‘Agreement as a continuum: The case of Spanish pronominal
clitics.’
Franks, S. 1995. Parameters of Slavic Morphosyntax. New York: Oxford University Press.
Franks, S. 1997. ‘South Slavic clitic placement is still syntactic.’ Penn Working Papers in
Linguistics 4: Proceedings of the 21st Annual Penn Linguistics Colloquium,
111–126.
Franks, S. 1998. ‘Clitics in Slavic.’ Position paper for the Comparative Slavic Morpho-
syntax Workshop, held at McCormick’s Creek State Park, 5–7 June. To appear in
Proceedings.
Franks, S. and T. King. In press. A Handbook of Slavic Clitics. New York: Oxford
University Press.
44 STEVEN FRANKS

Franks, S. and L. Progovac 1994. ‘On the placement of Serbo-Croatian clitics’ In: Indiana
Slavic Studies 7, Proceedings of the 9th Biennial Conference on Balkan and South
Slavic Linguistics, Literature and Folklore, 69–78.
Franks, S. and L. Progovac 1999. ‘Clitic second as verb second.’ Paper presented at the
75th Annual Meeting of the Linguistic Society of America, Los Angeles.
Golden, M. and M. Sheppard. This volume. ‘Slovene pronominal clitics.’
Golston, C. 1995. ‘Syntax outranks phonology: Evidence from Ancient Greek.’ Phonology
12: 343–368.
Grimshaw, J. 1997a. ‘The best clitic: Constraint conflict in morphosyntax.’ In: Haegeman,
L. (ed.) Elements of Grammar. Dordrecht: Kluwer, 169–196.
Grimshaw, J. 1997b. ‘Projection, heads, and optimality.’ Linguistic Inquiry 28: 373–422.
Groat, E. and J. O’Neil 1996. ‘Spell-out and the LF interface.’ In: Abraham, W., et al.,
(eds.) Minimal Ideas. Amsterdam: John Benjamins, 113–139.
Halle, M. and A. Marantz 1993. ‘Distributed morphology and the pieces of inflection.’ In:
Hale, K. and J. Keyser (eds.) The View From Building 20. Cambridge, Mass.: MIT
Press, 111–176.
Halpern, A. 1995. On the Placement and Morphology of Clitics. Stanford, Calif.: CSLI
Publications.
Halpern, A. and J. Fontana 1994. ‘X0 and Xmax clitics.’ In: Duncan, E., D. Farkas and P.
Spaelti (eds.) The Proceedings of WCCFL 12. Stanford: CSLI Publications, 251–266.
Harford, C. and K. Demuth. In press. ‘Phonology outranks syntax: an optimality theoretic
approach to subject inversion in Bantu.’ To appear in Linguistic Analysis.
Izvorski, R., T. King and C. Rudin 1997. ‘Against li lowering in Bulgarian.’ Lingua 102:
187–194.
Kallulli, D. This volume. ‘Direct object clitic doubling in Albanian and Greek.’
Kayne, R. 1975. French Syntax: the Transformational Cycle. Cambridge, Mass.: MIT
Press.
Kayne, R. 1994. The Antisymmetry of Syntax. Cambridge, Mass.: MIT Press.
King, T. H. 1994. ‘Focus in Russian yes-no questions.’ Journal of Slavic Linguistics 2:
92–120.
Klavans, J. 1982. Some Problems in a Theory of Clitics. Bloomington: Indiana University
Linguistics Club.
Krapova, I. and V. Petrov 1999. ‘Subjunctive complements, null subjects and case
checking in Bulgarian.’ In Dziwirek, K., H. Coats and C. Vakareliyska (eds.)
Proceedings of FASL 7: the Seattle Meeting. Ann Arbor: Michigan Slavic Publica-
tions, 265–287.
Legendre, G. 1999. ‘Morphological and prosodic alignment at work: The case of South-
Slavic clitics.’ In: Shahin, K., S. Blake and E.-S. Kim (eds.) The Proceedings of
WCCFL 17. Stanford: CSLI Publications, 436–450.
Munn, A. 1993. Topics in the Syntax and Semantics of Coordinate Structures. Doctoral
dissertation, University of Maryland.
 CLITICS AT THE INTERFACE 45

Percus, O. 1993. ‘The captious clitic: Problems in Serbo-Croatian clitic placement.’ Ms,
MIT.
Perlmutter, D. 1971. Deep and Surface Structure Constraints in Syntax. New York: Holt,
Rinehart and Winston.
Pesetsky, D. 1997. ‘Optimality theory and syntax: Movement and pronunciation.’ In:
Archangeli, D. and T. Langendoen (eds.) Optimality Theory: an Overview. Cam-
bridge, Mass.: Blackwell, 134–170.
Progovac, L. 1996. ‘Clitics in Serbian/Croatian: Comp as the second position.’ In:
Halpern, A. and A. Zwicky (eds.) Approaching Second: Second Position Clitics and
Related Phenomena. Stanford: CSLI Publications, 411–428.
Progovac, L. 1998. ‘Clitic-second and verb-second.’ Paper presented at the Comparative
Slavic Morphosyntax Workshop, held at McCormick’s Creek State Park, 5–7 June.
To appear in Proceedings.
Progovac, L. In press. ‘Eventive ‘to’ and the placement of clitics.’ In: Kenesei, I. (ed.)
Theoretical Issues in Eastern European Languages: Selected Papers from the
Conference on Linguistic Theory in Eastern European Languages (CLITE). Amster-
dam: John Benjamins.
Progovac, L. This volume. ‘Where do clitics cluster?’
Richards, N. 1997. What Moves Where When in Which Language? Doctoral dissertation,
MIT.
Rivero, M.-L. 1993. ‘Bulgarian and Serbo-Croatian yes-no questions: V0 raising to -li vs.
li-hopping.’ Linguistic Inquiry 24: 567–575.
Roberts, I. 1998. ‘Have/Be raising, move F, and procrastinate.’ Linguistic Inquiry 29:
113–125.
Rudin, C. 1997. ‘AgrO and Bulgarian pronominal clitics.’ In: Lindseth, M. and S. Franks
(eds.) Proceedings of FASL V: The Indiana Meeting. Ann Arbor: Michigan Slavic
Publications, 224–253.
Rudin, C., C. Kramer, L. Billings and M. Baerman. In press. ‘Macedonian and Bulgarian
li questions: beyond syntax.’ Natural Language and Linguistic Theory.
Rudnitskaya, E. 1999. ‘The yes/no question LI clitic placement in Russian.’ Paper
presented at Formal Approaches to Slavic Linguistics 8, Philadelphia.
Sadock, J. 1991. Autolexical Syntax. Chicago: University of Chicago Press.
Schick, I. This volume. ‘Clitic doubling contructions in Balkan Slavic languages.’
Schütze, C. 1994. ‘Serbo-Croatian second position clitic placement and the phonology-
syntax interface.’ MIT Working Papers in Linguistics 21: 373–473.
Schütze, C. 1996. ‘Serbo-Croatian clitic placement: An argument for prosodic movement.’
In: Toman, J. (ed.) Proceedings of FASL 3: the College Park Meeting. Ann Arbor:
Michigan Slavic Publications, 225–248.
Stepanov, A. 1998. ‘On Wh-fronting in Russian.’ In: Tamanji, P. and K. Kusumoto (eds.)
Proceedings of the North Eastern Linguistic Society 28. Amherst: University of
Massachusetts, 453–467.
46 STEVEN FRANKS

Tomić, O.M. 1996. ‘The Balkan Slavic clausal clitic clusters.’ Natural Language and
Linguistic Theory 14: 811–872.
Tomić, O.M. 1997. ‘Non-initial as a default clitic position.’ Journal of Slavic Linguistics
5: 301–323.
Tomić, O.M. This volume. ‘On clitic sites.’
Tranel, B. 1998. ‘Suppletion and OT: On the issue of the syntax/phonology interaction.’
In: Curtis, E., J. Lyle, and G. Webster (eds.) Proceedings of WCCFL 16. Stanford:
CSLI Publications, 415–429.
Travis, L. 1991. ‘Parameters of phrase structure and V2 phenomena.’ In: Friedin, R. (ed.)
Principles and Parameters in Comparative Grammar Cambridge, Mass.: MIT Press,
339–364.
Uriagereka, J. In press. ‘Multiple spell-out.’ To appear in: Epstein, S. and N. Hornstein
(eds.) Working Minimalism. Cambridge, Mass.: MIT Press.
Veselovská, L. 1995. Phrasal Movement and X0-Morphology. Doctoral dissertation,
Palacky University, Olomouc.
Zwart, C. J-W. 1997. Morphosyntax of Verb Movement. Dordrecht: Kluwer.
Zwicky, A. 1977. On Clitics. Bloomington: Indiana University Linguistics Club.
 Asymmetries in the Distribution of Clitics
The Case of Greek Restrictive Relatives*

Artemis Alexiadou Elena Anagnostopoulou

Zentrum für Allgemeine University of Crete
Sprachwissenschaft, Berlin

Abstract

Greek restrictive relative clause are of two types, one introduced by a wh-pro-
noun, the other by the complementiser pu ‘that’. Focusing on the latter, this
paper addresses an asymmetry in the distribution of object clitics. While direct
object clitics in pu-relatives are sensitive to the indefiniteness of the head,
indirect object clitics are obligatory in these constructions regardless of the
definiteness of the head. The paper proposes a head-raising analysis of relative
clause constructions to account for this asymmetry.

1. Restrictive relative clauses and clitics

Relative clauses in Greek are of two types: one type is introduced by a wh-
relative pronoun (cf. 1), and the other one is introduced by the complementiser
pu ‘that’ (cf. 2):
(1) diavasa ena vivlio to opioi mu sistise i Maria ti
read-1 a book- the-which - introduced Mary-
‘I read a book which Mary introduced to me’

* We would like to thank the audiences at the Workshop on Pied Piping in Jena in May 1997 and
the Workshop on Clitic Phenomena in English and other European Languages in Debrecen in
September 1997 and especially Marcel den Dikken, Melita Stavrou, and one anonymous reviewer for
comments and suggestions.
48 ARTEMIS ALEXIADOU AND ELENA ANAGNOSTOPOULOU

(2) diavasa ena vivlio pu mu sistise i Maria

read-1 a book- that - introduced Mary-
‘I read a book that Mary introduced to me’
Clitics may appear in both types of clauses, as illustrated in (3):
(3) a. agorasa ena vivlio to opio (to) diavasa me
bought-1 a book the which- (- read-1 with
megalo endiaferon
great interest
b. agorasa ena vivlio pu (to) diavasa me megalo
bought-1 a book that (- read-1 with great
endiaferon
interest
‘I bought a book which/that I read it with great interest’
In this paper we will concentrate on the distribution of clitics in pu relatives.
As first observed in Stavrou (1984), direct object (DO) clitics in pu relatives
are sensitive to the indefiniteness of the head. As shown in (4), clitics are
possible when the head of the relative is an indefinite NP, while they are not
possible when the head of the relative is definite.
(4) a. diavasa ena vivlio pu (to) pira apo ti vivliothiki
read-1 a book- that (- got from the-library-
‘I read a book that I got from the library’
b. *diavasa to vivlio pu to pira apo ti vivliothiki
read-1 the book- that - got from the library
The situation is quite different when the head of the relative corresponds to an
indirect object gap. In this case, indirect object (IO) clitics are obligatory
regardless of the definiteness of the head (cf. 5):
(5) enas/o mathitis pu *(tu) eftiaksa ena keik
a/the student- that *(-1 made-1 a cake
‘A/the student for whom I baked a cake’
In this paper we argue that the facts in (4) can be straightforwardly accommodat-
ed under a head-raising analysis for restrictive relative clauses (RRCs). We also
discuss the distribution of IO clitics, which does not immediately follow from a

1. We would like to point out that genitive and dative case have collapsed in Modern Greek, both
syntactic cases employing the form of the genitive. Here we gloss IOs uniformly as datives.
 ASYMMETRIES IN THE DISTRIBUTION OF CLITICS 49

head-raising analysis, and we explore a way in which the Greek facts can be
accounted for in terms of this analysis.

2. An account for direct object clitics

2.1 Direct object clitics

Let us take a closer look at the nature of DO clitics in RRCs. The sentences in
(6) and (7) show that restrictive relative clauses involving DO clitics are
selectively sensitive to islands. That is, they are sensitive to strong islands (cf. 6),
but not to weak islands (cf. 7) (see Cinque 1990 for a discussion of weak and
strong islands).
(6) Relative clause island
a. *gnorisa mja gineka pu sinandisa ton andra
got to know-1 a woman- that met-1 the man-
pu tin pandreftike
that - married-3
‘I got to know a woman that I met the man that married’
Complex NP-island
b. *vrika ena portofoli pu to gegonos oti to
found-1 a wallet- that the fact that -
ida me paraksenepse
saw-1 me surprised-3
Adjunct island
c. *pira mia efimerida pu o Petros apokimithike eno
got-1 a paper- that the-Peter- fell asleep while
tin diavaze
- read--3
(7) Wh-island
a. gnorisa mja gineka pu den ksero pjos tin
got-to-know-1 a woman that  know-1 who -
pandreftike
married-3
‘I got to know a woman that I do not know who married her’
50 ARTEMIS ALEXIADOU AND ELENA ANAGNOSTOPOULOU

Factive island
b. heretisa mia kopela pu harika pu tin
greeted-1 a girl- that was glad that -
ksanaida
saw-1 again
‘I greeted a girl that I was glad that I saw her again’
Being selectively sensitive to islands, this construction qualifies as a chain of the
Clitic Left Dislocation (CLLD) type discussed in detail in Cinque (1990). The
clitics exhibit the characteristics of A′-dependent clitics in the sense of Cinque
(1990) and Iatridou (1991). In general, in Greek and other languages, A′-depen-
dent clitics are licensed when their antecedent is referential/D-linked (cf. Iatridou
1991, Anagnostopoulou 1994 for Greek, Dobrovie-Sorin 1990 for Romanian
among others), while they are not licensed in chains headed by non-referential
elements (cf. Rizzi 1990 and Cinque 1990 for a systematic discussion). This is
illustrated in (8a–b) below for wh-questions.
(8) a. Pjon (*ton) ides?
who- (*- saw-2
‘Who did you see?’
b. Pjon apo tus filus su (ton) katalavenis kalitera?
which from the friends yours (- understand-2 better
‘Which one of your friends do you understand better?’
Given that, generally speaking, definite NPs are referential, it is not clear why in
RRCs they cannot license a clitic IP internally (compare (8a–b) with (4b)). Under
the standard analysis of relative clauses, in (4a–b) an empty operator has been
moved to SpecCP whose range is determined by the DPs ena vivlio, and to vivlio
respectively (cf. Browning 1987). The fact that in (4a) the clitic is legitimate is
not surprising given the well-formedness of the CLLD example in (9). What is
surprising, though, is that in (4b) a clitic is not permitted even though the ‘head’
of the relative is definite and the corresponding CLLD construction (10) is
grammatical:
(9) ena vivlio to pira apo ti vivliothiki
a book- - got-1 from the library
‘A book I got from the library’
(10) to vivlio to pira apo ti vivliothiki
the book - got-1 from the library
‘The book, I got it from the library’
 ASYMMETRIES IN THE DISTRIBUTION OF CLITICS 51

2.2 Direct object clitics under the head-raising analysis

2.2.1 The raising analysis

To account for the facts presented in the previous section, we adopt a structural
analysis of RRCs based on Kayne (1994) (cf. Vergnaud 1974; Schachter 1973),2
which relies on two related assumptions. First, the definite determiner that
appears to precede the head of the relative clause is external to the clause. In
other words, the relative clause is the complement of the determiner. Second, the
phrase which raises to SpecCP in pu-RRCs is an NP and not a null operator, as
in the standard analysis. In particular, as illustrated in (11), when the head of the
relative clause appears to be definite, what has raised to SpecCP is just the bare
NP pedi ‘child’. On the other hand, when the ‘head’ appears to be indefinite,
then what has raised to SpecCP is the DP ena pedi ‘a child’.
As shown in (11), ena and pedi form a constituent, the external D position
remaining empty. On the other hand, to and pedi do not form a constituent, since
the ‘head’ pedi raises from inside the clause to SpecCP, and the determiner to is
external to the relative. It follows from Kayne’s analysis, then, that there is a
structural difference between definite and indefinite NPs when they ‘head’ a RRC.
In the case of the definite head, what has been actually moved is only a bare NP.
As we shall see, this is crucial for our account of the occurrence of DO clitics.

2. Stavrou (1984) offers a different account from the one to be developed here. She distinguishes
between types of clitics that are present in relative clauses as the ones in (4) and clitics which are
present in RRCs as the ones in (5). With respect to the former she proposes that the presence of a
clitic can be explained by the notion of topicalisation. The presence of a clitic is dependent on
whether the head of the relative can be seen as a topic or not and the clitic is not involved in the
process of relativisation. This accounts also for the fact that in Greek appositive relatives, clitics are
always present (cf. i):
(i) to roloi tu, pu to agorase stin Elvetia
the watch his, that - bought-3 in the Switzerland
‘his watch, which he bought in Switzerland’
It is standardly assumed (see Sells 1985) that appositive relative clauses contain an element which
stands in a discourse anaphora relation with the NP they modify. It is precisely this anaphoric relation
that is expressed with the presence of a clitic in Greek appositive relatives.
Even though Stavrou’s analysis correctly captures the facts, there are some problems with the
particular implementation of her idea which are extensively discussed in Alexiadou & Anagnosto-
poulou (1997): a) first of all, as Lasnik & Stowell (1991) among others observe, in RRCs one cannot
dissociate the head of the relative from the clause to determine whether it is specific or not. It is
standardly assumed that the ‘head’ of the relative does not itself refer, but rather the reference of the
NP is determined jointly by the head noun and the relative clause. The reference of the NP is the
intersection of the sets defined by the head noun and the relative clause. b) Moreover, an account in
terms of topicalisation cannot explain why the distribution of clitics in RRCs is so different from all
other A′-dependencies, as shown in Section 2.1.
52 ARTEMIS ALEXIADOU AND ELENA ANAGNOSTOPOULOU

(11) DP

DP

D0 CP

to NP/QP CP
Ø
C0 IP
pedi
ena pedi pu
the/Ø child/a child that

Evidence that in English RRCs the definite determiner is external comes from
the fact that in certain cases it is licensed by a restrictive relative clause only (cf.
Vergnaud 1974, and recently Kayne 1994; Bianchi 1995 among others for
discussion). Consider the English examples in (12). In all three cases the definite
determiner cannot co-occur with the material following it, Paris, three books of
John’s and four of the boys, unless these phrases are modified by a restrictive
relative clause or some other restrictive modifier.
(12) a. the Paris *(I love)
b. the three books of John’s *(that I read)
c. the four of the boys *(that came to dinner)
That the XP raising to SpecCP is only an NP and not a full DP can be argued
for on the basis of an observation due to Browning (1987) (and see also
Vergnaud 1974; Schachter 1973). Browning notes that the trace within a
restrictive relative is interpreted as indefinite when the head of the relative
appears to be introduced by a definite determiner. The relative trace can appear
in contexts that typically exhibit the indefiniteness effect, like the existential
there construction (cf. 13b). A similar point can be made with respect to some
idiomatic expressions whose direct object is obligatorily a bare phrase (cf. 13c);
when the idiomatic object is relativised, it is introduced by a definite determiner
(cf. 13d):
(i) Definiteness restriction asymmetry
(13) a. *there were the men in the garden
b. [the men that there were t in the garden] were all diplomats
 ASYMMETRIES IN THE DISTRIBUTION OF CLITICS 53

(ii) Idiom asymmetry

c. *they made the fun of me
d. the fun that they made of me
The structure in (11) permits a structural implementation that readily captures the
above facts. In a feature checking approach (cf. Chomsky 1995) one could
motivate the obligatoriness of the raising of the bare NP in the presence of an
external definite D0 following Chomsky’s analysis of Definiteness Restriction
effects.3 The external determiner requires a NP complement, and as a result the
phrase undergoing raising can only be a NP and not a DP.4

2.2.2 Clitic doubling in restrictive relative clauses

Greek pu relatives headed by a definite determiner can be argued to fall under a
raising analysis on the basis of examples like the following:
(i) Definiteness restriction asymmetry
(14) a. *ihe tis gates ston kipo
had-3 the cats in the garden
‘There were the cats in the garden’
b. i gates pu ihe ston kipo
the cats that had-3 in the garden
‘The cats that there were in the garden’
(ii) Idiom asymmetry
c. *ekana tin plaka sto Jani
made-1 the fun to the John
‘I made the fun of John’
d. i plaka pu ekana sto Jani
the fun that made-1 to the John
‘the fun that I made of John

3. However, there are differences, since a bare NP cannot occur in postverbal position. Presumably,
this is related to the difference between an actual determiner the and a determiner-like element there.
(i) there arrived a man
(ii) *there arrived man
4. Zwart (1997) proposes that the higher CP attracts the head noun for semantic reasons: it needs an
element in its specifier in order to perform its function as a restrictor. After movement of the head
noun to the specifier position of CP has taken place, a configuration is created in which the relative
clause and the head noun can be interpreted as two constituents, one restricting the interpretation of
the other via set intersection.
54 ARTEMIS ALEXIADOU AND ELENA ANAGNOSTOPOULOU

As for English, we argue on the basis of the above contrasts that the definite
determiner is external to the constituent heading the relative, which is a bare NP
moving to SpecCP from the object position.
In such an analysis the clitic can be analysed as forming a constituent with
the raised NP. In other words, the clitic ‘doubles’ the raised NP. As we saw in
Section 2.1, the clitic construction is sensitive to strong islands, a fact which
supports a movement analysis from a clitic doubling input.5,6 In this view, the
restrictions on the distribution of clitics reduce to restrictions on clitic doubling.
There are a number of constraints on the morphological and semantic nature of
doubled NPs across languages.7 One such constraint is that DO doubling clitics
can never be associated with bare NPs. The Greek sentence in (15) illustrates this:
(15) *ta ida pedia na erhonde
- saw-1 children  come-3
‘I saw children coming’
Given what we have said so far, a RRC headed by a definite determiner results
from an input of the type illustrated in (15), i.e. it involves a bare phrase doubled
by an accusative clitic. Since this is an impossible structure to begin with, the
resulting relative is also impossible. Hence, the distribution of clitics in Greek
constitutes a language-specific argument in favour of a head raising analysis for
relative clauses.
Let us now come to the distribution of DO object clitics with indefinites,
which are well-formed. According to what we have said so far, RRCs headed by
indefinites which are optionally linked to a clitic must result from a well-formed
input involving an indefinite NP doubled by a clitic. Such strings are in principle
possible, though there are complications.
It is well known that clitic doubling of indefinites is possible in some
languages, if these are understood as specific or partitive. This is the case in
Romanian (cf. Dobrovie-Sorin 1990) and Spanish (cf. Suñer 1988). In Greek,
clitic doubling is restricted to certain types of discourse prominent definites (cf.
Anagnostopoulou 1994; Anagnostopoulou & Giannakidou 1995). Even in Greek,
however, it is possible for a DO clitic to co-occur with an indefinite. This

5. An alternative analysis would be that the clitic is a pronominal copy of the moved NP. However,
the pronominal copy hypothesis cannot offer a basis for an explanation of the asymmetry between
definite and indefinite headed relatives.
6. But cf. Cinque (1990) and Iatridou (1991) for a different conclusion concerning CLLD.
7. Cf. Jaeggli (1982), Borer (1984b), Suñer (1988), Anagnostopoulou (1994), Dobrovie-Sorin (1990)
among many others.
 ASYMMETRIES IN THE DISTRIBUTION OF CLITICS 55

happens in CLLD (cf. 9) and right dislocation, a construction in which a clitic

co-occurs with a DP to its right which is separated from the rest of the clause by
special intonation:8
(16) to PINO pu ke pu ena uzaki
- drink-1 every now and then an uzo
Recently Kayne (1994: 78ff.) and Cecchetto (1996) have argued that clitic
doubling, right dislocation and CLLD have the same underlying structure with a
clitic doubling a DP in all three constructions. The doubled DP starts out as a
complement, and is never generated in a left or right peripheral position. The
semantic and intonational differences among the three constructions are attributed
to the different sites to which the doubled DP moves overtly or covertly. If this
is so, then what is crucial for our analysis is that in CLLD and right dislocation
a clitic can co-occur with an indefinite, while it cannot co-occur with a bare NP.
(17) a. *Pedi to ida ktes
child - saw-1 yesterday
b. *to IDA pedi
- saw-1 child-
The sentences in (17) show that it is never possible for a clitic to co-occur with
a bare NP in CLLD and in right dislocation, in contrast with (9) and (16).
To conclude this section, we have argued that the restricted presence of DO
clitics in Greek RRCs can be accommodated under Kayne’s (1994) analysis of
relative clauses, providing further support for the raising analysis. In the next
section we turn to IO clitics.

3. Indirect object clitics

As mentioned in the introduction, in Greek the presence of a clitic is obligatory

when the extraction site of the RRC is an indirect object, regardless of definite-
ness. The relevant example is repeated below:
(18) ena/to koritsi pu *(tu) edosa to vivlio
a/the girl that *(- gave-1 the book-
‘A/The girl to whom I gave the book’

8. Cf. Zubizarreta (1998) and references provided there for a detailed description of the intonational
and discourse properties of right dislocation and a comparison between CLLD, right dislocation and
clitic doubling.
56 ARTEMIS ALEXIADOU AND ELENA ANAGNOSTOPOULOU

The ungrammaticality of (18) without the clitic is not surprising if the input to
(18) is a double object construction, because resistance to that- relativisation is
a well-known property of the double object construction in various languages (cf.
Larson 1988; Baker 1988; den Dikken 1995 among others and (19) below).
(19) *the man that I gave the book
In the next section we argue that the input to the structure in (18) without the clitic
is indeed a double object construction. In other words, it does not involve a morpho-
logical dative with the syntax of a prepositional dative, a possibility that cannot be a
priori excluded in view of the fact that morphological datives have been argued
to have the syntax of P-datives in some languages (cf. den Dikken 1995 based
on Müller 1995 for German datives and Falk 1990, and Holmberg & Platzack
1995 among others for Icelandic datives in so-called ‘inversion constructions’).

3.1 The Greek double object construction

Greek has an alternation between a P(reposition)-dative and P(reposition)-less

dative with morphological genitive case, which we are referring to as ‘dative’
(cf. note 1):
(20) a. PP-dative
Edosa to vivlio ston Janni
gave-1 the book- to-the-John-
‘I gave the book to John’
b. P-less dative
Edosa tu Janni to vivlio
gave-1 the-John- the book-
‘I gave John the book’
Evidence that (20b) is a double object construction has been presented in
Markantonatou (1994) and Anagnostopoulou (1997). Here, we partially recapitu-
late their arguments.
First of all, constructions like (20b) are more limited than those in (20a).
They are subject to several restrictions having to do with the animacy of the goal
and the semantic and morphological properties of the selecting predicates (cf.
Stowell 1981; Oehrle 1976; Pesetsky 1995; den Dikken 1995 among others).
Moreover, constructions of the type in (20b) provide evidence that the hierarchi-
cal relation between Theme and Goal is such that the Goal asymmetrically
c-commands the Theme (while in P-datives the Theme asymmetrically c-com-
mands the Goal).
 ASYMMETRIES IN THE DISTRIBUTION OF CLITICS 57

a. Animacy of the goal. A well-known property of the double object construction

is that it shows sensitivity to the [±human/animate] nature of the goal (cf.
Stowell 1981; Pinker 1989; den Dikken 1995 among others). Examples (21)–(22)
illustrate an asymmetry between the PP and the P-less construction depending on
animacy; while the former is possible with [+animate] and [−animate] goals, the
latter is grammatical only with [+animate] goals (cf. (21b) vs. (22b)):
(21) a. Estila to gramma ston Petro
sent-1 the letter- to-the Peter-
‘I sent the letter to Peter’
b. Estila tu Petru to gramma
sent-1 the Peter- the letter-
‘I sent Peter the letter’
(22) a. Estila to gramma stin Gallia
sent-1 the letter to-the-France-
‘I sent the letter to France’
b. *Estila tis Gallias to gramma
sent-1 the-France- the-letter-
‘I sent France the letter’
b. Restrictions depending on verb classes. As is well known, the double object
construction in English is not permitted with all verb classes. There are restric-
tions relating to the semantic properties of the selecting predicates (cf. Pinker
1989; Gropen et al. 1989; Pesetsky 1995) and their phonological/morphological
properties (cf. Oehrle 1976).
(i) Semantic. In Greek as in English, verbs that denote communication of
propositions do not permit the P-less dative construction:
(23) a. Ipostiriksa tin athootita mu ston dikasti
defended-1 the innocence my to-the-judge-
‘I defended my innocence to the judge’
b. *ipostiriksa tu dikasti tin athootita mu
defended-1 the judge- the innocence my
‘I defended the judge my innocence’
The same holds for Greek manner-of-speaking verbs, which in most cases do not
permit the P-less construction, similarly to their English counterparts:
(24) a. Ourliaksa ston Petro na viasti
screamed-1 to-the-Peter-  hurry-3
‘I screamed to Peter that he should hurry’
58 ARTEMIS ALEXIADOU AND ELENA ANAGNOSTOPOULOU

b. *ourliaksa tu Petru na viasti

screamed-1 the-Peter-  hurry-3
(ii) Phonological/morphological. It is well known for English that there are many
pairs of verbs which are equivalent in interpretation yet display an asymmetry
with respect to their occurrence in the double object construction. Oehrle
(1976: 124) and Pesetsky (1995) give examples of such pairs. Greek P-less
datives display similar effects. The contrasts below demonstrate that Greek verbs
allowing the dative alternation are morphologically simplex (cf. 25a–a′), while
verbs disallowing the dative alternation are morphologically complex, consisting
of a prefix and a stem (cf. 25b–b′).
(25) a. I Maria erikse tin bala ston Janni
the-Mary- threw-3 the-ball- to-the-John-
‘Mary threw the ball to John’
a′. I Maria erikse tu Janni tin bala
the-Mary- threw-3 the-John- the-ball-
‘Mary threw John the ball’
b. I Maria ek-sfendonise to paketo ston Janni
the-Mary- flung-3 the-parcel- to-the-John
‘Mary flung the parcel to John’
b′. *i Maria ek-sfendonise tu Janni to paketo
the-Mary- flung-3 the-John- the-parcel-
c. Asymmetries in the syntactic domain. A number of tests which diagnose c-
command relations have been taken to show that in English P-dative constructions,
the theme NP asymmetrically c-commands the goal NP inside the PP, while in
the double object construction, the goal c-commands the theme NP (cf. Barss &
Lasnik 1986). If we apply these tests to Greek dative constructions, we see that
the command relations in P-less datives are the reverse of the command relations
in P-datives, in other words the goal asymmetrically c-commands the theme.
(i) Weak crossover effects: as the examples in (26) show, a quantifier in NPgoal
position can bind a pronoun in NPtheme position; the reverse is not possible:
(26) a. ?Estila tu kathe ipallilui tin epitagi tui
sent-1 every employee the paycheck his
b. *?Estila tu katoxu tui kathe checki
sent-1 the-employee- his every check
Moreover, as the contrast between (27a) and (27b) shows, goal wh-extraction
when a theme contains a bound pronoun is possible, while theme wh-extraction
 ASYMMETRIES IN THE DISTRIBUTION OF CLITICS 59

across a goal containing a pronoun is not:

(27) a. ?Tinosipalliloui den edoses ton mistho tui?
which employee-  gave-2 the salary-his
b. *?Pjo checki den edoses tu katoxou tui?
Which paycheck  gave2–S the owner his
(ii) Superiority effects: as shown in (28), wh-extraction of themes across wh-in-
situ goals is not acceptable, while wh-extraction of goals across wh-in-situ
themes is well-formed:
(28) a. ?Tinos/pjas
ginekas estiles ti?
which woman sent-2 what
b. *Ti estiles tinos/pjas ginekas?
what sent-2 which woman
(iii) The each…the other construction: finally, a reciprocal reading is licit only
when the distributor each heads the dative and the reciprocal other is contained
within the theme as in (29a). Otherwise, the result is ungrammatical.
(29) a. ?Estila tis mias/kathe miteras to pedhi tis allis
sent-1 the one/every mother the child the other
b. *Estila tis miteras tu allou to ena/kathe pedhi
sent-1 the mother the other the one/every child
From the above discussion we conclude that example (20b) is indeed a double
object construction, and that consequently (18) involves a double object construc-
tion as the input to relativisation. Two questions now arise: (a) what explains the
ungrammaticality of (18) and (19), and (b) what explains the fact that the clitic
repairs the ungrammaticality. There is a large body of literature which tries to
account for the extraction facts in English dative shift constructions. In the next
section we take a closer look at two of these proposals and examine whether
they can account for the Greek facts as well.

3.2 Relativisation and dative shift

In English and other languages the theme argument can undergo A′-movement,
while the goal argument cannot in the double object construction. Most of the
approaches that try to account for this fact capitalise on the presence of an empty
preposition introducing the goal in the double object construction.
According to Baker (1988: 299), these extraction facts are accounted for in
terms of the non-oblique-trace filter in (30):
60 ARTEMIS ALEXIADOU AND ELENA ANAGNOSTOPOULOU

(30) *[Oi…Xj …[{−V}j ti] …] at S-structure

In (30), O stands for an operator, {−V} for a nonverbal category (i.e. a P or an
N) and X for a lexical category (usually V) which is co-indexed with the {−V}
element through reanalysis or incorporation. Baker assumes that double object
and applicative constructions involve reanalysis/incorporation into the verb of the
covert or overt preposition introducing the goal. In examples like (19) an empty
category appears with a different type of Case than expected given its thematic
role, and this may block its identification and recoverability in some way. Thus
the trace is an offending trace, because it is not obliquely case-marked contrary
to what is expected.
An alternative to Baker’s analysis is put forth in den Dikken (1995).
According to den Dikken, the ungrammaticality of (19) relates to the impossibili-
ty of pied-piping with null operator movement. As with Baker, a crucial assump-
tion behind this analysis is that the shifted dative is introduced by a zero P (cf.
Kayne 1984; Czepluch 1982). Unlike Baker, who assumes a ternary flat structure,
den Dikken assumes binary branching and proposes that the P-dative undergoes
leftward movement (dative shift) to a specifier position as illustrated in (31):
(31) ZP

PPi Z

P0 NP Z0 XP
Bob
NP X
a package
X0 PP
ti

In RRCs, the empty preposition must undergo pied-piping because otherwise the
null operator NP would be sub-extracted from a Left Branch, something which is
not permitted. However, null operator movement never pied-pipes more material.
(32) *John is not easy [[PP to OP]j to talk tj]
If we assume an analysis of the double object construction in terms of an empty
preposition, we can account for the ungrammaticality of the clitic-less construc-
tion either by appealing to recoverability, as suggested in Baker, or by appealing
to the impossibility of pied-piping with null operator movement, as suggested by
den Dikken. To decide between the two proposals is an empirical issue which
 ASYMMETRIES IN THE DISTRIBUTION OF CLITICS 61

cannot be fully discussed here. Den Dikken argues extensively that the problem
specifically applies to null operator constructions, while Baker’s filter predicts
that this restriction should be characterising all wh-movement. The Greek facts
seem to support den Dikken’s proposal, because the restriction shows up only in
RRCs and constructions involving focus movement9 but not in wh-interrogatives:
(33) a. *o anthropos pu edosa to vivlio
the man that gave-1 the book-
b. *TU PETRU edosa to vivlio
the Peter- gave-1 the book-
c. tinos edoses to vivlio
who- gave-1 the book-
d. tinos pistevis oti edosa to vivlio
who- believe-2 that gave-1 the book-
For this reason we mainly concentrate on den Dikken’s analysis. Specifically, we
assume that the clitic-less construction starts from a structure in which the dative
NP is preceded by a preposition which cannot be identified in situ. The PP
moves to a specifier position (i.e. undergoes dative shift) from which the empty
preposition can incorporate into the verb in order to be licensed. Null operator
movement will be impossible for the reasons suggested by den Dikken: sub-
extraction from a left branch is impossible and pied-piping with null operator
movement is also impossible. The ungrammatical structure is given in (34):
(34) [o anthropos] [CP OPi pu [edosa [PP Ø Pj t]i to vivlio]]
[the man that [gave-1 the book-
Let now us see whether the obligatoriness of the clitics can be successfully
accounted for in den Dikken’s terms. We must assume that the clitic identifies the
empty P in situ, so that no dative shift is triggered. As a result, the operator can
be extracted from the PP in complement position. From this position the operator
can be subextracted and therefore pied-piping of the zero P is not required.

9. It is standardly assumed in the literature on Greek that focus movement is not a null operator
construction (cf. Tsimpli 1990 among others). However, the Greek facts seem to show that focus
movement patterns with relative clause formation rather than with wh-question formation and it can
be argued that focus movement is like English topicalisation, which has often been analysed as a null
operator construction (cf. Chomsky 1977). There is an additional complication concerning relative
clauses which are ungrammatical even when the fronted phrase is an opio, i.e. ‘wh-’ phrase:
(i) *o anthropos tu opiu edosa to vivlio
the man the who- gave-1 the book-
We will come back to this in note 12.
62 ARTEMIS ALEXIADOU AND ELENA ANAGNOSTOPOULOU

(35) *[o anthropos] [CP OPi pu [tuj edosa to

[the man that [- gave-1 the
vivlio [PP ØPj [DP ti]]]]
book
The suggested move is possible if we adopt from den Dikken (1995: 135) the
following disjunctive condition on identification of zero prepositions:
(36) An empty dative preposition is licensed iff (i) or (ii)
a. [P Ø] is identified by dative Case morphology
b. [P Ø] is incorporated (at some level) into a verb.
In other words, we must assume that the clitic has dative Case morphology that
is sufficiently rich to identify an empty preposition in situ so that dative shift is
not necessary. This amounts to proposing that the clitic doubling construction is
not a true dative shift construction, despite appearances to the contrary.
That P-less datives with clitic doubling are quite different from the ones
without can be shown in the following environments.
a. Word order. First of all, as is well known about English, double object
constructions show strict ordering effects as in (37):
(37) *I described the book John
Consider now the Greek data in (38). While the theme cannot precede the goal
in (38a), this is possible in (38b) where the goal is doubled by a clitic.10
(38) a. ?*Periegrapsa tin embiria mu tu psychiatrou me
described-1 the experience my the-psychiatrist- with
poli kopo
great effort-
b. Tu periegrapsa tin embiria mu tu psychiatrou
- described-1 the experience my the-psychiatrist-
me poli kopo
with great effort-
‘I described my experience the psychiatrist with great effort’
Given that ordering restrictions are canceled in the presence of doubling, we can
conclude that in clitic constructions no dative shift takes place.

10. Unlike English, in Greek it is sometimes possible for an non-doubled dative to follow a theme
(cf. Markantonatou 1994), but it can be shown that this order is produced by A′-movement of the
theme over the dative (cf. Anagnostopoulou 1997).
 ASYMMETRIES IN THE DISTRIBUTION OF CLITICS 63

b. Licensing under theme passivisation. As is also well known, English double

object constructions resist theme passivisation (cf. 39):
(39) *The book was given John
The same restriction obtains in Greek. It is canceled, however, when the goal is
doubled by a clitic:
(40) a. *?To vivlio dothike tu Janni apo tin Maria
the book was given the-John- by the-Mary-
b. To vivlio tu dothike tu Janni apo tin Maria
the book - was given the-John- by the-Mary-
‘The book was given to John by Mary’
This is a second piece of evidence that these constructions do not involve dative
shift.
c. Weaker morphophonological restrictions when a doubling clitic is present.
Finally, the morphophonological restrictions observed above are weaker
when a doubling clitic is present (cf. (42a–b); compare these examples to
(25b′)):
(41) *Mary donated charity the money
(42) a. *eksfendonise tu Janni to packeto
flung-3 the-John- the-parcel-
b. ??tu eksfendonise tu Jani to packeto
- flung-3 the-John- the-parcel-
‘He/she flung John the parcel’
The facts above show that is not unjustified to propose this analysis for the clitic
doubled datives. The null operator analysis is successful.
Now, however, we seem to have led ourselves into a paradox: to account
for DO clitics we need to appeal to a raising analysis, but to account for IO
clitics we appeal to a null operator analysis.

4. Pied-piping as P-stranding

At this point, two possibilities suggest themselves: either the raising explanation
does not hold, in which case the distribution of DO clitics will be accounted for
on purely semantic grounds, or the null operator explanation does not hold, in
which case we have to revise our explanation for the distribution of IO clitics
64 ARTEMIS ALEXIADOU AND ELENA ANAGNOSTOPOULOU

and IO relativisation without clitics must be accounted for under a raising

analysis. Here we will explore the latter option.
The specific analysis we will put forward will be an attempt to restate the
restriction on pied-piping with null operators as a restriction on P-stranding in
. Specifically, we will adopt a proposal according to which RRCs have a
more articulated structure than previously thought. This proposal has been
developed by Bianchi (1995) for Italian and English and is in line with the view
put forth by a number of researchers that CP involves more that one functional
projection (cf. Reinhart 1981; Borer 1984a; Bhatt & Yoon 1991; Müller &
Sternefeld 1993; Rizzi 1995 among others). According to Bianchi, when a
wh-pronoun or a PP is fronted, it first moves as a whole to a SpecTopicP which
is situated below CP. From there the bare NP is subextracted to SpecCP. This
sub-extraction is necessary in order for the external D to get associated with an
NP complement:11
(43) DP

D0 CP

Spec CP

C0 TopicP

PP TopicP

P0 DP Topic0 IP

D tNP
the way in which he acted tPP

If we adopt Bianchi’s analysis, we can account for the ungrammaticality of IO

relatives in English by first looking at the restrictions on prepositions in the
sentences in (44):

11. Note that this analysis implies that the phrase in SpecTopicP is transparent for subextraction of
the NP head, unlike the dative shifted phrase which is not transparent. In GB literature it has been
argued that there is a difference between A and A′-positions concerning subextraction (cf. Chomsky
1986: 25–26, Lasnik and Saito 1992: 102 among others).
 ASYMMETRIES IN THE DISTRIBUTION OF CLITICS 65

(44) a. *the man that with I talked

b. *the with man that I talked
c. the man that I talked with
What (44) shows is that the only possibility for a preposition in PP relatives is
to remain in situ, as in (44c). (44a) and (44b), where the preposition has moved
along with the NP, are ungrammatical. (44b) is ungrammatical because SpecCP
by definition demands a bare NP and (44a) in which P has been stranded in
TopicP is ungrammatical as well. We propose that the reason for the ungrammat-
icality is that P-stranding requires reanalysis (van Riemsdijk 1978; Hornstein &
Weinberg 1981; Kayne 1984; Baker 1988 among others) and reanalysis requires
that the target be a lexical category and  is not lexical.
Let us now come to IO relatives of the type exemplified in (45).
(45) *the man that I gave the book
By hypothesis, IO objects are also PPs headed by a zero preposition. The facts
in (44a–b) show that it is impossible for the zero P to move along with the NP
for the reasons given above. The only other alternative is for the preposition to
remain in situ. Crucially, however, PPs in the double object construction are in
a position out of which extraction of the NP is not possible.
In this sense, the restriction on pied-piping with null operator movement is
restated as a restriction on P-stranding/reanalysis in .
Turning to Greek, first of all note that there is evidence in favour of the
structure in (43) (see Alexiadou 1998 for some discussion). (46) below shows
that in Greek pu relatives a PP may (as in 46b) or must (as in 46a) be fronted to
a position immediately following the complementiser:
(46) a. o anthropos pu mazi m’afton vgika ekso (*mazi m’afton)
the man that with him went out-1
b. o antropos pu (mazi tu) vgika ekso (mazi tu)
the man that (with him went-out (with him
Hence we can extend the explanation provided for English double object
constructions to the Greek datives without the clitic.12
When there is a clitic, then the empty headed-PP associated with it remains

12. Note that under this analysis we can derive the difference between wh-questions and wh-relat-
ives noted in note 9. For wh-relatives we must assume that the NP is not allowed to pied-pipe the
preposition because D must combine with an NP. The derivations that are possible are either one
involving subextraction from the dative shifted position or one involving subextraction from the topic
position. Both of them are impossible for the reasons explained in the main text.
66 ARTEMIS ALEXIADOU AND ELENA ANAGNOSTOPOULOU

in complement position and does not undergo dative shift, as we saw in the
previous section. The NP can be subextracted out of the PP complement because
this is not on a left branch. In other words, the constructions with clitics have the
same derivation as the English construction (44c). The only difference is that the
preposition is covert.13

5. A note on preposition stranding and cliticisation

We have claimed that the structure illustrated below in (47) in both its raising
and null operator counterparts involves P-stranding.
(47) a. [o anthropos] [CP Øi pu [tuj edosa to
[the man that [- gave-1 the
vivlio [PP ØPj [DP ti]]]]
book
b. o [CP anthroposi pu [IP tu edosa to
the man that - gave-1 the
vivlio [PP Ø [NP ti]]]]
book
Greek, however, does not seem to have P-stranding, as the ungrammaticality of
(48) shows:
(48) *pion horepses me
who danced-2 with
Nevertheless, Greek has some constructions that could be analysed on a par with
(47) as involving P-stranding. These cases involve complex prepositions formed
by a ‘heavy’ preposition that can also be used intransitively and which is

13. There is a tricky issue arising from our argumentation. On the one hand, we appeal to the
impossibility of clitic doubling of bare NPs to account for the distribution of DO clitics in RRCs. On
the other hand, however, we argue that the presence of the clitic is obligatory in the case of IO
extraction despite the fact that clitic doubling of bare IO NPs is also impossible in Greek. What we
have to say is that the nature of the clitic in the two cases is quite different. As is well known, in the
case of DO doubling the clitic is more like a marker of specificity (cf. Suñer 1988; Sportiche 1996;
Dobrovie-Sorin 1990) or referentiality (cf. Anagnostopoulou & Giannakidou 1995), thus being
incompatible with a bare phrase by its very nature. In the case of IO doubling the clitic is more of
an agreement marker (cf. Suñer 1988; Jaeggli 1982) and in Greek it functions as the piece of
morphology that identifies an empty preposition. Hence it is required by the syntax and therefore it
may co-occur with bare NPs in certain contexts such as RRCs.
 ASYMMETRIES IN THE DISTRIBUTION OF CLITICS 67

followed by a PP introduced by a ‘light’ or ‘simple’ preposition (cf. Theophano-

poulou-Kontou 1985; Terzi 1991 for discussion). The heavy preposition cannot
be followed by a DP regardless of the case it bears:
(49) a. epano sti Maria
over to-the Mary-
b. *epano tis Marias/ti Maria
over to-Mary-/Maria-
c. epano tis
over -
According to Anagnostopoulou (1997), (50a–b) is the structural representation of
(49a & c). In (50b) the clitic identifies an light empty preposition.
(50) a. [PP1 Epano [PP2 *(s)- ti Maria]]
over to-the-Mary
b. [PP1 Epano-tisj [PP2 Ø Pj]]
over-Cl()
In (50a), the intransitive P epano selects the transitive P se ‘to’ which Case-
licenses the NP. In (50b) the second P is lost and PP2 is replaced by a dative
clitic. Viewed this way, the alternation in complex prepositions is very similar
to the PP-dative alternation in the VP domain (cf. Terzi 1991 for a similar
conclusion).14
These constructions show an interesting behavior in RRCs. Consider the
examples below:
(51) a. i kopela pu horepsa mazi tis
the girl that danced-1 together -
b. *i kopela pu horepsa mazi
the girl that danced-1 together
c. i kopela pu mazi tis horepsa
the girl that together - danced-1
d. *i kopela pu mazi horepsa
the girl that together danced-1
In (51a) P-stranding is possible and a clitic is present. In (51b) P-stranding is not
possible and the clitic is absent. (51c) is the counterpart of (51a), and is again

14. Terzi (1991) analyses these constructions in the following way: a functional category FP,
selected by the ‘heavy’ P, is argued to be present between PP1 and PP2. The clitic adjoins to F0 and
receives dative case.
68 ARTEMIS ALEXIADOU AND ELENA ANAGNOSTOPOULOU

possible, except that the preposition has moved to the SpecTopicP position, while
(51d) is the counterpart of (51b) with movement to the Topic position and is
again impossible. The explanation that we have offered for the obligatoriness of
the clitic in double objects constructions can be extended to the complex
prepositions constructions: the clitic is obligatory to identify the zero P.
The generalisation derived both from the double object facts and the
complex preposition examples seems to be that in Greek P-stranding is possible
only when a clitic is present. This is reminiscent of the structural condition on
P-stranding in Dutch (cf. van Riemsdijk 1978; van Riemsdijk & Williams
1986: 296ff), namely r-cliticisation:
(52) a. dat ik niet [PP eri op ei] vuur
that I not it at fire
b. *dat ik hemi niet op ei vuur
that I him not at fire
As (52a) shows, a clitic-like pronominal element, an r-pronoun, is moved to a
special position to the left of the preposition. Stranding is possible only with
r-pronouns that are moved either into the clitic position or into . Moreover,
stranding under r-movement is possible only when the PP in question is an
argument of the verb, situated inside the VP. This is very close to the general-
isation we have reached for Greek P-stranding. According to van Riemsdijk &
Williams 1986, r-cliticisation is a way to license P-stranding without P reanaly-
sis. We hope that whatever explains the Dutch facts can be extended to Greek.

References

Alexiadou, A. 1998. ‘On the structure of Greek relative clauses.’ Studies in Greek
Linguistics 18: 15–29.
Alexiadou, A. and E. Anagnostopoulou 1997. ‘Clitics in restrictive relatives: an antisym-
metric account.’ Studies in Greek Linguistics 17: 304–317.
Anagnostopoulou, E. 1994. Clitic Dependencies in Modern Greek. Doctoral dissertation,
Universität Salzburg.
Anagnostopoulou, E. 1997. ‘The dative alternation in Greek.’ Ms, Tilburg University.
Anagnostopoulou, E. and A. Giannakidou 1995. ‘Clitics and prominence or why specifici-
ty is not enough.’ Proceedings of the Chicago Linguistic Society Meeting 1995,
Parasession on Clitics. Chicago: Chicago Linguistic Society, 312–331.
Baker, M. 1988. Incorporation: A Theory of Grammatical Function Changing. Chicago:
University of Chicago Press.
 ASYMMETRIES IN THE DISTRIBUTION OF CLITICS 69

Barss, A. and H. Lasnik 1986. ‘A note on anaphora and double objects.’ Linguistic
Inquiry 17: 347–354.
Bhatt, R. and J. Yoon 1991. ‘On the composition of comp and parameters of V2.’
Proceedings of WCCFL 10: 41–52.
Bianchi, V. 1995. Consequences of Antisymmetry for the Syntax of Headed Relative
Clauses. Doctoral dissertation, Scuola Normale Superiore Pisa.
Borer, H. 1984a. ‘Restrictive relatives in modern Hebrew.’ Natural Language and
Linguistic Theory 2: 219–260.
Borer, H. 1984b. Parametric Syntax: Case Studies in Semitic and Romance Languages.
Dordrecht: Foris.
Browning, M. 1987. Null Operator Constructions. Doctoral dissertation, MIT.
Cecchetto, C. 1996. ‘Clitic right dislocation is not right dislocation.’ GLOW Newsletter
36: 26–27.
Chomsky, N. 1977. ‘On Wh-movement.’ In: Culicover, P.W., T. Wasow and A. Akmajian
(eds.) Formal Syntax. New York: Academic Press, 71–133.
Chomsky, N. 1986. Barriers. Cambridge, Mass.: MIT Press.
Chomsky, N. 1995. The Minimalist Program. Cambridge, Mass.: MIT Press.
Cinque, G. 1990. Types of A′-Dependencies. Cambridge, Mass.: MIT Press.
Czepluch, H. 1982. ‘Case theory and the dative construction.’ The Linguistic Review 2:
1–38.
den Dikken, M. 1995. Particles: on the Syntax of Verb-Particle, Triadic, and Causative
Constructions. New York: Oxford University Press.
Dobrovie-Sorin, C. 1990. ‘Clitic doubling, Wh-movement and quantification in Roman-
ian.’ Linguistic Inquiry 21: 351–397.
Falk, C. 1990. ‘On double object constructions.’ Working Papers in Scandinavian Syntax
46: 53–100.
Gropen, J., S. Pinker, M. Hollander, R. Goldberg and R. Wilson 1989. ‘The learnability
and acquisition of the dative alternation in English.’ Language 65: 203–257.
Holmberg, A. and C. Platzack 1995. The Role of Inflection in Scandinavian Syntax. New
York: Oxford University Press.
Hornstein, N. and A. Weinberg 1981. ‘Case theory and preposition stranding.’ Linguistic
Inquiry 12: 55–91.
Jaeggli, O. 1982. Topics in Romance Syntax. Dordrecht: Foris.
Iatridou, S. 1991. ‘Clitics and island effects.’ Ms., MIT.
Kayne, R. 1984. Connectedness and Binary Branching. Dordrecht: Foris.
Kayne, R. 1994. The Antisymmetry of Syntax. Cambridge, Mass.: MIT Press.
Lasnik, H. and M. Saito 1992. Move a. Cambridge, Mass.: MIT Press.
Lasnik, H. and T. Stowell 1991. ‘Weakest crossover.’ Linguistic Inquiry 22: 687–720.
Larson, R. 1988. ‘On the double object construction.’ Linguistic Inquiry 19: 335–391.
Longobardi, G. 1994. ‘Reference and proper names: A theory of N-movement in syntax
and logical form.’ Linguistic Inquiry 25: 609–665.
70 ARTEMIS ALEXIADOU AND ELENA ANAGNOSTOPOULOU

Markantonatou, S. 1994. ‘Diptwta hmata thV N aV EllhnikhV: mia l xikosh-

masiologikh p os ggish.’; ‘Diadic verbs in Modern Greek: A lexical semantic
approach.’ Studies in Greek Linguistics 15: 362–371.
Müller, G. 1995. A′-Syntax. Berlin: Mouton de Gruyter.
Müller, G. and W. Sternefeld 1993. ‘Improper movement and unambiguous binding.’
Linguistic Inquiry 24: 461–507.
Oehrle, R. 1976. The Grammatical Status of the English Dative Alternation. Doctoral
dissertation, MIT.
Pesetsky, D. 1995. Zero Syntax: Experiencers and Cascades. Cambridge, Mass.: MIT
Press.
Pinker, S. 1989. Learnability and Cognition: the Acquisition of Argument Structure.
Cambridge, Mass.: MIT Press.
Reinhart, T. 1981. ‘A second  position.’ In: Belletti, A. et al. (eds.) Theory of
Markedness in Generative Grammar. Pisa, Scuola Normale Superiore, 517–557.
Riemsdijk, H. van 1978. A Case Study in Syntactic Markedness. Dordrecht: Foris.
Riemsdijk, H. van and E. Williams 1986. Introduction to the Theory of Grammar.
Cambridge, Mass.: MIT Press.
Rizzi, L. 1990. Relativized Minimality. Cambridge, Mass.: MIT Press.
Rizzi, L. 1995. ‘On the fine structure of the left periphery.’ Ms., University of Geneva.
Sells, P. 1985. Restrictive and Non-Restrictive Modification. CSLI Report No. 85–28.
Stanford, Calif.
Schachter, P. 1973. ‘Focus and Relativization.’ Language 49: 19–46.
Sportiche, D. 1996. ‘Clitic constructions.’ In: Rooryck, J. and L. Zaring (eds.) Phrase
Structure and the Lexicon. Dordrecht: Kluwer, 213–277.
Stavrou, M. 1984. ‘H klitikh antwnumia stiV p io istik V p otas iV m
xa thsh am sou antik im nou pou isagontai m to ‘pou’.’; ‘The clitic
prounoun in restrictive relative clauses that are introduced by ‘pu’.’ Studies in Greek
Linguistics 2: 121–136.
Stowell, T. 1981. Origins of Phrase Structure. Doctoral dissertation, MIT.
Suñer, M. 1988. ‘The role of agreement in clitic doubling constructions.’ Natural
Language and Linguistic Theory 6: 391–434.
Terzi, A. 1991. ‘Genitive clitics and prepositions.’ Ms., CUNY.
Theophanopoulou-Kontou, D. 1985. ‘Oi pou-anajo ik V p otas iV thV N aV
EllhnikhV kai h pa al iyh thV p oJ s wV.’; ‘Pu-relatives in Modern Greek
and preposition deletion.’ Studies in Greek Linguistics 3: 17–45.
Tsimpli, I. 1990. ‘The clause structure and word order of modern Greek.’ UCL Working
Papers in Linguistics 2: 226–255.
Vergnaud, J.-R. 1974. French Relative Clauses. Doctoral dissertation, MIT.
Zubizarreta, M.L. 1998. Word Order, Prosody and Focus. Cambridge Mass.: MIT Press.
Zwart, C. J-W. 1997. ‘A head raising analysis of relative clauses in Dutch.’ Ms.,
University of Groningen.
 Second Position Cliticisation
Syntax and/or Phonology?*

Željko Bošković
University of Connecticut

Abstract

Using a wide variety of empirical evidence as a testing ground, this paper

considers four different perspectives (the strong and weak syntax accounts and
the strong and weak phonology accounts) for the placement of second position
clitics in Serbo-Croatian, and argues for an analysis in which clitics undergo
movement in syntax, and phonology is responsible for the fact that they must
occur in the second position of the intonational phrase which they end up in
after syntactic movement. The paper presents a derivational model in which
syntax feeds phonology, locating the 2 effect in the phonology without
appealing to movement in this component.

1. Introduction

Second position (2) cliticisation has attracted a great deal of attention among
syntacticians, phonologists, and morphologists. This is understandable, given that
the solution to the 2 cliticisation puzzle has promised to shed light on such
serious theoretical issues as the nature of the phonology-syntax interface
(including the questions of whether the interface is derivational or co-present,
non-derivational and, if the former, whether the syntax needs to look ahead to

* For helpful comments and discussion I am grateful to Cédric Boeckx, Damir Ćavar, Marcel den
Dikken, Steven Franks, Sandra Stjepanović, Olga Mišeska Tomić, Chris Wilder, and an anonymous
reviewer. Portions of this material were presented at the Workshop on Clitic Phenomena in English
and other European Languages at Debrecen in 1997 and at the University of Novi Sad in 1998.
72 ŽELJKO BOŠKOVIĆ

the needs of the phonology), and the questions of whether movement is possible
in PF, what the internal structure of PF is, when lexical insertion can take place,
how much of morphology is syntactic, what the nature of V2 is, etc.
The bulk of recent work on 2 cliticisation has been done with respect to
Slavic languages, especially Serbo-Croatian (SC).1 This is not surprising, given
that most 2 cliticisation languages are either no longer spoken (Sanskrit, Ancient
Greek, Hittite, Old Spanish, among others) or, if they are, they are not as readily
accessible as SC (Warlpiri, Pashto, Tagalog, Luiseño, Mayo, Ngiyambaa, among
others).2 For this reason SC is increasingly becoming a testing ground for
theories of 2 cliticisation. As a result, the argumentation and the kind of data
examined with respect to 2 cliticisation in SC have reached the level of
subtlety not attested in the discussion of the phenomenon in other languages.
This paper focuses on 2 cliticisation in SC. The empirical basis of the
phenomenon will be examined in some detail and used as a testing ground for
different theories of 2 cliticisation. Previous arguments for and against differ-
ent theories of 2 cliticisation in SC will be summed up and a number of new
arguments will be adduced in an attempt to provide a complete picture of the
phenomenon and determine empirically and conceptually the most adequate
theory of 2 cliticisation in SC.3 The overarching theoretical concern during
the investigation will be issues concerning the relation of syntax to phonology,
particularly the nature of the interface and the question of whether typical
syntactic operations such as movement can apply in the phonology.4

1. Czech and Slovenian are also considered to have 2 cliticisation. However, ‘2’ clitics in these
languages can appear clause initially, which means that they are losing 2 clitichood. This makes
Czech and Slovenian much less profitable than SC in examining the 2 effect.
2. I have in mind here the availability of the relevant data, number of linguists working on the
language, and a relatively rich descriptive grammar tradition.
3. One aspect of 2 cliticisation in SC will be ignored here, namely, the order of clitics within the
clitic cluster. Giving this very intricate and not very well-understood issue proper attention would
require a paper in itself. However, as far as I can see, the conclusions concerning 2 cliticisation
reached below are independent of the issue of how the ordering of clitics within the clitic cluster is
achieved. (In this respect, see Stjepanović (1998a, b) for an interesting discussion of a peculiar
ordering requirement on the clitic je ‘is’.)
4. It is important to bear in mind that what I mean by phonology throughout this paper is the whole
PF component, which comprises the derivation from SS to the final phonetic representation. Some
morphological operations are clearly included here.
 SECOND POSITION CLITICISATION 73

2. Approaches to Second Position Cliticisation in Serbo-Croatian

The phenomenon of 2 cliticisation in SC is illustrated by (1a–d). Locating

clitics in any other position or splitting the clitic cluster in (1) would lead to
ungrammaticality. (2 clitics are given in italics.)
(1) a. Mi smo mu je predstavili juče
we are him her introduced yesterday
‘We introduced her to him yesterday.’
b. Zašto smo mu je predstavili juče?
why are him her introduced yesterday
‘Why did we introduce her to him yesterday?’
c. Ona tvrdi da smo mu je mi predstavili juče
she claims that are him her we introduced yesterday
‘She claims that we introduced her to him yesterday.’
d. Predstavili smo mu je juče
introduced are him her yesterday
‘We introduced her to him yesterday.’
A number of different theories of 2 cliticisation in SC have been proposed.
Here, I will concentrate on the approaches that taken together can give us a
complete picture of different angles one can take when examining the phenome-
non.5 The approaches can be broadly classified into 2 groups — syntactic and
phonological — depending on which component of the grammar plays the most
prominent role in the account. Each of these can be further subdivided into two
groups depending on whether they allow at least some amount of word reorder-
ing in PF. We then get the following 4 approaches:
a. The strong syntax approach, which holds that the syntax is completely
responsible for 2 cliticisation in SC, the linear position of the clitic cluster
being fully determined by the syntax. This approach has a number of
proponents. Some of the works representing this line of research are
Dimitrova-Vulchanova (1995), Progovac (1996), Rivero (1997), Roberts
(1994), Tomić (1996), and Wilder and Ćavar (1994a, b).
b. The strong phonology approach, which holds that the phonology is fully
responsible for 2 cliticisation in that it brings clitics into second position.

5. Several interesting approaches to 2 cliticisation in SC (see, for example, Zec and Inkelas 1990,
Anderson 1992, 1993, and Phillips 1996) cannot be easily categorised in the typology to be given
below, differing as they do from the works cited below in some basic underlying assumptions
concerning the nature of the phonology-syntax interface and/or lexical insertion of clitics.
74 ŽELJKO BOŠKOVIĆ

(Clitic placement is accomplished by applying Move in the phonology.)

This approach allows extensive word reordering in PF. The best known
representative of this line of research is Radanović-Kocić (1988) (see also
Hock 1992).
c. The weak syntax approach. Under this approach most movements of clitics
take place in the syntax. However, a small amount of word reordering is
still allowed in PF. More precisely, clitics are allowed to undergo phonolog-
ical movement in certain well-defined configurations. Halpern (1995) is the
first explicit proponent of this approach for SC. Other works along these
lines are Embick and Izvorski (1997), King (1996), Percus (1993), and
Schütze (1994).
d. The weak phonology approach, which holds that the phonology is responsi-
ble for the 2 effect. Under this approach, however, all movements of
clitics take place in the syntax. Phonology plays only a passive filtering role
by ‘selecting’ outputs of syntax; i.e. by ruling out certain syntactically well-
formed sentences due to violations of phonological requirements on clitics.
Bošković (1995) gives an outline of such an approach but does not fully
develop it. That task will be taken up below.6
I will start by discussing syntactic approaches.

3. Syntactic Accounts of Second Position Cliticisation

The strong and the weak syntax account of 2 cliticisation share the following
assumptions:
(2) a. Clitics cluster together syntactically, i.e. clause-mate clitics are
all located in the same position.
b. The position is structurally fixed for all constructions.
c. It is located high in the tree, so that there is no space for more
than one element to occur in front of the clitic cluster within
its clause.
Most proponents of the syntax approach assume that SC clitics are located under
C0 (Franks and Progovac 1995; King 1996; Progovac 1996; Schütze 1994; Tomić

6. An interesting, more syntactically oriented version of this approach is being developed in work
in progress by Steven Franks, who assumes both syntactic constraints on clitic placement and filtering
out by the phonology. Some syntactic accounts also assume a limited amount of filtering out by the
phonology (see Sections 3.1 and 3.2).
 SECOND POSITION CLITICISATION 75

1996; and Wilder and Ćavar 1994a, b). Rivero (1997) argues that SC clitics are
located in the Spec, and Percus (1993) and Roberts (1994) in the head position
of a phrase located between C and I. Halpern (1995) assumes that they are
adjoined to that phrase.
The strong and the weak syntax account differ with respect to the possibili-
ty of having no lexical material in front of clitics in the overt syntax. Under the
strong syntax account this possibility never arises: No grammatical construction
can contain sentence initial clitics in the output of the syntax. Clitics are placed
in 2 in the syntax. Under the weak syntax account, on the other hand, it is
possible to have grammatical constructions in which a clitic is located sentence
initially in the output of the syntax. If that happens, the clitic moves in the
phonology looking for an appropriate host. The underlying assumption here is
that SC 2 clitics have a lexical requirement that forces them to be suffixes on
a stressed element. Phonology has a filtering effect in that it rules out construc-
tions in which a clitic is found sentence initially, violating the suffixal require-
ment. Clitics are allowed to move in PF in order to satisfy this requirement. In
particular, PF movement takes place in SC when an enclitic is found stranded in
sentence initial position to ensure that the enclitic has an appropriate host.7
Given the well-defined motivation for PF movement, the movement ends up
being very local (it places the clitic in a position following the first stressed
word). Halpern (1995) calls the operation responsible for moving clitics in PF
Prosodic Inversion (PI) and formulates it as follows:
(3) For a DCL [directional clitic], X, which must attach to a w [phono-
logical word] to its left (respectively right),
a. If there is a w, Y, comprised of material which is syntactically
immediately to the left (right) of X, then adjoin X to the right
(left) of Y.
b. else attach X to the right (left) edge of the w composed of
syntactic material immediately to its right (left).
(Halpern 1995: 63)
Halpern formulates PI as a last resort operation that affects clitics only if their
prosodic requirements are not satisfied and moves them only the minimal
distance necessary to satisfy the requirements. In the following section I will
examine the theoretical and empirical validity of PI.

7. Given this line of reasoning we would also expect PF movement to occur in languages with
proclitics when a proclitic is found in sentence final position. I am not aware of any examples of this
kind. If there are no such examples, we would have a serious problem for this approach.
76 ŽELJKO BOŠKOVIĆ

3.1 The weak syntax account: prosodic inversion

Halpern proposes PI to account for the traditional observation (see Browne 1974)
that SC clitics can be located either after the first phrase of their sentence (1
environment), as in (4a), or after the first word (1 environment), as in (4b),
where a clitic appears to break up a phrasal constituent:
(4) a. Taj čovjek je volio Milenu
that man is loved Milena
‘That man loved Milena.’
b. Taj je čovjek volio Milenu.
In (4a), where a whole phrase precedes the clitic, syntactic movement could
provide a host for the clitic, which is assumed to be located outside IP under
syntactic approaches to 2 cliticisation. Halpern argues that in 1 environments
such as (4b), where the clitic appears to break up a phrasal constituent, PI
provides a host for the clitic. According to Halpern, the clitic is sentence initial
in the output of the syntax. PI then takes place in the phonology placing the
clitic after the first stressed word, namely taj.
(5) a. Syntax: je taj čovjek volio Milenu
b. Phonology: Taj je čovjek volio Milenu
Progovac (1996) and Wilder and Ćavar (1994a), however, show that the 1/1
dichotomy with respect to clitic placement in (4) is in fact an artifact of the
general possibility of separating SC determiners from nouns in the syntax, as
illustrated by (6a–b), which cannot be derived by PI and must involve syntactic
movement (left-branch extraction) of the determiner. In other words, Progovac
and Wilder and Ćavar argue that in (4b) we are dealing with 1 placement, with
the phrase preceding the clitic being located in front of the clitic at SS after
undergoing left-branch extraction. (Note that SC determiners are morphologically
adjectives, see Corver 1992.)
(6) a. Kojeg/Tog Milena voli čovjeka
which/that Milena loves man
‘Which man does Milena love?’/‘Milena loves that man.’
b. Kojeg/Tog je Milena volila čovjeka
which/that is Milena loved man
Progovac (1996) and Wilder and Ćavar (1994a) argue that only elements that can
be base-generated in front of clitics or can be independently shown to be able to
undergo syntactic movement can precede SC clitics within their clause. As a
result, they conclude, there is no need to appeal to phonological movement to
 SECOND POSITION CLITICISATION 77

account for 2 cliticisation in SC. To illustrate their point, in (7) I give two
more examples that appear to involve a clitic breaking up a phrasal constituent,
which are candidates for PI on the assumption that PI can, but syntactic move-
ment cannot do this. As shown in (8a–b), which cannot be derived by PI, it is
possible to show in all the relevant cases that the element preceding the clitic in
(7a–b) is capable of undergoing syntactic movement and therefore could be
placed in front of the clitic in (7a–b) by syntactic movement.
(7) a. Anina je sestra došla
Ana’s is sister arrived
‘Ana’s sister arrived.’
b. Zeleno je auto kupio.
green is car bought
‘He bought a green car.’
(8) a. Čija/Anina dolazi sestra
whose/Ana’s arrives sister
b. Čija/Anina je došla sestra
whose/Ana’s is arrived sister
‘Whose/Ana’s sister arrived’
c. Kakvo/zeleno Jovan kupuje auto
what-kind/green Jovan buys car
d. Kakvo/zeleno je Jovan kupio auto
what-kind/green is Jovan bought car
‘What kind of a car did Jovan buy?’/‘Jovan bought a green car.’
In (9a) we have an element that apparently cannot undergo syntactic movement.
As shown in (9b) and discussed by Progovac (1996) and Wilder and Ćavar
(1994a), although it is stressed, the element in question also cannot precede
clitics.8
(9) a. *Prema Milan i Jovan idu Mileni
toward Milan and Jovan walk Milena
‘Milan and Jovan are walking toward Milena.’
b. *Prema su Mileni Milan i Jovan išli
toward are Milena Milan and Jovan walked
‘Toward Milena Milan and Jovan walked.’
c. cf. Prema Mileni su Milan i Jovan išli

8. Most SC prepositions are lexically unaccented (they procliticise to the following stressed word) and
therefore cannot host clitics, which need a phonologically strong host. However, prema is accented.
78 ŽELJKO BOŠKOVIĆ

Under the PI analysis, the ungrammaticality of (9b) is surprising. It should be

possible for the syntax to provide the following output to PF:
(10) su prema Mileni Milan i Jovan išli
PI should then be able to apply to (10) placing the clitic after prema, thus
incorrectly deriving (9b.)9
Progovac gives several other examples where she claims that both syntactic
movement in front of a clitic and 1 placement fail, which raises the same
problem for the PI analysis. Consider (11). (The judgements in (11a–d) are
Progovac’s 1996, personal communication). As discussed below, there is some
variation with respect to (11a–c).)
(11) a. *Roditelji dolaze uspešnih studenata
parents arrive successful students
‘Parents of successful students are arriving.’
b. *Roditelji su došli uspešnih studenata
parents are arrived successful students
‘Parents of successful students arrived.’
c. *Roditelji su se uspešnih studenata razišli
parents are self successful students dispersed
‘Parents of successful students dispersed.’
d. cf. Roditelji uspešnih studenata su se razišli
e. SS: su se roditelji uspešnih studenata razišli
Under the PI analysis, (11c) could be derived by applying PI to (11e). (11c) is
thus incorrectly predicted to be good in the relevant dialect on the PI analysis,
but not on the syntactic movement analysis, which allows only elements that can
be placed in front of a clitic in the syntax to host it.
Interestingly, in my dialect, (11c) is acceptable, though somewhat marginal.
Significantly, the same holds for (11a–b). The correlation between syntactic
extractability and the ability to host a clitic thus holds. In fact, the dialectal split
provides a very strong confirmation of the generalisation that only elements that
can be placed in front of clitics by syntactic movement (or be base-generated in
front of clitics) can precede clitics in SC. (I am not aware of any dialect that
would have a difference between cliticisation and syntactic movement with
respect to the constructions under consideration.)

9. Halpern does attempt to provide an account of this kind of examples. However, Schütze (1994)
shows that the account is seriously flawed both empirically and conceptually.
 SECOND POSITION CLITICISATION 79

An interesting confirmation of the conclusion is provided by certain facts

discussed by Franks (1997). In SC it is possible in some cases to inflect for
structural case either one or both names in first+last name complexes. (Nomina-
tive is the default case in (12)–(14).)
(12) a. Lava Tolstoja čitam
Lav- Tolstoy- read
‘Lav Tolstoy I read.’
b. ?Lava Tolstoj čitam
Lav- Tolstoy- read
c. Lav Tolstoja čitam
Lav- Tolstoy- read
The first name can be separated from the last name by movement only when
both names are inflected.
(13) a. Lava čitam Tolstoja
b. *Lava čitam Tolstoj
c. *Lav čitam Tolstoja
Significantly, cliticisation patterns with movement in this respect.10
(14) a. Lava sam Tolstoja čitala
Lav- am Tolstoy- read
b. *Lava sam Tolstoj čitala
Lav- am Tolstoy- read
c. *Lav sam Tolstoja čitala
Lav- am Tolstoy- read
Franks observes that this pattern is expected under analyses that allow only
elements that can be base-generated or syntactically moved in front of a clitic to
precede the clitic. The pattern, however, raises a serious problem for the PI
analysis. Under this analysis we would expect all of the constructions in (14) to
be acceptable, since nothing blocks the derivation in which the names remain in
SpecIP overtly, with the clitic being located above the subject (C under most PI
analyses). PI would then apply in PF placing the clitic after the first name, the
first stressed word following the clitic, thus deriving (14a–c).
In conclusion, if only elements that can be independently shown to be able
to undergo syntactic movement can precede and host clitics in SC (from now on

10. Franks notes this with respect to examples in which only the first name is inflected. Nothing,
however, changes if only the second name is inflected.
80 ŽELJKO BOŠKOVIĆ

I will disregard elements that are base-generated in front of clitics), the PI

analysis should be rejected since it fails to capture this generalisation.
Schütze (1994), however, argues that the generalisation is incorrect.
Following up on some data discussed by Percus (1993), he gives one example in
which he claims the clitic host cannot be located in front of the clitic in the
syntax, since it can be independently shown that the relevant element is syntact-
ically immobile. Schütze bases his argument on the following contrast:
(15) a. U ovu je veliku sobu ušao
in that is big room entered
‘He entered that big room.’
b. ?*U ovu Jovan ulazi veliku sobu
in that Jovan enters big room
Schütze interprets the ungrammaticality of (15b) as indicating that in a sequence
P+Det+Adj+N, P+Det cannot be split from the rest of the sequence by syntactic
movement.11 Since P+Det precedes the clitic in (15a), we appear to have here
evidence that elements that cannot undergo syntactic movement can still precede
and host SC clitics, as expected under the PI analysis. Under this analysis, the
syntax could have (16a) as its output, with PI placing the clitic after the first
phonological word in PF. (Note that the preposition u is not stressed and
therefore not a phonological word.)
(16) a. Syntax: je u ovu veliku sobu ušao
b. PF: u ovu je veliku sobu ušao
The pair in (15) is generally cited as the strongest empirical evidence for PI in
SC. It is therefore important to examine it carefully.
Notice first that (15a–b) do not form a minimal pair in the relevant respect.
Whereas in (15a) the remnant of the split PP precedes the verb, in (15b) it
follows it. This is an important interfering factor, since these kinds of splits are
generally better when the remnant precedes the verb. (15b) actually improves if
the remnant is placed in front of the verb, which produces a closer minimal pair
with (15a).
(17) ??U ovu Jovan veliku sobu ulazi
in that Jovan big room enters

11. This is in contrast to P+Det in P+Det+N sequences.

(i) U ovu/u koju Jovan ulazi sobu
in that/in which Jovan enters room
 SECOND POSITION CLITICISATION 81

(15a) still appears to be somewhat better than (17), though the contrast is not
conclusive now. Notice also that (17) improves even more if Jovan is replaced
by a pronoun that is not a 2 clitic.
(18) (?)U ovu on veliku sobu ulazi
in that he big room enters
Since the pronoun in (18) is not an enclitic, there is no need to apply PI. The PI
analysis of (15a), then, cannot be extended to (18). The lack of a clear contrast
between (15a) and (18) thus remains unaccounted for under the PI analysis. (If
there is any contrast, it is very weak.)
Notice also that (15a) itself becomes degraded if the clitic is replaced by a
clitic cluster. (19) differs little, if at all, from (17) and is worse than (18). This
is unexpected under the PI analysis. Under this analysis we would not expect any
difference between (15a) and (19), since PI would be expected to affect the
whole clitic cluster (i.e. each clitic in the cluster). We would, however, expect
(19) to be better than (17) and (18).
(19) ??U ovoj si mi ga se velikoj sobi zasitio
in that are me it self big room fed-up
‘You got fed up with it in this big room.’
Given these facts, I think that we can safely conclude that the data concerning PP
splits provide no evidence for PI. It seems to me that the following scenario is
the most plausible here: Extraction of u ovoj out of the PP u ovoj velikoj sobi is
not allowed. When the material intervening between the parts of the split PP is
not very prominent phonologically (length and stress being relevant here), the
saliency of the split decreases. The sentence is then parsed as if it did not involve
a PP split. When we increase the saliency of the split by making the intervening
material phonologically more prominent (longer and/or stressed), this parse
becomes unavailable and the sentence is parsed as a real instance of movement
out of the PP, which is not allowed.12 If this analysis is on the right track, the

12. Under the most natural pronunciation, Jovan in (17) and the clitic cluster in (19) are followed
by a small pause, which causes the following adjective to be pronounced with a slightly higher pitch
and a slightly stronger stress, which is not the case with (15a), and does not have to be the case with
(18). It seems that the combined effect of the pause, pitch, and stress increases the saliency of the
split, which is confirmed by the fact that playing with the relevant prosodic properties (adding or
eliminating the combined effect of the pause, pitch, and stress) affects the acceptability of the
constructions under consideration. Thus, a small pause after the clitic, accompanied by the
pitch+stress effect, makes even (15a) bad.
Notice also that (i), where the PP split is most salient, is consistent with the above analysis.
82 ŽELJKO BOŠKOVIĆ

data in question in fact provide further evidence against PI since they show that
elements that cannot undergo movement syntactically cannot host clitics.13
As noted above, traditional 1 contexts provide a potential source of
evidence for PI. Browne (1975) argues that one such context involves clauses of
the form to be+predicate. He argues that such constructions allow only 1
placement based on examples such as (20a–f). I will examine such constructions
in some detail here since Schütze (1994) argues that they provide evidence for
PI. (I add lexical material following the auxiliary clitic since locating clitics in
clause-final position is often disfavoured.)
(20) a. *Prodao novine je (juče)
sold newspapers is (yesterday
‘He sold newspapers yesterday.’
b. Prodao je novine (juče)
c. *Sposoban direktor je (on)
capable manager is (he
‘He is a capable manager.’
d. Sposoban je direktor (on)
e. *Jako dosadni su (oni)
extremely boring are (they
‘They are extremely boring.’
f. Jako su dosadni (oni)
The fact that the complement of a clitic cannot be preposed could be interpreted
as an ECP-type phenomenon; i.e. it could be taken to indicate that clitics are too
weak to license their complement with respect to the ECP (the head government
part of the conjunctive ECP).14 When the clitic auxiliary is replaced by a full
non-clitic form of the auxiliary, (20a, c, e) become good, which on this analysis
would be interpreted as indicating that full non-clitic forms of auxiliaries are
proper governors, i.e. they can license their complement with respect to the ECP.
(This implies that full non-clitic forms are independent lexical items, as in

(i) ???U ovu si vjerovao da je veliku sobu ušao

in that are believed that is big room entered
‘You believed that he entered that big room.’
13. A number of interesting questions independent of 2 cliticisation remain to be answered
concerning extraction of P+Det, or, more generally, P+Adj sequences (recall that SC determiners are
morphologically adjectives). For some relevant discussion, see Franks and Progovac (1995).
14. See also Tomić (1996). Of course, it still remains to be seen how ECP-type phenomena are to
be captured in the current framework, which has no natural place for the notion of government.
 SECOND POSITION CLITICISATION 83

Bošković 1995 and King 1996, and not derived by incorporating a clitic into
another element, as in Rivero 1991; Tomić 1996 and Wilder and Ćavar 1994a.)15
(21) a. Prodao novine jeste/nije
sold newspapers is/is+not
‘He DID/did not sell newspapers yesterday.’
b. Sposoban direktor jeste/nije
capable manager is/is+not
‘He IS/is not a capable manager.’
c. Jako dosadni jesu/nisu
extremely boring are/are+not
‘They ARE/are not extremely boring.’
As for (20b), as shown in Bošković (1995, 1997a), such constructions simply
involve adjunction of the participle to the auxiliary, located in its base-generated
position. (The head-government part of the conjunctive ECP can be satisfied by
the moved element itself, an X0.) The adjunction is driven by feature-checking.
It is in fact argued in Bošković (1997a) that cross-linguistically, the head of the
complement of a modal or an auxiliary must adjoin to the modal/auxiliary,
languages differing only with respect to the level at which this adjunction takes
place. (See Bošković 1997a for discussion how this analysis applies to SC. For
relevant discussion, see also Wilder and Ćavar 1994b.)
Consider (20d) now. We could also be dealing here with adjunction of the
head of the complement of the auxiliary to the auxiliary under Abney’s (1987)
analysis, in which Adj takes NP as its complement. Another possibility (if
Abney’s analysis is not adopted) is phrasal left-branch extraction, in which case
we would not be dealing here with 1 placement at all. Evidence for this analysis
is provided by (22), where the element preceding the clitic is clearly an XP.16
(22) Izuzetno sposoban je (on) direktor
extremely capable is (he manager

15. According to Wilder and Ćavar (1994a), some speakers allow VP preposing with clitic
auxiliaries, but this appears to be the case only when the auxiliary is the 3.p.sg je. This is not
surprising, since it is well known that at least for some speakers, je is losing its clitichood (for
relevant discussion, see Tomić 1996).
16. Under the left-branch extraction analysis the extracted element could be licensed with respect
to the head government part of the ECP within the predicate, since it would not be a complement
of the auxiliary.
84 ŽELJKO BOŠKOVIĆ

Either way, the grammaticality of (20d), which contrasts with (20c), cannot be
used as evidence for PI. The PI derivation, on which (20d) would be analysed as
having the sequence je sposoban direktor as the output of overt syntax with PI
placing the clitic following sposoban in PF, is clearly not the only way of
analysing (20d). (Note also that (22) cannot be derived by PI.) The same holds
for (20f), which could be analysed in the same way as (20d). Notice furthermore
that when the first word of the predicate is syntactically immobile, it is not
allowed to precede the clitic auxiliary. Thus, speakers who find (11a–b) ungram-
matical also find (23a) ungrammatical. On the other hand, speakers who accept
(11a–b) also accept (23a). This strongly indicates that when a part of a predicate
precedes the clitic auxiliary, it is placed there by syntactic movement, not by PI.
(23) a. *Roditelji su uspešnih studenata
parents are successful students
‘They are parents of successful students.’
b. cf. Oni su roditelji uspešnih studenata
they are parents successful students
It is often claimed that when a predicate contains only one word, it can be
moved in front of the auxiliary, in contrast to (20a, c, e), where the predicate
contains more than one word.
(24) a. Zaspao je
fallen-asleep is
‘He fell asleep.’
b. Pametan je
clever is
‘He is clever.’
This is unexpected under the analysis sketched above. However, there is no need
to analyse (24a–b) as involving predicate preposing. The constructions could
simply involve adjunction of the head of the complement of the auxiliary to the
auxiliary, just like (20b). (See Bošković 1997a and the discussion above. See
also Lasnik 1995, who argues that auxiliary be in English constructions such as
John is smart is a light verb, so that the lower predicate must head-adjoin to it.
This happens in LF in English, but apparently can happen in overt syntax (at
least as an option) in SC.)
As for constructions such as (25), noted by Browne (1975), these could be
analysed as involving adjunction of the noun to the preposition, or adjunction of
the preposition, which is a (non-2) proclitic, to the noun (this would involve
lowering), followed by adjunction of the whole complex to the auxiliary. The last
 SECOND POSITION CLITICISATION 85

movement would take place for the same reason as in (24) and (20b).17
(25) U sobi su
in room are
‘They are in the room.’
That (25) involves head movement rather than fronting of the predicate phrase is
indicated by (26).
(26) *U drugoj sobi su
in other room are
‘They are in the other room.’
In conclusion, careful examination of split PP constructions such as (15) and to
be+predicate constructions such as (20), which appeared to be the only real
empirical evidence for PI in SC, has shown that they do not provide any
empirical support for PI. Given the lack of empirical evidence, as well as a
number of empirical problems that PI faces, the PI analysis must be rejected.
Since PI is the cornerstone of the weak syntax account, this means that if the
syntactic approach is to be maintained we are left with the strong syntax
approach: the syntax should do the whole job without appealing to any phonolog-
ical operations. I therefore concentrate on this approach from now on.

3.2 The strong syntax account

As noted above, the strong syntax account is crucially based on the assumptions
in (2). The idea behind pushing clitics high in the tree is to ensure that there is
not enough space for too many elements to precede clitics within their sentence.
Of course, it is also necessary to ensure that clitics do not occur sentence
initially. There are two ways of accomplishing this in the strong syntax approach
to 2 cliticisation. I will call these the filtering and the non-filtering strong
syntax approach.

17. The grammaticality of (i) indicates that there is a very close connection between the preposition-
al clitic and the following word in overt syntax.
(i) U veliku tvrdiš da ulazi sobu
in big claim that enters room
‘You claim that he is entering the big room.’
It is possible that the preposition adjoins to the following head in the syntax so that it is affected by
any XP movement that the maximal projection of this head undergoes. For analyses along these lines
for Polish, see Borsley and Jaworska (1988) and Corver (1992).
86 ŽELJKO BOŠKOVIĆ

3.2.1 Two types of strong syntax accounts

A non-filtering strong syntax account is proposed by Roberts (1994) (see also
Dimitrova-Vulchanova 1995 and Rivero 1997). Roberts assumes that SC clitics
are located in the head position of Voice Phrase, which functions as the comple-
ment of C. To account for the 2 effect, he posits a strong feature in SC matrix
C, which can be checked by either a head or a phrase of any type. SC embedded
C does not have this checking requirement. The reason why we find only one
element in front of clitics is strictly syntactic: one and only one element needs to
move in front of a clitic to check the strong feature of matrix C. In embedded
clauses nothing needs to move since lexically realised complementisers that
introduce embedded clauses in SC, like da ‘that’, do not have this checking
requirement. This account seems rather ad hoc. In particular, positing a feature
that can be checked by just about anything, and either an XP or a head, is very
ad hoc. I am not aware of any feature of this kind in any other language.
Filtering strong syntax accounts allow some phonological information, in
particular the enclitic status of the relevant elements, to be taken into consider-
ation. One such account is proposed by Progovac (1996) (see also Rivero 1991,
1994 and Wilder and Ćavar 1994a, b).
Progovac (1996) argues that one element needs to move in front of a clitic
to ensure that the clitic is not stranded in sentence initial position. This is the
motivation for movement in front of a clitic in constructions where no indepen-
dently motivated movement operation, such as wh-movement, or lexical insertion
(of an overt complementiser), locates lexical material in front of the clitic. (27a)
is then ruled out by the Last Resort Condition, which prohibits superfluous
operations. Since wh-movement has provided a host for the clitic, there is no
need for participle preposing, the only motivation for the operation, according to
Progovac, being to provide a host for clitics.18

18. There is actually plenty of evidence against the Last Resort view of participle preposing (see
Bošković 1995 and Embick and Izvorski 1997). Notice, for example, that participles can optionally
precede non-clitic auxiliaries. Since the auxiliary in (i) is not a clitic, we cannot be dealing here with
a last resort movement driven by the need to provide a host for a clitic.
(i) a. Zaspao bejaše Petar
fallen-asleep be Peter
‘Peter had fallen asleep.’
b. Bejaše zaspao Petar
Participles can also optionally precede auxiliary clitics in Slovenian, Macedonian, and Czech.
However, clitics in these languages can appear sentence initially, which indicates that they are not
necessarily enclitics. The grammaticality of Slovenian (iib) then provides evidence that participle
preposing in (iia) cannot be a last resort operation driven by the need to provide a host for clitics.
(ii) a. Videl sam ga
 SECOND POSITION CLITICISATION 87

(27) a. *Koga poljubio je Ivan?

who kissed is Ivan
‘Who did Ivan kiss?’
b. Koga je poljubio Ivan?
In this analysis, syntactic movement can be motivated by purely phonological
considerations, more precisely, by providing a host for a phonologically weak
element. If this were not the case, even (28) would be ruled out by the Last
Resort Condition in the syntax, because syntactically, participle preposing is
completely superfluous under Progovac’s analysis.
(28) Zaspao je Ivan
fallen-asleep is Ivan
‘Ivan fell asleep.’
If we accept Progovac’s analysis, we thus need to enable syntax to look ahead
to the needs of phonology. This introduces considerable globality into the system.
In the derivational model which Progovac adopts, it would certainly be preferable
to achieve the desired result without having syntax look ahead to the needs of
phonology. (In this particular respect, the PI analysis is conceptually more
appealing, since the rescuing movement operation takes place in the same
component in which the deficiency driving the movement operates, namely PF,
so that no look-ahead is needed.)
It should be emphasised here that in Progovac’s analysis, as well as other
filtering strong syntax analyses and all PI analyses, (29), with a sentence initial
clitic, is syntactically well-formed.
(29) *Je poljubio Anu
is kissed Ana
‘He kissed Ana.’
The only way to rule out (29) under Progovac’s analysis is to assume that it
violates the phonological requirement that SC clitics are suffixes. Progovac’s
approach is thus not strictly syntactic: It requires phonology to have a filtering
effect on syntax by ruling out some well-formed syntactic representations due to
violations of phonological requirements on clitics. We will see below that once
we fully embrace this filtering role of phonology, we can eliminate the globality
required in Progovac’s analysis.

seen am him
‘I saw him.’
b. Sam ga videl
88 ŽELJKO BOŠKOVIĆ

We have seen above how the strong syntax account of 2 cliticisation
works. Having examined some conceptual problems with this account I now turn
to the assumptions in (2a–c), the empirical backbone of the strong syntax
account.19 Bošković (1995, 1997a) and Stjepanović (1998a, b) have already
questioned the validity of (2a–c). Their arguments will be summed up in the next
section, where I also present new evidence against (2a–c). The general conclu-
sion that the data discussed in the next section will lead me to is that, conceptual
problems aside, current strong syntax accounts of 2 cliticisation must be
rejected on empirical grounds: they are simply empirically fatally flawed.

3.2.2 Empirical evidence against the strong syntax approach

3.2.2.1 Adverbs. Evidence that 2 clitics do not have a structurally fixed
position comes from the distribution of sentential adverbs. (30a) shows that the
adverb pravilno ‘correctly’ is ambiguous between the manner and the sentential
adverb reading. Only the former reading is available when the participle precedes
the adverb, which indicates that the landing site of participle preposing is below
sentential adverbs.
(30) a. Jovan je pravilno odgovorio Mileni
Jovan is correctly answered Milena
‘Jovan did the right thing in answering Milena.’ or ‘Jovan gave
Milena a correct answer.’
b. Odgovorio je pravilno Mileni
‘He gave Milena a correct answer.’ or *‘He did the right thing
in answering Milena.’
The fact that (30a) is ambiguous between the manner and the sentential readings
of the adverb shows that when preceded by an XP, SC clitics can be higher not
only than the attachment site of manner adverbs, which are generally assumed to
be adjoined to VP (or AgrOP under the Split I Hypothesis), but also higher than
the attachment site of sentential adverbs, which Bošković (1995) and Watanabe
(1993) argue are adjoined to TP. The non-ambiguity of (30b) shows that, when
preceded by a participle, SC clitics cannot be higher than sentential adverbs.
(30a–b) thus indicate that SC clitics are not always located in the same structural
position. In (30a) the clitic must be higher than TP, the attachment site of

19. Recall that (2a–c) are also needed under the weak syntax account, where they are implemented
in a slightly different fashion. Since this account has already been shown to be untenable I will
confine myself to examining the validity of (2a–c) under the strong syntax account.
 SECOND POSITION CLITICISATION 89

sentential adverbs, and in (30b) it must be below TP. Adopting the strong syntax
account assumption that SC clitics are always located in the same position would
get us into a paradox with respect to the data in (30). (30a) would have to be inter-
preted as indicating that the position is higher than sentential adverbs and (30b)
would have to be interpreted as indicating that the position is lower than sentential
adverbs. Based on constructions such as (30a–b), I argued in Bošković (1995)
that there is no fixed structural position for clitics in SC (see Bošković 1995 for
another argument to this effect based on double participle constructions).
(30b) also provides evidence against the assumption that SC clitics are
always located very high in the tree. Recall that under the strong syntax account
placing clitics very high in the structure is a way of ensuring that there is not
enough space for more than one element to precede SC clitics within their clause.
Since most proponents of the strong syntax approach assume that SC clitics are
located under C, I will take this position as the representative of this approach.

3.2.2.2 Participle movement: Participles are not in C. Given that sentential

adverbs clearly must be able to occur below C (cf. Jovan će vjerovatno otići ‘Jovan
will probably leave’) (30b) also provides evidence that SC clitics can be lower
than C. In fact, I show in Bošković (1995) that the auxiliary clitic in
participle+auxiliary clitic constructions is located very low in the tree: It is located
in the V node where it is generated, with the participle being adjoined to it.
(31) [VP Odgovorioi je [VP ti Mileni]]
answered is Milena
‘He answered Milena.’
Another empirical argument that SC participles cannot move as high as C, as a
result of which clitics that follow participles must be lower than C, is provided
by the question particle li, a 2 clitic that is standardly assumed to be located
under C.20 Consider (32):
(32) a. Ljubi li nju?
kisses  her
‘Does he kiss her?’
b. *Poljubio li je nju?
kissed  is her
‘Did he kiss her?’

20. There is also an archaic, non-interrogative usage of li, which I ignore here since it is not clear
whether li is located under C on this usage.
90 ŽELJKO BOŠKOVIĆ

The grammaticality of (32a), involving a finite verb serving as a host for the 2
clitic li, shows that finite verbs can move to C in SC. On the other hand, the
ungrammaticality of (32b) indicates that participles cannot move to C in SC. This
is not surprising, given that cross-linguistically, in languages in which V-to-C
movement clearly takes place in finite clauses, only finite verbs move to C,
which indicates that finiteness motivates the movement. Given that SC participles
cannot move to C, it follows that clitics following a participle cannot be located
under C.
It is worth mentioning here that the blocking effect of complementisers on
participle preposing is sometimes taken as evidence that the landing site of parti-
ciple movement is C (see Rivero 1991 and Wilder and Ćavar 1994a, among others).
(33) a. Otišao je juče
left is yesterday
‘He left yesterday.’
b. *Jovan misli da otišao je juče
John thinks that left is yesterday
c. cf. Jovan misli da je otišao juče
Jovan thinks that is left yesterday
The conclusion is clearly unwarranted. Note that under this line of reasoning we
are led to interpret the blocking effect of subjects (34a) on participle movement
as evidence that the participle lands in SpecIP, and the blocking effect of wh-
phrases on participle movement (35a) as evidence that the participle lands in
SpecCP, which is clearly undesirable.
(34) a. *Jovan otišao je juče
Jovan left is yesterday
b. cf. Jovan je otišao juče.
Jovan is left yesterday
‘Jovan left yesterday.’
(35) a. *Zašto otišao je juče?
why left is yesterday
b. cf. Zašto je otišao juče?
why is left yesterday
‘Why did he leave yesterday?’
The ungrammaticality of (33b) simply illustrates the 2 effect. As shown in
Bošković (1995), when the auxiliary is not a clitic, the presence of an overt
complementiser does not block participle movement, which indicates that the
participle does not land in C.
 SECOND POSITION CLITICISATION 91

(36) On tvrdi da istukao bejaše Petrovog prijatelja

he claims that beaten was Peter’s friend
‘He claims that he had beaten Peter’s friend.’
Notice also that, as observed by Embick and Izvorski (1997), Bulgarian (37),
which contains a clitic auxiliary that is not a 2 clitic, is good, which confirms
this conclusion.
(37) Razbrah če pročel e knigata
understood that read had book-the
‘I understood that he had read the book.’

3.2.2.3 Wh-superiority. Returning to 2 cliticisation in SC, we provide another

argument that SC clitics do not have to be located under C. This comes from the
syntax of wh-questions in SC. Rudin (1988) claims that multiple questions in SC
are not sensitive to the Superiority Condition. However, in Bošković (1997b, in
press) I show that SC does exhibit Superiority effects in certain contexts. In
particular, SC exhibits Superiority effects exactly in those contexts in which
French must have wh-movement, namely, in long-distance questions, embedded
questions, and root questions with lexical complementisers.21
(38) a. Ko je šta prodao?
who is what sold
‘Who sold what?’
b. Šta je ko prodao?
(39) a. [Ko koga voli], taj o njemu i govori
[who whom loves that-one about him even talks
‘Everyone talks about the person they love.’
b. *[Koga ko voli], taj o njemu/o njemu taj i govori
(40) a. (?)Ima
ko šta da ti proda
has who what that you sell
‘There is someone who can sell you something.’
b. *Ima šta ko da ti proda

21. I ignore here the irrelevant echo-question reading. Note that I do not give indirect questions as
examples of embedded questions in SC because such questions involve an interfering factor. Indirect
questions formally do not differ at all from matrix questions in SC. As a result, there is always a
danger that they could be analysed as matrix questions, with the superficial matrix clause treated as
an adsentential. The problem does not arise with the correlatives in (39) and the existentials in (40),
which also contain embedded questions (see Izvorski 1996, 1998). Note also that overt C questions
are not accepted in all dialects of French.
92 ŽELJKO BOŠKOVIĆ

(41) a. Ko si koga tvrdio da je istukao?

who are whom claimed that is beaten
‘Who did you claim beat whom?’
b. *Koga si ko tvrdio da je istukao?
(42) a. Ko li šta kupuje?
who C what buys
‘Who on earth buys what?’
b. *Šta li ko kupuje?
(43) Tu as embrassé qui?
you have kissed who
‘Who did you kiss?’
(44) a. *Pierre a demandé tu as embrassé qui
Peter has asked you have kissed who
b. Pierre a demandé qui tu as embrassé
(45) a. *Jean et Marie croient que Pierre a embrassé qui?
John and Mary believe that Peter has kissed who
b. cf. Qui Jean et Marie croient-ils que Pierre a embrassé?
(46) a. *Que tu as vu qui?
C you have seen who
‘Who did you see?’
b. Qui que tu as vu?
To account for the parallelism between the contexts in which SC exhibits
Superiority effects and the contexts in which French must have wh-movement,
I proposed in Bošković (1997b, in press) that SC is a French-type language with
respect to when it must have wh-movement: Like French, SC must have
wh-movement in long-distance, embedded, and overt C root questions, but not in
short-distance null C root questions. SC wh-movement is then well-behaved with
respect to Superiority: SC exhibits Superiority effects whenever it has wh-
movement. The only difference between SC and French is that even the wh-
phrases that do not move overtly to SpecCP still must be fronted in SC. This
also holds for echo wh-phrases (note the ungrammaticality of *On kupuje šta ‘He
buys what’ on the echo question reading), which indicates that the fronting is
independent of the +wh-feature. Following Stjepanović (1995), I argue that SC
wh-phrases not located in SpecCP overtly must move to a special focus position,
located above VP but below the CP projection (see Bošković 1998 for an
explanation why this focus movement does not exhibit Superiority effects). I
argue that the CP projection does not even have to be present overtly in (38) and
 SECOND POSITION CLITICISATION 93

(43). Since the complementiser is phonologically null and located at the top of
the tree nothing in the current framework prevents it from entering the structure
in LF in such constructions, given Chomsky’s (1995) derivational approach to
strength, where strong features are defined as elements that must be eliminated
immediately upon insertion into the structure. (I assume that the interrogative C
in SC and French has a strong +wh-feature.) The reason why matrix short-
distance null C questions in SC and French do not have to involve overt
wh-movement is then trivial: Its trigger (C) does not have to be present overtly.
I argue that the LF C-insertion derivation is the only way for French and SC to
avoid overt wh-movement. (Overt C-insertion triggers overt wh-movement.) In
constructions in which wh-movement is forced the derivation is blocked. With
embedded questions, the LF C-insertion derivation is blocked because it would
involve merger of the complementiser in an embedded position, which is
disallowed, Merge being allowed to take place only at the top of the tree. With
overt complementisers, the derivation is blocked because phonologically overt
elements cannot enter the structure in LF. If they do, the derivation crashes due
to the presence of phonological information in LF. I also show that with long-
distance questions, the LF C-insertion derivation fails. For an explanation, which
is a bit more involved, see Bošković (1997b, in press).
In conclusion, in French and SC matrix short-distance null C questions the
interrogative CP projection can be inserted in LF. As a result, wh-movement (i.e.
movement to SpecCP) does not have to take place overtly in such questions. This
is what ‘licenses’ wh-in-situ in French, and explains the lack of Superiority
effects in the relevant constructions in SC, which needs to front all wh-phrases
independently of the [+wh]-feature. Given this analysis, we are led to the
conclusion that the wh-phrase in constructions such as (38b) must be located in
a position that is lower than the CP projection overtly. It follows then that the
clitic must be lower than C too.

3.2.2.4 VP ellipsis. Having seen evidence against assumptions (2b) and (2c), let
us turn to (2a). Evidence against the assumption that SC clitics cluster under the
same node in the syntax is provided by Stjepanović (1998a, b), based on VP
ellipsis. Stjepanović shows that constructions like (48) involve VP ellipsis:
(47) Oni su kupili novine, a i vi ste (takodje)
they are bought newspapers and also you are (too
‘They bought newspapers and you did too.’
She observes that VP ellipsis in SC can delete part of a clitic cluster, leaving
some clitics behind.
94 ŽELJKO BOŠKOVIĆ

(48) Mi smo mu ga dali, a i vi ste (takodje)

we are him it given and also you are (too
‘We gave it to him, and you did too.’
The possibility of VP ellipsis in (48) raises a serious problem for the assumption
that clitics cluster under the same node in SC, which is crucial to the strong
syntax account of the 2 effect. Under this assumption, (48) has to involve
deletion of a non-constituent, which is standardly assumed not to be possible.
Stjepanović interprets the grammaticality of (48) as evidence that SC clitics do
not have to be located under the same node in the syntax, i.e. they can be located
in separate maximal projections. (48) can then be analysed as involving constitu-
ent deletion. (See Section 4.2. for another argument that SC clitics do not have
to cluster together in the syntax.)
Stjepanović notes that (48) also provides evidence that SC clitics can be
located fairly low in the tree. If they had to be located very high in the tree, as
is assumed under the strong syntax account, we would not expect it to be
possible for them to be affected by the process of VP deletion.
In summary, we have seen that none of the assumptions in (2a–c), which
are crucial to the strong syntax account of the 2 effect (without any of them
the account would fold), is warranted: SC clitics do not have to cluster in the
syntax, they do not have a fixed structural position, and they do not have to be
located under C. In the light of this all, I conclude that the strong syntax account
must be rejected. In the reminder of this section I will give an additional
argument against the syntax account which will also provide us with a clue
where to look for a solution to the 2 cliticisation puzzle.

3.2.2.5 Delayed clitic placement. It is well known that certain elements, such as
appositives, fronted heavy constituents, and parentheticals, can delay clitic
placement. This is shown by (49)–(52), where the clitics occur in the third and
the fourth position of their clause. (For discussion of delayed clitic placement,
see Bennett 1986; Bošković 1995; Browne 1974, 1975; Halpern 1995; Percus
1993; Progovac 1996; Radanović-Kocić 1988, 1996; Schütze 1994; Tomić 1996;
Ćavar and Wilder 1994; and Zec and Inkelas 1990.)
(49) Sa Petrom Petrovićem srela se samo Milena
with Peter Petrović, met self only Milena
‘With Peter Petrović, only Milena met.’
(50) Znači da, kao što rekoh, oni će sutra doći
means that as said they will tomorrow arrive
‘It means that, as I said, they will arrive tomorrow.’
 SECOND POSITION CLITICISATION 95

(51) Ja, tvoja mama, obećala sam ti sladoled

I your mother promised am you ice cream
‘I, your mother, promised you an ice cream.’
(52) Prije nekoliko godina sa Petrom Petrovićem srela se samo Milena
before several years with Peter Petrović met self only Milena
‘A few years ago, with Peter Petrović, only Milena met.’
The distribution of SC 2 clitics, illustrated above, can be stated in very simple
prosodic terms:
(53) SC clitics occur in the second position of their intonational phrase.
Nespor and Vogel (1982, 1986), Selkirk (1986), and Hayes (1989), among
others, have proposed a hierarchical theory of the prosodic structure, which is
determined by, but does not completely correspond to, the syntactic structure of
the sentence. The units of this prosodic structure from word level up are:
prosodic (phonological) word, phonological phrase, intonational phrase (I-phrase),
and utterance.22 I assume that, with some possible exceptions that need not
concern us here, unless interrupted by a special element that forms a separate
intonation domain, each clause is mapped to a single I-phrase. More precisely,
the left edge of a CP corresponds to an I-phrase boundary. Certain elements,
such as appositives, parentheticals, and heavy fronted constituents, are special in
that they form separate I-phrases, evidence for which is provided by the fact that
they are followed by pauses. Under the most natural pronunciation clitic second
constructions such as (28), repeated here as (53), then contain only one I-phrase.
(53) Zaspao je Ivan
fallen-asleep is Ivan
‘Ivan fell asleep.’
In (49)–(52), on the other hand, there is more than one I-phrase, since the
appositive in (51), the fronted heavy constituents in (49) and (52), and the
parenthetical in (50) form separate I-phrases. This means that a new I-phrase
starts after these elements. Note that the elements in question are obligatorily

22. There is disagreement on whether the clitic group exists as a prosodic unit, an issue on which
I remain silent here. Note that I will not be able to provide here a complete account of prosodic
phrasing in SC. Like other authors who discuss the relevance of the prosodic structure for 2
cliticisation in SC (Halpern 1995; Percus 1993; Radanović-Kocić 1988, 1996; Schütze 1994; and Zec
and Inkelas 1990), I will rely on some fairly widely accepted assumptions concerning prosodic
phrasing in general.
96 ŽELJKO BOŠKOVIĆ

followed by a pause, an indication of an I-phrase boundary. (For more phonolog-

ical evidence to this effect, see Radanović-Kocić 1988, 1996.) Given this, it is
clear that the clitics are located in the 2 of their I-phrase in (49)–(52). When
we attempt to place a clitic in the 3 of its I-phrase, we get an ungrammatical
sentence, as indicated by (54), which contains only one I-phrase, namely the
whole clause.23
(54) *Petru on će prodati knjige
Peter he will sell books
‘To Peter, he will sell books.’
We have seen, however, that clitics can be placed in the third, even the fourth
position of their clause. The correct descriptive generalisation for the distribution
of SC 2 clitics is then not that they are second within their clause, but within
their I-phrase, which strongly indicates that the 2 effect is phonological in
nature.24
Before showing how the role of phonology in 2 cliticisation can be imple-
mented formally, let us consider the standard syntactic account of the above

23. As noted by Browne (1975), even moved constituents that are not heavy can delay clitic
placement as long as they bear heavy emphatic stress and are followed by a pause, which indicates
that the relevant elements are forming separate I-phrases. I disregard this possibility here.
24. Certain facts concerning clitic placement after conjunctions confirm this conclusion. Browne
(1975) notes that some conjunctions must be followed by a pause. These conjunctions cannot host
clitics, in contrast to conjunctions that are not followed by a pause.
(i) a. *Dakle #su pozvali mnogo prijatelja i znanaca na ručak
so are invited many friends and acquaintances on lunch
‘So they invited many friends and acquaintances to lunch.’
b. cf. Dakle, pozvali su mnogo prijatelja i znanaca na ručak
c. Pojeli smo sav kruh, pa sam otišao da kupim još
eaten are all bread so am gone that buy more
‘We ate all the bread, so I went to buy more.’
Interestingly, pa can be used as an interjection, in which case it must be followed by a pause.
Browne (1975) notes that in that case pa cannot host a clitic.
(ii) Zašto nisi kupio voće? (Why didn’t you buy fruit?)
a. *Pa #sam kupio jabuke
but #am bought apples
‘But, I bought apples.’
b. Pa, #kupio sam jabuke
The fact that, in contrast to pa in (ic), dakle and pa in (ii) must be followed by a pause indicates that
they are followed by an I-phrase boundary. Given this, the data under consideration also indicate that
SC clitics must be second within their I-phrase rather than their clause. (For a prosodic explanation
why certain monosyllabic conjunctions, including pa, that normally do not bear noticeable accent can
still host clitics, see Hock 1996.)
 SECOND POSITION CLITICISATION 97

facts. Under syntactic accounts it is standardly assumed that elements that delay
clitic placement (e.g. the appositive and the fronted heavy constituents in
(49)–(52)) are CP-external and therefore do not count in determining 2. Given
that they are CP-external, the syntactic descriptive statement that clitics are
second within their clause could still capture the distribution of SC clitics.
However, it is unclear how the delaying effect of parentheticals, in particular,
the contrast between (50) and *Znači da oni će sutra doći (‘means that they will
tomorrow arrive’), can be captured under this analysis. On closer scrutiny, the
analysis also fails to account for the delaying effect of fronted heavy constituents
and appositives. This can be shown by considering obligatory clitic-third
constructions. It is well known that very heavy constituents obligatorily delay
clitic placement, as shown by (55).
(55) a. Njegovom najboljem prijatelju prodali su knjigu
his best friend sold are book
‘To his best friend, they sold the book.’
b. *Njegovom najboljem prijatelju su prodali knjigu
Under the prosodic account this would have to be interpreted as indicating that
njegovom najboljem prijatelju in (55) must be parsed as a separate I-phrase, in
contrast to, for example, tvome prijatelju in (56), which does not have to be
parsed as a separate I-phrase, as indicated by the fact that it does not obligatorily
delay clitic placement.25 (Note that a pause must follow tvome prijatelju in
(56a), but cannot follow it in (56b), as expected.)
(56) a. Tvome prijatelju prodali su knjigu
your friend sold are book
‘To your friend, they sold the book.’
b. Tvome prijatelju su prodali knjigu
Under the syntax account, the fact that, in contrast to tvome prijatelju, njegovom
najboljem prijatelju obligatorily delays clitic placement has to be interpreted as
indicating that, in contrast to tvome prijatelju — which can, but does not have to,
move to a CP-external position — njegovom najboljem prijatelju must move to
a CP external position. This, however, cannot be correct, as indicated by the fact

25. This seems plausible given that which constituents obligatorily delay clitic placement is
determined on prosodic grounds, prosodic heaviness being the determining factor (for discussion of
exactly how heavy a constituent must be to obligatorily delay clitic placement, see Radanović-Kocić
1988 and Schütze 1994).
98 ŽELJKO BOŠKOVIĆ

that njegovom najboljem prijatelju can follow wh-phrases, which strongly

suggests that it does not have to be sentence external, if it ever is.26
(57) a. Šta su (oni) njegovom najboljem prijatelju prodali?
what are (they his best friend sold
‘What did they sell to his best friend?’
b. Ko je njegovom najboljem prijatelju prodao knjigu?
who is his best friend sold book
‘Who sold a book to his best friend?’
A similar point can be made with respect to appositives. Radanović-Kocić
(1996) observes that adding an appositive to a subject NP also obligatorily delays
clitic placement:
(58) Ja sam ti obećala sladoled
I am you promised ice cream
‘I promised you an ice cream.’
(59) *Ja, tvoja mama, sam ti obećala sladoled
I your mother am you promised ice cream (cf. (51))
The contrast between (59) and (51) is readily accounted for under the prosodic
account. Appositives clearly must be parsed as separate I-phrases, as indicated by
the fact that they are obligatorily separated by pauses. The clitic is then located
in the 2 of its I-phrase in (51), but in the 1 in (59).
Under purely syntactic accounts, we have to assume that the subject NPs in
(51) and (59) must be located in different structural positions, which seems
rather implausible. In particular, it is necessary to assume that the subject in (59)
obligatorily moves to a CP-external position and therefore does not count for
determining second position. (60), however, provides strong evidence against this
analysis, since it shows that the relevant element does not have to be CP-external
(see also note 26).
(60) Šta sam ti ja, tvoja mama, obećala?
what am you I your mother promised
‘What did I, your mother, promise to you?’
I conclude, therefore, that syntactic accounts fail to account for delayed clitic
placement. Trying to rescue syntactic accounts by assuming that elements that

26. Notice that playing with CP Recursion cannot help here since we are dealing with an environ-
ment in which CP Recursion is not licensed (see Iatridou and Kroch 1992).
 SECOND POSITION CLITICISATION 99

delay clitic placement are CP-external, which would allow us to maintain the
generalisation that SC clitics are second within their clause, a syntactic unit,
clearly does not work. We are left with the generalisation in (53), which strongly
indicates that the 2 effect is phonological in nature.

3.2.2.6 VP ellipsis revisited. That the 2 requirement is a phonological rather

than a syntactic requirement is confirmed by VP ellipsis constructions such as
(61a), which contrasts with its non-elided counterpart (61b). (The relevance of
(61a) was pointed out to me by Marcel den Dikken. Note that ga is the only
clitic in (61).)
(61) a. Marija ga nije poljubila, a Ana jeste poljubila
————ga ——
Marija him is+ kissed, and Ana IS kissed him
‘Marija did not kiss him, and Ana did.’
b. *Marija ga nije poljubila, a Ana jeste poljubila ga
(61) shows that constructions that violate the 2 requirement can be rescued by
deleting the offending clitic in PF under VP ellipsis. (I assume that a copy of
moved ga is present in the antecedent VP.) This is expected if the 2 require-
ment is a phonological requirement (in PF, the clitic is second in (61a), but not
in (61b)), but not if it is a syntactic requirement. Under the PF deletion account
of ellipsis, the clitic is located in the same, non-2 in both (61a) and (61b) in
the syntax, since (61a–b) have the same structure in the syntax.
Having established that the 2 effect is phonological in nature, in the next
section I turn to phonological approaches to 2 cliticisation.27

27. Ćavar and Wilder (1994) observe that infinitival complements of verbs and nouns have different
possibilities for clitic placement and argue that this raises a problem for phonological statements of
the 2 effect. (Nothing changes in (ii) if the noun is accusative and follows the verb. Note that
infinitives are rarely used as nominal complements in SC, so the judgements, including the base-line
data, are somewhat murky here.)
(i) a. Želi ga vidjeti
wants him to-see
‘He wants to see him.’
b. Jovan ga želi vidjeti
c. *Želi vidjeti ga
(ii) a. Želja knjigu joj dati bila je velika
desire book her to-give been is great
‘the desire to give her a book was great.’
b. *Želja joj knjigu dati bila je velika
c. *Želja knjigu dati bila joj je velika
I assume that clitic climbing is a result of syntactic movement of clitics that takes place in restructur-
ing environments. This immediately accounts for (iic) since cross-linguistically, nominal complements
100 ŽELJKO BOŠKOVIĆ

4. Phonological Approaches to Second Position Cliticisation

4.1 The strong phonology approach: Move clitics in PF

Radanović-Kocić (1988, 1996) gives an account of the 2 effect in which all
clitic placement is accomplished in the phonology. The following rules are
responsible for clitic placement in her analysis:
(62) Assign the feature [+clitic] to the accusative, dative, and genitive
pronouns, and auxiliaries (except budem [a form of be]) and the
copula in all positions except when they are carrying phrasal stress
and when not preceded by an element that can serve as its host
(Radanović-Kocić 1988: 88).
(63) Move all [+clitic] elements within an IP [intonation phrase] into the
position after the first P [phonological phrase] of the same IP
(Radanović-Kocić 1988: 134).
Under Radanović-Kocić’s analysis, clitics and the corresponding full forms are
derived from the same elements. Clitics are identified as such through the
assignment of the feature [+clitic] via rule (62), and then moved to 2 in the

are non-restructuring environments. Consider now how the remaining constructions in (i)–(ii) fare
with respect to (53). (i) is not surprising since, as noted by Radanović-Kocić, the infinitive in (i) is
incorporated in the same I-phrase with the rest of the clause. (There can be no pause preceding the
infinitive.) The clitic is thus located in the 2 of its I-phrase in (i-a, b), but not in (i-c). As for (ii),
it seems to me that, in contrast to infinitival complements of verbs, infinitival complements of nouns
form separate I-phrases, as a result of which (ii-a, b) conform to (53). (Joj is located in the 2 of its
I-phrase in (ii-a), and in the 1 in (ii-b).) In my judgement, a small pause needs to follow želja in
(ii), an indication of an I-phrase boundary. Since this may not be obvious (Ćavar and Wilder seem to
disagree) I will give additional evidence that nominal infinitival complements form separate I-phrases.
Radanović-Kocić (1988, 1996) shows that I-phrase boundaries block degemination in SC, as shown
by the contrast between (iii-a) and (iii-b), where the heavy phrase must form a separate I-phrase.
(iii) a. Moj jorgan je od perja /mojorgan/
my comforter is of down
b. Za prošlogodišnji Prvi maj Janko je otišao u Paris /majjanko/ */majanko/
For last year’s first May Janko is gone to Paris
‘For last year’s May Day, Janko went to Paris.’
Significantly, degemination cannot take place in (iv), which can be accounted for if there is an
I-phrase boundary before the infinitive, i.e. if the infinitive forms a separate I-phrase. (This is not
surprising if clausal complements of nouns are actually adjuncts (see Grimshaw 1990), since adjuncts
often form separate I-phrases.)
(iv) Pokušaj juriti ga peronom je uzaludan /pokušajjuriti/ */pokušajuriti/
attempt to-chase him platform- is futile
‘The attempt to chase him down the platform is futile.’
 SECOND POSITION CLITICISATION 101

phonological component. The movement must be taking place in the phonology

rather than in the syntax because it refers to constituents (intonation phrases) not
present in the syntax. In the syntax itself, clitics and the corresponding full forms
are located in the same position.
The account has some serious empirical drawbacks. Consider (64).
(64) a. *Poljubio jesam Milenu
kissed AM Milena
‘I did kiss Milena.’
b. Poljubio sam Milenu
kissed am Milena
‘I kissed Milena.’
Since no relevant PF movement takes place in non-clitic constructions, (64a) should
be interpreted as indicating that syntax cannot place a participle in front of an
auxiliary. The conclusion should hold for both (64a) and (64b), since sam and jesam
are syntactically the same element (jesam) under Radanović-Kocić’s analysis. In
the syntax, the participle, then, must follow the auxiliary in both (64a) and (64b).
(65) Jesam poljubio Milenu
But then we cannot derive (64b) in PF by applying (62)–(63) because the
auxiliary is not preceded by an element that can serve as its host. (Radanović-
Kocić clearly intends each of the specified environments in (62) to block [+clitic]
assignment.) The grammaticality of constructions in which a participle precedes
an auxiliary clitic thus remains unaccounted for under Radanović-Kocić’s
analysis. There are several other problems with this analysis. Thus, to account
for the fact that both the Det--N (4b) and the Det-N- (4a) order are
possible, Radanović-Kocić assumes that the determiner in a Det+N sequence can
optionally form a phonological phrase on its own, which is far from obvious.
(4) a. Taj čovjek je volio Milenu
that man is loved Milena
‘That man loved Milena.’
b. Taj je čovjek volio Milenu
(14a–c) also raise a problem for Radanović-Kocić’s analysis.
(14) a. Lava sam Tolstoja čitala
Lav- am Tolstoy- read
b. *Lava sam Tolstoj čitala
Lav- am Tolstoy- read
102 ŽELJKO BOŠKOVIĆ

c. *Lav sam Tolstoja čitala

Lav- am Tolstoy- read
It appears that the only way to account for this data under her analysis is to
assume that Lava in (14a) does, and Lava in (14b) and Lav in (14c) do not, form
a phonological phrase, a distinction that seems rather implausible and which
leaves the correlation between syntactic movability and the ability to host the
clitic in the constructions in question (see (13)) completely unexplained.
In fact, quite generally, Radanović-Kocić’s analysis fails to capture the
generalisation that only elements that can be independently shown to be able to
undergo syntactic movement can host SC clitics (in addition to elements that are
base-generated in front of clitics), since rule (63) essentially washes away this
kind of syntactic effect. However, the most serious problem with Radanović-
Kocić’s analysis is the power of extensive non-local word reordering that the
phonology is invested with. Notice also that Progovac (1996) shows that the
locality constraints on at least some instances of clitic placement are the same as
the locality constraints on wh-movement (see the discussion below), which
Radanović-Kocić would consider a syntactic operation. This means that under
Radanović-Kocić’s analysis we are simply duplicating syntax by applying
syntactic operations in the phonology and subjecting them to the same locality
constraints, which is conceptually very unappealing, particularly in light of the
fact that Radanović-Kocić does not provide any independent motivation outside
SC cliticisation that non-local word reordering operations such as those she needs
in her analysis of SC cliticisation are indeed otherwise attested in the phonology.
Let us now examine Progovac’s (1996) evidence against Radanović-Kocić’s
analysis. Progovac observes that SC verbs fall into two groups: those that select
syntactically/semantically opaque complements (I-verbs), and those that select
transparent complements (S-verbs). Though SC does not have distinct subjunctive
morphology, the distinction goes along the lines of indicative/subjunctive
complements distinction in other languages. In fact, SC S-complements have
strong tense restrictions, which is a characteristic of subjunctive complements
cross-linguistically.
Progovac observes that clitic climbing is possible out of S-complements, but
not out of I-complements:
(66) a. Milan kaže da ga vidi
Milan says that him sees
‘Milan says that he sees him.’
b. *Milan ga kaže da vidi
 SECOND POSITION CLITICISATION 103

(67) a. Milan želi da ga vidi

Milan wants that him sees
‘Milan wants to see him.’
b. ?Milan ga želi da vidi
The possibility of clitic climbing in (67b) cannot be accounted for under
Radanović-Kocić’s analysis, since the embedded clause in (67b) forms a separate
I-phrase, just like the embedded clause in (66b). In fact, it is difficult to see how
the difference between (66b) and (67b) can be accounted for in a principled way
in phonological terms. The difference seems to be syntactic in nature. Thus,
Progovac observes that the difference between I-complements and S-comple-
ments can be observed with uncontroversially syntactic operations, such as
wh-movement:
(68) a. ?*Koga ne kažeš da voliš?
whom not say that love
‘Whom don’t you say that you love?’
b. Koga ne želiš da voliš?
whom not want that love
‘Whom don’t you want to love?’
Progovac observes that negative polarity items also extend their domain in
S-complements, but not in I-complements, like in many other languages:
(69) a. *Ne kažem da vidim nikoga
not say that see no-one
‘I do not say that I see anyone.’
b. Ne želim da vidim nikoga
not want that see no-one
‘I do not want to see anyone.’
All this indicates that the difference between I-complements and S-complements
is syntactic/semantic (see Progovac 1993 for an analysis). Since clitic placement
is sensitive to it, it follows that clitic placement is a syntactic rather than a
phonological operation, contrary to what Radanović-Kocić argues.28

28. Anderson (1993) presents an analysis that is in some respects similar to Radanović-Kocić’s
analysis. Instead of moving to 2 in the phonology, under Anderson’s analysis SC clitics are
introduced into the structure in 2 in PF (i.e. morphology). Anderson’s analysis has some of the
same empirical drawbacks as Radanović-Kocić’s analysis. Thus, clitic climbing facts discussed above
appear to remain unaccounted for under Anderson’s analysis. Anderson’s analysis also fails to
account for constructions in which traditional 1 placement is not allowed (cf. (9b), (11c) and
104 ŽELJKO BOŠKOVIĆ

The conclusion that the data discussed so far lead us to is that we need a
phonological account of the 2 effect in which clitic placement (i.e. clitic
movement) itself takes place in the syntax rather than in the phonology. One
such account is discussed in the next section.

4.2 The weak phonology account: Filtering out ungrammatical constructions at PF

Instead of investing the phonological component with the power to do operations

that belong to another domain and are otherwise not obviously applicable in the
phonology, let us try to capture the 2 effect by appealing to independently
motivated phonological mechanisms. It is clear that in every derivational model,
the phonology (by which I mean PF) should be allowed to have a filtering effect
on the syntax, i.e. it should be allowed to rule out some syntactically well-
formed constructions. Lasnik’s (1981) Stranded Affix Filter is an example of this
filtering effect of the phonology. It seems clear that we should not expect syntax
to rule out all constructions containing stranded affixes. In fact, it is not clear
how this could be do ne in a principled way. Such constructions, however, can
be ruled out in a principled way in PF due to the presence of an illegitimate PF
object, namely a phonologically weak element that does not have a host.
The same should hold for stranded clitics. PF should be able to filter out
well-formed syntactic outputs containing stranded clitics. Phonologically weak
elements clearly must be specified for the direction of their attachment to the
host. SC clitics are suffixes.29 I assume that they are specified as such in the
lexicon. Any syntactic output where this lexical requirement of SC clitics is not
satisfied will then be filtered out in PF. This accounts for the badness of
sentences in which clitics are found in the initial position of their I-phrase, given
the assumption that cliticisation cannot take place across I-phrase boundaries, as
argued for SC in Percus (1993) and Schütze (1994).30 What about other

(14b–c)), since under Anderson’s analysis it should always be possible to insert SC clitics after the
first word of their cliticisation domain. It is also difficult to see how Stjepanović’s (1998a, b) VP
ellipsis data discussed in Section 3.2.2.4 and Bošković’s (1995) data concerning double participle
constructions can be accounted for in a principled way under Anderson’s analysis. (To be fair to
Anderson 1993, none of the data claimed here to be a problem for Anderson’s analysis were
discussed in that work since they were not known at that time. Note also that Anderson is vague
about the syntax of the relevant constructions, so that it is difficult to be certain about the predictions
that his analysis would make.)
29. I am not claiming here that clitics and affixes are the same thing (for relevant discussion, see
Anderson 1992, 1993; Klavans 1985; and Zwicky and Pullum 1983). I use the term suffix merely to
indicate a phonologically weak element that follows its host.
30. This might not be true cross-linguistically, see Bresnan (1971).
 SECOND POSITION CLITICISATION 105

instantiations of the 2 requirement? Bošković (1995) argues that we do not

need to adopt any new phonological operations to capture them, as Radanović-
Kocić (1988) does, and suggests that the 2 requirement on SC clitics can be
captured in its entirety through a filtering effect of the phonology on the syntax.
Bošković’s (1995) analysis follows the line of work that originated with Klavans
(1985) (see also Anderson 1992, 1993). I will therefore discuss Klavans (1985)
before turning to Bošković (1995). (Klavans herself does not discuss SC.)
Klavans considers the 2 requirement a result of lexical properties of
clitics, some of which are syntactic and some of which are phonological in
nature. To account for cross-linguistic variation with respect to cliticisation,
Klavans proposes three parameters for clitic placement, which are instantiated as
lexical properties of clitics, i.e. clitics are lexically specified for their settings.
(70) a. Parameter 1 (Dominance): Initial/Final
b. Parameter 2 (Precedence): Before/After
c. Parameter 3 (Phonological Liaison): Proclitic/Enclitic
Parameter 1 determines whether a clitic attaches to the initial or final constituent
of a specified phrase. Parameter 2 specifies whether the clitic occurs before or
after the host chosen by Parameter 1. Parameter 3 gives the direction of phono-
logical attachment. The reason why both Parameter 2, which is essentially
syntactic, and Parameter 3, which is phonological, are needed is because, accord-
ing to Klavans, a clitic can have a different host in the syntax and the phonology.
Though Klavans does not state this explicitly, we also need a parameter that
would determine the specified phrase whose initial or final constituent the clitic
attaches to, i.e. we need to determine the domain of cliticisation for Parameter 1.
Klavans argues that the domain of cliticisation is determined syntactically. (Some
of the possibilities, according to Klavans, are sentence, NP, and V.) Since Klavans
does not explicitly propose a parameter for the domain, which is clearly needed,
and since SC does not exhibit any crucial difference between the syntactic and
the phonological attachment of clitics,31 I will adopt a slightly revised version
of Klavans’ parameterisation, which was also adopted by Anderson (1993).

31. In the current framework nothing special actually needs to be said about the syntactic attachment
of clitics that would not hold for other lexical items. Like all other lexical items, clitics can undergo
syntactic movement motivated by feature checking. If, for example, a clitic must be adjacent to a
verb or a functional head in the syntax, this could be interpreted as indicating that the clitic has a
feature to be checked against the verb or the functional head and adjoins to it to do that. Other non-
clitic lexical items can have the same property.
106 ŽELJKO BOŠKOVIĆ

(71) a. P1 (Domain): specifies the domain of cliticisation.

b. P2 (Dominance): specifies whether a clitic attaches to the initial
or final constituent of the domain specified by P1.
c. P3 (Precedence): specifies whether a clitic precedes or follows
the host determined by P2.
It is obvious that the value of P2 for SC 2 clitics is initial and the value of P3
suffix. What about P1? According to Klavans, the domain of cliticisation is
determined syntactically. We have seen that this cannot be correct for SC. The
domain of cliticisation for SC is determined prosodically, namely, it is an
I-phrase. SC 2 clitics then have the following values for the parameters in (71):
(72) a. I-phrase
b. Initial32
c. suffix
Following Klavans, I assume that we are dealing here with lexical properties of
clitics. Since the relevant properties of SC clitics are strictly phonological, it
follows that the 2 effect is a phonological effect. Klavans appears to treat her
parameters as constraints on attachment or insertion of clitics (she is not very
clear on this point though), i.e. she applies them derivationally. The parameters
in (72), on the other hand, are more naturally applied representationally in the
current system. I therefore assume that (72a–c) constrain PF representations:
Sentences violating the relevant lexical properties of clitics are filtered out in PF.
This way we can easily account for the fact that, as shown by the contrast in
(61), sentences violating the 2 requirement can be rescued by deleting the
offending clitic in PF: the clitic is located in 2 (satisfying (72a-c)) in the
output of PF in (61a), but not in (61b).
(61) a. Marija ga nije poljubila, a Ana jeste poljubila
————ga ——
Marija him is+ kissed, and Ana IS kissed him
‘Marija did not kiss him, and Ana did.’
b. *Marija ga nije poljubila, a Ana jeste poljubila ga

32. I assume that either an initial prosodic word or an initial phonological phrase can be the relevant
constituent for P2. Radanović-Kocić (1988, 1996) suggests that the host is always a phonological
phrase, which is unclear (see Zec and Inkelas 1992). Anyway, a modification of the current analysis
proposed below will explain in a principled way the possible disjunctivity in SC clitic placement.
 SECOND POSITION CLITICISATION 107

On the other hand, these facts remain unaccounted for if (72a–c) are applied
derivationally as constraints on attachment or insertion of clitics.33 Notice also
that, although under the current analysis the 2 requirement is considered to be
a phonological phenomenon, we do not need to appeal to PF movement of clitics
and no look-ahead from the syntax to the phonology is needed to account for it.
Syntax can do its job without worrying about what phonology will do: clitics are
present and undergo movement in the syntax, they are not lexically inserted and
do not undergo movement in PF, which enables us to account for the clitic
climbing facts from Section 4.1.
The phonological, representational version of Klavans’s analysis for SC can
be further simplified by eliminating one of the parameters in (72). Bošković
(1995) shows that it suffices to have the following as the relevant lexical
properties of SC clitics.
(73) a. #_
b. suffix
(73) states that SC clitics must be suffixes and right adjacent to an I-phrase
boundary. (73b) corresponds to P3, and (73a) is intended to capture the effects
of P1 and P2. It specifies the domain of cliticisation and states that SC clitics
must be located in the beginning rather than the end of that domain without
using a constituent of the domain as an intermediary in specifying the relation
between the clitics and the domain.
(73a) and (73b) appear to impose conflicting requirements: (73a) requires
that SC clitics immediately follow an I-phrase boundary and (73b) requires that
there be at least one phonologically overt element between the clitic and the
I-phrase boundary that can serve as a host for the clitic, given the natural
assumption that cliticisation cannot take place across I-phrase boundaries. In
Bošković (1995) I show that the conflict can be resolved by adopting Marantz’s

33. The same holds for the clitic climbing facts discussed in Section 4.1, which provide evidence
that clitics move in the syntax. Implementing (72a–c), which are phonological in nature, as
constraints on attachment or insertion of clitics would require clitics to be either inserted or undergo
movement in PF.
108 ŽELJKO BOŠKOVIĆ

(1988, 1989) notion of Morphological Merger. Consider the following definition

of Morphological Merger:34
(74) Morphological Merger
At any level of analysis, independent constituents X and Y standing
in a relation at that level (or heading phrases standing in a relation)
may merge into a single word X+Y, projecting the relation between
(the constituent headed by) X and (the constituent headed by) Y
onto the affixation relation X+Y.
The underlying assumption here is that Morphological Merger takes place in PF
under PF adjacency. Slightly departing from Marantz, I assume that after X and
Y merge, the derived element takes over the requirements of both X and Y. The
most important departure from Marantz is that I assume that Morphological
Merger cannot reorder elements; it simply puts two adjacent elements together
forming a single word out of them. This restrictive view of Morphological
Merger ensures that Morphological Merger has a very different effect from PI;
it is less powerful since it cannot affect linear order.
In constructions such as (75), the clitic merges with the preceding element,
thus satisfying its suffix requirement. Since the derived word, which takes over
the requirements of ‘its constituents’, is adjacent to an I-phrase boundary,
indicated by #, (73a) is also satisfied.35
(75) #Nju je Jovan poljubio#
#her is Jovan kissed
‘Her, Jovan kissed.’
More complex cases in which a branching element precedes a clitic can also be
readily accounted for given certain proposals made in Marantz (1989). Marantz

34. As discussed below, the view of Morphological Merger and assumptions relevant to it I take
here differs from that of Marantz (1988, 1989). This is necessary because several theoretical
assumptions I adopt differ from Marantz’s theoretical framework.
35. It is generally possible to assign more than one prosodic structure to a single syntactic structure,
depending on how it is pronounced. I disregard this below and show only the relevant prosodic
structures.
An anonymous reviewer observes that if all prosodic representations respect the prosodic hierarchy
I-phrase-phonological phrase-phonological word, and non-branching constituents project vacuously,
a phonological phrase boundary intervenes between nju and the I-phrase boundary in (75). To
account for the grammaticality of the construction we can then assume that prosodic boundaries do
not disturb the relevant adjacency relation. The assumption is unnecessary, however, under the
approach to PF headedness adopted below.
 SECOND POSITION CLITICISATION 109

argues that constituent heads at PF are the elements located at constituent edges,
an assumption that I adopt here.36 Marantz also assumes that bearing a relation
to a phrase is equivalent to bearing a relation to the head of the phrase, which I
implement and generalise by assuming that properties of a head can be satisfied
at the phrasal level and that properties of a phrase can be satisfied at the level of
its head. In light of this, consider (76).
(76) #[a Moju prijateljicu] je poljubio#
my friend is kissed
‘My friend, he kissed.’
The clitic in (76) merges with prijateljicu, which is the PF head of the prosodic
phrase a, thus satisfying (73b). Given that properties of a head can be satisfied
on the phrasal level, since a is right adjacent to an I-phrase boundary, (73a) is
also satisfied.
Turning now to constructions that do not obey the 2 requirement,
structures in which clitics are located in the third position of their I-phrase, such
as (54), are ruled out because (73a) is not satisfied. The suffix requirement can
be satisfied in (54) by merging the clitic with on. However, since on is not
adjacent to an I-phrase boundary (see the discussion above the example (54)),
(73a) cannot be satisfied.
(54) *Petru on će prodati knjige
Peter he will sell books
‘To Peter, he will sell books.’
Constructions with sentence initial clitics such as (29) are ruled out because they
violate (73b).
(29) *Je poljubio Anu
is kissed Ana
‘He kissed Ana.’
It is easy to verify that examples that are not in accordance with (53), the
descriptive statement of the 2 effect, inevitably violate either (73a) or (73b).
Constructions that are in accordance with it, on the other hand, satisfy (73a–b).

36. In all relevant cases discussed below the head of a phrase is the rightmost constituent, which is
in line with Nespor and Vogel’s (1982) system, in which in a right-branching structure (and SC is a
right-branching language) the strongest element prosodically is always the rightmost one. Notice also
that I assume that elements that are not prosodic words themselves (for example clitics) cannot head
PF phrases.
110 ŽELJKO BOŠKOVIĆ

Consider, for example, the delaying effect of elements that form separate
I-phrases. As discussed above, in (49), the fronted heavy constituent forms a
separate I-phrase.
(49) Sa Petrom Petrovićem srela se samo Milena
with Peter Petrović, met self only Milena
‘With Peter Petrović, only Milena met.’
Petrovićem is then followed by an I-phrase boundary, evidence for which is
provided by the fact that it is followed by a pause. This means that srela is
adjacent to an I-phrase boundary so that merging the clitic with srela can satisfy
(73a). The delaying effect of phonologically heavy elements on clitic placement
is thus accounted for. Note that this is accomplished in a derivational model in
which syntax feeds phonology, contra Zec and Inkelas (1990), who argue that we
need a non-derivational, co-presence model to account for the delaying effect.
Note also that no look-ahead from the syntax to the phonology is required to
account for the delaying effect under the current analysis.
(73) also accounts for clitic clustering. In (77), all the clitics merge with oni,
satisfying (73a–b).
(77) #Oni su mi ga predstavili#
#they are me him introduced
‘They introduced him to me.’
In (78), the first clitic merges with oni, which satisfies both (73a) and (73b). The
second clitic merges with the intervening element Anu. This satisfies (73b), but
not (73a).
(78) #Oni su Anu mi predstavili#
#they are Ana me introduced
‘They introduced Ana to me.’
Clitics (or, more precisely, clitics located in the same I-phrase-see (83) below)
are thus forced to cluster in PF by (73a–b). They are, however, not forced to
cluster in the syntax, since (73a–b) are PF requirements. This enables us to
account for Stjepanović’s ellipsis facts, which indicate that SC clitics do not have
to cluster in the syntax. The relevant descriptive generalisation is the following:
SC clitics must cluster (be adjacent) in the phonology, but do not have to cluster
(form a constituent) in the syntax. This is readily captured by the current
analysis, which forces clitic clustering in the phonology but is completely neutral
with respect to the issue of whether clitics need to cluster in the syntax.
One might think that split clitic constructions such as (79a), in which the
 SECOND POSITION CLITICISATION 111

clitics are located within two different I-phrases, are incorrectly allowed in the
current system.
(79) a. *Svome prijatelju će predstaviti ga sutra
his friend will introduce him tomorrow
‘To his friend, they will introduce him tomorrow.’
b. cf. Svome prijatelju, predstaviti će ga sutra
It is easy to verify that (79a) satisfies (73a–b). This, however, does not have any
undesirable consequences, since (79a) is ruled out on the relevant derivation in
the syntax. Recall that each clause forms a separate I-phrase and that certain
elements, such as appositives, parentheticals, and heavy fronted constituents,
form separate I-phrases, I-phrase boundaries coinciding with the boundaries of
these syntactic constituents. (79a) then must have the structure shown in (80).
The clitic must be located within the I-phrase introduced by the heavy element
to satisfy (73a). (Recall that cliticisation, or, to be more precise, Morphological
Merger, cannot take place across I-phrase boundaries.)
(80) #[NPSvome prijatelju će]# predstaviti ga sutra#
(80) involves movement of the auxiliary into the fronted NP. This movement is
clearly syntactically illegitimate since it does not have a proper driving force. In
the minimalist system the auxiliary in (80) has a syntactic reason to move to T0
and Agrs0, possibly C0. However, there is no syntactic requirement that could
plausibly motivate movement of the auxiliary into the fronted NP. The construc-
tion is then ruled out in the syntax by the Last Resort Condition, which prohibits
superfluous operations. (There are probably other syntactic violations in (80).) The
fact that (80) satisfies the phonological 2 requirement is then irrelevant. (80) is
in fact ruled out for the same reason as the corresponding English construction.
(81) *[PP To his friends will] they introduce him tomorrow
It seems safe to conclude that nothing goes wrong with (81) phonologically. The
construction then must be ruled out in the syntax. And we have seen that this can
be done straightforwardly.
It is easy to verify that the following constructions involving clitics located
in different I-phrases are also ruled out because the movement that introduces
the clitics into the additional I-phrase is syntactically superfluous and therefore
violates the Last Resort Condition.
(82) a. *#[PP Sa njegovim najboljim drugom će]# Milena se sresti#
with his best friend will Milena self meet
‘With his best friend, Milena will meet.’
112 ŽELJKO BOŠKOVIĆ

b. *#Znači da #[kao što rekoh će]# predstaviti ga Mileni#.

#means that #[as said will introduce him Milena
‘It means that, as I said, they will introduce him to Milena.’
The fact that (82a–b) satisfy the 2 requirement under the current analysis is
then irrelevant. In fact, this is desirable, since ruling (82a–b) out by an additional
requirement would be redundant.
As pointed out by Chris Wilder (personal communication), under the current
analysis we would then expect constructions with clitics split in different
I-phrases to be acceptable as long as nothing goes wrong with them in the
syntax. The expectation is difficult to test since, in my opinion, the syntax of SC
clitics is still largely mysterious. Due to extremely free word order in SC and the
phonological requirements on clitics it is very difficult to determine exactly
where SC clitics can move/be located in the syntax. (The same holds for other
relevant elements such as parentheticals.) As a result, for most of the potentially
relevant constructions it is very difficult to determine whether they are syntactic-
ally well-formed. However, it is still possible to construct acceptable construc-
tions of the relevant type. (Their acceptability indicates that they are syntactically
well-formed.) Consider the following example.37
(83) #Oni su, #kao što sam vam rekla#, predstavili se Petru#
#they are #as am you said introduced self Petar
‘They, as I told you, introduced themselves to Petar.’
Syntactic clause-mates su and se are split in two different I-phrases. However,
each clitic satisfies (73a–b) within its I-phrase. The acceptability of (83) is thus
accounted for under the current analysis.
Note that (83) provides conclusive evidence against the assumption that SC
clitics cluster together under the same node in the syntax, crucial to the syntax
account of the 2 effect. In fact, I see no way of accounting for (83) under any
of the purely syntactic approaches.
It is worth emphasising here that, although the current account of the 2
effect is phonological, it straightforwardly accounts for 1 fortresses (environ-
ments in which traditional 1 clitic placement is not allowed), or, to be more

37. Notice also the ungrammaticality of *Oni se, kao što sam vam rekla, predstavili su Petru, which
shows that su is syntactically higher than se.
Notice that some speakers prefer not to have clitics followed by an I-phrase boundary, which is an
interfering factor here (for the speakers in question). There may also be some variation with respect
to parenthetical placement that is independent of clitic placement. For another example in which
clitics originating in the same clause are split in different I-phrases, see Stjepanović (1998b).
 SECOND POSITION CLITICISATION 113

precise, the correlation between syntactic movability and the ability to host a
clitic, discussed in Section 3.1, because under the current phonological account
of the 2 effect clitics are inserted and undergo movement only in the syntax.
The ungrammaticality of (9b), (11b–c), and (14b–c) can then be accounted for in
the same way as under the strong syntax account. The element hosting the clitics
in (9b), (11b–c), and (14b–c) cannot be syntactically moved in front of the
clitics, as indicated by (9a), (11a), and (13b–c). Since no movement is assumed
to take place in PF there is then no way to obtain the word orders in (9b), (11c),
and (14b–c). 1 fortresses are thus accounted for in the same way as in the
strong syntax approach.
(9) a. *Prema Milan i Jovan idu Mileni
toward Milan and Jovan walk Milena
‘Milan and Jovan are walking toward Milena.’
b. *Prema su Mileni Milan i Jovan išli
toward are Milena Milan and Jovan walked
‘Toward Milena Milan and Jovan walked.’
(11) a. *Roditelji dolaze uspešnih studenata
parents arrive successful students
‘Parents of successful students are arriving.’
b. *Roditelji su došli uspešnih studenata
parents are arrived successful students
‘Parents of successful students arrived.’
c. *Roditelji su se uspešnih studenata razišli
parents are self successful students dispersed
‘Parents of successful students dispersed.’
(13) b. *Lava čitam Tolstoj
Lav- read Tolstoy-
c. *Lav čitam Tolstoja
Lav- read Tolstoj-
‘Lav Tolstoy, I read.’
(14) b. *Lava sam Tolstoj čitala
Lav- am Tolstoy- read
c. *Lav sam Tolstoja čitala
Lav- am Tolstoy- read
However, it is easy to verify that the current analysis readily accounts for the
following properties of SC 2 cliticisation, demonstrated in 3.2.2 and shown to
make the strong syntax approach itself untenable: SC clitics (i) do not have a
114 ŽELJKO BOŠKOVIĆ

fixed structural position, (ii) can be located very low in the tree, and (iii) do not
have to cluster together in the syntax. This is accomplished by removing the 2
requirement from the domain of syntax.38

5. Concluding Remarks

In conclusion, the analysis hinted at, but not fully developed in, Bošković (1995)
follows in the spirit of Klavans (1985). It simplifies the latter analysis, since it
replaces two of Klavans’s lexical parameters by a single lexical property. This is
achieved by avoiding an intermediary in specifying the relevance of the domain
of cliticisation, a move that seems to me conceptually appealing (see the
discussion below) and that has been made possible by adopting a modified
version of Marantz’s Morphological Merger. The current analysis of the 2
effect differs from Klavans’s in being representational and also more phonologi-
cal. As in Klavans (1985), the 2 effect is considered to be a result of lexical
properties of clitics. However, since the relevant properties are phonological,
under the current analysis the 2 effect is completely determined by the
phonology through a filtering effect of phonology on an overgenerating syntax.
Phonology was assumed to have a filtering effect on syntax even in some
syntactic accounts of the 2 requirement. The filtering effect was restricted,
however, to constructions with sentence initial clitics. The analysis argued for
here extends the filtering effect of phonology to constructions where clitics are
found in any other but the 2 of their I-phrase.
Although the current analysis places the 2 effect in the phonology, there
is no need to appeal to the assumption that clitics can undergo movement in the
phonology, in either its weak (PI) or strong (Radanović-Kocić’s) version.
Phonological movement accounts were shown to be both empirically and
conceptually untenable. I conclude, therefore, that the facts concerning 2
cliticisation in SC do not provide support for the possibility of movement in the
phonology, as was previously argued.
We have seen that the descriptive generalisation concerning the 2 effect
in SC is that SC clitics must be located in the 2 of the I-phrase in which the
syntax places them, which indicates that the 2 effect is phonological in nature
(I-phrases are phonological units) but that clitics undergo movement in the

38. The analysis does not make any predictions with respect to the relevant properties of clitics in
the syntax, in contrast to the strong syntax account, which, as shown above, makes incorrect
predictions.
 SECOND POSITION CLITICISATION 115

syntax. These are the defining characteristics of the current analysis. The
descriptive generalisation concerning the 2 effect in SC is thus straightfor-
wardly captured under the analysis argued for here, but not under alternative
accounts, as demonstrated above.
We have also seen that all the relevant facts concerning the 2 effect in
SC can be accounted for in a derivational model in which syntax feeds phonolo-
gy, contra Zec and Inkelas (1990), who argue for a co-presence model on the
basis of the delaying effect of phonologically heavy elements on clitic placement
in SC. Nothing in the data discussed above, including the delaying effect of
phonologically heavy elements, requires adopting a non-derivational, co-presence
model. (In fact, I am not aware of any data concerning 2 cliticisation in SC
that would require it.) In addition, we have seen that there is no need for a look-
ahead from the syntax to the phonology to account for 2 cliticisation in SC, as
a number of authors have previously assumed. The syntax can do its job without
caring about the needs of the phonology.39
It remains to be seen whether the current analysis — which is in a way a
combination of Radanović-Kocić (1988) and Klavans (1985), the emphasis on the
role of phonology coming from Radanović-Kocić, and its instantiation essentially
coming from Klavans — can be extended to other 2 cliticisation languages.
What I hope to have demonstrated here is that more attention has to be paid to
prosodic structure, as argued by Radanović-Kocić, in accounting for the 2
phenomenon. This, of course, can prove to be difficult in dealing with 2
languages that are no longer spoken.
Like Klavans’s analysis, the current analysis makes several predictions
concerning what kind of variation we would expect to find with respect to
cliticisation cross-linguistically. The predictions still remain to be tested, the
relevant prosodic information not being available in most cases. However, it is
worth noting here that Aissen (1992) has carried out an investigation of certain
clitics in Tzotzil and Jacaltec (Tzotzil un and Jacaltec an) and argues that the
distribution of these clitics cannot be accounted for without reference to the
prosodic structure of the sentence. In particular, the clitics in question must
encliticise to the final element of their I-phrase, which in the current terms
means that they have the following specification:40

39. The last two conclusions are also reached by Schütze (1994), though his analysis is not
completely trouble free with respect to either of the conclusions.
40. See also Hock (1996) for some discussion of cross-linguistic parameterisation of prosodic
domains of cliticisation.
116 ŽELJKO BOŠKOVIĆ

(84) a. _#
b. suffix
Pending a detailed cross-linguistic investigation of clitic types, I merely note here
that reference to a penultimate or second element as the anchor within the
domain of cliticisation, which would give us so far unattested (see Halpern 1995
and references therein) types of cliticisation (#X clitic+Y and Y+clitic X#, with
# being the boundary of the domain of cliticisation) and which in Klavans’s
system is blocked simply by saying that such reference is not possible, is ruled
out in a more principled way under the current analysis. To get these types of
cliticisation we need to be able to specify an intermediary in the relation between
the clitic and the domain of cliticisation, which, in contrast to Klavans’s analysis,
the current analysis simply does not do.
Finally, let me reiterate that under the current analysis nothing special needs
to be said about SC clitic placement syntactically. It is entirely possible (though
not necessary) that SC clitics behave like all other lexical elements in the syntax,
being subject to all and only the syntactic movements that their grammatical
category qualifies them for. It is the phonological component that is responsible
for the fact that in the surface they appear in a ‘special’ position. As pointed out
by Hock (1996), this seems desirable, given that clitics are defined in prosodic
terms.

References

Abney, S. 1987. The English Noun Phrase in its Sentential Aspect. Doctoral dissertation,
MIT.
Aissen, J. 1992. ‘Topic and focus in Mayan.’ Language 68: 43–80.
Anderson, S. 1992. A-morphous Morphology. Cambridge: Cambridge University Press.
Anderson, S. 1993. ‘Wackernagel’s revenge: Clitics, morphology, and the syntax of
second position.’ Language 69: 68–98.
Bennett, D. 1986. ‘Towards an explanation of word-order differences between Slovene
and Serbo-Croat.’ The Slavonic and East European Review 64: 1–24.
Borsley, R. and E. Jaworska 1988. ‘A note on prepositions and Case marking in Polish.’
Linguistic Inquiry 19: 685–691.
Bošković, Ž. 1995. ‘Participle movement and second position cliticization in Serbo-
Croatian.’ Lingua 96: 245–266.
Bošković, Ž. 1997a. The Syntax of Nonfinite Complementation: An Economy Approach.
Cambridge, Mass.: MIT Press.
Bošković, Ž. 1997b. ‘Superiority effects with multiple wh-fronting in Serbo-Croatian.’
Lingua 102: 1–20.
 SECOND POSITION CLITICISATION 117

Bošković, Ž. 1998. ‘Multiple wh-fronting and economy of derivation.’ Proceedings of

WCCFL 16: 49–63.
Bošković, Ž. In press. ‘Sometimes in SpecCP, sometimes in situ.’ In: Martin, R., D.
Michaels and J. Uriagereka (eds.) Step by Step: Essays on Minimalism in Honor of
Howard Lasnik. Cambridge, Mass.: MIT Press.
Bresnan, J. 1971. ‘Contraction and the transformational cycle.’ Ms., MIT.
Browne, W. 1974. ‘On the problem of enclitic placement in Serbo-Croatian.’ In: Brecht,
R. and C. Chvany (eds.) Slavic Transformational Syntax. Ann Arbor: University of
Michigan, 36–52.
Browne, W. 1975. ‘Serbo-Croatian enclitics for English-speaking learners.’ In: Filipović,
R. (ed.) Contrastive Analysis of English and Serbo-Croatian Vol. 1. Zagreb: Universi-
ty of Zagreb, 105–134.
Ćavar, D. and C. Wilder 1994. ‘”Clitic third” in Croatian.’ Eurotyp Working Papers 6:
19–61.
Chomsky, N. 1995. The Minimalist Program. Cambridge, Mass.: MIT Press.
Corver, N. 1992. ‘Left branch extraction.’ Proceedings of NELS 22: 67–84.
Dimitrova-Vulchanova, M. 1995. ‘Clitics in Slavic.’ Studia Linguistica 49: 54–92.
Embick, D. and R. Izvorski 1997. ‘Participle auxiliary word orders in Slavic.’ In: Browne,
W., N. Kondrashova and D. Zec (eds.) Formal Approaches to Slavic Linguistics: The
Cornell University Meeting 1995. Ann Arbor: Michigan Slavic Publications,
210–239.
Franks, S. 1997. ‘The secret life of Slavic clitics.’ Paper presented at the University of
Connecticut.
Franks, S. and L. Progovac 1995. ‘On the placement of Serbo-Croatian clitics.’ Indiana
Linguistic Studies 7: 69–78.
Grimshaw, J. 1990. Argument Structure. Cambridge, Mass.: MIT Press.
Halpern, A. 1995. On the Placement and Morphology of Clitics. Stanford, Calif.: CSLI
Publications.
Hayes, B. 1989. ‘The prosodic hierarchy in meter.’ In: Kiparsky, P. and G. Youmans
(eds.) Rhythm and Meter. Phonetics and Phonology, Vol. 1. San Diego: Academic
Press, 201–260.
Hock, H. 1992. ‘What’s a nice word like you doing in a place like this.’ Studies in the
Linguistic Sciences 22: 39–87.
Hock, H. 1996. ‘Who’s on first? Toward a prosodic account of P2 clitics.’ In: Halpern,
A. and A. Zwicky (eds.) Approaching Second: Second Position Clitics and Related
Phenomena. Stanford, Calif.: CSLI Publications, 199–270.
Iatridou, S. and A. Kroch 1992. ‘The licensing of CP-recursion and its relevance to the
Germanic verb-second phenomenon.’ Working Papers in Scandinavian Syntax 50:
1–24.
Izvorski, R. 1996. ‘The syntax and semantics of correlative proforms.’ Proceedings of
NELS 26: 133–147.
118 ŽELJKO BOŠKOVIĆ

Izvorski, R. 1998. ‘Non-indicative wh-complements of possessive and existential predi-

cates.’ Proceedings of NELS 28: 159–173.
King, T. H. 1996. ‘Slavic clitics, long head movement, and prosodic inversion.’ Journal
of Slavic Linguistics 4: 274–311.
Klavans, J. 1985. ‘The independence of syntax and phonology in cliticisation.’ Language
61: 95–120.
Lasnik, H. 1981. ‘Learnability, restrictiveness, and the evaluation metric.’ In: McCarthy,
J. and C. L. Baker (eds.) The Logical Problem of Language Acquisition. Cambridge,
Mass.: MIT Press, 1–21.
Lasnik, H. 1995. ‘Case and expletives revisited: On greed and other human failings.’
Linguistic Inquiry 26: 615–633.
Marantz, A. 1988. ‘Clitics, morphological merger, and the mapping to phonological
structure.’ In: Hammond, M. and M. Noonan (eds.) Theoretical Morphology:
Approaches in Modern Linguistics. San Diego: Academic Press, 253–270.
Marantz, A. 1989. ‘Clitics and phrase structure.’ In: Baltin, M. and A. Kroch (eds.)
Alternative Conceptions of Phrase Structure. Chicago: University of Chicago Press,
99–116.
Nespor, M. and I. Vogel 1982. ‘Prosodic domains of external sandhi rules.’ In: van der
Hulst, H. and N. Smith (eds.) The Structure of Phonological Representations.
Dordrecht: Foris, 225–255.
Nespor, M. and I. Vogel 1986. Prosodic Phonology. Dordrecht: Foris.
Percus, O. 1993. ‘The captious clitic: Problems in Serbo-Croatian clitic placement.’ Ms.,
MIT.
Phillips, C. 1996. Order and Structure. Doctoral dissertation, MIT.
Progovac, L. 1993. ‘Locality and subjunctive-like complements in Serbo-Croatian.’
Journal of Slavic Linguistics 1: 119–145.
Progovac, L. 1996. ‘Clitics in Serbian/Croatian: Comp as the second position.’ In:
Halpern, A. and A. Zwicky (eds.) Approaching Second: Second Position Clitics and
Related Phenomena. Stanford: CSLI Publications, 411–428.
Radanović-Kocić, V. 1988. The Grammar of Serbo-Croatian Clitics: A Synchronic and
Diachronic Perspective. Doctoral dissertation, University of Illinois.
Radanović-Kocić, V. 1996. ‘Placement of Serbo-Croatian clitics: A prosodic approach.’
In: Halpern, A. and A. Zwicky (eds.) Approaching Second: Second Position Clitics
and Related Phenomena. Stanford: CSLI Publications, 429–445.
Rivero, M.-L. 1991. ‘Long head movement and negation: Serbo-Croatian vs. Slovak and
Czech.’ The Linguistic Review 8: 319–351.
Rivero, M.-L. 1994. ‘Clause structure and V-movement in the languages of the Balkans.’
Natural Language and Linguistic Theory 12: 63–120.
Rivero, M.-L. 1997. ‘On two locations for complement clitic pronouns: Serbo-Croatian,
Bulgarian and Old Spanish.’ In: van Kemenade, A. and N. Vincent (eds.) Parame-
ters of Morphosyntactic Change. Cambridge: Cambridge University Press, 170–206.
 SECOND POSITION CLITICISATION 119

Roberts, I. 1994. ‘Second position effects and agreement in Comp.’ Paper presented at
the Third Annual Workshop on Formal Approaches to Slavic Linguistics, University
of Maryland.
Rudin, C. 1988. ‘On multiple questions and multiple wh-fronting.’ Natural Language and
Linguistic Theory 6: 455–501.
Schütze, C. 1994. ‘Serbo-Croatian second position clitic placement and the phonology-
syntax interface.’ MIT Working Papers in Linguistics 21: 373–473.
Selkirk, E. 1986. Phonology and Syntax: The Relation between Sound and Structure.
Cambridge, Mass.: MIT Press.
Stjepanović, S. 1995. ‘Short-distance movement of wh-phrases in Serbo-Croatian matrix
clauses.’ Ms., University of Connecticut.
Stjepanović, S. 1998a. ‘A note on placement of Serbo-Croatian clitics.’ Linguistic Inquiry
29: 527–537.
Stjepanović, S. 1998b. ‘On the placement of Serbo-Croatian clitics: Evidence from clitic
climbing and VP ellipsis.’ In: Bošković, Ž., S. Franks and W. Snyder (eds.) Formal
Approaches to Slavic Linguistics: The Connecticut Meeting 1997. Ann Arbor:
Michigan Slavic Publications, 267–286.
Tomić, O.M. 1996. ‘The Balkan Slavic clausal clitics.’ Natural Language and Linguistic
Theory 14: 811–872.
Watanabe, A. 1993. AGR-based Case Theory and its Interaction with the A-bar System.
Doctoral dissertation, MIT.
Wilder, C. and D. Ćavar 1994a. ‘Long head movement? Verb movement and cliticisation
in Croatian.’ Lingua 93: 1–58.
Wilder, C. and D. Ćavar 1994b. ‘Word order variation, verb movement, and economy
principles.’ Studia Linguistica 48: 46–86.
Zec, D. and S. Inkelas 1990. ‘Prosodically constrained syntax.’ In: Inkelas, S. and D. Zec
(eds.) The Phonology-Syntax Connection. Chicago: University of Chicago Press,
365–378.
Zec, D. and S. Inkelas 1992. ‘The place of clitics in the prosodic hierarchy.’ Proceedings
of WCCFL 10: 505–519.
Zwicky, A. and G. Pullum 1983. ‘Cliticisation vs. inflection: English n’t.’ Language 59:
502–513.
 Possessive Constructions and Possessive Clitics
in the English and Bulgarian DP*

Mila Dimitrova-Vulchanova
NTNU — Trondheim

Abstract

A typology of DP-internal possessive expressions is proposed featuring two

types: those realising arguments of the head noun and those realising non-
arguments (‘possessors’). These two types are licensed in different DP-internal
positions in Bulgarian and English. For the analysis of possessive clitics in
Bulgarian a more complex structure is posited than for English in that in
Bulgarian an extra outer DP-shell is needed. The difference between English
’s and the Bulgarian Dative possessive clitic is reduced to the difference
between non-referential clitics (i.e. those realising functional categories) and
referential (e.g. pronominal) clitics.

1. Introduction

In this paper I will explore the diversity of DP-internal possessive expressions in

English and Bulgarian. I will show that the two languages exhibit a considerable
similarity in respect of the overt realisation of the arguments of the head noun
and the mechanisms for the formal licensing of argument expressions. At the
same time, it will be argued that the variation between the two languages is to be
attributed to a difference in the setting of requirements on DP-licensing, as well
as a difference in the morphological status of the elements involved in the
licensing of arguments and quasi-arguments.

* I would like to thank the editors of this volume, Frits Beukema and Marcel den Dikken, and two
anonymous reviewers for thorough discussion and enlightening criticism. All remaining errors are my own.
122 MILA DIMITROVA-VULCHANOVA

Both English and Bulgarian exhibit the standard and well-attested possessive
construction featuring a prepositional phrase, as given in (1a–a′). Both languages
employ possessive pronouns ((1b–b′)), and in addition, one also finds a construc-
tion with an item which in current linguistic practice is referred to as a clitic, as
illustrated in (1c–c′).
(1) a. the leg of the table (English)
a′. pokrivut na kustata
roof-the to house-the (Bg)
b. my house (English)
b′. mojata kusta (Bg)
my-the house
c. John’s house (English)
c′. kustata mu
house-the he(,D) (Bg)
There are two main questions which arise with respect to the data in (1) above.
(1) To what extent are the three alternatives equivalent?
and
(2) What kind of licensing mechanism is employed?
The problem of equivalence in 1) splits up into two related, albeit distinct, issues:
a. whether it can be maintained that the possessive expressions in the three
construction types are all formally licensed in the same structural position,
and
b. whether the position in question can serve for the realisation and licensing of
all types of arguments or not.
In this paper I will address both the issues in 1) and 2). I will argue that there
are essentially two types of DP-internal possessive expressions, respectively
realising arguments proper, and quasi/non-arguments, such as those typically
called ‘possessors’. I will assume, following Grimshaw (1990), that the argument
structure of nominal expressions is contingent on the lexical conceptual structure
(lcs) of the head noun, and that the ways in which a DP-internal expression is
licensed depend entirely on the kind of relationship it has to the lcs of the head
noun. I will claim that so-called ‘possessors’ have no argument status, and that,
unlike arguments proper, they need not be base-generated NP-internally. Hence,
to avoid the inaccuracy associated with the use of ‘possessor’, from now on in
the text I will employ ‘possessive expression’ as a cover term for all types of
DP-internal DPs, PPs and APs which either
 POSSESSIVE CONSTRUCTIONS AND POSSESSIVE CLITICS 123

a. bear overt genitival morphology or occur in a possessive form,

and/or
b. serve as the realisation of an argument of the head noun.
It will be proposed that the structure of a DP with multiple possessive expres-
sions contains at least two licensing positions, a prenominal position and a
postnominal position. I will also argue that the elements in (1c–c′) above are
special clitics, following Zwicky’s (1977) terminology, and as such do not obey
the ‘non-clitic’ syntax of constituents in DP otherwise. It will be suggested that
these clitics serve as licensers in both languages, with a difference, however, in
the type of licensing involved.

2. DP-Internal Possessive Expressions

In this section, I discuss the types of possessive expressions found in DP and the
mechanisms for their formal licensing. I will claim that only agents and themes
are to be considered arguments proper of the head noun, subject to theta-marking
by the latter, while all other types of possessive expressions are formally licensed
by occurring in specific prenominal and postnominal structural positions.

2.1 Arguments and modifiers

The label ‘possessor’ has been used extensively in the literature on possessive
constructions to refer loosely to any type of satellite participant associated with
the referent of the head noun. I will claim that both the term and its content are
misleading and inaccurate. The notion ‘possessor’ generally presupposes a
relation of possession between two entities, the ‘haver’ and the ‘had’. However,
as pointed out by Grimshaw (1990), Barker (1995), among others, there are a
number of relations underlying what is usually expressed cross-linguistically
through a possessive construction, including adverbial satellites, as in e.g.
yesterday’s party in English. It is more than obvious that ‘possessor’ can hardly
apply to the role of yesterday with respect to the head party. The linguistic
tradition has long recognised the fact that possessive expressions can serve as the
realisation of both themes and agents among the arguments of the noun. This is
reflected in the traditional terminology by the labels subjective genitive, and
objective genitive (cf. Jespersen 1924, 1938; Benveniste 1971a, b). According to
Williams’ (1982) Det Rule, the relation between a possessive phrase and the
following N’ can be any relation at all. The question now arises whether
‘possessor’ is a role in its own right, and hence an argument of the head noun or
124 MILA DIMITROVA-VULCHANOVA

not: it has been argued convincingly by Grimshaw (1990) that nouns vary with
respect to whether they have an argument structure or not. It is only nominals
with a-structure which resemble verbs in that their argument slots must be filled,
all other types of nouns not specified for obligatory arguments. Nothing in the
lexical structure of dog suggests a priori that the referent is to be seen in relation
to some other entity, such as e.g. a ‘possessor’.
Grimshaw proposes that there are three ways of licensing DP-internal nomi-
nal expressions: theta-marking (for true arguments), such as in (2a); predication,
which creates modifiers, as the possessive in (2b); and a direct relationship to the
lexical conceptual structure of the noun, like the possessive in (2c).
(2) a. the frequent expression of one’s feelings
b. John’s dog
c. John’s murder
I follow Grimshaw’s main line of argumentation that it should not be taken for
granted that all nouns have argument structure, and that all DP-internal nominal
expressions are necessarily arguments of the head noun. Hence, I will disregard
the requirement that all such expressions are generated in NP-internal positions.
In particular, possessive expressions of the type John’s in (2b) above are
assumed to instantiate the category modifier, not argument. For my current
purposes, I will somewhat modify Grimshaw’s nominal typology,1 and propose
a distinction between argument-taking nouns, such as derived nominals and
relation nouns (cf. Barker 1995 for a discussion of the latter), and the non-
argument-taking ones, to which class all other nouns belong.
Argument-taking nominals are most typically those derived through
nominalisation, as e.g. destruction, examination, unistozavane (destruction),
subirane (gathering) etc. Nouns of this type inherit many of the properties of the
verb from which they are derived, including the theta-role grid. The expressions
realising roles from the thematic structure of such nouns will, as expected, be
generated NP-internally.
(3) a. the instructor’s examination of the papers (English)
b. Ivanovoto unistozavane na dokumentite (Bg)
Ivan- destruction to documents-the
‘Ivan’s destroying the documents’

1. Grimshaw’s categorisation system includes a number of major sub-types, such as complex event
nominals, result nominals, passive nominals, which I will ignore here.
 POSSESSIVE CONSTRUCTIONS AND POSSESSIVE CLITICS 125

The position where instructor and Ivan originate is SpecNP, whereas the PPs of
the papers and na dokumentite are complements of N. In both English and
Bulgarian there are alternative ways for realising both arguments. Thus, a theme
in English can occur as a genitival phrase, as in (4a), or as an AP, as in (4b),
while for agents the only alternative to the genitival phrase is an AP realisation,
as in (4c):
(4) a. the city’s destruction (English)
b. the French defeat
‘the defeat of France’
c. the American invasion
‘invasion by America(ns)’
d. *Ivanovoto unistozenie
Ivan- destruction (*theme) (Bg)
In Bulgarian, the alternative realisation for themes as an AP in prenominal
position is not available, as attested in (4d), while for agents the option illustrated
in (3b) is the only one. In 2.2.2 below I address the options involving a posses-
sive pronoun, and in 3.2, I turn to the option with a possessive clitic. I will argue
that generation in NP-internal positions is sufficient in terms of theta-role
assignment. What we have observed in the examples in (4) above, however, is
that most of these expressions raise to positions above NP, as e.g. city in (4a),
French in (4b). In 2.2 below I suggest a typology of possessive expressions rais-
ing based on three criteria: the need for formal licensing of the expression itself,
the overt presence of specific morphology on the head of the possessive expres-
sion, and the language-specific setting of the requirements on DP-licensing.
The second type of nominal is defined as non-argument-taking based on the
common assumption that such nouns are formally one-place predicates. Thus dog
has the representation D (x), where ‘D’ stands for the predicate ‘dog’, and ‘x’ for
potential referents to be associated with this predicate.2 In contrast to both
argument-taking nouns and verbs, whose arguments are realised as independent
expressions, non-argument-taking nouns always combine the expression of both
the predicate and the argument in a single word, i.e. the head noun dog. When
a possessive expression occurs with such nouns, the ‘hidden’ argument of the
noun is identified with the ‘hidden’ argument of the possessive predicate (cf.
Grimshaw 1990, among others), this operation not corresponding to any overt
process. Hence, as indicated above, I assume that possessive expressions

2. Cf. Higginbotham’s (1985) R-relation.

126 MILA DIMITROVA-VULCHANOVA

occurring in DPs with non-argument taking lexical heads are modifiers on a par
with APs and, as such, do not require NP-internal base-generation.3 Below I
propose mechanisms for their formal licensing.

2.2 The DP-internal licensing positions

I assume that there are essentially two modes of formal licensing available to
possessive expressions in DPs:
a. as a PP complement to N,
and
b. as an XP in the specifier of an Agr(eement)P(rojection) between N and D.
It has been maintained that, unlike verbs, nouns are defective theta-markers (cf.
Emonds 1985; Grimshaw 1990; Chomsky 1995a), and that as a result, they do
not take direct complements. Thus, the only way an argument can be licensed in
postnominal position is by the mediation of a preposition, i.e. the argument ought
to be realised as a DP embedded in a PP. I propose that option a) applies to the
licensing of themes and is sufficient as far as the formal needs of the possessive
expression occurring postnominally are concerned. Possessive expressions
generated prenominally are licensed much in the way subjects in clauses are, in
much the same way as through raising to designated Agr specifiers.
My proposal is an elaboration of Cinque’s (1994) argumentation that all
modifiers of the head noun are generated in the specifiers of agreement projec-
tions, rather than being adjoined. The exact number and labels of projections is
immaterial to our present purposes. What is relevant here are the consequences
of such an approach. Firstly, it allows for a straightforward account of the cross-
linguistically attested natural serialisation of the modifiers of the noun. Secondly,
the various features in nominal expressions, such as number, gender etc. can be
assumed to be associated with the respective Agr head positions. Moreover,
cross-linguistically, these features are related to overt morphology. Thus, both
XP-raising and N-raising, when overt, can be accounted for as driven by the
feature in question. Last but not least, each of the Agr projections can be thought
of as standing for Predication projections, as each modifier is predicated of the
head noun. Some authors have even argued for the need of a Pred(ication)P
inside DP (Camacho, Paredez and Sanchez 1997). While I do not see the

3. Giorgi and Longobardi (1991) analyse ‘possessors’ as adjoined to N″ which is clearly not
NP-internal. Likewise, de Wit (1997) proposes that in ‘picture’ nouns the highest possessive
expression is an adjunct.
 POSSESSIVE CONSTRUCTIONS AND POSSESSIVE CLITICS 127

presence of a predication projection in DP as fully justified, I find the analysis

in terms of Agreement projections quite flexible to cater for this need as well.
The two AgrPs that I adopt are the second projection above NP, which I
label Agr(Prov)P for reasons which will become clear soon (‘Prov’ stands for
‘provenance’) and Agr(Poss) P located immediately below D. While PossP has
been adopted in a number of analyses (cf. Delsing 1993; Veselovská 1995;
Schoorlemmer 1998, de Wit 1997), only de Wit has argued for the relevance of
a lower projection for formal licensing. Note, however, that there is a difference
in what she adopts as the label of the category (in her system NumP) and my
label, as well as the type of raising associated with this projection. In her system,
themes too are obligatorily checked in the prenominal SpecNumP position, which
I do not assume.4 Another difference between the two analyses resides in the
exact location of this projection. While de Wit assumes that NumP immediately
dominates NP, a finer analysis of the area immediately above NP suggests that
there is at least one agreement projection between the projection where lower
possessive expressions are licensed and NP. I will come back to this briefly
below.

2.2.1 The lower licensing position

I propose that Agr(Prov)P is the target of raising for agents in both English and
Bulgarian which raise from SpecNP to SpecAgr(Prov)P. Similarly,
SpecAgr(Prov)P is available for themes in English. As shown in example (4b),
this is the position where all arguments which bear overt adjectival morphology,
i.e. which are realised as APs, are forced to raise due to their morphology.
Raising to SpecAgr(Prov)P satisfies both the need of the expression for formal
licensing, and the need of the adjectival inflection for checking.
The label of this projection stems from the name of the adjectival category,
which in typological studies is recognised as occurring in the area immediately
before the head noun. Adjectives denoting nationality/class/regional belonging are
found here, such as Italian, French, local, liberal etc. There is support for the
position that the lower argument-licensing position coincides with the position
where provenance modifiers are found otherwise. Firstly, there is a lot of
notional similarity between what this category of modifier denotes and the nature
of arguments which I assume occur in SpecAgr(Prov)P. Provenance APs denote
inherent properties of the referent. Likewise, agents and themes — arguments

4. In this way a closer parallel is established between the CP-domain and the DP-domain, NumP
resembling in function AgrOP in the clausal system.
128 MILA DIMITROVA-VULCHANOVA

generated NP-internally — are inherently related to the noun. Secondly, there is

evidence that the two categories are realised in the same structural position,
judging from the linear ordering possibilities of both provenance APs and lower
possessive expressions with respect to another category of low modifier of N, i.e.
material-denoting adjectives. Observe the ill-formed phrases with the material AP
preceding the possessive expression amerikanska/American in (5b–b′), also
attested when it precedes the provenance AP in (6b–b′):
(5) a. amerikanska bombena ataka (Bg)
American bomb attack
a′. an American bomb attack (English)
b. *bombena amerikanska ataka (Bg)
bomb American attack
b′. *a bomb American attack (English)
(6) a. italianska durvena industrija (Bg)
a′. Italian wood industry (English)
b. *durvena italianska industrija (Bg)
b′. *wood Italian industry (English)
The data in (5)–(6) also serve the purpose of illustrating that material-denoting
adjectives immediately precede the head noun, which indicates that they are
located in the specifier of the projection immediately dominating NP. A partial
representation of the lower part of the extended nominal projection is given in
(7) below.
(7) Agr(Prov)P

Spec AgrP
(Poss)
Spec NP

Agent N

N PP

P DP
(Theme)

There is also diachronic support for the lower prenominal licensing position in
English. In Old English (OE), there were two linear positions for genitival
 POSSESSIVE CONSTRUCTIONS AND POSSESSIVE CLITICS 129

phrases: a position immediately preceding the head noun, and a postnominal

position (cf. Traugott 1992; Berndt 1984).5 The OE examples in (8a ,b) (from
Traugott 1992 and Fischer 1992) illustrate the prenominal position. Note that the
head in (8b) is a verbal noun and his feonda realises its theme.
(8) a. Þære geættredan deofles lare
this-. poisoned devil’s-. teaching-
(…head noun)6 (OE)
b. his feonda slege (OE)
his-. enemies-. killing (D)
‘the killing of his enemies’
Gradually, with the loss of case inflections, the prenominal position became unfit
for DPs; however, it can be argued that it was preserved as a position. The
difference between OE and Modern English is that in the latter, this position
split into two distinct projections, Agr(Prov)P, and the agreement projection
whose specifier accommodates material APs respectively. Such a development
makes sense in view of the functions of prenominal genitives in OE, ranging
from ‘possession’ proper, underlying ‘subject’, underlying ‘object’, ‘descriptive
genitive’, to ‘genitive of origin/source’ (cf. Berndt 1984). This suggests that there
was one form (the genitive) and one position (immediately prenominal) corre-
sponding to a number of distinct relations. It is to be expected that later stages
of the language introduced a hierarchical distinction among these categories,
hence, we have two separate projections in Modern English rather than just one,
as in Old English.
An independent argument in favour of Agr(Prov)P as the appropriate
location for both provenance APs and adjectival arguments of the head noun is
that both categories are inflected as adjectives, e.g. French, Italian, American.7

2.2.2 The higher licensing position

The higher licensing position is to be identified with SpecAgr(Poss)P. I assume
that this position is designed to accommodate non-argument possessives, which
are generated in its specifier. Thus, both possessive pronouns and possessive DPs

5. Note that the postnominal position can be accounted for in terms of partial (very short) movement of
the noun to the Agr position immediately dominating NP (cf. Cinque (1994) on partial N-movement).
6. The whole phrase headed by ‘lare’ is a modifier of another noun. Hence, the genitive marking on ‘lare’.
7. This could be accounted for in terms of a specific requirement of the head of this projection, due
to the presence of a [+N, +V] feature.
130 MILA DIMITROVA-VULCHANOVA

of the modifier type, discussed in 2.1 above, originate in SpecAgr(Poss)P, in

English and in Bulgarian as well.
(9) a. Peter’s dog (English)
b. moeto kuce
my-the dog (Bg)
The position of Peter in (9a) and moeto (9b) is the result of subsequent raising
from SpecAgr(Poss)P to SpecDP. This type of raising is for reasons of
DP-licensing, and in English is also due to the clitic properties of genitival ’s. I
come back the latter issue in Section 3. below.
Satisfying the formal definiteness feature of DP can be implemented in two
ways: through an XP base-generated in or moved to SpecDP, or through an item
base-generated in or moved to D. Theoretically, a third option exists with both
SpecDP and D filled; however, the latter seems a less favoured alternative across
languages, for reasons of e.g. economy (cf. Dimitrova-Vulchanova and Giusti
1998). Thus, raising of the possessive pronoun to SpecDP is called for by reason
of DP-licensing, whereas its own needs for formal licensing are already satisfied
by generation in SpecAgr(Poss)P. In Bulgarian, possessive pronoun raising fol-
lows the general mechanism of AP raising, whereby the head of the AP acquires
the article in its own functional projection FP (AP) as agreement with D.
I assume that SpecAgr(Poss)P is the generation site of possessive pronouns
also in the case of argument-taking nouns. The main reason for doing so is the
restrictions obtaining on the realisation of arguments by means of a possessive
pronoun. This is illustrated in (10) below.
(10) a. *negovoto unistozavane na dokumentite
his destroying to documents
(*negovoto=agent) (Bg)
b. his attempt to leave on time (English)
In the Bulgarian example in (10a), the possessive pronoun cannot realise the
agent of unistozavane (destruction) even if the theme is realised as a PP. This is
unexpected in view of Grimshaw’s (1990) generalisation that agent realisation in
event nominals is only possible when the theme is overtly present. Moreover, the
thematic hierarchy ‘agent > theme’ is observed in (10a). Likewise, in English,
(10b), taken from Grimshaw (1990), is only possible when attempt has a result
reading, i.e. functions as a non-argument-taking head. I take these data as an
indication that the possessive pronoun behaves in this way simply because it is
an instance of the modifier type of possessive and is created by predication, not
by theta-role assignment.
 POSSESSIVE CONSTRUCTIONS AND POSSESSIVE CLITICS 131

An alternative to generation in SpecAgr(Poss)P in both Bulgarian and

English is licensing a non-argument possessive in postnominal position. This
option is available, provided the possessive expression is embedded in a PP, as
illustrated in (1a–a′) above, repeated here as (11a–a′).
(11) a. the leg of the table (English)
a′. pokrivut na kustata (Bulgarian)
roof-the to house-the
b. kniga na Ivan
book to Ivan
c. na Ivan kniga
to Ivan book
‘a book of Ivan’s’
c′. *of the table the leg (English)
I assume that PP possessive expressions of the type in (11) above are base-
generated in the complement position of the nominal head (for a similar
proposal for Scandinavian cf. Delsing 1997). In this case the formal licenser of
the embedded DP is the preposition of, and na (to) respectively. Alternatively,
in Bulgarian, the possessive PP may raise to SpecAgr(Poss)P, as shown in
(11c). Unlike Bulgarian, overt raising to SpecAgr(Poss)P is not available in
English, as demonstrated in (11c′). This situation is strongly reminiscent of the
setting of V-raising in English, namely that, in the absence of V-to-I, only I-to-C
is available, thus rendering the lower part of the extended verbal projection
virtually inactive. Similarly, in the DP-domain, only the higher part of the
extended nominal projection is activated in non-argument possessive expression
raising.
Note that if the postnominal PP position was the result of adjunction,
nothing in principle could prevent of the table from adjoining to some other
intermediate or maximal projection inside DP. That the position of the prenom-
inal possessive PP in Bulgarian is not an instance of dislocation is shown by the
data in (12):
(12) a. ceta na Ivan knigata na studentite
read-1 of Ivan book-the to students-the
‘I am reading Ivan’s book to the students’
b. Tova e na Ivan kniga/(-ta)
this is of Ivan book/(-the)
‘This is a book of Ivan’s/Ivan’s book’
132 MILA DIMITROVA-VULCHANOVA

c. ot na basta ti supata
from of father your- soup-the
‘from your father’s soup’
In (12a) the naDP occurs as part of a constituent in object position, in (12b) it
is part of a constituent used predicatively, and in (12c) it is part of the comple-
ment of a preposition. All the data in (12) support treating the nominal possessor
phrase as constituting a part of the extended nominal projection.
So far I have proposed a typology of licensing for DP-internal possessive
expressions in Bulgarian and English, applying to both argument-taking nominals
and non-argument-taking nominals. In the former category arguments originate
NP-internally, and this is sufficient as far as their thematic status is concerned.
For formal licensing, two options are available, a postnominal position with
licensing mediated by a preposition, and a prenominal position in the specifier of
a projection labelled here Agr(Prov)P. The choice of licensing option depends on
the categorial status of the head of the possessive expression. Thus APs invari-
ably select SpecAgr(Prov)P, whereas DPs are either licensed postnominally
(embedded under a preposition) or in SpecDP. The latter option is also driven by
the formal requirements on DP-licensing referred to earlier in this section. For
non-argument nominals a higher licensing position, SpecAgr(Poss)P is employed
to accommodate possessive pronouns and DP possessive expressions. Both types
of expressions raise overtly to SpecDP, again for reasons of DP-licensing. In
addition, non-argument possessive expressions can be licensed postnominally as
PPs. Of the two languages, only Bulgarian displays overt raising of the PP to
SpecAgr(Poss)P. The structure I adopt is sketched in the phrase-marker represen-
tation in (13).
 POSSESSIVE CONSTRUCTIONS AND POSSESSIVE CLITICS 133

(13) DP

Spec D

D Agr(Poss)P

Spec AgrP

Agr(Prov)P

Spec AgrP

NP

Agent N

N PP

P DP
(theme/
non-argument)

2.3 Multiple possessive expressions

In this section I provide a number of examples of multiple possessive expres-

sions which corroborate the proposals made above. This situation is usually
obscured across languages by the fact that such phrases rarely co-occur, as
illustrated in (13a) by the ungrammatical construction with both a genitival
phrase/a possessive pronoun (realising a non-argument) and a PP (denoting the
agent) quoted in Giorgi and Longobardi (1991).8
(14) a. *his/*John’s books of my favourite writer
b. his/John’s books by my favourite writer
c. the American invasion of Vietnam (English)

8. Cf. the discussion in Dimitrova-Vulchanova and Giusti (1999) of more than one possessive phrase
in the same DP.
134 MILA DIMITROVA-VULCHANOVA

d. negovijat Rembrandov portret na Aristotel (Bg)

his- Rembrandt’s- portrait of Aristotle-
‘his portrait of Aristotle by Rembrandt’
However, both English and Bulgarian display multiple occurrences of possessive
phrases with argument-taking nominals, as shown in (14c–d). Note that portret
(portrait), being a relational noun, belongs in this group. The fact that possessive
expressions co-occur supports the idea that there are a number of generation and
licensing positions for DP-internal possessive expressions. The linear string in
(14d) demonstrates the licensing positions I have adopted, the postnominal PP
position for the theme na Aristotel, SpecAgr(Prov)P above NP for the AP- agent
Rembrandov, and SpecAgr(Poss)P for the generation of the possessive pronoun
negovijat (cf. Dimitrova-Vulchanova and Giusti 1999) for a discussion).9

3. The Clitics

3.1 The status of ’s in English

In the wake of Abney (1987), considering ’s as having head status and occupying
D is by far the most prominent assumption. This type of approach recognises the
properties of ’s already noticed in the linguistic tradition, namely that genitive ’s
in English, though a bound morpheme, functions at the ‘phrasal’ level, when
compared to other morphemes, as shown by the fact that units of XP-size are
attached to it ((15)). This has led to labelling ’s as some kind of ‘super’-mor-
pheme, phrasal marker and the like.10
(15) [DP [DP the king of England] [D′ [D ’s] [NP daughter]]]
However, there are alternative views, as the one put forward in Barker (1995)
where the whole possessive expression with ’s encliticised on it is assumed to be
in SpecDP. Thus, ’s is seen as agreement on the phrase in SpecDP with an

9. Possessive pronouns have been argued to occur in D (which is the favoured account, cf.
Bloomfield 1933; Hellan 1986; Radford 1993; Delsing 1997; den Dikken 1997, to mention some).
An alternative is SpecPossP/DP, which is what I adopt, primarily drawing on the parallel with APs
otherwise (cf. Löbel 1996; Giusti 1997; Schoorlemmer 1998 for a similar proposal for English/
German).
10. See Jespersen’s (1938) discussion of the ‘group genitive’ and the evolution of the ’s marker. For
recent morphological analyses treating ’s as a phrasal affix cf. Lieber (1992), Halpern (1995), and see
Jensen (1990) for viewing ’s as a clitic.
 POSSESSIVE CONSTRUCTIONS AND POSSESSIVE CLITICS 135

empty D marked for the feature [+poss]. Barker uses languages such as Hungari-
an as evidence for this type of analysis, where prenominal possessive expressions
co-occur with the article or demonstratives. In view of the possibilities for
licensing DP-internal possessive expressions proposed above, the data from such
languages presents no serious problem for the analysis of English ’s in D.
The advantages of the analysis of ’s as an instantiation of D are twofold. Firstly,
granting ’s head status in terms of X-bar theory presupposes the presence of a
specifier slot which can accommodate a phrasal unit of any size and complexity.
Secondly, with ’s being in a SpecHead configuration with the nominal expression in
its specifier, the obligatory dependence of ’s on a host is accounted for (cf. Radford
1993 for a discussion). It could be added, though, that ’s needs a specifier not
only because it is phonologically weak, but also because it lacks agreement
morphology, hence the obligatory SpecHead configuration. Note that heads which
bear overt agreement morphology may be exempt from having to raise overtly to
the projection in which they are checked against the XP in the specifier.
It now becomes necessary to compare ’s with other elements usually found
in D, such as the article. If we insist that genitive ’s in English obligatorily
projects a specifier, due to its morphological/phonological status — since it needs
to lean on an item in the phonological string, as suggested by Radford (1993) —
we are forced to assume a similar requirement for languages with article mor-
phemes in postposition, notably the Balkan11 and the Scandinavian languages.
It appears, though, that when the article is realised as a morpheme, SpecDP does
not have to be filled. On the contrary, most of the languages exhibiting an article
morpheme tend to have a requirement on DP similar to the doubly-filled comp
filter (e.g. Bulgarian, Rumanian). The article in such languages could be assumed
to be acquired through a mechanism of head-to-head (e.g. N-to–D) move-
ment.12 The latter mechanism highlights an essential difference between
genitive ’s and the article: While the article is to be conceived of as a piece of
morphology attaching to a constituent at the same level, i.e. X, genitive ’s stands
in a relation to an XP-level element. On these grounds, I propose that genitive ’s
is a clitic rather than a morpheme.13 In addition, Zwicky and Pullum (1983)

11. With the exception of Greek.

12. This is essentially the proposal in Delsing (1988), Taraldsen (1990), and Longobardi (1996).
However, it should be noted here that overt N-to-D movement is subject to parametric variation and
should not be taken for granted when a language exhibits an article in postposition (cf. Dimitrova-
Vulchanova and Giusti 1998).
13. If we refer to Anderson’s (1993) useful metaphor of clitics as ‘the morphology of phrases’, the
nature of ’s is insightfully revealed.
136 MILA DIMITROVA-VULCHANOVA

point out criteria by which genitive ’s is to be considered a special clitic, such as

e.g. the lack of a full (tonic) counterpart.
More evidence for the clitic status of ’s is found in data with the so-called
double possessive marking, as in (16a) below, and also in cases where ’s attaches
phonologically to a [−N,+V] category, as in (16b), and coordination structures,
as in (16c).
(16) a. [a friend of mine]’s car
b. [the woman who Mary hit]’s car
c. [John and Mary]’s mother
A clitic can be realised twice on the same attachment node (cf. clitic clusters
with two reflexive clitics in Bulgarian), but an affix cannot, as shown in (16a).
Clitics, but not affixes, exhibit a low selection for their hosts, as seen in (16b),
with ’s phonologically attached to the verb ‘hit’. Clitics, but not affixes, can be
affected by coordination, as affixes must be repeated on the heads of the
phrases in coordination. This is demonstrated by the option in (16c) in English
and in the Bulgarian data in (17) below:
(17) a. [majka-ta] i [dete-to]
[mother-the and [child-the
b. *[maika i dete]-to
‘the mother and the child’
c. [[PP na [DP Ivan i Maria]]…] kusta
to Ivan and Maria house
Note the close parallel between English ’s and the preposition na which licenses
postnominal non-argument possessive expressions in Bulgarian, as demonstrated
in (17c). Here the coordinated DP ‘Ivan i Maria’ is embedded under na, and na
may not be repeated. What is implicationally relevant for English ’s is that the
Bulgarian preposition na is morphologically a non-bound item, and phonological-
ly a simple clitic. Such a parallel in behaviour cannot be accidental.
I will conclude that English ’s has clitic status, and serves as an instantiation
of D. It provides for the SpecHead configuration to whose specifier both argu-
ment and non-argument possessive DPs raise for reasons of DP-licensing. The
latter, as I have proposed above, is related to the setting of the ‘overtly-filled D/
overtly-filled SpecDP’ parameter. I have also suggested that raising to SpecDP
in this case is obligatory for the sake of ’s itself, which, as already discussed,
needs an overt constituent in its specifier.
 POSSESSIVE CONSTRUCTIONS AND POSSESSIVE CLITICS 137

3.2 The Bulgarian possessive clitic

In Section 2 above I presented the possibilities for realising both arguments and
non-arguments in the extended nominal projection by means of APs and DPs. In
Bulgarian, in addition, there exists an independent option employing a Dative
possessive clitic, as shown in (18).
(18) a. knigata mi
book-the I-.
b. novata mi kniga
new-the I-. book
c. tazi mi nova kniga
this I-. new book
As the constructions in (18a, b) attest, the position of the clitic is always linearly
adjacent to (the item bearing) the article, or to the demonstrative, as in (18c).
Traditionally, these facts have been referred to as Wackernagel effects inside the
noun phrase, in that the possessive clitic invariably occurs in second position in
the DP. An interesting issue here is how to define the base position of the clitic.
One possibility is to say that the clitic is base-generated below N; however,
none of the constructions in (18), except for (18a), is explained in this way. An
alternative would be to assume base-generation in the position immediately
below D, thus reflecting the linear second position of the clitic. Note that the two
alternatives are equally defensible if, in addition, either raising of the clitic is
assumed or the analysis relies on overt N-movement. A principled solution will
have to take a stand with respect to these mechanisms. Whether clitic raising
obtains will depend on whether the clitic is viewed as an argument originating in
an NP-internal position or not. As for overt N-movement, the relevant question
is whether it obtains otherwise in the Bulgarian DP. These two issues will be
addressed below.
N-raising could account for all the data in (18) if the clitic is generated in
the position immediately below D. However, there is considerable evidence that
the head noun never raises overtly in Bulgarian, as argued in Dimitrova-Vulcha-
nova and Giusti (1998). The relevant facts are as follows.
Bulgarian exhibits a bound article in postposition, appearing on the first
adjective in the string of modifiers of the noun, or, in the absence of modifiers,
on the head noun itself. Some illustrations are given in (19).
(19) a. kusta-ta
house-the
138 MILA DIMITROVA-VULCHANOVA

b. xubava-ta kusta
nice-the house
c. *xubava kusta-ta
d. nova-ta durvena kusta
new-the wooden house
In Giusti and Dimitrova-Vulchanova (1996) and Dimitrova-Vulchanova and
Giusti (1998) the data in (19a–d) are accounted for in terms of two distinct
mechanisms. The general mechanism of DP-licensing when the modifier is
inflected for the article ((19b–d)) resides in AP raising to SpecDP, whereby the
definite article surfaces on the adjectival head as agreement with D. To ensure
that the article always attaches to the head and not to the AP constituent which
happens to be last in the string,14 we assume the presence of a functional pro-
jection FP dominating AP. As a default (last resort) mechanism, in the absence
of modifiers, the noun can be inflected for the article, as illustrated in (19a). This
option, however, does not involve overt raising of N to D, as the article is in-
serted in situ (also cf. Halpern 1995). Additionally, an analysis along these lines
finds diachronic support in the evolution of the DP in Bulgarian (cf. Dimitrova-
Vulchanova (1996), and Dimitrova-Vulchanova and Vulchanov to appear).
In the absence of N-raising elsewhere in the Bulgarian DP, I conclude that
the approach assuming that overt N-raising only obtains in the clitic construction
is inconclusive and ill-founded.
A clitic raising approach is justified if the clitic were an argument and
subject to theta-role assignment. However, the clitic may realise all three types
of satellites: agents ((20b)) and themes ((20c)), i.e. arguments proper, and non-
arguments, as shown in (20a).
(20) a. kuceto mu
dog-the he-.
‘his dog’
b. bjagstvoto mu
run-away he-.
‘his flight/running away’
c. unistozenieto mu
destruction he-.
‘his destruction’

14. This is especially relevant here, as Bulgarian APs allow for right branch complements (cf. Giusti
and Dimitrova-Vulchanova 1996).
 POSSESSIVE CONSTRUCTIONS AND POSSESSIVE CLITICS 139

How do we accommodate these possibilities with regard to the position in which

the clitic is generated? The clitic could come from any of the positions we have
assumed for the various categories. And if there is more than one potential
generation position, one would expect more than one clitic to be possible,15
which is contrary to fact. Moreover, the clitic is a light element with impover-
ished phonological form, and is more of a place holder of features rather than an
argument. In addition, the clitic plays an essential role in non-argument posses-
sive PP fronting, illustrated in (21):
(21) a. na Ivan knigata mu
to Ivan book-the he-.
b. na Ivan novata mu kniga
to Ivan new-the he-. book
c. *na grada unistozenieto mu
to city destruction he-.
In (21) the clitic mu doubles the fronted non-argument possessive na Ivan. As
attested by the ill-formed instance of doubling in (21c), with the clitic mu
doubling the theme na grada (to city), doubling only obtains with non-argument
possessive expressions. Hence, there is little motivation to assume that we are
dealing with two different clitics in the doubling and ‘non-argument’ construc-
tion, and in the non-doubling construction where the clitic allegedly realises an
argument (cf. (20b–c). This entails that the instantiations of the uniform construc-
tion type illustrated in (20) above correspond to different structures.
I suggest that the possessive clitic instantiates the head of a clitic projection
CLP dominating DP, which is activated for the licensing of possessive DPs
featuring the clitic and in all the cases of non-argument PP fronting. I assume
that in the latter case the function of the clitic is to provide for a specifier
(SpecCLP) position employed by the fronted possessive PP as an escape-hatch
for further raising.

3.2.1 The location of CLP

Schoorlemmer (1998) suggests that the clitic is generated NP-internally and
raises successively, changing category under way: it starts out as an XP which
is licensed in SpecPossP and ends up as a head licensed in D and cliticised on
the head of the constituent in SpecDP. I will demonstrate that an analysis in
terms of DP is insufficient to account for all occurrences of the possessive clitic.

15. This observation is due to an anonymous reviewer.

140 MILA DIMITROVA-VULCHANOVA

Hence, an additional layer of DP-structure appears to be needed. Consider the

contrast in (22) below:
(22) a. negovata (*mu) kniga
his-the (*he-) book
b. negovata si kusta
his . house
‘his own house’
(22a) demonstrates that the clitic does not double a non-argument possessive
pronoun, whereas in (22b), the two may co-occur, and this time the clitic is reflexive.
I take (21) to instantiate the regular case of DP-licensing with a modifier
inflected for the article in SpecDP, as possessive pronouns in Bulgarian behave
as adjectives. However, unlike the construction in (22b), the clitic may not occur
in (22a). One reason, of course, is that in (22a) the illicit clitic is a doubling
clitic, whereas in (22b) it is an anaphoric clitic. However, in terms of structural
position there should be no difference between the anaphoric and the non-
anaphoric possessive clitic, as both are marked for Dative (morphological) case.
Thus, the difference between the two constructions can receive a straightforward
explanation: below D there is no available position for the clitic. Following the
mechanism of DP-licensing with [AP+article] in SpecDP introduced briefly
above, in (22a) the pronominal adjective must be in SpecDP if it bears the
article. For the same reason, it is either in SpecDP or has passed through SpecDP
in (22b). Thus the difference between the constructions in (22a) and (22b) is
reduced to the absence of a clitic site below D vs. its presence above D. The
clitic site above DP is identified with the head of the clitic projection (CLP).
If we accept the view that the clitic raises to D and encliticises on the head
of the constituent in SpecDP (cf. Schoorlemmer 1998), the mechanism ensuring
encliticisation on the head has to be stipulated. Especially in view of the contrast
between the article, which is an affix, and the clitic, which resembles English ’s
in status and properties this would be an undesirable state of affairs. Now
consider the data with complements on a left branch in (23).
(23) a. [otdadenijat na deloto] mu prijatel
[devoted-the to cause-the] he-. friend
b. *[otdadenijat mu na deloto] prijatel
[devoted-the he-. to cause] firend
‘his friend devoted to the cause’
The possessive clitic clearly encliticises on the last constituent in the AP, which
is deloto (cause) in the case of (23a). As shown by the ill-formed (23b), the clitic
 POSSESSIVE CONSTRUCTIONS AND POSSESSIVE CLITICS 141

cannot encliticise on the adjectival head otdadenijat (devoted-the). I take this as

an indication that the whole AP moves as a constituent and the possessive clitic
relates to the phrase as a whole, as a head relates to the constituent in its
specifier. The data in (23) above also point up the substantial difference between
the article and the clitic in terms of mode and site of attachment. They highlight
once more the similarity between the possessive clitic in Bulgarian and ’s in
English in terms of structural status. The difference, however, is in their
locations, ’s instantiating D, the clitic heading its own projection above DP.
To sum up, I have proposed that the location of the possessive clitic is
above DP, drawing on data from clitics co-occurring with possessive pronouns,
and APs with right-branch complements. I have also taken into account the
argument status of clitics and concluded that they do not realise arguments of the
noun, and do not, therefore, require generation in an NP-internal position, which
is parallel to what I proposed for possessive pronouns in 2.2.2 above. I now turn
to the licensing mechanism and the motivation for non-argument possessive PP
raising to SpecCLP.

3.2.2 The motivation for a clitic projection

Following recent analyses in the theory of clitics, most notably Sportiche (1996),
it has become common to assume that clitics are the heads of their own function-
al projections. This type of approach accounts in a unified fashion for both
doubling constructions and non-doubling occurrences of clitics which is achieved
through a mechanism of agreement between a clitic and a full phrase attained in
a SpecHead configuration at some point of the derivation.16 I have argued
above that the possessive clitic in Bulgarian behaves as a head, and that it heads
its own projection. I have also indicated that the motivation for a clitic projection
resides, in part, in its role in possessive expression fronting, illustrated in (20)
above, and repeated here as (24).
(24) a. na Ivan knigata mu
to Ivan book-the he-.
b. na Ivan novata mu kniga
to Ivan new-the he-. book
The full phrase which is related to the clitic is a fronted non-argument possessive
PP, such as na Ivan in (24) above.17 A problem for the CLP analysis, however,

16. Note that either the clitic or the full phrase may be covert.
17. Doubling obtains only with non-argument constituents.
142 MILA DIMITROVA-VULCHANOVA

is the presence of a constituent (e.g., knigata (book-the) in (24a) and novata

(new-the) in (24b)) intervening between the items assumed to be in a SpecHead
configuration, i.e. the possessive expression and the clitic. The structure I adopt
accommodates these data by having a clitic projection with a recursive specifier,
as represented in (25a), illustrating (24b).
(25) a. CLP

SpecCLP CL′

SpecCLP AP CL DP

PP Spec D′

D AgrP

Spec NP
na Ivan novatai mu ti ti ... kniga

b. CLP

SpecCLP1 CL′

PP SpecCLP2 CL′

AP CL DP

Spec D′

D AgrP

Spec NP
na Ivan novatai mu ti ti ... kniga

In the structure in (25a) the possessive phrase na Ivan has moved from the
postnominal position where, as argued in Section 3.2.2 above, it is licensed by
the preposition na. It raises to SpecCLP where it is matched for identical
agreement and referential properties with the clitic mu which instantiates the head
of the clitic projection CL. The AP novata originates in SpecAgrP, and moves
 POSSESSIVE CONSTRUCTIONS AND POSSESSIVE CLITICS 143

successively through SpecDP to adjoin to SpecCLP, thus creating a multiply-

filled specifier. Passing through SpecDP, as already pointed out, is essential to
DP-licensing. This proposal is congenial to the analysis of Bulgarian clause
structure and clitic doubling constructions proposed in Dimitrova-Vulchanova and
Hellan (1999). Alternatively, the analysis can employ a multiple specifier
structure (cf. Chomsky 1995b), as given in (25b). Raising for both na Ivan and
novata will proceed as indicated for (25a), with one difference: in (25b), novata
will raise to SpecCLP2 by adjunction to an intermediate projection.18 I consider
both options equally appropriate.
The main motivation for the analysis proposed here is the fact that posses-
sive clitics serve as licensers to possessive PPs which subsequently raise to the
clausal fronting position. Thus, SpecCLP functions as an escape hatch for such
constituents, provided the clitic is realised overtly. The structure I have adopted
can also accommodate the non-doubling cases, such as novata mu kniga (new-the
cl book), on the assumption that the AP novata raises to SpecCLP in the absence
of a full possessive phrase. In this case, the motivation is similar to the situation
with English ’s when raising is, in part, forced by the needs of ’s itself, i.e. being
a phonologically weak element. A similar motivation underlies the AP raising to
adjoin to SpecCLP as a strategy of saving the construction when the possessive
PP moves out of the DP.

4. Conclusion

In this paper I have proposed a typology of DP-internal possessive expressions.

I have argued that there are essentially two types of such expressions, those
realising arguments of the head noun, and those realising non-arguments (i.e.
those commonly called ‘possessor’). I have proposed that the two types of
expressions are licensed in different DP-internal positions by employing
different formal mechanisms. Bulgarian and English display a similarity with
respect to the positions in question and the licensing mechanisms.
For the analysis of possessive clitics, Bulgarian has to employ a more
complex structure than the one in English. Thus, for Bulgarian an additional
layer of structure is needed, i.e. the outer DP-shell, dominating DP, whereas in
English the genitival construction is accommodated at the level of DP. In both
languages the analysis employs the SpecHead agreement configuration for an XP

18. This option was suggested to me by an anonymous reviewer.

144 MILA DIMITROVA-VULCHANOVA

and a clitic head. Both heads behave in a uniform way with respect to attachment
mode and attachment site, and both items attract phrases in their specifiers due
to their clitic status. The difference, however, resides in the nature of the clitic
heads in the two languages. Whereas English genitive ’s is not referential and is
on a par with the functional heads found in D (such as, e.g. the article), the
Bulgarian clitic head is not functional in nature. This is reflected in the function
that the two clitics have: in English, ’s provides for a SpecHead configuration
employed for the licensing of possessive DPs, whereas in Bulgarian, the posses-
sive clitic licenses possessive raising out of DP. Thus, the difference between
the English ’s and the Bulgarian Dative possessive clitic can be reduced to the
difference, found otherwise across languages, between non-referential clitics (i.e.
those realising functional categories) and referential (e.g. pronominal) clitics.

References

Abney, S. 1987. The English Noun Phrase in its Sentential Aspect. Doctoral dissertation,
MIT.
Anderson, S. 1993. ‘Wackernagel’s revenge: Clitics, morphology and the syntax of
second position.’ Language 69: 68–98.
Barker, Ch. 1995. Possessive Description. Stanford, Calif.: CSLI Publications.
Benveniste, E. 1971a. ‘Toward an analysis of case functions: The Latin genitive.’ In:
Problems in General Linguistics. Coral Gables, Florida: University of Miami Press,
121–130.
Benveniste, E. 1971b. ‘The linguistic functions of “to be” and “to have”.’ In: Problems
in General Linguistics. Coral Gables, Florida: University of Miami Press, 163–180.
Berndt, R. 1984. A History of the English Language. Leipzig: VEB Verlag Enzyklopädie.
Bloomfield, L. 1933. Language. New York: Holt.
Camacho, J., L. Paredez and L. Sanchez 1997. ‘The genitive clitic and the genitive
construction in Andean Spanish.’ In: Black, J. and V. Motapanyane (eds.) Clitics,
Pronouns and Movement. Amsterdam: John Benjamins, 23–37.
Chomsky, N. 1995a. ‘Bare phrase structure.’ In: Webelhuth, G. (ed.) Government and
Binding Theory and the Minimalist Program: Principles and Parameters in Syntactic
Theory. Oxford: Blackwell, 383–439.
Chomsky, N. 1995b. The Minimalist Program. Cambridge, Mass.: MIT Press.
Cinque, G. 1994. ‘On the evidence for partial N-movement in the Romance DP.’ In:
Cinque, G., J. Koster, J-Y. Pollock, L. Rizzi and R. Zanuttini (eds.) Paths Towards
Universal Grammar. Studies in Honor of Richard Kayne. Washington, D.C.: George-
town University Press, 85–100.
Delsing, L. O. 1988. ‘The Scandinavian noun phrase.’ Working Papers in Scandinavian
Syntax 42: 57–79.
 POSSESSIVE CONSTRUCTIONS AND POSSESSIVE CLITICS 145

Delsing, L.O. 1993. The Internal Structure of Noun Phrases in the Scandinavian Languag-
es: a Comparative Study. Doctoral dissertation, Lund University.
Delsing, O. 1997. ‘Possession in Germanic.’ Paper presented at the Annual DGfS
Conference, Düsseldorf, Februari 1997.
den Dikken, M. 1997. ‘Anti-agreement in possessive constructions.’ Paper presented at
the Workshop on Clitic Phenomena in English and other European Languages,
ESSE/4, Debrecen, Hungary.
Dimitrova-Vulchanova, M. 1996. Bulgarian Noun Phrase Structure. Paper presented at the
University of Trondheim.
Dimitrova-Vulchanova, M. and G. Giusti. 1998. ‘Fragments of Balkan nominal structure.’
In: Alexiadou, A. and C. Wilder (eds.) Possessors, Predicates and Movement in the
DP. Amsterdam: John Benjamins, 333–360.
Dimitrova-Vulchanova, M. and G. Giusti. 1999. ‘Possessors in the Bulgarian DP.’ In:
Dimitrova-Vulchanova, M. and L. Hellan (eds.) Topics in South Slavic Syntax.
Amsterdam: John Benjamins, 163–192.
Dimitrova-Vulchanova, M. and L. Hellan. 1999. ‘Clitics and Bulgarian clause structure.’
In: Van Riemsdijk, H. (ed.) Empirical Approaches to Language Typology, Clitics in
the Languages of Europe. Berlin: Mouton de Gruyter, 469–513.
Dimitrova-Vulchanova, M. and V. Vulchanov. To appear. ‘The evolution of the article
and the structure of the Bulgarian DP.’ Proceedings of XIV Nordiska Slavistmötet.
Helsinki. August 1997.
Emonds, J. 1985. A Unified Theory of Syntactic Categories. Dordrecht: Foris.
Fischer, O. 1992. ‘Syntax.’ In: Blake, N. (ed.) The Cambridge History of the English
Language, vol II., Cambridge: Cambridge University Press, 207–408.
Giorgi, A. and G. Longobardi 1991. The Syntax of Noun Phrases. Cambridge: Cambridge
University Press.
Giusti, G. 1997. ‘The categorial status of determiners.’ In: Haegeman, L. (ed.) The New
Comparative Syntax. London: Longman, 95–123.
Giusti, G. and M. Dimitrova-Vulchanova 1996. ‘Quantified noun phrase structure in
Bulgarian.’ In: Toman, J. (ed.) Formal Approaches to Slavic Linguistics 3. Ann
Arbor: University of Michigan Press, 123–144.
Grimshaw, J. 1990. Argument Structure. Cambridge, Mass.: MIT Press.
Halpern, A. 1995. On the Placement and Morphology of Clitics. Stanford, Calif.: CSLI
Publications.
Hellan, L. 1986. ‘The headedness of NPs in Norwegian.’ In: Muysken, P. and H. van
Riemdijk (eds.) Features and Projections. Dordrecht: Foris, 89–122.
Higginbotham, J. 1985. ‘On Semantics.’ Linguistic Inquiry 16: 547–593.
Jensen, J. 1990. Morphology: Word Structure in Generative Grammar. Amsterdam: John
Benjamins.
Jespersen, O. 1924. The Philosophy of Grammar. George Allen and Unwin: London.
Jespersen, O. 1938. Growth and Structure of the English Language. Oxford: Blackwell.
146 MILA DIMITROVA-VULCHANOVA

Lieber, R. 1992. Deconstructing Morphology: Word Formation in Syntactic Theory.

Chicago: Chicago University Press.
Longobardi, G. 1996. ‘The syntax of N-raising: A minimalist theory.’ Ms., University of
Utrecht.
Löbel, E. 1996. ‘Kategorisierung der Possessiva als Adjektive in der NP/DP.’ In: Tappe,
H. and E. Löbel (eds.) Die Struktur der Nominalphrase. Wuppertaler Arbeitspapiere
zur Sprachwissenschaft, vol. 12: 58–94.
Radford, A. 1993. ‘Head-hunting: On the trail of the nominal Janus.’ In: Corbett, G., N.
Fraser and S. McGlashan (eds.) Heads in Grammatical Theory. Cambridge: Cam-
bridge University Press, 73–113.
Schoorlemmer, M. 1998. ‘Possessors, articles and definiteness.’ In: Alexiadou, A. and C.
Wilder (eds.) Possessors, Predicates and Movement in the DP. Amsterdam: John
Benjamins, 55–86.
Sportiche, D. 1996. ‘Clitic constructions.’ In: Rooryck, J. and L. Zaring (eds.) Phrase
Structure and the Lexicon. Dordrecht: Kluwer, 213–277.
Taraldsen, K. T. 1990. ‘D-projections and N-projections in Norwegian.’ In: Mascaró, J.
and M. Nespor (eds.), Grammar in Progress. GLOW Essays for Henk van Riemsdijk.
Dordrecht: Foris, 419–431.
Traugott, E. 1992. ‘Syntax.’ In: Hogg, R. (ed.) The Cambridge History of the English
Language, vol I. Cambridge: Cambridge University Press, 168–289.
Veselovská, L. 1995. Phrasal Movement and X-morphology. Doctoral dissertation, Palacky
University, Olomouc.
Williams, E. 1982. ‘The NP cycle.’ Linguistic Inquiry 13: 277–295.
de Wit, P. 1997. Genitive Case and Genitive Constructions. Doctoral dissertation,
University of Utrecht.
Zwicky, A. 1977. On Clitics. Bloomington: Indiana University Linguistics Club.
Zwicky, A. and G. Pullum 1983. ‘Cliticization vs. inflection: English n’t.’ Language 59:
502–514.
 Agreement as a Continuum
The Case of Spanish Pronominal Clitics*

Jon Franco
Universidad de Deusto-Bilbao

Abstract

In this paper the (morpho)syntactic status of object clitic-doubled constructions

in Spanish is investigated. It is proposed on the basis of evidence gathered from
morphology and syntax that Spanish object clitics should be analysed as object
agreement morphemes on the verb, on a par with subject–verb agreement, and
not as phonologically dependent pronominal arguments. In clitic doubling
constructions the doubled NP object that the clitic is related to occupies the
specifier position of the AgrOP whose head harbours the object clitic.

1. Introduction

This paper investigates the (morpho)syntactic status of object clitic-doubled

constructions in Spanish. Specifically, I pursue and refine the widespread idea
that Spanish object clitics should be analysed as object agreement morphemes on

* This study has benefited from valuable suggestions and extremely insightful commentary from
Joseph Aoun, Bernard Comrie, Josep Fontana, Alazne Landa, Errapel Mejías-Bikandi, Mario
Saltarelli, Carmen Silva-Corvalán and María Luisa Zubizarreta. I am also grateful for their feedback
to the audience at the Workshop on Clitic Phenomena in English and other European Languages at
Debrecen in September 1997, and to the audience at the 1995–96 Linguistic Colloquium Series held
jointly at the University of the Basque Country and Universidad de Deusto. Comments from two
anonymous reviewers helped me clarify this paper throughout. Finally, Frits Beukema and especially
Marcel den Dikken have played a crucial role in bringing about the final version of this article. I am
indebted to them both for their logistic support as well as for their academic guidance. All errors are
nobody’s fault but mine.
148 JON FRANCO

the verb, on a par with subject–verb agreement, and not as pronominal arguments
that are phonologically dependent. This paper could be divided into two natural
bodies of argumentation. The first one is devoted to pinpointing pieces of
evidence that motivate the core insight that clitics are verbal agreement mor-
phemes and that their relation to the elements they ‘double’ is that of a verb-
argument agreement relationship. The second one concentrates on the technical
implementation of this hypothesis within the complex syntax of clitic-doubled
constructions in Spanish. To put it differently, there are two rivers running
through this paper, namely, the descriptive river and the formal river. The
descriptive one allows us to have a good panorama of the distribution of the
data, whereas the formal one, channelled in early minimalist theory, endows my
account of the phenomenon with coherence and explanatory adequacy.
Evidence for the agreement morpheme status of object clitics is drawn from
the morphological component as well as from the syntactic one. The fixed order
of clitics, the strict adjacency to the same host (i.e. the verb or auxiliary), and the
variation in the agreeing features, that is, the leísmo and lo/laísmo phenomena,
for instance, morphologically advocate the morpheme status. From the syntactic
point of view, the fact that the clitic-doubled elements originate in argument
positions also argues in favour of our hypothesis.
This study is organised in three sections. Section 2 is designed to review
some representative previous analyses in competition for Romance clitics, that is,
Kayne’s (1975) movement hypothesis, Jaeggli’s (1982, 1986) base-generation
hypothesis, Torrego’s (1994) and Uriagereka’s (1995) DP hypothesis and
Sportiche’s (1996) clitic voice approach. Along with the discussion of these four
proposals, I point out some of their conceptual and empirical problems, which in
turn motivate the present research. Section 3 analyses the morphological proper-
ties of object clitics as well as a number of outstanding characteristics of the
syntax of clitic doubling constructions. As a matter of fact, we observe that, from
the morphological point of view, Spanish object clitics do not differ much from
attested verb–object agreement systems such as that of Basque. Earlier objections
to the agreement analysis are discussed and shown to fall within the patterns of
cross-linguistic agreement relations. Section 4 develops the line of thought
defended in previous sections in the syntax. Thus, Spanish object clitics are
syntactically mapped as AgrO heads whose specifiers are to be occupied either
by a pro or by the doubled NP object that the clitic head is related to. The latter
option can be realised via movement from within the VP at some point of the
derivation, that is, covertly or overtly. This, for instance, enables us to give a
standard account of Case licensing in clitic doubling structures, along the lines
of Chomsky (1993, 1995).
 AGREEMENT AS A CONTINUUM 149

2. Theoretical Positions and Analyses of Romance Object Clitics in

Generative Grammar

2.1 Introduction

In this research, I have adopted as a framework some of the main assumptions

of the latest version of Principles and Parameters Theory (cf. Chomsky and
Lasnik 1993 and much related work) and its recent offspring, the Minimalist
Program. The core of the theory can be found in the works of Chomsky (1993,
1995). Thus, the hierarchical structural configuration for the syntactic realisation
of clausal agreement assumed here would be as in (1) below:
(1) AgrSP

Spec AgrS

AgrS TP

Spec T

T AgrOP

Spec AgrO

AgrO VP
As shown in (1), there are two Agr elements in the current articulated IP
structure,1 one involved in subject agreement and nominative Case, the other
involved in object agreement and objective Case. Chomsky (1993) points out the
advantages of adopting the structure in (1). In this configuration, AgrO selects a
VP, hence it must be ‘close to the verb’, thus providing us with a structural
landing site for verb raising in infinitival clauses and eliminating vacuous AgrS
in these clauses.

1. I am aware that Chomsky (1995: Chapter 4) has discarded independent Agreement Projections.
Still, there exists the possibility of articulating agreement heads adjoined to other functional
categories. For our purpose, if we posited a complex T/AgrS head and accommodated AgrO onto an
Aspectual head, we would obtain the same results (cf. Laka 1993); however, for the sake of
exposition, I have used in this paper early Minimalist functional projections which have become
relevant for the discussion of the primary data presented here.
150 JON FRANCO

Curiously, the proposals for the positing of the object Agreement node have
been built either on languages that have very restricted object agreement such as
French, which only has participial object agreement (see Kayne 1989b), or
languages like English that have abstract agreement (at the LF level, according
to Chomsky 1993). Less attention has been paid, however, to those languages
that exhibit a full-fledged verb object agreement paradigm that works on a par
with subject agreement. Be that as it may, the possibility of having two Agr
nodes together with the assumption that agreement relations hold in Spec–Head
configurations independently of whether this is done overtly or abstractly endows
the theory with greater descriptive and explanatory power.

2.2 Romance object clitics in generative grammar

There have been two main competing analyses with respect to Romance clitics
in the generative grammar literature. On the one hand, we have the analysis put
forward in Kayne (1975, 1989a) and also adopted by Rizzi (1986) in which,
roughly, object clitics are pronominal arguments generated in the canonical
position of the arguments of the verb; subsequently, clitics attach to the verb by
the rule of move-a abiding by all the theoretical constraints on movement. On
the other hand, we have Aoun’s (1979), Borer’s (1984), Jaeggli’s (1982, 1986),
and Suñer’s (1988) proposals, which, inspired by work of Strozer (1976) and
Rivas (1977) in the transformational framework, argue in favour of a solution
that generates clitics as affixes attached to their host (i.e. the verb), thus
constituting a chain with the argument positions of the syntactic categories they
stand for. The former analysis has been referred to in the literature as the
movement hypothesis, whereas the latter has been referred to as the base-genera-
tion hypothesis. These two competing analyses are reviewed and discussed in
detail in the present section. As will be shown, both hypotheses are going to turn
out to be descriptively faulty and incur principled violations in their own
theoretical framework when confronted with data from Spanish.
Furthermore, a more recent approach to object clitics in Spanish and verb-
argument agreement in Basque put forward in Torrego (1994) and Uriagereka
(1992b, 1995), respectively, will be examined here. According to these authors,
Spanish object clitics and Basque verb-argument agreement morphemes are
generated as determiner heads of DPs; therefore, I will refer to this proposal as
the Determiner Head Hypothesis. To finish the review of the most relevant
literature, I present a brief sketch of Sportiche (1996).
 AGREEMENT AS A CONTINUUM 151

2.2.1 The movement hypothesis

In this subsection, I will illustrate the claim that the movement hypothesis by
which clitics are lexical pronominal heads that project noun phrases is highly
jeopardised when confronting structures where the clitic is duplicated by a noun
phrase of the same grammatical function. Be that as it may, let us assume for the
sake of the discussion that on the movement analysis object clitics are generated
in the canonical position of the internal verb arguments and due to either their
weak phonological nature (they are always unstressed) or their affixal nature
(they are bound morphemes), they must attach to a host (presumably, a phono-
logical head). This is illustrated by the French example in (2):
(2) Pierre lai a vue ti
Pierre -3 have-3 seen
‘Pierre has seen her.’
According to my informants, the co-occurrence of a clitic with another NP
displaying the same Case or grammatical function is banned from French, as
shown in (3):
(3) *Pierre lai a vue Sandrinei
Pierre -3 have-3 seen Sandrine
‘Pierre has seen her Sandrine.’
The contrast between (2) and (3) is to be expected under Kayne’s (1975)
movement hypothesis since the structure-preserving principles would disallow the
nominal Sandrine to occupy the position of a trace that results from movement.
However, the Spanish equivalent of (3) is perfectly grammatical, as shown
in (4):
(4) Pedro lai ha visto a Sandrai
Pedro -3 have-3 seen to Sandra
The Spanish sentence in (4) and any other sentence in which the clitic and its
nominal counterpart are not in complementary distribution pose a problem for the
movement hypothesis from various perspectives. To begin with, the structure-
preserving principles would be nullified for Spanish at first sight, which is
unlikely and undesirable from the point of view of a formal grammar since we
assume that an element cannot occupy the position of the trace of another
element. Second, if the clitic la ‘her’ in (4) as well as the coreferential direct
object a María were endowed with argument status, this analysis would violate
152 JON FRANCO

the second clause of the q-Criterion of Chomsky (1981).2

In a last attempt to save the movement hypothesis, one could claim, in the
spirit of Aoun (1981) and Hurtado (1984) (both subscribers to the alternative
hypothesis), that the doubled NPs occupy A′-positions, so that they would be like
dislocated elements. This proposal has been proven to be untenable by Jaeggli
(1986) and Suñer (1988) in view of the extraction properties that these constitu-
ents display. In brief, Jaeggli shows that rightward extraction of the clitic-
doubled nominal is subject to subjacency, which is something unexpected for
dislocations, as shown in the contrast between (5a) and (5b):
(5) a. A Juani, que le/loi hayan visto en la fiesta no me molesta
(Jaeggli 32e)
‘Juan, that they have seen him- at the party does not bother me’
vs.
b. *Que loi hayan visto en la fiesta no me molesta, a Juani
‘That they have seen him- at the party does not bother me,
Juan’
Also, Suñer points out that wh-extractions such as the ones illustrated in (6)
would be rather unorthodox if the launching position were an A′-position:
(6) ¿A quiéni lei pegaste ei?
¿to whom -3 hit--2
‘To whom did you hit?’
With respect to Binding Theory, one could say that the doubled nominal has to be
in an A-position since it can serve as the antecedent of an anaphor as in (7), and
anaphors must be bound from A-positions according to Principle C of this theory:
(7) a. El decano (lesi) habló a los estudiantesi de
the dean (--3 talked to the students about
sí mismosi
themselves
‘The dean talked to the students about themselves.’

2. Basically, the Theta Criterion monitors the distribution of thematic roles:

Theta Criterion
Each argument is assigned one and only one theta role. Each theta role is assigned to one and
only one argument.
 AGREEMENT AS A CONTINUUM 153

b. *El decano lesi habló de sí mismosi a los

the dean --3 talked about themselves to the
estudiantesi
students
‘The dean talked about themselves to the students.’
The contrast between (7a) and (7b) illustrates that it is the doubled NP that
counts as the binder of the anaphor and not the clitic trace. If we destroy the
configuration in which the clitic-doubled NP can c-command the anaphor, the
sentence is ill-formed. Moreover, if the clitic or its trace were the valid c-com-
manding antecedent for the anaphor when clitic doubling occurs, (7a) would be
wrongly ruled out as a violation of Principle C since the clitic-doubled NP would
not be able to stay free in its governing category.
The last piece of evidence from Binding Theory arguing in favour of the
argumental status of the clitic-doubled element comes from the fact that object
anaphors in Spanish can and must be clitic-doubled:
(8) Juan sei perjudicó a sí mismoi
Juan  hurt himself
Crucially, the data in (8) constitutes an obstacle to a possible extension to
Spanish of Aoun’s (1993) analysis of clitic-doubled elements in Lebanese Arabic
as adnominals since object anaphors do not occupy non-argumental positions.
There are additional motivations that induce us to reject the idea that the
clitic-doubled element is an adjunct or a dislocated element, for instance; the
absence of a pause between this element and the rest of the sentence and the
possibility of embedding clitic-doubled constituents, both facts already pointed
out in Suñer (1988) and illustrated in (9):
(9) Lo último que escuché, claro que lai encontré
the last that listened-1 of course that --3 found
pesada la audicióni, fue el reportaje
boring the program was the report
‘The last thing I listened to, of course I found (it) boring the radio
program, was the interview.’ (Barrenechea and Orecchia 1979,
translation provided by Suñer 1988)
Moreover, at the theory level, the movement analysis adds a new and unwarrant-
ed dimension to the argument/adjunct distinction. That is, since clitic doubling in
Spanish is optional with nominals (cf. example 7a), identical objects occupying
similar positions will be adjuncts or arguments depending on whether these
154 JON FRANCO

objects are doubled by a clitic or not. Thus, in (7a) the indirect object a los
estudiantes ‘to the students’ would be a complement of the verb if the clitic is
not realised, and an adjunct if the duplicating clitic is present. Additionally,
assuming that clitic doubling was a strategy to encode nominal adjuncts, one may
wonder why Spanish has not kept (diachronically speaking) prepositional clitics
to do the same with prepositional phrases. Most importantly, the occurrence of
clitic doubling in ECM structures constitutes conclusive evidence against the
adjunct status of the clitic doubled element. Subjects of small clauses and
subjects of ECM verbs are unlikely to occupy an A′-position (Pat Schneider,
personal communication), and crucially, both elements can be clitic-doubled in
Spanish, as exemplified in (10) and (11):
(10) Los/lesi consideraron a los acusadosi culpables
- considered the defendants guilty
(11) Lei dejó a Pedroi terminar el asunto
- let-3 Pedroi finish the issue
Interestingly, in Catalan, a clitic doubling language across the board, i.e. with
NPs and PPs, the doubling of these subjects does not yield felicitous sentences
(Josep Fontana, personal communication). In fact, clitic doubled elements in
Catalan must be preceded by a small pause.
I would like to state clearly that the movement hypothesis and all its
variants has been dismissed here as a possible analysis for Spanish object clitics.
However, nothing has been said in this paper about the nature of French and Ital-
ian object clitics. As for Catalan, we could assume that clitic doubling construc-
tions in this language (or Lebanese Arabic for that matter) stand for something
else. Specifically, the ECM facts show that Catalan is definitely different from
Spanish when it comes to the analysis of clitic doubling. Actually, the traditional
right dislocation approach rejected above for Spanish does not seem too unrea-
sonable for Catalan. Therefore, our insight that not all languages make use of
clitic doubling strategies in the same way has a robust empirical basis.3

2.2.2 The base-generation hypothesis

The base-generation approach has meant a step forward with respect to previous
analyses of Romance clitics by overcoming quite successfully some of the
problems that the movement hypothesis had to face, yet, this more recent hypothe-
sis is not free from problems either. The basic tenet of the base-generation

3. I owe the clarification of this idea to Marcel den Dikken.

 AGREEMENT AS A CONTINUUM 155

hypothesis is that the clitic is originally generated to the left of the verb, that is,
the position where the clitic appears on the surface, as examples (6) through (11)
illustrate. In this way, the canonical object position would be available for a
coreferential NP if clitic doubling was to take place. Yet, we still have two tasks
pending: (i) to determine the nature of the structural position of the clitic, and (ii)
to specify how this position fits within the X′-schema. Along the lines of Rivas
(1977), Jaeggli (1982, 1986) and Borer (1984), Romance clitics should appear in
the following configuration:
(12) VP

Vn NP

CL V

Nevertheless, Borer (1986) and Saltarelli (1990) note the drawbacks encountered
by adopting the structure in (12) for clitics. Assuming that the pronominal clitics
enter into syntactic operations — and they do since they undergo movement such
as clitic climbing,4 and may serve as A′ binders (cf. Aoun, 1985) — the Lexical
Integrity Hypothesis of Chomsky (1970), which in broad terms states that
syntactic operations cannot look into the composition of lexical items, must be
abandoned if Vn equals V0, i.e. if the constituent [v CL + V] forms a lexical
item. If Vn equals V′, as proposed in Jaeggli (1982), we do not have an explana-
tion for the affix-like properties of clitics. In Section 4, I will give a solution to
this puzzle by considering Spanish object clitics functional heads (AgrO) in the
inflectional domain.
The main burden that the base-generation hypothesis had to face at the time
was that its subscribers still considered clitics as arguments. This implies that
their analyses must have a mechanism to assign Case to the clitic as well as to
the doubled NP in a sentence like (8) or (9) above. Given the structure in (12),
it seems unlikely that one can account for Case and q-role assignment without
abandoning the null hypothesis. Thus, Aoun (1981) and Hurtado (1984) propose
that clitics may be q-role absorbers, so the doubled NP becomes an adjunct. This
possibility was already argued against in Section 1.3.1. In parallel fashion,
Jaeggli (1982) suggests that clitics absorb government from the verb, hence, they
are Case absorbers.
In terms of Case assignment, leaving aside considerations of simplicity, the

4. See, however, the discussion of clitic climbing in Section 3.2 for an alternative view.
156 JON FRANCO

base-generation analysis would hold for example (4). Along the lines of Jaeggli
(1982), the verb gets the accusative Case absorbed by the clitic whereas,
simultaneously, the preposition-like element a discharges or — as refined in
Jaeggli (1986) — transmits it on the nominal. However, this assumption,
although accurate for Romanian which only allows clitic doubling with NPs
introduced by the preposition pe, is not sufficient to account for additional cases
of Spanish clitic doubling without a preposition, such as the one shown in (13):
(13) Lai comí la tortai (Southern Cone Spanish)5
it- ate-I the cake
‘I ate the cake.’
Be that as it may, the most significant insights of the base-generation hypothesis
(which I adopt in the sections that follow) are: (i) the idea that object clitics do
not move from canonical object positions in languages which exhibit clitic
doubling like Spanish, and (ii) the identification by the clitic of the empty
position of the object as pro, as claimed in Jaeggli (1986) and illustrated in (14):
(14) Juan lai miró proi
Juan her- saw
‘Juan saw her.’
This assumption indirectly leads us to give an equivalent treatment, saving the
obvious differences of course, to the clitic doubling phenomenon and subject–
verb agreement morphology since both license pronominal null verb-arguments
and satisfy the thematic properties of the verb in this way. This well-established
parallelism constitutes the backbone of my alternative analysis of object clitics
in Spanish, which is developed in Sections 3 and 4.

2.2.3 The determiner head hypothesis

A different solution to the problem of clitic doubling is sketched out in the
works of Torrego (1994) and Uriagereka (1992a, b). Both authors consider verbal
agreement markers and clitics participating in clitic doubling constructions as
equivalent, that is, as determiners that head their own projection. Hence, I will
refer to this analysis as the Determiner Head Hypothesis. Under this hypothesis,
one should assume that in clitic doubling constructions, the doubled phrase
would initially occupy the specifier position of the DP headed by the clitic, in

5. Southern Cone Spanish is a comprehensive cover term for some main stream Spanish varieties
spoken in Chile, Argentina and Uruguay. Needless to say, sometimes grammaticality judgements
might be subject to variation among the speakers from these countries.
 AGREEMENT AS A CONTINUUM 157

the same fashion as arguments occupy the specifier of a DP headed by an

agreement marker in Basque, as shown in the tree in (15) ((34) in Uriagereka,
1992a: 294):
(15) CP

Spec C

IP C

Spec I

VP Infl

DP

DP-E D V

NP D DP V
pro Agr-E
DP-A D

NP D
pro Agr-A

Notice that, since Basque is a head-last language, all X0 categories are projected
to the right. The Ergative subject in Basque DP-E appears as the specifier of the
highest DP within the VP and occupies the appropriate position in the configura-
tion to concord with a D ergative agreement head AgrE. An exact parallelism
can be established for the absolutive direct object DP-A, provided that the latter
is generated as a left-hand sister to V. A particular feature of this mapping is that
all the D heads take a pro as their sister complement.
The mapping of Spanish clitic doubling constructions onto the syntactic tree
would be the same as (15), but with heads projected to the left of their comple-
ments (see ex. 17 below). From the point of view of explanatory adequacy, this
proposal appears to have more advantages than the ones seen so far, since it
subsumes different possible analyses for pronominal clitics on the one hand, and
verbal agreement markers on the other. Despite the high degree of uniformity
across languages that this proposal accomplishes, it does not seem to be problem-
158 JON FRANCO

free, however. As we follow the mapping of (15) into further levels of derivation
along the lines of Uriagereka, a number of problems arise. Thus, Uriagereka
proposes the derivations via incorporation that are shown in (16) and (17) to
obtain the surface orders in Basque and Spanish, respectively:
(16) a. Zu-k Jon ikusi du-zu
you- Jon- seen 3-have-2
‘You have seen Jon.’
b. CP

Spec C

XP C

Zuk X

YP X

Jon Y X Agr-E
zu
VP Y

DP Y Agr-A
du
DP-E D

NP D V
pro
DP V
ikusi
DP-A D

NP D
pro

(17) a. Tu le/lo6 vi.ste a Jon

you 3 saw.2 to Jon
‘You saw Jon.’

6. See the discussion on the alternation of the lo and le forms in Section 3.2.1.
 AGREEMENT AS A CONTINUUM 159

b. XP

DP X′
Tu
X

X Clitic Tense′
le
Tense

Tense Agr Mood′

ste
Mood T Mood Aspect′
t
Aspect M Aspect Theme
t
Theme A [a/e/i] VP
t
V [a/e/i] DP V′
vi
DPS D′ Vt DP
t
D NP DPDO D′
pro Jon
D NP
t pro

According to Uriagereka, in Basque, the specifiers of the DPs as well as their

heads undergo ‘massive raising’ and end up in the same configuration in which
they were at D-structure, as can be seen in (16). A priori, there are at least a few
questionable points in this analysis. From the perspective of economy of
derivation (cf. Chomsky 1993), ‘massive raising’ in Basque seems to be far from
a least effort operation. To begin with, the movement of the arguments as a type
of NP-movement is totally unmotivated if one assumes the standard position that
NPs, or DPs for that matter, move from non-Case positions to Case positions to
receive their Case. Now, there is no reason why the DP arguments cannot get
Case in their initial configuration shown in (15). The ergative DP-E is able to
receive Case from the Agreement marker, the two being in a Spec–Head relation,
whereas the absolutive DP-A can obtain Case either by the Spec–Head relation
that it maintains with the absolutive agreement marker, or in former terms, by
government from the verb. Also, it is worth pointing out that ‘massive raising’
in (16) does not yield the final constituency of the elements since a supplementa-
ry movement of head to head incorporation is necessary in order to amalgamate
the absolutive and ergative agreement markers, which is the way they appear on
the surface in Basque.
160 JON FRANCO

As for Spanish, there is no independent evidence that supports the existence

of the abstract category X to which the clitic adjoins in (17). In any event, let us
take the correlation of a number of properties put forward in Uriagereka (1995)
with the elements landing in the XP projection at face value. Unfortunately,
under the proposal illustrated in (17), we would still fail to capture one property
of Spanish clitics, that is, they must exhibit strict adjacency to the clause’s major
inflected element in all the contexts, even when the verb moves. This suggests
that the clitic and the verb (or the auxiliary) form a constituent unit. In other
words, the real problem that a structure such as the one in (17) has to confront
is that nothing would prevent the occurrence of an adjoined intervening element
between the clitic and the verb (interpolation), a construction that, even though
it might be grammatical in Galician or other Romance languages, is ill-formed
in Spanish. Moreover, the motivation for a potential movement of the verb or the
auxiliary to X to fulfil the constituency unit requirement is also problematic. In
contrast, the agreement analysis of clitics defended in this paper warrants the
clitic-verb adjacency as a result of the standard head to head operation performed
by the verb to pick up all its inflectional morphology.
At any rate, there are two insights which are present in Torrego’s or
Uriagereka’s analyses as well as in Franco (1991) and are shared in this investi-
gation. First, Spanish clitics are functional heads with their own X-bar projection.
Second, languages with verbal object agreement and languages with object clitic
doubling on the verb should be treated similarly. Still, even though our insights
are the same, there are several ways in which my analysis departs from Torre-
go’s and Uriagereka’s pertaining to the nature of the functional category of
clitics as well as to the place they occupy in the phrase marker (cf. Section 4).

2.2.4 Sportiche (1996)

Sportiche’s analysis is without doubt the closest to the one I pursue in this
investigation, in the sense that Romance and, therefore, Spanish clitics head their
own functional projections, and that clitics can be analysed similarly to agree-
ment morphology, though not the same. However, our analyses also differ from
each other in some crucial points. Specifically, in Sportiche’s work, clitics head
Clitic Voice projections, and the asymmetries between accusative clitic construc-
tions and dative ones are captured in terms of an A/A′ distinction of the specif-
iers of the Voice phrases. Thus, the specifier of the Dative Voice is an A-
position whereas the specifier of the Accusative Voice is an A′-position.
My reservations about this type of analysis lie in the lack of compelling
arguments for this particular reduplication of functional categories, that is, Agr
Phrase as well as Clitic Voice Phrase, not to mention the problem of explanatory
 AGREEMENT AS A CONTINUUM 161

adequacy entailed by positing a new set of functional categories just to account

for Romance clitic phenomena.7 Furthermore, the motivation for postulating that
the specifier of the accusative clitic, unlike that of the dative clitic, is an A′
position is unclear as far as Spanish is concerned since his tests are either based
on French data or geared to the claim that the overt clitic doubled element moves
covertly. Bearing in mind that accusative clitics only double or stand for
presuppositional DPs, Sportiche’s claim that French clitic doubled elements move
covertly is clearly at odds with Diesing’s (1992) Mapping Hypothesis by which
presuppositional elements get scoped out of the VP via overt movement for their
logico-semantic representation. In the case of accusative clitic doubling, contra
Sportiche (1996), it is the strong agreement clitic that forces the doubled DP to
exit the VP overtly and land in its specifier, SpecAgrO, for the proper checking
of features. Furthermore, along the lines of Franco and Mejías (1997), it is this
movement that derives the presuppositionality condition on direct object clitic
doubling (see Section 4). In other words, the agreement analysis of direct object
clitic doubling put forward in this investigation leads us to the crucial conclusion
that Spanish object-verb agreement is not only a simple relationship of feature
matching, but a strategy to scope objects out of the VP in order to guarantee an
unambiguous presuppositional reading of the direct object.

3. Spanish Object Clitics as Verbal Agreement Morphemes

3.1 Introduction. Agreement as a continuum

The main challenge that clitics present for a theory of grammar is that they
exhibit mixed syntactic and morphological properties that make it difficult for a
linguist to classify them either as phonologically bound words or inflectional
agreement affixes. A further degree of difficulty for this classification task is
posed by pronominal clitics, whose feature resemblance with inflectional
agreement affixes makes them at first sight indistinguishable from one another.
For the sake of categorisation, let us assume that clitics, in general, are elements
intermediate between bound words and inflectional affixes or, more accurately
put for the pronominal clitics under consideration, between pronouns and
inflectional affixes. Now, since not all pronominal clitics in the languages of the

7. Dominique Sportiche himself acknowledges the issue of explanatory adequacy by raising the
following questions: Which functional categories occur overtly and how does the learner know?
162 JON FRANCO

world have the same behaviour or distribution, we can place them along a
spectrum such as the one given in (18), so that their position on this spectrum
will depend on how much the properties of each type of clitic resemble those of
a pronoun or those of a verbal inflectional agreement affix. This is represented
in (18):
(18) Infl. Affixes [–Z–Y–X–W–] Pronouns
pronominal clitics
In this section, I will focus my analysis on the syntactic and morphological
behaviour of object clitics in contemporary Spanish and will show that they
have gone beyond the realm of unstressed pronominal affixes to approach the
distributional patterns of desinential object-verb agreement morphemes. The two
basic approaches that I am going to take in order to verify this claim are the
following: (a) to compare the properties of Spanish object clitics to those of
agreement morphemes; (b) to contrast the behaviour of Spanish object clitics to
that of other clitics in Romance and non-Romance languages. Specifically for the
first step, we will look for similarities between Spanish object clitics and verbal
inflections in languages such as Basque, which have attested verb–object
agreement systems.8 For the second step, since not all pronominal clitics fall
uniformly under one single occurrence pattern, we will point out a number of
features of the Spanish object clitic system that are absent in clitic systems of
other languages and are responsible for the categorisation of the former as
agreement morphemes.

3.2 Object clitics: Pronominal affixes or verbal inflectional morphemes?

Romance pronominal clitics have moved in the direction of inflectional affixes

or, even further, they are in the process of acquiring the status of object verbal
inflection, as claimed in Saltarelli (1987). In what follows, I will show that

8. In favour of the Agreement Hypothesis analysis pursued in the present investigation, I must point
out that since clitic doubling is non-obligatory with nominals in Spanish, the strength of the
alternative analysis which considers clitics as verb arguments is slightly undermined due to the fact
that optional occurrence is not precisely a property of subcategorised arguments (I owe this
observation to Bernard Comrie). Actually. under this view, Spanish turns out to be a better candidate
to be analysed within the Agreement Hypothesis than Basque, which enforces obligatory object
agreement:
(i) Zuk Sandra ikusi *(d-u-zu)
You- Sandra- seen 3-have-you-
‘You have seen Sandra.’
 AGREEMENT AS A CONTINUUM 163

Spanish object clitics in comparison with some modern main stream Romance
languages are ahead in this process. For this purpose, I am going to single out
Spanish clitics in three varieties of Spanish from other clitics in other languages,
especially Romance languages, on the basis of a number of properties.
The phonological component of the grammar draws a firm line between
clitics, which normally have no independent stress, and full words, which are
always stressed; however, it does not shed much light on the categorisation of
pronominal clitics as unstressed pronouns or as inflectional agreement affixes.
Even though Spanish clitics and agreement morphemes alike do not bear
independent stress, this does not force us to group them together since stress
does not constitute a distinctive feature to differentiate ordinary clitics from
agreement morphology. It is a well-known fact that clitics in general are
unstressed; moreover, some languages, German for instance, have weak pronomi-
nal paradigms, which precisely owe their name to the fact that they are un-
stressed.9 Thus, in order to locate Spanish clitics along the spectrum in (18), we
are going to focus on the morphology, i.e. their distribution with respect to the
elements and features they occur with, and the syntactic operations into which
they may enter.

3.2.1 The categorial type of the clitic host, clitic ordering, adjacency conditions
and paradigm variation
Along the lines of Zwicky (1985), morphological rules specify the class of words
an inflectional affix can be attached to. Contrastively, the rules that account for
the distribution of words are not specified for word classes but for the XP
constituents and the structure in which these words can occur. Consequently, we
can assume that while words barely have adjacency constraints with respect to
the type of lexical items that can appear next to them, inflectional affixes are
very restricted in this regard.
In this way, let us take a look at clitics from some languages of the world
using this descriptive generalisation as a sounding board. Pronominal clitics in
Arabic can cliticise to verbs, nouns and prepositions (cf. Aoun 1993). Yagua
object clitics, according to Everett (1989), can be affixed to any constituent as
long as the clitic is minimally c-commanding its double. In Czech (Comrie
1989), the object clitic has to be positioned after the first constituent of the
clause — probably as a reflection of Wackernagel’s Law — regardless of its

9. For a thorough analysis of the distribution of weak pronouns in German, see Cardinaletti and
Starke (1999).
164 JON FRANCO

lexical nature. Almost the same constraint as in Czech holds for Old Spanish,
Old French, and Old Italian, a constraint that is known in the literature on
Romance languages as the Tobler-Mussafia Law (see Mussafia 1983 and
Wackernagel 1892). From a morpho-syntactic point of view, this freedom in
affixation places the clitics in Arabic, Czech, Old Spanish, etc. closer to a
wordlike status than to an inflectional status (see Rivero 1986 for this view of
Old Spanish clitics).
The pronominal clitics in contemporary French, Italian, and Spanish
contrast, however, with their medieval counterparts in terms of the type of host
they can cliticise to since they are restricted to preverbal or postverbal positions
depending on the verb’s feature [a finite] and have lost any effects of the
Tobler-Mussafia Law. One could conjecture that this strict morphological
dependency on the verb found in today’s Romance clitics with the exception of
those in European Portuguese (cf. Uriagereka 1992a) has reinforced the concep-
tion of the Agreement Hypothesis.
Also, on a closer look at the morphology of cliticisation of today’s Romance
clitics, we can see that they are subject to ordering constraints in relation to other
inflectional affixes. For instance, Modern Spanish clitics can no longer be
attached to the subject inflection of a finite verb, as opposed to Old Spanish
clitics. Example (19) is taken from Lapesa (1986: 58) (Apud. Cantar de Mio Cid):
(19) Acogensele omnes de todas partes (OK Old Spanish)
join .-.3 men from everywhere (*Modern Spanish)
‘Men join him from everywhere.’
(20) Se le acogen hombres de todas partes
- -.3 join men from everywhere
‘Men join him from everywhere.’ (Modern Spanish)
Furthermore, Spanish clitics not only appear in a fixed order with respect to their
host in tensed clauses, but also with respect to one another. Thus, the unalterable
order of Spanish argumental object clitics is: .-.-verb. Italian object
clitics also observe this order, whereas French inverts the order Dative-Accusa-
tive when both clitics are 3rd person:
(21) Juan se lo mandó (Spanish)
Juan - - sent
‘Juan sent it to him.’
vs.
(22) *Juan lo se mandó
Juan - - sent
 AGREEMENT AS A CONTINUUM 165

(23) Je le lui donnerai (Modern French)

I - - give-will
‘I will give it to her/him.’
Rigidity in order is a feature typical of inflectional affixation but rare for words,
which enjoy more freedom as regards the order in which they can appear. I will
propose that the strict clitic order Dative-Accusative-Verb can be obtained from
the syntax of verb movement and agreement feature checking in modern Spanish
since it mirrors the order of the arguments they represent and agree with. This
fact is paralleled by Basque data in which the agreement morphology on the verb
follows the order Absolutive–Dative–Ergative, i.e. the opposite of the order of
the verb arguments, which is Ergative–Dative–Absolutive. This parallelism
between these two languages is not due to coincidence but to principles that
ensure that the syntactic projection of agreement heads in configurational
languages form non-crossing referentiality chains with their antecedents.
Finally, contemporary Romance object clitics, with the exception of those
in Portuguese, require strict adjacency to their host, the latter being the main
verb or the auxiliary verb. There is no independent lexical material, not even
negative or emphatic particles, that can intervene between the clitic and the host
carrying all the other inflectional morphology, as we see in (24) and (25):
(24) *Juan lo ya vio (Spanish)
Juan - already saw
‘Juan already saw it.’
vs.
(25) Juan ya lo vio
Juan already - saw
However, unlike in Spanish, one can interpolate impersonal clitics and preposi-
tional clitics between the object clitic and the verb in Italian and in French,
respectively. Yet, I will not count this fact as a violation of the strict adjacency
condition since, first, the intervening elements are of the same nature and,
second, these languages have much more complex clitic systems than Span-
ish,10 and the morphological component of the grammar could possibly apply

10. Carstairs (1981) contrasts the properties of clitics with those of inflectional elements and
classifies the latter as ‘members of a relatively small closed system’. In this respect, Spanish clitics
qualify better for inflectional element status than their Romance counterparts since the Spanish clitic
inventory has been reduced to seven initial forms. The other Romance languages, on the other hand,
exhibit relatively large sets of clitics.
166 JON FRANCO

some reordering to these strings of clitics, as claimed in Bonet (1991).

In any case, the outcome of this discussion is that verbs and clitics form a
syntactic constituent unit. This can also be seen in instances in which the verb
undergoes syntactic operations, such as verb-subject inversion in questions: the
clitic and the verb remain together as a unit, which is how one would expect
agreement morphology to behave:
(26) a. Alfredo la trajo del Perú (Spanish)
Alfredo - brought from Perú
‘Alfredo brought her/it from Perú.’
b. ¿La trajo Alfredo del Perú de verdad?
¿- brought Alfredo from Peru truly
‘Did Alfredo really bring it from Peru?’
The restrictions on cliticisation we have seen above for Catalan, French, Italian,
Romanian and Spanish object clitics are closer to the morphological conditions
of the verb–object agreement morphology of Basque, for instance, than to those
of other clitic systems, such as the clitic systems of their corresponding medieval
counterparts.
The emergence of paradigm variation is also one of the characteristics of
inflectional affixation. Interestingly, contemporary Spanish, in addition to the
etymological clitic system, has three groupings with object clitics, namely, the
phenomena of leísmo, loísmo and laísmo. Leísmo is the substitution of an
etymologically accusative clitic by a dative clitic. That is, Spanish leísmo refers
to the cliticisation of an accusative object by the form le instead of lo or la, and
by les instead of los or las. In contrast, the loísmo and laísmo consist in substi-
tuting the etymological dative clitic le(s) by the accusative clitic forms lo(s) and
la(s). Landa (1995: 154) claims that this phenomenon is of a morphological
nature and by no means can the intrusive dative clitic forms alter the properties
of direct objects nor can the intrusive accusative clitic forms alter the syntactic
properties of indirect objects. These phenomena constitute a departure from the
etymological clitic system in the sense that the speakers perceive the need to
express on the clitic one of the features of its referent to the detriment of another
(see Landa 1995). Thus, most leísta speakers emphasise the animacy of the
referent of the accusative clitic,11 whereas loísta and laísta speakers emphasise

11. To be truthful to all the Spanish data, this description covers a good number of dialects, but is
not exhaustive. There are, however, several degrees of leísmo. In this way, two leísta dialects may
differ from one another in the extent to which they carry this phenomenon. For instance, one dialect
(or idiolect) may carry it across the board in such a way that the accusative clitic forms lo(s)/la(s) are
 AGREEMENT AS A CONTINUUM 167

the gender of the referent of the dative clitic. To the best of my knowledge, no
other European Romance language displays phenomena similar to these in their
clitic system. Lyons (1990) reports some clitic occurrences of the laísmo type in
Macedonian, a language very similar to Southern Cone Spanish as far as the
phenomenon of clitic doubling is concerned.
Most likely, the phenomena of loísmo and laísmo are linked to an attempt
to introduce the distinction of the feature gender in the dative clitic paradigm in
Spanish. On the other hand, there are several instances in which feature distinc-
tions are lost in Spanish. One is in the combination of dative and accusative third
persons in which the dative form le/les becomes the opaque form se, thus
eliminating the number distinction. Other examples of this process are found in
different dialects of Spanish such as colloquial Chilean or Caribbean Spanish in
which practically only the form le has survived for the third person dative clitic.
Regardless of whether this process takes place in the syntax or in the morpholo-
gy (as argued in Bonet 1991), feature erosion in pronominal affixes is a
characteristic typical of agreement systems. Contrastively, Italian has incorporat-
ed the gender distinction in the dative clitic paradigm. A conclusion that can be
drawn is that while leísmo is an agreement driven phenomenon — i.e. it is
geared to an animacy distinction that seems to govern agreement systems (cf.
Section 3.3.1.) — loísmo and laísmo are regressive moves of the agreement
system towards the pronominal one.

3.3 Clitic doubling and agreement with verbal complements

The term doubling stands for the duplication of a complement or an adjunct by

a clitic. This concept can overlap with that of verbal agreement, which is a cross-
reference marking (in f-features) of the verbal arguments on one of the clause-
inflection-bearing elements. The problem is that when clitic doubling is limited
to the arguments of the verb in A-positions, the distinction between agreement
and doubling turns out to be very fine, sometimes non-existent. Spanish and
Macedonian seem to be some of the languages that pose this problem, as
mentioned above.

thoroughly substituted by le. Another dialect may only allow leísmo to replace the masculine
accusative clitic. These two examples illustrate the two maximum and minimum realisations of the
phenomenon. Most leísta dialects I am familiar with or I have seen in the written language range
between these two poles and fall under the above description. However, let us not forget that there
are inconsistencies in the manifestation of this phenomenon, mostly due to the normative nature of
language teaching in schools.
168 JON FRANCO

Clitic doubling with verb complements in argument positions (cf. Jaeggli

1986 and Suñer 1988) is an outstanding characteristic of Romanian and Spanish
that distinguishes them from other Romance languages. Spanish, moreover,
unlike Romanian, does not have any independent clitic system other than the
reflexive and the object clitic paradigms, which makes the case for the existence
of an actual cliticisation process in Spanish not as clear as it is in other Romance
languages. Subject clitic doubling, on the other hand, is more common in
Romance, especially in French and Northern Italian dialects. However, it is not
obvious that Spanish clitic doubling and subject clitic doubling are two instances
of the same phenomenon. To begin with, the A/A′ status of the position of the
doubled subject is rather controversial. Moreover, an analysis of subject clitics as
agreement morphology, like the one I propose for Spanish object doubling, has
to face two pieces of counterevidence which are noted in Everett (1996: 124): the
additional independent presence of subject agreement morphology on the verb,
and the preverbal occurrence of subject clitics in declaratives versus the post-
verbal occurrence of subject clitics in interrogatives.
In the next subsection, I will point out some of the properties by which
established agreement systems and clitic doubling12 differ. The problem for
this kind of approach is that one may feel tempted to compare object clitic
doubling with subject agreement since the latter is the most ordinary form of
agreement. Notwithstanding the fact that there should be a point of contact
between the two, the relevant comparison for the issue at stake would seem to be
object clitic doubling versus object agreement. Interestingly, object agreement is
not as unrestricted nor as common as subject agreement.

3.3.1 Two issues in object agreement: The animacy hierarchy and the preposition a
Sportiche (1996) claims that agreement viewed as a Spec–Head(Agr) relation
must be semantically free. Hence, he concludes that accusative clitic doubling is
not an agreement relation since the clitic imposes certain semantic specifications
on the doubled DP. However, even though subject agreement in general or object
participial agreement in French do not seem to have any semantic restrictions, it
is not out of the question that agreement relations can be founded on the
semantics of the agreeing NP. What is hidden here is that, in many instances, the

12. From now on, I will refer to clitic doubling as the reduplication of an XP in an A-position by
a coreferential clitic attached to the verb. If the element doubled by the clitic is in a non-argumental
position, I will refer to it as left dislocation or right dislocation. This terminology basically follows
the nomenclature established in the literature on this topic.
 AGREEMENT AS A CONTINUUM 169

semantic conditions for an element to agree with the verb are derived from the
structural position of the element (cf. Section 4 for further discussion).
On the other hand, Comrie (1989), among many others, shows that many
agreement relations are driven by an Animacy Hierarchy which has an overall
cross-linguistic validity. Thus, arguments that occupy a high position in this
Animacy Hierarchy hold stronger or more uniform agreement relations than those
occupying a lower position. In broad terms, first, second and third person
pronouns in this order occupy the highest positions in this hierarchy followed by
definite human nouns, definite nouns etc, whereas inanimate generic nouns
occupy the lowest ones.13 Now, if Spanish clitics are agreement morphology,
and agreement is based on this Animacy Hierarchy we can correctly predict that
clitics should more easily double NPs ranking high in the Hierarchy.14
In Swahili, object agreement does not occur with lower elements of the
Animacy Hierarchy (cf. Wald 1979), i.e. with generic nouns or indefinite
inanimate nouns. Still, definite–accusative agreement and, as a small extension,
indefinite-human accusative object agreement take place obligatorily. Macedonian
exhibits a similar behaviour with respect to the direct object agreement pattern,
with the exception that in this language it is referred to as clitic doubling. Now,
Southern Cone Spanish clitic doubling (cf. note 5) is not far from the object
agreement patterns of Macedonian and Swahili. As a matter of fact, the distribu-
tion of Macedonian clitic doubling and that of this dialect of Spanish are almost
identical, that is, the lowest elements in the hierarchy that can be clitic doubled
are indefinite referential humans.
There is, however, a difference between Basque, Macedonian and some
varieties of Swahili on the one hand, and Spanish on the other. Whilst in the
former object agreement or clitic doubling is obligatory whenever possible, in
Spanish, object agreement is optional with most nominals. Still, even the obligat-
oriness of object agreement seems to be mapped onto the Animacy Hierarchy,
since clitic doubling is obligatory with pronouns as well as with animate indirect
objects in inverted predicates and other contexts in all dialects of Spanish.

13. Animacy can only be understood here as a cover term that subsumes the notions of saliency and
definiteness/referentiality. Any literal reading of the label animacy would be inaccurate for this
description. However, the term becomes handy since animate nouns and pronouns are usually
associated with a higher degree of saliency or referentiality than inanimate ones.
14. As one reviewer points out, in Colloquial French, ‘a strong pronoun in an argument position
always has to be doubled by a clitic.’ To my advantage, these facts fit perfectly in my Animacy
Hierarchy account of clitic doubling as agreement since pronouns are the highest ranking elements
in this scale. If agreement/doubling is going to start anywhere, it will be with pronouns.
170 JON FRANCO

The fact that in many languages clitic doubling cannot be realised with
prepositionless nominals has led many researchers, such as Lyons (1990) in
recent times, to believe that the NP which is coreferential with the clitic is not an
argument but some kind of PP adjoined to VP, whereas the clitic itself licenses
a pro in argument position. In order to validate this hypothesis for Spanish, one
should at least be able to show that, first, clitic doubling is impossible with
prepositionless NP objects and, second, that the presence of the preposition is
directly related to the phenomenon of clitic doubling. However, the adjunction
hypothesis cannot pass either of these tests.
As mentioned in Section 2, clitic doubling with prepositionless NP direct
objects is grammatical in Southern Cone Spanish. In addition, the occurrence of
the preposition a with clitic doubled NPs has nothing to do with the phenomenon
under scrutiny, as shown by the independent distribution of this preposition with
non-clitic doubled objects, for example:
(27) Vimos a María, a los niños, a los payasos, etc.
(we) saw to Mary to the kids to the clowns
‘We saw Mary, the kids, the clowns, etc.’
(28) Prohibieron a los niños jugar al balón
(they) forbade to the children play ball
‘They forbade the children to play ball.’
Conversely, clitic doubling constructions in Berber always demand the presence
of a supplementary preposition, i.e. in addition to the one in the clitic doubled
constituent. Without doubt the occurrence of this extra preposition is linked to
clitic doubling. To illustrate the point, notice the following contrast in the Berber
examples in (29) and (30) in which only the clitic doubled phrase repeats the
preposition twice:
(29) In-si n Munati
of-3 of Munat
‘Munat’s’
vs.
(30) In Munat
of Munat
Munat’s
Thus, an analysis along the lines of Lyons (1990) or Aoun (1993) for Lebanese
Arabic is more likely to hold for different dialects of Arabic than for Spanish,
which does not have prepositional occurrences conditioned by clitic doubling.
 AGREEMENT AS A CONTINUUM 171

3.4 Agreement features. Summary

If one considers agreement as an absolute, full-fledged, unrestricted verb-

argument relationship across the board, that is, as an either/or issue, there will
probably not be many languages that can be described as having an agreement
system, let alone an object-verb agreement system. On the other hand, one could
relativise the concept of agreement, going back to the view of agreement as an
increasing continuum, and propose that there are different degrees of agreement
that range from high to low.15 In the same way that there is a transitivity
continuum, as proposed in Hopper and Thompson (1980), i.e. the transitivity
degree of a verb is determined in relation to the number of properties out of a set
of transitivity characteristics which the verb exhibits, I claim that agreement is
also the fulfilment of a number of properties in the matching of phi features
between two elements within a domain. Thus, the parametric account of agree-
ment can be done with respect to how many agreement properties the two
elements involved in the putative agreement relationship exhibit.
Now, the issue at stake at this point is to determine when the ‘shade’ of
agreement is strong enough, that is, whether the number of properties is suffi-
cient to make itself visible in the syntactic component.
Bearing this in mind, let us summarise how differently object agreement or
clitic doubling systems, for that matter, pattern across the properties of agreement
elements we have seen above. In other words, we are going to determine how
strong the degree of agreement is in a number of languages by their compliance
with the properties listed in (31):
(31) Object Agreement Bearing Elements
Bq SC BqS AS F I P R Mac
A. Strict adjacency to [Aux/V] 1 1 1 1 1 1 0 0 1
B. Syntactic unit with host 1 1 1 1 1 1 0 1 1
C. Same specific host [Aux/V] 1 1 1 1 1 1 0 1 1
D. Fixed order 1 1 1 1 0 1 1 1 1
E. Feature erosion in the forms 1 1 1 1 0 1 0 1 1
F. Different paradigm selection 1 1 1 1 0 0 0 0 1
G. Co-occurrence with Acc arguments 1 1 1 0 0 0 0 1 1
H. Unrestricted co-occurrence 1 0 0 0 0 0 0 0 0
I. Obligatoriness of co-occurrence 1 0 0 0 0 0 0 0 1
J. Co-occurrence with prepositionless NP 1 1 0 0 0 0 0 0 1
TOTAL 100 8 7 6 3 5 1 6 9

15. Although conceived much earlier and on independent grounds, my proposal resembles Everett’s
(1996) analysis of clitics, agreement affixes and pronouns as instances of allomorphy of phi-features.
172 JON FRANCO

The languages contrasted in the table in (31) are from left to right: Basque (Bq),
Southern Cone Spanish (SC), Basque Spanish (BqS), other dialects of Spanish
(AS), French (F), Italian (I), European Portuguese (P), Romanian (R) and
Macedonian (Mac). A number of clarifications concerning the characteristics A.
to J. ought to be made. A. strict adjacency refers to the absence of interpolation
phenomena, which amounts to the occurrence of intervening elements between
the clitic and the host which is either the verb or the auxiliary. In Romanian, the
adverbial clitic mai ‘more’ occurs between object clitics and the verb (Dobrovie-
Sorin 1994). B. syntactic unit with host refers to whether or not the clitic is left
stranded after the host moves in the syntax. C. same specific host [Aux/V] refers
to the categorial type of the syntactic host, not the phonological host (we are not
at this point interested in the latter). In this regard, I subscribe here to Dobrovie-
Sorin’s (1994) analysis which claims that in addition to syntactic cliticisation,
Romanian clitics undergo prosodic cliticisation to a word that has nothing to do
with the selection of the clitic, as in (32), taken from Dobrovie-Sorin (1994):
(32) Maria-mi scrie des
Maria-1- writes frequently
‘Maria writes to me frequently.’
C. same specific host aims at isolating cases of cliticisation to hosts other than
the verb, such as cliticisation to a preposition in Galician and in European
Portuguese, as illustrated in (33) (example taken from Uriagereka 1992a):
(33) Pra lle (ti) (enton) falar(es),… (Galician)
to 3- (you (then talk-2
‘In order for you to talk to him,…’
D. Fixed order refers to whether there is an unaltered order of clitics or agree-
ment elements among themselves that applies everywhere. E. feature erosion in
the forms refers to the loss of feature distinctions, whereas F. refers to the co-
existence of different agreement paradigms in one language, for example, the
leísta paradigm versus the laísta paradigm and so on. G. co-occurrence with Acc
arguments refers to the co-occurrence of agreement elements with a verb internal
accusative argument in an A-position, so that we can differentiate dialects that
only have clitic doubling with dative nominals from those that have it with
datives and accusatives. Left-dislocations fall outside the domain of this property.
H. unrestricted co-occurrence refers to an across-the-board co-occurrence of the
agreement element and the argument, regardless of the semantic composition of
the latter. I. obligatoriness of co-occurrence refers to optional versus obligatory
co-occurrence of a clitic and an object nominal every time the conditions for
 AGREEMENT AS A CONTINUUM 173

agreement are met. In this case, we have left object pronouns out in order to
assign whole values, otherwise Spanish would obtain 1/2 for this row. Finally, J.
co-occurrence with a prepositionless NP is self-explanatory since it indicates
whether a pseudo-preposition must precede the argument for agreement with the
verb to take place. If we take Basque as a model of object agreement, the
languages that score closest to Basque in the table in (31) will rank higher in the
agreement continuum.

3.5 Object agreement in Spanish

3.5.1 Object agreement with strong pronouns and obligatory agreement

If we assume that ‘reanalysis’ occurs in languages, it would not be very illogical
to conjecture that Spanish banned subject clitics, eliminated the prepositional
clitics and reduced its argument clitic system to the object paradigm, so that the
function of argument clitics would be redefined for verb object agreement.
Another phenomenon that can be observed at first sight in the evolution from
Old Spanish to Modern Spanish is the birth of the obligatoriness of clitic
doubling with pronouns in Spanish. These two characteristics of contemporary
Spanish are absent in Old Spanish as well as in the other Romance languages (at
least in the written varieties), which have kept their prepositional clitics as well
as the option of non-doubling with non-dislocated pronouns.
Some literature on Spanish object clitics (such as Gerfen 1991, whose main
source of data is Andalusian, a dialect of Spanish whose clitic doubling is limited
to strong object pronouns and to nominal indirect objects) considers clitic
doubling with object pronouns merely as a device to place a mark of referential
contrast or contrastive focus on the object. This claim is reinforced by the fact
that the clitic-doubled pronouns, unlike nominals, often bear phonological
prominence in the sentence. This hypothesis is reminiscent of that of Lyons
(1990) under which the doubled pronoun in (34) below is not a syntactic
argument, but an adjunct to VP, in a structure such as (35):
(34) Te estoy hablando a tí
-2 am talking to you
‘I am talking to YOU’
(35) [ [VP i V proi][Pronoun]]
However, one of the problems that one encounters when dealing with strong
pronouns in Spanish has to do with the fact that Spanish has object and subject
pro (cf. Jaeggli 1986), hence we should expect that overt strong pronouns are not
used in completely free alternation with the less marked null pronominal forms.
174 JON FRANCO

This is a typical feature of subject and object pro-drop languages such as

Basque (cf. Ortiz de Urbina 1989). Now, there is a difference in descriptive
accuracy between claiming that strong argument pronouns often entail a
contrastive/emphatic interpretation and claiming that these elements exclusively
occur with this interpretation and have an adjunct status. Thus, in addition to
the normal native speakers’ intuitions against the unique reading of strong
pronouns as contrastive focus in isolated contexts,16 Franco (1993) and Suñer
(1999) put forward some pieces of evidence in which the clitic-doubled strong
pronoun receives a neutral interpretation. For instance, Franco (1993) points out
that the possibility that strong pronouns are exclusively contrastive focus
elements in Spanish is highly jeopardised in clauses that contain more than one
strong pronoun:
(36) Yo tek lai presenté (a) ellai a tik
I -2 -3f introduced her to you
‘I introduced her to you.’
Normally, elements bearing contrastive focus are correlated with one position in
the clause; hence, a sentence may have at most one focused element. Thus, the
sentence in (36) challenges Gerfen’s claim since not all three strong pronouns
can be used simultaneously for contrastive focus or referential contrast purposes.
Moreover, Suñer (1999) provides several contexts in which the clitic doubled
strong object pronoun receives the same intonation and stress as a lexical DP.
According to Suñer, coordination would be one of the clear environments in which
these pronouns are not contrastive, as illustrated in (37) ((15b) in Suñer 1999):
(37) Lasi invitamos a ellai y su hija para el próximo
. invited to her and her daughter for the next
martes
Tuesday
‘We invited her and her daughter for next Tuesday.’
In (37) the sentential stress falls on ‘next Tuesday’ and, the strong object
pronoun a ella remains neutral in the phonology.
In sum, the data show that overt strong object pronouns can be emphatic or

16. In this respect, Carmen Silva-Corvalán pointed out to me that in Chilean Spanish, the strong
object pronoun a él in a sentence like (i) is usually neither emphatic nor contrastive:
(i) Loi he visto a éli
.-3 have-1 seen him
‘I have seen him.’
 AGREEMENT AS A CONTINUUM 175

non-emphatic. In view of this data, the hypothesis I will pursue here is that
strong pronouns in Spanish are true arguments and, in the light of the Agreement
Hypothesis, their obligation to co-occur with a clitic is derived from the correla-
tion between agreement and the Animacy Hierarchy. In this way, in regard to
object agreement in Spanish, the initial hypothesis is that dialects that behave
conservatively with respect to clitic doubling such as Manchego or Andalusian
bear obligatory object agreement with the highest elements in the Animacy
Hierarchy, namely, pronouns and referentially definite humans, which are what
indirect objects of psych verbs usually stand for.

3.5.2 A difference in the Spanish three-way agreement system

One of the phenomena that the Agreement Hypothesis has to face when account-
ing for clitic doubling is to explain why direct object agreement is not always
possible, like indirect object agreement or subject agreement are. To put it
differently, it seems that only specific direct objects can be doubled, whereas
indirect objects and subjects are insensitive to any specificity constraints. This
semantic restriction on direct object clitic doubling has led Sportiche (1996) to
exclude Spanish direct object clitics from the object agreement paradigm.
However, there is nothing in principle that prohibits agreement systems from
having a semantic flavour. In fact, the presence of agreement morphology
correlates with the occurrence of definite or presuppositional arguments in
several languages: e.g. in Swahili (cf. Wald 1979) and Turkish (Diesing 1992),
to mention some. In any event, it will be shown in the next section that direct
object clitic doubling is not so much tied to specificity as it is to general
principles of syntactic mapping of presuppositional arguments.

3.5.3 Direct object agreement, specificity and the matching principle

Silva-Corvalán (1984) and Suñer (1988) put up some good arguments and
empirical evidence in favour of the feature [+specific] as the main trigger for
clitic doubling in Southern Cone dialects. For instance, examples (38) and (39)
below could be explained by saying that clitic doubling is possible when the
doubled element is [+specific]:
(38) Juan lai sacó la notai sin esfuerzo (Southern Cone)
Juan it-- got-3 the grade without effort
‘Juan got the grade without effort.’
(39) *Juan lai sacó una notai sin esfuerzo
Juan it-- got-3 a grade without effort
‘Juan got a grade without effort.’
176 JON FRANCO

Even though specificity seems to play a role in the clitic doubling phenomenon,
I believe it is the result of a side effect, i.e. all elements occupying a low
position in the Animacy Hierarchy are likely to be [−specific], and hence, cannot
be clitic doubled. Crucially, there is evidence that may upset the argument for an
exclusive correspondence between the feature [+specific] and clitic doubling.17
Consider:
(40) Juan loi invitaba a unoi y luego se olvidaba
Juan -him used to invite to one and then -3 forgot
‘Juan used to invite people and then forget all about it.’
(41) En ese departamento, loi admiten a cualquierai
in that department -him admit-3 to anyonei
‘In that department, they admit anyone.’
Examples (40) and (41) are clear cases of direct object clitic doubling with non-
specific NPs. In (40), a uno ‘to one’ is the prototypical pronoun of arbitrary
reference and, as in any object-verb agreement with a pronoun in Spanish, a fully
specified overt clitic is required.18 Similarly, a cualquiera ‘to anyone’ in (41)
is an animate indefinite pronoun, often called free-choice, and clitic doubling is
normally allowed. Crucially, both [−specific] direct objects uno and cualquiera
rank high in the Animacy Hierarchy as [+human] entities so their agreement with
the verb comes as no surprise.
There is supplementary evidence against Suñer’s claim that the referent of
the clitic has to be [+specific] or that the direct object clitic itself is inherently
[+specific]. Existential sentences with haber ‘there is/are’ in Spanish take

17. In this study, I have been using the notion of specifity in the sense that an element is specific if
we can recover its reference from an x that belongs to a known set or class with a property z (see Enç
1991 and the works cited in this article for a more thorough account of the semantics of specifity).
18. For the hard-core believers in specificity as the trigger for clitic doubling, one could argue that,
normally, generic elements like uno show specificity effects, but that nonetheless the core semantic
notion of specificity loses its root meaning (see note 17). In fact, it has been shown in Franco and
Mejías (1997) that it is the presuppositionality of the NP that determines clitic-doubling in Spanish.
Here, I must explicitly state, lest this analysis be misinterpreted, that I am not arguing against what
has been labelled in syntax as specificity effects, which generics and perhaps some clitic doubled
elements are subject to, but against the ability of the feature [+specific] to license accusative clitic
doubling in Spanish.
At any rate, under my approach, uno triggers clitic doubling because of its pronominal nature, not
because, as a generic, it may behave like specific elements. Otherwise, we should be able to clitic
double all generics, which is not the case.
 AGREEMENT AS A CONTINUUM 177

[−specific] complements obligatorily, as shown in (42); still, one can refer to

these complements by accusative clitics, as in (43), for instance:
(42) *Había los hombres vs. Había unos hombres
‘There were the men.’ ‘There were some men.’
(43) A: Parece que hoy no había mejillonesi en la pescadería
‘It seems that today there were no musselsi in the fish market.’
B: Sí que losi había, pero estaban ya vendidos
yes -3 there were but they were already sold
‘Yes, there were indeed, but they were already sold.’
To conclude, specificity by itself does not succeed in accounting for clitic
doubling phenomena; moreover, the notion of specificity has no independent
formal status in the grammar (cf. Mahajan 1992), i.e. it does not constitute a
primitive notion of the theory of Principles and Parameters or the Minimalist
program for that matter. At this point I should acknowledge that Animacy also
has a non-formal status in this framework. Thus, I contend that the notions of
Animacy or presuppositionality, important as they are for agreement relations,
only become theoretically enlightening when encoded in particular syntactic
positions that are mapped onto structural configurations. With this in mind, the
following section is devoted to the mapping of object agreement onto a syntactic
structure, which will be the starting point in the elaboration of an account of the
data above.

3.6 The structural mapping of the agreement hypothesis analysis

By adopting Pollock’s (1989), Chomsky’s (1993 and 1995), among others,

decomposition of the IP node into several functional categories, the theory of
Principles and Parameters indirectly provides us with a structural possibility to
solve the problem of the initial mapping of Spanish object clitics in the X′
schema onto non-argument positions (cf. Section 2). In this way, assuming these
analyses and the hypothesis that the so-called object clitics are verbal agreement
morphemes in Spanish, I am going to implement an analysis in which object
clitics are agreement heads (AgrO) that project agreement phrases (AgrOP), as
illustrated in (44) below (cf. page 35 for discussion on the order of phrasal
indirect and direct objects in Spanish). Based on the analysis represented in
structure (44), I will attempt to derive the Spanish data discussed so far.
178 JON FRANCO

(44) AgrSP

Spec Agr

Agr TP

Spec T

T AgrIOP

Spec Agr

AgrIO AgrDOP

Spec Agr

AgrDO VP

Spec V

V NPDO aNPIO
With respect to the two AgrOP projections, there is evidence beyond Baker’s
(1988) Mirror Principle, of course, that points at the order given in (44) as being
the correct one for Spanish. Here, it is worth taking a look at the examples in
(45) and (46):
(45) a. Juan lei dio la nota a Pedroi
Juan .-3 gave the grade to Pedro
‘Juan gave the grade to Pedro.’
b. *Juan lai dio proi a Pedro
Juan .-3 gave to Pedro
‘Juan gave it to Pedro.’
c. Juan sei lak dio prok a Pedroi
Juan .-3 .-3 gave to Pedro
‘Juan gave it to Pedro.’
(46) a. *Pedro sei mek entregó prok a la policíai
Pedro .-3 .-1 handed to the police
‘Pedro handed me to the police.’
 AGREEMENT AS A CONTINUUM 179

b. Pedroi sei me entregó a la policía

Pedro . .. handed to the police
‘Pedro turned himself in to the police for me.’
c. Pedro Ø me entregó a la policía
Pedro DO1 handed to the police
‘Pedro handed me to the police.’
d. Peruk ni polizei bidali nau [−dative ]
Peru- I- police- send 1--3
(Basque version of (46c))
In (45), we can see that in a ditransitive structure the indirect object Agreement;
may or may not subcategorise for direct object Agreement, however, direct
object agreement cannot stand without indirect object agreement in these double
object constructions, as shown in (45b). Actually, the data in (46) seem to
indicate that IO Agreement can only semantically select third person direct object
Agreement,19 hence the ungrammaticality of (46a); otherwise, the combination
3 1 can only stand for a reflexive and an ethical dative respectively, as in (46b).
Thus, in order to handle the meaning of the sentence Pedro handed me to the police,
the agreement chain with the indirect object argument must disappear, as in (46c).
Interestingly, the same anti-agreement effect with datives occurs in Basque, as
illustrated in (46d). Since Basque is a language with obligatory verb agreement
with IOs we conclude that this phenomenon can be derived from a universal
selectional pattern in the projection of Agr Phrases which is formulated in (47):
(47) In the person paradigm an agreement head cannot outrank a c-com-
manding agreement head of the same morphological type. (Taken
from Franco 1993: 96)

19. Bonet (1991) describes this phenomenon, that is, the *me lui/I-II Constraint, as taking place in
the Morphological Component. This constraint seems to have a universal character since it is found
in the morphology of languages as different as Catalan, Georgian and Greek. The stand I will take
on this issue along with that of Laka (1993) is that the object clitic morphology in Spanish is
restricted enough to read off this constraint from the syntax. Thus, Laka (1993) argues that if the
ungrammaticality of sentences such as (62a) was derived from a strictly morphological well-
formedness condition one could not explain why a dative clitic which does not correspond to a
thematic argument, that is, an ethical dative, is able to escape this constraint, as illustrated in (i),
taken from Laka (1993 (10)):
(i) Tei me han vendido proi al enemigo
2 1– have-3 sold to the enemy
‘They have sold you to the enemy (on me).’
The fact that the constraint under study only affects clitics that corefer with verb arguments crucially
indicates that it has its source in the syntactic structure.
180 JON FRANCO

In sum, indirect object Agreement imposes selectional restrictions on direct

object Agreement, but not vice versa, hence we assume the hierarchical organi-
sation given in (44). Bearing this in mind, as a first approach (cf. Section 3.1. for
a refined version), I propose the following derivation for a simple sentence in
Spanish with three-way agreement:
(48) a. Juan mei lask enviará proi prok
Juan 1 3 send-will-2
‘Juan will send them to me.’
b. Initial stage in the derivation:
AgrSP

Spec Agr′

AgrS TP
a
Spec T′

T AgrIOP
r′
Spec Agr′

AgrIO AgrDOP
me
Spec Agr′

AgrDO VP
las
Spec V′
Juan
V proDO proIO
envi-
 AGREEMENT AS A CONTINUUM 181

c. Input to Phonetic Form:

AgrSP

Spec Agr′
Juanm
AgrS TP
mek lai enviaráj
Spec T′

T AgrIOP
tj
Spec Agr′

AgrIO AgrDOP
[[tj]i]k
Spec Agr′

AgrDO VP
[tj]i
Spec V′
em
V proDO proDO
tj

In the mapping from the configuration of elements in (48b) to that of (48c), the
verb incorporates into the functional heads à la Baker (1988), picking up all the
inflectional morphology (agreement clitics included).20
To finish this section, a note on the phenomenon of clitic climbing might be
necessary. At first sight, clitic climbing seems to be a counterexample to the in
situ generation of clitics within the agreement analysis. In fact, the clitic
climbing operation in (49) could be analysed as an instance of incorporation of
the clitic into the upper verb:

20. Alternatively, one could claim, in the spirit of the Minimalist Program that the verb comes fully
inflected from the lexicon and checks its inflectional features in the pertinent functional projections.
At this stage, any discussion on this issue will take us too far afield.
182 JON FRANCO

(49) Sei lak hicieron mandar ti tk por fax

. . made send by fax
‘They made (somebody) send it to him by fax.’
Thus, under this view, the two clitics in (49) would have incorporated into the
causative verb from the enclitic position on the lower verb by head movement.
However, appealing as the clitic incorporation analysis might be, clitic climbing
per se does not undermine the Agreement Hypothesis at all. First, there is
evidence from languages with full-fledged verbal agreement that agreement
markers can be displaced within the inflectional amalgam (cf. Laka 1993).
Second, once a certain degree of restructuring between the upper verb and the
lower verb has been obtained, clitic climbing is optional. This is really a problem
for the syntactic incorporation analysis since under Minimalist assumptions,
elements move overtly because they have to. That is, they move in order to
satisfy the licensing of some feature. Furthermore, clitic climbing does not have
any semantic import in terms of scope relations or sentence interpretation, i.e. it
does not feed LF. These facts, in addition to some morphological idiosyncrasies
of clitics, have led Franco and Landa (1995) to claim that clitic climbing takes
place within an inflected atomic verbal unit at a PF that has inherited much
structure from the syntax, in the spirit of Aoun, Hornstein, Lightfoot and
Weinberg (1987) and Laka (1993). Clearly, this approach to clitic climbing as
subatomic movement weakens the commonplace conception of the clitic climbing
phenomenon as counterevidence to the Agreement Hypothesis since here the
notion of climbing is reinterpreted as morpheme reordering within a complex
verbal head. In the case of (49), the causative verb hacer and the subordinate
verb mandar, together with their corresponding inflectional morphology, have
undergone ‘restructuring’ and constitute an atomic unit. Crucially, this unit turns
out to be the domain of clitic climbing, that is, the so-called clitic climbing
operation can only occur within this unit. In fact, if there is an intervening
Causee splitting the verbal complex, the actual climbing operation is banned, as
shown in (49′):
(49′) *Sei lak hicieron al bedel mandar ti tk por fax
. . made the janitor send by fax
‘They made the janitor send it to him by fax.’
It is also useful to note that the data in (49′) does not fare well with the incorpo-
ration analysis of clitic climbing. Notice that, if clitic climbing were head to head
movement, an intervening XP should not interfere with this movement. On the
other hand, a syntax-PF interface approach to clitic climbing that contemplates
 AGREEMENT AS A CONTINUUM 183

this phenomenon as a morphonological tendency for proclisis that is realised at

a PF level with much inherited syntactic structure would straightforwardly
account for (49) as well as for the impossibility of (49′).

4. The Syntax of Clitic Doubled Object Constructions in Spanish

It has been claimed in Laka (1993) — and in Sportiche (1996) too — that
agreement does not affect the semantics of the sentence. This insight is clearly
reflected in sentences such as (51a), in which the presence or absence of the
clitic does not make any interpretative difference:
(50) (Lei) he llamado a Pedroi
(3 have.1 called to Pedro
‘I have called Pedro.’
So far, we could safely conclude that definite objects are semantically insensitive
to clitic doubling. However, this does not help us understand why direct object
clitic doubling is not licit across the board. Fortunately, to our advantage, the
phenomenon of clitic doubling with human indefinite objects paves the way
towards our understanding of the syntax of clitic doubled direct objects. Thus, in
the sentence in (51) below only the presuppositional reading of the direct object
is allowed when the overt clitic co-occurs with the direct object:
(51) a. Lei he visto a un estudiantei.
3 have.1 seen to a student
I have seen a student.
The direct object a un estudiante in (51) can only be any student whose face I
have previously seen but cannot associate with a name; however, never anyone
who happens to look like a student. In the light of these facts, we can see that
counterexamples to the specificity constraint may be found because specificity is
only a subcase of presuppositionality, which seems to be the accurate require-
ment for the direct object to be doubled by a clitic. This clarification does not
leave us in a better theoretical position. However, it is at this point of the
discussion that we can resort to Diesing’s (1992) Mapping Hypothesis and her
analysis of presuppositional elements,21 and claim along the lines of Franco and

21. Roughly put, with respect to the interpretation of indefinite NPs, Diesing’s analysis makes two
specific claims:
I. An indefinite outside the VP (at LF) will be interpreted as generic or presuppositional.
184 JON FRANCO

Mejías (1997) that Spanish direct object clitic doubling — in addition to being
an instantiation of object agreement — is mainly a strategy to scope objects out
of the VP and allow them to escape the nuclear scope. In this way, the clitic
does not only guarantee that the object will not be interpreted existentially, but
also functions as a disambiguator between the existential and presuppositional
readings that non clitic-doubled indefinite objects would otherwise have. Having
said this, the syntactic implementation of Spanish clitic doubled direct object
nominals for a sentence such as (50) would be as follows:
(51) b. TP

Spec T′
pro
T/AgrS AgrDOP
[lej [he vistoi]]k
Spec AgrDO′
a un estudiantej
AgrDO VP
tk
Spec V′

ti tj

In the overt syntax, the direct object a un estudiante raises to Spec of AgrDO
while the verb moves to AgrDO, and subsequently to T. I assume that both
movements are motivated by feature checking. In fact, one could hypothesise that
there is a feature [presuppositional] that needs to be checked before SPELL OUT
in the same way that the EPP has to be checked. In other words, the EPP is the
strong-D feature of T whereas [presuppositional] is the strong D-feature of
AgrDO when activated.
Incidentally, Franco (1993) claims that the difference between direct object
clitic doubling and indirect object clitic doubling lies in the point of the derivaton
where the checking of features takes place. In the case of indirect object clitic
doubling, the object (or a subset of its features) would move covertly at LF to
Spec of AgrIO. Positing this asymmetry, in addition to giving us some advantag-
es to explain some restrictions on direct object clitic doubling, also accounts for

II. An indefinite within the VP (at LF) will be interpreted as non-generic or non-presuppositional.
 AGREEMENT AS A CONTINUUM 185

the unmarked complement order in Spanish, V DO IO. Thus, the fact that Spec
of AgrIO is projected higher than the Spec of AgrDO, for the reasons outlined
before in Section 3.6, does not pose a problem anymore for overt word order in
Spanish since the Case in AgrIO is only active at LF.
Furthermore, this solution makes it feasible for us to offer a principled
syntactic explanation of why clitic doubling with direct objects is not even half
as productive as indirect object clitic doubling is.22 In the light of the Minimal-
ist Program, the question is answered by an economy principle such as the
Procrastinate principle, under which covert LF operations are ‘cheaper’ than
overt operations. Thus, assuming that the grammar observes these economy
constraints, it is logical to adopt the claim that ‘the system tries to reach PF ‘as
fast as possible’, minimising overt syntax’ (Chomsky 1993).
These theoretical assumptions have a direct bearing on Spanish clitic
doubling, provided that we adopt the analysis suggested above. First, since indirect
object clitic doubling can reach PF without resorting to any particular operation
of NP movement in the overt syntax, the system turns it into a productive
construction. Conversely, we have proposed that direct object clitic doubling
requires the NP object to be raised overtly to Spec of AgrDO; thus, this con-
struction becomes a costly derivation. Therefore, the system is compelled to limit
the production of direct object clitic doubling to a small percentage or to those
cases in which clitic doubling is necessary for the convergence of the derivation.

22. For the sake of clarity, in this case, the term ‘productivity’ refers to the frequency of the
occurrence of these constructions. That is, we are addressing the issue of why sentences like the one
exemplified in (i) occur only in 8.1% of the cases with the direct object clitic doubling construction
in spoken Porteño Spanish according to Barrenechea and Orecchia’s (1977) study, while sentences
with indirect objects like the one illustrated in (ii) show a 51.1% of clitic doubling occurrence (the
sentences are mine):
(i) (Loi) vi al decanoi esta mañana
-.3. saw-I to-the dean this morning
‘I saw the dean this morning.’
(ii) (Lei) hablé al decanoi esta mañana
-.3 talked-I to-the dean this morning
‘I talked to the dean this morning.’
There are other studies on the same dialect that show a higher occurrence of direct object clitic
doubling such as that of Suñer (1988) but, significantly, a steady gap in the frequency percentage
with respect to indirect object clitic doubling still remains.
186 JON FRANCO

5. Concluding Remarks

It might well be the case that the debate between the movement hypothesis and
the base generation hypothesis for Romance clitics will never be settled. This is
justified since once the execution of the technical details is figured out, the
implementation of either hypothesis is feasible. However, the data analysed in
this paper neatly falls within the notion of clitics as agreement morphemes
generated in situ as heads of functional projections. Furthermore, by parametris-
ing the strength of the features of the agreement heads, one can easily account
for the differences in (morpho)syntactic behaviour across pronominal clitics in
some European languages.
To approach this insight from a different perspective, let us say that if
pronominal clitic systems have developed diachronically into agreement systems,
one could conceive of agreement not in absolute terms, but as a gradable
continuum along which pronominal clitics are located, as illustrated in (18).
Thus, the highest degree of object-verb agreement would be represented by
cross-referential morphemes on the verb in Basque (‘agreement clitics’ in Laka
1993). Interestingly, the phenomenon of Spanish clitic doubling, in which a
verbal argument in an A-position co-occurs with a coreferential clitic, also fares
elegantly with cross-linguistic agreement patterns. To be more precise, this paper
puts forward diverse evidence and argumentation that support the old, but
underdeveloped, insight that the syntax and morphology of object clitics in
contemporary Spanish clitic doubling dialects can be accounted for without a
complicated machinery if these clitics are treated as object-verb agreement
morphemes. Furthermore, there is a subtle entailment that can be drawn from the
proposed analysis, i.e. languages will vary not only across the traditional
parameter of feature strength, but also with respect to the categorial status of the
element in which features get lexicalised, AgrO or D as a case in point.

References

Aoun, J. 1979. ‘On government, case marking and clitic placement.’ Ms., MIT.
Aoun, J. 1981. The Formal Nature of Anaphoric Relations. Doctoral dissertation, MIT.
Aoun, J. 1985. A Grammar of Anaphora. Cambridge, Mass.: MIT Press.
Aoun, J. 1993.‘The syntax of doubled arguments.’ International Journal of Basque
Linguistics and Philology 27: 709–730.
Aoun, J., N. Hornstein, D. Lightfoot and A. Weinberg 1987. ‘Two types of locality.’
Linguistic Inquiry 18: 537–578.
 AGREEMENT AS A CONTINUUM 187

Baker, M. 1988. Incorporation: A Theory of Grammatical Function Changing. Chicago:

University of Chicago Press.
Barrenechea, A. M. and T. Orecchia 1977. ‘La duplicación de objetos directos e indirec-
tos en el español hablado en Buenos Aires.’ In: Lope Blanch, J. M. (ed.) Estudios
sobre el Español Hablado en las Principales Ciudades de América. Mexico City:
Universidad Nacional Autónoma de México, 351–381.
Bonet, E. 1991. Morphology after Syntax. Doctoral dissertation, MIT.
Borer, H. 1984. Parametric Syntax: Case Studies in Semitic and Romance Languages.
Dordrecht: Foris.
Borer, H. and Y. Grodzinsky 1986. ‘The syntax of pronominal clitics: Introduction.’ In:
Borer, H. (ed.) The Syntax of Pronominal Clitics (Syntax and Semantics, vol. 19).
New York: Academic Press, 1–14.
Cardinaletti, A. and M. Starke. 1999. ‘The typology of structural deficiency: On three
grammatical classes.’ In: van Riemsdijk, H. (ed.) Clitics in the Languages of Europe
(vol. 8 of Language Typology). Berlin: Mouton.
Carstairs, A. 1981. Notes on Affixes, Clitics and Paradigms. Bloomington: Indiana
University Linguistics Club.
Chomsky, N. 1970. ‘Remarks on nominalization.’ In: Studies on Semantics in Generative
Grammar. The Hague: Mouton, 11–61.
Chomsky, N. 1981. Lectures on Government and Binding. Dordrecht: Foris.
Chomsky, N. 1993. ‘A minimalist program for linguistic theory.’ In: Hale, K. and S.
Keyser (eds.) The View from Building 20. Cambridge, Mass.: MIT Press, 1–52.
Chomsky, N. 1995. The Minimalist Program. Cambridge, Mass.: MIT Press.
Chomsky, N. and H. Lasnik 1993. ‘The theory of principles and parameters.’ In: Jacobs,
J., A. von Stechow, W. Sternefeld and T. Vennemann (eds.) Syntax: An International
Handbook of Contemporary Research. Berlin: Walter de Gruyter, 506–569.
Comrie, B. 1989. Language Universals and Linguistic Typology: Syntax and Morphology
(2nd edn.). Chicago: Chicago University Press.
Diesing, M. 1992. Indefinites. Cambridge, Mass.: MIT Press.
Dobrovie-Sorin, C. 1994. The Syntax of Romanian. (Studies in Generative Grammar, vol.
40). Berlin: Mouton.
Enç, M. 1991. ‘The semantics of specificity.’ Linguistic Inquiry 22: 1–25.
Everett, D. 1989. ‘Clitic doubling, reflexives and word order alternations in Yagua.’
Language 65: 339–372.
Everett, D. 1996. Why There are No Clitics. Arlington: University of Texas, Summer
Institute in Linguistics, Publication 123.
Franco, J. 1991. ‘Spanish object clitics as verbal agreement morphemes.’ MIT Working
Papers in Linguistics 14: 99–114.
Franco, J. 1993. On Object Agreement in Spanish. Doctoral dissertation, University of
Southern California.
Franco, J. and A. Landa 1995. ‘Conditions after SPELL OUT: The case of Spanish clitic
climbing.’ In: Dainora A., R. Hemphill, B. Luka, B. Need and S. Pargman (eds.)
188 JON FRANCO

Clitics. CLS 31–II: Papers from the Parasession on Clitics. Chicago: University of
Chicago Press, 105–117.
Franco, J. and E. Mejías 1997. ‘Overt and covert raising to (Spec AGRDO) and the
interpretation of objects.’ Linguistic Analysis 27: 79–107.
Gerfen, H. 1991. ‘Reciprocity in Spanish: A puzzle of scope.’ In: Hunt, K., T. Perry and
V. Samiian (eds.) Proceedings of The Western Conference on Linguistics (WECOL),
152–163.
Hopper, P. and S. Thompson 1980. ‘Transitivity in grammar and discourse.’ Language
56: 251–299.
Hurtado, A. 1984. ‘Clitic chains.’ Ms., Simon Fraser University.
Jaeggli, O. 1982. Topics in Romance Syntax. Dordrecht: Foris.
Jaeggli, O. 1986. ‘Three issues in the theory of clitics.’ In: Borer, H.(ed.) The Syntax of
Pronominal Clitics (Syntax and Semantics vol. 19). New York: Academic Press,
15–42.
Kayne, R. 1975. French Syntax: The Transformational Cycle. Cambridge, Mass.: MIT
Press.
Kayne, R. 1989a. ‘Null subjects and clitic climbing.’ In: Jaeggli, O. and K. Safir (eds.)
The Null Subject Parameter. Dordrecht: Kluwer, 239–261.
Kayne, R. 1989b. ‘Facets of Romance past participial agreement.’ In: Benincá, P. (ed.)
Dialect Variation and the Theory of Grammar. Dordrecht: Foris, 85–103.
Laka, I. 1993. ‘The structure of inflection: A case study in X0 syntax.’ In: Hualde, J.I.
and J. Ortiz de Urbina (eds.) Generative Studies in Basque Linguistics. Amsterdam:
John Benjamins, 21–70.
Landa, A. 1995. Conditions on Null Objects in Basque Spanish and their Relation to
“Leísmo” and Clitic Doubling. Doctoral dissertation, University of Southern
California.
Lapesa, R. 1986. Historia de la Lengua Española. Madrid: Editorial Gredos.
Lyons, C. 1990. ‘An agreement approach to clitic doubling.’ Transactions of the Philolog-
ical Society, Vol. 88.1: 1–57.
Mahajan, A. 1992. ‘The specificity condition and the CED.’ Linguistic Inquiry 23:
510–516.
Mussafia, A. 1983. Scritti di Filologia e Linguistica, Daniele, A. and L. Renzi (eds.)
Padua: Antenore.
Ortiz de Urbina, J. 1989. Some Parameters of the Grammar of Basque. Dordrecht: Foris.
Pollock, J.-Y. 1989. ‘Verb movement, UG and the structure of IP.’ Linguistic Inquiry 20:
365–424.
Rivas, A. 1977. A Theory of Clitics. Doctoral dissertation, MIT.
Rivero, M.-L. 1986. ‘Parameters in the typology of clitics in Romance and Old Spanish.’
Language 62: 774–807.
Rizzi, L. 1986. ‘On chain formation.’ In: Borer, H. (ed.) The Syntax of Pronominal Clitics
(Syntax and Semantics vol. 19). New York: Academic Press, 65–95.
 AGREEMENT AS A CONTINUUM 189

Saltarelli, M. 1987. ‘The acquisition of agreement.’ In: Bahner, W., J. Schildt and D.
Viehweger (eds.) Proceedings of the XIV International Congress of Linguists, Berlin:
Akademie-Verlag, 810–815.
Saltarelli, M. 1990. ‘La grámatica de los clíticos y sus parámetros: Concordancia y
cadenas funcionales.’ Paper presented at the Asociación de Lingüística y Filología
de América Latina. IX Congreso Internacional, August 1990.
Silva-Corvalán, C. 1984. ‘Semantic and pragmatic factors in syntactic change.’ In: Fisiak,
J. (ed.) Historical Syntax. Berlin: Mouton, 555–597.
Sportiche, D. 1996. ‘Clitic constructions.’ In: Rooryck, J. and L. Zaring (eds.) Phrase
Structure and the Lexicon. Dordrecht: Kluwer, 213–277.
Strozer, J. 1976. Clitics in Spanish. Doctoral dissertation, University of California, Los
Angeles.
Suñer, M. 1988. ‘The role of agreement in clitic-doubled constructions.’ Natural Lan-
guage and Linguistic Theory 6: 391–434.
Suñer, M. 1999. ‘Clitic doubling of strong pronouns in Spanish.’ In: Franco, J., A. Landa
and J. Martin (eds.) Grammatical Analyses in Basque and Romance Linguistics.
Amsterdam: John Benjamins, 233–266.
Torrego, E. 1994. ‘On the nature of clitic doubling.’ International Journal of Basque
Linguistics and Philology 28: 199–214.
Uriagereka, J. 1992a. ‘Extraction parameters. A case study on underspecification.’ Ms.,
University of Maryland.
Uriagereka, J. 1992b. ‘The syntax of movement in Basque.’ In: Lakarra, J. and J. Ortiz
de Urbina (eds.) Syntactic Theory and Basque Syntax. Donostia/San Sebastian:
International Journal of Basque Linguistics Monographs, 417–446.
Uriagereka, J. 1995. ‘Aspects of the syntax of clitic placement in Western Romance.’
Linguistic Inquiry 26, 79–123.
Wackernagel, J. 1892. ‘Über ein Gesetz der Indogermanischen Wortstellung.’ Indoger-
manische Forschungen 1: 333–435.
Wald, B. 1979. ‘The development of the Swahili object marker: A study of the interaction
of syntax and discourse.’ In: Givón, T. (ed.) Discourse and Syntax. New York:
Academic Press, 505–524.
Zwicky, A. 1985. ‘Clitics and particles.’ Language 61: 283–305.
 Slovene Pronominal Clitics*

Marija Golden Milena Milojević Sheppard

University of Ljubljana

Abstract

The distribution of Slovene second position pronominal clitics in single and

multiple-clause clitic clusters is argued to be derivable by a movement analysis
according to which these clitics originate in a VP-internal A-position and land
right-adjoined to C0. The paper adopts a derivation approach to the phonology-
syntax interface, in which syntax need not anticipate the constraints of
phonology.

1. Introduction

The principal aim of this paper is to indicate the basic properties of Slovene
clitics in order to gain an understanding of the principles that guide their
distribution in single and multiple-clause clitic clusters. In spite of renewed
interest in cliticisation in general, and in Slavic languages in particular, descrip-
tions of the clitic systems have not yet been worked out into a unifying and
coherent typology (cf., among others, Anderson 1993; Halpern 1995; Kayne
1989, 1991; Perlmutter 1971; Rivero 1997; Tomić 1996; Zwicky 1977). It is to
be hoped that the range of Slovene data considered here may contribute towards
the emerging typology of Slavic clitic systems, particularly in the direction of a
better understanding of the ways in which Slovene departs from such a pure
second-position language as Serbo-Croatian. Such a description may also shed
light on the properties of second position cliticisation in general.
Since we are at an early stage of investigation, we restrict the present

* We would like to thank two anonymous reviewers for their careful reading of the manuscript and
valuable suggestions.
192 MARIJA GOLDEN AND MILENA MILOJEVIĆ SHEPPARD

discussion to pronominal clitics. Syntactically, Slovene clitics are second position

(‘Wackernagel’) clitics. We argue that their domain is CP rather than IP; we
propose that the Wackernagel position be identified as adjunction to the head
(C0) of their domain (Ouhalla 1996; Wilder and Ćavar 1994; Tomić 1996).
Prosodically, they are hosted by the first, categorially unspecified, maximal
projection and are usually enclitic to the last word in it. Unlike Serbo-Croatian
clitics, Slovene clitics may also be proclitic.
The framework we would like to adopt is the minimalist theory of grammar
(Chomsky 1995), with a derivational approach to the phonology-syntax interface,
in which syntax need not anticipate the constraints of phonology. We assume that
the clitic forms of the genitive, dative, and accusative personal pronouns are
identified as such in the lexicon ([+clitic]), and are inserted in the syntactic tree
by Merge. More specifically, we propose that Slovene pronominal clitics are
VP-internal arguments. We pursue the idea that the placement of clitics is
determined by regular feature-checking operations in syntax, which they share
with their corresponding full forms, although the two sets of pronouns may
differ in their feature content. As Slovene is not a clitic-doubling language, we
adhere to the assumption that the placement of clitics, as well as the internal
order of clitics in clitic clusters, is accomplished through syntactic movement
alone, with no reordering taking place in prosodic structure.
This paper is organised as follows. Section 1 is a brief overview of the
basic empirical observations of Slovene single-clause pronominal clitics. While
outlining their general properties, we compare Slovene clitics with the clitics in
Serbo-Croatian (SC). In Section 2, we provide empirical support for the proposed
landing site of the clitics and the movement approach to clitic placement. In 2.1,
we argue that Slovene clitics right-adjoin to the C node of their CP domain, and
in 2.2, that they assume this position by Move a. In the final part of the paper,
we sum up the differences between the descriptive properties of Slovene and
Serbo-Croatian clitics.

2. Standard properties of Slovene single-clause clitics

Slovene pronominal clitics, listed in (1), are allomorphs of strong pronouns. They
show morphological case {genitive, dative, accusative}, number {singular, dual,
plural}, and gender {masculine, feminine, neuter} distinctions. They are recipi-
ents of q-roles and enter agreement relations. Thus in (2), the clitic ga ‘him’
agrees in its features (Sg., 3rd person, masculine, Accusative) with the predica-
tive adjective domišljavega ‘conceited’; the use of the non-agreeing feminine
 SLOVENE PRONOMINAL CLITICS 193

form domišljavo, for instance, is unacceptable. (Throughout the text, pronominal

clitics will appear in boldface italics.)
(1) Genitive Dative Accusative1
strong clitic strong clitic strong clitic
1st person  mêne me mêni mi mêne me
 nàs nas nàm nam nàs nas
 náju naju náma nama náju naju
2nd person  têbe te têbi ti têbe te
 vàs vas vàm vam vàs vas
 váju vaju váma vama váju vaju
3rd person  njêga ga njêmu mu njêga ga
masculine  njìh jih njìm jim njìh jih
 njìju ju/jih njìma jima njìju ju/jih
feminine  njé je njéj/ ji njó jo
njèj/njí
(2) Uspeh ga je naredil domišljavega (*domišljavo)
success him- is- made conceited
‘The success made him conceited.’
Single-clause clitics do not appear in a random order. They form a cluster with
mutually exclusive fillers from each positional class. The clitic template (3a)
indicates that of the pronominal clitics, the reflexive clitic appears first, with the
dative clitic preceding the accusative clitic (3b). The clitic cluster begins with the
present tense forms of the auxiliary verb biti, except for the 3rd person singular
form je, which ends the clitic cluster. In (3c) the position of the first Aux is taken
by the conditional form bi. The relative order of pronominal clitics in clitic
clusters is partially determined by case and reflects the unmarked order of their
non-clitic counterparts (cf. (3c) with (3d)).2

1. Toporišič (1976): 240, 535.

2. Although the order of the pronominal clitics in (3a) appears to be constrained by case, it is not
inviolable. For example, when the reflexive clitic is the accusative argument, it will precede the
dative clitic. To our ears, the ordering of clitics in such a multiple-clause clitic cluster as in (i), i.e.
‘accusative + dative’/‘embedded clause clitic + matrix clause clitic’ is more usual than the order in (ii),
whereby the dative clitic would precede the accusative clitic.
(i) Peter se jim je hotel predati brez boja
Peter se- them- is- wanted to-surrender without fight
‘Peter wanted to surrender himself to them without a fight.’
(ii) ?*Peter jim se je hotel predati brez boja
194 MARIJA GOLDEN AND MILENA MILOJEVIĆ SHEPPARD

(3) a. Aux1 Reflexive Dative Accusative Genitive Aux2

b. Peter se ji ga je napil
Peter [Refl-Acc se-] [Dat her] [Acc him] [Aux 2 is] drunk
‘Peter got drunk on her.’
c. Peter naj [Aux 1 bi] [Dat ji] [Acc jo] [Acc jo] učil
d. Peter naj bi učil ŠpeliDat hčerkoAcc matematikoAcc
‘Peter should teach mathematics to Špela’s daughter for her.’
Slovene clitics are usually enclitic to the last word of the leftmost maximal
projection, which can be of any size and category. In (4), the second position for
the main-clause clitic cluster mu jo je follows the initial constituent, an adverbial
clause; the clitic cluster is proclitic to the participle ponudil. The second position
of the embedded-clause clitic cluster ga je is after the initial adverbial comple-
mentiser ker, and is enclitic to it.3
(4) [[Ker] ga je zeblo], mu jo je
[[because him- is- cold] him- her- is-
ponudil vročo
offered hot
‘Because he was cold he offered it to him hot.’
The first position can be filled by the first XP of the clause (zanimivo pismo
‘interesting letter’ in (5a)) or by the first YP of XP (zanimivo ‘interesting’ in
(5b)). However, it is rare in present-day Slovene for the clitic to appear after the
first phonological word, thus breaking up the sentence-initial constituent ((5b); cf.
also (6a)). What some authors claim to be valid for SC (Halpern 1995; Schütze
1994) does not hold for Slovene: there is no first prosodic word/first syntactic

In addition, the clitic cluster may contain more than a single clitic in the same case, if one is
reflexive (iii). The reflexive clitics, however, do not co-occur: in combining the clitics in (iv) into a
single cluster, the dative reflexive would be omitted (v).
(iii) Drznil si ji ga je hvaliti
dared  her him is to-praise
‘He dared praise him to her.’
(iv) Drznil si je predstavljati se kot njen prijatelj
dared  is to-introduce himself as her friend
‘He dared introduce himself as her friend.’
(v) Drznil se je predstavljati kot njen prijatelj
3. We thank one of the anonymous reviewers for making us aware that our original description
needed clarification.
 SLOVENE PRONOMINAL CLITICS 195

constituent alternative. The host of the Slovene enclitic is always a syntactic

constituent.4
(5) a. Zanimivo pismo ji je napisal
interesting letter her- is- written
‘He wrote her an interesting letter.’
b. ?Zanimivo ji je pismo napisal
If the sentence-initial prosodic word were a legitimate target of encliticisation in
Slovene, the preposition blizu in (6b) could act as a phonological host to the
clitic. When clitics seem to appear after the first word of the matrix sentence, as
in (6c), the first word turns out to be a constituent that has been targeted by
syntactic movement. In (6c), for example, a subconstituent of a quantified noun
phrase has been preposed. The unacceptable placement of the clitic cluster in
(6b) can be ruled out by assuming that syntactic movement cannot strand
prepositions in Slovene (cf. Golden 1996b).
(6) a. ?Mojega mu kolesa že ne bom posodil
my him- bike  not shall lend
‘I will definitely not lend him my bike.’
b. *Blizu jo je gozda ustavil
close her- is- forest stopped
‘Close to the forest he stopped her.’
c. Veliko/Koliko/Toliko ji je kupil knjig
many/how many/so many her- is- bought books
(, da jih ne utegne prebrati)
(, that them- not manage to-read)
‘He bought her many books/How many books/So many books
did he buy her (that she doesn’t have the time to read them).’
Slovene clausal clitics do not belong, in an immediately obvious way, to any of
the three clitic systems proposed so far in the clitic typology (cf., among others,
Anderson 1993; Bošković this volume; Klavans 1982, 1985; Rivero 1997; Tomić
and Beukema 1997).
(i) Slovene, like modern Czech (cf. Toman 1993), cannot be characterised
as a pure Wackernagel language. As the literature on clitics notes in passing (cf.,
among others, Bennett 1987; Bošković this volume), Slovene clitics may appear

4. That the host of the enclitic is a syntactic constituent also in SC has been argued for by authors
like Bošković and Progovac (this volume).
196 MARIJA GOLDEN AND MILENA MILOJEVIĆ SHEPPARD

first in the prosodic structure of their matrix CP (7), the position in which true
Wackernagel clitics are prohibited. The Slovene clitic appears in sentence-initial
position when neither independently motivated syntactic movement such as
wh-movement, topicalisation, or (long-head) participle preposing, nor base-
generation such as lexical insertion of an overt complementiser or an adverb
provides lexical material to the left of the clitic as its prosodic host. When clitics
appear in the initial position of the matrix sentence with an appropriate intonation
pattern, the sentence is interpreted as a yes/no question.5
(7) Ga še nisi srečal?
him- yet not-are- met
‘Haven’t you met him yet?’
However, sentence-initial clitics are not restricted to verbal questions without an
overt interrogative complementiser, as in (7). In the spoken register, they
frequently occur in declaratives as well. Thus (8a) can be a response to a
question such as Kje je pismo? ‘Where is the letter?’, or (8b), to a question, such
as Kakšen se ti zdi Peter? ‘How does Peter strike you?’
(8) a. Sem ga že oddal6
am- it- already mailed
‘I have already mailed it.’
b. Ga še nisem srečal
him- yet not-am- met
‘I haven’t met him yet.’
(ii) If the Slovene clitic system were of the Romance type, as is the case with
the I0-oriented verbal clitics of Bulgarian and Macedonian, the placement of
clitics would be restricted to a preverbal or postverbal position, depending on the
finiteness of the verb. In (9) the clitic cluster ji ga is separated from the finite
verb by a sentential adverb (očitno ‘obviously’) and an aspectual/VP (vedno
‘always’) adverb. Additionally, treating Slovene clitics as I0-oriented would leave
the second position effects of Slovene, observed in sentences (4) to (6), unac-
counted for.

5. As in other Slavic languages, embedding a clause in Slovene requires an overt complementiser.

6. Traditional Slovene grammar assumes that sentence-initial clitics result from deleting the material
preceding the clitics, frequently a complementiser or a conjunction. In (8a), for example, the deleted
material may be saj ‘but’: Saj sem ga že oddal.
 SLOVENE PRONOMINAL CLITICS 197

(9) Peter ji ga očitno vedno hvali

Peter her- him- obviously always praises
‘Peter apparently always praises him to her.’
(iii) Finally, Slovene clitics cannot be identified as Romance second position
clitics (the so-called Tobler-Mussafia clitics, cf. Comrie 1989). For Slovene
clitics to be identified as TM clitics, nothing would be allowed to come between
the clitics and the verb, which is contradicted by (9); at the same time, the clitic
would be prohibited from appearing in clause-initial position, contrary to the
distribution in (7) and (8).

3. The distribution of single-clause pronominal clitics

The claim that the second position constraint is a syntactic restriction in Slovene
means that the position Slovene clitics occupy in their CP would be identified in
syntactic terms. In spite of claims in the literature which systematically rule out right-
adjunction (Kayne 1994), we think that the empirical data presented here give support
to the assumption that Slovene clitics right-adjoin to the C node of their CP domain.

3.1 Right-adjunction to C0

The first piece of supporting evidence for the C0 node as the proposed adjunction
site is the neutral word order. In discourse-neutral contexts, clitic clusters immedi-
ately follow complementisers (da in (10a); ali in (10b)) and must precede subject
NPs, when the latter are lexical NPs (10c). Identifying the clitic landing site as
right adjunction to C0 in the suggested phrase structure hierarchy adequately
discriminates between the acceptable and unacceptable clitic orderings in (10).
(10) a. Mislil je, [CP [C da se ji je … [VP prikupil…]]
thought is- that  her- is- endear
‘He thought that he had endeared himself to her.’
b. Ali si mu ga dal? Pravi, da si mu ga dal
whether are- him- it- given?
‘Did you give it to him?’ ‘He says that you gave it to him.’
c. Pravi, [CP [C da (*Peter) mu ga] [IP (ØPeter) prinese
says that (*Peter) him- it- (ØPeter) brings
jutri]]
tomorrow
‘He says that Peter brings it to him tomorrow.’
198 MARIJA GOLDEN AND MILENA MILOJEVIĆ SHEPPARD

More support for the syntactic approach adjoining clitics to the C node of their
CP domain comes from the observation that a clitic may interrupt a string of
interrogative wh-phrases beginning multiple questions (11). As argued in Golden
(1995) and (1996a), in Slovene multiple questions, the first wh-phrase lands in
the SpecCP position, while the remaining interrogative phrases adjoin to the IP
node. Under this analysis, the clitic cluster ti ga je in (11) appears outside the IP
projection, and right-adjoining it to C0 would yield the desired order. Similarly,
clitics also follow any topicalised phrases (12a). More specifically, if the
topicalised structure contains an overt complementiser, as in (12b), they appear
immediately after the complementiser.
(11) [SpecCP Kdo ti ga je [IP za kakšno ceno [IP prvi ponudil]]]?
‘Who offered it to you first for what price?’
(12) a. [TopicPi Knjigo] sem mu samo posodil [ti]
book am- him- only lent
‘As for the book, I only lent it to him.’
b. [[TopicPi Knjigo] [CP [C da sem mu] dal [ti], ni
book that am- him- given is-not
res, samo posodil sem mu jo7
true only lent am- him- her-
‘As for the book, that I gave it to him is not true, I only lent it
to him.’
The claim that second position placement, identified here as right-adjunction to
C0, is a syntactic restriction in Slovene predicts the absence of delaying effects.
SC clitics occupy the third position in the sentence when preceded by a preposed
heavy constituent, a parenthetical or an appositive (Bošković this volume; Wilder
1996). In Slovene, preposing a heavy constituent does not demote the clitic to the
third position (13). Similarly, the Slovene clitic, but not the SC clitic, retains its
second position when preceded by an appositive (14) or a parenthetical (15).
(13) a. Sa Petrom Petrovićem srela se samo Milena. (SC; Bošković
this volume, (57))
‘With Peter Petrović, only Milena met.’

7. For the purposes of this paper, we assume that topicalised phrases are located in a Topic Phrase
dominating CP. Cf. the possibility of intervening material, as in Knjigo, praviš, da si mu samo
posodil? ‘As for the book, you say, that you only lent it to him?’
 SLOVENE PRONOMINAL CLITICS 199

b. Z Janezom Drnovškom se je srečala samo Milena

with Janez Drnovšek se- is- met only Milena
‘With Janez Drnovšek only Milena met.’
c. Z Janezom Drnovškom, sedanjim predsednikom slovenske
with Janez Drnovšek, present President of the Slovene
vlade, # se je srečala samo Milena
government se- is- met only Milena
‘With Janez Drnovšek, the present President of the Slovene
government, only Milena met.’
(14) a. *Ja, tvoja mama, sam ti obećala igračku. (SC; Bošković this
volume, (51))
‘I your mother promised you a toy.’
b. Jaz, tvoja mama, sem ti obljubila igračko
I your mother am- you- promised toy
‘I, your mother, promised you a toy.’
(15) a. Kdo — za božjo voljo — ti je razbil avto?
who — for God’s sake — you- is- ruined the car
‘Who, for God’s sake, ruined your car?’
cf. Kdo ti je — za božjo voljo — razbil auto?
b. *Tko — za boga — je razbio auto?
who for God be-3rd -cl ruined car
(SC; Wilder 1996, (101a, 101b))
‘Who, by God, ruined the car?’
cf. Tko je — za boga — razbio auto?
Since preposing a heavy constituent, inserting a parenthetical or adding an
appositive splits the intonation phrase, the placement of the SC clitic after the
first prosodic element of the second intonation phrase (after the preposed/inserted
material) is evidence for the claim that SC clitics are restricted to occur in the
second position of their intonation phrase. The assumption that Slovene clitics
follow the first syntactic constituent of their CP makes the correct prediction that
they will appear immediately after the preposed/inserted constituent. Additionally,
if cliticisation, universally, does not cross the boundary of the intonation phrase,
their distribution as initial elements immediately after the break supports the
claim that in the prosodic structure they may be proclitic (15).
In (13a), with a preposed prepositional phrase, Bošković (this volume)
shows that the SC clitic cannot occupy the second position in its clause. Rather,
it is delayed and appears in what is syntactically the third position. Preposing an
equivalent prepositional phrase in Slovene has no such delaying effects on clitic
200 MARIJA GOLDEN AND MILENA MILOJEVIĆ SHEPPARD

placement. The clitic cluster in (13b) appears in the expected second position.
(13c) shows that the preposed constituent can be even heavier, without necessari-
ly postponing the occurrence of the clitic to the third position.
Given the assumption that clitics have a syntactically fixed position, this
position is then defined for all CP structures, including ‘be+predicate’ construc-
tions.8 For Slovene, the normal order in such constructions is for the predicate
phrase as a whole to precede the clitic forms of the copula biti (16a), (17a),
(18a). The ordering where the clitic follows the first prosodic element of its
intonation phrase, which is normal in SC, may occasionally be encountered in
Slovene with premodified adjectival predicates (17b); it is rare/obsolete with
nominal and prepositional predicates (16b), (18b).
(16) a. Sposoben direktor je
capable manager is
‘He is a capable manager.’
Cf. SC (Bošković (this volume) (20c)):
*Sposoban direktor je (on)
capable manager is (he)
‘He is a capable manager.’
b. ?*Sposoben je direktor
Cf. SC (Bošković (this volume) (20d)):
Sposoban je direktor (on)
c. Je sposoben direktor Cf. SC: *Je sposoban direktor
(17) a. Zelo dolgočasen si
very boring are
‘You are very boring.’
Cf. SC (Bošković (this volume) (20e)):
*Jako dosadni su (oni)
extremely boring are (they)
‘They are extremely boring.’
b. Zelo si dolgočasen
Cf. SC (Bošković (this volume) (20f)):
Jako su dosadni (oni)
c. Si zelo dolgočasen Cf. SC: *Su jako dosadni

8. As demonstrated by Browne (1975) and discussed further by Bošković (this volume), ‘be +
predicate’ clauses are one of the two typical 1W contexts that provide empirical grounds for prosodic
inversion in SC (cf. Halpern 1995).
 SLOVENE PRONOMINAL CLITICS 201

(18) a. V drugi sobi sem

in other room am
‘I am in the other room.’
Cf. SC (Bošković (this volume) (26),(25)):
*U drugoj sobi su; U sobi su
in other room are
‘They are in the other room.’
b. ?*V drugi sem sobi Cf. SC: U drugoj su sobi
c. Sem v drugi sobi Cf. SC: *Sam u drugoj sobi

3.2 Move a in syntax

In the approach adopted in this paper, Slovene clitics are not generated in C0;
they move there as pronominal arguments from their base positions within the
VP, through regular checking operations in the syntax.
The movement approach to clitic placement is supported first by the
observation that clitic clusters contain clitics of various syntactic categories, yet
they all surface in the same position in the sentence. Additionally, there may be
a categorial diversity of clitics occupying a single position in the cluster, i.e., the
positional class may contain homophonous morphemes, as in (19).
(19) a. da se ji je posmehoval
‘that he made fun of her.’
b. da se mu je kolcalo
‘that he had hiccups’
c. da se ji je hvalil
‘that he praised himself to her’
d. da se mu je govorilo o njej same laži
‘that people were telling him only lies about her’
In all the examples in (19), the reflexive morpheme se is ordered before the
dative and/or accusative clitic, irrespective of its morphosyntactic function. In
(19a) it is part of the lexical entry of an obligatorily reflexive verb (posmehovati
se ‘make fun of’). In (19b) the morpheme se manipulates the predicate-argument
structure of the verb by turning a nominative subject–predicate agreement clause
(On je kolcal ‘He hiccuped’) into an impersonal one with the dative logical
subject. In (19c) se is a contrastive reflexive argument of the verb (On je hvalil
sebe), and in (19d) it denotes a generic human subject. Irrespective of its
morphosyntactic category and function, the morpheme se has its fixed and unique
position in the clitic cluster. If the morphemes se were base-generated in the
202 MARIJA GOLDEN AND MILENA MILOJEVIĆ SHEPPARD

position they occupy in S-structure, it would be difficult to see what the

categorical specification of a node dominating such diverse elements should be.
More evidence for the movement approach is provided by idiomatic VPs
with an object clitic. Although in surface structure the verb and its idiomatic
clitic complement do not form a constituent because of the second position
constraint, there must be a point in the derivation of such idiomatic VPs where
they are treated as constituents, in order to state their non-compositional mean-
ing. The VP idiom mahniti jo in (20) has the meaning ‘to walk’.
(20) Peter jo je mahnil čez travnik
Peter her- is- waved across meadow
‘Peter walked across the meadow.’
Additional evidence supporting the movement approach can be gleaned from
clitic climbing (cf., among others, Kayne 1989; Roberts 1997; Uriagereka 1995).
Under the assumption that pronominal clitics undergoing clitic climbing receive
their q-roles and check their case features when they are still arguments of the
embedded predicates, neither their thematic status nor case features call for any
extension of the main-stream q-theory or case theory. If they were to be base-
generated in the position they occupy in S-structure, additional mechanisms
would be required.
Slovene pronominal clitics, objects of infinitives which are complements to
subject-control verbs, almost obligatorily join the matrix-clause cluster, in which
they assume their usual position. In (21a) the dative clitic ji and the accusative
clitic ga, underlyingly the indirect and the direct objects of the most deeply
embedded infinitive opisati ‘describe,’ join the matrix clause verbal clitic sem
‘am’. In (21b) the clitics ji and ga join the matrix clause with the raising
predicate utegniti ‘may’ (Bolta 1985). The resulting multiple-clause clitic cluster
follows the order of the clitics in the single-clause clitic cluster, as established in
(3a). In (21c), for example, both the matrix-clause clitics, the auxiliary clitic sem
and the reflexive particle se, precede the embedded-clause clitics ji and ga. Clitic
climbing out of subject control and raising structures, as in (21), seems to be
almost obligatory. It applies even to clitics contained in idiomatic VPs (21d).
(21) a. …ker proi sem sklenil [PROi poskusiti [PROi opisati ji ga]]
…ker sem ji ga sklenil poskusiti opisati
‘… because (I) decided to try to describe him to her’
b. …ker proi sem utegnil [ti opisati ji ga slabo]
…ker sem ji ga utegnil opisati slabo
‘…because (I) may have described him to her badly’
 SLOVENE PRONOMINAL CLITICS 203

c. …ker sem sei jij gak [prol [VP naveličal ti [PROl hvaliti tj tk]]]
…because am- se- her- him- tired to-praise
‘…because (I) grew tired of praising him to her’
d. Dovolil nam jo je ucvreti domov
proi je dovolil namj [PROj [ucvreti jo] domov]
allowed us- her- is- to-speed home
‘He let us speed towards home.’
If clitic placement is the result of syntactic movement, then clitic movement is
expected to be constrained by restrictions valid for movement in general, such as
the Condition on Extraction Domains, and to show minimality effects.9 The
sentence in (22a) shows that Slovene clitics move out of governed positions in
adjectival phrases in order to assume their position in the clitic cluster. In
(22b) the adjectival phrase is an adjunct in the structure of the NP; this time no
clitic movement is possible. As (23) shows, the constraint is also valid for
wh-movement.10
(22) a. Politiki so jei vedno [AP lačni [NP ti]]
politicians are her- always hungry
(Cf.: Politiki so vedno [AP lačni [NP slave]], i.e., ‘Politicians are
always hungry for glory.’)
‘Politicians are always hungry for it.’
b. *Bil jei je [NP [AP ti lačen] politik]
been is- her- hungry politician
(Cf.: Bil je [NP [AP slave lačen] politik])
‘He was a politician, hungry for it’

9. In other words, the account of why the clitic cluster sem se ji ga is acceptable in (21c) and
unacceptable in (i), although it does not err against the ordering given in (3a), is expected to be
syntactic in nature.
(i) *…ker sem sek jij gai [slišal ti [PROi dovoliti tj [PROj igrati tk na dvorišču]]]
…because am- se- her- him- [heard to-allow to-play on courtyard
‘…because I heard him allow her to play in the courtyard’
10. This, of course, should not suggest a claim that the two instances of movement share all of their
properties. One immediately obvious difference relates to the observation that wh-movement may
(i-a), but clitic climbing may not (ii-b), cross over a [+Tense]-clause boundary.
(i) a. To so ključi, katerei je Peter mislil, da je Špela [V′ izgubila ti] na poti domov
‘These are the keys which Peter thought that Špela lost on her way home.’
(ii) a. To so ključi, ki jihi je Špela [V′ izgubila ti] na poti domov
b. *To so ključi, ki jihi je Peter mislil, da je Špela [V′ izgubila ti] na poti domov
cf.: To so ključi, ki je Peter mislil, da jihi je Špela [V′ izgubila ti] na poti domov
204 MARIJA GOLDEN AND MILENA MILOJEVIĆ SHEPPARD

(23) a. Česai so politiki vedno [AP lačni [NP ti]]?

what- are politicians always hungry
‘What are politicians always hungry for?’
b. *Česai je bil [NP [AP ti lačen] politik]?
what- is- been hungry politician
‘What was he a hungry politician for?’
The sentence in (24) illustrates the effects of the Complex NP Constraint. As
shown in (21) above, multi-clause clitics may reorganise into a single clitic
cluster if they are embedded under control or raising predicates. The sentence in
(24) shows that when clitics are arguments of infinitives which are clausal
complements to control nouns, no clitic climbing may take place. Further data
show that clitic climbing is sensitive not only to such syntactic information as the
value finite of the embedded clause, but also to subject vs. object control and
negation. As noted in the literature, these are syntactic constraints and are not
replaceable by prosodic restrictions (cf. Progovac 1993a, 1993b).
(24) a. Peteri prikriva [svojoi pripravljenost [PROi predstaviti mu jo ta
večer]
‘Peter is concealing his willingness to introduce her to him this
evening.’
b. *Peter mu jo prikriva [svojo pripravljenost [predstaviti ta večer]]
One last piece of evidence presented here is the use of the genitive of negation
in Slovene. In one of its uses, the genitive case replaces the accusative case of
the object when the latter is in the domain of sentential negation. The accusative
clitic jo in (25a) is assigned its q-role as a complement to the verb prositi ‘ask’.
In order to receive genitive case in (25b), the clitic must be in the c-commanding
domain of negation. In the suggested hierarchical structure, the second position
clitic je in (25b) is not c-commanded by the negative operator ne.11 The move-
ment approach can easily provide the required hierarchical environment: there is
a stage in the derivation of sentence (25b), during which the clitic is in the
c-command domain of the negative operator. The same argument also applies to
the genitive clitic as the subject of the small clause in (26b) and the genitive
clitic as the object of the control infinitive in (27b). (28) makes the same point
with the use of the genitive of negation with nominative subjects of existential
sentences.

11. The negative operator ne fuses with the auxiliary clitic sem into nisem.
 SLOVENE PRONOMINAL CLITICS 205

(25) a. [CP [C da sem jo] [AgrSP jaz [prosil za pomoč]]]

‘that I asked her for help.’
b. [CP [C da jei [AgrSP jaz [NegP ni- [TP -sem …[V′ prosil ti] za
pomoč]]]]]]
‘that I did not ask her for help.’
(26) a. Uspeh bo naredil [SC ŠpeloAcc lahkomiselno]
‘The success will make Špela irresponsible.’
b. Uspeh [jeGen]i ne bo naredil [ti lahkomiselne]
(27) Peteri [jeGen]j ne želi [PROi razočarati tj]
‘Peter does not want to disappoint her.’
(28) a. Rekel je, da je bila na mizi [knjigaNom]
‘He said that there was a book on the table.’
b. Rekel je, da jeGen ni bilo na mizi

4. Conclusion

In this paper we have argued for a movement approach to second position

Slovene pronominal clitic placement. Clitics are taken to move overtly out of
their VP-internal position, ultimately landing right-adjoined to C0. In our future
work, we will take up the view that clitics move out of the VP to the relevant
specifiers for feature-checking reasons, yet the exact mechanism responsible, in
particular, for the movement of clitics to C0 has yet to be worked out.
A summary of the descriptive properties of Slovene clitics with respect to
SC clitics is given in (29).
(29) domain dominance precedence
S: CP-phrase Initial suffix/prefix
SC: I-phrase Initial suffix
In terms of Anderson’s (1993) domain parameter, Slovene clitics select the value
‘CP’. Bošković (this volume) specifies the domain of SC clitics in terms of
prosodic structure, i.e., they select the value ‘Intonation-phrase’ (I-phrase). In
terms of the Bošković’s dominance parameter, clitics in both languages select the
value Initial. The relevant initial constituent for the second position effects in
Slovene is the first syntactic constituent of the CP domain containing the clitic.
For SC clitics, the relevant constituent is the initial element of their I-phrase; it
may be either the first prosodic word or the first prosodic phrase. In terms of the
direction of phonological attachment (the precedence parameter), Bošković (this
206 MARIJA GOLDEN AND MILENA MILOJEVIĆ SHEPPARD

volume) lexically specifies SC clitics as inherently enclitic; they are clearly

suffixes. As phonologically weak elements, they follow their host. In Slovene,
clitics have a structurally fixed position, without being phonologically specified
as inherently enclitic.

References

Anderson, S. R. 1993. ‘Wackernagel’s revenge: clitics, morphology, and the syntax of

second position.’ Language 69: 68–98.
Bennett, D. C. 1987. ‘Word-order change in progress: the case of Slovene and Serbo-
Croat and its relevance for Germanic.’ Journal of Linguistics 23: 269–287.
Bolta, M. 1985. Nekateri Vidiki Stavčnega Dopolnjevanja v Angleški in Slovenski Skladnji.
Doctoral dissertation, University of Ljubljana.
Bošković, Ž. This volume. ‘Second position cliticisation: syntax and/or phonology?’
Browne, W. 1975. ‘Serbo-Croatian enclitics for English-speaking learners.’ In: Filipović,
R. (ed.) Contrastive Analysis of English and Serbo-Croatian, Vol.1. Zagreb: Institute
of Linguistics, Faculty of Philosophy, University of Zagreb, 105–134.
Chomsky, N. 1995. The Minimalist Program. Cambridge, Mass.: MIT Press.
Comrie, B. 1989. Language Universals and Linguistic Typology (2nd edn.). Chicago:
Chicago University Press.
Golden, M. 1995. ‘Multiple Wh-questions in Slovene.’ In: Browne W., E. Dornisch, N.
Kondrashova and D. Zec (eds.) The Cornell Meeting 1995. Ann Arbor: Michigan
Slavic Publications, 240–266.
Golden, M. 1996a. ‘Interrogative Wh-movement in Slovene and English.’ Acta Analytica
14: 145–186.
Golden, M. 1996b. ‘K-premik in skladenjski otoki v slovenski skladnji.’ In: Razprave XV,
Slovenska akademija znanosti in umetnosti, Razred za filološke in literarne vede,
Ljubljana, 237–253.
Halpern, A. 1995. On the Placement and Morphology of Clitics. Stanford, Calif.: CSLI
Publications.
Huang, C.T.J. 1982. ‘Move wh in a language without wh movement.’ The Linguistic
Review 1: 369–416.
Kayne, R.S. 1989. ‘Null subjects and clitic climbing.’ In: Jaeggli, O. and K. Safir (eds.)
The Null Subject Parameter. Dordrecht: Kluwer, 239–261.
Kayne, R.S. 1991. ‘Romance clitics, verb movement, and PRO.’ Linguistic Inquiry 22:
647–86.
Kayne, R.S. 1994. The Antisymmetry of Syntax. Cambridge, Mass.: MIT Press.
Klavans, J. L. 1982. Some Problems in a Theory of Clitics. Bloomington: Indiana Universi-
ty Linguistics Club.
Klavans, J. L. 1985. ‘The independence of syntax and phonology in cliticization.’
Language 61: 95–120.
 SLOVENE PRONOMINAL CLITICS 207

Ouhalla, J. 1990. ‘Clitic movement and the ECP.’ Lingua 79: 165–215.
Perlmutter, D. 1971. Deep and Surface Structure Constraints in Syntax. New York: Holt,
Rinehart and Winston.
Progovac, L. 1993a. ‘Locality and subjunctive-like complements in Serbo-Croatian.’
Journal of Slavic Linguistics 1: 119–145.
Progovac, L. 1993b. ‘Clitics in Serbian/Croatian: Comp as the second position.’ In:
Halpern, A. and A. Zwicky (eds.) Approaching Second: Second Position Clitics and
Related Phenomena. Stanford, Calif.: CSLI Publications, 411–428.
Roberts, I. 1997. ‘Restructuring, head movement, and locality.’ Linguistic Inquiry 28:
423–460.
Rivero, M.-L. 1997. ‘On two locations for complement clitic pronouns: Serbo-Croatian,
Bulgarian and Old Spanish.’ In: van Kemenade, A. and N. Vincent (eds.) Parame-
ters of Morphosyntactic Change. Cambridge: Cambridge University Press, 170–206.
Schütze, C. 1994. ‘Serbo-Croatian second position clitic placement and the phonology-
syntax interface.’ MIT Working Papers in Linguistics 21: 373–473.
Toman, J. 1993. ‘A note on clitics and prosody.’ Eurotyp Working Papers, Clitics in
Germanic and Slavic 4: 113–118.
Tomić, O. M. 1996. ‘The Balkan Slavic clausal clitics.’ Natural Language and Linguistic
Theory 14: 811–873.
Tomić, O. M. and F. Beukema 1997. ‘On clitic sites.’ Paper presented at the Workshop
on Clitic Phenomena in English and other European Languages, ESSE/4, Debrecen,
Hungary.
Toporišič, J. 1976. Slovenska slovnica. Maribor: Založba Obzorja.
Uriagereka, J. 1995. ‘Aspects of the syntax of clitic placement in Western Romance.’
Linguistic Inquiry 26: 79–123.
Wilder, C. 1996. ‘English finite auxiliaries in syntax and phonology.’ Paper presented at
the Workshop on Clitic Phenomena in English and other European Languages,
ESSE/4, Debrecen, Hungary.
Wilder, C. and D, Ćavar 1994. ‘Long head movement? Verb movement and cliticization
in Croatian.’ Lingua 93: 1–58.
Zwicky, A. 1977. On Clitics. Bloomington: Indiana University Linguistics Club.
 Direct Object Clitic Doubling
in Albanian and Greek*

Dalina Kallulli
University of Durham

Abstract

Unlike object agreement markers, direct object clitics in Albanian and Greek
are restricted in their distribution and display operator-like properties. This
paper shows that direct object clitic doubling in Albanian and Greek produces
information structure in a systematic way, in that doubled DPs are unambigu-
ously interpreted as topics. Hence, topichood is syntactically encoded in these
languages. Specificity cannot be bestowed on an argument by a doubling clitic
or by scrambling but is fundamentally related to the D-head.

1. Introduction

A pervasive phenomenon in the languages of the Balkans is clitic doubling. This

study investigates clitic doubling of direct objects in two of these languages:
Albanian and Greek. This undertaking is motivated by the need to gain deeper

* This paper deals with topics that are further elaborated in the second and third chapters of my
doctoral dissertation (Kallulli 1999). Versions of it were presented at the GLOW-workshop ‘The
Syntax of Balkan Languages’ (Athens, April 1996) and at the workshop on Clitic Phenomena in
English and other European Languages, ESSE/4 (Debrecen, September 1997). I am grateful to these
audiences for their comments. In particular, I wish to thank Frits Beukema and Marcel den Dikken
for making possible its presentation at the Debrecen workshop in my absence. I am indebted to
Antonia Androutsopoulou for her invaluable help with the Greek data. I also thank Lars Hellan, Joe
Emonds, Georg Niklfeld and an anonymous reviewer for detailed comments.
Parts of the material contained in this article appear in the Proceedings of ConSOLE V and the
Proceedings of WCCFL 17.
210 DALINA KALLULLI

insight into the nature of clitic doubling constructions, and in turn contributes to
the general question of why clitic doubling appears at all. Doubling constructions
are by their nature strongly reminiscent of object agreement constructions. Yet,
there are essential differences between the two that beg for explanation. The
Albanian and Greek patterns confirm the idea that in spite of certain similarities
between clitic doubling and object agreement phenomena, the two cannot be
equated. For instance, unlike object agreement markers, direct object clitics in
Albanian and Greek have a restricted distribution and operator-like properties. It
will be shown that the factors determining clitic doubling of direct object DPs in
both languages are by and large identical and can be captured by a uniform
syntactic analysis. Crucially, I argue that direct object clitics in both languages
unequivocally mark the DPs they double as [−Focus], which in analogy with the
[+Focus] feature on phrases (cf. Jackendoff 1972; Horvath 1986; Rochemont
1986; Brody 1990, i.a.), will be defined as a syntactic feature interpretable at
both the LF and PF interfaces. Consequently, clitic doubling of direct object DPs
does not induce specificity on these DPs, as has been claimed for Romance (cf.
Sportiche 1996; Uriagereka 1995, i.a.). It will be argued instead that the locus of
specificity is the D-position (cf. Abney 1987), which for noun phrases underlies
argumenthood (cf. Longobardi 1994). The view that direct object clitics in
Albanian and Greek mark the DPs they double as unambiguously [−Focus] may
be implemented successfully within the minimalist framework (cf. Chomsky 1995)
by preserving Sportiche’s (1996.) basic assumption that clitics head their own
maximal projections and that direct object clitics in particular are heads with
operator-like properties. Importantly, it will be argued that argument clitics carry
a D-feature, which is why they may double only DPs, not NPs,1 and that specific-
ity, presuppositionality and/or strength effects often attributed to clitic construc-
tions (cf. Sportiche 1996; Uriagereka 1995; Anagnostopoulou 1994 i.a.) are only
epiphenomenal, straightforwardly derived because of the feature-matching need.
This paper is organised as follows. I start out in Section 2 by outlining and
scrutinising the general properties of Albanian and Greek clitic doubling. This is

1. Here I depart from the view that an NP is exclusively a complement of D (cf. Abney 1987) and
more generally from the implication that once a functional projection is available at least within a
given language, it is always present/syntactically active in that language even though at times it may
be inert/morphologically empty (cf. Chomsky 1995). Note, however, that I am not claiming that the
D-position cannot be morphologically empty. For discussion, see sections 4.2 and 4.3 where I argue
that countable bare singulars and existential bare plurals are not DPs with a morphologically null D, but
NPs lacking a D-projection altogether. Consequently, they are not arguments, but predicates at LF. By
contrast, generic bare plurals are DPs with morphologically null Ds. The advantage of this distinction
between DPs and NPs is that it allows for a more principled mapping between syntax and semantics.
 DIRECT OBJECT CLITIC DOUBLING IN ALBANIAN AND GREEK 211

motivated by the need to comprehend the factors that are important for the so-
called Clitic Doubling Parameter. In Section 3 the interaction of focus and
doubling is discussed. Finally, Section 4 deals with matters of representation. In
this section I also investigate the parallels between doubling constructions in
Albanian and Greek and scrambling constructions in Germanic and discuss in
some detail the internal structure of noun phrases. In addition, I provide an
account of the phenomenon of specificity which rests on the individual vs.
property-denotation distinction.

2. Preliminaries

Albanian and Greek are so-called free word order, null subject languages with
rich morphology. Both languages have object pronominal clitics with distinct
morphological inflections for accusative and dative/genitive2 cases; both lack
subject clitics. In Greek, clitics follow only gerunds and imperatives. In Albanian
they may precede, follow or be infixed in imperatives. As in French, clitics in
both languages immediately precede all other verb forms both in matrix and
embedded clauses.3 The relative order of clitics is rigidly fixed for all combinations
of Person(s): dative/genitive followed by accusative. Clitic climbing is absent, as
are infinitives, which have historically been supplanted by the subjunctive form.4
Perhaps the most striking property of Albanian and Greek clitic doubling is
the fact that it violates Kayne’s generalisation which, informally stated, says that
clitic doubling is possible whenever a noun phrase can get case by means of
some non-verbal device which has case-assigning properties, namely, preposi-
tions.5 The Albanian and Greek examples below show that doubled DPs are not

2. Albanian and Greek have identical case systems except for the fact that the Greek counterpart of
the Albanian dative is the genitive.
3. For a detailed description of the positioning of clitics in several types of clauses in Albanian and
Greek, see Joseph and Philippaki-Warburton (1987), Rivero (1994), Rivero and Terzi (1994), Kallulli
(1995, 1997a).
4. For an analysis as to why clitic climbing is absent across all Balkan languages, see Terzi (1992).
5. Suñer (1988: 399–400) provides the following examples from Porteño Spanish as empirical
evidence against viewing the prepositional element a in Spanish, a language where Kayne’s
generalisation seems to be generally operative, as a case assigning device; she argues instead that a
is an animacy marker, which is why it is missing in the examples below in spite of the fact that the
object DPs are doubled here.
(i) Yo lo voy a comprar el diario justo antes de subir
I it am going to buy the newspaper just before coming up
212 DALINA KALLULLI

preceded by prepositions. In fact, prepositional objects may not be clitic doubled

in these languages.6
In Albanian, dative DPs are invariably clitic doubled. In (1a) this applies to
a definite expression, in (1b, c) to an indefinite expression, in (1d) to a wh-dative,
in (1e) to a quantified dative. The opposition (1b) vs. (1c) shows that dative clitic
doubling is insensitive to so-called ‘VP-internal scrambling of objects’ (cf.
Massey 1991).
(1) a. Ev-a *(i) dërgoi An-ës lule
Ev-the (*her. sent An-the. flowers
‘Ev sent Ann flowers’
b. Ben-i *(i) dërgoi një vajze lule
Ben-the (*her. sent a girl. flowers
‘Ben sent a girl flowers’
c. Ben-i *(i) dërgoi lule një vajze
Ben-the (*her. sent flowers a girl.
‘Ben sent a girl flowers’
d. Kujt *(i) foli mësues-i?
who. (*him/her. talked teacher-the
‘Who did the teacher talk to?’
e. Ben-i *(u) blen gjithë vajza-ve(t) lule
Ben-the (*them. buys all girls-(the) flowers
‘Ben buys all (the) girls flowers’
Both in Albanian and Greek, quirky subjects are invariably clitic doubled both when
marked for dative/genitive or accusative case. Examples are given in (2) and (3).
(2) a. Al: Jan-it *(i) mungojnë dhjetë libra7
Jan-the. *(him.. miss-they ten books.
b. Gr: Tu Yanni *(tu) lípun dheka vivlia
the Yannis. *(him.. miss-they ten books.
‘John is missing ten books’

(ii) Yo la tenia prevista esta muerte

I it had foreseen this death
(iii) Ahora tiene que seguir usndolo el apellido
now she has to go on using-it the surname
6. On the significance of violations of Kayne’s generalisation for the Clitic Doubling Parameter, cf.
Anagnostopoulou (1994).
7. Throughout the paper I use the symbols Al, Gr, Ge and Du as abbreviations for Albanian, Greek,
German and Dutch, respectively.
 DIRECT OBJECT CLITIC DOUBLING IN ALBANIAN AND GREEK 213

(3) a. Al: Ben-in *(e) mërzit vetmia

Ben-the. *(him.. bores solitude.
‘Solitude bores Ben’
b. Gr: Ton Yánni *(ton) ponái to kefáli tu
the Yánnis. *(him.. hurts the head. his
‘Yannis has a headache’
The examples in (4) instantiate clitic doubling of direct object DPs.8
(4) a. Al: *(E) pashë Jan-in9
b. Gr: *(Ton) idha ton Yánni
(him. saw-I the Yannis
‘I did see John’
As indicated by the English translation, (4a, b) cannot mean: ‘I saw John’
(uttered as out-of-the-blue sentences or as sentences in which either the whole
VP or the direct object DP is focused), which would be their meaning in the
absence of the doubling clitic. As such, (4a, b) are not felicitous answers to a
question like: ‘Who did you see?’, which they would be in the absence of the
doubling clitic. In other words, clitic doubling of direct object DPs in Albanian
and Greek is not an optional phenomenon, strictly speaking. For the moment, let
us just note this fact in passing; I will resume the discussion in detail in the next
section.
The examples in (5) show that unlike doubling in Romance, doubling of
direct objects in Albanian and Greek is not restricted to [+animate] or [+human]
DPs.10 Nor is it restricted to [+definite] DPs.
(5) a. Al: Do t-a pija me kënaqësi një uiski
fut-it. drink with pleasure a whisky
b. Gr: To pino eukharistos ena ouiskáki
it. drink with pleasure a whisky
‘I would gladly drink a whisky’

8. Albanian and Greek are pro-drop, null-subject languages and nothing stops clitics from appearing
sentence initially.
9. In Albanian the definite article is suffixed to the noun stem (indicated by the use of hyphens in
the Albanian examples); in Greek, like in English, it is a separate phonological entity and precedes
the noun stem.
10. For instance, doubling is sensitive to the feature human in Romanian and animacy in Spanish
(cf. Jaeggli 1986; Borer 1984; Suñer 1988; Dobrovie-Sorin 1990).
214 DALINA KALLULLI

It has been claimed for Greek that clitic doubling of direct object DPs is subject
to definiteness, in the sense that only definite DPs may be clitic doubled (cf.
Anagnostopoulou 1994).11 The example in (5b) (from Kazazis and Pentheroud-
akis (1976)) is then a counterexample to this claim since the doubled DP here is
clearly indefinite.12 This counterexample is in fact acknowledged by
Anagnostopoulou, who writes:
‘At first sight, sentences like [5b] seem to contradict the view that Modern
Greek doubling is subject to definiteness… Utterances like [5b] have a clear
modal reading, the verbal form used is subject to various aspectual restrictions
(imperfective aspect is systematically chosen: this type of aspect is typical of
conditionals) and the clitics in them seem to have a kind of “sentential”
function… These constructions are extremely interesting because the function
of the clitics in them is not clear. However, they are, in many respects,
different from the doubling constructions of the type examined here and, from
this point of view, beyond the scope of the present discussion… The fact that
the adverbial elements … must be heavily stressed and that they typically
precede the doubled DPs seems to indicate that structures like [5b] are right
dislocations. Furthermore, note that examples of this type are only possible in
“ordering-contexts” where … it is quite common to use attributive definites
instead of indefinites.’ (Anagnostopoulou 1994: 4, footnote 5)

Let me point out several inaccurate claims in the quote. First, doubled indefinite
DPs need not occur in constructions where the verb has imperfective aspect; the
Greek example in (6) contains a perfective aspect form. Secondly, adverbial
elements do not necessarily precede the indefinite DPs, as (6) also shows.
Thirdly, (6) shows that doubling of indefinite DPs is possible outside of ‘order-
ing-contexts’. Even if examples as in (5b) were only possible in ‘ordering-
contexts’, where it is claimed to be common to use attributive definites instead
of indefinites, doubling should still be unexpected for Anagnostopoulou, who
claims that attributive definites, as a subclass of novel definites, may not be clitic
doubled in Greek.13

11. Note, however, that the implication is only one way: definite direct object DPs may be doubled
but need not be. As not all definites can be clitic doubled in Greek (cf. e.g. (9b), (10b)), Anagnosto-
poulou tries to relate direct object clitic doubling in this language to Heim’s (1982) Familiarity
Condition. However, this analysis is untenable in the face of doubling of indefinites unless Heim’s
crucial claim that all indefinites represent novel information is rejected.
12. Cf. also Agouraki (1993) who provides several other parallel examples of doubling of indefinites.
13. Anagnostopoulou’s claim that attributive definites may not be clitic doubled in Greek is not
uncontroversial, though. In this context, according to Anagnostopoulou (1994), while clitics
 DIRECT OBJECT CLITIC DOUBLING IN ALBANIAN AND GREEK 215

(6) Akoma ke i Anna katafere na to ekdosi ena vivlio prin

still and the Anna managed  it. publish a book before
na pethani
 die
‘Even Anna managed to publish a book before she died’
Finally, I reject the idea that indefinite DPs may be clitic doubled only when
they are right dislocated, as has been claimed by Anagnostopoulou for Modern
Greek. The main argument against the view that clitic doubled indefinites are
exclusively right dislocated phrases comes from the fact that, just like clitic
doubled definite DPs, they may occur in both languages in positions that are
typically associated with theta marking and case marking, i.e., in A-positions
such as ECM complements and subjects of small clauses, as in (7) and (8).14
(7) a. Al: Jan-i e pret një gjë të tillë të ndodhë
Jan-the it. expects a thing such  happen
b. Gr: O Yannis to perimeni kati tetio na simvi
the John it. expects something such  happen
‘John expects something like this to happen’
c. Al: Jan-i e pret Mer-in të ankohet
Jan-the her. expects Mary-the.  complain
d. Gr: O Yannis tin perimeni [tin Maria na paraponethi
the John her. expects [the Mary].  complain
‘John expects Mary to complain’
(8) a. Al: Jan-i nuk e konsideron një vajzë të tillë/
Jan-the not her. consider a girl such/
Mer-in inteligjente
Mary-the. intelligent
b. Gr: O Yannis dhen tin theori kamja tetia kopela/
the John not her. consider no such girl
tin Maria eksipni
[the Mary]. intelligent
‘John does not consider any such girl/Mary intelligent’

necessarily license familiarity on the direct object DPs they double, these DPs may be either novel
or familiar if not doubled. This is clearly imperfect, as clitic doubling emerges in her treatment not
only as a totally optional but also as an entirely redundant phenomenon if clitics may double definite
DPs which are non-novel/familiar even when not doubled.
14. An additional argument against the right dislocation hypothesis is presented in Section 3.3.
216 DALINA KALLULLI

It is thus my contention that clitic doubling constructions of the type in (5b) do

not differ from the doubling constructions involving doubling of definite direct
object DPs other than with respect to the definiteness feature, which is irrele-
vant. The factors that determine clitic doubling of direct object DPs are the
same, irrespective of the [±definite] status of these DPs. In this way clitic
doubling of direct object DPs emerges as a uniform phenomenon and should be
treated as such.
The fact that both definite and indefinite direct object DPs may be doubled
does not mean that they always can be. The data in (9) show that even definite
DPs cannot be doubled invariably.
(9) A: Do you walk to school or do you take the bus?
B: a. Al: Nuk shkoj në këmbë, (*e) marr autobus-in
b. Gr: Dhen pigheno me ta podhja, (*to) perno to leoforio
not walk with feet (*it. take the bus
‘I don’t walk, I take the bus (to school)’
The fact that the definite DPs in these examples cannot be doubled is problemat-
ic for the specificity/presuppositionality/familiarity/d-linking/strength approaches
to doubling (cf. Sportiche 1996; Anagnostopoulou 1994; Uriagereka 1995), if we
assume with Enç (1991) and Diesing (1992) that all definites are specific/
presuppositional/strong. While the claim that all definites are specific will be
challenged (cf. Section 4.3.4), there are indisputably specific/presuppositional/
strong definites (and indefinites) that need not and/or cannot be doubled. (10) is
a case in point.
(10) A: What happened?
B: a. Al: Jan-i (#i) hëngri fasule-t/ (#e) piu
b. Gr: O Yánnis (#ta) éfaye ta fasólia/ (#tin) ipje
the Yannis (#them. ate the beans/ her. drank
një birrë
mia bira
a beer.
‘Yannis ate the beans/drank a beer’
Finally, referentiality of the doubled DP is also irrelevant for direct object clitic
doubling in Albanian and Greek. This is indicated by the fact that quantified
 DIRECT OBJECT CLITIC DOUBLING IN ALBANIAN AND GREEK 217

expressions may also be doubled, as in (11).15 Note that doubling in Albanian

and Greek does not suppress the attributive reading of definite DPs; the doubled
DP in (12) may receive both a referential and a non-referential/attributive
interpretation.16
(11) a. Al: An-a i urrente të gjithë djem-të
b. Gr: I Anna ta misouse ola ta aghorja
the Ann them.  all the boys
‘Anna  all the boys’
(12) a. Al: I dua mace-t e vogla
them.  cats-the small
b. Gr: Tis aghapos tis mikres (tis) ghates
them.  the small (the cats
‘I  small cats’
Now that we have seen that clitic doubling of direct object DPs in Albanian and
Greek cannot be adequately described in terms of any of the features highlighted
so far by various theorists as significant for the Clitic Doubling Parameter, let
me turn to the identification of the factors determining direct object clitic
doubling in Albanian and Greek.

3. The Non-optionality of direct object clitic doubling

A variety of facts converge in showing that clitic doubling of direct object DPs
systematically yields ungrammaticality when these DPs are in focus or part of
the focus domain, that is, when they are marked [+Focus].17 In this section, I
show that clitic doubling of direct object DPs in Albanian and Greek unambigu-
ously marks these DPs [−Focus].

15. However, clitic doubling in Albanian and Greek is incompatible with focus DPs, as will become
clear in Section 3. According to the view that any constituent that can be raised by QR can serve as
focus (cf. Chomsky 1976), quantifiers in general are default foci. In (11) I have tried to control this
factor by focusing the verb. This is indicated in the English translation by the use of small capital
letters. The interaction of clitic doubling and focusing will be discussed at length in Section 3.
16. Anagnostopoulou (1994) claims that doubled DPs in Greek may only receive a referential
interpretation. A. Androutsopoulou (personal communication), however, points out to me that the
doubled DP in (12b) can receive an attributive interpretation (e.g. when the verb is focused).
17. Except where indicated otherwise, I will only be concerned with doubling of direct objects, not
of accusative quirky subjects.
218 DALINA KALLULLI

3.1 Justifying [ Focus]

In defining the feature [−Focus], I will proceed indirectly by defining the notion
focus first.
Informally speaking, focus is viewed as the most informative part of an
utterance. Hence, any definition of focus is sensitive to the speech act and varies
according to it. For instance, the notion information or information structure for
a question does not make sense unless one defines information structure as the
type of answer one expects (cf. Sperber and Wilson 1988). So, for wh-questions,
focus is the variable represented by the wh-element; this also holds for
echo-questions. For a yes-no question focus is either the assertion (i.e. the given
polarity), or the negation (i.e. the opposite polarity). Focus can also be an
element which is contrasted. Finally, focus can be the item that fills in a slot in
an information structure where other slots have already been filled. In this latter
function, focus is close to the notion ‘new information’. The definitions above
are quasi-collectively reflected in the following quote from Vallduví (1994: 575):
‘… focus, an informative, news-bearing, dominant, or contrary-to-expectation
part…’.18 The complement of focus is topic. Following a long-established
tradition in generative grammar, I assume that focus is a syntactic feature on
phrases interpretable at both the LF and the PF interfaces as [+Focus] (cf.
Jackendoff 1972; Rochemont 1986; Horvath 1986; Brody 1990).
In view of the fact that a sentence may lack a topic (e.g. out-of-the-blue
sentences) but will always have a focus, I assume that the [+Focus] feature is in
fact the unmarked value in a markedness theory for natural language and that the
[−Focus] feature is the marked value. Derivational syntax then renders this
feature significant. I argue that clitic doubling is one of the means by which this
feature gets licensed.
Consider the examples in (13).
(13) a. Al: An-a lexoi libr-in
b. Gr: I Ana diavase to vivlio
the Ana read the book
‘Ann read the book’
The undoubled Albanian example (13a) is a felicitous answer to either (14a) or
to (14b), but not to (14c) or (14d).

18. For details on the formalisation of focus (i.e. its formal representation in lambda-reduced
intensional logic), see Jacobs (1986), Rooth (1996), Krifka (1996).
 DIRECT OBJECT CLITIC DOUBLING IN ALBANIAN AND GREEK 219

(14) a. What did Ana do?

b. What did Ana read?
c. Who read the book?
d. What did Ana do to/with the book?
The Greek example (13b) may be a felicitous answer to either of the questions
in (14). However, (15b), the doubled version of (13b), is preferred as an answer
to (14c) and (14d) even in Greek. Crucially, (15a, b), the doubled version of
(13a, b), may in both languages only be a felicitous reply to (14c, d) but not to
(14a, b). This latter fact suggests that direct object clitic doubling in Albanian and
Greek is incompatible with direct object DPs that are marked [+Focus] (or
alternatively, are contained in focus domains).
(15) a. Al: An-a e lexoi libr-in
b. Gr: I Ana to diavase to vivlio
the Ana it. read the book
‘Ann did read the book’/‘ read the book’
I devote the next two sub sections to a brief review of some recent ideas on the
syntactic encoding of focus and to how focus interacts with clitic doubling of
direct object DPs in Albanian and Greek.

3.2 Focus, wh-elements and clitic doubling

Following Horvath (1988), Brody (1990) assumes that just like there is a feature
[+wh], which marks phrases as wh-elements, there is a feature [+f ] that indicates
focushood;19 wh-phrases are argued to be necessarily [+f ] and the conditions on
[+wh] and (+WH) CPs are generalised to the [+f ] and F(ocus) P(hrase) so that
they will entail (16), which may be regarded as a Focus Criterion. The unavoid-
able implication is that Rizzi’s (1991) Wh-criterion is a subcase of the Focus
Criterion and that Spec of (a root) CP is one of the canonical positions for focus.

19. Brody claims that the S-structure presence of the [+f] feature shows up as heavy stress at PF.
According to him, the stressed [+f]-marked category is not necessarily the same as the [+f]-phrase,
but the [+f]-phrase will always contain a [+f]-marked element. While he does not define the notion
of ‘heavy stress’, I take it to be phonetic prominence, probably indicated by a pitch accent. Unlike
Brody, I wish to leave open the possibility that focus may have other PF correlates even if phonetic
prominence/pitch accent is absent.
220 DALINA KALLULLI

(16) (Brody 1990: 208)

a. At S-structure and LF the Spec of an FP must contain a [+f]-
phrase
b. At LF all [+f]-phrases must be in an FP
Just as the corresponding notion on +WH CPs is parametrised, it is assumed that
(16a) may or may not hold in a given language. (16b), on the other hand, like
the condition on wh-elements, should be universal. Further, a distinction is drawn
between + and −WH FPs.
If my claim is correct that direct object clitics license non-focusing of the
DPs they double and if we assume with Brody (1990) that wh-elements are
necessarily foci, then clitic doubling of wh direct object DPs in Albanian and
Greek is bound to yield ungrammaticality. The examples in (17) show that this
is indeed the case.
(17) a. Al: Kë/çfarë (*e) pe?
[who/what]. it/him/her. saw-you
b. Gr: Pjon/ti (*ton/to) idhes?
[who/what]. him/it. saw-you
‘Who/what did you see?’
Direct object DPs in Albanian are obligatorily clitic doubled in constructions
with wh-subjects, as well as in yes/no questions, as shown in (18) and (19),
respectively. Similar facts are reported for Greek by Agouraki (1993), who notes
that in questions, either yes/no or wh-questions, a doubling clitic is strongly
preferred.20 These facts are also predicted under the hypothesis that clitic
doubling exempts direct object DPs from focus domains (that is, from phrases
that are marked [+Focus]).
(18) a. Al: Kush *(e) pa fëmijë-n?
b. Gr: Pios *(to) ídhe to pedhí? (Agouraki 1993: 154)
who it. saw the child
‘Who has seen the child?’

20. At this point, it should be clear that direct object clitic doubling is somehow less strict in Greek
than in Albanian since only in the latter does it obligatorily occur whenever the direct object DP is
outside the focus domain. The fundamental point to note, however, is that in both languages direct
object clitic doubling indisputably marks the direct object DP as [−Focus]. In other words, while
doubling of direct object DPs in Albanian and Greek necessarily marks these DPs as [−Focus], it is
not the case that for the direct object DP to be interpreted as [−Focus], it has to be clitic doubled
(e.g. Greek).
 DIRECT OBJECT CLITIC DOUBLING IN ALBANIAN AND GREEK 221

(19) a. Al: (A) *(e) pe Jan-in?

[+]21 him. saw Jan-the
b. Gr: (Ton) idhes ton Yánni? (Agouraki 1993: 170)
(him. saw the John
‘Have you seen John?’
In (20a) and (20b) the whole VP is contrastively focused. Since the direct object
here is part of the focus domain (i.e. is marked [+Focus]), it cannot be doubled.22
(20) a. Al: An-a nuk (*i) zjeu fasule-t, por
b. Gr: I Anna dhen (*ta) mayirepse ta fasólia, alá
the Ann not (*them. cooked the beans but
(*i) hëngri fiq-të
(*ta) éfaye ta sika
(*them. ate the figs
‘Anna didn’t [cook the beans]F; she [ate the figs]F’
Likewise, direct object DPs in out-of-the-blue sentences may not be doubled, as
the examples in (21) show.23
(21) A: What happened here?
B: a. Ben-i (*e) ka thyer termometr-in/një pjatë (Al)
Ben-the (*it. has broken thermometer-the/a plate
‘[Ben has broken the thermometer/a plate]F’
b. O Yánnis (*ta) éfaye ta fasólia/ (*tin) ipje
the Yannis (*them. ate the beans/ (*her. drank
mia bira (Gr)
a beer.
‘[Yannis ate the beans/drank a beer]F’
Focus (i.e. a [+Focus] phrase) is most clearly brought out in association with so-
called focus particles, such as even and only, otherwise referred to as scalar
particles by Jacobs (1984), or as focusing adverbs by Rooth (1996). In the next

21. Albanian has an optional question particle for yes/no questions.

22. The sentences in (20) are grammatical also when the direct object (in the first conjunct) is clitic
doubled under an interpretation which can be roughly rendered in English as: ‘As for Anna and the
beans, she didn’t cook them, rather she ate the figs’. But notice that under this interpretation, ‘the
beans’ is indisputably outside the focus domain. Hence, doubling exempts the direct object from the
focus domain.
23. (21) is analogous to (10).
222 DALINA KALLULLI

section, I use this diagnostic to identify [+Focus] phrases and investigate the
effects of their interaction with direct object clitic doubling.

3.3 More [+Focus] phrases and their interaction with doubling

In the examples in (22a, b) the direct object DP Tiranën is a [+Focus] phrase, as

the English translation indicates.24 As such, it cannot be clitic doubled either in
Albanian or in Greek.
(22) a. Al: Pap-a (*e) vizitoi madje Tiranë-n (jo vetëm Shkodrën)
Pope-the it. visited even Tirana-the (not only Shkodra)
b. Gr: O Papas (*ta) episkeftike akoma ke ta Tirana…
the Pope them. visited still and the Tirana
‘The Pope visited even [Tirana]F (not only Shkodra)’
Likewise, the direct object DP in (23a, b) cannot be clitic doubled, since it is
marked [+Focus]. The fact that the direct object DP ‘a beer’ in (23) may not be
clitic doubled is not related to its being [−definite]; the examples in (24) show
that in both languages constructions involving doubled indefinites are fully
grammatical if (and only if) the direct object is construed as outside the focus
domain, a point which was already made earlier in the discussion.25
(23) a. Al: Jan-i (*e) piu madje një birrë para se të
Jan-the (*it. drank even a beer before that 
shkonte
went
b. Gr: O Yánnis (*tin) ipje akoma ke mja bira prin
the Yannis (*her. drank still and a beer. before
na fighi.
 went
‘John drank even [a beer]F before he left’

24. In Albanian, focusing adverbs can attach to different sites without necessarily affecting the
interpretation of phrases in terms of the [±Focus] feature. That is, unlike in English, it is not
necessarily the constituent which the focus particle immediately precedes that constitutes the focus
domain. Because of this complexity, I provide the intended interpretation in the English translations
of the Albanian and Greek examples by employing square brackets followed by the subscript ‘F’ (to
indicate focus domains).
25. Again, in Albanian, clitic doubling of direct object DPs is obligatory when the object is outside
the focus domain. A. Androutsopoulou (personal communication) points out that clitic doubling of
the object when the direct object is outside the focus domain is optional in Greek; however, she notes
that (24b) and (25b) are strongly preferred with the doubling clitics.
 DIRECT OBJECT CLITIC DOUBLING IN ALBANIAN AND GREEK 223

(24) a. Al: Jan-i *(e) piu madje një birrë para se të

Jan-the *(it. drank even a beer before that 
shkonte (jo vetëm e porositi)
went (not only it. ordered
‘John even [drank]F a beer before he left (not only did he
order it)’
b. Gr: O Yánnis ?(tin) IPJE mja bira prin na
the Yannis ?(her. drank a beer. before 
fighi…
went…
‘John [DID drink] a beer before he left (he didn’t just
order it)’
Similarly, the clitic doubled versions of the sentences in (22) are grammatical
under an interpretation in which the direct objects are construed outside the focus
domain; in these cases, doubling is indeed obligatory in Albanian. This is shown
in (25).
(25) a. Al: Pap-a *(e) vizitoi madje Tiranë-n.
Pope-the *(it. visited even Tirana-the
b. Gr: O Papas os ke ?(ta) episkeftike ta Tirana
the Pope till and ?(them. visited the Tirana
‘The Pope even [visited]F Tirana’ i.e. ‘As for Tirana, the
Pope even visited it’
The clitic doubled versions of the objects in (22) also become grammatical if the
subject DP is marked [+Focus], a fact which is indicated in the examples in (26)
by the focus particles in front of the subject DP.26
(26) a. Al: Madje Pap-a *(e) vizitoi Tiranë-n
even Pope-the *(it. visited Tirana-the
b. Gr: Akoma ke o Papas (ta) episkeftike ta Tirana
still and the Pope (them. visited the Tirana
‘Even [the Pope]F visited Tirana’
The data thus systematically reveal that clitic doubling of direct object DPs that
are marked [+Focus] or are contained in [+Focus] phrases is disallowed in
Albanian and Greek. The question then arises as to whether the function of direct

26. In fact, as the notation in (26) indicates, clitic doubling of the direct object DP is obligatory in
Albanian when the subject is focus; in Greek, however, clitic doubling of the direct object DP is only
optional when the subject is focus.
224 DALINA KALLULLI

object doubling clitics is to license verb/subject-focusing or object non-focusing.

The fact that verb/subject focusing may still be achieved in intransitive construc-
tions decides the issue in favour of the latter alternative.
In sum, we may state that clitic doubling of direct object DPs in Albanian
and Greek is not optional: [+Focus] DPs cannot be clitic doubled. Thus, direct
object clitics in Albanian and Greek have interpretive import; they mark the DPs
they double as unambiguously [−Focus], which is interpreted as an operator
feature.27 In this respect, direct object clitic doubling in Albanian and Greek is
different from clitic doubling in Spanish, which does not necessitate a [−Focus]
reading (cf. e.g. (27) from Porteño Spanish), but is strongly reminiscent of so-
called clitic right dislocation structures in French, Spanish and Italian, which are
incompatible with [+Focus] phrases.28
(27) La nombraron a Maria. (Suñer 1988: 419)
her they-nominated a Maria
‘They nominated MARIA’
However, the fact that clitic doubled DPs in Albanian and Greek may occur in
positions where adjuncts are simply not tolerated, as was shown in Section 2 (cf.
e.g. (7) and (8)), ultimately rules out a right dislocation approach to these
constructions. Further evidence can be adduced to this effect. For instance, if the
doubled direct object DPs in Albanian were indeed right-dislocated, a [+Focus]
phrase to the right of a right dislocated direct object would be precluded. This
prediction is not borne out, however, as the example in (28) demonstrates.

27. This feature (i.e. [−Focus]) could alternatively be represented formally as [+Topic]. Recall that
in Section 3.1 I defined topic as the complement of focus, not as necessarily old/familiar information.
In this context, see also Reinhart (1982, 1995), who crucially points out that defining topic as old/
familiar information in keeping with the Prague school is not only conceptually clumsy, but also
empirically incorrect. In view of the fact that topic is the counterpart of focus, it makes little
difference whether we choose to represent it formally as [−Focus] or as [+Topic]. For the sake of
symmetry in representation, however, the postulate of one binary feature (here: [±Focus]) might be
preferable. Hence my choice of label: [−Focus]. As Reinhart remarks, ‘even in view of the massive
varieties of opinions regarding what topics are, there is one context all studies agree upon: the NP in
there-sentences can never be topic…’ (Reinhart 1995). We thus expect that objects of the verb ‘to
have’ may not be clitic doubled in Albanian and Greek existential constructions. This is indeed the
case, as shown by the examples under (i) and (ii) below:
(i) Al: (*I) kishte minj në gjithë apartament-in
(ii) Gr: (*Ta) ixe pontikia se olo to diamerisma
(*them. had mice. in all the apartment
‘There were mice all over the apartment’
28. I thank the anonymous reviewer for pointing this out.
 DIRECT OBJECT CLITIC DOUBLING IN ALBANIAN AND GREEK 225

(28) Al: I-a dhashë libr-in BEN-it.

him.-it. gave book-the. BEN-the.
‘I gave the book to BEN’ i.e. ‘It was BEN that I gave the book to’
The question then remains whether the Albanian and Greek doubling construc-
tions constitute yet a third type of clitic constructions with properties distinct
from those of the two others–that is, clitic doubling constructions in Spanish/
Romanian on the one hand and clitic right dislocation constructions in Romance
on the other–or whether it can subsume, or be subsumed, under either of the
two. To address this question one has to look at all the properties of the other
two constructions in detail, as well. Such a task is well beyond the scope of the
study at hand.
Direct object clitic doubling in Albanian and Greek is also strongly reminis-
cent of scrambling of direct objects in Germanic (cf. Webelhuth 1989), as will
be more closely discussed in Section 4.2.

3.4 Summary

In concluding this section, it may be stated that direct object doubling clitics in
Albanian and Greek are characterised by the fact that they have (i) a restricted
distribution, and (ii) operator-like properties. These two properties suggest that
direct object doubling clitics in these languages cannot be treated as mere object
agreement markers, that is, as spell-outs of e.g. AgrO heads. Yet, there is little
doubt that clitic doubling is a form of agreement between an X0 and an XP,
namely the clitic head and the DP it doubles and with which it agrees in phi-
features. The next section is devoted to how this cluster of properties can best be
represented.

4. Issues of representation

4.1 Spec–Head licensing, feature checking and doubling

The view that accusative clitics mark the DPs they double as [−Focus] may be
implemented formally in terms of the theory of spec-head licensing (cf. Chomsky
1995), if we assume with Sportiche (1996) that a clitic heads its own maximal
projection in whose specifier position it licenses a particular property/feature
226 DALINA KALLULLI

F.29 For the derivation to converge, this feature has to be saturated or checked
off (cf. Chomsky 1995). Since features may only be checked off in spec-head
configurations, the (doubled argument) XP* in (29) must by LF move to the XP^
position so as to establish the relevant spec-head configuration.
(29) ClP

XP^ Cl

Cl0 VP

V0 XP*

In Sportiche’s terms, movement of XP* to the XP^ position is motivated by the

Clitic Criterion, an analogue of Rizzi’s (1991) wh-Criterion, and yet another
instantiation of the so-called Generalised Licensing Criterion, according to which
feature-licensing may only obtain in spec-head configurations.
Further, Sportiche (1996) sets the following parameters of clitic constructions:
(30) Clitic construction parameters (Sportiche 1996: 36)
(i) Movement of XP* to XP^ occurs overtly or covertly
(ii) H is overt or covert
(iii) XP* is overt or covert
By these parameters, the cases in (31), among others, are predicted.
(31) a. Clitic doubling constructions (as in Spanish, Romanian, Greek,
Albanian) arise when an overt XP* moves covertly with an
overt Cl.
b. Scrambling in Dutch/German arises when an overt XP* moves
overtly with a covert Cl.

29. With respect to the property they license, according to Sportiche, clitics subdivide into two types.
The first type (typically accusative clitics) assimilates to such functional heads as [+wh] complement-
isers or [+negative] heads, which license some operator-like properties (e.g. wh or negative
quantifiers). Sportiche argues that the operator-like property these clitics license is specificity in DPs.
The second type of clitics (typically nominative and dative Romance clitics) is claimed not to be
linked to specificity. Concerning this second type of clitics, Sportiche suggests that they should be
analysed as pure agreement markers, that is, as elements devoid of interpretive import, presumably
responsible for dative case assignment (i.e. AgrIO-heads in the sense of Chomsky 1995).
 DIRECT OBJECT CLITIC DOUBLING IN ALBANIAN AND GREEK 227

As for direct object clitic constructions, Sportiche claims that the property which
the clitic head licenses in the specifier of the phrase it heads is invariably
specificity, irrespective of whether the direct object clitic is overt (as in doubling
constructions) or covert (as in scrambling constructions). As discussed above,
this cannot possibly be the case for Albanian and Greek direct object clitics. The
feature that Albanian and Greek direct object clitics license in the specifier of the
phrase they head is what was defined in Section 3 as [−Focus]. By the theory of
spec-head licensing, for the derivation to converge, the feature values on the
clitic head and those of the DP in its specifier must match. Since the attracting
feature is [−Focus], a clitic doubled [+Focus] direct object DP would invariably
cause the derivation to crash. In this way, doubling of [+Focus] direct object DPs
is of necessity ungrammatical.
While the idea that the same syntactic configuration underlies both doubling
and scrambling constructions is desirable conceptually and attractive theoretically
(cf. Chomsky 1995), I argue that the property F, whose need to be licensed
motivates the postulated maximal projections (that is, Sportiche’s ClP(s) or Voice
Phrases), is identified incorrectly by Sportiche. In Section 3, I demonstrated that
the feature that Albanian and Greek direct object clitics license on the DP they
double is not specificity but topichood. In the next section I show that this is also
the case for Germanic scrambling.30

4.2 Parallels with scrambling

Like doubling of direct objects in Albanian and Greek, scrambling of direct

objects in Germanic applies both to definite DPs as well as to a-expressions.31
That is, the [±definite] feature of the DP is not relevant for scrambling. This is
illustrated in (32b), (33b) and (34b).
(32) a. Ge: Anna hat gestern das Buch gelesen
Anna has yesterday the book read

30. The idea that focus is involved in scrambling phenomena is extensively discussed in Reinhart
(1995). While Reinhart argues that a scrambled constituent cannot be focus, she favours a PF
approach to focus (cf. Cinque 1993), which crucially involves the notion of stress prominence.
However, as stated in note 19, I wish to leave open the possibility that the syntactic feature focus
may have PF correlates that are different from (and perhaps exclude) stress prominence. Therefore,
I will not undertake to present Reinhart’s account.
31. Throughout, I use the term a-expression (cf. Chastain 1975) to refer to non-quantified singular
indefinite noun phrases with articles.
228 DALINA KALLULLI

b. Ge: Anna hat das Buch gestern gelesen

Anna has the book yesterday read
‘Ann read the book yesterday’
(33) a. Ge: Ich habe gestern eine Zeitung gelesen.
I have yesterday a newspaper read
b. Ge: Ich habe eine Zeitung gestern gelesen.
I have a newspaper yesterday read
‘I read a newspaper yesterday’
(34) a. Du: dat de politie gisteren een kraker opgepakt heeft
that the police yesterday a squatter arrested has
(de Hoop 1992: 50)
b. Du: dat de politie een kraker gisteren opgepakt heeft
that the police a squatter yesterday arrested has
(de Hoop 1992: 50)
It was shown in Section 2 that definite direct object DPs cannot always be
doubled. The data in (35) and (36) show that they cannot always scramble either.
This fact is problematic for the specificity/presuppositionality/strength related
approaches to scrambling (cf. Sportiche (1996), Diesing (1992), de Hoop (1992)
i.a.) if we assume with Enç (1991) and Diesing (1992) that all definites are
specific/presuppositional/strong. While the claim that all definites are specific
will be challenged (cf. Section 4.3.4), there are unequivocally specific/presup-
positional/strong definites (and indefinites) that cannot scramble (cf. e.g. (37)).
Hence, scrambling emerges even in these analyses as an optional phenomenon.
(35) Ge: Er sagte, daß er nicht zu Fuß in die Schule geht, sondern
he said that he not on feet in the school walks but
a. daß er immer den Bus nimmt
that he always the bus takes
‘He said that he doesn’t walk to school but always takes
the bus’
b. *daß er den Bus immer nimmt
that he the bus always takes
(36) Du: a. dat ik altijd de bus neem (Reinhart 1996: 4)
that I always the bus take
b. *dat ik de bus altijd neem
that I the bus always take
 DIRECT OBJECT CLITIC DOUBLING IN ALBANIAN AND GREEK 229

(37) A: What happened?

B: a. Ge: Hans hat heute das Thermometer/einen Teller
Hans has today the thermometer/a plate
zerbrochen
broken
#
b. Ge: Hans hat das Thermometer/einen Teller heute
Hans has the thermometer/a plate today
zerbrochen
broken
‘Hans broke the thermometer/a plate today’
It is easy to notice in the (grammatical) examples above that the direct object
DPs are marked [+Focus]. I propose that this is why these DPs cannot undergo
scrambling.32 Further evidence that can be adduced to this effect is the fact
that +wh direct object DPs cannot scramble, as (38) shows.
(38) (Sternefeld 1990)
a. Ge: Wem hat der Student welche Frage beantwortet?
whom has the student which question answered?
‘To whom did the student answer which question?’
b. Ge:*Wem hat welche Frage der Student beantwortet?
whom has which question the student answered
Now consider the German examples in (39).
(39) A: Hat der Papst Tirana endlich besucht?
has the Pope Tirana finally visited
‘Did the Pope finally visit Tirana?’

32. The anonymous reviewer points out that scrambled noun phrases may have contrastive focus, as
in the Dutch example below:
(i) Ik heb slechts EEN van de boeken nog niet gelezen
I have only ONE of the books yet not read
Here the DP ‘the books’ is marked [−Focus], but ‘one’ is [+Focus]. However, in Albanian and Greek
contrastively focused direct object DPs are incompatible with doubling. The reason as to why the
parallel between scrambling and doubling breaks down when contrastive focus is involved is not
entirely clear to me. It might be stipulated, though, that contrastive focus is fundamentally correlated
with stress prominence at PF. However, since clitics are incompatible with PF stress (i.e. marked
[−stress]), the derivation crashes because of value divergence with respect to PF stress. The non-overt
clitic head in the case of scrambling might, however, be totally underspecified for the PF stress value;
as such, a [+stress] element moved to its specifier position in the syntax will not render the derivation
illicit at PF.
230 DALINA KALLULLI

B: a. Der Papst hat Tirana noch immer nicht besucht

the Pope has Tirana yet always not visited
‘The Pope has not visited Tirana yet’
b. #Der Papst hat noch immer nicht Tirana besucht
the Pope has yet always not Tirana visited
‘The Pope has not visited Tirana yet’
The examples in (39) show that scrambling of direct objects is obligatory in
answers to yes/no questions. This fact can be accounted for in a straightforward
manner under the hypothesis that scrambling of direct object DPs licenses a
[−Focus] feature on these phrases. As was pointed out in Section 3.1, for yes/no
questions (and answers to yes/no questions) focus is either the assertion or the
negation of the event expressed by the verb, whereas direct object arguments are
outside the focus domain. i.e. , they are not marked [+Focus]. Consequently,
there is no feature clash between the (covert) clitic head and the scrambled DP
in the specifier of the ClP in the diagram in (29) with respect to the feature
[±Focus]. Therefore the derivation will converge (provided that the covert clitic
head and the XP* do not show a mismatch with respect to other features). Note
that the specificity/presuppositionality/strength approaches to scrambling cannot
account for the fact that scrambling of direct objects in answers to yes/no
questions is obligatory, since ‘Tirana’ as a proper noun is referentially specific
also in the unscrambled version.
While definite and indefinite DPs with overt determiners may be doubled
and scrambled, bare indefinites cannot. For bare plurals this is shown in (40);
doubled and scrambled bare plurals are ungrammatical in any context.33 The
sentences in (41a–c) show that this also holds for countable bare singular direct
objects.34

33. In fact, this claim only holds for those bare plurals that receive an existential interpretation. This
is explained in Section 4.3.2.
34. As it happens, even closely-related languages differ with respect to the possibility of instantiat-
ing their direct objects by countable bare singulars. Thus, while countable bare singulars are virtually
non-existent as direct objects in English, across Balkan and Mainland Scandinavian languages they
may occur as direct objects of all predicates whose bare plural direct objects cannot get a generic
(either referential/kind-denoting or quantificational) interpretation but get an existential interpretation
instead. In German, on the other hand, countable bare singulars do occur as direct objects, but are
much more restricted than in Balkan and Mainland Scandinavian. Note in this context that of all the
languages mentioned above, only English disallows countable bare singulars in predicate nominal
position. Finally, note that countable bare singulars are found also in English as objects of certain
prepositions; e.g. go to school/church/market; travel by train/plane etc.
 DIRECT OBJECT CLITIC DOUBLING IN ALBANIAN AND GREEK 231

(40) a. Al: An-a nuk (*i) zjeu fasule, por

b. Gr: I Anna dhen (*ta) mayirepse fasólia, alá
the Ann not (*them. boiled beans, but
Al: (*i) hëngri fiq
Gr: (*ta) éfaye sika
(*them. ate figs
c. Ge: Anna hat nicht Bohnen gekocht, sondern sie hat Feigen
Anna has not beans boiled but she has figs
gegessen
eaten
d. Ge:*Anna hat Bohnen nicht gekocht, sondern sie hat Feigen
Anna has beans not cooked but she has figs
gegessen
eaten
‘Anna didn’t [cook beans]F but [ate figs]F’
(41) a. Al: An-a donte t-(*a) blente fustan
b. Gr: I Anna ithele na (*to) aghorasi forema
the Ann wanted -it. buy dress
‘Anna wanted to buy a dress’
c. Ge: Ich habe (*Zeitung) nicht/im Garten (Zeitung)
I have (*newspaper not/in the garden (newspaper
gelesen
read
‘I have not read a newspaper’/‘I have read a paper in the
garden’
vs.
a′. Al: An-a donte t-(a) blente një fustan
b′. Gr: I Anna ithele na (to) aghorasi ena forema
the Ann wanted -it. buy a dress
‘Anna wanted to buy a dress’
c′. Ge: Ich habe (eine Zeitung) nicht/im Garten (eine
I have (a newspaper not/in the garden (a
Zeitung) gelesen
newspaper read
‘I have not read a paper’/‘I have read a paper in the garden’
(41a, b) are ungrammatical when the bare singular objects are doubled in spite of
the fact that the clitics and the direct object bare singulars agree in phi-features
(that is, in number, person and gender, since bare singulars, like a-expressions,
232 DALINA KALLULLI

are not marked for morphological case in Albanian and Greek).

The question arises as to why bare indefinites cannot be doubled/scrambled.
I will approach this question by considering first why bare singulars cannot be
doubled/scrambled.35 To the extent that this question has been addressed at all,
bare singulars have been treated as forming a complex predicate with the clausal
predicate (cf. Haiden 1996), that is, as incorporating semantically. While this
seems intuitively correct, the fact that countable bare singulars need not be
adjacent to the clausal predicate but may be moved to Spec of CP, as in (42),
shows that this semantic incorporation does not result from syntactic incorpora-
tion of the bare singular into V.36
(42) a. Al: Fustan doja të bleja
dress wanted  buy
‘It was a dress that I wanted to buy’
b. Ge: Zeitung habe ich gestern gelesen
newspaper have I yesterday read
‘It was a newspaper that I read yesterday’
I propose that the impossibility of doubling and scrambling bare singulars is due
to feature mismatch between the clitic head and the direct object bare singular
with respect to the D-feature: while clitics carry a D-feature (cf. Emonds (1992),
Uriagereka (1995)), bare singulars are NPs that altogether lack a D-projection.
Clitics are listed in the lexicon as separate morphophonological units. That clitics
carry a D-feature (alternatively: are specified in the lexicon as elements of
category D0 or are underlying determiners (cf. Postal 1969; Raposo 1997) is not
surprising, in view of the fact that they originate from personal and demonstra-
tive pronouns which are prototypical D-heads (cf. Abney 1987 and subsequent
literature). This means among other things that only DPs but not NPs may be
doubled and scrambled, since the [−D] feature of the latter will clash with the
[+D] feature on the clitic head, thus causing the derivation not to converge. This
reasoning, however, rests on the assumption that bare singulars are NPs lacking
a D-projection. This is problematical, as it seems to run counter to Longobardi’s

35. The relation between bare singulars and bare plurals is discussed in detail in Section 4.3.2.
36. Here I am not implying that if a constituent occurs clause-initially it necessarily occupies the
Spec of CP. I am only assuming with Brody (1990) that Spec of CP is one (of the) canonical
position(s) for [+Focus] phrases and since the fronted constituents in (42) are indisputably [+Focus],
it makes sense to assume that they occupy precisely this slot. However, I remain open to the idea that
above there is the CP-node a projection headed by some operator which licenses D-linking in its
specifier position (cf. Pesetsky 1987).
 DIRECT OBJECT CLITIC DOUBLING IN ALBANIAN AND GREEK 233

(1994) claim that only DPs but not NPs may function as arguments, his idea
being that bare noun objects have a morphologically null D-head. Therefore, the
assumption that bare singulars are NPs and not DPs with a morphologically null
D is in need of some justification. Is there any evidence that legitimises the claim
that bare singulars lack a D-projection? In what follows, I will argue that there is.
First, note that bare singulars occur only as predicate nominals and as direct
objects. Crucially, they cannot occur as subjects.37 Further, bare singulars do
not occur as direct objects of just any predicate; they may occur as direct objects
of only those predicates whose bare plural direct objects cannot get a generic
interpretation (either referential/kind-referring or quantificational), but they get
only an existential interpretation.38
This fact alone raises an important question: what are the factors that govern
the distribution of bare singulars? The importance of addressing this question is
twofold. On the one hand, it has a bearing on the study of bare singulars. On the
other hand, it also relates to the study of bare plurals, given the distributional
parallels in languages between the bare singulars and the existential bare plurals.
With regard to the factors governing the distribution of bare singulars, I claim
that these are semantic in nature. However, on the assumption that a given
syntactic construction cannot be systematically ambiguous, my basic working
hypothesis is that semantic interpretations for noun phrases are fundamentally
dependent on their internal structure. On this view, I crucially claim that whereas
DPs may be either arguments or predicates, NPs translate as predicates at LF
irrespective of whether they occur as predicate nominals or as direct objects.
Consequently, they do not translate as variables or restricted modifiers. For bare
singulars (and existential bare plurals), this amounts to the claim that they are
predicates, not arguments. In other words, while subjects are always DPs (since
they are arguments, not predicates), direct objects (and predicate nominals) may
be either DPs or NPs, i.e., direct objects are not always arguments, they can also
be predicates. Thus, I argue that countable bare singulars cannot be doubled/

37. In some (though not all) Balkan languages (e.g. Greek, Bulgarian), bare singulars may occur as
subjects of unergative predicates. However, it is essential to note that unlike non-subject bare
singulars, bare singular subjects are referentially specific and therefore interchangeable with singular
definite descriptions and/or a-expressions on the referential reading of the latter (cf. also Marinis
1997). (These crucial meaning differences are detailed in Section 4.3, where, among other issues, I
investigate in detail the meaning of bare singulars.) Since I am advocating a systematic mapping
between syntax and semantics, I contend that bare singulars occurring as subjects are DPs (not NPs)
with a morphologically null D (cf. Section 4.3.3).
38. Throughout this paper the term ‘existential’ is used in opposition to ‘presuppositional’. (Thus
‘existential bare plurals’ should be understood as ‘non-generic bare plurals’ only.)
234 DALINA KALLULLI

scrambled because they are not arguments but predicates; they denote properties,
not individuals, and translate therefore as predicates, not as variables or restricted
quantifiers at LF. Drawing on work by Zimmerman (1993), I argue that most
natural language predicates can take both individuals and properties as their
internal arguments.
In the next section, I show that it is precisely in terms of the distinction
individual vs. property denotation that the distinction specific vs. non-specific for
noun phrases should be understood. Countable bare singulars provide an excel-
lent tool for this. A discussion of the phenomenon of specificity is essential for
this study, as I intend to show that specificity cannot be bestowed on an argu-
ment by a clitic. I will also demonstrate eventually that specificity-related effects
in clitic doubling and scrambling constructions arise only as epiphenomena since
argumenthood for noun phrases is defined by specificity.

4.3 Specificity, individuation, argumenthood39

4.3.1 The meaning of bare singulars

The a-expressions një fustan in (41a′) and ena forema in (41b′) might denote:
(43) a. some particular dress that Ann has seen on some display
b. some particular kind of dress (e.g. some Dior vs. some Versace
dress)
c. some/any object which classifies as a dress; that is, any dress at
all
With respect to specificity, the (43a) and (43b) readings are both specific
readings and can be continued by (44):40

39. This section builds on earlier work (cf. Kallulli 1997b, 1998). For reasons of space, I do not
discuss data from Mainland Scandinavian (MS) here. However, whatever is said in this paper about
the meaning of bare singulars in German and Balkan languages holds for MS as well. For details, see
Kallulli (1997b, 1998).
40. Note that the referential/attributive dichotomy (cf. Donellan 1966) divides the three readings in
(43) in a different manner. The reading in (43a) is referential, while the (43b) and (43c) readings are
attributive. This is so because only in (43a) has Ann established a direct relationship with some
particular dress. This is not the case in (43b); any Dior dress, not just a particular one, is sufficient
for Ann under the reading in (43b). Yet, the indefinite noun phrase in (43b) receives a specific
interpretation, because Ann is not interested in any dress; she wants a specific type of dress, a Dior
one, but obviously she does not mind as to what particular sample (e.g. with respect to colour, cut,
production year, etc.) she gets. Thus, specific noun phrases may be intended as either referential or
attributive (cf. also Ioup 1977). In other words, the distinction referential vs. attributive makes sense
for specific noun phrases only.
 DIRECT OBJECT CLITIC DOUBLING IN ALBANIAN AND GREEK 235

(44) She may find it in House of Fraser

Only the (43c) reading is non-specific, and (44) is not an appropriate continua-
tion for it. One could continue the (43c) reading as in (45):
(45) She may find one in House of Fraser
Importantly, the bare singulars fustan in (41a) and forema in (41b) cannot refer
to some particular dress or to some particular kind of dress. So, the bare
singulars in (41a, b) lack the readings given under (43a, b) that obtain for the
a-expressions in (41a′) and (41b′). This means that the bare singulars in (41a)
and (41b) may not receive specific interpretations. Thus, a-expressions and bare
singulars are not fully synonymous; they are so only on the non-specific readings
of the former.
As Ioup (1977) points out, certain inferences follow on a specific reading
which are invalid on a non-specific reading. In the specific readings (43a, b), the
existence of the items referred to by the a-expressions is presupposed. Given
(43a) and (43b), the given sentence in (46) will be true.
(46) There is a certain dress that Ann wants to buy
No existence claims follow from the non-specific reading in (43c), i.e. (46) is not
a valid inference from (43c). Instead, we can paraphrase (43c) as in (47):
(47) Ann wants there to be some dress or other that she can (find and) buy
Thus, what Ann is interested in (in (41a) and (41b)) is some individual or other
which embodies a certain property, namely that of being [+dress] and not, say,
[+book]. The identity of the item that Ann wants, beyond its being [+dress], is
irrelevant here. Assuming that properties do not exist outside individuals (that is,
that properties are not ontological primitives), Ann is interested in some individu-
al or other that has the property [+dress]. But, each individual that has the
property [+dress] in addition has other properties, at least one, that make it
distinct from other individuals that have the same property [+dress]. The very
existence of distinct individuals possessing the same basic property (here:
[+dress]), which causes them to be regarded as members of the same class (here:
the class of dresses), is due to the existence of at least one distinct property.
Being a distinct individual itself is a property. These extra properties of individu-
als, beyond the property [+dress], are not only irrelevant to Ann in (41a, b), but
indeed unable to be expressed by the bare singulars here. The bare singulars in
(39a, b) do not denote individuals but properties, which is why (41a) and (41b)
get an event-related reading which may be paraphrased as in (48):
236 DALINA KALLULLI

(48) Ann wants to engage/is interested in dress-buying

It is my contention that while direct object a-expressions may denote individuals,
direct object bare singulars may not; the latter invariably denote properties. The
distinction between properties and individuals may be represented as in (49):
(49) P vs. P ∩ pi
(where ‘P’ is the fundamental property that identifies individuals as
members of the same class and ‘pi’ is a property that does not
contradict ‘P’)
It is by now a well-established view in the semantic literature that specific
readings are presuppositional and non-specific readings are not (cf. Enç 1991;
Diesing 1992). The hypothesis that bare singulars are property-denoting expres-
sions, i.e. predicates, can account for the fact that they are not presuppositional
if we assume that presupposition is about saturated structures, that is, about
individuals (and propositions), not about properties. It then follows that
specificity involves individuation; individual-denoting expressions are always
specific, irrespective of the fact that they may be used referentially or attribu-
tively. On the other hand, property-denoting expressions are non-specific. Since
arguments are saturated structures, noun phrase arguments denote individuals,
that is, are specific.
To summarise: On their specific reading, noun phrases always denote
individuals, not properties. Individuals translate as arguments (they are saturated
structures), never as predicates at LF. Therefore, noun phrase arguments are
always specific (irrespective of the fact that as such they may be used referent-
ially or attributively). On their non-specific reading, noun phrases invariably
denote properties, not individuals. Properties translate as predicates at LF; they
are unsaturated structures. Bare singulars are non-specific (i.e. property-denot-
ing); they translate as predicates.
Given that direct objects may be instantiated by bare singulars, which
invariably denote properties, it follows that direct objects are not always argu-
ments; they may also be predicates. I claim that when direct objects denote
properties, not individuals (i.e. when they are predicates, not arguments),
doubling and scrambling cannot apply to them.
Consider the examples in (50) and (51).41 In (50), the bare singular piano
is a predicate, not an argument. Therefore it cannot scramble past the high
adverb probably. In (51) piano occurs to the left of the adverb. Yet, the meaning

41. Many thanks to Marcel den Dikken (personal communication) for pointing out these data to me.
 DIRECT OBJECT CLITIC DOUBLING IN ALBANIAN AND GREEK 237

of (51) suggests that piano is a predicate here as well, as indicated by its English
translation. Observe, however, that here the predicate (namely: play or take)
which selects piano as its internal argument is deleted at PF; that is, no adverbial
intervenes between the bare singular piano and the predicate whose internal
argument it is. Note also that piano playing or taking piano lessons is a gerundive
argument of the clausal predicate find. As such, it may scramble.
(50) Du: dat Jan (*piano) waarschijnlijk (piano) speelt
that Jan (*piano probably (piano plays
(51) Du: dat Jan (piano) waarschijnlijk (piano) leuker zal vinden dan
that Jan (piano probably (piano nicer will find than
viool
violin
‘that Jan will find playing the piano/taking piano lessons nicer
than playing the violin/taking violin lessons’
Adapting the formalisation of Dobrovie-Sorin and Laca (1995) for existential
bare plurals, it may be stated that direct object bare singulars are predicates
restricting an existentially bound argument variable that is independently
introduced in the LF representation as the placeholder of the theta-slot (that is,
the internal argument) of the clausal predicate. It is important to note that the
argument variable that the bare singular restricts does not arise via the translation
of the bare singular itself. The clausal predicate may then be translated as an
open formula whose open positions are bound by existential quantification, as
given in (53) for the (German) sentence in (52):
(52) Ich lese Zeitung
I read newspaper
(53) lx1 [∃ (x1 read x2 at e ∧ zeitung (x2))] (ICH1)
Let me now explain where the existential quantification in (53) comes from.
A striking property of bare singulars is that they invariably take (existential)
narrow scope in the presence of other scopal items in the sentence. Thus, the
Albanian sentence in (54a) only has the reading in (54b) but lacks the reading in
(54c) where the bare singular has scope over negation.
(54) a. Nuk dua biçikletë
not want-I bicycle
‘I don’t want a bicycle’
b. It is not the case that I want a bicycle
c. #There is a bicycle that I don’t want
238 DALINA KALLULLI

Likewise, the Albanian sentence in (55), unlike its English translation, can only
mean that there is a one-to-one correspondence between children and bicycles.
That is, (55) cannot mean that a bicycle was such that it was bought by many
children.
(55) Shumë fëmijë blenë biçikletë dje
many children bought bicycle yesterday
‘Many children bought a bicycle yesterday’
The data in this section unequivocally show that bare singulars cannot take wide
scope. In this respect, they differ both from definite descriptions and a-expres-
sions, which may, but need not, take wide scope. This fact immediately reminds
one of Carlson’s (1977) observation that the English bare plural always takes
narrow scope with respect to negation. He accounts for this by suggesting that
the existential force of the bare plural in non-generic contexts comes from a
source external to the bare plural itself, namely from the verb. I adopt this
proposal for bare singulars as well.42
What then is the relation between count bare singulars and bare plurals? I
address this question in the next section.

4.3.2 On the relation of count bare singulars to bare plurals

Recall from Section 2 that bare plural direct objects cannot be clitic doubled in
Albanian and Greek. Nor can they scramble in German or Dutch. For the
explanation that I will propose for this phenomenon, it is essential to point out
the distinction between generic and existential bare plurals (cf. Carlson 1977).
This distinction, which holds across the Germanic languages, does not, however,
hold for the Balkan languages. In the Balkan languages, generic readings (either
referential/kind-denoting or quantificational) are incompatible with bare plurals.
Bare plurals in these languages can only have an existential interpretation. The
same holds for Romance bare plurals (cf. Laca 1990; Longobardi 1994). Thus,
individual-level predicates, which, as is well known, force generic readings on
their direct objects, are incompatible with bare plurals in these languages. Some
examples are love, respect, admire, adore, etc. Generic readings in Balkan (and
Romance) languages require an overt determiner; the definite determiner for
plural noun phrases and either the definite or the indefinite determiner for
singular noun phrases.

42. As van Geenhoven (1996) shows, the arguments that have been brought against Carlson’s
lexicalised existential quantifier vanish if this quantifier is granted dynamic instead of static force.
 DIRECT OBJECT CLITIC DOUBLING IN ALBANIAN AND GREEK 239

My proposal rests on the claim that generically and existentially interpreted

bare plurals differ with respect to their D-feature: generic bare plurals are DPs
with a morphologically null D, whereas existential bare plurals are NPs lacking
a D-projection altogether. Consequently, generic and existential bare plurals
differ with respect to their specificity feature: generic bare plurals are [+specific]/
individual-denoting, whereas existential bare plurals are [−specific]/property-
denoting.43
What does it mean for generic bare plurals to be individual-denoting? It
means that generic bare plurals denote kinds (in non-quantified contexts), as in
I love dogs, or (in quantified contexts) denote (quantified) instantiations of kinds,
as in (Most) dogs are clever. This implies that generic bare plurals are either con-
stants or variables, depending on whether they name a kind or in (quantified
contexts) denote instantiations of it. I claim that existential bare plurals, on the
other hand, denote properties. As such, they are not constants or variables but
predicates. I argued above that bare singulars denote properties as well.
What, then, is the difference (if any) between bare singulars and existential
bare plurals, given that all languages that have bare singulars also have existen-
tial bare plurals? While both (56a) and (56b) necessarily have an event-related
reading, it seems to me that the difference between bare singulars and existential
bare plurals has to do with event reference. Thus, while the meaning of the
sentence in (56a) can be rendered as in (56c) or (56d), the minimally different
(56b) containing an (existential) bare plural instead of the bare singular can be
rendered as in (56d), not as in (56c). Thus, (56a) can, though need not, be
synonymous with (56b), whereas (56b) can only mean that Eva might engage in
several events of newspaper reading. Strictly speaking, there is no small event in
which a person can read more than one newspaper at a time. Hence, it is as if
the bare plural in (56b) has scope over the whole VP.44 Whether this is an
instance of genuine wide scope of the bare plural or some kind of a pseudo-
scope effect, this paper will not contribute to assessing.
(56) a. Eva will morgen Zeitung lesen
Eva will tomorrow newspaper read

43. This is independently proposed by Kiss (to appear). However, Kiss relies on Enç’s (1991)
account of specificity which is rather problematic. Space considerations prevent me from dealing with
this point in detail, though some problems with it will be identified in Section 4.3.4. For further
discussion see Abbott (1995).
44. M. Krifka (personal communication) points out to me that in the same vein, number words can
have wide scope, as in his example: ‘Four thousand ships passed through the lock’, which means:
‘There were four thousand ship-passings’.
240 DALINA KALLULLI

b. Eva will morgen Zeitungen lesen

Eva will tomorrow newspapers read
c. Tomorrow Eva will engage in (at least) one newspaper-reading
event
d. Tomorrow Eva wants to engage in several events of newspaper
reading
Thus, I am claiming that existential bare plurals are the plural counterparts of
bare singulars. On the one hand, the fact that bare singulars occur as direct
objects of only those predicates whose bare plural direct objects cannot get a
generic interpretation supports this claim. On the other hand, however, the
reverse does not hold across all the languages that have bare singulars in object
position. German is a case in point. How are we to account for the lack of (total)
distributional compatibility between bare singulars and existential bare plurals
then? For one thing, we saw in (56a) to (56d) above that the meaning of bare
singulars is a subset of the meaning of existential bare plurals. We saw (in
Section 4.3.1) that the meaning of bare singulars is also a subset of (i.e. con-
tained in) the meaning of a-expressions. It is only reasonable then to try to relate
the lack of (one-to-one) distributional parallelism between bare singulars and
existential bare plurals within and across languages to economy considerations.
If existential bare plurals are the plural counterparts of bare singulars, they
should not occur as subjects of unergative predicates. At first sight, this predic-
tion is easily falsified by data like the one in (57).
(57) Studenten lärmen auf der Strasse
students make noise in the street
‘Students are making noise in the street’ or: ‘Students make noise in
the street’45
However, I maintain that this prediction is indeed borne out and that construc-
tions like the one in (57) are instances of subjectless sentences when Studenten
is interpreted existentially (though not when it is interpreted generically). I
propose that when interpreted existentially, Studenten in (57) is a predicate
nominal in the Spec of CP, probably derived from a cleft construction. Crucial
evidence for this view comes from another Germanic language, Norwegian.
Hellan (1986) observes that in Norwegian, adjective phrases (APs) in predicative
position agree in gender and number with their subject. In (58), however, they do

45. It is well known that, unlike, simple present tense in English, the simple present in German can
have both an episodic and a generic meaning. Hence the double translation in English.
 DIRECT OBJECT CLITIC DOUBLING IN ALBANIAN AND GREEK 241

not: the predicative adjective is marked for neuter gender and singular number,
while the noun is masculine, and can be either singular or plural. If the bare
singular in (58) were really the subject of the sentence, this construction would
be a counterexample to the theory of agreement.46
(58) Bil(er) er dyr-t. (Hellan 1986: 95)
car..(s) is expensive-neut
I propose that the NP bil(er) ‘car(s)’ is not the subject of the sentence, but
instead occupies the Spec of CP. This view is corroborated by the fact that the
NP in constructions like (58) is necessarily interpreted as a [+Focus] phrase; the
meaning of the sentence in (58) can be rendered as in (59).
(59) It is expensive to have/keep/run/manage/buy a car/cars
In this section I have argued that just like bare singulars, existential bare plurals
are not DPs with a morphologically empty D, but NPs that lack a D-projection.
As such, they cannot be doubled (in Albanian/Greek) or scrambled (in German/
Dutch). On the other hand, generic bare plurals are DPs with a morphologically
empty D. They are always specific (i.e. individual-denoting). As such, they can
scramble unless they are marked [+Focus].47 The syntactic distinction NP vs.
DP (with morphologically null D) that I have drawn between existential and
generic bare plurals, respectively, in addition to representing a principled
mapping between syntax and semantics, is also motivated by the (morphological)
fact mentioned above, that generic plural nominals in Balkan (and Romance)
languages necessarily require the presence of the definite determiner.

4.3.3 Bare singular subjects

As already pointed out, in Albanian bare singulars are confined to predicative
and direct object positions. In Greek, bare singulars may also occur as subjects,
in which case they are necessarily focused, as the English translation of the
Greek example in (58) indicates.
(60) FIDHI ton íkhe dhagósi ton Cósta (Agouraki 1993: 170)
SNAKE him. had bitten the Costas
‘It was a snake that had bitten Costas’

46. Note also that Norwegian is not a pro-drop language.

47. Since bare plurals are incompatible with generic readings in the Balkan languages, the question
of doubling them does not even arise.
242 DALINA KALLULLI

In this context, it is fundamental to note that while all transitive verbs may take
a-expressions as their direct objects, not all may take bare singulars. Besides,
while a-expressions in direct object position may be ambiguous between a
specific and a non-specific interpretation, bare singular objects may only receive
a non-specific interpretation. In other words, direct object bare singulars are not
always interchangeable with direct object a-expressions. However, when bare
singulars occur as subjects (as in (60)), they are fully interchangeable with either
a definite DP or an a-expression. This is natural, in view of the fact that subjects
are necessarily specific. These facts suggest that subject bare singulars are
structurally different from direct object bare singulars. I claim that, while
subjects are always DPs, direct objects may be either DPs or NPs.

4.3.4 Definite expressions

Consider the example in (61):
(61) I shall kiss the first woman to enter this room
In line with what was stated in Section 4.3.1, the definite expression in (61) is
also specific, though it may have both a referential and an attributive reading,
depending on whether or not the speaker knows beforehand who the first woman
to enter the room will be. In other words, the definite expression in (61) may
denote either a particular individual in relation to the speaker, namely, the type
of ‘first woman to enter the room’ (as opposed to, say, the type of ‘second
woman to enter the room), or the type of ‘no woman to enter the room’. The
type of ‘first woman to enter the room’ is an individual with respect to the
concept/property ‘woman’. So, independently of whether the definite expression
is intended to refer or not, it is specific, which also accords with Enç (1991).
The question arises, however, as to whether definite noun phrases in direct
object position can ever be predicates, that is, denote properties (like bare
singulars and a-expressions on a non-specific reading). I will argue that they can.
Examples are definite noun phrases in object position in set expressions like:
take the bus in (62a), play the violin in (62b).48
(62) a. I like to take the bus
b. Ben has played the violin beautifully at times
It is true that the definite expression the bus in (62a) may have both a referential-

48. J. Emonds (personal communication.) points out to me that definite expressions in some locative
phrases (e.g. I am going to the airport/to the doctor’s/to the shore/to the hospital) have a predicative
reading as well. Note that these are not generic: The only time in my life I went to Texas I took the plane.
 DIRECT OBJECT CLITIC DOUBLING IN ALBANIAN AND GREEK 243

specific and an attributive-specific reading (as paraphrased in (63a, b)), but what
is important to note is that it also has a non-specific reading, as paraphrased in
(63c). Likewise, the violin in (62b) also has a non-specific reading which may be
paraphrased as in (64).
(63) a. There is a bus-vehicle, always the very same, that I like to take
b. There is a bus-line that I like to take
c. I like to travel by bus (I don’t like to walk, drive, take the train
or fly)
(64) Ben is a talented violin-player
The fact that not only indefinite expressions but also definite expressions may
have both a specific and non-specific reading constitutes a counterexample to the
claim that all definites are specific (cf. Enç 1991). It suggests that the class of
definite expressions is far from homogeneous semantically (cf. also Vergnaud
and Zubizarreta 1992). Above I argued that specific readings arise when noun
phrases denote individuals and non-specific readings when they denote proper-
ties. Note, however, that both a-expressions and definite expressions may only be
interpreted non-specifically when they occur as predicate nominals or as direct
objects (sometimes also as objects of certain prepositions), but not when they
occur as subjects. That subjects invariably denote individuals when they are
instantiated by noun phrases should not be a matter of controversy in a frame-
work like Principles and Parameters.49
The fact that bare singulars are synonymous with a-expressions on their
non-specific reading only suggests that a-expressions are potential designators of
either properties or individuals (that is, a-expressions may be predicates or
variables). However, postulating that a-expressions are intrinsically ambiguous
between a specific (i.e. individual-denoting) and a non-specific (i.e. property-
denoting) interpretation (that is, correspond to two distinct logical types, viz. 〈e〉
vs. 〈e,t〉) cannot account for two things in a principled manner. The first is why
a-expressions occurring as subjects and datives lack a non-specific (i.e. property-

49. In the Principles and Parameters framework the subjects of sentences like Being wise/To be wise
is crazy or Being crazy is crazy (examples from (Chierchia, 1985: 418) are clausal from a syntactic
point of view and propositional from a semantic point of view (Koster and May 1982). For cases like
Wisdom deserves reward, I agree with Chierchia in that it cannot be stressed too strongly that the
subject of this sentence is a property-like element, because ‘the realm of nominalisations such as
[wisdom] … is still largely unknown, which relegates our considerations to the realm of intuitions’
(Chierchia 1985: 418). Such examples do not therefore necessarily constitute counterexamples to my
claim that subjects invariably denote individuals.
244 DALINA KALLULLI

denoting) interpretation and why. The second is why the ambiguity in terms of
the distinction specific vs. non-specific for a-expressions arises only when they
occur as direct objects of certain predicates (e.g. want, buy, draw, hunt, smoke,
find, get, etc.) but not of others (e.g. love, hate, admire, adore, etc.). These facts
can be accounted for if we assume that many (and perhaps most, though not all)
natural language predicates of the type 〈e,〈e,t〉〉 (e.g. buy) can be raised to type
〈〈e,t〉,〈e,t〉〉 meaning:
(65) lP lx ∃y [P(y) ∧ BUY (x,y)]
This means that certain predicates that take individuals as their internal argu-
ments may also take properties as their internal arguments. In addition, we need
to assume that the dual nature of a-expressions is due to their lexical under-
specification with respect to specificity (i.e. individual vs. property-denotation).
Hence, they can oscillate between type 〈e〉 and 〈e,t〉. Given the fact that many
definite noun phrases may also be interpreted non-specifically/predicatively when
objects of verbs and prepositions, we need to assume that the-expressions are
also underspecified with respect to individual vs. property-denotation and can
therefore oscillate between type 〈e〉 and 〈e,t〉. To generalise, we may then state
that while NPs (e.g. bare singulars and existential bare plurals) are unambiguous-
ly type 〈e〉, DPs may be of type 〈e〉 or 〈e,t〉.50
Consider the examples in (66).
(66) Ge: a. weil ich morgen den Bus nehme
because I tomorrow the bus take
b. weil ich den Bus morgen nehme
because I the bus tomorrow take
‘because I will take the bus tomorrow’
In line with what was stated above, den Bus ‘the bus’ in (66a) can denote either
an individual (that is, some bus-vehicle or other or some bus-line or other) or a
property. In other words, both (67a) and (67b) are valid paraphrases for (66a).
(67a) is an event-related reading; that is, den Bus here denotes a property and
translates therefore as a predicate at LF.
(67) a. because, as for me, I will engage in bus-taking tomorrow
b. because, as for (me and) the bus, I will take it tomorrow

50. Alternatively, it might be that both the indefinite and the definite article are not exclusively
generated under D but may also be generated NP-internally. It is beyond the scope of this study to
decide between these alternatives.
 DIRECT OBJECT CLITIC DOUBLING IN ALBANIAN AND GREEK 245

In (66b), on the other hand, the scrambled DP den Bus denotes an individual
only; that is, it denotes some bus or other or some bus-line or other. In other
words, the scrambled DP den Bus in (66b) is specific/presuppositional. Since
specificity/presuppositionality is a property of arguments not of predicates, den
Bus in (66b) is an argument variable, not a predicate, as it can (though need not)
be in (66a). Crucially, (66b) lacks the event-related reading that obtains for (66a).
This suggests that scrambling applies to arguments only, not to predicates. Hence
the unavailability of the reading in (67a) for the sentence (66b). The same pattern
obtains with clitic doubling of definites in Albanian and Greek.
In sum, it may be stated that definite noun phrases and a-expressions are
semantically (and perhaps syntactically) non-homogeneous; they are not always
syntactic arguments when objects of verbs (and prepositions) but may translate
both as arguments or as predicates at LF depending on whether the clausal
predicate selects an individual (type 〈e〉) or a property (type 〈e,t〉) as its internal
argument (cf. also van Geenhoven (1996) for a similar treatment of indefinites).
The type shifting mechanism (cf. Partee 1987) allows for this duality. This
creates the illusion that scrambling/doubling of definites and a-expressions is
optional. In fact, scrambled/doubled objects are always syntactic arguments.
Since argument noun phrases are always specific (i.e. individual-denoting),
specificity effects will be observed in scrambling constructions. Non-scrambled/
non-doubled objects may but need not be arguments.

5. Conclusion

In this paper, I have shown that direct object clitic doubling in Albanian and
Greek produces information structure in a systematic way: doubled DPs are
unambiguously interpreted as topics. This suggests that topichood is, at least in
part, encoded in the syntax for these languages. Whether this is the case univer-
sally and whether the representation of topics involves the same syntactic
configuration cross-linguistically, remains an issue for further study. I have also
shown that specificity cannot be bestowed on an argument by a doubling clitic/
scrambling; instead specificity is fundamentally related to the D-slot. Specificity
effects in doubling and scrambling constructions are only by-products of deeper
triggering properties.
246 DALINA KALLULLI

References

Abbott, B. 1995. ‘Some remarks on specificity’. Linguistic Inquiry 26: 341–347.

Abney, S. 1987. The English Noun Phrase in its Sentential Aspect. Doctoral dissertation,
MIT.
Agouraki, Y. 1993. Spec–Head Licensing: The Scope of the Theory. Doctoral dissertation,
University College London.
Anagnostopoulou, E. 1994. ‘On the representation of clitic doubling in Modern Greek.’
EUROTYP Working Papers, Theme Group 8, Vol. 5: 1–66.
Borer, H. 1984. Parametric Syntax: Case Studies in Semitic and Romance Languages.
Dordrecht: Foris.
Brody, M. 1990. ‘Some remarks on the focus field in Hungarian.’ UCL Working Papers
in Linguistics 2: 201–225.
Carlson, G. 1977. Reference to Kinds in English. Doctoral dissertation, University of
Massachusetts.
Chastain, C. 1975. ‘Reference and context.’ In: Gunderson, K. (ed.) Minnesota Studies in
the Philosophy of Science, vol VIII: Language, Mind and Knowledge. University of
Minnesota Press, 194–269.
Chierchia, G. 1985. ‘Formal semantics and the grammar of predication.’ Linguistic
Inquiry 16: 417–443.
Chomsky, N. 1976. ‘Conditions on rules of grammar.’ Linguistic Analysis 2: 303–351.
Chomsky, N. 1995. The Minimalist Program. Cambridge, Mass.: MIT Press.
Cinque, G. 1993. ‘A null theory of phrase and compound stress.’ Linguistic Inquiry 24:
239–297.
Diesing, M. 1992. Indefinites. Cambridge, Mass.: MIT Press.
Dobrovie-Sorin, C. 1990. ‘Clitic doubling, wh-movement and quantification in Roman-
ian.’ Linguistic Inquiry 21: 351–397.
Dobrovie-Sorin, C. and B. Laca 1995. ‘Generic bare NPs.’ Ms., Université Paris and
Université Strasbourg.
Donellan, K. 1966. ‘Reference and definite descriptions.’ The Philosophical Review 75:
281–304.
Emonds, J. 1992. ‘How clitics license null phrases: A theory of the lexical interface.’ In:
van Riemsdijk, H. (ed.) EUROTYP Working Papers. University of Tilburg.
Enç, M. 1991. ‘The semantics of specificity.’ Linguistic Inquiry 22: 1–25.
van Geenhoven, V. 1996. Semantic Incorporation and Indefinite Descriptions: Semantic
and Syntactic Aspects of Noun Incorporation in West Greenlandic. Doctoral disserta-
tion, University of Tübingen.
Haiden, M. 1996. ‘The aspect of short scrambling.’ Wiener Linguistische Gazette 57:
121–145.
Heim, I. 1982. The Semantics of Definite and Indefinite Noun Phrases. Doctoral disserta-
tion, University of Massachusetts.
Hellan, L. 1986. ‘The headedness of NPs in Norwegian.’ In: Muysken, P. and H. van
Riemsdijk (eds.) Features and Projections. Dordrecht: Foris, 89–122.
 DIRECT OBJECT CLITIC DOUBLING IN ALBANIAN AND GREEK 247

de Hoop, H. 1992. Case Configuration and Noun Phrase Interpretation. Doctoral disserta-
tion, University of Groningen.
Horvath, J. 1986. FOCUS in the Theory of Grammar and the Syntax of Hungarian. Dor-
drecht: Foris.
Ioup, G. 1977. ‘Specificity and the interpretation of quantifiers.’ Linguistics and Philoso-
phy 1: 233–245.
Jackendoff, R. 1972. Semantic Interpretation in Generative Grammar. Cambridge, Mass.:
MIT Press.
Jacobs, J. 1986. ‘The Syntax of focus and adverbials in German.’ In: Abraham, W. and
S. de Meij (eds.) Topic, Focus and Configurationality. Amsterdam: John Benjamins,
103–127.
Jaeggli, O. 1986. ‘Three issues in the theory of clitics.’ In: Borer, H. (ed.) The Syntax of
Pronominal Clitics (Syntax and Semantics, vol. 19). New York: Academic Press,
15–42.
Joseph, B. and I. Philippaki-Warburton 1987. Modern Greek. London: Croom Helm.
Kallulli, D. 1995. Clitics in Albanian. University of Trondheim Working Papers in
Linguistics, vol. 24.
Kallulli, D. 1997a. ‘Optional verb movement: Albanian imperatives.’ In: Bruening, B.
(ed.) Proceedings of SCIL 8. MIT Working Papers in Linguistics, vol. 31: 225–234.
Kallulli, D. 1997b. ‘Bare singulars and bare plurals: mapping syntax and semantics.’ In:
Cambier-Langeveld, T., J. Costa, R. Goedemans and R. van de Vijver (eds.)
Proceedings of ConSOLE V. Leiden, 153–168.
Kallulli, D. 1998. ‘NP Predicates.’ In: Blight, R. and M. Moosally (eds.) Proceedings of
the 1997 Texas Linguistic Society conference “The Syntax and Semantics of Predica-
tion”. Texas Linguistic Forum 38: 173–188.
Kallulli, D. 1999. The Comparative Syntax of Albanian: On the Contribution of Syntactic
Types to Propositional Interpretation. Doctoral dissertation, University of Durham.
Kazazis, K. and J. Pentheroudakis 1976. ‘Reduplication of indefinite direct objects in
Albanian and Modern Greek.’ Language 52: 398–403.
Kiss, K. To appear. ‘On generic and existential bare plurals and the classification of
predicates.’ In: Rothstein, S. (ed.) Events. Dordrecht: Kluwer.
Koster, J. and R. May 1982. ‘On the constituency of infinitives.’ Language 58: 116–143.
Krifka, M. 1996. ‘Frameworks for the representation of focus.’ Ms., University of Texas.
Laca, B. 1990. ‘Generic objects: some more pieces of the puzzle.’ Lingua 81: 25–46.
Longobardi, G. 1994. ‘Reference and proper names: A theory of N-movement in syntax
and logical form.’ Linguistics and Philosophy 25: 609–669.
Marinis, T. 1997. ‘The acquisition of expletive definite articles in Modern Greek.’ Paper
presented at ConSOLE 6, Lisbon.
Massey, V. 1991. Compositionality and Constituency in Albanian. Doctoral dissertation,
University of North Carolina.
Partee, B. 1987. ‘Noun phrase interpretation and type shifting principles.’ In: Groenen-
dijk, J., D. de Jongh and M. Stockhof (eds.) Studies in Discourse Representational
Theory and the Theory of Generalized Quantifiers. Dordrecht: Foris, 115–143.
248 DALINA KALLULLI

Pesetsky, D. 1987. ‘wh-in-situ: Movement and unselective binding.’ In: Reuland, E. and
A. ter Meulen (eds.) The Representation of (In)definiteness. Cambridge, Mass.: MIT
Press, 98–129.
Postal, P. 1969. ‘On so-called “Pronouns” in English.’ In: Reibel, D. and S. Schane (eds.)
Modern Studies in English: Readings in Transformational Grammar. Englewood
Cliffs: Prentice Hall, 201–224.
Raposo, E. 1997. ‘Postal’s theory meets the minimalist program: some observations on
the pronominal system of Portugese.’ Paper presented at ConSOLE 6, Lisbon.
Reinhart, T. 1982. Pragmatics and Linguistics: An Analysis of Sentence Topics. Blooming-
ton: Indiana University Linguistics Club.
Reinhart, T. 1995. ‘Interface economy: Focus and markedness.’ Ms., University of
Utrecht.
Rivero, M.-L. 1994. ‘Clause structure and V-movement in the languages of the Balkans.’
Natural Language and Linguistic Theory 12: 63–120.
Rivero, M.-L. and A. Terzi 1994. ‘Imperatives, illocutionary force, and V-movement.’
Ms., University of Ottawa.
Rizzi, L. 1991. ‘Residual verb second and the Wh-criterion.’ Technical Reports in Formal
and Computational Linguistics 3. University of Geneva.
Rochemont, M. 1986. Focus in Generative Grammar. Amsterdam: John Benjamins.
Rooth, M. 1996. ‘Focus.’ In: Lappin, S. (ed.) The Handbook of Contemporary Semantic
Theory. Oxford: Blackwell, 271–297.
Sperber, D. and D. Wilson 1988. Relevance: Communication and Cognition. Oxford:
Blackwell.
Sportiche, D. 1996. ‘Clitic constructions.’ In: Rooryck, J. and L. Zaring (eds.) Phrase
Structure and the Lexicon. Dordrecht: Kluwer Academic Publishers, 213–277.
Sternefeld, W. 1990. ‘Scrambling and minimality.’ In: Grewendorf, G. (ed.) Scrambling
and Barriers. Amsterdam: John Benjamins, 239–257.
Suñer, M. 1988. ‘The role of agreement in clitic-doubled constructions.’ Natural Lan-
guage and Linguistic Theory 6: 391–434.
Terzi, A. 1992. PRO in Finite Clauses: A Study of the Inflectional Heads of the Balkan
Languages. Doctoral dissertation, City University of New York.
Uriagereka, J. 1995. ‘Aspects of the syntax of clitic placement in Western Romance.’
Linguistic Inquiry 26: 79–123.
Vergnaud, J-R. and M. L. Zubizarreta 1992. ‘The definite determiner and the inalienable
constructions in French and English.’ Linguistic Inquiry 23: 595–652.
Vallduví, E. 1994. ‘Detachment in Catalan and information packaging.’ Journal of
Pragmatics 22: 573–601.
Webelhuth, G. 1989. Syntactic Saturation Phenomena and the Modern Germanic Languag-
es. Doctoral dissertation, University of Massachusetts.
Zimmerman, E. 1993. ‘On the proper treatment of opacity in certain verbs.’ Natural
Language Semantics 1: 149–179.
 Where do Clitics Cluster?

Ljiljana Progovac
Wayne State University

Abstract

Critically examining Bošković’s treatment in this volume of second position

clitic placement in Serbo-Croatian, this paper argues for a conceptually
desirable tight fit between syntactic position and intonation boundaries, on the
basis of the strong correlation between these two. The approach proposed
accounts for the interaction between wh-formation, comma intonation and clitic
placement, and sheds light on the nature of the syntax/phonology interface.

1. Introduction

This paper is a brief response to the conclusion reached in Bošković, this

volume, that second position cliticisation in Serbo-Croatian (SC) is phonological
in nature. For the relevant background on Clitic Second in SC, the reader is
referred to Bošković’s paper.
There are two basic sets of data on which Bošković’s conclusion relies: (i)
the availability of partial clitic clusters, and (ii) apparent sensitivity of clitic
placement to intonation boundaries. The first set of data has been addressed in
Progovac (1998a), the basic argument of which will be reviewed in Section 2.
This paper addresses the latter set of data, and argues that the influence of
intonation boundaries on clitic placement is only apparent (see also Progovac 1998b
for additional arguments against phonological placement of clitics, based on the
interaction between the distribution of the (eventive) pronominal to and clitics).
If A and B consistently coincide, there are two possible reasons: (i) A
causes/determines B, or vice versa; or, alternatively, (ii) there is a third factor,
factor C, that causes/determines both A and B, giving the appearance that A and
250 LJILJANA PROGOVAC

B are directly related. Bošković goes for the first scenario, and does not consider
the logical alternative. The main purpose of this paper is to point out the
availability of the latter (more complex) scenario, to explore what the third
(common) factor C may be in the case of clitic placement and intonation
patterns, and to compare the two logical alternatives. Specifically, both clitic
placement (A) and comma intonation (B) are argued to be sensitive to a third
factor, the edges of the ‘kernel clause’ or the extended projection of V, normally
CP (factor C).

2. Partial clitic clusters

This section is a brief sketch of one of the arguments given in Progovac (1998a)
against the proposal that SC clitics sit in separate functional projections.
A recent proposal by Stjepanović (1997) challenges the assumption that
clitics cluster in one position (see also Bošković 1995, this volume). Instead, she
argues that each clitic sits in a separate functional projection, in which it checks
its (strong) features. Although it is not clear to me why, Bošković (this volume)
takes this analysis as support for his phonological approach to clitics, as opposed
to syntactic placement (see note 1. for a possible explanation); if clitics indeed
sit in separate functional projections, then they are syntactically placed. The
reasons need not concern us here, since my intention is to show that the analysis
itself does not hold.
In the data below, the integrity of clitic clusters does not seem to be
preserved, since only parts of clitic clusters surface in second conjuncts:
(1) ?Vi ste me videli, a i oni su, takodje
you are me seen but and they are too
‘You saw me, and they did, too.’
Stjepanović (1997) proposes to analyse such data as deletion/ellipsis of a
functional projection (in the extended projection of VP) selected by the auxiliary
clitic. In Progovac (1998a) I argue that this data is also consistent with another
analysis, in particular, base generation of a silent VP. Suppose that no (surface)
deletion rule applies to the data above. Instead, a silent VP (e) is base generated
in second conjuncts:
 WHERE DO CLITICS CLUSTER? 251

(2) XP

Vi X

X AuxP

Aux VP

ste e

If pronominal clitics in SC are generated in the corresponding argument positions

inside VP (Kayne 1989), they cannot be generated with the silent VP. Since
auxiliary clitics are not generated within VP, but rather in a functional projection
above VP, they can surface with empty VPs.
The placement of the third person auxiliary clitic je follows directly from the
proposal in (2), but is problematic for Stjepanović’s proposal. The analysis in (2)
predicts that it is possible to drop pronominal clitics, while retaining only the
auxiliary, regardless of the relative ordering of auxiliary and pronominal clitics
at Spell-Out. This is supported by the data. Unlike other auxiliary clitics, je
surfaces after the pronominal clitics (witness the different orders in (1) and (3);
see Tomić 1996 for discussion and an analysis). Stjepanović’s proposal predicts
that the deletion of the pronominal clitics will necessarily imply the deletion of
je as well. This prediction is contrary to fact; (3) is grammatical, although, under
this analysis, it involves deletion of a non-constituent, italicised in (4).
(3) ?Novak mi ga je predstavio, a i Goran je, takodje
Novak me him is introduced, but and Goran is too
‘Novak introduced me to him, and Goran did, too.’
(4) ?Novak mi ga je predstavio, a i Goran mi ga je predstavio, takodje
Under Stjepanović’s analysis, as the last element in the string, je should be able to
delete if it is in a separate AuxP. The example below is ungrammatical, however:
(5) *Novak mi ga je predstavio, a i
Novak me- him- is- introduced but and
Goran mi ga, takodje
Goran me- him- too
The ungrammaticality of (5), again, follows directly from the analysis in (2), but is
not accounted for in Stjepanović’s analysis, at least not without a major stipulation.
252 LJILJANA PROGOVAC

The reader is referred to Progovac (1998a) for additional arguments against

placing clitics in separate projections at Spell-Out.

3. Clitics and intonation

It is claimed in Bošković (this volume) that purely syntactic approaches to clitic

placement in SC are ‘fatally flawed’, e.g. the ones in Ćavar and Wilder (1994),
Wilder and Ćavar (1994a, b), Franks and Progovac (1994), Rivero (1994),
Dimitrova-Vulchanova (1995), Tomić (1996), Progovac (1996), etc.
First, it is worth pointing out that these approaches do not claim that ‘syntax
is completely responsible for the phenomenon of second position cliticisation in
SC’ (p. 3.), and therefore cannot be classified as ‘purely syntactic’. Rather, the
basic argument in these approaches is that the first position (the host) is uniform-
ly a syntactic phrase, rather than a phonological unit/word, as might seem to be
the case in some examples (see e.g. Browne 1974; Comrie 1989).1 Thus, they
present an argument that certain aspects of cliticisation in SC are syntactic in
nature, which is very different from claiming that all of cliticisation is syntactic
in nature. Although they do not offer an explanation for it, these approaches do
not deny that the very requirement on clitics to be phonologically supported may
be phonological in nature. Of course, adopting a primarily phonological approach
is not a guarantee that this phonological requirement will be captured successful-
ly, not to mention the problem of capturing the generalisation that the host has
to be a syntactic phrase. So far, the literature has been successful in recognising
phonological aspects of cliticisation and syntactic aspects of cliticisation. The
challenge remains, however, of putting the two together into a coherent theoreti-
cal and empirical picture; as far as I can tell, this has not happened yet (but see
a recent attempt in Franks 1998).
Third position placement of clitics, such as (6) below, has been used as an
argument for the claim that the position for clitics is not a well-defined syntactic
projection, such as the highest functional projection in an extended projection
of V (say C), but rather that it is determined by phonological considerations,

1. In fact, in his paper Bošković seems to adopt this result of the syntactic approaches since he uses
it to demolish the purely phonological approaches to cliticisation. He also uses the results of these
same phonological approaches to demolish what he regards as purely syntactic approaches. This is
obviously a dubious strategy, and what is crucially missing in his paper is a demonstration of how
his own specific proposal, which adopts elements of all these analyses, fares with respect to each, and
why it is superior.
 WHERE DO CLITICS CLUSTER? 253

more specifically, intonation boundaries (Radanović-Kocić 1988; Bošković, this

volume).2
(6) Svojim rodjacima po majci, Jovan će prodati knjige
self’s relatives after mother John will sell books
‘To his maternal relatives, John will sell the books.’
Based on the fact that wh-words can follow heavy preposed phrases, as in (7), it
was argued in Progovac (1996) that any third position placement correlates with
the first constituent being CP external, either adjoined to CP, or in a projection
above CP:
(7) Svojim rodjacima po majci, šta li će Jovan prodati?
self’s relatives after mother what  will John sell
‘To his maternal relatives, what is John going to sell?’
Given that a wh-word and the question particle li, uncontroversially in C, follow
the preposed phrase, it must be that this phrase is higher than SpecCP. To
exclude the placement of a clitic after this phrase (8), or inside this phrase (9),
it is enough to assume that this phrase is not in SpecCP and that C is the highest
functional head in the extended projection of V, in this case. Thus, the clitic in
(8) is not supported within the relevant projection; the clitic in (9), on the other
hand, is placed beyond the highest functional head in the extended projection of
the verb.
(8) *Svojim rodjacima po majci, će Jovan prodati knjige
self’s relatives after mother will John sell books
‘To his maternal relatives, John will sell the books.’
(9) *Svojim će rodjacima po majci, Jovan prodati knjige
self’s will relatives after mother John sell books
The judgements are different without comma intonation after the preposed
phrase. Without a pause, as in (10), the clitic can be placed after the first
element of the preposed phrase. It is still rather marginal, although not ungram-
matical, to place the clitic right after the heavy phrase (11).

2. Caink (1997) and Franks (1998) reach the conclusion that SC clitics are in the highest functional
head of the clause, whether this is C or not. Notice that Bošković (1995, this volume) puts
considerable weight on his argument that not all clitic placement is in C. However, the issue of
whether clitics are uniformly in C, or in the highest functional head of a clause, or in distinct
functional heads, a la Stjepanović (see Section 2), has no direct or decisive bearing on the phonology/
syntax debate. In fact, each of the scenarios is consistent with syntactic placement of clitics.
254 LJILJANA PROGOVAC

(10) Svojim će rodjacima po majci Jovan prodati knjige

self’s will relatives after mother John sell books
‘To his maternal relatives, John will sell the books.’(no pause)
(11) ??Svojim rodjacima po majci će Jovan prodati knjige
self’s relatives after mother will John sell books (no pause)
Thus, indeed, intonation breaks correlate with the placement of clitics. This can
mean two things: (i) either clitic placement is indeed defined on the basis of
intonational boundaries, as pursued in Radanović-Kocić (1988) and Bošković
(this volume); or (ii) both clitic placement and the position of intonational
boundaries are due to a third factor.
The latter scenario is supported by the data below. Another fact is correlated
with comma intonation, too: the ability to insert wh-words after the heavy
phrases. While wh-words can follow preposed phrases in case of comma
intonation, wh-words cannot follow such phrases where there is no pause:
(12) Svojim rodjacima po majci, šta će Jovan prodati? (pause)
‘To his maternal relatives, what će will John sell?’
(13) ?*Svojim rodjacima po majci šta će Jovan prodati? (no pause)
The contrast above does not follow from Radanović-Kocić’s or Bošković’s
account, but it does follow straightforwardly from an account that recognises a
tight fit between intonation patterns and syntactic positions, insuring that
intonation patterns reflect syntactic structure. In fact, Syntax and Phonology
converge here in a rather coherent picture, given the following assumption: the
material that is set off with commas in the above examples is necessarily outside
of the kernel clause, i.e. the extended projection of V, which explains why it can
be followed by a wh-word:
(14) A kernel clause (the highest extended projection of V) forms an
intonation unit. Anything preceding the kernel clause is set off by
comma intonation.
The material not set off by commas in the above examples is then inside the
kernel CP, and occurs in the SpecCP, prohibiting the appearance of a wh-word
in this position. Given this much, clitics follow an expected pattern. With comma
intonation following the preposed phrase, clitics can attach neither inside that
phrase (9), nor immediately behind it (8), since it is external to their kernel CP
domain. Without the comma intonation, clitics can attach inside the phrase, as
illustrated in (10). If clitics are indeed in a fixed position, then this means that
the AP svojim appears in the Spec of the highest functional head, perhaps a
 WHERE DO CLITICS CLUSTER? 255

TopicP, or a CP, and the rest of the phrase is below this head, say IP-adjoined.
Why is it, then, that the preposed phrase in (11) can support the clitic only
marginally? Given the analysis developed here, the question reduces to: why is
the preposed phrase not comfortable in SpecCP? Suppose that a long/heavy
phrase tends to impose an intonation boundary after it, for reasons that may well
have to do with how much material one can pronounce ‘in one breath.’ As soon
as this boundary is phonologically marked, the phrase has to be interpreted as
outside the kernel clause, by force of (14). This would account for the more
general ban on ‘heavy’ material within the kernel clause, without involving a
‘look-ahead’ by syntax. Instead, the pronunciation will determine what kind of
syntactic structure is to be built, which reflects the way we process language.
Just like encountering an article is a signal that one needs to build a Determiner
Phrase, encountering a comma boundary after a single phrase is a signal that the
kernel clause has not begun yet.
Bošković (this volume) finds it undesirable to place heavy vs. light phrases
into different syntactic positions. However, this is empirically a rather common
phenomenon. The English examples below illustrate that heavy direct objects are
normally placed after the PP (16), while such placement for light direct objects
is prohibited (15). The grammatical status of (17) largely depends on whether or
not it is pronounced ‘in one breath,’ that is, as a single intonation unit, which is
also the case with SC (11).3 (Of course, the longer the phrase, the less possible
it will be to pronounce it ‘in one breath.’)
(15) ?*John read to the class the book.
(16) John read to the class the book that his teacher loaned to his best
friend.
(17) ??John read the book that his teacher loaned to his best friend to the
class.
Thus ‘heaviness’ is not an absolute criterion for external placement, but rather a

3. In the light of the analysis proposed here, the contrast between (i) and (ii), adapted from Bošković
(this volume), illustrates that there is no special position for heavy/external phrases inside the IP
kernel. In fact, this seems true regardless of clitic placement, as illustrated in (iii):
(i) Šta su oni njegovom najboljem prijatelju dali? (no pause)
what are they his best friend given
‘What did they give to his best friend?’
(ii) ?*Šta su oni, njegovom najboljem prijatelju, dali? (pause)
(iii) ?*Oni, njegovom najboljem prijatelju, dadoše auto (pause)
they his best friend gave car
‘They gave a car to his best friend.’
256 LJILJANA PROGOVAC

tendency, as expected under this approach. Also, as pointed out by Wayles

Browne, quoted in Bošković (this volume), even very short phrases can be set
off by comma intonation, in which case they show the same behaviour as heavy
phrases illustrated above: clitics occur in the third position (18), and wh-words
can follow the phrase (19):
(18) Petru, Jovan će prodati knjigu
‘To Peter, John will sell the book.’
(19) Petru, šta će Jovan prodati?
‘To Peter, what will John sell?’
In sum, the data discussed in this section are captured rather straightforwardly
given the following natural assumptions. Comma intonation after a phrase signals
that the phrase is outside the kernel clause, due to the principle in (14). Any
preposed phrase that is set off from the rest of the clause by comma intonation,
and that can be followed by material in either C or SpecCP, is not part of the
kernel CP. Alleged instances of third position clitic placement are therefore still
instances of second position placement within the kernel CP.4
Bošković (this volume) in fact observes that Intonation phrases normally
correspond to CPs. The results of this paper follow from simply extending this
assumption to all the data. Without empirical evidence to the contrary, it is
conceptually desirable to have intonation phrases reflect syntactic structure. It is
likewise desirable for basic intonation phrases to be associated with an identifiable
clausal projection, say the highest extended projection of V. The approach outlined

4. Appositive phrases can be analysed in the same way. The data below are from Bošković (this
volume); question marks in (ii), (iii), and (iv) were added to reflect my judgments of slight marginality:
(i) *Ja, tvoja mama, sam ti obećala igraćku
I your mom am you promised toy
‘I, your mom, have promised you a toy.’
(ii) ?Ja, tvoja mama, obećala sam ti igraku
(iii) ?Šta sam ti ja, tvoja mama, obećala?
‘What did I, your mom, promise you?’
(i) is ungrammatical because the clitics are not supported within the kernel clause. This follows if the
appositive phrase is pre-CP, given that it is set off with comma intonation. This explanation is
corroborated by the fact that wh-words and question words can follow the appositive:
(iv) ?Ja, tvoja mama, šta sam ti ono obećala?
*‘I, your mother, what is it that I promised you?’
(v) ?Ja, tvoja mama, da li sam ti nešto obećala?
*‘I, your mother, did I promise you something?’
On the other hand, (ii) is grammatical because the verb supports the clitic from within the kernel CP.
(iii) can be analysed as involving IP adjunction of the appositive phrase, another external position in
the extended projection of V; the clitics are supported by the wh-word.
 WHERE DO CLITICS CLUSTER? 257

in this section meets both of the above conceptual requirements. In addition, it

has wider empirical coverage. It accounts not only for the facts concerning clitic
placement and comma intonation, but also for the interaction between wh-for-
mation, comma intonation, and clitic placement. Empirically, the approach that
defines clitic placement solely on the basis of intonation boundaries has nothing
to say about the correlations with wh-word placement discussed above.

4. Concluding remarks

The arguments presented in this paper remove the motivation for an intonation-
based characterisation of clitic-second in SC. Bošković’s arguments for an
intonation-based placement of SC clitics, contra a syntactic placement, can only
make sense if the two can be shown to work autonomously and independently of
each other, and if a theory of intonation boundaries can be developed that is
blind to syntactic structure.
The approach outlined in this paper reveals a striking correlation between
syntactic positions and intonation boundaries. It confirms the conceptually most
desirable scenario: that there is a tight fit between syntactic positions and
intonation boundaries. Pursuing the study of clitic placement in SC along these
lines can give invaluable insight into the nature of the syntax/phonology inter-
face, the insight that no other phenomenon in the grammar may be able to
provide.

References

Bošković, Ž. 1995. ‘Participle movement and second position cliticization in Serbo-

Croatian.’ Lingua 96: 245–266.
Bošković, Ž. This volume. ‘Second position cliticisation: Syntax and/or phonology?’
Browne, W. 1974. ‘On the problem of enclitic placement in Serbo-Croatian.’ In: Brecht,
R. and C. Chvany (eds.) Slavic Transformational Syntax. Ann Arbor: University of
Michigan, 36–52.
Caink, A. 1997. ‘Extended projections in South Slavic.’ Paper presented at the Second
Conference on Formal Approaches to South Slavic Linguistics, Sofia.
Ćavar, D. and C. Wilder 1994. ‘Long head movement? Verb movement and cliticization
in Croatian.’ Lingua 93: 1–58.
Comrie, B. 1989. Language Universals and Linguistic Typology: Syntax and Morphology
(2nd edn.) Chicago: Chicago University Press.
Dimitrova-Vulchanova, M. 1995. ‘Clitics in Slavic.’ Studia Linguistica 49: 54–92.
258 LJILJANA PROGOVAC

Franks, S. 1998. ‘Clitics in Slavic.’ Position paper for the Comparative Slavic Morpho-
syntax Workshop, held at McCormick’s Creek State Park, 5–6 June. To appear in
Proceedings.
Franks, S. and L. Progovac 1994. ‘On the placement of Serbo-Croatian clitics’ In: Indiana
Slavic Studies 7, Proceedings of the 9th Biennial Conference on Balkan and South
Slavic Linguistics, Literature and Folklore, 69–78.
Kayne, R. S. 1989. ‘Null subjects and clitic climbing.’ In: Jaeggli, O. and K. J. Safir
(eds.) The Null Subject Parameter. Dordrecht: Kluwer, 239–261.
Progovac, L. 1996. ‘Clitics in Serbian/Croatian: Comp as the second position.’ In:
Halpern, A.L. and A. Zwicky (eds.) Approaching Second: Second Position Clitics
and Related Phenomena, Stanford, Calif.: CSLI Publications, 411–428.
Progovac, L. 1998a. ‘Clitic clusters and coordination’ In: Austin, J and A. Lawson (eds.):
Proceedings of the Fourteenth ESCOL 1997, Ithaca: Cornell University, 161–169.
Progovac, L. 1998b. ‘Eventive to and the placement of clitics in Serbian.’ Paper presented
at the First Conference on Linguistic Theory in Eastern European Languages
(CLITE 1), Szeged, Hungary.
Radanović-Kocić, V. 1988. The Grammar of Serbo-Croatian Clitics: A Synchronic and
Diachronic Perspective. Doctoral dissertation, University of Illinois.
Rivero, M.-L. 1994. ‘Negation, imperatives and Wackernagel effects.’ Rivista di Linguist-
ica 6.1: 39–66.
Stjepanović, S. 1997. ‘Clitics and VP ellipsis in Serbo-Croatian.’ Paper presented at
FASL (Formal Approaches to Slavic Linguistics), University of Connecticut, Storrs.
Tomić, O. M. 1996. ‘The Balkan Slavic clausal clitics.’ Natural Language and Linguistic
Theory 14: 811–872.
Wilder, C. and D. Ćavar 1994a. ‘Long head movement? Verb movement and cliticization
in Croatian.’ Lingua 93: 1–58.
Wilder, C. and D. Ćavar 1994b. ‘Word order variation, verb movement, and economy
principles.’ Studia Linguistica 48: 46–86.
 Clitic Doubling Constructions
in Balkan-Slavic Languages*

Ivanka Petkova Schick

University of Potsdam

Abstract

Clitic pronouns in Bulgarian are shown to be adjuncts to functional categories:

F (or Mod) in CP and D in DP. The functional category F triggers raising of
topics from VP and predetermines their information-structural interpretation as
discourse-given entities which represent uniquely identifiable individuals.
Using a Bierwisch type semantics, Bulgarian clitic pronouns are shown to
consistently act as topic markers and in this participate in the interpretation of
information structure.

1. Introduction

In morphologically rich languages like German or the Slavic languages, gram-

matical functions such as ‘subject’ or ‘object’ are not as rigidly expressed by
fixed word order as they are in English. Recent analyses show that variability in

* This research was supported by the German Science Foundation (DFG) and is part of work in
progress. It has benefited from presentations at the Second Conference on Formal Approaches to
South Slavic Languages (Sofia, September 26–28, 1997), the Second European Conference on Formal
Description of Slavic Languages (University of Potsdam, November 20–22, 1997), the SFB-Kollo-
quim KLITIKA (Heinrich-Heine-Universität Düsseldorf, May 15, 1998), the Comparative Slavic
Morphosyntax Workshop (McCormick’s Creek State Park, Spencer, Indiana University, June, 5–7,
1997) as well as the 31th Annual Meeting of the Societas Linguistica Europaea (St Andrews, August
26–30, 1998). Frits Beukema’s and Gerd Jägers’s helpful comments on various aspects of this paper
have contributed greatly to an improved presentation of the issues. The responsibility for errors is of
course my own.
260 IVANKA PETKOVA SCHICK

word order is relevant with respect to the principle of information structure. In

the spirit of Steube (1997) information structure is understood as the situational
and textual positioning of utterances in coherent utterance sequences. Beside
variability in word order, Bulgarian, which represents a typical Balkan-Slavic
language, uses articles and pronouns as well as intonation, and especially,
accentuation strategies for the purpose of information structuring.
In what follows I will discuss the phenomenon of pronominal clitic dou-
bling, focusing on its relation to the information structure of Modern Bulgarian.
According to a number of recent publications, topical arguments of this language
with specific reference can be substituted or doubled in sentences (CPs) by the
Dative and Accusative clitic forms of personal pronouns (cf. Rudin 1990/91,
1997; Penčev 1993; Guentchéva 1994; Angelova 1994; Dimitrova-Vulchanova
1995; Dimitrova-Vulchanova and Hellan 1996). Consider the correspondence
between the clitic pronouns1 and their counterparts in (1a) go–Ivo; in (1b) the
clitic sequence mu go–na nego pismoto; in (1c) mu–na Ivo; and also in (1d)
vi–na vas:
(1) a. Ivo *(go) običa Rada (≠ Ivo običa Rada)
Ivo *(--3-/ loves Rada
‘Rada loves Ivo.’ (≠ ‘Ivo loves Rada.’)
b. RADA *(mu go) dade *(na) nego
Rada *(--3-/ --3-/ gave *( him
pismoto;
letter-the
*Mu go dade Rada …; Rada bi mu go dala …; ??Rada mu go bi
dala …; Včera mu go dade Rada …
‘RADA gave him the letter’; ‘Rada would give him the letter’;
‘Yesterday Rada gave him the letter’
c. (Na) Ivo/nego *(mu) e studeno; Studeno*(mu)
( Ivo/him *(--3-/ is cold
e (na) Ivo
‘Ivo/He is cold.’
d. POMAGA *(vi) (na) vas Ivo; (Na) vas *(vi)
helps-he *(--2 ( you Ivo
pomaga IVO
‘Ivo HELPS you.’; ‘IVO helps you.’

1. In this paper the pronominal clitics are indicated by boldface italics, while capital letters indicate
the focused constituent.
 CLITIC DOUBLING CONSTRUCTIONS IN BALKAN-SLAVIC 261

The comparison of the examples in (1) and (2) demonstrates that in noun phrases
(DPs) clitic doubling is reserved for the Dative clitics which co-occur with
Possessor na-phrases,2 as given in (2):
(2) a. [[[kola][ta]] [mu]] (na Ivo); *mu kolata na Ivo
[[[car-the [--3-/ ( Ivo
‘Ivo’s car’
b. [[[nova][ta]] [mu]] bjala kola (na Ivo)
[[[new-the [--3-/ white car ( Ivo
‘the new white car of Ivo’
c. [[tazi] [mu]] nova kola (na Ivo); *novata mu
[[this [--3-/ new car ( Ivo
tazi kola
‘this new car of Ivo; this new car of Ivo’s’; ‘*the new this car
of Ivo’
The clitic pronouns in Bulgarian belong to the group of so-called special clitics
(Zwicky 1977, 1985) which exhibit a peculiar syntactic behaviour. The expres-
sions in (1) and (2) show that the clitic pronouns tend to be on the left periphery
of CPs and DPs, respectively. At the same time, however, they do not occur on

2. Bulgarian exhibits a lack of formal difference between the Dative clausal (CP) clitics and the
Dative possessive (DP) clitics. This phenomenon can be traced back to the unification of Genitive
and Dative (Mirčev 1963). Semantically, the possessive clitics relate to the corresponding possessive
pronouns, as in (i) — cf. also appendix 1 and appendix 2 at the end of this paper:
(i) negovata kola/kolata mu
his-the car/ car-the --3-/
‘his car’
In contrast to the examples in (2) the expression in (ii) shows that in written Bulgarian the non-
reflexive possessive clitics do not co-occur with possessive pronouns (which are bearers of adjectival
inflection and represent NP-modifiers):
(ii) negovata *mu kola
his--the --3-/ car
‘his car’
Why should this be so? I suggest that doubling clitics relate to (a part of) the reference situation
variable (note 15). They are event-oriented. If we accept that prepositional phrases yield events and
the doubled Possessor na-phrase is topical, then the inherently topical doubling clitic mu is
compatible with the Possessor DP na Ivo, as in (2). In (ii), however, clitic doubling is not allowed,
because possessive pronouns, such as negova(ta), do not satisfy the requirement for a proper
structural environment to yield event entailment (cf. Kratzer 1994). They are predestined to get a
focal interpretation (Schick 1996b). In the context in which the syntax allows the possessive pronoun
as well as the possessive clitic to be coindexed with a given antecedent, the choice between them is
motivated by discourse considerations (cf. Ariel’s 1990 analysis of anaphora, defined in terms of
sentence Topics).
262 IVANKA PETKOVA SCHICK

the left edge of CPs and DPs, as shown in (1b) and (2a): In CP-structures the
pronominal clitics are syntactically construed with the main verb and are oriented
towards the inflected forms of its functional projections. This is illustrated by the
periphrastic constructions with the auxiliary ‘be’ in conditionals — cf. bi (2/3 ps.
sg.) in (1b). Given that clitic pronouns are usually enclitics (Tilkov 1977;
GSBKE 1983), we see that in (1) they are attached to the right of whichever
preceding accented unit such as subjects (RADA) and adverbs (včera), which do
not form part of the noun phrase whose properties the clitics mark. Within
DP-structure, the Dative clitic cliticises to the lexical head, signalling definite-
ness, which is marked as a rule on the leftmost i.e. hierarchically highest lexical
head in the nominal string — cf. the noun kolata and the adjective novata
incorporating the definite article -ta or demonstrative determiners, such as tazi.
In order to explain the underlying structures of expressions in which a clitic
co-occurs with a coreferential lexical or pronominal DP, I will address the
properties of clitic pronouns using some recent proposals about the role of syntax
and lexicon in the division of labour between sound and meaning (cf. Bierwisch
1987, 1988, 1990a, 1996, 1997). It will be shown a) what lexical information is
relevant; b) which syntactic movements take place; c) how agreement works
between the Determiner, Adjective and Noun on the one hand, and between the
doubled (na)-phrase and the doubling clitic on the other; and d) what the
semantic contribution of the constituents of such expressions is. It is important
to note here that in cases like (1) the use of pronominal clitics is obligatory. This
fact will be a central concern in the analysis of the properties of these lexical
items and the related mechanism of clitic doubling of object DPs as a means of
information structuring in the corresponding Bulgarian expressions. Additional
support for this analysis is found in sentences such as (1c) and (1d), which
demonstrate the possibility of omitting the preposition na.
The analysis of the Bulgarian constructions with clitic doubling will be
developed in a comparison with Macedonian, a closely related Slavic language.
The idea behind this is to highlight a number of important typological aspects of
the phenomenon within the context of the Balkan languages (Schick 1996b).

2. Theoretical Background

I will adopt the Minimalist Program of Chomsky (1995), enriched by the level
of Semantic Form (SF) in the sense of Bierwisch’s two-level semantics (1987,
1988, 1990a, 1996, 1997). In contrast to the level of Conceptual Structure, SF
 CLITIC DOUBLING CONSTRUCTIONS IN BALKAN-SLAVIC 263

represents the grammatically determined meaning of linguistic expressions.3

According to Bierwisch’s framework, provided in Figure 1, sound and meaning
are paired by way of the interaction between Phonetic Form (PF) and SF. The
lexicon plays a key role in providing the relevant information.

Lexicon
Merge
Move
PF SF

Figure 1
The operations Move and Merge are crucial for the morphosyntactic structure
and are therefore activated at the interface levels in both directions. The organis-
ation of the surface structure is affected by the transition from the morphosyn-
tactic structure to PF, while SF relates grammatical structures to a more articulat-
ed level of meaning.
I assume that when pronominal clitics are used, general, well-motivated
principles apply with regard to the saturation of the argument structure of lexical
heads, the binding of referential arguments, the merging of modifiers and heads
and the information structure from the standpoint of the grammatical system. I
consider information structure to be determined by the interplay of discourse-
related aspects of utterances with the grammatical sentence representations on
different grammatical levels.

3. Lexical Entries for Pronominal Clitics

According to the Minimalist point of view on lexical information (Bierwisch

1996, 1997), the lexicon exclusively comprises all the idiosyncratic properties of
the lexical items of individual languages. Lexical items are represented as
complex data structures. Their basic components and general organisation are
provided by the principles of Universal Grammar. Lexical entries consist of basic
units in terms of which the correspondence between PF and SF is to be comput-
ed. They contain information about the specific contribution of each unit to PF

3. ‘SF is considered to be a computational level of grammar, and CF is the level of reasoning that
may draw on any kind of mental operation.’ (Wunderlich 1997: 30)
264 IVANKA PETKOVA SCHICK

and SF, as well as about each unit’s grammatically determined ability to combine
with other elements in more complex expressions. Thus, lexical entries can be
considered as programs themselves which determine local conditions for the
combinatorial processes.
The lexical properties of pronominal clitics are assumed to be derived from
the essential structural parallelism between CPs and DPs involving such ele-
ments, and are represented in their lexical entries, as shown below for the Dative
clitic 3 person, singular, masculine/neuter mu:4
(3) /mu/, [[ ]p _ ]p v [ _ [ ]p ]p
+D −V+N +MAX +max +definite +specific +topic [+hr, +lr]
x with x ∈ O
In addition to the phonological form of this item, the first line provides the
prosodic context in which the clitic needs to find a host (cf. Inkelas 1990). The
data in the second line refer to its categorial and grammatical properties (mor-
phological, syntactic, semantic): Pronominal clitics belong to the functional
category [+D(eterminer)] and represent D-heads of DPs. The features [+MAX]
and [+max] specify that clitic pronouns are syntactically non-projecting and
morphologically non-analysable formatives (cf. Muysken 1982). They are
referentially definite, specific, and they are anaphoric entities signalling Topics.
Below I discuss these features in detail and also focus on the features [+h(igher)
r(ole)] and [+l(ower) r(ole)] which pertain to the Case-properties of lexical
entries as a part of the linking mechanism of the Lexical Decomposition
Grammar (LDG) developed by Joppen and Wunderlich (1995) and Wunderlich
(1997). According to Minimalist Morphology, which makes extensive use of
underspecification, the lexical entry of mu does not include explicit information
about ‘m(asculine)/n(euter)’, ‘3rd person’ and ‘singular’, because features with
negative values are added by rule, whereas those with positive values are part of
the lexical entry. The representation of meaning in the third line of the lexical
entry states that pronominal clitics are interpreted as individual variables. Hence,
they may saturate argument positions

4. See the detailed theory of lexical structure developed in Bierwisch (1996, 1997), Wunderlich (1997)
and Stiebels (1996), as well as the Minimalist Morphology in Wunderlich (1994) and Wunderlich and
Fabri (1995). The lexical entry in (3) represents a slight modification of the proposal in Schick and
Zimmermann (1995, 1997), in which we subscribe to the credo one form — one meaning.
 CLITIC DOUBLING CONSTRUCTIONS IN BALKAN-SLAVIC 265

4. Referential Properties

In the Balkan-Slavic languages clitic pronouns occur as lexical items with

independent reference in first and second person and also as replacements of
proper names and descriptive DPs. In traditional grammars, it is claimed that
clitics are preferred to full pronominal forms because of the Economy Principle
(Andrejčin 1952; Nicolova 1979), which Cardinaletti and Starke (1995) aptly dub
Minimise Structure. Pronominal clitics can indeed be regarded as deficient
structural units with severely restricted structural properties, a fact that lies at the
heart of their prosodic ‘need-of-a-host”, and especially their idiosyncratic
syntactic behaviour.5

4.1 Specificity

In Bulgarian grammar the crucial condition for the use of clitic pronouns appears
to be the presence of specific (in the sense of ‘sufficiently individualised’
referents — cf. Ivančev 1957/78, Pašov 1978; Nicolova 1986; Penčev 1993).
First and second person clitics always meet this condition; however, it also has
to be satisfied when third person clitics substitute or double DPs.
As for the much-discussed notion specificity, I share the view that this
notion involves the referential properties of noun phrases and clauses, where the
identification of the item’s uniqueness by the speaker is central (cf. von Heu-
singer 1997):6 non-specific DPs may not be doubled. The close relationship
between the categories of specificity and definiteness is shown by the fact that
referential definite DPs normally appear to be specific (cf. Donnellan 1966).
Bulgarian also distinguishes between specific and non-specific indefinite
7
DPs. The comparison between (4b) and (5) demonstrates that indefinite indirect
object DPs, such as na edno dete in (5), need to be specific, whereas direct object
DPs marked by the indefinite article edin may have specific or non-specific

5. Schick (1994, 1996b) and Schick and Zimmermann (1995) deal extensively with the distribution
of pronominal clitics.
6. The necessity of a discourse-pragmatic anchoring of reference is accounted for in von Heusinger
(1997) by way of a semantic representation of the reference of linguistic expressions in which
semantic aspects and discourse-pragmatic aspects are combined in a unified analysis.
7. ‘Specific readings of indefinite NPs are understood as those in which the NP denotes a specified
object, while an indefinite NP in a non-specific reading refers to an arbitrary object which cannot be
further identified’ (von Heusinger 1997).
266 IVANKA PETKOVA SCHICK

referents, such as edin lekar in (4b).8 I take the occurrence of the so-called
indefinite article edin ‘a, one’ as a diagnostic for specificity, while a phonetically
empty D usually indicates non-specific reference (cf. Longobardi 1994; Penčev
1993).9 Thus, in (4)–(6) ungrammaticality ensues if there is no agreement
between the doubling clitic and the doubled DP in terms of the D-feature
[+specific]:
(4) a. Rada (go) tǎrsi pismoto/ edno pismo/ *pismo
Rada (--3-/ looks for letter-the a letter *letter
‘Rada is looking for the letter/a certain letter/*letter.’
b. Tărsja (go) edin lekar ≠ Tărsja (edin) lekar
seek-I (--3 / a doctor seek-I (a doctor
‘I am looking for a certain doctor (= a specific one).’ ≠ ‘I’m
looking for a doctor (= any)’ ‘(= any doctor).’
(5) Rada (mu) pomaga na deteto/ edno dete/ *dete
Rada (--3-/ is helping  child-the a child *child
‘Rada is helping the child/a certain child/*child.’
(6) uredi*(te) ì na laboratorijata/ edna
instruments-the --3-f  laboratory-the a
laboratorija/ *laboratorija
laboratory *laboratory
‘the instruments of the laboratory/of a laboratory/*of laboratory’
In (6), DPs containing Dative clitics are affected by the definiteness effect
(Budagov 1958), so that indefinite DPs are excluded. In the Balkan-Slavic
languages definiteness has to be marked on the hierarchically highest lexical head
of DPs. The expressions in (2) and (6) demonstrate that the possessive Dative
clitic always appears to the right of the definite D-head, thus occupying the right
edge of D. Within DP-structures the Dative clitic pronoun appears in second
position and together with the definite marked item forms one prosodic group, as
was shown in (2).
The occurrence of Bulgarian pronominal clitics is not restricted by the type

8. As to the ambiguity of the expressions in (4b), the [+specific] reading of the indefinite article edin
depends on the discourse-reference characteristics of the lexical head-noun. In such cases, the use of
anaphoric pronominal clitics acts as a disambiguating device, because these explicitly mark uniquely
and existentially presupposed individuals.
9. The statement that the article may be missing must be qualified to the extent that generically
referring na-phrases or DPs containing numerals may also get a [+specific] interpretation, as shown
in (14) and (16b) below.
 CLITIC DOUBLING CONSTRUCTIONS IN BALKAN-SLAVIC 267

of referentiality found in DPs with a definite article, demonstrative pronouns,

proper names, quantifiers such as vsički ‘all’ or full forms of personal pronouns
in Accusative and Dative. Rather, they are capable of relating to referents which
may encode specificity, even minimally so.10 The expressions in (7), (8), (13)
and (16b) demonstrate that interrogative (Drubig 1994), indefinite and negative
pronouns as well as DPs containing numerals (cf. Ritter 1995) which are
doubled by clitics are also to be regarded as having specific reference. In the
case of negative pronouns their presupposed existence is rejected. In DPs with
partitive Dative clitics the definiteness requirement is implicitly satisfied at the
semantic level:
(7) a. Koja (mu) kola na săseda
which (--3-/ car  neighbour-the
‘Which car of the neighbour’s cars.’
b. Malkite (ì) na koja/ njakoja/ nikoja kotka
young-the (--3-f  which some no cat
‘The kittens of which/some/no cat.’
c. Koj/ Njakakăv/ Nikakăv (mu) prijatel na Ivo
which some no (--3-/ friend  Ivo
‘What friend of his (Ivo’s)? / Some friend of his (Ivo’s) / None
of Ivo’s friends.’

10. The use of deficient pronouns such as reflexive clitics ‘instead of’ pseudo-arguments is very
typical of Slavic languages (cf. Cinque 1988; Růžička 1992; Reinhart and Reuland 1993; Schick
1996b; Progovac 1998). In this class we find, for example, expressions with a zero subject position,
as in (i), and also the anti-causative (ii) and impersonal (iii) constructions in which the reflexive clitic
se is event-related. This expletive se cannot be replaced by the full pronominal form of the reflexive
pronoun, nor appear together with it. I suppose that se represents a legitimate entity with a zero
interpretation at LF:
(i) Svečeri se
fell-night --
‘It got dark./Evening fell.’
(ii) Kolata se povredi
car-the -- broke down
‘The car broke down.’
(iii) Tuk ne se puši
here  -- smokes
‘No smoking here.’
268 IVANKA PETKOVA SCHICK

(8) a. Kogo (go) čaka Ivo? colloquial: Koj11

whom (--3-/ awaits Ivo who
*(go) čaka Ivo?
*(--3-/ awaits Ivo
‘Who(m) is Ivo waiting for?’
b. Na kogo (mu) pomaga Ivo? colloquial: Na
 whom (--3-/ is helping Ivo 
koj (mu) pomaga Ivo?
who (--3-/ is helping Ivo
‘Who(m) is Ivo helping?’

4.2 Topicality

The examples given so far emphasise the relation between doubling clitics and
specificity. A more elaborate analysis, however, points to one very essential
property of the anaphoric clitic pronouns in Bulgarian: with respect to the
referential properties of DPs, they signal in a consistent way that a new interpre-
tational perspective is introduced in which the referentially specific doubled
object phrase is presupposed in the discourse as a given entity (cf. Krifka
1991/92, Jäger 1995). A comparison with Macedonian will elucidate this claim.

4.2.1 Clitic doubling differences in the Balkan-Slavic languages

In marking the referential properties of DPs, Bulgarian and Macedonian differ
with respect to clitic doubling primarily in that in written Macedonian (Mac.) this
phenomenon is found in all contexts with definite object phrases — whether
topical or not. The omission of pronominal clitics leads to ungrammaticality:
(9) a. Rada *(go) bara Ivo (Mac.)
Rada *(--3-/ is looking for Ivo
‘Rada is looking for Ivo.’
b. Rada *(mu) pomaga na deteto (Mac.)
Rada *(--3-/ is helping  child-the
‘Rada is helping the child.’

11. In everyday Bulgarian and Macedonian the widespread use of nominative forms of non-clitic
pronominals like e.g. koj (Nominative) and na koj ( + Nominative) instead of the standardised
kogo (Accusative) and respectively na kogo ( + Accusative for Dative) is characteristic for the
ousting of Case-forms also in the area of pronominals. The same applies to the pronouns nikoj ‘no
one, no(t) (any)body’ and njakoj ‘some, somebody, someone’. In this way, the doubling clitic
contributes to the identification of the q-role of the doubled phrase.
 CLITIC DOUBLING CONSTRUCTIONS IN BALKAN-SLAVIC 269

c. *(Mu go) dadov pismoto nemu (Mac.)

(--3-/ --3-/ gave-I letter-the him
‘I gave him the letter.’
Macedonian exhibits a type of tautological marking of definiteness by pronomi-
nal clitics. This effect shows similarities to the repetition of the definite article
in the modifier phrases of complex DPs in Modern Greek. Conversely, clitic
doubling in Bulgarian (Bg.) systematically interacts with information structure:
In (10a) and (11a) below, the object phrase is the Focus of the missing clitic
pronoun in the base-generated SVO word order. The same applies to all cases
with inverted word order in which the focused object phrase is marked prosodic-
ally. Analogously, the object phrases in (10b) and (11b), assigned to the Focus
domain by the Q(uestion)-particle li, do not allow doubling by a clitic pro-
noun.12 Thus, according to this Focus effect, clitic doubling can be perceived
as mandatory if the subject DP as well as the predicate or an adverbial phrase
are in Focus, as in (1b), (1d),13 (10c) and (11c):
(10) a. Ivo *go napisa PISMOTO; PISMOTO *go napisa
Ivo *--3-/ wrote letter-the
Ivo (Bg.)
‘Ivo wrote the LETTER.’
b. PISMOTO li *go napisa Ivo? (Bg.)
letter-the -part *--3-/ wrote Ivo
‘Did Ivo write the LETTER?’
c. IVO go napisa pismoto; Pismoto IVO go napisa; Ivo go
NAPISA pismoto (Bg.)
‘IVO wrote the letter.’; ‘Ivo WROTE the letter.’
(11) a. Rada *mu pomaga na IVO; Pomaga *mu
Rada *--3-/ is helping  Ivo
Rada na IVO (Bg.)
‘Rada is helping IVO.’

12. Cf. Schick (1996b) on expressions with negation or other operator-like modifiers such as samo
‘only’. For more details on the properties of so-called Focus-sensitive scope particles, see Reis and
Rosengren (1997).
13. In the Balkan-Slavic languages, the doubling of direct object phrases is similar to that found in
Albanian, see Kallulli (1995). On more differences, cf. Tomič (1996) and Billings and Baermann
(1997).
270 IVANKA PETKOVA SCHICK

b. Na IVO li ??mu pomaga Rada? (Bg.)

 Ivo -part --3-/ helps Rada
‘Is Rada helping IVO?’
c. RADA mu pomaga na Ivo; Pomaga mu RADA na Ivo; Na Ivo
mu pomaga RADA; Rada mu POMAGA na Ivo; POMAGA
mu Rada na Ivo (Bg.)
‘RADA is helping Ivo’; ‘Rada is HELPING Ivo.’
Sentences with clitic doubling encode different presuppositions from those
without. A discussion of Focus–Background structure (cf. Drubig 1994; Reis and
Rosengren 1997; Steube 1997 and Zybatow 1997) falls outside the confines of
this paper, but it is crucial to observe that the Focus of the question yields a
partition of the interpretation of an expression into a Background part and Focus
part. In (12), the Focus of the answer does not participate in the Background of
the question in (12b). For that reason only the questions in (12a) are felicitous,
whereas the questions in (12b) violate the required congruence of non-focal
constituents and their presuppositions (cf. Rooth 1992):
(12) Ivo go napisa pismoto (Bg.)
Ivo --3-/ wrote letter-the
‘Ivo wrote the letter.’
a. What happened to the letter?; What did Ivo do with the letter?;
Who wrote the letter?
b. #What happened?; #What is new?; #What did Ivo write?
In contrast to Bulgarian, the Macedonian clauses in (9), (10′) and (11′) reveal
that clitic pronouns cannot function as Topic markers in the mandatory doubling
of [+definite] object DPs. Additional support for this analysis is provided by
comparing the presuppositions of (12) and (12′). In Macedonian all the questions
in (12a–b) are perfectly appropriate:
(10′) a. Ivo go napisa PISMOTO; PISMOTO Ivo go napisa (Mac.)
b. PISMOTO li go napisa Ivo? (Mac.)
c. IVO go napisa pismoto; Pismoto IVO go napisa; Ivo go
NAPISA pismoto; NAPISA go Ivo pismoto (Mac.)
(11′) a. Rada mu pomaga na IVO; Na IVO Rada mu pomaga (Mac.)
b. Na IVO li mu pomaga Rada? (Mac.)
c. RADA mu pomaga na Ivo; Na Ivo mu pomaga RADA;
Pomaga mu RADA na Ivo; Rada mu POMAGA na Ivo;
POMAGA mu Rada na Ivo (Mac.)
 CLITIC DOUBLING CONSTRUCTIONS IN BALKAN-SLAVIC 271

(12′) Ivo go napisa pismoto (Mac.)

Ivo --3-/ wrote letter-the
‘Ivo wrote the letter.’
a. What happened to the letter?; What did Ivo do with the letter?;
Who wrote the letter?
b. What happened?; What is new?; What did Ivo write?
Thus, the widespread application of clitic doubling in Macedonian can be
accounted by recognising that pronominal clitics are primarily and consistently
used as a grammaticalised means of emphasising the referential definiteness of
the doubled object phrase (cf. Koneski 1966; Anderson 1993; Dimitrova-Vulcha-
nova 1995; Schick 1994, 1996a, b).14
In contrast, Bulgarian pronominal clitics only double definite object phrases
which are topical. This explains why the Bulgarian clitic pronouns are inherent
carriers of the operator-like feature [+topic] (cf. Section 7) and represent a means
of clarifying how the pragmatically determined informational load is distributed
over the Bulgarian clause structure.

4.2.2 Clitic doubling similarities in the Balkan-Slavic languages

With the exception of the differences mentioned above, clitic doubling is similar
in the Balkan-Slavic languages. Bulgarian and Macedonian share complement
constructions headed by an indefinite article, as in (4) and (5), or complement
constructions with interrogative pronouns, as in (8), in which clitic doubling is
optional. Cases like these involve the interaction of doubling clitics with the
information structure, where the semantic and aspectual properties of the VP (cf.
Verkuyl 1993; Schmitt 1996; Steube and Späth 1998) and the actual characteris-
tics of the determiners in its DP-complements also affect the characterisation of
the reference situation variable in CPs.15

14. This is also borne out by the loss of person and number agreement between the clitic and the
doubled DP, which Koneski (1966) points out for some Macedonian dialects. In the following
sentences, person and number indications must be inferred from the relevant nominal or pronominal
full forms:
(i) mu rekov na majka mi
--3-/ told-I  mother --1
‘I told my mother’
(ii) mu rekov nim
--3-/ told-I --3
‘I told them’
15. The reference situation is determined by the actual context of discourse and may be (partly)
constrained by linguistic means. In line with Bierwisch’s (1987, 1988) compositional point of view,
272 IVANKA PETKOVA SCHICK

In Bulgarian the relation between direct object doubling and information

structure matches similar patterns in Macedonian and, to a lesser degree, in
Albanian. In these Balkan languages, the doubling phenomenon is not restricted
to definite DPs, as it is in Modern Greek. In Bulgarian, clitic doubling systemati-
cally serves as a means of information structuring, involving both the hierarchi-
cally higher indirect object DP and the lower direct object DP. In the other
Balkan languages these two complement DPs are treated differentially. In
Macedonian, only the more deeply embedded direct complement DP is sensitive
to the use of doubling clitics as a marker of the information structure. This is for
two reasons. The first, as I have already pointed out, is the tautological marking
of definiteness performed by pronominal clitics. The second reason, which
Macedonian shares with Bulgarian, is that an indefinite direct object phrase can
take a specific as well as a non-specific referent, while an indirect object DP is
always [+specific], as shown by the comparison of the data in (4) and (5). In DP,
the phenomenon of clitic doubling is clearly most strongly represented in
Bulgarian. We saw in (6) above that participants in what may be called general
possessor (pertinence) relations are not restricted to nouns denoting family or
family-like relationships as in Macedonian (Koneski 1966).
Given the data sketched here, Macedonian takes up an intermediate position
in the Balkan languages, while Bulgarian represents a characteristic ‘specimen’
of the Balkan area with respect to the information structuring in CPs involving
clitic doubling constructions (cf. Schick 1994, 1996b).

5. Pronominal Clitics as Inherent Topic and Case Markers

5.1 Doubling clitics in impersonal constructions with psych predicates

While the signalling of referential characteristics is robust, it is not the only

structural property of clitic pronouns. The following examples indicate that
pronominal clitics serve as inherent markers of existentially and uniquely
presupposed individuals and as Case-bearers.
In CP-constructions with psych predicates and a semantically empty argument
in subject position, as in (13), (14) and (1c), the pronominal clitic is non-omissible:

the (reference) situation variable ‘s’ always represents the highest argument of the verb and indicates
the instantiation between situations and propositions (cf. Wunderlich 1997; Steube 1997). Maienborn
(1997) provides a compositional account, which integrates the reference situation into the SF of CPs
— cf. also Bach’s (1986) notion of eventuality covering events, states and processes.
 CLITIC DOUBLING CONSTRUCTIONS IN BALKAN-SLAVIC 273

(13) (Na) kogo/ njakogo/ nikogo ne *(mu)

( whom somebody nobody not *(--1
se spi
-- sleeps
‘Who is not sleepy /somebody is not sleepy /nobody is sleepy.’
(14) (Na) čovek(a) *(mu) stiga samo edna dobra duma
( man-(the) *(--3-/ suffices just one good word
‘One needs (just) a good word.’
To the experiencer argument the clitics mark a true Accusative or Dative object
position which is subordinate to a non-thematic subject position occupied by a
semantically empty argument (cf. Bierwisch 1996). The clitic reduplicating the
complement represents an obligatory component, whereas the spell-out of the
doubled DP is optional. The non-omissibility of the clitic pronoun is configura-
tionally determined since it is only through the use of anaphoric clitics that basic
clausal requirements can be satisfied: using a clitic indicates that the structurally
“deepest” argument is the essential and the only true argument of the referent-
ially deficient expression that has a semantically empty, but hierarchically
“highest” subject position.
Furthermore, it is crucial that the experiencer is always presupposed as
given and relates to the reference situation of clausal propositions. Hence, neither
the doubled lexical head-noun by itself nor the Case-marked full pronominal
form, but only the clitic can fix the discourse-bound characteristics of the
referent: the anaphoric clitic pronoun is an inherent bearer of strong Case-feature
which at the same time it carries the operator-like [+topic] feature that signals
existentially and uniquely presupposed individuals.
To round off this part of the discussion, let me point out the following
important distinctions between clitic and full pronominal forms: Although non-
clitic pronouns — as opposed to nouns — are also Case-marked, their Case-
specification is weak, whereas the clitics bear the required strong Case-feature of
the experiencer argument (see Section 5.2). Moreover, only anaphoric clitics con-
tribute to the characterisation of the reference situation as Topic markers (cf. Sec-
tions 6.2 and 7), because they have event-related clausal propositions in their scope.

5.2 Case-marking in the Balkan-Slavic languages

Bearing in mind that word order in the Balkan-Slavic languages is not entirely
fixed, I will very briefly consider another group of expressions in which the need
for pronominal clitics also relates their q-role assigning function as Case bearers
274 IVANKA PETKOVA SCHICK

and Topic markers. In CPs with transitive predicates of the type ‘love’, as in (1a)
and (15), the preposing of the object phrase does not allow the correct identifica-
tion of the internal argument on the basis of constituent order alone: it is
necessary to double the DP by the clitic with the corresponding morphological
Case. Furthermore, the clitic pronoun indicates that the doubled DPs are given
entities, i.e. existentially and uniquely presupposed individuals:
(15) Edna žena *(ja) risuva edin hudožnik
one woman *(--3-f paints one artist
‘An artist is painting a woman.’ (≠ ‘A woman is painting an artist.’)
With regard to the lexical-grammatical contribution of pronominal clitics to the
informational structuring of the Bulgarian clause, clitic doubling is also consid-
ered to be natural (Ivančev 1957/78) in expressions like (16).16 In these cases,
in which the clitic is missing, the object function of the relevant non-focused DP
cannot be explicitly signalled (GSBKE 1983). This may result in subject–object
ambiguities:
(16) a. Edna kruša/ Krušata *(ja) risuva edno dete/ deteto
one pear pear-the *(--3-f paints one child/ child-the
‘A child/The child is painting a pear/the pear.’
b. Dve(te) knigi *(gi) četat CJAL mesec
Two(-the) books *(--3 read-they whole month
‘They has been reading (the) two books for a WHOLE month
now.’
In the type of analysis proposed in this article the selection and the hierarchical
ordering of argument positions and their modifiers as well as the morphological
and syntactic conditions imposed by the lexical head are crucial. The complete
representation of the information in the Argument Structure (AS) of the predicate
is achieved by a language-specific parameter setting of the morphological Case
associated with argument positions. In (3) I suggested that pronominal clitics
bear the relational Case-features [+hr] and [+lr] for structural Cases as a part of
the Argument linking approach in Lexical Decomposition Grammar (Wunderlich
1997). As first proposed by Kiparsky (1992), a single feature system expresses
both the abstract argument positions of a lexical item (represented by the
interfacing level of q-Structure) and the morphosyntactic properties of the linkers.

16. Vallduví (1992) provides an analysis of the remarkably similar behaviour of clitic doubling in
Catalan. Meinungen (1995) points out the parallelism between this phenomenon and scrambling in
German as a means towards the informational structuring of the clause (cf. also Haftka 1988).
 CLITIC DOUBLING CONSTRUCTIONS IN BALKAN-SLAVIC 275

In this way, the feature system determines which syntactic complement is

attached to which q-role. Strictly speaking, the hierarchy of argument variables
in SF determines the AS of the verb, as shown in (17a) below. In SF the
reference situation variable is denoted by ‘s’. According to LDG (Joppen and
Wunderlich 1995; Wunderlich 1997) the two binary features are interpreted as
follows: [± hr] means ‘there is a/no higher role’, while [± lr] means ‘there is a/
no lower role’. These features are not an addition to the q-roles but simply a
device that inherently encodes the position of structural arguments, as shown in
(17b) for (canonical) verbs with one, two, and three structural arguments. We see
that each position in the AS is specified by features indicating whether there
exists a lower and/or a higher position.17 The same features realise the structur-
al Cases which are not fully specified but only assigned [+]-valued features, as
given in (17c) for Bulgarian: Dative is fully specified with [+hr, +lr], Accusative
only gets [+hr], whereas Nominative is least restricted and remains unmarked.
The linking mechanism consists in the unification of features outlined in (17d).
Notice, that the linking of canonical verbs to the structural Cases takes place in
accordance with the Structural Linking principle,18 Specificity Principle19 and
Uniqueness Constraint20 (Wunderlich 1997):
(17) a. give SF: CAUSE (x, BECOME (POSS (y, z))) (s)
AS: lz ly lx ls
b. Features of the AS-positions of canonical verbs
i. intransitive ii. transitive iii. ditransitive
lx V(x) ly lx V(x, y) lz ly lx V(x, y, z)
−hr +hr −hr +hr +hr −hr
−lr −lr +lr −lr +lr +lr
c. Structural Case features in Bulgarian
: [+hr, +lr] : [+hr] : [ ]

17. The argument of an intransitive verb is represented as [−hr, −lr], because there is neither a higher
nor a lower role. The direct object of a transitive verb is assigned as [+hr, −lr] since there is no lower
role and because of the subject being the higher role. The subject in turn is represented by the reverse
feature values. Indirect objects are assigned with [+hr, +lr] because of the subject being the higher
role and the direct object being the lower role.
18. Only structural arguments have to be associated with structural linkers.
19. In each context, the most specific compatible linker is chosen.
20. Each linker can only be assigned once.
276 IVANKA PETKOVA SCHICK

d. Linking for canonical verbs (in Accusative–Nominative systems)

lx V(x) ly lx V(x, y) lz ly lx V(x, y, z)
−hr +hr −hr +hr +hr −hr
−lr −lr +lr −lr +lr +lr
↓ ↓ ↓ ↓ ↓ ↓
     
As Case morphology is lost over time, the Dative is replaced by the analytical
periphrases with the preposition na. In order to safeguard the required agreement
between the na-DP and the doubling Dative clitic, it is essential to distinguish
between the meaningful locative preposition na and the grammaticalised phrasal
affix na as a marker of Case-distinctions. This can be covered by two different
characterisations of the categorial status of na, as proposed in (18):
(18) /na/, [ _ [ ]p ]p
(−V−N)a ( [+hr, +lr])−a
(ly lx [x  [UP y]])a
In the SF of the locative preposition na, the value of the variable x in the
relation  is related to the variable y of the functional value UP. This means
that a locative preposition with the two argument positions lx and ly shows up
if in the semantic structure the argument x is located in the UP domain of y
(Bierwisch 1988). However, the element na, which has become devoid of
meaning and has specialised in signalling Dative Case, now only carries the
characterisation [+hr, +lr] and is consequently confined to the status of a phrasal
affix indicated as −a. As opposed to prepositions like săs in gord săs ‘proud of’,
which helps in identifying non-structural arguments, the phrasal affix na is
predictable and indicates a structural Case (Schick 1994, 1996b).
Since all prepositions govern Accusative, the interpretation of the noun
phrases na nego ‘to him’ or na Ivo ‘to Ivo’ must be marked as [+hr, +lr]. This
marking is determined by the characteristics which the na-formative conveys to
its nominal complement, as in (19):
(19) a. [DP na [DP nego]
[+hr, +lr] [+hr, −lr]
na nego ‘to him’
b. [DP na [DP Ivo]
[+hr, +lr]
na Ivo ‘to Ivo’
 CLITIC DOUBLING CONSTRUCTIONS IN BALKAN-SLAVIC 277

Na-dropping in CPs with a doubled object, as in (1c), (1d), (13) and (14),
implies the elimination of redundant pieces of information. Notice that in
comparison with the phrasal affix na the Dative clitic displays the stronger Case
feature morphologically. From the point of view of the information structure,
na-drop is fundamentally restricted to topical na-phrases (Vakareliyska 1995;
Schick 1994, 1996b, 1997).
The data in Macedonian support the following conclusion: na-drop is
exclusively allowed in the characteristic Topic-position on the left edge of the
clause, but only if na co-occurs with the Case-marked full form of the personal
pronouns. Thus, the Dative clitic, by bearing the strong Case-feature [+hr,+lr]
and the operator-like feature [+topic], inherently satisfies the structural require-
ments of expressions as in (1c) and (1d) in an optimal way.
Summarising, we conclude that in Bulgarian the spell-out of doubling clitics
can also be regarded as obligatory in all cases with na-drop.

6. The Base Position of Pronominal Clitics

The distribution of pronominal clitics in Bulgarian CPs and DPs illustrated in (1)
and (2), as well as the ill-formedness of constructions like (20), indicate that they
normally appear in the extended projections of nouns and verbs relatively far to
the left in relation to the other constituents in these constructions (cf. Pašov
1978; Rivero 1994; Dimitrova-Vulchanova 1995; Rudin 1997). Moreover, cases
like (21) and (22), in which adverbial phrases may intervene between the
pronominal clitic and the base position of the verb, demonstrate this tendency
quite clearly (cf. Schick 1997):
(20) a. *Roditelite pomagat vi na vas
parents-the help-they --2  you
‘Your parents help you.’
b. *Az davam mu go na Ivo pismoto
I give --3-/ --3-/  Ivo letter-the
‘I am giving Ivo the letter.’
(21) Edna mu s bilki ranata vărže
a- --3-/ with herbs wound-the dresses
‘One of them dresses his wound with herbs’
(22) treta go v usta celune bărže
third --3-/ in mouth kisses quickly
‘Third one kisses him on the mouth quickly’
278 IVANKA PETKOVA SCHICK

Notice, however, that differences in the base position of pronominal clitics

between languages have been postulated elsewhere in the literature. Using
Scandinavian data, Hellan and Platzack (1995) show that clitic pronouns may
occur in the canonical base positions of DPs.

6.1 General structural scheme

Bearing in mind that CPs and DPs are extended functional projections (FPs) of
lexical projections (LPs) of which V and N are the heads, our empirical findings
so far can be captured in the general structural schema in (23). The position
assigned to the clitic pronouns is represented first and foremost in keeping with
the traditional view that Bulgarian pronominal clitics are topical entities (cf.
Ivančev 1957/78, Pašov 1978; Nicolova 1986; Rudin 1990/91, 1996, 1997;
Guentchéva 1994; Schick 1994; Dimitrova-Vulchanova 1995) which usually
appear as enclitics (Tilkov 1977; GSBKE 1983). Secondly, and more in particu-
lar, the clitic position is uniform across the various syntagmatic types and reflects
the specific semantic and informational contribution of anaphoric clitics to the
interpretation of Bulgarian clauses (cf. Schick and Zimmermann 1995, 1997;
Schick 1996a):
(23) [ … [FP (XP) [F′ [F cl F ] … [LP … L … ] … ]] … ]
In (23) it is assumed that the pronominal clitic is base-generated as an adjunct to
a separate operator-like functional category F which is the head of FP, outside
the scope of the lexical projection LP. XP in SpecF represents a topical DP
doubled by the clitic.

6.2 Pronominal clitics in CP-structure

In CPs, FP is situated in the immediate vicinity on the right edge of the function-
al category Mod(al)P. I assume that the binding of the referential argument of
the verb ‘s’ takes place in ModP (note 15), because the event-related functional
projection ModP contains linguistic information about tense, aspect and verbal
mood and provides semantic constraints on the reference situation variable ‘s’
indicating the instantiation between situations and propositions (cf. Bierwisch
1988, 1990a; Wunderlich 1997; Steube 1997; Maienborn 1997; Frey and Pittner
1998). The configuration in (23) arises at LF at the latest. According to this
view, movement of the clitic is unnecessary.
However, with regard to the base position of the pronominal clitic in
Bulgarian CPs, I take into account a slightly modified version of (23), according
 CLITIC DOUBLING CONSTRUCTIONS IN BALKAN-SLAVIC 279

to which the non-projecting clitic entity is in the domain of ModP as an adjunct

on the right edge.21 In this approach there is no a separate functional category
F (contra Rizzi 1997). The topical XPs exhibit the operator-like feature [+topic]
which allows them to raise to the left periphery of CPs and activates the
mechanism of topical interpretation. In Section 7, I will focus on the first
version, which demonstrates the semantic aspects of clitic doubling.
The analysis of heterogeneous clitic clusters, as in (24a), provides support
for both proposals. For the multiple occurrence of pronominal clitics I assume
that several FPs enter the extended projection above LP successively. The order
of the clitic pronouns given in (25) is based on the Parallel Movement Principle
(Müller 1997), which is also responsible for the sequence of multiple interroga-
tive, relative, indefinite or negative pronouns, i.e. the argument hierarchy as fixed
by the semantic structure of the lexical head in VP. Thus, the possessive Dative
clitic mu ‘his’ which refers to a Possessor DP and is subordinated to the direct
object DP, appears to the left of the Accusative clitic go. In terms of LDG and
in accordance with Wunderlich (personal communication), (24b) represents the
Possessor extension of the verb complex săm dala ‘have given’. The order of l-
abstractors in LF corresponds to the hierarchy of the arguments and their
modifiers in the derived SF: go and mu, respectively, saturate z and its modifier
u, whereas ti saturates y.22 The clausal interpretation of the SF conjunction
follows from the addition of the semantic predicates BECOME POSS(ESSOR):
(24) a. [CP če ne [ModP săm ti mu
that not have-I --2 --3-/
go dalai] [VP ti pismoto na Ivo]]23
--3-/ given- letter-the  Ivo
‘That I have not given it to you (the letter of Ivo).’
b. lz lu ly lx {ACT(x) & BECOME POSS(y, z) & POSS (u, z)}
(25) … ( …  ) > indirect object > possessive > direct object > …

21. My thanks to Anita Steube (personal communication) for showing me her suggestion that this
proposal is indeed more economical (cf. also Steube 1997).
22. The reference situation variable ‘s’ is disregarded in (24b).
23. The sentence in (24) is a colloquial expression which is common in the region of Burgas, Varna,
etc. Specific constraints on the interaction of morphosyntax and PF are involved in non-homogeneous
clitic sequences containing the question particle li and the negation marker ne in CPs, as well as in
the choice of the definite article in DPs. In the latter case, it is primarily vowel harmony that
determines the form of the Balkan-Slavic definite article (cf. Schick 1994, 1997; Schick and Zimmer-
mann 1995).
280 IVANKA PETKOVA SCHICK

6.3 Dative clitics in DP-Structure

In DPs, the placement of the Dative clitic requires adjacency with a definite
determiner, as can be gathered from the comparison between (2) and the more
complex attributive expression in (26b) in which coordinated adjective phrases
also display the operation explained in (30a) below:
(26) a. grižlivo napisanoto mu na ruski pismo
accurately written-the --3-/ in Russian letter
na Ivo
 Ivo
‘Ivo’s accurately written letter in Russian’
b. [[jasno napisanoto i sgănatoto] mu] pismo
[[clearly written-the and fold up-the --3-/ letter
na Ivo; [[jasno napisanoto] mu]] i sgănato pismo na Ivo
 Ivo
‘Ivo’s clearly written and folded letter’
The proposed analysis is based on the empirical observation that the structure of
the NP and the characteristics of the D-head determine the structure of DPs. In
accordance with (23), I propose that in the underlying structure of Bulgarian DPs
the clitic pronoun is base-generated as an adjunct to the functional head D over
NP, and occupies a position on the right edge of D:24
(27) [DP D (XP) [D′ [D⁰ D0 (cl)] [NP [N′ … N …] ([DP na DP ])]]]
The doubled Possessor na-phrase explicating the doubling clitic is interpreted as
a daughter of NP.25 It represents an attributive postposed phrasal constituent
(XP)26 which participates in the characterisation of a reference situation variable,

24. Cf. the tree-representation of (27):

DP

D′

D0 NP

D0 cl

25. Cf. a similar proposal considered by Szabolcsi (1983, 1987).

26. The clitic doubling of the Possessor na-DP is optional. Because of the relatively free word order
in Bulgarian, however, we observe two cases in which the possessive clitic is non-omissible (Schick
 CLITIC DOUBLING CONSTRUCTIONS IN BALKAN-SLAVIC 281

because of its status as an event nominal. The examples in (2) are a special case
of existential constructions in which the Possessor represents a uniquely presup-
posed individual. In the spirit of Kayne (1993), den Dikken (1995, 1997) and van
Riemsdijk (1998), I assume that the Possessor is the predicate in the subject-
predicate relation between the Possessum and the Possessor. In the Balkan-Slavic
languages, the Dative case of the Possessor is identical with the [+hr, +lr] Case
marker, which is in keeping with the suggestion that in double object construc-
tions the theme and the goal enter into a possessive relationship.27 It was
already shown that in DPs the Dative clitic cliticises on a [+definite] host. In
Bulgarian, as opposed to other Balkan languages, definiteness marking is not
tautological. D remains phonetically empty if definiteness is not signalled by a
DP with a demonstrative determiner, as in (2c), but rather by the enclitic Balkan-
Slavic formative -t.28 As shown in (2a), -t belongs to the morphological struc-
ture of the lexical head N in NP (cf. Bierwisch 1990b). In more complex DPs,
-t appears on the hierarchically highest phrase with an adjectivally inflecting
head, cf. (2b).29 I assume that the morphological configuration with the definite

1997). The first one represents a colloquial construction which is restricted to Possessor nouns
denoting family relationships, as in (i). If a Possessor na-phrase with an inherently relational noun as
‘uncle’ overtly raises to the left edge of the embedding DP, the anaphoric Dative clitic is needed for
marking the designated Possessor of an additional modifying Possessor phrase. Its landing site is a
DP-internal position, as shown by the test of embedding it under a preposition:
(i) pred [na čičo *(mu) kăštata]; pred [na čičo kăštata *mu]
in front [ uncle --3-/ house-the
‘in front of his uncle’s house’
The phenomenon represented in the second case concerns both the Possessor and the object-status of
Dative clitics. If the doubled na-phrase is omitted, the anaphoric Dative clitic in examples of the type
in (ii) is also obligatory. Its use indicates that the non-pronominal non-focused DP in the sentence-
initial position Ivo, which sets the framework within which the main predication is contained, is
coreferential with the omissible doubled na-DP. I analyse such a sentence-initial DP as an instance
of external Topic: it is non-argumental and base-generated as an adjunct to the left edge of CP (Alter
and Junghanns 1997):
(ii) Ivo, az *(go) vidjah proekta *(mu) (na Ivo)./*Na Ivo, …
Ivo I *(--3-/ saw-I project-the *(--3-/ ( Ivo
‘Concerning Ivo: I saw his (Ivo’s) project.’
27. See den Dikken (1995).
28. Macedonian has two more variants of the definiteness marker at its disposal: -v and -n. The
Balkan-Slavic definiteness markers represent grammaticalised formatives which have become integrat-
ed in the morphological structure of the noun as clitic affixes. In Bulgarian -t is semantically empty
nowadays, while this process has not been completed yet in Macedonian, given the existence of the
three variants. These mark spatial differentiations originating in three deictic roots: -t-, -n- and -v-.
29. The position of the possessive Dative clitic in constructions with a complex adjective of the type
‘proud’ containing an event-related complement is determined by the requirements of the Parallel
282 IVANKA PETKOVA SCHICK

article is base-generated. The feature [+definite] is licensed either by adjunction

of N to D (cf. Longobardi 1994; Ćavar and Wilder 1994) or by raising the
adjective phrase to SpecD (cf. Gallmann 1996 for German). These operations,
which are comparable with the movement of heads to C or of XPs to CP, are not
visible in SF, see also (32). As shown in (28a) and (29a), cliticisation of the
pronominal clitic to the definite determiner satisfies the adjacency requirement.
The clitic pronoun stays in its base position as a non-projecting D-adjunct. The LF
of the examples in (2a) and (2b) is represented in (28b) and (29b), respectively:
(28) a. [DP [D Ni [D Ø cl] [NP … ti …]]] SS
[DP [D kolatai mu] [NP ti [DP na Ivo]]]
b. [DP D na Ivoi [D Ø mu] [NP kolata ti]] LF
kolata mu (na Ivo)
‘Ivo’s car’
(29) a. [DP APi [D′ [D Ø cl] [NP ti … N … ]]] SS
[DP [AP novata]i [D′ [D Ø mu] [NP ti kola [DP na Ivo]]]]
b. [DP D na Ivoi [D Ø mu] [NP novata kola ti]] LF
novata mu kola (na Ivo)
‘his (Ivo’s) new car’
In more complex attributive structures, as in (26), the phrase raised to SpecD and
its trace are subject to a complementary deletion of identical peripheral parts in
lexical chains, in accordance with Wilder’s (1994) operation in (30):
(30) [XP Y X — Z]i′ [XP Y
— X Z]i
[grižlivo napisanoto na
————]
ruski i′ [grižlivo
———— napisanoto
————— na ruski]i

6.4 The definite article

From a semantic point of view, definiteness is achieved in the functional head D

only after N’s referential argument is bound by the D-element. The definiteness
of the respective [+N] units is marked by the enclitic affix -t and is licensed by
a phonetically empty D. The [+definite] marker -t has no meaning of its own.
The proposal in (31) of a lexical entry for the definite article is based on the

Movement Principle. Hence, in the expression (i) the correct word order results only if the possessive
Dative clitic si cliticises to the [+definite] Possessum-noun pobedata ‘the victory’:
(i) gord(ijat) ot pobedata si narod; gord(ijat) *si ot pobedata narod
proud-(the) of victory-the -- nation
‘the nation proud of its victory’
 CLITIC DOUBLING CONSTRUCTIONS IN BALKAN-SLAVIC 283

assumption that the phonetically empty D stands for a complex of features which
determines the form, meaning and referential status of DPs (cf. Grimshaw 1991).
The -operator describes an ontological object with the property P (see Bier-
wisch 1987); ! means that the expression bound by has the interpretation of
an existentially and uniquely presupposed individual (cf. Steube and Späth
1998):30
(31) /Ø/
+D -V+N +MIN +definite +specific atopic
lP ! x [P, x]

6.5 Towards the interpretation of non-relational nouns

In the compositional view of the sound-meaning computation of linguistic

expressions adopted here (Bierwisch 1987, 1988, 1990a, 1996, 1997; Wunderlich
1991, 1997), the interpretation of non-relational nouns (cf. Löbner 1985) can be
achieved by Ortmann’s (1994) procedure of argument structure enriching,
described very briefly in (32a). In (32b) a non-relational noun kola ‘car’ has an
additional argument position besides its own referential position, so that in
expressions with a Dative clitic and a doubled na-phrase the noun ‘car’ acquires
the same morphosyntactic, semantic and information-structural status as relational
nouns and adjectives such as prijatel ‘friend’, gordost ‘pride’, zakăsnenie ‘delay’,
ljubim ‘favourite’ etc. in analogous expressions (Schick 1997):
(32) a. lQ lx lr [Q r] & [x R r]
−V+N
with R = general Possessor (Pertinence) relation
b. lx lr [CAR r] & [x R r]

7. The Lexical Entry of the Functional Category F

The placement of the pronominal clitic in a special functional domain FP over LP

has semantic effects, which are based on the information-structural status of the
clitic and the doubled DP. As mentioned above, Topics are those entities which
are presupposed in the relevant discourse as given. Following Jäger’s (1995)
analysis of the interrelatedness of scrambling, topicality and presuppositions of

30. Cf. also Heusinger’s (1997) notation i.

284 IVANKA PETKOVA SCHICK

definite and certain indefinite DPs, the lexical entry of the functional category F
in (33) is intended to be supplementary to the characterisation of pronominal
clitics as topical. Since the feature [+topic] has a function similar to [+wh] or
[+focus] features and triggers information-structural strategies in which the
occurrence of the Bulgarian doubling clitic is determined by discourse-bound
arguments, the semantics of the functional category F is a central issue. When
the SpecF position is filled by the doubled phrase, at LF at the latest, the
operator-like meaning of F as a topicaliser is activated. F acts as a mediator
between the lexical projection of the doubled DP and the pronominal clitic. In
keeping with the meaning components of F, the clitic pronoun and the doubled
DP (where this applies) have the status of Topics. In the lexical entry, both
expressions are represented by two argument places of F — lx and ly — and
are entities which identify each other semantically:
(33) /Ø/
+topic
ly lP (lx)a lr ([y = x] :)a [P y r]
k k +definite
k = +topic +governed ±oblique ±feminine ±neuter ± ±1 ±2
The asymmetric connector ‘:’ in the lexical entry of the functional category F
indicates that in relevant cases, the pronominal clitic, together with the explicat-
ing DP, is taken as given with respect to the rest of the semantic structure. In the
case of CPs, lP is specified by the meaning of VP, and in the case of DPs, lP
is specified by the meaning of NP. In addition to the referential argument, there
could be another unspecified argument of V or N which only emerges in FP. The
definiteness feature, which is linked to the argument position lr, requires that the
argument position has to be bound by a [+definite] DP. The argument address ‘k’
lists the requisite properties of the co-addressed argument positions lx and ly.
This treatment is based on the assumption that semantically the information
structure follows the division Topic/Comment (cf. Zybatov and Junghanns 1998),
which forms an argument/function structure. The Topic is the ontological object
that the sentence is about, whereas the Comment is the predication about the
Topic.31 On the SF-level topical DPs are interpreted as generalised quantifiers
by means of type-shifting operations. The use of anaphoric clitics bring in the
information that the corresponding DPs represent existentially and uniquely
presupposed individuals.

31. Cf. Späth’s (1997) consideration on this issue.

 CLITIC DOUBLING CONSTRUCTIONS IN BALKAN-SLAVIC 285

Since scrambling in the syntax can be considered to be type-shifting in the

sense of Partee (1987), the instantiation of a as ‘+’ yields a natural connection
between topicality, scrambling and pronominal clitics.32 In activating the
respective information-structure strategies of Modern Bulgarian, the doubling
clitics figure as licensers of topical DPs even though they do not necessarily
have to be overt (cf. also Rudin 1997).

8. Conclusions

In my analysis clitic pronouns in Bulgarian are adjuncts to functional categories:

F (or ModP) in CPs and D in DPs, contrary to other views in which they are
heads in the extended lexical projections of the verb or the noun. The functional
category F triggers raising of Topics from the lexical projection VP and thus
determines their information-structural interpretation as discourse-given entities
which represent uniquely identifiable individuals.
The clitic pronouns are base-generated in functional projections as a direct
consequence of their status as non-projecting adjuncts. This assumption, together
with the collocational and combinatorial requirements in the argument addresses
of the lexical items, allows for an economical deployment of functional domains
in extended projections. Agr-phrases are not postulated (cf. Wunderlich 1994).33
Anaphoric entities such as the pronominal clitic34 determine the form,
meaning and information-structural status of DPs. In the functional projection
they represent counterparts of topical DPs, which induce the binding of N’s
referential argument. In this way the clitic pronouns contribute to the character-
isation of the reference situation as a contextual parameter of CPs.
We have shown that the clause structure of Bulgarian contains explicit
linguistic information about the reference status of internal argument DPs, as
indicated by clitic doubling. Bulgarian clitic pronouns consistently act as Topic
markers and participate in this way in the information structure interpretation.

32. In relating SF-representations to surface structure, we assume that anaphor resolution takes place
in surface structure. In such cases the compositional structure of the SF-representation of sentence
sequences will be realised by type-shifting. All anaphoric SF-components, anaphoric topics included,
are treated on a par. Thus, the referents of all anaphors are asserted as existential. These operators
have scope over the operator which binds the reference situation variable in the Comment part of CPs
(cf. Steube 1997).
33. Chomsky (1995), while proposing the elimination of AGR as a category, suggests that all
functional categories may have a semantic connection.
34. Reinhart and Reuland (1993) argue that only an anaphor allows a syntactic binding relation.
286 IVANKA PETKOVA SCHICK

Appendix 1. Bulgarian

Table 1. Personal Pronouns

Num   Full Forms Clitics
Nominative Accusative (Dative) Accusative Dative
 1 az mene (men) mene me mi
2 ti tebe (teb) tebe te ti
3 m toj nego nemu go mu
f tja neja nej ja ì
n to nego nemu go mu
 1 nie nas nam ni ni
2 vie vas vam vi vi
3 te tjah tjam gi im

Table 2. Reflexive Personale Pronouns

Casus Full Forms Clitics
Nominative – –
Accusative (sebe), sebe si se
Dative (sebe, sebe si) si

Table 3. Possesive Pronouns

Num   Full Forms Clitics
/ / / 
 1 moj moja moe moi mi
2 tvoj tvoja tvoe tvoi ti
3 m negov negova negovo negovi mu
f nein nejna nejno nejni ì
n negov negova negovo negovi mu
 1 naš naša naše naši ni
2 vaš vaša vaše vaši vi
3 tehen tjahna tjahno tehni im

Table 4. Reflexive Possessive Pronouns

Casus Full Forms Clitics
/ / / 
Nominative svoj svoja svoe svoi si
Accusative (svoego) – – – –
 CLITIC DOUBLING CONSTRUCTIONS IN BALKAN-SLAVIC 287

Appendix 2. Macedonian

Table 5. Personale Pronouns

Num   Full Forms Clitics
Nominative Accusative Dative Accusative Dative
Sg 1 jas mene mene Me mi
2 ti tebe tebe Te ti
3  toj / on nego nemu Go mu
 taa / ona nea nejze Ja i
 toa / ono nego nemu Go mu
Pl 1 nie nas nam Ne ni
2 vie vas vam Ve vi
3 tie niv nim Gi im

Table 6. Reflexive Personale Pronouns

Casus Full Forms Clitics
Nominative – –
Accusative (sebe), sebe si se
Dative (sebe, sebe si) si

Table 7. Possesive Pronouns

Num  
/ / / 
Sg 1 moj moja moe moi
2 tvoj tvoja tvoe tvoi
3  negov negova negovo negovi
 nejzin nejzina nejzino nejzini
 negov negova negovo negovi
Pl 1 naš naša naše naši
2 vaš vaša vaše vaši
3 nivni nivna nivno nivni

Table 8. Reflexive Possesive Pronouns

/ / / Pl
svoj svoja svoe svoi
288 IVANKA PETKOVA SCHICK

References

Alter, K. and U. Junghanns 1997. ‘Topic-related prosodic patterns in Russian.’ In: Kosta,
P. and M. Unger (eds.) Proceedings of the Second European Conference on Formal
Description of Slavic Languages, Potsdam.
Anderson, S. 1993. ‘Wackernagel’s revenge: Clitics, morphology and the syntax of
second position.’ Language 69: 68–98.
Andrejčin, L. 1958. ‘Enklitični pritežatelni mestoimenija.’ In: Izsledvanija v Čest na
Akademik Dimităr Dečev po Slučaj 80godišninata mu. Sofia: BAN, 199–200.
Angelova, I. 1994. Sintaksis na Bălgarskata Razgovorna Reč. Sofia: Universitetsko
Izdatelstvo ‘Sv. Kliment Ohridski’.
Ariel, M. 1990. Accessing Noun-Phrase Antecedents. London: Croom Helm.
Bach, E. 1986. ‘The algebra of events.’ Linguistics and Philosophy 4: 5–16.
Bierwisch, M. 1987. ‘Semantik der Graduierung.’ In: Bierwisch, M. and E. Lang (eds.)
Grammatische und Konzeptuelle Aspekte von Dimensionsadjektiven. Berlin:
Akademie-Verlag, 91–286.
Bierwisch, M. 1988. ‘On the grammar of local prepositions.’ In: Bierwisch, M., W.
Motsch and I. Zimmermann (eds.) Syntax, Semantik und Lexikon. (Studia Gramma-
tica 29). Berlin: Akademie-Verlag, 1–65.
Bierwisch, M. 1990a. ‘Event nominalization: Proposals and problems.’ Acta Linguistica
Hungarica 40: 19–84.
Bierwisch, M. 1990b. ‘Verb cluster formation as a morphological process.’ In: Booij, G.
and J. van Marle (eds.) Yearbook of Morphology 3. Dordrecht: Foris, 173–199.
Bierwisch, M. 1996. ‘Lexikon und Universalgrammatik.’ In: Weber, N. (ed.) Semantik,
Lexikographie und Computeranwendungen. (Sprache und Information 33) Tübingen:
Max Niemeyer Verlag, 129–164.
Bierwisch, M. 1997. ‘Lexical information from a minimalist point of view.’ In: Wilder,
C., H.-M. Gärtner and M. Bierwisch (eds.) The Role of Economy Principles in
Linguistic Theory. (Studia Grammatica 40). Berlin: Akademie-Verlag, 227–265.
Billings, L. and M. Baermann 1997. ‘Macedonian clitics: More evidence for a Wacker-
nagel position.’ Paper presented at the Second Conference on Formal Approaches to
South Slavic Languages, Sofia, Bulgaria.
Budagov, P. A. 1958. Etjudy po Sintaksisu Rumynskogo Jazyka. Moskva: Vysšaja Škola.
Cardinaletti, A. and M. Starke 1995. ‘The typology of structural deficiency: On three
grammatical classes.’ FAS Papers in Linguistics 1: 1–55.
Ćavar, D. and C. Wilder 1994. ‘Clitic third in Croatian.’ Eurotyp Working Papers 6:
19–61.
Chomsky, N. 1995. The Minimalist Program. Cambridge, Mass.: MIT Press.
Cinque, G. 1988. ‘On si constructions and the theory of arb.’ Linguistic Inquiry 19:
521–581.
Dimitrova-Vulchanova, M. 1995. ‘Clitics in Slavic.’ Studia Linguistica 49: 54–92.
 CLITIC DOUBLING CONSTRUCTIONS IN BALKAN-SLAVIC 289

den Dikken, M. 1995. Particles: on the Syntax of Verb-Particle, Triadic, and Causative
Constructions. New York/Oxford: Oxford University Press.
den Dikken, M. 1997. ‘Predication in DP.’ Paper presented at DGfS, Düsseldorf.
Dimitrova-Vulchanova, M. and L. Hellan 1996. ‘Clitics and Bulgarian clause structure.’
In: Dimitrova-Vulchanova, M. and L. Hellan (eds.). Papers from the First Confer-
ence on Formal Approaches to South Slavic Languages (Plovdiv 1995). University of
Trondheim Working Papers in Linguistics 28: 363–409.
Donnellan, K. S. 1966. ‘Reference and definite descriptions.’ The Philosophical Review
75: 281–304.
Drubig, H. 1994. ‘Island constraints and the syntactic nature of Focus and association
with Focus.’ Arbeitspapiere des Sonderforschungsbereich 340, Bericht Nr. 51,
Universität Tübingen.
Frey, W. and K. Pittner 1998. ‘Zur Positionierung der Adverbiale im Deutschen Mittel-
feld. Linguistische Berichte 176: 489–534.
Gallmann, P. 1996. ‘Die Steuerung der Flexion in der DP.’ Linguistische Berichte 164:
283–314.
Grimshaw, J. 1991. ‘Extended projections.’ Ms., Brandeis University.
GSBKE 1983. Gramatika na Săvremennija Bălgarski Knižoven Ezik. Vol. II, Vol. III.
Institut za Bălgarski Ezik. Sofia: BAN.
Guentchéva, Z. 1994. Thématisation de l’Objet en Bulgare. Bern: P. Lang S.A.
Haftka, B. 1988. ‘Linksverschiebungen: Ein Beitrag zur Konfigurationalität des Deut-
schen.’ In: Bierwisch, M., W. Motsch and I. Zimmermann (eds.) Syntax, Semantik
und Lexikon. (Studia grammatica 29). Berlin: Akademie-Verlag, 25–58.
Hellan, L. and C. Platzack 1995. ‘Pronouns in Scandinavian languages: An overview.’
Working Papers in Scandinavian Syntax 56: 47–69.
Heusinger, K. von 1997. Salienz und Referenz. (Studia Grammatica 43). Berlin: Akade-
mie-Verlag.
Inkelas, S. 1990. Prosodic Constituency in the Lexicon. New York: Garland Publishing.
Ivančev, S. 1957/78. ‘Nabljudenija vărhu upotrebata na člena v bălgarskija ezik.’
Bălgarski Ezik 7: 500–524.
Jäger, G. 1995. Topics in Dynamic Semantics. Doctoral dissertation, Humboldt Universität,
Berlin.
Joppen, S. and D. Wunderlich 1995. ‘Argument linking in Basque.’ Lingua 97: 123–169.
Kallulli, D. 1995. Clitics in Albanian. University of Trondheim Working Papers in
Linguistics, vol. 24.
Kayne, R. 1993. ‘Towards a modular theory of auxiliary selection.’ Studia Linguistica 47:
3–31.
Kiparsky, P. 1992. ‘Structural case.’ Ms., Institute for Advanced Study, Berlin.
Koneski, B. 1966. Gramatika na Makedonskiot Literaturen Jazik. Skopje: Izdavačko
pretprijatie ‘Prosvetno delo’.
Kratzer, A. 1994. ‘The event argument and the semantics of voice.’ Ms., MIT.
290 IVANKA PETKOVA SCHICK

Krifka, M. 1991/92. ‘A compositional semantics for multiple focus constructions.’ In:

Jacobs, J. (ed.) Informationsstruktur und Grammatik. Linguistische Berichte.
Sonderheft 4.
Löbner, S. 1985. ‘Definites.’ Journal of Semantics 4: 279–326.
Longobardi, G. 1994. ‘Reference and proper names: A theory of N-movement in syntax
and logical form.’ Linguistic Inquiry 25: 609–665.
Maienborn, C. 1997. ‘On the meaning of sentence modifiers: Semantic indeterminacy and
grammatically induced specification.’ In: van der Sandt, R., R. Blutner and M.
Bierwisch (eds.) From Underspecification to Interpretation (Working Papers of the
Institute for Logic and Linguistics. IBM Deutschland, Heidelberg), 183–201.
Meinungen, A. 1995. Discourse dependent DP (De-)Placement. Doctoral dissertation,
Universität Potsdam.
Mirčev, K. 1963. Istoričeska Gramatika na Bălgarskija Ezik. Sofia: Nauka i Izkustvo.
Müller, G. 1997. ‘Parallel movement principle.’ Paper presented at ZAS, Berlin.
Muysken, P. 1982. ‘Parametrizing the notion “Head”.’ Journal of Linguistic Research 2:
57–75.
Nicolova, R. 1979. ‘Njakoi osobenosti v semantikata i distribucujata na pritežatelnite
mestoimenija v bălgarskija ezik v săpostavka s polskija ezik.’ Săpostavitelno
Ezikoznanie 3: 58–66.
Nicolova, R. 1986. Bălgarskite Mestoimenija. Sofia: Nauka i Izkustvo.
Ortmann, A. 1994. Possessorkongruenz. Eine Fallstudie zum Verhältnis von Semantik,
Morphologie und Syntax. Doctoral dissertation, Heinrich-Heine-Universität, Düssel-
dorf.
Partee, B. 1987. ‘Noun phrase interpretation and type-shifting principles.’ In: Groenen-
dijk, J., D. de Jongh and M. Stokhof (eds.) Studies in Discourse Representation
Theory and the Theory of Generalised Quantifiers. Dordrecht: Foris, 115–143.
Pašov, P. 1978. ‘Za “padežite” na mestoimenijata v săvremennija bălgarski ezik. In:
Pašov, P. (ed.) Pomagalo po Bălgarska Morfologija. Imena. Sofia: Nauka i Izkustvo,
340–355.
Penčev, J. 1993. Bălgarski Sintaksis. Upravlenie i Svărzvane. Plovdiv: Plovdivsko
Universitetsko Izdatelstvo.
Progovac, L. 1998. ‘Events in Serbian.’ In: Dimitrova-Vulchanova, M., L. Hellan, I.
Kasabov and I. Krapova (eds.) Papers from the Second Conference on Formal
Approaches to South Slavic Languages, Sofia 1997. (University of Trondheim Working
Papers in Linguistics 31): 79–116.
Reinhart, T. and Reuland, E. 1993. ‘Reflexivity.’ Linguistic Inquiry 24: 657–720.
Reis, M. and I. Rosengren 1997. ‘A modular approach to the grammar of additive
particles: the case of German AUCH.’ Journal of Semantics 14: 237–309.
van Riemsdijk, H. 1998. ‘Head movement and adjacency.’ Natural Language and
Linguistic Theory 16: 633–678.
Ritter, E. 1995. ‘On the syntactic category of pronouns and agreement.’ Natural Language
and Linguistic Theory 13: 405–443.
 CLITIC DOUBLING CONSTRUCTIONS IN BALKAN-SLAVIC 291

Rivero, M.-L. 1994. ‘Clause structure and V-movement in the languages of the Balkans.’
Natural Language and Linguistic Theory 12: 63–120.
Rizzi, L. 1997. ‘The fine structure of the left periphery.’ In: Haegeman, L. (ed.) Elements
of Grammar. Dordrecht: Kluwer, 281–337.
Rooth, M. 1992. ‘A theory of focus interpretation.’ Natural Language Semantics 1:
75–116.
Rudin, C. 1990/91. ‘Topic and focus in Bulgarian.’ Acta Linguistica Hungarica 40:
429–447.
Rudin, C. 1996. ‘On pronominal clitics.’ In: Dimitrova-Vulchanova M. and L. Hellan
(eds.) Proceedings of FASS, Plovdiv, 1995. (University of Trondheim Working Papers
in Linguistics 28): 229–246.
Rudin, C. 1997. ‘AgrO and Bulgarian pronominal clitics.’ In: Lindseth, M. and S. Franks
(eds.) FASL V: The Indiana Meeting. Ann Arbor: University of Michigan Slavic
Materials, 224–252.
Růžička, R. 1992. ‘Slavic and Italian impersonal constructions with reflexive clitics.’ In:
Zimmermann. I. and A. Strigin (eds.) Fügungspotenzen. Zum 60 Geburtstag von
Manfred Bierwisch. (Studia grammatica 34). Berlin: Akademie-Verlag, 133–161.
Schick, I. 1994. Besonderheiten der Bulgarischen Sprache aus der Sicht der Balkanologie
und der Slavistik in Konfrontation mit dem Deutschen. Universität Leipzig, Institut für
Slavistik.
Schick, I. 1996a. ‘Spezifische Konstruktionen mit pronominalen Klitika im Balkan-
slavischen.’ In: Pittner, R. J. and K. Pittner (eds.) Beiträge zu Sprache und Sprachen.
B. 3 (Vorträge der 6. Münchener Linguistik-Tage, München) München: Lincom
Europa.
Schick, I. 1996b. ‘Spezifische Konstruktionen mit duplizierenden pronominalen Klitika im
Balkanslavischen.’ Sprache und Sprachen, München: Gesus.
Schick, I. 1997. ‘Clitic-doubling constructions in modern Bulgarian.’ In: Kosta, P. and M.
Unger (eds.) Proceedings of the Second European Conference on Formal Description
of Slavic Languages, Potsdam.
Schick, I. 1998. ‘Pronominale Klitika des Balkanslavischen.’ In: Rothe, H. and H.
Schaller (eds.) Beiträge zum XII. Internationalen Slavistenkongreß (Krakau 1998).
Marburger Abhandlungen zur Geschichte und Kultur Osteuropas, vol 38. München:
Otto Sagner Verlag, 91–114.
Schick, I. and I. Zimmermann 1995. ‘Flexive und klitische Annexe in der Definitheits-
kennzeichnung des Bulgarischen.’ Paper presented at ZAS, Berlin.
Schick, I. and I. Zimmermann 1997. ‘Das dativische pronominale Klitikum in der
Substantivgruppe des Bulgarischen.’ In: Junghanns, U. and G. Zybatow (eds.)
Formale Slavistik. (Leipziger Schriften zur Kultur-, Literatur-, Sprach- und Überset-
zungswissenschaft 7) Frankfurt am Main: Vervuert Verlag, 49–61.
Schmitt, C. 1996. Aspect and the Syntax of Noun Phrases. Doctoral dissertation, University
of Maryland at College Park.
292 IVANKA PETKOVA SCHICK

Späth, A. 1997. ‘On definite reference of determinerless nouns in Russian.’ In: Kosta, P.
and M. Unger (eds.) Proceedings of the Second European Conference on Formal
Description of Slavic Languages, Potsdam.
Steube, A. 1997. ‘Ein kognitionswissenschaftlich basiertes Modell für Informations-
strukturierung. Paper presented at Sächsischen Akademie der Wissenschaften, June
13, 1997 and at Institut für Deutsche Sprache, Köln, July 2, 1997.
Steube, A. and A. Späth 1998. ‘DP-Semantik und Informationsstrukturierung im Russisch-
en (auf der Basis eines Vergleichs mit dem Deutschen).’ In: Sprache und Pragmatik
46. Lund: Germanistisches Institut der Universität Lund.
Stiebels, B. 1996. Lexikalische Argumente und Adjunkte: zum Semantischen Beitrag von
Verbalen Präfixen und Partikeln. (Studia grammatica 9). Berlin: Akademie-Verlag.
Szabolcsi, A. 1983. ‘The possessor that ran away from home.’ The Linguistic Review 3:
89–102.
Szabolcsi, A. 1987. ‘Functional categories in the noun phrase.’ In: Kenesei, I. (ed.)
Approaches to Hungarian 2, Theories and Analyses. Szeged: Jate, 167–189.
Tilkov, D. 1977. ‘Učastieto na klitikite v izgraždaneto na akcentnata edinica v bălgarskija
knižoven ezik.’ Bălgarski Ezik 1977: 177–185.
Tomić, O. M. 1996. ‘The Balkan-Slavic clause clitics.’ Natural Language and Linguistic
Theory 14: 811–872.
Vakareliyska, C. 1995. ‘Na-drop in the context of Bulgarian case marking: Data from
aphasia.’ In: Yokoyama, O.T. (ed.) Harvard Studies in Slavic Linguistics. Vol. III,
Harvard University, Slavic Dept. Cambridge, Mass., 234–242.
Vallduví, E. 1992. The Informational Component. Doctoral dissertation, University of
California, Santa Cruz.
Verkuyl, H. J. 1993. A Theory of Aspectuality. The Interaction between Temporal and
Atemporal Structure. (Cambridge Studies in Linguistics 64). Cambridge: Cambridge
University Press.
Wilder, C. 1994. ‘Coordination, ATB and ellipsis.’ In: Zwart, C. J-W. (ed.) Minimalism
and Kayne’s Asymmetry Hypothesis. (Groninger Arbeiten zur Germanistischen
Linguistik 37): 231–329.
Wunderlich, D. 1991. ‘How do prepositional phrases fit into compositional syntax and
semantics?’ Linguistics 29: 591–621.
Wunderlich, D. 1994. ‘Towards a lexicon-based theory of agreement.’ Theoretical
Linguistics 20: 1–35.
Wunderlich, D. 1997. ‘CAUSE and the structure of verbs.’ Linguistic Inquiry 28: 27–68.
Wunderlich, D. and R. Fabri 1995. ‘Minimalist morphology: An approach to inflection.’
Zeitschrift für Sprachwissenschaft 14: 236–294.
Zwicky, A. 1977. On Clitics. Bloomington: Indiana University Linguistics Club.
Zwicky, A. 1985. ‘Clitics and particles.’ Language 61: 283–290.
Zybatow, G. and U. Junghanns 1998. ‘Topiks im Russischen.’ Sprache und Pragmatik 47.
Lund: Germanistisches Institut der Universität Lund.
 On Clitic Sites*

Olga Mišeska Tomić

University of Novi Sad

Abstract

Paying attention to the gradual change from Wackernagel to verbal clitics in a

number of Balkan Slavic languages, an analysis of clitic clusters is developed
in which Case and specificity features of the pronominal clitics are checked by
raising the XP*s to the specifiers of the agreement phrases in which the clitics
are generated. Third person auxiliary clitics are analysed as heads of AuxPs
c-commanding VP while other auxiliary clitics are analysed as AgrS/TensePs.

1. Introduction

Clitics can be defined as function words that are syntactically and phonologically
dependent on neighbouring words. Zwicky (1977) distinguishes between two
types of clitics: (a) clitics which appear in positions relative to the rest of the
structure where rules of ‘normal’ syntax would put them; and (b) clitics whose
positions within given phrasal units are determined by principles other than those
of nonclitic syntax. This paper will be concerned with the positions or sites of
clitics of the latter type, which Zwicky called ‘special clitics’.1

* This paper is partially based on Tomić and Beukema (1997), a paper presented at the Debrecen
Workshop on Clitic Phenomena in English and other European Languages. Thanks for comments are
due to the participants of the Workshop. Special thanks are due to Marcel den Dikken and Damir
Ćavar, who both read a previous version of this paper and gave suggestions that led to clarification
and strengthening of many points. I am immensely indebted to Frits Beukema for turning my
attention to many important details in need of explanation.
1. Zwicky’s labels ‘simple’ and ‘special’, referring to the clitics of type (a) and (b), respectively,
could be disputed on substantive grounds (cf. Tomić in press). We should, rather, treat them as
reference labels.
294 OLGA MIŠESKA TOMIĆ

Following Klavans (1982, 1985), Stephen Anderson (1992, 1993) identifies

three parameters along which ‘special clitics’ vary; he refers to these parameters
as ‘domain, anchor and orientation’. The domain is the syntactic constituent
within which the clitic is located; the anchor is the element within the given
domain to which the clitic relates; the orientation is the position of the clitic
relative to the anchor. Anderson does not make a distinction between syntactic
and phonological orientation. For Klavans, however, the choice of whether the
clitic forms part of a phonological word with material on its right vs. left is an
additional parameter of clitic placement.
Linguistic analyses of those Indo-European ‘special clitics’ that have as their
domain the clause (henceforth ‘clausal clitics’) have identified two major types of
such clitics: (a) second position or Wackernagel clitics and (b) clitics that are
oriented towards the verbal head of the clause and are referred to as ‘verbal’.2
While the Wackernagel clitics sit in the highest available site in the extended
projection of the verb, the verbal clitics are left-adjacent to the verb. The clausal
clitics in most European languages belong to one of these types. I will, however,
provide evidence that both types can be exemplified in a single language.
Historical evidence has shown that at the turn of the millennium the Indo-
European clitics were phonologically enclitic Wackernagel clitics. Over the
centuries, there has been a shift towards their becoming more and more oriented
towards the head of the clause and more and more phonologically proclitic. As
a result of this shift the pronominal clitics abandon the argument positions where
they had originally been generated and come to be generated directly in AgrO
and AgrIO. At the same time, the available landing sites for the clitics get
gradually closer to the verb, until finally the clitics end up invariably left-adjacent

2. Ćavar, personal communication, pointed out that the Polish auxiliary clitics, traditionally referred
to as ‘person and number morphemes’, are verb oriented. Note, however, that the Polish auxiliary
clitics can appear not only on the verb, as in (i), but also on any XP to the left of the verb, as in
(ii)–(iv):
(i) My znowu wczoraj poszliśmy do parku
we again yesterday went--+1-- to park-
(ii) Myśmy znowu wczoraj poszli do parku
we+1-- again yesterday went-- to park-
(iii) My znowuśmy wczoraj poszli do parku
we again+1-- yesterday went-- to park-
(iv) My znowu wczorajśmy poszli do parku
we again yesterday+1-- went-- to park-
‘We went to the park yesterday again.’
Franks, in the paper from which these examples were taken (Franks 1988), analyses the Polish
auxiliary clitics as agreement heads which are on their way to becoming verbal affixes.
 ON CLITIC SITES 295

to it. While many European languages have already gone through the change, in
some of them the change is still under way; hence the dual — Wackernagel and
verbal — clitic behaviour. I shall argue that the dual behaviour follows from
different values for the features [±V] and [±N] of the head of the clause.

2. Interdependence between types of clitics and the morphological

features of the head of the clause

The clausal clitics of a language can in most cases be qualified as clitics of a

given type or subtype. In Macedonian this is not the case. Consider the Macedo-
nian clauses in (1):
(1) a1. Si mu go dal
are-2-- him-- it-- given-.
včera
yesterday
‘You gave it to him yesterday, I understand.’
a2. *Si mu go včera
are-2-- him-- it-- yesterday
dal
given-.
b1. *Si mu tatko
are-2-- him-- father
b2. Tatko si mu
father are-2-- him--
‘You are his father (so you have to take care of him).’
b3. Ti si mu tatko
you are-2-- him-- father
‘YOU are his father (and not anybody else)!’
c1. ??Si mu mil
are-2-- him-- dear
‘He likes you.’
c2. Mil si mu
dear are-2-- him--
‘He likes you.’
c3. Ti si mu mil
you are-2-- him-- dear
‘It is you he likes!’
296 OLGA MIŠESKA TOMIĆ

d1. ?Mu e skinato paltoto

him-- is-3- torn-. coat+the-
‘His coat is torn out.’
d2. Skinato mu e paltoto
torn-. him-- is-3- coat+the-
‘His coat is torn out.’
d3. Na Petreta mu e skinato
to Peter- him-- is-3- torn-.
paltoto
coat+the-
‘Peter’s coat is torn out.’
f1. ?Mu e rečeno da bide
him-- is-3- told-. to be-3-
točen povek¢ e pati
punctual more times
‘He was told to be punctual more than once.’
f2. Rečeno mu e da bide
told-. him-- is-3- to be-3-
točen povek¢ e pati
punctual more times
‘He was told to be punctual to be punctual more than once.’
f3. Na Petreta mu e povek¢ e pati
to Peter- him-- is-3- more times
rečeno da bide točen
told-. to be-3- punctual
‘Peter was told to be punctual more than once.’
The grammaticality of (1a1), where the clitics are clause-initial and left-adjacent
to the l-participle, in contrast with the ungrammaticality of (1a2), where the
clitics are clause-initial but separated from the l-participle by an adverb, tells us
that we are dealing with prototypical verbal clitics, i.e. clitics that are left-
adjacent to the verb and procliticise to it. The well-formedness of (1b1) and
(1b2), where the clitics are preceded and followed by a nominal, in contrast with
the ill-formedness of (1b3), where they are placed clause-initially to the left of a
nominal, suggests that the clitics can be hosted by a nominal to their left but not
by one to their right. The relativised acceptability of (1c1), (1d1) and (1f1)
compared to the unconditional acceptability of the other clauses in (1c), (1d) and
(1e), indicates that in environments such as the ones illustrated in these clauses,
the clitics can with some (but not all) speakers of the language occur in initial
 ON CLITIC SITES 297

position and be hosted to their right, though non-initial clitic positioning, with
hosts to their left, is preferred.
The choice of anchor and the orientation of the Macedonian clausal clitics
actually correlates with the morphological properties of the head of the clause. In
Tomić (1997a, 1997b), I pointed out that tensed lexical verbs, l-participles, past
participles, passive participles and adjectival and nominal predicates have distinct
values for the features [±V] and [±N].3 These are represented in (2):
(2) V N
tensed verbs + −
l-participles + −
past participles + +
passive participles + +
adjectives + +
nouns − +
Note that l-participles have positive values for V, negative for N, as do tensed
verbs. As argued in Tomić (1996, 1997b) the behaviour of l-participles in the
neighbourhood of clitics is analogous to that of tensed verbs. Like the tensed
verbs, the l-participles form enlarged local domains with the clitics to their left,
to the extent that, when they move, the clitics get a free ride with them. This is
exemplified in (3a) and (3b), where an l-participle and a tensed verb, respective-
ly, have moved to the left of the interrogative clitic li, which in all the Slavic
languages is uncontroversially located in C:
(3) a. Bi (si) mu go
would-- (are-2-- him-- it--
dala li peroto?
given-- - pen+the-
‘Would you be willing to give him the pen?’
b. 7́e mu go dadeš li
will-- him-- it-- give--2 -
peroto?
pen+the-
‘Will you give him the pen?’

3. Frits Beukema, personal communication, points out that the adjectival and nominal predicates are
actually small clause predicates and suggested that prepositions should be included in the list. This
would not affect the generalising power of the analysis, since the behaviour of adjectives is distinct
from the behaviour of nouns. The examples in (1) actually demonstrate that adjectives behave to
some extent like participles, while nouns do not. The fact that the nominal and adjectival predicates
are small clause predicates, as are prepositional phrase predicates, is thus immaterial.
298 OLGA MIŠESKA TOMIĆ

The nouns have positive values for N, negative for V; and, when in the position
of clause nuclei, they do not form enlarged local domains with clitics. Accord-
ingly, clauses such as (4a), where the clitics in initial position occur to the left
of a nominal predicate, contrast with clauses such as (4b), in which the clitics in
initial position occur to the left of an l-participle; while the former are ill-formed,
the latter are well-formed:
(4) a. *Si mi tatko
are-2-- me-- father
b. Si mi go dal
are-2-- me-- it-- given-.
‘You have allegedly given it to me’
Predicate adjectives, past participles and passive participles, which are [+V, +N]
categories, show dual behaviour: when occurring in the position of clausal heads,
they may form enlarged local domains with the clitics to their left, like the
tensed verbs and l-participles, but more often they do not and we find them in
clause-initial position, to the left of the clausal clitic cluster. Thus, both (1f1),
repeated as (5a), where the clitics are in clause-initial position, to the left of the
past participle rečeno, and (1f2), repeated as (5b), where the past participle is in
clause-initial position, to the left of the clitics, are well-formed Macedonian
clauses, though (5b) is more ‘usual’:
(5) a. ?Mu e rečeno da bide
him-- is-3- told-. to be-3-
točen povek¢ e pati
punctual more times
‘He was told to be punctual more than once.’
b. Rečeno mu e da bide
told-. him-- is-3- to be-3-
točen povek¢ e pati
punctual more times
‘He was told to be punctual more than once.’
A discussion of the qualification ‘more usual’ is in order here. In the languages
of the Balkans the clausal clitic clusters have undergone a change from second
position to verbal. In some of these languages, including Macedonian, the clausal
clitics have come to form enlarged local domains with the verb and are phono-
logically proclitic, rather than enclitic. The dual behaviour of the clitics in
Macedonian clauses in which V is instantiated by past or passive participles (and
marginally by adjectives) indicates that in this language the change from 2 to
 ON CLITIC SITES 299

verbal clitics is still under way. With those speakers with whom, or in those
environments in which the Macedonian clausal clitics are allowed to procliticise
to past or passive participles (and marginally to adjectives), the [+N] value of
these participles (and adjectives) seems to be undergoing a change to [−N].
In Tomić (1997a, 1997b) I treated the clitics in clauses whose heads are
instantiated by [+N] categories as a subtype of the verbal clitics. Note, however,
that the clitics in such clauses do not satisfy the criterion for ‘verbal clitichood’:
non-separability from the verb. As illustrated in (6), in clauses whose heads are past
participles, the Macedonian clausal clitics can be separated from the clausal head:
(6) a. Na Petreta mu e povek¢ e pati
to Peter- him-- is-3- more times
rečeno da bide točen
told-. to be-3- punctual
‘Peter was told more than to be punctual.’
b. Na Petreta mu e od strana na
to Peter- him-- is-3- from side of
komisijata veteno deka…
commission+the- promised-. that
‘Peter was promised by the commission that….’
On the other hand, as illustrated in (7), the clitics in clauses whose heads are
past or passive participles, can occur in clitic-third and clitic fourth positions,
thus contrasting with the Wackernagel clitics:
(7) a. Na Petreta povek¢ e pati mu e
to Peter- more times him-- is-3-
rečeno da bide točen
told-. to be-3- punctual
‘Peter was told more than once to be punctual.’
b. Na Petreta od strana na komisijata povek¢ e
to Peter- from side of commission+the- more
pati mu e rečeno da bide
times him-- is-3- told-. to be-3-
točen
punctual
‘Peter was more than once told by the commission to be punctual.’
These data indicate that the Macedonian clausal clitics in clauses whose heads
are past or passive participles behave sometimes as verbal, sometimes as
Wackernagel clitics. With those speakers with whom, or in those environments
300 OLGA MIŠESKA TOMIĆ

in which the value of the Macedonian past and passive participles is undergoing
a change from [+N] to [−N] they behave as verbal clitics; otherwise they behave
as Wackernagel clitics.4

3. The derivation and the site of the Macedonian verbal clitic cluster

In Macedonian, the clausal clitic cluster consists of pronominal, auxiliary, modal

and negative clitics.5 Tomić (1996, 1997a, 1997b) proposes that the Macedonian
verbal clitics are derived as heads of a range of functional projections. The
negative particle clitic is derived as head of NegP, the modal clitics as heads of
ModP, the first and second person auxiliary clitics as heads of Tense/AgrSP,6
the dative pronominal clitics as heads of AgrIOP, the accusative pronominal
clitics as heads of AgrOP, and the third person auxiliary clitics as heads of
AuxP. Accordingly, the relevant structure of (8a), with an l-participle as head,
and a clitic cluster consisting of the negative clitic ne, the modal clitic ¢
ke, the
7
auxiliary clitic sum, the dative pronominal clitic mu and the accusative pronom-
inal clitic go, would be (8b):
(8) a. Ne ¢ke sum mu go
not- will-- am-- him-- it--
dadela proektot
given-- project+the-
‘Rumour has it,8 I would be unwilling to give the project to him.’

4. The clitics in clauses with nominal predicates always behave as Wackernagel clitics.
5. In interrogative clauses introduced by raised nominals, such as (i), the interrogative clitic li
appears adjacent to the other clausal clitics:
(i) Po Marija li ¢ke ti go prati?
by Maria - will-- you--- it-- send-3
‘Is it with Maria that (s)he will send it to you?’
But even in cases such as this one, the Macedonian interrogative clitic li behaves (from a syntactical
and phonological point of view) distinctly — it may be hosted by a variety of hosts and is always
enclitic, while the other clitics are hosted only by verbs or predicate nominals and are most often
proclitic. Thus, it contrasts with the Serbo-Croatian interrogative clitic li, which is part of the Serbo-
Croatian clitic cluster.
6. A joint Tense and Subject Agreement Phrase is argued for by the portmanteau morph which
represents tense, person and number in all the Balkan Slavic languages.
7. While diachronically the Macedonian clitics sum ‘am’, si ‘are-2’, sme ‘are-1’, ste ‘are-2’,
se ‘are-3’, are analysable into root + affix, synchronically they are non-analysable morphs.
8. In Macedonian, the co-occurrence of forms of be with forms of the l-participle signals reporting
modality.
 ON CLITIC SITES 301

b. NegP

Spec Neg′

pro Neg ModP

ne Mod Tense/AgrSP

κ¢ e Tense/AgrS AgrIOP

sum AgrIO AgrOP

mu AgrO VP

go V DP

dadela
The structure of (9a), where the clausal head is instantiated by a tensed verb and
the clitic cluster consists of the negative clitic ne, the dative pronominal clitic mu
and the accusative pronominal clitic go, would be (9b):
(9) a. Ne mu go davam proektot
not him-- it-- give-1 project+the
‘I won’t give him the project.’
b. NegP

Spec Neg′

pro Neg Tense/AgrSP

ne Tense/AgrS AgrIOP

AgrIO AgrOP

mu AgrO VP

go V DP

davam proektot
302 OLGA MIŠESKA TOMIĆ

The structure of (10a), where the clausal head is instantiated by a passive

participle whose nominal feature is undergoing a change from [+N] to [−N], and
the clitic cluster consists of the modal clitic ¢
k e, the dative pronominal clitic mu
and the third person auxiliary clitic e, is represented in (10b):
(10) a. 7́e mu e izpraznet
will-- him-- is- emptied-.
stanot
apartment+the
‘His apartment will be vacated.’
b. ModP

Mod Tense/AgrSP

κ¢ e Tense/AgrS AgrIOP

AgrIO AuxP

mu Aux VP

e V XP

izpraznet stanot

When we compare (8b) to (10b), we observe that the projection of the first and
third person auxiliary clitic in (10b) differs from the projection of the first
person auxiliary clitic in (8b). The first (and second) person auxiliary clitics
(singular and plural) are analysed as heads of Tense/AgrSPs, whereas the third
person auxiliary clitics (also singular and plural) are analysed as heads of AuxPs.
Crucial evidence for the distinct analysis of the third person auxiliary clitics is
the fact that, in the clitic cluster, they occur to the right of the pronominal clitics,
whereas the first and second person ones occur to the left of the pronominal
clitics. Moreover, the third person auxiliary clitics do not occur in clauses in
which V is instantiated by l-participles. Thus, while (11a) with a second person
auxiliary clitic is a well-formed sentence of Macedonian, (11b) with a third
person auxiliary clitic is not; in its well-formed counterpart, (11c), there is no
auxiliary clitic at all:
 ON CLITIC SITES 303

(11) a. Ne si mu go
not are-2-- him-- it--
pratil proektot
sent-l--- project+the-
‘As I understand, you have not sent him the project.’
b. *Ne mu go e pratil
not him-- it-- is- sent-l---
proektot
project+the-
c. Ne mu go pratil
not him-- it-- sent-l---
proektot
project+the-
‘As I understand, you have not sent him the project.’
If the third person singular auxiliary clitic e or the third person plural auxiliary
clitic se are situated in the head of the projection to the immediate left of VP, the
fact that they appear in clauses in which the head is instantiated by a passive
participle, such as (10a), but not in clauses in which the head is instantiated by
an l-participle, such as (11c), can be accounted for through a subcategorisation
requirement: while passive participles (as well as past participles and adjectival
and nominal predicates) are selected by Aux, l-participles are not. The distinct
subcategorisation requirements of the past and passive participles, on the one
hand, and the l-participle, on the other, contribute to the fact that, in all environ-
ments and with all speakers of Macedonian, the behaviour of the l-participles
with respect to the clitics patterns with that of the tensed verb. This is not the
case with the behaviour of the past and passive participles.

4. Specificity and case checking

The derivation of the Macedonian pronominal clitics in AgrIO and AgrO is

argued for by the obligatory co-occurrence of these clitics with definite nominal
and pronominal objects, known as clitic doubling.9 This is illustrated by the

9. The clitic doubling argument for the derivation of the Macedonian pronominal clitics in AgrOP is also
found in Spencer (1991: 388), whose analysis of the Macedonian pronominal clitics was motivated by
Suñer’s (1988) analysis of clitic pronouns in a variety of Latin American Spanish. Rudin (1997) likewise
argues that the pronominal verbal clitics should be derived as heads of agreement projections.
304 OLGA MIŠESKA TOMIĆ

unacceptability of (12a), which contrasts with the acceptability of (8a), for

convenience repeated as (12b):
(12) a. *Ne ¢ke sum dadela proektot
not- will-- am-- given-- project+the
nemu
him-
b. Ne ¢ke sum mu go
not will-- am-- him-- it--
dadela proektot nemu
given-- project+the him-
‘As reported, I would be unwilling to give the project to him.’
Let us now tackle the problem of the relationship of the pronominal clitics to
XPs which they ‘double’. As Sportiche (1996: 234) points out, if the clitic is not
an element moved from an argument position, we need to account for the fact
that the pronominal clitics agree with XP*s in argument positions, though these
XP*s may be lexically empty.10 Referring to Chomsky’s (1992) proposal that
lexical items are inserted in the syntax with all their morphological affixes, and
that morpheme properties and ordering are checked under successive head
movement to functional projections, he proposes that the agreement of the XP*s
with the pronominal clitics, which he generates in the heads of clitic projections
to the left of I, is checked by raising the XP*s to the specifiers of the clitics
projections, referred to as ‘clitic voices’. Sportiche maintains that individual
clitics in individual languages differ as to the A/A′-status of the ‘clitic voices’
and whether XP* can or is required to raise overtly. He argues that in French the
specifier of an Accusative clitic is an A′-constituent, and the XP* raises at LF
from a complement position, skipping AgrO,11 whereas the specifier of a
Dative clitic, which, unlike the Accusative clitic, has no interpretative import, is
an A-constituent, and the XP* raises from AgrIO. The different treatment is
accounted for by the fact that Dative clitics express overt morphological agree-
ment, whereas the Accusative clitics do not. Hence, the Dative clitics carry an
agreement feature, whereas the Accusative clitics carry a specificity feature, also
carried by the (covertly) raised XP*.
If we take a broader view, we observe that Macedonian data indicate that

10. In French the XP*s are lexically empty, but in the Balkan languages, where we have clitic
doubling, they are lexically filled (though their cases are not morphologically marked).
11. As argued by Sportiche (1996: 244), skipping AgrO is a consequence of the lack of object agreement.
 ON CLITIC SITES 305

specificity is not incompatible with overt Case marking. Consider the Macedo-
nian clauses in (13):
(13) a. Im gi dadov knigite na
them-- them-- gave-1 books+the- to
edni deca
some- children
‘I gave the books to some (specific) children.’
b. ??Gi dadov knigite na edni deca
them-- gave-1 books+the- to some- children
c. *Im gi dadov knigite na
them-- them-- gave-1 books+the- to
deca
children
d. Gi dadov knigite na deca
them-- gave-1 books+the- to children
‘I gave the books to (non-specific) children.’
In (13a) the clitic im is associated with the specific DP edni deca, with which it
shares Dative Case, thus showing that Case and specificity are fully compatible.
The grammaticality of (13b) is questionable since it presupposes that the salience
of knigite is prominent in the discourse. In (13c) deca, although marked with
Dative Case from the preposition na, cannot be doubled by the dative clitic
because it is non-specific. In (13d) the XP na deca is non-specific and is not
doubled by a clitic pronominal; the grammaticality of this clause shows clearly
that not only in the case of accusative clitics, but also with dative clitics, clitic
doubling is contingent on specificity.
Agreeing with Sportiche that the features of the pronominal clitics are
checked by raising the XP*s with which they associate to the specifiers of the
phrases projected by the clitics, I argue that (a) these projections are the agree-
ment phrases themselves, rather than clitic phrases to the left of I; and (b) the
raised XP*s check both Case and specificity, with both Dative and Accusative
pronominal clitics.12 Whether the XP* is lexically filled, as in the clitic doubling

12. Although specificity and Case could in principle be checked in different positions in the tree,
the fact that clitic doubling is contingent on specificity makes it reasonable to check specificity in the
nodes projected by the clitics, which is where Case is also be checked. If the Dative pronominal clitic
projects AgrIOP and the Accusative pronominal clitic projects AgrOP, the Case of the XP* sitting in
the Indirect Object position of the verb has to be checked in AgrIOP and the Case of the XP* sitting
in the Direct Object position in AgrOP.
306 OLGA MIŠESKA TOMIĆ

cases, or empty, as in French, the checking of both Case and specificity is done
by covert movement, in LF.13

5. The site of the Tobler-Mussafia clitics

The analysis of the Macedonian clitic clusters in clauses headed by [+V, −N]
categories is applicable to all ‘verbal’ clitic clusters, irrespective of whether they
are phonologically proclitic, or else are syntactically oriented towards the head
of the clause in that nothing can come between them and that head, but are
restricted in that they cannot appear in first position in the clause, and are
phonologically enclitic. As a matter of fact, the latter clitic type, to which
Romance linguists refer as Tobler-Mussafia (TM) clitics (for a recent reference
see Rivero 1997), is a subtype of the ‘verbal’ clitics.
Rivero (1997) derives the TM clitics in a TMP which takes IP as its
complement. In this TMP the clitics can be assigned to two locations: SpecTMP
or TM. Rivero’s analysis is represented in (14):

13. As illustrated in (i), the checking of the [+topic] feature, carried by the Bulgarian pronominal
clitics, may involve overt movement of the XP*:
(i) Na Ivan knigata az mu ja dadox
to Ivan book+the- I him-- it-- gave-1
‘As for giving the book to Ivan, I did so.’
To accommodate moved XPs such as na Ivan in (i), Vulchanova-Dimitrova and Hellan (1996) design
a specific site, called FRONT. This site is divided into a high field and a low field. The elements in
the high field are taken to stand in an agreement relation to the pronominal clitics. Thus, in (i) na
Ivan and knigata stand in an agreement relation to the clitics mu and ja, respectively.
Agreement is, however, a broader phenomenon than suggested by these authors specifically for
Bulgarian. Moreover, even in Bulgarian, clitic doubling does not occur only with topicalised phrases;
it also occurs with empty pro-XPs from which the clitics derive their semantic import. Thus, my
approach, in which agreement is not made dependent on topicalisation, subsumes the more limited
approach that Vulchanova-Dimitrova and Hellan take.
 ON CLITIC SITES 307

(14) a. CP

C YP

Y TMP

Spec TM

clitic cluster TM IP

b. CP

C YP

Y TMP

Spec TM

TM IP

clitic cluster

Rivero assigns the TM pronominal clitics to SpecTMP or TM solely on the basis

of their surface ordering with respect to other constituents. Thus, she says that
the position of the adverb in the Old Spanish sentence in (15) strongly suggests
that the clitic is in the head of TMP:
(15) mas, mal pecado, [XP algunos de los señores grandes [TMP
and bad sin some- of the- lords big-
mas aina [TM’ se inclinan]]]…
more now  tend-
‘And, unluckily some of the nobles now tend to…..’
Rivero asserts that (14a) is the prototypical structure for ‘Balkan languages
including Bulgarian’, though she consistently disregards the auxiliary clitics,
which in the South-Slavic languages co-occur with the pronominal clitics in
clusters. Consider the Bulgarian clause (16):
308 OLGA MIŠESKA TOMIĆ

(16) Az sǎm mu dala knigata

I am-. him-. given-. book+the
‘I have given him the book.’
The auxiliary clitic sǎm in (16), like any other tensed auxiliary, can only be in
AgrSP or AgrSP/Tense, which makes the word order of the clitics in this clause
impossible to capture in Rivero’s system, since there is no reason to move the
auxiliary to a position to the left of the pronominal clitic mu.
If the clitics are derived as heads of functional projections — the auxiliary
clitic sǎm in AgrS/Tense and the pronominal clitic mu in AgrIO, in this case —
and the clitic cluster is formed to the immediate left of V, the clause in (16)
would present no problem. In this case, V attracts the clitics to its left, but does
not form a local domain with them. The phonological requirements of the
Bulgarian clitics would then be satisfied by the presence of the lexical non-clitic
pronominal subject az, to which the clitics would encliticise.
The contrast between the ‘true’ verbal and TM verbal clitics can best be
illustrated by the comparison of the Macedonian and Bulgarian clauses in (17):
(17) M Mu go dadov
him-- it-- gave-1
‘I gave it to him.’
B1 *Mu go dadox
him-- it-- gave-1
B2 Dadox mu go
gave-1 him-- it--
‘I gave it to him.’
In the Macedonian clause in (17), the clitics are to the left of the verb. The
Bulgarian clause with the same word order, B1, is ungrammatical, since the
clitics require a phonological host to their left, which is not there. In the
Bulgarian grammatical clause B2 a phonological host for the clitics is provided
by the raised verb. Rivero (1995) suggests that this raising is a Last Resort
movement which left-adjoins the verb to the head of the functional projection
harbouring the clitics, in order to provide a phonological host for them. This
suggestion, however, assumes that syntax looks into phonology. It is more
plausible to maintain that the Bulgarian tensed verb moves to AgrS/Tense in
order to check its [+Tense] feature.14 The distinct behaviour of the Macedonian

14. Since the movement of the verb places the verb to the immediate left of the Bulgarian
pronominal clitics, it can, in PF, act as a host to which these clitics encliticise.
 ON CLITIC SITES 309

and Bulgarian tensed verbs would then follow from the ability of the Macedonian
verb to form an extended local domain with the clitics to its left, which the
Bulgarian verb does not have. We are, however, faced with the problem of
explaining why the Bulgarian tensed verbs in clauses such as (18), in which there
is another constituent to the left of the clitics, the verbs do not raise overtly to
check their [+Tense] feature:
(18) Včera go vidjax
yesterday him-. saw-1
‘I saw him yesterday.’
Petko mi go pokaza
Petko me-. it-. showed-1
‘Petko showed it to me.’
I have no definitive solution for this problem, though it seems to be a relic of the
Wackernagel behaviour of the Bulgarian clitics centuries ago.

6. From Wackernagel to verbal clitics

Let us now consider the shift from Wackernagel to verbal clitic clusters. Up until
recently, the Wackernagel clitic clusters have been located in one and the same
syntactic position — in CP (cf. Wilder and Ćavar 1994; Franks and Progovac
1994; Progovac 1996; Tomić 1996), or at least in a functional projection to the
left of IP (cf. Roberts 1994; Rivero 1994, in press; Halpern 1992; Percus 1993;
Schütze 1994). Boković (1995, 1997, this volume), however, challenges the
assumption that the Wackernagel clitics are located in the same position. Accor-
ding to him, the clitic cluster is not formed in C, but rather in different positions
in the structure; in Bošković (this volume) he argues that they are placed behind
the first prosodic boundary and the host is provided by the phonology.
Bošković’s principal argument against the assumption that clitics sit in C
hinges on the blocking potential of sentential adverbs; they prevent a verb from
moving to C. Bošković’s example is given in (19):
(19) *Istukao je nesumnjivo Petra
beaten-3 is- undoubtedly Peter
Undoubtedly, (19) is ungrammatical. Note, however, that not only clauses such
as (20a), in which the participle hosting the clitic cluster has crossed over the
sentential adverb, but also clauses such as (20b) where a tensed verb has crossed
over a sentential adverb, are unacceptable.
310 OLGA MIŠESKA TOMIĆ

(20) a. *Dao mi ga je nesumnjivo juče

given-3 me-- it-- is- undoubtedly yesterday
b. *Dade mi ga nesumnjivo juče
gave-3 me-- it-- undoubtedly yesterday
If we leave out the sentential adverb nesumnjivo, however, both clauses in (20)
would be fine:
(21) a. Dao mi ga je juče
b. Dade mi ga juče
‘He gave it to me yesterday.’
Accordingly, with respect to crossing over a sentential adverb, tensed verbs and
participles behave uniformly.
Notice now that participles and tensed verbs do not behave uniformly with
respect to movement to C: in (22a) the tensed verb is to the left of the interroga-
tive particle li, which by all accounts is located in C. However, the ungrammat-
icality of (22b) shows that in interrogative clauses, the participle cannot be
located to the left of li.15
(22) a. Dade li mu ga juče?
gave-3 - him-- it-- yesterday
‘Did he give it to him yesterday?’
b. *Dao li mu ga je juče?
given-3 - him-- it-- is- yesterday.
The contrast between (22a) and (22b) cannot be used as an argument against
placing the clitic cluster in C, leading to the assumption that the clitics are placed

15. Ćavar, personal communication, pointed out that the particle li is not the element that excludes
fronting of the participle, but rather the question operator involved. According to him (22b) is
‘perfect’ as a declarative clause. It may be so, but only in some specific dialects of Serbo-Croatian.
In any case, this does not affect the conclusion that in Serbo-Croatian the participle cannot be located
to the left of li in interrogative clauses.
Note, that with respect to the relationship of li and the participle, Serbo-Croatian differs from
Macedonian, where the participle can be located in clause-initial position to the left of li, as
exemplified below:
(i) Dojden li e?
come-.-- - is-
‘Has he come?’
(ii) Rečeno li mu e da ne docni?
said-. - him-- is- to not be late-3
‘Has he been told not to be late?’
 ON CLITIC SITES 311

behind the first prosodic boundary, the host being provided by the phonology.16
It only indicates that the clitic cluster need not be formed in C. As a matter of
fact, the data presented lead to the conclusion that, rather than in C, the clitic
cluster sits in the highest available site in the extended projection of the verb.
Adopting a Copy-and-Delete analysis (cf. Chomsky 1995), Franks (1998)
contends that the Wackernagel clitics move to the highest available site in the
extended projection of the verb, head to head, leaving copies at all intermediate
sites. In PF all but the highest copy deletes, unless retention of the highest copy
results in a PF crash. Economy considerations then dictate that the next highest
copy is pronounced, unless again the result fails to converge. As a matter of fact,
the highest clitic copy is pronounced that satisfies the clitic’s phonological
requirement for encliticisation. Thus, while clitic movement is syntactic, which
copy is pronounced is determined by phonology.
Franks argues that Clitic-Second essentially follows from and is related to
Verb-Second. The argument goes like this: The overt movement of the clitics
towards the highest functional head position available has to involve strong
feature checking. Taking strong features to be features which, following
Chomsky (1995), have to be checked as soon as possible against the licensing
lexical head (V, for clausal clitics); and assuming it to be a fact of phrase
structure that in the course of the derivation lexical heads raise (covertly or
overtly) to the top of their extended projections, it is to be expected that each
time a new functional head is merged into the tree, the clitics raise to it as the
immediate next step in the derivation. While the features do not get erased until
the verb reaches the clitics (in LF), the clitics still move each time they have the
opportunity.
Progovac (1998) modifies Frank’s analysis. She argues that the clitics move
all the way overtly together with the verb, rather than moving only part of the
way along with it and then continuing by themselves to the ultimate LF position
of V (commonly assumed to be the highest head in the extended projection),
where they ‘wait’ for the verb to reach them. The clitics are then pronounced in
the head of the chain, while the verb is normally pronounced in a functional head
with a strong feature, i.e. in the position in which it is usually pronounced.
If Franks and Progovac are on the right track, the change from second
position clitics to verbal clitics is due to the loss of the V2 effect, which
eliminates the propensity of the clitics to move overtly to the highest available

16. See Progovac (this volume) for arguments against the motivation for an intonation-based
characterisation of clitic-second in Serbo-Croatian.
312 OLGA MIŠESKA TOMIĆ

site in the extended projection of the verb. In the case of the ‘true’ verbal clitics,
the V2 effect has been completely lost and the clitics remain in the position
from which they started to move along with the verb. In the case of the Tobler-
Mussafia clitics, however, there is a remnant of the Clitic-Second effect,
reflected in the fact that, in clauses with pro subjects and no topicalised or
focused XPs, the verb appears to the left of the clitics.17
On the basis of the analysis of the clitics in Serbo-Croatian — the proto-
typical instance of a Wackernagel language, let us now reconstruct the clitic
cluster formation of a model Indo-European language (of several centuries,
perhaps as much as a millennium ago) and trace the changes that have lead to
the verbal clitic clusters. The pronominal clitics of that language would be
generated in argument positions in VP, most probably as Case heads. The
auxiliary clitics, on the other hand, would be generated in AgrSP/TenseP. Thus,
a clause with a participle as head, a first person auxiliary clitic, and dative and
accusative pronominal clitics, would have an underlying structure along the lines
of (23):
(23) AgrS/TenseP

Spec AgrS/Tense′

AgrS/Tense AgrIOP

AgrIO AgrOP
1Sg-Cl
AgrO vP

Spec v′

pro v VP

DP1 V′

Dat-Cl V DP2

Partic Acc-Cl

17. This is exemplified in Section 5.

 ON CLITIC SITES 313

The Dative and Accusative pronominal clitics would move to AgrIO and AgrO,
respectively, to check their agreement features. The participle would move first
to v and then to AgrS/Tense to check its [+Aux] feature, picking up the pronomi-
nal and Auxiliary clitics on its way. Both the participle and the clitics would be
pronounced in AgrS/Tense, which is where the clitic cluster would actually sit.
If the model language is very much like Serbo-Croatian, the third person
singular clitic (je in Serbo-Croatian) would not be derived in AgrS/TenseP, but
rather in AuxP, since it has mixed clitic and non-clitic properties and can appear
in clause-initial position when it raises to C, by itself or else along with some of
the other clitics, which it picks up along the way (for detailed argumentation cf.
Tomić 1996). A sentence such as the Serbo-Croatian sentence (24a) would have
an underlying structure such as (24b):
(24) a. Jesi li mu ga dao?
je+2-- - - - given--
‘Did you give it to him?’
b. CP

Spec C′

C AgrS/TenseP

Spec AgrS/Tense′

AgrS/Tense AgrIOP

2Sg-Cl AgrIO AgrOP

AgrO AuxP

Aux vP

je Spec v′

pro DP1 V′

Dat-Cl V DP2

Partic Acc-Cl

As in (23), the Dative and Accusative pronominal clitics would move to AgrIO
and AgrO, respectively. After that, je would move to C, picking up the pronominal
314 OLGA MIŠESKA TOMIĆ

and auxiliary clitics on its way. Both the verb and the clitics would be pro-
nounced in the highest available extended projection of V, in C, where the clitic
cluster would be located.
As a result of the shift from Wackernagel to verbal clitics, the pronominal
clitics abandon the argument positions where they were originally generated and
come to be generated directly in AgrO and AgrIO. At the same time, the number
of available landing sites for the clitics in the extended projections of the verb gets
gradually reduced, until finally the clitics invariably end up left-adjacent to it.

7. Conclusion

Linguistic analyses dealing with the clitics of the Indo-European languages have
identified two types of clitics: Wackernagel and verbal. While the Wackernagel
clitics sit in the highest available site in the extended projection of the verb, the
verbal clitics are left-adjacent to the verb. The clitics in most Indo-European
languages belong to one of these types. It is possible, however for both types to
be exemplified in a single language; I have shown that this occurs in languages
in which the change from Wackernagel to verbal clitics, through which many
Indo-European languages have gone, is still under way. I argued that the dual
behaviour of the clitics follows from different values for the features [±V] and
[±N] of the head of the clause.
In the structural analyses of the clitic clusters, provided in the paper, special
attention has been paid to the pronominal and auxiliary clitics. More specifically,
it has been argued that the Case and specificity features of the verbal pronominal
clitics are checked by raising (covertly or overtly) the XP*s with which they
associate to the specifiers of the agreement phrases in which the clitics are
generated. Arguments have also been provided for analysing the third person
clitics (singular in Serbo-Croatian and singular and plural in Macedonian) as
heads of AuxPs c-commanding VP, whereas the other auxiliary clitics are
analysed as heads of AgrS/TensePs.

References

Anderson, S. R. 1992. A-Morphous Morphology. Cambridge: Cambridge University Press.

Anderson, S. R. 1993. ‘Wackernagel’s revenge: Clitics, morphology, and the syntax of
second position.’ Language 69: 68–98.
 ON CLITIC SITES 315

Bošković, Ž. 1995. ‘Participle movement and second position cliticization in Serbo-

Croatian.’ Lingua 96: 245–266.
Bošković, Ž. 1997. The Syntax of Nonfinite Complementation: An Economy Approach.
Cambridge, Mass.: MIT Press.
Bošković, Ž. This volume. ‘Second position cliticisation: syntax and/or phonology?’
Chomsky, N. A. 1993. ‘A minimalist program for linguistic theory.’ In: Hale, K. and S.
J. Keyser (eds.) The View from Building 20. Cambridge, Mass.: MIT Press, 1–52.
Chomsky, N. A. 1995. The Minimalist Program. Cambridge, Mass.: MIT Press.
Dimitrova-Vulchanova, M. and L. Hellan 1996. ‘Clitics and Bulgarian clause structure.’
University of Trondheim Working Papers in Linguistics 28: 363–409.
Franks, S. 1998. ‘Clitics in Slavic.’ Position paper for the Comparative Slavic Morpho-
syntax Workshop, held at McCormick’s Creek State Park, 5–6 June. To appear in
Proceedings.
Franks, S. and L. Progovac 1994. ‘On the placement of Serbo-Croatian clitics’ In: Indiana
Slavic Studies 7, Proceedings of the 9th Biennial Conference on Balkan and South
Slavic Linguistics, Literature and Folklore, 69–78.
Halpern, A. L. 1995. On the Placement and Morphology of Clitics. Stanford, Calif.: CSLI
Publications.
Klavans, J. 1982. Some Problems in a Theory of Clitics. Bloomington: Indiana University
Linguistics Club.
Klavans, J. 1985. ‘The independence of syntax and phonology in cliticization.’ Language
61: 95–120.
Percus, O. 1993. ‘The captious clitic: problems in Serbo-Croatian clitic placement.’ Ms,
MIT.
Progovac, L. 1996. ‘Clitics in Serbian/Croatian: Comp as the second position.’ In:
Halpern, A. L. and A.M. Zwicky (eds.) Approaching Second: Second Position Clitics
and Related Phenomena. Stanford Calif.: CSLI Publications, 411–428.
Progovac, L. 1998. ‘Clitics second and verb second.’ Paper presented at the Comparative
Slavic Morphosyntax Workshop, held at McCormick’s Creek State Park, 5–6 June.
To appear in Proceedings.
Progovac, L. This volume. ‘Where do clitics cluster?’
Rivero, M.-L. 1994. ‘Clause structure and V-movement in the languages of the Balkans.’
Natural Language and Linguistic Theory 12: 63–120.
Rivero, M.-L. 1995. ‘Diachrony and the status of greed as an economy principle.’ Paper
presented at the Fourth Diachronic Syntax Conference, Montreal.
Rivero, M. -L. 1997. ‘On two locations for complement clitic pronouns: Serbo-Croatian,
Bulgarian and Old Spanish.’ In: van Kemenade, A. and N. Vincent (eds.) Parame-
ters of Morphosyntactic Change. Cambridge: Cambridge University Press, 170–206.
Roberts, I. 1994. ‘Second position effects and agreement in Comp.’ Paper presented at
the Third Annual Workshop on Formal Approaches to Slavic Linguistics, University
of Maryland.
316 OLGA MIŠESKA TOMIĆ

Rudin, C. 1997. ‘AgrO and Bulgarian pronominal clitics.’ In: Lindseth, M. and S. Franks
(eds.) Proceedings of FASL V: the Indiana Meeting. Ann Arbor, University of
Michigan Slavic Materials, 224–253.
Schütze, C. T. 1994. ‘Serbo-Croatian second-position clitic placement and the phonology-
syntax interface.’ MIT Working Papers in Linguistics 21: 373–473.
Spencer, A. 1991. Morphological Theory. Oxford: Blackwell.
Sportiche, D. 1996. ‘Clitic constructions.’ In: Rooryck, J. and L. Zaring (eds.) Phrase
Structure and the Lexicon. Dordrecht: Kluwer, 213–277.
Suñer, M. 1988. ‘The role of agreement in clitic-doubled constructions.’ Natural Lan-
guage and Linguistic Theory 6: 391–434.
Tomić, O. M. 1996. ‘The Balkan Slavic clausal clitics.’ Natural Language and Linguistic
Theory 14: 811–872.
Tomić, O. M. 1997a. ‘Finiteness and type of clitichood.’ Paper presented at the Genera-
tive Grammar Conference Oviedo, Spain.
Tomić, O. M. 1997b. ‘Non-first as a default clitic position.’ Journal of Slavic Linguistics
5(2): 1–23.
Tomić, O. M. In press. ‘On clitichood.’ In Kenesei, I. (ed.) Theoretical Issues in Eastern
European Languages; Selected Papers from the Conferences on Linguistic Theory in
Eastern European Languages (CLITE). Amsterdam: John Benjamins.
Tomić, O. M. and F. Beukema 1997. ‘On clitic sites.’ Paper presented at the Workshop
on Clitic Phenomena in English and other European Languages, ESSE/4, Debrecen,
Hungary.
Wilder, C. and D. Ćavar 1994. ‘Long head movement? Verb movement and cliticization
in Croatian.’ Lingua 93: 1–58.
Zwicky, A. 1977. On Clitics. Bloomington, Indiana University Linguistics Club.
 Subject Index

This index indicates pages where subjects/terms are defined or discussed in

some detail.

Agreement - and intonation boundaries 249,

leísmo/laísmo/loísmo - 148, 166, 253-255
167 - and intonation phrases 256
object - 149, 150, 160-162, 166, - and movement 150-155,
168-177, 179, 180, 184 159-161, 182, 184, 185
Animacy 57 - and phonological reordering 3
Animacy Hierarchy 168-170, - and right-adjunction 197-200
175-177 - and specificity 265-267,
- and the preposition ’a’ 303-305
168-170 - as (functional) head 14, 16-21,
Appositive 23, 24, 75, 86, 109
- and delaying effect on clitic - as argument 201, 202, 204
placement 198, 199 - as pronominal affix 162-167
Bare plurals - climbing 17-19, 102, 103, 107,
- and existential interpretation 202, 204
233, 237-241 - cluster 3, 5, 6, 249, 250, 298,
- and generic interpretation 233, 300-302, 306, 308-314
238-241 - doubling 40, 41, 53-55, 62,
Bare singulars 231-242 153-157, 160, 161, 167-173,
Binding Theory 152, 153 175-177, 183-185, 260-262,
Clitic 268-272, 303, 305
- anchor 294, 297 - left dislocation (CLLD) 50, 55
- and agreement 147-150, - movement 74, 75, 101, 104,
160-162, 166, 168-170, 107, 111, 114
173-177 - orientation 294, 297
- and base-generation 154-156 - phrase 4
- and Case 264, 272-277 - placement 3, 5-11, 17-20, 28,
- and idiomatic VPs 202 249-252, 254, 255, 257
- projection (CLP) 139-142
318 SUBJECT INDEX

- raising 137, 138 Clitic doubling 40, 41, 53-55, 62,

- sites 300-303, 306-309 153-157, 160, 161, 167-176,
- splitting 5-9, 11 183-185, 303, 305
- third 6, 7, 19, 252, 253, 256 - and d-linking 216
- voice 160 - and definiteness 214, 216
auxiliary - 4, 15, 21, 82-84, 89, - and information structure
90, 101, 111, 251, 300, 302, 260-263, 269-271
303, 307, 308 - and object agreement 210, 225
clause-initial - 296-298, 313 - and presuppositionality 210,
dative - 261, 262, 264, 266, 267, 216, 228, 230
276, 277, 279-281 - and referentiality 216
direct object (DO) - 48-55 - and relative clauses 53-55
indirect object (IO) - 55-63 - and specificity 210, 216, 227
multiple-clause - clusters 202 - and strength 210, 216, 228,
non-referential - 144 230
order in - clusters 193, 200-202 - and topicality 268-272
partial - clusters 249-251 - and topichood 227
possessive - 137-141, 143, 266, - and [+Focus] 210, 217,
279, 281 220-225, 227, 229, 230, 241
pronominal - 4, 21, 22, 251, - and [-Focus] 210, 218, 219,
300-305, 307, 308, 312-314 224, 225, 227, 230
referential - 144 - datives 212
second position (2P) - 252, 256 - of direct objects 209-245
simple - 10, 12, 13 - parameter 211, 217
single-clause - clusters 193, 197, Clitic placement 3-11, 17-20, 28
198, 202 - as second daughter (2D) 3, 6,
special - 8, 9, 11, 13, 20, 34, 9
261, 262, 293, 294 - as second word (2W) 3, 6, 9
Tobler-Mussafia (TM) - 197, appositive and delaying effect on
306-309, 312 - 198, 199
verbal - 294-296, 298-303, 306, parenthetical and delaying effect
308-314 on - 198, 199
Wackernagel - 17, 192, 196, phonological properties of - 10,
294, 295, 299, 300, 309-314 13
Clitic clusters 4-6, 249, 250, 298, prosodic properties of - 2, 30
300-303, 306-312 second position (2P) - 3, 14-20,
multiple-clause - 202 24, 26-29, 31, 191, 192, 194,
order in - 193, 200-202 196-200, 202, 204, 294, 298,
partial - 249-251 311
single-clause - 193, 197, 198, syntactic properties of - 8, 14
202 Dislocation 152-154
clitic left - (CLLD) 50, 55
right - 214, 215, 224, 225
 SUBJECT INDEX 319

Distributed morphology 23 German 226-230, 232, 237, 244

DP-structure 262, 266, 280-282 Greek 47-68, 209-214, 216-230,
Feature movement 20-26 232-245
Features Italian 19, 172
formal - (F-features) 20, 22-24, Macedonian 32-36, 172,
26 259-285, 295, 297-300,
semantic - (S-features) 22-27, 302-304, 306, 308
35, 36 Norwegian 240
Focus 217-225, 227, 229, 230, 241, Old English (OE) 128, 129
269, 270 Old Spanish 164, 307
Fortresses 12-14 Portuguese 172
Generalised quantifiers Romanian 172
- and type-shifting 284 Russian 10, 11
Genitive Serbo-Croatian 4-8, 15-17,
- ’s 134-136, 144 28-31, 35, 36, 72-95, 97-116,
objective - 123 198-201, 205, 249-257
subjective - 123 Sesotho 40
Head Movement Constraint (HMC) Slovene 191, 192, 194-206
18, 19 Slovenian 7-9, 30, 31, 35, 36
Head-to-head movement 135 Southern Cone Spanish 156,
Information structure 170, 172
- and word order 260, 269 Spanish 19, 147-151, 153-158,
Interface Phenomenon 20-29 160-170, 172-178, 180-186,
Intonational phrases 6, 7 224
Languages Last resort 35, 36
Albanian 209-220, 222-245 Last Resort Condition 86, 87, 111
Basque 172 Lexical Decomposition Grammar
Basque Spanish 172 (LDG) 264, 274, 275
Berber 170 Lexical Integrity Hypothesis 155
Bulgarian 10-13, 29-32, 37, 38, Minimalist program 2, 9, 28, 182,
91, 121-128, 130-132, 185
134-137, 139-141, 143, Noun phrases
259-285, 307-309 internal structure of - 233
Catalan 154 Optimality Theory (OT) 26-29, 38,
Chishona 40 39
Czech 164 Parenthetical
Dutch 68, 226-228, 237 - and delaying effect on clitic
English 52, 54, 57-66, 84, 111, placement 198, 199
121-131, 133-136, 140, 141, Participles
143, 144, 255 - and movement to C 196, 310
French 91-93, 151, 164-166, Pied-piping 22, 23, 35, 36
168, 172, 304 - and relative clauses 60, 61,
Galician 172 63-65
320 SUBJECT INDEX

last resort - 35, 36 234, 245

Possessive - and topicality 283, 285
- clitic 137-141, 143 Second position (2P) cliticisation
- constructions 122, 123 - and adverbs 88, 89
- expressions 122-127, 132-134, - and delayed clitic placement
136, 141 94-99
Possessor 123, 124, 132, 143 - and filtering at PF 104-114
Preposition ’a’ 168-170 - and move clitics in PF
Pronoun 100-104
strong - 173-175, 192 - and participle movement 89-91
weak - 163 - and VP ellipsis 93, 94, 99
Prosodic Inversion (PI) 5-14, 76-85 - and wh-superiority 91-93
- as last resort 75 the strong phonology approach to
Relative clauses - 100-103
- and clitic doubling 53-55 the strong syntax approach to -
- and clitic left dislocation 85-88
(CLLD) 50, 55 the weak phonology approach to
- and definite NP head 50 - 104-114
- and direct object (DO) clitics the weak syntax approach to -
48-55 76-85
- and double object constructions Spec-head agreement 143
56-60, 62, 63, 65 SpecAgr(Poss)P 129-132, 134
- and indefinite NP head 48, 51, Specificity 175-177, 265-268, 275
54 - and case marking 303-305
- and indirect object (IO) clitics - and clitic doubling 305
55-63 - and individual
- and islands 49, 50 denotation/individuation
- and pied-piping 60, 61, 63-66 234-239, 241-245
- and preposition stranding - and property denotation
63-68 234-237, 239, 242-245
- dative shift 59-63 Syntax/phonology interface 192,
- in Greek 47-49 254
- introduced by ’pu’ 48 Tobler-Mussafia 164, 197,
- introduced by ‘pu’ 47 306-309, 312
- introduced by a wh- relative Type shifting 245
pronoun 47 - and generalised quantifiers 284
head-raising analysis of - 48, Verb movement
51-55, 63, 64 short - 19
Remnant topicalisation 8, 13 verb second (V2) 17-19, 311,
Scrambling 212, 225-230, 232, 312
In the series LINGUISTIK AKTUELL/LINGUISTICS TODAY (LA) the following titles
have been published thus far, or are scheduled for publication:
1. KLAPPENBACH, Ruth (1911-1977): Studien zur Modernen Deutschen Lexikographie.
Auswahl aus den Lexikographischen Arbeiten von Ruth Klappenbach, erweitert um drei
Beiträge von Helene Malige-Klappenbach. 1980.
2. EHLICH, Konrad & Jochen REHBEIN: Augenkommunikation. Methodenreflexion und
Beispielanalyse. 1982.
3. ABRAHAM, Werner (ed.): On the Formal Syntax of the Westgermania. Papers from
the 3rd Groningen Grammar Talks (3e Groninger Grammatikgespräche), Groningen,
January 1981. 1983.
4. ABRAHAM, Werner & Sjaak De MEIJ (eds): Topic, Focus and Configurationality.
Papers from the 6th Groningen Grammar Talks, Groningen, 1984. 1986.
5. GREWENDORF, Günther and Wolfgang STERNEFELD (eds): Scrambling and Barri-
ers. 1990.
6. BHATT, Christa, Elisabeth LÖBEL and Claudia SCHMIDT (eds): Syntactic Phrase
Structure Phenomena in Noun Phrases and Sentences. 1989.
7. ÅFARLI, Tor A.: The Syntax of Norwegian Passive Constructions. 1992.
8. FANSELOW, Gisbert (ed.): The Parametrization of Universal Grammar. 1993.
9. GELDEREN, Elly van: The Rise of Functional Categories. 1993.
10. CINQUE, Guglielmo and Guiliana GIUSTI (eds): Advances in Roumanian Linguistics.
1995.
11. LUTZ, Uli and Jürgen PAFEL (eds): On Extraction and Extraposition in German. 1995.
12. ABRAHAM, W., S. EPSTEIN, H. THRÁINSSON and C.J.W. ZWART (eds): Minimal
Ideas. Linguistic studies in the minimalist framework. 1996.
13. ALEXIADOU Artemis and T. Alan HALL (eds): Studies on Universal Grammar and
Typological Variation. 1997.
14. ANAGNOSTOPOULOU, Elena, Henk VAN RIEMSDIJK and Frans ZWARTS (eds):
Materials on Left Dislocation. 1997.
15. ROHRBACHER, Bernhard Wolfgang: Morphology-Driven Syntax. A theory of V to I
raising and pro-drop. 1999.
16. LIU, FENG-HSI: Scope and Specificity. 1997.
17. BEERMAN, Dorothee, David LEBLANC and Henk van RIEMSDIJK (eds): Rightward
Movement. 1997.
18. ALEXIADOU, Artemis: Adverb Placement. A case study in antisymmetric syntax.
1997.
19. JOSEFSSON, Gunlög: Minimal Words in a Minimal Syntax. Word formation in Swed-
ish. 1998.
20. LAENZLINGER, Christopher: Comparative Studies in Word Order Variation. Ad-
verbs, pronouns, and clause structure in Romance and Germanic. 1998.
21. KLEIN, Henny: Adverbs of Degree in Dutch and Related Languages. 1998.
22. ALEXIADOU, Artemis and Chris WILDER (eds): Possessors, Predicates and Move-
ment in the Determiner Phrase. 1998.
23. GIANNAKIDOU, Anastasia: Polarity Sensitivity as (Non)Veridical Dependency. 1998.
24. REBUSCHI, Georges and Laurice TULLER (eds): The Grammar of Focus. 1999.
25. FELSER, Claudia: Verbal Complement Clauses. A minimalist study of direct perception
constructions. 1999.
26. ACKEMA, Peter: Issues in Morphosyntax. 1999.
 °
27. RUZICKA, Rudolf: Control in Grammar and Pragmatics. A cross-linguistic study.
1999.
28. HERMANS, Ben and Marc van OOSTENDORP (eds.): The Derivational Residue in
Phonological Optimality Theory. 1999.
29. MIYAMOTO, Tadao: The Light Verb Construction in Japanese. The role of the verbal
noun. 1999.
30. BEUKEMA, Frits and Marcel den DIKKEN (eds.): Clitic Phenomena in European
Languages. 2000.
31. SVENONIUS, Peter (ed.): The Derivation of VO and OV. 2000.
32. ALEXIADOU, Artemis, Paul LAW, André MEINUNGER and Chris WILDER (eds.):
The Syntax of Relative Clauses. 2000.
33. PUSKÁS, Genoveva: Word Order in Hungarian. The syntax of È-positions. 2000.
34. REULAND, Eric (ed.): Arguments and Case. Explaining Burzio’s Generalization.
2000.
35. HRÓARSDÓTTIR, Thorbjörg. Word Order Change in Icelandic. From OV to VO.
2000.
36. GERLACH, Birgit and Janet GRIJZENHOUT (eds.): Clitics in Phonology, Morphol-
ogy and Syntax. 2000.
37. LUTZ, Uli, Gereon MÜLLER and Arnim von STECHOW (eds.): Wh-Scope Marking.
2000.
38. MEINUNGER, André: Syntactic Aspects of Topic and Comment. 2000.
39. GELDEREN, Elly van: A History of English Reflexive Pronouns. Person, ‘‘Self’’, and
Interpretability. 2000.
40. HOEKSEMA, Jack, Hotze RULLMANN, Victor SANCHEZ-VALENCIA and Ton
van der WOUDEN (eds.): Perspectives on Negation and Polarity Items. 2001.
41. ZELLER, Jochen : Particle Verbs and Local Domains. n.y.p.
42. ALEXIADOU, Artemis : Functional Structure in Nominals. Nominalization and
ergativity. n.y.p.
43. FEATHERSTON, Sam: Empty Categories in Sentence Processing. 2001.
44. TAYLAN, Eser E. (ed.): The Verb in Turkish. n.y.p.
45. ABRAHAM, Werner and C. Jan-Wouter ZWART (eds.): Issues in Formal German(ic)
Typology. n.y.p
46. PANAGIOTIDIS, Phoevos: Pronouns, clitics and Empty Nouns. ‘Pronominality’ and
licensing in syntax. n.y.p.