Studies in Greek Syntax

STUDIES IN GREEK SYNTAX
Studies in Natural Language and Linguistic Theory

VOLUME 43
Managing Editors
Liliane Haegeman, University of Geneva
Joan Maling, Brandeis University
James McCloskey, University of California, Santa Cruz
Editorial Board
Carol Georgopoulos, University of Utah
Guglielmo Cinque, University of Venice
Jane Grimshaw, Rutgers University
Michael Kenstowicz, Massachusetts Institute of Technology
Hilda Koopman, University of California, Los Angeles
Howard Lasnik, University of Connecticut at Storrs
Alec Marantz, Massachusetts Institute of Technology
John J. McCarthy, University ofMassachusetts, Amherst
Ian Roberts, University of Wales, Bangor
The titles published in this series are listed at the end of this volume.
STUDIES IN
GREEK SYNTAX
edited by
ARTEMIS ALEXIADOU
ZAS.
Berlin. Germany
GEOFFREY HORROCKS
University of Cambridge.
Cambridge. U.K.
and
MELITA STAVROU
Aristotle University ofThessaloniki.
Thessaloniki. Greece
SPRINGER-SCIENCE+BUSINESS MEDIA. B.V.
A C.I.P. Catalogue record for this book is available from the Library of Congress.
ISBN 978-90-481-5117-2
ISBN 978-94-015-9177-5 (eBook)
DOI 10.1007/978-94-015-9177-5
Printed on acid-free paper
Ali Rights Reserved

1999 Springer Science+Business Media Dordrecht
Origina11y published by Kluwer Academic Publishers in 1999
Softcover reprint of the hardcover 1st edition 1999
No part of the material protected by this copyright notice may be reproduced or
uti1ized in any form or by any means, electronic or mechanical,
inc1uding photocopying, recording or by any information storage and
retrieval system, without written permission from the copyright owner
TABLE OF CONTENTS
Introduction
Artemis Alexiadou, Geoffrey Horrocks and Melita Stavrou
1
Propositional Operators
Yoryia Agouraki
23
On the Properties of some Greek Word Order Patterns
Artemis Alexiadou
45
On Experiencers
Elena Anagnostopoulou
67
On Clitics, Prepositions and Case Licensing in Standard and
Macedonian Greek
Alexis Dimitriadis
95
Weak and Strong Polarity: Evidence from Greek
Anastasia Giannakidou
113
Clitic Placement and the Projection of Functional Categories
Michael Hegarty
135
On Control in Greek
153
Irene Philippaki-Warburton and Georgia Catsimali
Modals and the Subjunctive
Anna Roussou
169
Theta-Role Saturation in Greek Compounds
Anna-Maria di Sciullo and Angela Ralli
185
The Position and Serialization of APs in the DP: Evidence from Greek
Melita Stavrou
201
Cypriot Greek Clitics and their Positioning Restrictions
Arhonto Terzi
227
Null Operators, Clitics and Identification: a Comparison between Greek
and English
Ianthi-Maria Tsimpli
241
Tense and Temporal Adverbials in Greek
George J. Xydopoulos
263
List of Contributors
277
Subject Index
279
INTRODUCTION!
Artemis Alexiadou Geoffrey Horrocks

Z4S, Berlin
University of Cambridge
Melita Stavrou
University of Thessaloniki
The descriptive power of early Generative Grammar was tested and established
through detailed analyses of phenomena in many languages including Greek. 2 But
although it is undeniable that Greek (Gr) has been subject to extensive analysis
within the generative framework, it is also true that some twenty years ago its study
was still in its infancy; practitioners were few and far between, and there were correspondingly few opportunities for work to be presented to, and discussed by, an informed audience. Today things are very different. Due to the rapid developments
within generative theory itself, conferences dedicated to the theoretical implications
of the syntactic analysis of the language have become routine, and a sizeable community of linguists with a primary interest in Gr has grown up both in Europe and
the United States. Thus, the study of Greek, a major European language, on a par
with that of Romance, Germanic and increasingly Slavic languages, has become
important for the development of generative theory. Older patterns of thinking and
more traditional issues are being reshaped in the light of new perceptions of modem
grammatical theory, and phenomena discussed earlier primarily in discourse/pragmatic terms, e.g. free word order, are treated consistently today in formal
syntactic terms. Many of these issues are currently highly relevant for the development of a satisfactory theory of comparative syntax and we believe the analyses put
forth here will contribute to the elaboration of such a theory.
This volume contains 13 papers that were presented at the Workshop on Greek
Syntax at the Zentrum fUr Allgemeine Sprachwissenschaft ('Research Center for
General Linguistics') in Berlin in December 1994. It is the purpose of this book to
present some of the results of recent work in a number of central areas of current
interest and controversy, emphasising throughout the importance of the Greek facts
for our understanding of the theoretical issues at stake and, even more importantly,
for the development of theoretical linguistics.
The thematic organization of the present book reflects the major characteristics
of Greek. Accordingly, in section 1, we attempt to sketch' an overview of the most
basic features of the language to help those who may be unfamiliar with the
language, before we present the contents of the book. In sections 2 through 6 we
illustrate how the individual papers fit into the general picture.
A. Alexiadou et al. (eds.), Studies in Greek Syntax, 1-21.

1999 Kluwer Academic Publishers.
ARTEMIS ALEXIADOU, GEOFFREY HORROCKS AND MELITA STAVROU
1. PROPERTIES OF GR
1.1 Word Order and Morphology

Gr is a relatively free word order, pro-drop language, showing all the properties of
this type (see Chomsky 1981, Rizzi 1982 and others); the lexical subject can be
omitted, and, when present, it can appear in postverbal position (see 1a&b).
Moreover, Gr unlike Italian but like Spanish permits VSO orders (see 1c):
irthe3
the-John-nom came-3sg
(0 Janis)
b. irthe
came-3sg the-John-nom
'John came'
c. ftlise 0 Janis
ti Maria
kissed the-John-nom the-Mary-acc
'John kissed Mary'
(1) a.
(0 Janis)
After the work of Rizzi (1982) on similar Italian patterns, it has been assumed that
in (la) the overt subject occupies Spec,IP, while (lb) involves an empty expletive
pro in preverbal position. When the lexical subject is omitted, referential pro
occupies the subject position which is licensed and identified via rich agreement
(Rizzi 1986). The problem that arises with structure (lb) is how the postverbal
subject receives Case. It has been proposed (see Koopman & Sportiche 1991 among
others) that postverbal subjects receive case via government from INFL, while
preverbal subjects are assigned nominative in Spec,IP via spec-head agreement.
Within the minimalist program (Chomsky 1995), (lb) could involve covert
movement of the case features of the subject DP to INFL, while in (1a) the subject
could be argued to check its case overtly.
In the literature on Greek, some researchers have argued that preverbal subjects
are best analysed as being left dislocated (see Philippaki- Warburton 1985, Tsimpli
1990, Drachman & Klidi 1992, Horrocks 1994, Anagnostopoulou 1994, Alexiadou
1994, Alexiadou & Anagnostopoulou 1998 and others for variants). Moreover, it
has been argued that, since verbal agreement is rich enough to identify pro, lexical
subjects are necessarily always adjuncts. Such an analysis brings Greek in some
respects close to so called non-configurational languages (Philippaki-Warburton
1987, Catsimali 1990). As shown in (2), the Greek verb is fully inflected for person
and number, displaying three distinct personal endings for the singular and three for
the plural.
(2) aghap-o
aghap-as
aghap-a
'I love'
'you love'
'he, she, it loves'
aghap-a-me 'we love'

aghap-a-te 'you love'
aghap-un 'they love'
INTRODUCTION
On the other hand, others have pointed out that preverbal subjects are at least partially distinct from other topicalised constituents in displaying residual properties
characteristic of items in an A-position (Horrocks 1994).
Although there is a general consensus that inverted orders as in (lb-c) arise
when the verb occupies a slot higher than the structural position hosting the subject,
in the light of an articulated INFL projection (see below), there is no agreement
concerning the nature of the head hosting the verb, the nature of the projection hosting the subject and whether or not an expletive is required.
Greek has no infinitives. All its moods (indicative, subjunctive, imperative) are
expressed by fmite forms throughout; specifically, the 'subjunctive' mood consists of
the particle na plus fully inflected forms of the verb, being the common translational
equivalent of the English infmitive. A number of researchers have analysed na as a
subjunctive marker (Philippaki-Warburton & Veloudis 1984, Terzi 1992, Rivero
1994 among others), while others have argued that it behaves like a complementizer
(Agouraki 1991, Tsoulas 1993). We will return to the properties of na-clauses later
on.
(3) a.
b.
thelo
na fIjis
want-lsg subj go-2sg
'I want you to go'
Fije
go-2sg
'Go!'
Subjunctive
Imperative
There is also an indeclinable verbal form, the so-called gerund or active participle.
Rivero (1994) takes the gerund morpheme -ondas to occupy the same structural slot
as the subjunctive particle na (cf. also Drachman 1994, Philippaki-Warburton
1994):
(4)
treh-ondas i Maria,
heretise to Jani
running
the-Mary-nom greeted the-Jobn-acc
Aspect is a prominent feature of the language, as the perfective/imperfective distinction, present in all three moods and always marked through suffIxation (or a
different verbal stem), is fundamental (cf. (5); Rivero 1990, Joseph &
Smirniotopoulos 1994).
(5) a.
b.
elise
tin askisi
solved-perf-3sg the-problem-acc
'He/she solved the problem'
eline
tin askisi
solved-imp-3sg the problem
'He/she was solving the problem'
c.
d.
zlnse
na ti
lisi
asked-perf-3sg subj cl-acc solve-perf-3sg
'He/she asked to solve it'
zimse
na ti
lini
asked-imp-3sg subj cl-acc solve-imp-3sg
'He/she used to ask to solve it'
In current syntactic theory the clausal structJ.Ire is represented as a cascade of functional projections terminating in a verbal projection. In the work of Ouhalla (1988),
Pollock (1989) the need was recognized to break down INFL into its functional
components AGRS, Tense, AGRO. Moreover, the categories Mood, Aspect and
Voice were further suggested; it has been claimed that all of these are instantiated in
Gr, since they are part of the morphology of the Greek verb (see Rivero 1988/94,
Tsimpli 1990, Philippaki-Warburton 1989, Drachman 1991, to mention a few).
While the exact number and order of the functional categories still remains a matter
of debate, researchers agree on the fact that Gr involves V-raising to the highest
INFL head.
In addition, the categories Negation, Topic and Focus have been added to the
inventory of functional heads and their properties have been extensively discussed.
Gr is a negative concord language. It has two negation particles: dhen which cooccurs with indicative forms and mi which co-occurs with subjunctive forms, i.e.
those involving na:
(6) a.
b.
dhen thelo
neg want-lsg
'I do not want'
thelo
na min erthis
want-lsg subj neg come-2sg
'I do not want you to come'
An issue that has attracted a lot of attention in the literamre on Gr is whether or not
these two negation particles occupy the same slot in the clause strucmre and whether
Neg O precedes Moodo or the opposite, with reference also to the fact that
imperatives in Greek cannot be negated. (Rivero 1994, Drachman 1991, PhilippakiWarburton 1996 among others). Another issue that has been a matter of debate is the
namre of licensing of negative polarity items in Greek, in view of the difference in
licensers between Gr and other languages (Veloudis 1981, Klidi 1994, Tsimpli and
Roussou 1996, Agouraki's and Giannakidou's contributions). Polarity items are
partly homophonous to the negative quantifiers, the difference being that the latter
are emphatically stressed:
(7) a.
b.
na min erthi
kanislKANIS
subj neg come-3sg anybody/nobody
KANIS/*kanis
na min erthi
Nobody/anybody subj neg come-3sg
INTRODUCTION
The question of whether these two items are simply two instances of the same
lexical entry or constitute two distinct lexical entries has not yet been settled.
Moreover, there is no agreement concerning the nature of the functional category
which hosts the emphatic item, when this occurs in preverbal position. It has been
taken to be a Focus projection by some, and a NegP by others.
In the literature, one of the interpretations associated with Focus is that of an
operator in a functional projection of its own. The focused element is identified by
the feature [+F(ocus)]. Focused elements in Gr are discussed in the work of Tsimpli
(1990) and Agouraki (1990) who argue that Gr has a focus projection and point out
the similarities and differences between focus phrases and wh-phrases. Moreover,
research has shown that topicalized elements are sharply different from focused
items.
1.2 Clitics
Gr has a rich system of clitics, though there is nothing equivalent to Romance ne, y,
en clitics. Greek clitics are the weaklshortlnon-emphatic forms of the personal pronouns of the first, second and third person singular and plural, encoding features of
direct and indirect object and the possessor (with nouns), the first manifesting accusative case and the latter two genitive case (see 8), at least in Standard Greek:
(8) Sg. Gen mu 'to me'/'my'
su 'to you'/'your'
tultis/tou 'to himlher/it' 'hislhers/its'
Ace me 'me'
se 'you'
ton/tin/ton 'himlher/it'
PI Gen mas 'to usl'ours' sas 'to you/'yours'
sas 'you'
Ace mas 'us'
tus
'to them' 'theirs'
tus/tis/ta 'them'
Some linguists studying the language assume that the person and number indication
of the verbal ending identifying pro 'corresponds' to a nominative clitic (PhilippakiWarburton 1987 who follows Safrr 1985, Alexiadou & Anagnostopoulou 1998 for a
variant). True nominative clitis have been also argued to be present in the language
(see Joseph 1994).
(9)
na tos
here cl-nom
'Here he is'
Clitics precede the finite verb forms but are enclitics with (non-finite) gerunds
and the imperative, although there is a variation geographically conditioned, allowing for enclitics with both fmite and non-finite verbs.
(10)
a.
to vlepo
it see-lsg
'I see it'
b. dhjavase to
read-2sg it
'Read it!'
c. dhjavazondas to
reading
cl-acc
The relative order of the preverbal clitics is strictly indirect object-direct object, but
there is no strict order between enclitics.
(11)
a.
mu to
edhose
cl-gen cl-acc gave-3sg
'He/she gave it to me'
b. dhose
mu
to
give-2Ssg cl-gen cl-acc
c. dhose
to
mu
give-2sg cl-acc cl-gen
The appropriate derivation of the various orderings isa matter of debate among linguists (see Rivero 1988, Drachman 1991, Philippaki-Warburton 1994, Terzi 1992,
Rivero and Terzi 1995, among others).
Clitics, as in Spanish, can 'double' the corresponding arguments of the verb
(both direct and indirect objects; see Drachman 1983, Theophanopoulou-Kontou
1986, Philippaki-Warburton 1987, Iatridou 1990, Tsimpli 1992, Agouraki 1993).
Clitics in Gr double both animate and non-animate NPs. Moreover, direct object
clitic doubling, as discussed in detail in Anagnostopoulou (1994), seems to be
semantically conditioned.
(12)
a.
to
idha
to trapezi
cl-acc saw-l sg the-table-acc
b. tis
milisa
tis Marias
cl-gen talked-lsg the-Mary-gen
In the GB literature on clitic-doubling, it is often assumed that the clitic 'absorbs'

Case and the doubled-NP violates the Case Filter. To solve the problem many researchers have capitalized on what is known as 'Kayne's Generalization'. Specifically, it has been assumed that clitic-doubling is licensed in languages where a
dummy preposition (a in Spanish, pe in Romanian) can Case-license the NP. However, as shown in (12), languages like Greek have clitic-doubling even though the
NP is not introduced by a preposition, thus providing a counterexample for Kayne's
Generalisation. Because of this, Drachman (1983), Philippaki-Warburton (1987)
have suggested that the doubled NP has the status of a right dislocated phrase.
However, research has shown that it might be misleading to stress the importance of
INTRODUCTION
Kayne's generalisation over the other characteristic properties of clitic

concstructions (see Sportiche 1992 and others).
Finally, a number of issues still remain a matter of debate. For example: are
clitics best analysed as pure agreement markers or not? Do clitics move? If so, why
and how? In which ways does this movement correlate with verb-movement?
1.3 The Noun Phrase

One major issue extensively discussed within recent formulations of the theory concerns the possible parallelisms between the structure of the clause and other major
categories. Thus the field of work inaugurated by Chomsky's Remarks on Nominalisation and further elaborated in Iackendoff's X'-theory in the 1970s still feeds a
considerable amount of contemporary research. In particular, Szabolcsi (1983), Horrocks and Stavrou (1987), Abney (1987) among others have pointed out that numerous advantages are gained if NPs are analysed as being headed by a functional category, D, for determiner.
Research on the noun phrase subsequently opened up questions similar to those
being asked about the number and types of functional projections in the CPo For
example, which are, crosslinguistically, the functional projections in the DP? What
is their exact function? How strict is the parallelism between the clausal structure
viewed as an ordered series of functional categories built on top of the lexical VP,
and the nominal structure (DP), considered as the building of functional categories
on top of the lexical NP? Is it true, as has been suggested in recent literature, that
this internal parallelism between clauses and noun phrases is reflected, among other
things, in the parallel between the serial ordering of AdvPs in the clausal projection
(see Alexiadou 1994, Cinque 1995), and the ordering of APs in the nominal one?
Are adjectives maximal projections or heads, as argued for in Abney (1987)? Does
N-movement parallel V -movement and in what ways?
In Gr, nouns are fully inflected for number (singular-plural) and case. Nouns
belong to several inflectional classes, which determine the appropriate endings,
while gender is lexically determined. Adjectives are also inflected for number and
gender, every adjective belonging to its own inflectional class, which is independent
of the class to which the modified noun belongs. Adjectives overtly agree with
nouns in case, gender and number in both attributive and predicative position. The
same rules of agreement apply to determiners and pronouns.
(13)
a.
b.
i
eksipni
jineka
the-fem-sg clever-fem-sg woman-fem-sg
i
eskipnes
jinekes
the-fern-pI clever-fern-pI women-fern-pI
There are two articles, defmite and indefmite (the latter coinciding with the numeral
'one' as well as the indefmite pronoun 'some'). Defmiteness is always marked by the
defmite article but indefmiteness is performed either by the indefinite article, or
indeed by any indefinite pronoun, or through zero determination. Adjectives always

precede nouns (but see 14 below), suggesting that here is no overt movement of the
noun for checking of morphological agreement features. In this respect Gr behaves
like the Germanic languages where adjectives also precede nouns, and unlike
Romance where adjectives can follow them. In the literature on Romance, it is often
assumed that these serial orders are a result of noun movement which places the
noun in a different position relatively to the adjective, which is analysed as a
maximal projection.
However, adjectives can follow the noun in an indefinite DP. This fact has been
analysed as involving noun-movement related to the checking of (in)defIniteness
(see Karanassios 1992):
(14)
a.
*ijineka eksipni
the woman clever
b. mia jineka eksipni
a woman clever
However, this view is currently disputed by Alexiadou & Wilder (1997) and reconsidered in Alexiadou & Stavrou (to appear).
The possessor is embodied either in a full DP or in a clitic, both in the genitive
case. There are no prenominal (genitive) possessives of the Germanic or the Romance type.
(15)
to vivlio mu
thebook my
'my book'
In Greek, much as in Arabic, Hebrew and Scandinavian languages, defInite nouns

may optionally be modified by defInite adjectives, in which case the defInite adjective may either precede or follow the defInite noun, a situation which
Androutsopoulou (1994) analyses as a defIniteness agreement phenomenon:
(16)
i jineka i eksipni
the woman the clever
1.4 Argument Structure

Within Principles and Parameters Theory, it is standardly assumed that the thematic
structrure associated with a lexical item, i.e. the number and the types of roles which
a predicate assigns, must be saturated in the syntax. It is also generally recognized
that thematic information and syntactic conflguration are systematically related. In
particular, it is assumed that initial syntactic representations are built on the basis of
the thematic representations stored in the lexicon. However, there are important
problematic areas which challenge these standard views. Moreover, with the advent
INTRODUCTION
of minimalism which dispenses with the levels of Deep and Surface-structure, it is

no longer clear what the status of principles such as the theta-criterion exactly is.
Within Greek generative literature, with the exception of Markantonatou
(1992), little attention has been devoted to issues of argument structure. Thus, Anagnostopoulou's contribution dealing with Greek psychological predicates (a class
which presents non-trivial problems for theories of mapping) is important for our
understanding of the differences and similarities between Greek, English and Italian.
Moreover, there has been much work on the derivation of the argument
structure of compounds (see the references in di Sciullo and Williams 1987). Gr
compounding and its properties, have recently become a topic of study among
researchers, also stemming from the rich morphological system of the language.
1.5 Subordination
1.5.1
Complementation
Complement clauses are introduced by one of the complementizers/subordinators

and can manifest either the indicative or the subjunctive (see 17a&b).
(17)
a.
b.
ipe
oti tha erthi
0 Janis
said-3sg that fut come-3sg the-John-nom
'He/she said that John will come'
thelo
na figho
want-lsg subj go-lsg
'I want to go'
An issue of controversy among linguists is whether or not na- clauses involve

Control or not. That na-clauses do not always involve Control can be illustrated with
cases where the embedded subject need not be interpreted as coreferential with the
matrix subject and where lexical NPs can be licensed in the embedded subject
position.
(18)
0 Janis
elpizi na fiji
i Maria!
na fighun
the-John-nom hopes subj go-3sg the-Mary-nom subj go-3pl
'John hopes that Mary goes/John hopes that they go'
Some researchers (Iatridou 1993, Terzi 1993, Varlokosta 1994 and others), on the
basis offacts such as the ones presented in (19), have argued that despite the lack of
infinitives, the subject of the embedded clause must still be PRO (whether in its
original conception as simultaneously anaphoric and pronominal, or, in the spirit of
Bouchard (1984), as simply anaphoric), and have accordingly attempted to motivate
various structural modifications that permit the appearance of PRO in a non-conventional context.
10
(19)
a.
b.
ksero
na kolimbo
know-lsg subj swim-lsg
'I know how to swim'
*ksero
na kolimba (0 Petros)
know-lsg subj swim-3sg the-Peter-nom
However, others have pointed out that such control cases seem to have a rather more
semantic/pragmatic than a strictly syntactic flavour (Philippaki-Warburton 1987).
1.5.2
Relativization
Relative clauses are introduced either by the declinable relative pronoun preceded
by the defmite article o/i/to opi- 'who' or by the invariable complementizer/relativizer pu ('that'). The latter are examples of a so called null operator construction (NOC). If the relativized element represents the direct object, then a clitic
is licensed, under specific conditions (see next paragraph). If the relativized element
stands for the possessor or the object of comparison (for some speakers the indirect
object too), the clitic obligatorily copies it.
(20)
a.
b.
enas anthropos ton opio/pu

ton sinandisa
a man-nom
the who-ace/that cl-acc met-lsg
'The man who/that I met'
i kopela pu tis
milisa
the girl that cl-gen talked-lsg
The exact role of clitics in relative constructions, has been extensively discussed in
the generative literature. With respect to Greek, it has been pointed out that clitics
are better tolerated with indefmite heads which function as the direct object of the
verb of the relative clause (see (2la), Stavrou 1984, and more recently Alexiadou &
Anagnostopoulou 1996b, Tsimpli's contribution). Moreover, the presence of clitics
with defmite heads, is only possible with appositive relatives and not with restrictive
relatives.
(21)
a.
b.
*0 andhras pu ton idha

the man that cl-acc saw-lsg
0 andhras, pu ton idha,
the man that cl-acc saw-lsg
Furthermore, it has been pointed out that clitics in relative clauses are best analysed
as A'-dependent clitics (see Cinque 1990) rather than resumptive clitics, as they exhibit a selective sensitivity to islands. As (22a-b) show, the construction is sensitive
to strong islands (in (22a) a complex DP of the relative clause type) and not to weak
islands (in (22b) a tensed wh-island):
INTRODUCTION
(22)
b.
11
*ghnorisa
mja jineka
pu sinandisa ton andhra
got to know-lsg a woman-acc that met-lsg the man-acc
pu tin
pandreftike
that cl-acc married-3sg
'I got to know a woman that I met the man that married her'
ghnorisa
mja jineka pu dhen ksero
got-to-know-lsg a woman that neg know-lsg
pjos tin
pandreftike
who cl-acc married-3sg
'I got to know a woman that I do not know who married her'
In general, in Greek as well as in other languages, A'-dependent clitics (the term is

taken from Cinque 1990) are licensed when their antecedent is referentiallD-linked
(cf. Iatridou 1991, Anagnostopoulou 1994 for Greek, Dobrovie-Sorin 1990 for
Romanian among others) while they are not licensed in 'operator-variable' chains
headed by non-referential syntactic operators (cf. Rizzi 1990 and Cinque 1990 for a
systematic discussion of the properties of 'binding chains' as opposed to 'operator
variable' chains). Given that, generally speaking, defmite DPs qualify as referential
syntactic operators while indefmites tend to qualify as non-referential syntactic
operators, it is not clear why we obtain the reverse effect in restrictive relatives.
1.6 This Volume
The articles of this book deal with many of the above mentioned major features of
the language, and focus on the following central issues:
(1) word order patterns, problems of clausal and nominal structure
(2) clitics in standard Gr and its dialects
(3) the nature of sentential operators and the licensing of negative polarity items
(4) control and non-fmite clauses revisited
(5) thematic roles and their grammatical realisation
Though not every paper addresses directly the often difficult empirical and theoretical issues that arise in the wake of Chomsky's Minimalist Programme, the impact of
this rapidly evolving paradigm is apparent in a number of the contributions. The
articles included, apart from demonstrating the current vitality and quality of syntactic and syntax-related studies based on the analysis of Gr, also highlight the importance of the language for the formulation of current theory. The papers not only
present new facts about this language, but also, and even more importantly, provide
important insights into the whole range of issues at the heart of contemporary
syntactic research, thus making a distinctive contribution to the ultimate quest for a
satisfactory characterisation ofUG.
We tum now to the survey of the individual papers.
12
2.
CLAUSE AND DP STRUCTURE: ALEXIADOU, STAVROU,

XYDOPOULOS
The paper by Alexiadou addresses the issue of word order alternations in Gr,
arguing that the various patterns are informationally specialized. The paper ftrst
points to the topic status of the preverbal subject and to some differences between
Gr and Italian preverbal subjects. Second, evidence is provided that in inverted
constructions lexical subjects are VP internal and that no expletive is present.
Crucially, the paper argues that no functional specifter is occupied by subjects in
Greek as a result of A-movement or as a result of expletive insertion. In this respect
Gr differs from languages like Icelandic and Celtic which in inverted constructions
show instances of short A-movement to Spec,TP. Furthermore, the paper suggests
that in Gr the distinction between 'thetic' and 'categorical' statements is mediated by
the various word orders, vsa orders being appropriate for the former, sva orders
being appropriate for the latter. An important new conclusion for Gr, based on
binding asymmetry facts, is that vas order arises via leftward scrambling of the
object to a higher Spec position rather than, as conventionally assumed, through
right-adjunction of the subject to VP. Since VP fmal subjects are necessarily
focused, the movement is motivated by the necessity that the most prominent accent
must fall on the most deeply embedded element within the clause nucleus. In this
way the relatively 'free' constituent order of Gr turns out to be essentially
grammatically controlled.
Concerning the structure of noun phrases and the issue of noun-movement,
Stavrou's article constitutes a wide-ranging discussion of the semantic and pragmatic
factors involved in the positioning and ordering of sequences of AdjPs in Gr. By
linking her conclusions to a motivated syntactic framework, which allows for Nmovement in indefmite DPs (see Karanassios 1992) to the head position of a functional projection (FP) located between DP and NP, and which also permits AdjP to
appear both in Spec of NP and in the Spec of this FP, many previously odd facts
about Gr are given a natural explanation. In particular, the restriction that post-NAdjPs occur only in indefmite DPs follows from the assumption that the defmite
article normally stands in the head position of FP; this is therefore filled by Nmovement only in indefmites, thus leaving behind the AdjP in the Spec of NP.
Similarly, the fact that the number of adjectives involved in 'stacked' (layered) readings is normally restricted to two follows directly from the existence of only two
possible c-commanding Spec positions (SpecFP c-commanding SpecNP, which in
tum c-commands the modified N). Since, however, there are cases where more than
two AdjPs can precede the noun under the 'stacked' reading, it is further assumed
that the lower (NP) part of the whole DP projection is structured in a layered
fashion, being in certain respects parallel to the Larsonian VP shell (1988). In this
way more specifters are provided in order to host as many classifying adjectives as
may be syntactically realized. This assumption further explains rare cases in which
the noun, after moving to the intermediate functional projection (F) appears to be
followed by more than one AdjP. The sharp contrast between the 'conjoined' and the
'stacked' readings of a series of adjectives is also linked to a clear difference of
INTRODUCTION
l3
syntactic structure (the former involving linear iteration in a single position).

Finally, unexpected sequences consisting of a classifier-like adjective and a noun
followed by another adjective are assumed to involve the formation of compoundlike structures between the classifier-like adjective and the noun, the resulting
elements denoting specific subsets of the denotation of the noun in isolation and
being able to move as a single unit.
Xydopoulos's paper discusses the tense system and the distribution and semantics of some temporal adverbs in Gr clause structure. The author analyses the Gr
tense system in neo-Reichenbachian terms (cf. Hornstein 1990). This model recognizes three primitive entities of time, the Speech time (or S point), the Event time
(or the E point), and the Reference time (or the R point). These are related to each
other only in the form of SIR and EIR according to the theory. Xydopoulos proposes
an inventory for Gr Tenses, but leaves aside the Future Perfect and the Future
Perfect in Past, which, according to him, have a modal interpretation. Moreover, he
suggests a mapping mechanism for the syntactic representation of tense departing
from the proposal put forth in Giorgi and Pianesi (1991). Though he accepts their
view that temporal relations are linked to different morphological instantiations, he
considers, contrary to Giorgi & Pianesi, that tense is a purely functional and not a
lexical head. In his analysis of adverb placement, he follows the general tendency in
the literature towards treating adverbs as referential expressions (see among others,
Em;: 1987, and Alexiadou 1994, 1997 for Gr). He exemplifies the sensitivity some
deictic adverbs have to the temporal specification of the predicate they modify (see
also Smith 1981, and Alexiadou 1994, 1997 for Gr) and he argues against a
syntactic treatment of adverbs it la Kayne (1994), proposing an alternative analysis
to that of Alexiadou (1994). In particular, following a theory of phrase structure
such as the one proposed in Brody (1994) and Manzini (1994), he argues that
adverbs, being co-referential 'antecedents' of Tense (see Partee 1973), are adjoined
to the TPs of the clause they modify. Thus, they determine the temporal reference of
the associated Ts in a variety of ways.
3. CLITICS IN STANDARD GR AND DIALECTS: HEGARTY, TERZI,

DIMITRIADIS, TSIMPLI
Hegarty's paper addresses the familar problem of different clitic placement in fmite
(proclitic) and non-finite (enclitic) clauses within a Minimalist framework, using
data from Gr and Romance. Hegarty adopts the mechanics of Madeira's (1992) and
Manzini's (1994) approach and implements them in terms of the distinction between
the checking of V-features and N-features in the framework of Chomsky (1993,
1995a, 1995b), based on an exclusive Agro-orientation of the object clitic in
Spanish, Standard Italian, and Greek (see also Tsimpli's contribution). Hegarty
argues that, in these languages, the alternation between [c1- V] and [V-cl] orders
arises from the possibility of dissociating the checking of N-features on nominals
against a functional head from the context of theta-role assignment (or checking) in
the presence of the verb; the possibility of this dissociation is tied to the null subject
14
property of these languages. Hegarty points out that in gerunds Agro, lacking Vfeatures, does not attract the verb; Agro has N-features attracting the clitic, so the
clitic alone moves to Agro. Tinf/ger has V-features attracting the verb. However,
under the Shortest Move condition, the verb cannot skip Agro on its way to T since
Agro is a nearer L-related head. Therefore, verb movement to T is effected by leftadjunction of [V+Iinf/ged to [cl+Agro], followed by left adjunction of
[V+Iinf/ger+cl+Agro] to Tinf/ger, yielding the order V-cl. The specific
implementation of these derivations of [V-cl] and [cl-V] also provides a basis for
deriving differences in multiple object clitic orders both in fmite clauses and
gerunds in Greek, with no further assumptions beyond ones concerning the
organization ofN-features on a functional head, depending on whether that head has
V-features as well. Gerunds have no object agreement features, and they allow for a
free order of merging of direct object Agr and indirect object Agr into the phrase
structure of the gerund, thus allowing both clitic orders with non-fmite forms.
Terzi's contribution discusses clitic placement in Cypriot Greek (CGr) fmite clauses.
CGr shows Verb-clitic order in fmite clauses, whereas Standard Greek exhibits CIVerb orders. Though CGr is subject to the ban on first position clitics, it is not a
Wackernagel language, since clitics are shown to appear in various positions provided that they are adjacent to the verb. The paper's central claim is that clitics both
in Standard and CGr adjoin to a functional head, which occupies the same position
in the clause structure, and takes IP as its complement. Their different surface
position in each variety is seen as the result of Verb-movement: in CGr, but not in
Standard Greek, the verb can move to Mood. Given that clitics adjoin to F,
adjunction of the verb to F would constitute multiple adjunction, a configuration
ruled out by Kayne's (1994) Linear Correspondence Axiom. It is argued that verb
movement is related to the strong features that clitics have in CGr which must be
licensed in the internal domain of a functional head with operator like-properties
(Negation, Focus); it is only in the absence of a functional head of the appropriate
type that verb movement to Mood takes place.
The paper by Dimitriadis also focuses on geographical varieties of Gr by considering the problems raised by 'dative' clitics in Northern and Southern Gr dialects. Indirect objects in Gr surface either as PPs headed by a 'light' (semantically empty)
preposition, or as DPs with overt genitive in Southern Greek or accusative case in
Norther Greek. When objects surface as PPs, there is no clitic doubling (contrary to
the familiar Spanish facts), while when they surface as DPs doubling is, if not
obligatory, then at least routine. These facts suggest that Gr violates Kayne's
generalization. On the assumption that a bare NP cannot appear because it fails to be
Case-licensed, Dimitriadis argues that the clitic here is in effect a 'mediating element', in some ways analogous to the light preposition of the alternative
construction, which is needed to assign case to the associated DP, thus avoiding
Case Filter violations. The analysis Dimitriadis presents treats the difference
between Greek, which violates Kayne's generalization, and Spanish, which obeys it,
INTRODUCTION
15
as a difference not in the presence or absep.ce of Case, but in the identity of the Case
assigned by indirect object clitics. This reduces Kayne's generalization to an
accident of morphology: languages that obey it have clitics that assign Case which
cannot be realized by bare NPs. Languages that violate it have clitics that assign
better-behaved Case.
Tsimpli's paper provides an interesting theoretical and empirical link between cliticcontaining and operator-variable structures, examining clauses involving null and
overt operators in English and in Gr, alongside the distribution of clitics in Deleted
Object Constructions (of the 'tough'- type), purposive clauses, matrix interrogatives,
restrictive and appositive relative clauses. All of these involve the presence of an
operator, realized or null. Tsimpli shows that an object clitic is obligatory whenever
an operator can bind the clitic (or the empty subject, in the constructions where there
is one), as when it is non-quantificational; coreferentiality in these cases between the
clitic and its antecedent in an argument position in the matrix clause is further due to
the predicational relationship between the antecedent and the CP clause. In the congeneric English sentences there is a null epithet with a fixed reference rather than an
overt pronoun, because feature identifiers are not available. The clitic is excluded,
however, when the relevant operator is quantificational (thus lacking a referential
index), since the clitic is a specificity marker. In relative clauses with indefinite
heads, a clitic is optionally present, and this is due to the fact that the indefmite head
of the RC in these cases gets a specific reading (Ene; 1991); given further the predicational relationship between the RC and the head of which this is predicated, it is
assumed that the clitic enters a predication relationship with the indefmite NP head.
This relationship is mediated through the (null) operator in the CP which A'-binds
the c1itic. In the case of appositive relatives, the operator is assumed to be non-quantificational, so that it does not contribute to the recoverability of the reference and
Case feature; thus, the clitic acquires a referential index with the head NP in A-position via A'-binding plus predication indexing. Similarly, in embedded interrogatives
the clitic can be present because there is a non-quantificational operator in the local
CP which can A'-bind the clitic. Tsimpli's account of the availability vs. non-availability of an empty category in the two languages makes further use of Cardinaletti
& Starke's (1994) theory of strong-weak and clitic pronouns. Thus, for Tsimpli,
clitics are feature-identifiers in AGR-O position in compliance with the 'weakest'
form of pronouns in C&S's system. English pronouns and Greek clitics differ in that
in the former there are no clitics (i.e. the 'weakest' pronouns), whereas Greek has
strong and clitic pronouns, lacking the 'weak' -intermediate- pronoun type. A cluster
of otherwise puzzling differences between the two languages follows from this
'asymmetry' .
4. NEGATIVE POLARITY ITEMS AND SENTENTIAL OPERATORS:

GIANNAKIDOU, AGOURAKI, ROUSSOU
The properties of operators is the main subject of the papers of this section.
16
ARTEMIS ALEXlADOU, GEOFFREY HORROCKS AND MELITA STAVROU
Giannakidou's article, which helps our understanding of the licensing of negative

polarity items in Gr, builds on work by Ladusaw (1992, 1994) by distinguishing two
mutually exclusive kinds of polarity involving the strong ('non') and weak ('any/a')
construal of negative indefInites (the same words are used in both functions in Gr).
She therefore rejects approaches such as that of Zwarts (1993), which propose a
sliding scale of licenser strength. Strong negative polarity items (NP1s) bear emphatic stress in Gr and it is shown that they are licensed in the restrictor position of a
semantic structure within the scope of a strong polarity operator (a spec-head relation, as Greek shows negative concord). Weak NP1s by contrast lack emphatic
stress, require a weak polarity operator and are licensed within the matrix of a
semantic structure. The relevant operators are classifIed into non-veridical ones (not
entailing the truth of their argument proposition), and averidical (entailing the falsity
of their argument proposition), and it is shown that non-veridical operators in Gr can
only license weakly, with negative polarity items always providing existential
quantifIers to the interpretation (e.g. the translation of 'John denied that he said
anything' can only mean 'John denied that there was anything that he said'). By
contrast, averidical operators in Gr allow either strong or weak construal, with the
former involving a universal quantifIer ('For all people, it is not the case that I saw
any'), the latter an existential quantifIer as before.
The article by Agouraki argues, primarily on the basis of data from Gr, that there is
a natural class of propositional operators (compare the list given in connection with
Roussou's contribution) which constitutes a primitive of VG. The elements in question are functional heads which, inter alia, select propositional arguments, license
polarity items, induce island effects, and affect c-selection. An important secondary
observation concerns the fact that arguments may be s-selected by one head but cselected by another (including here a propositional operator). Thus, only the fIrst of
the following pair of sentences is grammatical because, even though 'wonder' has
the required interrogative complement in both cases, it is the presence of the overt
wh-complementizer that licenses the complement infmitive.
(20)
a.
b.
I wonder [whether to go]

*1 wonder [if to go]
The paper presents a number of counterarguments to Giannakidou's with respect to

the role of non-veridical operators in negative polarity item licensing.
Roussou's article provides an interesting link with Philippaki-Wabrurton and Catsimali's paper which deals with Control. The basic claim is that embedded subjunctives in Gr are licensed by matrix sentential operators such as negative, Q, conditional, necessity and possibility, and that they therefore have much in common with
indefInite expressions (i.e. by functioning as a kind of polarity item, cf. Manzini,
1994b). This is expressed by means of a local head dependency between the
relevant matrix and the embedded (subjunctive) T. An important distinction is
INTRODUCTION
17
drawn, however between epistemic ('think' etc.) and volitional predicates in that the
fonner are not normally inherently modal and so may take subjunctive complements
only in the context of an appropriate matrix modality (neg, Q, etc.). Furthennore,
they alone permit independent time reference in their complements, thus permitting
the appearance of past tense forms in combinaton with the subjunctive-marking
particle na. This difference is explained as follows. Like true epistemic modals
expressing possibility and necessity, epistemic predicates like 'think' necessarily
display a 'pleonastic' present tense (the verb fonn regularly used for temporally
'neutral' situations, reflecting here the fact that the expression of epistemic modality
is time-independent). This means that the normal time-anchoring of an embedded
verb to the time reference of the matrix fails; instead, the T-dependency is headed
here by an expletive T in the matrix, and the temporal evaluation of the chain that it
heads must then be derived from the time reference of the embedded T. The
temporal independence of the subjunctive complements of epistemic verbs is thus a
natural consequence of the analysis.
5. CONTROL AND NON-FINITE CLAUSES REVISITED: PHILIPPAKIWARBURTON & CATSIMALI

Philippaki-Warburton and Catsimali argue, on the basis of evidence that the controlled subject must be assigned nominative case, that the non-realized subject in
subjunctive embedded clauses should be treated as an instance of pro (with the usual
agreement/case assignment properties of subjects in fmite clauses) and not PRO.
Building on Borer (1989), the authors suggest that a distinction must therefore be
drawn between the semantic aspects of control (the obligatory co-reference of the
embedded subject with a matrix argument) and its syntactic aspects (the appearance
of PRO with an infmitival). Since Gr lacks infmitives, its treatment of control
phenomena can involve only the ftrst of these, thus permitting the obligatorily
anaphoric use of pro.
6. THEMATIC STRUCTURE: ANAGNOSTOPOULOU, DI SCIULLO &

RALLI
Anagnostopoulou's paper deals with an interesting set of problems that arise from
the analysis of psychological predicates. Proposals such as that of Baker (1988)
involve an essentially algorithmic correspondence between arguments ranked on a
thematic hierachy and syntactic projections (the Universal Thematic Assignment
Hypothesis). Psychological predicates, however, split with respect to the treatment
of experiencer arguments, which apparently appear as subjects with verbs like 'fear'
but as objects with verbs like 'frighten'. Anagnostopoulou argues on the basis of Gr
data that the properties of Experiencer-Object predicates support the view advocated
in Grimshaw (1990) that the mapping between thematic structure and syntactic
structure is not as straightforward as the UTAH stipulates. She demonstrates, by
presenting a battery of tests, that certain properties of 'frighten' type predicates do
18
not follow from a causative analysis. In particular, she shows that the 'object' in
these cases in fact patterns with the dative experiencers of verbs like 'please' Cit
pleases me' etc.) in displaying subject-like properties and fails to pattern with
'normal' objects by requiring obligatory clitic doubling. Assuming Marantz's (1991)
theory of Case, Anangostopoulou proposes that in the case of psychological
causative predicates, the experiencer is lexically specified or quirky and appears in
initial position to satisfy the EPP. Hence, the theme argument surfaces in nominative
(the unmarked case). On the other hand, in the case of agentive psychological
predicates, the experiencer can receive 'dependent' case, since the Agent cannot have
lexically specified case.
The article by Di Sciullo and Ralli deals with the problem of theta-role saturation in
compounds. Their principal argument is that not only may theta-roles be saturated
inside such compounds but that the roles in question are in principle unrestricted;
the apparent restriction in e.g English are shown to be the product of 'weak'
inflectional morphology. Having shown that argument structure in compounds is
configurationally defmed, with the availability of additional structure (via leftward
recursion) following from the presence of strong morphology, the authors argue that
the number of roles may readily exceed one if left recursion is permitted, since
elements may be moved to pre-head 'adjunct' positions. The Gr facts are then
shown to follow directly from the configurational properties of compounds and the
strong morphological character of the language, with the presence of an overt
inflectional suffIx in Gr compounds permitting non-restricted theta-role saturation.
NOTES
For advice and assistance in connection with the preparation of this volume, we would like to
thank Marcel den Dikken, Eric Haeberli, Ursula Kleinhenz, and Jamal Ouhalla. We are grateful to Annie
Kuipers, Polly Margules and Laura Walsh for editorial advice and to two anonymous reviewers for
helpful comments and suggestions. The workshop was partly funded by the Research Center for General
Linguistics (Forschungsschwerpunkt flir Allgemeine Sprachwissenschaft) whose support is hereby
gratefully acknowledged.
2
Householder, F., A. Koustoudas & K. Kazazis (1964), Koutsoudas, A. (1966) Warburton, I. (1970),
among others in both the US and Europe; see also the overview in Theophanopoulou-Kontou (1994:
11).
3
A note in the mode of transcribing the Greek data is in order here. For typographical reasons
conventional letters (or combinations ofJetters) of the Latin alphabet (and not the IPA) have been used
throughout the volume in a uniform way to represent a broad phonetic transcription. The following
monophonernic sounds of Greek have been represented as follows: the velar voiced fricative by gh, the
dental voiced fricative by dh, the dental voiceless fricative by th, the labio-dental voiceless fricative by
f, the velar voiceless fricative by h. [b], [d], [g] are used for the bilabial, dental and velar stop
respectively, whereas [v] denotes the labio-dental voiced fricative. The letter-symbol j is used for (a) the
palatal variant of gh in front of high front vowels (i, e) and (b) the glide (voiced/voiceless) before
unstressed high front vowels (a variant of the vowel [iD.
References
Abney, S. (1987) The English Noun Phrase in its Sentential Aspect, Ph.D. Dissertation, MIT.
INTRODUCTION
19
Agouraki, Y. (1990) On the Projection of Maximal Categories: the Case ofCP and FP in Modem Greek,
UCL Working Papers in Linguistics 2,183-199.
Agouraki, Y. (1991) A Modem Greek Complementizer and its Significance for Universal Grammar,
UCL Working Papers in Linguistics 3, 1-24.
Agouraki, Y. (1993) Spec-head Licensing: the Case of Foci, Clitic-constructions and Polarity Items: a
Case Study of Modem Greek, Ph.D. Dissertation, UCL.
Alexiadou, A. (1994) Issues in the Syntax of Adverbs, Ph.D. Dissertation, Universitat Potsdam.
Al~xiadou, A. (1997) Adverb Placement: a Case Study in Antisymmetric Syntax, John Benjamins,
Amsterdam.
Alexiadou, A. & E. Anagnostopoulou (1996) Clitics in Restrictive Relatives: an Antisymmetric
Account, to appear in Studies in Greek Linguistics 17.
Alexiadou, A. & E. Anagnostopoulou (1998) Paremetrizing AGR: Word Order, Verb-movement and
EPP- checking, to appear in Natural Language and Linguistic Theory.
Alexiadou, A. & M. Stavrou (1997) Crosslinguistic Asymmetries in N-movement: a View from
Morphology, ZAS Papers in Linguistics 8, 1-16.
Alexiadou, A. & C. Wilder (1997) Adjectival Modification and Multiple Determiners, to appear in A.
Alexiadou & C. Wilder (eds.) Possessors, Predicates and Movement in the DP, John Benjamins,
Amsterdam.
Anagnostopoulou, E. (1994) Clitic Dependencies in Modern Greek, Ph.D. Dissertation, Universitat
Salzburg.
Androutsopoulou, A. (1994) The Distribution of Definite Determiners and the Syntax of Greek DPs, to
appear in Proceedings ofCLS.
Baker, (1988) Incorporation: a Theory of Grammatical Function Changing. Chicago University Press.
Borer, H. (1989) Anaphoric AGR, in O. Jaeggli & K. Safir (eds.) The Null Subject Parameter, Kluwer,
Dordrecht, pp. 69-9\.
Bouchard, D. (1984) On the Content ofEmpty Categories, Foris, Dordrecht.
Brody, M. (1994) Phrase Structure and Dependence, UCL Working Papers in Linguistics 6,1-33.
Cardinaletti, A. & M. Starke (1994) The Typology of Structural Deficiency: on the Three Grammatical
Classes, University of Venice Working Papers in Linguistics 4.
Catsimali, G. (1990) Case in Modem Greek: Implications for Clause Structure, Ph.D. Dissertation,
University of Reading.
Chomsky, N. (1981) Lectures on Government and Binding, Foris, Dordrecht.
Chomsky, N. (1993) A Minimalist Program for Linguistic Theory, in K. Hale and S. J. Keyser (eds.),
The View from Building 20, MIT Press, Cambridge, Mass, pp. 1-52.
Chomsky, N. (1995) The Minimalist Program, MIT Press, Cambridge, Mass.
Cinque, G. (1990) Types ofA-bar Dependencies, MIT Press, Cambridge, Mass.
Cinque, G. (1995) Adverbs and the Universal Hierarchy of Functional Projections, GLOW Newsletter
34,14-15.
Dobrovie-Sorin, C. (1990) Clitic Doubling, Wh Movement and Quantification in Romanian, Linguistic
Inquiry 21, 351-397.
Drachman. G. (1983) Parameters and Functions ofClitic Doubling in Modem Greek, paper presented at
the 4th Meeting on Greek Linguistics, University of Thessaloniki.
Drachman, G. (1991) Clitic Placement, unpublished manuscript, Universitlit Salzburg.
Drachman, G. (1994) Verb Movement and Minimal Clauses, in I. Philippaki-Warburton, K. Nikolaidis
& M. Sifianou (eds.) Themes in Greek LinguistiCS, Amsterdam, John Benjamins, pp. 45-52.
Drachman, G & S. Klidi (1992) The Proper Treatment of Adverbial Questions in Greek: the Extended
Minimal Structure Hypothesis, Studies in Greek Linguistics 13, 371-389.
En~, M. (1987) Anchoring Conditions for Tense, Linguistic Inquiry 18, 633-657.
En~, M. (1991) The Semantics of Specificity, Linguistic Inquiry 22,1-25.
Giorgi, A. & F. Pianesi (1991) Towards a Syntax of Temporal Representations, Probus 3,1-27.
Grimshaw, 1. (1991) Argument Structure, MIT Press, Cambridge, Mass.
Hornstein, N. (1990) As Time Goes by: Tense and Universal Grammar, MIT Press, Cambridge, Mass.
Horrocks, G. (1994) Subjects and Configurationality, Journal ofLinguistics 30, 81-109.
Horrocks, G. & M. Stavrou (1987) Bounding Theory and Greek Syntax: Evidence for wh-movement in
NP, Journal ofLinguistics 23, 79-108.
20
Householder, F., K. Kazazis & A. Koutsoudas (1964) Reference Grammar of Literal Dhimotiki, Indiana
University Press, Indiana.
Iatridou, S. (1991) Clitics and Island Effects, unpublished manuscript, MIT.
Iatridou, S. (1993) On Nominative Case Assignment and a few Related Things, MIT Working Papers in
Linguistics 19, 175-198.
Joseph, B. (1994) On Weak Subjects and Pro-drop in Greek, in I. Philippaki-Warburton, K. Nikolaidis
& M. Sifianou (eds.) Themes in Greek Linguistics, John Benjamins, Amsterdam pp.21-32.
Joseph, B. & J. Smimiotopoulos (1994) The Morphosyntax of the Modem Greek Verb as Morphology
and not Syntax, Linguistic Inquiry 24, 388-398.
Karanassios,G. (1992) Syntaxe Comparee du Groupe Nominal en Grec et dans d'autres Langues,
Doctorat d'etat, Universite de Paris VIII, Vincennes.
Kayne, R. (1994) The Antisymmetry of Syntax, MIT Press, Cambridge, Mass.
Klidi, S. (1994) l:-rOlXtta AQV1]u1cr)C; IIoAtlCOTrJ-rac;, AQV1JUlCOt IIoaolittlC"t'tC; lCat AQV1JUlCOt
TtAtmtC; ma Nta EAA1]vtlCa, Studies in Greek Linguistics 15, 451-461.
Koopman, H. & D. Sportiche (1991) The Position of Subjects, Lingua 85, 211-285.
Koutsoudas, A. (1966) Writing Transformational Grammars: An Introduction, McGraw-Hill Book
Company, New York.
Ladusaw, W.A. (1992) Expressing Negation, Proceedings of SALT 11,237-259, Ohio State University.
Ladusaw, W.A. (1994) Thetic & Categorical, Stage & Individuals, Weak & Strong, Proceedings of
SALT IV, 220-229.
Larson, R. (1988) On the Double Object Construction, Linguistic Inquiry 19, 335-392.
Madeira, A. M. (1992) On Clitic Placement in European Portuguese, UCL Working Papers in
Manzini, M. R. (1994a) Triggers for Verb-Second: Germanic and Romance, The LinguisticReview 11,
299-314.
Manzini, R. (1994b) Syntactic Dependencies and their Properties, UCL Working Papers in Linguistics
6,205-217.
Marantz, A. (1991) Case and Licensing, Proceedings of ESCOL, 234-253.
Markantonatou, S. (1992) The Syntax of Modem Greek Noun Phrases with a Derived Nominal Head,
Ph.D. Dissertation, University of Essex.
Ouhalla, J. (1988) The Syntax of Head Movement, Ph.D. Dissertation, UCL.
Partee, B. (1973) Some Structural Analogies between Tenses and Pronouns in English, The Journal of
Philosophy 70,601-9.
Philippaki- Warburton, I. (1985) Word Order in Modem Greek, Transactions of the Philological Society
2,113-143.
Philippaki-Warburton, I. (1987) Empty Categories and the pro-drop Parameter in Greek, Journal of
Philippaki-Warburton, I. (1989) Subjects in English and in Greek, unpublished manuscript, University
of Reading.
Philippaki-Warburton, I. (1994) V-movement and Clitics in Modem Greek, in I. Philippaki-Warburton,
K Nikolaidis and M. Sifianu (eds.) Themes in Greek Linguistics, John Benjamins, Amsterdam, pp.
53-60.
Philippaki-Warburton, I. (1996) Functional Categories and Modem Greek Syntax, GLOW Newsletter
36,89-90.
Philippaki-Warburton, I. & B. Joseph (1987) Modern Greek, Croom Helm, London.
Philippaki, I. & J. Veloudis (1984) The Subjunctive in Complement Clauses, Studies in Greek
Pollock, J-Y. (1989) Verb Movement, UG and the Structure ofIP, Linguistic Inquiry 20, 365- 424.
Rizzi, L. (1982) Issues on Italian Syntax, Foris, Dordrecht.
Rizzi, L. (1986) Null Objects in Italian and the Theory of pro, Linguistic Inquiry 17, 501-558.
Rizzi, L. (1990) Relativized Minimality, MIT Press, Cambridge, Mass.
Rivero, M-L. (1990) The Location of Non-Active Voice in Albanian and Modem Greek, Linguistic
Inquiry 21, 135-146.
Rivero, M-L. (1988/1994) Verb Movement and the Structure of IP in the languages of the Balkans,
Natural Language and Linguistic Theory 12, 63-120.
INTRODUCTION
21
Rivero, M.-L. & A. Terzi (1995) Imperatives, V-movement and Logical Mood, Journal of Linguistics
33,301-332.
Sciullo di, A-M & E. Williams (1987) On the Definition of Word, MIT Press, Cambridge, Mass.
Smith, C. (1981) Semantic and Syntactic Constraints in Temporal Interpretation, in P. Tedeshi & A.
Zaenen (eds.) Syntax and Semantics 14, Academic Publishers, New York, pp. 213-237.
Sportiche, Dominique (1992) Clitic Constructions, unpublished manuscript, UCLA.
Stavrou, M. (1984) H KAL'ttlCl] Avtrovullla on~ IIeeLOel<TtllCe~ AvacpOellCe~ IIeOTaoel~ j.!e
e~aQTl]<Jl] Aj.!eoou AvtLlCelj.!evou :n:ou eloayovtat j.!e TO ':n:ou', Studies in Greek Linguistics 2,
121-136.
Szabolsci, A. (1983) The Possessor that ran away from Home, The Linguistic Review 3,89-102.
Terzi, A. (1992) PRO in Finite Clauses: a Study of the Inflectional Heads of the Balkan Languages,
Ph.D. Dissertation, CUNY.
Theophanopoulou-Kontou, D (1986) Keve~ KaTl]yoele~ lCal KMTLlCa <Ttl] Nea EUl]VLlCl],
Glossologia 5.
Theophanopoulou-Kontou, D. (1994) Transformational Grammar and Modern Greek Syntax: an
Overview and some problematic Cases, in I. Philippaki-Warburton, K. Nikolaidis & M. Sifianou
(eds.) Themes in Greek Linguistics, John Benjamins, Amsterdam, pp. 11-20.
Tsimpli, I.-M. (1990) The Clause Structure and Word Order of Modern Greek, UCL Working Papers in
Tsimpli, I.-M. (1992) Focussing in Modem Greek, unpublished manuscript, UCL.
Tsimpli, I-M. & A. Roussou (1996) Negation and Polarity Items in Modem Greek, The Linguistic
Review 13,49-81.
Tsoulas, G. (1993) Remarks on the Structure and the Interpretation of na-clauses, Studies in Greek
Linguistics
14,196-207.
Varlokosta, S. (1994) Issues on Modern Greek Sentential Complementation, Ph.D. Dissertation
University of Maryland.
Veloudis, I. (1981) Negation in Modem Greek, Ph. D. Dissertation, University of Reading.
Warburton, I. (1970) On the Verb in Modern Greek, Mouton, The Hague.
Zwarts, F. (1993) Three types of Polarity, to appear in F. Hamm & E. Hinrichs (eds.) Semantics.
PROPOSITIONAL OPERATORS!
y oryia Agouraki
University of Cyprus
1. INTRODUCTION
The paper takes a class of logical/semantic operators, which share the property of
being higher predicates, and shows that they also share a number of syntactic properties. On the basis of these syntactic properties it is suggested that we should reexamine the syntax of these operators and investigate whether they are also syntactic
predicates. All the proposals in the paper are advanced as part of the common core
of human languages. None is specific to Greek (Gr)2, which is the main language of
exposition. For the crosslinguistic validity of the proposals to become apparent the
reader should take into consideration not only the Greek data but also their English
translations. This is not directly possible in two cases, i.e Polarity Item Licensing
and Root Infinitives, where English data are provided separately.
More analytically, the paper argues that NEG, Q, F(ocus), the C(onditional),
the N (ecessity) and the P( ossibility) O(perators) form a natural class (claim (1 in
virtue of having the properties in (2). Properties (2a-e) only obtain with NEG, Q, F,
CO, NO and PO. It is next argued that (2a) is the defming property of the class and
(2) is replaced with (3). Sections 2 to 6 discuss properties (2a-e). Finally, in section
7, (3) is replaced with a more economical schema and possible extensions of the
analysis are considered.
(1) NEG, Q, F, CO, NO and PO form a natural class.

(2) NEG, Q, F, CO, NO and PO:
a. modify the proposition;
b. license Polarity Items (PI's);
c. induce weak island effects;
d. license root infmitives;
e. license embedded interrogatives.
(3) Propositional Operators:
a. license PI's;
b. induce weak island effects;
c. license root infmitives;
d. license embedded interrogatives
23
A. Alexiadou et al. (eds.J, Studies in Greek Syntax, 23-44.
24
YORYIA AGOURAKI
2. PROPOSITIONAL OPERATORS
I will explain property (2a) repeated below.
(2) a.
NEG, Q, F, CO, NO and PO modify the proposition.
In virtue of modifying the proposition NEG, Q, F, CO, NO and PO will be hence

referred to as Propositional Operators. 'Proposition' is used in the logician's / semanticist's sense, to mean the semantic content of a sentence disambiguated with
reference assigned (cf. Allwood, Andersson and Dahl 1977). I follow Dowty
(1982), Eny (1981), (1987) and Hornstein (1977), (1981), (1990), who argue that
Tense is not a propositional operator. In Intensional Predicate Calculus (IPC)
Logic, NEG, CO, NO and PO are analysed as higher predicates that take other
predicates as arguments. In this way they form compound sentences. For instance,
NEG forms a compound sentence, the truth-value of which is the opposite of that of
the simple sentence it operates on. Mutatis mutandis for CO, NO and PO. Property
(2a) can be therefore reformulated as follows: NEG, Q, F, CO, NO and PO are
higher predicates. Interrogative sentences fall entirely outside the system of IPC
Logic. The same applies to F. I propose that the propositional operator analysis
extends to Q and F and will proceed to show how Q and F also belong to the class
of propositional operators.
The notion of Focus Operator needs to be adequately explained, particularly in
view of the numerous proposals on Focus in the literature. The Focus Operator
obtains only when it is the proposition as a whole that is focused. The Focus Operator discussed here is not an identical notion with the head F ofFP (see Brody 1990)
or foci in general. What about focused XP's other than the IP, i.e. the proposition?
According to the above defInition of the Focus Operator, focused XP's other than IP
do not involve the Focus Operator. Focused XP's could very well be syntactic
operators themselves. The issue is open to argumentation (cf. Agouraki 1990, 1993
where it is argued that focused XP's are syntactic operators). But focused XP's do
not involve the Focus Operator in the sense developed here and are not propositional operators. The distinction I draw between the Focus Operator and foci is the
same as the distinction between Q and wh-phrases. As far as I know this notion of
Focus Operator is not proposed anywhere else in the literature. As presented, it is
part of the defInition of the Focus Operator that it exclusively modifIes the proposition. The Focus Operator forms a compound sentence with the proposition it operates on.
Now with respect to Q it is intuitively clear that it modifIes the proposition.
Lyons (1977) already treats Q as a propositional operator. The representation of an
interrogative sentence he gives is of the form? [p], where p stands for proposition.
Arguably, Q forms a compound sentence, taking as predicate the corresponding
declarative sentence.
I will briefly return to NO and PO to specify that I follow their analysis
presented in Kratzer (1981). Kratzer's NO and PO include not only the
homonymous operators of other analyses but also the Epistemic, the Gnomic and
PROPOSITIONAL OPERA TORS
25
the Habitual operators. Kratzer describes the semantic field of modal expressions
along three axes specifying a modal relation, conditions for the modal base and
conditions for the ordering source. The modal relation is either necessity or
possibility. There are two major types of modal bases: circumstantial modal bases,
which are constituted by certain relevant facts or actual circumstances, and
epistemic modal bases, in which we rely on what we know and are interested in
what can tum out to be the case, given what we know. The function of the ordering
source is to rank the modal base according to certain standards.
So far, I have illustrated property (2a) of the {NEG, Q, F, CO, NO and PO} set,
which can argue on its own for claim (1). Property (2a) is taken to be the defining
property of the set. The list in (2) is replaced with that in (3). The semantic analysis
of propositional operators is summarized in (4).
(3) Propositional Operators:
a. license PI's;
b. induce weak island effects;
c. license root infinitives;
d. license embedded interrogatives.
(4) Propositional Operators are semantic higher predicates.
Having completed the presentation of the Propositional Operator notion, the
next question I would like to address, which will lead us to the discussion of properties (3a-d), is whether propositional operators were ever seen to form a natural
class in Syntax. Only partly. Klima (1964) refers to NEG, Q and CO as Affective
Operators and attributes to them the property of licensing Existential Polarity Items
(EPI's) in English. The term affective operator is used since to refer to these three
operators, mainly in EPI licensing environments. Notably, Rizzi (1990) uses it to
describe the environments triggering inner island effects as well.
An open question in Linguistics is the relation between Logic/Semantics and
Syntax. The relationship between the semantic representation of propositional operators (cf. (4 and their syntactic representation needs to be examined. The task
becomes more pressing given that propositional operators are shown to have syntactic effects (cf. 3(a)-(d; notably the same crosslinguistically. Observe that in
standard theory the syntax of these operators presents differences. In particular, Q
is a morpheme under C while NEG and F are functional heads. NO and PO, when
overt, are part of the verbal complex. The differences in the syntactic representation
of propositional operators sometimes produce theoretical distinctions, perhaps otherwise unmotivated. In Section 4 it is argued that the distinction between weak and
inner islands is a case in point.
Propositional operators are a subset of syntactic operators. Moreover, they are
the only logical operators that Syntax discusses. Syntax also discusses a number of
other operators that are not operators in Logic, namely quantifiers, wh-phrases, foci
and PI's. I will call these other non-logical operators nonpropositional operators to
distinguish them from logical! propositional operators. The proposed distinction
between propositional and nonpropositional operators has the following basis.
26
YORYIA AGOURAKI
Propositional operators are semantic higher predicates. Nonpropositional operators

are not. 3 Nonpropositional operators have the characteristics in (5), which contrast
them with propositional operators.
(5) NonpropositionalOperators:
a. are arguments
b. bind variables.
3. POLARITY ITEM LICENSING

One of the basic questions addressed in PI licensing is to specify the formal property shared by all polarity licensing environments. PI's standardly include NPI's,
licensed by NEG, EPI's licensed by NEG, Q and CO, and Free Choice Polarity
Items, licensed by PO (cf. (6. In Agouraki (1993), I argued that PI's also include
indefmite pronouns of the one type, licensed by NO, PO and F (cf. (7a-c. On the
basis of the different PI types and their licensers we draw the generalization in (2b),
the fIrst syntactic property of the {NEG, Q, F, CO, NO and PO} set to be discussed.
The equivalent formulation in (3a) is my proposal for the formal property shared by
all polarity licensing environments.
(6) Anyone will tell you where the Department is
(7) a. One should be more careful with personal feelings
b. One cannot always fmd time for reading.
c. One DID try to stop them
(2) b. NEG, Q, F, CO, NO and PO license PI's
(3) a. Propositional Operators license PI's
PO
NO
PO
F
EPI licensing in Gr is a case where there is full correspondence between a PI

type and the whole set of propositional operators (cf. (8 (cf. Agouraki 1993 for an
analysis of EPI licensing contexts in Gr). EPI Licensing in Gr is particularly interesting in view of the difference in licensers between Gr and many other languages.
I illustrate (8) in (9).
(8) Propositional Operators license EPI's in Gr
(9) a. *(dhent
idha
kanena
not
saw-l sg
anyone-acc
'I did not see anyone'
b. thelis
tipota
*(?)
want-2sg anything-acc
'Do you want anything?'
c. *(F)IPE
ke
kamia kuvenda
SAID-3sg FocusParticle any-acc word-acc
'He DID say something'
27
PROPOSITIONAL OPERATORS
d.
e.
f.
*(an) thelisis
tipota
tilefonise mu
if
need-2sg
anything-acc
call-Imp-2sg cl-gen
'If you need anything, call me'
*(prepi)
na
dhis
kanena
apopse
must-3sg
subj
see-2sg
anyone-acc tonight
'You should see friends tonight'
mi njazese
pu
dhen majirepse
not worry-Imp-2sg that
not
cooked-3sg
*(PO) trome tipota ekso
eat-1 pI anything-acc
out
'Don't worry that he has not cooked; we can eat out'
co
Licensing of EPl's by the Gnomic (cf. (10 or the Habitual Operator (cf. (11
is not a problem for (8). As explained in Section 2, I follow Kratzer (1981) who
analyses the Gnomic and Habitual Operators in terms of either NO or PO.
(10)
(11)
afu katalavi
kanis
to provlima
prohori
after realize-3sg anyone-nom
the problem-acc go-3sg on
sti
lisi
tu
to-the solution cl-gen
'After realizing the problem, one goes on to its solution'
0 prothipurghos
vlepi pu ke pu
kanena ipurgho
the prime minister-nom sees
occasionally any
minister-acc
'The prime minister occasionally sees some minister'
Having defended claims (3a) and (8) for all languages and Gr especially, I will
compare claim (8) with Giannakidou's (this volume) claim (12). Giannakidou's
weak NPl's are the same as the EPl's of this paper. I will have argued against claim
(12), if I show either that some veridical operators license EPl's or that some nonveridical operators do not license EPl's. Both these counterarguments can be
advanced with respect to claim (12). Giannakidou's analytic characterization ofEPI
contexts will not be considered except where it is necessary to do so for the
discussion of(12).
(12)
Non-veridical operators license EPl's in Gr
The fITst argument against (12) is that some veridical operators, namely F (cf.
(9c, the Habitual Operator (cf. (11 and the Gnomic Operator (cf. (10, license
EPl's in Gr. Giannakidou includes the Habitual Operator among EPI licensers but
fails to recognize that claim (12) is incompatible with the Habitual Operator being
an EPI licenser. Giannakidou does not include Focus and the Gnomic Operator
among EPI licensers. I consider the argument just presented the most important
argument against claim (12).
The second argument against (12) is that some non-veridical contexts do not
license EPl's in Gr. Giannakidou includes predicates that take subjunctive clauses9
28
YORYIA AGOURAKI
among EPI licensing contexts. These predicates are taken to be non-veridical

predicates. Crucially, not all predicates that take na-clauses license EPI's. pr6kite
in (13) is a non-veridical predicate; still it cannot license the EPI kanena in its
complement clause. The claim in (12) seems to be further undermined.
(13)
*prokite
na
sinadiso
is about
subj
meet-1 sg
'*1 will meet anyone at five'
kanena
stis pede
anyone-acc at five
I want to point out however that this second argument against Giannakidou's (12) is
not a strong argument. It is not advanced against claim (12) as such but rather
against Giannakidou's characterization of EPI licensers, which according to her
include predicates that take na -complements. If it turns out that these predicates
are not EPI licensers after all, the argument they provide against (12) will cease to
exist. The problem with (13) is that, contrary to what Giannakidou claims,
predicates that license na-clauses, pr6kite in (13) is one of them, are not EPI
licensers. (13) is a counterexample against Giannakidou's claim that EPI triggers
include predicates that license na -clauses. Also consider main na-clauses,lo which
depend on no predicate and yet constitute EPI licensing contexts. Giannakidou
must additionally assume with respect to predicates that take na-complements that
they are necessarily superordinate licensers because they never seem to license
EPI's in their own clause, as example (14) shows. I wish to argue that those nclauses where EPI's are licensed involve one of the propositional operators and it is
this propositional operator, and not the predicate selecting the na-clause or the na
-clause itself, that licenses the EPI inside the na-clause. But if the predicates that
take na-clauses are not, as I have argued, EPI licensers, the second argument
against (12) is not a valid argument because it depends on the wrong, in my view,
assumption that predicates that take na-clauses are EPI triggers.
(14)
*thelo
kanena
want-1 sg
anyone-ace
'*1 want anyone'
The same inaccurate, in my view, inclusion of predicates that take na -clauses

among EPI contexts, gives rise to another objection against (12). Predicates that
take na -complements, e.g. volitive or directive predicates, are not operators,
unless we equate operator with predicate. (12), if true, would have to be replaced
by the generalization that non-veridical contexts and not operators license EPI's in
Gr. To conclude the discussion, the first argument I presented against
Giannakidou's (12), i.e. the fact that F, the Habitual and the Gnomic Operators
license EPI's, is a strong argument against (12). It should be obvious how the same
argument (cf. additionally the data in (6) and (7 also blocks the claim that nonveridical operators license PI's crosslinguistically, put forward by Giannakidou
(idem), from being an alternative to claim (3a).
29
The preceding discussion on whether na-clauses are EPI licensers was

triggered by Giannakidou's paper but has been useful with respect to this paper as
well. Predicates that take na -clauses were argued to be neither EPI licensers, nor
non-veridical, as a general characteristic of their class, nor operators. What is
important for us is that they are not EPI licensers. Notice that if predicates that take
na-clauses were indeed EPI licensers, this would be a major problem for claim (8),
which I have advanced but also for (3a).
4. WEAK ISLANDS
Weak islands standardly include Wh-Islands, Neg Islands, Factive Islands, Extraposition Islands and Pseudo-Opacity Islands. 1112 Weak islands block extraction of
adjuncts but not of arguments. I am only interested in Wh-Islands (cf. (15 and Neg
Islands (cf. (16 among weak islands. I will adduce three more types of weak islands, i.e. Focus Islands, Necessity Islands and Possibility Islands, already presented in Agouraki (1993). In this way I argue for the second syntactic property of
propositional operators repeated in (2c).
(15)
(16)
(2)
a. *Howi do you wonder whether he repaired the car ti?

b. ?Whati do you wonder whether he repaired ti?
a. Howi do(*n't) you think he repaired the car ti?
b. Whati do(n't) you think he repaired ti?
c. NEG, Q, F, CO l3 , NO and PO induce weak island effects.
Rizzi (1990) gave the argument / adjunct asymmetry induced by Neg Islands a
unified treatment with the argument / adjunct asymmetry induced by Wh Islands
under ECP. This is presented next. Rizzi (1990, 1991) argues for re1ativized
minimality and proposes that the following minimality condition holds on
antecedent government: a certain type of government relation is blocked by the
intervention of a potential governor of the same type. In the case of a Wh-chain,
potential interveners are A'-specifiers, i.e. Specs of CPo Neg Islands can reduce to
the same explanation as Wh Islands, if we assume that negative clauses involve a
NegP. According to Rizzi (1991), 'The Spec ofNegP may be filled by a sentential
negation operator, phonetically realized as French pas, presumably English not,
or phonetically null, as in Italian, Spanish etc .. Alternatively, it may be filled (at the
latest at LF) by the movement of a negative quantifier. The negative head may be
phonetically realized as French ne, Italian non, or null, as in English, German, etc.
Such a uniform substructure will have an effect analogous to the effect of Wh CP's:
the intervention of the A' specifier will block an antecedent government relation in
A' chains. Neg Islands and Wh Islands thus receive a uniform treatment under
Relativized Minimality.
Rizzi (1990:19) also observes: 'It would then seem that inner-island effects are
determined by "affective" operators, in Klima's (1964) sense - that is, operators licensing negative polarity items.' Rizzi's observation is more limited than claims (2)
YORYIA AGOURAKI
30
and (3). According to (2) there is a set of elements with five properties. Two of
these properties are PI licensing and weak island effects. Neither of these two properties is the defming property of the set. The term 'affective' operator Rizzi, among
others, uses is not a primitive notion, as the term propositional operator is. The term
affective operator is not defmed independently of licensing PI's.
Separately from his discussion of weak island effects Rizzi (1990: 116, ft. 16)
remarks that modals in some cases appear to block embedded construals of preposed adjuncts. Rizzi does not state a connection between this observation and the
observation that Q and NEG induce weak island effects. In my proposal the two
observations are related. What those cases are is precisely weak island effects
induced by modals. The sources Rizzi quotes for claiming that modals may block
embedded construals of preposed adjuncts are Travis (1984) and Pesetsky (p.c.).
Travis (1984:171-2) gives an example of a weak island induced by Negation (cf.
(17)) and contrasts it with an example with no Negation from Lasnik and Saito
(1984) (cf. (18)). Travis goes on to observe that the contrast between (17) and (18)
'is not a core case since, as soon as the matrix verb is given more content, the
construction worsens'. She gives examples (19) and (20) to illustrate her point.
Crucially, examples (19) and (20) involve the Necessity Operator and the
Possibility Operator, respectively. I want to argue that what examples (19) and (20)
show is not that the contrast between examples (17) and (18) is not a core case, as
Travis argues, but rather that modal operators give rise to weak island effects.
(17)
*[How slowlyJi won't you say Mary gave the book to Joan ti?
(18)
[How slowlyJi did Sean say that Matthew gave Jared the book ti?
(19)
*[How slowlyJi should Sean have said that Matthew gave Jared the
(20)
book ti?
*[How slowlyJi might Sean believe that Matthew gave Jared
the book ti?
In examples (21) to (25) I illustrate claim (2c). Weak island effects are shown
to be triggered in many more environments than NEG and Q. In particular, F, NO
and PO also give rise to weak island effects. The set of weak island triggers14 is no
different than the propositional operators set. Concerning Wh-Islands, let me point
out a difference in emphasis between the standard account and property (2c). In the
standard account it is the filled [Spec,CP] that is responsible for the Wh-Island. The
wh-phrase in the Spec of the lower CP is both the cause of and the account for the
island. Rizzi brings Neg Islands within the same reasoning. The negative operator
in [Spec,NegP] is both the cause of and the account for the Neg Island. In (2c) it is
rather assumed that it is Q of the lower CP and NEG that are responsible for the
Wh-Island and the Neg Island respectively. An account for weak islands is not provided.
(21)
(22)
(23)
(24)
(25)
31
a.
tii
dhen lei
oti eftjiakse
ti?
NEG
what-acc
not
say-3sg that fixed-3sg
'Whati doesn't he say that he fixed ti?'
b. *jatii dhen lei oti eftiakse
to aftokinito ti ?
why
not say-3sg that fixed-3sg the car-acc
'*WhYi doesn't he say that he fixed the car ti?'
a. tii
anarotjese an
eftjiakse
ti ?
Q
what-acc
wonder-2sg whether
fixed-3sg
'Whati do you wonder whether he fixed ti?'
b. *posi anarotjese an
eftjiakse
to aftokinito ti ?
how
wonder-2sg whether fixed-3sg the car-Acc
'*Howi do you wonder whether he fixed the car ti?'
a. tii
LEI
oti
eftjiakse
ti?
F
what-acc
SAYS-3sg that
fixed-3sg
'Whati DOES he say that he fixed ti?'
b. *jatii LEI
oti eftjiakse
to aftokinito ti ?
why
SAY-3sg that fixed-3sg the car-acc
'*WhYi DOES he say that he fixed the car ti ?'
a. tii
prepi
na lei
oti eftjiakse
ti? NO
what-acc
must-3sg
subjsay-3sg that fixed-3sg
'Whati must he say that he fixed ti?'
b. *jatii prepl
na lei
oti eftjiakse
to aftokinito ti ?
why
must-3sg
subj say-3sg that fixed-3sg the car-acc
'*WhYi must he say that he fixed the car ti?'
a. tii
bori
na lei
01 eftjiakseti?
PO
what-acc
can-3sg subjsay-3sg that fixed-3sg
'Whati can he say that he fixed ti?'
b. *jati
bori
na
lei
oti eftjiakse
to aftokinito ti ?
why
can-3sg subj
say-3sg that fixed-3sg the car-ace
'*WhYi can he say that he fixed the car ti?'
It is not in the aims of this paper to offer an account for Focus Islands, Necessity Islands and Possibility Islands by extending the standard treatment of the argument / adjunct asymmetry induced by Wh-Islands to the three new types of weak
islands, as Rizzi did for Negative Islands. It should be obvious to the reader how
Relativized Minimality could also account for the new weak islands. I would simply like to comment briefly on the remark in Section 2 that differences in the syntactic representation of propositional operators sometimes produce theoretical distinctions, perhaps otherwise unmotivated. I believe that the distinction between
weak and inner islands is a case in point. In standard syntactic terminology inner
islands are weak islands that happen to be structurally inside IP (below the complementizer). In line with this terminological distinction NEG and F induce inner
32
YORYIA AGOURAKI
island effects. The same applies to NO and PO, as long as these operators are part
of the INFL associated with the verb and not separate verbs. In the case where NO
and PO are modal verbs heading their own clause, NO and PO will induce weak,
but not inner, island effects. As for Wh Islands, they are also taken to induce weak,
but not inner, island effects. I would like to point out that if propositional operators
turned out to have a common syntactic representation, as seems to be argued for by
the list of syntactic properties attributed to them (cf. (3)), the distinction between
weak islands and inner could no longer hold. All propositional operators would be
taken to induce either weak island effects or inner island effects. It seems to me that
what would be lost would be the inner island type. But even without questioning
the standard syntax of propositional operators I wonder what sense it makes to talk
about inner islands in a V2 language.
5. ROOT INFINITIVES
I use the term root infmitives IS as a cover term for root clauses with dependent time
reference. In particular, root na -clauses in Gr and actual root infmitives in other
languages. na is a particle that has been analysed as a Subjunctive marker
(Veloudis and Philippaki-Warburton 1983), an infmitive marker (Efthimiou 1989)
or a complementizer (Agouraki 1991). I will argue for property (2d) of propositional operators. The discussion does not rely on a specific analysis of na -clauses
and is compatible with all three of them.
(2) d. NEG, Q, F, CO, NO and PO license root infmitives.
Let me clarify three points concerning claim (2d). First, I am not claiming that
the tense dependence 'value' of root infmitives, or whatever manifests this in the
clause, e.g. -Tns, na, is an operator. Second, propositional operators in root
infmitives can be either overt or not. And thirdly, claim (2d) does not underestimate
or substitute the role played by Pragmatics and / or Intonation. In the case
propositional operators are not overt, it is the context and / or the intonation that
make it clear which propositional operator obtains. Let me stress that what claim
(2d) asserts is that root infmitives necessarily have one of the propositional
operators readings. Though the context and / or intonation will determine which
propositional operator reading it will be in the absence of an overt operator, the
context and / or intonation cannot be taken to be responsible for the obligatoriness
of a propositional operator reading for all root infmitives. In particular, a pragmatic
analysis alone of root infmitivals readings will never be able to account for the
unavailability of declarative readings in root infmitives. Declarative readings
should be the easiest to get and certainly should be compatible with some contexts.
But they never obtain. And this is a property a pragmatic analysis cannot account
for but also a property that needs to be explained. (2d) is part of an explanation.
For a first form of claim (2d) see Agouraki (1991), who argues root naclauses have a Q, NEG, CO, NO or PO reading, but never a declarative reading.
33
Also that na-clauses, though tensed, do not have independent time reference.
Complement na-clauses acquire time reference on the basis of the time reference
of their selecting verb. Root na-clauses acquire time reference with respect to
speech time, as ordinary tensed clauses, or context information. It is pointed out
that a Q , NEG, CO, NO or PO reading arises obligatorily in root na-clauses alone.
These readings are attributed to the lack of a higher verb that could provide a time
frame for the verb in the na -clause and are considered to be a property of main
clauses which lack independent time reference. Crosslinguistic evidence is
provided. In particular, root infmitives in English are shown to license Q and NO.
Rizzi (1993) also argues that crosslinguistically root infmitives are compatible with
a Question, a Counterfactual or a Jussive reading but never with a declarative
reading. 16 Rizzi's counterfactual and jussive readings are respectively the NEG and
the NO reading of this paper. He provides examples from Italian. Rizzi offers an
account for the distribution of readings in root infmitivals. According to Rizzi
(1993: 17), 'A natural idea is that clauses have a tense variable that must be flXed
somehow. In fmite clauses the overt fmite morphology fIxes the value of the tense
variable; in embedded infmitives the tense variable is bound by the main clause
tense value; main clause infmitives are then excluded because their tense variable
would remain unbound, ultimately a violation of Full Interpretation at LF. What
about main clause infinitives with the question, counterjactual or jussive reading?
It is arguable that they differ from declaratives in that they involve some kind of
operator, as is overtly the case for questions and not implausible for the other
cases... One possible line of thought is that the operators involved may
unselectively bind the tense variable~ We may think of this option as a marked
option that adult languages mayor not adopt (e.g. English does not seem to
naturally allow root infinitival questions.).' I will disagree with Rizzi's view that the
'special' readings of root infmitives are a marked option, at least on the basis of
English (cf. (26).17 (27) illustrates (2d) for Gr. 18 Notice the ungrammaticality of
(27g) in the declarative reading, which is due to unavailability of a propositional
operator in the sentence.
(26)
a.
b.
c.
d.
(27)
a.
b.
Give up after all these efforts?

What say?
You must be mad, to go out in the rain in your bikini
' ... if you went out .. .'
To be or not to be?
'Should one be .. .'
na
zitiso egho rusfeti
?
subj
ask-lsg
for political favour-acc
'I would never ask for a political favour'
na
ftjiakso
kafe
?
make-lsg COffee-acc
subj
'Shall I make coffee?'
NEG
Q
CO
NO
NEG I9
YORYIA AGOURAKl
34
c.
ke epipleon
na
prokite
ke pali
Jla JI0 F20
and moreover subj
be-3sg about once again a son
'Let alone that it is another SON?lt
d. na
iha
lefta
tha aghoraza dheka ikopedha CO
subj
had-1sg money-acc would buy-1sg ten building plots
'Had I the money, I would buy ten building plots'
e. na erhjese
pjo sihna
NO
subj come-2sg more often
'You should come more often'
f. na
su
po egho pu
ta vrike
P022
subj
you-cl tell I
where them-cl found-3sg
'I can tell you where he found them'
g. *na
ftjakso
kafe
*DECLARATIVE
subj
make-1sg coffee-acc
'I will make coffee'
As already pointed out, claim (2d) is not explanatory. It gives us a property of
propositional operators, namely that of licensing root infmitives, without accounting for it. This is what I will try to do next. Let me first point out that Agouraki
(1991) and Rizzi (1993) put forward accounts for the obligatoriness of a
propositional operator reading in root infmitives (cf. the underlined parts in the
presentation of these theories). Basically both accounts do the same thing, which is
to relate the lack of independent time reference with the triggering of propositional
operator readings. I want to argue against such a line. As already claimed, root naclauses acquire time reference with respect to speech time or context information.
The same is true of root infmitives in general. An example of time reference
established with respect to speech time is 27(d), where the Past Tense of the verb in
the na -clause establishes time reference. To sum up, speech time and / or context
information provide time reference for root infinitives. If this is so, propositional
opeator readings in root infmitives cannot / should not be linked to time reference.
Time reference is acquired through other means / anyway. This is the reason why
the accounts in Agouraki (1991) and Rizzi (1993) for the triggering of
propositional operator readings seem not convincing. The triggering of
propositional operator readings and time reference of root infmitives should be kept
separate.
I will propose that in root infmitives we do not simply have triggering of
propositional operator readings. From the point of view of claim (2d) now, propositional operators do not simply license root infmitives. They actually select them.
You see the new proposal which replaces (2d) in (28) below. Can we talk about
selection? I believe we can, since propositional operators are (at least semantic)
higher predicates. 23 Assume it is so. But isn't 'root' in contradiction with 'select' in
(28)? How can you have selection of root infmitives? The contradiction is resolved
in (28'). I will argue that in this contradiction (28) lies the account of (2d), i.e. the
answer to why propositional operator readings obtain in root infmitives. My claim
is that there are no root infinitives (cf. 29a)). They exist only in form but not in
35
essence. Languages seem to have a ban against true root inflnitives, which would
be root inflnitives with a declarative reading (cf. (30. Prima facie root infInitives
are in essence embedded infInitives selected by propositional operators (cf. 29(b.
The propositional operator readings of root infInitives is nothing more than a consequence of the natural language ban on root infmitives. Claims (28) and (29) extend
to root infmitivals (cf. (31, examined in the fIrst footnote of this section. Root
infmitivals were shown to be licensed by propositional operators. In (32) below I
repeat the root infmitival example of the relevant footnote.
(28)
(28')
(29)
(30)
(31)
(32)
Propositional Operators select root infInitives

Propositional Operators select infmitives24
a. Root infinitives are not root~
b. Prima facie root infinitives are in essence embedded infinitives
selected by propositional operators~
Languages have a ban against root infmitives
Propositional Operators select infmitival relatives
thelo
enan andra
pu na
majirevi
PO
want-lsg a
man-acc thatsubj
cook-3sg
'I want a man who can cook'
The next task is to explain the natural language ban on root infmitives. The
term 'root infmitives' is, I believe, a fIrst descriptive approximation of a more basic
concept, something like the structural realization of an s-selected entity. Let me
decompose the term 'root infmitive' into its constituent parts, i.e. 'root' and 'infmitive', to see if we get anywhere. It cannot be that the key notion is 'infmitive' or Tense because na-clauses are +Tense. Nor is the key notion 'Tense Dependence'
because tensedependent root infmitives have been shown to be able to acquire time
reference. It is not the case either that the key notion is that of 'root' because of
claims (28) and (29).
I will next present another argument against 'root' being the key notion of the
semantically vague concept of 'root infmitives'. I would like to draw the reader's
attention to embedded wh-infmitivals. There is no doubt that embedded wh-infmitivals are embedded clauses. Interestingly, propositional operators obtain in embedded wh-infmitivals (cf. (33) and example (34. In example (34) NO or PO obligatorily obtain in the embedded wh-infmitival. (33) argues not only against 'root being the key notion in the 'root infmitives' concept but also in favour of claims (29a)
and (b). How does (33) fIt in with (28)? I will claim that in the case of embedded
wh-infmitivals the higher predicate selects a wh-clause and not a wh-infmitival. The
ban against unselected infmitives triggers the propositional reading of the infmitival
included in the wh-infmitival. In (28) 'root' means nothing different than 'unselected'. Propositional operators select otherwise unselected infmitives (cf. (35.
Claim (35) applies to root infmitives, infmitival relatives and embedded wh-infmitivals.
(33)
Propositional operators obtain in embedded wh-infmitivals
36
YORYIA AGOURAKI
(34)
dhen ksero
ti
know-lsg what
not
'I do not know what to do'
na
subj
(35)
Propositional operators select (otherwise unselected) infinitiv.es
kano
do-lsg
NO/PO
To come back to the question raised with respect to the ban on root inftnitives,
I have argued that 'root infmitives' in (30) is a non-felicitous approximation. The
category the term seeks to characterize is neither root nor infmitive. It seems to be
'unselected infmitive' . The 'infmitive' part continues being an unsatisfactory
descriptive term for the same reasons it was an unsatisfactory description when it
formed part of the term 'root infmitive'. The question raised earlier with respect to
root infmitives should rather be raised about unselected inftnitives. Namely, why is
there a ban on unselected inftnitives (cf. (36? I want to propose that the answer to
this question is a topic of inquiry for Theta Theory. Some predicates subcategorize
for infmitives. Infmitives involve c-selection. I believe that we will have an account
for (36) if we fmd which s-selection requirement of predicates the c-selection of
infmitives satisftes. Infmitives seem to me to be a type of propositional argument
tied with a particular theta-role. The account for (36) presupposes answering the
question why some predicates select infmitives only, other predicates select tensed
complements and yet other predicates select both infmitival and tensed
complements. I will not continue the discussion about the nature of propositions
realized as infmitives. Claims (37a) and (b) provide the selectional properties of
propositional operators. According to claim (37a) propositional operators are oneplace predicates. The approach nicely incorporates the analysis of these elements in
Logic as (higher) predicates. Properties (2a) and (d) can be said to be unifted under
(37a). (37b), on the other hand, describes the occurrence of propositional operators
in +/- infmitival clauses.
(36)
(37)
Languages have a ban against unselected infmitives

a. Propositional Operators s-select propositions
b. Propositional Operators c-select +/- infmitival clauses
I will next return to the discussion of property (28'). There is an important

difference between claims (2d) and (28'). While (2d) simply attributes a syntactic
property to a semantic set, (28') not only attributes a syntactic property to a semantic set but additionally argues that we also have to do with a syntactic set and puts
forward a particular syntactic analysis as the characteristic property of the syntactic
set. In particular it claims that the members of the set are syntactic predicates. (2d)
is compatible with the standard syntactic analyses of the propositional operators
involved. (28') is not. (28') is the ftrst attempt in the paper to translate a logical /
semantic claim into a syntactic claim. Propositional operators were shown to have a
c-selectional property, namely that of c-selecting infmitives. Among the syntactic
properties of propositional operators, this is the strongest evidence in favour of a
syntactic analysis of propositional operators as predicates.
37
6. EMBEDDED INTERROGATIVES
The fmal syntactic property of propositional operators that will be discussed is that
oflicensing embedded interrogatives (cf. (2e. Before looking at the data it will be
useful to remember what the theory says about embedded interrogatives.
(2) e.
NEG, Q, F, NO and PO license embedded interrogatives 25
It is generally assumed that it is specified in the lexical entry of some verbs that
they s-select interrogative propositions and c-select +WH CP's. Alternatively, following Pesetsky (1982), these verbs are only specified for s-selecting interrogative
propositions. The Canonical Structural Realization (CSR) of an interrogative
proposition is a +WH CP. The class of verbs s-selecting interrogative propositions
is very small. Notably, it includes the verbs wonder, ask and a few others. Interestingly and in suprising contrast with the small class of verbs that s-select an interrogative proposition, the class of verbs that can take a +WH CP is fairly big. This
discrepancy is in need of an explanation. I will argue that the interrogative marking
on a proposition is not part of s-selection. What (2e) claims instead is that verbs can
take a +WH CP as long as they are associated with a propositional operator. Example (38) illustrates the above claim. Contrary to examples (38a-e), example (38f)
does not have interrogative force 26 and is not here considered as introducing a +WH
CPo
(38)
a.
b.
c.
d.
e.
f.
ipan
erthi
NEG
dhen
pjos
tha
come-3sg
not
said-3pl
who-nom will
'They did not say who is corning'
ipan
pjos
tha erthi?
Q
who-nom will come-3sg
said-3pl
'Did they say who is coming?'
F
IPAN
pjos
tha erthi
SAID-3pl who-nom will come-3sg
'They DID say who is coming'
prepi
na
ipan
pjos
tha erthi
NO
must-3sg
subj
said-3pl
'They must have said who is coming'
na
ipan
pjos
tha erthi
PO
bori
rnay-3sg
subj
said-3pl
'They may have said who is coming'
ipan
pjos
tha erthi
*Interrogative Reading
said-3pl
'They said who is coming'
A verb like leo 'say', the past tense of which you see in (38) s-selects a proposition. The CSR of a proposition is a -WH CPo Examples (38a-e) should therefore
38
YORYIA AGOURAKl
be ungrammatical, but they are not. This is a problem for standard theory. It is obvious that a story based on CSR's of theta-roles will not work. (2e) accounts for
their grammaticality. But we also need an account for (2e). This is what I attempt to
do next.
In the previous section, I argued that propositional operators may c-select infmitives. In the same spirit, it could be argued that what we have in (2e) is another
c-selectional property of propositional operators. There seems to be an important
difference between (2d) and (2e) however. And that has to do with the link between
s-selection and c-selection. It is standardly assumed that an argument is s-selected
and c-selected by the same head. With respect to (2d), I have argued that propositional operators s-select and c-select the infmitival clause. On the basis of the examples in (38), we would perhaps have to challenge the assumption that s-selection
and c-selection are necessarily linked in this way. We would probably need to argue that it is possible for an argument B to be s-selected by a head A but be c-selected by a head C. The embedded clause in examples (38a-e) would be s-selected
by the main verb but it would be c-selected by the propositional operator. One
could draw a parallel between theta vs. formal (:WH) properties of CP's and theta
vs. formal (:Case) properties of DP's. In both cases the two properties may have a
common or distinct assigners. For Pesetsky (1982) the syntactic mechanism of Case
assignment interacts with CSR. In the present case we would have a syntactic
mechanism, i.e. licensing by propositional operators, interacting with c-se1ection /
CSR of propositions. Another problem for this approach, provided we follow the
analysis of propositional operators as syntactic predicates, is that of locality. In each
of the examples in (38a-e) the propositional operator s-selects the ipan predicate,
which in turn s-selects the embedded interrogative. The propositional operator
predicate in each case would c-select both the ipan predicate and the embedded
interrogative. Moreover, c-selection of the embedded interrogative would take
place across the intervening ipan predicate. The picture is too complicated to be
true. It seems to me to be defendable only at a price.
I will instead put forward the following account for (2e). The examples in (38ae) have shown that the +WH marking on a clause is not associated with s-selection.
If it is not, there is no compelling reason for assuming that the +WH marking in
embedded interrogatives is indeed marking on the embedded clause. I am
suggesting that +WH marking in embedded interrogatives is perhaps exclusive
marking on the wh-phrase itself and not on both the clause and the wh-phrase introducing it. Suppose that wh-phrases are a type of polarity items, that are licensed by
propositional operators. This situation, if true, would be described by claim (2b)
repeated below. A piece of evidence in support of this hypothesis is the fact that in
some languages, e.g. Chinese, wh-phrases are the same lexical items as polarity
items. I started this section by claiming that that the interrogative marking on a
proposition is not part of s-se1ection. If the account for (2e) that has been sketched
in this paragraph turns out to be the correct one, this will mean that the +WH marking on an embedded clause is not c-selection either, in fact that there is no +WH
marking on an embedded clause. The interrogative force of the embedded clauses
39
in (38a-e) with then be a consequence of the interrogative force on the wh-phrases

introducing these clauses. All licensing of wh-phrases would be superordinate.
(2) b.
NEG, Q, F, CO, NO and PO license PI's.
What about the small class of verbs, e.g. wonder, ask, that seem to s-select
interrogative propositions? Will we have to say that they have no place in the presented system if only as a small exception? I will next claim otherwise. And I
would like to point out a parallelism between apparently s-/c-selected +WH CP's
and PI licensing by inherently negative verbs (cf. also the inherent possibility verbs
of Gr presented in Section 3). I propose that c-selection of +WH CP's by verbs like
wonder and ask should be analysed in similar terms. Namely, wonder and ask
could be considered inherently interrogative verbs functioning like the Q operator
in licensing an embedded interrogative (or in line with the proposed account for
(2e), the wh-phrase introducing the embedded interrogative). Given that the class of
verbs that can take +WH complements without this being specified in the Lexicon
is crucially very large, while the class of verbs that must take a +WH complement
is fairly small, I propose that the +WH marking need not be specified as a lexical
property for the latter class either. The proposal about inherently interrogative verbs
would ensure grammaticality for those verbs.
7. CONCLUSIONS
I have taken a class of logical I semantic operators, which share the property of
being semantic higher predicates, and have shown that they also share a number of
syntactic properties, which are repeated in (39b-e) below. On the basis of property
(39d) and in line with property (39a) it was proposed that NEG, Q, F, CO, NO and
PO are syntactic one-place predicates. An account for property 39(e) was offered,
according to which wh-phrases introducing embedded interrogatives are a PI type
licensed by the {NEG, Q, F, CO, NO and PO} set. If we incorporate the two proposals in the list of properties in (39), we obtain a new characterization (cf. (40
for the syntactic properties of propositional operators, a term justified by property
(39a). The lists of properties in (39) and (40) illustrate how the class of propositional operators 'unifies' a number of allegedly unrelated syntactic phenomena.
(39)
(40)
NEG, Q, F, CO, NO and PO:

a. are semantic one-place predicates
b. license PI's
c. induce weak island effects
d. license root infmitives
e. license embedded interrogatives
NEG, Q, F, CO, NO and PO:
a. are one place predicates
b. license PI's
40
YORYIA AGOURAKI
c. induce weak island effects

Let me briefly recapitulate the three arguments for developing a syntactic
analysis of propositional operators as syntactic one-place predicates. The ftrst was
developed in Section 5 and was based on property (39d). Propositional operators
were argued to select root inftnitives. The second argument is the fact that propositional operators have a number of syntactic properties (cf. (39b-e)), which demand
a syntactic analysis for propositional properties. The absence of crosslinguistic
differences with respect to these syntactic properties adds to the force of the
argument. The third argument is the generally asssumed correspondence between
semantic structure and syntactic structure. Semantic property (39a) of propositional
operators seems to demand a compatible syntactic analysis for propositional operators. A syntactic one-place predicate analysis seems to be the obvious choice.
NOTES
I wish to thank M. Brody for discussing the paper with me and Y. Caranassios for his intuitions on
some of the examples. Thanks are also due to A. Dimitriades, M. Hegarty, S. latridou and D.
Theophanopoulou-Kontou for comments during the presentation of the paper at FAS and to A.
Alexiadou for excellently organizing the FAS Workshop on Modem Greek Generative Syntax. I would
like to express my indebtedness to the two reviewers whose comments contributed considerably to
presenting the logic of the arguments in a better way.
2
It should be specified that all Gr data come from the Standard Greek (SGr) dialect.
Lewis (1975) treats adverbs of quantification as unselective quantifiers and contrasts them with the
familiar 'selective' quantifiers. Unselective quantifiers can bind an unlimited number of variables. The
distinction I propose between propositional and nonpropositional operators is more general than Lewis'
distinction. The proposed distinction covers all operators and not just quantifiers and adverbs of
quantification, as is the case with Lewis' distinction.
4
For each EPI licenser the notation *(Licenser) is used in order to show that the element in
parentheses is what licenses the EPI. If the particular licenser in each of these examples is deleted and
no other licenser is introduced, the sentence becomes ungrammatical.
5
Superordinate licensing is possible. For instance, sentential negation can be a superordinate EPI
licenser (cf. ia). So can inherently negative verbs like arnume 'deny' (cf. ib).
(i) a. dhen
ipa
oti idhe
kanena
not
said-Isg that saw-3sg anyone-acc
'I did not say he saw anyone'
b. amithika
oti idhe
kanena
denied-lsg
that saw-3sg anyone-acc
'I denied he saw anyone'
Contrast (9c) with ungrammatical (9c'). It is only focused propositions, more accurately the Focus
Operator, that license EPI's; focused XP's do not. The stress on the verb here marks wide IP focus.
ipe
kamia kuveda
(9) c'. *0 YORGHOS
the YORGHOS-nom
said-3sg any word-acc
An anonymous reviewer asks whether future marker tha is an EPI licenser. The answer is no.
What the reviewer has in mind is examples like (I) below. Particle tha is not always/only a future
marker. It may be a Necessity Operator or a Possibility Operator marker as in example (i). In (i) it is not
the future marker tha, but the NO / PO marker tha that licenses the EPI.
41
afti
tha
ehi kanena provlima
she-nom must / may
has any problem-acc
'She must / may have a problem.'
Gr has inherent possibility verbs, e.g. leo na 'think of, may', that can be superordinate (cf. ia), and
crucially not clausemate (cf. ungrammatical ib), EPI licensers. Inherent possibility verbs also include
skeptome na 'think of, may', pistevo na 'maybe, may'.
(i) a. lei
na pai puthena
apopse
is thinking of subj goes anywhere
tonight
'He is thinking of going somewhere tonight.'
na pai sinema
apopse
b. *lei
kanis
is thinking of anyone-nom subj goes to the movies tonight
'Someone is thinking of going to the movies tonight.'
9
By subjunctive clauses Giannakidou refers to clauses introduced with the particle na.
(i)
10
On main na -clauses see section 5.
Pseudo-opacity islands are outside the scope of the paper. I would simply like to point out a
difference between pseudo-opacity islands and the other types of weak islands. Pseudo-opacity islands
do not show the argument / adjunct asymmetry that the rest of weak islands manifest and which
crucially is taken to be the defining property of weak islands.
12
A reviewer remarks that pseudo-opacity is a reason against assimilating EPI licensers and weak
island triggers. The reviewer points out that as discussed by Obenauer and then Rizzi, some sorts of
adverbs create weak islands, typical among them are French souvent 'often' and beaucoup 'much'.
Now such adverbs as souvent, which do trigger islands, do not license EPl's ( John often meets Mary is
not an EPI licensing context.) Therefore, the property of triggering weak islands and the property of
licensing EPl's seem distinct (and in fact Rizzi keeps them separate).
Notably, the observation that Rizzi keeps the property of triggering weak islands separate from the
property of licensing EPl's is not true. In fact Rizzi does not keep the two properties separate, with
respect to the one propositional operator he examines, namely NEG. According to Rizzi (1990:19), 'It
would then seem that inner-island effects are detennined by affective operators, in Klima's (1964) sense
- that is, operators licensing negative polarity items.' It is the spirit of Rizzi's comment that I want to
draw the reader's attention to. Strictly speaking, the generalization he draws concerns only NEG and
does not apply to the other two affective operators, i.e. Q and CO. Q does not give rise to inner islands,
a subtype of weak islands. Also, as explained in the next footnote, CO cannot in principle give rise to
any island effect for the simple reason it is incompatible with embedded clauses.
With respect to the specific points the reviewer raises, I believe that the beaucoup case and the
souvent case should be kept separate. Concerning the beaucoup case first, let us briefly remember the
data. In French a VP-initial adverbial QP selectively blocks extraction of certain VP-intemal elements.
In particular extraction ofa direct object is possible (cf. ia) while extraction of the specifier of the direct
object is not (cf. ib). The asymmetry is between argument vs. specifier of argument and not between
argument vs. adjunct. I am only interested in islands that present the argument / adjunct asymmetry.
French beaucoup does not present this asymmetry. Souvent, to the extent it patterns like beaucoup in
(i), is also outside the scope of the discussion in the main text.
(i) a.
[Combien de Iivresli a-toil
beaucoup consultes ti?
how many books
has-he
much
consulted
a-toil beaucoup consulte [ti de Iivres1?
b. Combieni
how many
has-he
much
consulted books
To the extent souvent presents the argument / adjunct asymmetry, and it does, as I will argue in
this section, it does not belong to the same category as beaucoup in (i) and will be discussed in the
main text. Souvent is an adverb that appears with the Habitual Operator. MG sihna 'often', more
accurately the Necessity or Possibility Operator it is compatible with, is an EPI licensing context (cf. ii).
Also, souvent in French licenses PIon.
(ii) pijeni
sihna
kamia ekdhromi
goes
often
any excursion
'He often goes on an excursion'
11
42
YORYIA AGOURAKI
In the spirit of the reviewer's remark about Pseudo-Opacity Islands, if I understand it well, one
could argue that the existence of Factive Islands and Extraposition Islands argues against assimilating
EPI (or more generally PI) licensers and weak island triggers. Not necessarily. It could be the case, as I
argue it is, that PI licensers assimilate with one subcategory of weak island triggers, that gives WhIslands, Negative Islands, Focus Islands, Necessity Islands and Possibility Islands. It is at a later stage
that we need to address the question what this subcategory of weak islands has in common with Factive
islands and Extraposition islands as well as why the last two categories of weak islands do not assimilate
with PI triggers. At present I have nothing to say about this.
13
Checking the Conditional Operator is nonapplicable in the present case.
With the exception of Pseudo-Opacity Islands, Factive Islands and Extraposition Islands, discussed
in earlier footnotes. I am only interested in one subcategory of weak islands.
15
I take root infinitives to include infinitivltl relatives. An anonymous reviewer observes that most
discussions of root infinitives would be incompatible with taking root infinitives to include infinitival
relatives because they capitalize on the non c-commanded property of the root. I assume that infinitival
relatives adjoin to DP because they modify the whole DP. In that position they are not c-commanded by
the lexical head N. In any case, I take the definitional property of root infinitives to be not the non ccommanded property but absence of selection by a head and this requirement is met by infinitival
relatives. Moreover, infinitival relatives also have property (2d), like the rest of root infinitives (cf.
example (i) below and compare with (26f) in the text).
(i) thelo
enan andra
pu na majirevi
PO
want-I
a man-ace that na cooks
'I want a man who can cook.'
16
Rizzi in that paper mainly argues that contrary to the situation in adult language root infinitives in
child language are incompatible with Q, present limited compatibility with NEG and standardly have a
declarative reading. Child language is outside the scope of the present study.
17
English infinitives can be bare.
14
18
With respect to the earlier discussion of EPI licensers in Gr note that EPI's are licensed in root naclauses.
19
A reviewer observes with respect to (27a) that 'the negative reading of (27a) is simply the effect of
the modality of na , as in English Would I ask for a political favour? , the pragmatic inference being
'Of course I wouldn't.'. As I have already explained, I am not underestimating the contribution of
pragmatics and intonation in the interpretation of root na -clauses in general. Likewise in the case of
(27a). In fact, the pragmatics and the intonation can be manipulated to derive all propositional operators
readings for this example, each at a time. My point is that no matter how much native speakers
manipulate the pragmatics and the intonation in (27a) they will not be able to derive a declarative
reading.
Attributing the reading of (27a), or the readings of root na -clauses in general, to 'the modality of
na ' is an alternative account that needs however to be defended. It is not self-evident. Moreover, it
relies on a particular analysis of the particle na (cf. Veloudis and Phillipaki-Warburton 1983) and is for
this reason less easy to extend to root infinitives in other languages. I will stress that the account in (2d)
does not rely on or presuppose any of the analyses for na . Nor do I claim that the readings of main na
-clauses have anything to do with the status of particle na. One question that will need to be answered
with respect to the 'modality analysis' is why the 'modality' of na is different, in fact absent, in
embedded clauses.
The reviewer also raises the question why it is that (27a) 'contains' the NEG Operator while (27b)
contains the Q Operator. I have already answered this in the discussion of the reviewer's first comment.
It is in fact possible that (27a) and (b) contain the Q Operator and the NEG Operator respectively, like
(27f) is compatible with either the Possibility Operator or the Necessity Operator. Which propositional
operator obtains in a particular root infinitival clause has to do with the context as it also has to do with
intonation. What is not a context requirement, however is that every root infinitival clause has a
propositional operator reading. It is this property that is described in (2d).
20
A reviewer observes with respect to example (27c) that 'particle ke may be a focus-like marker,
which would suggest that example (27c) contains no Focus Operator in the sense intended in the paper'.
43
I agree that particle ke may be a Focus Particle, although I do not think that this is the case with either
instance of ke in (27c). The first ke modifies epipleon while the second ke modifies pali. Neither
instance of ke modifies the proposition and cannot therefore be taken as an overt marker of the Focus
Operator. But I take the point made by the reviewer. ke would indeed be a Focus Particle in example (i)
below, which differs only minimally from (27c). Claim (2d) does not exclude that root infinitives
contain overt markers for propositional operators. Consider also example (2) which, like (27c), contains
no Focus Particle. The na -clause in (ii) differs minimally from ungrammatical (27g) in that the verb is
focused. Grammaticality in (ii) is due to the presence of the Focus Operator.
jia jio
(i) na PROKITE
ke
na IS ABOUT
FocusMarker a son
'Let alone that it is a SON.'
(ii) na FTJAKSO kafe
na MAKE-I
coffee-acc
ti
sinevi
tha mu pis
omos
will me-cl tell-you however what-nom
happened-it
'I WILL make coffee; but you will tell me what happened.'
21
In the MG example the emphasis is on the whole IP. The natural way to render this in English is by
stressing the final word of the IP constituent, i.e. SON, to get wide IP focus.
22
Wishes and curses, which in Gr are na -clauses (cf.i), involve either NO or PO.
(i) na ta
ekatostisis
na them-cl (:your years)
make-you a hundred
'Happy Birthday!' (Literally: 'Make your years a hundred.')
23
Selection is used in the strong sense to mean selection by a predicate (i.e. c-selection) and not in
the weaker sense that also includes selection by a functional head.
24
An anonymous reviewer raises two questions with respect to claim (28'). The first one is how (28')
is to be intergrated into (or made compatible with) the general syntax of Negation and the other
propositional operators beyond the contexts of root infinitives. If I understand the question well, it asks
whether (28') is problematic in view of the fact that propositional operators may also obtain in noninfinitival clauses. I claim it is not. What (28') stresses is the obligatoriness of propositional operator
readings in root infinitives. It makes no claim as to the exclusive occurrence of propositional operators
in root infinitives.
The second question is how Negation can be treated as an entity which selects IP since it is inside
IP. Let me add that the same question arises with respect to the rest of propositional operators. The
problem seems to be due to the fact that propositional operators, though semantic predicates, are not
standardly taken to be syntactic predicates. I can only observe that the problem would cease to exist if
propositional operators were taken to be syntactic predicates as well. In fact this is suggested in (28')
and will be later defended also on the basis of all the syntactic properties of propositional operators (cf.
(3a)-(d. One objection to a syntactic analysis of propositional operators as higher predicates could be
the following. Ifwe take NEG, for instance, we see that the subject ofthe clause precedes NEG. Mutatis
mutandis for the rest of propositional operators. As an answer, one could suggest that propositional
operators may be raising predicates. The second question of the reviewer really addresses the
compatibility between the logical/semantic analysis of propositional operators and their syntactic
properties and I wish to point out that there is a problem there for the standard analysis of propositional
operators as well.
25
The Conditional Operator is not considered because it does not obtain in embedded clauses.
26
A reviewer suggests that (39f) could be considered a semi-question, as in Sutler (1993).
References
UCL Working Papers in Linguistics 2,183-200.
Agouraki, Y. (1991) A Modem Greek Complementizer and its Significance for UG, UCL Working
Papers in Linguistics 3, 1-24.
44
YORYIA AGOURAKI
Agouraki, Y. (1993) Spec-Head Licensing: The Scope of the Theory, Ph.D. dissertation, UCL.
Agouraki, Y. (1996) Embedded Questions, unpublished manuscript, University of Cyprus.
Allwood, J., Andersson, L.-G. & O. Dahl (1977) Logic in Linguistics, Cambridge University Press.
Brody, M. (1990) Some Remarks on the Focus Field in Hungarian, UCL Working Papers in Linguistics
2,201-225.
Chierchia, G. & McConnell-Ginet (1990) Meaning and Grammar, The MIT Press, Cambridge, Mass.
Chomsky, N. (1986a) Knowledge ofLanguage: Its Nature, Origin and Use, Praeger, New York.
Dowty, D. (1982) Tenses, Time Adverbs and Compositional Semantic Theory, Linguistics and
Philosophy 5, 23-55.
Efthimiou, E. (1989) Ni-Clauses: the Infinitival Structure in Modem Greek, Studies in Greek
Linguistics 10,187-204.
En~, M. (1981) Tense without Scope, Ph. D. Dissertation, Universtity of Wisconsin at Madison,
Published by Indiana University Linguistics Club, 1986.
Giannakidou, A. (this volume) Weak and Strong Polarity: Evidence from Greek.
Hornstein, N. (1977) Towards a theory of Tense, Linguistic Inquiry 8,521-557.
Hornstein, N. (1981) The Study of Meaning in Natural Language: Three Approaches to Tense, in N.
Hornstein and D. Lightfoot (eds.), Explanations in Linguistics, Longman, London.
Hornstein, N. (1990) As Time Goes By, MIT Press, Cambridge, Mass.
Klima, E. (1964) Negation in English, in J. Fodor and J. Katz (eds.), The Structure of Longuage,
Prentice-Hall, Englewood Cliffs, pp. 246-323.
Kratzer, A. (1981) The Notional Category of Modality, in H. Eikmeyer and H. Rieser (eds.), Words,
Worlds and Contexts, Mouton de Gruyter, Berlin.
Lasnik, H. & M. Saito (1984) On the Nature of Proper Government, Linguistic Inquiry 15,235-289.
Lewis, D. (1975) Adverbs of Quantification, in E. Keenan (ed.), Formal Semantics of Natural
Language, Cambridge University Press, Cambridge.
Lyons, J. (1977) Semantics, Cambridge University Press.
McCawley, J. (1981) Everything that Linguists have Always Wanted to Know about Logic but were
ashamed to ask, The University of Chicago Press, Chicago.
Pesetsky, D. (1982) Paths and Categories, Ph.D. dissertation, MIT.
Rizzi, L. (1991) Argument / Adjunct (A)symmetries, unpublished maunscript, University of Geneva.
Rizzi, L. (1993) The Case of Root Infinitives, GenGenP 1,16-25.
Rothstein, S. (1991) Syntactic Licensing and Subcategorization, Syntax and Semantics 25,139-157.
Travis, L. (1984) Parameters and Effects of Word Order Variation, Ph.D. dissertation, MIT.
Veloudis, I. & I. Philippaki-Warburton (1983) The Subjunctive Mood in Modem Greek, Studies in
Greek Linguistics 4, 151-168.
ON THE PROPERTIES OF SOME GREEK WORD ORDER PATTERNS'

Artemis Alexiadou
ZAS, Berlin
1. INTRODUCTION
Crosslinguistic variation with respect to the syntactic positions occupied by arguments, e.g. subjects and objects, within clauses has traditionally been in the center of
research within syntactic theory, and has given rise to important developments in
relation to the parametrization of Universal Grammar.
The present study aims at providing an overview of the main aspects which
characterize the basic word order patterns in Greek (Gr) in the light of the view, put
forth in Chomsky (1995), that argument displacement is related to the presence vs.
absence of (morphological) features on functional heads which regulate the application of operations such as Merge and Move; under this view, optionality is severely
restricted.
Gr is a null subject language (NSL) in which various arrangements of the arguments are possible. This property is illustrated in (1):2
(1) a.
h.
c.
0 Aleksandhros nikise to Dhario

the-Alexander-nom won-3sg the-Darius-acc
nikise 0 Aleksandhros to Dhario
nikise to Dhario 0 Aleksandhros
'Alexander won against Darius'
sva
vsa
vas
The orderings presented in (1) are a feature of all clause types in Gr and are not restricted in the way characteristic embedded root phenomena are. Consider (2),
which exemplifies that the SVONSO alternation (VOS is also possible) occurs in
complement clauses (2a) and in complex NP structures (2b):
(2) a.
b.
ipe
oti (0 Arthuros)
sinandise (0 Arthuros)
to Merlin
said-3sg that the-Arthur-nom met-3sg the-Arthur-nom the-Merlin-acc
'He/she said that Arthur met Merlin'
to jeghonos oti (0 Arthuros)
sinandise (0 Arthuros) to Merlin
the fact
that the-Arthur-nom met-3sg the-Arthur-nom the-Merlin-acc
45
A. Alexiadou et al. (ells.), Studies in Greek Syntax, 45-65.

46
ARTEMIS ALEXIADOU
As is well known certain structures, while they are generally restricted to root
clauses, can also appear, under certain conditions, in a restricted subset of complement clauses. Left Dislocation in English is a well known case. However, there is a
fundamental difference between this construction and the patterns in (1): the latter
are found in a wider variety of clauses, even in complex NP structures which typically resist all root phenomena (see Iatridou & Kroch 1992, McCloskey 1996 for a
recent discussion and references therein).
As (1) and (2) also reveal, Gr differs from Italian in an important way: Gr permits VSO orders. As is well known (cf. Rizzi 1982), Italian does not allow VSO
orders in declarative sentences, whether matrix or embedded. In this respect, Gr
word order patterns are closer to the Spanish ones (see Zubizarreta 1992). Thus,
NSLs differ in significant ways from each other with respect to word order.
The Gr patterns in (1) differ from the more or less comparable word order alternations in Germanic (cf. 3&4) in the following obvious respects: a) unlike English
and Icelandic subject inverted orders, no overt expletive is present in Gr (cf. Ib); b)
unlike English, but like Icelandic, transitive inverted orders are possible; c) unlike in
English and Icelandic, inverted subjects in Gr need not be indefmite.
(3) a.
b.
(4) a.
John arrived
There arrived a man
Einhverjir srudentar lasu b6kina
'Some students read the book'
b. pao lasu einhverjir srudentar b6kina
there read some students the book
There are a number of ways in which word order alternations as the ones in (1) can
be accounted for. For instance, (la) could be argued to be similar to (3a-4a), i.e. to
involve subject movement to [Spec,IP] to satisfy the formal EPP requirement; (lb)
could be argued to be similar to (3b-4b), i.e. to involve a covert expletive in
Spec,IP; (lc) could be argued to be derived by rightward movement of the subject
(cf. Rizzi 1982). All these positions have been defended in the literature. It is the
aim of this paper to examine some alternatives to these analyses. The following
considerations underlie the investigation undertaken here: i) subject systems need
not be identical crosslinguistically, and ii) optional word order alternations are
specialized informationally; this specialization is structurally implemented in certain
languages.
The paper is broken down into four parts. In section 2 the clitic left dilsocated
character of preverbal subjects in Gr is briefly discussed. In section 3 VSO orders
are examined. It will be shown that Gr inverted constructions differ from their
counterparts in Icelandic. In section 4 it will be shown how some aspectual
restrictions on word order which can be accounted for under the proposed analysis
for VSO and SVO orders. Finally, in section 5, an attempt is made to account for the
properties ofVOS patterns in Gr, arguing that object movement is PF-driven.
GREEK WORD ORDER PATTERNS
47
2. PROPERTIES OF PREVERBAL SUBJECTS

The intuition that preverbal subjects in Gr are syntactically distinct from preverbal
subjects in non-NSLs goes back to Philippaki-Warburton (1985). There it was
argued that SVO orders involve a topic subject (see also Tsimpli 1990, Drachman
1991, Horrocks 1994, Rivero 1994, and Alexiadou 1994, 1996, 1997, Alexiadou &
Anagnostopoulou 1995, 1998 for a more recent discussion).
In particular, Alexiadou (1994, 1996), Alexiadou & Anagnostopoulou (1995,
1998) present a number of arguments for a left dislocation analysis (in the sense of
Cinque 1990) of the preverbal subject in Gr. The argumentation runs in the following way: the preverbal subject patterns like a dislocated element. Thus, it follows
that it itself is in a base generated dislocated position (see also Barbosa 1994 for
Romance). The evidence is based on distributional and interpretational facts
concerning preverbal quantificational and indefmite subjects, and also on the
interference between preverbal subjects and wh-phrases. For the sake of
concreteness consider (5). (5) shows two things. First, in Gr, for which it has been
argued that it involves V-raising to AgrSO, SVO does not involve a configuration in
which the subject is located in the specifier of AgrSP and the verb is in AgrSo. As
shown in (5a) a number of adverbs intervene between the preverbal subject and the
verb, while (5b) shows that adverbs cannot intervene between the subject and the
verb in French, another V-raising language. Second, (5) also reveals that preverbal
subjects have a strong (partitive/specific) interpretation (cf. 5c). On the other hand,
the subject has weak, existential interpretation in (5d), where it is in postverbal
position. This is not the case in English, where preverbal QPs are actually
ambiguous. Moreover, the subject in (5c) has a similar interpretation to the clitic left
dislocated (CLLDed) object in (5e).
(5) a.
b.
c.
d.
e.
0 Arthuros
hthes
meta apo poles prospathies
the-Arthur-nom yesterday after from many efforts
sinandise to Merlin
met
the-Merlin-acc
lit.'Arthur fmally met Merlin'
*Jean probablement aime Marie
Enas sinandise to Merlin
one met-3sg the-Merlin-acc
'A certain person/one of the people met Merlin'
sinandise enas to Merlin
'Someone met Merlin'
?Enan ton sinandise 0 Merlin
one-acc c1-acc met-3sg the-Merlin-nom
'Merlin met one of the people'
On the basis of the assumption that the preverbal subject in Gr is directly merged in
preverbal position, we can account for the fact that it is unambiguous.
48
ARTEMIS ALEXIADOU
Here, I will not proceed into a detailed presentation of all the arguments
pointing to the CLLDed nature of the preverbal subject in Gr. The reader is referred
to the works cited above. Rather, I will discuss a number of counterarguments
against this position which have been brought up for other NSLs (see Cardinaletti
1995 for a recent discussion based on Italian facts, Motapanyane 1994 for
Romanian), and see how Gr preverbal subjects fare.
Cardinaletti (op.cit.) observes, that although full subjects in NSLs can be leftdislocated, they need not be left dislocated. 3 In fact, in Italian a full subject can be
found in a position internal to the sentence, comparable to the position of the subject
in English. This is more evident in contexts such as Aux-to-Comp (see Rizzi 1982)
and complementizer deletion (see (6) below). This is exactly the position where a
quantifier like nessuno 'nobody' which cannot be left dislocated can also occur:
(6) a.
b.
c.
Credevo Gianni/nessuno avesse telefonato a Maria

I believed Gianni/nobody had called
to Mary
??Credevo a Roma Gianni ci
avesse vissuto per venti anni
I believed in Rome Gianni there had
lived for twenty years
Credevo che a Roma Gianni ci avesse vissuto per venti anni
Gr subjects fare differently. First of all, comparable contrasts as the ones illustrated
above are not found in Gr (see 7). Both preverbal subjects and CLLDed elements
are accepted in complementizer deletion contexts:
(7) a.
Nomizo (oti) 0 Janis

tilefonise
think-lsg that the-John-nom called
tulahiston 20 hronia
b. Nomizo (oti) sto Parisi zisane
think
in Paris lived -3pl at least 20 years
c. Nomizo oti sto Parisi 0 Janis ezise tulahiston 20 hronia
Moreover, negative quantifiers can be CLLDed. As Giannakidou (1997) points out,

emphatic negative indefmites in Gr e.g. kanis 'nobody', which represent the counterpart of the Italian quantifier, show a behavior similar to CLLDed elements.
Emphatic preposing occurs in root and embedded clauses; emphatics, like CLLDed
elements, are insensitive to CP-recursion. Consider (8a-b) (from Giannakidou 1997:
185) and compare (8a-b) with (8c-d) which involve a CLLDed object:
(8) a.
b.
c.
epidhi KANENAN 0 Janis

dhen idhe
because nobody
the-John-nom neg saw-3sg
na
ghnorisa ton antbropo pu KANENAN dhen katafere
met-l sg the man
that nobody
neg managed-3sg subj
kani eftihismeno
make happy
epidhi aCta ta pragmata 0 Janis ta
aghnoi
because these-things-acc the-John c1-acc ignores
d.
49
ghnorisa ton anthropo pu afta ta pragmata ta

kseri kala
met-lsg the person that these-the things-acc cl-acc knows well
Furthermore, as Cardinalletti (1995) points out, in Italian, depending on the type of

the wh-constituent, a full subject is sometimes allowed to occur between the whphrase and the verb. This position however is not available for a left dislocated element, which tends to precede the wh-phrase:
(9) a.
b.
A chi
Giannilnessuno ha consegnato l'invito
to whom Gianni nobody has given
the invitation
l'invito,
a chi (?? l'invito) Gianni l'ha consegnato
the invitation to whom
Gianni it has given
In Gr, however, both preverbal subjects and left dislocated elements are acceptable
in such contexts (see Anagnostopoulou 1994, Horrocks 1994, Drachman and Klidi
1992 for extensive discussion):
(10)
a.
b.
Pjon apo tus filus

tu
0 Petros sinanda sihnotera
who from his friends cl-gen the-Peter meets more often
pjos apo tus fitites
tin askisi tin elise amesos
who from the students the problem cl-acc solved immediately
Thus, the above presentation leads us to conclude that preverbal subjects in Gr can
be argued to be left dislocated and not to involve NP-movement to Spec,IP.4
Crucially, the preverbal subject is situated in the specifier position of a Topic
Phrase, where it is directly merged and [Spec,IP] is not present.
(11)
[TopicP
[IP
VO]]
3. VSO ORDERS
3.1 VP Internal Subjects
VSO orders in Gr occur with any type of eventive predicate. s Consider (12). As is
clear, postverbal subjects occur with accomplishment (12a), achievement (12b) and
process (12c) predicates:
(12)
a.
h.
ehtise i Maria
to spiti
built the-Mary-nom the-house-acc
'Mary built the house'
kerdhise i Maria
ton aghona
won
the-Mary-nom the-race-acc
'Mary won the race'
ARTEMIS ALEXIADOU
50
c.
eghrafe
i Maria
to ghrama
010 to proi
wrote-imp-3sg the-Mary-nom the-letter-acc all the morning
'Mary was writing the letter the whole morning'
There is a consensus in the literature that VSO orders occur when the subject
appears in a position below the functional head in which the fInite verb appears.
However, there is no consensus with respect to the exact location of the verb and the
location of the postverbal subject. A third related issue concerns the presence of an
expletive pro in [Spec,IP] in case the subject does not appear in this position.
There are a number of ways in which VSO linear strings can be analysed. A
fIrst possibility that comes to mind is to analyse them as involving V-to IO-to-C
movement, the subject remaining in [Spec, IP]. This analysis has been proposed, for
instance, for Celtic VSO by Deprez & Hale (1986 among others) and for Greek in
Varlokosta and Hornstein (1993). However, such an account is disfavored, as there
is no root vs. non-root asymmetry with respect to VSO patterns (see the examples in
(2) above).
Another possibility would be to propose that the verb raises to the highest Infl
head, but the subject does not move to the highest Spec in the IP. Rather, it moves to
a lower specifIer position. In other words, under such an approach Greek would
have one other than [Spec,IP] VP external specifIer available for subjects. Tsimpli
(1990) proposes that the subject moves to Spec,AgrSP and the verb to the higher TO
head, crucially making Greek VSO orders somehow similar to Celtic VSO (see
McCloskey 1996) or Icelandic TECs. 6 According to certain proposals in the literature (Bobaljik & Carnie 1992, McCloskey 1994, Jonas & Bobaljik 1993 among others), Celtic dialects and also Icelandic license Spec,TP as an intermediate subject
position. Spec, TP is the position subjects overtly occupy in Celtic VSO and in Icelandic TECs.
However, there are considerable differences between these language types and
Gr. First of all, note that in VSO orders in Gr no adjacency between the verb and the
subject is necessary as in Celtic dialects. Moreover, in Gr, the subject is always
lower than aspectual adverbs, assumed to be situated in Spec,AspP (see Alexiadou
1994), while it is higher than these adverbs in Celtic languages (cf. the Irish example
in (13a) from McCloskey 1994):
(13)
a.
Ni bhfuair aon bhean riamh roirnhe greim laihrnhe air

neg took any woman ever before-it grip hand-gen on-him
'No woman had ever before taken his hand'
b. dhjavaze
sihna 0 Aleksandhros
tin Iliadha
read-imp-3sg usually the-Alexander-nom the-Iliad-acc
Furthermore, as shown in (14a) the order of constituents in periphrastic

constructions in Gr is auxiliary, aspectual adverb, participle, light manner adverb
and subject. In Alexiadou (1994) the relative order between the light manner adverb,
situated in Spec,VoiceP, the projection which immediately precedes the VP, and the
participle is taken as evidence showing that the participle has moved outside the VP
51
domain. The participle is argued to raise obligatorily to Aspo. The subject in Gr

follows both the light manner adverb and the participle. From this, one can conclude
that the subject occupies its VP internal position. 7 (14b), where the subject
intervenes between the auxiliary and the participle, is ungrammatical. Note that
there is no strict adjacency requirement between the participle and the auxiliary,
since aspectual adverbs can intervene:
(14)
a.
b.
an ehun idhi mathi kala i Kokini

to sistima tus
if have already learnt well the-Reds-nom the system cl-gen-pl
'If the Reds have already learnt their system well .. .'
*an ehun idhi i Kokini mathi kala to sistima tus
(15) illustrates that in Icelandic the participle follows the manner adverb. That the
participle does not move in Icelandic has been argued for by a number of people
(see Holmberg 1986). Yet, the subject precedes both the manner adverb and the
participle, thus being VP external.
(15)
pa6 hefur sennilega einhver alveg

loki6 verkefninu
there has probably someone completely ftnished the assignment
Similar observations have been made for Irish (see Carnie 1993). From the above
discussion, it is clear that Greek VSO orders are not similar to Celtic VSO orders or
Icelandic inverted constructions, given that subjects are VP internat.s As Jonas &
Bobaljik (1993) argue, the following generalization holds: subject inverted orders
with transitive predicates and object shift can only exist in languages which license
Spec,TP as an intermediate landing site for the subject. Thus, Icelandic licenses
[Spec,TP] and permits TECs, while English does not license [Spec,TP] and shows
an intransitivity constraint on inverted orders. Gr does not show an intransitivity
constraint on inverted orders (matters are a bit complicated when it comes to object
shift; see the discussion in section 5). Thus, in Gr Spec, TP as a subject position is
not licensed.9 The previous section established that [Spec,IP] is not present in Greek.
Given the results of this section, I conclude that Greek does not make use of
functional specifters as landing sites for A-movement of subjects. In other words, it
seems to be the case that the N-features of T are weak in Gr and subject
displacement is not triggered. As far as the D-features of AgrS are concerned, the
remarks made in note 4 apply (and see Alexiadou & Anagnostopoulou 1995, 1998
for details).
(16) below illustrates the structural representation proposed for Gr VSO orders:
52
ARTEMIS ALEXIADOU
(16)
C
CP
AgrSP
/"--....
an AgrSO TP
ehun ~
Spec
idhl
AspP
A AspP
A;:;:----VoiceP
mathij
Spec
VoiceP
kala
VoiceD
VP
~
i Kokini ~ to sistima tus
In the following section I will present another characteristic property of Greek
VS(O) linear strings which further distinguishes them from Icelandic TECs and
English inverted orders.
3.2 Remarks on Expletive Pro

A characteristic property of Gr inverted orders is that they do not show any defmiteness restriction effects. As is well known, DR effects are present with 'there' type of
expletives and 'iI' type of expletives (cf. 17a&b). The former do not trigger agreement with the verb, while the latter do. Given that in Gr agreement is always with
the associate, if there was a covert expletive present (as it has been assumed for
inverted orders in NSLs, see Rizzi 1982), it would have to be a 'there' type one (see
Chomsky 1995 for discussion).
(17)
a.
b.
*there arrived every child
il est arrive un homme/*Jean

is arrived a man/*John
It has been observed (see Chomsky 1981, Safu 1985, Calabrese 1992 and references
therein) that DR effects in unaccusative constructions are systematically absent in
NSLs. The following data show this property of Gr VS orders. to
(18)
53
irthe
kathe pedhilo Janis sto jipedho
arrived-3sg every child/the-John to the stadium
Turning now to transitive inverted orders, as (19a) shows DR effects are absent in
Gr, while they are present in Icelandic (cf. 19b):
(19)
a.
dhjavase ena pedhilkathe pedhi to vivlio

read-3sg a child/every child the-book-Acc
'A/every child read the book'
b. pao lasu einhverjir studentar bokina
there read some students the book
In Alexiadou & Anagnostopoulou (1995, 1998) the non-universality of DR effects

in transitive (and unaccusative and some unergative) constructions was taken as
evidence that DR effects are syntactically triggered in these contexts (unlike
existential and 'donkey anaphora' contexts). For this reason, the authors adopted, for
these specific constructions, Chomsky'S (1995) analysis of the DR effects. More
specifically, they assumed that 'there' is a determiner head which takes an NP
complement, hence the presence of DR effects (cf. Chomsky 1995, Frampton 1995,
and unlike Hoekstra & Mulder 1990, Moro 1997 among others). Note that
morphologically 'there' is very close to the determiner 'the' in English, hence an
account along those lines seems plausible.
If a covert expletive 'there' were present in the Greek examples, it would have to
behave very differently from its overt counterpart. However, the unmarked option
would be that overt and covert expletives behave alike. On the basis of this reasoning, one can conclude that the lack of DR effects is an argument that there is no expletive in VSO orders in Gr (see also McCloskey 1994 for a similar conclusion concerning Irish VSO orders which also lack DR effects). Note that its presence would
not be needed for formal reasons either, since V-raising checks the EPP feature (see
note 4). Moreover, if the argumentation in favour of the absence of expletive pro in
VSO orders is on the right track, the analysis of SVO orders in terms of left dislocation is more or less forced within the framework of Chomsky (1995): since no
expletive is merged in [Spec,IP], no overt subject movement can occur to check the
EPP feature; SVO and VSO orders start with the same numeration and subject
movement is ruled out as a Procrastinate violation.
4. ASPECTUAL RESTRICTIONS
The discussion so far has shown the following two things: a) SVO orders in Gr involve left dislocation, and b) VSO orders involve VP internal subjects. In this
section I will argue that these syntactic structures express also the distinction
between categorical and thetic judgments.
ARTEMIS ALEXIADOU
54
As noted in the previous section, structures with postverbal subjects seem to be

constrained to eventive predicates. Stative predicates only occur in SVO order, apparently also in Italian and Spanish (see Calabrese 1992, Zubizarreta 1994):
(20)
a.
b.
c.
I Meropi
kseri Ispanika
the-Meropi-nom knows Spanish
'Meropi knows Spanish'
*kseri Ispanika i Meropi
*kseri i Meropi Ispanika
I will argue that this restriction is related to the interpretation that VSO orders receive in Gr (see also Alexiadou 1996). Before I proceed, some notes on the nature of
thetic and categorical statements are in order.
Kiss (1995) among others, following Kuroda (1972), points out that categorical
judgments displaying a notional subject-notional predicate structure consist of two
acts: the act of recognition of that which is made the notional subject and the act of
affrrming or denying that which is expressed by the predicate of the notional subject. Notional subjects/topics are [+specific] referring expressions and they foreground a particular individual as the subject of the predication. Thetic judgments, on
the other hand, contain only a mere notional predicate which consists of a single act:
the act of recognition of the material of a judgment. In other words, thetic judgments
simply communicate events (on this distinction see also Ladusaw 1994 and Giannakidou this volume). Given that thetic judgments communicate events, they are
assumed to be appropriate answers to the question 'what happened?'.
It has been observed that VSO orders in Gr and in NSLs in general are
understood as answers to the question 'what happened?' (cf. Philippaki 1985,
Comorovski 1991, Anagnostopoulou 1994, Zubizarreta 1994, Alexiadou &
Anagnostopoulou 1995, Alexiadou 1996). SVO orders, analyzed here as involving a
'topic' subject, are unacceptable in these contexts (cf. 21 b).
(21)
what happened?
a. molis erikse i ghata
to potiri
just threw the cat-nom the glass-acc
b. *molis i ghata erikse to potiri
If the above observations are on the right track and given the discussion in the previous sections, we can account for the facts in (20) as follows: stative predicates are
inherently incompatible with the discourse function of VSO contexts; they correspond to categorical statements. In other words, they always need a notional topic so
that this can act as the subject of the predication. In support of this claim note that
(22), which involves a CLLDed object, is grammatical:
(22)
ta Ispanika
ta
kseri i Meropi
the-Spanish-acc-pl c1-acc-pl knows the-Meropi-nom
55
Crucially, dislocation of one of the two argumental DPs is necessary for the
sentence to receive the appropriate discourse interpretation. Thus, Gr qualifies as a
language that distinguishes between categorical and thetic judgments syntactically.
Since DP movement does not take place for reasons of feature-checking these
aspectual/discourse contrasts are more obvious in a language like Greek. Similar
contrasts are not expected to exist with Celtic VSO orders or Icelandic TECs, that is
in languages where NPs move for reasons offeature checking (see Kiss 1995).
As noted in Alexiadou (1996), the above restrictions on word order are
cancelled, when the verb is focalized:
(23)
KSERI Ispanika i Meropi
Similar contrasts are reported in Kiss (1996) for Hungarian. As is clear, focusing has
the effect of neutralizing the requirement for a specific subject. The above facts can
be interpreted as follows: stative predicates always demand a [+strong] DP in
preverbal position. In focus environments, i.e. in quantificational structures, the focus operator can be argued to play the role of the [+strong] item necessary for the
intended interpretation. As a result, inverted orders become acceptable. II
Given the above argumentation, Gr seems to pattern like a topic prominent language, in which the notional subject, or topic, does not have to coincide with the
grammatical subject at all. This is taken in Kiss (1995) to illustrate that the language
does not distinguish between thetic and categorial structures syntactically. English,
where both specific and non-specific subjects share the same structural position,
does not show such a behaviour. Another argument pointing to the topic prominent
character of Gr is demonstrated below. As (24) shows, the order of dislocated
elements in Gr is free. According to Kiss, if a language shows this free ordering,
then this language qualifies as a topic prominent language:
(24)
a.
b.
tis Marias
to vivlio
tis
to
edhose
the-Mary-gen the book-acc cl-gen cl-acc gave-3sg
to vivlio
tis Marias
tis
to
edhose
the book-acc the Mary-gen cl-gen cl-acc gave-3sg
To conclude, Gr SVO and VSO orders seem to be specialized informationally. The

next section deals with VOS orders. As it will tum out, these seem to perform an
informational 'task' as well.
5. VOS ORDERS
5.1 Description of the Facts
VOS orders in Gr have been analysed as involving right adjunction of the subject to
the CP (see Tsimpli 1990). Here, I will present evidence, as it has also been argued
56
ARTEMIS ALEXIADOU
for Spanish by Ordonez (1994), that this view is not correct and that leftward object
movement over the subject is involved (see also Alexiadou 1994).
First of all, if subjects in VOS orders were right adjoined, we would expect that
the following order V-CompIClause-S to be possible. However, (25b) is not grammatical:
(25)
a.
rotise i Meropi
an
tha erthi
0 Ahileas
asked the-Meropi-nom whether fut come-3sg the-Achileas-nom
'Meropi asked whether Achileas will come'
b. *rotise an tha erthi 0 Ahileas i Meropi
V COMP-Clause
S
The claim that the object has moved to a higher Spec position leaving the subject
behind, explains the following binding asymmetries,12 which remain unaccounted
for under the right adjunction hypothesis:
(26)
(27)
*sinandise 0 pateras
~
[to kathe koritsi]i;
met-3sg the father-nom cl-gen the each girl
a. sinandise [to kathe pedhi]i 0 pateras tui;
met-3sg
the each boy the father-nom cl-gen
[kathe pateras]i
b. sinandise to pedhi ~
met-3sg the child cl-gen every father
(26) is ungrammatical since it fails to meet the c-comrnand requirement on the pronominal binding of the possessive. On the other hand, in (27a) the object has been
scrambled to the left and thus c-commands the subject possessive pronoun. Moreover, (27b) shows that the object may bind the subject, but not vice-versa. Given that
this type of scrambling feeds binding, it shows properties of A-movement and not of
A'-movementY Thus, the above facts indicate that overt leftward object movement
takes place. In (28) the proposed structural configuration is given where it is shown
that the object has overtly scrambled over the subject:
(28)
XP
D0x'
~VP
XO
/'-....
DPs
V'
A number of questions arise immediately: a) what are the similarities between this
type of object movement and Germanic scrambling/object shift; why doesn't Gr
have an active [Spec,TP] for subjects, if the presence of object shift determines
activation of [Spec,TP]; b) what is the nature of the XP projection in (28), and c)
what is the trigger for this movement? The next sub-sections deal with these issues.
57
5.2 Greek VOS vs. Germanic Scrambling

Let us fIrst examine how similar this type of scrambling is to the one found in Germanic. Germanic Scrambling/Object-shift is characterized by a number of
properties. First, Scandinavian Object Shift is characterized by de-focusing and has
been argued to involve object deplacement in the PF component, since it does not
feed binding relations (cf. Holmberg 1996). However, Greek object movement has
effects on binding relations. Thus, it qualifIes as an instance of syntactic Amovement. Contrast (27) with (29) (from Holmberg 1996: 7):
(29)
a.
b.
Vi ansag till deras/*sin; besvidelse [Per och M~

we considered to their/refl disappointment Per and Martin
vara odugliga
to be incompetent
Vi ansag deIllt till deras/*sin; besvidelse
we considered them to their/refl disappointement
[~ vara odugliga
to be incompetent
Moreover, as is well known, Germanic Scrambling/Object Shift is sensitive to the

referential nature ofNPs (cf. Johnson 1991, Diesing & Jelinek 1993, Vikner 1994
a.o.), and it is subject to several restrictions pertaining to their defmiteness. In some
Germanic languages, the class of elements that may undergo scrambling/object shift
is limited. In Icelandic, for instance, object shift of defmite NPs is grammatical
(30a) while object shift of bare plurals is ungrammatical (cf. 30b; data from Vikner
1994):
(30)
a.
b.
Eg las bokina ekki

I read book-the not
Hann las brekur ekki
he
read books not
Furthermore, Scrambling/Object-shift is associated with strong/specifIc interpretation of NPs (cf. de Hoop 1992, Adger 1993, Abraham 1994, Diesing 1992,
Meinunger 1996 among many others). This is shown in the paradigm in (30) from
Dutch, where scrambling triggers referential (31a), partitive (31b) and generic (31c)
readings on weak NPs (cf. de Hoop 1992):
(31)
a.
b.
c.
dat de politie een kraker gisteren opgepakt heeft

that the police a squatter yesterday arrested has
dat de politie twee krakers gisteren opgepakt heeft
that the police two of the squatters yesterday arrested has
dat de politie krakers altijd oppakt
that the police squatters always arrests
58
ARTEMIS ALEXIADOU
As argued for in detail in Alexiadou & Anagnostopoulou (1997c), the Greek

counterpart of Germanic scrambling/object shift is clitic doubling. Gr Clitic
doubling, much like object movement phenomena in Germanic, is associated with
similar interpretational effects. For instance, doubling of defmite NPs is well formed
(32a), while doubling of indefmites is ungrammatical (32b). Most importantly, in
doubling movement of the NP does not take place, as shown in (32c), where the
doubled NP follows the postverbal subject, the light manner adverb and the
participle:
(32)
a.
to
dhjavasa to vivlio
me prosohi
cl-acc read-lsg the-book-acc with care
b. to
dhjavasa kapjo vivlio
me prosohi
cl-acc read-lsg some book-acc with care
c. to
ehi idhi dhjavasi kala 0 Janis
to vivlio
cl-acc has already read
well the-John-nom the book-acc
Greek VOS orders, however, lack the above presented properties; weak NPs are
possible and actually preferred (see 33). Nothing changes in the interpretation of the
NP in these cases, i.e. the NP does not have a 'strong' reading contrary to the Dutch
facts above:
(33)
eghrafe ghramata 0 Janis

wrote letters
the-John-nom
'John was writing letters'
Moreover, unlike DPs in Germanic, which precede aspectual adverbs, the position to
which the object moves is rather low, lower than the aspectual adverb situated in
[Spec, AspPJ. As (34c) shows, objects in VOS orders follow the participle which is
situated in Aspo (cf. 34c):
(34)
a.
aghorase amesos
to vivlio
0 Petros
bought-3sg immediately the book-acc the-Peter-nom
b. *aghorase to vivlio amesos 0 Petros
Petros
c. ehi idhi
aghorasi to vivlio 0
has already bought the vivlio-acc the-Peter-nom
'Peter has already bought the book'
From the above discussion, I conclude that VOS orders are sharply different from
object movement constructions in Germanic. In the following, I will argue that this
type of movement is actually PF-driven. 14
59
5.3 Toward an Explanation

A characteristic property of vas constructions in Greek is that in those the subject
carries the most prominent accent of the sentence (see Joseph & Philippaki-Warburton 1987).15
(35)
nikise tus andipalus

a ALEKSANDHRaS
won the opponents-acc the- Alexander-nom
As is well known, in phonological terms there is a close connection between phrasal

accent and focus. The focus of the sentence is intonationally prominent. The concept
of focus has been a subject of extensive debate in the literature (see Selkirk 1984,
Cinque 1993, Zubizarreta 1994, Vallduvi 1993 among others; for Greek see Tsimpli
1995, Agouraki 1990). As also pointed out in Kiss (1995), the term focus is used in
two different senses: it can denote the part of the sentence which carries new information and also it can mean an operator expressing identification. The focus
operator is associated with a particular structural position in sentence structure, a
functional projection of its own. The focused element is identified by the feature
[+F(ocus)]. The prosodic properties (stress and so forth) are argued to be linked to
the presence of this feature. For the discussion here the first sense is relevant.
According to Zubizarreta (1994), who builds on Cinque (1993), what is
characterized as the focus of the sentence must carry the main stress of the sentence
as well.
As has been noted, the notions 'new' and 'old' information are reflected in the
manner in which phrases are structured in the sentence. In Romance, in sentences
with unmarked intonation postverbal subjects constitute new information and as a
result are focused or part of the focus (see Calabrese 1992, Zubizarreta 1994, and
section 4 for discussion). Unmarked focus intonation is used when a sentence is
uttered out of the blue. vsa strings, as pointed out in section 4, are the ones that can
be uttered 'out of the blue', and the whole sentence is asserted as 'new'. In vsa
orders, both the subject and the object remain VP internal. Thus, it is VP internal
material that constitutes new information, a conclusion which has been reached in
the relevant literature. Hence, the VP domain could be associated with the presence
of [+foc] feature (see Diesing 1992, Abraham 1994, and Holmberg 1996 for a recent
discussion).
I would like to propose, in the spirit of Vallduvi (1993), that what happens in
vas orders in Greek is the following: a syntactic operation is performed through
which the association of focus and intonational prominence is mediated. Crucially,
the object, a non-focal element, is forced to move out of the focus domain. Clearly,
the property that forces DPs to leave the VP, when they do not have to for reasons
of formal feature checking, must have to do with their semantic and structural
incompatibility with this VP domain. Moreover, their structural properties have to
be such that movement is legitimate. When the object moves out of the VP, the
subject is the most embedded constituent and receives the main accent of the
sentence. As discussed in Vallduvi (1993), the necessary syntactic operations take
60
ARTEMIS ALEXIADOU
place so that the syntactic configurations fit the prosodic structure. In this respect
Greek object movement is different from the type of object shift discussed in
Holmberg (1996), which is described as a case of movement in the PF component.
What type of elements are incompatible with this VP domain? Elements with
deficient structure are possible candidates (see Cardinaletti and Starke 1995, Chomsky 1995). As Chomsky (1994, 1995) proposes, fmal elements that have a complex
structure, which focused constituents plausibly have, can be linearly ordered by the
Linear Correspondence Axiom (LCA), which applies on the way to PF. However,
when these elements are non-complex, then they cannot be ordered by the LCA and
as a result they have to move overtly. I would like to propose that exactly this property of objects forces them to move overtly in VOS orders. That is they are noncomplex terminals, found in a domain which demands for complex structures and
thus they move obligatorily. This movement is similar to the one weak adverbs undergo in Greek (see Alexiadou 1994, 1997):
(36)
a.
b.
apandise
kala i Meropi
tin erotisi
answered-3sg well the-Meropi-nom the question-acc
apandise i Meropi tin erotise KALAI*kala
(36b) shows that only the focused variant of the adverb is acceptable in VP internal
position. The position to which weak manner adverbs moves to in Gr has been identified with Voice Phrase in Alexiadou (1994, 1997). Now, consider the following
sentences:
(37)
a.
b.
c.
d.
*eghrapse amesos
prosektika ta ghramata 0 Aleksandhros
wrote-3sg immediately carefully the-letters-acc the-Alexander-nom
*eghrapse amesos ta ghramata
prosektika 0 Aleksandhros
eghrapse amesos
ta ghramata 0 Aleksandhros PROSEKTIKA
wrote-3sg immediately the letters the-Alexander-nom carefully
*eghrapse amesos prosektika 0 Aleksandros ta ghrama~_focus)
As the contrast in the above data shows, the scrambled object and the manner
adverb seem to compete for the same position. If this position, that is Voice Phrase
or Johnson's (1991) flPhrase, can be identified as the one that hosts 'weak' elements,
then the facts are straightforwardly accounted for. In support of this, note that the
object in VOS strings cannot be modified and coordinate (cf. (38) below), as one
would expect if it were a weak element in Cardinaletti and Starke's (1995) terms.
Since all non-focal elements must move out of this domain, it is kind of expected
that the adverbial facts should pattern the way they do and most importantly that
object movement should interact with adverbial movement: 16
(38)
a.
*eghrapse ta pjo katapliktika ghramata pu eho

wrote-3sg the most amazing
letters
that have-Isg
dhjavasi 0 Aleksandhros
read -lsg the-Alexander-nom
61
b. *egbrapse ta gbramata ke ta vivlia 0 Aleksandhros

wrote-3sg the letters and the books the-Alexander-nom
Summarizing, the overt deplacement of the object in Gr has to do more with its
structural incompatibility with the focus domain. After object movement, the subject
is left as the most embedded constituent in the sentence and necessarily receives the
main accent. Under such an analysis, Gr qualifies as a language that has instances of
object shift. Thus, if Jonas & Bobaljik's (1993) generalization were on the right
track, one would expect that Gr should show instances of subject displacement as
well, contr~ to fact. Hence, the licensing of Spec,TP for subjects seems to be
related to 'some other properties of the languages in question. 17
6. CONCLUSION
In this paper, I have examined the three major word order patterns (SVa, vsa,
VaS) of Gr. I have presented evidence that (a) preverbal subjects behave like c1itic
left dislocated elements, (b) in inverted orders subjects remain VP internal and no
expletive is present, and (c) in vas orders leftward object movement takes place
which is necessary so that the syntactic structure fits the prosodic structure. All these
structural configurations were contrasted with their counterparts in languages like
English, Germanic or Celtic languages and the sharp differences between these and
Gr were illustrated. Crucially, sva orders in Greek unlike their counterparts in English or French involve left dislocated subjects; Greek vsa strings, unlike Celtic
vsa orders and Icelandic TECs, involve VP internal subjects; Greek object
scrambling does not have the properties of Germanic Scrambling, but it is rather PF
driven.
Moreover, I have shown that in a language like Gr where no DP movement is
necessary for reasons of feature checking, discourse notions such as 'topic', and 'focus'/'old' and 'new' information enter special syntactic structures. These syntactic
configurations reflect the difference between thetic and categorical judgments.
Crucially, the paper has shown that word order is not as 'free' as generally assumed,
but rather controlled by specific principles of the Grammar; thus, word order
alternations do not seem to be derived from 'optional' movements.
NOTES
Parts of the material included in this paper were presented at the Workshop on Greek Syntax in
Berlin (December 1994), and at the 5th Colloquium on Generative Grammar in La Coruna (April 1995).
I would like to thank Elena Anagnostopoulou, Anastasia Giannakidou, Andre Meinunger, Jamal
Ouhalla, Irene Philippaki-Warburton, Melita Stavrou, Dhimitra Theophanopoulou-Kontou, Spyridoula
Varlokosta, and two anonymous reviewers for comments and suggestions.
2
O(c1)VS orders are also possible. They are either a result of object left-dislocation (when the c1itic is
present) or of object focalization (when no c1itic is present).
62
ARTEMIS ALEXIADOU
Cardinaletti (1995) points out that the CLLD analysis is wrong for 'weak' subjects, i.e. the 'egli'
series in Italian. Since Gr lacks overt weak subjects, this issue will not be discussed here.
4
The question that arises is whether Gr qualifies as a non-EPP language. As proposed in Alexiadou
& Anagnostopoulou (1995,1998) V-raising is sufficient to check the EPP feature in 1. In other words,
verbal agreement in Gr being [+nominal] can erase the nominal feature of 1. Moreover, since verbal
agreement has [+interpretablel phi features, there is no issue concerning the existence of referential pro;
this is replaced by rich Agr (see also Philippaki-Warburton 1989, Ouhalla 1994).
The analysis of preverbal subjects as left dislocated elements raises non-trivial questions with
respect to Control, ECM and Raising structures (cf. Motapanyane 1994). Note that as far as Raising is
concerned, subject movement might not be necessary to ensure agreement between the raising verb and
the preverbal subject as in (i).
(i)
(i fitites)
fenonde na
orghanonun/*orghanoni mia apeIjia (i fitites)
the-students seem-3pl subj prepare-3pl/prepares a strike
'The students seem to prepare a strike'
I would like to suggest that obligatory agreement is actually a result of a 'reanalysis' of the two
predicates at LF. This reanalysis can only take place under specific conditions. As SoJa (1992),
following Borer (1989), proposes, in his discussion of similar constructions in Catalan, the embedded
AgrO becomes anaphoric to the matrix Agro. Such a process is possible only when the embedded AgrO is
defective, that is non case marked. That the lower Agr is defective in these structures has been argued
for Gr in Iatridou (1993). As a result of this process, the two Agrs end up co-indexed and agreeing.
Thus, in raising constructions main and embedded AgrO are reanalysed as an extended head-chain. That
is precisely what happens in the Greek examples: the two predicates are reanalysed as one (see also
Philippaki- Warburton & Catsimali this volume).
For Control and ECM structures see Alexiadou & Anagnostopoulou (1997b).
Stative verbs are preferred in SVO order (see section 4). With unergatives grammatical aspect, i.e.
the distinction between Perfective vs. Imperfective seems to influence word order (see Alexiadou 1996
for discussion).
6
In Tsimpli's system Tense is higher than AgrSO and the subject occupies Spec, AgrSP. Here, I
assume the order of functional categories put forth in Belletti (1990), Chomsky (1992) among others
where AgrS precedes Tense.
7
Apart from Tsimpli, most of the researchers (see the references cited) assume, without presenting
structural arguments though, a VP adjoined position for subjects in VSO orders in Or, where the subject
is assigned Case under governemt from Infl (see Koopman and Sportiche 1991).
As proposed in Alexiadou & Anagnostopoulou (1998), in VSO orders the verbal agreement and the
,
overt DP form a doubling configuration in which the agr (=clitic) is responsible for the checking of the
DP Case. It is argued that the clitic forms the non-trivial chain necessary for case-checking.
9
This might amount in saying that ro and AgrO are 'fused' in Greek. See latridou (1990), Rivero
(1994) and Thniinsson (1995) for a proposal regulating the (un)availability of a split !NFL across
languages.
10
As discussed in Belletti (1988), the counterpart of (18c) is ungrammatical in Italian. As is clear from
section 2, a number of differences exist between Or and Italian (see also Anagnostopoulou this volume).
A closer study of all contrasts between these two NSLs is needed.
11
Kiss (1996) points out that in similar contexts in Hungarian the focus operator represents the main
assertion, hence the neutralizing effects.
12
Similar binding facts are found in Spanish VOS orders (data from Ordofiez 1994):
(i) a.
A quien Ie presento [cada nifio]i SUi madre?
to whom c1 introduced each boy his mother
b. *A quien Ie presento SUi madre [cada nifio]i?
13
Ordofiez (1994) points out that object-raising favours wide scope of quantifiers. Similar
observations can be made on the basis of Greek facts. Thus, (ia) contrasts with (ib) in that only the latter
is ambiguous- the universal quantifier taking wide or narrow scope. In (ia) wide scope is preferred:
(i)
a. sinodhepsan kathe pedhi 3 anthropi sto sholio
accompanied each child 3 persons to school
= each child was such that 3 persons accompanied it to school
63
b.
sinodhepsan 3 anthropi kathe pedi sto sholio

accompanied 3 persons each child at school
14
Note that licensing of floating quantifliers, which is possible with scrambled objects in Gennanic, is
marginal with VOS orders. As Anagnostopoulou (1994) points out, stranded quantifiers in Greek must
be always doubled (cf. ii):
(i)
*?dhjavase ta vivlia
0 Janis
ola
read-3sg the books-acc the-John-nom all-acc
(ii) ta
dhjavase 0 Janis
ola
c1-acc read-3sg the-John-nom all
See Anagnostopoulou (1994) for a discussion on the exceptional properties of the Greek quantifier ola.
15
See also Bonet (1990) and Vallduvi (1993) who report similar facts for Catalan VOS orders.
16
Melita Stavrou pointed out to me that VOS orders with heavy stress on the object are also possible.
Similar facts are also discussed in Zubizarreta (1994) for Spanish VOS orders. Zubizarreta proposes that
these cases involve subject right dislocation, so that the object remains the most embedded constituent.
Given Kayne's (1994) Antisymmetry Hypothesis, these facts require an alternative analysis. Presumably
in these cases the subject is not part of the same VP or CP as the object.
17
Alexiadou & Anagnostopoulou (1997a) deal extensively with these issues. Thus, I leave the topic
open here.
REFERENCES
Abraham, W. (1994) Fokusgramrnatik und Indefinitheit, in B. Haftka (ed.), Was determiniert
Wortstellung, Westdeutscher Verlag, Opladen, pp. 235-245.
Adger, D. (1993) Functional Heads and Interpretation, Ph.D. Dissertation, University of Edinburgh.
UCL Working Papers in Linguistics 2, 183-199.
Alexiadou, A. (1994) Issues in the Syntax of Adverbs, Ph.D, Dissertation, Universitiit Potsdam.
Alexiadou, A. (1996) Aspectual Restrictions on Word Order, Folia Linguistica XXX, 35-46.
Alexiadou, A. (1997) Adverb Placement: a Case Study in Antisymmetric Syntax, John Benjamins,
Amsterdam.
Alexiadou, A. & E. Anagnostopoulou (1995) SVO and EPP in Null Subject Languages and Germanic,
FAS Papers in Linguistics 4, 1-21.
Alexiadou, A & E. Anagnostopoulou (1997a) Covert Feature Movement and the Placement of
Arguments, to appear in The Minimalist Paramete: Papers Deriving from the Open Linguistics
Forum "Challenges ofMinimalism", John Benjamins, Amsterdam.
Alexiadou, A & E. Anagnostopoulou (1997b) Notes on ECM, Control and Raising, ZAS Papers in
Alexiadou, A & E. Anagnostopoulou (1997c) Toward a Uniform Account of Scrambling and Clitic
Doubling, in W. Abraham & E. van Gelderen (eds.) German: Syntactic Problems-Problematic
Syntax, Max Niemeyer Verlag, Tilbingen, pp. 143-161.
Alexiadou, A & E. Anagnostopoulou (1998) Parametrizing Agr: Word Order, V-Movement and EPPChecking, to appear in Natural Language and Linguistic Theory.
Anagnostopoulou, E. (1994) Clitic Dependencies in Modem Greek, Ph.D. Dissertation, Universitiit
Salzburg.Anagnostopoulou, E. (this volume) On Experiencers.
Barbosa, P. (1994) A new Look at the Null Subject Parameter, Paper presented at CONSOLE III,
Venice.
Belletti, A. (1988) The Case of Unaccusatives, Linguistic Inquiry 19, 1-34.
Belletti, A. (1990) Generalized Verb Movement: Aspects of Verb Syntax, Rosenberg & Selier, Torino.
Bobaljik, J.D. & A. Camie (1992) A Minimalist Approach to Some Problems of Irish Word Order,
unpublished manuscript MIT.
Bonet, E. (1990) Subjects in Catalan, MIT Working Papers in Linguistics 13, 1-26.
Borer, H. (1989) Anaphoric AGR, in O. Jaeggli & K. Safir (eds.) The Null Subject Parameter, Kluwer
Academic Publishers, Dordrecht, pp. 69-109.
Burzio, L. (1986) Italian Syntax, Kluwer Academic Publishers, Dordrecht.
64
ARTEMIS ALEXIADOU
Cardinaletti, A. (1995) Subjects and Clause Structure, to appear in L. Haegeman (ed.) The New
Comparative Syntax, Longman, London.
Cardinaletti, A & M. Starke (1995) The Typology of Structural Deficiency: On the three grammatical
classes, FAS Papers in Linguistics 1, I-55.
Carnie, A. (1993) Nominal Predicates and Absolutive Case Marking in Irish, MIT Working Papers in
Cinque, G. (1990) Types ofA-bar Dependencies, MIT Press, Cambridge, Mass.
Cinque, G. (1993) A Null Theory of Phrase and Compound Stress, Linguistic Inquiry 24, 239-297.
Comorovski, I. (1991) Partitives and the Definiteness Effect, Proceedings ofWCCFL 10, 91-102.
Deprez, V. & K. Hale (1986) Resumptive Pronouns in Irish, Proceedings of the 1985 Harvard Celtic
Colloquium.
Diesing, M. (1992) Indefinites, MIT Press, Cambridge, Mass.
Diesing, M. & E. Jelinek (1993) The Syntax and Semantics of Object Shift, Working Papers in
Scandinavian Syntax 51.
Drachman, G & S. Klidi (1992) The proper treatment of adverbial questions in Greek: the extended
minimal structure hypothesis, Studies in Greek Linguistics 13, 371-389.
Frampton, J. (1995) Some Current Issues in the Minimalist Program, paper presented at the Role
Economy Principles in Linguistic Theory Workshop in Berlin, 9-12.2.95.
Hoekstra, T. & R. Mulder (1990) Unergatives as Copular Verbs; Locational and Existential Predication,
The Linguistic Review 7,1-79.
Giannakidou, A. (this volume) Weak and Strong Polarity: Evidence from Greek.
de Hoop, H. (1992) Case Configuration and NP Interpretation, Ph.D. Thesis, University of Groningen.
Horrocks, G. (1994) Subjects and Configurationality in Journal of Linguistics 30, 81-109.
Johnson, K. (1991) Object Positions, Natural Language and Linguistic Theory 9,577-636
Jonas, D & J. Bobaljik (1993) Specs for Subjects, MIT Working Papers 18, 59-98.
Joseph, B. & I. Philippaki-Warburton. (1987) Modern Greek, Croom Helm, London.
Iatridou, S. (1990) About AgrP, LinguisticInquiry 21, 551-577.
Iatridou, S. (1993) On Nominative Case Assignment and a few Related Things, MIT Working Papers in
Iatridou, S & A. Kroch (1992) On the Licensing of CP-recursion and its relevance to the Germanic
Verb-second Phenomena, Working Papers in Scandinavian Syntax 50
Kayne, R. (1994) The Antisymmetry of Syntax, MIT Press, Cambridge.
Kiss, K. (1995) Introduction, in K. Kiss (ed.) Discourse Configurational Languages, Oxford University
Press, Oxfrod pp. 3-27.
Kiss, K. (1996) Two Subject Positions in English, The Linguistic Review 13, 119-142.
Koopman, & D. Sportiche (1991) The Position of Subjects, Lingua 85,211-285.
Kuroda, S.Y. (1972) The Categorical and the Thetic Judgement, Foundations of Language 9, 153-185.
Ladusaw, W. (1994) Thetic and Categorial, Stage and Individual, Weak and Strong, Proceedings of
SALT IV 220-229.
McCloskey, J. (1994) Subjects and Subject Position in Irish, unpublished manuscript, Santa Cruz.
McCloskey, J. (1996) On the Scope of Verb Movement in Irish, Natural Language and Linguistic
Theory 14, 47-104.
Meinunger, A. (1996) Discourse Dependent DP Deplacement, GAGL 39.
Moro, Andrea. 1997. The Raising of Predicates: Predicative Noun Phrases and the Theory of Clause
Structure, Cambridge University Press, Cambridge.
Motapanyane, V. (1994) An A-position for Romanian Subjects, Linguistic Inquiry 25, 729-734.
Ordoilez, F. (1994) Post-verbal Asymmetries in Spanish, paper presented at the Vienna GLOW
Colloquium.
Philippaki- Warburton, I. (1985) Word Order in Modern Greek, Tra,!sactions of the Philological Society
2, 113-143.
Philippaki-Warburton, I. (1989) Subjects in English and in Greek, unpublished manuscript, University
of Reading.
Philippaki-Warburton, I. & G. Catsimali (this volume) On Control in Greek.
65
Reinhart, T. (1995) Focus- the PF Interface, unpublished manuscript, University of Utrecht.

Rizzi, L. (1982) Issues in Italian Syntax, Foris Publications, Dordrecht.
Rivero, M-L. (1994) Verb Movement and the Structure of IP in the languages of the Balkans, Natural
Language and Linguistic Theory 12, 63-120.
Safir, K. (1985) Syntactic Chains, Cambridge University Press, Cambridge.
Selkirk, E. (1984) Phonology and Syntax: the Relation between Sound and Structure, MIT Press.,
Cambridge, Mass.
SoIa, J. (1992) Agreement and Subjects, Ph.D. Dissertation, Universitat Autonoma de Barcelona.
Tsimpli, I-M. (1990) The Clause Structure and Word Order of Modem Greek, UCL Working Papers in
Vallduvi, E. (1993) Information Packaging: a Survey, research paper, University of Edinburgh.
Varlokosta, S & N. Hornstein (1993) A Bound Pronoun in Modern Greek, Natural Language and
Linguistic Theory 11, 175-195.
Vikner, S. (1994) Scandinavian Object Shift and West Germanic Scrambling, in N. Corver and H. van
Riemsdijk (eds.), Studies on Scrambling, Mouton de Gruyter, Berlin, pp. 487-517.
Zubizarreta, M-L (1992) Word Order in Spanish and the Nature of Nominative Case, unpublished
manuscript, USC.
Zubizaretta, M-L. (1994) Grammatical Representation of Topic and Focus; Implications for the
Structure of the Clause, Cuadernos de Linguistica del/'U. Ortegay Gasset 2,181-208.
ON EXPERIENCERS i
ML T. & University of Tilburg
1. THE STATUS OF EXPERIENCERS AND THE RELEVANCE OF

GREEK
A central concern of linguistic theory has been to detennine the algorithm by virtue of
which the arguments of a given lexical item are projected syntactically. Most theorists
within Principles-and-Parameters-style frameworks have been assuming that this
mapping is mediated by a universal thematic hierarchy determining the ranking of
thematic roles according to a constraint as below (cf. Baker 1993:31 for discussion and
references):
(1) If the a-role of an argument X is higher than the a-role of a second
argument Y, then Xc-commands Y at the level ofD-structure
Psychological predicates present a challenge for theories of alignment of this sort
because as is well known, they systematically split into several sub-classes depending
on how they project the so-called experiencer argument. With verbs like fear the
experiencer is the subject; with frighten the experiencer is the object. This is quite
unexpected if the view that the mapping of argument structure to syntactic structure is
regulated by (1) is correct. In the recent literature, there are two approaches towards
the fear vs. frighten puzzle. According to the more traditional view, Experiencer
Subject (ES) and Experiencer Object (EO) predicates have identical theta grids
containing an experiencer and a theme. Assuming a version of the thematic hierarchy
as in (2), ES predicates can be said to preserve relative prominence. On the other hand,
EO predicates have an inverse structure, the reason being either syntactic (Stowell
1986, Belletti & Rizzi 1988, a.o.) or aspectual (cf. Grimshaw 1990 a.o.).
(2) Agent> Experiencer >GoallSourcelLocation >Theme
Altematively, it has been proposed that the fear and frighten classes do not have
identical theta roles and, therefore, the problem they pose for (1) is only apparent.
More specifically, EO-predicates of the frighten-class have been treated as causative
constructions licensing a nominative causer argument and an accusative causee. Under
this view, the experiencer/causee is thematically less prominent than the theme/causer.
67
@ 1999 Kluwer Academic Publishers.
68
ELENA ANAGNOSTOPOULOU
This is ensured by the more elaborated hierarchy exemplified in (3) (pesetsky

1992:52):
(3) Causer> Experiencer > TargetlSubject Matter
In this article, I will present evidence from Greek in favor of the former class of
approaches. Object experiencers of the frighten-class have a number of properties

which follow from the inverse-structure analysis. 2 More specifically, accusative
experiencers of the frighten class pattern with dative experiencers of the
piacere/appeal to-class in displaying subject-like behavior; they differ from structural
objects with respect to the properties of clitic doubling. I will conclude that frighten
EO-predicates lack an external argument due to the reasons discussed in Grimshaw
(1990).
2. ACCUSATIVE EXPERIENCERS BEHAVE LIKE DATIVES
2.1 Subject Properties of Dative Experiencers: Evidence

Like Italian (Belletti & Rizzi 1988), Greek has three classes of Experiencer Predicates.
Class 1 includes verbs like agapo (love), latrevo (adore), antipatho (dislike), miso
(hate) featuring an experiencer-subject and theme-object:
(4) 0 Petros
aghapai ta skilja
The Peter-nom loves the dogs-acc
'Peter likes dogs'
EO-predicates like anisixo (worry), provlimatizo (puzzle), enoxlo (bother), diaskedazo
(amuse), fovizo (frighten), endiafero (interest) form Class 2. The experiencer has
morphological accusative case; the theme/cause has nominative case and agrees with
the verb:
(5) TonPetro
ton
anisihi
The Peter-acc cl-acc worry-3sg
'The situation worries Peter'
i katastasi
the-situation-nom
Finally, there is one more class of constructions, corresponding to Italian piacerepredicates (Class 3). This includes expressions like aresi(1ike), fiei (bothers/matters)
selecting for a dative experiencer (pP as in (6a)or morphological genitive as in (6bW
and a nominative agreeing theme:
ON EXPERIENCERS
(6) a.
b.
To krasi
aresi
ston
The wine-nom like-3sg to-the
'Peter likes the wine'
To krasi
tu
aresi
tu
The wine-nom c1-dat 1ike-3sg the
-Peter likes the wine'
69
Petro
Peter
Petru
Peter-dat
Belletti and Rizzi (1988) and Masullo (1993) have argued for Italian and Spanish
respectively that the fronted datives in psych verb constructions qualify as quirky
subjects in the sense that they are not characterized by properties typically associated
with topics and left-dislocated constituents. 4 A number of arguments show that the
same point can be made for Greek dative experiencers as well:
(a) Word Order: the orders EXPdat-V-TH"om, TH"om-V-EXP dat are equally neutral, as
(7a) and (7b) show; in fact, (7a) is slightly preferred. This is in contrast with normal
(direct and indirect) object fronting, which feels like a c1itic left dislocation (CLLD, cf.
Cinque 1990, latridou 1991 a.o.).
(7) a.
b.
Tu Petru
tu
The Peter-dat c1-dat
To krasi
tu
The wine-nom c1-dat
aresi
to krasi
1ike-3sg the wine-nom
aresi
tu Petru
1ike-3sg the Peter-dat
(b) Internal CLLD is generally possible in Greek. However, in contexts where CLLD
is odd, presumably due to discourse factors, Experiencer Object (EO)-fronting is
felicitous:
(8) a.
b.
#Ta viv1ia
pu tu Janni
tu edhosa ine
#'The books that the John-dat c1 gave-lsg are
'The books I gave John are literature'
tu aresun ine ta
Ta vivlia pu tu Janni
The books that the John-dat c1 1ike-3p1 are the
'The books John likes are literature'
10ghotehnika
literary
10ghotehnika
literary
(c) A third asymmetry between CLLD and EO-fronting concerns bare quantifiers and
indefinites. There is a difference in acceptability between CLLD of such elements and
experiencer fronting: 5
(9) a.
??Kanenos
dhen
leo
efko1a
tu
??Nobody-dat neg
cl-dat say-lsg easily
ston dhjao10
to-the hell
'It is not easy for me to send somebody the hell'
na pai
subj go-3sg
70
b.
Kanenos
dhen
tu aresi
na ton stelnun ston dhjaolo
Nobody-dat neg
c1-dat like-3sg subj c1-acc send-3pl to-the hell
'Nobody likes to be sent to hell'
(d) Like subjects, experiencers are able to control PRO in absolute constructions. 6
(10)
a.
b.
c.
[AkughontasPROil*j tin istoria] 0 Petrosj arhise na tis perigrafi

[Hear-ing PROi/oj the story]
the Peterj started subjc1 describe
tis Mariasrdat tin embiria
tu
to Maryrdat the experience his
'While he was listening to the story/after he listened to the story, Peter
started describing his experience to Mary'
[Akughontas PROi/oj tin istoria] tis Mariasrdat 0 Petrosj arhise na tis
perighrafi tin embiria tu
[Akughontas PROjfOj tin istoria] tis Mariasj (0 Petrosj)
PROjfOj the story]
the Maryrdat (the Peterj-nom)
[Hear-ing
arhise na min
tis
aresi
(0 Petrosj)
started subj not
cl-dat-fem likej-3sg
(the Peterj)
'While she was listening to the story/after she listened to the story,
Mary started not liking Peter (lit: John started not appealing to
Mary)'
As indicated by (IOa,b), object control is not possible in absolute constructions,

indepenent of whether the object is left dislocated or not (compare (lOb) to (lOa.
(lOc) shows that dative experiencers pattern with subjects and nominative themes
pattern with objects in this respect. We thus have a more direct piece of evidence in
favor of the "subjecthood" of the former.
(e) "Ellipsis". 7 The paradigm in (11) indicates that in Greek coordination constructions,
a subject "gap" (or pro) in the second conjunct must have a subject and not a (direct or
indirect) object as its antecedent (compare (Ha) to (Hb) and cf. the indices in (Hc.
Switching the order of subject and object in the first conjunct as in (12) does not alter
the indexing possibilities; however, it does yield a slightly more awkward output:
(H)
a.
b.
0 Jannisj aghorase
to vivlioj ke ej
The John
bought
the book and e
'John bought the book and was pleased'
*0 Jannisj aghorase
to vivlioj ke ej
The John
bought
the book
and e
'John bought the book and pleased him'
harike
was pleased
t0I1.i evharistise
cl pleased
ON EXPERIENCERS
c.
(12)
a.
b.
71
0 Jannisj tis
milise tis Mariasj
ke ej/Oj
harike
cl-dat talked the Mary-dat
The John
ande
was
pleased
'John talked to Mary and was pleased'
*To vivlioj to aghorase
0
Jannisj ke ej
*Thebook cl bought
the John
ande
evharistise
tOIl;
cl
pleased
#Tis Mariasj
tis milise 0 Jannisj
ke ej xarike
#The-Mary-dat cl talked the-John-nom ande was pleased
We may conclude that structures of this sort require that the antecedent in conjunct-l
be syntactically parallel with the dependent in conjunct-2. Let us now tum to EOstructures:
(13)
a.
b.
c.
Tis Mariasj
tisj
aresun ta mathimatika
The-Mary-dat cl-dat like-3pl the mathematics-nom-pl
ke ej sihenetellatrevi-3sg tin musiki
and ej detest-/adore-3sg the music-acc-sg
'Mary likes mathematics and hates/adores music'
0
Petrosj
alla ejl*j
Tis Mariasj
tisj
aresi
The-Mary-dat cl-dat like-3pl the Peter-nom-pl but
e
sihenete-3sg tin Katerina
detest-3sg the Katerina-acc
'Mary likes Peter but hates Katerina'
0 Petrosj
tisj
aresi
tis Mariasj
ke e#jl"i
The-Peter-nom cl-dat like-3sg the-Mary-dat
and e
sihenetellatrevi tin Katerina
detest-/adore-3sg the-Katerina-acc
'Mary likes Peter and hates/adores Katerina'
(13a,b) show that a dative experiencer in the ftrst conjunct licenses a nominative
gap/pro in the second conjunct while a nominative theme does not. Moreover, altering
the relative order of experiencer and theme in the frrst conjunct, as in (13c), has the
same effect as switching the subject-object order in (12c). In this regard too, dative
experiencers pattern with subjects. s
(t) Pro vs. "aftosli". The following sentences can be viewed as evidence that preverbal
dative experiencers are in anA position: 9
(14)
a.
I Marillj aghapuse
ton Hari
prin
*afti/proj
The Mary loved
the Harry before *afti/pro
erotevti
ton Andrea
fall in love with Andreas
'Mary loved Harry before she fell in love with Andreas'
72
b.
c.
d.
Ton Hari;
ton
aghapuse
i Maria
prin
The-Harry-acc cl-acc loved
the Mary-nom before
aftos/proj erotefti
tin Katerina
aftos/pro fall in love with Katerina-acc
'Mary loved Harry before he fell in love with Katerina'
Tis Mariasj tis
milise 0 Haris
prin
afti/proj
The-Mary-dat cl-dat talked the-Harry-nom before afti/pro
prolavi na ton
apofiji
had time subj cl-acc avoid
'Harry talked to Mary before she had time to avoid him'
Tis Mariasj
tis
aresan ta mathimatika prin
The-Mary-dat cl-dat like-3pl the mathematics before
*afti/pro
anakalipsi tin musiki
*afti/pro
discovered the music
'Mary liked mathematics before she discovered music'
The full pronoun afti in subject position of the adjunct-temporal clause headed by prin
cannot corefer with the main clause-subject (14a) but it can corefer with a CLLDed
object (14b,c). Once again, experiencers pattern with subjects (14d). These facts can be
interpreted as follows. There is strong evidence that aftos/i is a demonstrative, thus
falling under Principle c. 1O (14a) under the relevant interpretation can accordingly be
ruled out as a Principle C violation (afti would be A-bound by the subject). On the
other hand, in (14b) aftos is A' bound by the CLLDed object and, therefore, the
coreference reading is permitted. 11 Given this reasoning, the fact that afti is excluded in
(14c) can be viewed as a strong argument that dative experiencers occupy an A
position (thus qualifying as subjects).12
(g) Finally, the fact that dative experiencers can act as binders for anaphors in Greek
(Anagnostopoulou 1993) can be viewed as one more argument for their subjecthood:
(15)
a.
b.
Tis Marias tis

aresi
0
eaftos
tis
The-Mary-dat cl-dat like-3sg the self-nom
her
'Mary likes herself
*Tis-Marias
tis
milai
0
eaftos tis
*The-Mary-dat cl-dat talk-3sg
the-self-nomher
'*Herself talks to Mary'
Note that Icelandic quirky subjects can also be binders for anaphors provided that the
latter are non-nominative (cf. Everaert 1990). Assuming that the ban on nominative
anaphors in certain languages is an independent restriction which doesn't have to do
with binding theory per se (cf. Everaert 1990), an explanation for the grarnmaticality
of (15a) would be that the dative is in a position that feeds binding. \3
ON EXPERIENCERS
73
2.2 Accusative Experiencers and Subjecthood

Interestingly, applying the tests developed in section 2.1. to Greek accusative
experiencers in psychological predicates where the nominative argument is interpreted
as a Theme/Cause (and not as an Agent) unequivocally shows that they also qualify as
"subjects":
(a') Word Order:

(16)
a.
b.
Ton Petro
ton
endhiaferun
ta mathimatika
The-Peter-acc cl-acc interest-3pl
the mathematics-nom
Ta mathimatika ton endhiaferun ton Petro
the Peter
The mathematics cl interest
'Mathematics interests Peter'
(b') Internal CLLD vs. EO-jronting:

(17)
a.
b.
#Ekini pu ton Petro ton

fovunte
#Those that the-Peter-acc cl-acc fear
'The ones that fear Peter are his students'
Eleino pu ton Petro ton
fovizi
That
that the-Peter-acc cl-acc frighten
'What frightens Peter is the future'
ine i mathites tu
are the students his
ine to mellon
is the future
(c') CLLD vs. EO-jronting of Quantifiers/Indefinites:

(18)
a.
b.
*?Kapjon tha ton vro -ime

siguri (alla den ksero pjon)
*Someone fut cl fmd-I-am sure (but I don't know whom)
'I'll find someone, I am sure (but I don't know whom)'
?Kapjon
tha ton anisihisun ta nea-ime
siguri
?Someone fut cl worry
the news-I-am sure
(alIa den ksero pjon)
(but I don't know whom)
'The news will bother someone, I am sure (but I don't know whom),
(d') Absolute Constructions:

(19)
a.
[Akughontas PROj tin istoria] 0 Petrosj arhise na antipathi

[Hear-ing PROj the story] the Peterj started disliking
Maryj
tinM~racc
b.
[Akughontas PROj tin istoria] tin Mariay_ -acc 0 Petrosj arhise

na tin antipathi
ELENA ANAGNOSTOPOUWU
74
c.
[Akughontas PROj tin istoria] tin Mariayacc (0 Petrosj) arhise

[Hear-ing PROj the story] the-Maryyacc the-Peterj started
na tin ghoitevi (0 Petrosj)
to her attract
'Having PRO heard the story, Mary started being attracted by Peter'
(e') Subject "ellipsis":

(20)
a.
b.
*Tin Maria; tin aghapai 0 Petros

*The-Maryj c~ loves the Peter-nom
ke ej antipathi
tin Katerina
the Katerina
and ej dislikes
'Peter loves Mary and dislikes Katerina'
Tin Maria; tin eknevrizun ta mathimatika
The-Maryj c~ irritate-3pl the mathematics
ke ej latrevi
ti musiki
and ej adore-3sg the music
'Mary gets irritated by mathematics and she loves music'
(f) Pro vs. "ajtosli" in "before"-clauses:

(21)
a.
b.
c.
I Maria; aghapuse
ton Hari
prin
The-Mary loved
the Harry before
*afti/proj
erotevti ton Andrea
*afti/pro
fall-in-Iove with Andreas
'Mary loved Harry before she fell in love with Andreas'
Tin Maria;
tin
idha
prinafii/proj
prolavi... .
The-Mary-acc c1-acc saw-1sg before afti/pro had time ... .
'I saw Mary before she had time to ..... '
Tin Maria; tin
eknevrizan ta mathimatika
The-Maryacc c1-acc irritated-3pl the mathematics
*afti/pro
prin
*afti/pro
anakalipsi tin musiki
she
before she
discovered the music
'Mary was irritated by mathematics before she discovered music'
(g') Anaphor Binding:

(22)
a.
Tin Maria
tin
provlimatizilenoxlilanisihi
The-Mary-acc c1-acc puzzleslbothers/worries
eaftos
tis
self-nom
her
'Mary is puzzled/botberedlworried with/atlby herself'
o
the
ON EXPERIENCERS
b.
75
*Tin Maria
den
tin
thavrnazi/aghapai 0 eaftos
cl-acc admiresllikes
the self
*The-Mary-acc not
'*Herself doesn't admirellike Mary'
tis
her
Summarizing, so far we have seen that there is no difference between dative EOs and
accusative EOs with respect to the "subjecthood"-criteria (a)-(g). In turn, this implies
that the experiencer argument of psychological predicates of Class 2 has quirky (i.e.
lexically specified, non-structural) accusative. The existence of case alternations as in
(23) further supports this claim:
(23)
a.
b.
TuPetru
tu
ponai to kefali
The-Peter-gen cl-gen aches the head
Ton Petro
ton
ponai to kefali
The-Peter-acc cl-acc aches the head-nom
'Peter's head aches'
It is not implausible to analyse these constructions as instances of dative (inalienable)

possessor-raising of the type recently discussed by Kayne (cf. Kayne 1993, Hoekstra
1994 and Mahajan 1994 among others).14 If this is correct, then the fact that the
possessor may have morphological accusative (23b) strongly supports the conclusion
that besides structural accusative, Greek also has an accusative with the properties of a
dative (oblique accusative).
2.3 Experiencers and Clitics

The fact that accusative experiencers in Greek differ from direct objects and behave
like dative objects can be established independently. In this section, I will present
evidence supporting both points.
(i) Accusative experiencers differ from direct objects. The properties of elitic doubling
in the constructions under consideration provide the main evidence for this. As
discussed in Anagnostopoulou (1994), Greek direct object (DO) clitic doubling is
characterized by two properties, namely it is optional and it is regulated by the
Prominence Condition. Accusative EO-doubling differs from simple DO-doubling in
both respects. The asymmetry between optional doubling of structural accusatives and
obligatory doubling of accusative experiencers lS is illustrated in (24):
(24)
a.
b.
0 Jannis (tin)
ghnorise
tin Maria se ena
TheJohn
(cl-acc) met
the-Mary-acc in a
'John met (her) Mary at a party'
0 Jannis ?*(tin)
endhiaferi tin Maria
The-John ?*(cl-acc) interests
the-Mary-acc
ap'ola
than everything
'John interests Mary more than anything else'
party
party
pano
more
76
Moreover, DO-doubling in Greek is felicitous only with anaphoric defmites, not with
"novel" or "accommodative" defmites (i.e. it is subject to the Prominence Condition,
cf. Anagnostopoulou 1994, for details). EO-doubling, on the other hand, violates the
Prominence Condition. This difference is exemplified in (25):
(25)
a.
b.
c.
Prin apo ligo kero eghrapsa mia vivliokrisia jia ena kenoufjo vivlio
pano sto clitic doublingj
'Some time ago, I reviewed a new book on clitic doubling'
sinandisa ton
sigrafeak se ena taksidhimu
#Arghotera ton
#Later on cl-acc met-I the author-acc
in a
trip
my
'Later on, I met him-the author during a trip of mine'
I kritiki mu ton
enoWise
ton sigrafeak
The criticism my cl-acc bothered
the-author-acc
toso
oste na paraponethi
ston
ekdhoti
such
that subj
complain
to-the editor
'My criticism bothered the author so much that he complained about
it to the editor'
As (25b) shows, doubling of the direct object ton sigrafea is infelicitous in a context
where the defmite may satisfy the Familiarity Condition only via accommodation (i.e.
linking of the index k of "the author" to the index i of "the new book on clitic doubling
that the speaker reviewed some time ago" which has already been introduced in the
fIle). The acceptability of (25c) in the same context indicates that EO-doubling is not
subject to this restriction.
(ii) Accusative experiencers behave like P-less dative objects. A well known property
of the double object construction in English and its counterparts crosslinguistically is
that 'shifted' datives cannot undergo relativization or other instances of what has been
traditionally analyzed as empty operator movement (cf. Larson 1988, Emonds 1993,
den Dikken 1995 a.o. for discussion):
(26)
a.
b.
*Billlikes the man that John gave a book

Bill likes the man that John gave the book to
Greek bare IO-datives (i.e. morphologically genitive NPs) pattern with English shifted
datives in this respect (cf. Stavrou 1984, Alexiadou & Anagnostopoulou to appear) for
discussion). (27a) is a null operator restrictive relative clause (RRC) in which the
extraction site is a bare dative, while (27b) is an instance of a wh-RRC in which a PPdative is moved pied-piping the overt preposition:
77
ON EXPERIENCERS
(27)
a.
b.
*Simbatho ton antbropo

pu 0 Petros
edhose
Like-lsg
the man-acc
that the Peter-nom gave
to vivlio
the-book-acc
'I like the man that Peter gave the book'
Simbatho ton antbropo
s-ton opio 0 Petros
Like-lsg
the man-acc
to-the which the-Peter-nom
edhose to vivlio
gave
the book-acc
'I like the man to whom Peter gave the book'
No such restriction applies to direct objects crosslinguistically, and Greek does not
present an exception to this:
(28)
a.
b.
I like the man that John met

pu sinantise
that met-3sg
Like-lsg
the man-acc
'I like the man that Peter met'
o Petros
the Peter-nom
Constructions like (27a) can be rescued if a dative clitic is present inside the RRC:
(29)

pu 0 Petros
Like-lsg
the-man-acc
that the Peter-nom
edhose to vivlio
gave
the book-acc
'I like the man that Peter gave him the book'
tu
c1-dat
Such a clitic is ungrammatical in RRCs whose 'head' functions as a direct object of the
embedded verb whenever the head of the RRC is definite (cf. Stavrou 1984, Alexiadou
& Anagnostopoulou to appear for discussion):16
(30)
pu ton
*Simpatho ton antbropo
Like-lsg
the-man-acc
that c1-acc
'I like the man that Peter met him'
sinandise
met-3sg
o Petros
the-Peter-nom
EO-relativization is identical to IO-relativization regardless of whether the experiencer
has dative or accusative case. The experiencer cannot undergo relativization unless a
clitic is present inside the RRC. Note, furthermore, that in all examples below the head
of the relative is a definite NP which, as previously mentioned, does not license clitics
when it is a direct object:
(31)
a.
0 antbropos pu *(tu)
aresi
Maria
The man
that (c1-dat) like-3sg the-Mary-nom
'The man that Mary appeals to is stupid'
ine ilithios
is stupid
78
b.
c.
0 anthropos pu *(ton) endhiaferi i

Maria
ine ilithios
The man
that (cl-acc) interests
the-Mary-nom is stupid
'The man that Mary interests is stupid'
0 anthropos pu *(ton) provlimatizun ta nea
The man
that (cl-acc) puzzle
the-news-nom
bike
mesa
came in
'The man that the news puzzles came in'
This pattern strongly supports the view that accusative experiencers are not licensed as
structural objects but rather as bare dative objects. Most of the accounts for the
ungrammaticality of examples like (26a) crucially rely on the presence of an empty
dative P in the structure of the double object construction (cf. Czepluch 1982, Kayne
1984, Baker 1988, den Dikken 1995). As is shown in Alexiadou & Anagnostopoulou
(to appear), this line of explanation can be extended to Greek (27a). In turn, this
implicates the presence of an empty dative P in all structures in (31 ).17
(iii) Agentive psych predicates and the direct-object properties of experiencers. It is

well known that most EO-predicates of Class 2 are systematically ambiguous between
agentive and psychological readings. This alternation may show up only when the
nominative argument is animate so that it can be interpreted as an agent. It is not a
priori evident that the agentive-non-agentive alternation is a matter of grammar rather
than, for instance, a pragmatic effect. The fact that backward binding is permitted only
with non-agentive readings (cf. Belletti and Rizzi 1988 a.o.) could be viewed as an
argument that the alternation is syntactically determined, but not necessarily so if
Reinhart & Reuland (1993) are right that in the contexts under discussion, the anaphors
are actually logophors whose distribution is determined by discourse factors like "point
of view". Greek provides strong evidence that the agentive alternation is a matter of
syntax. Consider fIrst (32):
(32)
I Maria
enohli ton Petro
The-Mary-nom bothers the-Peter-acc
'Mary bothers John'
Unlike English or Italian, (32) is unambigous in Greek. I Maria can only be

understood as somebody who bothers Peter deliberately; in other words, the
interpretation 'there is something about Mary that bothers Peter' is excluded. This is
due to the fact that the nominative argument is animate and the accusative argument is
undoubled. Accordingly, (33) is ruled out without doubling because the inanimate
nominative cannot be interpreted as an agent:
ON EXPERIENCERS
(33)
Ta epipla
*?(ton)
The-furniture-nom *?(cl-acc)
'The furniture bothers Peter'
79
enohlun ton Petro

bother the Peter-acc
Clitic doubling is permitted in a construction like (32) and the agentive interpretation
can be preserved. However, this doubling is subject to the Prominence Condition, as
(34) indicates:
(34)
a.
Prin apo ligo kero egrapsa mia vivliokrisia jia ena
b.
'Some time ago, I reviewed a new book on clitic doubling

#Epitidhes ton
enohlisa
ton sigrafea
Deliberately cl-acc annoyedlbothered-lsg
the author
kenourjo vivlio pano sto clitic doubling
Recall, furthermore, that the tests presented in section 2.3. which illustrate the
"subjecthood" of accusative experiencers concern the psychological and not the
agentive reading of Class 2-predicates.
Finally note that relativization of the accusative experiencer is fine when the
agentive interpretation of the nominative argument is forced:
(35)
anthropos pu eknevrise
Maria
aplos
that irritated-3sg
the-Mary-acc
simply
The man
ja
na dhi tis antidhrasis tu apodhihtike
poli
anektikos
in order subj see the reactions his proved-3sg
very
tolerant
'The man that Mary irritated just to see his reactions proved
to be very tolerant'
The conclusion I draw from all these facts is that experiencers have oblique accusative
case when the subject is a theme/cause while they have structural accusative when the
subject is an agent. Oblique accusative is manifested as obligatory doubling; structural
accusative permits doubling under conditions that have to do with anaphora resolution
in discourse (degrees of familiarity). Whether or not oblique accusative and dative
experiencers are subjects or objects is an unclear and theory-dependent issue. We saw
a number of arguments in support of their "subjecthood". However, we also pointed
out that these arguments are not as strong as the ones that have been presented for
Icelandic quirky subjects.
Interestingly, it has been reported about Spanish that "...many dialects of Spanish
do not have a clear cut between the second class and the third class. A significant
number of speakers, on the other hand, do not acknowledge any Experiencer argument
bearing the accusative Case. Intriguingly, there are some dialects of Spanish, especially
from the Southern Cone of Latin America and some areas of Spain, in which
homophonous forms of the Experiencer verbs allow an alternation accusative-dative in
the Case marking ofExperiencer arguments ... " (Franco 1990:46 citing Jaeggli 1984):
80
(36)
a.
b.
Ese
tipo de commentarios Ie
That
type of comments
cl-dat
'That type of comments anger Juan'
enojo a Juan
Maria 10
Maria cl-acc anger to Juan
'Maria angered Juan'
enojan a Juan
anger to John
A Juan has dative when the subject is a theme/cause while it has accusative when the
subject is interpreted as an agent. This is indicated by the case marking of the doubling
clitics in (36a) and (36b) respectively. What I am suggesting here is that even though,
superficially, Greek looks like Italian it is actually more like Spanish. Despite the fact
that there is a morphological distiction between Class 2 and Class 3 in Greek, there is
no corresponding syntactic distinction between the two classes. Moreover, the Spanish
alternation in case marking of the experiencer depending upon the agentivity of the
subject also exists in Greek, but it manifests itself in the form of clitic doubling. 18
3. TOWARD AN ANALYSIS
3.1 Goals
We are now in a position to address the problem posed by Class 2 EO-predicates,
which was briefly presented in section 1, on the basis of the Greek facts. 19 What we
need is an analysis that will be able to capture the generalizations presented in the
previous sections without making it impossible to also account for the properties of, for
instance, English and Italian Class 2-verbs which are clearly different. The Greek facts
that call for an explanation are the following: (a) The similarity between Class 2 and
Class 3 with respect to (i) the quirky subjecthood of the experiencer argument and/or
(ii) the fact that the experiencer argument has oblique case. (b) The agentive-non
agentive alternation which systematically characterizes Class 2-predicates, in
particular the fact that this alternation is visible through the different morphological
case and structural properties of the experiencer. I will argue that an elegant way to
reconcile the (a)-facts with the (b)-facts is by adopting the analysis of Class 2predicates put forward in Grimshaw (1990). I will assume that oblique accusative and
dative experiencers are indeed subjects, in order to sharpen the argument. But my
results would be essentially the same even if they weren't.
The theory of case I will be assuming is essentially the one developed in Marantz
(1991). According to Marantz, the relation between abstract Case, and case and
morphological case and agreement is at best indirect: quirky subjects in Icelandic
strongly argue for a separation of licensing and morphological case realization.
Nominal arguments are licensed by (extended) projection, not by case or
morphological properties. Case and agreement merely interpret syntactic structure;
ON EXPERIENCERS
81
they do not drive syntax. Case realization takes place at Morphological Structure,
which is part of the PF branch of the syntax; it obeys the disjunctive hierarchy in (37):
(37)
case realization disjunctive hierarchy
lexically governed case

dependent case (accusative or ergative)
unmarked case (environment-sensitive: NOM in IPs, GEN in DPs)
default case (when no other case realization principle is available)
The notion "dependent case" is defined as follows:
(38)
Dependent case is assigned by V+I to a position governed by V+I when

a distinct position governed by V+I is:
i. not "marked" (not part of a chain governed by a lexical case determiner)
ii. distinct from the chain being assigned dependent case
Burzio's Generalisation effects are partly reduced to the EPP and partly to the
conditions under which dependent case can be assigned. (38i) prevents ACC case on
an object when the subject has lexically specified case. I will follow Marantz in
assuming that the EPP and Case/Agr-checking should be kept apart. I will furthermore
assume that dative case is lexically specified/oblique case and that lexically specified
case is case governed by V. If an oblique argument moves overtly to a position in the
functional domain this must be due to the EPP.
3.2 The Paradox
-----
Belletti and Rizzi (1988) have proposed the following D-structure for EO-predicates:
(39)
NP
VP
V'
/'"'--.....
ec
I
preoccupa
NP
...............
NP
I
questo
Gianni[+ inherent Accusative)

a Gianni
piace
According to (39), the only difference between Class 2 and Class 3 is the case marking
of the experiencer argument: inherent accusative vs. dative. Being unaccusative,
experiencer verbs cannot assign structural case to the theme argument; therefore, the
latter moves to the subject position. Instrumental for this analysis are three
assumptions:
82
(40)
(41)
(42)
Case is assigned to the object iff a a-role is assigned to the subject

(Burzio 1986)
V is a structural Case assigner iff it has an external argument (B&R 1988)
"Inherent Accusative will be specified in the Case grids of these Verbs as
related to the Experiencer slot in the a-grid, so that the Theme will be
skiped and will have to move" (B&R 1988: 333)
(40) is Burzio's Generalisation (BG) as originally stated, (41) is Belletti and Rizzi's
(B&R) interpretation of BG, while (42) shows why it is crucial for B&R that the
experiencer have inherent accusative. This analysis implies that the agentive-non
agentive alternation is a matter of syntax: when the subject is an agent, it qualifies as
an external argument, hence the experiencer may be assigned structural accusative.
There is no doubt that Greek strongly supports B&R's approach. We saw ample
evidence that there are no real asymmetries between Class 2 and Class 3; we also saw
that accusative experiencers do not behave like structural objects under the
psychological reading while they do behave like structural objects under the agentive
reading. This is precisely what B&R's analysis predicts. Furthermore, the subjecthood
of accusative experiencers is expected under a theory of quirky case according to
which, (i) arguments with lexically specified case may satisfy the EPP (Marantz 1991,
cf. above) and (ii) these arguments are structurally higher than the ones having
nominative case as in Schiitze (1993). In turn, this implies that the experiencer
argument is thematically more prominent than the theme/cause given the algorithm in
(1). Finally, the fact that accusative experiencers never qualify as subjects under the
agentive reading is fully compatible with the observational generalization that agentive
predicates never license quirky subjects.
Nevertheless, there are at least two problems for this analysis. 20 One problem has
to do with languages like Italian in which datives behave like subjects but accusatives
don't. If the language permits quirky subjects as a parametric option, and if B&R are
correct that accusative experiencers have lexically specified case in Italian, then it is
not clear why datives may surface as subjects while inherent accusatives may not. If,
on the other hand, the accusative in question is structural (or dependent) and the
difference between Greek and Italian reduces to the fact that Greek has lexical
accusative while Italian doesn't, then this strongly reduces the conceptual basis on
which B&R's analysis relies. BG is simply not correct on this point and the Greek data
are less revealing than was originally suggested. A more severe problem has to do with
the very existence of the agentive-non agentive alternation. Under B&R's approach, it
is not clear why this altemation systematically characterizes Class 2 but not Class 1/3.
But there is a related asymmetry between Class 2 and Class 113: the former can be
eventive, the latter are inherently stative. This is clear in English where Class 1 verbs
never occur in the progressive while Class 2 verbs do (cf. Dowty 1979:55, Grimshaw
1990:23):21
ON EXPERIENCERS
(43)
a.
b.
83
*We were fearing the stonn

The stonn was frightening us
This difference, which correlates with the intuition that the theme can be interpreted as
causer in the case of Class 2-verbs, has no fonnal status within B&R's analysis.
A causative analysis is superior on precisely this point: the correlation between
agentivity, causative interpretation and eventive predication can be expressed.
According to standard assumptions (cf. Alsina 1992), causative verbs are complex
predicates consisting of a functor CAUSE which takes two arguments: a causer and
the caused event which is brought about by the causation:
(44)
CAUSE<ag!causer PRED<.....
Treating Class 2-EO predicates as causative is compatible with the fact that (45a) can
be paraphrased as (45b) (cf. Grimshaw 1990:22):
(45)
a.
b.
The stonn frightened us

The stonn caused us to experience fear
In turn, this analysis provides a natural explanation for the fact that they can be
agentive: the caused (emotional) event/state is brought about by either a volitionar
agent or a non-volitionar causer. On the other hand, however, (44) cannot adequately
characterize the other properties of Class 2 EO-predicates in Greek. The reason is that
under this representation, the theme/cause is higher than the experiencer: the fonner is
an argument of the higher CAUSE predicate, the latter is an argument of the predicate
embedded in the caused event. Moreover, if (44) was correct there should be no
syntactic asymmetries between agentive and non-agentive readings because in both
cases, the nominative argument would qualify as the external causer.
Suppose that we adopted instead Pesetsky's (1995) analysis according to which,
Class 2 EO-predicates involve affixation of an empty preposition CAUS to a bound
root like "annoy followed by subsequent movement of the NP-argument of CAUS to
the subject position for Case reasons (46):
(46)
V'
/ \
V
PP
"annoy / \
Exp P'
/ \
DP
CAUS Causer
This is not unproblematic either. First of all, it is not clear why the experiencer in
direct object position has oblique accusative and it may surface as a subject, especially
since the causer qualifies as a derived external argument, according to Pesetsky's
84
tenninology. A more complex problem arises in connection with agentive readings. It

is not clear how these should be analysed. If it is correct that annoy-type verbs are
based on bound roots which cannot occur by themselves, then it is reasonable to
assume that their agentive variants involve the same type of aftixation. This is
supported by the observation that the Target/Subject Matter restriction applies when
the nominative is interpreted as an agent:
(47)
*John angered Mary at the government
Given Pesetsky's (1995) reasoning, the ungrammaticality of (47) seems to suggest that
agentive readings involve CAUS affIxation as well. In other words, the problem posed
by (44) can be reproduced for (46): treating Class 2 EO-predicates as causatives leads
to the expectation that the agentive alternation is not part of their grammar but rather it
has to do with pragmatic inferencing. Yet Greek shows that this is not correct. The
altemative, namely that agentive readings involve an external agent while
psychological readings arise from a process like the one depicted in (46), is not
unlikely. However it would lead us to do away with the assumption that the surface
form annoy is invariably based on Vannoy and that the Target/Subject Matter
restriction reduces to the HMC. But this would amount to abandoning essential
ingredients of Pesetsky's analysis.22
Summarizing, neither of the alternatives we have considered can accommodate the
Greek facts. What we have so far is a paradox. 23 On the one hand, an unaccusative
analysis is needed to account for quirky subjecthood, oblique accusative and the
syntactic and moxphological asymmetries between agentive and non-agentive
structures in languages like Greek. On the other hand, a causative analysis may
provide a more satisfactory explanation for the emergence of agentive structures with
Class 2 verbs crosslinguistically as well as the syntactic and moxphological properties
of Class 2 verbs in languages like English and Italian. As we shall see next, this
paradox is resolved if the EO-problem is treated on the basis of two hierarchies, a
thematic and an aspectual hierarchy.
3.3 An Articulated A(rgument)-Structure Approach

Grimshaw (1990) proposes that the exceptional properties of Class 2-verbs can be
dealt with once we acknowledge that argument structure has internal organization
according to (i) a thematic hierarchy and (ii) an aspectual hierarchy. The two
dimensions do not always coincide, i.e. an argument of a predicate can be the most
prominent in the thematic hierarchy while another argument of this predicate is most
prominent in the aspectual hierarchy. External arguments are defined as arguments
that are maximally prominent in both dimensions and they are distinguished from dstructure subjects. An argument can be a d-structure subject without necessarily
ON EXPERIENCERS
85
qualifying as an external subject. Phenomena such as the causative/anticausative

alternation, nominalization and passivization affect external subjects rather than dstructure subjects. D-structure realization of arguments is determined by the aspectual
hierarchy rather than the thematic hierarchy in situations where the two are in conflict.
On the other hand, quirky argument realization can occur only when the thematically
most prominent argument is not also the most prominent in the aspectual dimension.
Consider now the analysis Grimshaw provides for Class 2 verbs in English. These
verbs have an Experiencer argument and a Theme argument. Since the Theme role is
lower in the thematic hierarchy than the Experiencer role, the th-prominence relations
are as in (48) with the Experiencer as the most prominent argument and the Theme as
the less prominent:
(48)
(x (y))
ExpTheme
However, Class 2 verbs are also causative, as already shown. The causative character
of these predicates is further supported by the fact that they can license causal adjuncts
(50), a property limited to causative predicates (49):
(49)
(50)
a.
b.
a.
b.
John killed Bill by his foolish action

*Bill died by his foolish action
?The storm frightened the children by its noise
*The children feared the storm by its noise
Recall the representation of causative predicates in (44) repeated here as (51a):

(51)
a.
b.
CAUSE<ag/causerPRED<.....
x killed yl xfrightened y
The causer argument John or the storm is involved in an activity, x engages in

breaking or frightening, and the resulting state is one in which y is broken or
frightened. The causal analysis is translated in event-structure terms as follows: the
causer is associated with the first sub-event (activity) which is causally related to the
second sub-event (the psychological state of being frightened in the case offrighten).
Grimshaw defines aspectual prominence as follows:
(52)
"..An argument which participates in the frrst sub-event in an event

structure is more prominent than an argument which participates in the
second subevent. ..{Grimshaw 1990: 26) Or, alternatively, ... an argument
associated only with the frrst event counts as more prominent than an
argument associated with both subevents"
According to this deftnition, the causer argument will always qualify as more
prominent than the causee.
86
Grimshaw's analysis makes a number of predictions for psychological predicates

of the frighten-class.
(a) Psychological causative predicates will not have an external argument while
agentive psychological causatives will have an external argument. In the former case,
neither argument has maximal prominence, in the latter case the Agent argument has
maximal prominence:
(53)
a.
Psychological causative b. Agentive Psychological causative

(Exp (Theme
(Agent (Exp
2 1
1 2
(b) (53a) presents a mismatch. The Exp is thematically more prominent, while the
Theme is aspectually more prominent. Neither will qualify as an external argument.

However, the theme will be realized as a D-structure subject because the aspectual
dimension determines the linking of arguments. Hence the D-structure representation
will be as in (54):
(54)
VP
/'-...
Theme
V
V'
Experiencer
No such mismatch is present in (53b) where the Agent qualifies as maximally

prominent in both dimensions. The D-structure representation of (53b) will be as in
(55):
(55)
VP
/"'-...
Agent
V
V'
Experiencer
Thus, there is no difference in D-structure representations between psychological

causatives and agentive psychological causatives. This accounts for the syntactic and
morphological properties of Class 2 verbs in Italian and English.
(c) Assuming that quirky argument realization can occur when the thematically most
prominent argument is not the most prominent aspectually, leads to the prediction that
in (53a) the Exp can be quirky while in (53b) the Agent cannot. For Grimshaw, a
quirky argument will not be specified by the aspectual analysis. Thus, we are led to the
following a-structure representation:
ON EXPERIENCERS
(56)
a.
Psychological causative b.
(Exp (Theme))
(1 )
87
Agentive Psychological causative

(Agent (Exp))
I propose that this is the situation in Greek. The Exp is lexically specified or quirky in
(56a) and thus it can satisfy the EPP. The Theme cannot bear accusative by virtue of
Marantz's defmition of dependent case in (38). Hence, it will be realized as nominative
(environment sensitive or default case). On the other hand, in (56b) the agent must be
specified by the aspectual analysis because it is thematically and aspectually the most
prominent argument. The Exp can be assigned dependent case because the agent
cannot have lexically specified case.
Treating Class 2 predicates as semantically causative naturally accounts for the
fact they can be agentive. This is a general property of causative verbs. Admitting that
they are psychological means that they select an experiencer, i.e. an argument which
experiences an emotion. In the case of Class 2 this emotion is viewed as a state
resulting from an activity; Class 1 and Class 2 are not complex events, they represent
simplex states. This describes correctly what the three classes have in common and
where they differ. Being stative, Class 1 and Class 3 verbs cannot have an agent
because agentivity is a property associated with dynamic predicates. Treating the
Greek problem as an a-structure problem along the lines of Grimshaw gives us the
facts correctly. This does not exclude the possibility that in some languages, Class 2
verbs are syntactically causative, as seems to be the case in Georgian (cf. Nash 1994).
This is one advantage of this proposal. An important consequence of a more articulated
theory of a-structure representation is that it permits a solid characterization of the
notion external argument. Moreover, this theory permits a defmition of quirky subject
realization which captures the generalization that agents are never quirky subjects.
Grimshaw draws a distinction between external argument, which is an a-structure
notion, and d-structure subject, which is a configurational notion determined by
aspectual rather than thematic prominence. In this manner, she basically rejects (1).
This is in line with much current research which seeks to replace or reduce the
thematic hierarchy by capitalizing on event structure instead.
The difference between this analysis and a causative alternative along the lines of
Pesetsky lies in the way the agentive alternation is treated. Within the two-dimensional
approach, it is expected that experiencers may surface as quirky subjects when the
nominative argument is a theme/cause but not when it is an agent. The similarity
between Class 2 psychological causatives and unaccusatives is expressed in terms of
the assumption that neither of the two has an external argument. Within the simple
causative approach, the asymmetry concerning the realization of experiencers cannot
be expressed as straightforwardly.
Whether or not it is correct that themes are projected as subjects is an issue
requiring further investigation. One could attempt to derive this result from a more
primitive structure reflecting thematic prominence by incorporating into the syntactic
representation an additional layer of structure (cf. Bennis 1993).
88
As for the D-structure representation of Greek Class 2-predicates, this depends on

our overall theory concerning the way quirky subjects start out. One way of
interpreting Grimshaw's proposal is that skipping the Exp in the aspectual analysis
permits a situation in which the Exp rather than the theme is projected as a subject.
This would be in line with Schiitze's (1993) treatment of quirky subjects. Ambiguous
linking would be in principle possible only if there is no external argument. The option
will be taken only if the realization of the thematically most prominent argument can
be lexically specified because the language allows it: Greek would permit lexical
specification in this case, Italian would not.
4. CONCLUSION
To conclude, in this paper, I have presented evidence from Greek that the a-role of the
experiencer argument of frighten-predicates is ranked higher than the a-role of the
theme/cause in the thematic hierarchy. In addition, I have argued that the properties of
Greek EO-predicates strongly support the view advocated in Grimshaw (1990) that the
mapping between thematic structure and syntactic structure is not as straightforward as
hypotheses like the Universal Thematic Hierarchy (cf. Baker 1988) stipulate. In
situations where thematic prominence is in conflict with aspectual prominence both
dimensions playa significant role. The realization of bare DP arguments is determined
by the aspectual dimension; on the other hand, there is a close connection between
oblique argument realization and thematic prominence.
NOTES
An earlier version of this paper was presented at the Workshop on Greek Syntax which was held in
Berlin (December 1994). I would like to thank the participants for their comments and the organizer Artemis
Alexiadou for the great atmosphere she created. I have benefited from discussions with and comments from
Artemis Alexiadou, Hans Bennis, Hans Broekhuis, Norbert Corver, Alexis Dimitriadis, Martin Everaert,
Anastasia Giannakidou, Riny Huybregts, Sabine Iatridou, Sila Klidi, Paul Law, Josep Quer, Henk van
Riernsdijk, Dimitra Theophanopoulou-Kontou, Chris Wilder and two anonymous reviewers.
The version presented here follows very closely the 1994 presentation. Later research (Alexiadou and
Anagnostopoulou 1996a, 1996b, Anagnostopoulou and Everaert 1996a, I 996b) has shown that some of the
facts discussed here could receive an alternative explanation. However, I believe that the main point of the
article can be maintained and for this reason, I have not changed anything substantial from the original
paper.
2
Interestingly enough, the conclusion reached here for Greek is in agreement with Hornstein and
Varlokosta (1992), though independently motivated. Bennis (1993) investigates psychological adjectives in
Dutch and pursues a line of argumentation very similar to the one presented here for Greek (cf. below).
3
Most datives in Greek can either be PPs or have motphological genitive case. Markantonatou (1994)
and Anagnostopoulou (in progress) demonstrate that in ditransitives, this alternation semantically and
syntactically corresponds to the dative shift alternation in English. In the text, I am glossing this genitive as
ON EXPERIENCERS
89
'dative'.
It should be kept in mind that "quirky subjects" in Italian, Spanish and Greek differ from quirky
subjects in Icelandic. Belletti and Rizzi (1988) are aware of this fact which Masullo (1993) attempts to
provide an account for. It would lead us too far afield to address this issue here (cf. Anagnostopoulou &
Everaert 1996b for some further discussion).
5
This contrast is not sharp in Greek. Belletti & Rizzi (1988:338) present a clearer contrast in Italian:
(i) a. *A nessuno gli hanno detto di andare al diavolo
To nobody to him they said to go to hell
b. ?A nessuno gli piace esser mandato al diavolo
To nobody to him pleases to be sent to hell
This seems to be related to a more general difference between Greek and Italian CLLD. As discussed in
Cinque (1990:15), CLLD of bare quantifiers in Italian does not require a resumptive clitic:
(ii) Qualcuno, (10) troveremo
someone we (him) will find
When the clitic is present the quantifier is interpreted as specific, when the clitic is absent the quantifier is
interpreted as non-specific (cf. also Dobrovie-Sorin 1992 for Romanian).
These facts do not seem to hold in Greek. Bare quantifiers either undergo CLLD, in which case a clitic
is required and the quantifiers are generally interpreted as specific, or they undergo focus-movement, in
which case the clitic is necessarily absent (this is always the case with focus-movement in Greek, cf. Tsimpli
1995 for discussion and references):
(iii)
Kapjon i Maria *(ton) epjase
na antighrafi
Someone, the Mary *(him) found-3sg to cheat
'Mary found somebody cheating'
b. *Kapjon tha vrume (alIa den kserume pjon)
Someone will find-we (but we don't know whom)
'We'll find someone (but we don't know who)'
c. KAPJON tha vroume (alIa den kseroume pjon)
It seems to me that (iiib) improves considerably when the verb receives heavy stress for reasons that I don't
understand. More research on the topic is needed.
6
Note that the examples below contain the verb arhizo 'start' which on the basis of anaphora facts can
be shown to be a raising verb in Greek. The fact that arhizo agrees with the nominative therne argument
shows that the theme qualifies as a subject with respect to raising. On the other hand, the fact that the dative
experiencer argument acts as a controller shows that the experiencer qualifies as a subject w.r.t. control.
Hence, it is not easy to decide which of the two arguments is the subject in these constructions (cf. Campbell
and Martin 1989 for a proposal).
7
I would like to thank Chris Wilder (p.c.) for discussions and suggestions concerning the form of the
argument.
s An anonymous reviewer points out that the Italian counterparts of (13) seem systematically worse. The
reviewer suggests, furthermore, that this entails a scale of 'quirkiness' that would place Greek between
Icelandic and Italian.
9
lowe the interpretation of the facts and the formulation of the argument to Sabine latridou (p.c.).
10
The evidence that aftos is a demonstrative rather than a personal pronoun is the following:
a. Aftos can be a demonstrative, as shown in (i):
(i) Tha pam afto to molivi
ke ohi ekino
rut take this the pencil
and not that one
'I will take this pencil and not that one'
b. Moreover, aftos is characterized by the properties which regularly differentiate demonstratives from
personal pronouns (cf. Cardinaletti & Starke 1995). Specifically:
-It may refer to non-human entities in contexts requiring strong forms while personal pronouns cannot:
(ii) a.
Tha pam
mono
afto
fut take-l sg only
this-neut-sg
'I will take only this'
4
90
b.
Mu
aresi
afti
(. .... ke oxi i alii karekla)
CI-dat.
like-3sg this-fem-sg
(.....and not the other chair-fern)
'I like only this
(....and not the other chair)'
-Coordinated personal pronouns cannot refer to non-human entities. No such restriction applies to
demonstratives. Aftos qualifies as a demonstrative according to this property. It doesn't have to refer to
humans under coordination:
(iii) Tha paro afto ke ohi ekino
fut take this and not that one
'I will take this and not that one'
-Demonstrative systems typically make spatial distinctions of the near/far type while personal pronouns do
not. Also in this respect afto qualifies as a demonstrative: in (iii) afto implies lack of distance (i.e. 'near') and
ekino implies distance (i.e. 'far').
-Finally, demonstratives are subject to Principle C while personal pronouns are only subject to Principle B.
Aftos must be disjoint from any c-commanding antecedent:
(iv) *1 Maria; ipe oti (aftij)tha erthi
(aftij)
The Mary said that (she) fut come
(she)
'Mary said that she will come
11
Note that non-dislocated objects can also co-refer with aftos; we would have to assume that the
temporal clause is attached higher than the object, thus treating these cases as instances of pragmatic
coreference. For discussion of Principle C in connection with the conditions under which [+R] name-like
elements can be A' bound, cf. Lasnik and Stowell (1991). A potential objection to the analysis presented
here might be that the phenomenon illustrated in (14) is a problem which can be more appropriately handled
in terms of anaphora resolution in discourse. Given that aftos is a demonstrative and thus, a medium
accessibility marker in the sense of Ariel (1990) it cannot have a sentence topic as its antecedent. For this
reason, the coreference option is excluded when the antecedent is a subject (a "topic" in inforrnationstructure terms) while it is not excluded when the antecedent is an object (non-topic). The relative
distribution of pro could be accounted for along similar lines (pro is a high accessibility marker thus being
selected in the appropriate contexts, cf. also Montalbetti's facts). Even if this explanation is correct, the main
point of the text still remains, namely that dative experiencers and subjects pattern alike. However, (14b)
would be problematic for such an approach if it is correct (as is standardly assumed) that CLLDed elements
are topics.
12
An anonomous reviewer points out that this account leads to the incorrect prediction that
demonstratives can be A' bound by CLLDed elements. According to the reviewer, this would be surprising
in view of the fact that the resumptive c1itic c-commands the demonstratives thus violating principle C and
it is also empirically false in Romance languages where a CLLDed element cannot bind a demonstrative it
c-commands.
There is an easy way to test the validity of the argument for Greek. The element ekinos is clearly a
demonstrative (cf. note 10) and it can be A'bound by a CLLDed element:
(i) Tis Mariasj tis
milise
0
Haris
prin
ekini/proj prolavi
The-Mary-gen c1-gen
talked
the- Harry-nom before
ekini/pro had time
na ton
apofiji
subj him
avoid
'Harry talked to Mary before she had time to avoid him'
This suggests that the temporal clause is attached higher than the c1itic.
13
However, cf. Anagnostopoulou & Everaert (1 996a,b) for an alternative.
14
Cf. Campbell and Martin (1989) for arguments that intransitive sensation predicates like "aches" are
unaccusative.
15
Under the psych interpretation (cf. below). Note, furthermore, that c1itic doubling of dative
experiencers is also obligatory.
16
When the head of the RRC is indefinite the clitic may be present:
ON EXPERIENCERS
(i)
91
Dhiavasa ena vivliopu (to)

pira
apo tin vivliothiki
Read-Isg a
book that(cl-acc) got-Isg
from the library
"I read a book that I got (it) from the library"
17
Alexiadou & Anagnostopoulou (to appear) follow den Dikken's (1995) explanation for (26a) which
relies on the impossibility of pied-piping in the case of empty operator movement in combination with the
ban of sub-extraction from a left branch after dative shift has taken place. They argue that the presence of a
dative clitic serves to license the empty preposition in situ so that dative shift does not have to take place
and the problem of sub-extraction from a left branch does not arise. It is not entirely obvious that this line of
explanation can be extended to EO-constructions of the frighten class. It might tum out that Baker's
(1988:299) Non-Oblique Trace Filter is more promising.
18
It is interesting to note that Macedonian Greek makes use of morphological accusative instead of
morphological genitive in P-Iess dative constructions. It is even more interesting that clitic doubling is
obligatory in such constructions in Macedonian Greek while it is much more free with P-Iess datives of
standard Greek which have morphological genitive case (cf. Dimitriadis this voluroe).
19
Although Greek doesn't have any "safe" tests for unaccusativity, I will accept the arguments from the
literature (Belletti and Rizzi 1988, Pesetsky 1995 a.o.) that Class 3 EO-predicates are unaccusative. I believe
that a promising way of treating the alternation between Class I and Class 3 would be to apply a "possessor
raising"-style analysis to them (cf. Kayne 1993 a.o.). Given that the semantics of these two classes are
identical, such a treatment would do away with the stipulation that for soroe arbitrary reason, the
Experiencer is the external argument of Class I-verbs while it is an internal argument of Class 3-verbs, a
stipulation which seems otherwise unmotivated. Though conceptually attractive, such an approach would
have to deal with the problem that Class I-verbs may undergo passivization.
20
Cf. Pesetsky (1995) for an extensive criticism ofBelletti & Rizzi's proposal.
21
The progressive test does not apply in Greek. However, there are other tests distinguishing Class 113 as
from Class 2 with respect to stativity (cf. Dowty 1979:55,60):
(a) Class 2 verbs can occur as complements offorce, persuade, Class 113 verbs cannot:
(i).
0 Jannis epise
tin Maria na fovisi ton Petro
The John persuaded the Mary to frighten the Peter
'John persuaded Mary to frighten Peter'
#0 Jannis epise tin Maria na fovithi ton Petro
The John persuaded the Mary to fear the Peter
'John persuaded Mary to fear Peter'
#0 Jannis epise
tin Maria na tis
aresi 0 Petros
The John persuaded the Mary subj to-her Iike-3sg the Peter-nom
'John persuaded Mary to appeal-to her Peter'
(b) Class 2 verbs can co-occurwith deliberately (cf. examples 34, 35 in the text) Class 113 verbs cannot:
(ii) a. #0 Jannis epitidhes
aghapai tin Maria
The Jannis deliberately loves the Mary
'John deliberately loves Mary'
b. #Tu Janni
epitidhes tu arese to krasi
The Jannis-dat deliberately cl-dat Iike-3sg the wine-dat
'John deliberately likes the wine'
(c) Class 2 verbs can occur as imperatives while Class 113 verbs cannot:
(iii) a. #Andipathise ton Petro!
Dislike-2sg the-Peter-acc
'Dislike Peter!'
b. Fovise
ton Petro!
Frighten-2sg the-Peter-acc
'Frighten Peter!
22
In Pesetsky (1995), the issue of agentive readings is addressed in footnote 179. Pesetsky does
acknowledge the necessity of distinguishing between agentive and non-agentive readings, the reason being
mainly that backward binding with agentive uses of causative Experlencer verbs is less acceptable. His
92
solution to the problem is to assume that "... the upper occurrence of Causer may bear agentive features that
the lower occurrence fails to bear. [.... J Binding might not be evaluable in a "reconstructed" position where
the semantics of the phrase when it occupied that position were in some fashion less complete than the
semantics of the same phrase in its surface position." (pesetsky 1995: 321).
23
Bennis (1993) examines psychological adjectives in Dutch and shows that they can either be simplex,
i.e. the Experiencer is the subject, or complex, i.e. the Theme is the subject. The complex cases present a
similar paradox: they are unaccusative with respect to their argument structure and transitive with respect to
their syntax. He proposes that this paradox may be solved if they are treated as lexically unaccusative and
syntactically unergative.
References
Alexiadou, A. & E. Anagnostopoulou (1996a) Synunetries, Asymmetries and the Role of Agreement,
GLOW Newsletter 36, 12-13.
Alexiadou, A. & E. Anagnostopoulou (I 996b) On the Conunon Formal Properties of Scrambling and Clitic
Doubling, ZAS Papers in Linguistics 6, 2-16.
Alexiadou, A. & E. Anagnostopoulou (to appear) Clitics in Restrictive Relatives: an antisymmetric account,
to appear in Studies on Greek Linguistics 17.
Alsina, A. (1992) On the Argument Structure ofCausatives, Linguistic Inquiry 23, 517-556.
Anagnostopoulou, E. (1993) On the Distribution of the Nominative Anaphor "0 eaftos mu" in Modem
Greek, presented at the First International Conference on Greek Linguistics, Reading.
Anagnostopoulou, E. (1994) Clitic Dependencies in Modem Greek, PhD disssertation, University of
Salzburg.
Anagnostopoulou, E. (in progress) The Dative Alternation in Greek, unpublished manuscript, University of
Tilburg.
Anagnostopoulou, E. & M. Everaert (1996a) Asymmetries in Binding: Configurational and Thematic
Effects on Anaphora, GLOW Newsletter 36,14-15.
Anagnostopoulou, E & M. Everaert (l996b) How Exceptional are Nominative Anapbors? A Case Study of
Greek, to appear in Actes du Deuxieme Colloque Langues & Grammaire 1995.
Ariel, M. (1990) Accessing Noun-Phrase Antecedents, Croom Helm, London.
Baker, M. (1988) Incorporation: A Theory of Grammatical Function Changing, University of Chicago
Press, Chicago.
Baker, M. (1993) Why Unaccusative Verbs cannot Dative-Shift, Proceedings ofNELS 23, 33-47.
Belletti, A. & L. Rizzi (1988), Psych Verbs and a-theory, Natural Language and Linguistic Theory 6,292352.
Bennis, H. (1993) Adjectives and Argument Atructure, HIL Manuscripts Vol. 1
Burzio, L. (1986) Italian Syntax: A Government and Binding Approach, Reidel Dordrecht.
Campbell, R & 1. Martin (1989), Sensation Predicates and the Syntax of Stativity, Proceedings of WCCFL
8,44-55.
Cardinaletti, A. & M. Starke (1995) Reply to Comments on The Typology of Structural Deficiency: On the
Three Grammatical Classes', in H. van Riemsdijk, ed., Clitics in the Languages of Europe, Vol III of
Language Typology, Mouton de Gruyter, Berlin (to appear).
Chomsky, N. (1993) A Minimalist Program for Linguistic Theory, MIT Occassional Papers in Linguistics.
Cinque, G. (1990) Types ofA '-Dependencies, MIT Press Cambridge Mass.
Czelpuch, H. (1982) Case Theory and the Dative Construction, The Linguistic Review 2,1-38.
Dikken, M. den (1995) Particles. On the Syntax of Verb-Particle, Triadic, and Causative Constructions.
Oxford University Press, New York-Oxford.
Dimitriadis, A. (1994) Dative Clitics and Case Licensing in Standard and Macedonian Greek, paper
presented at the Workshop in Greek Syntax, Berlin, this volume.
ON EXPERIENCERS
93
Dobrovie-Sorin, C. (1992) The Syntax ofRomanian. Comparative Studies in Romance, Mouton de Gruyter,
Berlin.
Dowty, D. (1979) Word Meaning and Montague Grammar, D. Reidel Publishing Company, Dordrecht.
Emonds,1. (1993) Projecting Indirect Objects, The Linguistic Review 10, 211-264.
Everaert, M. (1990) Nominative Anaphors in Icelandic: Morphology or Syntax?, in W. Abraham, W.
Kosmeijer and E. Reuland, (eds.), Issues in Germanic Syntax, Mouton de Gruyter, Berlin.
Franco, 1. (1990) Towards a Typology of Psych Verbs: Evidence from Spanish, in T. Green & S. Uziel, eds.,
MITWPL 12,46-62.
Grimshaw, 1. (1990) Argument Structure, MIT Press, Cambridge, Mass.
Hoekstra, T. (1994) Transitivity and Possession, unpublished manuscript, HIL, Leiden University.
[atridou, S. (1991) Clitics and Island Effects, unpublished manuscript, MIT.
Kayne, R. (1984) Connectedness and Binary Branching, Foris, Dordrecht.
Kayne, R. (1993) Toward a Modular Theory of Auxiliary Selection, Studia Linguistica 47, 3-31.
Larson, R. (1988) On the Double Object Construction, Linguistc Inquiry 19, 335-391.
Lasnik, H. & T. Stowell (1991) Weakest Crossover, Linguistic Inquiry 22, 687-720.
Mahajan, A. (1994) Ergativity and have-be Selection, Proceedings of NELS 24,317-331.
Marantz, A. (1991) Case and Licensing, Proceedings of ESCOL, 234-251.
Markantonatou, S. (1994) Diptota Rimata: Mia Lexico-Semasiologiki Prosegisi, Proceedings of the 15th
Annual Meeting of the Department ofLinguistics of the University ofThessaloniki, Thessaloniki.
Masullo, PJ. (1993) Two Types of Quirky Subjects: Spanish vs. Icelandic, Proceedings of NELS 23,303317.
Nash, L. (1994) On the Categorial Specification of Causative Morphemes, Proceedings of NELS 24,411425.
Pesetsky, D. (1995) Experiencers and Cascades, MIT Press, Cambridge Mass.
Reinhart, T. & E. Reuland (1993), Reflexivity, Linguistic Inquiry 24,657-720.
Stavrou, M. (1984) H KAl'tllCll AvrrovUJ.lla <m~ IIEQLOQl<mKE~ AvaqJoQlKE~ IIQ01:aoEl~ J.lE E~aQTllOT]
AJ.lEOOU AvtlKElJ.lEVOU nou Eloayovrm J.lE TO 'nou', Studies in Greek Liguistics 2, 121-136.
Stowell, T. (1986) Psych-Movement in the Mapping from D-structure to Logical Form, GLOW abstract.
Schiitze, C. (1993) Towards a Minimalist Account of Quirky Case and Licensing in Icelandic, MITWPL
19,321-376.
Tsimpli, L-M. (1995) Focussing in Modem Greek, in K. Kiss (ed.), Discourse Configurational Languages,
Oxford University Press, Oxford, pp. 176-206.
Varlokosta, S. & N. Hornstein (1993) A Bound Pronoun in Modem Greek, Natural Language and
Zaenen, A., J. Maling & H. Thniinsson (1985) Case and grammatical functions: The Icelandic Passive,
Natural Language and Linguistic Theory 3,441-483.
ON CLITICS, PREPOSITIONS AND CASE LICENSING IN STANDARD

AND MACEDONIAN GREEKI
Alexis Dimitriadis
University ofPennsylvania
I.
INTRODUCTION
Doubling in Romance languages is subject to Kayne's generalisation, which states

that a doubled object must be introduced by a preposition.
(1)
Lo vimos *(a) Juan.

cl we-saw Juan
(Spanish)
This behavior suggests that clitics absorb Case, requiring the presence of another
Case assigner in order for doubling to be licit. But other languages systematically
violate Kayne's generalisation:
(2)
Ton idhame to Yorgho

c1-acc saw-3pl the George
(Greek)
Indirect objects in Greek can be expressed either in the genitive, or as a preposition

plus an accusative NP. But only the non-prepositional alternative can be doubled by
a clitic, which is precisely the opposite to that predicted by Kayne's generalisation.
(3) a.
b.
(Tu) eghrapsa tu Yorghu

c1-gen wrote-lsg the George-gen
(*Tou) eghrapsa s-to Yorgho
cl-gen wrote-lsg to-the George-acc
Bulgarian follows a mixed pattern; doubled indirect objects are preceded by a

preposition, in accordance with Kayne's generalisation, but doubled direct objects
are not. Such phenomena call into question not only the cross-linguistic validity of
Kayne's generalisation, but also the explanation provided for it, namely, that
doubled objects require a preposition in order to receive Case.
In this paper I argue that verbs hosting a clitic do assign Case to a doubled
object, and that the doubling clitic can crucially participate in the assignment of
95
96
ALEXIS DIMITRIADIS
Case. The argument is based on a number of constructions which require a clitic,

sometimes allowing, but at others prohibiting a doubled NP. The existence of such
constructions can be taken to demonstrate that clitics participate in Case assignment.
In answer to the question of why the clitics are obligatory, I will argue that Greek
verbs are defective Case assigners; in obligatory-clitic constructions, indirect object
NPs must receive Case through the mediation of a clitic or a "light" preposition, with
the result that genitive or accusative is assigned instead of dative.
I will assume the analysis of clitics proposed by Sportiche (1992), and adopted
for Greek by Anagnostopoulou (1993, 1994). Clitics are functional heads that head
their own projection, and never appear as complements of the verb. The verbcomplement position is occupied by a doubled XP*, overt or null, which must be
licensed through specifier-head agreement by moving, overtly or covertly, to the
specifier XP of the projection headed by the clitic. The structure relevant to direct
object clitics is given below.
(4)
--------..
CLPacc
"XP
CL'acc
~
CLoacc
VP
6
V XP*
The Doubly Filled Voice Filter (Sportiche 1992:28) guarantees that when both CL
and XP* are overt (that is, in clitic-doubling configurations), movement to XP can
only occur at LF. When Clo is overt and xp* is null we have cliticization without
doubling, while null ClO and overt XP*, which can involve overt movement to XP,
corresponds to certain scrambling constructions.
Sportiche's analysis involves an important difference between direct and
indirect object clitics. He provides syntactic evidence that movement to the specifier
of the projection headed by direct object clitics is A' movement, while movement to
the specifier of dative clitics is A movement. Accordingly, he concludes that direct
object clitics head a separate "Accusative Voice" projection (denoted by CLPacc
above), but indirect object clitics appear in the head of the dative equivalent of
AgrO. The resulting structure is shown below. 2
CLITICS, PREPOSITIONS AND CASE LICENSING
(5)
97
CLPacc
XP
----
CL'acc
CLoacc
AgrPIO
/'-
yP
Agr'Jo
/'"--..
AgrOIO VP
I
~
CLdat
V XP* yP*
Clitic doubling has semantic consequences that have been studied at some length
(e.g., see Anagnostopoulou (1993, Uriagereka 1995). One relevant aspect is the fact
that doubling of direct objects has certain well-studied interpretational effects,
involving specificity and status in discourse, which are absent when indirect objects
are doubled. (cf. Suiier 1988 for Spanish). Here, however, I will focus exclusively
on issues of Case.
2. INDIRECT OBJECTS IN GREEK

Greek has lost the morphological dative case of Ancient Greek; formerly dative
constructions are typically expressed with the preposition se Cat' or 'to') followed
by an NP carrying morphological accusative. When followed by a determiner, se
obligatorily incorporates into it and appears as the prefix s-.
(6) a.
b.
Edhosa ta vivlia s-to Yorgho

gave-lsg the books to-the George-acc
Eghrapsa s-ti Maria
wrote -1 sg to-the Mary-acc
Depending on the verb, a number of prepositions are used. Of these se is by far the
most common, but apo 'from' and me 'with' are also seen with some frequency.
(7)
Zita apo to Yorgho

ena potiri nero
Ask from the George-acc one glass water-acc
'Ask George for a glass of water'
I will refer to this construction as "periphrastic dative", a term intended to be purely

descriptive: I will leave open for now the question of whether se should be
considered a Case assigner or, as suggested by the name I have chosen, a reflex of
dative morphology.
Indirect objects are not always expressed in the periphrastic dative. In the
standard dialect, spoken in Athens and most of southern Greece, (abbreviated SGr),
98
ALEXIS DIMITRIADIS
an indirect object may instead carry morphological genitive. 3 This construction is

slightly awkward with some verbs, clitic doubling being the preferred alternative,
but is quite grammatical. Genitive and periphrastic dative constructions are
synonymous, and the use of one or the other seems to be an optional stylistic matter.
(8) a.
Edhosa tu Yorghu
ta vivlia
gave-Isg the George-gen the books-acc
'I gave George the books'
b. Eghrapsa tis Marias
wrote-Isg the Maria-gen
'I wrote to Mary'
(SGr)
If the indirect object is a clitic pronoun, it invariably receives morphological

genitive:
(9) a
Tu
edhosa ta vivlia
cl-masc-gen gave-Isg the books-acc
'I gave him the books'
b. Tis
eghrapsa
cl-fem-gen wrote-Isg
'I wrote to her'
(SGr)
In doubling constructions, clitic and overt NP must always receive the same Case;
when a dative clitic is doubled, both the clitic and the overt object must appear in
the genitive, that is, periphrastic dative is disallowed on the doubled NP:
(10)
a.
b.
(11)
a.
b.
Tu
edhosa ta vivlia
tu Yorghu
(SGr)
cl-gen gave-lsg the books-acc the George-gen
'I gave George the books'
*Tuedhosa ta vivlia
sto Yorgho
cl-gen gave-Isg the books-acc to-the George-acc
Tis
eghrapsa tis Marias
cl-gen wrote -Isg the Maria-gen
'I wrote to Maria'
*Tis
eghrapsa sti Maria
cl-gen wrote-Isg to-the Maria-acc
This in effect makes Greek diametrically opposite to the pattern described by

Kayne's generalisation: far from being necessary to clitic doubling, prepositions are
actually incompatible with it.
In the dialect spoken in most of northern Greece, especially rural areas of
Macedonia, periphrastic dative is again universally available. But instead of using a
genitive clitic to refer to an indirect object, the Macedonian dialect uses accusative
clitics. Clitic doubling is readily available; as might be expected, the doubled object
must also receive (bare) accusative.
99
(12)
a.
b.
(13)
a.
b.
Ton edhosa ta vivlia

(to Yorgho)
cl-acc gave-lsg the books-acc (the George-acc)
'I gave him (George) the books'
sto Yorgho
* Ton edhosa ta vivlia
him-acc gave-lsg the books-acc to-the George-acc
Tin egrapsa
(ti Maria)
cl-acc wrote-lsg (the Maria-acc)
'I wrote to her (to Maria)'
*Tin
egrapsa sti Maria
cl-acc wrote-lsg to-the Maria-acc
(NGr)
So far Macedonian Greek is just like the southern version, with the substitution of
accusative for genitive as the reflex of dative. But while a bare genitive object is
grammatical in the southern dialect (SG), as in sentence (8), the Macedonian dialect
typically does not allow a dative object to carry bare accusative in the absence of a
clitic. 4 Thus sentences (14b) and (15b) are ungrammatical, unlike their SGr
counterparts in (8).
(14)
a.
b.
(15)
a.
b.
Edhosa sto Yorgho

ta vivlia
gave-l sg to-the George-acc the books-acc
* Edhosa to Yorgho
ta vivlia
gave-lsg the George-acc the books-acc
Eghrapsa sti Maria
wrote-lsg to-the Maria-acc
*Eghrapsa ti Maria
wrote-lsg the Maria-acc
Thus the presence of the clitic adds to the options for Case assignment, suggesting
that clitics participate in the assignment of Case in doubled constructions.
To summarize the data: In both the standard and the Macedonian dialects, an
indirect object can carry periphrastic dative. In addition, the Macedonian dialect
allows an indirect object to be in the accusative, but only if it is doubled by an
accusative clitic; the standard dialect allows genitive indirect objects, which can, but
need not, be doubled by a genitive clitic. The following table lists the types of
oblique objects allowed in the two dialects:
(16)
Standard/Southern Greek (SGr)

a. se + NPacc ("periphrastic dative")
b. clgen...
(NPgen)
c. *cl-gen .. se + NPacc
d. NPgen
("bare" genitive)
ALEXIS DIMITRIADIS
100
(17)
MacedonianlNorthern Greek (NGr)

a. se + NPacc ("periphrastic dative")
b. elacc...
(NPacc)
c. *elacc ... se+ NPacc
d. *NPacc
(''bare'' accusative)
In view of the structure in (5) , there is a straightforward account of the role of the
clitic in the Macedonian dialect. In order to receive Case in a specifier-head
configuration, the indirect object must move to Spec(AgrIO); the verb incorporates
into AgrOIO' which is already occupied by the indirect object clitic. This
independently motivated process places the clitic in a position to participate in Case
assignment. Suppose that the verb, in the absence of the clitic, is a defective Case
assigner, unable to Case-mark an indirect object. This explains the ungrammaticality
of (14b) and (15b). In the presence of the elitic, the clitic-verb complex assigns
accusative Case, hence (12a) and (13a) are well-formed.
It remains to explain the incompatibility of the clitic with periphrastic dative, as
in (12b) and (13b). We can appeal to reasons of Case, simply requiring that if yP*
appears in Spec(AgrIO)' it must appear in the Case assigned by the verb-elitic
complex. Sportiche's account requires a doubled YP* to check a certain [+F] feature
at Spec(AgrIO), hence this requirement effectively renders ungrammatical any
configuration where yP* cannot check its Case in this way.
Ifwe accepted (which we should not, as I will argue below) that the preposition
of periphrastic dative is an ordinary Case assigner, it would follow that periphrastic
datives are prevented from appearing in Spec(AgrIO)' and hence cannot be doubled.
This would explain the asymmetry between Spanish and Macedonian Greek, with
respect to Kayne's generalisation, as a difference in the Case-assignment properties
of the verb-clitic complex: in Spanish, which requires doubled NPs to be doubled by
a preposition, the complex does not assign Case; in Greek, which prohibits this, it
does. This would be consistent with the traditional explanation for why a preposition
is obligatory for doubled objects in Spanish, namely, that such objects do not
receive Case from the verb.
However, things are a bit more complicated. The preceding analysis, which ties
the need for a dative clitic in Gr to the need of the full NP object for Case, cannot
account for the fact that indirect objects in Spanish must also be obligatorily
doubled by a elitic. As Jaeggli (1986) reports, indirect objects in most dialects of
Spanish are markedly degraded unless doubled by a dative clitic. (The clitic is in
principle optional).
(18)
??(Lej) entregue el libro al

professorj.
el gave-I sg the book to-the professor
We must abstract away from the fact that indirect objects in Greek have an alternate
mode of realization (periphrastic dative), which is incompatible with clitic doubling.
We can then say that in Spanish, as in Macedonian Greek, the presence of the elitic
licenses the indirect object NP. In Macedonian Greek, the elitic licenses an object
101
NP in the accusative, while in Spanish the clitic licenses the only available type of
dative object, which happens to be introduced by a preposition.
In Spanish, then, the preposition is not in itself sufficient to license the indirect
object; if the preposition was the sole source of Case for the indirect object, we
would have to appeal to reasons other than Case for the obligatoriness of the dative
clitic. Jaeggli (1986) concluded that verbs do assign Case to their indirect object; the
preposition a is either a morphological reflex of dative case (hence not a true
preposition at all) , or else it serves to transmit Case to its complement. I will
propose modifying our provisional assumption, that se is an ordinary Case-assigning
preposition, and arguing instead that se transmits the Case assigned by the verb, or
rather that it mediates, that is, participates along with the verb in the assignment of
Case to its complement. A periphrastic dative phrase must receive defective Case;
since a dative clitic in Macedonian Greek causes accusative Case to be assigned to
its doubled complement, the latter cannot be in the periphrastic dative, just as a
direct object, doubled or undoubled, could never be in the periphrastic dative. This
analysis is defended in the following sections.
3. THE DOUBLE OBJECT CONSTRUCTION IN GREEK
A discussion of Case assignment to indirect objects must depend on the structure

assumed for indirect objects in general, and double objects in particular. English
ditransitive verbs can express their indirect object in two ways reminiscent of the
alternation between genitive and the prepositionally introduced periphrastic dative
of Greek:
(19)
a.
b.
I gave John the book

(double object)
I gave the book to John (oblique dative)
It is well known that these two constructions are structurally different; the problems
posed by their properties were pointed out by Barss and Lasnik (1986), and received
a widely accepted analysis by Larson (1988, 1990).
Given the obvious resemblance of the Greek ditransitive alternations to the
English double object/oblique dative constructions, it is tempting to analyze the
Greek periphrastic dative as the analogue of the English oblique dative, and genitive
indirect objects as parallel to English double objects. Such an analysis was argued
for by Catsimali (1990), and is apparently assumed by Campos (1991). However, on
closer inspection such a parallel turns out to be untenable: the structural contrast that
motivated Larson's analysis is systematically absent in Greek. In the absence of any
evidence of a structural difference, we can conclude that genitive (or accusative) and
periphrastic dative indirect objects occupy identical structural positions. In the
English double object constructions, the indirect object necessarily appears ftrst, and
can bind the direct object.
102
ALEXIS DlMITRIADlS
(Order: V 10 DO)
(20)
a. I showed Mary herself (in the mirror)
b. *1 showed herself Mary
a. I gave every workerj hisjpaycheck
(21)
b. * I gave itsjowner every paychec~
10>DO
In oblique dative structures, both linear order and binding possibilities are reversed:
the direct object precedes and can bind the indirect object.
(Order: V DO to 10)
(22)
a. *1 showed herselfto Mary
b. I showed Mary to herself
a. *1 gave hisj paycheck to every workerj
(23)
b. I gave every chec~ to itsj owner
DO > 10
In Greek, we fail to fmd a similar contrast: reflexive binding in sentences with

(bare) genitive, periphrastic dative and doubled complements consistently patterns
with the English double-object sentences, not the obliques. Sentence (24) shows that
a periphrastic-dative indirect object binds a reflexive direct object, but not viceversa. Sentence (25) shows that the same is true of bare-genitive and doubled
indirect objects in the standard (SG) dialect; the Macedonian dialect behaves
similarly.
(Order: V 10 > DO)
(24)
(25)
a.
Edhiksa
sti Meri ton eafto
tis (ston katbrefti)
showed-lsg to-the Mary the self-acc her (in-the mirror)
'I showed Mary herself (in the mirror)'
b. * Edhiksa ston eafto tis ti Meri
(ston kathrefti)
showed-lsg to-the self her the Mary-acc (in-the mirror)
'I showed herself Mary (in the mirror),
a. (Tisj) edhiksa tis Merisj
ton eafto tis
(SGr)
cl showed-lsg the Mary-gen the self-acc her
'I showed Mary herself
tu ea~
tis ti Meri
b. *(Tllj) edhiksa
cl-gen showed-lsg the self-gen her the Mary-acc
'I showed herself Mary'
Greek, which in general has relatively free word order, can realize the direct object
before the indirect object without reversing their dominance relationship. S Since
Greek nouns are clearly inflected for Case, none of these examples is ambiguous
between the DO-IO and the IO-DO orders.
(Order: V DO C 10)
a. Edhiksa
ton eafto tis sti Meri
(26)
showed-l sg the self her to-the Mary
b. * Edhiksa ti Meri ston eafto tis
showed-lsg the Mary to-the selfher
a. (Tisj) ediksa
ton eafto tis tis Merisj
(27)
cl-gen showed-lsg the self her the Maryj
b. *(Tllj) edhiksa
ti Meri tu eaftu j tis
cl-gen showed-lsg the Mary the self-gen her
103
(SGr)
In all the Greek examples shown so far, the indirect object can bind a reflexive
direct object but not vice versa, regardless of linear order or the form of the indirect
object. Surprisingly, a quantifier in the direct object can bind an indirect object
appearing to its right. But again, there is no contrast between the different forms of
indirect object: examples (28) and (29) show this for periphrastic datives and bare
genitives, respectively.
(28)
a.
b.
(29)
a.
b.
Edhosa tin kathe epitaghij ston idhioktiti tis

gave-Isg the each check
to-the owner its
'I gave each check to its owner'
*Edhosa ston idhioktiti tisj tin kathe epitaghij
gave-lsg to-the owner its the each check
'I gave its owner each check'
Edhosa tin kathe epitaghij tu idhioktiti tisj
gave-lsg the each check
the owner-gen its
'I gave each check to its owner'
*Edhosa tu idhioktiti tisj tin kathe epitaghij
gave-Isg the owner- gen its the each check
'I gave to its owner each check'
(SGr)
The analysis of Greek ditransitives must remain beyond the scope of this paper. For
my present purposes, what matters is the consistent lack of a structural contrast
between genitive and periphrastic dative indirect objects. It is reasonable, then, to
conclude that all types of oblique complement NP in Gr occupy the same structural
position. 6
The binding properties of the Greek and English objects are summarized in the
following table.
(30)
English:
Double objects: 10 > DO
Oblique datives: DO > 10
NP-dat > NP-acc

NP-acc > to NP-dat
104
ALEXIS DIMITRIADIS
(31)
Greek (Reflexive binding):

Periphrastic dative: 10 > DO
indirect object clitic: 10 > DO
10 > DO
doubling clitic:
(bare) genitive object: 10 > DO
se NPacc
> NP-acc
cl-gen
> NP-acc
cl-gen NPgen > NP-acc
NP-gen
> NP-acc
(The Greek binding facts are insensitive to constituent order).
4. THE CASE OF PERIPHRASTIC DATIVES

What is the source of the Case assigned to the NP appearing in periphrastic dative
constructions? Given that the Case filter requires NPs to receive Case from
somewhere, there are two logical possibilities.
a. There is no extrinsic source of Case, i.e., the preposition heading the periphrastic
dative phrase is the sole source of Case. Verbs do not assign Case to periphrastic
dative objects.
b. Case is assigned extrinsically (e.g., by the verb or AgrO).
In section 2 we considered one problem associated with option (a), namely, the fact
that if we take the preposition to be the sole source of Case for its complement, we
must look elsewhere for an explanation of the obligatoriness of dative clitics in
Spanish. Even if we restrict our attention to Greek, there are problems with option
(a), which can be summarized with the observation that periphrastic dative objects
behave as if they occupy a Case-assigned, not a Case-less, position.
Indirect objects in Greek never raise to a non-thematic, Case-assigned position.
For example, they cannot become the subjects of passives:
(32)
a.
*0 Nikos dhjavastike ta dhikeomata tu

the Nikos was-read the rights
his
'Nikos was read his rights'
b. *0 Nikos apanditbike
the Nikos was-answered
If verbs in Macedonian Greek did not assign Case to their indirect object, we would
expect these sentences to be grammatical.
Conversely, why can't Case-less direct objects ever be licensed by se? For
example, why can't the underlying direct objects of unaccusatives and passives stay
in situ?
(33)
* pro irthe sto
Niko
expl. came to-the Niko-acc

'Nikos came'
105
It seems that periphrastic dative objects are assigned Case extrinsically by the verb,
that is, that the preposition of periphrastic datives is not an independent Case
assigner. This allows for the following two possibilities:
a. The elements heading periphrastic dative phrases are not really prepositions; they
are just bits of dative morphology, expressing the Case assigned directly by the verb
or locative.
b. Case is assigned extrinsically, but cannot be "realised" directly by the NP; se or
another "light" preposition heads a Case-assigned PP and "mediates" the assignment
of Case to its complement.
Since several different prepositions can head a periphrastic dative complement, and
the choice between them is not free, they cannot be treated as morphological
reflexes of Case unless we postulate a different "Case" for each one. It is preferable
to assume that a single form of Case is assigned by the verb, and that it is
transmitted to the NP by the preposition heading the periphrastic dative
construction, along the lines suggested by Jaeggli (1986) for the preposition a in
Spanish. A particular preposition can be used as long as its semantics are compatible
with those of the relationship it expresses. Thus most obliques are compatible with a
goal or benefactive interpretation and are introduced with se 'at/to'; those expressing
source are introduced by apo 'from'; and some arguments, not being compatible with
any of the available prepositions, cannot be expressed through periphrastic dative at
all.
5. MORE OBLIGATORY CLITICS

In the preceding sections, I argued that clitics or "light" prepositions mediate in the
assignment of Case to the indirect object NPs of verbs in obligatory-clitic
constructions. But the indirect objects of Macedonian Greek and Spanish are not the
only context in which a c1itic argument is required or preferred. A variety of marked
constructions in both Greek and Spanish also require c1itics. While many of those
employ genitive clitics in standard Greek and accusative clitics in the Macedonian
dialect, others use genitive c1itics in both dialects.
5.1 Datives of Possession

The so-called "datives of inalienable possession" (which does not actually need to be
inalienable) must also be expressed with a clitic. Jaeggli (1986) proposes that the
clitic augments the thematic grid of the verb it is used with. As the following
example shows, datives of possession can be doubled.
(34)
a.
Lej examinaron los dientes al caballoj

'They examined the horse's teeth'
ALEXIS DIMITRIADIS
106
b.
Le duele la cabreza a Juan

'John has a headache'
Gr also have datives of possession; like indirect objects, they are expressed in the
genitive in southern Greek, and in the accusative in Macedonian. As in Spanish, the
clitic is obligatory (in both dialects).
(35)
a.
Tu Yorghu
tu
ponai to kefali (tu)
the George-gen cl-gen hurts the head (his-gen)
'George has a headache'
b. Ton Yorgho
ton ponai to kefali (tu)
the George-acc cl-acc hurts the head (his-gen)
'George has a headache'
(SGr)
(NGr)
The preposed NP is not a scrambled possessor: note that the possessive clitic (which
is optional) is in the genitive in both dialects, while the preposed NP tou Yorghulton
Yorgho agrees in Case with the verbal clitic. Datives of possession should not be
confused with ethical datives, which both Greek and Spanish also have.
(36)
a.
Juan me lez arruin6 la vida a esa chicai

Juan cl cl ruined the life to that girl
'Juan ruined that girl's life (and this affects me)'
b. Mi mu
stenachorite to pedi
(SGr)
not cl-gen-Isg upset
the kid
'Don't upset the kid (which concerns me)'
In such constructions the referent of the clitic is not an argument of the verb, but is
an entity somehow interested in the situation being described. An ethical dative
clitic must be in the ftrst or second person, and can never be doubled. 7
5.2 Various Oblique Complements

Both standard and Macedonian Greek have numerous verbs which require that an
oblique complement be expressed as a clitic. Some allow a doubled NP, but others
do not. Some, but not all, allow a full NP to be introduced by a preposition. The
following examples, both in the standard dialect, allow a clitic or a clitic-doubled
NP. Sentence (37c) shows that the argument of epese 'fell' cannot be introduced by
the preposition apo 'from' (or any other); while as (38b) shows, the object of
andistathike 'resisted' can be introduced by the preposition se. (As before, a clitic
cannot double an NP introduced by a preposition).
(37)
a.
*(Tis) epese to potiri

tu papa
(tis Marias)
cl-gen fell the glass-nom the- priest-gen the Maria-gen
'She (Maria) dropped the priest's glass'
107
b.
(38)
*Epese tis Marias

to potiri
tu papa
fell
the Maria-gen the glass-nom the priest-gen
'Maria dropped the priest's glass'
c. *Epese apo ti Maria
to potiri
tu papa
fell
from the Maria-acc the glass-nom the priest-gen
a. 0 Petros *(tis) andistathike (tis Marias)
the Petros cl-gen resisted
the Maria-gen
'Petros resisted Maria'
b. 0 Petros (*tis) andistathike (s-ti Maria)
to-the Maria-acc
the Petros c1-gen resisted
'Petros resisted Maria'
In the Macedonian dialect, the genitive would be replaced by accusative, as before,

and the judgments would be the same as above.
5.3 Locative Prepositions

Greek allows the complement of most complex locative prepositions ("near the
house", "with the children") to be expressed with a genitive clitic. 8 Such prepositions
categorically prohibit bare genitive objects: full NP complements must be
introduced by a "light" preposition, just like the indirect objects of verbs in
Macedonian Greek. Several prepositions are possible in this context.
(39)
(40)
a.
Brosta s-to spiti

in-front at-the house-acc
'In front of the house'
b. Brosta tu
in-front c1-gen
c. *Brosta tu spitiu
in-front the house-gen
a. Mazi me ti Maria
together with the Maria-acc
'With Maria'
b. Mazi tis
with c1-gen
c. *Mazi tis Marias
together the Maria-gen
Thus locative prepositions behave very similarly to the obligatory-c1itic verbs: an

argument must be expressed either as a genitive c1itic (in all dialects) or as a
periphrastic dative NP. 9 Clitic doubling in these constructions is usually degraded,
but seems to be at least sometimes possible:
ALEXIS DIMITRlADIS
108
(41)
Ta klidhja tis Marias

itan brosta ~
tu Yorghu;,
the keys the Maria-gen were in-front c1-gen the George-gen
alla dhen ta
evlepe
but not them saw
'Mary's keys were in front of George, but he did not see them'
These constructions carry genitive case, even in the Macedonian dialect. Thus
Macedonian distinguishes between the indirect objects of verbs, which must be
expressed as prepositional phrases or accusative clitics, and the complements of
locative prepositions, which must be expressed as prepositional phrases or genitive
clitics. We see then the same alternation as with the indirect objects of Macedonian
Greek: an argument can be expressed only as a clitic, possibly doubled, or as a
prepositional phrase, never as an undoubled bare NP. The clitic complements of
locatives were studied by Terzi (1991), who proposed an analysis much along the
lines of the account developed here. Terzi points out that locatives in Ancient Greek
could license genitive complements directly, and argues that Greek locatives have
lost the ability to do so. Her analysis involves an empty functional head below the
locative preposition, between the locative and the light preposition. This head is
assigned genitive Case (through government) by the locative, thus discharging the
latter's Case feature. The NP complement is independently assigned accusative by
the light preposition that governs it.
Since current views of Case require that it always be assigned in a specifierhead configuration, I will take Case assignment to proceed via an agreement
projection appearing above, not below, the locative. I have argued that the light
prepositions that introduce periphrastic dative objects are' not independent Case
assigners, but mediate in the assignment of Case to their complement. The PP
complements of locatives can be taken to have the same status, with Case assigned
by the light preposition in conjunction with the locative.
6. CONCLUSION
In the constructions presented in the previous sections, accusative or genitive
oblique NPs are in complementary distribution with periphrastic datives: the former
can co-occur with clitics but not with clitic-less verbs or locatives, while periphrastic
datives co-occur with locative prepositions and with verbs without a clitic, but not
with clitics.
I have argued that what is at issue is Case licensing: a ''bare'' NP cannot appear
because it fails to be Case-licensed. Since the presence of a clitic licenses an NP
object in most of these environments, we conclude that clitics contribute to the
Case-licensing capacity of the verb or locative. By themselves, such verbs and
locatives are defective Case assigners, which cannot directly assign Case to an NP
complement.
It is possible to provide a concrete interpretation for the "defectiveness" of the
Case assigned by these verbs. Recall that Greek lacks the morphological dative of
109
Ancient Greek. Suppose that Gr verbs assign (abstract) dative to their oblique
complements; because Gr no longer has morphological dative case, NPs cannot
directly receive dative, and Case assignment must be mediated by a clitic or "light"
preposition. Perhaps by virtue of being closed-class items, clitics and light
prepositions are capable of receiving dative Case from the verb, and in turn they
assign genitive or accusative to their complement.
Unfortunately, this story cannot be straightforwardly extended to locative
prepositions, since these assigned genitive, not dative, in Ancient Greek. Moreover,
as noted in section 5.3, they contrast with verbs in Macedonian Greek, since
periphrastic datives alternate with accusative clitics as the complements of verbs,
but with genitive clitics as the complements of locative prepositions. We can only
conclude that verbs and locatives have slightly different properties, which determine
whether the "defective" Case they assign is realized by the host-clitic complex as
genitive or accusative. But the identity of this defective Case must remain rather
abstract in nature.
One more messy part remains to the story: recall that the indirect objects of
verbs in standard Greek can generally appear carrying "bare" genitive. We must
assume that most verbs of standard Greek have the option of assigning genitive to
their indirect object instead of dative, often somewhat marginally. Similarly, but
much less often, some verbs of Macedonian Greek may assign accusative to their
indirect object.
The account developed here explains a characteristic of obligatory-clitic
constructions that I have not dwelled on until now: although direct objects can be
doubled in a number of languages, including Greek and some dialects of Spanish,
no obligatory-clitic constructions involve direct objects. to The reason can now be
readily seen: NPs can realize accusative Case directly, so the mediation of a clitic or
preposition is never needed for direct objects, which are assigned accusative. Recall
also that according to Sportiche (1992), dative clitics occupy Agr0 10 while
accusative clitics head their own "Clitic Voice" projection. Although it is in
principle possible that Case features could be checked by specifier-head agreement
at the Clitic Voice projection, it remains true that dative clitics are intimately
involved with the locus of Case assignment.
This framework works for Spanish dative clitics as well: In Spanish, dative is
always realized through an a phrase. The Case assigned by verbs to their indirect
objects is slightly "defective", hence clitic doubling is the preferred way to license
dative objects. But Spanish, unlike Greek, has "real" dative clitics: Indirect object
clitics in Spanish assign dative to their doubled NPs, which is once again realized
prepositionally.
Kayne's generalisation draws a sharp line between Spanish, which obeys it, and
Greek, which violates it. The analysis I presented here treats the difference between
the two languages as a difference not in the presence or absence of Case, but in the
identity of the Case assigned by indirect object clitics. This reduces Kayne's
generalisation to an accident of morphology: languages that obey it have clitics that
assign Case that catinot be realized by bare NPs. Languages that violate it have
clitics that assign better-behaved Case. The correctness of this viewpoint can be seen
ALEXIS DIMITRIADIS
110
by examining the Bulgarian clitic system, which appears to partly obey and partly
violate Kayne's generalisation.
Bulgarian has almost completely collapsed the dative and genitive cases
(although the name "genitive" is traditionally used for possessives, and the name
"dative" for oblique objects). Dative/genitive NPs (with the exception of prenominal
genitives, which I will ignore) are expressed periphrastically, by prefixing the NP
with the preposition na. Doubled dative/genitive NPs are expressed in exactly the
same way.
(42)
a.
prijatelkata (mUj) na Ivan;

the-girlfriend c1-gen of Ivan
'Ivan's girlfriend'
b. Dadoh (mUj) pismoto na Ivan;
gave-lsg cl-dat the-letter to Ivan
'I gave the letter to Ivan'
Thus, indirect objects and possessives appear to obey Kayne's generalisation.

However, direct objects also appear in the same form whether or not they are
doubling a clitic:
(43)
Vidjah (go) Ivan

I-saw c1-acc Ivan
'I saw Ivan'
In this case Bulgarian appears to violate Kayne's generalisation, since the doubled
NP is not introduced by a preposition. It should be clear that there is no real
inconsistency in the Bulgarian pattern. Bulgarian consistently puts doubled objects
in the same form they take when they are undoubled; it so happens that Bulgarian
expresses dative/genitive periphrastically, and accusative holophrastically. Clitic
doubling is simply irrelevant to this pattern, i.e., clitics in Bulgarian do not modify
the Case assigned by the verb.
NOTES
I am grateful to Sabine Iatridou, Michael Hegarty, Spyridoula Varlokosta, Elena Anagnostopoulou

and Roumyana Izvorski for their considerable contributions to the conception and expression of the
ideas presented in this paper. I also wish to thank an anonymous reviewer for a multitude of constructive
suggestions. I remain solely responsible for all errors.
2
In Sportiche's (1992) account, clitic heads cliticise to higher functional categories, with the result
that this structure does not necessarily make a prediction about their surface order; if left adjunction is
assumed, we predict a surface order that is the reverse of the order of projections, namely CL-dat CLacc, which is the usual order for Greek clitics.
3
Greek has lost the morphological distinction between genitive and dative case and has
generalized the use of genitive. Here I use both terms interchangeably.
III
Many speakers accept bare accusative with some verbs (usually very common ones). There is quite
a bit of speaker variation on this point, although to my knowledge no speaker finds bare accusative
universally acceptable as a way to express indirect objects.
5
The alternative orders are not necessarily related to each other through scrambling. Alexiadou
(this volume) shows that the binding relationship of a post-verbal subject to the direct object is
dependent on their relative order.
(i) a.
Se pjon parusiase i mitera tuj to kathe agorij ?
to whom presented the mother his the each boy
'To whom did hisj mother present each boYj'
b. Se pjon parusiase to kathe agorij i mitera tu j?
to whom presented the each boy the mother his
'To whom did hisj mother present each bOyj?'
It should be noted that (ia) becomes grammatical if the object is doubled by a clitic:
(ii)
Se pjon tOj parusiase i mitera tuj to kathe agori j
to whom cl presented the mother his the each boy
'To whom did hisj; mother present each boy?'
6
After this paper was presented, Anagnostopoulou and Everaert (1996) proposed an analysis of the
phenomena presented here, which retains the treatment of the two types of indirect object in Greek as
corresponding to the two types present in English. The lack of a binding contrast is explained within the
framework of reflexivity, by claiming an asymmetry between Greek, in which binding is said to be
determined by th-prominence. and English, in which it is not. I will not address their analysis here; but
if it is correct, it is clear that the present account of the Case licensing of periphrastic datives should be
revised accordingly.
7
Pace Warburton (1977), the clitic in (36b) cannot be replaced by the NP ja mena 'for me' without
some change in meaning:
(i) Ja mena, mi to stenachorite to pedhi
For my sake, don't upset the kid'
There is no clear pattern as to which of these prepositions exceptionally prohibit clitics. For
example brosta 'in front of and piso 'behind' allow clitics, while kato 'below' disallows them and pano
'on/above' allows them only when it means 'on'.
9
A word may be necessary about the difference between prepositions that do not allow bare NP
arguments (presumably because they do not assign Case to them) and prepositions that are introduced in
order to assign Case to NPs. It may be simplest though not necessarily correct to think of the former as
adverbs with arguments, not true prepositions. They have clear semantics (mostly locative), they can be
used without arguments as (conventional) adverbs, and they are phonologically strong. Case-assigning
prepositions, on the other hand, tend to be semantically vague, must always have a complement, and are
phonologically weak.
The reader is referred to Theofanopoulou-Kondou (1995), who studies this issue in detail. She
adopts Starke's (1993) distinction of colourful and colourless prepositions (the former basically
comprising the complex, contentful prepositions), and argues that colorless prepositions may occupy a
Co head, while the colorful ones appear in po.
Schneider-Zioga (1994) also treats both types as prepositions; she notes that locative prepositions
cannot take a reflexive complement, and that conversely "lighter" prepositions, which do not accept
clitics, allow reflexives.
At any rate it is clear that the two types of preposition differ systematically in fundamental ways.
10
Certain NPs, for example the bare quantifier ola 'all', cannot be used as direct objects unless they
are doubled by a clitic. In these cases the requirement for a clitic is conditioned on the identity of the
object rather than that of the Case assigner. Since they differ so markedly from the phenomena
presented in this paper, I will assume that they are indicative of a different phenomenon. See
Anagnostopoulou (1993) for discussion of these examples.
112
ALEXIS DIMITRIADIS
References
Anagnostopoulou, E. (1993) On the Representation of Clitic Doubling in Modem Greek, unpublished
manuscript, University of Salzburg! University of Tilburg.
Anagnostopoulou, E. (1994) Clitic Dependencies in Modern Greek, Ph.D. Dissertation, Universitat
Salzburg.
Anagnostopoulou, E. & M. Everaert (1996) Asymmetries in Binding: Configurational and Thematic
Effects on Anaphora, paper presented at the 19th GLOW Colloquium, Athens.
Barss, A. & H. Lasnik (1986) A Note on Anaphora and Double Objects, Linguistic Inquiry 17, 347-354.
Bruge, L. & G. Brugger (1993) On the Accusative 'A'in Spanish, unpublished manuscript, Universita di
Venezia.
Campos, H. (1991) Indirect Object Alternations in Modem Greek, unpublished manuscript, Georgetown
University.
Catsimali, G. (1990) Case in Modern Greek: Implications for Clause Structure, Ph.D. Dissertation,
University of Reading.
Izvorski, R. (1993) Genitive Clitics, unpublished manuscript, University of Pennsylvania.
Jackendoff, R. (1990) On Larson's Treatment of the Double Object Construction, Linguistic Inquiry 21,
427-456.
Jaeggli, O. (1986) Three Issues in the Theory of Clitics: Case, Doubled NPs, and Extraction, in Hagit
Borer (ed.), The Syntax 0/ Pronominal Clitics, volume 19 of Syntax and Semantics, Academic
Press, New York, pp. 15-42.
Kayne, R. (1994) The Antisymmetry 0/Syntax. MIT Press, Cambridge, Mass.
Larson, R. (1988) On the Double Object Construction, Linguistic Inquiry 19,335-392.
Larson, R. (1990) Double Objects Revisited: Reply to Jackendoff, Linguistic Inquiry 21, 589-632.
Schneider-Zioga, P. (1994) The Syntax of Clitic Doubling in Modern Greek, Ph.D. Dissertation,
University of Southern California.
Sportiche, D. (1992) Clitic Constructions, unpublised manuscript, UCLA.
Suffer, M. (1988) The Role of Agreement in CIitic-Doubled Constructions, Natural Language and
Starke, M. (1993) Notes on Prepositions and Clause Structure, Mini memoire, Universire de Geni:ve.
Terzi, A. (1991) Genitive Clitics of Prepositions, unpublished manuscript, CUNY Graduate Center.
Theophanopoulou-Kontou, D. (1995) Complex Prepositions in Modern Greek: Notes on the Modern
Greek Prepositional System, unpublished manuscript, University of Athens.
Uriagereka, J. (1995) Aspects of the Syntax of Clitic Placement in Western Romance, Linguistic
Inquiry 26, 79-123.
Warburton, I. (1977) Modern Greek Clitic Pronouns and the 'Surface Structure Constraints' Hypothesis,
Journalo/Linguistics 13, 259-281.
WEAK AND STRONG POLARITY: EVIDENCE FROM GREEK)
University ofAmsterdam
1. INTRODUCTION
This paper highlights some central aspects of negative polarity and negative concord in
Greek (Gr). The principal goal is to attempt a solution to the so-called diversity
problem, i.e. to the fact not all polarity items are licensed in the same environments, by
motivating, on both syntactic and semantic grounds, a distinction between strong and
weak polarity licensing. The proposal builds on a conjecture expressed in Ladusaw
(1992, 1994) that there might be two mechanisms involved in the licensing of negative
polarity items (NPIs) and negative indefinites. I will show that in Gr strong and weak
licensing are explicitly marked at s-structure by means of emphatic stress. It will be
suggested that the analysis presented here can prove helpful in accounting for the
diverse distribution ofNPIs in languages other than Gr.
2.
BACKGROUND CONSIDERATIONS
A wealth of literature on negative polarity and negative concord is devoted to the

licensing problem, that is, to the specification of what the formal property is that all
NPI-licensing environments share. Some of the proposals view NPI-licensing as a
phenomenon being properly talked about in pragmatic terms (Linebarger 1980, Krifka
1994), some others appeal to semantic conditions (Ladusaw 1979, Zwarts 1986,
Kadmon & Landman 1993, van der Wouden 1994, Giannakidou 1994, 1995, 1997
Israel 1994).2 With the exception of Giannakidou 1994, 1995 the common
denominator in the semantic theories listed here is that they attribute the licensing
potential to the monotone decreasing MD (or scale-reversing) properties of the
triggering contexts. This assumption however is shown to be empirically and
conceptually refutable. For reasons I cannot go into in this paper, I take it that NPIlicensing is regulated by a semantic condition and I adopt the theory I proposed in
Giannakidou 1997) according to which the negativeness of the polarity contexts can be
properly subsumed under the wider notion ofnonveridicality.
As regards the syntax of the phenomena at hand, a great number of compelling
proposals is well-documented in the literature (Progovac 1988, 1994, Zanuttini 1991,
Haegeman & Zanuttini (H&Z, 1991), Ladusaw 1992, 1994, Aquaviva 1993, Quer
113
A. Alexiadou et al. (eels.), Studies in Greek Syntax, 113-133.
ANASTASIA GIANNAKIDOU
114
1993, Deprez 1995 among others). The interpretation of negative indefinites (NIs) has
been a central issue in almost all of these studies. NIs form the subset of NPls that
comprises atomic NPls (DPs and adverbs) which are used in negative concord
structures and of which the Romance n-words are the most celebrated member. Often,
authors make recourse to the opposition between NPls and negative quantifiers (NQs)
in order to capture the fact that NIs are sometimes interpreted as negative (for instance
under negation) and sometimes as existential nonnegative (for instance in questions
and conditionals). Opinions differ and according to some authors NIs are NQs (cf.
Zanuttini 1991, H&Z, 1991), and according to some others, NIs are NPls (cf. Laka
1990, Progovac 1988,1994, Aquaviva 1993, Deprez 1995).
As I mentioned above, the difference underlying the NPI versus NQ opposition is
a difference between an existential nonnegative and a universal negative quantifier for
NPls and NQs respectively. However, if we assume, quite intuitively, that the most
obvious way to understand negative polarity is as involving expressions, i.e. NPls,
whose occurrence is sanctioned in negative contexts but banned in nonnegative
contexts, then a discussion in terms of the NPIINQs opposition does not prove very
helpful because it is based on the wrong assumption that the term NPI bears on the
interpretation. An NPI is not an expression which is interpreted as an existential
nonnegative quantifier (if it were, then singular indefinites would be NPls too) but
rather, NPI is a label that relates to distribution and it refers to expressions which are
licensing dependent on negation. Once we realize that the terms NPI and NQ apply on
different levels, i.e the former to the level of distribution and the latter to the level of
interpretation, the opposition appears meaningless. Because they may require the
presence of overt negation, ''NQs'' may be NPls in negative concord languages
(Romance, Serbian Croatian, Russian and, as I will argue below, Greek) and may not
be in others (English, German, Dutch). For this reason I will abstain from the NQINPI
debate and instead, I will insist on the contrast between existential nonnegative and
universal negative which characterizes the interpretation of the items under
consideration. For space considerations, I will also refrain from tackling the specifics
of the syntactic proposals presented above and I take the Ladusaw type of reasoning as
the point of departure.
Ladusaw (1992, 1994) sketches a theory for the interpretation ofNIs of which the
most important features can be summarized as follows (the proposal is couched in
terms of the NPIINQ opposition which I will rephrase here as existential nonnegative
versus (universal) negative for NPI and NQ respectively):
(i) The semantic representation of NIs involves an analysis of them as indefmites
in the strict Heimian sense. Just like prototypical indefinites, i.e. singular indefmites
such as a student, they do not really have any quantificational force of their own but,
rather, they contribute (a) a variable, and (b) a descriptive condition that has to be met
by that variable:
(1)
a
b
Ia student I =>
Inessuno I
student' (x)
=> person' (x)
WEAK AND STRONG POLARITY: EVIDENCE FROM GREEK
115
Under this supposition, the attested parametrized meanings of NIs (existential

nonnegative/universal negative) are viewed as the result of the composition with
different licensers (negative or nonnegative).
(ii) NIs are systematically ambiguous between existential nonnegative and
universal negatives interpretation. This ambiguity can be captured if we assume that
the two readings correspond to two distinct If-formatives3 In view of the fact that lfs
are conservative, this claim intimates a relaxed version of conservativity which allows
for cases in which one s-structure unit is mapped onto two (or possibly more) Ifexpressions. The English item nobody, for instance, is ambiguous between nobody [+]
(the strong construal) and nobody [-] (the weak construal). Mapping onto nobody [+]
results to the negative interpretation. Mapping onto nobody [-] brings up the existential
nonnegative interpretation attested in non-standard English.
(iii) Given the ambiguity of NIs between the negative and nonnegative readings,
NC is reduced to negative polarity. Languages with NC differ from languages without
in that the former allow for the existential interpretations of their NIs whereas the latter
do not.
(iv) It is plausible to invoke two distinct mechanisms for the licensing of the strong
and the weak construal respectively. Ladusaw hypothesises that the strong construal
may be seen as licensing via Spec-Head agreement (in the spirit of the NEG-criterion as
conceived of in H&Z 1991 and Zanuttini 1991) while the weak construal can be
regarded as licensing via roofing by an operator of the appropriate semantic type.
Being roofed by an operator is a synonym for being (unselectively) bound by that
operator. Although Ladusaw is not clear with respect to this point, I assume that he
takes binding to be represented as some kind of co-indexation. This way, the NQ and
the NPI readings are derived as products of two distinct syntactic mechanisms.
3. RESOLVING THE AMBIGUITY IN GREEK
In Greek we find a series ofNIs reproduced under (2):
(2)
kanenas / KANENAS
tipota / TIPOTA
pote/POTE
puthena / PUTIIENA
'anyone, anybody' fno one, nobody'

'anything' / 'nothing'
'ever' fnever'
'anywhere' / 'nowhere'
These indefInites can either bear emphatic stress (the capital-items) or not (the lowcase items). It has been observed (Veloudis 1981, Giannakidou 1993, Quer 1993, Klidi
1994) that only the emphatic items can answer in isolation with the meaning of no N,
which suggests that only these expressions are interpreted as universal negatives. In
addition, emphatic items require the co-occurrence of sentential negation in order to be
licensed, in any other context they are ill-formed:
ANASTASIA GIANNAKIOOU
116
(3)
(4)
KANENAS *(dhen) ipe

TIPOTA
nobody
NOT said-3sg nothing
'Nobody said anything'
ipes * TIPOTA?
said-2sg nothing
(3) shows that Greek exemplifies the pattern of what I call strict negative concord, i.e.
the variety of negative concord which always requires the presence of the sentential
negative marker.4 As we see, lack of sentence negation dhen rules out the occurrences
of the emphatic indefinites. On the other hand, the items of the non-emphatic paradigm
are licensed in an ample variety of constructions, sentence negation being one of them.
Besides negation, the list includes yes/no and rhetoric wh-questions, the first argument
of universal quantifiers and the scope ofMD DPs in general, the CP argument ofMD
predicates, the comparative, the superlative, prin 'before' and xoris 'without' clauses,
subjunctive main and complement clauses, imperatives, the scope of modal verbs and
the habitual (see Giannakidou 1993, 1997. for a detailed description of the data).
(5)
(6)
(7)
(8)
(9)
thelo
na pjo
* KAMJA BIRITSA / karnja biritsa
want-1sg SUB drink-1sg any beer
'I want to drink a beer'S
rotise
* KANENAN IDHIKO / kanenan idhiko
ask-2sg-IMP any specialist
'Ask a specialist'
idhes
kanenan sto parko?
INTER saw-you anyone in the park
'Did you see anyone at the park?'
perna
pu ke pu
*KANENAS FITITIS / kanenas fititis
pass-by-3sg where and where any student
taproina
the mornings
'In the mornings, a student would pass by every now and then'
0 Janis
arnithike oti idhe *KANENAN / kanenan
the-John denied-3sg that saw-3sg anyone
'John denied that he saw anyone'
This remarkable disparity in the distribution of the items in discussion has led to the
conclusion that in Greek emphatic stress draws the demarcation line between the
universal negative (V'x) and the existential nonnegative (3x) interpretation. The
emphatic paradigm corresponds to the former, the nonemphatic to the latter.
Nonemphatic indefinites contribute existential quantifiers in all nonveridical
environments as shown in (10) where NONV stands for the nonveridical operator:
(10)
I. KANENAS N 1=> AP ( 1N 1n P = 0) corresponding to V'x (x:N)

1NONV kanenas N 1=> AP ( 1N 1n P * 0) corr. to NONV 3x (x:N)
117
The distinct semantic import of KANENAS (no one) and kanenas (anyone) can be
detected in a number of assymetries between the two which may obtain even under
negation. I will consider here some of them.
One asymmetry concerns modification by almost and absolutely. Since Dahl
(1970) and Hom (1972), almost/absolutely modification is used to diagnose universal
quantifiers, or at any rate, high scalar values. 6 Observe that almost cannot modify
existentially quantified DPs like something in (lla) but it can modify universally
quantified DPs like everything in (lIb):
(11)
a.
b.
*Electra was willing to accept almost something

Electra was willing to accept almost everything
The emphatics can be modified by almost/absolutely but the nonemphatics cannot:

(12)
dhen idha shedhon KANENAN / *kanenan

not saw-Isg almost nobody/ anybody
'I saw almost nobody'
KANENAN in (12) is translated as "nobody". The contrast suggests that, under

negation, emphatics are being interpreted as a universal negatives (\7'-,) whereas the
nonemphatics are interpreted as an existential nonnegatives (-,3). Of course, quantifier
negation will ultimately render the structures containing either item truth conditionally
equivalent, but in the scope of operators other than negation (cf. (5)- (9 the existential
nonnegative interpretation of kanenas is more salient.
A second asymmetry involves modification by intensifiers of existential
quantifiers such as ke "and". This item, comparable to Dutch ook maar "too + focus
particle" and German auch nur "too + only" only combines with existential quantifiers,
as is illustrated in (13):
(13)
010 ke kapjos/*kathenas
erhete to proi
all and someone /everyone come-3sg the moming
'Someone usually comes in the moming'
ke is grammatical with the nonemphatics but ungrammatical with the emphatics:

(14)
dhen ipe
ke tipota / *TIPOTA spudheo
not said-3sg and anything! nothing important
'He didn't see anything important'
The contrast in (14) can be understood as a contrast between an existential and a

universal quantifier in terms of their compatibility with ke.
Finally, evidence in favour of the distinct interpretation of emphatic and
nonemphatic items comes from donkey-anaphora. Nonemphatics support donkey
anaphora and exhibit all the empirical characteristics of it aligning thereby with
indefInites such as kati "something" in (15a). Emphatics do not support donkey-
118
anaphora. They display all the blocking effects we observe with "real" quantifiers such
as every and no. For example, in (15a,b) we see that the nonemphatics can establish
anaphoric links from a relative clause, thus from a non c-commanding position, while
the emphatics cannot:
(15)
a.
1 fitites
pu ehun
tipo~ / katl;
na pun,
the students that have-3pl anything/something subj say-3pl,
as to;pun
tora
let it say-3pl now
'The students that have anything (something) to say can say it now'
b. *1 fitites pu dhen ehun TIPOT~ na pun, as to; pun tora
(* The students that have nothing to say, let them say it now)
Moreover, in (16), we observe that the nonemphatic can be modified by a pseudorelative clause which contains a pronoun anaphoric to it. This is not an option for the
emphatic (the example is from Quer 1993):
(16)
Mi fas
tipo~
ke se piraksi
pro;
not eat-imp-2sg anything and you upset-3sg
Don't eat anything; in case i~ upset your stomach'
b. *Mi fas TIPOTA ke se piraksi
(*Eat nothing;; i~ might upset your stomach)
a.
Third, in (17a), the nonemphatic scopes over a conjunct, just like indefinites do, but
the emphatic item cannot and the anaphoric link between the emphatic and the
pronoun in (17b) is ill-formed.
(17)
a.
Illektra tha aghorasi kanena vivlio; ke i Kleo tha to; katastrepsi

the Electra fut buy-3sg any
book and the Cleo fut it destroy-3sg
'Electra will buy a book and Cleo will destroy it'
b. *1 Ilektra dhen tha aghorasi KANENA vivlio; ke i Kleo tha to;
katastrepsi
(* Electra will buy no book and Cleo will destroy it)
The pronouns linked to the nonemphatics are inteIpreted just like the pronouns linked
to indefinites in the regular cases of donkey-anaphora (possibly as E-type although
nothing crucial for our discussion hinges on this).
In view of the contrasts above, 1 conclude that emphatic and nonemphatic items
receive semantically distinct inteIpretations. 7 It can also be shown that these
expressions are also syntactically different. First, nonemphatics require that negation
(as well as any other licensing operator) precede them at s-structure. Emphatics do not
pose such a requirement and they may precede negation at s-structure. In this respect,
emphatics behave in tandem with bare NPs used in the so-called minimizers such as
vazo bukia 'eat a bite' which, crucially, also bear emphatic stress:
(18)
119
a.
KANENAN dhen idha (KANENAN)

nobody
not saw-I sg
'I saw nobody'
b. BUKIA dhen evale sto stoma tu
(BUKIA)
bite
not put-3sg in-the mouth his
'He didn't eat a bite'
c. *kanenan dhen idha
The second difference concerns sensitivity to island constraints. Emphatics, just like
wh-pbrases, are shown to be sensitive to syntactic islands such as, inter alia, relative
clauses (cf. (20 and sentential adjuncts (cf. (19. Nonemphatics do not display such
sensitivity and are freely licensed there. The examples below provide a partial
illustration. The ungrammaticality of the c-sentences suggests that overt preposing of
the emphatics is also ungrammatical:
(19)
(20)
a.
Dhen itan
isihi epidhi fovithike
kanenanl *KANENAN
not was-3sg quiet because was-scared-3sg anyone
'(S)he wasn't quiet because (s)he was scared of anybody'
b. *PjOl1; dhe itan
isihi epidhi fovithike t j?
who not was-3sg quiet because was-scared-3sg
c. *KANENANj dheh itan isihi epidhi fovithike ~
a. Dhen prodhosa
mistika pu eksethesan kanenan
not betrayed-lsg secrets that exposed-3pl anybody
'I didn't reveal secrets that exposed anybody'
b. * PjOl1; dhen prodhosa mistika pu eksethesan ~?
c. * Kaneflall; dhen prodosa mistika pu eksethesan ~
In face of the contrast above, it is feasible to invoke a movement analysis for emphatic
items and an in situ analysis for the nonemphatics.

A related difference, further illustrating the point that nonemphatics may be
licensed in situ, concerns the licensing of these items long distance. Nonemphatics are
typically licensed by superordinate negation in embedded clauses but emphatic items
are not. Exceptional licensing of emphatics long distance occurs with predicates that
induce tense or world dependency like want but I cannot go into this here (see
Giannakidou & Quer 1995, to appear for an extensive discussion):
(21)
I Ilektra dhen ipe

oti idhe
tipotal*TIPOTA
the Electra not said-3sg that saw-3sg anything
'Electra didn't say that he saw anything'
Given the semantic and syntactic differences between emphatic and nonemphatic
indefinites, I will conclude that these expressions form two distinct paradigms. I will
consider emphasis a quasi-morphological feature which reflects the two different
"meanings" involved. Below, I will argue that the two different "meanings" correspond
to two different syntactic representations.
120
4. THE PROPOSAL
Building on Ladusaw's insights for negative licensing, I put forward the following
proposal for the licensing ofNIs:
(i) NIs are heimian indefmites, hence predicate-and-variable contributing
elements with no quantificational force of their own.
(ii) There exist two types of licensing, weak and strong. NC is an instance of
strong polarity licensing. The two types of licensing correspond to different semantic
structures. When NIs are licensed strongly, their variable is bound in the restriction by
the relevant operator and they are interpreted as universal negatives (V'x-,). The
binding of the NI-variable is illustrated (22a), where binding is represented, as usual, in
terms of co-indexation. In weak licensing, the NI-variable is existentially closed in the
scope of the operator yielding thereby the corresponding existential interpretation for
the NIs as we see in (22b):8
(22)
a.
b.
STRONG OPx (restriction" .x...)

WEAK OP [stope.. 3x...]
[scope Main Predication]
(Strong)
(Weak)
These representations imply that when OPs license weakly they are not
quantificational, whereas when they license strongly they are. Hence the pattern
restriction-scope is available only in the latter cases. The intuition behind the
quantificational / nonquantificational contrast postulated here can also be stated in
terms of the categorical (quantificational) versus thetic (nonquantificational)
distinction as addressed in Kuroda (1992). For the moment, I use the neutral terms
weak and strong to refer to the relevant operators. I will show below that, from the
semantic point of view, it is possible to properly defme classes of operators which can
license strongly, weakly or both strongly and weakly (as we have noticed already from
the Greek facts, negation is such an operator).
(iii) There is a direct mapping between syntax and semantics in the sense of
Ladusaw (1994) where a correspondence is established between the syntactic specifier
and the quantificatiOnal restriction on the one hand, and between the syntactic
complement and the scope on the other.9 Movement for the satisfaction of the NEGcriterion is taken to be the syntactic analogue of strong licensing. I assume here the
version of the NEG-criterion put forth in Zanuttini 1991 and H&Z 1991 given in (23)
and I regard it as a condition applying either at s-structure or at LF. Given that in
Greek preposing of emphatics is optional, I assume that in this language the NEGcriterion is satisfied at LF. Overt preposing (cf. (18 may be treated as an instance of
topicalization (see Giannakidou 1997 for arguments and discussion): 10
121
The NEG-criterion (Haegeman & Zanuttini 1991, Zanuttini 1991)

a. A NEG-operator must be in Spec-head agreement relation with an XO
[NEG]
b. An XO [NEG] must be in Spec-head agreement relation with a NEGoperator.
(23)
Where the following defInitions obtain:

c.
d.
NEG-operator: a negative phrase in a scope position;

Scope position: left-peripheral A'-position [Spec,XP] or [YP,XP].
Weak licensing, on the other hand, is in situ licensing via application of existential
closure under the relevant operator which must c-command the NI at some syntactic
level. In other words, the universal negative interpretation of NIs is taken to be the
semantic reflex of syntactic movement, whereas the existential interpretation suggests
that no such movement has occurred.
The bulk of the present paper illustrates how this proposal can extend to the whole
class of NPls and account for the diversity problem in Greek. I will not insist on the
syntactic details for reasons of space. I will focus on the distribution and interpretation
ofNPls aiming towards a semantic characterization of the licensers.
4. 1 A Hierarchical Typolofy of Triggers Based on Monotonicity

What I call, following Israel (1994), the diversity problem refers to the fact that not all
NPls are licensed in the same environments. Which items will be allowed for in which
environments is subject to a great number of subtleties and sometimes one tends to
think that the distribution of NPls is highly idiosyncratic therefore no valid
generalizations can be drawn. More optimistic researchers assume that the
dissimilarities can be captured once we appeal to the relative strength of different NPls
as combined to the relative strength of the licensers. Yet, it is rather unclear whether
NPls can be neatly ordered from weak to strong or whether the diverse spectrum of
triggers can be reduced to one dimensional gradient of licensing force.
Zwarts (1993) attempted to resolve the problem by developing a hierarchical
typology of NPls based on the restrictions concerning the cooccurrence with a MD
licenser. According to Zwarts (1993) there are three types of triggers, ordered along a
dimension of strength which is derived from the number of the DeMorgan relations
they satisfy: (i) MD, the general set of triggers in the sense of Ladusaw (1980), (ii)
antiadditive (AA), and (iii) antimorphic (AM) triggers. MD expressions satisfy the fIrst
and fourth of the DeMorgan relations and are the vehicle of minimal negation. AA
phrases satisfy the fIrst, second and fourth of the DeMorgan relations and convey
regular negation and, fInally, AM expressions are the typical instances of classical
negation; they denote set-theoretic complementation and satisfy all four DeMorgan
relations:
122
(24)
monotone decreasingness:
antiadditivity:
antimorphicity
(a) f{X u Y) ~ f{X) n feY)

(b) f(X) u f{Y) ~ f(X n Y)
(a) f(X u Y) B f(X) n feY)
(b) f{X) u feY) ~ f(X n Y)
(a) f{X u Y) B f{X) n f{Y)
(b) f(X n Y) B f(X) u f(Y)
AA functions form a subset of MD functions and AM functions form a subset of AAs.

Expressions such as few N, at most n N, deny and forget in English denote MD
functions. no N, before and constituent negation not obey the AA pattern. Sentence
negation and without NP / without S are AM. Zwarts (1993) postulates that there are
three types of polarity items and motivates the following hypothesis:
(25)
(i) Weak NPIs occur in the scope ofMD operators.

(ii) Strong NPIs occur in the scope of AA operators.
(iii) Superstrong NPIs occur in the scope of AM operators.
(25) can successfully accommodate data like (26)-(29) in Dutch and English:
(26)
(27)
(28)
(29)
*Weinigl*hoogsteens zes kinderen hebben ook maar iets bemerkt

Few children / at most six children noticed anything
Geen kinderen hebben ook maar iets bemerkt
No children noticed anything
a. De kritiek was niet mals
the criticism was-3sg not soft
The criticism was not tender
b. * Geen kritiek / De kritiek was mals
a. * Few people lifted a finger
b. NoonelHe didn't lift(ed) a finger
The Dutch oak maar iets is characterized as a strong NPI because it requires an AA
trigger in order to be licensed, hence the grammatical (27). (26) fails to license the oak
maar indefmite due to the absence of any AA trigger. maTs is a superstrong NPI since
it is well formed only in the scope of sentence negation. As shown in (28b), AA
constructions cannot host maTs. The same rationale applies to (29a,b).
Things are not so c1earcut however. First of all, from the logical point of view, the
connection between the DeMorgan laws and degrees of strength of negation is not an
uncontroversial issue, and an appeal to the former in order to identify the later is not
entirely justified (for recent criticism see Atlas 1996). From the linguistic point of
view, Zwarts's hypothesis, proposed as a universal hypothesis, meets a number of
serious difficulties.
First, examination within and across languages suggests that a typology based on
monotonicity is inadequate on empirical grounds. If we attempt to extend Zwarts's
hierachy in Gr, for instance, empirical problems arise in view of the fact that
imperatives, subjunctives, the protasis of conditionals, questions and habituals, to
123
mentioned just some of the licensing environments, allow for NPls without being MD
in any obvious sense (cf. Giannakidou 1994, 1995 for criticism). This is not only a
problem with respect to Greek but also with respect to English, and as I show in (30)
with respect to Dutch:
(30)
a.
b.
Heb je ook maar iemand gezien?

Did you see anybody?
Als je ook maar iemand ziet.. ..
If you see anyone ....
Monotonicity is a notion of inference and it is far from clear how it applies to

nondeclaratives such as interrogatives and conditionals, if at all. We cannot
characterize the question in (30a) as MD, let alone as AA, as it would be required in
Zwart's terms. The licensing of ook maar iemand in (30a) is therefore not explained
under Zwarts's account, unless recourse to circular reasoning is made: interrogatives
are AA because they license ook maar indeftnites (which is in fact a claim made in van
der Wouden 1994). The problems with monotonicity patterns in conditionals were
pointed out in Heim (1984), and given the discussion in that study there is no
explanation for the grammaticality of ook maar iemand in (30b) within Zwarts's
framework either.
Second, the split between antiadditivity and antimorphicity is not strongly
motivated on the empirical side. Antiadditivity can be dispensed with when it comes to
negative concord languages like Gr which, as we have seen, strictly require the
presence of (antimorphic) sentence negation to license AA readings like no N.
Moreover, in English and in Romance, most NPls which are licensed in AM contexts
are licensed in AA contexts as well (He didn't budge an inch, Ifyou budge an inch I'll
kill you, He didn't say a word, Before you say a word listen to me).
Third, the difference in the interpretation between types of NPls as well as the
syntactic means of expressing these differences are completely ignored. "Occurring in
the scope of" is just a semantic requirement and it makes no appeal to syntactic or
interface conditions.
Finally, there is a conceptual problem with Zwarts's typology: monotonicity
provides no room for addressing the issue of the licenser-licensee relation, that is, the
relation between NPls and their triggers. It is not clear why monotonicity should be
relevant for NPI-licensing.
To sum up, although Zwarts's typology contributes to our understanding of how
degrees of MD affect NPI-licensing in Dutch, it does not give us the whole picture
either in Dutch or crosslinguistically, thus it cannot supply the basis for a
comprehensive theory of negative polarity (cf. Krifka 1994 for analogous reasoning).
Alternatively, a typology based on nonveridicality can be more successful.
124
4. 2
A Typology of Negative Polarity Based on Non-veridicality
As I mentioned in the beginning of this paper, I take it here that affective licensing, i.e.
NPI-licensing, is regulated by the condition in (31) (see Giannakidou 1997 for
extensive discussion)
(31)
Licensing condition on NPIs

NPIs are grammatical if they are in the scope of nonveridical operators.
(31) states that NPls are grammatical in nonveridical domains, but ungrammatical in
veridical ones. Montague (1969) talks about veridicality in an attempt to characterize
the semantics of perception predicates like see which entail the existence of the
individuals involved in their complement proposition. Because if I see a student
running, I also see a student, therefore a student exists, see is veridical. Zwarts (1986,
1995) builds on Montague's observations and puts forth the definitions we see in (32):
(32)
Definition
Let OP be a monadic sentential operator. The following statements hold:
(i) OP is veridical just in case OP p -+ p is logically valid. Otherwise,
OP is nonveridical: OP p -1-+ p.
(ii) A nonveridical operator OP is averidical just in case OP p -+ -, p is
logically valid.
A propositional operator OP is veridical iff OP entails p, that is, an operator OP is

veridical iff whenever OP p is true, p is true too (where p is an arbitrary proposition).
OP is nonveridical iff OP does not entail p, i.e. iff whenever OP p is true, p is not
necessarily true. Note that nonveridical operators do not entail the falsity of p.
Entailing the falsity of p is the defining property of averidical operators. As conceived
of in (31), averidical operators form a subset of the nonveridical, so every averidical
operator is also nonveridical but not vice versa. The defInition in (32) can be extended
to dyadic propositional operators too, but I will not deal with the extension here. II
Adverbs such as yesterday obey the veridical pattern:
(33)
Yesterday, John discovered a snake

John discovered a snake
-+
Predicates that license subjunctive in Greek and Romance (volitive, directive, modal),
imperatives, the habitual, questions, negative predicates, MD (not AA) NPs and factual
conditionals are nonveridical but not averidical. 12 Note in passing that the question
operator and the factual ifalso license subjunctive in their domain:
(34)
a.
thelo
na horepso -/-+horevo
want-lsg subj dance-Isg
dance-lsg
'I want to dance'
b. Open the door!
-1-+ you open the door
c.
d.
e.
Did you see him?

Ifhe comes by
John denied that he left
125
-/~ you
saw him
comes by
-/~ he left
-/~he
Sentence negation, without, AA NPs, counterfactual conditionals are averidical:

(34')
a.
b.
c.
d.
He didn't come
without looking at me
No woman strolls
If it had rained
It is not the case that he came

It is not the case that he looked at me
~ It is not the case that women stroll
~It is not the case that it rained13
~
~
A close examination of the distribution ofNPIs in Greek shows that the only option for
NPIs triggered by nonveridical operators is to contribute existential quantifiers.
Expressions corresponding to the strong construal and minimizers are not acceptable in
non-veridical environments. Hence the ungrammaticality of KAMJA BIRITSA, pjo
ghoulja 'drink a drop', pja 'anymore' and ipe lruvenda / vghali ahna 'say a word' in the
examples below:
(35)
(36)
(37)
(38)
(39)
a.
thelo
na pjo
*KAMJA BIRITSAI * ghoulja / kamja biritsa
want-lsg subj drink-lsg no beer/ drop / any bier
'I want to drink a beer'* I want to drink a drop
b. * pjes ghoulja
*Drink a drop!
perna
pu
ke pu
*KANENAS FITITIS / *pja lkanenas fititis
passes-by where and where no student! anymore/any student
ta proina
the mornings
'In the mornings, a student would pass by every now and then'
0 Janis
arnithike oti ipe *TIPOTA / *kuvenda / tipota
the- John denied that said nothing! word / anything
'John denied that he said anything'
LIn fitites ipan *TIPOTA / tipota
few students said nothing! anything
'Few students said anything'
an erthi kanenas / *KANENAS / * vghali AHNA, idhopiise me
if comes anyone / *noone / *takes-offbreath, let me know
'If anyone comes let me know'
*'Ifhe says a word, let me know'
There seems to be a correlation between nonveridical licensing and the exclusive

availability of the existential interpretation for the licensees. In terms of the proposal
put forth in this paper, this amounts to saying that nonveridical operators can only
license weakly, that is, NPIs can only enter their scope. This motivates the hypothesis
in (40):
126
(40)
(a) Nonveridical OPs license NPIs only weakly.

(b) Averidical OPs license NPIs strongly and/or weakly.
(c) In Greek, emphatic stress flags the performance of strong licensing
and lack thereof signals weak licensing
That (a) holds has already been shown. What about (b)? Let us take three averidical
operators: sentence negation, hons 'without' and counterfactual conditionals. In Gr
emphatic as well as nonemphatic indefinites, minimizers, KAN "not even" and
superlatives are all licensed with the characteristic emphatic stress:
(41)
(42)
(43)
(44)
(45)
(46)
dhen vlepo POTE KANENAN I * POTE kanenan I *kanenan

not see-lsg never noone I never anyone I anyone
'I never see anyone'
b. *(Dhen) irthe
KAN
not
came-3sg even
'He didn't even come'
dhen patise
PSIHI I eho STON ILIO MIRAleho KUKUTSI mjalo I
not stepped-3sg soul lhave-lsg in-the sun destiny lhave-lsg pit brains I
evala BUKJA sto stoma mu
oli meraleho TIN PARAMIKRI idhea
put-Isg bite
in-the mouth mine all day lhave-lsg the smallest idea
'Not a living soul arrived I I have no place under the sun I I don't have a
grain of sensell haven't had a bite all dayll don't have the faintest idea'
evale tis fones horis KAN na tu kejete KARFIja to ti
for what
put-3sg the voices without even subj him burn nail
tha pi 0 kosmos
fut say the world
'He started screaming without giving a damn about what people
would say'
horis
KAMIA PROKATALIPSI I horis kamja pseftodhikeolojia
without no prejudice I without any false excuse
'Without any prejudice anymore I without any false excuse'
an vghalis AHNA I pis tipota I*TIPOTA 1* KAN tha se skotoso
'If you breath a sound/lf you say anything you're dead'
an ihe KUKUTSI mjalo/an me *KANENAN/*KAN I *pja kanenan...
if had-he pit brains
'Ifhe had a grain of sense... .'
a.
Similar examples abound. It was shown in the previous section that nonemphatic
indefmites are interpreted weakly under negation. The examples above suggest that
this observation be extended to averidical contexts in general. On the other hand,
emphatic items and minimizers correspond to the strong construal and they are
interpreted as strong, emphatic expressions. Note that there are more rigid restrictions
on the distribution of emphatic indefinites and KANN not even' than on the
distribution of minimizers. As we see in (45), (46), this set of items is excluded from
conditionals altogether. The reason may be a morphological ban which requires that
127
these items be triggered by an averidical expression which bears overt negative

morphology. Counterfactual conditionals or conditional-like prohibitions (cf. (45 do
not meet this requirement since they are not overtly negative. But we can also provide
a more semantic explanation. If we highlight the interpretation of NPls when licensed
in conditionals in general, we see that only the possibility of the existential reading is
available. The application of the almost/absolutely test can verify that. In the same
vein, modification by ute 'neither' (for universal quantifiers) and esto ke 'even' (for
existential quantifiers) can be used as another convincing diagnostics:
(47)
(48)
dhen ipe
ute
leksi
not said-3sg neither word
'He didn't say a word'
an pis
esto ke mia leksi / * ute leksi tha se skotoso
if say-2sg even and one word! neither word fut you-ace kill-I
'If you say even a word I'll kill you'
Modification by esto ke is in principle operative under negation too. This squares with
the fact that negation is a double trigger. Conditionals are not like negation in that they
only permit existential readings for the NPls in their scope. Before is like conditionals
whereas without behaves in much the same way as sentence negation:
(49)
(50)
prin pis esto ke mia / * ute leksi ...

Before you say a word...
horis na pis esto ke mia / ute leksi ...
Without saying a word....
Finally, the contrast between minimizers and emphatics can be explained by saying
that the later are licensed only when averidicality comes from an entailment and not
from an implicature (cf. footnote 12).
Using the interpretational effects as the guide for the characterization of NPls, I
propose the following classification:
(51)
(a) There are three types ofNPls: weak, strong, superstrong.

(b) Weak NPls are weakly licensed by nonveridical OPs. They are only
interpreted as existential quantifiers.
(c) Strong NPls (i) are licensed weakly or strongly by averidical OPs.
They are interpreted as existential (weak licensing) or as universal
(strong licensing).
(d) Superstrong NPls are only strongly licensed by averidical OPs. They
are interpreted as universal quantifiers.
(51) partitions the class of Greek NPls in the following way. Nonemphatic indefmites
are weak NPls. They are licensed in all nonveridical contexts and they always
contribute 3x. Minimizers are strong NPIs since they are only licensed in averidical
contexts and obtain both interpretations (3x, \fx). Emphatic indefmites and KAN 'not
ANASTASIA GIANNAKlDOU
128
even' are superstrong NPIs. Whenever the Vx interpretation is not accessible, they are
ruled out. (51) can naturally be extended to characterize the corresponding classes of
NPIs (at least) in English, Italian, French, Spanish and Catalan. Furthennore, (51) can
also be effective in dealing with long-distance licensing (for an analysis see
Giannakidou & Quer 1995).
A fmal remark. The split between weak and strong licensing can be understood in
relation to a distinction argued for in Israel (1994) between emphatics and
understaters. Sentences with emphatic NPIs produce strong/general statements
whereas sentences with understaters constitute weak statements. Essentially, weak
licensing as conceived of in this paper gives rise to weak, non-general statements while
strong licensing yields strong, general statements.
5. AN INTERESTING CONSEQUENCE
The postulation of strong and weak polarity depending on whether licensing takes
place in the restriction or in the scope of an operator has some important consequences,
the most important of which is that it makes predictions with respect to the
presuppositionality of the DPs involved. It is generally assumed that the restriction but
not the scope is the target of presuppositions, hence a corollary of the theory developed
in this paper seems to be that NPIs licensed in the scope will not be presuppositional
while NPIs licensed in the restrictor will be. I examine this issue here by focusing on
emphatic and nonemphatic DPs.
That weakly licensed DPs are not presuppositional becomes apparent once we
recall that they can never be referential. This follows from their being NPIs: they have
to be interpreted in the semantic and syntactic scope of the licensing operator. The
question is: can we justify the claim that KANENAS, under the assumption that it is
licens~d in the restriction, is presuppositional? Consider the negative sentence in (52a)
and its (thetic) representation in (52b):
(52)
a.
Dhen irthe
kanenas fititis
not came-3sg any
student
'No student came'
b. ....,3e (3x (student'(x) at e) A came' (x, at e
(Weak licensing)
(52b) states that there were no coming events that involved students and it can be given
as a successful answer to a question like What happened? If the question is fonnulated
in such a way as to include a subject, i.e. What about the students? then only the use of
the emphatic is felicitous, and the answer is understood as a negative claim about a set
of students (cf. (53b:
(53)
a.
Dhen irthe
KANENAS FmTIS
not came-3sg any
student
'Nobody came'
b. ...., (3x (student'(x [came' (x)]
(Strong licensing)
129
But is the existence of students presupposed in (53a)? Can the sentence be taken as a
quantificational claim presuming that there is a set of students such that none of them
came? The answer is yes, but not necessarily so. (53a) can be true in a situation in
which no comings occur. So KANENAS FITITIS is not presuppositional in the sense of
denoting a set of students whose existence is taken for granted.
However, as noted in Philippaki & Veloudis (1984), KANENAS can be left
dislocated and co-indexed with a clitic contrary to kanenas for which none of these
options is available:
(54)
a.
b.
KANENOS j / *kanenos dhen ~ aresi i kakometahirisi

noone-gen
not cl-gen like the bad-treatment
dhen ~ aresi KANENOS j / *kanenos i kakometahirisi
'Nobody likes to be treated badly'
This suggests that they are specific indefinites which can be topicalized by left
dislocation (more on this in Giannakidou 1997.). I propose that emphatic indefInites
are "presuppositional", in the sense that they are the subjects of categorical statements.
This characterization underlies the representation in (53b) and the contrast between
(52) and (53). Statements expressing categorical judgement are structured statements.
They consist of a subject (corresponding to what is called topic in the theories about
the informational structure of sentences) and a predicate, such that given the subject,
the predicate tells something about it. Statements expressing thetic judgement are
structureless, they simply communicate events. Thinking in these terms, we can say
that in Greek, emphasis distinguishes between the two modes of negating. Negating
with an emphatic gives rise to a categorical statement , whereas negating with a
nonemphatic results in a thetic judgement. 14
Additional evidence comes from generic negative sentences:
(55)
a.
b.
KAMJA AJELADHA / *kamja ajeladha dhen troi kreas

no cow / any cow
not eat-3sg meat
'No cows eat meat'
dhen troi kamja ajeladha kreas
not eat-3sg any cow
meat
'No cow is eating meat (right now)'
(55a) and (55b) have the representations in (55a)' and (55b)' respectively:
(55)'
a.
b.
GENx,s (cow'(x) in s) [. eat-meat' (x, in s)]

.3x (cow'(x) 1\ eat-meat'(x
(55a) can only be interpreted as the generic statement in (55'a). The nonemphatic
indefInite is ruled out because it is not c-commanded by negation at s-structure (cf.
section 3, (18. Crucially, in (55b) where the right c-command relation obtains, the
nonemphatic indefinite is licensed but then only the eventive reading is available as we
130
ANASTASIA GIANNAKlDOU
see from (55'b). The sentence is true under the following scenario: being an
incredulous person who does not believe that cows do not eat meat, I conducted an
experiment in my friend John's farm by exposing his cows to meat. On the 15th of
June, at four 0' clock I checked John's cows and to my disappointment none of them
was eating meat. Evidence from Japanese (Kuroda 1992) shows that generic
statements are categorical statements. Generic subjects are suffIxed by the morpheme
which generally marks categorical subjects.
In the light of the above considerations, I conclude that emphatic indefInites can
be viewed as topic markers in negative sentences. Once we associate the notion of
topic with aboutness and pragmatic referentiality (Strawson 1954, Reinhart 1982) we
have a natural way to understand the "presuppositionality" of these indefInites.
To conclude, this paper has provided direct evidence for Ladusaw's (1994)
conjecture that two syntactic mechanisms are involved in NPI-licensing. Building on
Ladusaw's insights as confirmed from Greek data, I have proposed a theory of weak
and strong NPI-licensing which combines Ladusaw's conjecture with the proposal I
made in Giannakidou (1994, 1995) that NPI-licensing is sensitive to nonveridicality.
Negative concord has been treated as an instance of strong polarity. The analysis
pursued here makes a great number of crosslinguistic predictions with respect to the
diversity problem and I am confIdent that it will yield fruitful results when applied to
languages other than Greek.
NOTES
I want to express my gratitude to a number of people for encouragement and comments: Elena
Anagnostopoulou, Donka Farkas, Jack Hoeksema, Sabine labidou, Sila Klidi, Bill Ladusaw, Jason
Merchant, Josep Quer, and Frans Zwarts. My thanks also to the audience of the Workshop on Greek Syntax
in Berlin and to Artemis Alexiadou for organising an exciting Workshop and for giving us aJl the chance to
be there.
The term NPI-licensinge is to be understood in the general sense, as the licensing of what Klima
(1964) chacacterizes as affective items which are subject to a negation related (but not necessarily
negative) structural dependency.
3
We should not confuse if with LF. if stands for the input material to semantic interpretation, it is the
structure to be semanticaJly interpreted. As Ladusaw puts it, lfs are conservative in that, to the extend
possible, there is a one-to-one correspondence between s-structure formatives and the formatives oflf.
4
WeJl-known examples ofsbict negative concord languages are the languages of the Slavic family (cf.
Progovac 1988, 1994 for Serbian/Croatian and references therein).
S
The distributions of the Greek nonemphatic indefinites and any in English do not overlap, although
they do intersect to a great extend. In subjunctive clauses parallel to (5), any may be licensed but it receives
the so-caJled free-choice interpretation:
(i) I insist he allow anyone in
On the other hand, any is licensed in emotive factive complements, which are veridical, and which do not
aJlow for Greek nonemphatics (or emphatics):
(ii) *Metanjosa pu milisa me kanenan/KANENAN
'I regret that I talked to anybody'
Although any and the Greek NIs are polarity items in the sense that they are not grammatical everywhere
but they have restricted distribution, it is plausible to assume that they are not subject to the same kind of
semantic dependency. The acceptability of any in (ii) clearly points out towards the conclusion that the
131
licensing of any involves some additional mechanism (possibly a negative implicature a la Linebarger
1980) which is not operative in this particular Greek case. I cannot deal with the issue here, but the reader
should keep in mind that the area of polarity sensitivity does not form a homogeneous domain. The
semantic condition on nonveridicality can be given as a general semantic condition, but auxiliary
conditions, directly or indirectly related to nonveridicality, should be postulated to account for individual
classes of items across languages (Nls in Greek, n-words in Romance, any), and, as I will show below,
within languages (cf. Giannakidou 1997 for a fully-fledged analysis along these lines).
6
But see Hom and Lee (1995) for a more critical view on this diagnostics. I will not deal with the
possible objections here as the main goal is to simply point out the different behavior of emphatics and
nonemphatics.
7
Another difference between emphatics and nonemphatics includes the possibility to use them as
predicate nominals:
(i) Dhen ine kanenas/* KANENAS idhikos
He is no specialist
Space prevents me from elaborating on this issue.
8
The proposed representations for Nls come as a consequence of their being indefinites. Recall that
similar suggestions proliferate in the literature on bare plurals and singular indefinites. As is well-known,
bare plurals and a Ns give rise to two distinct interpretations, one universal (or generic) and a second
existential. Krifka et al. 1995, Wilkinson 1991, Condoravdi 1994, among others, propose that the switch
from one interpretation to the other can be captured by means of postulating distinct semantic structures.
The universal/generic reading is taken to reflect licensing of the indefinites in the restriction whereas the
existential reading is derived from their being licensed in the scope of the generic operator.
9
The mapping suggested here is not to be confused with the mapping hypothesis advanced in the work
of Diesing (1992).
10
The NEG-criterion should be viewed as one instance of the more general AFFECT-criterion (cf.
Rizzi 1991, Haegeman 1995).
II
Nonveridicality is defined here on propositional operators, and in this sense, a condition like (31)
subsumes the requirement on the licensing operators to be propositional proposed in Agouraki (this
volume) by further restricting the set of propositional operators which will be appropriate licensers. That
is, according to (31), licenser is not just any propositional operator, but a propositional operator which is
nonveridical. That we need to constraint the set of licensers in terms of nonveridicality and not simply
in terms of 'propositionality' becomes obvious in the following two cases. First, veridical operators likee
the focus operator and the adverb xthes 'yesterday' (see discussion below) and panda 'always' do not
license NPls although they are propositionals (for this see Giannakidou 1995):
(i) *IDHE i Maria kanenan
saw-3sg the-Mary anybody
(ii) *hthes
mas episkeftike kanenas filos
yesterday us visited-3sg any friend
Agouraki presents a petrified construction with focus leo ke kamia kuvenda 'to say something
(in)appropriate every now and then' as an instance oflicensing by focus but was we see from (i), focus
does not generally allow NPls. Sentence (i) and its counterpart without focus *idhe i Maria kanenan are
equally bad. Second, there are licensing operators which do not operate on propositions but on
properties. Such operators are the definite determiner and the universal determiner kathe 'every':
(iii)
i fitites pu ghnorizun tipota shetiko na to anaferun stin astinomia
the students that know anything relavant (to the cas) please report it to the police
A condition like (31) can be modified to account for (iii) because it gives priority to nonveridicality (cf.
Giannakidou 1997 for an extension of nonveridicality to determiners). A licensing condition stated in
terms of propositional operators predicts that cases like (iii) should not exist.
12
Not all verbs that license the subjunctive license NPls, however. As argued extensively in
Giannakidou (1994, 1997), it is not the subjunctive per se that licenses NPls but the semantics of the
selecting verb. Veridical verbs, e.g. vlepo 'see' license the subjunctive, but do not sanction NPls in their
domain.
13
As suggested to me by Larry Hom (p.c.), the averidicality of the counterfactual conditional is not
derived via entailment but via a negative implicature that may occasionaly be cancelled, as in (i):
(i) Ifhe had committed the crime the evidence would show exactly what it does
132
ANASTASIA GIANNAKIOOU
As we shall see below, the difference entailment versus implicature affects the distribution ofNPIs.
14
Greek exploits emphasis to distinguish between the thetic and the categorical modes of negating, but
other languages exploit word order for this purpose. We can envision, for instance, the preverbal versus
postverbal placement of n-words in Spanish and Italian being related to categorical and thetic negation
respectively. The idea is explored in Tovena (1996) for Italian.
References
Aquaviva, P. (1993) The Logical Form of Negation. A Study of Operator-Variable Structures in Syntax,
Ph.D. Dissertation, Scuola Normale Superiore, Pisa.
Atlas, J. (1996) A typology of negative quantifier noun phrases, or What did they mean, "Don't be so
negative!"?, unpublished manuscript, Pomona CoUege.
Condoravdi, C. (1994) Descriptions in Contexts, Ph.D. Dissertation, Yale University.
Dahl, Oe. (1970) Some Notes on Indefinites, Language 46, 33-41.
Deprez, V. (1995) The Roots of Negative Concord in French and French based Creoles, to appear in
Parametric Approaches to Creole Syntax, M. deGraff (ed.), MIT Press, Cambridge, Mass.
Farkas, D.F. (1992) On the Semantics of Subjunctive Complements, in Hirschbuehler P. et al.(eds),
Romance Languages and Modem Linguistic Theory, John Benjamins, Amsterdam, pp. 69-104.
Farkas, D. F. (1994) Specificity and Scope, unpublished manuscript, UCSC.
Giannakidou, A. (1993) KANIS I kanis, a Case of Polarity Sensitivity in Modem Greek, Studies in Greek
Giannakidou, A. (1994a) The Semantic Licensing of NPls and the Modem Greek Subjunctive, Language
and Cognition 4, 55-68, Yearbook of the Research Group for Theoretical and Experimental Linguistics,
University of Groningen.
Giannakidou, A. (1995) Subjunctive, Habituality and the Licensing ofNPIs, Proceedings of SALT V, 132150.
Giannakidou, A.(1997) The Landscape of Polarity Items, Ph.D. Dissertation, University ofGroningen.
Giannakidou, A. & J. Quer. (1995) Two Mechanisms for the Licensing of Negative Indefinites, FLSM VI:
Papers from the 6th Annual Meeting of the Formal Linguistics Society ofMidamerica, ed. L. Gabriele
et aI, IULC Publications, Bloomington, Indiana, pp. 103-114.
Giannakidou, A. & J. Quer (to appear) Long-distance Licensing of Negative Indefinites, in Negation: Syntax
and Semantics, P. Hirschbuhler et al. (eds), Cambridge University Press, Cambridge.
Haegeman, L. (1995) The Syntax ofNegation, Cambridge University Press, Cambridge.
Haegeman, L. & R Zanuttini (199) Negative Heads and the Neg-criterion, The Linguistic Review 8, 233251.
Hom, L. (1972) On the Semantic Properties of Logical Operators in English, Ph.D. Dissertation, UCLA.
Heim, I. (1982) The Semantics of Definite and Indefinite NPs, PhD Dissertation, UMass, Amherst
Heim, I. (1984) A Note on Negative Polarity and Downward Entailingness, Proceedings of NELS 14, 98107. Umass, Amherst.
Israel, M. (1994) Polarity Sensitivity as Lexical Semantics, unpublished manuscript, UCSD.
Kadrnon, N. & Landman, F. (1993) Any, Linguistics and Philosophy 15, 353-422.
Klidi, S. (1994) l:'tOLXEUl AQVI1U1CTJ~ IIoAL1c6nj'ta~, AQVI1UlCOL IIQ(Jo/)EllC'tE~ lCaL AQVI1UlCOL TEAEo-re~
o-ra N ea EAAl]VLlCIl, Studies in Greek Linguistics 15, 451-461.
Klima, E.S. (1964) Negation in English, in J. Fodor and G. Katz (eds.), The Structure ofLanguage, Prentice
HaU, New York. pp. 246-323.
Kritka, M. (1994) The Semantics and Pragmatics of Weak and Strong Polarity Items in Assertion,
Proceedings ofSALT IV, 195-219.
Kritka, Manfred et allii. (1995) Genericity. An introduction, In G. Carlson and F.G. PeUetier (eds), The
Generic Book. Chicago University Press, Chicago, pp. 1-124.
Kuroda, S-Y. (1992) Japanese Syntax and Semantics, Reidel, Dordrecht
Ladusaw, W.A. (1980) Polarity Sensitivity as Inherent Scope Relations, Ph.D. Dissertation, University of
Texas at Austin.
Ladusaw, W.A. (1992) Expressing Negation, Proceedings ofSALTll, 237-59.
WEAK. AND STRONG POLARITY: EVIDENCE FROM GREEK
133
Ladusaw, W.A. {l994) Thetic & Categorical, Stage & Individual, Weak & Strong, Proceedings of SALT IV,
220-29.
Laka, I. (1990) Negation in Syntax: the Nature of Functional Categories and Projections, Ph.D. Dissertation,
MIT.
Lee, Yung-Suk & Larry Horn (1995) Any as Indefinite plus even, unpublished manuscript, Yale University.
Linebarger, M. (1980) The Granunar of Negative Polarity, Ph.D. Dissertation, MIT.
McCawley, J.D. (1980) Everything that Linguists have always wanted to know about Logic (and were
afraid to ask), Basil Blackwell, Oxford.
Montague, Richard. 1969 On the Nature of certain Philosophical Entities, The Monist 53, 159-94, Reprinted
in 1974 in Formal Philosophy. Selected papers of Richard Montague, RH.Thomason (ed), Yale
University Press, New Haven, pp. 148-87.
Quer, J. (1993) The Syntactic Licensing of Negative Items, MA thesis, Autonomous University of
Barcelona.
Philippaki-Warburton, I. & Veloudis, J. (1984) The Subjunctive in Complement Clauses, Studies in Greek
Progovac, L. (1988) A Binding Approach to Polarity Sensitivity, Ph.D. Dissertation, University of Southern
California.
Progovac, L. (1994) Positive and Negative Polarity: a Binding approach, Cambridge University Press,
Cambridge.
Reinhart, T. (1982) Pragmatics and Linguistics: an Analysis of Sentence Topics, Philosophica 27, 53-94.
Rizzi, L. (1991) Relativized Minimality ,MIT Press, Cambridge, Mass.
Strawson, P.F. (1954) Identifying Reference and Truth-values, Reprinted in D. Steinberg et al. (eds),
Semantic,s Cambridge University Press, Cambridge, pp. 86-99.
Tovena, L. (1996) Issues on Polarity Sensitivity, Ph.D .Dissertation, University of Edinburgh.
Veloudis, I. (1981) Negation in Modern Greek. Ph.D. Dissertation, University of Reading.
Wilkinson, Karina. {l991) Studies in the Semantics of Generic Noun Phrases, Ph.D. Dissertation, UMAss,
Amherst.
van der Wouden, T. (1994) Negative Contexts, Ph.D. Dissertation, University ofGroningen.
Zanuttini, R (1991) Syntactic Properties of Sentential Negation: a Comparative Study of Romance
Languages, Ph.D. Dissertation, University of Pennsylvania.
Zwarts, Frans. (1986) Categoriale Grammatica en Algebraische Semantiek. Een studie naar negatie en
polariteit in het Nederlands, Ph.D. Dissertation, University of Groningen.
Zwarts, F. (1993) Three types of Polarity, to appear in F. Hamn and E. Hinrichs (eds), Semantics.
Zwarts, F. (1995) Nonveridical Contexts, Linguistic Analysis 25, 286-312.
CLITIC PLACEMENT AND THE PROJECTION OF FUNCTIONAL

CATEGORIESl
Michael Hegarty
University ofMinnesota
1. INTRODUCTION
Instances of verb-clitic order in Galician and European Portuguese main clauses are
taken in recent work to be the product of overt verb and clitic movement to a position high in the clause. The verb and clitic move to CO according to Madeira (1992)
and Manzini (1994), and to Uriagereka's (1988) F node according Uriagereka
(1995a, 1995b), while Martins (1994) proposes that the verb moves to Laka's (1990)
L node and the clitic to the highest Infl head. Assuming that any of these is correct,
the alternation in languages such as Spanish and Standard Italian between [cl-V]
order in ftnite clauses and [V-cl] in infmitivals, as well as the alternation between
[cl-V] in fmite clauses and [V-cl] in gerunds in Standard Italian and Greek, might be
due to the features on the verb to be checked against C, L, or F being strong on infmitival verbs (and gerunds), but weak on fmite verbs. Or the alternations in Spanish, Standard Italian, and Greek might reflect an Infl-orientation of the clitic,
realised in the Minimalist framework of Chomsky (1993, 1995a, 1995b) as a need
for the clitic to check agreement N-features against an inflectional head, with the
alternation resulting from different orders of adjunction of V and the clitic to I in
fmite clauses versus infmitives (and gerunds) along lines laid out in Madeira (1992)
and Manzini (1994); specifically, these alternations might reflect an Agroorientation of the clitic, realised as a need for the clitic to check agreement Nfeatures against Agro?
In the present paper, the mechanics of the MadeiraIManzini approach are
adopted and implemented in terms of the distinction between the checking of Vfeatures and N-features in the framework of Chomsky (1993, 1995a, 1995b), based
on an exclusive Agro-orientation of the object clitic in Spanish, Standard Italian,
and Greek. The alternation between [cl-V] and [V-cl] orders will arise from the
possibility of dissociating the checking of N-features on nominals against a functional head in these languages from the checking of V-features by the verb; the
possibility of effecting this dissociation is tied to the null subject property of these
languages. The invariable [cl- V] order in French fmite clauses and infmitivals will
135
A. Alexiadou et aI. (eds.), Studies in Greek Syntax, 135-152.
136
MICHAEL HEGARTY
follow directly, assuming, as proposed in Uriagereka (1995b), that French object

clitics are Agro-oriented. Granting a dual XPlXo nature to clitics, following Chomsky (1995a), so that clitic movement can be realised as XP movement followed by a
culminating XO movement at the end, thus instantiating a proposal originally made
in Sportiche (1996, distributed 1992), excorporation in the sense of Roberts (1991),
invoked for clitic placement by Rooryck (1992), Martins (1994), and other researchers, can be dispensed with for these purposes. The implementation of these
derivations of [V-cl] and [cl-V] will also provide a basis for deriving differences in
multiple object clitic orders in fmite clauses versus gerunds in Greek, with no further assumptions beyond ones concerning the organisation ofN-features on a functional head, depending on whether that head also has V-features on it.
For the points to be made in this introduction, it is worth pointing out that the
present work is part of a larger project seeking an account of clitic behaviour in the
terms described above. Another part, which cannot be given an extended discussion
here, but which deserves mention, is clitic climbing. If the object clitic is Agro-oriented, needing to check N-features against Agro, the occurrence of clitic climbing
poses a particular problem for Minimalist syntax since, on standard accounts, there
are no features on functional categories in the higher clause that should attract the
clitic. Furthermore, the optionality of clitic climbing, if it is realised as optional
movement of the clitic, is problematic in Minimalism where movement for featurechecking purposes cannot be optional. The latter point is acknowledged and addressed in Rooryck (1994); however, Rooryck's analysis of clitic climbing as excorporation of the clitic out of [cl-V-T-Agrs] in C under government by the higher VT-Agrs is not available under the strict version of Minimalism under consideration
here. Nevertheless, clitic climbing can be seen to be possible in a Minimalist
framework, in quite narrow circumstances, by appealing to the way in which functional categories are projected in a derivation of phrase structure from a numeration
along the lines of Chomsky (1995a), by allowing that Agro can be a syncretic category, in the sense of Giorgi and Pianesi (1996), optionally combining N-features of
the clitic in an infmitival complement clause with Agro-features from the higher
clause; see Hegarty (1996) for detailed exposition of this approach to clitic climbing
within a strict Minimalist syntax.
What emerges is a position which defers to Madeira (1992), Manzini (1994),
Martins (1994), and Uriagereka (1995b) for the derivation of [V-cl] in European
Portuguese and Galician (though recasting the latter two with exclusive left-adjunction of head movement and without excorporation), and which accounts for the most
basic facts of clitic placement and clitic climbing (or the lack thereof) in Spanish,
Standard Italian, Modem French, and Greek in quite strict Minimalist terms, dispensing with excorporation, maintaining uniform left-adjunction of verb and clitic
to a head or head complex, adhering strictly to the Shortest Move provision of the
Economy of Derivation, and taking an object clitic to be Agro-oriented in the lattermentioned languages, with its basic properties to be accounted for in terms of its
need to check N-features on Agro, given the way in which V-features and N-fea-
CLITIC PLACEMENT
137
tures are organized on functional heads, under Chomsky's (1995) formulation of the
basis of syntactic movement as an operation, Attract-F, involving the attraction of
features. This yields a new distribution of the burden on theoretical principles, with
a heavier part of the burden being placed on aspects of the Minimalist framework,
especially on the treatment of features within a Minimalist syntax as they appear in a
numeration, are projected into syntax on functional heads, and are checked off by
attracting corresponding features on lexical items within the numeration, projected
lower in the tree. Of these topics, some issues concerning verb and clitic order, and
the relative ordering of multiple object clitics, will be taken up here.
This paper does not purport to be a substantive new investigation of the syntax
of pronominal clitics. It is, rather, an examination of some consequences of adopting
the Minimalist framework of Chomsky (1993, 1995a, 1995b), and an attempt to cast
a certain comer of recent work within a constrained version of that framework,
shifting the explanatory burden from framework-exterior principles and processes,
such as excorporation, to aspects of the articulation of the Minimalist framework
itself. The main purpose of this exercise is to confront some instances of the fundamental tension in grammatical theorizing, which is between descriptive coverage of
data and the maintenance of well-constrained theoretical devices and principles.
Work on the syntax of pronominal clitics has exploded in the past decade, greatly
expanding empirical coverage in this area, both within languages and cross-linguistically. The bulk of this work can be characterized as GBlMinimalist, and much of
it has been cast, nominally or fully, in Minimalist terms. The question of to what
degree the theoretical devices and principles appealed to are internal to the framework, or can be cast as such, is, in itself, worth pursuing. The review of the totality
of recent work in these terms is a very large undertaking, and well beyond the scope
of this paper; the present paper merely seeks to address several aspects of the
behaviour of object clitics in languages for which the Agro-orientation of these
clitics has been proposed by more than one researcher in recent work.
2. VERB AND CLInC ORDER

In languages with the most well-established Infl-orientation of object clitics, including Spanish, Standard Italian, French, and Greek, object clitics are uniformly proclitics on the highest, tensed, verb in a fInite clause, and are uniformly enclitics on
infInitives and gerunds. Assuming an Agro-orientation of object clitics, these facts
must be addressed in the context of an articulation of clausal phrase structure. The
structure in (I) will be adopted here for simple tense (nonparticipial) fmite and infmitival clauses, and that in (2) for participial clauses; these derive in large part from
Belletti (1990) and Chomsky (1991), based on argumentation of a sort developed in
Emonds (1978) and Pollock (1989); corresponding structures for Greek contributed
by PhiIippaki-Warburton (1994) and Rivero (1994) are discussed further below.
MICHAEL HEGARTY
138
(2) Participial clause
(1) Simple tense
AgrsP
AgrsP
A
Agrs
TP
Agrs
TP
AgroP
Agro AuxP
Agro AspP
Asp VP
AgroP
DP
Aux AspP
Asp
VP
DP
Some aspects of these structures are innovative and require comment. A tensed
verb, V or Aux, moves through Inf1 heads, checking tense and agreement features,
through successive head movement, of V/Aux to Agro, of Agro to T, and of T to
Agrs (pollock 1989, Belletti 1990). (In Greek finite clauses, the tensed verb moves
to an intermediate-level functional head at or above T/Agr; see below.) Cast in
terms of the feature checking mechanisms of Chomsky (1995b), the tensed verb has
all tense and agreement features registered on it, and functional heads are merged
successively into phrase structure, attracting features of the verb, causing it to raise
under head movement. This entails checking of object agreement features on this
verb against Agro; hence Aux must originate below Agro in (2). This is problematic
for the common assumption that Accusative case is checked against the object and
the main verb in a spec-head relation established by movement of the main verb to
Agro and the object to spec-AgroP. The problem consists of the fact that, in (2),
where Aux, not the main verb (= the participle), moves through Agro, the participle
V has no feature-matching needs that would drive its movement to Agro, and furthermore, movement of the object to spec-AgroP would contravene the
"equidistance" requirement on syntactic movement in Minimalist theory. However,
following studies in various languages (cf. Lee 1993, Kratzer 1993, and Runner
1993), Accusative case exhibits an affinity with aspect that motivates checking of
Accusative case features against a head Asp directly above VP. Accusative case can
then be checked by movement of the main V to Asp and the object to spec-AspP,
which is unproblematic in the structures in (1)-(2).
Furthermore, essentially following Kayne's (1991) adaptation of a proposal of
Raposo (1987), infmitival and gerundive morphology will be taken to be distinct
from tense and entered into phrase structure on an inflectional head below Agro,
above the highest verb. This node, which Kayne calls Infn, will here be labelled linf
CLITIC PLACEMENT
139
or Iger. The tense node, higher in the tree, holds the tense feature or operator that
distinguishes the temporal interpretations of infinitives and gerunds; for discussion
of the content of T and temporal aspects of the interpretation of infinitives and gerunds, see Stowell (1982), Lujan (1980), Belletti (1990), Kayne (1989), and Kayne
(1991).
We will assume that the sequences of heads given in (1)-(2) are present in
Greek as well. Rivero (1994) argues for the structure in (3) below for Greek, where
11M (which Rivero labels simply 'I') is an inflectional head for mood, in which the
subjunctive marker na appears.
If T/AgrP can be articulated at least as containing T and Agro in the order given
within the TP of (1)-(2), then Greek phrase structure will be identical for present
purposes to the phrase structure adopted here for Romance. Furthermore, Rivero
argues for the head movements required for the present discussion, namely of the
fmite verb to T/Agr and of a gerundive verb to 11M, above T/Agr. However, where
Rivero takes the gerund morpheme -ondas to occupy 11M, driving movement of the
gerundive verb to 11M, we take this movement to be driven by feature checking requirements, and put the morpheme -ondas (or its features) on a lower head, Asp,
below T/Agr, in line with assumptions for Romance given above. Reasons for this
will surface in section 2.2 below. Philippaki-Warburton (1994) proposes the structure in (4) below for Greek, where I combines T and Agr(slo), and the subject is in
spec-IP.
(4)
[IP 10 [VoiceP Voice o [AspP Aspo [yp VO DP]]]]
Tense and mood particles, which would appear in 11M in Rivero's structure, are clitic
particles standing at the periphery of the clause, above IP, on Philippaki-Warburton's account. The structure in (4) could be adopted for Greek as well for present
purposes provided I in (4) can be factored into the heads Agrs, T, and Agro, with
relative positioning as in (1)-(2). In Greek, where there are gerunds but no infmitives, Iger would be Asp in the structure ofPhilippaki-Warburton, consistent, under
the different identification and labelling of nodes, with Iger being projected between
T/Agr and VP in Rivero's structure, as assumed above.

2.1 Finite Clauses
The facts of proclisis in fmite clauses in languages addressed here with Agro-oriented object clitics are illustrated below; the facts of Spanish match those of Italian.
Corresponding forms of fmite clauses with enclisis are ungrammatical. 6b) and
(8) are from Kayne (1991).)
MICHAEL HEGARTY
140
Greek
(5) a.
Ton sinandisa
him met-lsg
'I met him'
b. To eho
fai
it have-l sg eaten
Italian
(6) a.
ti
vedo
you see-lsg
'I see you'
b. Sarebbe
assurdo che tu gli
parlassi
it-would-be absurd that you him-dat spoke
(7)
I'ha
sempre fatto
it-has-3sg always done
'He has always done it'
French
(8) a.
b.
Marie Ie connait
Marie him knows
Marie nous a parle
Marie us-dat has spoken
These results will be obtained if, adopting the approach of Madeira (1992) and
Manzini (1994), and adapting it to split-Infl, the tensed verb moves to Agro fIrst,
left-adjoining to it, followed by left-adjunction of the clitic to V+Agro, yielding
[cl+V+AgroJ in Agro; this ordering will then be preserved when the [cl+V+AgroJ
complex moves to T, and the [cl+V+Agro+TJ complex moves to Agrs. But this
order of verb and clitic adjunction must be invariant in fInite clauses, so that the [VclJ order, which surfaces in infmitivals and gerunds, is not available in fmite clauses.
So a fIxed order of operations must be encoded. In the Minimalist framework, the
order of movement operations is determined by the order of features on lexical items
and functional heads. This actually overdetermines the order of operations, since
the order of features on lexical items must match the order of corresponding features
on functional heads, against which they check. So a nonredundant specifIcation of
the order of movement processes in the derivation of a sentence can be given by the
ordering of features on lexical items in the numeration with which the derivation
starts. Alternatively, if movement is interpreted as an operation Attract-F, involving
attraction by features on a functional category of features on elements lower in the
tree, as proposed in Chomsky (1995b), then the order of movement operations in the
derivation of a sentence can be given by the order of "attractor" features on functional categories. 3 The latter view will be adopted here since it will be more natural
CLITIC PLACEMENT
141
for the analysis of the derivation of [V-cl] order in infmitivals and gerunds, considered in section 2.2 below.
Returning to [cl-V] order in fmite clauses, what distinguishes the fmite Agro is
that strong V-features of the verb and strong N-features of the object are present
together on Agro. The invariant [cl-V] order in this case indicates that, in the ordering of features on Agro, the strong V-features of the verb are "outside" the strong Nfeatures of the clitic, requiring them to be checked off ftrst in a checking process
that operates on the feature sequence of a functional head, proceeding from the
outside or end of the feature sequence towards the beginning. The local feature
structure at issue is then as given in (9) below, where CPobj are the cp-features on
Agro to be checked against the clitic, and v-obj is the object agreement V-feature on
Agro; generalizing, this is a special case of (9b), in which, whenever strong Nfeatures and strong V-features appear together on a functional head X, the V-features are ordered after or "outside" the N-features, so that the V-features FV must be
checked off by a verb before the N-features FN are checked off by a nominal.
(9) a.
b.
Agro,
X,
<CPobj, v-obj>
<FN,FV>
As a result of this organization of the features on Agro, the fmite verb in (5)-(8)
must move to Agro ftrst, with movement of the clitic to Agro then following. With
both movements taking place by left-adjunction, by hypothesis, this yields a
[cl+V+Agrol complex in Agro. Subsequent movement of Agro to T, and of T to
Agrs, preserves the [cl-V] order within the successive head complexes.
Of course, the speciftcation of the feature order in (9) is not principled. 4 However, as the discussion of multiple clitic orders in section 3 below will further attest,
the ordering of features in fmite clauses seems to be quite rigid, so that a speciftc
order of features on functional heads is invariably manifested. This is in contrast to
infmitives and gerunds where, as discussed in the next section, the lack of V -features on some functional heads leads to limited underdetermination of feature order
in certain cases, which is reflected in the grammar. The more rigid order of features
in fmite clauses can thus be inferred, with a particular order being invariably manifested, even if a principled basis for the manifested order is not available, except on
a speculative level, as in note 4.
2.2 Infinitives and Gerunds

Nonfmite forms, including infmitives, gerunds, and imperatives, behave differently.
The present discussion will be mainly confmed to the infmitive and gerund data for
languages with Agro-oriented object clitics, given below. s
MICHAEL HEGARTY
142
Greek gerunds: [Vel]
(10)
(11 )
troghondas to
eating
it
ehondas to fai
having it eaten
Italian infinitives: [Vel]
(12)
(13)
Parlargli
sarebbe un errore
to-speak-him-dat would-be an error
Gianni sostiene di averlo letto
Gianni elaims to-have-it read
Italian gerunds: [Vel]
(14)
dicendoglie1o
saying-him-dat it-acc
French infinitives:
(15)
(16)
[el V]
Lui
parler serait
une erreur
him-dat to-speak would-be an error
l'avoir donne serait
une erreur
it-have given would-be an error
In Italian infmitives, the verb or auxiliary moves through an inflectional head linf
with infmitival morphology (Kayne 1991) to T (Belletti 1990). In Greek gerunds,
essentially following Rivero (1994), the verb or auxiliary moves to an inflectional
head Iger at which it checks the features of -ondas (we take Iger to be below Agro),
and then moves to T; the same is plausible for Italian gerunds. Pursuing the Agrooriented analysis of object elitics in these languages, and adapting once again the
approach of Madeira (1992) and Manzini (1994), the [V-ell orders in (10)-(14) will
be produced under uniform left-adjunction of verb and clitic provided the elitic leftadjoins to Agro fIrst, followed by movement of V+Iinf or V+lger to Agro. Elaborating the account by appeal to the organization of features on functional heads, an
explanation for this order of operations is available. Consider the attraction of features by successive functional heads, as these heads are merged into phrase structure
one by one as the phrase structure tree is constructed, along the lines proposed in
Chomsky (1993, 1995a, 1995b), for the sequence of heads T-Agro-linf/gerV/Aux
under consideration here for the languages in question. First, linf/ger is merged into
phrase structure, taking the highest VP (labelled AuxP in (2) as complement;
linf/ger has no N-features, so it does not attract the elitic, but it has V-features for
the infmitival verb (corresponding to infmitival or gerundive morphology), so it
CLITIC PLACEMENT
143
attracts the verb alone, which left adjoins to it, yielding V+Iinflger. Then Agro is
merged into phrase structure, taking IinflgerP as its complement. In nonfmite
forms, Agr heads have no V-features, as evidenced by the lack of subject-verb
agreement. 6 So Agro, lacking V-features, does not attract the verb; but Agro has Nfeatures attracting the c1itic, so the clitic alone moves to Agro, left adjoining to it,
yielding [c1+AgroJ. In this step of the derivation, the c1itic moves ftrst as an XP,
crossing V+Iinf/ger and adjoining to Iinf/gerP, then moves as an XO, left-adjoining
to Agro. Then Tinf is merged into phrase structure, taking AgroP as its complement. At this point in the derivation, the structure is as given in (17) below.
(17)
Tinf/ger
[c1+Agrol
[V+Iinf/gerl
[yp tv tel
Tinf/ger has V-features attracting the verb. However, under the Shortest Move condition, the verb cannot skip Agro on its way to T since Agro is a nearer L-related
head. Therefore, verb movement to T is effected by left-adjunction of [V+Iinf/gerl
to [c1+Agrol, followed by left adjunction of [V+Iinf/ger+cl+Agrol to Tinf/ger,
yielding the complex in (18) below.
(18)
[T V+Iinf/ger+c1+Agro+Tinflgerl
Thus, [V-cll order is produced in infmitives and gerunds for languages with Agrooriented object clitics, excluding French.7 The order of elements in the head complex in (18) also reflects accurately the order of infmitival or gerundive morphology
with respect to the verb root and the object c1itic, as in the Greek troghondas to and
the Italianparlargli and dicendoglielo from (10), (12) and (14). The fact that [V-cll
order in these cases is produced on a single functional head, as in (18), means that
further movement of the verb will consist of movement of this entire complex, carrying the object c1itic along; this is the case in Italian Aux-to-Comp, as in (19) below.
(19)
Avendolo Gianni fatto, .. .

having-it Gianni done, .. .
Such data pose no problem for the present account. They do pose a problem for an
account such as that of Hegarty (1993) in which the infmitival or gerundive verb
and the clitic surface on different functional heads, as noted by a reviewer for this
volume. See also Madeira (1992) for general arguments against splitting up a verb
and clitic on separate functional heads.
The [c1-V] order in French infmitives does not fall in line with the above. It is
tempting to attribute this fact to the more restricted verb movement of French infmitives detailed in Pollock (1989); if the verb does not raise above Agro, it does not
have the opportunities of the Italian infmitive or Spanish inftnitive, or the Italian or
MICHAEL HEGARTY
144
Greek gerund, to get ahead of the clitic. However, this approach runs into trouble
with the data in (20), where in (20b), which is common in informal French, the Aux
has, for whatever reason, raised above Agro to T, carrying the clitic along with it in
the order [cl-V], invariant for French.
(20)
a.
ne pas l'avoir
donne serait
une erreur
neg
it-acc-have given would-be an error
b. ne l'avoir pas donne ...
[informal French]
c. * n' avoir Ie pas donne .. .
d. * avoir Ie donne ...
If French were like Italian and Spanish, Agro would be merged into phrase structure
without V-features, and the clitic alone would move to it, yielding [cl+Agro]; the
subsequent movement of [Aux+Iinf] to T would have to proceed via left-adjunction
to [el+Agro], yielding a head complex with [Aux-el] order, producing (20c) instead
of (20b), or, in the absence of negation, producing (20d). Forms such as (20b)
would not be derivable.
In an approach such as the present one, which derives from that of Madeira
(1992) and Manzini (l994) with strict left-adjunction to heads, the [cl-V] order in
French infmitives results from a derivation in which the highest verb, VlAux, leftadjoins to an inflectional head Ix, followed by left-adjunction of the clitic to
[V/Aux+Ix]. The task of the present paper is to reduce this sort of derivation to
Minimalist principles. The correct results will be obtained if French requires that
strong N-features on a functional head be checked against a nominal a only when a
is in the checking domain of a verb, as suggested for fmite clauses in other languages in note 4 above. However, it will turn out to be more productive to suppose
that, in French, strong N-features on a functional head can be checked off against a
nominal a only in spec-head or head-head relation with a verb that checks correspondingly V-features. In a French infmitive, these might be features that drive
Pollock's (1989) short verb movement, or possibly features for the theta-role assigned to a. Thus, Agro attracts V-features (or theta-features) of the verb as a prerequisite to attracting the N-features of the clitic. The verb then left-adjoins to Agro
first, followed by left-adjunction of the elitic to V+Agro, yielding eel-V] order. In
participial constructions, this approach requires that the auxiliary check the feature
for short verb movement or be in a theta-chain with the main verb, so that the presence of Aux on Agro allows strong N-features of the clitic to be checked against
Agro. When the verb does not undergo short verb movement in French infmitives, a
possibility demonstrated by Pollock (1989), the N-features on Agro must be correpondingly weak; the clitic must then move directly to the highest verb to satisfy PF
requirements, and the [cl-V] or [cl-Aux] complex can move covertly to Agro. A
principle of the sort proposed above, tying the checking of nominal features of the
clitic to the presence of a verb bearing strong V-features or transmitting the thetarole of the clitic, cannot be operative in Spanish, Standard Italian, and Greek, since
CLITIC PLACEMENT
145
it would preclude the derivations of [V-cl] orders in infmitives and gerunds in these
languages, given above.
An interesting consequence of this approach is that clitics with more adjunct
status, such as y and en in French, should not be subject to the principle of French
which ties the checking of strong N-features to the theta-structure provided by a
verb, or transmitted by an auxiliary. This is pertinent to the form in (21) from
Kayne (1991), which is possible in literary, but not colloquial, French.
(21)...
en bien parler ...

of-it well to-speak
In the derivation of this form, the clitic has moved overtly to Agro, but is not accompanied by the verb, which has remained in-situ. This form might still violate
other requirements on clitic structure, such as a requirement of adjacency between
the clitic and a verbal host for PF merger of the clitic and its host; this, and perhaps
other problems with (21), would account for limitations on its acceptability. Corresponding forms with fully thematic clitics are entirely impossible however, and on
the present account, the residue of difference between (21) and these cases can be
traced directly to the difference in thematic status of the clitic.
The property of French invoked above, that strong N-features on a functional
head can be checked off against a nominal a only in spec-head or head-head relation
with a verb checking a strong V-feature or bearing the theta-role of a, has further
consequences. In Hegarty (1996), it is tied to the fact that French, in contrast to
Spanish, Standard Italian, and Greek, is not a null subject language, and it is is used
to explain why French lacks clitic climbing, illustrated in the Italian examples in
(22).
(22)
a.
Mario vuole leggerlo

Mario wants to-read-it
b. Mario 10 vuole leggere
Mario it wants to-read
Clitic climbing has various restrictions and special properties; see Rizzi (1982),
Lujan (1980), Napoli (1981), Burzio (1986), Kayne (1989), Terzi (1993), Rooryck
(1994), and references cited therein. From the present perspective, in which object
clitics are Agro-oriented, and their movement in both fmite and infmitival clauses is
driven by the need for checking of agreement features against Agro, and by no other
factors, the central problem of clitic climbing is that there would be no features projected on functional heads in the higher clause that would attract (the N-features of)
the clitic. 8 A solution to this problem can be given by articulating the projection of
phrase structure along the lines given in Giorgi and Pianesi (1996), in which features are linearly ordered in a numeration, and features which are consecutive in this
linear order can be projected into phrase structure in various ways, whether bundled
up on a single functional head, or scattered over a sequence of functional heads
which are merged into phrase structure consecutively. The fact that object agree-
MICHAEL HEGARTY
146
ment N-features in an infmitival are not accompanied by object agreement V-features gives them some flexibility within the ordering of features in the numeration,
allowing for the possibility that the N-features of the lower clause are consecutive
with the Agro features of the higher clause, and can therefore be projected with
them on a syncretic Agro in the higher clause. 9 For further details, and the application of this view to various properties of clitic climbing described in the references
mentioned above, the reader is referred to Hegarty (1996). To return to the point at
issue in the present paper, the checking ofN-features of the clitic against a syncretic
Agro in the higher clause requires the strong agreement features of the clitic to be
checked independently offeatures of the lower verb, and precisely this is disallowed
in Modern French on the proposal advanced here. The proposal is thus integrated
with wider proposals concerning the behaviour of French, in contrast to Spanish,
Standard Italian, and Greek. lo
This section will conclude with a note on Greek imperatives, which uniformly
exhibit [V-cl] order. I I The imperative verb in Greek is inflected for the number, but
not the person, of the subject. The following examples illustrate this.
(23)
a.
b.
dhjavase to!
read-2sg it
dhjavaste to!
read-2pl it
As argued in Rivero (1994), the verb moves to C in Balkan imperatives. The discussion to this point might lead one to expect [cl-V] order from this movement. In
particular, if the verb's inflection for the number of the subject entails that it has
Agrs V-features, then these must be checked against Agrs, and this can be accomplished by the verb moving through Agrs on its way to C. If Agrs and Agro have
the same dispensation in the grammar, then the imperative verb should move
through Agro as well, checking V-features against Agro, producing a head complex
of the form [cl-V]. But there is a technical consideration to keep in mind with respect to this wrong prediction. In the framework ofChornsky (1993 1995b), Agr Vfeatures co-occur with tense. In fmite participial clauses, for example, Agr V-features are registered on the tensed auxiliary verb, not on the main verb (= the participle), and are checked against Agro and Agrs as the auxiliary raises. If the co-occurrence of Agr V-features with tense holds generally, as a reflection of the fundamental unity of Agr-T inflection, then the imperative verb lacks Agr V-features
since it lacks tense. If this is the case, Agr nodes in an imperative clause are entered
into phrase structure only for the checking of N-features on the arguments. Thus
Agro is projected into phrase structure in an imperative with object N-features only.
In a structure with an object clitic, Agro attracts the clitic only, producing cl+Agro.
Subsequently, the imperative C is merged into phrase structure and attracts the verb,
which must then move through the nearer head Agro, via left-adjunction by hypothesis, on its way to C; this produces [V-cl] order in Greek imperatives.
CLITIC PLACEMENT
147
3. MULTRIPLE OBJECT CLmCS IN GREEK

Now consider the situation in which both the direct and indirect objects of a ditransitive verb are expressed as clitic pronouns. In Minimalist syntax, verbs and noun
phrases are selected from the lexicon with all their syntactic features registered on
them. Suppose, as in section 2 above, that these features are organized in sequences
of the form <Ft, F2, ... , Fn>, where in such a sequence the feature Fi must be
matched and deleted before Fi-l, for each i = n, n-l, ... , 2. In fmite clauses with a
ditransitive verb, the tensed verb, whether it is the main verb or Aux, has object
agreement features for both the direct and indirect objects, and these must be ordered within the feature sequence of the tensed verb. Since the verb checks object
V-features against Agro, the order of object V-features on the verb should reflect the
order of object features in the V-feature sequence on Agro. Assuming, as in section
2, that strong object N-features on Agro accompany the corresponding strong Vfeatures on Agro, it follows that the order of object N-features on Agro is
determinate, matching the order of object V-features on Agro; therefore, when two
objects are both expressed as clitics, the two clitics should emerge from the syntax
in a particular order with respect to one another, reflecting the order of the corresponding object N- and V-features on Agro, which matches the order of object Vfeatures on the verb. This seems to be the case in Greek, as illustrated in (24)-(25).
(24)
(25)
a.
tu
to
edhosa
him-gen it-acc gave
'I gave it to him'
b. * to tu edhosa
to
eho dhosi
a. tu
him-gen it-acc have given
'I have given it to him'
b. * to tu eho dhosi
For example, if the verb edhosa in (24) is selected from the lexicon with the V-feature sequence shown in (26a) below (only aspects relevant to the current discussion
are shown), then Agro should have the feature sequence given in (26b), where v-10
and v-DO are V-features, and <PIO and <PDO are corresponding N-features.
(26)
a.
b.
edhosa,
Agro,
< ... , v-10, v-DO, ...

< ... , <PIO, v-10, <PDO, v-DO, ...
In a fInite clause, strong V-features must be checked before the corresponding Nfeatures, so, after Agro is merged into phrase structure, it attracts the verb, and
checks off v-DO. This leaves <PDO as the outermost remaining feature bundle on
Agro, so Agro next attracts the accusative direct object clitic, which, upon left-ad-
MICHAEL HEGARTY
148
junction, yields [c1ACC+V+Agro], checking <PD~. Then v-IO is checked, leaving

<PIO as the outermost remaining feature bundle on Agro, so Agro next attracts the
genitive indirect object clitic, yielding [cIGEN+cIACC+V+Agro], checking <PIO.
Therefore, the two object clitics must cliticize in the order in (24a), in which the Nfeatures <PD~ are matched and deleted by adjunction of the direct object clitic to
Agro, and then the N-features <PIO are matched and deleted by subsequent adjunction of the indirect object clitic to Agro. In (25), the auxiliary verb Aux has agreement features registered on it, and Aux, not V, moves through Agro and other inflectional heads. In this case, Aux has a V-feature sequence of the form shown in
(26a).
In infmitivals and gerunds in Greek, and arguably imperatives, the verb has no
agreement features, and the two object clitics therefore emerge from the syntax with
no determinate relative ordering imposed on them. This means that v-IO and v-DO
do not occur in the feature sequence of the verb, or on Agro; thus no order is imposed on the N-features <PIO and <PD~ on Agro. They may then be taken to occur
freely in either order, with each of the freely occurring orders attracting a corresponding ordering of clitic movements to Agro. The results are reflected in the
Greek gerund data in (27)-(28), and imperative data in (29).12
(27)
(28)
(29)
dhinondas tu
to
glvmg
him-gen it-acc
b. dhinondas to tu
to
dhosi
a. ehondas tu
having him-gen it-acc given
b. %ehondas to tu dhosi
a. stiltu
to
send-him-gen it-acc
b. stilto tu
a.
The account of (24)-(29) given here would not change substantially if it were
assumed that there are two separate Agr nodes for the direct and indirect objects in
these examples. In the finite case, the object agreement V-features would still appear in a fixed order on the fmite verb, as in (26a), necessitating a fixed order of
merging of direct object Agr and indirect object Agr, resulting in a fixed order of the
two object clitics surfacing on the verb in (24)-(25). It would still be the case that
gerunds have no object agreement features, allowing for a free order of merging of
direct object Agr and indirect object Agr into the phrase structure of the gerund, thus
allowing both possible orders of clitics to surface in (27)-(28); likewise in the imperative in (29). As noted by Terzi (1994), separate heads to host direct object clitics and indirect object clitics are required by the assumption of Kayne's (1994)
Linear Correspondence Axiom The account Terzi (1994) provides of the Greek
multiple object clitic facts of the sort given above crucially appeals to the existence
of separate hosts for these clitics; her account holds to the uniform left-adjunction
CLITIC PLACEMENT
149
hypothesis, but otherwise differs substantially from the present one in basic assumptions about the nature of the clitic hosts and the grammatical principles responsible for the orders of verbs and clitics, and multiple clitics.
The results obtained above do not generalize to Standard Italian and other languages considered here with Agro-oriented object clitics. In Standard Italian, the
indirect object clitic usually precedes the direct object clitic in nontensed as well as
tensed clauses.
(30)
a.
b.
dicendoglielo
saying-him-dat-it-acc
'saying it to him'
* dicendologli
This indicates either that other syntactic processes than those considered here are at
play in Standard Italian, or it reflects regimentation of the forms of the clitics beyond the syntax, in the mapping to PF, perhaps tied to the regular phonological
changes that Standard Italian indirect object clitics undergo when they co-occur with
direct object clitics; in particular, gli changes to glie, as in (30), and mi, ti, si, ci, vi
change to me, te, se, ce, ve. This issue is beyond the scope of the present paper and
will not be settled here. \3
4. CONCLUSION
Following, and slightly refIning, proposals in the literature, object clitics in Spanish,
Standard Italian, Modem French, and Greek are taken to be Agro-oriented, in the
sense that they must move to Agro to check N-features against Agro. The various
orderings of object clitics with respect to the verb in fmite and infmitival clauses and
gerunds in these languages, as well as the very regular ordering of multiple object
clitics with respect to one another in Greek, all discussed in previous literature, follow within a strict formulation of Minimalist syntax, without excorporation, by appeal to the way in which features are organized within a numeration and projected
onto functional categories which are merged into phrase structure in the course of a
syntactic derivation.
NOTES
I am indebted to participants in the Workshop on Greek Syntax at the FAS for their response when
I presented this paper, and to an anonymous reviewer of this volume for detailed comments and
criticisms. I have benefitted from comments and questions elicited by presentations of other versions of
this paper at the University of Pennsylvania Linguistics Colloquium, the 1994 Linguistic Society of
America Meeting in Boston, the University of Maryland MinimaJism Workshop in 1994, and at the City
University of New York, the University of North Carolina at Chapel Hill, and the University of
Connecticut. I have especialJy benefitted from discussion with Giorgia Agouraki, Zeljko Boskovic,
150
MICHAEL HEGARTY
Robert Frank, Sabine latridou, Richard Kayne, Anthony Kroch, Roger Martin, Donna Jo Napoli, Irene
Philippaki-Warburton, Arhonto Terzi, and Gert Webelhuth. In its earliest version, this work was
supported by a postdoctoral fellowship at the Institute for Research in Cognitive Science at the
University of Pennsylvania, under NSF grant IRl 90-16592.
2
See also Terzi (1994) for an account of clitic placement involving uniform left-adjunction to heads,
but which invokes an empty place holder head as the host of clitics in finite clauses rather than an 1head, as assumed here and in Manzini (1994) for languages like Spanish, Standard Italian, and Greek,
where object clitics are I-oriented rather than C-oriented, as in Galician and European Portuguese.
Manzini (1994) assumes, following Madeira (1992), that different orders of overt adjunction of verb and
clitic to I in Spanish and Italian yield the alternation between [cl-V] and [V-cl], but assumes,
furthermore, that the clitic has a weak operator feature which has to be checked against C, thereby
determining covert movement of [cl-V] or [V-cl] to C. But on this account, it is the Infl-orientation of
the strong features of the clitic which is responsible for the alternations in verb and clitic order that are
observed at PF.
Following Giorgi and Pianesi (1996), this could be reduced to the ordering of features in a
numeration. See the summary discussion of cli~c climbing in section 2.2 below.
4
It might express a preference for checking of a strong N-feature of a nominal against Agr to take
place within the checking domain of the verb. This is not required by the Minimalist framework: an
element has to be within the checking domain of a head against which it checks features in this
framework, but the clitic is not checking features against the finite verb in Agro, it is checking features
against Agro itself. Furthermore, the checking of features of a nominal against a functional head does
not take place within the checking domain of the verb when the verb has no strong features for that
head; this is the case with subjects of English finite clauses, and, as will be discussed below, it is the
case for clitics in Spanish and Italian infinitives and Italian and Greek gerunds.
5
(12) and (15) are from Kayne (1991), (13) is from Belletti (1990), and (14) is from Lepschy and
Lepschy (1988).
6
Agr heads are nevertheless present in infinitives and gerunds, with N-features to check against
subject and object NPs and clitics.
7
In languages with inflected infinitives, such as European Portuguese, Agr heads evidently have Vfeatures. Note that no prediction is made here concerning the pattern of cliticization in this language if
object clitics in European Portuguese are not Agro-oriented, but rather, oriented towards Ctr.IF. The
coordination of verb and clitic movement to C or F, in the references given in section I above, is not
regulated by the principles that coordinate verb and clitic movement to Agro' discussed here.
An analogous problem for Minimalism is posed by accounts of clitic climbing such as that of
Roberts (1994), in which the clitic moves to satisfy its own needs (however those are formulated) as a
clitic. Under the Shortest Move provision of the Economy of Derivation, such movement should not be
able to skip the local Agr, as assumed by Roberts. See also latridou (1994) for comments on Roberts
(1994).
9
The possibility of short verb movement in the sense of Pollock (1989) for verbs taking a clausal
complement establishes the existence of Agr0 in the clause of such a verb.
10
On this approach, the lack of clitic climbing in Greek does not depend on a general principle of
the sort proposed here for French, but on aspects of the account that block clitic XP movement out of
gerunds.
11
In Italian and Spanish, the paradigms for imperatives do not exhibit a uniform order of the verb
with respect to object clitics, but rather orders that are variable between positive and negative
imperatives and forms of the imperative determined by person and mood. They are beyond the scope of
this paper.
12
The ACC-GEN order of enclitics on the auxiliary in the participial gerund (28b) is good for some
speakers, but not others. For some speakers of each sort, the judgment is not very robust.
13
For an account of (27) not based on the interaction of verb movement with clitic placement, see
Warburton (1977). In more recent work, Philippaki-Warburton (1994), addressing Greek data,
discusses problems with accounts of clitic placement which treat the clitics as affixes on the verb in the
context of verb movement. Philippaki-Warburton (1994) proposes that object clitics in Greek are not
subject to head movement, but rather are maximal projections placed at the periphery of a tensed clause
CLiTIC PLACEMENT
151
along with the negative particles den and min, the subjunctive particle na, and the future particle tha.
One argument she gives for this position is that [cl V/Aux] order in finite clauses could not be obtained
without stipulation on a head movement account; this argument is countered to the extent that the
proposal in note 4 above is principled. Another argument cites the failure of head movement to produce
the verbal complex in (i) below, in which the object clitics are outside the incorporated adverb ksana
and the augment e- of the past tense marking.
(i) Tu to ksana-eghrapsa
Tu
to ksana- e- ghraf -s -a -1/1
him-gen it-acc again-acc -write-perf-Act-Pastlsg
But this is not problematic for a head movement account of clitic placement, or an account of clitic
placement as XP-movement followed by XO-movement, if the adverb incorporates to VlAux before
raising of VIAux, and if the augment, which exceptionally appears preverbally on verb roots beginning
with a consonant for verbs in a certain class, and which is accompanied by regular tense suffixes, is not
a regular inflectional affix involved in syntactically conditioned affixation driven by verb movement.
Nevertheless, Philippaki-Warburton's proposal remains as a viable alternative to clitic placement
through the interaction of verb movement with XP/XO clitic movement.
References
Belletti, A. (1990) Generalized Verb Movement, Rosenberg and SeIlier, Torino.
Burzio, L. (1986) Italian Syntax: A Government-Binding Approach, Reidel, Dordrecht.
Chomsky, N. (1991) Some Notes on the Economy of Derivation and Representation, in R. Friedin (ed.),
Principles and Parameters in Comparative Grammar, MIT Press, Cambridge, Mass.
Chomsky, N. (1993) A Minimalist Approach to Linguistic Theory, in K. Hale and S. J. Keyser (eds.),
The View from BUilding 20, MIT Press, Cambridge, Massachusetts.
Chomsky, N. (1995a) Bare Phrase Structure, in G. Webelhuth (ed.), Government and Binding Theory
and the Minimalist Program, Blackwell, Oxford.
Chomsky, N. (1995b) Categories and Transformations, in N. Chomsky, The Minimalist Program, MIT
Press, Cambridge, Massachusetts, pp. 219-394.
Emonds, J. (1978) The Verbal Complex V'-V in French, Linguistic Inquiry 9,151-175.
Giorgi, A. & Pianesi, F. (1996) Verb Movement in Italian and Syncretic Categories, Probus 8,137-160.
Hegarty, M. (1993) The Derivational Composition of Phrase Structure, unpublished manuscript,
University of Pennsylvania, Philadelphia.
Hegarty, M. (1996) Clitics and the Organization of Features in Minimalist Syntax, unpublished
manuscript, University of Minnesota.
Iatridou, S. (1994) Comments on the Paper by Roberts, in D. Lightfoot and N. Hornstein (eds.), Verb
Movement, Cambridge University Press, Cambridge.
Kayne, R. S. (1989) Null Subjects and Clitic Climbing, in O. Jaeggli and K. Safir (eds.), The Null
Subject Parameter, Reidel, Dordrecht.
Kayne, R. S. (1991) Romance Clitics, Verb Movement, and PRO; Linguistic Inquiry 22, 647-686.
Kayne R. S. (1994) The Antisymmetry of Syntax, MIT Press, Cambridge, Mass.
Kratzer, A. (1993) On Extemal Arguments, in E. Benedicto and J. Runner (eds.), Functional
Projections, University of Massachusetts Occasional Papers 17, Amherst, Massachusetts, pp.103130.
Laka, I. (1990) Negation in Syntax: On the Nature of Functional Categories and Projections, Ph.D.
dissertation, Massachusetts Institute of Technology, Cambridge, Mass.
Lee, Y.-S. (1993) Scrambling as Case-Driven Obligatory Movement, Ph.D. Dissertation, University of
Pennsylvania, Philadelphia.
Lepschy, A. and Lepschy, G. (1988) The Italian Language Today, New Amsterdam Books, New York.
Lujan, M. (1980) Clitic Promotion and Mood in Spanish Verbal Complements, Linguistics 18, 381-484.
Madeira, A. M. (1992) On Clitic Placement in European Portuguese, UCL Working Papers in
Linguistics 4, University College London, pp. 97-122.
152
MICHAEL HEGARTY
Manzini, M. R. (1994) Triggers for Verb-Second: Germanic and Romance, The Linguistic Review 11,
299-314.
Martins, A. M. (1994) Enclisis, VP-deletion and the Nature of Sigma, Probus 6 173-205.
Napoli, D. J. (1981) Semantic Interpretation vs. Lexical Governance: Clitic Climbing in Italian,
Language 57,841-887.
Philippaki-Warburton, I. (1994) Verb Movement and Clitics in Modem Greek, in Proceedings of the 1st
International Conference on Greek Linguistics.
Pollock, J.-Y. (1989) Verb Movement, Universal Grammar, and the Structure of IP, Linguistic Inquiry
20, 365-424.
Raposo, E. (1987) Case Theory and Infl-to-Comp: The Inflected Infinitive in European Portuguese,
Linguistic Inquiry 18, 85-109.
Rizzi, L. (1982) Issues in Italian Syntax, Foris, Dordrecht.
Rivero, M.-L. (1994) Clause Structure and V-Movement in the Languages of the Balkans, Natural
Roberts, I. (1991) Excorporation and Minimality, Linguistic Inquiry 22, 201-218.
Roberts, I. (1994) Two Types of Head Movement in Romance, in D. Lightfoot and N. Hornstein (eds.),
Verb Movement, Cambridge University Press, Cambridge.
Rooryck, J. (1992) Romance Enclitic Ordering and Universal Grammar, The Linguistic Review 9,219250.
Rooryck, J. (1994) Against Optional Movement for Clitic Climbing, in M. L. Mazzola (ed.), Issues and
Theory in Romance Linguistics: Selected Papers from the Linguistic Symposium on Romance
Languages XXIII, Georgetown University Press, Washington, D.C.
Runner, 1. (1993) A Specific Role for AGR, in E. Benedicto and 1. Runner (eds.), Functional
Projections, University of Massachusetts Occasional Papers 17, Amherst, Massachusetts, pp. 153177.
Sportiche, D. (1996) Clitic Constructions, in J. Rooryck and L. Zaring (eds.), Phrase Structure and the
Lexicon, Kluwer Academic Publishers, Dordrecht.
Stowell, T. (1982) The Tense ofInfinitives, Linguistic Inquiry 13, 561-570.
Terzi, A. (1993) Clitic Climbing from Finite Clauses, unpublished manuscript, University of Ottawa.
Terzi, A. (1994) Double Object Clitics, their Hosts and their Ordering, unpublished manuscript,
University of Ottawa.
Uriagereka, J. (1988) On Government, Ph.D. Dissertation, University of Connecticut, Storrs,
Connecticut.
Uriagereka, J. (1995a) An F Position in Western Romance, in K. E. Kiss (ed.), Discourse
Configurational Languages, Oxford University Press, Oxford.
Uriagereka, 1. (1995b) Aspects of the Syntax of Clitic Placement in Westem Romance, Linguistic
Inquiry 26, 79-123.
Warburton, I. P. (1977) Modem Greek Clitic Pronouns and the 'Surface Structure Constraints'
Hypothesis, Journal ofLinguistics 13, 259-281.
ON CONTROL IN GREEKl
Georgia Catsimali
University of Crete
Irene-Phipippaki-Warburton
Reading University
1. INTRODUCTION
The tenn 'control' has been used in the literature to refer to two logically separate
phenomena, namely: a) the semantic phenomenon of obligatory coreference between
a higher clause argument and the empty subject in the complement clause, and b) a
syntactic configuration with the empty category (ec) PRO in subject position of a
complement clause following a verb of control (verb with the feature specification
[+subject control] or [+object control]). The fact that in English, as Chomsky (1981)
has argued, these phenomena coincide has led to the view that whenever there is
obligatory coreference between main clause argument and a missing embedded
subject, the latter must necessarily be a PRO. In this paper we will argue that the
semantic and syntactic correlates of control are separable by providing evidence from
Greek (Gr) which shows that, although obligatory coreference between main clause
argument and an embedded empty subject obtains, the controlled ec has the
properties of pro rather than PRO. In this context, it will also become clear that the
functional category responsible for the characterisation of a clause as fmite in Gr is
that of person and number agreement (AGR). The question of fmiteness is crucial in
discussing the distribution of PRO because, in its standard defmition, being
[+anaphor, +pronominal], it must obey the "PRO theorem", i.e. it must be
ungoverned and uncased. This, in tum, means that PRO can only occur as the subject
of a non-fmite verb and that since all clauses in Gr, with the exception of gerunds not
discussed here, contain verbs with rich AGR, PRO cannot be licensed as their
subject. Even in theories such as Bouchard (1982), Manzini (1983), Koster (1984),
Hornstein and Lightfoot (1987) and others, where it is argued that controlled PRO
should be interpreted as an ec governed by the higher verb (more accurately the
higher V') establishing coreference via Principle A of the Binding Theory, PRO must
still be ungoverned from within its own clause and must also be uncased.
There are two opposite views concerning the status of the empty category PRO
in Gr. According to one of these views, expressed in Philippaki-Warburton (1987),
this ec does not occur in a fmite clause, (where by 'fmite clause' in Gr we mean one
whose verb exhibits personal agreement), not even in clauses introduced with the
particle na (i.e. subjunctive complement clauses), precisely because, here too, the
153
A. Alexiodou et al. (eds.), Studies in Greek Syntax, 153-168.
154
IRENE PHILIPPAKI-WARBURTON AND GEORGIA CATSIMALI
verb has inflectional endings indicating personal agreement and thus it is, according
to our interpretation, fInite. It is further claimed that in Gr PRO does not exist except perhaps in small clauses and gerunds. The counter proposal (cf. Iatridou 1988
and subsequently Terzi 1991, 1993, Varlokosta 1993 and others) claims that PRO is
the appropriate ec to represent the missing subject in a subset of subjunctive clauses,
specifIcally those na complements which exhibit obligatory control. Such
complement clauses occur after the main verbs like: arhizo 'begin', ksero 'know',
vlepo 'see' etc.
The motivation for postulating subject PRO in this subset of na complement
clauses is the fact that in these cases there is obligatory coreference between a main
clause argument, subject or object, and the missing embedded subject. In what
follows we will fIrst present briefly the main points of the above three studies, which
argue for the presence of PRO, and then we will provide evidence which undermines
them and which strengthens our earlier position (Philippaki-Warburton 1987),
according to which the missing subject of control clauses is not the ec PRO but a
nominative case pure pronominal pro. We will examine the studies arguing for PRO
chronologically.
2. IATRIDOU'S 1988 ANALYSIS

Iatridou (1988) divides the na complement clauses into two subvarieties: those where
we fmd obligatory control, such as after main like verbs vlepo 'see' with object
control (but see below), and ksero 'know' with subject control, as in (1-2).
(1) a.
b.
c.
(2) a.
b.
Vlepo to Jani
na
kolimbai
see-lsg the Jani
subj
swim-3sg
'I see John swimming'
*Vlepo
to Jani
na
kolimbao
see-lsg
the Jani
subj
swim-lsg
*Vlepo
to Jani
na
kolimbai
see-lsg
the Jani
subj
swim-3sg
Ksero
na
kolimbao
know-lsg subj
swim-lsg
'I know how to swim'
*Ksero
na
kolimbai
know-lsg subj
swim-3sg
iMaria
the Mary
There are also na clauses where we do not fmd control, such as the examples with
main verb thelo 'want', elpizo 'hope' etc. as in (3):
(3) a.
Thelo/elpizo
na
want/hope-l sg subj
'I want/hope to leave'
fIgho
leave-lsg
ON CONTROL IN GREEK
b. Thelo
na
fiji
want-lsg subj
leave-3sg
'I want you to leave'
c. Thelo
na
fiji
want-lsg subj
leave-3sg
'I want Mary to leave'
155
i Maria
the Mary
It must be pointed out that verbs of perception such as vlepo 'see', which Iatridou
considers typical control verbs, are not in fact straightforward examples of what is
considered a control configuration because they can occur in constructions such as
(4), where the na clause contains an obligatorily controlled argument but this need
not be the subject.
(4) a.
b.
Vlepo to Jani na ton sprohnun

see-lsg the Jani subj cl-acc push-3pl
'I see John being pushed'
dhinun ta lefta
sto hjeri
Idha
to Jani na tu
saw-lsg the Jani subj cl-gen give-3sg the money in-the hand
'I saw John being handed the money'
In (4) the embedded clauses are in the active voice and their subject is a third person
plural pronominal. Coreference obtains between the main clause object to Jani on the
one hand and the embedded clause object clitic pronouns on the other, ton 'he-acc' in
(4a) and tu 'he-gen' in (4b). Data such as these clearly undermine the strong position
that control is completely tied up with a subject PRO.
Returning now to the main issue, Iatridou claims that the subject in complement
na clauses, as those in (1)-(2), is the ec PRO while in those in (3) the subject, if not
lexical, is pro. To justify the presence of PRO in (1-2) she suggests that fmiteness has
to do not with the presence of personal agreement on the verb but rather with the
category of tense and consequently that the appropriate nominative case assigner in
Greek is tense and not AGR. She observes that tense can vary in constructions such
as those in (3), which do not exhibit control and, therefore, such constructions are
said to have tense. More precisely, they are said to have the features [+T, +A] (where
T=Tense and A=Agreement). On the other hand, tense cannot vary freely in na
complements, such as those in (1-2), which exhibit control. It is assumed that the
latter do not have the property of tense and are characterized as [-T, +A]. The [+T]
feature of the complement clauses in (3) assigns nominative case to the subject
which, if missing is pro, while the [-T] feature of (1-2) cannot assign nominative and
therefore, Null case is assigned giving thus the ec PRO.
In Iatridou' s proposal the standard analysis of PRO as combining simultaneously
the features [+anaphor, +pronominal] and obeying the PRO Theorem, which requires
that it is ungoverned and uncased, is not modified. The only adjustment made in
order to give a syntactic account for the control constructions is to attribute case
assignment not to the typical category for checking nominative, AGR, but to tense.
156
3. TERZI'S (1991, 1993) ANALYSIS

Terzi also adopts the standard view of PRO as the ec which is simultaneously
anaphoric and pronominal. She considers the proposal in Iatridou, where the [+,-T]
property is held responsible for nominative case assignment and where a defective
tense, represented as [-T] of the verb, licenses Null case, but she rejects it because, as
she correctly points out, we can have defective tense even after other verbs, as in (3)
above, without necessarily obtaining a controlled embedded subject, as exemplified
in (5).
(5) a.
b.
0 Janis
the Jani
0 Janis
the Jani
theli
want-3sg
theli
want-3sg
na
subj
na
subj
erthi/ *irthe
come/*came-3sg
ertho/*irtha-lsg
come/*came-lsg
In (5a) and (5b) the na complement clauses have defective tense but (5b) shows
clearly that the subject of the complement clause can be different from that of the
main clause. We may, therefore, conclude that the defectiveness of tense in na
clauses does not guarantee obligatory coreference. This observation leads Terzi to the
conclusion that, although tense in Greek subjunctives is, in many cases, defective, it
is not clear how exactly defective tense correlates with the Null case checking and the
licensing of PRO. In addition, Terzi fmds theoretical problems with the view that
tense is a head governor for the subject NP. She points out that, within a minimalist
framework (Chomsky 1993), head government is not the appropriate mechanism for
case checking. What is needed is an appropriate functional category against which to
check the case of an NP. Terzi's own view is that it is not the tense but the
subjunctive marker na itself which is responsible for licensing Null case. The
specific details of this proposal are as follows.
In subjunctive clauses the subject NP moves from a basic VP internal position
via Spec of AGR to the Spec of the functional category of modality, namely to the
subjunctive marker na. Here, according to this analysis, subject NP is ungoverned, as
required, because it has moved away from the appropriate governor, and it is thus
uncased. This movement of subject NP to an ungoverned position, Spec of M,
justifies in tum its interpretation as the ec PRO.
This proposal, however, leads to the undesirable conclusion that all na clauses,
including those which do not exhibit control, as in (3), should have subject PRO. To
explain the fact that verbs such as thelo, elpizo (as in (3 and many others, followed
by na complements, may not exhibit control she proposes that, in the non control use
of these verbs, and only those, a further movement applies. The combination
modality marker and Verb, [M+V], (which have been brought together by some
previous V-movement to M ) is said to undergo a further head movement to the head
position of the complementizer C. The two alternatives are shown in (6): (6a) and
(6b) represent control non-control constructions respectively:
(6) a. V!helo [CP [C' 0 [MP PRO [M' M [JP [1' ..V ... ]]]
vlepo
ON CONTROL IN GREEK
157
b. Vthelo [CP [C' M+V [MP pro/lexical [M' [IP [I' ... e ... )]] elpizo
As a consequence of the [M+V] to C movement the postverbal subject NP in (6b) is
now in a governed (by the moved embedded verb) and thus cased position, and this
allows it to be either lexical or pro. The solution in (6b) for non control na clauses,
however, involves again head government from V to an element to its right, the NP
in Spec of M', an analysis which Terzi herself had found problematic in Iatridou's
proposal.
4. VARLOKOSTA'S (1993) ANALYSIS

Varlokosta (1993) puts forward an analysis which combines features of both Iatridou
(1988) and Terzi (1993), but it departs from both in some specific ways. Firstly,
Varlokosta repeats Terzi's observation that tense cannot by itself and in a direct way
correlate neatly with the presence vs. absence of obligatory coreference and,
therefore, she agrees with Terzi in rejecting Iatridou's solution. She then proceeds to
divide the na complement clauses into two types, those with obligatory coreference
and those whose subject may refer freely. According to her view only the former are
said to have PRO.
Before we proceed to the details of Varlokosta's justification of PRO for control
constructions, we must point out that she attempts to dissociate the two properties of
PRO by reducing it to an anaphor, an ec with the features: [+anaphor, -pronominal]
and to account for its referential properties not via control but via binding, following
thus the position advocated by Bouchard (1982), Manzini (1983), Koster (1984) and
Hornstein and Lightfoot (1987) among others.
Let us consider this proposal. Even in a theory of a governed PRO the pure
anaphor proposed for the missing subject of control structures (sentences in (1)-(2
must be still ungoverned within its own clause and it must not receive case either
from a case checking property from its own clause or from the higher clause.
According to this view control verbs must be marked as having the property which
allows them to delete the C" element of their complement clauses (S' in older
formulations) thus reducing them to Exceptional (IP) Clauses and, like other ECM
verbs, control verbs are supposed to govern the subject of the embedded clause which
occurs in a position immediately across the IP boundary. As a consequence, the
proposed ec in subject position of a control complement has as its governing category
the main clause and it must, therefore, establish coreference with one of the
arguments of the main clause through Binding.
In order to accept such an analysis Varlokosta has to explain how this
[+anaphor,-pronominal] PRO which she claims is governed from an external
governor has failed to receive government and case from within its own na clause
where the verb shows full personal agreement (it is thus finite, in our terms). The
solution given is as follows:
It is proposed that the difference between control and non-control na
complements correlates with the difference between a na complement without an
158
lRENEPHILIPPAKI-WARBURTON AND GEORGIA CATSIMALI
independent event interpretation (control) and a na complement with an independent

event interpretation (non-control). The na complements with independent event
interpretation are said to be subject to 'tense sequencing' (in the sense of Hornstein
1990) where the tense of main verb influences and restricts the tense of the verb in
the na clause. Tense sequencing is said to motivate a movement of the embedded
verb to the head of C and this in turn enables V from the position of C to govern and
to assign nominative to a following subject NP. This NP, being cased, can either be
lexical or pro. Non-control na clauses have been thus accounted for. Control
constructions, on the other hand, are said to have the following characteristics: they
are not subject to tense sequencing because main and complement clause here
constitute a single event. Therefore the V to C movement does not take place and
thus the subject remains to the left of na+V. Here it is not governed by the embedded
verb and thus cannot receive nominative. It is, therefore, PRO.
4.1 Some Theoretical Problems with the Governed PRO Analysis
One theoretical problem with Varlokosta's proposal is that it crucially relies on the
following assumptions: that in Gr (as in English) the clause is appropriately
interpreted as the projection IP; that the clause initial position is that of Spec of IP
and that it is the position structurally designated for subjects. These assumptions,
however, have been challenged by Philippaki-Warburton (1985, 1987, 1990), who
provides further evidence to show: a) that the clause initial position is not a structural
position for subjects but a position for topics (see also Alexiadou 1994, this volume,
who argues that Gr preverbal subjects are left dislocated), which is only optionally
filled, and it is not exclusive for subjects; other constituents may occupy it, and b)
that nominative is checked by the appropriate feature of the verb in a position far
lower than this topic position because the clause in Gr is more appropriately
interpreted as Mood Phrase. The projection of Mood precedes the projection of
Negation which is followed by the projection of the Future Tense Particle, while the
projection more appropriately labelled IP follows these projections and has as its
head the afflXal inflectional categories [AGR-Tense). Verb movement takes the verb
from VP only as far as these inflectional heads. (Whether AGR is a separate
projection from Tense or the two constitute a single projection or whether the AGR is
a feature of the verb itself carried by the verb to the head of Tense (Chomsky 1995)
is irrelevant to the issues explored here). Furthermore, there is no evidence that the
subject NP has any structural reason to move to the topic position either to satisfy a
[+D] feature of the verb or to check its own feature of case. Thus the motivation for
the EPP which makes it necessary for a subject to occur in a position immediately
after the clause boundary is not present in Greek. If this analysis of the Greek clause
is correct, and the evidence seems overlwhelming that it is so, the subject consituent
of the na clause is neither necessarily nor exclusively in the appropriate position
immediately across the control clause boundary and, therefore, it cannot be
systematically exposed to the governing properties of the main verb (see PhilippakiWarburton 1992 and 1994).
ON CONTROL IN GREEK
159
It is however possible, as has been shown (Philippaki-Warburton 1987), that a

fronted topicalised NP of a complement clause may be affected by the main verb in
some circumstances. We have examples of apparent 'object raising constructions'
where after certain main verbs an NP with the thematic properties appropriate for the
subject of the embedded appears with accusative case as if this NP were the object of
the main verb.
(7) a.
0 Hatzidhakis theli
i Faranduri
na traghudhai
the Chadzidakis want-3sg the-Faranduri subj sing-3sg
mono dikatu traghudhja
only his-own songs
b. 0 Hatzidhakis theli
ti Faranduri
na traghudhai
the Chatzidakis want-3sg ti Faranduri-acc subj sing-3sg
mono dika tu tragudhja
only his own songs
'Chatzidakis wants Faranduri to sing only his own songs'
In view of these examples it is possible to argue that an embedded subject in topic

position though it could licence a nominative case by coindexation with the subject
pro of its own clause at LF, as in (7a), it may instead license accusative case in the
higher clause. This phenomenon of case attraction however is sporadic and it does
not correlate with control verbs. Case attraction may occur with subjunctive
complements of non-control verbs such as thelo 'want' or even with indicative
complements, as in (8):
(8)
to Jani
oti lei
pola psemata
Ksero
know-lsg the Jani-acc that tell-3sg many lies
'I know that John lies a lot'
The phenomenon of case attraction (for an interpretation of this within the minimalist
program see Philippaki-Warburton & Spyropoulos 1996) offers no support for an
analysis of control complements in Gr involving a governed PRO accross IP, as
required by the theories of governed PRO because such a PRO, is governed by the
control verb but it is not case marked by it; control verbs are not case assigners
within this analysis. What we have shown is that some topicalised NPs, either
embedded subjects or embedded objects, may receive case from the main verb. This
situation is the reverse of what the theories of governed PRO require. We may
therefore conclude that, given the well supported analysis of the Greek clause, the
subject element is not always and obligatorily occupying a constituent immediately
after the initial clause boundary. Consequently a main verb of control cannot
systematically govern it, so as to alter its status to an anaphor.
160
5. OBSERVATIONAL AND DESCRIPTIVE PROBLEMS OF ALL THREE

ANALYSES POSTULATING PRO
There are various objections to the proposals presented above all of which support an
uncased ec PRO, ungoverned, as in Iatridou (1988) and Terzi (1993) or governed, as
in Varlokosta (1993), both on observational as well as descriptive grounds. We will
start with the observational weaknesses. The main line of our argumentation will be
based on the crucial assumption about PRO, namely that it must not be governed
within its own clause and that its own clause does not provide the category necessary
to check a nominative case. We will show that the missing subject in the control
clauses in Gr is an ec with structural nominative and that therefore it is the governed
pronominal pro.
5.1 Evidence that the missing Subject receives, in fact, Nominative Case
One of the main objections to the analyses presented above derives from the fact that,
when we extend the data basis, we realize that there is overwhelming evidence
showing that the missing subject, even in control constructions, is not uncased but
that it has in fact nominative case. Examples which argue for this are the following:
5.1.1
(9)
(10)
(11)
The Case of the Predicate

0 Janis
kseri
na
ine
panda evjenikos
the Jani-nom sg know-3sg subj
be-3sg always polite-nom-sg
'John knows how to be always polite'
I Eleni
arhise
na jinete
poli omorfi
the Eleni-nom-sg
started-3sg subj become-3sg very pretty-nom-sg
'Helen has started to become very pretty'
Hiles tis
iposhonde na ine
fronimes
the friends cl-gen promise-3pl subj are-3pl
well behaved
nom-fern-pI
nom-fern-pI
'Her friends promise to behave themselves'
In these constructions, the predicate adjective phrase in the embedded clause carries
clearly not only the gender and number features of AGR (as observed by Terzi 1993)
but also a morphologically distinct nominative case.
(12)
(13)
0 Janis
ine
kalos
the Jani nom sg is-3sg good-nom-sg
'John is nice'
I Eleni
ejine omorfi
the Eleni-nom-sg
became beautiful-nom-sg
'Helen has become beautiful'
ON CONTROL IN GREEK
( 14)
161
I files tis
me
fronimes
the friends-nom-pl cl-gen
are-3pl well behaved-nom-pl
'Her friends are well behaved'
In (12)-(14) above, where the subject NP receives nominative, it would seem

appropriate to say that the predicate phrase licences its case, also nominative, from
the subject NP of which it is predicated through predication indexing. This simple
and intuitively satisfying account can only be entertained for sentences (9)-( 11) if we
were to assume that, here too, the missing subject of the complement clause is pro
with nominative case. It may be countered that in (9)-(11) the case of the predicate
derives from the controling main clause subject by some sort of long distance
coindexation, but this possibility must be rejected, because of the counter evidence
provided by object control examples, like (15)-(16) below:
(15)
(16)
0 Janis
idhe
to Niko
na
ine
stenohorimenos
the Jani- nom saw-3sg the Niko-acci subj
be-3sg sad; -nom
'John saw Nikosl being sad;.
Evala
to Niko
na dithi
Meghas A1eksandhros
made-lsg the Niko-acc1 subj dress-3sg Alexander the Great-noml
'I made Nikos dress up as Alexander the Great'
In the above constructions the embedded predicate refers to a main clause object
which has accusative rather than nominative case. In spite of this, the embedded
predicate carries nominative whose only source has to be a nominative cased
embedded missing subject to which it refers directly. Notice that a predicate which is
not within a na clause is in the accusative when it modifies an accusative object NP.
(17)
akatalilo
Theoro
to Jani
consider-lsg the Jani-acc inapprpopriate-acc
This shows that the predicate licences its case through predication indexing with the
NP which serves as its subject, but if the predicate is inside a control na clause, then
its case is not licenced directly by the main clause NP of which it is predicated, but
indirectly, i.e. by the nominative of the missing pro subject of the na clause which in
turn may co-refer either with a main clause nominative or accusative NP according to
whether we have a subject or object control structure.
5.1.2
NP modifiers and Intensifiers
The elements, monos-clitic, and idhjos modify NPs and agree in gender, number and
case with them:
(18)
o Janis
irthe
the Jani-nom
came
'John came by himself'
monos-to
by himself-nom
162
IRENE PHiLIPPAKl-WARBURTON AND GEORGIA CATSIMALI
(19)
(20)
(21)
(22)
(23)
o idbjos
Janis
irthe
ke mu to
ipe
the same-nom the Jani-nom
came-3sg and cl-gen cl-acc told
'John himself came and told me this'
o Janis 0 idbjos
irthe
ke mu
to
ipe
the Jani nom the same-nom came-3sg and cl-gen cl-acc told
'John himself came and told me this'
ton idjo
Milisa
sto
J ani
the same-acc
spoke-lsg to-the Jani-acc
'I spoke to John himself
Milisa
ston idbjo
to Jani
spoke-lsg to-the same-acc the Jani-acc
tu idhju
Milisa tu Jani
spoke-l sg the John-gen
the same-gen
0
The most natural source for the features of gender, number and case on monos and
idhjos is the host NP. Let us now consider control constructions like those in (24)(25):
(24)
(25)
kolimbai
o Janis
kseri
na
monos tu
know-3sg subj
swim-3sg alone-nom
the Jani-nom
'John knows how to swim by himself
na
fai
monos tu
to Jani
Evala
subj
eat-3sg alone-nom
made-lsg John-acc
'I made John eat by himself
In the above constructions the item monos tu, in nominative case, refers to main
clause subject in (24) but to main clause object in (25), yet, in both cases, it is in the
nominative, a fact which indicates that its case is not licensed by some long distance
association with the main clause argument to which it refers but by reference to a
nominative marked missing embedded clause subject.
Consider also the following constructions:
(26)
(27)
(28)
Sto telos
arhise
0 Janis
na lei
in the end began-3sg the Jani-nom! subj say-3sg
o idhjos
psemata
the same-nom! lies
'In the end John started to tell lies himself
Arhise 0 Janis
na kani
ki 0 idhjos
sahlamares
started the John-nom! subj do-3sg the same-nom! stupidities
'John started to do stupidities himself
Anangase ti Maria
na kuvalai i idhja
ta vivlia
forced-3sg the Maria-acc! subj carry-3sg the same-nom! the books
'He forced Mary to carry the books herself
163
ON CONTROL IN GREEK
In the above sentences we see that the element idhjos, in the complement clause, is in
the nominative although it refers to a main clause nominative subject in (26)-(27) but
to a main clause accusative object in (28). Again this nominative of the embedded
element idjos is naturally explained if we consider that the missing subject of the
complement clause to which idhjos refers has been assigned nominative case. Notice
also that we cannot avoid this conclusion by saying that the elements monos and
idhjos somehow belong to the main clause, but have moved via scrambling to their
position within the embedded. This explanation is not possible because, as we
pointed out, these items are in nominative even when they refer to main clause
accusative object. If they had started as modifiers of main clause argument NPs they
would have licensed their case from a subject NP host nominative for (26)-(27) but
from a DO host NP accusative in (28). Our data show that this is not the case. In
addition, semantically, they modify the lower verb, a fact which strengthens our
hypothesis that they must license their nominative case within the embedded clause.
5.1.3
Control na Clauses with lexically filled emphatic Subject Pronoun
In addition to the above evidence we may also fmd cases where the complement
clause of control constructions has, in fact, a clear nominative emphatic subject
pronoun optionally accompanied by idhjos or monos, as in (29).
(29)
Anangasan tin Eleni

na
milisi
forced-3pl the Eleni-acc subj
speak-3sg
'They forced Helen to speak herself'
afti i idhja
she herself-nom
Examples such as the above show clearly that, when there is a need to emphasize the
embedded subject, it is even possible to use a strong pronoun which is in nominative
case. This indicates that, even when missing, the subject must be in nominative.
5.1.4
Evidence/or Nominativefrom Conjunction o/two na Complement Clauses
Consider also the following:

(30)
0 Janis kseri
rna kathete
(aftos)] ke
the Jani knows-3sg subj sit-3sg
(he-nom) and
[i ali
na dhulevun]
the others-nom subj work-3pl
'John knows how to sit while the others do the work'
In (30) we have a conjunction of two na complement clauses after a control main

verb. Even when the first na complement has a missing subject the second na clause
may have a lexically filled nominative one. This also shows that the subject of a
control na complement clause is a nominative cased NP.
164
5.2 The Floating Subject

There is further evidence which shows that control in Gr cannot be accounted for in
terms of an obligatorily missing ec PRO, as the following examples show.
(31)
(32)
a. 0 Janis kseri tora na kolimbai ke horis sosivio

b. Kseri tora 0 Janis na kolimbai ke horis sosivio
c. Kseri tora na kolimbai ke horis sosivio 0 Janis
d. Kseri tora na kolimbai 0 Janis ke horis sosivio
'John knows now how to swim even without a life-jacket'
(0 J) Arhise(o J) na thavmazi (0 J) to dhaskalo tu (0 J)
'John started to admire his teacher'
In (31a) the lexical nominative NP 0 Janis appears in main clause initial position, a
position which has been extensively argued to be topic. In (31b) this NP occurs in
main clause fmal position, probably a right dislocation site; in (31c) the lexical
subject NP occurs at the right periphery of the sentence. This position may be
interpreted as an adjunction to the main clause C", possibly also a right dislocation
site. I~ all three of these sentences the NP 0 Janis belongs structurally to the main
clause and it is possible to view this NP as the main clause subject controller of the
missing controlled embedded subject. However, the variant (31 d) challenges this
analysis. In (31d) the lexical subject NP 0 Janis occurs apparently within the VP of
the embedded clause while it is the main clause that has an ec as subject. Control is
still obligatory, in that the main clause missing subject must be coreferential with the
explicit embedded one. It would be absurd to propose a PRO here.
Examples of this sort also undermine the claim that control in Gr is a structural
relation between a main clause controller and an embedded PRO. However,
constructions such as those in (31d) also appear to present a problem for Principle C
of the Binding Theory according to which, a referential expression (0 Janis) must be
free (not bound either in its own governing category or by an argument in the higher
clause). In (31d) it would seem that 0 Janis is the subject argument of the embedded
clause in a postverbal position and that it is bound by a c-commanding argument, the
pro of the main clause.
The data is quite fIrm and principle C is well supported especially in comparable
cases from English and other languages, therefore, the explanation for constructions
of this sort must be sought in the interpretation of the status of lexical subject NPs in
Greek. Following earlier claims in Philippaki-Warburton (1987) and more recent
studies, there is strong evidence that the subject in all fInite clauses in Greek is
always the ec pro and that lexical subject NPs are A-bar elements (cf. Alexiadou this
volume, Alexiadou and Anagnostopoulou (1996 which are licenced at LF by
coindexation with pro. Under this interpretation the control relation between main
and subordinate clause in (31 d) holds between main and subordinate pro while the
lexical subject NP is irrelevant to this relation. Furthermore since the lexical NP is an
ON CONTROL IN GREEK
165
A-bar element, it is possible to claim that the pro of its clause is bound by this
adjunct NP at LF without violation of the Binding Theory.
6. DESCRIPTIVE PROBLEMS WITH THE PRO ANALYSIS OF CONTROL
SENTENCES
Chomsky's theory of a [+anaphoric, +pronominal], ungoverned and uncased PRO
derives further strength and elegance by the fact that this PRO occurs in non-fInite
CP clauses, whether main or subordinate, with or without control verbs. The
environment of a control verb in main clause provides simply a control interpretation
(co-indexation) for this PRO. In non control constructions it has, as it is natural, an
arbitrary reference interpretation. If it were the case that na clauses in Gr also had a
PRO we should be able to fmd the arbitrary interpretation for such an item in
environments where we do not have a control main verb. However, the
corresponding constructions are as in (33).
(33)
Dhen ine
fanero ti
na kan-o I-is I-i I-ume/-ete I-un
not be-3sg clear what subj do-l,2,3sg, 1,2,3pl
'It is not clear what I, you, he, we, etc., should do'
In (33) the missing embedded subject may vary in person and number as indicated by
the inflectional ending and has defmite and specifIc reference established in the
linguistic or extra linguistic context. When we need to give the arbitrary
interpretation, we must resort to the use of the indefmite pronoun kanis 'one' as in
(34)-(35).
(34)
(35)
Dhen ine
fanero ti
na kani kanis
not be-3sg clear what
subj do-3sg someone
'It is not clear what one should do'
Na zi
kanis i na mizi
subj live-3sg one
or subj not live-3sg
'To be or not to be'
Similarly in English where PRO is well justifIed, we also fmd it in various types of
adjunct clauses with infmitive verbs. In Gr however we fmd no evidence of PRO in
adjunct clauses. For example, the missing subject of a purpose clause is not
controlled.
(36)
Aghorasa ena vivlio ja na

to dhoso/is/i
sti Maria
bought-lsg a book
for subj cl-acc give 112/3 sg to-the Maria
'I bought a book in order that I/you/he etc give it to Mary'
Other adverbial clauses too either choose their subject freely or if this subject needs
to be interpreted as arbitrary the indefmite pronoun kanis must be present:
166
(37)
a.
b.
Tha figho prin

metanjoso/is/si
'I will leave before I /youlhe/ect change my/yourlhis mind'
Tha figho
prin me dhi kanis
'I will leave before somebody sees me'
This shows that there are no factors (either lack of tense or presence of na) within the
na clause itself which could license PRO, as in the case of English, where the
infmitive being unable to govern and to assign case conditions subject PRO.
We saw that neither arbitrary reference constructions nor adjunct clauses of any
type show any evidence that PRO exists in subjunctive clauses and consequently we
must reject it. Descriptively it would be a strange situation indeed to have a language,
Greek, which possesses the category PRO but chooses to use it only in the more
marked environment of control while no evidence for a more general unmarked
occurence of this element is attested.
7. CONCLUSIONS
We have presented evidence from a variety of constructions which shows the
following: in all Greek complement clauses, including subjunctive ones after control
main verbs, the ec in subject position cannot be a PRO, whether in its standard
interpretation as [+anaphor, +pronominal] or in a true anaphor version as [+anaphor,
-pronominal]. The reason for excluding PRO from these constructions is the fact that
such a category must be ungoverned and uncased within its clause and, in Gr, this
possibility does not exist. We saw that there is strong evidence that the missing
embedded subject of subjunctive complements, including those after control verbs,
does receive nominative and that it is therefore pro. (For similar evidence from other
languages see Borer 1989). We suggest that the main factor in establishing control,
i.e., obligatory coreference between a main clause argument and the typically missing
embedded subject, is the lexical characteristics of the main verb involved. It is of
course natural that control verbs in Gr are a subset of those taking subjunctive in their
complement. This is because subjunctive is the mood of a clause which is most
dependent on the main clause; this dependence is expressed in a number of ways; for
example, the interpretation of the tense in a subjunctive complement depends on the
tense of the main verb. This dependence is also reflected in the fact that the reference
of the subject argument of a subjunctive complement is determined by an argument
in the main verb. This is optional for non control verbs but obligatory for control
verbs. Such coreference requirement makes the explicit expression of the controlled
subject redundant and therefore, following the principle of least effort, or Grice's
maxim of quantity, a controlled embedded subject should not be explicitly
mentioned. This makes an infmitive construction with PRO the most appropriate
expression for control in English and other languages including Ancient Greek,
which have the category of infmitive (see Philippaki-Warburton and Catsimali to
appear) but Gr does not have infmitives. The next best choice is the subjunctive with
a missing, pro subject. Thus, Greek syntax aids control as far as it can by using more
ON CONTROL IN GREEK
167
reduced complement clauses, i.e., subjunctives rather than indicatives, missing

embedded subject rather than lexically filled one, etc., but control, in the absence of
the category of infmitive does not involve PRO in Gr. This leads to the more general
theoretical conclusion that the semantic and the syntactic correlates of control are not
always satisfied by the same structural features. The semantic dimension of control
seems to be a property of the lexical verb which requires the obligatory coreference
interpretation of the missing subject. This information, and this demand from main
verb can be satisfied and needs to be satisfied at the semantic level (as suggested also
by Joseph 1992) even when the syntax does not afford a specific configuration and
specific categories for it. To put it in another way, a na complement clause following
a control verb cannot have a subject which may be given independent reference. It
must have a subject which can establish coindexation with the appropriate main
clause argument in accordance with what the main verb demands. For this to be
satisfied the embedded clause subject must be the weakest pronominal element
available in the language. This is pro for Greek and not PRO.
NOTES
We would like to thank two anonymous reviewers for their valuable comments and criticisms which
have contributed to the clarity of the ideas presented here. Irene Phillipaki-Warburton's research was
supported by the Reading University which is hereby gratefully acknowledged.
References
Alexiadou, A. (1994) Word order alternations in Modem Greek: looking at Subject Positions, paper
presented at the Workshop on Modem Greek Syntax, Berlin.
Alexiadou, A. & E. Anagnostopoulou (1996) Symmetries, Asymmetries and the Role of Agreement,
Glow Newsletter 36,12-13.
Borer, H. (1989) Anaphoric AGR, in O. Jaegli and K. Safir (eds.), The Null Subject Parameter: Studies in
Natural Language and Linguistic Theory, Kluwer Academic Publishers, Dordrecht, pp. 69-91.
Bouchard, (1982) On the Content of Empty Categories, Ph.D. Dissertation, MIT.
Chomsky, N. (1993)A Minimalist Program for LinguistiC Theory, in K. Hale and J. S. Keyser (eds.), The
view from Building 20, MIT Press, Cambridge, Mass, pp. 1-52.
Chomsky, N. (1995) Bare phrase structure, in G. Webelhuth (ed.) Government and Binding Theory and
the Minimalist Program, Blackwell, Oxford, pp. 383-441.
Hornstein, N. & D. Lightfoot (1987) Predication and PRO, Language 63, 23-52.
Iatridou, S. (1988) On Nominative Case Assignment in Modem Greek and a few Related Things,
unpublished manuscript, MIT.
Joseph, B.(1992) Diachronic Perspective of Control, in R. Larson, S. Iatridou, U. Lahir and 1.
Higginbotham (eds.), Control in Grammar: Studies in Linguistics and Philosophy 48, Kluwer
Academic Publishers, Dordrecht, pp. 195-234.
Koster, J. (1984) On Binding and Control, Linguistic Inquiry IS, 417-59.
Manzini, R (1983) On Control and Control Theory, Linguistic Inquiry 14, 421-46.
168
Philippaki-Warburton, I. (1987) The Theory of Empty Categories and the pro-drop Parameter in Modem
Greek, Journal ofLinguistics 23, 289-318.
Philippaki-Warburton, I. (1992) On Mood and Complementizers in Modem Greek, in I. PhilippakiWarburton and R. Ingham, (eds.), Reading University Working Papers in Linguistics 1,5-41.
Philippaki-Warburton, I. (1994) Verb Movement and the Distribution of Clitic Pronouns, in PhilippakiWarburton I., K. Nicolaidis and M. Sifianou (eds.), Themes in Greek Linguistics, John Benjamins,
Amsterdam, pp. 53-60.
Philippaki-Warburton, I. & G. Catsimali (1989) Accusativus cum infinitivo in Ancient Greek, Studies in
Philippaki-Warburton & G. Catsimali (to appear) On Control in Ancient Greek, Studies in Greek
Linguistics 16.
Philippaki-Warburton, I. & V. Spyropoulos (1996) Problems of Case within the Minimalist Program, to
appear in Studies in Greek Linguistics 17.
Terzi, A.(1991) PRO and Obviation in Modem Greek Subjunctives, Proceedings of WCCFL 10,471482.
Terzi, A. (1993) PRO and Null Case in Finite Clauses, unpublished manuscript, University of Ottawa.
Varlokosta, S. (1993) Control in Modem Greek, University of Maryland Working Papers in Linguistics
1,144-163.
MODALS AND THE SUBJUNCTIVE!
Anna Roussou
University o/Wales, Bangor
1. SUBJUNCTIVE COMPLEMENTS IN GREEEK

A property of the Greek (Gr) complementation system, shared to a greater or lesser
extent by all Balkan languages, is the lack of infmitives and the use of the
subjunctive in the corresponding infmitival Romance (and English) constructions.
The examples in (1a) and (1b) from Gr and Italian respectively illustrate this
typological difference:
(1) a.
b.
Thelo
na figho
want-1sg subj leave-1sg
'I want to leave'
Voglio andare
want-1sg go
'I want to go'
As (1a) shows subjunctive clauses in Gr are introduced by the particle na. (For the
loss of the infmitive in the Balkan languages see Joseph (1983)). Following Ingria
(1981), Veloudis and Philippaki-Warburton (1983), among others, I will assume that
na is the subjunctive marker itself, i.e. an I(nflectional) element. As far as their
distribution is concerned, na-clauses occur as complements to the following classes
of verbs: volitionals, modals, aspectuals, causatives, perception and experiencer
predicates.
Regarding their tense properties, the contrast in (2a) and (2b) shows that nacomplements cannot be inflected for past tense:
(2) a.
b.
Thelo I ithela
na figho
want-1 sg/wanted-1 sg subj leave-1 sg
'I want/wanted to leave'
*Thelo I ithela
na efigha
want-1sg/wanted-1sg subj left-1sg
'I want/wanted to have left'
The na-clause in (2) remains inflected for present subjunctive despite the
morphological changes of the matrix V which may be inflected for present or past
169
A. Alexiadou et al. (eds.), Studies in Greek Syntax. 169-183.
ANNA ROUSSOU
170
tense, as in (2a). Furthermore, the V in the na-clause is inflected for perfective (cf.
(2a, or imperfective (cf. (3 aspect. In the latter case the na-clause has an iterative
reading:
(3) Thelo / ithela
na fevgho
want-lsglwanted-lsg subj leave-lsg
Morphological tense restrictions of this type hold for the na-complements of the
volitional predicates, as well as of the aspectuals, causatives, perception and
experiencer verbs. On the basis of data similar to those in (2)-(3) PhilippakiWarburton and Veloudis (1984) argue that na-complements are essentially [-Tense],
like their inftnitival counterparts in other languages. Picallo (1985), on the other
hand, argues with respect to the Romance subjunctives that the Tense of the
subjunctive is anaphoric/or dependent to that of the matrix clause. In either approach
the conclusion is that subjunctives do not have 'independent' time reference.
There are two points that need to be considered in relation to this claim. The
ftrst is more theoretical in nature and concerns the notion of anaphoric Tense.
Kempchinsky (1986) argues against the claim that the subjunctive T is anaphoric, or
to be more precise that it is more anaphoric than the one in indicative complements.
Following En~'s work on this topic (appearing as En~ (1987, she argues that it is a
general property of complement clauses that their time reference is evaluated
(partially or exclusively) with respect to that of the matrix predicate. In other words,
the embedded T has to be linked to the matrix T through which it is anchored to the
speech/utterance time so that temporal evaluation of the embedded clause is
ultimately achieved. On this basis then an analysis of an anaphoric subjunctive T is
not tenable, and the distinction between subjunctives and indicatives in the
following section has to be considered along different lines.
The second point, concerns the existence of na-complements which are clearly
inflected for past tense and perfective aspect (this is referred as the aorist in
traditional grammars), as already noted by Veloudis (1985):
(4) a.
Elpizo na eftje
noris hthes
to vradhi
hope-Isg subj left-3sg early yesterday the night
'I hope that he left early last night'
b. Pistevo
na pije
sto jatro
believe-lsg subj went-3sg to the doctor
'I believelhope that he went to the doctor'
The examples in (4) show that in some cases past tense subjunctive is indeed
possible, yielding a clear contrast with respect to the data in (2)-(3) and (4). In order
to account for this pattern we need to identify the properties of the matrix predicates,
or to be more precise of the matrix T, in (4) that permit a past tense subjunctive in
the complement clause. In other words it appears to be the case that in certain
contexts, i.e. with volitionals, etc., past tense subjunctive is excluded (cf. (2, while
MODALS AND THE SUBJUNCTIVE
171
in others, i.e. with epistemics, this is indeed possible (cf. (4)). It is worth noting
though that the possibility of having past tense subjunctive is very limited indeed, as
we will see in section 3. Before we consider the T-properties of the na-complements
we should provide a more detailed consideration of the general properties of the
subjunctive. This will be discussed in the following section.
2.
THE SUBJUNCTIVEIINDICATIVE DISTINCTION
Let us fIrst start by considering the data in (2): the na-clause in (2) functions as the
complement of a volitional predicate, i.e. want. Subjunctive complements to
volitionals have traditionally been analysed as embedded imperatives (cf. Jacobs
1981, Huntley 1984), in that they have the effect of switching the time reference of
the embedded clause to the future, i.e. posterior to the matrix time (Kempchinsky
1986). This is due to the fact that predicates of this type embed an abstract modal
operator, as has been proposed in the traditional literature (cf. Lakoff 1968). In Gr
this state of affairs is best illustrated with predicates like elpiza 'hope' that
subcategorise for either a na- or ati-complement, as in (5a) and (5b) respectively:
(5) a.
b.
Elpizo na fIji
hope-l sg subj leave-3sg
'I hope he leaves'
Elpizo
oti tha fIji
hope-lsg that willleave-3sg
'I hope he will leave'
As (5b) shows, when ati is used the embedded V is marked for future, indicated by
the presence of the future marker tha. Thus, the indicative and the subjunctive
complements in (5) have quite similar interpretations in terms of time reference.
Nevertheless the sentences in (5) are not synonymous. Lightfoot (1975) argues
that despite the similarities between the future and the subjunctive their distribution
is regulated by the presence vs. absence of 'existential presupposition' respectively.
To be more precise, according to his analysis the future particle tha in (5b) gives
rise to an existential reading, while the subjunctive particle na in (5a) yields a nonexistential reading. Distinctions of this type between indicatives and non-indicatives
have been discussed quite extensively in the literature mainly from a
semantic/pragmatic point of view. The idea is that indicatives refer to an actual
world, while non-indicatives refer to possible worlds (cf. Farkas (1992), and for the
Gr data see Veloudis and Philippaki-Warburton (1983) and Rouchota (1991)).
Closely related to this type of formulation is the analysis that attributes to nonindicatives the lack of a referential index (Huntley 1984).
Going back to the syntax of subjunctives, it has been argued that subjunctives
are similar to indefmites (Manzini forthcoming, Tsoulas 1994). In particular,
Manzini argues that the subjunctive T corresponds to an indefmite, hence to a free
ANNA ROUSSOU
172
variable in the sense of Heinl (1982). Thus it is licensed by a sentential/intensional

operator, such as negation, question, conditional, as well as necessity and possibility
operators; hence the 'non-existential' interpretation. Alternatively, a subjunctive T is
like a polarity item in that it needs to be licensed by a sentential operator. An
indicative T, on the other hand, is to be compared to an existential
quantifier/positive polarity item. Although this formulation differs from the one put
forward by Lightfoot (1975), it preserves the same intuition with respect to the
properties of the subjunctive. Syntactically, licensing of the subjunctive requires the
formation of a (head) dependency between the (intensional) operator presumably
located in the matrix C and the embedded T, i.e (Op, .... ,T). This dependency must
satisfy locality conditions, otherwise ungrammaticality arises.
Assuming then that the subjunctive is triggered by a sentential/ intensional
operator, the modal reading attested in matrix subjunctives follows:
(6) a.
Na fijis
subj leave-2sg
'You may/shall/must leave'
b. Na efije
araje?
subj left-3sg possibly
'Can it be the case that he left?'
c. Na erhotane tora 0 Janis!
subj came-3sg now the John
'I wish John would come'
In (6a) the utterance has imperative force, in (6b) it expresses possibility and in (6c)
it is an optative. Given the preceding discussion, the natural assumption to make is
that matrix subjunctive is triggered by the relevant sentential operator/feature in
each case: the Q(uestion) or F(ocusing) which are associated with the force of the
clause and form part of the feature specification of the matrix C (cf. Roussou 1994,
Roberts and Roussou 1996). The absence of these features in matrix declaratives
excludes the possibility of having a matrix subjunctive with declarative force, as is
indeed the case. Furthermore, it follows why, in the absence of infmtivals in Gr, naclauses appear with modals, given that the latter are essentially lexicalised operators.
Similar considerations are expected to extend to the other classes of predicates that
take a na-complement. However, a discussion of these constructions is beyond the
scope of the present paper.
Moreover, there are a number of other contexts where the licensing operator is
not part of the lexical properties of the matrix predicate, but a sentential operator,
such as NEG or Q in the matrix clause. This is the case of epistemic predicates as
the examples below illustrate:
(7) a.
Dhen thimame
na efije
not remember-lsg subj left-3sg
'I don't remember him leaving'
b.
(8)
173
Dhen nomizo na ef~e

not think-lsg subj left-2sg
'I don't think that he left'
Thimase
na efije?
remember-2s subj left-3sg
'Do you remember him leaving?'
In (7) negation is present and the embedded clause is introduced by na. The same
holds if there is a yeslno question operator, as in (8).
Interestingly though, when NEG or Q are absent the sentences in (7)-(8)
become ungrammatical. In this case the result is grammatical only if the embedded
clause is an indicative introduced by the complementiser oti, as in (9):
(9) a.
b.
Thimame
*na loti efije
remember-lsg subjlthat left-3sg
Nomizo *na loti efije
think-lsg subj I that left-3sg
Thus as the contrast between (7)-(8) and (9) clearly shows the na-complement with
episternic predicates is due to the NEG/Q operator. In the absence of the relevant
operator ungrammaticality arises, as is indeed the case in (9); the subjunctive is then
excluded and the only alternative is to use an oti-complement. Notice though that
the presence of such operators may be a necessary but not a sufficient condition for
licensing of the subjunctive in (7)-(8). There is one more condition that needs to be
met, namely that there is a matrix present T. If this condition is not met the result is
ungrammatical, even if there is tense agreement between the matrix and the
subjunctive V's, as in (10):
(10)
*Dhen thirnithika
na efije
not
remembered-lsg subj left-3sg
The subjunctive in (8) will then have to be interpreted as the result of a conspiracy
between negation and the present tense specification in the matrix clause. The
immediate question that arises of course is why present tense turns out to be crucial
for the licensing of the subjunctive with epistemic predicates, as opposed to the case
of volitionaIs as in (2) above.
The same results hold for other cases of episternic predicates, which can license
the subjuctive in the absence of an operator such as negation. Nevertheless, and
crucial to our discussion, the requirement is, once again, that they are inflected for
present tense:
(11)
a.
Pistevol *pistepsa
na fiji
believe-l sglbelieved-l sg subj leave-3sg
'I believe/*believed (expect) that he will leave'
174
ANNA ROUSSOU
b.
Fandazomel *fandastika na fiji

imagine-Isg/imagined-Isg subj leave-3sg
'I imagine/*imagined (expect) that he will leave'
As the ungrammatical versions of the examples in (11) show the matrix predicate
cannot be inflected for past tense. Interestingly, when the complementiser oti is
used, then the T-specification of the matrix clause is irrelevant:
(12)
Pistevo I pistepsa
oti tha fiji
believe-I sglbelieved-lsg that willleave-3sg
'I believelbelieved that he will/would leave'
Thus the requirement on present tense with the matrix predicate strongly correlates
with the presence of the na-complement. What we need to consider next then is why
this type of correlation holds with epistemic predicates.
To summarise the discussion so far, the subjunctive vs. indicative distinction
has been formalized along the lines of the analysis of Manzini (forthcoming): the
subjunctive T is licensed by a sentential/intensional operator. In syntactic terms
licensing of this type involves the formation of a dependency between the Operator
and the subjunctive T. An analysis of this type then explains the presence of the
subjunctive in a number of syntactic contexts at least as far as complementation is
concerned. What remains to be clarified though are the apparently special conditions
that trigger the subjunctive with epistemic predicates.
3. EPISTEMIC MODALITY AND NA-COMPLEMENTS

The discussion so far shows that epistemic predicates may take ana-complement
provided they are inflected for present tense, and in some cases there is NEG/Q
present in the matrix clause. Recall also that it is with this class of predicates that
past tense subjunctive is possible, as already mentioned in section I with respect to
the examples in (4). In particular as opposed to most other cases of
complementation, the na-clause in this case seems to exhibit independent time
reference. Thus subjunctive complements to epistemic predicates allow for the
following tense alternations, schematically represented in (13):
(13)
Matrix Present Present PastPast-
Na-clause
present
OK
past
OK
present
*
past
*
According to (13) the only possible sequences are present-present and present-past.
Veloudis (1985) notes that past tense subjunctive (along with NEG/Q in some cases)
175
is found systematically with predicates that encode epistemic modality. Regarding

verbs like thimame (remember), nomizo (think) it is clear that to some extent NEG
may also contribute to the derived reading. The interpretation in this case is "I can't
tell for sure/ I don't think it's possible", i.e one that has an implicit epistemic modal.
A similar interpretation is derived when the question operator is used, as in (8)
above. On the basis of these data Veloudis (1985) suggests that it is the presence of
epistemic modality in the matrix clause which allows for past tense specification in
the embedded clause. However, this observation although intuitively correct it is not
formalized in clear syntactic and/or semantic terms, that is it does not explain why
such a correlation should hold. If it is true that epistemic predicates under these
conditions are interpreted as (implicit) modals, then we would expect to fmd the
same tense pattern with the 'pure' epistemic modals. Let us then consider the
following examples:
(14)
a.
Prepi
na fiji
must-3sg subj leave-3sg
'He must/should leave'
b. Bori
na fiji
may-3sg subj leave-3sg
'He may/can leave'
The modals 'prepi' and 'bori' in (14) give rise to both epistemic and non-epistemic
(i.e root) readings, as shown in the English paraphrases in (15) and (16)
respectively:
(15)
(16)
a.
b.
a.
b.
It must be the case that he leaves

He is obliged to leave
It is possible that he leaves
He is able/allowed to leave
(Epistemic)
(Non-epistemic)
(Epistemic)
(Non-epistemic)
In (16b) the interpretation can be that of ability or permission. In either case though
the modal has the non-epistemic reading. Note crucially that in (14) both the matrix
and the embedded T are inflected for present indicative and subjunctive
respectively. Suppose next that the modal is inflected for past tense as in (17):
(17)
a.
Eprepe na fiji
must-3sg subj leave
'He was obliged to leave'
b. Boruse
na fiji
could-3sg subj leave-3sg
'He had the ability/was allowed to leave'
Interestingly the sentences in (17) are no longer ambiguous: as the English

translation indicates the presence of past tense in the matrix clause blocks the
ANNA ROUSSOU
176
epistemic reading. Consider next the alternative case where the subjunctive T is
inflected for past, as in (18):
(18)
a.
b.
Prepi
na efije
must-3sg subj left-3sg
'It must be the case that he left'
Bori
na efije
might-3sg subj left-3sg
'It is possible that he left'
The pattern in (18) is the reverse of that in (17), in that the presence of past tense in
the embedded clause allows for the epistemic reading only. If, on the other hand,
both the matrix and the embedded clause are in the past tense, then the result is
ungrammatical, giving rise to uninterpretability:
(19)
a.
*Eprepe na efije
must-3sg subj left-3sg
b. *Boruse na efije
could-3sg subj left-3sg
Neither reading is available in (19).

Thus the discussion so far shows that the pattern summarised in (13) is indeed
found with the modals 'prepi' and 'bori' when they express epistemic modality. In
their non-epistemic reading they disallow past tense subjunctive, in a way similar to
volitional predicates (cf. (2) above). In short, the sequence present-present allows
for both the epistemic and the non-epistemic reading; the sequence present-past
allows for the epistemic reading only, while the sequence past-present allows for the
non-epistemic interpretation only. Similar results are found with epistemic
impersonal constructions such as apokliete (be impossible),jenete/miazi (to appear/
seem), ine pithanolapithano (be possible/impossible), etc., and their non-epistemic
counterparts such as epitrepete (be allowed), prokite (it is going to), ine anageol
aparetito (be necessary), etc. The data discussed so far fall into two classes: the first
class involves epistemic modals, while the second involves epistemic verbs that take
a subjunctive complement ('subjective epistemics' in the sense of Lyons 1977). The
common properties of these two classes of predicates are: frrst that epistemic
modality requires present tense, and secondly that the na-clause allows for past
tense.
Thus, the presence of na-complements with epistemic predicates is to be
expected assuming that in this case the latter have an interpretation similar to
epistemic modals. The question that remains though is why present tense is
obligatory and consequently why the na-clause can be in the past tense.
177
4. EPISTEMIC MODALITY AND EXPLETIVE TENSE

In all the cases of epistemic modality discussed so far the matrix T is
morphologically realised as present, while the embedded subjunctive T can be either
present or past. What I would like to suggest in this section is that past tense
subjunctive in Gr is only found in these contexts due to the fact that the matrix T is
an expletive (cf. latridou 1990). As a result, the embedded T (i.e the na-c1ause)
acquires matrix scope, so that it is directly anchored to the speech/utterance time;
hence the independent time reference exhibited by the na-complement with
epistemic predicates. Let us now look at the details of this account more carefully.
Consider frrst the difference between epistemic and non-epistemic modals.
Kratzer (1981) argues that these two types of modality are based on different
premises: "if we use an epistemic modal, we are interested in what else mayor must
be the case, given everything we know already." (1981: 52). With non-epistemics on
the other hand, "we are interested in what can or must happen, given circumstances
o/a certain kind." (1981: 52). Thus non-epistemic modality can be more dependent
on the special conversational circumstances, while this is not the case for epistemic
modality. In latridou's (1990) terms possibility and necessity, as these are expressed
in Gr by the modals prepi (must) and bori (can) in their epistemic reading, are
notions which are not sensitive to time. Her argument is based on the
incompatibility of past tense with adjectives like 'possible', as in (20) below:
(20)
a. It is possible that John stole the tapes

b. #It was/ will be possible that John stole the tapes
On the basis of the oddness of (20b) (indicated by the # sign) latridou (1990) argues
that the matrix predicate lacks a time variable, since possibility and necessity do not
change over time. 2 Thus in latridou's terms the present tense in (20a) is pleonastic in
that it has no semantic function.
As the Gr examples show it is only present tense that is compatible with an
expletive T (but see latridou 1990 for the Basque data where a distinct irrealis tense
is used). This is probably due to the fact that the present tense appears to have a
default interpretation, i.e. the speech time, while past tense is interpreted as a deictic
element as it denotes an interval (Partee 1984). Further morphological specifications
for Mood and Aspect in Gr come into play of course. In particular, present
indicative is always inflected for imperfective aspect which allows for a modal
reading as well, at least with non-stative predicates. The present subjunctive, on the
other hand, can bear either perfective or imperfective aspect. The so called aorist
(past tense), however, in both indicatives and subjunctives is inflected for +past and
+perfective features: the combination of these two positive values gives rise to the
notion of temporal specificity (Tsimpli and Roussou 1993), and an iterative reading
is excluded. The latter is achieved if past tense is inflected for imperfective aspect.
Further morphological combinations of Tense (+/-past), grammatical Aspect (+/perfective), and Mood (+/-indicative) allow for a number of possible interpretations
178
ANNA ROUSSOU
which need not concern us at present. What is of interest in the current paper is the
fact that the morphological realisation of an expletive T as present is to be expected
if we consider further (morphological) properties of the language in question. As far
as past tense subjunctive is concerned we still need to account for the conditions that
trigger it in the presence of an expletive T in the matrix clause.
Let us then assume that the present tense found with epistemic modality is an
expletive T, i.e. it is specified for categorial features but has no temporal variable. In
that respect, an expletive T is similar to other expletives, such as there which is also
specified for categorial D features but has no agreement and/or Case features
(Chomsky 1995). If this assumption is correct, then an expletive T is expected to
show syntactic properties similar to those of an expletive D. In order to understand
the function of an expletive T, let us first look at the case of D-expletives.
According to Chomsky (1995), the expletive there in (21) is inserted under Merge to
satisfy the strong D-feature ofT (the Extended Projection Principle):
(21)
There arrived three men
What is usually referred to as the expletive-associate pattern, arises from the fact
that the (nominative) Case-features associated with T cannot be checked by there
which is not specified for any of these properties. As a result T attracts the NPassociate. Assuming that this is covert movement, then it is only sufficient for the
relevant features to move, under the Move-F operation, avoiding pied-piping of
lexical/phrasal material. Under Move-F then the relevant features enter a checking
configuration with T and therefore the derivation converges. This operation explains
then why agreement comes from the associate and not from the expletive in theretype constructions. What is interesting to note is that under this account, there is no
need to postulate expletive-replacement procedures as in Chomsky (1986), or
Chomsky and Lasnik (1993). Crucially, any kind of movement (a kind of raising) is
not triggered to satisfy the needs of the expletive, but of the associate in
combination with T.
Similar considerations can naturally extend to expletive T in that it does not
have to be replaced and therefore any kind of T-raising from the embedded clause
takes place in order to satisfy certain properties (which can be formulated in terms
of feature checking) relating to the embedded T. Furthermore if this is not a
requirement on the matrix T, then there must be some other head that attracts the
embedded T. Let us assume that the presence of T categorial features in a clause
stems from the requirement that the C head which is responsible for T-anchoring
and Force specification enter a dependency (checking) with a T-head. Recall that
under En~'s (1987) approach T-anchoring, the temporal evaluation of clauses, is
mediated via the C position. This (C, T) dependency is overtly manifested in V-2
languages for example in which case T has moved to C (cf. Roberts and Roussou
(1996) and references therein). Regarding complement clauses, T-anchoring takes
place indirectly via the matrix T; this is achieved to the extent that locality
conditions are satisfied (Hornstein 1990, Manzini 1995). In particular, the idea is
179
that the embedded T forms a dependency with its C. Given that the embedded C
forms a dependency with the matrix V (selection), T-anchoring proceeds through
the embedded C to matrix V and from there to matrix T which ultimately enters a
dependency with the root C. The crucial step is that the embedded T via its own C
reaches the next T up. Stating this in terms of Feature-attraction we could say that it
is essentially the topmost C that attracts all T-features. Alternatively, we could say
that the root C binds all T-variables, provided locality conditions are satisfied.
Suppose next that the matrix clause has an expletive T. T-anchoring proceeds in
the way mentioned above as far as the embedded clause is concerned: the embedded
T reaches its C and from there matrix V. However, once the dependency reaches
matrix T there is no variable available due to the expletive nature of T in this case.
Thus the only variable available that can be bound by root C is that provided by the
embedded T. In that respect then we could say that the embedded T acquires matrix
scope and the independent time reference of the complement clause is derived.
Alternatively, we could assume that there are no properties of the matrix T that
attract the embedded T, apart from the fact that the root C requires a variable: in this
case there is only one provided and that comes from the embedded clause. Once Traising has taken place in the context of epistemic modality, the matrix and the
embedded clause can be taken to form a single proposition in terms of their tense
properties (see also Rizzi 1978, Burzio 1986 and Zubizarreta 1982), among others
for treating epistemic modals as raising predicates). Matrix scope of the embedded T
is then expected on the assumption that the higher T is an expletive. Interestingly a
similar phenomenon is found with D-expletives: raising of the NP-features of the
associate allow it to take matrix scope. This is attested in cases of anaphor-binding
as illustrated in the following example (Chomsky 1995):
(22)
a.
b.
There arrived three men without introducing themselves

[IP There [I+NP[FF] ][arrived three men.... ]]
Once the features of the associate have risen to T they are in the right configuration
(c-command) to bind the anaphor in the adjunct clause. In other words binding is
performed from the T position. Thus T- and D-expletives pattern alike, as expected.
Regarding na-complements to epistemic modals the question that arises is why
independent time reference of the complement clause is only attested in this case.
The answer is rather straightforward given the previous discussion: it is only in this
context that there is an expletive matrix T. Indeed this excludes the possibility of
having independent time reference with oti-complements, given that the latter being
in the indicative mood are never found with modals in the frrst place. As far as Tanchoring of oli-clauses is concerned, this proceeds in the standard way: the
embedded T is bound by its own C which in turn forms (a T-anchoring) dependency
with the matrix T (and ultimately the root C). Finally, T-raising is simply not
attested in non-epistemic modals given that these do not have an expletive T: ability,
permission, and obligation are properties that can change over time, or to be more
precise are temporally defmed, as opposed to the notions of necessity and possibility
180
ANNA ROUSSOU
which are not temporally evaluated but are dermed in terms of possible worlds (cf.
Iatridou 1990 for further discussion). In other words, the lower T acquires higher
scope with respect to the higher T, only when the latter is an expletive, i.e. it does
not introdue a temporal variable.
There is one more point that needs to be considered and which in fact offers
further support for the claim that the expletive T associated with epistemic modality
exhibits properties similar to those of D-expletives. It is well-known that when two
modals co-occur their relative order is fixed, so that the epistemic modal precedes
the non-epistemic one. This is illustrated by the following Gr example:
(23)
a.
Bori
[na eprepe
[na fighun]]
may-3sg subj must-3sg subj leave-3p
'It is possible that they should have left'
b. *Boruse [na prepi
[na efighan]]
could-3sg subj must-3sg subj left-3p
'*They could should have left'
Picallo (1985) argues, on the basis of the Romance data, that modals in their nonepistemic (i.e root) reading are analysed as VP elements. Epistemic modals, on the
other hand, are base-generated in !NFL. Thus under her analysis the order where the
epistemic modal precedes the non-epistemic one is well-formed because the former
is under I and the latter is under V. However, if the non-epistemic precedes the
epistemic modal then the construction is ill-formed, since it would imply that I is
lower than V in the clause structure. Hence the ungrammaticality of sentences like
the one in (23b). Picallo's analysis is rather difficult to maintain for the Gr data,
since the two modals occur in different clauses anyway, as the presence of na
indicates. In terms of the present analysis the order in (23a) as well as the
ungrammaticality of (23b) are predicted. Consider (23a) first with the order
epistemic - non-epistemic modal, schematically represented in (24):
(24)
[T\ [bori] [T2 [na eprepe] [T3 [na fighun]]]
epistemic - non-epistemic
Given the previous discussion, Th the one associated with epistemic modality is an
expletive, while T2 and T3 are not. This is illustrated by the fact that T2 is inflected
for past tense (subjunctive) in (23a). Following our previous assumption about the
temporal evaluation of clauses, T-anchoring proceeds as follows: T3 forms a
dependency via its C with T2, satisfying locality conditions. Moreover, T2 introduces
a temporal variable and therefore it contributes to the anchoring of T3. Linking of
T2 to Tb however, is different given that T\ is an expletive. Thus T-anchoring
proceeds in the way already described: T2 acquires matrix scope and is directly
anchored to the root C. In fact this is reminiscent, despite other apparent differences,
oflong-distance expletive-associate pairs in there-constructions shown in (25):
(25)
181
There seem to have arrived three girls
In (25) the NP-associate three girls is ultimately attracted to matrix T for Casefeature checking. This is achieved to the extent that the intervening T is an
infmitival and has no (nominative) Case-features that need to be checked, allowing
for the NP-associate to raise to matrix T.
Consider next the ungrammaticality of (23b), schematically represented in (26):
(26)
*[ Tl [boruse] [T2 [na prepi] [T3 [na efighan]]]

*non- epistemic - epistemic
As (26) shows, Tl is associated with the non-epistemic modal, while T2 is associated

with the epistemic and, according to the present discussion, is an expletive element.
Thus the difference between (24) and (26) is that in the former case the T-anchoring
dependency is headed by an expletive element, while in the latter the expletive
element occurs in an intermediate position yielding an ungrammatical output. To be
more precise, it seems to be the case that expletive elements can only appear in the
highest position of the dependency.3 This is true of D-expletives as well, as the
example in (27) shows:
(27)
a. *A man seems that there has arrived

b. It seems that there has arrived a man
In (27a) super-raising has taken place: the NP associate has raised to the matrix
clause to check the Case feature ofT. Raising has taken place over the embedded T,
whose Case-features remain unchecked, giving rise to ungrammaticality. In other
words, raising of this type violates minimality conditions on movement, since the
lower T is a closer attractor for the associate NP. Regarding (27b), the NP-associate
(its [FF] features) raises up to the lower T, and the Case features of the matrix Tare
checked under merge with the appropriate expletive, that is it; hence the
grammaticality. Similar considerations extend to (26): raising of the most embedded
T3 over the expletive T2 is ruled out, under minimality conditions. To be more
precise, assuming that there is a (C, T) dependency formed which is essential for Tanchoring purposes as well, and that in complement clauses it is essentially the
embedded C that establishes a link with the higher Y, the problem in (26) is that T3
has skipped the intermediate C position. In other words the intermediate C is a
closer attractor and T-anchoring to the root C ultimately has to proceed through that
position. It is in that respect then that the ungrammaticality of (26) is similar to
super-raising constructions. Thus once again the similarities between T- and Dexpletives are illustrated.
The discussion provided so far shows that the postulation of an expletive T
associated with epistemic modality receives support on independent grounds: the
association of an expletive T with an argumental T satisfies all locality conditions
that relate to D-NP expletive-argument associates. The primitive notions of
182
ANNA ROUSSOU
attraction and minimality apply in the same way in both cases. Furthermore, the
tense sequences summarised in (13) above with respect to epistemic modals are now
derived on the basis that the embedded T acquires matrix scope, giving rise to a
reading where the na-clause shows independent time reference.
5. CONCLUSION
To summarise, in the present paper some of the tense properties of na-complements
were discussed. The presence of the subjunctive in certain contexts was accounted
for on the assumption that the subjunctive T is triggered by the appropriate
sentential/intensional operator (Manzini (forthcoming. The crucial cases were
those in which ana-clause occured as a complement to an epistemic predicate. It
was argued that the epistemic predicate in this case has an (implicit) modal reading,
triggering therefore a subjunctive complement. Furthermore this assumption was
conftrmed by the fact that the tense sequences betwen the matrix and the embedded
clauses were the ones found with other epistemic modals (cf. (13) above). In either
case the na-complement exhibits independent time reference which was argued to
be due to the fact that the T associated with epistemic modality is essentially an
expletive element. Following Iatridou (1990), we assumed that the notions of
necessity and possibility are not sensitive to time and therefore do not introduce a
temporal variable. The presence of an expletive T, which in languages like Gr is
morphologically realised as present tense, triggers then T-raising so that the
embedded T acquires matrix scope, yielding the effect of independent time
reference of the na-clause. In that respect the tense sequences given in (13) with
epistemic modals are derived. Furthermore, the present analysis points towards
certain (expected) similarities between T- and D-expletives especially as far as
locality conditions are concerned: feature attraction and minimality.
NOTES
I would like to thank Rita Manzini, Ian Roberts, and the anonymous reviewers for comments and
suggestions, as well as the audiences at the LAGB Auturrm meeting in Middlessex, the Berlin workshop
on Modern Greek Syntax and the Bangor Research Seminar.
2
As a point of clarification: Iatridou refers to this modality as metaphysical and she uses the term
'epistemic' in a slightly different way.
3
For Chomsky (1986) this follows from the Theta-Critierion and the impossibility of movement
from a theta- to another theta-position. Brody (1995) derives this effect under the assumption that only
root positions of chains can be thematic. Thus expletives, being non-thematic, cannot occur in root
positions. This claim extends to Tense, assuming that a T head that introduces a temporal variable is an
argument.
183
References
Brody, M. (1995) Lexico-Logical Form: A Radically Minimalist Theory. MIT Press, Cambridge, Mass.
Chomsky, N. (1986) Knowledge ofLonguage. Praeger, New York.
Chomsky, N. and H. Lasnik (1993) Principles and Parameters Theory, in J. Jacobs, A. von Stechow, W.
Stemefeld, and T. Vennemann (eds) Syntax: An International Handbook of Contemporary
Research, Mouten de Gruyter, Berlin, pp. 506-569.
En~, M. (1987) Anchoring Conditions for Tense, Lingustic Inquiry 18, 633-657.
Farkas, D. (1992) On the Semantics of Subjunctive Complements, in P. Hischbiilher and K. Koerner
(eds.) Romance Languages and Modern Linguistics, John Benjamins, Amsterdam, pp. 69-104.
Heim, I. (1982) The Semantics of Definite and Indefinite Noun Phrases, GSLA, University of
Massachussets, Amherst.
Huntley, M. (1984) The Semantics of English Imperatives, Linguistics and Philosophy 7,103-134.
Iatridou, S. (1990) The Past, the Possible, and the Evident,Linguistic Inquiry 21,123-129.
Ingria, R (1981) Sentential Complementation in Modem Greek, Ph.D. Dissertation, MIT.
Jacobs, R (1981) On being Hypothetical, in R Hendrick, C. Masek and M. Miller (eds.) Papers from
the 17th Regional Meeting, Chicago Linguistic Society, Chicago University Press, Chicago, pp.
Joseph, B. (1983) The Sychrony and Diachrony of the Balkan Infinitives, Cambridge University Press,
Cambridge.
Kempchinsky, P. (1986) Romance Subjunctive Clauses and Logical Form, PhD Dissertation, UCLA.
Kratzer, A. (1981) The Notional Category of Modality, in H. Eikmeyer and H. Rieser (eds.) Words,
Worlds and Contexts: New Approaches in Word Semantics, Mouton de Gruyter, Berlin.
Lakoff, R (1968) Abstract Syntax and Latin Complementation, MIT Press, Cambridge, Mass.
Lightfoot, D. (1975) Natural Logic and the Greek Moods, Mouton de Gruyter, The Hague.
Lyons, J. (1977) Semantics, Vol. 2, Cambridge University Press, Cambridge.
Manzini, M.R (1995) From Merge and Move to Form Dependency, UCL Working Papers in Linguistics
7,323-345.
Manzini, M.R. (forthcoming) The Subjunctives, in P. Coopmans, M. Everaert, and J. Grimshaw (eds)
Lexical Specification and Lexical Insertion. LEA, Hillsdale.
Partee, B. (1984) Nominal and Temporal Anaphora,Linguistics and Philosophy 7, 243-286.
Philippaki-Warburton, I. & I. Veloudis (1984) The Subjunctive in Complement Clauses, Studies in
PicaBo, C. (1985) Opaque Domains, Ph.D. Dissertation, CUNY.
Rizzi, L. (1978) A Restructuring Rule in Italian Syntax, in S. Keyser (ed.) Recent Transformational
Studies in European Languages, MIT Press, Cambridge, Mass.
Roberts, I. and A. Roussou, A. (1996) Speculations on the Phenomenon of Verb Second, Bangor
Research Papers in Linguistics 7, 48-61.
Rouchota, V. (1991) The Interpretation of Na-Clauses in Modem Greek: a Relevance Theoretic
Approach, UCL Working Papers in Linguistics 3, 255-277.
Roussou, A. (1994) The Syntax ofComplementisers. PhD Dissertation, UCL.
Tsimpli, I-M. and A. Roussou (1993) Polarity Items in Modem Greek: their Distribution and
Interpretation, UCL Working Papers in Lingustics 5,129-159.
Tsoulas, G. (1994) The Subjunctive as an Indefinite, unpublished manuscript, University of Paris VIII.
Veloudis, I. (1985) Tensed oa-complements in Modem Greek, Studies in Greek Linguistics 6,183-198.
Veloudis, I. & I. Philippaki-Warburton (1983) The Subjunctive Mood in Modem Greek, Studies in
Greek Linguistics 4,151-168.
Zubizaretta, M.L. (1982) On the Relationship of the Lexicon to Syntax, Ph.D. Dissertation, MIT.
THETA-ROLE SATURATION IN GREEK COMPOUNDS'

Anna-Maria di Sciullo
Universite du Quebec a Montreal
Angela Ralli
University ofAthens
1. INTRODUCTION
This study deals with theta-role saturation in deverbal and verbal compounds in
Greek. We claim that theta-role saturation inside compounds is related to the
configurational properties of argument structure, as well as to the properties of rich
morphology, particularly to strong inflection. This claim, combined with the fact
that several semantic roles may be expressed in an adjunct position whereas a more
restricted set of semantic roles is represented in a complement position, can account
for theta-role saturation inside Greek compounds.
The paper is organized as follows. The frrst section provides general
information about the structural properties of Greek deverbal and verbal compounds
and the kind of theta-roles saturated inside these compounds. The second section
contains the basic assumptions and claims concerning morphology and argument
structure in general. An analysis of Greek deverbal and verbal compounds is given
in the last section.
2. THE STRUCTURE OF GREEK DEVERBAL AND VERBAL

COMPOUNDS
Greek deverbal compounds with internal theta-role saturation are basically nouns or
adjectives. In this paper, however, we will also account for verbal compounds since
they display internal theta-role saturation.
Nominal deverbal compounds are realized as one-string formations containing
at least two stems and two suffixes, one derivational and one inflectional (cf. (1.2
The constituent parts are overtly realized in the following order: the inflectional
suffix, responsible for the morpho syntactic features of number and case, appears at
the righthand edge of the compound next to the deverbal item;3 the derivational
suffix, transforming the verbal stem into a nominal stem, is attached to the verbal
stem and a bare nominal stem is the lefthand constituent:
185
A. A1exiadou et al. (eds.), Studies in Greek Syntax, 185-199.

ANNA-MARIA DI SCIULLO AND ANGELA RALLI
186
(1) a.
ihthiokalierjia4
lit. 'fish-culture'
-a
< ihthi- kaliergh- S -i'fish'-'cultivate'- der.suf.<noun, fem> infl.suf.<nom. sing.>
b. kozmohalazmos
'world-destruction'
< kozm- xalas-6 -m-os
'world' 'destroy' der.suf.<noun,masc> infl.suf.<nom.sing.>
c. nerovrastos
'boiled in the water'
< ner- vras- -t-os
'water' 'boil' der.suf.<adj.> infl.suf.<nom., sing.>
In verbal compounds, on the other hand, a derivational suffix may not be

present in the structure (2a) but the inflectional suffix is always present denoting
agreement features. If, however, a derivational suffix appears attached to the
righthand member of the compound, this suffix is responsible for the verbal
category of the member (2b).
(2) a.
hartopezo
< hart- pez- -0
'I play cards'
'card' 'play' 1st per. sing. pres. active
b. Haropalevo
'I fight with death'
< har- pal-ev-0
'death' 'fight' <N> der. suff.<V> 1st per. sing. pres. active
It should be noticed that between the members of both deverbal and verbal
compounds, the surface ordering is fixed, that is, the verbal, or the deverbal element,
occupies the second position. Furthermore, a vowel -0- appearing between the frrst
and the second member of the compound constitutes a transition vowel the presence
of which is independently motivated from the two members participating in
composition, as shown in Ralli (1988, 1992V
Greek deverbal compounds, as is the case of deverbal compounds in English
and other languages, have both XO and XP properties. They are formally similar to
affixed words and dissimilar to phrases at the interface with the performance
systems. The compounds are head fmallike affixed words, but not like phrases. On
the other hand, deverbal compounds share some properties of phrasal structure with
respect to theta-role saturation, since a considerable number of deverbal compounds
are object-verb structures and theta-role saturation occurs in these structures:
(3) a.
nihokoptis < nih- koptis

lit. nail-cutter 'nail' 'cutter'
'nail-clipper'
b. hrimatodhotisi < hrimat- dhotisi
'money-giving' 'money' 'giving'
GREEK COMPOUNDS
187
According to Di Sciullo & Williams (1987), English deverbal compounds (e.g.

tree-eater, thanks-giving, etc.) are morphological objects, that is expressions that are
generated in morphology, whereas French compounds such as essuie-glace "wiper"
and tire-bouchon "corkscrew" are syntactic words, that is expressions generated in
syntax even though they are syntactic atoms. Syntactic atomicity holds for these
objects that do not have a morphological form but otherwise display the general
properties of XOs. In Di Sciullo & Williams' theory, syntactic words are fIrstly
generated in syntax and then reanalyzed as words. They differ from idioms though,
which are listed phrases, since their meaning is not compositional. Certain idioms
are not syntactically atomic, however, since parts of idioms can be extracted (4a,b)
or be subject to independent referential relations (4c):
(4) a. English: tabs were put on DP
b. French: en voir de toutes les couleurs
c. Greek: evale nero sto
krasi tu
lit. he put water in the wine his
'his attitude became softer'
Similarly, parts of deverbal compounds cannot be extracted. This is the case for
English deverbal compounds, as discussed in Di Sciullo & Williams (1987), and
they are islands with respect to XP anaphoric relations, as noted in Giorgi &
Longobardi (1991) (cf. (5)). This is also the case in Greek (cf. (6)):
(5) a.
b.
(6) a.
*The [who-killer] did the police catch?

*A [person-informer] about himself
*Ton [ti-pehti] epjase i astinomia
the what-player did the police catch?
vs. i astinomia epjase ton hartopexti
the police
caught the card gambler
b. * Ejine
anthropofaghos
tu eaftu tu
he became a man-eater (cannibal) of himself
vs. ejine anthropofaghos
he became a man-eater
In this paper, we would like to take a unifIed approach to compound formation

and adopt the idea that there is only one computational space for the derivation of
the linguistic expressions which are formed by autonomous components: the
morphology, the syntax and the phonology (cf. Chomsky 1995, Di Sciullo 1996a,b).
We will thus take deverbal compounds to be generated in the computational
component of the grammar in a way that we will make precise below.
As has been noted elsewhere, (cf. Di Sciullo & Williams 1987, on the basis of
English and Italian; Roeper 1987 and Grimshaw 1990, on the basis of English; Di
Sciullo 1992, on the basis of Italian and English), and it has been proved by Ralli
(1989, 1992) for Greek, theta-roles may be saturated inside compounds. This is
evidenced in the following examples, where the admissibility of a by-phrase (7a)
188
and the presence of an aspectual modifier (7b) indicate that, depending on the case,
some thematic roles are saturated inside the compound.
(7) a.
i eleokalierjies
apo tus agbrotes
'the olive-cultures by the farmers
esosan tin ikonomia tu nisju
saved the economy of the island'
b. i
sinexis
hartopeksia ton katandise ftoho
'the continuous card-playing has made him poor'
If this were not the case, it would be impossible to explain why some theta-roles
of the verb included in the deverbal compound may not be saturated outside of the
compound:
(8) a.
b.
*aftos ine enas hartopektis pebnidhjon

card-player of games'
'this is a
*i simerini laosinaksi
ton anthropon ine ekpliktiki
is surprising'
'today's people-reunion of men
As a matter of fact, some deverbal compounds in -tis, like the one in (9a), are
complete functional complexes with all arguments saturated inside the
morphological structure of the compound. This is also the case in French, as noted
in Di Sciullo (1992).
(9) a.
o/enas kardhjokataktitis molis ilthe

'thela heart conqueror just came in'
b. *0 kardhjokataktitis apo to Jani bori na ine epikindhinos
'the heart conqueror by John may be dangerous'
c. *natos
0 kardhjokataktitis ton kardhjon
'here comes the heart conqueror of hearts'
A closer examination of deverbal compounds (cf. (10 reveals that in Greek,

the range of theta-roles saturated inside seems not to be restricted, contrary to what
has been claimed until now on the basis of data from other languages (cf. Pesetsky
1994).
(10)
psihopsaksimo < psix- psaksimo

(Theme)
lit.'soul-searcbing' 'soul' 'searching'
b. anemodharmenos < anem- dharmenos (Agent)
'beaten by the wind' 'wind' 'beaten'
c. kondarohtipima < kondar- htipima (Instrument)
'pole-stroke'
'pole'
'stroke'
d. anthospartos
< anth- spartos (InstrumentlMaterial)
'flower-strewn'
'flower' 'strewn'
a.
GREEK COMPOUNDS
e.
f.
189
ematokilizma
< emat- kilizma
(Location)
'wallowing in blood' 'blood' 'wallowing'
poltopiisi
< polt- piisi
(Result)
lit. 'pulp making'
'pulp' 'making'
3. THE FRAMEWORK
Along the lines of Williams & Di Sciullo (1987), Chomsky (1995) and Di Sciullo
(1996a, b), we accept here that morphology is an autonomous component of the
grammar providing structural descriptions for word formations. Assuming that there
is only one computational space for the generation of linguistic expressions, word
formations are not accessible to the syntactic operations of MERGE and MOVE, as
defmed in Chomsky (1995). Contrary to MERGE, the morphological operations of
composition and linking do not create new categories. Contrary to MOVE, the
morphological operations are not subject to the Minimal Link Condition, again as
defmed in Chomsky (1995). See Di Sciullo (1996b) for discussion.
We will take the morphological component to generate visible, i.e., interpretable
head adjunction structures at the interface with the C-I system. We will refer to this
interface as to Morphological Form (MF) and assume the defmition given by Di
Sciullo (1996b). MF is the XO dimension of LF interfacing with the C-I system for
XO interpretation. The existence of an XO dimension to LF is motivated in Di Sciullo
(1996b), where it is shown that word internal interpretation, i.e., conceptual and
referential opacity is not obtained in phrasal structure. In phrasal structure, the
predicate argument structure interpretation is obtained, thus defmite as well as
indefmite reference for nominal expressions, and truth values for phrasal
expressions.
Assuming the architecture in (11), XO expressions may have a phrasal structure
at Spell-Out, but not at MFILF or at MFIPF, which we will not discuss here.
(11)
Spell-Out
I \
MFIPF MFILF
At MF/LF, XO expressions have an adjunct structure and not a specifier-headcomplement structure. This is a consequence of the hypothesis that the performance
system interprets canonical target configurations, that is configurations the form of
which is non-ambiguous with respect to the object generated by the grammar, either
word structure or phrase structure.
190
4.
COMPOUNDS AND THEMATIC STRUCTURE
In this paper, the study of theta-role saturation in Greek compounds relies on the
following assumptions:
a) Theta-roles are not linearly represented but only configurationally. As it will
become clear below, this assumption may partly account for the representation of
theta-roles in the grammar and their saturation inside compounds.
b) The configurational approach of theta-roles should be viewed in conjunction
with Kayne's (1994) proposal according to which, in a linguistic expression, the
order of constituents determines the structural relation between them. That is, an
element that precedes the head of a configuration is an adjunct and an element that
follows the head is a complement. It follows then, that in a compound, a category
that precedes the head is an adjunct. This also follows from Di Sciullo & Williams'
(1987) Relativized Head Hypothesis, according to which the head of a word with
respect to a given feature F is the rightmost F-marked constituent.
c) Arguments in compounds are considered to be nominal categories in adjunct
position. However, an argument may also appear in head position when it is
represented by a suffix, as shown in Di Sciullo (1992). Thus, in the simplified
representation given under (12), the noun stem orghan, being in the adjunct position
with respect to the derived-inflected noun pehtis, saturates the Theme role of the
base verb. Moreover, the suffix -tis saturates the Agent role of the verb but, this
time, it is the suffix which is in head position. 8 The fact that the suffix -tis saturates
the Agent theta-role in this type of compounds has been already pointed out by
Kakouriotis (1993) for Greek and by Di Sciullo (1992) for Italian and English. 9
(12) N (orghanopehtis "instrument-player" 10)
/ \
N N
I / \
I V N
I I I
orghan peh tis

d) With respect to the semantic interpretation of compounds whose nominal
constituents have no independent definite reference in syntax and generally have a
generic/indefmite reading, it is assumed, following Di Sciullo (1993), that this
interpretation is not descriptive. As stated above, this is considered to be a
consequence of the fact that compound structures, as other word formations, are
adjunct-head structures at MF/LF where they receive an interpretation and that only
XO predicative meaning can be obtained at that interface.
According to Di Sciullo (1996a), compounds are adjunct-head structures at the
interface (cf. (13 but they may not be at Spell-Out (cf. (14, given independent
GREEK COMPOUNDS
191
properties of the languages. Thus, in (13a), the adjunct is not related/linked to an

argument position e, as it is the case in (13b). Furthermore, no legitimate compound
structure may include overtly both adjunct and complement positions at Spell-Out,
as in (15), the core cases of compounds being two membered categories.
(13)a.
X
/ \
adjunct X
X
/ \
adjunct X
/ \
b.
X e
(14)
X
/ \
X Compl
(15) * X
/ \
adjunct X
/ \
X Compl
Since compounds in Greek, as it is the case in English, have the adjunct-head

structure at Spell-Out and the adjunct mayor may not be linked to an argument
position as schematized in (13), we predict that various semantic roles may be
expressed in these compounds:
(16)
(17)
Greek:
< alogh- kleftis
(Theme)
a. aloghokleftis
'horse-thief, lit. stealer'" 'horse' 'thief
< Turk- kratumenos
(Agent)
b. Turkokratumenos
'Turk' 'occupied'
'occupied by Turks'
c. oksighonokolisi
<
oksighon- kolisi (Instrument)
'welding, lit. attaching with oxygen' 'oxygen' 'attaching'
< plakstrosi
(InstrumentlMaterial)
d. plakostrosi
'flat-stone-paving'
'stone' 'paving'
(Location/Source)
e. uranokatevatos < uran- katevatos
'sky-corne-down'
'sky' 'come-down'
f. aghrotodhaniodhotisi < aghrot- dhani- -dhotisi (Goal, Theme)ll
'farmer-loan-giving'
'farmer' 'loan' 'giving'
English:
a. exam-giving
(Theme)
b. student-exam-giving
(Goal, Theme)
c. wind-blown (flyer)
(Agent)
d. mountain-sky-watching (can be fun) (Location, Theme)
This is not the case in Romance languages, where the head-complement

configuration is obtained at Spell-Out (cf. (14)), and thus a restricted set of semantic
192
roles may be saturated within the compounds, usually the Theme role, as
exemplified in (18).
(18)
Italian:
porta-ombrelli "umbrella-holder"
French:
essuie-mains "hand-towel"
Spanish:
lavaplatos "dish-washer"
Portuguese: abre-latas "tin-opener,,12
We assume that, like in English, Greek compounds have the adjunct-head

configuration at Spell-Out because their lefthand noun must not be in a position
where case is assigned. As stated in the second section of the paper, this noun is a
bare stem without an overt inflectional suffIx and, as such, it cannot undergo
Longobardi's (1994) overt N to D movement, as proposed in Di Sciullo (1996a) for
Romance compounds, which are syntactic words. Thus in Greek, this noun must
appear in adjunct position at Spell-Out, a position where case is not visible and a
position where a large variety of roles can be expressed.
In a language with weak morphology, such as English, compound internal thetarole saturation is restricted, although not limited to only one theta-role, as is the
case for most Romance languages, given the possibility of recursion to the left. On
the contrary, in Greek, which is morphologically rich, bearing overt verbal and
nominal inflection, as well as a significantly rich derivation, compound internal role
saturation is particularly extensive.
We will see below that the interaction of rich morphology with the
configurational properties of compound structures allows us to account for the facts
in Greek.
5. AN ANALYSIS OF GREEK DEVERBAL AND VERBAL COMPOUNDS

As we have seen in the second section, in a typical Greek deverbal compound, a
bare noun precedes a deverbal inflected category. Moreover, the language allows
various theta-roles to be saturated inside the compound.
The bare noun in the adjunct position acts as a modifier of the morphologically
complex head and may assume an additional semantic role to the one already
expressed by the theta-role originating in the lower complement position. This can
be seen in a number of compounds where the nominal fIrst member may denote
both a theta-role and a Possessor role. For example, in a compound such as
horofilakas 'country-guard/country-keeper', the nominal stem hor- 'country' can be
an argument of the predicative head filak- 'guard/keeper', and thus related to the
Theme role of that head, as well as a modifier of the derived-inflected nominal
filakas, where then denotes a Possessor role. This fact further motivates the structure
given under (13b), where the nominal underived stem hor-, being an adjunct to the
derived-inflected nominal elementfilakas, is linked to the argument position of the
verbal base filak-. The position of the adjunct next to the complex nominal is also
GREEK COMPOUNDS
193
supported, if we assume, following Grimshaw (1990), that, contrary to verbs, nouns

do not theta-mark.
(19)
N I3
/ \
N N
1/\
N
hor- V
/\ I
V e -as
filakLet us see now how rich morphology, namely suffIxation, can affect theta-role
saturation inside compounds.
Until now, we have considered suffIxes to be carriers of both derivational and
inflectional properties (cf. (12), (19. In what follows, we will make a distinction
between derivational and inflectional suffIxes since in most Greek morphologically
complex words, the two types of suffIxes are clearly separated and do not
intermingle. See Ralli (1994b) for a detailed study of these two types of suffIxation.
In general, the relation between suffIxation and the presence of arguments
within the structure of Greek deverbal compounds has been taken in consideration
by Roeper's (1987) statement according to which, an argument-head relation is
allowed in compounds if a suffIx is present.
With respect to derivation, in Greek deverbal compounds, a derivational suffIx,
transforming the verb into a noun, is generally present. 14 As we see below, Greek
data support the view that it is the presence of this suffIx that extends the saturation
of various theta-roles inside compounds. For instance, the examples listed under
(20) show that the AgentlInstrument role may be saturated inside compounds
containing specific derivational suffIxes: 15
(20)
a.
b.
c.
d.
e.
thalasodharmenos < thalas- dhar- -men- -os

'beaten-by-the-sea' 'sea'
'beat' -en
nom., sing.
iljolustos
< ili- lus- -t- -os
'washed-by-the-sun' 'sun' 'wash' -ed nom., sing.
pondikofaghoma < pondik- fagho- -ma- -/iJ
'rat-eating'
'rat'
'eat'
-ing nom., sing.
aetopetaghma
< aetpetagh- -ma--/iJ
'eagle-flying'
'eagle' 'fly'
ing nom., sing.
lemonostiftis
< lemon- stiv-ti- -s
'lemon-squeezer'
'lemon' 'squeeze' er nom., sing.
The fact that derivational suffIxes such as -men-, -t- and -ma- (cf. (20a-d allow
theta-role saturation inside compounds has already been observed by Ralli (1992),
where it is proposed that these suffIxes perform a lexical operation on the argument
194
structure of the verb, either by suppressing the real subject argument of the verb or
by transforming it into a complement. On the other hand, it is already an established
fact that the -ti- suffix (cf (20e is linked to the Agent role of the verbal base (cf.
(12) above). Several works exist on this matter, particularly with respect to the
status of the derivational suffix '-er', the English correspondant of -ti(s}. To mention
a few, cf. Fabb (1984), Grimshaw (1990), Oi Sciullo (1992) for English, and
Kakouriotis (1993) for Greek.
Therefore, Roeper's statement about the role of suffixes in deverbal compounds
is valid, but also too general because it does not explain why, for example,
saturation of the Agent role inside deverbal compounds occurs only in the presence
of specific suffixes (cf. (20. As a matter of fact, this statement has been refmed by
Oi Sciullo (1991) who showed that the simple presence of a suffix is not sufficient
to determine whether an argument-head relation holds in the compound structure.
For Oi Sciullo, the determinant factor about the presence of argument-head relations
in compounds is the nature of the suffix. By following this proposal, we can predict
why other Greek derivational suffixes, such as -sri} and -sim(o}, do not allow the
Agent role saturation inside deverbal compounds. 16 The compound words
fidhosirsimo 'snake-crawling', jinekojersimo, 'woman-behavior', trihoptosi, 'hairfall', filoroi 'leaf-flow', proposed by some native speakers as counter-examples to
this statement, constitute only apparent exceptions. In fidhosirsimo and
jinekojersimo we have middle verb formations, and in the examples trihoptosi and
filoroi, the verbal bases belong to the ergative class of verbs. That is in both cases,
we deal with real internal arguments appearing in complement position.
At this point, we are allowed to say that Greek data enable us to pursuie Oi
Sciullo's refinement even further. We will claim that the inherent properties of
derivational suffixes make possible and, at the same time, restrict the choice of
theta-roles within compounds, depending on the nature of the suffix, but the
presence of an inflectional suffix allows the occurence of non-restricted theta-role
saturation. That is why in languages weak in inflectional morphology, like English,
theta-role saturation is possible but limited, contrary to what happens in languages,
like Greek, with strong inflectional morphology, where more theta-roles may be
saturated by the nominal non-head within compounds. It is this property of
inflectional affixes to extend theta-role saturation within compounds, independently
of the presence of derivational affixes, that can explain the presence in Greek of a
considerable number of verbal compounds, as in (21), whose second member is an
underived inflected verbal form and the first member assumes the role of an
argument of the verb. 17
(21)
hartopezo
< hart- pez- -0
(Theme)
'I play cards'
'card' 'play' 1st per. sing. present, active
b. ematokilo
'I (make something) wallow in blood'
-0
(Location)
< emat- kil'blood' 'wallow' 1st per. sing. present, active
a.
GREEK COMPOUNDS
195
thalasodhernome
< thalas- dhern- -orne (Agent)
'I am beaten by the sea' 'sea' 'beat'
1st per.sing. pres. passive
d. dhjaolostelno
'I send (something/somebody) to hell'
< dhjaol- steln- -0
(Goal)
'hell'
'send' 1st per. sing. pres. act.
e. oksighonokolo
lit. 'I attach (something) by using oxygen'
< oksighon- kol-0
(Instrument)
'oxygen'
'attach' 1st per. sing. pres. active
= 'I weld'
c.
The rich internal theta-role saturation in compounds like the ones above suggest that
constraints such as the First Sister Principle (cf. Roeper & Siegel 1978), or the most
recent proposal that a Locative role cannot appear in compounds without the
presence of the Theme role (cf. Pesetsky 1994), cannot be universal. Their
application in a particular language depends on the general configurational
properties of this language as well as to its morphological variation.
Before concluding, we would like to propose the morphological structures in
(22), as possible generation schemas for Greek deverbal (22a) and verbal
compounds (22b). These structures are motivated by the following basic
assumptions:
a) Inflection is a word internal process (cf. Booij 1993, Chomsky 1995, Ralli
forthcoming, among others).
b) Derivational suffixes are distinct from inflectional ones, and both of them are
heads of the structures into which they participate (cf. above, and Di Sciullo &
Williams 1987).
c) In the morphological component, structures are right-headed (cf. sections 2 & 4,
and Williams 1981; Di Sciullo & Williams 1987; Di Sciullo 1996).
(22)
a.
N
/ \
N Ninfl. suf.
/ \
N N
/ \
V Nder.suf.
/ \
V e
b.
V
/ \
V Vinfl.suf.
/ \
N V
/ \
V e
In the structures given above, the bare nouns appear in adjunct position, but are
linked to the verbal complement position e. Thus, the range of theta-roles saturated
inside the compounds, as well as all possible modifier roles expressed by the bare
nouns, are accounted for. Both derivational and inflectional suffixes are represented
196
word internally, and their presence contributes to the extensive saturation of thetaroles.
Notice that the inflectional suffIx consitituent appears at the topmost level of word
formation, and not at a lower level as the structures in (23) below suggest:
(23)
a.
N
/ \
N N
/ \
N Ninfl.suf.
/ \
V Nder.suf.
/ \
V e
b.
V
/ \
N V
/ \
V Vinfl.suf.
/ \
V e
Were the structures in (23) the correct ones, it would be diffIcult to explain why
the lefthand noun is always a stem and is never combined with an inflectional
ending. 18 Moreover, a representation that displays the inflectional suffIx to have
scope over the two members of the compound structure, provides a better account of
the fact that, in some cases, the inflectional endings of the compound words are
different from those which are borne by the second member taken in isolation. The
examples below illustrate this last remark
(24)
ksifomaho < ksif-mah- -0

'I fence'
'sword' 'tight' 1st pers. sing.
vs.
maxome < max- -ome
'I tight'
'tight' 1st per. sing.
6. SUMMARY
In this paper, it has been argued that argument structure defmed in purely
configurational terms is related to rich morphology, both derivational and
inflectional. Evidence is presented to the effect that the conjunction of the
configurational properties of deverbal compounds with the properties of their
morphological system allows to make predictions on the internal saturation of
arguments/theta-roles.
NOTES
IThis work has been supported in part by the grant from the Social Sciences and Humanities Council
(441-92-0012) attributed to Anna Maria Di Sciullo and by a grant from the University of Athens
research Commission (70-4-1681) attributed to Angela RaUi.
GREEK COMPOUNDS
197
More than two stems and more than one derivational affixes may be recursively added to the basic
structure of compounds. Note, however, that only one inflectional suffix is allowed in one-string
compound formations, as in any other lexical morphologically complex item.
3
As shown in Ralli (I 994a), the feature of gender inherently marks stems and/or derivational
suffixes, but not inflectional suffixes, contrary to what traditional grammars describe. The inflectional
properties of Greek compounds are extensively discussed in Di Sciullo & Ralli (1994). We will relate
the inflectional properties to thematic constituents in the last section of the paper.
4
Greek examples will appear unstressed. For a description of the stressing procedure in Greek
compounds, cf. Nespor & Ralli (forthcoming). Conventionally, all nominal compounds will be given in
nominative singular forms, while verbal compounds will appear in the present tense, particularly, in the
form of the first person singular.
S
Phonologically, [gh] is palatalized before [iJ.
[s] becomes [z] before the voiced [mJ.
According to a proposal forwarded by Ralli (1992), the structure of Greek deverbal compounds is
either as in (ia) or as in (ib), assuming that categories such as 'word', 'stem' 'sur (derivational and
inflectional) and 'prer constitute primitives of the morphological analysis (cf. also Ralli 1989 and
I 994b):
b.
N<word>
(i)a.
N<word>
/ \
N<stem> N<word>
/ \
/ \
N<stem> N<infl. suf.>
/ \
N<stem> N <infl. suf.>

N<stem> N <stem>
/ \
/ \
V<stem> N<der.suf>
V<stem> N deriv. suf.
However, only the structure under (ia) is adopted in a phonologically motivated proposal forwarded by
Nespor & Ralli (forthcoming), where it is claimed that deverbal compounds and, generally, compounds
containing a derived-inflected item as their second member behave differently from other compounds
(i.e., compounds whose second member is not derived-inflected) with respect to stressing procedure.
That is, the former preserve the stress of their second member since they are of a [stem word] type, as
opposed to the latter, belonging to a [stem stem] type, which are subject to the application of a special
stress rule bound only to these formations. We will not discuss these properties here.
8
It should be noticed that -tis is not a single suffix in Greek since it contains both derivational (-t-)
and inflectional material (-is). In the last section of the paper, we will make a distinction between
derivational and inflectional suffixes.
In the modular approach assumed here, theta-role saturation by affixes is allowed because
arguments/theta-roles are generally saturated by head chains in word structure and not only bare nouns
but affixes too may be part ofa chain saturating an argument (cf. Di Sciullo 1993).
10
As stated in the second section, the -0- appearing between the first and the second member of the
compound constitutes a transition vowel.
11
Although this compound is not an attested word of the language, it is not considered to be an
ungrammatical formation by most native speakers.
12
For a detailed description of these compounds, cf. Di Sciullo 1991, 1992; Scalise 1992 (Italian),
Zwanenburg 1992 (French), Rainer & Varela 1992 (Spanish) and Villalva 1992 (Portuguese).
13
For simplicity reasons, -as is considered to be an inflectional affix with category-changing
properties. See, however, note 13 for a different view.
14
In some cases though, the derivational or the inflectional suffixes are not overtly realized. These
cases are analyzed by Ralli (1988) as containing a -iii suffix.
An example of a iii derivation process is provided by the noun filakas "guardlkeeper" above (cf.
(19) horofilakas "country-guardlkeeper"), where the nominal stem filak- derives from a verbal stem by
the addition of a iii suffix. -as constitutes the inflectional suffix which is attached to the nominal stem in
order to form the word filakas. A iii inflectional process is illustrated by the examples given under
(20cd).
Notice, however, that no Greek word may display both derivation and inflection realized as iii
suffixation processes.
198
ANNA-MARIA Dr SCIULLO AND ANGELA RALLI
15
It should be noticed that beside the Agent, that is the person who carries out the act, the -tis derived
nominals may also express the Instrument that does the act (cf. Triantaphyllidis, 1941 for a list of these
derived formations in Greek). This type of instrument which may assume the function of a grammatical
subject is called by Maranz (1984) an intermediary instrument.
16
Inflectional suffixes following derivational ones appear in parentheses.
17
Notice that also in English there are verbal compounds with internal theta-role saturation (e.g., to
bar-tend). However, their number is rather restricted compared to the considerable occurrence of verbal
compounds in Greek.
18
A different view is expressed by Nespor & Ralli (forthcoming) who opt for the structures given in
(23) as possible generation schemas of Greek verbal and deverbal compounds. Nespor and Ralli
consider these compounds to belong to the [Stem Word] type. Under this assumption, the lefthand noun
(i.e., the bare noun stem) is added to a fully inflected word containing a stem, of either a nominal or a
verbal category, and a corresponding inflectional suffix. In other words, in Nespor & Ralli's work, the
inflectional suffix is represented as attached to the righthand member of the compound. As stated before
(cf Note 6), Nespor & Ralli's claim is motivated on phonological grounds, while our proposal here
relies on morphological and syntactic considerations. According to recent developments of the theory,
the structures of linguistic expressions do not have to be isomorphic at both interfaces, the PF and the
LF (cf. Chomsky 1995, for discussion).
References
Booij, G. (1993) Against Split Morphology, in G. Booij and J. van Marie (ed.) Yearbook of
Morphology, Kluwer Academic Publishers, Dordrecht.
Chomsky, N. (1994) Bare Phrase Structure, MIT Occasional Papers in Linguistics, no 5. MIT,
Cambridge, Mass.
Di Sciullo, A.M. (1991) On the Structure of Deverbal Compounds, Working Papers in Linguistics, no 3,
University of Venice.
Di Sciullo, A. M. (1992) Deverbal Compounds and the External Argument, in I. Roca (ed.) Thematic
Structure, Its Role in Grammar, Foris, Dordrecht.
Di Sciullo (1993) The Complement Domain ofa Head at Morphological Form, Probus 5, 95-125.
Di Sciullo, A.M. (1994) Selection and Derivational Affixes, in Progress in Morphology, Mouton de
Gruyter, Berlin.
Di Sciullo, A.M. (1996a) Modularity and X/XP Asymmetries, Linguistic Analysis 26,1-26.
Di Sciullo, A.M. (1996b) Atomicity and Relatedness in Configurational Morphology, in A.M. Di
Sciullo (ed.) Configuration, Cascadilla Press, Somervile.
Di Sciullo, A.M. & A. Ralli (1994) Argument Structure and Inflection: Some Differences between
English, Italian and Modern Greek Compounds, Proceedings of the Third Meeting on Multilingual
Aspects ofNominal Composition, ISSCO, Geneva.
Di Sciullo, A.M. & E. Williams (1987) On the Definition of the Word, MIT Press, Cambridge, Mass.
Fabb, N. (1984) Syntactic Affixation, Ph.D. Diss., MIT, Cambridge, Mass.
Giorgi, A. & G. Longobardi (1991). The Syntax of Noun Phrases: Configuration, Parameter and Empty
Categories, Cambridge University Press, Cambridge.
Grimshaw, J. (1990) Argument Structure, MIT Press, Cambridge, Mass.
Kakouriotis, A. (1993) On the so-called Agentive Nominals in English and their Counterparts in
Modern Greek, Proceedings of the 7th Symposium on English and Greek Linguistics, University of
Thessaloniki, Thessaloniki.
Kayne, R. (1994) The Antisymmetry of Syntax, MIT Press, Cambridge, Mass.
Longobardi, G. (1994) Reference and Proper Names, Linguistic Inquiry 25,609-667.
Maranzt A. (1984) On the Nature of Grammatical Relations, MIT Press, Cambridge, Mass.
Nespor, M. & A. Ralli (forthcoming) Morphology-Phonology Interface: The Case of Stress in Greek
Compounds, The Linguistic Review.
Pesetsky, D. (1994), Zero Syntax, MIT Press, Cambridge, Mass.
Rainer, F and S. Varela (1992) Compounding in Spanish, Rivista di Linguistica 4,117-143.
GREEK COMPOUNDS
199
Ralli, A. (1988) Elements de la Morphologie du Grec Modeme: la Structure du Verbe, Ph. D.

Dissertation, Universite de Montreal, Montreal.
Ralli, A. (1989) Greek Verbal Compounds, Studies in Greek Linguistics.
Ralli, A. (1992) Compounds in Modem Greek, Rivista di Linguistica 4, 143-174.
Ralli, A. (1994a) Feature-representations and Feature-passing Operations: The Case of Greek Nominal
Inflection, Proceedings of the 7th Symposium on English and Greek Linguistics. University of
Thessaloniki, Thessaloniki.
Ralli (1994b) Affixation in Lexical Morphology, Studies in Greek Linguistics, Kyriakides Thessaloniki.
Ralli, A. (forthcoming). On the Morphological Status of Inflectional Features: Evidence from Modem
Greek, in G. Horrocks, B. Joseph and I. Philippaki-Warburton (ed.) Themes on Greek Linguistics,
John Benjamins, AmsterdamlPhiiadelphia.
Roeper, T. (1987) Implicit Arguments and the Head-Complement Relation, Linguistic Inquiry 18, 267311.
Roeper, T. & D. Siegel (1978) Transformations and the Lexicon, Linguistic Inquiry 9, 199-260.
Scalise, S. (1992) Compounding in Italian, Rivista di Linguistica 4, 175-201.
Triantaphyllides, M. (1941) Modem Greek Grammar, OEDV, Athens.
Villalva, A. (1992) Compounding in Portuguese, Rivista di Linguistica 4,201-221.
Williams, E. (1981) On the Notions 'Lexically Related' and 'Head of the Word', Linguistic Inquiry 2,
245-274.
Zwanenburg, W. (1992) Compounding in French, Rivista di Linguistica 4, 221-240.
THE POSITION AND SERIALIZATION OF APs IN THE DP:

EVIDENCE FROM GREEK t
Melita Stavrou
Aristotle University of Thessaloniki
1. INTRODUCTION
The recent growth of interest in the syntactic relationship between adjectives and
nouns within DPs is due to the extended research in the structure of both the DP and
the clause in terms of functional categories and their interaction. The history of this
approach goes back to the 'split' of the NP in one functional (D) and one lexical (N)
domain, thus becoming in effect an exocentric construction paralleling the 'exocentricity' of the clause. The establishment of the 'DP' as a functional projection, to which
the lexical NP is embedded in order to be able to function as an argument, has been
independently argued for by Abney (1987), Szabolsci (1983, 1987, 1989) and
Horrocks & Stavrou (1985, 1987). Ever since, there has been a growing literature
concerning the number and content of the possible functional projections which can
intervene between D and NP2, the evidence adduced being focused on the number,
scope and lexical content of modifying adjectives.
In this paper I would like to contribute to the discussion of this evidence; in
particular, I shall consider the issue of the number and order of adjectives that can
modify the noun in the DP, research which is currently guided by the widely accepted
claim that the "internal structure of Noun Phrases is strictly parallel to that of CP in all
respects" (Cinque 1993; Valois 1990: 1; Crisma 1993; Zamparelli 1994; also
Alexiadou 1994: 222). The primary aim of the paper is to present evidence which
quite strongly suggests that, unlike what has been recently proposed for Romance
(Cinque 1993; Zamparelli 1994; Bosque and Picallo 1994), there is no need for more
than one functional projection between D and NP in Greek (Gr); this projection,
roughly similar to Ritter's Number Phrase (Ritter 1991), is independently needed to
account for the movement of the noun in indefmite DPs in Or (Karanassios 1992;
Stavrou 1996).
201
202
MELITA STAVROU
2. BACKGROUND LITERATURE
The point to which one can push the parallelism between nominal and clausal elements, beyond the well established parallelism between DP and CPIIP, is both an
empirical and a theoretical issue, and discussion of it continues the research inaugurated by Chomsky's pivotal 1970 article Remarks on Nominalization. Chomsky
dealt there specifically with derived nominals (and their comparison to gerunds), but
today's extended research into clausal and nominal structure seeks to collapse both
'derived' and 'common' nominal structure with that of clauses. This is not an easy
enterprise (see Cinque 1993: 32), given the frequently illusory nature of the empirical
foundation of such a hypothesis. In what follows I will concentrate on the difficulties
that arise in handling the data of multiple adjectival modification, commenting also on
the most recent literature. In essence, I will examine whether there is a fixed order for
APs in the DP, and if so, what this is, and how it can be syntactically accounted for. I
shall confme myself to the evidence provided by common (object/non-derived) nouns,
since these have not been examined as thoroughly as derived or eventive nominals,
and are apparently more resistent to the 'parallel structure' hypothesis.
The general framework which the present discussion assumes is that of the DP
hypothesis (Abney 1987; Horrocks & Stavrou 1987; Szabolcsi 1983, 1987). But more
particularly, I shall rely on those analyses which focus on the position and interaction
of adjectives in the DP, taking their position to be indicative of the movement of the
noun (Bernstein 1993; Picallo 1991, Cinque 1993; Giusti 1992, Crisma 1993, 1995;
Ritter 1991, a.o.). This issue will be taken up in section 4. In section 3 the patterns of
adjectival modification of nouns, the adjectival orderings and the way-basically
semantic in nature-in which they have been handled by the relevant literature in terms
of such notions as 'absoluteness', 'predicativeness', 'immanence' etc. will be discussed.
In section 5 certain unsolved problems concerning a particular type of adjectival
ordering will be offered a tentative solution and section 6 will sketch the conclusion
of the whole discussion.
3. THE ORDERING RESTRICTION ON APs

Our data, the discussion of which constitutes the major point of this article, will be
contrasted with the following orderings found in the literature:

Cinque (1993): (for object nominals) cardinal>ordinal>quality>size>shape>colour>nationality.
Sproat & Shih (1987, 1991): size>colour>provenance; size>shape>; quality>shape;
quality colour.
Valois (1990) (for event nominals): group 1 (speaker orientedgroup 2 (subject
orientedgroup 3 (mannerV
Halliday, M.A.K. (1985): numerative>epithet 1 (evaluative/attitudinalepithet 2
(objective property/experiential)>classifier.
APs IN THE DP
203
Pedagogical English Grammars: number>opinion>size>weight>age>shape>colour>participle>originlnationality>material. 4

Factoring out the elements from all the above, we get: numericals (not dealt with
here quality (evaluative/modalssubject oriented (mannersize>shape>colour,
origin or nationality, material (the order between the last three types being largely
interchangeable).
3.1 Patterns of Adjectival Modification
Our examination of the data will make use of the standard distinction between the
scope and the conjoined reading of a series of APs in the DP (cf. Cinque 1993; Zamparelli 1994, a.o.). A conjoined reading ('asyndeton', loose, paratactic or 'asyndetic'
coordination' in Cinque's 1993 terms) is assumed to admit any ordering of the APs
involved. Although conjoinability of APs needs separate study, I would like to point
out that what is called a conjoined reading of a series of APs represents only one out
of four, possibly five, structural patterns of the distribution of A(P)s (in the nominal
projection)5 to be found in the general literature (including traditional grammars of
various languages). The most common patterns encountered are (the patterns are
stated in terms of linear order, no 'AP' symbol is given):
(1) A.
B.
C.
D.
E.
(Art) (A)+(A)+(A)+ ...... (N)

(Art) (A+(A.(A(A ..... (A-N)
(Art) (A and A) (N)
(Art) (A and A) (N)
? (Art) A-A) (N
(A) represents the 'conjoined APs' pattern: each A(P) modifies the noun independently of the modifying function, and presence, of the other(s); usually there is a pause
between the A(P)s (an optionally inserted comma in writing reflecting their status of
A(P)s as separate intonational phrases (Sproat & Shih 1991 :578, and the possibility
of insertion of the conjunction ke ('and'). Following Sproat & Shih, I shall call this
pattern 'parallel modification'. (B) represents the scope or hierarchical reading,
whereby every prior A(P) modifies (has scope over) the following (A(P)+N) combination. I shall call this pattern scopal or hierarchical modification. This is the kind of
modification taken as basic in the recent literature, reflecting the structural notion of
c-command, and it is the pattern with which we shall be concerned here. More formally, we could say of the difference between (A) and (B) that in the latter progressive modification by APs delimits an ever smaller subset, with the leftmost adjective
designating the smallest relevant subset of the set denoted by the noun,6 while in the
former, parallel modification creates a conjunction of entities, in formal terms, an
intersection of the individual denotations of the noun and the adjectives involved. We
might also express the difference between (A) and (B) in terms of how the adjective
assigns a thematic role to the modified noun: in (B), each adjective assigns its thematic role to its sister-a projection of N-, in the case of (A), it does so directly to the
204
MELITA STAVROU
head noun (Sproat & Shih 1991: 567-8).7 Pattern (C) is relatively rare, exemplified by
strings like mia aspri kje mavri blusa ('a white and black blouse'). Here the conjunction ke cannot be omitted (unlike in (A without the outcome being ungrammatical,
probably because it conjoins two adjectival heads and in essence it equals to the
compounding of the stems of these heads (cf. aspromavri 'black&white'). In (0) it
also seems that what is conjoined are adjectival heads rather than full APs. It is typographically represented in the same way as (C), but the intensional entity (exemplified
by the English and Chinese traders (from Ferris 1993: 135), refers to two extensions
of the same noun (ibid.); in other words, what is conjoined in the end is the entity
being modified (there are two groups ofpeople--English traders and Chinese traders-rather than just one with two properties). (E), a much rarer pattern still, is exemplified
by ena vathi kokjino forema ('a deeply red dress'), where the first adjective modifies
the second, rather in the way that an adverb might. I shall not deal with either (C) (0)
or (E) here. (A)-(E), incidentally, do not, necessarily, represent all the possible combinations of A(s) and a N. B
3.1.1
The Conceptual Basis ofthe Restriction
What is of interest at this point is the correlation of the patterns in which As are found
and the order in which they appear crosslinguistically. The correlation is not immediately obvious. The crosslinguistic order in the way it has been standardly, and traditionally, presented seems in the first place to reflect a different descriptive/theoretical
basis from the one assumed in many current discussions of the ordering problem of
the APs, which is UG. This descriptive basis is rather perceptual and semantic and has
been described in the relevant literature in terms of 'inherence', 'immanence',
'permanence', 'absoluteness', 'apparentness' (cf. Quirk et al. 1972; Ferris 1993: 144;
Halliday 1985; Sproat & Shih 1987, 1991), notions which point to the kind of property assigned by an adjective, its semantic content and its processing or computation
with respect to the modified noun (Lamarche 1991).9 We shall tum to the issue of
permanence/immanence in 4.1. Here I would like to discuss briefly the issue of
absoluteness/apparentness and introduce my basic assumption which will be that the
relevant distinction encapsulating all the above notions is sensitive to an opposition
between what I call speaker and object oriented adjectives.
From the perceptual angle then, the ordering restrictions have been largely guided
by considerations which have to do with the degree of absoluteness inherent in the
meaning of an adjective, and the complexity of fixing the type of ascription of a nonabsolute adjective in terms of the number of comparisons required (Sproat & Shih
1987, 1991); in particular, the more absolute the meaning of an adjective is taken to
be, the closer it is expected to be found to the head noun which it modifies. Consider
for example adjectives/adjective phrases denoting a physical property of the object
they modify, such as shape or absolute size (three feet tall) (Sproat & Shih 1987:
467);10 the computation of such adjectives is simple, since it does not entail comparisons which are necessary for computing the meaning of less absolute adjectives, such
as those denoting non-absolute size and, even more so, quality. For this type, one has
to proceed by making comparisons in order to establish the appropriateness of the
APs IN THE DP
205
ascription of such a property to an entity/object (i.e. to fix the meaning of big ant as
being different from that of big elephant).ll
The notion of (degree of) 'absoluteness' has been also dealt with in formal semantics. The corresponding logical notion is that of 'vagueness' (Hoepelman 1983:
194); the less absolute adjectives are, the vaguer they are in semantic terms, as they
force us "to reset our standards again and again according to the sets we are dealing
with" (ibid.: 195-6).12
My assumption is that the distinction which is syntactically most relevant and
which in fact reflects a more primitive contrast is the distinction between speaker- and
object-oriented adjectives. The former subsume the class of 'less absolute' adjectives,
which appear further away from the modified noun, the latter subsume the 'absolute'
ones in that they denote an a measurable/physical property of the referent of the noun.
For instance, in eftelis politiki satira 'mean political satire' while the first adjective is
speaker oriented in that it reflects the speaker's opinion/evaluation of the referent
denoted by the nominal head, the second adjective denotes an inherent part/property
of this referent. Similarly in endiposiako kristalino potiri 'impressive crystal glass'.
Put in other words, while the object oriented adjectives are extensionally the speaker
oriented ones are intensionally defmed. With these remarks in mind, there is no need
for further discriminations in either adjectival class, as the speaker oriented adjectives
encompass older labels such as evaluative, descriptive, qualitative, etc., and within the
object-oriented class adjectives appear basically interchangeable, as we shall see in
examples (4e-g), a fact which will be offered an explanation in section 4.3.
The speaker/object oriented distinction further reflects the possibility for an
adjective being modified by degree words/adverbs (like much, relatively, etc.) and
forming a comparative and a superlative; object oriented ('classifier-like' adjectives in
Bosque & Picallo's 1994 terms) do not normally seem to allow for any kind ofmodification.
My primary concern in what follows will be the mapping of the afore mentioned
distinction on syntax.
4. TOWARDS AN ACCOUNT OF THE ORDERING ISSUE

In the first part of this subsection the distinction between speaker/object oriented
adjectives will be illustrated by means of examples, and my basic assumption about
the formation of a compound out of the adjective found immediately in front of the
noun and the noun will be spelled out. In the third part the proposed DP structure will
be shown to accommodate the facts discussed so far.
4.1 Permanence and the 'Speaker/Object Oriented' Adjectives Distinction
The potential of an adjective to identify the referent of the modified noun is commonly held to be "one of the main purposes of the use of an adjective in attributive
position" (Kamp 1975:153). In other words, such adjectives "contribute to the delineation of the class of objects that the complex noun-phrase of which it is part is
206
MELITA STAVROU
designed to pick out-or, alternatively, help determine the particular individual which
is the intended referent of the description in which the adjective occurs" (ibid.).
Interestingly, the order of the adjectives, which I take to belong to the speaker/object
distinction appears to be the reverse of the identifying potential that each adjective
class has; i.e., the less object-like (starting from left to right) an adjective is the more
efficiently it appears to identify the noun; conversely, the more object-like it is, the
less identifying it is. This correlation is expressed in the traditional 'morelless permanent' distinction as follows: "By and large, the more permanent the attribute of a
thing, the less likely it is to identify it in a particular context. So we proceed with the
very impermanent, quantitative characterization (... ); through various qualitative
features such as new in new ball; and end up with the most permanent, the assignment to a class (emphasis mine), e.g. tennis ball.13 Within the qualitative characteristics, if more than one is specified there is again a tendency to move from the less
permanent to the more permanent...." (Halliday 1985: 166). The picture that emerges
then is that as far as the primary function of the adjectives, i.e. the delimitation of the
reference of the modified noun, is concerned, speaker oriented ones are further away
from the noun they modify, whereas object oriented adjectives, since they characterize nouns in a more stable fashion, are found closer to the head noun. In the light of
these remarks then, it seems that the whole issue of the restrictions on the hierarchical
ordering of adjectives amounts to head proximity (rather than simple linear ordering).
The emphasized sequence in the above citation brings us to the second function of
prenominal adjectives which is the classification of the modified noun. Apparently,
object oriented adjectives, those adjacent to the modified noun, end up in classifying
it. In the light of this observation, it is interesting to note that the last of a series of
adjectives with the most 'intimate' relation to the modified noun is called a 'classifier'
by Halliday14 and a 'denominal' adjective by Quirk et al. Obviously, this is the
adjective that may (although not necessarily) modify the noun in the more 'objective'
way, so that it effectively subcategorizes it;15 e.g. a eksotiko puli ('exotic bird')
denotes a proper subset of the set of birds-a particular kind of bird, or, else, a natural
class. But this is much less true of evaluative/speaker oriented adjectives: a dhinamiko
('active')/oreo ('nice') pedhi ('kid') does not constitute a natural class of children.
Likewise, mia thrakjiotildforesja ('a Thracian costume') refers to a particular kind of
folk-costume, but mia eksohiforesja ('a wonderful costume') clearly does not.
The classificatory function of attributive adjectives is also noticed by Sadler &
Arnold (1994), cf.: " ..prenominal adjectives are better when they provide a 'natural'
classification in combination with the noun" (: 193). This point is also briefly made
by Sproat & Shih (1987: 469-70) too,16 who consider adjectives of provenance in
particular to be typical servers of a taxonomy (ibid.). They notice that since the
ordering of such adjectives cannot be derived from the comparison scale their classificatory role has to be considered in addition. Interestingly, Zamparelli (1993; although
in Zamparelli 1994 he takes a different stand) views the adjectives that immediately
precede the noun as appositives, in a way reminiscent of Givan's (1993:268) view
that, since such adjectives subcategorize the noun, they form almost a unitary concept
with it, almost a new lexical item, and so do not restrict its denotation (see note 7).
APs INTHEDP
207
The classificatory role of certain adjectives aslo constitutes a major point of Bosque &
Picallo's (1994) analysis of the order of adjectives.
But here, on the basis of this classificatory/subcategorizing role of object oriented
adjectives, I shall make a further assumption, namely that such an adjective constitutes in effect a compound with the noun immediately following it, i.e. together they
form a unitary concept. For instance, in the aforementioned examples, the combinations thrakjiotiki foresia and eksotiko puli each constitutes a compound both semantically (: a single conceptual entity) and syntactically (each combination counts as a N-( see also Ralli&Stavrou to appear for a detailed morphological account of this phenomenon). As a further illustration of this consider (2):
(2) a.
b.
i theoritiki
sinhroni
ghlosoloji
the theoretical
contemporary linguists
i sinhroni
theoritiki ghlosoloji
the contemporary theoretical linguists
In (a) contemporary linguists forms a larger set (created out of the sets of linguists and
contemporary things), ultimately the single entity 'contemporary linguists'. This
denotatum is restricted by the adjective theoretical. In (b) the relevant concept is that
of 'theoretical linguists' and it is this denotatum which is further restricted by the
adjective contemporary. The choice of the relative linear position of the adjectives lies
with the speaker and the way she wishes to classify the world so as to indicate the
contextually relevant (sub)classes. E.g. she may want to express the meaning 'modem
linguists' (i.e. a broader set) and then make a contrast between theoretical and, say,
applied modem linguists (i.e. restricting the set), or that she takes the central notion to
be that of theoretical linguists (larger set), around which she fixes the contrast contemporary vs. earlier theoretical linguists (a smaller set).
The question that arises at this point is how these two basic functions of the
adjectives (the restrictive modifier and the classificatory one) relate to the distinction
between pattern (A) and pattern (B). The classificatory one must be independent of
the kind of reading a series of adjectives gets before an adjective plus noun compound, but, crucially, if such a compounding has taken place and there is only one
adjective preceding this compound (as ih (2a-b, then this adjective will, naturally
modify the whole combination (: the compound), as is the case with (2). This is
illustrated in Diagram (1):
208
MELITASTAVROU
Diagram (1)
In (l), the scope reading gives a subset of the set denoted by the noun and the adjective immediately preceding it under the basic assumption that these two form a syntactic compound.
If there are more than one adjectives these yield, as said, either a paralleVconjoined or
a scopal reading over the noun that is modified, independently of whether this noun
has formed a compound with the adjective immediately preceding it or not. Diagram (2) illustrates the conjoined reading of adjectives when there is no such compounding.
Diagram (2)
In (2), the conjoined reading yields an intersection of the sets {contemporary things},
{linguists}, {theoretical things}.
Notice crucially that under (1) there are no contemporary linguists that are not
also theoretical linguists, but under (2) there are contemporary things/entities that are
not theoretical things/entities-or, in fact, even linguists.
To summarize: there are two ways in which multiple adjectival modification is
accomplished; the one is the so-called parallel or non-hierarchical pattern (A) under
(1), the other is the hierarchical or scope pattern (B) under (1). In the case of pattern
(A) there is considerable freedom as to the relative order between the modifying APs
(see 4e-g) below). In the case of (B), speaker oriented adjectives precede object
oriented ones. Moreover, discourse-pragmatic considerations very often interfere, in
that the speaker may wish to 're-group' the adjective-noun combinations in such a
way as to make the last adjective of the series the least identifying one, i.e. to form a
sub-category of the denotation of the noun. 17 Thus I take the pair ofpbrases like those
APsINTHEDP
209
in (2a-b) to represent equally possible instantiations of the scopal pattern, but with a
clear difference in meaning between them, as expected.
As a fmal note, the existence of a scope relation is further confmned by the fact
that very similar,18 and apparently synonymous, APs cannot constitute a series determined by relative scope; cf. (3):
(3) ??i kali pisti hristi
polites
the good faithful virtuous citizens
''The good, faithful (and) virtuous citizens'
Such a sequence, on the other hand, is expected to constitute an acceptable string
under an asyndeton/coordination reading (i kali, pisti, hristi polites.), as the gloss
indicates, forced by the similarity Isynonymy of the adjectives.
4.2 The Data
In order to address the issue of the structure of a DP with several APs we must fITst
establish the evidence provided by the data. In (4) examples containing two or more
prenominal adjectives are presented. In (4a-d) the scopal reading is exemplified, in
(4e-g) the parallel or conjoined one:
(4) a.
b.
c.
d.
e.
f.
g.
profilaghmeni kali thesi

protected
good position
mya iposinidhiti ghlikja via
a
subconscious sweet violence
ta nea pragbmatika tekmiria
the new real
proofs
mia melancholiki fthinoporini imera
a
sad
ununnal
day
to panemorfo mikro, skalisto, dhriino, mavro(,) japoneziko trapezi
the very-nice small curved oak
black Japanese table
Ito mikro, skalisto, mavro, dhriino, panemorfo japoneziko trapezi
the small curved black oak
very-nice Japanese
table
tria japonezika, mikra, omorfa, ksilina trapezja
three japanese
small nice
wooden tables
Itria omorfa, mikra, ksilina japonezika trapezja
three nice small wooden Japanese tables
Itria mikra, omorfa, japonezika, ksilina trapezja
three small nice
Japanese
wooden tables
i ksanthja,
psili, simbathitiki jineka tu
the fair-haired tall nice
wife- his
Ii simbathitiki, psili, ksanthja jineka tu
the nice
tall fair-haired wife- his
I i psili, ksanthja, simbathitiki jineka tu
the tall fair-haired nice
wife-his
MELITA STAVROU
210
Some general comments are required here concerning the order of APs exemplified in
(4). Though examples can easily be multiplied, even this limited list is sufficient to
show that the unmarked order of APs denoting (a) the speaker's judgement and (b) a
physicaVobject-oriented property appears to conform to the orderings given in section
3, when the adjectives in question are interpreted under the scope or hierarchical
reading; but when they are read as conjoined, their relative order appears almost
entirely free (4e-g). Greek, then, displays both AP patternings, and, consequently, one
expects to fmd the concomitant differences inherent to these patterns. Thus, in the
case of parallel or flat adjectival modification, where each adjective modifies the noun
independently of the others, as illustrated by Sproat & Shih (1987: 478; 1991: 568),
the ordering restrictions are relaxed and the number of APs is open-ended, as expected, if what is termed 'parallel' modification is equivalent to coordination. On the
other hand, when the hierarchical pattern is involved, the number of modifying APs is
much smaller, as the relevant examples show; taking the data in (4a-d) at face-value,
there are no more than two, and the ordering restrictions are more severe. This limitation on the number of adjectives involved in hierarchical modification is also pointed
out by Sproat & Shih for Mandarin, and we shall see below that it fits well with the
limited number of functional projections independently argued for the Gr DP.
The next question to be addressed concerns the issue of the structural account of
pattern (B). The discussion will be crucially based on the evidence provided by Nmovement in indefmite DPs.
4.3 The Syntactic Account

As has already been said, pattern (B), unlike pattern (A), conforms to the structural
relation of c-command; it is therefore expected that in (A), a change in the order of
the coordinated adjectives will not affect the c-command relation, while in (B) it will,
leading to the change of meaning we observed above. Sproat & Shih claim that in the
case of coordinated adjectives (our pattern (A, each adjective foms part of an independent, intonationally separate, intermediate phrase. For the pupose of the present
paper it is sufficient simply to adopt this view and to proceed to the account of pattern
(B).
4.3.1
The Basic Assumptions about the Status a/the Adjectives in the DP
Following the develpoment of'DP theory', in the relevant literature one can fmd three
major hypotheses about the status and position of the A(P) in the DP; the one is the
assumption put forth by Abney (1987) himself taking adjectives to be heads which
select a NP, or another DP, as their complement, being themselves selected by the
(highest) Determiner. Though this assumption has since been adopted for a number of
languages (also Radford 1993; Kester 1993; Androutsopoulou 1994, 1995 for Gr.
Delsing 1993 takes the adjective to be a head, but the NP instead of being its complement is taken to be its right hand specifier), it does present certain difficulties,
APsINTHEDP
211
especially as concerns Gr, where prenominal adjectives can routinely be specified/complemented. Furthermore, (full) agreement between adjectives and nouns
holds equally for both pre- and post-head adjectives (the latter occurring only in
indefinite DPs), so the essence of Radford's argument that agreement is obligatory
only for prenominal (heads) adjectives is called into question by the Greek evidence. 19 Moreover, from a semantic point of view, it is natural to consider the combination of an adjective and a noun as a nominal rather than adjectival phrase, as the
outcome of the combination denotes an object, not a property (Lamarche 1991),
contrary to the view of those favouring the adjective-as-head analysis. There is,
however, another sense of the term head in this context, according to which the
prenominal adjective is still a head but only in terms of its phrasal status, i.e. it is a
bare, zero -level category which does not branch. Under this view, the adjectival head
does not select another category but it is adjoined to the left of the nominal head. This
is the view underlying Lamarche's (1991) and Sadler & Arnold's (1994) analyses. 2o It
is also the explicit assumption of Higginbotham's (1985: 562-565, see espec. p. 563)
claims about the syntax and semantics of the relation between the adjective and the
noun called theta-identification. It is this sense of head that underlies the way in which
compounding of A plus N is assumed to take place in this paper and which forms the
basis ofRalli & Stavrou's (to appear) analysis.
The other basic assumption is the one according to which APs are specifiers
adjoined to the functional categories which intervene between the selecting DP and
the selected NP (Giusti 1992; Cinque 1993; Crisma 1993; Zamparelli 1994, among
others; Kayne's view is taken up in detail by Bosque & Picallo 1994). In some
versions of this view adjectives are specifiers of the NP as well (Valois 1990; Stavrou
1996; Alexiadou & Stavrou 1996a, b, next to others); this will be the version adopted
here. The obvious advantage of the adjective-as-specifier view is that the
morphosyntactic agreement which typically occurs between the adjective and the
noun is accounted for straightforwardly in terms of Spec-Head agreement (for a
detailed analysis of such an agreement see Zamparelli 1994 and Bosque & Picallo
1994).
According to the strictest version of the adjective-as-spec hypothesis, the specifier
of NP is not a suitable position for APs, mainly because the head of this category,
being lexical, does not participate in the spec-head relationship (Zamparelli 1994: 28).
Nevertheless, there is a generally held view that the spec,NP position is reserved for
thematic adjectives, i.e. for those adjectives that absorb a thematic role from the noun
(Radford 1993; Bosque & Picallo 1994; Valois 1990; but see Alexiadou & Stavrou
for discussion).
Finally, the adjective-as-specifier hypothesis further accounts for the restricted
number and (unmarked) ordering of (a series of) adjectives (Cinque 1993:5-6).
Intermediate assumptions are also encountered, as for example, the one according
to which only certain adjective-types are heads (selectors), while others are specifiers.
This is Bernstein's (1992, 1993) view, who maintains that so-called intensional adjectives (of the mere, alleged ...type) are heads that select a NumP Phrase2 1, while other
adjectives are in the specifier position of NumP (and Word Marker P). Radford's
(1993) independently developed analysis is in the same spirit, as he takes only the-
212
MELITA STAVROU
matic (,referential' in the sense of Giorgi & Longobardi 1991) adjectives to be internal
to the NP, while all the attributive/modificatory ones are treated as the "heads of their
containing phrases, whereas post-nominal Adjectives would be adjuncts to the nominals they follow" (Radford 1993: 85).
Finally, accounts less elaborate with respect to the status of adjectives simply take
them to be adjuncts of maximal projections (including the NP) (Ouhalla 1991;
Karanassios 1992; also Lamarche 1991). The drawbacks, both empirical and conceptual (see Cinque 1993:5 for a fuller discussion) of these accounts are first that they
cannot handle agreement between the adjective and the noun in a straightforward
way, or, to the extent that adjunctions are generally free, explain the ordering restrictions on a series of APs; secondly, multiple adjunction, which would be necessary to
account for multiple instantiations of APs, is not only theoretically avoidable but
widely regarded as something to be avoided (Kayne 1994). Cinque (1993) also draws
attention to the fact that the adjunction hypothesis does not naturally account for the
left position (relative to the noun) in which the adjectival phrase is found, at least in
languages with prenominal adjectives, whereas under the Spec hypothesis this fact is
automatic. 22
4.3.2
N-movement as Evidence for the Position(s) ofAdjectives
As said above, the view which will be adopted and further explored here (cf. also
Stavrou 1996) is the one according to which adjectives are specifiers. This hypothesis
seems to me to be more challenging, in that, apart from its inherent merits with
respect to spec-head agreement and ordering, it also involves consideration of a
process independently attested in a number of languages, namely N-movement. Most
recent discussions of the issue of the relative order of APs are crucially based on the
evidence provided by the position that adjectives are found after N has moved from its
basic position in NP (Cinque 1993; Crisma 1993; Zamparelli 1994; Ritter 1991;
Bernstein 1992, 1993; Giusti 1992; Bosque & PicaBo 1994; Stavrou 1996).23 The
crucial question of how far N goes is addressed in the following subsection.
In Gr, the noun, despite the apparently 'strong' morphological features (for gender, case, number) it seems to share with Romance languages, does not move (in
contrast with the verb) in DPs; at least it does not do so in overt syntax. 24 As we shall
see immediately below, there is evidence about the noun moving past some AP in an
indefmite DP. In fact, this 'asymmetry' is a familiar problem, which amounts to the
fact that, as traditionally described, in a defmite NP the AP is not allowed to follow
the noun; cf: *0 dhemonas Jterotos (the demon winged), *i thesi kali prophilafgmeni
('the place good protected'), *ta 10 meghalajeghonota ekdhotika ('the ten big events
editorial').25 This means that, considering the (relative) order of a series of APs, the
basic order is established, as was done above, without evidence from N-movement
(cf. Cinque 1993 and Bosque & Picallo 1994 for English). However, in indefmite DPs
the noun can be followed by an AP, as shown in (5):
APs IN THE DP
213
(5) a.
enas dhemonas fterotos

a
demon
winged
'a winged demon'
b. ena trapezaki tetraghono
a small-table rectangular
'a small rectangular table'
c. pola ruha
ftina
many clothes cheap
'many cheap clothes'
d. dhimarhos entimos
mayor
honest
'an honest mayor'
e. sinohi
kjinoniki
coherence social
'social coherence'
f. pedhi kjefato
child cheerful
'a cheerful child'
g. epanastasis italikjes
revolutions italian (from Mouma 1993)
'Italian revolutions'26
It is clear from the above data that the (im)possibility of noun movement is determined by the (in)defmitemess of the whole DP. This 'asymmetry' was ftrst addressed
by Horrocks & Stavrou (1986); a satisfactory solution was put forth in Karanassios
(1992) and this has been further expanded in Stavrou (1996). According to this
analysis, there is a functional projection (FP) between DP and NP, the head of which
may contain the defmite article as a gender and defmiteness marker (for the whole
DP). When empty, however, as in indefmite DPs, this position is available as a landing site for a moved noun. The auxiliary assumption made by Karanassios (1992) and
Stavrou (1996) is that, while indefmite determiners are generated in D when this is [~
de], the defmite article, as the head of the intermediate FP, is subsequently moved to
D to check its defmiteness feature. In other words, when D is marked [+de], the FP
head is instantiated by the defmite article, but when D is [-de] , the head of FP is
empty and N moves to if7, giving the order (indef. det)-N-AP. After N has moved out
ofNP, a predicative-like indexing between N (the head of the chain) and AP accounts
for the 'predicative' (small clause) reading that we get in (indef. art)-N-AP sequences,
as well as other differences in interpretation between this sequence and the
corresponding one involving a prenominal AP (Stavrou 1996).
Taking the standard semantics for adjectives in their attributive role as functions
from common nouns denotations to common nouns, i.e. as forming noun meanings
from noun meanings28 , I will follow the more traditional syntactic account which, in
my view, best reflects the semantics of adjectival modiftcation, i.e. the one according
to which there is at least one AP position which is part of the lexical NP (Jackendoff
1977; Leffel & Bouchard (eds.) 1991; Sproat & Shih 1987, 1991; Valois 1990;
MELITA STAVROU
214
Karanassios 1992; Stavrou 1996). I also assume that APs are specifiers adjoined to the
relevant categories and as such are licensed by their heads, which by the LCA (Kayne
1994) they obligatorily precede.
The above assumptions are illustrated in the tree-structure in (6):
(6)
DP
D
[ def]
FP
/"'-....
AP
FP
~
NP
def. art
AP
NP
4.3.2.1 How far does N go?

As noted above (4.3.2), the examination of N-movement has been taken to be an
important means of shedding light on the question of the position(s) of adjectives that
(appear to) follow the noun in Romance languages. The effect of such a movement is
seen in the following examples:
(7) a.
b.
la stupida aggressione brutale/italiana all'Albania (from Cinque 1993:3)

the stupid attack
brutaVof-ltaly to-the Albania
la loro aggressione all'Albania
brutale (from Cinque 1993:2)
the their attack
to-the Albania brutal
It is clear from these that the noun moves only a short distance upwards so as to leave
behind thematic adjectives and, according to Cinque, also subject oriented ones. 29 But
the exact nature of the projection(s) to which N moves is quite unclear and has given
rise to dispute (cf. Cinque 1993:5). Cinque and Zamparelli, for example, do not name
the projections that intervene between the determiner and the noun, Valois (1990)
labels two such projections, as Case and Number Phrase, Picallo (1991) and Bosque
& Picallo (1994) label three projections, as Gender, Number and Kase, while
Bernstein (1992, 1993) uses NumP and Word Marker Phrase as intermediate projections. Finally Ritter (1991) establishes Number Phrase as an intermediate phrase in
Hebrew.
At this point I would like to consider the evidence provided for the position(s) of
APs by N-movement in indefmite DPs in Gr.
Movement of N to an intermediate functional head in indefmite DPs is extremely
common and straightorward when it leaves behind one AP only (see again the data in
APsINTHEDP
215
(5). The data in (8) suggests that it is difficult for the noun to leave behind two and
moreAPs:
(8) a.
b.
??simvan ekpliktiko mireo

incident
unexpected fatal
*ghlipta politima mikinaika elefantosteina3o
sculptures precious Mycenaean ivory
(contrast with: politima mikinaika ghlipta eleJantosteina)
Thus, the most common situation is for the N to be preceded by one AP and be followed by another:
(9)
a.
dhemoniko teras
mithiko
demonic monster mythological
b. mia kali fili
adhiakriti
a good friend indiscreet
c. nees times almires
new prices high
d. arhaiko aghalmatio asvestolithiko
archaic small-statue limestone
e. ena panemorfo
forema letko
a very-beautiful dress white
It is fairly clear that there must be two structural positions for APs, one lower and one
higher, as reflected correctly in (6), so that when the noun moves to an intermediate
position between D and NP it can leave behind the lower one. Thus, N-raising in
indefmite DPs operates in a way similar to that proposed for Romance languages.
Even more interestingly, the fact that there is usually just one AP left behind appears
to parallel the facts described by Cinque (1993) for Italian, where the noun skips one
AP position,3! as it moves upwards.
The AP that appears on the left of the moved noun must be located in a projection
either higher than the one reached or, in fact, within the one reached. Given the analysis of N-movement outlined above (see also Stavrou 1996), the AP preceding the
moved noun must be taken to be the one in the spec of FP, to whose head N has
moved.32 In other words, there may always be an AP never crossed by N.
In line with the 'AP ordering' hypothesis outlined above, we should be able to
predict that the 'higher' AP, which is in the spec ofFP, will be of the speaker-oriented
type, while the adjective left behind is of the object-oriented type; this is suggested by
the data in (10):
(10)
a.
b.
ena oreo forema metaksoto

a
nice dress silk
mja kompsi tsanda italiki
a elegant bag Italian
MELITA STAVROU
216
So far then we have seen that the independently motivated structure in (6),
(Karanassios 1992), appears to account for the data concerning not only the position
of an adjective relative to a defmite/indefmite noun, but also for the order and number
of adjectives (if more than one). In particular, the intermediate FP projection in the
Greek OP, in combination with the spec-N position, is sufficient to account for the
adjectives that are .ordered under a scope reading. What is even more interesting
perhaps, is that the claim made by Sproat & Shih about the number of adjectives
under a scope reading being limited to two (in Mandarin) not only fmds an exact
parallel in Gr, but also follows directly from the preceding discussion. Of course, one
should not forget that any number of adjectives can precede (or follow, under the Nraising analysis) the noun if they are interpreted as coordinated.
4.3.2.2 Classifier-like Adjectives and N-movement

In this subsection I would like to consider counterevidence to the above conclusion
and extend my analysis to capture first data like (1Ia), involving what are called
classificatory adjectives, the number of which can in principle be infmite, and, second, data like (lIb) in which the adjectives appear in the reverse order vis-it-vis that
exemplified by (lOa-b) above:
(11)
a.
b.
ena endhjaferon amerikaniko metapolemiko politiko ergho

post-war
political fIlm
an interesting american
ena metaksoto forema oreo
a
silk
dress nice
Given the 'deviance' of such adjectives from the speaker>object-oriented order, as

well as in terms of their role (further specified below) I shall work from the minimal
structure in (6) and adopt Bosque & Picallo's (1994) analysis for classifier-type APs
{like amerikaniko, politiko, metapolemiko in (lta, according to which NP is taken as
a Larsonian layered projection (cf. Larson 1988). This provides for us as many layers
as we need to accommodate classifier-like adjectives; by 'classifier-like adjectives' I
mean (see also section 4.1 above) those adjectives about which Bosque & Picallo
point out "the semantic links that we can establish between C-adjectives and Nouns
can easily become a taxonomic classification of relations between entities expressed
with th-role labels" (1994: 18,9).33 Following the preceding comments on the
perceptually based ordering of APs, I consider these classifier-like adjectives as
largely coinciding with what I called object-like adjectives (e.g. those denoting origin,
material, shape, colour) and I further take the relative order (scope) of the various
Specs of the multi-layered NP to be pragmatically conditioned according to what was
said in sections 3 and 4.1: i.e. the adjective found closer to the head noun is the one
which forms a contextually relevant taxonomy/class, and more importantly, the
largest class, ultimately a compound. 34 I shall rely on such a compounding process
between the object-oriented adjective adjacent to the noun and the noun in order to
account for (11 b), i.e. the fact that the object-oriented adjective plus noun have moved
as a unit leaving behind a non-object-oriented one. We shall come to this again in
APs IN THE DP
217
section 5. Coming back to (11a), the adjectives further to the left are the ones that
successively render this class smaller (consider eamples 2 again), i.e speaker-oriented
ones. Concerning the layered NP-shell, it can have as many empty 'light' heads, as we
need specifier positions to place classificatory APs into; the lowest NP is the one
which is lexically realized, and this moves to the head F of FP in (6). This revised
analysis of the lowest part (NP) of the DP given in (6), illustrated in (12) below, has
the advantage of avoiding proliferation of FPs with unknown content and of
overcoming the problem of multiple adjunctions (to NP), in a way that accords with
the general theoretical framework (Kayne 1994).35
(12)
NP
~
AP
NP
[e)
NP
~
AP
NP
I
N
In this structure, crucially, it is thematic adjectives that are taken care of, much as
argumental DPs are treated by Larson in the layered VP (see Bosque and Picallo for a
detailed account of thematic APs in this kind of structure in Spanish). I shall not
discuss thematic adjectives here (see Alexiadou & Stavrou 1996a&b for an account of
so-called thematic adjectives), but I will tentatively assume that, contra Bosque &
Picallo, these are hosted by the lowest Spec, just above the lexical N, as the neutral
order in sequences like takti/d proedhriki episkepsi ('usual presidential visit'), suggests, with the classifier-like adjective preceding the thematic one.
Given the evidence exemplified by (8), however, in combination with the evidence of (11a) and the prediction implied by the structure in (12), that N can move to
F leaving behind more than one adjective (: the classificatory ones, since there can be
more than one of these), we must fmd a solution to the contradiction created by our
successive assumptions. In other words, we have created room to accommodate more
than two adjectives in front of the noun under the scope reading (as required by (12,
but we still need to explain why the restriction exemplified by (8) holds true for the
majority of these cases, while strings like (13a-b) below are also quite common.
(13)
a.
b.
fita
dhikotilidhona apetala
plants dicotyledonous without petals
sedonja polihroma
mona
sheets multicoloured single
In the light of sequences like (13a-b), involving two stranded adjectives and still being
grammatical, we can still make sense of our structures in (6) and (12). In (13a-b) the
noun has skipped two adjectives placed within the NP shell-notice that both skipped
MELITA STAVROU
218
adjectives in (13) are classifying ones. So the conclusion at this point would be that
the noun, when found under F, is preceded by a qualitative adjective (or descriptive or
speaker-oriented one) and followed by object-oriented ones. This conclusion is in line
with the data under (5), (8), (9), (lOa-b) and (13). However, it is not in line with the
fact that (13a-b) displays the maximum number of classifying adjectives that can be
skipped by the noun, neither with (lIb), in which the classifying adjective precedes
and a speaker-oriented one follows the noun. In the following section I shall propose a
tentative solution for this ordering problem.
5. REMAINING PROBLEMS
The analysis presented in the preceding sections further entails that the moved noun
can be followed by many (in principle infmite in number) classifying adjectives.
Interestingly though, consideration of the relevant (if scanty) data reveals that in
practice there are no sequences involving more than two classifier-like, or in general,
object-oriented adjectives following a noun. Thus (13a-b) seems to represent the set
of data displaying the maximum number of 'postnominal' (classificatory) adjectives,
on a par with the evidence adduced by (8). My tentative explanation for the order of
adjectives in (13a-b) relies on my account of the order 'class. adj-noun-descr. adj' in
(lIb). Thus, the fact that in (lIb) the whole of classif.adj.-noun is followed by a
descriptive adjective suggests that this complex forms a unit which is further modified
by another, speaker-oriented, adjective. In the spirit of the proposals made by Ralli &
Stavrou (to appear), I assume that the classifying adjective which immediately precedes the head noun incorporates with it so as to form a construct (ibid.).36 It is
important to recall the basic assumption that the adjective which is adjacent to the
noun is taken by general consensus to form a compound with the noun. It is this
construct that moves upwards to the head F, leaving behind another adjective. To
return now to (13), I assume that if incorporation of the adjective to the noun has not
taken place before the noun raises, it can happen after raising, yielding the order
displayed by (13a-b). This implies that the order of the relevant entities before Nraising and incorporation is as in (14a)-(14b) below:
(14)
a.
b.
polihroma
mona sedonja
multi-coloured single sheets
dhikotilidhona apetala
fita
dikotylidonous without-petals plants
where the noun first moves and then incorporation takes place between it and the
adjective immediately follows. If incorporation takes place before the noun moves,
the resulting sequences is as in (15):
(15)
a.
mona sedonja polihroma

single sheets multi-coloured
APs INTHEDP
b.
219
apetala
fita dhikotilidhona
without-petals plants dikotylidonous (cf. ( 11 b)
Of course, the lack of a mirror effect of N-raising, which is expected under a Nmovement analysis (see also Lamarche 1991; Cinque 1993), changes the initial scope
relations, but in either case, i.e. whether incorporation takes place before movement or
after, the adjective incorporated is the one adjacent to the noun.
We can summarize the proposal as follows: in the case of classificatory adjectives
in the spec positions of the NP-shell, the one adjacent to the noun can incorporate
either before N-raising, (lIb), after which the adjective+noun moves as a unit to F, or
after (as in 13a-b), in which case the adjective incorporates to the right of the noun. I
do not, however, have an answer for why even under the incorporation/compounding
analysis there cannot leave more than two classifying adjectives behind the noun,
neither can I give a more precise answer to the question of when incorporation applies.
But there is still a problem with examples like (13b). Ideally, the adjective that
follows the (raised) noun should always be of the object-oriented type. Unfortunately,
this is not so in every case; very often, ordinary speaker-oriented adjectives are
crossed by the noun, despite the assumption that they are in the spec of FP-i.e. in a
higher position than classifying ones. Consider (5d,f) above and (16) below:
(16)
ena vivlio endhiaferon

a book interesting
This is a problem that has to be faced also in the case of Italian (cf. un libra interessante) and Spanish (for the latter Bosque & Picallo (1994: 45) observe that qualitative
adjectives may precede or follow the moved noun). It could be suggested that these
adjectives (of the nice, interesting, etc. type) are generated in the specifier of two
different projections, one higher and one lower 7 , or even that N moves further than F
in (6), but then again we would run the risk of inventing projections that do not have
any other purpose or content but to 'give shelter' to adjectives found in unexpected
positions. Such projections, however, are not available under the present analysis
(under (6) & (12)), so I do not as yet have an answer to this problem.
6. CONCLUSION
We have seen that although the universally stated strict order of APs cannot be disputed, it is very often obscured either by the conjoined reading of a series of APs,
which frees it (a fact that should be recognized and given formal status in the account
of AP ordering restrictions), or even by the scope reading, which often suggests that a
pragmatically based re-ordering is carried out by the speaker to exploit the classificatory role of certain object-oriented adjectives. The nominal structure given in (6) and
(12) is an attempt to capture the restrictions on AP placement. Thus, the higher FP,
independently argued for in the relevant literature, hosts APs that are most remote
from the noun (NP), namely speaker-oriented ones, whereas APs of the more abso-
220
MELITA STAVROU
lute--oblect-oriented--type are placed in (the specifiers of) the NP-shell. The placement of classifying APs in the specs of a multi-layered NP seems to account satisfactorily for the semantic classification they impose on the noun. There is no more
specific order assumed for the APs in the NP-shell, as their relative scope seems to be
regulated by pragmatic criteria. Thus, the AP adjacent to the head is the one that is
intended to 'classify' the noun, while every preceding AP has scope over the following adjective+(adjective+) noun. Generally, the higher an AP appears, the less objectlike the property it denotes. Evidence coming from N-raising in indefmite DPs was
invoked in order to delimit further the ordering restrictions on APs. It was shown that
N really moves only a short distance, specifically to the head of an FP selected by D
and lying immediately above NP. The adjective that is never skipped by the noun
must be located in a high position; I take this to be the position adjoined to the FP, to
the head of which the noun moves.
There is no evidence for the existence of further 'higher' positions which can
accommodate prenominal adjectives, as there are rarely more than two such adjectives
under a scope reading (although there may be an infmite number of APs under a
conjoined reading). Apparent counterevidence, corroborated by the facts concerning
N-raising can be explained on the basis of syntactic compounding, or incorporation,
according to which an adjective can form a 'phrasal compound' (i.e. a zero level
category) with the adjacent noun, which raises as a single unit past the lower AP;
interestingly, an adjective that typically forms such a compound has a classifying
function (Ralli & Stavrou (to appear.
The general conclusion drawn from the whole discussion concerning Gr agrees
with Cinque's (1993, 1995) and Zamparelli's (1994) conclusion concerning Italian,
viz. the evidence discussed here confirms the "existence of a clear limit on the number
of non--coordinated attributive APs within DP" (Cinque 1993:5), which is due to the
"limited number of functional projections independently available between D and NP"
(ibid.); we differ however as regards the number of these projections, therefore in the
number of non-coordinated adjectives, amounting to six or seven in Italian (ibid.) but
to no more than three (including the one forming a compound with the noun) in Gr.
There are, however, a number of residual questions that remain to be explored,
such as how the conjoined reading can best be structurally represented, and, more
importantly perhaps, what is the exact content of the FP projection taken to be headed
by the defmite article in defmite DPs.
NOTES
I would like to thank the following people for their helpful comments and insightful suggestions on
earlier drafts of this paper and during long discussions they had with me concerning various points of the
ideas developed here: Dhimitra Theophanopoulou-Kontou, Artemis Alexiadou, Angeliki Ralli, Geoffrey
Horrocks, Guglielmo Cinque, Anna Cardinaletti, Giuliana Giusti, Nuria Marti, Anna Roussou and two
anonymous reviewers. Special thanks go to Anastasia Giannakidou for patiently helping me out with the
semantics part of the presentation. Of course, the usual disclaimers apply.
2
For example, Ritter (1988,1991) argues for a Number Phrase above NP in Hebrew and Karanassios
(1992) maintains that there is a Definite Projection selected by D and hosting the def. article in Gr, in a
APsINTHEDP
221
way that parallels Ritter's NumP. Bernstein (1992, 1993) establishes a Word Marker Phrase and Picallo
(1991) a Gender Phrase below NumP.
According to the aforementioned analyses promoting the specifier analysis on adjective placement,
the constraints on the ordering of APs follow from the constraints on the order of the relevant FPs.
4
For a fuller list of such orderings ee also Sproat & Shih (1987: 1991) and references cited there.
See Sadler & Arnold (1994) as well as Ferris (1993) for more details in this connection.
Cf. also Bosque & Picallo (1994: 24): "the adjective adjacent to the head denotes the larger class,
the folIowing one a subclass" (in the case of Spanish, where the adjectives in question appear
postnominally). In the case of the English big brown dogs and its Or counterpart meghali kafe skili, the
outer adjective denotes the subset 'big' of the subset 'brown dogs'. In fact, I shall claim below that under
the scope reading the noun and the adjective immediately preceding it form a cornnpound.
7
Givan (1993) further correlates the distinction between parallel and scope interpretation of
adjectives with the restrictive/non-restrictive dichotomy: when adjectives modify the noun in a parallel
fashion they are non-restrictive; but ''when no intonational separation occurs, the adjectives are more
likely to be interpreted as a progression of restrictive modifiers" (Givan 1993:268). I am not touching
the distinction restrictive/non-restrictive adjectives here, as this would take me too far afield and would
further complicate an already complicated issue.
8
E. g.it is not clear how 'intensional', or 'sycategorernatic' in Higginbotham's terms, adjectives -like
former, alleged,falaH:an fit in any of (A)-(D).
9
A satisfactory account of universal ordering stated in such terms will probably relate it not only to
the conceptual-intentional interface system of Chomsky (1992, 1994), but also to other cognitive
systems with which language interacts (Chomsky 1995: 3; cf. Androutsopoulou 1995: 3). The crucial
question, then, seems to be how structural properties of (particular) languages affect, or may affect, this
order. It is expected, of course, not only that the order will be expressible structurally (even if not
motivated on purely syntactic grounds), but also that independently motivated (syntactic) processes may
naturally disturb this order (cf. Bosque & Picallo 1994: 25).
10
Sproat & Shih consider colour- the object's surface reflectance-as an absolute property on the basis
of relevant data from a number oflanguages; cf.: "... to establish that a car is red, one has to establish that
a sufficiently large amount of its surface looks red" (Sproat & Shih 1987:467). However, as one
anonymous reviewer remarks, this view has been recently challenged. Since further pursuit of this point
would take us too far afield, I leave the issue of the degree of absoluteness of colour adjectives open.
\I
See Hoepelrnan (1983: 194 fl) for a detailed exposition of this procedure of measuring. Such
adjectives belong to the non-intersective type, the sycategorematic one in Higginbotham's (1985) terms.
12
Although, in the area of formal semantics, the analysis based on the notion of vagueness is not
confined only to linear (tall, hot) and extensional adjectives (like fat}-in Hoepelrnan's (1983) terms-,
but extends even to predicative ones (like four legged and red), i.e. to those which are absolute, in
Sproat & Shih's terms.
13
Givan (1993: 268) makes a very similar observation, saying that the last adjective, the one
immediately before the noun, is non-restrictive because rather than identifying the referent of the noun
forms a new concept with it (see also fn. 7).
14
Cf. the term classifier-like/classifying adjectives used by Bosque & Picallo.
15
Cf.: "The Classifier indicates a particular subclass of the thing in question" (Halliday 1985: 164).
16
Cf.: "under appropriate conditions it is possible to interpret ( ... ) size adjectives as setting up a
taxonomy" (ibid.), but only when the combination of 'adjective-noun' constitutes a taxon in the context
of utterance and some contrastive stress is placed on the adjective.
17
"A sensible taxonomic type" (Sproat & Shih 1987: 470).
18
According to the recent literature, similarity is taken to be same-class-mernbership (e.g. two or more
adjectives denoting colour, or size or agency cannot co-occur in non-coordinated seqences, see Crisma
1993: 91-2 and her fn. 20).
19
Moreover, a head-movement analysis (for indefinite DPs) captures automatically the agreement
facts.
20
This view is also discussed be Radford (1993), but it is rejected in favour of the view of the
adjective as a head of its containing phrase (a selector).
21
Roughly corresponding to Picallo's (1991) Gender Phrase.
222
MELITA STAVROU
This objection refers, of course, to adjunction structures where the adjunct is not theoretically
restricted to the left of the 'host' category.
23
However, Lamarche (1991) argues against a N-movement analysis in the French OP, showing that
pre-and post-nominal adjectives are not syntactically related (through movement), but post-nominal ones
are base-generated in that position, hence the attested differences of meaing between the two occurences
of adjectives. It is also worth pointing out that Giorgi & Longobardi's analysis of the noun phrase and of
the position of adjectives in particular does not assume N-movement either; it is the AP that can, in
specific cases, move rightwards from its base position to the left of the noun.
24
See Alexiadou & Stavrou 1997 for an account.
25
Notice, however, that the same facts can be put in a different way, as in Androutsopoulou (1994,
1995), nam. in terms of the phenomenon of' detenniner spreading (OS)', which amounts to the fact that
in definite OP in Or the postnominal adjective(s) are obligatorily preceded by their own definite
determiner. In the starred examples in this text the postnominal adjective is not preceded by the definite
article. This restriction does not hold in indefinite OPs (for a discussion of these facts from a different
angle see also Karanassios 1992).
26
The absence of the article results in an indefinite interpretation in the case of singular; in the case of
plural nouns there is no indefinite article available ('bare plurals ').
27
Since in the minimalist framework optional movement is not allowed, there must be a trigger for the
movement just described. Given the iterpretative differences between sequences before N-movement and
those resulting after N has moved to F as presented in Stavrou (1996), it can be argued that N moves
obligatorily. However, I leave the issue of the exact trigger to further study.
28
In the spirit of Kamp (1975) and Hoepelman (1983). Cf. also the common term 'adnominal
adjective' .
29
And see Crisma (1995) and Bernstein (1992,1993) for an interesting discussion on the interaction of
structural positions and the lexical meaning of adjectives (and nouns) yielding the desired interpretation.
30
(8a-b) are fine, though, with a comma between the two As, indicating that these can be interpreted
as conjoined. Cf. further cases like: htipima aprosmeno, frikto 'stroke unexpected, terrible'), or the data
from Tzartzanos (1946: 69): ena trapezaki tetraghono, ksilino, aplo, avafo (a little-table square,
wooden, simple, colourless), bales pikres, farmakomenes ('bullets bitter, poisoned'), mia pipa hondri ,
mavri ('a pipe thick, black'). These examples show that in the case of conjoined APs, the N can skip all
the conjuncts, since these are not in a c-commanding relation to each other. Another, expected, fact that
is crucially relevant for the interpretation of several (asyndetically) conjoined APs 'left behind' after the
noun has moved, is that they are not prone to the scope reading, according to which every following
adjective modifies every preceding noun+adjective unit. So they involve no ambiguity as to whether
pattern (A) or (B) is involved; the only way to interpret them is under loose coordination, and to assume
that the noun has essentially raised over one structural. position. Sproat & Shih (1991: 578-9 in
particular) assume that in the case of our pattern (A) "the adjectives each constitute a separate
(minimally) intermediate phrase".
31
Notice, however, that Cinque's data differ from those in Crisma, since for the latter, sequences
involving two postnominal adjectives are also possible (see the relevant comment in Cinque (1993: 3334), who considers the co-existence of the two postnominal adjectives as marginal and probably as
involving a comma intonation between them).
32
It is further assumed (Karanassios 1992; Stavrou 1996) that in definite OPs, the AP in this position
must be embedded to its own FP headed by the def. art., due to the double definiteness phenomenon
existing in Greek (Androutsopoulou 1994, 1995).
33
The reader is referred to their work for a detailed analysis of these adjectives, which are considered
to refer to entities rather than properties. It is worth drawing attention to the fact that in Greek these-also
called relational-adjectives have a distinct derivational affix, -ik-os, whereby "they relate the noun to a
domain according to which the NP is classified and its denotation restricted" (Bosque & Picallo 1994:
17) and thus they function as elements that "add a 'with respect to' function to the lexical head" (ibid.:
18).
34
In the spirit of Given (1993), this adjective which is adjacent to the noun functions non-restricitively
(see fn. 7, 13 and section 4.1).
35
An anonymous reviewer remarks that the layered NP hardly avoids proliferation (of 'layers'). My
answer to this remark is that it is not simply the name of the 'proliferated' category that changes but,
22
APs IN THE DP
223
crucially, the purpose and, therefore, the justification of what is proliferated. The NP-shell is
independently motivated (Carstens 1993; Valois 1990; Bosque & Picallo 1994) and needed anyway for
thematically marked constituents, whereas the functional projections established in the DP by many
researchers do not have any independent existence but they only serve to accommodate adjectives, hence
they remain unlabelled, since their content is as yet unknown.
36
It is alsoworth while mentioning that Crisma (1993, 1995) takes it for granted that this sort of
classifying adjectives routinely incorporate into the modified noun. The same assumption is explicitly
made by Zamparelli (1994).
37
Guglielmo Cinque (p.c.) assumes that adjectives like interessante may be generated in the spec of
either a high category, bearing the feature {+'speaker oriented'], in which case they are not crossed, or a
lower one, below the position where the noun moves. Crisma (1995) proposes an analysis according to
which differences, or even nuances, in the interpretation of adjectives are assigned not only to the
'lexical meaning of adjectives but to the combined effect of this meaning with the structural position in
which the adjective is found; Bernstein's analysis (1992,1993) lies in the same spirit.
References
Abney, S. (1987) The English Noun Phrase in its Sentential Aspect, Ph.D. dissertation, MIT.
Alexiadou, A. (1994) Issues in the Syntax of Adverbs, Ph.D. Universitlit Potsdam.
Alexiadou, A. & M. Stavrou (1996a) On derived nominals in Greek, to appear in Horrocks, G., Joseph,
B. & Philippaki-Warburton (eds) Themes in Greek Linguistics II, John Benjamins, Amsterdam.
Alexiadou, A. & M. Stavrou (1996b) (A)symmetries in DPs and Clauses: Evidence from Derived
nominals, to appear in The Linguistic Review.
Alexiadou, A. & M. Stavrou (1997) Cross-linguistic Asymmetries in N-movement: a view from
morphology, to appear in ZAS Working Papers 8.
Androutsopoulou, A. (1994) The Distribution of the Definite Determiner and the Syntax of Greek DPs,
to appear in Proceedings ofCLS 30.
Androutsopoulou, A. (1995) The Licensing of Adjectival Modification, to appear in Proceedings of
WCCFL 14.
Bernstein, J. (1992) On the Syntactic Status of Adjectives in Romance, paper preseented at the CUNY
Forum 17.
Bernstein, 1. (1993) Topics in the Syntax of Nominal Structure across Romance, Ph.D. Dissertation
CUNY, New York.
Bosque, I. & M. C. Picallo (1994) Postnominal Adjectives in Spanish Indefinite DP, unpublished
manuscript, Universidad Comlutense de Madrid & Uninersitat Autonoma de Barcelona.
Carstens, V. (1993) On Grammatical Gender and NP-internal Subjects, GLOW Newsletter 30,64-65.
Chomsky, N. (1992) A Minimalist Program for Linguistic Theory, MIT Occasional Papers in
Linguistics, 1.
Chomsky, N. (1994) Bare Phrase Structure, MIT Occasional Papers in Linguistics S.
Cinque, G. (1993) On the Evidence of Partial N-Movement in the Romance DP, Working Papers in
Crisma, P. (1993) On Adjective Placement in Romance and Germanic Event Nominals, Working Papers
in Linguistics 3,81-107.
Crisma, P. (1995) On the Configurational Nature of Adjectival Modification, unpublished manuscript,
University of Venice & UCLA.
Delsing, L.-O. (1993) On Attributive Adjectives in Scandinavian and other Languages, Studia
Linguistica 47,105-126.
Ferris, C. (1993) The Meaning of Syntax. A study in the Adjectives of English, Longman, London.
Giorgi, A. & G. Longobardi (1991) The Syntax of Noun Phrases, Cambridge University Pres,
Cambridge.
Giusti, G. (1992) La Sintassi dei Determinanti, UNIPRESS, Padova.
Givon, T. (1993) English Grammar. A Function-based Introduction, YoU, John Benjamins,
Amsterdam/Philadelphia.
224
MELITASTAVROU
Halliday, M.A.K. (1985) Functional Grammar, Cambridge University Press, Cambridge.

Higginbotham, J. (1985) On Semantics, Linguistic Inquiry 16, 547-593.
Hoepelman, J. (1983) Adjectives and Nouns: a new Calculus, in Bauerle, R, Schwarze C.& Stechow A
von (eds.) Meaning, Use, and Interpretation ofLanguage, Walter de Gruyter, Berlin, pp. 190-220.
Horrocks, G. & M. Stavrou (1985) I 0ecoQla ,(cov L1eal1eooecov Kat EAAlJVlKO IUVCa1C"ClKo: Evoelsel~
'Yla Me"CalClVlJOlJ "ClJ~ Ava!poQlKoeQCO"ClJl1anKlJ~ !PQaOlJ~ O"ClJV 0<1>, Studies in Greek Linguistics S.
Horrocks, G. & M. Stavrou (1986) Mla 3tQCO"ClJ IIQoae'Y'YlOlJ Kat eQl1lJvela '(cov em{)enKcov
3tQoaoloQlal1COV O"Ca Nea EA.A.lJVlKa, Studies in Greek Linguistics 6.
Horrocks, G. & M. Stavrou (1987) Bounding Theory and Greek Syntax: Evidence for wh-movement in
NP, Journal ofLinguistics 23, 79-108.
Jackendoff, R. (1977) X'-Syntax: a Study ofPhrase Structure, MIT Press, Cambridge, Mass.
Kamp, H. (1975) Two Theories of Adjectives, in Keenan, E. (ed.), Formal Semantics of Natural
Language, Cambridge university Press, Cambridge, pp. 123-155.
Karanassios,G. (1992) Syntaxe Comparee du Groupe Nominal en Grec et dans d'autres Langues,
Doctorat d'etat, Universite VIII. Paris, Vincennes.
Kayne, R (1994) The Antisymmetry ofSyntax, MIT Press, Cambridge Mass.
Kester, E.P. (1993) The Inflectional Properties of Scandinavian Adjectives, Studia Linguistica 47, 139153.
Larson, R (1988) On the Double Object Construction, Linguistic Inquiry 19,335-391.

Lamarche, J. (1991) Problems for NO-Movement to NumP, Probus 3, 215-236.
Leffel, K. & D. Bouchard (eds.) (1991) Views on Phrase Structure, K1uwer Academic Publishers,
Dordrecht.
Marti, N. (1993) Adverbial Modifiers in DP, paper presented at the second Workshop on the Syntax of
Central Romance Languages, Universitat Autonoma de Barcelona.
Napoli, D.-J. (1989) Predication Theory, Cambridge University Press, Cambridge.
Ouhalla, J. (1991) Functional Categories and Parametric Variation, Routledge, London & New York.
Picallo, C. (1991) Nominals and Nominalizations in Catalan, Probus 3, 279-316.
Quirk, R., S. Greenbaum, G., Leech & J. Svartvic (1972) A Grammar of Contemporary English.
Longman, London.
Radford, A. (1993) Head-hunting: on the Trail of the Nominal Janus, in Corbett, G., N. Frazer & S.
McGlashan (eds) Heads in Grammatical Theory, Cambridge University Press, Cambridge, pp. 73113
Ralli & Stavrou (to appear) Morphology-Syntax Interface: two Cases of Phrasal Compounds, Yearbook
ofMorphology.
Ritter, E. (1988) A Head-movement Approach to Construct State Noun Phrases, Linguistics 26, 909-929.
Ritter, E. (1991) Evidence for Number as a Nominal Head, paper read at the GLOW meeting, University
of Leiden.
Sadler, L. & D. Arnold (1994) Prenominal Adjectives and the PhrasallLexical Aistinction, Journal of
Sproat, R & C. Shih (1987) Prenominal Adjectival Ordering in English and Mandarin, Proceedings of
NELS 12, 465-489.
Sproat, R. & C. Shih (1991) The Cross-linguistic Distribution of Adjective ordering Restrictions, in
Georgopoulos, C. & R. Ishihara (eds.) Interdisciplinary approaches to Language, K1uwer Academic
Publishers, Dordrecht, pp. 565-593.
Stavrou, M. (1996) Adjectives in Modem Greek. An instance of Predication, or an old Issue Revisited,
Journal ofLinguistics 32, 79-111.
Szabolcsi, A. (1983) The Possessor that ran away from home, The Linguistic Review 1, 89-102.
Szabolcsi, A. (1987) Functional Categories in the Noun Phrase, in I. Kenesei (ed.), Approaches to
Hungarian 2, JATE, Szeged, pp. 167-190.
Szabolcsi, A. (1989) Noun Phrases and Clauses. Is DP analogous to CP or IP?, to appear in J. Payne
(ed.), The Structure ofNoun Phrases, Mouton de Gruyter, Berlin.
Szabolcsi, A. (1994) The Noun Phrase, in F. Kiefer and K. Kiss (eds.), Syntax & Semantics 27 The
Syntactic Structure ofHungarian, Academic Press, New York, pp. 179-274.
Tzartzanos, A. (1946) Neoelliniki Syntaxis, Athens.
APs IN THE DP
225
Valois, D. (1990) The Internal Syntax of DP and Adjective Placement in French and English,
Proceedings of NELS IS, 367-382.
Zamparelli, R.(1991) Cardinal Expressions and non-restrictive Adjectives, Proceedings ofFLSM II.
Zamparelli, R.(1994) Correspondences in Functional Structures across Major categories, unpublished
manuscript, University of Rochester.
CYPRIOT GREEK CLITICS AND THEIR POSITIONING RESTRICTIONS
Arhonto Terzi
Technological Educational Institute (T.E.l) ofPatras
1. INTRODUCTION
This paper is a study of clitic placement in Cypriot Greek (CGr) fInite clauses. 2 Its
primary purpose is to provide a systematic description of the positions in which
pronominal clitics surface in CGr, compare them with the counterpart structures of
Standard Greek (SGr), and attempt to offer a formal account of the differences in
clitic positioning between the two varieties. It will be shown that the ban on fIrst
position clitics that CGr demonstrates is unlike that manifested by the Slavic (the
sometimes-called Wackemagel) languages. Rather, the environments from which
CGr pronominal clitics are excluded are similar, although not identical, to those of
Portuguese and Galician.3 A central claim that will be advanced is that clitics adjoin
to a functional head that occupies the same position in the clausal structure in both
CGr and SGr; the fact that they often surface in a different position in each variety is
seen as the result of the overt movement which the fInite verb undergoes in CGr. I
consider ~ (the head of M(ood) P(hrase as the landing site of fmite verb
movement and conjecture that V-to-M movement is related to the licensing
requirements of CGr clitics which, unlike those of their SGr counterparts, have to be
satisfIed before spell-out.
2. CYPRIOT GREEK IS NOT A WACKERNAGEL LANGUAGE

Contrary to SGr, where clitics always precede the fmite verb and surface fIrst in
linear order when no other element precedes the verb, (la), clitics may not appear
sentence initially in car as the contrast between (la) and (2a) illustrates (Terzi
1992).4
(I) a.
b.
(Ego) to diavasa
(I)
it read-I sg
*Diavasa to
read-Isg it
'I read it'
(SGr)
227

ARHONTO TERZI
228
(2) a.
*To edhkiavasa
it read-lsg
b. Edhkiavasa to
read-lsg
it
'I read it'S
(CGr)
As a result of this prohibition, CGr clitics often surface second in the clausal
structure (2b), reminding us of the clitic ordering manifested by languages with
second position (2P) restrictions, such as SerbolCroatian (SIC), (3) (or Ancient
Greek, see Rivero and Terzi (l995.
Vidam ga svaki dan
I-see him every day
b. *Ga vidam svaki dan
him I-see every day
'I see him every day'
(3) a.
(SIC)
It will be demonstrated that, despite the existence of examples such as (2b), CGr is
not a language in which clitics occupy the second position in the clause. A closer
look at the facts suggests that what seems to be the general requirement, is that
clitics appear immediately after the fInite verb regardless of their linear position
with respect to the rest of the constituents. (4a) illustrates, for instance, that clitics
follow the verb when an overt subject is present, thus they end up in third position,
and (Sa) that clitics are postverbal (and adjacent to the verb) even when more
material than just the subject precedes the verb. 6
(4) a.
I Maria edhkiavasen to
Maria read-3sg
it
b. *1 Maria to edhkiavasen
Maria it read-3sg
'Maria read it'
(5) a. Poli anthropi panda kamnoun to sosta
Many people always do-3pl it correctly
b. *Poli anthropi panda to kamnoun sosta
Many people always it do-3pl correctly
'Many people always do it correctly'
Therefore, CGr differs signifIcantly from Serbo/Croatian (and Ancient Greek), in

which clitics must always follow the fIrst constituent of the clause but do not have to
be adjacent to the verb.
(6)
Maria ga je citala
Mary it has read-3sg
'Mary was reading it'
(SIC)
CYPRIOT GREEK CLITICS
(7)
229
Mnogi ljudi ga pazljivo citaju

many people it carefully read-3pl
'Many people read it carefully'
In Rivero and Terzi (1995) the second position of SIC clitics is accounted for in
terms of movement of the element that surfaces fITst, to Spec(CP) or to Co,
depending on its X-bar status. The clitic occupies the Specifier of W(ackemagel)
Phrase, that is, of the Phrase that CO takes as its complement:
This proposal is supported by the fact that clitics immediately follow the
complementizer in embedded contexts:
(9)
(10)
Znam
da ga mnogi ljudi pazljivo citaju
know-1sg that it many people carefully read-3pl
'I know that many people read it carefully'
Ivan kaze da nam Olga nista ne daje
Ivan says that us Olga nothing neg gives
'Ivan says that Olga is not giving us anything'
The sentences in (3) and (4), but also the ones that follow immediately below,
indicate that this line of reasoning does not extend to Cypriot Greek, where clitics,
rather than being adjacent to Co, are adjacent to the verb, and, more precisely, they
follow it.
(11)
(12)
Ksero
oti i Maria edhkiavasen to
know-1sg that Mary read-3sg
it
'I know that Mary read it'
Ksero
oti poli anthropi kamnoun to sosta
know-1sg that many people do-3pl
it correctly
'I know that many people do it correctlyf7
The examples in (11) and (12) also argue that the postverbal position of clitics
in Cypriot Greek fmite clauses is unlikely to be the result orv-to-C movement since
CO is occupied by a lexical complementizer (see Beninca 1984 where a V-to-C
proposal is made to account, among other things, for partially similar clitic ordering
in medieval Romance, and also Fontana 1993).8 Therefore, adopting Kayne's
(1994) antisymmetry proposals that disallow right adjunction (of clitics to the verb
in this particular case) and, consequently, derive postverbal clitics from verb
movement past the functional head to which clitics adjoin, we are led to conclude
that the landing site offmite verb movement in Cypriot Greek is lower than Co.
ARHONTO TERZI
230
To summarise, the discussion in this section intends to establish that despite the ban
on ftrst position clitics, to which it is subject, CGr should not be considered a
Wackernagellanguage, that is, a language which is subject to 2P restrictions. While
in a 2P language clitics must appear second, in CGr clitics simply cannot appear ftrst
but they may surface in various positions in the clause provided they are adjacent to
the verb. This contrast supports the claim that a higher position serves as host for
SIC clitics, namely, the WP projection following CP; thus clitics immediately
follow CO in embedded contexts but they may be separated from the ftnite verb,
(10). On the other hand, CGr clitics are to be found in a lower position in the clausal
structure (Rivero 1994b).9
3. PREVERBAL CLITICS IN CYPRIOT GREEK

Clitics are not always postverbal in CGr, however. In fact, in a number of
environments they can only surface preverbally, patterning in this respect with the
order manifested by Standard Greek and Spanish/Italian. 10 The sentences below
show that clitics must be preverbal when they are preceded by Neg O(14), by the
subjunctive marker M O(15) or by a Focused XP, such as the DP in (16).
(14)
a.
b.
(15)
a.
b.
(16)
a.
b.
En ton iksero
neg him know-l sg
*En iksero ton
neg know-Isg him
'I don't know him'
Thelo
na ton dho
want-lsg subj him see-lsg
*Thelo na dho
ton
want-lsg subj see-lsg him
'I want to see him'\l
Tuto to vivlio su edhoken i Maria
This the book you gave-3sg Mary
*Tuto to vivlio edhoken su i Maria
This the book gave-3sg you Mary
'This book Mary gave you'
In the light of facts such as the above, I would like to propose that, despite the
different positions in which pronominal clitics surface in a number of contexts in
CGr, they occupy the same position in the clausal structure as in SGr (but also as in
Spanish or Italian). The plausibility of this claim is confrrmed at fIrst by the
paradigm in (17), where clitics appear preverbally and adjacent to the verb in all the
above languages.
(17)
a.
b.
Lipume pou i Maria en to djkjavazi

Lipame pou i Maria den to dhjavazi
(CGr)
(SGr)
c.
Lamento que
Maria no 10 lea
regret-lsg compl Mary neg it read-3sg
'I regret that Mary doesn't read it'
231
(Sp)
In order to account for the precise location of clitics in CGr, I will follow proposals
in Terzi (1996a) based on SGr, and hold that CGr pronominal clitics - which I
consider to be XO elements 12 - adjoin to a featureless functional head FO which takes
IP as its complement: 13
Subsequently, I would like to attribute the V-cl order of CGr fmite contexts to
verb movement past the functional head FO. As noted earlier, fmite verb movement
to CO is excluded since postverbal clitics and overt complementizers may cooccur
(see (11), repeated below): 14
(11)
Ksero
oti i Maria edhkiavasen to
know-lsg that Mary read-3sg
it
Movement of the fmite verb to FO is ruled out by Kayne's (1994) Linear

Correspondence Axiom, which we are adopting in the present work. Since FO is the
functional head to which clitics adjoin, adjunction of the verb to it amounts to
mUltiple adjunction, a configuration which is illicit because linear order cannot be
derived by hierarchical structure. Furthermore, V-to-F movement via substitution is
not an option to consider either, since it results to the order cl- V.
I would like to propose that the fmite verb movement of CGr (which is
responsible for the postverbal position of clitics) has MO as its landing site - the
functional head that encodes modality and is lexically realised in both SGr and
CGr: 15
Evidence for this claim (or, to be precise, evidence that the fmite verb does not
move higher than MO) is provided by the identical order of adverbs and MO when the
latter dominates the subjunctive marker, as in (20), or the verb that moves to it (21).
(20)
(21)
panda na to dhkiavazi
always subj it reads
, always to read it'
I Maria panda dhkiavazi to
Mary always reads
it
'Mary always reads it'
232
ARHONTO TERZI
Another position to be considered as a possible landing site for fmite verb

movement in CGr is some functional head between MO and FO, as is understood to be
the case in Martins' (1994) analysis of Portuguese. Martins (1994) considers Laka's
head of Sigma Phrase to be the landing site of fmite verb movement in Portuguese.
She bases her claims mainly on VP-deletion facts like in (22b).
(22)
a.
Deste-lhe
0 livro?
gave-2sg him the book
'Did you give him the book?'
b. (Sim,) dei
yes, gave-l sg
'Yes, I did'
Since no comparable evidence is present in Cypriot Greek, however, I regard it

unlikely that option is viable. Although Sigma Phrase might be present in CGr, it is
difficult to justify verb movement to r,0 in this language. In Portuguese, the V-to-r,
movement responsible for enclisis is triggered by strong S features evidenced by
(22).
Notice that my proposal concerning the location of the functional head to which
clitics adjoin in CGr is in many respects similar to that of Uriagereka's (1995).
Discussing Portuguese and Galician, on the one hand, and Spanish and French, on
the other, Uriagereka considers clitics to adjoin to the functional head ~ located in
the same position in both language types, despite the different surface position of
clitics in each one of them.17 In this respect we do not differ from Martins (1994)
either.
By contrast, I consider CGr to differ significantly from SIC, where clitics are in
Spec(wp), (Rivero and Terzi 1995):
(23)
Zalim
8to ga Maria nije
kupila
regret-l sg that it Maria not has-3sg bought
'I regret that Mary hasn't bought it'
(SIC)
That the WP of 2P languages is not the same as the FP of CGr becomes more
explicit when negation is inserted in the clausal structure. While NegP precedes FP,
(18), it follows WP:
(24)
... re, [C' COMP [wp CL [W' [NegP [IP [yp ... ]]]]]]]
4. FINITE V-TO-MOVEMENT AND LICENSING OF CLIncs

The reasons that trigger fmite verb movement in CGr are less clear than its landing
site. It is not immediately obvious at this point what type of features the verb has in
CGr which require checking in the overt syntax, and motivate V-to-M movement.
Notice that in Portuguese/Galician the strong F features of the fmite verb (suggested
233
in Uriagereka 1995) fmd a reasonable amount of empirical support in view of the

inflected infmitives with which the language(s) are associated (see also Martins' SO).
As a result of this, recasting the driving force of Portuguese fmite verb movement in
minimalist terms (Chomsky 1993, 1995) is not an unjustified direction.
It is not clear, however, that a similar approach is easily plausible for Cypriot
Greek. Considering movement to be driven by the morphological requirements of
the element that moves (Chomsky 1991, 1993) it is not easy to see, for instance,
how the morphological requirements of the Cypriot Greek fmite verb differ from
those of Standard Greek in a way that triggers V-to-M movement in the former but
not in the latter. Moreover, if V-to-M movement in CGr were triggered by strong
features (supposedly ~/modality features, since the landing site of verb movement
is MO) that require checking in the overt syntax, why is no such movement required
when a functional head is present in the clause, as in the paradigm (14)-(16) to
which we will return shortly? Finally, it is not obvious whether the CGr Infl has
features unique to it, which distinguish it from those of the SGr Infl and motivate
the proposed V-to-M movement. IS
A tentative solution I will consider here is that the fmite verb movement that
takes place in CGr (but not in SGr) is related to a difference in the level of
representation at which clitics are formally licensed in each variety. That is, I will
hold that CGr clitics have strong features that must be licensed in the internal
domain of a functional head with operator-like properties before spell-out. This is a
type of licensing that involves a head-complement rather than Spec-head relation
and must be satisfied before spell-out (see Rivero 1994b).
In (14)-(16), repeated immediately below, the functional heads licensing the
clitics in their internal domain are the negative marker, or NegO, the subjunctive
marker, or MO, and the head of Focus Phrasse respectively.
(14)
(15)
(16)
a.
En ton iksero
neg him know-lsg
b. *En iksero
ton
neg know-lsg him
'I don't know him'
a. Thelo
naton dho
want-lsg subj him see-Isg
ton
b. *Thelo na dho
want-lsg subj see-Isg him
'I want to see him'
a. Tuto to vivlio su edhoken i Maria
This the book you gave-3sg Mary
b. *Tuto to vivlio edhoken su iMaria
This the book gave-3sg you Mary
'This book Mary gave you'
ARHONTO TERZI
234
In the absence of such a functional head, verb movement to MO takes place and
gives rise to V-cl as the only acceptable order, as is illustrated by (2a)-(4a) repeated
below. Verb movement along these lines is a last resort operation, hence is excluded
from (14b)-(16b), where clitics surface preverbally.
(2) a.
Edhkiavasa to
read-lsg
it
b. *To edhkiavasa
it read-lsg
'I read it'
(3) a. I Maria edhkiavasen to
Maria read-3sg
it
b. *1 Maria to edhkiavasen
Maria it read-3sg
'Maria read it'
(4) a. Poli antbropi panda kamnoun to sosta
Many people always do-3pl it correctly
b. *Poli antbropi panda to kamnoun sosta
Many people always it do-3pl correctly
'Many people always do it correctly'
Support for the claim that the presence of a functional head, rather than of any
element preceding IP, is the crucial factor for the licensing of CGr clitics in
preverbal position is offered by the contrast between Clitic Left Dislocated
(CLLDed) and Focused constructions:
(25)
(26)
a.
*Tuto to vivlio to edhkiavasa

this the book it read-lsg
b. Tuto to vivlio edhkiavasa to
this the book read-l sg it
'As for this book, I read it'
a. Tuto to vivlio su edhoken
this the book you gave-3sg
b. *Tuto to vivlio edhoken su
this the book gave-3sg you
'This book (s/he) gave you'
CLLDed XPs, such as Tuto to vivlio 'This book' in (25), are CP adjoined (Cinque
1990, latridou 1991), or IP-adjoined (Anagnostopoulou 1994). This amounts to
saying that no functional head is present to license the clitic in its internal domain
and V-to-M movement takes place satisfying the licensing requirements and
resulting in the order V-c1 of (25b). The focused Tuto to vivlio 'This book' of (26),
on the other hand, is in the Specifier position of Focus Phrase, a phrase headed by
the Focus operator (Tsimpli 1994, Agouraki 1993). The Focus head licenses the
strong features of the clitic in its internal domain with the result that the latter
235
surfaces preverbally. As expected, verb movement is excluded in (26), hence the

ungrammaticality of (26b). There is also evidence that not any functional head is
capable of licensing clitics in its internal domain but only heads with operator-like
properties. This was the case with NegO, MO, Focuso in the examples (14)-(16)
earlier. In addition to those, the following contrasts provide additional support for
this idea:
(27)
(28)
Pjos ton idhe?

who him saw
b. *Pjos idhe ton
who saw him
'Who saw him?'
a. *1 Maria to edhkiavasen
Maria it read-3sg
I
Maria edhkiavasen to
b.
Maria read-3sg
it
'Maria read it'
a.
While a CO with wh-features licenses CGr clitics in matrix environments with the
result that they appear preverbally, (27a), an empty CO cannot perform the same role
as the ungrammaticality of (28a) illustrates. V-to-M movement takes place giving
rise to the order V-cl (28b).
Along the same lines, the (lexical or not) CO of embedded declaratives cannot
license CGr clitics, hence they do not surface preverbally (29a) and (30a). Verb
movement to MOresults in the grammatical order V-cl, (29b) and (30b).
(29)
(30)
a.
*Ksero (oti) i Maria to edhkiavasen

know-lsg that Mary it read-3sg
b. Ksero
(oti) i Maria edhkiavasen to
it
a. *Ksero
(oti) poli anthropi to karnnun sosta
know-lsg that many people it do-3pl correctly
b. Ksero
(oti) poli anthropi karnnun to sosta
know-lsg that many people do-3pl it correctly
'I know that many people do it correctly'
By contrast, the CO of embedded interrogatives licenses the strong features of CGr

clitics in its internal domain, by virtue of hosting the wh-operator in its Specifier
position. This is indicated by the preverbal position of clitics in (31a):
(31)
a.
En ksero
ti
su edhoken
Neg know-is what you gave-3sg
ARHONTO TERZI
236
b.
*En ksero
ti
edhoken su
Neg know-lsg what gave-3sg you
'I don't know what s/he gave you'
The factive complementizer pu (31), behaves along similar lines supporting the
view in Anagnostopoulou (1994) that pu-CPs (in SGr) are identical to wh-CPs, in
the sense of being complements of the matrix verb with their Specifier occupied by
an empty operator.
(32)
a.
b.
Lipume pu
to dhkiavazi
regret-lsg comp it read-3sg
*Lipume pu
dhkiavazi to
regret-lsg comp read-3sg it
'I regret that (s/he) reads it'
(CGr)
If we consider the CGr factive complementizer to also host an empty operator in its
Specifier position, thus being able to license pronominal clitics in its internal
domain, we explain why clitics are preverbal in (32a).19
5. CONCLUSIONS
In this paper I focused on the clitic positioning restrictions of Cypriot Greek and
compared them with other languages where various restrictions apply but also with
Standard Greek which is not subject to similar constraints. I proposed that the
different positions in which clitics surface in SGr and CGr do not reflect different
adjunction sites for clitics in each variety but a parametric variation with respect to
whether the finite verb undergoes overt movement to MO. I argued that V-to-M
movement is responsible for the V-cl order of Cypriot Greek fmite contexts and
speculated that this movement is related to the fact that pronominal clitics have
strong features in CGr which must be licensed in the internal domain of a functional
head with operator-like properties before spell-out; it is only in the absence of a
functional head of this type that verb movement to ~ takes place. However, I leave
open for future research the possibility that V-to-M movement is related to the
features of MO in Cypriot Greek. This idea is primarily motivated by the fact that
Cypriot Greek, contrary to Standard Greek, lacks the future particle, which, along
with the subjunctive particle, is standardly considered to occupy the MO position (but
see also the discussion in note 18).
NOTES
Research for this paper was conducted when I was a postdoctoral FeIJow at the University of
Ottawa and was funded by grant SSHRCC Grant 410-91-0178. I would like to thank Richard Kayne,
Maria Luisa Rivero and two anonymous reviewers for comments and suggestions. My deepest thanks to
237
Kakia Petinou (CUNY/Speech and Hearing), Pavlos Pavlou (GeorgetownlLinguistics) and Kiriakos
Savvas (Ottawa) for patiently providing most of the Cypriot Greek data.
2
For a detailed discussion of clitic positioning in the context of imperatives and gerunds of both
Standard and Cypriot Greek see Rivero and Teni (1995) and Terzi (1996).
3
This amounts to saying that we do not group together languages with strict second position
restrictions, such as Serbo/Croatian, with Portuguese and Galician as is implicitly done in Rouveret
(1992) and Uriagereka (1995).
4
See Drachman (1993) for a brief discussion of similar facts encountered in the dialects of Chios
and Rhodos.
Compound tenses are not available in Cypriot Greek, thus (i)=(2a) stands for the Standard Greek
(ii) as well.
(i) Edhkiavasa to
read-lsg it
'I readlhave readlhad read it'
(ii)
(To) eho/iha
dhjavasi
it have-l sg/had-l sg read
'I havelhad read it'
Notice that quantified subjects do not result ina different position of clitics in CGr; thus, CGr
differs in this respect from Portuguese and Galician, in which the status of the subject affects the
position of clitics (cf. (i) vs. (iii) from Barbosa (1993) and Uriagereka (1995).
(i)
Oli idhan ton
all saw-3pl him
'All of them saw him'
0 kathenas theli
to
(ii)
the everyone want-3sg it
'Everyone wants it'
(iii)
*Todos virarn-no
everyone saw it
(iv)
Todos no viram
everyone it saw
'Everyone saw it'
The precise impact of subject quantifiers on clitic positioning in Portuguese and Galician is not clear
(Uriagereka 1995), but see also Raposo (1995). The contrast with CGr, however, strengthens the idea
(elaborated in section 4) that the presence of some lexically filled functional head is of crucial
importance for the licensing of preverbal clitics in CGr.
As for negative quantifiers in subject position, they are subsumed under the paradigm in (14)-(16)
of the following section, since they require the presence ofNegO (in both Cypriot and Standard Greek):
(v)
Kanenas *(en) ton idhen
(CGr)
Nobody dhen him saw
'Nobody saw him.'
The same holds when the embedded clause is introduced by the complimentizer pos.
(i)
Ksero
pos i Maria edhkiavasen to
it
However, proclisis is also possible.
s
Agouraki (1994) claims that enclisis is due to V-to-C movement in Cypriot Greek. She does not
consider examples such as (11)-(12), however, neither does she compare the second position restrictions
of Cypriot Greek with those of SIC.
9
Focusing on Slavic and Old Spanish, Rivero (1994b) argues for two different positions that clitic
pronouns may surface in the clause: a Complementizer or C-oriented position and an Inflection or 1oriented position. A similar view is held by Martins (1993) for Portuguese clitics; the interpolated
clitics of Portuguese (see note 10 below) belong, according to Martins, to the first type. As to why
238
ARHONTO TERZI
Portuguese clitics (and for that matter cOr c1itics as well) cannot possibly be in Co, see Rouveret's
(1992) arguments against such a line proposed by Madeira (1992).
10
More precisely, clitics must immediately precede the verb. That is, Cypriot Greek differs from
Old Romance and some varieties of Northern Portuguese (Rouveret 1992, Barbosa 1993, Martins 1993)
in that it lacks the interpolation facts of the latter. An example of interpolation is the one that follows,
from Old Spanish, where the clitic is separated from the verb by Negation.
(i)
Que ellos te non digan en que puede finar ...
that they you not tell in what can end
'Let them not tell you how it can end' (from Rivero 1994b)
11
The position of clitics with respect to the future particle tha cannot be tested as tha is
unavailable in cOr.
Its counterpart structure utilizes an embedded 'na'-introduced sentential
complement of what seems to be an impersonal verb.
(i)
Tha pao
(sOr)
fut go-lsg
(ii)
E
na pao
(cOr)
is(?) subj go-1 sg
'I will go'
12
On the XO status of Standard Greek clitics, see Anagnostopoulou (1994).
13
See Terzi (1996a) for a detailed discussion of the pO head to which clitics adjoin. In the spirit of
Kayne's (1994) anti symmetry program, it is proposed that po is a featureless functional head in finite
clauses - unlike Kayne (1989,1991) - but amounts to in gerunds and imperatives.
14.
A reviewer correctly points out that the incompatibility of enclitics and lexical complementizers
does not necessarily excludes a V-to-C approach, as there is always the possibility of recursive CPs,
discussed extensively in the context of Germanic languages (Vikner 1995 among others). It is not only
the lack of independent evidence for recursive CPs in cOr, however, that makes the above suggestion
rather implausible, but also the fact that there is no obvious reason why the finite verb would undergo
movement to Co.
15
See Rivero (1994a) and Terzi (1992) for a detailed discussion of M Owith reference to sOr and the
other Balkan languages.
16
I will not discuss here whether or how the verb skips over the clitic on its way to M O(in apparent
violation of the HMCIECP) but see Beninca and Cinque (1993) and Terzi (l996a) where this and related
issues are addressed in detail.
17
There are a number of differences concerning the nature of po between Uriagereka's (1995)
approach and that in Terzi (1996a), which, nevertheless, are not immediately relevant for the present
work.
18
Some additional facts are to be considered in more detail, however, before disregarding the idea
that cOr Infl is involved in the process ofV-to-M movement which is responsible for the V-cl order that
cOr manifests in a number of finite contexts. This approach, which is not necessarily compatible with
Chomsky (1993), is in line with Lasnik's (1995) Enlightened Self-Interest, and is tackled in Terzi
(1996b). As mentioned in note 11, cOr lacks the future particle tha which is available in sOr.
According to standard views of the Balkan clausal structure (Terzi 1992, Rivero 1994a) the future
particle, as well as the sUbjuntive marker, is in MO. The fact that tha is unavailable in cOr, may suggest
that cOr MOhas different features than sOr, and this is related to V-to-M movement. The lack of
compound Tenses in cOr (note 4) also supports such an idea.
Pinally, this would also explain why finite verb movement in GGr has MOas its landing site
rather than adjoining to the clitic that has itself adjoined to po. As a reviewer notices, the latter option is
not disallowed by Kayne's antisymmetry proposals that I am assuming here.
19
It should be pointed out that not all speakers of cOr share the judgements in (32). Por a number
of them, clitics can be postverbal in the presence of the factive complementizer, in apparent conflict
with our analysis. Interestingly, however, this disagreement in judgements correlates with a
disagreement concerning the status of pou (in sOr). While for Anagnostopoulou (1994) pou has
operator status, this is not the case for Varlokosta (1994).
ro
239
References
Agouraki, Y. (1993) Spec-head Licensing: The Cases of Foci, Clitic Constructions and Polarity Items. A
Study of Modern Greek, Ph.D. Dissertation, University College, London.
Agouraki, Y. (1994) On the Enclisislproclisis Alternation, paper given at the Second International
Conference on Greek linguistics, University of Salzburg.
Anagnostopoulou, E. (1994) Clitic Dependencies in Greek, Ph.D. Dissertation, University of Salzburg.
Barbosa, P. (1993) Clitic Placement in Old Romance and European Portuguese, unpublished
manuscript, MIT.
Beninca, P. (1986) Un' Ipotesi sulla Sintassi delle Lingue Romanze Medievali, Quaderni Patavini di
linguistica 4, 3-19.
Beninca, P. & G. Cinque (1993) Su a1cune Difference fra Enc1isi e Proc1isi, in Ommagio a Gianfranco
Folena, Editoriale Programma, Padova, pp. 2313-2336.
Chomsky, N. (1991) Some Notes on Economy of Derivation and Representation, in R. Freidin (ed.),
Principles and Parameters in Comparative Grammar, MIT Press, Cambridge, Mass.
Chomsky, N. (1993) A Minimalist Program for Linguistic Theory, in K. Hale and S.-1. Keyser (eds.),
The view from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger, MIT Press,
Cambridge, Mass, pp. I-52.
Drachman, G. (1993) The attachment of (Greek) Pronominal Clitics, paper given at the First
International Conference n Greek Linguistics, University of Reading.
Fontana, J. M. (1993) Phrase Structure and the Syntax of Clitics in the History of Spanish, PhD
Dissertation, University of Phi ladephi a, Pennsylvania.
Iatridou, S. (1991) Clitics and Island Effects, unpublished manuscript, MIT.
Kayne, R. (1989) Null Subjects and Clitic Climbing, in O. Jaeggli and K. Safir (eds.), The Null Subject
Parameter, Kluwer, Dordrecht, pp. 239-261.
Kayne, R. (1991) Romance Clitics, Verb Movement, and PRO, Linguistic Inquiry 22,647-686.
Kayne, R. (1994) The Antisymmetry of Syntax, Linguistic Inquiry Monographs 25, MIT Press,
Cambridge, Mass.
Lasnik, H. (1995) Case and Expletives Revisited: On Greed and Other Human Failings, Linguistic
Inquiry 26,615-633.
Madeira, A. (1992) On Clitic Placement in European Portuguese, UCL Working Papers in Linguistics 3.
Martins, A. M. (1993) Clitic Placement from Old to Modern European Portuguese, unpublished
manuscript, University of Lisbon.
Martins, A. M. (1994) Enclisis, VP-deletion and the Nature of Sigma, Probus 6, 173-205.
Raposo, E. (1995) Pr6clise, Enclise, e a Posi9ao do Verbo em Portugues Europeu, in J. Camilodos
Santos and F.G Williams (eds.), 0 Amor das Letras e das Gentes, Center for Portuguese Studies,
University of California, Santa Barbara, pp. 455-481.
Rivero, M. L. (1994a), Clause Structure and V-Movement in the Languages of the Balkans, Natural
Rivero, M. L. (1994b) On two Locations for Complement Clitic Pronouns: Serbo/Croatian, Bulgarian
and Old Spanish, paper presented at the Third Diachronic Generative Syntax Conference, Vrije
Universiteit, Amsterdam.
Rivero, M. L. & A. Terzi (1995) Imperatives, V- Movement and Logical Mood, The Journal of
Rouveret, A. (1992) Clitic Placement and the Wackernagel Position in European Portuguese, in L. Rizzi
(ed.), ClWcs in Romance and Germanic, Eurotyp Working Papers, pp. 103-139.
Terzi, A. (1992) PRO in Finite Clauses. A Study of the Functional Heads of the Balkan Languages,
Ph.D. Dissertation, GUNY Graduate Center.
Terzi, A. (1996a) The Linear Correspondence Axiom and the Adjunction Site of Clitics, in A. M. Di
Sciullo (ed.), Configurations, Cascadilla Press, pp. 185-199.
240
ARHONTO TERZI
Terzi, A. (1996b) Cypriot Greek and Serbo/Croatian: Clitic Positioning and Modal-Aspectual Features,
GLOW Newsletter 36, 93-94.
Tsimpli, I. M. (1994) Focusing in Modem Greek, in K. Kiss (ed.), Discourse Configurational
Languages, Oxford University Press, pp. 176-206.
Uriagereka, 1. (1995) Aspects of the Syntax of Clitic Placement in Western Romance, Linguistic
Inquiry, 26, 79-123.
Varlokosta, S. (1994) Issues on Modem Greek Sentential Complementation, Ph.D. Dissertation,
University of Maryland at College Park.
Vikner, S. (1995) Verb Movement and Expletive Subjects in the Germanic Languages, Oxford
University Press, Oxford.
NULL OPERATORS, CLITICS AND IDENTIFICATION:

A COMPARISON BETWEEN GREEK AND ENGLISH
lanthi-Maria Tsimpli
1. INTRODUCTION
The aim of this article is to discuss the properties of English and Greek
constructions involving null and overt operators. Assuming that operator-variable
structures are LF objects, differences in the overt/null realisation of the operator
and/or the element in the position of the variable between the two languages will be
argued to stem from independent differences in the properties of the Case and the
Agreement systems. The underlying motivation for this correlation between Case
and Agreement, on one hand, and the properties of A'-chains, on the other, is
claimed to be the identification requirement on A'-bound empty categories.
In section 2, I outline the theoretical background on operators and the
distinction between quantificational and referential operators. In section 3, Greek
data involving null operator structures (NOS) is discussed in relation to congeneric
English data; the differences pointed out concentrate on the obligatory presence of a
pronominal clitic in the Greek data and the nature of the empty category in the
English data. In section 4, the properties of purposive clauses in English are
investigated and a parallelism between purposive and relative clauses is suggested
on the basis of the properties of the operators involved in each case. In section 5, the
distribution of clitics in Greek interrogatives and relatives is presented; fmally, an
analysis of clitics as feature-identifiers in AGR-O is suggested in syntactic contexts
where clitics in Greek appear to be A'-bound.
2. THEORETICAL BACKGROUND
According to standard assumptions in the Principles and Parameters framework,
operators are subject to a licensing condition to satisfy Full Interpretation (FI) at LF,
namely that they bind a variable. Among other consequences, this condition bans
vacuous quantification in natural language. Consistent with this assumption is the
configurational defmition of operators (Browning 1987) which defmes operators as
241
IANTHI-MARIA TSIMPLI
242
elements occupying operator-positions, e.g. SpecCP, which structurally defme their

domain of scope. Thus, for an operator-variable structure to converge at LF,
operators must appear in SpecCP binding a variable within their domain of scope.
As for the interpretation possibilities of such a structure, operators are responsible
for the specification of a range of possible referents associated with the variable
they bind. For example, in an interrogative such as the one in (1) below, the whoperator restricts the set of possible referents that the variable can pick out:
(1) a.
b.
Who did you meet t?

for which x, x a person, you met x
Variables are subject to the Strong Binding Condition (Chomsky 1986) which
combines the requirement on the configurational constraints on the A'-chain they
form with their operators, with the interpretive possibilities specified by the range
defmed by the operator in question. That variables should be strongly bound has
interesting implications as regards the difference between null and overt operators;
instances of the former are assumed to be found in easy-to-please, toughconstructions, and purposive clauses, as illustrated by the examples in (2)-(5):
(2)
(3)
(4)
(5)
John is easy Op to please t

John is too stubborn Op for us to talk to t
Thessaloniki is pretty Op to look at t
John left a book Op for me to read t
In these null operator structures (NOS), the operator does not specify the range of
possible referents for the variable it binds. On the other hand, in all these cases,
there is an antecedent in an A-position responsible for restricting the reference of
the variable. Assuming that variables are r-expressions in that both are specified [-p,
-a], we can formulate the distinction between them in terms of the following
principles, where (6ii) applies when (6i) is inapplicable (Chomsky 1986):
(6) (i) An r-expression must be A-free in the domain of its operator.
(ii) An r-expression must be A-free.
(6i) is also relevant to restrictive relative clauses where the operator is null. The
semantic role of the variable in this case, according to Chomsky (1986), is fixed by
the reference of the head of the relative clause. In other words, the variable is Abound by the head of the relative clause, assuming a predication relation between
the CP predicate and the 'subject' NP of which the clause is predicated. I
However, this formulation does not distinguish further between (a) the semantic
properties of wh-operators in interrogatives, for example, and operators in NOS, and
(b) the empty categories involved in each case, both being variables, i.e. [-p,-a].
Such further distinctions have been suggested in Lasnik & Stowell (1991), where it
is argued that operators differ in quantificational force. In NOS of the type in (2)(5), the operator is non-quantificational, the antecedent being the noun phrase in the
NULL OPERATORS, CLITICS AND IDENTIFICATION
243
A-position, whereas in interrogatives and restrictive relatives the operator involved

is quantificational, i.e. it is a true QP, specifying a possible range of referents for the
variable it binds. 2 They extend this distinction such it has implications for the nature
of the empty category in the two sets of constructions, by introducing a new type
referred to as null epithet; this is available in cases where the operator is nonquantificational. With regard to relative clauses (see the examples in (7) below), the
distinction between appositives and restrictive relatives is also based on the nature
of the operator involved: in appositives (cf. (7b, the operator is not a QP and the
empty category is a null epithet, whereas in restrictives (cf. (7a, there is a 'true'
operator-variable structure:
(7) a. The books which/Op (that) John read
b. The books, which/*Op (*that) John read
Note, crucially, that in appositives the wh-phrase is obligatorily overt. This implies
that there is no necessary correlation between null and non-quantificational
operators, i.e. the latter can be either null as in NOS or overt, as in appositive
relatives. 3
Rizzi (1994), based on Lasnik & Stowell's idea, introduces the [variable]
feature to distinguish between true variables and null constants, whereby the former
are specified as [-a,-p,+v] whereas the latter has negative specification for all three
features. Rizzi's account capitalises on the two conditions responsible for the
distribution of non-pronominal empty categories. The licensing requirement
formulated in terms of the ECP and the identification requirement which can
independently be satisfied by a discourse antecedent in root clauses in colloquial
German which exhibit topic-drop and in subject-drop in diaries (Haegeman 1990).
This distinction between the licensing and the identification requirements. will
become significant in the account of the difference between English and Greek
NOS with respect to the (un)availability of an empty category in these contexts.
In what follows, we will consider structures involving null and overt operators
in English and Greek (Gr). The aim is to show that the distinction between
quantificational and non-quantificational operators is not sufficient to account for
the Greek data; additional considerations which involve parametric differences
between the two languages with regard to the Case and pronominal systems are
necessary to capture the range of the constructions under discussion.
3. THEDATA
Greek sentences involving adjectival predicates, tough-constructions and purposive
clauses with adjectival or nominal predicates differ from the English data in (3)-(4)
with regard to the obligatory presence of a clitic pronoun coreferential with the
antecedent in the matrix clause. This is shown by the examples below
(Theophanopoulou-Kontou 1986):4
244
(8)
a.
b.
(9)
a.
b.
(10)
a.
b.
(11)
I Maria ine omorfi na *(tin) kitas

the Maria is pretty subj cl-acc 100k-at-2sg
'Maria is pretty to look at'
I
filosofia ine vareti na *(ti) dhiavazis
the philosophy is boring subj cl-acc-read-2sg
'Philosophy is boring to read'
To panteloni ine poli fardhi (ja) na *(to)
foresis
the trousers is too big for subj cl-acc- wear
'The trousers are too big to wear'
I portokaladha ine poli kria (ja) na *(tin) pjis
the orange-juice is too cold for subj cl-acc-drink-2sg
'The orange-juice is too cold to drink'
0 musakas ine etimos (ja) na *(ton) valume sto furno
the musaka is ready for subj cl-acc-put-lpl in-the oven
'The musaka is ready to put in the oven'
Ta Aglika ine dhiskola (ja) na *(ta) mathis
the English are difficult for subj cl-acc-Iearn-2sg
'English is difficult to learn'
Tha aghorasi to vivlio (ja) na *(to)
dhosi
sti Maria
will buy-3sg the book for subj cl-acc-give-3sg to-the Maria
'He will buy the book to give to Maria'
Two general points are relevant here: in Gr, the distinction between tense-dependent
and tense-independent clauses is largely expressed through the use of the
subjunctive and the indicative, respectively. In all of the above examples, the
embedded clause is in the subjunctive form. The second point concerns the
purposive and the degree-clauses in (9)-(11), which can optionally be introduced by
the preposition ja 'for' as indicated in the examples. Note that the optionality of the
preposition cannot be generalised to all purposive clauses:
(12)
a.
b.
Prepi
na dhiavasume ta arthra *(ja) na *(ta)
should-imp subj read-lpl
the articles for subj cl-acc
hrisimopiisume stin analisi
use-lpl
in-the analysis
'We should read the articles to use them in the analysis'
Estile simera to ghrama *(ja) na *(to) lavi
sent-3sg today the letter for subj cl-acc receive-3sg
o Petros avrio
the Petros tomorrow
'She sent the letter today for Peter to receive it tomorrow'
The examples in (12a&b) are similar to the ones in (9)-(11) with regard to the
obligatory presence of the clitic pronoun. The two sets differ, however, as far as the
optional vs obligatory realisation of the prepositional complementiser is concerned.
An interesting contrast between the two sets can also be found in the English
245
congeneric examples provided in the translations above. In particular, the English

sentences require an overt pronoun in the object position, coreferential with the
object NP in the matrix clause. The question is what the differences between the
two types of purposive clauses are, in the two languages.
Contreras (1993) argues that the CP in NOS is always an adjunct (see also
Wilder 1991 for the adjunct status of CP in tough constructions). His argument is
based on 'antecedent-contained deletion' structures and on Condition on Extraction
Domain (CED) phenomena, although he points out that judgements on the latter are
not uniformly unacceptable: easy-to-please and some purposive constructions seem
to block adjunct construal with the embedded predicate as in (13a&b) whereas
tough-constructions and some examples of purposive clauses are not as clearcut, as
shown by (13c&d):
(13)
a. ?*When was Mary hard to please t

b. ?*When did you buy a book to give to Bill t
c. Who did you buy this book to give to?
d. Who did you buy a book to please?
That judgements are not clearcut suggest "... that apparent absence of CED effects
(at least when it involves extraction of complements) is not a reliable diagnostic for
complementhood." (Contreras op.cit., p.9). In other words, properties independent
of the complement/adjunct distinction may affect the discrepancy in judgements on
sentences like (13c&d).
The corresponding Greek data from degree-clauses and ordinary purposive
clauses are ungrammatical:
(14)
a.
*Pjos ine i portokaladha poli kriaja na tin

pji t
who is the orange-juice too cold for subj cl-acc-drink-3sg
'The orange-juice is too cold for who to drink?'
b. *Pote ine i portokaladha poli kria ja na cl-acc pjis t
when is the orange-juice too cold for subj her- drink-2sg
'*When is the orange-juice too cold for you to drink?'
c. *Pote aghorases ena vivlio (ja) na (cl-acc) dhosis
sto Vasili t
when bought-2sg a book for subj cl-acc- give-2sg to-the Vasili
'*When did you buy a book to give to Vasili?'
d. *Se pjon aghorases to vivlio (ja) na todhosis?
to whom bought-2sg the book for subj cl-acc-give-2sg
'*To whom did you buy the book to give to?'
The data in (14) clearly pattern with the extraction constraints associated with
adjunct clauses. If the CP in all NOS is an adjunct, however, then the difference
between the English translation of the data in (12) and the ones in (2)-(5) involving
NOS should be accounted for by invoking some difference other than the
complement/adjunct distinction. With regard to the Greek data, the obligatory
246
presence of the clitic remains to be accounted for, if the constructions under

discussion are accurately described as NOS in both languages.
4. CP IN PURPOSIVE NOS
4.1. English
The English constructions in (2)-(S) indicate that in NOS the antecedent of the
operator-variable structure c-commands it either strongly or weakly (Chomsky
1986).5 Assuming an adjunct status of the CP including the Null Operator (NO), the
contrast between the purposive clauses in (S) and (12a) repeated here as (ISa&b) is
problematic:
(IS)
a.
b.
John bought a booki [Opi for me to read ti]

We should read the articlesi [pRO to use themi in the analysis]
As mentioned in the previous section, in (ISb) an overt pronoun is required. The

contrast between the obligatory presence of a pronoun in this example and the use
of an empty category, i.e. a null epithet, in (2)-(S) indicates that a null operator is
not available. Let us assume that in (ISb), the purposive clause is VP-adjoined
whereas in (ISa) the CP is adjoined to the NP complement of the matrix verb. The
difference is schematised as follows:
(16)
a.
b.
John bought [NP [NP a book [CP Op for me to read tm

We should [vp [vp read the articles [CP PRO to use them
in the analysis]]]
In (16a) there is a predication relation between the antecedent and the CP clause
whereas this configuration is not available in (l6b). The unavailability of NOS in
(16b) implies that weak c-command is not sufficient to license NOS. It thus follows
that the strong binding condition on variables in NOS can be satisfied if either of
the two conditions hold: strong c-command as in (2)-(4) or a predicational relation
similar to the one found in relative clauses as in (S).
Evidence for the structure in (16a) comes from the constituent behaviour of the
NP as it appears in clefts, pseudo-clefts and reverse-clefts illustrated by the
examples in (17 a, b&c) respectively:
(17)
a.
b.
c.
It was [a book for Peter to read] that John bought

What John bought was [a book for Peter to read]
[A book for Peter to read] is what John bought
What we need to consider next is the suggested similarity between relatives and
purposive clauses in constructions such as (16a). The prima facie similarities
247
include the operator-variable structure and the predication relation between the NP
'subject' and the CP predicate. One of the differences between the two structures is
the infmitival status of the purposive NOS as opposed to the fmite status of relatives
that can be introduced by an overt wh-phrase.
With regard to both restrictives and appositives I will assume that there is a
predication relation involved between the head of the relative and the CP. As
already mentioned in section 1, Lasnik & Stowell suggest that there is a difference
between the wh-operator in restrictive and appositive relatives in that in the former
it is a true QP whereas in the latter it is a non-quantificational operator binding a
null epithet. This implies that identification of the empty category in an appositive
relative, in the sense of strong binding, is met by it being A-bound by the head of
the relative clause whereas in a restrictive the operator is a QP binding a variable. 6
According to Lasnik & Stowell, the head of the restrictive relative does not have
independent reference; rather, its reference is derived by the two sets denoted by the
head of the relative and the QP, which intersect to give an index to the whole NP
(see n.lV
Consider now purposive clauses of the type in (16a). The property that NOS in
English have in common is that identification of the empty category is met by coindexation with the antecedent in the A-position. In that sense, then, the null
operator in NOS, including purposive clauses as in (16a), is non-quantificational
and the empty category is a null epithet (see Lasnik & Stowell's claim for
topicalisation structures which also involve a null operator). If this conclusion is
correct then (16a) is similar to appositive and not restrictive relatives.
4.2. Greek
Recall that the properties of the Greek examples of tough-constructions and
adjective-CP structures differ from their English counterparts with respect to the
obligatory presence of a clitic pronoun coreferential with the antecedent in the
matrix clause, as in the examples reproduced below in (18):
(18)
a.
I Maria ine omorfi na *(tin) kitas

the Maria is pretty subj cl-acc-Iook-at-2sg
b. To panteloni ine poli fardhi (ja) na *(to) foresis
the trousers is too big
for subj cl-acc-wear-2sg
'The trousers are too big for you to wear'
c. 0 musakas ine etimos (ja) na *(ton) valume sto
furno
the musaka is ready for subj cl-acc-put-l pI in-the oven
'The musaka is ready for us to put in the oven'
I will assume the relevant structure to be similar to the English NOS equivalent,
where the clause is AP-adjoined:
(19)
I Maria ine [Ap [Ap omorfi [CP na tin kitas]]]
248
The frrst question we need to address is whether the Greek data also involve a null
operator in the embedded clause. In theoretical terms, the default hypothesis is that
they do: given that the motivation for assuming a null operator in such cases partly
stems from the semantic properties of the matrix predicate, expressed at LF in terms
of an operator-variable structure, the implication is that this representation should
be crosslinguistically available. More precisely, the licensing properties of the
operator involved and the condition that it binds a category with a certain featurespecification, is expected to hold in both languages. On the other hand, the
identification requirement can be satisfied in different ways crosslinguistically, on
the assumption that the category in question is not a true variable but a null rexpression with fixed reference (cf. Rizzi 1994).
With respect to the optional realisation of the prepositional complementiser ja
in (18b&c) and its obligatory presence in (12a&b) I suggest that the difference is
similar to the congeneric English examples; in (12) the purposive clause is VPadjoined whereas in (18b&c) it is adjoined to the AP predicate. The quasi-purposive
interpretation associated with the second pair is attributed to the semantic properties
of the predicate and the predication relation it enters with the matrix subject which
functions as the antecedent of the clitic element. 8 In the absence of a predication
relation in (12), the purposive function of the clause is conveyed by the realisation
of the prepositionja in C.
Going back to the examples in (18a-c), notice that the clitic must be
coreferential with the antecedent in the matrix clause. The adjectival predicate in
(18c) etimos 'ready' can also involve a coreferential subject or an oblique object in
the embedded clause, as is the case with similar English examples. In the former
case, the null subject is a pro and in the latter, it is a genitive clitic as shown by the
examples in (20):9
(20)
a.
b.
I ghata ine etimi na fai

to psari
the cat is ready subj eat-3sg the fish
'The cat is ready to eat the fish'
0 musakas ine etimos na tu- valume ti besamel
the musaka is ready subj cl-gen-put-l pI the bechamel
'Musaka is ready to put the bechamel sauce on'
The English equivalent of (20a), given in the translation, has been argued to involve
a control structure where the PRO subject of the infmitival is controlled by the
matrix subject. Given the fmite nature of the embedded clause in the Greek
sentences however, there is no reason why the subject/object/oblique object should
be treated in a different way. I will thus assume that the obligatory coreferentiality
of the clitic or the null subject with a matrix clause antecedent in the examples in
(18) and (20) is forced by the presence ofa NO in the embedded SpecCP. The fact
that coreferentiality is obligatory in these cases implies that in (18) and (20) there is
a binding relation between the NO and the clitic. If Lasnik & Stowell's distinction
between the two types of operators is correct, the implication is that, in the Greek
249
examples, the empty category in subject position and object clitics are of the same
type.
5. CLITICS IN INTERROGATIVES AND RELATIVES

Clitics in Gr can have a bound variable reading as shown by the following
examples:
(21)
a.
Pjosi aghapai ti mitera tui/j

who love-2sg the mother cl-gen
'Who loves his mother?'
b. Kathenasi aghapai ti mitera tui/j
everyone love-2sg the mother c1-gen
'Everyone loves his mother'
In (21) the c1itic pronoun is A-free in its governing category, which is the DP
object, thus satisfying BC(B). I will assume that, in such cases, coindexation
between the c1itic and the operator is an instance of coreferentiality rather than
binding (Reinhart 1983), hence the optionality of the bound variable reading for the
clitic.
Clitics coreferential with a wh-phrase may not appear in matrix interrogatives
as the data in (22) show: 10
(22)
a.
b.
*Pjon ton idhes?

whom c1-acc saw-2sg
'Who did you see him?'
*Pjon fititi
ton
idhes?
whom student c1-acc saw-2sg
'Which student did you see him?'
On the other hand, genitive clitics appear to differ in this respect from accusative
clitics:
(23)
Pjanu (tu)
ldepsan ta ldidhia?
whose c1-gen stole-3pl the keys
'Who did they steal the keys from?
b. Pjanu (tu)
aresi
i musiki?
whose cl-gen please-3sg the music
'Who likes music?'
c. Pjanu fititi (tu)
eftiakses e1iniko kafe?
whose student cl-gen made-2sg greek coffee
'For which student did you make Greek coffee?'
a.
250
Similar data seem to be available in Spanish (Suner 1988, Torrego 1994). A

possible account for the difference between the data in (22) and (23) may have to do
with the position that the clitic is associated with. II This difference is reflected in the
case features of the DP. The examples in (22) show that genitive, but not accusative,
clitics in Gr can be A'-bound by an overt wh-operator. Given that the distinction
does not involve D-linking as was suggested with regard to the examples in (22)
(see fn. 10), a plausible suggestion would be to distinguish between genitive and
accusative clitics in terms of structural differences in Case-features. 12 Notice that in
all the examples in (24), the genitive DP can, alternatively, be realised as a PP; in
this case, the clitic is excluded:
(24)
a.
Apo pjon (*tu) eklepsan ta klidhia?

from whom cl-gen stole-3pl the keys
b. Se pjon (*tu) aresi
i musiki?
to whom cl-gen please-3sg the music
c. Selja pjon fititi
(*tu) eftiakses eliniko kafe?
tolfor whom student cl-gen made-2sg greek coffee
Following Cardinaletti & Starke (1994) I will assume that clitics are XO elements
necessarily associated with a functional head with phi- and Case features.
Simplifying Cardinaletti & Starke's arguments, clitics, unlike strong pronouns, head
XPs that lack the nominal functional projection (CNP in the authors' terms)
responsible for the referential/specific index and phi-/Case features.13 The reason
why clitics need to appear on a functional head which belongs to the set of verbal
functional projections has to do with the recoverability requirement on deficient
pronouns which postulates that "features missing in a deficient structure must be
recoverable at all levels of representation" (Cardinaletti & Starke, op.cit., p. 78).
AGR-O is, thus, a likely host for accusative clitics. With regard to genitive clitics,
on the other hand, I will assume that they are heads of XPs that are associated with
a nominal functional projection within the DP. The availability of this functional
projection is responsible for associating the pronoun with a cluster of case and phifeatures. The functional head associated with case can either be realised as a
preposition or as genitive case-features realised on the clitic itself. Thus, the
distinction between genitive and accusative clitics can be reduced to the former
being part of a nominal functional projection bearing case-features. 14
An object (accusative) clitic coreferential with a wh-phrase is optionally
available in embedded clauses, as shown by the examples in (25) below (cf. (22:
(25)
a.
b.
Pjon fititi
ipes
oti tha (ton) aporipsis?
whom student said-2sg that will cl-acc reject-2sg
'Which student did you say that you will reject him?'
Pjon nomizis
oti (ton) prosevalan horis logo?
whom think-2sg that cl-acc insulted-3pl without reason
'Whom do you think that they insulted him without reason?'
NULL OPERATORS, CLiTICS AND lDENTIFICA TION
251
The relevant constraint on data such as (22) presumably involves the clausemate cooccurrence of the clitic and the wh-operator. Let us then assume that there is a
locality condition on a clitic interpreted as a bound variable, in that the specifier of
CP containing the (accusative) clitic should be occupied by a null operator. In other
words, whenever an accusative clitic is A'-bound in an interrogative structure, it
cannot appear in a root clause as this would be incompatible with the indefinite,
non-specific nature of the existential wh-operator (see Chierchia 1992 and Reinhart
1992 for a semantic analysis of wh-phrases as indefInite quantifiers). I will follow
Rizzi (1990) and Haik (1990) in assuming that the operator in the intermediate
SpecCP, i.e. the intermediate trace, does not have quantificational features, hence
the availability of the clitic in the embedded clause. IS
Summarising then the facts of object clitics in interrogatives, it appears that
accusative clitics can be A'-bound as long as there is a non-quantificational operator
in the local CP. The data that allow the bound variable reading for the clitic in a
matrix interrogative, as in (21), can be interpreted as involving a coreferentiality
rather than a binding relation. Notice crucially, however, that in all the cases
discussed in this section, the occurrence of clitics is possible, but not obligatory.
This point will become relevant with regard to the analysis of NOS where absence
of the clitic leads to ungrammaticality.
Let us now turn to the case of restrictive and appositive relatives in Gr (26a &
b) are examples of a restrictive and an appositive relative, respectively:
(26)
a.
Ta epipla
pu (*ta) eho
stin apothiki
ine
the furniture that cl-acc have-1sg in-the store-room are
ja petama
for throwing away
'The furniture I have in the store-room is for throwing away'
b. Ta epipla, pu *(ta) eho
stin apothiki,
ine
the furniture that cl-acc have-1sg in-the store-room are
ja petama
for throwing away
The contrast between (26a&b) concerns the presence of the object clitic: in the
restrictive relative the clitic is excluded whereas in the appositive the clitic is
obligatorily present (cf. also Stavrou 1984).16 Both types of relative clauses are
introduced by the complementiser pu, also used to introduce factive complements
and exclamatives but not ordinary [-wh] complement clauses (see Roussou 1992,
1994 and Varlokosta 1994 for a discussion offactives).17
The contrast between clitics in restrictive and non-restrictive relatives is not so
clearcut, however. In certain restrictives, a clitic may optionally be present as shown
by the examples below:
(27)
a.
Aghorase ena vivlio pu tha (to)

dhiavasi sto
taksidhi
bought-3sg one book that will cl-acc read-3sg at-the trip
'He bought a book that he will read on the trip'
252
b. Mia kopela pu (tin) simbathun oli ine i Anna

one girl
that cl-acc like-3pl all is the Anna
'A girl that everyone likes is Anna'
The optional presence of the clitic in these examples is associated with the
indefmite nature of the NP head of the relative. Witness the fact that if the indefinite
determiner is changed into a defmite one, the clitic is excluded (if the interpretation
is to remain that of a restrictive relative clause):
(28)
a.
Aghorase to vivlio pu tha (*to) dhiavasi sto

taksidhi
bought-3sg the book that will cl-acc read-3sg at-the trip
b. I kopela pu (*tin) simbathun oli ine i Anna
the girl that cl-acc like-3pl all is the Anna
A possible account of these data could be formulated in terms of the notion of

specificity as in (En9 1991) where the distinction is drawn between specific and
non-specific indefmites. 18 Assuming that the semantic function of a restrictive
relative is to restrict the range of possible referents for the interpretation of the head
NP, presumably via the predication relation outlined in section 1 above, the
quantificational properties of an indefmite NP containing a relative clause could
not, in any case, be those of a non-specific indefmite. 19
If something along these lines is on the right track, the implication is that in
predication relations involving quantificational operators, as in restrictive relatives,
the clitic is a specificity marker which, being A'-bound by the operator, enters a
predication relation with the indefmite NP. The specific interpretation, however, is
not due to the clitic itself; rather, it stems from the predicational relation whereby
the indefmite NP acquires specific status. This is precisely the difference between
wh-operators in interrogatives and operators in restrictive relatives; although both
are quantificational, the restriction on the possible range of referents responsible for
the interpretation of the variable is provided by the A'-Chain formed in the case of
wh-interrogatives, whereas in relatives, the restriction is provided by the 'subject' of
predication, i.e. the NP in A-position. 20
This is an important distinction between the unavailability of an accusative
clitic in matrix interrogatives (see examples in (22) above) and its optional presence
in restrictive relatives with indefmite NP 'heads'. In particular, whereas wh-phrases
are indefmite quantifiers lacking a referential index, the operator in relative clauses,
by virtue of the predication relation it enters with its 'subject', necessarily acquires a
specific status. We can interpret this as a difference between the lack of a referential
index in the former case and its availability in restrictive relatives.
This is an apparent contradiction in view of what was suggested earlier with
respect to the differences between genitive and accusative clitics, whereby the latter
must appear on AGR-O for identification purposes. This contradiction could be
resolved if we assume that accusative clitics can acquire a referential index in two
ways: either by appearing in AGR-O or by being A'-bound by an operator
predicationally related with an NP in an A-position.
253
Nevertheless, the point is that data such as (27) cannot be analysed in a way
similar to the appositives in (26b) given that in (27) the clitic is optionally rather
than obligatorily present. Recall that the distinction between the operator in
restrictive and appositive relatives according to Lasnik & Stowell is assumed to be
based on its quantificationallnon-quantificational properties, respectively. If this is
correct, then we can associate the obligatory presence of the clitic in (26b) and its
unavailability in (26a) with the difference in the operator involved in each case. In
particular, if a clitic can acquire a referential index via the second route, i.e. A'binding plus predication, the non-QP status of the operator will be irrelevant to the
referential index of the clitic. In other words, an operator which is not a quantifier
does not contribute to feature-recoverability with regard to referentiality and Case.
Therefore, in appositives, the requirements on the part of the features on the clitic
will only be sensitive to the 'head' in the A-position. 21
The facts of appositive relatives can thus provide a pattern very similar to what
we found in Greek NOS, namely a non-quantificational operator in a CP adjoined to
a predicate and a coreferential clitic obligatorily present. The difference between the
two structures concerns the antecedent responsible for fixing the reference of the
clitic: in appositive relatives the antecedent is the head of the relative whereas in
NOS with adjectival predicates the antecedent is a c-commanding NP in the matrix
clause.
To sum up, the distribution of clitics in interrogatives and relative CPs exhibits
the following pattern: where the operator is quantificational as is the case in matrix
interrogatives and restrictive relatives, the presence of a clitic coreferential with the
operator is disallowed, unless the 'head' of the restrictive relative is an indefmite
NP; in this case the clitic is optional. The optionality of the clitic in these cases was
attributed to a further distinction between operators with a QP status, namely their
participation in a predication relation which would thus entail a 'subject' NP with
independent feature-specification. In appositives (and NOS), the non-QP status of
the operator renders it irrelevant insofar as the recoverability requirement for the
features on the clitic are concerned. In this respect, the operator is a dummy element
and the clitic requires an A-binder to satisfy recoverability.
The issues that remain to be addressed are the following: assuming that the
distinction between the NOS under discussion, and structures with true
quantificational operators is valid in Greek too, as evidenced by the distinct
behaviour of clitics respectively, why does this language not use an empty category,
e.g. a null epithet in the relevant structures? Also, what is the role of the NO in
Greek NOS?
6. CLITICS AS FEATURE-IDENTIFIERS
On the assumption that (accusative) clitics in Gr are deficient, it was suggested in
the previous section that recoverability of Case and referential features can be met
either via the clitic incorporating into AGR-O or via A'-binding by an operator
predicationally related to a noun phrase in an A-position. Note that the second route
254
does not imply that the clitic does not appear on AGR-O; the difference regards the
presence of a referential index on this functional head. If a clitic is not A'-bound,
referentiality is exclusively associated with the features on AGR-O. On the other
hand, if the clitic is A'-bound by an Operator, then recoverability is associated with
the nature of the operator in question: if it is not a QP, as in NOS and appositives,
recoverability is met via the element in A-position, whereas if the operator is
quantificational, as in interrogatives and restrictive relatives, a clitic is excluded
unless the quantifier can be associated with a specific reading, with which clitics are
compatible.
If the NO in NOS does not contribute to feature-identification of the clitic
element, then what is its role? Recall that the obligatory presence of the clitic in
NOS correlates with its obligatory coreference with the antecedent in the matrix
clause. A plausible way of interpreting this correlation is to say that obligatory
coreference always involves a binding relation. More precisely, for the clitic to be
interpreted as a bound element coreferential with an antecedent in the matrix clause,
an intermediate operator is necessary to provide the 'open' sentence in which the
clitic is A'-bound. What this implies is that all NOS are 'predicational' in the
relevant sense (Contreras 1993).
Some evidence for the suggestion that accusative clitics are necessarily
associated with a referential/specific index is shown by the unavailability of CLLD
and clitic-doubling structures involving a universal quantifier:
(29)
a.
b.
*Kathenan, ton- simbatho

everyone cl-acc-like-l sg
'Everyone I like him'
*To-idha
kathe pedhi
cl-acc-saw-1sg every child
'I saw every child'
If universal quantification is a priori incompatible with a specific reading, the data

in (29) provide further support for the clitic in AGR-O being referential. On the
other hand, existential quantifiers can, in certain contexts, be associated with a
specific interpretation (see Cinque 1991 for similar effects in Italian CLLD
structures; see also Anagnostopoulou 1994, Roussou & Tsimpli 1994).
(30)
Merikus fitites, tus-sinandisa sto sinema

some students cl-acc-met-1sg in-the cinema
'I met some students in the cinema'
Let us therefore conclude that accusative clitics can be A'-bound by a dummy NO,
i.e. a non-QP, or by an existential quantifier if the latter can be associated with a
specific/referential reading. The distinction between QPs of the latter type and nonQP operators concerns the optional vs obligatory presence of the clitic.
Consider now the English picture of QP and non-QP operators. In both cases
the category bound by the relevant operator must be empty, albeit of different
255
feature-specification, namely a true variable and a null epithet, respectively. Note

that the fact that an overt pronoun can never occur in either of these structures in
English can be taken to show that pronouns cannot be A'-bound in the language.
Strong pronouns are also banned from the equivalent Greek structures:
(31)
a.
b.
c.
*Pjo apo ta pedhia

idhes afto?
which from the children saw-2sg it
'Which of the children did you see?'
*0 Janis, pu afton aghapane oli, ...
the Janis that him love-3pl all
'John, who everyone loves, ... '
*1 Maria ine omorfi na kitas
aftin
the Maria is pretty subj 100k-at-2sg her
The ungramrnaticality of the examples in (31) and the congeneric English structures
involving a pronoun can be accounted for in terms of the A'-disjointness
Requirement formulated as in (32):
(32)
The A'-disjointness Requirement (adapted from Aoun & Li 1990)

A pronoun must be free in the smallest Complete Functional Complex
(CFC) which contains it.
Notice that (32) would correctly exclude the presence of a pronoun in the English
structures, including NOS, but would not distinguish between pronominal clitics
and strong pronouns in Gr, in that both types should give rise to ungrammaticality,
if they are to be construed as A'-bound pronouns. This conclusion is obviously
falsified by data which allow or even require the presence of a clitic as is the case in
Greek NOS.
A possible way of resolving this problem while maintaining (32) is in terms of
the properties of 'deficient' and 'strong' pronouns, discussed above. Recall that
clitics lack functional nominal projections that would allow them to be associated
with Case and a referential index, thus forcing them to appear on some verbal
functional projection, AGR-O in our terms. Assuming that (32) refers to 'strong' and
'weak' but not 'severely deficient' pronouns, English pronouns and Greek clitics
differ in that the English pronominal system distinguishes between 'strong' and
'weak' pronouns but lacks the 'weakest' option, which Greek has, namely clitics.
This trivial observation becomes significant in connection with the distribution
of English pronouns and Greek clitics in the syntactic contexts depicted by (32). Let
us assume that the difference in the nominal functional projections of strong
pronouns and clitics is as in (33) (Cardinaletti & Starke 1994):
256
(33)
a.
Strong Pronouns
CPL
CL~SPL
S
b.
------------ IPL
Clitic Pronouns
IP
~P
where L stands for lexical category, S for prosody-related features (Laka 1990) and
C for the functional head bearing Case features (and the referentiality index).
Feature-checking for strong pronouns will involve a Spec-head relation with an
appropriate functional projection, e.g. AGR-O, at LF, where the nominal projection
of the strong pronoun in (33a) appears in the Spec position of this projection.
Presumably, whether this configuration holds at PF as well will be determined by
parametric properties as is standardly assumed for all overt movement processes
(Chomsky 1995). Clitics, on the other hand, will have to appear in AGR-O to meet
recoverability at PF and LF, thus implying that for a clitic structure to converge at
both levels, spell-out of clitics should be a head-adjunction structure in AGR-O.
Assuming that overt verb-raising to AGR-O is motivated by strong features on this
functional projection, we can interpret the morphophonological properties of clitics
as responsible for rendering the set of verbal features on AGR-O strong. Note that
this does not imply that in a language like Gr, overt verb-raising to AGR-O is
always available. If morphological properties determine parametric variation, the
strong/weak alternation can depend on the element that identifies the relevant
features, namely the clitic. The role of the clitic in other words, is that of a featureidentifier, where the relevant features are phi-features and Case. I suggest that this
can be taken to imply that 'severe deficiency' in the pronominal system reduces to
feature-identification whereas strong pronouns, being independent functional XPs
are relevant to whatever binding constraints are associated with nominal
expressions, one of which is the A'-disjointness requirement in (32).
If this line of reasoning is correct, then clitics should be allowed to be A'-bound
as long as the features they identify are compatible with those of the operator. In the
case of quantificational operators the referential/specific index is missing, hence the
only possible A'-bound element is a true variable. In NOS, the operator inherits the
features from the predicational relation it enters with the antecedent in A-position,
hence the availability of the clitic.
257
To summarise, the reason why English uses a null epithet rather than an overt
pronoun in NOS is thus reduced to the unavailability of feature-identifiers in
English and the constraint on strong pronouns in (32). The remaining question is
why Gr does not use a null epithet in these structures, i.e. why Gr does not seem to
allow for an empty category with fixed reference. If identification of the null epithet
in English NOS is due to the antecedent in the matrix clause, and the same
configuration is available in Gr, as I have argued so far, an independent reason for
the absence of an empty category with the properties of a null epithet is called for.
A possible answer to this question can be formulated according to some
principle of economy in conjunction with recoverability. Feature-checking with
respect to Agreement, for example, involves an appropriate LF configuration in
both languages but overt verb-raising to AGR-O at PF is only required in Gr
assuming clitics to render AGR-O strong and thus visible at PF. The lack of overt
raising in English is the result of Procrastinate, according to which the
computational system prefers covert to overt movement.
On the other hand, it was previously suggested that 'deficient' forms of
pronouns, such as clitics, are preferred over strong pronouns on the basis of some
economy condition which Cardinaletti & Starke (1994) refer to as Minimise
Structure. Notice, however, that morphological identification and its corresponding
structure can be minimised, provided FI is preserved. The implication is that the
choice between an empty category and a clitic in Gr could not be free but regulated
by distinct feature-specification. Thus, in Greek NOS an empty category will give
rise to different feature-specification in the interpretive component and, in any case,
should not convey the referentiaVspecific interpretation required by the construction
in question. This is precisely what we fmd in cases of purposive clauses with nonspecific indefmite objects in Gr:
(34)
a.
Tha paro
mazi mu kanena vivlio na (*to) dhiavaso
will take-1sg with me some book subj cl-acc read-1sg
'I will take some book with me to read'
b. Thelo
kati
na (*to) fao
want-1sg something subj it eat-1sg
'I want something to eat'
(34a) involves a quantified NP, but the quantifier is an existential polarity item
which disallows the specific interpretation of the indefmite NP, unlike ordinary
quantified NPs with pure existential quantifiers (see Tsimpli & Roussou 1996). In
(34b) the existential quantifier, having no restricted reference, favours the nonspecific reading. A clitic is disallowed in both cases although, as previously
discussed, a clitic is obligatory in purposive clauses with referentiaVspecific NPs.
Given the non-referential nature of the NPs in (34), the implication is twofold; first,
a clitic is excluded due to a clash of features and, secondly, the empty category in
(34) is licensed by the NOS as before, but lacks the relevant referential features.
The only possible interpretation in this case is that of arbitrary reference. Thus, the
claim with respect to the examples in (34) is that the empty category in the NOS
258
fails to be identified by the antecedent which lacks specific reference itself. The
relation is therefore iloose' in that the empty category is a pro with arbitrary
reference predicationally related via the NO with the indefinite quantifier in the
matrix clause:
(35)
TAKE 3x (xlbook) [NO ...proarbl
What (35) implies is that, in the case of a purposive NOS where the antecedent does
not allow a specific/referential interpretation, the empty category in the purposive
clause is a null pronominal with arbitrary reference, licensed by the NO and
'loosely' connected with the antecedent insofar as the latter involves a restrictor on
the choice of its variable x; in (35) the restrictor is 'book', as in (34a), whereas in
(34b) there is no such restrictor.
7. CONCLUSION
In this article, Greek and English constructions involving quantificational and nonquantificational operators have been discussed with the aim of accounting for
certain differences between the (un)availability of an empty category in the two
languages. It was argued that NOS involve a dummy operator responsible for
binding the clitic/null epithet in the two languages. Greek clitics were analysed as
feature-identifiers in AGR-O, this status being compatible with the morphosyntactic properties of the 'weakest' form of pronouns in Cardinaletti & Starke's
terms.
The presence of clitics in A'-bound positions is associated with the features of
the antecedent and the nature of the operator involved: if the latter is a QP, as in
interrogatives and restrictive relatives, a clitic can appear, provided the operator is
compatible with a specificity index, as is the case, for example, with D-linked whphrases. The unavailability of an empty category in Greek NOS is argued to stem
from the recoverability condition on feature-specification whereby a clitic, but not
an empty category, can be associated with referential features. The distinction
between strong/weak features on AGR-O associated with morphological differences
in Greek has been argued to be responsible for the differences between Greek
PFILF and English PFILF representations. Finally, this distinction also accounts for
the lack of a referential reading of an empty category, i.e. a null epithet, in Greek.
NOTES
I assume that the variable is also A'-bound by the operator in SpecCP as a result of the raising of
the wh-operator internal to CPo Note that (6) refers to A-binding and as such it intends to cover the
requirement on strong binding for variables. In other words, co-indexation between the head of the
relative and the variable is there to satisfy the interpretation restrictions on the variable, hence the
distinction between nun and overt operators. In what fonows, this suggestion win be discussed and
modified.
NULL OPERA TORS, CLIncs AND IDENTIFICA nON
259
Lasnik & Stowell's discussion is largely based on crossover phenomena which appear to
distinguish between QPs and NOs in the structures under discussion. The contrast is shown by data such
as the ones in (i) below:
(i) a.
"Whoi does hisi boss dislike ti
b. Johni was hard [NOi [PRO to persuade hisi boss [PRO to vouch for eilll
Their claim is that the lack of WCO effects in (ib) is due to the properties of the NO and the empty
category which is not a variable.
3
Cleft constructions in English are similar to restrictive relatives with regard to the overt/null nature
of the operator:
(i) It is John who/Op I saw
It seems to me, however, that their interpretation differs from that of restrictive relatives in that the
operator in (i) is not quantificational. Consider the pair in (ii) where (iia) is a cleft and (iib) a relative,
with the pronominal subject it a referential element:
It is THE BOOK that I read.
(ii) a.
b. It is the book that I READ.
Lasnik & Stowell claim that predication in the case of restrictives which involve a QP is a process of
relating the head of the relative with index j with the operator bearing index i. This results in a new
index k for the whole NP. This suggestion intends to capture the different quantificational properties
associated with the operator clause-internally and the determiner of the head NP. In the case of
appositives where the operator is, according to Lasnik & Stowell, non-quantificational, co-indexation is
direct, in that the operator acquires the index of the head of the relative.
If this analysis is correct, it can be extended to the distinction between (iia&b) to describe the
difference between clefts and restrictive relatives. What needs further explanation, however, is the
optionally overt nature of the operator in clefts but not in appositive relatives, assuming the same type
of operator is involved in both.
4
It should be mentioned that Greek does not have the equivalent of easy-ta-please constructions.
Predicates like easy are always impersonal in that the subject position cannot be occupied by a
referential NP:
"0 Janis ine efkolos na ton etbaristisis
(i) a.
the Janis is easy-masc subj cI-acc please-2sg
'Janis is easy to please'
b. Ine efkolo na etbaristisis to Jani
is easy subj please-2sg the Jani
'It is easy to please Jan'
Weak and strong c-command correspond to m-command and standard c-command:
a weakly c-commands 13 iff:
(i) a does not dominate; 13 and
(ii) there is no y, y= Xmax, such that y dominates aand excludes 13
a strongly c-commands 13 iff:
(i) a does not dominate 13; and
(ii) the first branching category dominating a dominates 13.
If this distinction to hold, the prediction is that the NO in non-restrictive relatives should not give
rise to WCO effects whereas in restrictive relatives it should:
Johflj, whoi hisi mother loves ti, ...
(i) a.
b.
The mani whoi hisi mother loves ti
Higginbotham (1980), Safir (1986) and Lasnik & Stowell agree that there are WCO effects in (ib) but
not in (ia), contrary to Chomsky's (1982) claim that WCO effects are absent in restrictive relatives too.
7
The predication relation in appositives then would be consistent with Chomsky's (1982) rule of
predication, operating on LF structures at LF', which coindexes the head of the relative with the index
of the operator-variable chain in the open sentence. In Lasnik & Stowell's analysis, however, the
distinction between the operators involved in restrictive and appositive relatives is responsible for the
difference in indexing in the two constructions. Thus, no need to postulate an additional level is
invoked.
260
The distinction between stative and non-stative predicates is relevant here; the quasi-purposive
interpretation is associated with the latter only:
(i) 0 Petros kseri
tin istoria (*ja) na ti
lei
sta pedhja
the Petros know-3sg the story for subj cl-acc-tell-3sg to-the children
*'Petros knows the story to tell the children'
In (i) the purposive interpretation is excluded as evidenced by the unavailability of the prepositional
complementiser ja. The Greek sentence is not ungrammatical because it can be construed in a different,
albeit irrelevant, way whereby the NP the story is a CLLD object of the embedded predicate. In other
words, the complement of the matrix predicate is not the NP but the CPo This structure is not possible in
English, hence the asterisk on the English translation of (i).
9
Given that the subjunctive form always bears agreement morphology I will assume that the null
subject in these cases is a pro rather than a PRO (cf. Terzi 1992).
10
Iatridou (1991) considers examples such as (22b) grammatical. Her own example is provided in (i)
below:
(i) Pja pedhja (ta) ma10ses?
which children c1-acc scolded-2sg
'Which children did you scold?'
To me there is no difference between 'which-NP' and 'who' in these interrogatives. A clitic becomes
relatively acceptable if the wh-phrase has a partitive reading as in (ii) (see also Anagnostopoulou 1994
for a detailed discuusion):
(ii) Pjo apo afta ta vivlia (to) ehis aghorasi?
which from these the books cl-acc have-2sg bought
'Which of these books have you bought?'
In my opinion this is due to the partitive phrase, necessarily encoding a specific reading (Diesing 1992)
and hence allowing for a coreferential clitic. So, for speakers like myself, whose judgments on 'whichNP' phrases are similar to those on 'who' phrases, the former are not 'specific' or not specific enough, in
the sense of D-Iinking, for a clitic to co-occur. For data like (ii), I will assume, following Iatridou
(op.cit.) that, in such cases, movement of the wh-phrase occurs from a dislocated position, hence no
variable is associated with the VP-intemal position. The possibility of a wh-phrase to appear in a
dislocated position is associated with D-Iinking (Pesetsky 1987, Anagnostopoulou 1994).
11
'Ethical datives' are realised in Greek as genitive clitics (see Dimitriadis this volume). In some
cases, they may be interpreted as benefactives, in which case they can double a genitive NP. In other
cases they cannot:
(i) Tha tuftiakso ena kafe tu Jani
will cl-gen-make-1sg one coffee the-Jani-gen
'I will make him a coffee'
(ii) Arghises
na mu- ksipnisis
pali
were-late-2sg subj c1-gen wake-up-2sg again
'You were late in getting up again'
Catsimali (1989) has suggested that such clitics have an adjunct status, contrary to my claim that clitics
must be directly or indirectly theta-marked. Note, however, that bare NP-adverbs, which are considered
true adjuncts, can never be doubled by a clitic:
to Jani
(iii) *Tu hronu, tha tu-dho
the year-gen will cl-gen-see-1 sg the Jani
'I will see Yani next year'
I will thus maintain that some form of theta-marking is relevant to the (un)availability of a cliticdoubling structure in Greek.
12
I will assume with Iatridou that, whenever a genitive clitic is associated with a wh-operator,
extraction of the wh-phrase is from a dislocated position rather than the base position (see also n.1 0).
13
In fact, the presence vs absence of a referential index, is assumed to be syntactically related to the
case properties of a nominal functional head, in Cardinaletti & Starke's analysis. Severe 'deficiency'
which is characteristic of clitics is thus associated with the absence of case within their functional
projection.
261
14
Notice that this distinguishes between genitive and accusative clitics in Cardinaletti & Starke's
analysis in terms of weak vs clitic pronouns, respectively. Both sets are deficient but only the latter
require a verbal functional projection to meet well-formedness criteria on recoverability.
IS
Rizzi (1990) argues that a wh-trace in the intermediate SpecCP is [-wh] in order for the featurematching specification between the [-wh] C and its Spec to be met. My interpretation of this suggestion
is that, for locality constraints on extraction to be satisfied, we need the embedded SpecCP to be filled
by an Operator which is not a copy of the moved wh-phrase. A null operator in such cases can thus be
interpreted as non-quantificational, assuming that the wh-feature is responsible for the quantificational
force in a wh-construction, at the LF level.
16
There are certain cases of restrictive relatives, however, where a clitic is possible (cf. Stavrou
1984):
Ta pedhja pu (ta)
aghapane oli ine kala
(i) a.
the children that cl-acc-love3pl all are good
'The children that everyone loves are good'
b. Aghorasa tis kremes pu (tis)- forane
ijinekes
bought-Isg the creams that cl-acc-wear-3pl the women
'I bought the creams that women wear'
It seems to me that most cases like (i), namely restrictives which optionally allow a pronominal clitic,
have a generic/gnomic interpretation associated with the relative clause. The quantifier oli in (ia), and
the generic subject phrase in (ib), clearly favour the generic reading. On the other hand, examples like
(26a), which do not have such an interpretation, disallow the clitic in the relative clause. I will thus
maintain that independent properties of the subject phrase inside the CP-relative may be responsible for
the gramrnaticality of (i). Clearly, such cases of restrictive relatives require further investigation. They
are, however, beyond the scope of the present discussion.
17
Restrictive and appositive relatives can also be introduced by a full DP which consists of a
determiner and a pronoun as shown in (i). The occurrence of clitics in these cases is exactly parallel to
the examples where the relative NP is absent, i.e. (26):
(i) a.
Ta epipla ta opia
eho
stin apothiki
ine ja petarna
the furniture the which have-l sg in-the store-room are for throwing-away
'The furniture I have in the store-room is for throwing away'
b. Ta epipla, ta opia eho stin apothiki, ine ja petama
the furniture the which cl-acc have-Isg in-the store-room are for throwing-away
In the following discussion I
only use examples with pu rather than the full relative pronoun for
consistency. None of the data on clitics depends on such a difference.
18
This is compatible with Cardinaletti & Starke's idea of a nominal functional projection in strong
pronouns associated with a referential index and its absence in the case of deficient clitic pronouns.
19
That clitics can only be associated with specific NPs has been suggested in the literature with
regard to topicalisation and left-dislocation constructions (see Cinque 1991 for Italian,
Anagnostopoulou 1994 and Roussou & Tsimpli 1994 for Greek).
20
Ouhalla (1996) argues that wh-phrases consist of the wh-feature, responsible for the
quantificational reading, and a pronoun which in some languages is morphologically realised and in
some is a null pronominal. This is consistent with Reinhart's (1992) suggestion that the existential
indefinite interpretation of wh-phrases is associated with the wh-feature, giving rise to a functional
reading and any possible restriction on its range is provided by the XP associated with the variable in
the position of the copy (see also Chierchia 1992 for the functional reading of wh-phrases).
21
If this conclusion is correct then the nulVovert nature of the operator is irrelevant insofar as the
quantificational properties of it are independently determined.
will
References
Anagnostopoulou, E. (1994) Clitic Dependencies in Modern Greek. Ph.D. Dissertation, University of
Salzburg.
Aoun, Y. & A. Y.-H. Li (1990) Three Cases of Logical Relations: Relative Scope, Bound Pronouns and
Anaphoric Relations, unpublished manuscript, University of Southern California.
262
Browning, M. (1987) Null Operator Constructions., Ph. D. Dissertation, MIT.

Cardinaletti, A. & M. Starke (1994) The Typology of Structural Deficiency. On the Three Grammatical
Classes, Working Papers in Linguistics, Vol. 4, n. 2.
Catsimali, G. (1989) Genitives Governing VP/Sentence. Studies in Greek Linguistics, 257-275.
Chierchia, G. (1992) Functional Wh- and Weak Crossover, Proceedings of the Tenth West Coast
Conference on Formal Linguistics, 75-90.
Chomsky, N. (1982) Some Concepts and Consequences of the Theory of Government and Binding, MIT
Press, Cambridge, Mass.
Chomsky, N. (1986) Knowledge ofLanguage: Its Nature, Origin and Use, Praeger, New York.
Chomsky, N. (1995) The Minimalist Program,. MIT Press, Cambridge, Mass.
Cinque, G. (1991) Types ofA '-Dependencies, MIT Press, Cambridge, Mass.
Contreras, H. (1993) On Null Operator Structures, Natural Language and LinguistiC Theory, 11, 1-30.
Diesing, M. (1992) Indefinites, MIT Press, Cambridge, Mass.
En~, M. (1991) The Semantics of Specificity, Linguistic Inquiry 22,1-25.
Haegeman, L. (1990) Non-overt Subjects in Diary Contexts, in J. Mascaro & M. Nespor (eds.)
Grammar in Progress, Dordrecht, Foris, pp. 167-174.
Haik, I. (1990) Anaphoric Pronominal and ReferentiallNFL, Natural Language and Linguistic Theory,
8,347-374.
Higginbotham, J. (1980) Pronouns and Bound Variables, Linguistic Inquiry 11, 679-708.
Iatrldou, S. (1991) Clitics and Island Effects, unpublished manuscript, MIT.
Laka, I. (1990) Negation in Syntax: On the Nature of Functional Categories and Projections, Doctoral
Dissertation, MIT.
Lasnik, H. & T. Stowell (1991) Weakest Crossover, Linguistic Inquiry, 22,687-720.
Ouhalla, J. (1996) Remarks on the Binding Properties ofwh-pronouns, (to appear in) Linguistic Inquiry.
Pesetsky, D. (1987) Wh-in-situ: Movement and Unselective Binding, in E. Reuland & A. Meulen (eds.)
The Representation of(In)definiteness, MIT Press, Cambridge, Mass, pp ..
Reinhart, T. (1983) Anaphora and Semantic Interpretation, Croom Helm, London.
Reinhart, T. (1992) Wh-in-situ: An Apparent Paradox, in P. Dekker & M. Stokhof (eds.), Proceedings
ofthe Eighth Amsterdam Colloquium.
Rizzi, L. (1994) Early Null Subjects and Root Null Subjects, in T. Hoekstra & B. Schwartz (cds.)
Language Acquisition Studies in Generative Gramma,. J. Benjamins Publishers, Amsterdam.
Roussou, A. (1992) Factive Complements and Wh-Movement in Modem Greek, in UCL Working
Papers in Linguistics, 4, 123-147.
Roussou, A. (1994) The Syntax ofComplementisers, Ph. D.Dissertation, University College London.
Roussou, A. & I.M. Tsimpli (1994) On the Interaction o(Case and .Definiteness in Modem Greek, in I.
PhiJippaki-Warburton, K. Nicolaidis & M. Sifianou (eds.), Themes in Greek Linguistics, John
Benjamins Publishers, Amsterdam, pp. 69-76.
Safir, K. (1986) Relative Clauses in a Theory of Binding and Levels, Linguistic Inquiry 17, 663-689.
Stavrou, M. (1984) 'R KALrUClJ avtrovul1La 1Ytl.~ XQlOQLIYtl.K~ avaq>0QLK~ xQoTaoL~ 11 SIlQT1JOlJ
al1oou avtLKLI1VOU xou LoaYOvtllL 11 TO "xou", Studies in Greek Linguisitcs 5, 121-136.
Suner, M. (1988) The Role of AGR(eement) in Clitic Doubled Constructions, Natural Language 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, CUNY.
Theophanopoulou-Kontou, D. (1986) Empty Categories and Clitics in Modem Greek: The Object Case,
Glossologia 5/6, 41-68.
Torrego, E. (1994) On the Nature of Clitic Doubling, unpublished manuscript, University of
Massachusets at Boston.
Tsimpli, I. M. & A. Roussou (1996) Negation and Polarity Items in Modem Greek, The Linguistic
Review, 13,49-81.
Varlokosta, S. (1994) The Properties of Pu-complements in Modem Greek, in I. PhiJippaki-Warburton,
K. Nicolaidis & M. Sifianou (eds.) Themes in Greek Linguistics, John Benjamins Publishers,
Amsterdam, pp. 61-68.
Wilder, C. (1991) Tough Movement Constructions, Linguistische Berichte 132, 115-132.
TENSE AND TEMPORAL ADVERBIALS IN GREEKI
George J. Xydopoulos
1. INTRODUCTION
The purpose of this paper is to discuss the tense system and some deictic temporal
adverbials (DTAdvs) in Greek (Gr). I will represent tense by applying the neoReichenbachian model. Furthermore, I will suggest a mapping mechanism for the
syntactic representation of tense departing from Giorgi & Pianesi's (1991) theory.
Finally, I will discuss the relation of tense and DTAdvs with the aim of providing a
syntactic analysis for the latter.
2. SEMANTICS OF TENSE: mEN NEO-REICHENBACIDAN MODEL

2.1 Preliminaries
Hornstein (1990) extends a theory of Tense Representation introduced by
Reichenbach (1947). This model recognises three primitive entities of time, the
Speech time (or S point), the Event time (or E point), and the Reference time (or R
point). S refers to the time of utterance (or to the narration time). E refers to the time
the event expressed by the predicate takes place. Finally, R refers to some kind of
"intermediate" time that is involved in complex tenses, or to the time fixed by
temporal adverbials (TAdvs).
S, Rand E are primitive objects related to each other. The system recognises
only SIR and EIR relations. Consequently, there is no SIB relation; S and E are
indirectly related. Their relation is mediated by R. Relations between the three
points are understood in terms of temporal precedence (symbolised by a line) and of
contemporaneity (symbolised by a comma). As an example consider now the
English sentence in (1):
(1) John had eaten the apple at 8pm.
The tense is Past Perfect modified by the TAdv at 8pm. In neo-Reichenbachian

terms, the representation of (1) will be as in (2):
263
GEORGEJ,XYDOPOULOS
264
The event of eating the apple took place before 8pm, 8pm is before the utterance
time. So E precedes R and R precedes S; so, indirectly, E is located before S.
Tense structures are not primitives themselves. They are compositions of the
relations that hold between SIR and EIR. So (2) is the result of composing the
relation "R precedes S" and the relation "E precedes R". (3) illustrates the whole
compositional procedure that yields the Basic Tense Structure (BTS) for the Past
Perfect:
Accordingly, Hornstein (1990: 117-8) proposes the Universal Tense Inventory given
in (4):
(4) Present
: (S,R) 0 (R,E) = S,R,E (i)

or (R,S)O (E,R) =E,R,S (ii)
Past
: (R_S) 0 (R,E) = E,R_S
Future
: (S_ R) 0 (R,E) = S_ R,E
Present Perfect : (S,R) 0 (E_R) = E_S,R (i)
or (R,S) 0 (E_R) = E_R,S (ii)
Future Perfect : (S_ R) 0 (E_ R)
Past Perfect
: (R_S) 0 (E_R) = E_R_S
Future in Past : (R_ S) 0 (R_E)
Proximate Future: (S,R) 0 (R_E) = S,R_E (i)
or (R,S)O (R_E) =R,S_E (ii)2
2.2 The Tense System of Greek: Application and Problems

Let me now proceed with the application of the model sketched so far to the tense
system of Gr. Gr displays ten different tense forms. Six are one-word (or simple)
tenses (i.e. Imperfect, Aorist, Future lterativelProgressive in Past, Present, Future,
and Future lterativelProgressive). The remaining four are two-word (or complex)
tenses (i.e. Present Perfect, Past Perfect, Future Perfect in Past, and Future Perfect).
Simple tenses are manifested as morphemes on the verb stem. Complex tenses are
represented by periphrastic forms of the type "auxiliary + indeclinable perfect
formative"; the tense morpheme appears on the auxiliary.
Ignoring the aspectual differences, the corresponding tense inventory for Gr,
based on the inventory in (4), will be as in (5):
(5) Present
(S,R)
0 (R,E) =
S,R,E
TENSE AND TEMPORAL ADVERBIALS IN GREEK
AoristlImperfect
FuturelFuture 1t.lPr.
Present Perfect
Future Perfect
Past Perfect
Future Perf. in Past
(R_ S) 0 (R,E) =
(S_R) 0 (R,E) =
(S,R) 0 (E_R) =
(S_R) 0 (E_R)
(R_S) 0 (E_R) =
(R_S) 0 (R_E)
265
E,R_S
S_R,E
E_S,R
E R S
The inventory in (5) looks to me problematic. First, there is no representation

for the Future IterativelProgressive in Past. This tense is distinct from the Future
Perfect in Past, just as Past Perfect differs from Imperfect (or from Simple Past).
Second, both Future Perfect and Future Perfect in Past do not have composed SREforms; this is a problem as the composed form is the interpretable form. 3 The fIrst
problem, as I expressed it above, is not that difficult to solve. In other words, I could
provide a representation for the Future IterativelProgressive in Past by assuming,
intuitively, that it is, a composition of the relations "R precedes S" and "R is
contemporaneous with E" arriving at a possible structure given in (6):4
(6) (R_S) 0 (R,E)
However, the real problem is with the nature of these two Futures in Past. I believe
that these two "Futures" have a modal interpretation. Can we say that they also have
a temporal interpretation ? So the question we need to ask is whether they are real
tenses or just modal forms. In the former case the neo-Reichenbachian model must
account for them, in the latter case the model does not need to provide BTSs for
them and maybe some other model will have to handle them. I will leave this
question open given the limitations of space and the scope of this paper. The second
problem is of technicaVmechanical nature. I believe that the system needs to be
uniform for all tenses for practical reasons, with no exception (cf. to account for
complex tense structures, for adverbial modifIcation, and for "Sequence of Tense"
phenomena; see Hornstein (1990. So, it must provide composed forms for all
tenses. Ignoring the case of the Futures in Past, I will try to provide a composed
form for the Future Perfect which arguably is a real tense. According to the
inventory in (5), the Future Perfect is represented as follows:
If we try to complete the procedure in (7) it will yield four possible composed SREforms:
The possibilities above imply more than one SIB relationship. This, according to
Hornstein (1990) and Comrie (1985), implies that the SIB relation is vague so the
system should not determine it. In my opinion, the possibilities in (8) are not all
266
GEORGE J. XYDOPOULOS
true. In (7), SIR and EIR are said to be "intrinsically ordered", that is, S precedes R
and E precedes R. Specifically, the system distinguishes between "intrinsic" and
"extrinsic" ordering. The former is temporally significant since the relation ItS
precedes R" (i.e. S_R ~ "future") is temporally opposite from the relation "R
precedes SIt (i.e. R_S ~ "past"). However, the latter is not temporally significant
since the relation ItS precedes R" (i.e. S,R) and "R precedes SIt (i.e. R,S) are
temporally equivalent (for discussion see Hornstein (1990), and Xydopoulos (1996)
among others). Respecting this distinction, only (8a&c) are possible since they
maintain the intrinsic ordering of points in (7). 5 I believe we can further reduce these
possibilities if we assume that the EIR relation is dependent on the SIR relation. This
dependence follows indirectly from the principle in (9) proposed by Hornstein
(1990:113):
(9) In a given BTS, if linear order is not intrinsically determined, assume that
the linear order of RE is identical to the linear order of SR.
This is an identity principle meaning that whenever we have the ER-points
extrinsically ordered (i.e. related with a comma) they will obtain the same order as
that of the SR-points. If S is to the left ofR in the one pair, R will be to the left ofE
in the other pair. So, in the SR-relation, R is dependent on S; in the ER-relation, E is
dependent on R. I would like to generalise this by assuming that SIR is the primary
relation and that EIR is the secondary relation. In other words, I will assume that the
primary relation (i.e. SIR) fixes the two extreme points of the composed
representation and that the secondary relation (i.e. EIR) is fitted into the boundaries
defmed by the primary relation. Given the composition in (7), S precedes R and E
precedes R. The boundaries are fixed by S_ R, so E will follow S and will precede R.
Accordingly, the completed procedure for Future Perfect will be:
There is empirical support for the composed form of the Future Perfect in (10).
Consider the sentence in (11):
(11)
0 Janis
tha ehi
fiji
otan tha
the-John-nom fut have-3sg leave-perf when fut
erthi
i Maria
come-3sg the Mary-nom
'John will have left when Mary comes'
This is a complex clause involving two sub-clauses linked by the temporal

connective otan (when). Its interpretation follows from the linkage of the BTSs of
the first and second predicates that will give the Derived Tense Structure (DTS) for
(11). This linkage is governed by the principles given below (Hornstein (1990: 15,
43):
267
(12)
(13)
a.
"X associates with Y" is defmed as: X is separated from Y by a

comma.
b. Constraint on DTS (CDTS): DTS must preserve BTS.
c. BTSs are preserved iff
(i) No points are associated in DTS that are not associated in BTS.
(ii) The linear order of points in DTS is the same as that in BTS.
a. Rule for Temporal Connectives (RTC)
In (13b), write the BTS ofTNS under the BTS ofTNS . Associate
2
the R points by moving R to R , placing E accordingly.

2
Movement ofR to a position associated with R must obey CDTS.

2
b.
(14)
[s ... TNS 1
[adjunct
TempConn [s ... TNS 2
S- 1
E -1
R
S
RTC~
R,E
2
]]
E 1 -R1
S- - -R,E
22
(14) is a well-formed DTS. The E point of the second predicate is associated with R
and consequently it is interpreted as taking place after the E of the fIrst predicate;
this is the desired interpretation. So (14) supports my assumption about the Future
Perfect composed SRE-form.
As a conclusion to this section, I am proposing the inventory for the Gr tenses
as given in (15), leaving aside the two Futures in Past for the reasons I mentioned
earlier:
( 15)Present
Aorist (& Imperfect):
Future (& Fut. 1t.1Pr.):
Present Perfect
Future Perfect
Past Perfect
(S,R)
(R_S)
(S_R)
(S,R)
(S_R)
(R_S)
o (R,E) =
o (E,R) =
o (R,E) =
o (E_R) =
o (E_R) =
o (E_R) =
S,R,E
E,R_S
S_R,E
E_S,R
SER
E R S
3. THE SYNTAX OF TENSE IN GREEK: MAPPING SRE-RELATIONS TO

MORPHEMES
My next task will be to link the SRE-representations I gave in (15) to appropriate
morphological instantiations for Gr. I will follow Giorgi and Pianesi's (1991) theory
whereby SIR and EIR relations are mapped into syntactic entities under the Biunique
Mapping Principle given below (Giorgi & Pianesi (1991: 5):
268
GEORGEJ. XYDOPOULOS
(16)
Biunique Mapping Principle

Temporal morphemes and T-relations are in biunique correspondence,
everything being equal.
(16) states that for a particular SRE-relation, there is one and only one morpheme J..l
that corresponds to it and that for a morpheme J..l there is one and only one SRErelation that corresponds to it. In particular, Giorgi and Pianesi claim that SIR and
EIR relations are syntactically realised as TNS morphemes projected as the lexical
heads T1 and T2 respectively.
First, I will differ from them in the categorial characterisation of these heads.
Following the standard claim, I will assume that they are functional and not lexical
categories in terms of X-bar theory (see Belletti (1990) or Ouhalla (1991) among
others). Consequently, these heads will lack any type of a-role assignment
properties (or T -role) as Giorgi & Pianesi claim.
Second, I will agree with them in that the SIR relation corresponds to a. TNS
morpheme. Third, following a suggestion in Belletti (1990), I will assume that the
EIR relation is realised as an ASP head.
So, different SIR relations will correspond to different tense morphemes. In
particular, contemporaneity of SIR will give the present tense morpheme.
Precedence of R to S will give the past tense morpheme. Precedence of S to R will
give the future tense morpheme.
A parallel strategy can be assumed for the EIR relations. Having assunied that
the EIR relation corresponds to an ASP head, different EIR relations are mapped
onto appropriate aspectua1 morphemes following the Biunique Mapping Principle
given in (16) above (or some adapted version of it). The "E_R" and "E,R" relations
will give the perfective aspectual morpheme borne by the main verb - either by the
declinable form or by the indeclinable perfect formative. The "R,E" relation will
give the imperfective aspectual morpheme borne - only - by the declinable form of
the main verb. (17) below illustrates the whole structure for both Tense and Aspect
(irrelevant details omitted):
(17) ... [TP Spec
Cr T [ASPP Spec [ASP' ASP [vP Spec [v' V ... ]]]]]]
Summarising, so far I viewed the tense system of Gr from a neoReichenbachian perspective. I provided SRE-representations for the tenses and I
also discussed the mapping mechanism from semantic representations to
morphological/syntactic realisation.
269
4. DEICTIC TEMPORAL ADVERBIALS AND TENSE

4.1 Temporal Sensitivity
Next, I will examine some DTAdvs in Gr, like hthes 'yesterday', avrio 'tomorrow',
simera 'today' . DTAdvs, in general, are said to be sensitive to the temporal
specification of the predicate they modify (see also Alexiadou 1994 for discussion).
So, yesterday is only compatible with a Past time reference, and tomorrow is only
compatible with a Future time reference. Consider the examples in (18) and (19)
that illustrate this sensitivity:
(18)
a.
b.
c.
(19)
a.
b.
c.
0 Janis
efije
hthes
the-John-nom left-3sg yesterday
'John left yesterday'
*0 Janis fevji hthes
'*John is leaving yesterday'
*0 Janis
tha ehi
fiji
hthes
the-John-nom fut have-3sg leave-perf yesterday
'*John will have left yesterday'
*0 Janis efije avrio
'*John left tomorrow'
0 Janis
fevji
avrio
the-John-nom PRES-Ieave-3sg tomorrow
'John is leaving tomorrow'
0 Janis
tha ehi
fiji
avrio
the-John-nom fut have-3sg leave-perf tomorrow
'John will have left tomorrow'
AORIST
PRESENT
FUT.PERF
AORIST
PRESENT
FUT.PERF.
In (18) I combined the adverbial hthes (yesterday) with three tenses of distinct time
reference. Only (18a) is well-formed, thus yesterday is compatible only with the
past. In (19) I did the same with the adverbial tomorrow. (19a) is ill-formed as the
adverbial and the tense are incompatible in time reference. (19b) is well formed
since the present tense may also have future reference. 6 (19c) is well formed for
obvious reasons. So, tomorrow is compatible only with future time reference.
However, things are rather different with a DTAdv like today. Consider (20)
below:
(20)
a.
b.
o Janis
efije
simera
the-John-nom PAST-Ieave-3sg today
'John left today'
o Janis
fevji
simera
the-John-nom PRES-Ieave-3sg today
'John is leaving today'
AORIST
PRESENT
GEORGEJ.XYDOPOULOS
270
c.
0 Janis
tha fiji
simera
the-John-nom fut leave-3sg today
'John will leave today'
FUTURE
As (20) shows, today is compatible with either future or past time reference. So,
compared to yesterday and tomorrow it is not equally sensitive to the temporal
specification of the predicate.
4.2 Semantics of Tense and Temporal Adverbs

I will follow a tendency in the literature in treating DTAdvs as referential
expressions (in the sense of En~ 1986, 1987). There is evidence that in some
circumstances they can function as ordinary arguments of the verb, that is as
ordinary DPs (see Larson 1985 and En~ 1986, 1987) among many others).7 Since
they are temporal expressions, they will not be assumed to refer to individuals but to
time intervals (see for instance Dowty 1979 or En~ 1987). So, an adverbial like
tomorrow will refer to the time interval of twenty-four hours, namely the interval of
a day, which follows the interval that contains the speech time. Similarly, the
DTAdv yesterday will refer to the interval that precedes the interval containing S.
In these terms, tense is also assumed to refer to some interval that is possibly
unspecified, at least in one direction. The role of the TAdv is to specify that interval.
Consider the examples:
(21)
a.
b.
0 Petros
efije
the-Peter-nom left-3sg
'Peter left'
0 Petros
efije
hthes
the-Peter-nom PAST-leave-3sg yesterday
'Peter left yesterday'
In (21a) the past tense refers to some unspecified interval prior to the speech time.
An available way to specify this interval is, say, by recourse to the context (cf.
Partee (1973. In (21b) the past tense refers to the same unspecified interval prior to
the speech time as (21a). However, here it is its clause-mate DTAdv that specifies
that interval by restricting it within the frame of the twenty-four hour interval (i.e.
one day) which is just prior to the one-day interval containing the time of speech.
Things are somehow different with the DTAdv today. Although it refers to a
time interval of twenty-four hours, that interval contains or else "surrounds" the
speech point, instead of being located prior or after it. The sentences given in (20),
which contain simera (today), are true at some unspecified subinterval within the
interval denoted by today. The fact that one can use either the present, the past, or
the future with it will be now linked to the choice of appropriate subintervals.
Specification of these subintervals can be done by the use of clock adverbials.
Consider (22) below:
(22)
271
0 Janis
eftje
simera stis okto m.m.
the-John-nom left-3sg today at-the eight after noon
'John left today ~'
This sentence is true at the interval denoted by the DTAdv simera (today) which
surrounds the speech point and contains a subinterval denoted by the past tense.
This subinterval is prior to the speech time and is specified by the point adverbial at
8pm.
What these observations really mean is that a DTAdv like yesterday or
tomorrow denotes an interval that is contained within the past or the future
respectively. On the contrary, the DTAdv today denotes an interval that overlaps
with that of past or future. For instance, when we use past tense with today, the
interval of the adverbial will still be "running", so to say, at the time of speech.
Let us now see how the semantics of DTAdvs are handled by the neoReichenbachian framework. As we saw in section 2.1, Hornstein claims that TAdvs
modify a sentence by anchoring to the R point of the BTS of the tense. The actual
DTS in (2) obeys the CDTS. Let us take a concrete example to illustrate this.
Consider again sentences in (20a), (20b) and (21b) in section 4.1. The
corresponding SRE-representations are given by (23a), (23b) and (24) respectively:
(23)
a.
b.
E,R_S
-yesterday ~
S,R,E
yesterday
-yesterday ~
E,R_S
*ER
, -S
yesterday
(24)
S,R,E
--tomorrow~
S_R,E
tomorrow
What (23) states is that given the BTS for Simple Past, the DTAdv yesterday is
associated to the R point. 8 This association has the result of shifting the tense
structure, yielding a DTS. This shift is either vacuous as in (23a) or non-vacuous as
in (23b). Of course (23b) is ill-formed since it alters the linear order of the BTS and
it associates R and S, not associated in the BTS, thus violating both clauses of the
CDTS. The derivation in (24) involves the BTS for Present and the adverbial
yesterday. It results in a non-vacuous shift of the BTS yielding a well-formed DTS.
Observe that the linear order of the SRE-points and the relations to each other
correspond to the SRE-representation for Future. This captures the idea that Present
tense can have future time reference. 9 As we just saw, Hornstein's model represents
the modification by DTAdvs through association of the adverbial to a single point of
the tense structure and simultaneous shift of the tense structure. In other words the
adverbial alters the structure of the tense as soon as it associates to it. However,
GEORGEJ.XYDOPOULOS
272
recall that I assumed the relation between tense and TAdvs to be established by the
latter specifying the interval of the former.
I believe that the assumptions I made following the insights of Partee, Dowty
and Eny could, at least partially, be translated into a model that treats TAdvs as
antecedents of tense and not as elements that alter its semantic structure (as
suggested by Hornstein). Assuming that a verb will always have a time argument in
its argument structure, tense will saturate that argument. If a sentence does not
contain a TAdv the tense will refer to an unspecified interval. If a TAdv is present,
the tense will get the index of the TAdv (i.e. coindexation) and its interval will be
specified; thus its range will be restricted. A number of problems nevertheless
remain open under this approach. Thus it is important to note that here coindexation
denotes coreference but not in terms of identity (as it is generally assumed for
pronouns) but in terms of inclusion (i.e. the adverbial being included within the
tense). Furthermore, the solution applies to the case of DTAdvs like hthes
(yesterday) and avrio (tomorrow) and not that of simera (today). In a nutshell, it
seems that my view of the relation between tense and DTAdvs as the latter being
antecedents of the former cannot be represented by the mechanism proposed by the
neo-Reichenbachian model discussed earlier.
4. THE SYNTAX OF TEMPORAL ADVERBIALS IN GREEK

In this section I will translate the attested semantic relation between TAdvs and
tense into syntactic terms. In particular, I will propose appropriate structural
positions for the DTAdvs. In order to do that I will assume that SRE-relations are
mapped onto morphemes as discussed in Section 3 yielding the corresponding X-bar
structure (cf. the structure in (17. My discussion will be based on the distribution
ofDTAdvs in the Gr clause as illustrated below:
(25)
a.
b.
c.
d.
0 Janis
eghrapse
hthes
ti dhjathiki
tu
the-John-nom PAST-write-3sg yesterday the-will-acc cl-gen
'John wrote his will yesterday'
0 Janis eghrapse ti dhjathiki tu hthes
?I*O Janis hthes eghrapse ti dhjathiki tu
?I*hthes 0 Janis eghrapse ti dhjathiki tu
(2Sc&d) are odd (or ungrammatical) without a pause (signalled by commas). Thus, I
consider these two possibilities non-typical as they involve pause intonation. I will
assume that they should be analysed as instances of topicalisation. Here, my
attention will focus on the distribution illustrated by (2Sa&b) which I take to be
typical. So, the typical positioning options are after the verb (as in 2Sa) and sentence
[mally (as in 25b).
Recall that the SIR relation is mapped onto the tense morpheme which is
realised as the head of a TP in terms of X-bar theory. Also recall that a TAdv will
serve as the antecedent of a given tense by coindexation. In structural terms, I will
273
assume that this is represented by projecting the TAdv as an adjunct to the TP node
as shown below: 10
(26)...
hp TempAdv hp Spec [T' T ... ]]] ...
Now, consider again the distributional data presented in (25), and especially
(25b) that illustrates the sentence-final position. This position is the most frequent of
all (this is also attested by Tzartzanos (1946) and Nacas (1987. Consequently, I
will not impose any directionality restrictions to the TP-adjunction. In other words, I
will permit for a TAdv to adjoin also to the right ofTP.1I So, (25a) and (25b) will be
accounted for in terms of left and right adjunction to TP respectively.12 My
proposal, at this point, goes against Kayne's (1994) theory. Kayne restricts all
adjunctions to the left, thus excluding right-adjunction. The sentence-fmal position
for him could be analysed, for instance, as a result of dislocating the sentence
around the adverbial without any obvious motivation (for discussion see Brody
(1994) and Manzini (1994.
Another available position for the adverbial would be that of the Specifier of the
TP as proposed by Alexiadou (1994).\3 This position seems to me impossible
whichever the phrase structure model one might assume. I can see two reasons for
that. First, if we exclude rightward direction of a phrase XP which is dependent to a
phrase YP (excluding complements) word-order instances like that illustrated by
(25b) cannot be given a straightforward account, in my opinion. Second, the (single)
Specifier of TP, in the minimalist programme, might not be available, after all, for
the TAdv as it may be used as a site where the subject DP of a sentence checks its
morphological features against that of Tense in order to be Case-licensed (see
Chomsky (1993), Bobaljik and Carnie (1992) among others), unless of course
multiple specifiers are allowed and the function of each one of them is clearly
defmed (see Chomsky (1995a, 1995b.
Finally, I must mention that Stroik (1990, 1992) provides independent evidence
that the TAdv is generated inside a Larsonian type of nested verb phrases. For him
the TAdv will be sister to V under V'. as illustrated below:
(27)...
[vp Spec [V' eV [vp DPDO [V' V Adv]]]]
However, a typical Gr sentence like the one in (28) below cannot be accounted for:
(28)
0 Janis
elise hthes
tin askisi
tis fIsikis
the-John-nom solved yesterday the-exercise-acc of-the physics
'Yesterday John solved the physics exercise'
This sentence illustrates the postverbal position of the TAdv in Gr. In the system I
am assuming here it would be assigned the structure in (29):
274
GEORGEJ.XYDOPOULOS
(29)
[AGRsP 0 Janisk elisei [TP hthes [TP

askisi]]]]]]
h 1i ] [AGRoP [vp tk [V' 1i tin
Here, the verb moves overtly from its original position, within VP, up to AGRs via
other intervening functional heads. The subject DP moves overtly, too, from [Spec,
VP] to the [Spec, AGRsP] (maybe via the [Spec,TP)). Given that the object DP
moves, only, at LF from inside V' to the Spec of AGRoP, we will get the derived
(PF) word-order in (28). Note crucially that all movements assumed in (29) are
necessary in order to satisfy morphological properties of the moved elements under
Greed. In addition to that, I wish to stress that in (29) the TAdv originates as
adjunction to TP (and not in the [Spec, TP)). Under Stroik's analysis, the claim
would be that hthes originates inside V' and then moves and adjoins to TP. I do not
see any reason for it to move at all. On the one hand, the adverbial is not interpreted
as a topic (no pause intonation and the like) so topicalisation is excluded. On the
other hand, the adverbial has no real morphological features to check (i.e. no Case
or cp-features as argument DPs), so we cannot assume that it is forced to move either
in the syntax or at LF.141S
5. CONCLUSION
I discussed the neo-Reichenbachian system of tense representation by applying it to
Gr. I examined a mechanism for the mapping of SRE-representations onto
morphemes. Furthermore, I dealt with the relation of DTAdvs to tense which I
viewed as involving coreference whereby tense refers to the TAdv in order to
restrict its range. Finally, I investigated the distribution of these adverbials and I
suggested an exact structural representation for them in terms of left and right
adjunction to TP.
NOTES
I wish to thank Rita Manzini and Neil Smith for commenting on earlier drafts of this paper. My
special thanks go to Rita Manzini for her insightful comments and suggestions and for her patience and
encouragement. I am also grateful to the audiences of the LAGB Conference at Middlesex University in
September 1994 and of the Workshop on Greek Syntax at FAS Berlin in December 1994 for their
comments. Finally, I also would like to thank an anonymous reviewer for hislher constructive comments
and suggestions. All mistakes remain my own.
2
A given language must choose either option (i) or option (ii) of these tenses but not both.
In all operations involving temporal connectives like before or after, "sequence of tense"
phenomena, and adverbial modification, the model makes use of the composed SRE-form.
4
This is the same structure as that proposed in (4) for the Past. As it will become clear later in the
paper, this should not be the case if we take into consideration the difference between the "E,R" and the
"R,E" relations in the representation of tenses.
5
An anonymous reviewer objects to this claiming that the relation "S precedes R and E precedes R"
can be equally expressed by all four possibilities in (8). This would be so only if the representation in
275
(7) included both intrinsic and extrinsic orderings of the points (see e.g. (6. However, this is not the
case here since the points are only intrinsically ordered. I believe that in a strongly ordered theory only
(8a&c) can be the completed forms for (7).
6
Given the limitations of space and the scope of this paper, I will not discuss the interesting issue of
the present tense having future reference in some cases. The reader is referred to the works of Comrie
(1985) and Binnick (1991) for some discussion.
7
As reminded to me by an anonymous reviewer, another piece of evidence for the "NP-argument"
nature of these adverbials, comes from locality issues whereby A-bar movement of such adverbials is
sensitive to weak islands; this is not the case with non-temporal adverbials.
8
An anonymous reviewer correctly notes that DTAdvs have an obvious link with S (i.e. deixis).
This does not mean, however, that DTAdvs modifY the S point, hence their association to R in the
system I am assuming here (cf. Hornstein 1990: 197).
9
Note that a hypothetical reference of present tense to past is ruled out by (25) since such an
interpretation could only be yielded if the points were reordered in violation of the CDTS.
10
Adjunction to TP is within the limits of the "Adjunction Principle" proposed in Chomsky (1986).
In addition to that, Bennis & Hoekstra (1989), among others, also propose that DTAdvs are generated as
adjunctions to TP (for Dutch).
II
In linear terms this means that a TAdv will have to precede (left-adjoined) or follow (rightadjoined) other adverbials (of other semantic classes). This is true for Gr as a TAdv is hierarchically
higher that an aspect-sensitive adverbial and a manner adverb (see Alexiadou 1994 and Xydopoulos
1996 for discussion).
12
Right and left adjuncts to TP can be viewed as right and left non-argument specifiers in the sense
of Laenzlinger (1993) (and Chomsky 1995b if we lift the corresponding restrictions); see Xydopoulos
(1996).
13
Alexiadou assumes that a phrase can have a single specifier position, also excluding any adjuncts
to it. She proposes that the DTAdv originates inside the VP and passes through [Spec, TPj for licensing
reasons (covertly in Or).
14
The current version of checking theory is restricted to the evaluation of pure morphological
features for convergence reasons (cf. Greed or Last Resort). Therefore, I believe, it cannot be extended
to features of semantic content (e.g. [temporalj, etc.).
15
Note that, at present, I do not have any analysis for the binding phenomena that led Stroik to claim
a VP-internal position for TAdvs.
References
Alexiadou, A. (1994) Issues in the Syntax of Adverbs. Ph.D. Diss. Universitiit Potsdam.
Belletti, A. (1990) Generalised Verb Movement, Rosenberg & Sellier Torino.
Bennis, H. & T.Hoekstra (1989) Generatieve Grammatica, Foris, Dordrecht.
Bobaljik, J. & A.Carnie (1992) A Minimalist Approach to Some Problems of Irish Word Order,
unpublished manuscript, MIT.
Brody, M. (1994) Phrase Structure and Dependence UCL Working Papers in Linguistics 6, 1-33.
Chomsky, N. (1986) Barriers, MIT Press, Cambridge, Mass.
Chomsky, N. (1993) A Minimalist Program for Linguistic Theory, in K. Hale and S. J. Keyser (eds.),
The View from Building 20. MIT Press, Cambridge, Mass, pp. I-50.
Chomsky, N. (1995a) Bare Phrase Structure, in G. Webelhuth (ed.), Government and Binding Theory
and the Minimalist Program: Principles and Parameters in Syntactic Theory, Blackwell, Oxford.
Chomsky, N. (1995b) The Minimalist Program, MIT Press, Cambridge, Mass.
Dowty, D. (1979) Word Meaning and Montague Grammar, Reidel, Dordrecht.
En~, M. (1986) Towards a Referential Analysis of Temporal Expressions, Linguistics and Philosophy 9,
405-26.
Giorgi, A. & F. Pianesi (1991) Toward a Syntax of Temporal Representations, Probus 3,1-27
Hornstein, N. (1990)As Time Goes By: Tense and Universal Grammar, MIT Press Cambridge, Mass.
276
GEORGEJ.XYDOPOULOS
Kayne, R. (1994) The Antisymmetry of Syntax, Camdridge, MIT Press.

Larson, R. (1985) Bare-NP Adverbs. Linguistic Inquiry, 16: 595-621.
Manzini, R. (1994) Syntactic Dependencies and their Properties, UCL Working Papers in Linguistics, 6:
205-17.
Nacas, A. (1987) The Adverbials of Modem Greek., Ph.D. Diss. University of Athens.
Ouhalla, J. (1991) Functional Categories and Parametric Variation, Routledge, London.
Partee, B. (1973) Some Structural Analogies between Tenses and Pronouns in English, Journal of
Philosoph, 70,601-9.
Reichenbach, H. (1947) Elements of Symbolic Logic, Free Press, London (1966).
Stroik, T. (1990) Adverbs as V-sisters, Linguistic Inquiry 21,654-661.
Stroik, T. (1992) On the Distribution of Temporal and Locative NP Adverbials. Linguistic Review 9,
267-84.
Tzartzanos, A. (1946) Modern Greek Syntax (of Common Demotic), Volumes 1 & 2, Kiriakides
Thesaloniki (1989).
Xydopoulos, GJ. (1996) Tense, Aspect, and Adverbials in Modern Greek, PhD Thesis, University
College London.
LIST OF CONTRIBUTORS
ILLC
Department of Philosophy
University of Amsterdam
Nieuwe Doelenstr. 15
1012 CP Maasterdam
The Netherlands
giannakiou@Philo.uva.nl
Y oryia Agouraki
Dept. of Hellenic Studies, Philosophy
and History
Kallipoleos 102100
Aglantzia, Nicosia, Cyprus
gpyoryia@Zeus.cc.ucy.ac.cy
Artemis Alexiadou
ZAS
Jaegerstr. 10/11
10117 Berlin
Germany
artemis@zas.gwz-berlin.de
Michael Hegarty
ILASIL
University of Minnesota
192 Klaeber Court
320 16th Avenue S.E.
Minneapolis MN 55455
USA
hegarOO l@gold.tc.umn.edu
Dept. of Linguistics and Philosophy
MIT
BLDG 20D-219
77 Massachusetts Avenue
Cambridge MA 02139
USA
eleni@mit.edu
&
Grammatica Modellen
KUB de letteren
Postbus 90153/5000
LE Tilburg, The Netherlands
Geoffrey Horrocks
University of Cambridge
st. John's College
Cambrdige CB2 1TP
U.K.
gch1000@cam.ac.uk
Irene Philippaki-Warburton
University of Reading
Department of Linguisticcs
PO Box 218
Reading RG2 2AA UK
I.PhilippakiWarburton@reading.ac.uk
Georgia Catsimali
Department of Linguistics
University of Crete
Rethymno, 74100 Crete
Alexis Dimitriadis
619 Williams Hall
U. Penn, Philadelphia
PA 19104-6305
U.S.A
alexis@unagi.cis.upenn.edu
Angela Ralli
University of Athens
Dept. of French
Panepistimioupoli
15784 Athens
Greece
aralli@atlas.uoa.ariadne-t.gr
277
278
LIST OF CONTRIBUTORS
Anna Roussou
Linguistics Department
Universtiy of Wales at Bangor
Gwynedd LL 59 5 QP UK
a.roussou@bangor.ac.uk
Anna-Marie di Sciullo
Dept. of Linguistics
Universite du Quebec aMontreal
Montreal H3C 3P8
Canada
Melita Stavrou
Dept. of Linguistics
Aristotle University ofThessaloniki
54006 Thessaloniki
Greece
staurou@lit.auth.gr
Arhonto Terzi
T.E.! ofPatras
Dept. of Speech Therapy
Meg. Alexandrou 1
Koukouli, Patras 26334
Greece
aterzi@teipat-gw.teipat.gr
Ianthi-Maria Tsimpli
Dept. of English
54006 Thessaloniki
Greece
George Xydopoulos
P.O. Box 51306
14510 Kifissia
Greece
SUBJECT INDEX
A'-chain 241,242,252
A'-disjointness Requirement 255,256
absoluteness 202, 204, 205, 222
accommodation 76
accusative
inherent 81, 82
oblique 75, 83
structural 75
adjectives 201-225
classifier-like 12,205,216-218,222
evaluative 202, 203, 205-206
object oriented 205,210,216-218,
220
patterns of adjectival modification
202-205
speaker oriented 216,217,218,219,
220
thematic 212, 214, 217
adjunction
left 13,136,140,141-144,146,148
multiple 212, 217, 231
right 55-56, 274-276
adjuncts 2
clauses 72, 119, 165
adverbs
deictic 12,265
temporal 12, 272
agreement
AgrIO 97, 100
AgrO 4, 135-150,250,252-257,
139
AgrS 4, 47,50,51,62, 139, 158
spec-head 2, 115,211
Ancient Greek 108-109
antecedent government 29
antecedent-contained deletion 245
antiadditive 121
antimorphic 121, 123
argument
adjunct asymmetry 29
structure 8-9, 87
aspect 3, 4, 139, 177
aspectual restrictions 53-55
attract-f 137,140,178,182
Balkan Languages 169

bare quantifiers 69, 89
binding
anaphor 74, 102-103
asymmetries 56
backwards 78, 92
condition, strong 242
binding theory 153, 164, 165
principle C 72, 90
Biunique Mapping Principle 269-270
Bulgarian 109-110
Burzio's generalisation 81,82
canonical structural realisation (CSR)
37-38
case
abstract 80
alternations 75
assignment 96, 99-101, 108-109
attraction 159
dependent 81
morphological 80-81
null 155-156
realization, disjunctive hierarchy 81
causative
constructions 67, 83, 86-87
interpretation 83
checking domain 144, 150
classifier
clitic 5-6
A' -dependent 10
climbing 136, 145-146, 150
doubling 6, 68, 75-76,80,91,254,
261
left dislocation (CLLD) 47, 48,54,
62,60,72,73,89,91,234,254,260
movement 135, 136, 148, 150
resumptive 10
voice 109
clitics as feature identifiers 13,241,
253
complementizer deletion 48
CP-recursion 48
complete functional complex 188
complex NP 44-46
279
280
SUBJECT INDEX
compound
deverbaI185-188,192-195
verbal 185-186, 194-195
compounding with adjective
interpretation 204.207,208,211,
217,219,200
computational component 187
conditionals 114, 122-124, 127
counterfactual125-126
conjunction 163
conservativity 115
control 9, 153-167
c-command 118, 121, 129
strong 246, 259
weak 246, 259
Cypriot Greek 227-238
episternic predicates 171-182

Extended Projection Principle (EPP)
46,53,62,81,82,87,158,178
equidistance 138
excorporation 136, 137, 149
existential
presupposition 171
experiencer
argument 67-68, 70-71, 73, 75-85
object fronting 69
expletive 46-48
transitive expletive constructions
46,50-51
replacement 178
external argument 68, 82, 84-86
Familiarity Condition 76
features
N- 51,135,136,137,141-150
V- 135-137, 141-50
First Sister Principle 195
Focus
information 59
operator 55
Phrase 4, 233, 234,
French 47, 140, 142, 144, 145, 187
Full Interpretation 33,241
dative
bare 76, 78
ethical 106
periphrastic 97-108, 111
of possession 105-106
shifted 76
deficient pronouns 250
D-linking 11,250,258,260
DeMorgan relations 121-122
determiner 7
phrase 201, 211- 222,250
directive 124
diversity problem 113, 121
donkey anaphora 53, 117-118
DR-effects 52-53
D-structure 67,81,85-88
double object construction 76, 78, 100101,147-149
Doubly Filled Voice Filter 96
Dutch 57, 122
head movement 136,138, 139, 151

head proximity 206
hierarchy
aspectual84,85
thematic 67, 85, 87-88
Exceptional Case Marking (ECM) 157

Economy of Derivation 136,150
ellipsis 70, 74
embedded interrogatives 23, 25, 37-38,
236,241-243,249,251-254,260
emphatic subject pronoun 71-72, 163
Empty Category Principle (ECP) 243
Icelandic 46,51,53,57, 72
identification requirement 241,243,
258
imperative 3, 141, 146, 148, 150
inalienable possession 105
indefmite 113-119, 123, 125, 126
inference 123
Galician 135,237
genitive 95, 97-103, 105, 107-110
gerund 3, 141-146, 148
SUBJECT INDEX
infmitives 3, 141-146, 169

root23,32-26,40,42,43
inherence 203
Intensional Predicate Calculus 24
internal theta-role saturation 185-186,
190, 192,-19
Irish 50
islands
focus 29, 31, 42
inner 25
necessity 29,31,42
weak 10,25,29,30-32
factives 29
extraposition 29
negation 29
pseudo-opacity 29
wh-29
Italian 48,49,81,89, 115, 140, 142,
143,145,214
Kayne's generalization 6,95,98, 100,
109, 110
light preposition 96, 105, 107-108
Linear Correspondence Axiom 13, 60,
locative prepositions 109, 107, 108,
109, 111
Logical Form (LF) 33, 159, 189,241,
242,248,256
logophors 78
maximal prominence 86
Minimal Link Condition 189
minimizers 118, 125-127
monotone decreasing 113, 121
modal predicate 116, 124
mood phrase 13, 139, 156, 158, 227ff.
Morphological Form 189
Morphological Structure 81
negation 4, 114-130
classical 121
concord 4, 113, 114, 116, 123, 130
Neg-criterion 115,121,131
minimal 121
281
negative polarity items 4,23,26-29,

113-130
strong 127
superstrong 127
weak 127
negative quantifiers 48, 114
regular 121
phrase 4
neo-Reichenbachian 265,276
non-configurational language 2
nonveridicatliy 123, 130-131
noun
raising 210, 212-216, 219, 222
intensifiers 161-162
modifiers 161-162
phrase 7-8
shell 217, 219, 220, 223
n-to-D movement 192
null constants 243
null epithets 243, 246, 247, 253, 255,
257-258
null subject language 2, 45
null operator constructions 10,241-262
numeration 53, 136, 137, 140, 145,
146, 150
oblique accusative 79,80,84
operator
affective
averidical 124-125
conditional 23
empty
movement 76, 91
epistitemic
focus 23, 24
gnomic 23, 27, 28
habitual 24, 27, 28
modal 30
necessity 23, 30, 40, 41
possibility 23
propositional 15, 24-26, 28, 30-36,
37,39
variable chain 241-242, 246-247,
260
veridical 15, 27, 131
282
SUBJECT INDEX
non-veridical 15,27, 116, 123-125

parallel modification 203
parallel structure hypothesis 202
Phonetic Form (PF) 59-61, 144, 145,
150, 189,256,257
pied-piping 76,91
polarity licensing
strong 113, 120, 122, 130
super-strong 122, 127
weak 113, 120, 122
Portuguese 150, 192,237,238
possessives
possessor raising 75, 91
presuppositionality 128
pro
expletive 2, 50, 52-53
referential 2 153-155, 157-158, 160,
161, 165, 167
PRO 9, 153-167
theorem 153
Prominence Condition 75,76, 79
prosodic structure 59
psychological predicates 8, 16,67- 88
purposive clausses 14, 165,243,246
quirky subjects 16,69, 72, 79, 80-82,
84,87-88
raising 62
super raising 181
relative clauses 10-11
appositives 243, 247, 249, 251-254
pseudo 118
restrictives 11, 76, 77,243,247,
279,251-254
relative pronoun 262
Relativized Head Hypothesis 190
Relativized Minimality 29, 31
restriction-scope 120
restrictor 258
right dislocation 258
roofmg 115
Scandinavian object shift 57
scope 201. 203, 208, 209, 216, 218,

220,221,223
scrambling 56-61, 163
Serbo/Croatian 228, 229, 232
selection
c- 36
s- 36
set-theoretical complementation 121
shortest Move 136, 143, 150
Sigma Phrase 135, 232
Spanish 62, 95, 100, 105-106
Speech
spell-out 189, 190-192,227,233,236
strong construal 115, 125-126
subjunctive 3, 4,62,27,32, 116, 122,
124, 130, 131, 156, 167, 169-182,
233,244
subordinators 9
substitution 231
suffIx
derivational 185, 186, 193, 194,
195, 197
inflectional 185, 186, 193, 194, 195,
197
syntactic atom 93, 198
syntactic word 187, 192
Target/subject matter restriction 84
temporal
connectives rule 269
specificity 177
variable 178, 180, 182
Tense 155, 156, 158, 166, 170, 173,
174-176
anaphoric 170
anchoring 170, 178-181
dependence 119
derived structure 268
expletive 177, 182
morphologica1170
phrase 4,50-51, 139,274-276
universal 266
T-raising 178-179, 182
thetic vs. categorical12, 53-55, 120,
128, 129
SUBJECT INDEX
Time
Event 265
Reference 170, 171, 174, 177, 179,
182,265
Speech 170,177,265
topic
drop 243
markers 129
phrase 2, 4, 49
prominent character 55
tough constructions 245, 247
unaccusatives 88, 104
Uniformity of Theta Assignment
Hypothesis (UTAH) 16,88
verb
ditransitive 101
experiencer170
raising 4, 47
V-to-C movement 50,229,237-238
V-to-M 227,232-236
phrase, nested 275
VP internal subject hypothesis 2
voice
accusative 96
c1itic 96, 109
phrase 60, 139
volitional predicates 15, 124
Wackernagel
languages 227, 230
phrase 229
weak construal 115
wh-phrase 23, 24, 47, 49, 243, 349,
250-252, 258, 260
283

Managing Editors
Liliane Haegeman, University of Geneva

Joan Mating, Brandeis University
James McCloskey, University of California, Santa Cruz
Publications
1. L. Burzio: Italian Syntax. A Government-binding Approach. 1986.
ISBN Hb 90-277-2014-2; Pb 90-277-2015-0
2. W.O. Davies: Choctaw Verb Agreement and Universal Grammar. 1986.
ISBN Hb 90-277-2065-7; Pb 90-277-2142-4
3. K. E. Kiss: Configurationality in Hungarian. 1987.
ISBN Hb 90-277-1907-1; Pb 90-277-2456-3
4. D. Pulleyblank: Tone in Lexical Phonology. 1986.
ISBN Hb 90-277-2123-8; Pb 90-277-2124-6
5. L. Hellan and K. K. Christensen: Topics in Scandinavian Syntax. 1986.
ISBN Hb 90-277-2166-1; Pb 90-277-2167-X
6. K. P. Mohanan: The Theory ofLexical Phonology. 1986.
ISBN Hb 90-277-2226-9; Pb 90-277-2227-7
7. J. L. Aissen: Tzotzil Clause Structure. 1987.
ISBN Hb 90-277-2365-6; Pb 90-277-2441-5
8. T. Gunji: Japanese Phrase Structure Grammar. A Unification-based Approach.1987.
ISBN 1-55608-020-4
9. W. U. Wurzel: Inflectional Morphology and Naturalness. 1989
ISBN Hb 1-55608-025-5; Pb 1-55608-026-3
ISBN 1-55608-042-5
10. C. Neidle: The Role of Case in Russian Syntax. 1988
11. C. Lefebvre and P. Muysken: Mixed Categories. Nominalizations in Quechua.
1988.
ISBN Hb 1-55608-050-6; Pb 1-55608-051-4
12. K. Michelson: A Comparative Study ofLake-Iroquoian Accent. 1988
ISBN 1-55608-054-9
13. K. Zagona: Verb Phrase Syntax. A Parametric Study of English and Spanish.
1988
ISBN Hb 1-55608-064-6; Pb 1-55608-065-4
14. R. Hendrick: Anaphora in Celtic and Universal Grammar. 1988
ISBN 1-55608-066-2
15. O. Jaeggli and K.l. Safrr (eds.): The Null Subject Parameter. 1989
ISBN Hb 1-55608-086-7; Pb 1-55608-087-5
16. H. Lasnik: Essays on Anaphora. 1989
ISBN Hb 1-55608-090-5; Pb 1-55608-091-3
17. S. Steele: Agreement and Anti-Agreement. A Syntax of Luisefio. 1990
ISBN 0-7923-0260-5
18. E. Pearce: Parameters in Old French Syntax. Infmitival Complements. 1990
ISBN Hb 0-7923-0432-2; Pb 0-7923-0433-0
19. Y.A. Li: Order and Constituency in Mandarin Chinese. 1990
ISBN 0-7923-0500-0

20. H. Lasnik: Essays on Restrictiveness and Learnability. 1990
ISBN 0-7923-0628-7; Pb 0-7923-0629-5
21. M.J. Speas: Phrase Structure in Natural Language. 1990
ISBN 0-7923-0755-0; Pb 0-7923-0866-2
22. H. Haider and K. Netter (eds.): Representation and Derivation in the Theory of
Grammar. 1991
ISBN 0-7923-1150-7
23. J. Simpson: Warlpiri Morpho-Syntax. A Lexicalist Approach. 1991
ISBN 0-7923-1292-9
24. C. Georgopoulos: Syntactic Variables. Resumptive Pronouns and A' Binding
in Palauan. 1991
ISBN 0-7923-1293-7
25. K. Leffel and D. Bouchard (eds.): Views on Phrase Structure. 1991
ISBN 0-7923-1295-3
26. C. Tellier: Licensing Theory and French Parasitic Gaps. 1991
ISBN 0-7923-1311-9; Pb 0-7923-1323-2
27. S.-Y. Kuroda: Japanese Syntax and Semantics. Collected Papers. 1992
ISBN 0-7923-1390-9; Pb 0-7923-1391-7
28. I. Roberts: Verbs and Diachronic Syntax. A Comparative History of English
and French. 1992
ISBN 0-7923-1705-X
29. A. Fassi Fehri: Issues in the Structure ofArabic Clauses and Words. 1993
ISBN 0-7923-2082-4
ISBN 0-7923-2649-0
30. M. Bittner: Case, Scope, and Binding. 1994
31. H. Haider, S. Olsen and S. Vikner (eds.): Studies in Comparative Germanic
ISBN 0-7923-3280-6
Syntax. 1995
32. N. Duffield: Particles and Projections in Irish Syntax. 1995
ISBN 0-7923-3550-3; Pb 0-7923-3674-7
33. J. Rooryck and L. Zaring (eds.): Phrase Structure and the Lexicon. 1996
ISBN 0-7923-3745-X
34. J. Bayer: Directionality and Logical Form. On the Scope of Focusing Particles
and Wh-in-situ. 1996
ISBN 0-7923-3752-2
35. R. Freidin (ed.): Current Issues in Comparative Grammar. 1996
ISBN 0-7923-3778-6; Pb 0-7923-3779-4
36. C.-TJ. Huang and Y.-H.A. Li (eds.): New Horizons in Chinese Linguistics.
1996
ISBN 0-7923-3867-7; Pb 0-7923-3868-5
37. A. Watanabe: Case Absorption and WH-Agreement. 1996
ISBN 0-7923-4203-8
38. H. Thrainsson, S.D. Epstein and S. Peter (eds.): Studies in Comparative
ISBN 0-7923-4215-1
Germanic Syntax. Volume 11.1996
39. CJ.W. Zwart: Morphosyntax of Verb Movement. A Minimalist Approach to
the Syntax of Dutch. 1997
ISBN 0-7923-4263-1; Pb 0-7923-4264-X
40. T. SHoni: Noun Phrases and Nominalizations. The Syntax of DPs. 1997
ISBN 0-7923-4608-4
41. B.S. Vance: Syntactic Change in Medieval French. 1997 ISBN 0-7923-4669-6
42. G. Muller: Incomplete Category Fronting. A Derivational Approach to
Remnant Movement in German. 1998
ISBN 0-7923-4837-0

Studies in Greek Syntax

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Studies in Greek Syntax

Uploaded by

Copyright:

Available Formats

STUDIES IN GREEK SYNTAX

Studies in Natural Language and Linguistic Theory

SPRINGER-SCIENCE+BUSINESS MEDIA. B.V.

Printed on acid-free paper

Ali Rights Reserved

Artemis Alexiadou Geoffrey Horrocks

A. Alexiadou et al. (eds.), Studies in Greek Syntax, 1-21.

ARTEMIS ALEXIADOU, GEOFFREY HORROCKS AND MELITA STAVROU

1.1 Word Order and Morphology

aghap-a-me 'we love'

ARTEMIS ALEXIADOU, GEOFFREY HORROCKS AND MELITA STAVROU

ARTEMIS ALEXIADOU, GEOFFREY HORROCKS AND MELITA STAVROU

In the GB literature on clitic-doubling, it is often assumed that the clitic 'absorbs'

Kayne's generalisation over the other characteristic properties of clitic

1.3 The Noun Phrase

ARTEMIS ALEXIADOU, GEOFFREY HORROCKS AND MELITA STAVROU

indeed by any indefinite pronoun, or through zero determination. Adjectives always

In Greek, much as in Arabic, Hebrew and Scandinavian languages, defInite nouns

1.4 Argument Structure

of minimalism which dispenses with the levels of Deep and Surface-structure, it is

Complement clauses are introduced by one of the complementizers/subordinators

An issue of controversy among linguists is whether or not na- clauses involve

ARTEMIS ALEXIADOU, GEOFFREY HORROCKS AND MELITA STAVROU

enas anthropos ton opio/pu

*0 andhras pu ton idha

In general, in Greek as well as in other languages, A'-dependent clitics (the term is

ARTEMIS ALEXIADOU, GEOFFREY HORROCKS AND MELITA STAVROU

CLAUSE AND DP STRUCTURE: ALEXIADOU, STAVROU,

syntactic structure (the former involving linear iteration in a single position).

3. CLITICS IN STANDARD GR AND DIALECTS: HEGARTY, TERZI,

ARTEMIS ALEXIADOU, GEOFFREY HORROCKS AND MELITA STAVROU

4. NEGATIVE POLARITY ITEMS AND SENTENTIAL OPERATORS:

ARTEMIS ALEXlADOU, GEOFFREY HORROCKS AND MELITA STAVROU

Giannakidou's article, which helps our understanding of the licensing of negative

I wonder [whether to go]

The paper presents a number of counterarguments to Giannakidou's with respect to

5. CONTROL AND NON-FINITE CLAUSES REVISITED: PHILIPPAKIWARBURTON & CATSIMALI

6. THEMATIC STRUCTURE: ANAGNOSTOPOULOU, DI SCIULLO &

ARTEMIS ALEXIADOU, GEOFFREY HORROCKS AND MELITA STAVROU

ARTEMIS ALEXIADOU, GEOFFREY HORROCKS AND MELITA STAVROU

(1) NEG, Q, F, CO, NO and PO form a natural class.

NEG, Q, F, CO, NO and PO modify the proposition.

In virtue of modifying the proposition NEG, Q, F, CO, NO and PO will be hence

PROPOSITIONAL OPERA TORS

Propositional operators are semantic higher predicates. Nonpropositional operators

3. POLARITY ITEM LICENSING

EPI licensing in Gr is a case where there is full correspondence between a PI

Non-veridical operators license EPl's in Gr

among EPI licensing contexts. These predicates are taken to be non-veridical

The same inaccurate, in my view, inclusion of predicates that take na -clauses

The preceding discussion on whether na-clauses are EPI licensers was

a. *Howi do you wonder whether he repaired the car ti?

Give up after all these efforts?

Propositional Operators select root infInitives

Propositional operators obtain in embedded wh-infmitivals

Propositional operators select (otherwise unselected) infinitiv.es

Languages have a ban against unselected infmitives

I will next return to the discussion of property (28'). There is an important

PROPOSITIONAL OPERA TORS

NEG, Q, F, NO and PO license embedded interrogatives 25

PROPOSITIONAL OPERA TORS

in (38a-e) with then be a consequence of the interrogative force on the wh-phrases

NEG, Q, F, CO, NO and PO license PI's.

NEG, Q, F, CO, NO and PO: