Documente Academic
Documente Profesional
Documente Cultură
Series Editor
Werner Abraham
University of Vienna
Volume 49
Theoretical Approaches to Universals
Edited by Artemis Alexiadou
Theoretical Approaches
to Universals
Edited by
Artemis Alexiadou
University of Potsdam
TM
The paper used in this publication meets the minimum requirements of American
National Standard for Information Sciences Permanence of Paper for Printed
Library Materials, ansi z39.48-1984.
2002021464
Table of contents
List of contributors
Introduction
Artemis Alexiadou
Universal features and language-particular morphemes
Maya Arad
Agree or attract? A relativized minimality solution to a proper
binding condition puzzle
Cedric Boeckx
Distributed deletion
vii
1
15
41
65
109
139
165
211
237
259
Index
315
List of contributors
Artemis Alexiadou
University of Potsdam
Institute of Linguistics
Postfach 601553
14415 Potsdam, Germany
artemis@ling.uni-potsdam.de
Maya Arad
University of Geneva
Dept. of Linguistics
2, rue de Candolle
CH - 1211 Genve 4
arad@lettres.unige.ch
Cedric Boeckx
Department of Linguistics
4088 Foreign Language Building
707 South Mathews Avenue, MC-168
University of Illinois at
Urbana-Champaign
Urbana, IL 61801, USA
cboeckx@ux1.cso.uiuc.edu
Damir Cavar
Dresdner Bank
CC IT, enateg
Research & Innovations
60301 Frankfurt a.M., Germany
damir.cavar@t-online.de
Gisbert Fanselow
University of Potsdam
Institute of Linguistics
Postfach 601553
4415 Potsdam, Germany
fanselow@rz.uni-potsdam.de
Robert Frank
Department of Cognitive Science
Johns Hopkins University
243 Krieger Hall
3400 N. Charles St. Baltimore,
MD 21218-2685, USA
rfrank@vonneumann.cog.jhu.edu
Paul Hagstrom
Department of Modern Foreign
Languages & Literatures
Boston University
718 Commonwealth Ave.
Boston, MA 02215, USA
hagstrom@bu.edu
Murat Kural
University of California, Irvine
Department of Linquistics
3151 Social Science Plaza
Irvine, CA 92697-5100, USA
mkural@earthlink.net
Luis Lpez
University of Illinois-Chicago
Dept. of Spanish, French, Italian and
Portuguese
College of Liberal Arts and Sciences
601 South Morgan St.
Chicago, IL 60607-7117, USA
luislope@uic.edu
Jaume Mateu
Departament de Filologia Catalana
Facultat de Filosofia i Lletres Edifici B
Universitat Autnoma de Barcelona
E-08193 Bellaterra, Spain
Jaume.Mateu@uab.es
List of contributors
Gemma Rigau
Departament de Filologia Catalana
Facultat de Filosofia i Lletres Edifici B
Universitat Autnoma de Barcelona
E-08193 Bellaterra, Spain
Gemma.Rigau@uab.es
Joachim Sabel
ZAS
Jgerstr. 10-11
10117 Berlin
Germany
sabel@zas.gwz-berlin.de
Juan Romero
Dept. de Filologa Espaola
Universidad Autnoma de Madrid
28049 Madrid, Spain
juan.romero@uah.es
Vijay K. Shanker
Department of Computer and
Information Science
University of Delaware
Newark, Delaware 19716, USA
vijay@cis.udel.edu
Introduction
Artemis Alexiadou
University of Potsdam
The present volume has its origin in the GLOW conference on Universals organized by the Research Center for General Linguistics (ZAS, Berlin), the Linguistics Department of the University of Potsdam and the Dutch Graduate School
in Linguistics (LOT) and hosted in Berlin in March 1999.1 In the first part of
this introduction, I offer a brief overview of the main issues involved in our understanding of and quest for universals, by presenting some points of controversy concerning the proper characterization of universal and language specific
properties. In the second part, I summarize the contributions to this volume.
The search for universals has always been at the center of interest in linguistic theory. Two main approaches can be recognized. On the one hand, work
by Greenberg (1966), Comrie (1981), Croft (1990) and others searches for surface properties that would be common to all languages and attempts to identify
patterns of regularities. For the generative linguist, on the other hand, fundamental claims about universal properties of language are build into the very
architecture of the theory of Universal Grammar (UG) in various forms as we
will see below. Alongside formal universals, generative linguists also seek substantive universals in inventories, markedness patterns, feature hierarchies etc
of the type explored in Greenbergs work. As Croft (1990) points out, both
approaches deal with the question what is a possible language? and believe
that there are universal constraints that define the answer to this question.
Moreover, both believe that the answer to this question is reached at by the
comparative study of language.
Artemis Alexiadou
SVO
OVS
The basic word order should be the one most frequently appearing in a language. English is unambiguously SVO, Turkish, and Japanese are SOV (cf. 3),
Welsh is VSO, Malagasy is VOS. But some Australian languages such as Warlbiri
permit all possible permutations (cf. 4):
(3) Ken-ga
Naomi-wo miru Japanese
Ken-Nom Naomi-acc sees
Ken sees Naomi
(4) a.
Introduction
c.
Greenberg (1966) took a sample of about thirty languages from different families and different parts of the world. He observed that although there was considerable variation in word order, the variation was structured in the sense that
certain properties varied together. Greenberg put forth several universals that
capture this structured variation. These are of the type in (5):
(5) With overwhelmingly greater than chance frequency, languages with normal SOV are postpositional.
(5) simply states that languages with normal SOV order are in their majority postpositional. In fact the order of V and O can be taken to be
the central parameter, which determines the serialization of the language of
modifier(operator)-head (operand) or head-modifier, as shown in (6).
(6) SOV
AN
GN
DetN
RelN
Post
SVO
NA
NG
NDet
NRel
Prep
where O = object, V = verb, N = noun, A = adjective, G = genitive, Det = determiner (article), Rel = relative clause, Post = postpositions, Prep = prepositions.
After Chomsky (1986), the difference between SVO and SOV languages
such as e.g. English and Japanese was seen as the result of a difference in the
values of a parameter of X -theory. OV and VO patterns result from choosing
different values for the head-parameter that regulates the position of head categories in relation to their complements. This parameter can take two values:
head first and head last, each accounting for the two patterns of word order,
head complement and complement head found in languages such as English
and Japanese. In other words OV arises when a language opts for follows in
(7b) and VP when a language opts for precedes.
(7) a. X follows its specifier
b. X precedes its complement
The hypothesis here is that when learning a language the child has to determine
the word order. Word order variation is seen as a parameter according to which
languages vary. The parameter can have a limited set of values, i.e. either VO
or OV. Sentences like the ones in (1) provide evidence to the English child that
Artemis Alexiadou
in her language the verb precedes the object, while (3) provides evidence to the
Japanese child that in her language the object precedes the verb.
The question that arises is whether parameters such as those can handle
all the results of typological work. Consider again (5). With the same token,
we take OP patterns to arise when a language opts for follows in (7b) and
PO when a language opts for precedes. On this view, one deep property, the
choice of precede or follow in (7b), is responsible for the derivation of several
other properties. As (6) suggests, OV languages tend to have PossN and AN
orders. Roberts (1997), however, points out that these conclusions are slightly
tricky for the principles and parameters approach: possessors and adjectives
are taken to occupy specifiers within the DP/NP. In principle they should be
independent of the ordering patterns we find within X .
On the view just presented, parameters are macroparameters, that is they
constitute a small set of binary-valued parameters with very far reaching consequences (see Baker 1996 for a recent discussion in defense of this view). However, more recent work in generative syntax has re-examined the status of such
parameters, and the components of grammar these operate in. For instance,
Kayne (1994) proposes that the head parameter can be dispensed with. In order
to derive OV patterns from a universal SVO order, one needs to make use of extensive movement operations. In this respect, Kaynes proposal for a universal
ordering merely shifts the burden from phrase structure to movement. Others
pursue the idea that macroparameters can be dispensed with and that linguistic variation is best understood in terms of microparameters. Microparameters
are local, low-level phenomena which can partially obscure macroparametric
variation.
The view that parametric variation affects only the inflectional system of
languages, as proposed in Borer (1983), has been recently receiving a lot of
attention among generative linguists. On this view, parameters are associated
with individual lexical items. As a result, the structure-building apparatus is
simple, unified and universal. Well-formedness conditions on phrase-structure
such as X theory are eliminated in favor of a system that incorporates the operations Merge and Move (Chomsky 1995). Fundamental claims about universal
properties of language are build into the very architecture of the theory of UG
in the form of e.g. primitives (features), combinatorial operations (Merge), the
operation Move, interfaces with extra-linguistic systems (LF, PF), and so forth.
Language variation is a reflex of the interaction of language specific properties
of lexical items, and in particular of morpho-lexical features, operating in the
structure building apparatus seeking to satisfy the requirements imposed by
the interface (see Arad; Mateu & Rigau).
Introduction
Several issues arise. First, how are the primitive notions of the structure
building apparatus, Merge, Move, Agree or Attract defined? How do these apply, and what are the properties they are sensitive to? Second, what features are
relevant for Agree or trigger overt displacement? What does a typology of features look like? Third, how are we to understand variation in morpho-lexical
features exactly? Finally, is it true that morpho-lexical variation dispenses with
the need for structural variation?
All these questions are taken up in the contributions to this volume. In
what follows I briefly turn to some of the issues concerning clause structure
and universal properties that have preoccupied the recent literature. What has
become a more or less standard view by now is to assume that functional morphemes occupy different syntactic slots in the structural representation of the
(verbal and nominal) clause. A first issue concerns the types of features that we
can assume to project in the functional domain. According to some authors,
the types of features that are present in the syntactic terminals are those that
are relevant for semantic interpretation at LF, as in (8) (see Halle & Marantz
1993; Chomsky 1995; Embick 1997). The underlying assumption is that there
is a universal set of features, and each language draws from that pool. Languages will differ as to whether they will realize the feature at all, and whether
they will realize it by means of an auxiliary, an affix, a particle and so on (see
Arad; Romero for further discussion).
(8) Tense Neg
Aspect Force
Mood Number etc.
Artemis Alexiadou
On such views, the overt morphological instantiation of some feature, i.e. rich
number or gender or even tense inflectional morphology, is correlated with
the presence of overt movement. This correlation between syntactic movement
and the presence of overt morphology could be taken to be a principle of UG
(see the discussion in and the contributions to Haegeman 1997). However, as
Snyder (1995) points out, it is not clear whether there is a principled reason to
expect the particular feature combinations distinguished by a given morphological paradigm to have direct consequences for language specific properties
of syntax. One could imagine that the implications are completely the reverse.
A related point of controversy concerns the number and the order of the
functional projections associated with these features. Some authors take the
number and the order of these projections to be universal, see Chomsky (1995),
and Cinque (1999), for an elaborated CP-IP domain, shown in (12):
(11) [TP [Asp [VoiceP [ V]]]
(12) [Mood speech act [Mood evaluative [Mood evidential [Mood epistemic
[T (Past) [T (Future) [(Mood irrealis) [Mood necessity [Mood possibility
[Mood volitional [Mood obligation [Mood ability/permission [Asp habitual [Asp repetitive [Asp frequentative I [Asp celerative [T (anterior)
[Asp terminative [Asp continuative [Asp perfect? [Asp retrospective [Asp
proximative [Asp durative [(?) Asp generic/progressive [prospective [Asp
sg Completive I [Asp Pl Completive I [Voice [Asp Celerative II [Asp sg
Completive II [Asp repetitive II [Asp frequentative II
Introduction
But, others take the order and number of functional projections to be subject to cross-linguistic variation. For instance, Ouhalla (1991) argues that the
order of functional projections is parameterized. Iatridou (1990) and Thrinsson (1995) among others argue that one should assume only those functional
projections one has evidence for in a given language. On this view, languages
may vary as to whether they have a pre-pollockian unsplit IP or an IP containing an Agreement phrases distinct from Tense (the so called Split Infl Parameter). In particular, Bobaljik and Thrinsson (1998) argue that there are a
series of straightforward consequences of assuming such a parameter, both for
the syntax and for the morphology, namely there are more specifier positions
in (13b) than in (13a), there are non local relations among Infl-type heads
in (13b), and there are more terminal nodes in (13b) than in (13a). On this
view, on might expect two VP external subject positions (the specifiers of AgrP
and TP) and perhaps a VP external DP object position in languages that have
structure (13b), but not in those that have (13a).
(13) a.
b. AgrP
IP
Agr'
I'
I
VP
Agr
TP
T'
T
AgrP
Agr'
Agr
VP
The authors argue that this is the correct interpretation of multiple subject
position and object shift phenomena in Icelandic as opposed to Mainland
Scandinavian and English. Hence Icelandic licenses Spec, TP as an intermediate subject position, allows for object shift, and exhibits transitive expletive
constructions (TECs). On the other hand, languages such as English, lack object shift (OS), transitive expletive constructions and do not license Spec, TP as
a further subject position (14) (see Lpezs contribution).
(14) a.
TEC
Artemis Alexiadou
OS
. The papers
The contributions to this volume all attempt to identify universal properties of
the language faculty, as well as the source of cross-linguistic variation. Some
of the articles pay particular attention to the organization of the grammar, the
type of operations that are effective, the role of features in determining variation, and primitive notions of phrase-structure. Others show how structural
differences capture semantic and morphological differences within a language
and across languages, and how these are ultimately responsible for variation.
In sum, the papers in this volume are concerned both with formal as well as
with substantive universals. I turn to a brief summary below.
Arad argues that there are three sources for language variation: the inventory of roots the language has, the features it has selected, and the way these
features are bundled together. In particular, Arad following Marantz (1997)
and Halle and Marantz (1993), views the Lexicon of a language as a set of roots
and (possibly bundled) features. There are three ways in which languages vary:
Introduction
(i) Root inventory, i.e. signs or lexical pieces available in a language. Variation
in roots includes also the variation in contextual meanings that roots can be assigned. (ii) Subset of features selected from the universal pool: i.e. the features
a language uses in building its lexicon. More precisely, does it have morphological case features, gender features, aspect (perfective/imperfective) etc? (iii)
The ways in which features are bundled together. Languages may put together
different sets of features into morphemes. This allows for a further source of
lexical variation: the bundles of features available in a language. Specifically,
Arad argues that the functional head little v comes in two types. Generalizing
the case of little v, she further argues that verbal heads in general are features
bundles. Languages may thus have different verbal morphemes, i.e. different
feature bundles, and the same root can form different types of verbs when
combining with verbal morphemes of different types. The empirical discussion
of verb creating morphemes concentrates on psychological verbs.
Boeckxs contribution is concerned with a critical evaluation of the organization of the grammar, and the movement operations (Agree vs. Attract
F(eature)) that could be effective. In particular he shows that Lasniks argumentation of favor of Attract F and against Agree can be dealt with within a
One-cycle model, as the one put forth in Chomsky (1999), hence Agree is superior to Attract F. A further question that Boeckx addresses is whether the
One-cycle model is a notational variant of the Single output model, proposed
in Bobaljik (1995), Groat and ONeil (1995) among others. Boeckx outlines
empirical arguments based on the how likely paradigm that Lasniks analysis
is inadequate. He offers an alternative that implicates Relativized Minimality in
support of the One-cycle model and the Agree operation.
Fanselow and Cavars contribution deals with XP-split constructions in
German and Slavic languages. These splits have the following properties: they
arise in the context of operator movement only. XP-splits can retain or invert
the order of the elements found in the continuous counterpart. The latter type
of split cannot show up with PPs it is replaced by a construction that differs from XP-splits only in the presence of copies of the preposition in all slots
where parts of the PP appear. Pull splits do not show up for all types of operator movement in German. The authors point out that movement analyses
of such splits face serious problems with respect to syntactic islands and the
phonetic shape of the parts of the split phrase (the regeneration problem discovered by van Riemsdijk). They argue that these problems render a simple
movement analysis of the XP-split construction impossible. But these patterns
are not amenable to a treatment in which both parts are base-generated in situ,
either. The authors propose that the copy & deletion (CD) approach to move-
Artemis Alexiadou
ment (Chomsky 1995) offers a way to account for the paradox. If the CD is
implemented in such a way that the deletion operation following the copying
step of movement may affect both copies. The CD-approach offers a unified
analysis for both DP and PP splits.
Frank, Hagstrom, and Vijay-Shankers article is concerned with the proper
characterization of grammatical structures, and the primitive notion in the determining the properties of tree structures. They demonstrate that there is no
way for grammar to refer to dominance. More specifically, dominance does not
figure into grammatical explanation. The authors show that notions for the
definition of which dominance has been used were can be translated into statements about c-command. Two concepts are considered: roots and constituents.
The authors show how these can be described in terms of c-command. The ccommand based view on roots distinguishes between the categorial root and
the site of cyclic attachment. Thus no need to refer to the dominance relation is necessary in order to determine the root of a tree structure. The ccommand based definition of constituents further enables an understanding
of cases where movement is blocked.
Kurals paper deals with certain problems that emerge from the two-way
classification of monadic verbs as unaccusative and unergative verbs, as proposed in e.g. Perlmutter (1978) and Burzio (1986). Kural argues that (a) the
tests used to distinguish between the two classes do not all test the same structural properties, and (b) the discrepancies in the behavior of some monadic
verbs across these tests can be explained naturally by positing a four-way classification rather than the traditional two-way classification. Kural points out
that some of the tests, e.g. there insertion make reference to some VP external
position, while others e.g. resultatives, cognate objects, relate to the VP internal base position of the object. The four classes are: verbs of being, change of
state verbs, change of location verbs, and finally verbs of creation. Each class
is associated with a distinct structure. The syntactic behavior of these verbs
becomes more transparent, but it is also shown that the classification of these
verbs along syntactic lines fully coincides with their broad semantic properties
such as denoting a change of location or creation of an abstract entity.
Lpez proposes a new look at the operations Agree and Move. He argues
that (i) the operation Agree is strictly local, and (ii) the operation Move is triggered by the instability created in the system by unvalued features (following
similar ideas in Frampton & Gutmann 1999). The paper introduces the concept
of co-valued features: two terms with unvalued features of the same type that
are related by the operation Agree must have their features valued in tandem.
On the conceptual side, these alterations allow us to revisit and eliminate some
Introduction
Artemis Alexiadou
The CAM predicts that the only existing intermediate traces of a moved element are traces in specifier positions. The empirical evidence against intermediate adjunction is formulated with respect to different movement types,
such as wh-movement, empty-operator-movement, A-movement, extraposition, quantifier raising, scrambling, and head movement. Furthermore, data
that were traditionally used as providing evidence for intermediate adjunction have been explained as involving movement via a second specifier position. For example, the analysis of scrambling in German and Japanese rests on
the assumption that Japanese allows for multiple A-specifiers whereas German
does not. On the other hand, multiple (CP) A -specifiers seem to be the unmarked case in languages as argued in connection with the proposed analysis
of extraction out of wh-islands.
Acknowledgements
I wish to thank the Zentrum fr Allgemeine Sprachwissenschaft in Berlin,
the Linguistics Department of the University of Potsdam, the Dutch graduate school OTS, and the DFG for their support in organizing GLOW in Berlin
in March 1999. I thank the authors of this volume for their co-operation, as
well as the participants of the conference. I am grateful to Werner Abraham
and Kees Vaes for their assistance in the preparation of this volume.
Notes
. The papers by Fanselow & Cavar and Boeckx were not presented at that event.
References
Baker, M. (1996). The Polysynthesis Parameter. Oxford: Oxford University Press.
Bernstein, J. (1993). Topics in the Syntax of Nominal Structure across Romance. Ph.D. Diss.,
CUNY.
Bobaljik, J. (1995). Morphosyntax: The Syntax of Verbal Inflection. Ph.D. Diss., MIT.
Bobaljik, J., & H. Thransson (1998). Two heads arent always better than one. Syntax, 1,
3771.
Borer, H. (1983). Parametric Syntax: Case studies in Semitic and Romance Languages. Foris:
Dordrecht.
Burzio, L. (1986). Italian Syntax: A Government and Binding Approach. Kluwer.
Introduction
v
n
(1)
fish: a.
b.
v
fish
to fish (v)
fish
a fish (n)
The root fish is the lexical and phonological core shared by the noun a fish
and the verbs to fish. Speakers thus have access to words complex entities built
of roots and features but not to the roots themselves. Perhaps for that reason,
Maya Arad
the idea that in all languages the (category neutral) root is distinct from the
word creating morphology may not seem straightforward. Indeed in English,
where word-creating morphology is non-obligatory, the root is not morphologically distinct from the nominal or verbal morpheme. However, other languages (e.g. Romance), always have some overt verbal morphology on the verb
stem. In Semitic languages, going one step further, the root is mostly easily distinguishable from the word creating morphology: on their own, roots units of
three consonants are neither pronounceable, nor belong to any grammatical
category. Only when put into word creating morphology (known as patterns),
roots become nouns, verbs or adjectives. Consider the following example from
Hebrew:
(2)
lmd lamad (learn, v, pattern CaCaC)
talmid (student, n, pattern taCCiC)
limudi (pertains to learning, adj., pattern CiCCuCi)
Consider, next, the second participant in word formation features. I will rely
here on two complementary assumptions:
1. UG makes available a universal set of features (Chomsky 1998).
2. Languages may select a subset of these features. In particular, a language
may bundle together some of these features into morphemes, i.e. featurebundles (cf. Marantz 1999).
While the first claim is a conceptual necessity, the second claim could be subject to debate: do all languages have all possible features? I will assume here
that languages select a subset of the feature inventory offered by UG. For example, Russian has instrumental case features and grammatical Gender features, while English does not. Similarly, English, but not Russian, has progressive marking for verbs. So here is an important source for language variation:
which morpho-syntactic features has the language selected out of the universal pool? Suppose, furthermore, that language may bundle together different
properties. If this is the case, then here is another source for variation: two languages with the same set of features could bundle these features in different
ways, thus ending up with a different set of morphemes.
Following the essence of Borers (1984) proposal, I argue that language
variation is restricted to the lexical and morphological component, and has
three sources: first, languages differ as to their inventory of roots (namely, what
signs a language has). Second, language may select different features out of
the universal pool made available by UG. Finally, different languages may bundle features in different ways, thus having different morphemes. I will illus-
trate this with a case study of verb creating morphemes. My starting point is
a specific verbal head, little v (cf. Chomsky 1995, 1998). Following Marantz
(1999), I claim that this head bundles together two sets of features: semantic
contents (giving the event an agentive interpretation) and transitivity properties (case checking properties). These two features are separable. They may
be bundled with other features, into different verbal morphemes. Generalizing the case of little v, I will argue that verbal heads in general are features
bundles. Languages may thus have different verbal morphemes, i.e. different
feature bundles, and the same root can form different types of verbs when
combining with verbal morphemes of different types.
In my discussion of verb creating morphemes I will concentrate on psychological verbs. This is because a sub-group of these verbs alternates between
an agentive and a non-agentive reading, which enables us to see the interaction
of a single root with two different verbal morphemes. Furthermore, I argue
that in many cases both types of psych verbs Subject Experiencer and Object
Experiencer are formed from the same root, combined with different verbal
morphemes. So, because of their special properties, these verbs make a good
case study of verbs and features bundles. But before getting to psych verbs, a
few words on the best known verbal morpheme, little v.
. Little v
Little v (Chomsky 1995, 1998; also Collinss 1997, Transitivity Phrase) is commonly taken to be the upper head in the VP-shell, and is often referred to as a
transitivity head, which heads transitive constructions. This head introduces
an external argument in its specifier and enters into a relation with the object
(Agree, or checks structural Case):2
(3)
vP
external argument
VP
V
Object
Maya Arad
(4) a.
yv + CaCaC
= yaav (be seated, stative)4
b.
yv + hiCCiC
= hoiv (make someone sit down, causative)5
c.
yv + hitCaCeC = hityaev (sit down, inchoative)
(5) a.
rgz + CaCaC
= ragaz (be angry, stative)
b.
rgz + hiCCiC
= hirgiz (anger, causative)
c.
rgz + hitCaCeC = hitragez (get angry, inchoative)
(Hebrew)
A single root can form different types of verbs: stative, causative or inchoative.
These verbs share something (in (4), for example, they all refer to an event
of sitting), but they also differ: one refers to a stative event, being at a state
of sitting, another, to a causative event making someone sit down, and yet
another to an inchoative event of change of state: sit down. How can one root
form several verbs? My hypothesis is that the root is combined in each case
with a verbal morpheme of a different type or different semantic flavor, thus
forming different verbs:
(6) a.
root + Va = stative.
b.
root + Vb = causative.
c.
root + Vc = inchoative.
The hypothesis that verbal heads have different semantic flavors raises (at least)
two questions:
1. What types of verbal morphemes exist in natural language?
2. What are the contents of verbal morphemes?
In what follows, I will examine these questions through the specific case of a
class of verbs that alternate between an agentive and a non-agentive reading:
Object Experiencer verbs. I will argue that the two readings are achieved by
combining the same root with different verbal morphemes, one agentive and
one stative.
The starting point of my discussion is Belletti and Rizzis (1988) seminal work
on Italian psych verbs. Belletti and Rizzi (henceforth B&R) show that although
ObjExp verbs seem identical to standard transitive verbs, they differ from them
syntactically in a number of ways. Qualifying B&Rs typology further, I will
argue the following:
1. ObjExp verbs such as frighten can have a stative reading or an agentive
reading. This ambiguity is pervasive through the class of ObjExp verbs:
their subjects can be interpreted as either agentive or non-agentive.6
2. The two readings of ObjExp verbs differ syntactically: only the stative reading has the syntactic effects noted by B&R. On the agentive reading, the
verb behaves like a standard transitive verb.
3. The stative and the agentive readings of ObjExp verbs are formed by the
same root (e.g. fright). The difference between them arises from the type
of verbal morpheme with which the root is combined. On the agentive
reading the root is combined with (standard) little v. On the stative reading
it is combined with a verbal head which is stative and causative (in a sense
to be made explicit below). I call this head stative little v.
Maya Arad
On this reading, the agent in this case Anna intends to bring about a state of
fright in the experiencer. The experiencer, Laura, undergoes a change of state
and becomes frightened. In contrast, the stative reading has neither an agent
nor a change of mental state in the object.7 For example, in (9a) below there is
no single point in time in which Laura turns from unconcerned into concerned. Rather, perception of the problem by Laura triggers a concomitant
state of concern. When she happens to think of the problem, she experiences a
spell of concern (cf. Pylkknen 1997):
(9) a. This problem concerned Laura.
b. Anna/Annas behavior frightens Laura.
c. Blood sausage disgusts Laura.8
Several points have to be noted about this reading. First, since there is no agent
on the stative reading, the triggering of the mental state by the stimulus is neither volitional, nor under the stimulus control. Even when the subject is human, he or she do not act on purpose: in (9b) it is something about Anna that
frightens Laura, rather than Annas attempts to frighten. Second, there is no
change of state in the experiencer.9 The stative reading only asserts that the
experiencer is at a specific mental state as long as she perceives the stimulus
(or has it on her mind). Another difference concerns the relationship between
the entity that brings about the mental state and the mental state itself. On the
agentive readings the agent only brings about the resulting state, which holds
independently, and is not part of the event of mental state (cf. Pesetsky 1995).
On the stative reading the stimulus has to co-occur with the mental state in
order for it to hold: the experiencer is at a specific mental state for as long as it
perceives the stimulus (cf. Pylkknen 1997). The stimulus is thus an inherent
part of the event of mental state. The schema in (10) describes the difference
between the two readings:
Finally, note that both readings are causative.10 Following Pylkknen (1997),
I assume that causation can be active or stative. Active causation involves an
agent, who acts and brings about a change of state, while stative causation involves a stative causer (or a stimulus) which triggers a state (whose existence is
co-extensive with that of the stimulus).
. The syntactic realization of the stative and the agentive readings
So far we were concerned with the semantic properties of ObjExp verbs. However, these verbs are also known for exhibiting some syntactic peculiarities. In
this section I argue that such peculiarities only occur with the stative reading
of ObjExp verbs. When the subject is interpreted as agentive, ObjExp verbs
lose their psych properties. I will illustrate this on the properties pointed out
by Belletti and Rizzi (1988) for Italian reflexive clitics, causativization and
extraction from the object but this observation holds across a number of
languages (English, Hebrew, Spanish, Greek see Arad (1998) for a detailed
account).
Consider reflexivization first. As pointed out by B&R, ObjExp verbs cannot appear with a reflexive clitic si in Italian (11a). However, (11a) is crucially
interpreted as a stative reading of frighten. If an agentive reading is forced (11b)
the sentence is fine:11
(11) a.
??Gianni
si spaventa.
Gianni self frightens
b. Gli studenti si spaventano prima degli esami
the students self frighten
before the exams
per indursi
a studiare di pi
to urge-refl to study
more
The students frighten themselves before exams in order to urge themselves to study harder.
Next, B&R show that embedding an ObjExp verb under the causative construction in Italian is ungrammatical. However, if a minimal pair of stative and agen-
Maya Arad
Finally, B&R show that extraction from the object is ungrammatical with
ObjExp verbs (13a). Again, on the agentive reading, extraction is grammatical (13b):
(13) a. *La ragazza
the girl
b. La ragazza
the girl
di cui
of which
di cui
of which
perch gliela
facessero
sposare.
for
him-dat-her-acc make (3rd pl) marry
The girl whose parents G. frightens so that they will allow him to
marry her.
el nio/la musica le
molest.
the boy/the music her-dat bothered
The boy/the music bothered her.
b. el nio/* la musica la
molest.
the boy
the music her-acc bothered
The boy/*the music bothered her.
c. Lo hice para
molestarla/lo
(*le).
it I did in order to bother her/him (acc) (dat)
I did it in order to bother him/her.
(stative)
(agentive)
Speakers interpret (14a) as non-agentive: something about the boy or the music
triggered a mental state of bothering. (14b), on the other hand, is interpreted
as unambiguously agentive: the boy intended to bother (and the music can-
b. frighten (stative)
/fright/
mental state
causative
stative
stimulus
The similarities between the two readings seem related to the root they share.
The differences between them in particular having an agent or a stimulus
seem related to their different external arguments. External arguments are assumed in current theory to be assigned by some functional head (cf. Kratzers
1996 Voice head). Suppose that heads introducing external arguments could
belong to more than one type. Specifically, in the case of ObjExp verbs, this
head could be agentive or stative. We can then explain the two readings of Ob
jExp verbs by assuming that the root fright can combine with two types of
verbal heads, which introduce an agentive or a stative external arguments. As
a result, the same root forms two types of ObjExp verbs: agentive or stative.
Let me elaborate now on this hypothesis. What is shared by the two readings is
precisely the root, for example fright. This is the smallest kernel, referring to
some event of fright, which both readings share. The root fright forms both
the stative and agentive readings. The identity of the verbal head with which the
Maya Arad
vP
NP
Agent
fright
NP
Experiencer
notion.12 The head v combines with the root phrase, P. Below v there can
be a predicate of change of state (as in the case of frighten) or other types of
predicates. For example, psych predicates can have the form of a change of
possession predicate (cf. give fright) or change of location (e.g. French mettre
en colre, literally put into anger).
Consider now the stative reading. I suggest that on this reading the root
is merged with a verbal head that I will call stative little v. This head has
the following properties: first, it gives the event the interpretation of stative
causation (unlike standard little v, which is active). Second, the argument in
its specifier is interpreted as a stative causer. Finally, its object is marked with
dative case (cf. Spanish) rather than accusative. Put more explicitly, I suggest
that little v comes in two flavors, active and stative:
v1
(17) a.
b.
v1
agent
v1
ACC
v2
v2
stative
causer
v2
DAT
Stative little v is responsible for the stative reading of psych verbs (a stative
causative construction):
v2
(18)
v2
stative
causer
v2
fright
NP
Both active and stative v introduce an external argument in their specifier and
check object case. However, they differ with respect to their semantic content:
one gives the event an interpretation of an agentive or active event, while the
other gives the event an interpretation of stative causation. The arguments in
their specifiers are interpreted accordingly, as an agent or as a stative, nonagentive causer. The heads also differ with respect to the morphological spell
out of the object-case they assign (dative vs. accusative). In the next section I
will further examine the similarities and differences between these two heads.
Maya Arad
Consider the first property. All verbal morphemes create verbal environments,
as part of their defining properties: they make roots into verbs, rather than
nouns or adjectives. I believe that the verbalizing property could be reduced
to a formal requirement of merging with T (or, possibly, Asp), but I leave the
issue open here (other properties of verbs, such as person features, could also
be associated with T).
The head little v also has semantic content. Possible characterizations of
this content are agentivity (gives the event agentive interpretation, cf. Kratzers
(1996) VoiceP), causation (the event is interpreted as a causative event, cf.
Harleys 1995 CAUS) or some aspectual content, such as process, giving the
event a durative interpretation (cf. Borers 1998 AspP head).
Finally, consider the transitivity property of v. It has an external argument in its specifier and forms a relation with the object (Agree Chomsky
1998). Taken together, these properties capture the essence of Burzios (1986)
generalization, that is, the correlation or the dependency between an external
argument and structural object case.13
Suppose now that these three features verbalizing, semantic content, and
transitivity are features of the universal set made available by UG. Let these
features be the ingredients that are used for building verbal morphemes (of
which little v is just an example). The verbalizing property distinguishes the
verbalized root from a noun. Semantic content has several values or flavors:
agentive, stative etc. Transitivity is a general property any head that introduces an external argument in its specifier and agrees with the object has the
transitivity property (I will discuss this in detail below). Languages can bundle
together any of these features, thus forming verbal heads of different types. In
other words, different verbal morphemes are different bundling of the features
available in the language. Possible examples of such feature bundles are:
bundle 1
bundle 2 bundle 3 bundle 4
verby
verby verby verby
agentive
agentive stative
inchoative
transitivity
bundle 5
verby
causative
transitivity
What counts as a possible V bundle? I assume that all v heads have the
verby property: all verbal morphemes make roots into verbs. All heads also
have some semantic content: heads with no content do not enter into the computational system. However, not all heads have the transitivity property, because not all verbs in language have an external argument and object case (cf.
unaccusatives).
We could, in principle, find heads that share only a subset of their properties. For example, heads sharing their semantic content but not transitivity, or
heads sharing transitivity but not semantic content. In fact, this is exactly the
case, as I will illustrate through several salient cases.
.. Sharing semantic content
Consider the contrast between transitive and reflexive verbs in Romance languages:
(19) a. Gianni lava Maria
b. Gianni si lava
(Italian)
A reflexive verb does not agree with its object (no object case is assigned). Also,
at least according to some analyses of Romance reflexives (e.g. Marantz 1984),
si verbs have no external argument. However, transitive and reflexive wash bear
many similarities: both refer to a washing event and both have an agent, a
washer, even if it is not syntactically projected. I assume that transitives and
reflexives are made from the same root, combined with different verbal morphemes. These morphemes share their semantic content (in this case agentive),
but differ precisely with respect to their transitivity: one has an external argument and checks object case, while the other does not. Passives are a similar
case. One of the best known generalizations about passives is that they do not
have a (syntactically projected) external argument and that they do not check
object case:
(20) a. He was hit
b. The book was read
However, another well-established claim about passives is that they share many
characteristics of their active counterparts. In particular, passives retain their
agentive interpretation: in (20) there was an agent who did the hitting or read
the book, even if it is not projected in the syntax. I assume that active and passive verbs are formed from the same root (note that they do share their basic
verbal morphology, as in eat and eaten). The root may be combined with two
morphemes, both sharing the property of being verbal and having some se-
Maya Arad
mantic content. However, one also has the transitivity property (active) while
the other lacks it (passive). Passive morphology may thus be only an overt manifestation for the lack of external argument and object case (in many languages
passive morphology is similar to that of reflexives or unaccusatives, all three of
them lacking object case).
.. Sharing transitivity
Consider now the opposite case: two heads sharing the transitivity property
but not their semantic content. This is the case of active and stative little v,
which was presented above in relation to ObjExp verbs. The two heads do not
share their semantic content one is active while the other is stative. However,
both have the transitivity property: they merge with an external argument (an
agent in one case, a stimulus in the other) and check object case (spelled out as
accusative in one case, dative in the other). It is possible, thus, for two verbal
morphemes to share transitivity but not semantic content. I take transitivity
to be a general property: any head with a filled specifier and a relation (agree)
with a lower element has the transitivity property.
v
(21)
filled spec
v
v
Another head that has the transitivity property is the applicative element in
Double Object Constructions (DOC). According to some analyses (Marantz
1993; McGinnis 1998), DOC involve an applicative head which introduces a
benefactive argument and checks the case of the lower object:
v
(22)
agent
v
v
vapplicative
benefactive
vapplicative
vapplicative
give
theme
The applicative head does not share the semantic content of either active or
stative little v, but it does share the transitivity property with them. It merges
with an external argument (benefactive; cf. Marantz (1993), where it is argued
that the benefactive argument is external to the inner event of change of state)
and checks the case of the lower object.
The examples discussed here represent the kind of morphemes that can
be made from the ingredients described above: verbalizing property, semantic
content and transitivity. The features selected from the lexicon of a particular
language can be bundled in different ways, thus creating verbal heads of different types. Active little v, stative little v and heads creating passives are examples
of such morphemes. Crucially, as argued above, morphemes may differ with
respect to one feature only (transitivity or semantic content). In the next section I will come back to the topic of Italian psych verbs, and consider another
case of verbal heads that differ with respect to some of their properties.
. The case of Italian piacere and preoccupare verbs
It was argued above that B&Rs effects hold only for the stative reading of ObjExp verbs. The question remains: what is it about the stative that reading triggers these effects? I suggest that this may be related to the type of (languageparticular) morpheme that creates ObjExp verbs in Italian. The crucial facts
are related to a sub-group of Italian ObjExp verbs, the piacere group, which
will be discussed below.
As shown by B&R, beside the worry-type (preoccupare) class, Italian has
verbs of the please-type (piacere). Apart form piacere, this small group includes
scocciare (displease) and interessare (displease):
(23) a.
Questo
this
b. Questo
this
preoccupa
worries
piace a
pleases to
Gianni.
G.
Gianni.
G.
Follow B&R, I argue that piacere verbs are similar in every respect to preoccupare verbs, except for their object case which is dative (but cf. Pesetsky 1995).
Interestingly, piacere verbs do not exhibit the same syntactic behavior as preoccupare verbs. They can form reflexives and be embedded under a causative verb:
(24) a.
??Gianni
G.
si preoccupa.
self worries.
Maya Arad
b. Gianni si piace.
G.
self please
G. likes himself/thinks highly of himself.14
(25) a. *Gianni ci
ha fatto preoccupare Maria.
Gianni us-dat has made worry
Maria
Gianni made us worry Maria.
b. Gianni ci
ha fatto piacere il gelato.
Gianni us-dat has made please ice cream
Gianni made us like ice cream.
So the situation in Italian is as follows: preoccupare verbs on their agentive reading do not exhibit psych effects. Piacere verbs do not exhibit psych effects. Only
preoccupare verbs on their stative reading exhibit psych effects.
Psych effects exist only when accusative case is assigned to the experiencer
in the absence of an agent. My assumption is that this is related to the properties
of the verbal head that is involved in forming each of these verb groups. Active
v merges with the root and forms the agentive reading of preoccoupare verbs.
Stative v forms dative ObjExp verbs like piacere:
v1
(26) a.
v1
agent
v1
v2
b.
ACC
preoccupare (agentive)
v2
stative
causer
v2
DAT
piacere
Unlike B&R, I assume that the dative case on piacere verbs is not inherent, but
is assigned by stative little v. Note, for example, that it can be absorbed with a
reflexive clitic, as in (24b).15 Thus, both agentive preoccupare and piacere have
structural case on their object.
I assume that the stative reading of preoccupare verbs is formed by combining the root with a defective v head: this morpheme shares the semantic
property of stative little v (giving the event a stative causative interpretation),
but not its transitivity property (i.e. does not agree with the object):
(stative) preoccupare:
verby
stative-causative
piacere:
verby
stative-causative
transitive
Stative preoccupare verbs are thus semantically similar to piacere verbs. However, the ACC case on the object of stative preoccupare is lexically marked
(or inherent, as suggested by B&R). Stative preoccupare verbs thus differ both
from agentive preoccupare and from the piacere group, in that they are formed
through a head that does not have a transitivity property.
My hypothesis is that the peculiar behavior of stative preoccupare verbs in
Italian is related to this lack of transitivity. Suppose that the ability to form
reflexives or to causativize is related to the transitivity property of the head. The
presence or absence of the transitivity property on the head may explain why
agentive preoccupare verbs and piacere verbs can form reflexives and causatives,
while stative preoccupare verbs cannot.
Let us see how this should work for the syntactic effects noted above. Consider reflexivization first. Reflexivization affects the transitivity property of the
verb and object case is absorbed (cf. Section 4.1.1 above):
(27) a.
Suppose that the process of reflexivization involves dispensing with the transitivity property of v. Taking this as my hypothesis, I assume that it is only
verbs that are formed through verbal heads possessing the transitivity property
which may appear as reflexives. In other words, in order to suppress the transitivity property, a head must first have the option to have it. Agentive preoccupare and piacere verbs can form reflexives, when their object case is absorbed.
The case on the object of stative preoccupare verbs cannot be absorbed, as it
is not assigned by a transitivity head, and thus the process of reflexivization
cannot take place.
B&R note that passives and raising verbs in Italian cannot reflexivize, and
assume that this is related to the fact that they lack external arguments. My hypothesis is that what passives, raising verbs and stative preoccupare verbs share
is this: they are formed by combining the root with a verbal head which does
not have any transitivity properties, hence their failure in reflexivization.
Consider next causativization. I assume that the formation of causatives
involves placing a verbal head on top of another head, in a process similar to
restructuring. If the causativized verb is mono-argumental, then the subject of
the lower verb is assigned ACC by the upper v:
Maya Arad
vcause
(28)
Maria
vcause
vcause
vagentive
vagentive
Gianni
work
vagentive
If the lower v is transitive, then the causative form is essentially similar to double object constructions (Marantz 1993), with the upper (causative) v checking
structural DAT, instead of ACC:
vcause
(29)
Maria
vcause
vcause
vagentive
vagentive
to Gianni
vagentive
eat
P
NP
an apple
rather than with the syntax. Recall that causativization in this case involves
merging a causative head that assigns structural case with a defective v (i.e., a v
which lacks transitivity), whose object is lexically marked with accusative case.
In the morphology, dative is inserted for the object of the upper verb, following the morphological accusative of the lower verb. This dative is infelicitous,
perhaps because the subjects of stative little v are not meant to be marked with
dative, as this makes them interpreted as subjects of agentive v.
The argument above is tentative, but I believe that there is some evidence
that the issue is indeed morphological. Italian has one verb that belong to both
the piacere and preoccupare groups, interessare, interest, which can take either
a dative or accusative object (30a). I assume that such roots are compatible with
both types of stative little v: transitive (with structural accusative) or defective
(with lexically marked accusative). Interestingly, the two variants exhibit a difference in causativization: the dative variant can undergo causativization, but
the accusative one cannot (30bc):
(30) a.
la politica / Maria lo
/ gli
interessa.
Politics
/ Maria he-acc / he-dat interests
Politics/Maria interest him.
b. Gianni ha fatto interessare Maria a Paolo.
Gianni made interest
Maria to Paolo.
Gianni made Maria interest Paolo.
c. *Gianni ha fatto interessare Paolo a Maria.
Gianni made interest
Paolo to Maria
The behavior of ObjExp verbs in Italian provides another example of the types
of verbal morphemes found across languages. In the next section I will look at
another case of variation across languages, Subject Experiencer verbs. There,
too, much of the variation can be traced into the verbal morphemes that create
the verbs.
Maya Arad
If we take roots to be the basic elements in the lexicon, the relation between
SubjExp and ObjExp verbs is straightforward. I suggest that (in many cases, in
many languages) SubjExp and ObjExp verbs are formed from the same root.
The combination of this root with different verbal morphemes yields verbs
of different types.
SubjExp and ObjExp verbs share some of their morphology in many languages:16
(33) a. ragaz (be angry)
b. hirgiz (anger)
c. hitragez (get angry)
( rgz)
( inho)
( ton)
b. udivljat
(surprise)
( udivl)
(Hebrew)
Like all verbal heads, the verbal morpheme that makes roots into SubjExp verbs
is a bundle of features. Consider now the features of this morpheme. It certainly
has the verbalizing feature, namely, making roots into verbs. It also has some
semantic content. In this case, the content is stative, non-causative. Like locative prepositions, it establishes a static relation between the experiencer and
a mental state (the experiencer is at some mental state). This head makes the
root into a stative event, and introduces an external argument that is stative and
non-causative. I assume that these features are universally associated with morphemes that make roots into SubjExp predicates: these verbs have to be stative,
as part of their content. However, the third property that verbal morphemes
may have, transitivity, is not obligatory.
Interestingly, as far as transitivity is concerned, the head forming SubjExp
predicates exhibits cross-linguistic variation. In some languages, like English
and Italian, it has the transitivity property and behaves like standard agentive
little v. Thus, SubjExp verbs in Italian and English have accusative object marking. Furthermore, SubjExp verbs in Italian behave like agentive transitive verbs
in allowing reflexive si and causativization (cf. B&R):
(37) a.
Gianni si
ama / apprezza / teme.
Gianni refl love / appreciates / fears
Gianni loves/appreciates/fears himself.
b. Gianni ha fatto amare /apprezzare /temere Paolo a Maria
Gianni made love /appreciate /fear
Paolo to M.
Gianni made Maria love/appreciate/fear Paolo.
In other languages stative verbs are realized differently from active verbs. In
Hindi and Georgian, for example, statives mark their subjects with dative case:
(38) a.
Gela-s
Gela-dat
b. Ram-ko
Ram-dat
nino
Nino-nom
Sita-se
Sita-instr
uqvars.
love
(Georgian: McGinnis 1998)
pyaar hai.
love be-Present
(Hindi)
In Irish and Scottish Gaelic SubjExp predicates are expressed through the verbs
be or have, combined with a noun or an adjective:
(39) a.
T fuath do Y ag X.
is hatred to Y at X
X hates Y.
b. Tha eagal orm.
is fear on me.
I am afraid.
I assume that the head that forms SubjExp verbs has the same semantic content (stative, non-causative) in all languages. The syntactic realization of the
stative head (its argument structure, case marking properties and transitivity)
is decided by each language separately, depending on which features it chooses
to bundle together. Some languages (English, Italian) bundle the transitivity
property and stative content under the same morpheme, while others (Georgian, Hindi) do not. The kind of morpheme a language has will affect the type
of verbs (the combinations of roots with morphemes) it has. A variation in
morphemes thus leads to syntactic variation in the realization of verbs across
languages.
The Experiencer argument of SubjExp verbs is an external argument, introduced by a verbal head. It is a subject of a state, while the Experiencer of
ObjExp verbs is the argument that is being put into a state. The two groups of
verbs also differ with respect to the type of event they encode. SubjExp verbs
are static: the Experiencer is at a certain state. ObjExp verbs (on both readings)
are dynamic: the Experiencer is being put into a state (with or without change
of state). We thus expect the two groups to have their Experiencer syntacti-
Maya Arad
cally realized in different positions subject (in static relations) or object (in
dynamic relations).
It is important to note at this point that cross-linguistic variation occurs
with SubjExp verbs and other stative verbs, but not with standard, agentive
verbs. Quirky (dative marked) subjects are often subjects of stative verbs, never
agents. I assume that this fact is related to stativity in some way. Perhaps because stativity is a property shared by verbs and prepositions of certain kind,
there are less restrictions on the way in which it is realized (unlike agentive
predicates, which must be realized as verbs). I leave the question open for future
research.
. Summary
Regarding the Lexicon of a language as a set of roots and (possibly bundled)
features enables us to give a more precise content to the claim that language
variation is restricted to lexical items. There are three ways in which languages vary:
1. Root inventory: what signs or lexical pieces does the language have? Variation in roots includes also the variation in the meanings that are assigned to
roots in different environments, or contexts. For example, cake is assigned
different meanings in eat a cake, take the cake or piece of cake (cf. Marantz
1997).17
2. Subset of features selected from the universal pool: what features does the
language employ in building its lexicon? Does it have morphological case
features, gender features, aspect (perfective/imperfective)?
3. The ways in which features are bundled together. As argued above, languages may put together different sets of features into morphemes. This
allows for a further source of lexical variation: what bundles does the
language have.
In this paper I concentrated specifically on verbal morphemes, in order to account for variation in verb types. I argued that there are three types of properties that serve as ingredients of verbal morphemes: verbalizing property
(making roots into verbs), semantic content (which may come in several flavors agentive, stative, stative-causative) and transitivity (formal features, external argument and case checking, which are optional). Different phenomena across languages (passivization, reflexivization and the formation of psych
verbs) have been argued to result from the combination of roots with different
morpheme types.
Notes
. For a similar proposal, assigning a morpho-syntactic origin for language variation, see
Mateu and Rigau (this volume).
. Some theories postulate a richer structure of the verbal projection, correlating with different verb classes. See, in particular, Hale and Keyser (1998), Kural (this volume). In this
paper I argue that such richness is best captured in terms of features of the verbal head that
merges with the root. However, this approach shares much with the structural approach
mentioned above.
. This syntactic and semantic alternation of the root is reflected in the morphology of
the verb. The root appears with a different verb-creating morpheme, or pattern, in each
of its appearances as a stative, inchoative or causative verb. Note that the same type of
alternation exists in Romance languages, although its morphological effects are less rich.
The causative-inchoative alternation is morphologically expressed by the presence or absence of the pronominal clitic SE (e.g. French rchauffer, heat-causative vs. se rchauffer,
heat-inchoative).
. b, k and p are spirantized in Hebrew in post-vocalic positions, yielding v, x and f respectively.
. Initial y gives rise to phonologically contracted forms in certain contexts, such as hoiv.
. This ambiguity dates back to Ruwet (1972), who notes that verbs such as strike cannot be
interpreted as psych verbs if they are agentive (strike someone with your intelligence vs. strike
someone with a bat).
. There is also a non-stative non-agentive reading, or an eventive reading, which has a
change of mental state in the experiencer, but no intentional agent:
(i)
This reading patterns with the agentive reading in some languages, and with the stative reading in others. In this paper I abstract away from this issue, as it does not bear directly on my
analysis. See Arad (1998) for discussion.
. Note that although the stative reading is easier to get with present tense or habitual aspect,
it cannot be reduced to it. As shown in Pylkknen (1997), the stative reading can also refer
to a single event, or spell of mental state.
. In languages in which object case marking is sensitive to change of state, such as Finnish,
the objects of these verbs are marked with partitive case (instead of accusative).
. In languages in which psych verbs bear causative morphology, such as Hebrew or
Finnish, both readings carry a causative morpheme (cf. Pylkknen 1997).
Maya Arad
. B&R show, in fact, that verbs such as colpire, strike, can take a reflexive on their physical,
non-psych reading:
(i) Gianni si
colpito a. con un bastone.
Gianni self is struck
with a
stick
b. *per la
sua prontezza.
by
the his quickness
. In fact, as shown by Ruwet (1972), many ObjExp have a non-psych reading, in which
the same participant is interpreted not as an experiencer, but as a patient: shake, disturb,
move etc.
. Note that this is not an explanation of Burzios generalization, but rather, a description.
In line with much work on case (e.g. Marantz 1991; Laka 1993) I think this generalization
could be derived from independent principles (such as the manner in which case is assigned
and the relation between v and T and the EPP).
. Note that this reading is non-agentive: it does not mean that John is trying to please
himself, but rather, that he likes himself, thinks highly of himself, etc.
. Note that dative is assigned structurally in causative constructions in Italian when the
causativized verb is transitive (see discussion in the text below).
. English has very few pairs of SubjExp and ObjExp verbs formed from the same root with
References
Arad, M. (1998). VP structure and the Syntax-Lexicon Interface. Doctoral dissertation,
University College London.
Belletti, A. & L. Rizzi (1988). Psych verbs and theta theory. NLLT, 6, 291352.
Borer, H. (1984). Parametric Syntax. Dordrecht: Reidel.
Borer, H. (1998). Passive without Theta Grids. In S. Lepointe (Ed.), Morphology and its
Interfaces with Phonology and Syntax. Stanford, CA: CSLI.
Burzio, L. (1986). Italian Syntax. Dordrecht: Kluwer.
Chomsky, N. (1995). The Minimalist Program. Cambridge, MA: MIT Press.
Chomsky, N. (1998). Minimalist inquiries: The framework. MITOPL. MIT.
Collins, C. (1997). Local Economy. Cambridge, MA: MIT Press.
Grimshaw, J. (1990). Argument Structure. Cambridge, MA: MIT Press.
Hale, K. & J. Keyser (1998). The Basic Elements of Argument Structure. In H. Harley (Ed.),
MITWPL 32. Papers from the Upenn/MIT Roundtable on Argument Structure and
Aspect (pp. 73118).
Halle, M. & A. Marantz (1993). Distributed Morphology and Pieces of Inflection. In K. Hale
and J. Keyser (Eds.), The View from Building 20 (pp. 111176). Cambridge, MA: MIT
Press.
Agree or attract?
A Relativized Minimality solution to a
Proper Binding Condition puzzle
Cedric Boeckx
University of Illinois at Urbana-Champaign
I examine a paradigm first discussed by Kroch and Joshi (1985) which Lasnik
(in press) takes as an argument for feature-movement. I show that Lasniks
solution is problematic on both conceptual and empirical grounds. I offer an
alternative approach that in contrast to previous solutions does not rely on
the Proper Binding Condition, but instead deeply implicates Relativized
Minimality.
Introduction
Considerable insight has been gained in recent years into the nature of word
order by focusing on possibilities of remnant movement.1 Abstractly, remnant
movement takes the form in (1). An element is moved out of an element ,
which is subsequently moved to a position higher than (and featurally distinct
from) s derived position.
(1) [[ ... t ...] ... [ ... ... [ ... t ...]]]
Cedric Boeckx
(Fiengo 1977)
The influence of remnant movement analyses in recent years has grown steady.
For instance, projects are being developed to show that most (perhaps all) instances of head-movement can, and should be reanalyzed in terms of remnant
movement (see Koopman & Szabolcsi 2000 and Mahajan 2000, among others).
Kayne (1998, 2001) has even suggested that traditional instances of covert processes such as scope reversal via QR or reconstruction be captured by making
massive use of remnant movement, with no need for a distinct LF component.
Remnant movement raises many intriguing properties which I will not discuss here (see Mller 1998 for what is to date the most detailed study of remnant movement).2 Rather, I will focus on a fairly narrow issue that arises in
the remnant movement approach, and suggest that once properly analyzed the
problem is only apparent. In addition, our investigation will enable us to draw
some important conclusions about the structure of the grammar; in particular,
about the status of a distinct LF component.
. How likely
Consider the contrast in (4)(5), originally reported in Kroch and Joshi (1985).
(4) a. John is likely to win
b. There is likely to be a riot
(5) a. How likely to win is John
b. *How likely to be a riot is there
Surprisingly, an example like (4b), which is minimally different from (4a), fails
to yield a grammatical output if the how likely ... chunk is raised, as in (5b)
(compare (6) and (7)).
Agree or attract?
(7) (=5b) [CP [how likely [t i to be a riot]]j [C isk [IP therei t k [VP t j ]]]]
The crucial difference between (4a) and (4b) is that John raises to matrix SpecIP
in (4a), but not in (4b). Instead in (4b) an expletive is inserted in the embedded
SpecIP (due to the preference of Merge-over-Move, see Chomsky 1995: 348;
2000: 111), and undergoes movement to SpecIP. (I return to the status of expletive raising below. Note that if expletives are taken to undergo predicate
movement, as in Moro 1997 and related work, (5b) is equally puzzling.)
The contrast in (5) is puzzling in more than one respect. Mller (1998: 7,
n.10) mentions the paradigm in (5) and suggests we treat (5b) as an unexplained exception. It is indeed difficult to see what grammatical property would
exclude (5b) while ruling in (5a).
Further, it has been suggested in Vukic (1998) and Bokovic (2001a) that
expletives are merged in their surface position without undergoing any movement (both Vukic and Bokovic treat expletives as grammatical formatives that
are merged as late as possible. See Bokovic for evidence that there is no EPPchecking in the infinitival complement of raising predicates). If correct, the
late-insertion view of expletives replaces (7) with (8).
(8) [CP [how likely [to be a riot]]j [C isk [IP there t k [VP t j ]]]]
That (8) yields an ungrammatical output, and (6) doesnt is clearly unexpected.
If anything, one would expect the reverse pattern of grammaticality, as (6) appears to violate the PBC, whereas (8) does not (the raised predicate does not
contain any trace at all).
It is this puzzle that I will concentrate on in this paper. On grounds that
I have discussed elsewhere (see Boeckx 2001), I will assume that the derivation of (5b) given in (8) is the correct one (i.e., expletives are merged in their
surface positions). The thesis I will entertain here is that the contrast in (5)
has nothing to do with remnant movement or the Proper Binding Condition.
Whatever their ultimate status, these will remain unaffected by (5), for reasons
to be developed shortly. I will also argue, contra Lasnik (in press), that the contrast in (5) fails to provide an argument for the existence of feature movement
(Chomsky 1995) and for a distinct LF component.
Cedric Boeckx
. More facts
Before offering a solution to the contrast in (5), let me expand the data base by
showing that the contrast in (4)(5) is not limited to existential constructions.
A effect similar to (5b) is found with idiom chunks (9b).3
(9) a. Advantage is likely to be taken of John
b. *How likely to be taken of John is advantage
Relying on Kroch and Joshis original observations, Lasnik and Saito (1992: 141)
take the how-likely paradigm to require an explanation in terms of the PBC. According to them, (5b) and (9b) are out because the trace (of there and of advantage, respectively) fails to be properly bound after predicate raising. (Note that
when only part of the predicate raises, as in (10), all traces are bound, as the
portion of the predicate containing the trace remains in situ.) According to Lasnik and Saito, what saves (5a) is the existence of an alternative derivation that
does not contain an unbound trace. Lasnik and Saito appeal to a long-standing
view that modal predicates such as likely are ambiguous between raising and
control predicates. Thus, (4a) may be represented as (11) or (12).
(11) John is likely [t to win]
(12) John is likely [PRO to win]
If the derivation in (11) is chosen, (5b) will violate the PBC, as shown in (13).
(13) [CP [how likely [*t i to win]]j [C isk [IP Johni t k [VP t j ]]]]
If, however, the control derivation is chosen, no trace will be contained in the
raised predicate, and the sentence will be grammatical.
(14) [CP [how likely [PRO to win]]j [C isk [IP John t k [VP t j ]]]]
Lasnik and Saito note that since expletives and idiom chunks cannot control
PRO, a derivation like (14) is unavailable to them.
Agree or attract?
The three sentences just given mirror the contrast between (5a) and (5b)(9b).
The importance of the present cases is that they do not involve the remnant
Cedric Boeckx
movement part (predicate raising). Hence they cannot be ruled out via the
PBC.
. A Move-F account
Recently, Lasnik (in press) has revisited the how-likely paradigm, and argued
that it provides an argument for Chomskys (1995) treatment of covert movement as feature movement.
To understand the argument, it is useful to retrace certain developments
in the minimalist program concerning the nature of covert movement and the
timing of operations.
. Theories of movement
Whereas the first Minimalist paper (Chomsky 1993) had taken over the socalled Y-model (Chomsky & Lasnik 1977) to some extent (compare (20) and its
minimalist variant (21)), subsequent research within the Minimalist Program
(henceforth MP) has led to interesting modifications of the central architecture
of the language faculty.
(20)
D(eep)-Structure
(21)
Initial Array
S(urface)-Structure
LF
PF
Spell-Out
LF
PF
Chomsky (1995) still adopts the temporal asymmetry between overt and
covert movement (i.e., pre- vs post-Spell-Out operations), much like in Chomsky (1993), but suggests that we view covert operations as consisting not of
movement of categories that happen to receive no pronunciation (as was the
case in 1993, and in work within the GB-framework), but rather of movement
of formal features. For Chomsky (1995), the operation Move (...) seeks to raise
just F[eature] (p. 262). Chomskys reasoning is that movement is triggered to
check features. We therefore expect under Minimalist assumptions that if the
computational component can raise just what is needed (features to carry out
the checking operation), it will do so. The question now arises as to why sometimes whole categories, and not just formal features, move. Chomskys answer
(in 1995) is that overt movement is to be decomposed in the following way.
Agree or attract?
Applied to the feature F, the operation Move creates at least one, perhaps two
derivative chains alongside the chain CHF = (F, tF ) constructed by the operation itself. One is CHFF = (FF[F], t FF[F] ), consisting of the set of formal features
FF[F] and its trace; the other is CHCAT = (, t), a category carried along by
generalized pied-piping.
(Chomsky 1995: 265)
Chomsky assumes that any overt operation is the result of moving features first
(so far, overt and covert movements are indistinguishable, except perhaps in
terms of timing, a point I will discuss extensively below), and then an operation
of pied-piping which carries along the remnants of the item from which the
features have been moved.
Chomsky assumes that
For the most part perhaps completely it is properties of the phonological component that require pied-piping. Isolated features and other scattered
parts of words may not be subject to its rules, in which case the derivation is
canceled; or the derivation might proceed to PF with elements that are unpronounceable, violating F[ull] I[nterpretation].
(Chomsky 1995: 266)
As emphasized in the first quote, within a Move-F framework, overt and covert
operations are indistinguishable up to a certain point (formation of a derivative chain/pied-piping). This leads Chomsky to claim that such considerations could permit raising without pied-piping even overtly, depending on
morphological structure. (p. 266)
What Chomsky means by overtly here is the overt (pre-Spell-Out) component in (21). Before Move-F, it was assumed without discussion that covert
operations took place after Spell-Out. Given Move-F, it is now possible to view
covert movement as the first part of overt movement, that is, as an operation
that does not require any distinct component. This is indeed the conclusion
that Chomsky (2000, 2001a, b) embraces:
There is a single cycle; all operations are cyclic. Within narrow syntax, operations that have or lack phonetic effects are interspersed. There is no distinct
LF component within narrow syntax.
(Chomsky 2000: 131)
Cedric Boeckx
S(urface)-Structure
LF-expletive replacement
This analysis was criticized as soon as it was proposed (apparently, first, by Lori
Davis; Howard Lasnik, personal communication): the expletive replacement
analysis gets the scope facts wrong. As is well-known, indefinites in subject
positions are scopally ambiguous (see (23)). (22b) predicts that such ambiguity
exists in existential constructions. But this is not the case. The associate in (24)
only has a low reading.
Agree or attract?
There are many problems with this analysis, and I wont review them here.
They are thoroughly discussed in Lasnik (1992).
Chomsky (1995) proposes a much more satisfactory account. Chomskys
reasoning is that movement is triggered to check features. We therefore expect
under Minimalist assumptions that demand minimization wherever possible
that if the computational component can raise just what is needed (features to
carry out the checking operation), it will do so (recall the operation Move (...)
seeks to raise just F[eature] (Chomsky 1995: 262)). Thus, Chomsky argues
for the existence of feature movement (Move-F). Relying on the Move-F hypothesis, Chomsky proposes that in existential constructions only formal (-)
features of the associate NP move (head-adjoin) to Infl0 , leaving all phonological and semantic features behind. Raising of -features immediately accounts
for the fact that finite agreement in existential constructions is controlled by
the feature specification of the associate, as illustrated in (26). (I here set aside
semi-formulaic examples like theres two men in the garden.)
(26) a. there is/*are a man in the garden
b. there *is/are two men in the garden
As Lasnik has extensively discussed (see the essays in Lasnik 1999b), the Move-F
account provides a straightforward explanation for the narrow scope of the
associate NP in these constructions if we assume, quite plausibly, that the establishment of scopal relations is more than a matter of formal features, and
requires phrasal displacement (see Pesetsky 2000: 25 for some discussion).9
On largely conceptual grounds, Chomsky (2000:123) dispenses with feature movement altogether and captures its effects via the operation Agree.
The latter amounts to a process of feature checking (in his terms, valuation)
at a distance.
The summary of minimalist views on covert movement given here, especially the covert process that accounts for agreement in existential sentences
Cedric Boeckx
will now enable us to examine Lasniks (in press) argument in favor of feature
movement based on the how likely paradigm discussed in previous sections.
. The Proper Binding residue
Lasnik (in press) uses the paradigm above in support of the Move-F hypothesis.
His argument runs as follows. Barss (1986) rules out (5b) by capitalizing on
Chomskys (1986) expletive-replacement analysis of existential constructions.
Barss notes that if a riot must replace there at LF, the movement will be illicit
because it is sidewards. Lasnik agrees that the expletive replacement account
cannot be correct, but he claims that Barsss analysis can be maintained under
the feature movement analysis, crucially not under an Agree analysis.
Assume that the -features of a riot are attracted in (5b), the element becomes PF-deficient (see Chomskys 1995 quote above; see also Ochi 1999a, b;
Lasnik 1999c and Uriagereka 1999). How likely-fronting removes the category
from the c-command domain of the moved features, the necessary repair strategy cannot be carried out, and the derivation crashes due to the presence of
scattered features at the interfaces.
Lasnik claims that an Agree account cannot capture the fact in (5b), as in
the absence of (feature) movement, there is no feature scattering to start with,
hence no requirement for the associate NP to remain within the c-command
domain of the expletive.
Lasniks analysis is appealing for the level of subtlety it reaches. In contrast
to many other studies (see, e.g., Wurmbrand 2001), it is not concerned with
whether or not some non-phrasal covert process exists, but with the more
difficult question of which form that process takes.
However, Lasniks account faces many problems.
First, by adopting a feature movement analysis, Lasnik inherits the conceptual difficulties that led Chomsky to reject feature chains in favor of Agree.
In particular, it is not clear what feature scattering (crucial for Lasnik) means
under the copy theory of movement. Movement of the feature will leave a copy
behind, rendering the need for repair obscure. Further, if feature movement
chains reduce to head-chains (Chomsky 1995; Bokovic 1998), they inherent
the problems associated with the latter (see Chomsky 2000, 2001a; Brody 2000;
Boeckx & Stjepanovic 2001; and Mahajan 2000).
Second, in order to account for why feature movement is not accompanied by repair in standard existential sentences, Lasnik has to assume that that
instance of feature movement takes place after Spell-out, in a separate LF com-
Agree or attract?
ponent (where feature scattering does not cause any crash). He is thus forced
to a return to the Y-model.
Third, Lasnik has to postulate that feature movement out of the copy left by
remnant movement is impossible (contra Bokovic 1997 and Nishioka 1997).
If it were, the remnant movement case (5b) would be virtually identical to (4b),
as illustrated in (27).
(27) [how likely to be a riot] [is [there [<how likely to be a riot]>]
|
F-movement
|
Fourth, Lasniks solution says nothing about the badness of (17), as it crucially
relies on remnant movement to exclude (5b).
Fifth, it is not clear how the feature movement account of (5b) extends to
(9b) (let alone (18)). Unlike the expletive-associate relation, the raising of an
idiom chunk to the idiom remnant has never been treated in terms of feature
movement as far as I know.
On the basis of the problems it faces, I think it is fair to say that Lasniks
analysis is inadequate.
. Relativized Minimality
We have seen that neither a raising vs. control/PBC account nor a feature movement account adequately captures the how likely paradigm. In this section I
propose a novel way of looking at the facts that not only accounts for the whole
range of data, but also allows us to preserve the arguably more elegant single output model of syntax, and does not jeopardize any conclusions about
remnant movement reached by previous studies.
The format of the solution I would like to argue for is well-known. It essentially amounts to a Relativized Minimality violation (Rizzi 1990). An element
enters into a relation with an element if there is no that meets the requirement(s) of (i.e., that matches ), and either c-commands . The illicit
situation is schematized in (28).
(28) [ ... [ ... ...[ ... ...]]] ( c-commands )
In the following I will adopt Starkes (2001) conception of Relativized Minimality, as it leads to what I think is a clearer solution. The portion of Starkes
view on chains that will be relevant for us is roughly as follows.
(29) a. ... ...
b. ... ...
Cedric Boeckx
In (29a) we have three elements of the same type (). Attempting to relate the
first and the third (in linear order) leads to a violation of Relativized Minimality. The situation in (29b) is more complex. The intervening element (second
element in linear order) is of type (). The first and the third elements which
the grammar is trying to relate both contain an feature. In addition, they contain a feature which is missing from the intervener. Starkes point is that if the
first and the third element are -related, the situation that obtains is equivalent
to that in (29a), and is thus ruled out by Relativized Minimality. If, however,
the first and the third elements are -related, no intervention effect emerges, as
the potential intervener is not of the same type.
A concrete case of (29a) is a superiority condition of the type we found in
(30)(31).
(30) *whati did who buy t i
(31) [C [ who T [buy what]]]
+wh +wh
+wh
Starkes view is that the good instances of extraction (32b, 33b) correspond to
the situation in (29b) when the moving element -relates to its final landing
site. The bad instances of extraction correspond to an -relation.
Agree or attract?
In what follows, I argue that the /-relation in (29b) plays a role in the
how-likely paradigm. The reader should understand that I will not try to rule
out the bad cases by appealing to some inadequacy of the remnant movement
step fronting how likely .... The ungrammaticality of (17)(18) suffices, in my
view, to show that the badness of (5b)(9b) is independent of remnant movement. In other words, they should be ruled out prior to the application of
remnant movement. Put differently, in order to derive the contrast between
(4b)/(9a) and (5b)/(9b), we must find a difference between the following stages
of the derivation:
(34) a. is [likely to be a riot]
b. is likely to be advantage taken of John]
(35) a. is [how likely to be a riot]
b. is [how likely to be taken advantage of John]
With Chomsky (2000) I assume that in existential constructions, Infl0 and the
associate stand in a checking relationship by Agree, and that feature matching
takes place as the derivation unfolds, not in a distinct LF-component. It will
therefore be crucial to examine the various cases step by step.
The good cases (4a, 4b, 9a) are derived straightforwardly, as in (38)(39).
(38) [ T [likely [... NP ...]]]
|
|
(39) a. [ T [likely [John to win]]] Agree (T,John) + Move (John)
b. [ T [likely [to be a riot]] Agree (T,[a riot]) + Merge there
_: there-insertion
c. [ T [likely [to be taken advantage of John]]] Agree (T,advantage) +
Move (advantage)
The proposal I would like to make to rule out the cases in (5b, 9b) is that
the presence of how in how likely ... blocks the Agree relation between T and
some NPs.
It has often been argued that that wh-words are decomposable into a whpart and an indefinite part (an idea going back to Chomsky 1964 and Katz &
Postal 1964). Suppose that how in how likely actually consists of a wh-part and
an indefinite part, roughly as wh-indefinite (degree).12 I would like to argue
Cedric Boeckx
It may be objected that the indefinite part of the wh-phrase does not ccommand the associate, hence should not be a blocker for Agree (recall (28)).
However, there are various ways around this well-known almost c-command
problem. For concreteness, I will assume that c-command out of the specifier
of an XP is possible (Kayne 1994). (Note that some features of how must head
the whole phrase to trigger pied-piping under wh-movement.)
Although (40) rules out the crucial Agree relation in (39b, c), it appears
to do so in (39a) as well, predicting (5a) to have the same status as (5b)(9b),
contrary to fact. However, here Starkes characterization of weak islands (29b)
comes handy. Recall that blocking is obviated if there is an alternative agreerelation involving a feature that is absent from the blocker. I will argue that
the [+NP] corresponds to in (29b). The idea now being that there is another
feature that is present and can partake in Agree in (39a), but not in (39b, c).
The feature that I will make use of is [+D]. It is often assumed that definite
noun phrases are DPs, while indefinites are NPs (see, e.g., Chomsky 1995: 342,
350). Further, indefinites are ambiguous between a DP reading (e.g., specific
indefinites) and an NP reading. The ambiguity may account for the two readings in (41).
(41) someone is likely to win the lottery (someone likely/likely someone)
Let us take the absence of the wide scope reading in (42) to mean that the
indefinite NP in existential constructions is a pure NP (it lacks the DP reading).
(This restriction may underlie the well-known definiteness effect.)
Now let us go back to the examples in (39) and the intervention effect in
(40). The presence of the indefinite feature on how blocks the Agree relation
that relates T and the noun phrase in the infinitive complement. This is the
-relation in (29b). I propose that Agree can succeed if a D-feature is involved.
This would correspond to the -relation in (29b). The D-feature is absent from
the intervener how. Having established that the indefinite noun phrase in existential construction is a pure NP, the -relation (Agree [+D]) cannot be estab-
Agree or attract?
(39a) is rescue by the presence of a D-feature on the noun phrase being attracted.
(44) [ T [how
likely [... DP ...]]]
[+DP] [+WH,+NP]
[+DP]
|
|
So far our proposal is able to capture the basic how-likely paradigm without
any appeal to remnant movement, distinct LF-component, or move-F, which I
take to be desirable.
What remains to be explained is the improvement in (10b, c) (repeated).
(46) a. how likely is there to be a riot
b. how likely is advantage to be taken of John
In an earlier version of the present work (Boeckx 1999), I claimed that extraposition took place in (46) (more precisely, I adopted Larsons 1988 treatment
of extraposition as resulting from Light-Predicate Raising (see already Fiengo
1977, see also Kayne 1994, and, for a precise formulation of the Light-Predicate
Raising rule, Runner 1995). The derivation I assumed is given in (47).
(47) a. [how [likely [to be a riot]]] predicate-raising/extraposition
b. [to be a riot]i [how [likely [t i ]]]
Cedric Boeckx
The crucial step was (47b), which brings the NP past the intervener, allowing the establishment of the -relation (Agree [+NP]), thus avoiding the fate
of (39b, c).
However, as pointed out in Nomura (2001) there are at least two problems
with the extraposition analysis (problem #1 was also brought to my attention
by Klaus Abels, p.c.). First, what prevents extraposition from taking place in
(17), repeated here as (48).
(48) *who said that there was how likely to be a riot
The problematic step here is (49c), where an Agree/-relation can be established. By remnant-moving how likely ... in (49b), the (almost) c-command
relation between how and the lower NP (a riot) is broken, preventing intervention from taking place. (49) thus predicts (17) to be grammatical (a similar
problem arises for (18), which I wont illustrate here).
A second problem for the extraposition approach is that the extraposition
step in (47a) patterns unlike familiar instances of extraposition (see Nomura
2001). (The contrasts are sublte, but nonetheless significant.)
As shown in (51), wh-movement out of an extraposed infinitival is degraded.
(50) whati did Bill ask Mary [to fix t i ] yesterday
(51)
??what
Agree or attract?
Crucially, the deviance in (52) is not replicated to the same degree in parallel
cases involving how likely.
(54)
??what
(55)
???what
i
i
As I see no straightforward solution to the problems raised for the extraposition account by Nomura (2001), I reject the solution to (10b, c) offered in
Boeckx (1999), and turn to an alternative.
The proposal I would like to make is based on an intuition going back to
Rosenbaum (1967: 108, n. 1).14 The idea is that what is traditionally referred to
as raising adjectives (likely, certain) are in fact peculiar adverbs (Rosenbaums
term). Treating raising adjectives as adjuncts opens up a new possibility of
dealing with (10b, c). It has become popular since Lebeaux (1988) to view adjuncts as being inserted acyclically (in contrast to complements, which conform to a strict view of the cycle, along the lines of Chomskys 1993 Extension
Condition). (See, e.g, Chomsky 1993 and much subsequent work). Suppose
then that there are two ways of inserting how likely: cyclically, in which case it
takes the to-infinitive as its complement), or acyclically; separately from its alleged complement (to-infinitive). If the latter option is chosen, a sentence like
(56) is derived as in (57).
(56) how likely is John to win
(57) a.
b.
c.
d.
Acyclic insertion essentially allows us treat how likely as a constituent, independent from the to-infinitive. This option therefore does not affect cases where
wh-movement of how likely pied-pipes the infinitive. This is an important
point because if acyclic insertion could be involved in pied-piping cases, nothing would prevent a derivation like (58), which is clearly unwanted, as it would
incorrectly rule in cases like (5b).16
(58) a. is [to be a riot] Agree (T,riot) (+ there-insertion)
b. [there [is [to be a riot]]] acyclic insertion of how likely
c. [there [is [how likely] [to be a riot]]] wh-movement (+ piedpiping)
d. [[how likely [to be a riot]]j isk [there [t k [t j ]]]]
Cedric Boeckx
Acyclic insertion of how likely does not face the second problem raised by Nomura for the extraposition account. In the absence of extraposition, we do not
expect (52) and (54) to pattern the same way. As for the first problem (how
likely in situ, as in (17)), it also does not arise under acyclic insertion. If how
likely is an adjunct, (17) reduces to (59). That is, whatever excludes in-situ whadjuncts in multiple questions will exclude the acyclic insertion option for a
case like (17). The only option available will be the cyclic insertion, which gives
rise to intervention, as demonstrated above.
(59) *who left why
The acyclic insertion analysis of how likely thus seems superior to the extraposition analysis in accounting for the grammaticality of (10b, c), without running
afoul of (17)(18).
. Conclusion
To conclude, I have examined a paradigm first discussed by Kroch and Joshi
(1985) which Lasnik (in press) took as an argument for feature-movement. I
have shown that Lasniks solution is problematic on several grounds. In particular, it fails to provide a solution for part of the paradigm (the idiom case)
and moreover leads to an organization of the grammar that contains a separate LF component and feature chains. I have offered an alternative approach
to the how likely paradigm that deeply implicates Relativized Minimality. Not
only does the present analysis capture the full paradigm straightforwardly, it
also need not assume the existence of feature-movement chains or of a distinct LF-component.17 In so doing, the present analysis lends credence to the
conceptually more elegant mechanism of Agree and the One-cycle model of
syntax. Finally, the account reconciles the how likely paradigm with independent conclusions about remnant movement, which previous analyses (Mller
1998) had failed to do.
Notes
. For helpful comments on an early version of this paper, I thank eljko Bokovic, Howard
Lasnik, Adolfo Ausn, Koji Sugisaki, and especially Jairo Nunes. I particularly appreciate
comments by Klaus Abels and Masashi Nomura which led to a sharpening of the hypothesis I first entertained, and considerable improvement of its technical implementation. I
Agree or attract?
I will not discuss such examples here, as the simple cases of raised predicates like (i-a) pose
non-trivial questions (why is there no Relativized Minimality effect when an NP, the winner,
crosses another one, John?), to which I cannot provide an answer here. If the exact nature of
the first step of movement out of the small clause to SpecIP is ignored, the solution I offer
in the text for (5b)(9b) extends to (i-b).
. The raising-control ambiguity appears to be supported by Martins (1992, 1996) argument, based on observations in Barss (1986), that a sentence like (i) allows for only one
reading of the indefinite, in contrast with regular raising cases (see (ii)), but alongside with
control predicates (see May 1985) (iii).
(i)
(ii) someone from NY is likely to win the lottery (someone likely/likely someone)
(iii) A unicorn is eager to be apprehended
(a unicorn eager/*eager a unicorn)
Setting aside the issue of whether control predicates do indeed block reconstruction (see
Hornstein 2001: 139140), I note that Sauerland (1999) provides an analysis of the contrast
between (i) and (ii) that is independent from raising vs control. The explanation I provide
below for the how likely paradigm also captures the asymmetry.
. Contrast:
(i)
who said that it was how likely that there will be a riot
. Some speakers find (19) marginal (see note 14 below for a possible explanation for this).
Crucially, they still perceive a contrast between (17)/(18) and (19). (Thanks to Masashi
Nomura for discussion of the relevant examples.)
Cedric Boeckx
. For a different tack on the overt/covert asymmetry, see Nissenbaum (2000) and Chomsky
(2001b), where covert movement is taken to be phrasal movement taking place after (cyclic)
spell-out. I will not discuss this option here, as many details still remain to be worked out.
. The LF-intervention effects on A-bar movement discussed in Pesetsky (2000) seem to
provide another argument in favor non-phrasal covert operations.
. Lasnik also points out that the Move-F analysis captures the paradigm discussed in Den
Dikken (1995) (see also Lasnik & Saito 1991 for similar examples in ECM-contexts) which
is problematic under expletive-replacement.
(i)
a.
b.
(ii) a. *there seem to each other to be some applicants eligible for the job
b. *there seem to any of the deans to be no applicants eligible for the job
As the data in (ii) show, the associate is incapable of licensing an NPI/anaphor located in the
matrix clause, which is unexpected under the expletive replacement analysis since according
to the latter (i) and (ii) share the same LFs. Lasnik takes the ungrammaticality of the sentences in (ii) to mean that such licensing mechanisms require more than formal features.
But see Branigan (1999), Yatsushiro (1999), and Watanabe (2000) for some arguments that
binding (but not scope) can be established via feature movement.
. Strictly speaking, this is not Cinques (1990) view. According to him, all islands are absolute. Apparent (good) extraction out of islands are cases of base-generation and (nonovert) resumption. I ignore this detail here (for arguments that resumption demands a
movement-approach, see Boeckx 2001).
. I will not be concerned here with defining the relevant factors easing extraction out
of weak islands. For valuable discussion, see Frampton (1999), Szabolcsi and Den Dikken
(1999), and Starke (2001).
. Masashi Nomura (p.c., attributing the original observation to Howard Lasnik) points
out that if the proposal I make is correct, we expect no substantial difference between (i)
and (ii). My proposal indeed seems to predict a Relativized Minimality effect triggered by
somewhat in (ii).
(i)
Agree or attract?
ment matches the featural requirements of a probe P, but fails to agree with it. (In other
words, blocks the raising of to even though itself cannot raise to .)
. Howard Lasnik (p.c.) informs me that Noam Chomsky has made a proposal similar to
Rosenbaums at various times. Neither Lasnik nor I have been able to locate the proposal in
Chomskys writings. It must therefore have been made in class lectures.
. Alternatively, one could follow Rizzi (1990), Uriagereka (1988), Law (1991, 1993),
Boeckx (2001), and Starke (2001), among others, and take wh-adjuncts to be directly inserted in COMP. As far as I can see, this possibility yields identical results to the text discussion (i.e., it is unavailable in the pied-piping cases and in the in-situ examples, for reasons
I discuss in the text immediately below).
Klaus Abels points out that many speakers of English find long-distance questions with how
likely like (i) deviant.
(i) *?how likely did John say that there was to be a riot
This suggests that how likely patterns like how come, which is known to lack long-distance
construal.
(ii) *how come did John say that Mary left t
The facts in (i) and (ii) may demand an analysis in terms of base-generation in [+wh]
SpecCP for how come and how likely. Such an analysis may account for the fact that some
speakers find (19) deviant.
. Unless Relativized Minimality is viewed as a condition on representation (see Rizzi
1986). I reject this option as I follow Chomsky (2000) in taking Agree to be derivationally
established.
. Lasniks (in press) second argument in favor feature movement (based on the interaction of head-movement and ellipsis) is equally problematic, as discussed in Boeckx and
Stjepanovic (2001).
References
Barbiers, S. (1995). The syntax of interpretation. The Hague: HAG.
Barss, A. (1986). Chains and Anaphoric Dependence. Doctoral dissertation, MIT.
Bobaljik, J. D. (1995). Morphosyntax: The syntax of verbal inflection. Doctoral dissertation,
MIT.
Bobaljik, J. D. (2001). A-chains at the Interfaces: Copies, agreement, and covert
movement. Ms., McGill University.
Bobaljik, J. D. & S. Wurmbrand (2000). Modals, Raising, and Reconstruction. Ms., McGill
University.
Boeckx, C. (1999). Agree or Attract? Ms., University of Connecticut.
Boeckx, C. (2001). Mechanisms of Chain Formation. Doctoral dissertation, University of
Connecticut.
Boeckx, C. & S. Stjepanovic (2001). Head-ing toward PF. Linguistic Inquiry, 32, 345355.
Cedric Boeckx
Agree or attract?
Kayne, R. S. (2001). Raising, Reconstruction, and Remnant Movement. Talk given at the
Asymmetry conference, May 2001, UQAM.
Katz, J. & P. Postal (1964). An Integrated Theory of Linguistic Descriptions. Cambridge, MA:
The MIT Press.
Koopman, H. & A. Szabolcsi (2000). Verbal Complexes. Cambridge, MA: The MIT Press.
Kratzer, A. (1991). Modality. In A. von Stechow and D. Wunderlich (Eds.), Semantics. An
international handbook of contemporary research (pp. 639650). Berlin: de Gruyter.
Kroch, A. & A. Joshi (1985). The Linguistic Relevance of Tree-adjoining Grammar. Ms.,
University of Pennsylvania.
Larson, R. (1988). On the Double Object Construction. Linguistic Inquiry, 19, 335391.
Lasnik, H. (1992). Case and Expletives: Notes toward a parametric account. Linguistic
Inquiry, 23, 381405.
Lasnik, H. (1999a). Chains of Arguments. In S. D. Epstein and N. Hornstein (Eds.), Working
Minimalism (pp. 189215). Cambridge, MA: The MIT Press.
Lasnik, H. (1999b). Minimalist Analysis. Oxford: Blackwell.
Lasnik, H. (1999c). On Feature Strength: Three minimalist approaches to overt movement.
Linguistic Inquiry, 30, 197217.
Lasnik, H. (In press). Feature Movement or Agreement at a Distance? In A. Alexiadou,
E. Anagnostopoulou, S. Barbiers and H.-M. Gaertner (Eds.), Remnant Movement, Fmovement and the T-model. Amsterdam: John Benjamins.
Lasnik, H. & M. Saito (1991). On the Subject of Infinitives. In L. Dobrin, L. Nichols and
R. Rodriguez (Eds.), Papers from the 27th Regional Meeting of CLS (pp. 324343).
University of Chicago, IL.
Lasnik, H. & M. Saito (1992). Move . Cambridge, MA: The MIT Press.
Law, P. S. (1991). Effects of Head-movement on Theories of Subjacency and Proper
Government. Doctoral dissertation, MIT.
Law, P. S. (1993). On the Base Position of Wh-adjuncts and Extraction. Paper presented at
the 67th Annual Meeting of the Linguistic Society of America, Los Angeles.
Mahajan, A. (2000). Eliminating Head-movement. Ms., University of California, Los
Angeles.
Martin, R. (1992). On the Distribution and Case Features of PRO. Ms., University of
Connecticut.
Martin, R. (1996). A Minimalist Theory of PRO and Control. Doctoral dissertation,
University of Connecticut.
May, R. (1985). Logical Form. Cambridge, MA: The MIT Press.
Moro, A. (1997). The Raising of Predicates. Cambridge: CUP.
Mller, G. (1998). Incomplete Category Fronting. Dordrecht: Kluwer.
Newmeyer, F. (1975). English Aspectual Verbs. The Hague: Mouton.
Nishioka, N. (1997). On Trace Movement. English Linguistics, 14, 182202.
Nissenbaum, J. (2000). Explorations in Covert Phrase Movement. Doctoral dissertation,
MIT.
Nomura, M. (2001). Extraposition or Scattered Deletion. Ms., University of Connecticut.
Nunes, J. (1999). Linearization of Chains and the Phonetic Realizations of Chain Links. In
S. D. Epstein and N. Hornstein (Eds.), Working Minimalism (pp. 217249). Cambridge,
MA: The MIT Press.
Cedric Boeckx
Nunes, J. (To appear). Sideward Movement and Linearization of Chains in the Minimalist
Program. Cambridge, MA: The MIT Press.
Ochi, M. (1999a). Constraints on Feature Checking. Doctoral dissertation, University of
Connecticut.
Ochi, M. (1999b). Some Consequences of Attact-F. Lingua, 109, 81109.
Pesetsky, D. (2000). Phrasal Movement and its Kin. Cambridge, MA: The MIT Press.
Rizzi, L. (1986). On Chain Formation. In H. Borer (Ed.), Syntax and Semantics, 19: The
syntax of pronominal clitics (pp. 6595). Orlando: Academic Press.
Rizzi, L. (1990). Relativized Minimality. Cambridge, MA: The MIT Press.
Rosenbaum, P. S. (1967). The Grammar of English Predicate Complement Constructions.
Cambridge, MA: The MIT Press.
Runner, J. (1995). The Licensing of Noun Phrases. Doctoral dissertation, University of
Massachusetts, Amherst.
Saito, M. (2001). A Derivational Approach to the Interpretation of Scrambling Chains. Ms.,
Nanzan University.
Sauerland, U. (1999). Scope Reconstruction Without Reconstruction. In Proceedings of
WCCFL 17 (pp. 582596). Stanford CA: CSLI.
Starke, M. (2001). Move Dissolves into Merge: A theory of locality. Doctoral dissertation,
University of Geneva.
Szabolcsi, A. & M. den Dikken (1999). Islands. Ms., New York University and CUNY.
Szabolcsi, A. & F. Zwarts (1993). Weak Islands and Algebraic Semantics for Scope Taking.
Natural Language Semantics, 1, 235284.
Uriagereka, J. (1988). On Government. Doctoral dissertation, University of Connecticut.
Uriagereka, J. (1999). Minimal Restrictions on Basque Movements. Natural Language and
Linguistic Theory, 17, 403444.
Vukic, S. (1999). Attract F and the Minimal Link Condition. Linguistic Analysis, 28, 185226.
Watanabe, A. (2000). Feature Copying and Binding: Evidence from Complementizer
Agreement and Switch Reference. Syntax, 3, 159181.
Wurmbrand, S. (1998). Infinitives. Doctoral dissertation, MIT.
Wurmbrand, S. (2001). Move or Agree? Paper preresented at WCCFL 20, February 2001,
University of Southern California.
Yatsushiro, K. (1999). Case Licensing and VP Structure. Doctoral dissertation, University of
Connecticut.
Distributed deletion*
Introduction
(Croatian)
(Polish)
(German)
(Croatian)
(Polish)
The empirical focus of the present article lies on the constructions in (5)(6),
which we will call (XP-) split constructions.1 The standard analysis of (5) was
proposed by van Riemsdijk (1989): the part of the XP that appears in clauseinitial position is moved out of XP, stranding the material left behind. If left
branch extraction is impossible, the analysis of (6) must be more complex. It
involves remnant movement of an XP out of which some material has been
extracted before it was placed into the clause-initial slot.
However, movement analyses face serious problems with respect to syntactic islands and the phonetic shape of the parts of the split phrase (the regeneration problem discovered by van Riemsdijk). We will argue that these
problems render a simple movement analysis of the XP-split construction impossible. However, it does not seem amenable to a treatment in which both
parts are base-generated in situ, either. A way out of this apparent paradox is
offered by the copy & deletion (cd-) approach to movement (Chomsky 1995) if
it is implemented in such a way that the deletion operation following the copying step of movement may affect both copies. The cd-approach offers a unified
analysis for both type of constructions, i.e., DP-splits as in (5) and PP-splits as
in (6). How such a derivation may proceed is illustrated in (7) for Croatian.
(7) mi je Marija zanimljive knijge preporucila
Distributed deletion
This account for XP-splitting may suggest itself, so the major virtue of the
present paper lies in the presentation of the empirical arguments in its favor,
and in developing the approach in some detail. Our account may also be applicable to (3) and (4), and its general idea seems helpful for a number of further
puzzles of syntax.
The article is structured as follows. Section 2 introduces core properties
of the split construction, and distinguishes two types of splits. Section 3 is
dedicated to a discussion and refutation of previous analyses. Sections 4 and
5 presents the distributed deletion theory in some detail. Section 6 briefly
discusses loose ends and possible extensions of the present approach.
(9) a.
Na kakav
je Ivan krov skocio?
on what-kind has Ivan roof jumped?
On what kind of roof has Ivan jumped?
hat er Schweine gekauft?
b. Wieviel
how many has he pigs
bought
How many pigs has he bought?
(Croatian)
(German)
(11) a.
dass er teure
Bcher wahrscheinlich der
Frau
keine
that he expensive books probably
the.dat woman no
schenken wollte.
give
wanted
. . . that he probably did not want to give the woman expensive books
as a presents.
b. ?*dass er wahrscheinlich teure Bcher der Frau keine schenken wollte.
XP-splits come in two varieties. XPs can simply be pulled apart (Pull-splits),
leaving XP-internal order intact. This is illustrated in (12). German differs from
the Slavic languages in allowing pull-splits for simple wh-extraction only (13).
(12) Na kakav
je Ivan krov skocio?
on what-kind has Ivan roof jumped?
On what kind of roof has Ivan jumped?
(Croatian)
Wieviel
hat er Bcher gelesen?
how many has he books read
How many books has he read?
a . Wieviel Bcher hat er gelesen?
b. *Keine hat er Bcher gelesen.
no
has he books read
He has not read any books.
b . Keine Bcher hat er gelesen.
(13) a.
The internal order of the XP can also be inverted in the split construction, as
illustrated in (14). Inverted splits are well-formed for noun phrases only, but not
for PPs (15). Therefore, PP-splits are confined to wh-movement in German.
Distributed deletion
(14) a.
(Croatian)
Na kakav
je Ivan krov skocio?
on what-kind has Ivan roof jumped?
On what kind of roof has Ivan jumped?
b. *Krov je Ivan na kakav skocio?
(Croatian)
(15) a.
(German)
PPs can be torn apart in a different way in German (and certain dialects of
Croatian), however. In (16), the prepositional head of the PP appears in both
parts, while the DP-part of the PP is split in the inverted way (as compared to
in keinen Schlssern in no castles, *in Schlssern keinen).
(16) In Schlssern habe ich noch in keinen gewohnt
lived
in castles
have I yet in no
I have not yet lived in any castles.
. Previous analyses
. Simple movement theories
The standard account for XP-discontinuity is movement. In (17a) the verb
phrase is serialized discontinuously, because who has been extracted from it.
That (17b) involves movement, too, seems to be the standard view, though
alternative accounts have been proposed (see Horn 1975). The null hypothesis for XP-splits thus should also involve the creation of discontinuity by
movement, as has been proposed for German by van Riemsdijk (1989), Tappe
(1989), Diesing (1992), Kniffka (1996) among others, and by Franks and Progovac (1994) for Croatian, or Yearley (1993) and Sekerina (1997) for Russian.
(17) a. who did you [VP see t] ?
b. who did you see [DP a picture of t] ?
At early stages of generative theory, movement analyses for XP-splits were confronted with the problem that movement is restricted to minimal or maximal
projections, while the analysis of split noun phrases seems to presuppose that
submaximal projections are moved, cf. (18) for an illustration, and Fanselow
(1988) for the pertinent argument.
(18) a.
As (18) shows, any segment of [keine [interessanten [neuen [Bcher]]]] can undergo movement in an extraction account of XP-splits, and at first glance, only
one of these segments can be maximal. But, as was noted by Tappe (1989) and
Kniffka (1996), this line of reasoning is problematic because of the additional
layers of functional structure that have been discovered in the DP, following the
seminal work of Abney (1987) in fact, the movement facts of (18) themselves
constitute evidence for an elaborate internal structure of noun phrases, which
might look as in (19). A movement analysis of inverted splits can thus pick any
of the functional projections in the noun phrase, and move it to the front.
(19) [DP [D keine] [AGR-A1-P [AP interessanten] [[AGR-A1 e] [AGR-A2-P [AP neuen ]
[[AGR-A2 e][Nom-P Bcher]]]]]]
Pull splits require a slightly more complex derivation. Since P+Det does not
form a constituent, the derivation of (20) must involve remnant movement in
the sense of den Besten & Webelhuth (1990), Mller (1998), see, e.g., Corver
(1990) and the discussion in Sekerina (1997): first, krov is extracted from na
kakav krov (this involves an inverted split), then [na kakav t] is moved to
sentence initial position.
Distributed deletion
(20) Na kakav
je Ivan krov skocio?
on what-kind has Ivan roof jumped?
On what kind of roof has Ivan jumped?
(Croatian)
This analysis has the advantage of reducing pull splits to inverted splits followed
by remnant movement, and thus seems to explain why languages allow pull
splits only if inverted ones are licensed, too but it faces the problem that PPs
disallow inverted splits though pull splits of PPs are fine.
Simple movement theories face at least two kinds of problems, both of
which have already been alluded to. First, inverted splits can be imperfect
in the sense that the two parts contain more phonetic material than fits into a
single constituent. The case of preposition doubling (21a, b) has been discussed
above,3 but a similar constellation arises with determiners, too, as (21cd)
illustrate.4
(21) a.
The indefinite article and the negative quantifier kein do not go together in
German noun phrases (as (21d) shows), because they compete for the same
structural position, but they may occur in different parts of a split noun phrase
(21c). Imperfect splits such as (21a, c) have no well-formed source in a movement account there is not enough space in a single continuous XP for the
material present in the split case.
Van Riemsdijk (1989) attributes the imperfection of the split in (21c) and
similar examples to a regeneration process: according to his theory, what
moves to first position in (21c) is just amerikanischen Wagen. This sequence,
however, is not a legal independent noun phrase in German.5 Therefore, phrase
structure rules re-apply after movement and insert an indefinite article in order to guarantee well-formedness. At the present moment, one can at least say
that regeneration adds a complication to the movement analysis, which one
would hope to be able to avoid.
The second problem of the movement account of split XPs has also been
mentioned already. Recall that, (6), (8b), or (10) require a remnant movement
analysis, in which a nominal projection is moved out of PP. But this ingredient
of an extraction analysis of splits is confronted with the serious problem that
PPs are islands for movement in Croatian otherwise, as (22) illustrates:
(22) a.
b.
c.
d.
e.
The examples in (22) show that PPs are islands for wh-extraction (b), topicalization (c), and scrambling (d). However, a split of the complex PP is possible,
as (22e) shows. Thus, the movement step necessary for creating the discontinuous PP is not well-formed, since it violates a strong island restriction.6
The problem is not confined to split PPs. In German, split noun phrases do
not respect at least three types of islands, as the following data illustrate. First,
(23) shows that subjects (of non-unaccusative verbs, at least) are islands for the
extraction of PPs (cf. e.g. Mller 1996). Nevertheless, subjects can be split up,
as Fanselow (1988, 1993) observes (24).7
(23) a. *[An Maria] haben mir [keine Briefe t] gefallen.
to Mary have me no letters
pleased
No letters to Mary have pleased me.
b. *[An Maria] hat mich [kein] Brief t] erschreckt.
to Mary has me
no letter
frightened
No letter to Mary has frightened me.
Briefe an Maria gefallen mir keine.
letters to Mary please me no
As for letters to Mary, they do not pleaseme.
b. Briefe an Maria haben mich keine erschreckt.
letters to Mary have me no frightened
As for letters to Mary, they have not frightened me.
(24) a.
Kniffka (1996: 52) shows that subjects can be split up even when they precede
modal particles which are often claimed to mark the boundary of VP. Like-
Distributed deletion
Dative indirect objects (26ab) and many genetive (26cd) noun phrases illustrate essentially the same point. They are islands for movement (Mller 1996;
Vogel & Steinbach 1998), yet split noun phrases can be formed on their basis
(Fanselow 1993; Kniffka 1996: 33).
(26) a. *[ber Polen ] ist hier noch [keinen Bchern t ] ein Preis
about Poland is here yet
no
books-dat a prize
verliehen worden
awarded been
No books about Poland have been awarded with a prize here.
b. Interessanten Bchern ber Polen ist hier noch keinen ein
interesting books about Poland is here yet no
a
Preis verliehen worden.
prize awarded been
As for interesting books about Poland, no prize have been awarded
to any of them here so far.
c. *[An Studenten] habe ich ihn [schrecklicher Morde]
at students have I him horrible-gen murders-gen
angeklagt
accused
I have him accused of horrible murders of students.
d. Schrecklicher Morde an Studenten ist er vieler beschuldigt
horrible
murders at students is he many accused
worden.
been
He has been accused of many horrible murders of students.
Similar arguments can be formulated with respect to the Specific Subject Condition and pragmatic constraints on movement. Thus, a number of stable generalizations concerning extraction8 are not fulfilled by split noun phrases in
German.
Mohawk is also in line with this picture. As Baker (1991, 1995) shows, whmovement is subject to standard CED effects in Mohawk, cf. (27) (= (28), (29),
and (30a) in Baker 1991).
(27) a.
Noun phrases are intransparent for movement (28) (= (34) in Baker 1991), but
they can be split up irrespective of grammatical function (29) (= (40)(41) in
Baker 1991).
(28) a. *uhkai se-nuhwe-s ne ti ako-kara
who 2sS-like-hab NE
FsP-story
Whose story do you like?
b. *uhka we-sa-tsituni-
ne ti ako-kara
FsP-story
who fact-2sO-make.cry-punc NE
Whose story made you cry?
to ni-hati
wa-she-kv-
rati-ihn-a-rakv
how part-MpS fact-2sS/3pO-see-punc MpS-skin-be-white
How many white men did you see?
b. to ni-hati wa-esa-kv- rati-ihn-a-rakv
How many white men saw you?
ne kweskwes
c. ka nikayv wa-hse-nut-e
which
fact2sS-feed-punc NE pig
Which pig did you feed?
ne kweskwes
d. ka nikayv wa-ka-nvst-a-k-e
which
fact-ZsS-corn-eat-punc NE pig
Which pig ate the corn?
(29) a.
The same problems arise in Slavic languages, as Sekerina (1997) shows. One of
the most fundamental predictions of a movement account of split constituents,
namely those of the bounding theory, is thus not borne out.9
There are further data that require additional complications in a movement account. In (30), one part of a split noun phrase occupies a position in a
VP moved to clause-initial position, whereas the other part is left behind.
Distributed deletion
(30)
On the other hand, the overtly legal split operation (34a) does not feed remnant
VP topicalization (34b) in contrast to all other movement types.
(34) a.
weil
Bcher selbst der Fritz noch keine t geschrieben hat
because books even the F.
yet no
written
has
Because even Fritz has not yet written any books
b. *[[ noch keine t ] geschrieben ] hat Bcher selbst der Fritz
c. Bcher geschrieben hat selbst der Fritz noch keine.
books written
has even the F.
yet no
Even Fritz has not yet written any books.
If one finds a way of blocking (32), one could try harder and attempt to explain well-formed (30) or (34c) by two rather than one steps of movement
preceding remnant VP topicalization. Thus, one could first move Bcher out
of keine Bcher, so as to yield (35b). If it is now possible to front the remnant noun phrase keine t, as in (35c), one would have produced a constituent
(underlined in (35c)) that contains exactly the phonetic material one needs to
front in remnant VP topicalization for (30).
We have seen, then, that simple movement theories of XP-splits face at least
three types of problems:
a. They cannot account for the repetition of phonetic material in imperfect
splits.
b. They cannot cope with the fact that XP-splits disrespect standard islands
for movement (PP-islands, barriers by lack of l-marking).
c. They cannot handle the existence of XP-splits in VP-fronting constructions
easily.
. Base generation theories
At least the first two of the three problems for movement accounts would not
arise if the parts of split constituents would be base-generated in place. The
idea that discontinuous phrases are generated as two (or more) independent
constituents goes back to Hale (1983). According to him, split noun phrases
(in Warlpiri) are a diagnostics for non-configurationality.
Thematic theory seems to militate against the view that more than one
phrase is linked to a single thematic role, but whether this constitutes a problem
depends on the nature of thematic linking. Hale (1983) proposed a theory of
Lexical Conceptual Structure and its relation to phrase structure in which multiple linking of more than one NP to a single role is unproblematic. Furthermore, NPs fulfill functions other than the referential closing of argument slots
in Warlpiri. It may even be the case that the only function of non-pronominal
NPs in Warlpiri (Jelinek 1984) or Mohawk (Baker 1995) is that of adjuncts,
so that no conflict with standard theta-theory arises. Van Geenhoven (1998)
presents a semantic theory that is able to handle multiple XPs that are linked
to the same argument slot, at least for the case of direct objects.
Distributed deletion
(39) a.
The Japanese XP-splits (38), (39a) involve two independent noun phrases, one
generated in an A-position, the other being merged in a Topic position. XPsplits of the Slavic and German type differ from Japanese, however, in that
certain kinds of islands have to be respected. This fact can be accounted for
in a base generation account only indirectly. (40a) illustrates the fact that the
relation between the parts of an XP-split respects the Complex Noun Phrase
Constraint in German. Similarly, the complex noun phrase (40b) is an island
for both movement (40c) and split constituent formation (40d) in Croatian.
(40) a. *Bcher habe ich [eine Geschichte dass sie keine liest ] gehrt.
books have I a
story
that she no reads heard
I have heard a story that she does not read any books.
je Marija svojoj sestri
b. Ivan je vidio [NP auto [RelCP koji
car
which is M.
her sister
I.
is seen
kupila ]]
bought
Ivan has seen the car which Mary bought for her sister.
je Marija ti
sestri ] je Ivan vidio [NP auto [RelCP koji
c. *[NP Cijoj
auto
which is M.
whose sister is I.
seen
kupila]]?
bought
Whose sister is such that Ivan saw the car which Mary bought for
her?
i je Ivan vidio [NP auto [RelCP koji je Marija ti sestri kupila ]]?
d. *Cijoj
Fanselow (1988) tries to account for such facts by assuming that one of the two
parts of an XP-split has to obligatorily undergo movement to Spec,CP after
having been merged independently of the other part of the split construction.
Since the two NP parts are merged independently of each other, it is obvious why the relation between keine and Bcher itself need not respect islands
for movement as such (41a). If Bcher has to undergo later movement (41b),
the relation between Bcher and its trace must be compatible with subjacency,
however a fact Fanselow claims is able to capture (40a).
(41) a.
This argument is valid, however, only if there are additional constraints on the
distance at which two XPs may be merged independently of each other when
they are linked to the same thematic role. Otherwise, one could circumvent all
islands constraints by simply merging the XP-parts at any distance. Even if such
locality constraints on merger can be identified,10 a base-generation account of
XP-splits for German leaves it open why one of the XPs must move to Spec,CP
or the sentence internal topic-position in the sense of Frey (2000). Similarly,
a theory which merges the parts of XP-splits in Croatian and Polish in situ
would leave it unexplained why at least one part of a split DP and PP must
appear in front of VP proper, in a focus position. There may be technical ways
Distributed deletion
to guarantee that such a kind of movement takes place (see Fanselow 1988) but
they are certainly not satisfactory.
A further disadvantage of base generation solutions is that they have little
to say about a phenomenon favoring movement analyses. Riemsdijk (1989) observes that some linear order facts are unexpected in base generation theories.11
Order is not free in German noun phrases. As (42a) illustrates, there is only
one option for arranging the prenominal elements keine, zwei, grne which
is mirrored in the discontinuous case, as (42b) shows. We can explain (42b) if
the source of a split noun phrase is its continuous counterpart (=42a) while it
is not obvious how a base generation might capture (42b): noun phrases such
zwei Bcher, keine grnen, keine Bcher or zwei Grne are perfect if they form
a single complete phrase. Thus, they should be able to co-occur within a single
clause if they can be generated independently of each other.
(42) a.
Similarly, adjective order is not free. (43a) is unmarked while (43b) is not the
latter is fully acceptable only if amerikanische bears focal stress.
(43) a.
It is hard to imagine that such restrictions12 can be made follow from semantic
or related considerations in base generation accounts, at least, no such accounts
have been proposed so far.
. A prosodic option?
Zec and Inkelas (1990) assume that syntactic constituents may be split by enclitics in Serbo-Croatian. They claim that such data provides evidence for a
phonological or prosodic placement of enclitics. For Croatian, (45) shows that
the clitic-cluster may appear after a complex DP (45a), or apparently inside
the complex DP, as in (45b).
(45) a.
As Cavar
(1999) points out, the same type of syntactic discontinuity is possible
with the constructions discussed in Zec and Inkelas (1990), in (47), a demonstrative is topicalized, being separated from the head noun of the complex DP
by the subject Ivan, and not just by clitics.
(47) Taj je
Ivan kupio auto.
this be3sg I.
buyptc car
Ivan bought this car.
Distributed deletion
Prosodic placement of clitics thus cannot be the general analysis of split constituents. Whether some XP-splits emerge as a result of prosodic clitic placement is an open issue, however. Thus, Browne (1975) argues that (48b) must
be due to a non-syntactic clitic placement, because (48c) suggests that proper
names can only be split in sentence initial position quite unlike what we have
seen in (46) and (47).
(48) a.
Lav Tolstoj je
veliki ruski
pisac.
Leo Tolstoy be3sg great Russian writer
Leo Tolstoy is a great Russian writer.
Tolstoj veliki ruski
pisac.
b. ?Lav je
Leo be3sg Tolstoy great Russian writer
bio Tolstoj veliki ruski
pisac.
c. *Lav je
L. be3sg beptc T.
great Russian writer
(50) a.
Lava
L.
b. *Lava
L.
citam
read1sg
citam
read1sg
Tolstoja.
T.
Tolstoj.
T.
Cavar
(1999).
For the contrast in (52), the following characterization suggests itself: a movement barrier does not block the formation of a split XP if and only if
itself is the XP to be split up. This follows if (a) splitting up involves movement (then, (52b) is explained), but (b) not movement out of .13 If splits are
not formed by moving something out of the category that will be split up, the
subject condition has no chance to block (52a).
The idea that split formation may involve movement, but not movement
of part of XP out of XP is enigmatic at first glance only. It makes sense if we
assume that a chain <, > is formed (so barriers dominating must be respected), in which the phonetic material of is partially realized in the upper
position, and partially in the lower copy. This is the core idea of the partial
(distributed) deletion account of split constituents.
Distributed deletion
(53) a.
There is evidence that the deletion of the lower copy is not an automatic sequel
to movement. Rather, as was argued, e.g., by von Stechow (1992), Groat and
ONeill (1996), Pesetsky (1998) and Sabel (1998), among others, at least some
instances of covert movement14 are better analysed as movement in the overt
component, with the upstairs rather than the downstairs copy being made
invisible to the phonological component:
(53) Full Deletion of upper Copy
c. . . . . . .. . . . . .
and Cavar
(2001), and Nunes (2001) for analyses.
(54) wer denkst du denn wer du bist?
who think you ptc who you are?
Who do you think you are?
Furthermore, Pesetsky (1998) argues that (certain) resumptive pronouns reflect the failure of copies of movement to delete completely. Thus, there seems
to be some evidence that (53c) and (53c ) are not the only legal modes of treating chains in terms of phonological realizations. What we would like to add to
this picture is the idea that, under certain conditions, deletion may affect both
the upstairs and the downstairs copy, but in a partial way so, which yields the
split XP construction. Thus, simplifying matters first, assume that a movement
step maps (55a) onto (55b), by copying a noun phrase. If the downstairs copy
Under such a view,15 it is obvious why simple island effects fail to arise with split
XPs: the step from (55b) to (55d) does not involve movement at all (so the XP
to be split up cannot be a barrier), but split formation involves movement, so
that barriers containing the lower XP have an effect on well-formedness (52b).
Partial or distributed deletion as envisaged here is an extension of partial
reconstruction at LF to the overt component of grammar.16 Reflections on the
failure of quantifier raising or LF-wh-movement to bleed Principle C effects
(see, e.g., Fox 1995; Nunes 1995; Pesetsky 2000) and further considerations
(see Chomsky 1995) suggest that the semantic material of a phrase may end up
being distributed to more than one position in a chain. It has been observed,
e.g., that LF quantifier raising normally does not bleed the effects of Principle
C of the Binding Theory. Thus, him and John cannot be coreferent in (56),
although Quantifier Raising of the object should yield an LF-representation
such as (57), in which John is no longer c-commanded and bound by him. If,
however, as much semantic material of the quantified NP is reconstructed after
LF-movement as is compatible with the necessity to keep the quantificational
head in place, as in (57 ), a structure arises that represents scope, does not fail
to imply the Principle C effect, and which is, in effect, identical with the kind
of structure that arises by partial deletion in overt syntax, according to our
account.17 See Fox (1995), Pesetsky (2000), Wilder (1997) for arguments that
show that partial reconstruction is superior to an analysis in which Principle C
is checked before LF-movement.
(56) *I sent himi [every letter Johni expected]
(57) *[every letter Johni expected]k I sent himi tk
(57 ) *[every]k I sent himi [tk letter Johni expected]
Distributed deletion
. Pragmatic conditioning
Both in Croatian and in German, XP-splits go hand in hand with a particular pragmatic structure that was studied in detail by Kniffka (1996) and de Kuthy (2000)
for German, and for Slavic languages, e.g., by Siewierska (1984) (Polish), Lapteva
(1976) and Sekerina (1997) (Russian). In a split construction, the right part of XP
must be focal, while the lefthand part may be a (link-) topic or a second focus. Note
that both Croatian and German sentence structure offer a number of positions reserved for YPs with specific pragmatic functions, such as focus and topic positions
(see Frey 2000; Pili 2001). Bringing these observations together, the following generalization suggests itself: the XP-split construction is grammatical only if a single
XP must fulfill two different positional requirements defined by pragmatic constraints on order.18 In other words: Suppose that XP bears a feature f1 that requires
that XP be overtly realized in position A, and an additional feature f2 that forces
XP into position B. Then XP is split up in languages like Croatian or German.
(58) [[A XP] ..... [[B XP] ......]]
This general idea can be made precise along the following lines. Suppose that an
XP = [ap [b c]q ] bears two semantic or pragmatic features p, q, such as [+wh],
[+focus], [+link-topic], etc., and suppose that these feature are checked by corresponding heads Hp and Hq in the standard way: the head attracts a phrase bearing
a corresponding feature. Consider now a structure such as (59a). If the features
p and q must be checked on Hp and Hq , respectively, (59b) will arise after two
instances of movement/ attraction.
(59) a. [Hp .... [Hq ... [XP ap [b c]q ]]]
b. [[XP ap [b c]q ] [Hp .... [[XP ap [b c]q ] [Hq ... [XP ap [b c]q ]]]]]
In the approach proposed here, the strength of the attracting feature does not
determine whether movement (copying) applies before Spellout or not. Rather,
copying always takes place as soon as possible. The strength of the attracting feature
rather determines which of the copies created by movement is spelt out. In the
easy case, the attracting features of both Hp and Hq are weak, so that the lowest
copy is spelt out (=59c) if (60) holds.
(59) c.
[[XP ap [b c]q ] [Hp .... [[XP ap [b c]q ] [Hq ... [XP ap [b c]q ]]]]]
(60) In a chain C = <C1 ,C2 > of XP, C1 is not spelt out if the feature attracting
XP to C1 is weak.
For heads with strong attracting features, the most simple implementation of standard ideas would seem to be (61). (61) yields correct results when only one attracting feature is strong (as sketched in (59d)), but problems arise as soon as two
attracting heads have strong features, as seems to be the case in the split construction (both parts appear in positions related to semantic/pragmatic features). (61)
would then require that XP be spelt out in both positions.
(61) In a chain C = <C1 ,C2 > of XP, C1 is spelt out if the feature attracting XP
to C1 is strong.
(59) d. [[XP ap [b c]q ] [Hp .... [[XP ap [b c]q ] [Hq ... [XP ap [b c]q ]]]]]
Multiple full copies of a single phrase (caused by the presence of two strong features
of different heads attracting the same XP) seem non-existent in natural languages.
Thus, (61) cannot be maintained. A situation in which both p and q of [XP ap [b
c]q ] are attracted by corresponding strong features of different heads either implies ineffability, or the XP-split construction. The former situation holds in Dutch
(where a constellation in which one part of an NP is focal, the other topical, simply
cannot be expressed, as Henk van Riemsdijk (p.c.) points out), the latter in German
and Croatian.
(62) Suppose C = <C1 ,C2 > is formed because a strong feature of H has attracted XP and suppose that H checks the operators features f1 ... fk of XP.
Then the categories bearing f1 ... fk must be spelt out in C1 .
According to (62), operator positions checked by strong features must be filled
by phonetic material bearing the corresponding operator feature. This implies an
XP-split construction whenever the operator features are checked in two different
specifier positions.
When a phrase bears only one operator feature, it is not split up, even if not all
of its parts bear that feature. This is guaranteed if the phonetic spellout is governed
. Anti-freezing
In the preceding section, we have seen why XP-splits arise only if XP bears two
different pragmatic or semantic functions. The mechanism of splitting XPs that is
implicit in (62) makes a stronger claim: it presupposes that phrases that are split
are moved to specifier positions linked to operator features. Is this stronger claim
really justified? At least for Croatian, the answer seems to be positive.
Distributed deletion
In Section 3.1, we have observed that barriers such as subject islands, dative
islands or PP islands are not respected by split constituent. Croatian obeys a further restriction: descriptively speaking, a constituent cannot be split up in its root
position, rather, a split is possible just in case both (all) parts of the XP occupy
derived positions. Thus, as (63) illustrates, a PP cannot be split up if part of it remains in the base position following the verb (recall Croatian is an SVO language
underlyingly):
(63) a. *na kakav je Ivan bacio loptu krov
on what is Ivan thrown ball roof
b. *na kakav je Ivan bacio krov loptu
c. na kakav je Ivan krov bacio loptu
Dative DPs share these properties. They are islands for extraction, as (64a) illustrates. Nevertheless, they can be split up, as expected (64b), but only so if no part
of the discontinuous NP follows the verb (65).19
(64) a. *cega
je policajac pokazao oferu put za Split
of what is policeman shown driver way to split
The policeman has shown the way to Split to the driver of what?
b. oferu je policajac autobusa pokazao put za Split
driver is policeman of-bus shown way to Split
The policeman has shown the driver of the bus the way to Split.
(65) a. *cijoj je Ivan dao knjigu sestri
whose is Ivan given book to sister
Whose sister has Ivan given the book to?
b. *cijoj je Ivan dao sestri knjigu
c. cijoj je Ivan sestri dao knjigu
It thus appears as if splitting up DPs and PPs is possible in derived positions only.
This is predicted if the spellout principle (62) refers to chains and attractors, and
not to focus or topic positions. However, in Croatian (and in Polish), there is
an exception to the generalization just presented: accusative noun phrases can be
discontinuous even if part of the DP follows the verb:
Distributed deletion
That relativized minimality considerations imply that XP-splits are of the inverted
type is a welcome consequence, given that inverted splits are the default version of
the split construction.
Pull splits preserve the c-command relations among the overt elements of the
continuous XP: an XP merged as [a [b [c]]] appears as [a [X [b [Y [c] ....]]]] at the
surface. Therefore, pull splits may be related to the Parallel Movement Constraint
(PMC) proposed by Mller (2001).
(68) Parallel Movement Constraint
If A c-commands B at level L, then A c-commands B at level L
The PMC requires that c-command relations generated in the base should be preserved (to the extent that this is possible). If the PMC is interpreted as a principle
governing phonetic realizations, pull splits will be generated.
Having identified the two principles of grammar that might be made be responsible for inverted and pull splits, respectively, one has to identify the traffic
rules for their interaction. Initially, one might suspect that the choice between
pull and inverted splits is correlated with the relative ranking of the relativized
minimality/A-over-A condition and the PMC in the spirit of Optimality Theory.
Structures that respect one of the two constraints inevitably violate the other.
What is grammatical and what not would thus be a function of which of the two
principles has priority over the other.
This simplistic account fails for two reasons, however. First, it predicts that
there are languages in which only pull splits exits (in which PMC outranks the
A-over-A condition), and this does not appear to be the case. Second, it ignores
the fact that the choice among pull and inverted split seems to be correlated with
the operator features involved at least in German. When the split XP involves
a wh-feature and a topic/focus feature, the choice of split type must reflect the
hierarchical relations among the attracting heads:
Bcher weiss ich nicht wieviel
er gelesen hat.
books know I not how many he read has
as for books, I do not know how many of them he has read
denkst du dass er tglich Bcher liest?
b. wieviel
how many think you that he daily books reads
how many books do you think that he reads every day?
(69) a.
When no wh-feature is involved, splits are inverted. This observation suggests a refinement of an assumption made above. Recall that the A-over-A-condition and/or
the Minimal Link Condition affect features only that are identical from a grammatical perspective. The distribution of split types in German suggests that topic and
focus features are identical from the perspective of the Minimal Link Condition,
while the wh-feature is different from the topic-focus feature. Therefore, wh-splits
do not have to be inverted.
For Croatian (and perhaps Slavic languages in general), we then only have to
add the assumption that topic and focus features may optionally be treated as distinct. If they are, the A-over-A condition/the MLC will no longer force an inverted
serialization of the split construction, as required.
[das wievielte
Buch] ist das?
the how-many-eth book is that
how many books does that make
b. [ein wie teueres
Buch] hat sie gekauft?
a
how expensive book has she bought
how expensive a book has she bought
b . [wessen Buch] hat er gekauft
whose book has he bought
b . [wem sein Buch] hat er gekauft
who his book has he bought
(sie wollte wissen) (she wanted to know)
c. [den wievielten
Geburtstag] er heute feiert
the how-many-eth birthday
he today celebrates
how old did he get today
d. [ein welcher Student] das geschrieben hat
a
which student that written
has
e.
(dialectal)
(dialectal)
On the other hand, there is no pied-piping for features that follow the noun, that
is, for features c-commanded by the lexical noun.
(71) es ist egal it does not matter
a. *einen Bruder von wem
a
brother of whom
whose brother she loves
b. *einen Bruder wessen sie
a
brother whose she
sie liebt
she loves
liebt
loves (=a.)
Distributed deletion
(72) a.
Distributed deletion
er
he
b. er
he
c. er
he
hat
has
hat
has
hat
has
kein Geld.
no money
keines/*kein.
none
keines/*kein aus Deutschland.
none
from Germany
(79) a.
. Overt determiners
The discussion in the preceding paragraph helps to understand a fundamental restriction concerning the formation of pure DP-splits in German. As had already
been observed in the early work concerning DP-splits (Fanselow 1988; van Riemsdijk 1989), DP-splits can be wellformed in certain varieties of German only if the
split phrase is a plural DP, or is projected from a mass noun. Thus, contrasts such
as (82) can be observed:
(82) a. *Alten Professor kennt sie keinen
old professor knows she no
she knows no old professor
b. Alte Professoren kennt sie keine
old professors knows she no
she knows no old professors
Distributed deletion
ich
I
b. *ich
I
c. ich
I
kenne
know
kenne
know
kenne
know
Professoren
professors
Professor
professor
einen Professor
a
professor
Whatever the nature of the restriction exemplified in (84) is, it creates a problem
for the phonetic realization of a split construction [[einen Professor] . . .. [einen
Professor] . . ..], because it implies in conjunction with (81) that the first occurrence
of the copied DP cannot be pronounced without a determiner. In this situation van
Riemsdijks regeneration idea comes into play. Since singular count nouns do not
constitute well-formed DPs by themselves, alten Professoren in (82a) must not be
part of a split DP. The problem can be circumvented, however, by inserting an
indefinite article into the determiner position of the left copy of the DP.
(82) c.
If we follow the standard idea that keinen is the spellout of a negative operator
merged with an indefinite determiner, the following description seems natural: An
abstract DP [neg [indef [alt [professor]]]] is copied to two operator positions. If
two operator features are present, the constellation [[neg [indef [alt [professor]]]]
[neg [indef [alt [professor]]]] . . .. ] arises, in which the abstract morphemes neg
and indef have been marked as [-pronounced] in the lefthand copy. This implies a
conflict between requirement (85) blocking the realization of singular count DPs
without overt determiners, and the pronunciation principle (86) that requires that
no material be pronounced twice. If the former principle is stronger than the latter,
the [-pronounced] instruction for abstract [indef] is ignored in the left copy this
is the most economical way of respecting DP well-formedness. Consequently, [indef] is pronounced as einen as in (82c). This happens in the dialect of most speakers of German. The minority dialect ranks pronunciation economy (86) (=nonpronunciation of indef ) higher than the determiner requirement (85a). For these
speakers, (82a) is grammatical.
(85) Singular count DPs start with a determiner
(86) Do not pronounce material twice
This dialectal difference constitutes an aspect of the split construction that is easy
to account for in Optimality Theory. The same holds for (87). This sentence has
been rated as ungrammatical by only 2 of the 45 consultants (9 found the sentence questionable, and 34 grammatical) which is surprising given the number
mismatch between the left and the right part of the split DP.
(87) Zeitungen liest er nur eine die taz
newspapers reads he only one the taz
As for newspapers, he only reads one: the taz
Such constructions are grammatical only if the left DP is plural, and the right DP
singular. The reverse constellation is strongly ungrammatical. Assume that the DP
that was merged originally is nur eine Zeitung only one newspaper-sg or rather
a constellation of abstract morphemes corresponding to that. After copying and
partial deletion, the configuration (88) arises. The left copy in (88) violates (85).
One way of dealing with the problem is to insert an article (if (85) outranks (86)),
which leads to (90b), while the minority dialect tolerates (90a) since (86) (85).
But there is a further way of dealing with the problem constituted by (85): one
can realize the lefthand DP in the plural, so that (85) is not violated at all. Such a
strategy obviously violates a further principle of spellout: abstract formal features
should find the proper phonetic realization. But if (85) (89), the slight deviation
from the input is warranted.
(88) [DP zeitung, sg] liest er [DP nur eine]
(89) Feature Faith:
The Phonetic Realization must respect the formal features of the input
Zeitung
liest er nur eine
newspaper-sg reads he only one
liest er nur eine einzige
b. eine Zeitung
a
newspaper reads he only a
single
(90) a.
(91) exemplifies a number of puzzles that arise in the context of DP splits in German. The structures exemplified in (91) have been used as arguments in favor of
Distributed deletion
base-generation by Fanselow (1988, 1993), but they can be dealt with successfully
in the present theory as well if one assumes more pronunciation principles like
(85). Thus, the order of the words in the two copies of the split DPs is not fully
inverted sometimes. The sequence that keine nur Bcher that underlies (91a) in
our account is ungrammatical, as (91b) shows, because nur only must be leftperipheral in a DP, while relative clauses (91c, d) have to appear at the right edge.
Does this argue against deriving (91a, c) from the sources like (91b, d)? If the two
serialization constraints just mentioned do not govern the construction process of
noun phrases, but rather apply to DPs in isolation at surface structure, the contrast
in (91ad) is explained: by splitting it up, the DP loses its offending properties.
(91) a.
b.
c.
d.
e.
f.
Likewise, welche some cannot co-occur with an overt noun in German, a problem that welche Bcher manages to solve in (91e) by splitting up.
. Loose ends
. The EnglishGerman/Slavic contrast
So far, we have focused on problems that may arise in the left copy of a split DP.
The restrictions affecting the righthand copy of an inverted split seem to have more
severe consequences.
In an inverted DP-split, the right copy has no overt nominal head. In German and Slavic, this cannot create a problem because the overt realization of a
noun in DP is never necessary. Likewise, Warlbiri noun phrases need no overt
noun (Hale 1983), the same holds for West Greenlandic (Fortescue 1984), Latin
(Khner-Stegman 1976, 61, 247; Ostafin 1986), and Dyirbal and Yidi (Dixon
1972, 1979). The option for omitting a noun, that is, the option for an ellipsis of
the complement of some functional category in the DP, is certainly related to the
strength of agreement in the noun phrase in these languages. Lobeck (1991) suggests that only agreeing functional heads permit ellipsis of their complements. Languages tolerating noun ellipsis allow a split construction. In contrast, most English
noun phrases need an overt nominal head. The literal translation (92b) of German
(92a) is ungrammatical the empty nominal position must be filled by one, as in
(92c). No XP-splits exist (92d). Noun ellipsis is also impossible in Japanese, which
has no movement-based split construction either.
(92) a.
b.
c.
d.
Fanselow (1988) tries to derive the grammaticality of XP-splits from the independent existence of DPs lacking an overt noun. Because of the repair strategies discussed in 5.2, the present model does not correlate XP-splits and the existence of
noun phrases without nouns.
. PP-splits
Since local wellformedness requirements as discussed in 5.2. imply that the parts of
a split DP should come as close as possible to the shape that complete independent
DPs have, it is natural to suspect that the same holds for split PPs. (93) seems to be
an obvious and trivial condition for the phonetic realization of PPs. It implies that
PPs cannot be split in the strict sense (94): only one of the two copies can fulfill
(93) if distributed deletion is maximal.
(93) Left Edge of PP
PPs begin with an overt preposition
(94) *Bcher hat er in keine geschaut
books has he in no looked
He has not looked into any books.
Just as in the cases discusses in 5.2., the problem can be repaired by choosing a less
economical pronunciation, that is, by realizing the preposition in both copies:
(95) in Schlssern habe ich noch in keinen gewohnt
in castles
have I yet in no
lived
So far, I have not yet lived in any castle.
However, two copies of the preposition are retained in inverted splits only, and not
in pull splits, as (96) shows:
Distributed deletion
(96) Na kakvo
se Ivan stablo penje?
on what-kind-of self I.
tree climbs
On what kind of tree does Ivan climb?
(Croatian)
. A mystery
Some speakers of German (10 out of the 45 informants)26 find structures such
as (97a) unobjectionable a construction which cannot be integrated easily into
the present framework because no speaker of German accepts noun phrases with
more than one nominal head, that is, *nur Bussarde Raubvgel is completely ungrammatical.
#
Raubvgel
kennt Gereon nur Bussarde
birds of prey knows Gereon only buzzards
As for knowing birds of prey, Gereon knows just buzzards.
b. *er kennt nur Bussarde Raubvgel
greifen den Gereon immer nur Bussarde an
c. # Raubvgel
birds of prey attack the Gereon always just buzzards ptc
as for being attacked by birds of prey Gereon is always attacked by
buzzards only
(97) a.
Similar problems arise in the analysis of noun incorporation, as Mithun (1984: 870)
and Anderson (2000b) point out: the incorporation of fish does not preclude the
appearance of a head noun in the object DP in Mohawk (98).
(98) shat:ku snik:ti rabahbt wahutsyahn:nu ki
rakeniha
eight
of.them bullhead he.fish.bought thus my.father
my father bought eight bullheads
How can these structures be analyzed in incorporation models? In many languages,
N2 frequently bears a possessor relation to the incorporated noun N1 in the incorporation structure [VP [V V-N1] [DP . . . N2 . . . ]]. We may analyse such a constel-
Gisbert Fanselow and Damir Cavar
. Other constructions
It is tempting to explain a further construction type that is characterized by properties much similar to the one we have discussed here by distributed deletion, namely
extraposition. Haider (1997) notes a number of problems concerning the assumption of an extraposition operation for CPs in German: clauses in extraposed position are not barriers, as they should be in a derived position (but see Mller 1998
for considerations weakening this argument), they are c-commanded by the elements preceding them according to evidence involving polarity items, and the
movement that extracts them out of their host noun phrase would violate conditions on movement more often than not: In (100a), the relative clause would have
been moved out of a PP. This problem shows up with the apparent extraposition
of PPs, too, as (100b) shows.
gedacht, die Bcher liest
ich habe an eine Frau
woman thought who books reads
I have at a
I have thought about a woman who reads books
b. ich habe ber den Titel nachgedacht von deinem Buch
of your
book
I have about the title thought
I have reflected about the title of your book
(100) a.
Distributed deletion
Notes
* This article has its roots in a joint presentation at the 1997 International Conference on
Pied Piping held at the Friedrich Schiller University Jena. Parts of the paper have been presented at workshops and conferences at the universities in Leipzig, Osijek, Poznan, and
Stuttgart. For inspiring discussions and helpful hints, we are grateful to Artemis Alexiadou, Josef Bayer, Caroline Fry, Gereon Mller, Henk van Riemsdijk, Peter Staudacher, and
Masatoshi Tanaka. A particular thank goes to Anoop Mahajan, who suggested distributed
deletion as an analysis of split DPs in the discussion period of a 1995 talk by Fanselow. The
research reported here was partially supported by grants of the German Research Foundation (DFG) to the Innovationskolleg Formale Modelle kognitiver Komplexitt (INK 12) at
the University of Potsdam, and the Graduiertenkolleg konomie und Komplexitt in der
Sprache at the Humboldt University Berlin and the University of Potsdam.
. (5) and (6) belong to the class of separation constructions in the sense of Pesetsky
(2000). Split topicalization and split scrambling have been used as further labels for (5)
and (6) in the literature. The Slavic constructions have also been discussed as a subtopic of
left-branch extractions.
. Speakers differ in the extent to which they accept or reject multiple splits presumably,
because multiple splits involve a highly complex pragmatic structure. Furthermore, multiple
splits necessarily involve one phrase which is completely split within IP. Such splits were
considered ungrammatical in the early literature (see, e.g., Fanselow 1988; Kniffka 1996),
but such claims were based on data that were constructed in a less than optimal way.
. The existence of this construction type has been brought to our attention by Josef Bayer.
Gisbert Fanselow and Damir Cavar
. See Kniffka (1996) for an assessment of the dialectal distribution of imperfect DP splits
with two determiners.
. More precisely, Riemsdijk (1989) assumes that the clause initial position of German,
nowadays the specifier of CP, can host maximal projections only, and considers amerikanische Wagen not to be one certainly a necessary assumption in models of noun phrase
structure that did not assume the fine functional structure related to the DP models.
. Of course, the problem will not be solved if we assume extraction of non-constituents.
. In a questionnaire study carried out together with Reinhold Kliegl and Matthias Schlesewsky, we found a certain nestedness of judgments: there are some (few) speakers who
accept splitting for accusative noun phrases only, others accept splits of nominative and accusative noun phrases, and a third group accepts discontinuity for accusative, nominative,
and dative phrases. The nesting of the judgements in our questionnaire study reflects the
development of judgements in the literature, to a certain extent.
. DeKuthy (2000) has argued that German noun phrases are islands for extractions of PPs.
Structures showing apparent PP-extraction from NP involve an underlying structure [VP NP
PP V] in her approach. A discussion of this view is beyond the scope of the present paper.
Note that our major point is not affected if her analysis is correct: movement processes
have to obey the island conditions, independent of whether the construction one wants
to compare XP splits with is a movement construction or not. In fact, we might say that
our general point would rather be strengthened, because it would be fairly unclear why noun
phrases should be islands for extractions of PPs but not for extractions of, say, NomPs. Given
that splits affect subjects and indirect objects, a reanalysis option is ruled out as an account
for split constituents immediately, because reanalysis processes are assumed to involve direct
objects only (if the process exists at all).
. The only way to counter this argument against the movement approach of split constituents would be to claim that the bounding theory does not hold for the extraction of
XPs from a YP that is an extended projection of XP. We do not think that such a proposal
could be spelt out in a convincing way, and it would be incompatible with the observation
that the extraction of VPs out of IPs or CPs essentially respects bounding theory.
. Doing so may in fact be simple: if an XP can be linked thematically to predicate P only
if XP is merged in the projection of P, then two XPs sharing a thematic role must be merged
in the same maximal projection.
Alternatively, one can assume that one part of the DP merges in VP, the other in the projection of the functional head which licenses the formal features of the DP. If, as argued in
Fanselow (2001a, in press) the checking of certain formal features implies theta-role assignment, and if the two DP-parts both check features with the relevant functional head, they
share a thematic role and their relation is correctly predicted to be local one. In the interest
of space, we will not pursue this idea here.
. Note, however, that there are also word order facts that are unexpected in simple movement theories at least, as we show in Section 5.
. Given the freedom of word order in Slavic noun phrases, no similar argument can be
made easily for noun phrase splits in Slavic. For PP splits, the relevant point is obvious
Distributed deletion
however; base generation does not readily explain why the highest part of a PP must contain
the preposition.
. In this respect, the present account is much in line with base generation theories as
proposed in Fanselow (1988) or van Geenhoven (1998): there is movement, but splitting
itself is not caused by extraction from something.
. Pesetsky (2000) argues that featural movement of the kind introduced in Chomsky
(1995) is nevertheless necessary, in addition to the phonological deletion of the upstairs
copy. If Chomsky (2000) is correct in replacing feature movement by agreement at a distance, the analysis sketched above is, of course, the only kind of covert movement.
. For a similar approach developed independently of us, see Hinterhlzl (1999, to appear).
. Partial deletion effects might be reanalyzed as involving (partial) reconstruction of phonetic material in the overt component: phrase first moves completely from a to b, later,
a part of is reconstructed to a. At a purely descriptive level, this approach and the theory
proposed here have fairly similar consequences.
. The effects of partial reconstruction can be reanalyzed as being due to distributed deletion after LF copying. An analysis of all reconstruction phenomena in terms of distributed
deletion seems possible, but is well beyond the scope of the present paper. A few remarks
can be found in Fanselow (2001b).
. In this respect, it resembles other types of NP-discontinuity, cf. DeKuthy (2000).
. Ungrammaticality may, however, be repaired for certain speakers with heavy stress.
. Note that reference to the fact that an accusative NP occupies a non-derived focus position cannot account for the contrast discussed above: PP objects can occupy non-derived
postverbal focus positions, too, yet they cannot be split up there.
. Recall that the Minimal Link Condition (MLC) requires that K attracts only if there is
no , closer to K, such that K attracts (Chomsky 1995: 310). is closer to target K than
if c-commands (Chomsky 1995: 358).
. The A-over-A condition (Chomsky 1964) reduces to the MLC if the feature triggering
the locality effect is sitting on a syntactic head which projects this feature.
. That is, we assume that contiguity is a graded constraint.
. Croatian adds a difficulty, however, (i) shows that PPs may be split off noun phrases in
a process of partial deletion recall that dative DPs are islands for movement. At present,
we have no account for this difference between German and Croatian.
(i)
. The morphological facts of German are mirrored in other languages with DNP. In Warlbiri, noun phrases are morphologically well formed if they begin with a (possibly empty)
sequence of words not bearing Case morphemes followed by a (necessarily non-null) sequence of words (including the final one) that are Case-marked. This condition must be
respected by the parts of a DNP individually. See Nash (1980).
Gisbert Fanselow and Damir Cavar
(i)
References
Abney, S. (1987). The English Noun Phrase in its Sentential Aspects. PhD dissertation, MIT.
Anderson, S. (2000a). Towards an Optimal Account of Second-Position Phenomena. In
J. Dekkers, F. van der Leeuw, and J. van de Weijer (Eds.), Optimality Theory: Phonology,
syntax and acquisition (pp. 302333). Oxford: OUP.
Anderson, S. (2000b). Some Lexicalist Remarks on Incorporation Phenomena. In B. Stiebels
and D. Wunderlich (Eds.), The Lexicon in Focus (pp. 123142). Berlin: Akademie Verlag.
Austin, P. & J. Bresnan (1996). Non-Configurationality in Australian Aboriginal Languages.
Natural Language and Linguistic Theory, 14, 215268.
Baker, M. (1991). On Some Subject/object Non-asymmetries in Mohawk. Natural Language
and Linguistic Theory, 9, 537576.
Baker, M. (1995). The Polysynthesis Parameter. Oxford: OUP.
den Besten, H. & G. Webelhuth (1990). Stranding. In G. Grewendorf and W. Sternefeld
(Eds.), Scrambling and Barriers (pp. 7792). Amsterdam: John Benjamins.
Bokovic, . (1997). Second Position Cliticization: Syntax and/or phonology? Ms.,
University of Connecticut.
Browne, W. (1975). Serbo-Croatian Enclitics for English-Speaking Learners. In R. Filipovic
(Ed.), Contrastive Analysis of English and Serbo-Croatian. Zagreb: Institute of
Linguistics.
Browne, W. (1976). Two Wh-fronting Rules in Serbo-Croatian. Junoslovenski filolog, 32,
194204.
Cavar,
D. (1999). Aspects of the Syntax-Phonology Interface. Doctoral dissertation.
University of Potsdam.
Chomsky, N. (1964). Current Issues in Linguistic Theory. The Hague: Mouton.
Chomsky, N. (1995). The Minimalist Program. Cambridge, MA: The MIT Press.
Chomsky, N. (2000). Minimalist Inquiries: The framework. In R. Martin, D. Michaels and J.
Uriagereka (Eds.), Step by Step: Essays on minimalist syntax in honor of Howard Lasnik
(pp. 89155). Cambridge, MA: The MIT Press.
Culicover, P. & M. Rochemont (1990). Extraposition and the Complement Principle.
Linguistic Inquiry, 21, 2347.
Corver, N. (1990). The Syntax of Left Branch Extractions. Doctoral dissertation, Tilburg
University.
De Kuthy, K. (2000). Discontinuous NPs in German. A case study of the interaction of
syntax, semantics, and pragmatics. Doctoral dissertation, Saarbrcken.
Diesing, M. (1992). Indefinites. Cambridge, MA: The MIT Press.
Distributed deletion
Gisbert Fanselow and Damir Cavar
Hinterhlzl, R. (In press). Remnant Movement and Partial Deletion. In A. Alexiadou,
E. Anagnostopoulou, S. Barbiers and Hans-Martin Gaertner (Eds.), Dimensions of
Movement. Amsterdam: John Benjamins.
Hhle, T. (1996). German w...w-constructions. In U. Lutz and G. Mller (Eds.), Papers
on Wh-Scope Marking. Arbeitspapier 76 des Sonderforschungsbereich 340. Stuttgart and
Tbingen.
Horn, G. (1975). The Noun Phrase Constraint. Doctoral dissertation, University of Mass.,
Amherst.
Jelinek, E. (1984). Empty Categories, Case, and Configurationality. Natural Language and
Linguistic Theory, 2, 3976.
Kayne, R. (1994). The Antisymmetry of Syntax. Cambridge, MA: The MIT Press.
Kniffka, G. (1996). NP-Aufspaltung im Deutschen [KLAGE 31]. Hrth: Gabel.
Kuhn, J. (1998). Resource Sensitivity in the Syntax-semantics Interface and the German
Split NP Construction. In T. Kiss and D. Meurers (Eds.), Proceedings of the ESSLLI
X Workshop on Current Topics in Constraint Based Theories of Germanic Syntax.
Saarbrcken.
Kuhn, J. (To appear). The Syntax and Semantics of Split NPs in LFG. In Proceedings of CSSP
1997.
Khner, R. & C. Stegmann (1976). Ausfhrliche Grammatik der lateinischen Sprache. Zweiter
Teil: Satzlehre. Darmstadt: Wissenschaftliche Buchgesellschaft.
Lapteva, O. A. (1976). Russkij razgovornyj sintaksis Russian Colloquial Syntax. Moscow:
Nauka.
Lobeck, A. (1991). Phrase Structure of Ellipsis in English. In S. Rothstein (Ed.), Perspectives
on Phrase Structure [Syntax & Semantics 25] (pp. 81107). San Diego, CA: Academic
Press.
Massam, D. (1986). Case Theory and the Projection Principle. PhD dissertation, MIT.
Mithun, M. (1984). The Evolution of Noun Incorporation. Language, 60, 847894.
Mller, G. (1996). A Constraint on Remnant Movement. Natural Language and Linguistic
Theory, 14, 355407.
Mller, G. (1998). Incomplete Category Fronting. Dordrecht: Kluwer.
Mller, G. (2001). Order Preservation, Parallel Movement, and the Emergence of the
Unmarked. In G. Legendre, J. Grimshaw and S. Vikner (Eds.), Optimality Theoretic
Syntax. Cambridge, MA: The MIT Press. (Also on ROA ROA-275-0798.)
Nash, D. (1980). Topics in Warlpiri Grammar. New York: Garland.
Nunes, Jairo (1995). The Copy Theory of Movement and Linearization of Chains in the
Minimalist Program. Doctoral dissertation, University of Maryland.
Nunes, J. (2001). Sideward movement. Linguistic Inquiry, 32, 303344.
Ostafin, D. M. (1986). Studies in Latin Word Order: A transformational approach. Doctoral
dissertation, University of Connecticut, Storrs.
Pesetsky, D. (1998). Some Optimality Principles of Sentence Pronunciation. In P. Barbosa,
D. Fox, P. Hagstrom, M. McGinnis and D. Pesetsky (Eds.), Is the Best Good Enough?
(pp. 337383). Cambridge, MA: The MIT Press.
Pesetsky, D. (2000). Phrasal Movement and its Kin. Cambridge, MA: The MIT Press.
Pili, D. (2001). On A- and A -dislocation in the left periphery. A comparative approach to
the cartography of the CP-system. Doctoral dissertation, University of Potsdam.
Distributed deletion
Riemsdijk, Henk van (1989). Movement and Regeneration. In P. Beninc (Ed.), Dialectal
Variation and the Theory of Grammar (pp. 105136). Dordrecht: Foris.
Sabel, J. (1998). Principles and Parameters of Wh-movement. Habilitation thesis,
Frankfurt/M.
Sekerina, I. (1997). The Syntax and Processing of Scrambling Constructions in Russian.
Doctoral dissertation, CUNY.
Siewierska, A. (1984). Phrasal Discontinuity in Polish. Australian Journal of Linguistics, 4,
5771.
Sportiche, D. (1988). A Theory of Floating Quantifiers and its Corollaries for Constituent
Structure. Linguistic Inquiry, 19, 425449.
Stechow, A. von (1992). Kompositionsprinzipien und grammatische Struktur. In
P. Suchsland (Ed.), Biologische und soziale Grundlagen der Sprache (pp. 175248).
Tbingen: Niemeyer.
Tanaka, M. (In preparation). Anti-Bewegungsanalyse der japanischen Topik-Transformation im Vergleich mit dem Deutschen (working title). Doctoral dissertation,
University of Potsdam.
Tappe, T. (1989). A Note on Split Topicalization in German. In C. Bhatt, E. Lbel and
C. Schmidt (Eds.), Syntactic Phrase Structure Phenomena in Noun Phrases and Sentences
(pp. 159179). Amsterdam: John Benjamins.
Thiersch, C. (1985). VP and Scrambling in the German Mittelfeld. Ms., University of
Tilburg.
Vogel, R. & M. Steinbach (1998). The Dative an Oblique Case. Linguistische Berichte, 173,
6590.
Wilder, C. (1997). Phrasal Movement in LF: de re readings, binding and ellipsis. In
K. Kusumoto (Ed.), NELS 27 [Proceedings of the North East Linguistic Society] (pp.
425439). GSLA, Amherst.
Yearley, J. (1993). Discontinuity in the Russian Noun Phrase. Ms., University of Massachusetts, Amherst.
Zec, D. & S. Inkelas (1990). Prosodically Constrained Syntax. In S. Inkelas and D. Zec (Eds.),
The Phonology-Syntax Connection. Chicago: The University of Chicago Press.
Background
A
C
B
D
In grammatical explanation, the relation of c-command is of fundamental importance; movement is allowed if and only if the moved element c-commands
its trace, antecedents must c-command pronouns and anaphors, and so forth.
Traditionally, c-command is defined in terms of dominance: c-commands
iff every node dominating dominates (and neither dominates the other).
However, the dominance relation is much less central in syntactic explanation than c-command. In fact, our hypothesis here (following Frank & VijayShanker 2001) is that syntactic structures ought to be characterized directly in
terms of a primitive c-command relation, as opposed to a primitive dominance
relation.
Frank and Vijay-Shanker (2001) show that the class of tree structures that
can be characterized in terms of primitive c-command is a subclass of those
that can be characterized with primitive dominance including all of the lin-
x y iff
xCy
z [ z C x
zCy]
There are two parts to the definition in (2). First, it says that x can only be less
embedded than y if y is indeed embedded; if x c-commands y, then y is certainly not embedded below x x could not dominate y in traditional terms.
The second part says that x can only be less embedded than y if everything
A
C
B
D
In terms of primitive c-command, this structure is characterized by the following set of c-command relations:
(3) D
B
B
B
mutually c-commands
mutually c-commands
c-commands
c-commands
E (D c-commands E, E c-commands D)
C (B c-commands C, C c-commands E)
D
E
Given the definitions in (2), we can see that C is less embedded than D here
because (i) C does not c-command D, and (ii) everything that c-commands
C (namely, B) also c-commands D. The reverse does not hold; D is not less
embedded than C because there is something which c-commands D (namely,
E) that does not c-command C. By the same reasoning, we can see that A is
less embedded than B but not vice-versa, since (vacuously) everything that ccommands A also c-commands B, but there is a node (C) which c-commands
B but not A.
To identify the root node of (1), we can use the intuitive idea that the root
is the least embedded node in terms of this formal definition of relative embeddedness and conclude that A is the root; for any node N in (1), A is less
embedded than N and N is not less embedded than A. The complete list of
c-command relations from (1) is given in (3).
At least for the case of the structure (1), the definition of less embedded in
(2) gives us the same results we would have had using traditional dominance.
C
C
B
D
(5) D
B
B
B
mutually c-commands
c-commands
c-commands
c-commands
E (D c-commands E, E c-commands D)
C
D
E
Running through the same sort of calculation as before, we discover unsurprisingly that C is less embedded than both D and E: C does not c-command
either D or E, and the only node c-commanding C, namely B, also c-commands
both D and E. More interestingly, we also see that B is not less embedded than
C, since B c-commands C, but neither is C less embedded than B since there is
a node (B) which c-commands C but not B. The question then arises: What is
the root of the structure in (4)? There is no node which is less embedded than
all other nodes; it cannot be C (since C is not less embedded than B), nor can
it be B (since B is not less embedded than C). We would not like to consider
(4) to be a rootless structure, since it must always be possible to determine the
category of such a tree, and this is one of the functions of the root node.
There is a slightly weaker way we can think of the root of a structure. There
are two nodes in (4) for which we can say that there is at least no other node
less embedded than it. There is no node less embedded than B, nor is there
a node less embedded than C. Following this idea, we define the root node
as follows.
(6) A node N is a root iff
there is no other node M such that
M is less embedded than N.
N is a root
M.M N
Using this definition, both B and C are roots of (4). In fact, generally, this way
of looking at roots means that more adjuncts directly leads to more roots in
the structure. So, in a structure like (7), the roots are C, F, and B.
(7)
C
C
F
B
C
D
However, even in multi-rooted structures like (7), two of these roots have
a distinguished status. Specifically, C is the root which does not c-command
any other nodes (whereas both B and F c-command C), and F is the root
which is not c-commanded by any other nodes (whereas both B and C are
c-commanded by F).
Recall that there are two things that the root is important for: determining
the label or category of the tree as a whole, and determining the site of cyclic
attachment. In (7), C is the node which determines the category, so it is good
that we can distinguish C from among the three roots of (7).
(8) The categorial root is
the root which does not c-command any other nodes.
Since we can also distinguish F in an equally general way, we expect to see that F
plays a similarly important role. We will see shortly that it plays the other major
role of the traditional root node, determining the site of cyclic attachment.
Before we make this connection, however, we must take a detour to discuss
how these structures come about derivationally.
Substitution
Adjunction, on the other hand, as in (10), only asserts c-command in one direction; so, for every pair of roots (one from <X> and one from <Y> and
where the <X> subtree is being adjoined to the <Y> subtree), the <X> root
will c-command the <Y> root.
(10)
Adjunction
Notice too that if we accept that structures are combined using c-command
relations (only), then these two modes of attachment, substitution and adjunction, exhaust the logically possible means of combination: c-command can be
established either in both directions or in just one direction.
Another interesting result follows from this and from a natural wellformedness condition on trees stated in (11), requiring that the categorical
status of the tree be determinable.
(11) Categorial Identity Condition
A well-formed tree has a unique category-determining root
(a categorial root).
Recall that the category of the tree is determined by the categorial root, the
root that does not c-command any other nodes. This means that an adjunction to a tree cannot change the well-formedness of that tree with respect to
(11). In (10), when <X> is adjoined to <Y>, the categorial root of <Y>
does not c-command any new nodes (since c-command is only asserted from
<X> roots to <Y> roots), so it remains the root which does not c-command
any other nodes. The <X> root(s) on the other hand now (each) c-command
the root(s) of <Y> and so can no longer serve as categorial roots.
Substitution, however, establishes a mutual c-command relation between
the roots of the two subtrees. Saying nothing further, the result of (9) would
be a tree which does not satisfy the Categorial Identity Condition (11). In order
to have a well-formed tree, a new node must be added to the representation,
one which does not c-command any other nodes. This new node is the label
of the combined tree; it is the node which projects (in the terminology of
Chomsky 1995).
As a consequence of this interaction between the need for a categorial
root and the mechanisms underlying adjunction and substitution, we no
longer have any need to posit a distinction between segments and categories (May 1977; Chomsky 1986). In effect, segments dont exist; no new
node is necessary in an adjunction structure. Only under substitution is a new
node added to the representation.
YP
WP
YP
Y
(13)
ZP
XP
X
XP
YP
WP
X
YP
YP
WP
YP
ZP
As a result of the need to satisfy the Categorial Identity Condition, a node (XP)
must be added to the structure. At this point XP is not in a c-command relation
with any other node. It becomes the only root of the resulting structure. One
question that arises in this connection is how the properties of the newly projected node are determined; that is, why in (13) does the node labeled XP take
its features from X rather than YP? One possibility is that the choice is free in
the syntax, with incorrect choices being filtered out by uninterpretability at the
LF interface. An alternative, suggested by Chomsky (2000), is that the possibility of substitution is regulated by the existence of a selection relation. Since this
mode of combination does not impose any structural asymmetry, there must
be some substantive asymmetrical relation between the combined elements, a
relation that Chomsky takes to be selection.3
YP
Y
ZP
(15)
YP
YP
XP
XP
YP
WP
YP
Y
YP
WP
YP
Y
ZP
The interesting thing about the structure in (15) is that the c-command relations are the same as those in the structure in (16). To put it another way, (15)
is nondistinct from (16).
(16)
YP
WP
XP
YP
WP
ZP
We are used to thinking of (16) as coming about as a result of a different derivational history, one in which XP first adjoins to WP and then the complex adjoins to YP (e.g., in the proposals about multiple wh-movement in Ackema
& Neeleman 1998; Grewendorf & Sabel 1996). People have argued for structures like (16), although the structures have always appeared somewhat odd,
seemingly resulting from movement which targets things which should not,
cyclically speaking, be targets.
However, if representations are described only in terms of c-command, as proposed here, no such operation is required, since the structures are already
nondistinct.
Another well-known example of a structure like (16) was proposed by
Rudin (1988) to account for the properties of multiple wh-movement in certain Slavic languages, including Bulgarian and Romanian. In these languages,
all wh-words move to the front of their clause, forming an unbreakable constituent. Thus, (18b) is degraded compared to (18a) because the wh-words are
separated by the adverb vchera yesterday.
e udaril
koj kogo vchera
who whom yesterday has hit
Who hit whom yesterday?
kogo e udaril
b. ??koj vchera
who yesterday whom has hit
Who hit whom yesterday? (Marina Todorova, p.c., cf. Rudin 1988)
(18) a.
Rudin (1988) analyzed this as multiple adjunction of wh-words to SpecCP, although from our understanding of other cases of wh-movement, we expect to
find wh-words moving to, or targeting, CP itself, rather than SpecCP. Additionally, movement that targets SpecCP also constitutes a violation of cyclicity when understood in terms of the extension condition of Chomsky (1992).
From our perspective, however, these two apparently distinct sorts of movement are actually the same (under the further assumption that there is no distinction between specifiers and adjuncts, e.g. as argued by Kayne 1995). We can
thereby avoid the need to posit a movement that targets an embedded element.
On the basis of this discussion, one might be tempted to conclude that in
our system the locus of attachment for a second adjunction is indeterminate
among the multiple roots, with all possibilities yielding identical results. The
Bulgarian data just reviewed, however, indicate that that multiply fronted whelements form a single, unbreakable constituent, suggesting that each successive adjunct attaches to the one that precedes it. Observe that the first adjunct,
WP in (14), is not only a root of its structure, but is in fact the unique root
We will see that the attachment root provides the locus of cyclic attachment (the role of the traditional notion of root not covered by the categorial root).4
XP
WP
XP
ZP
X
Y
The complex head, considered alone as a subtree, has two roots, X and Y.
Recall that the definition of root is based on the intuition that roots are minimally embedded; this means that in (20), X and Y are (equally) minimally
embedded in the complex head subtree.
There is reason to think that both X and Y are local enough to WP in
SpecXP to check features. For example, consider the licensing conditions on Nwords like French personne. As is well-known, personne requires the presence of
a local negative element. This requirement can be instantiated by the assertion
that personne contains an uninterpretable negative feature ([Neg]) that must be
checked for convergence. We assume that the negative head ne also contains an
instance of [Neg] that is capable of checking personnes feature. Following Pollock (1989) in the assumption that the negative head ne is generated between
T and V, this would imply that in an example like (21), the Neg head adjoins to
T (perhaps in a complex together with the verb), after which personne moves
to SpecTP.5
(21) Personne nest venu.
No one neg-is came
No one came.
Such a derivation, combined with the empirical fact in (21), entails that the
[Neg] feature of ne is close enough to personne in SpecTP to check, despite
the fact that the [Neg] feature is part of a complex head with T and the copula (nest).
Returning to (20), notice that, once this complex head <X> (containing X
and Y) has been merged with the complement <ZP> (the subtree with categorial root ZP), both X and Y c-command <ZP> (X and Y are both roots of
<X> and merging asserts mutual c-command relations between every root of
<X> and every root of <ZP>). However, thinking further back in the derivation, if Y had moved to adjoin to X from within <ZP>, this means that Y still
c-commands its trace.6 Looking at it this way allows us to maintain the view
that a moved element must c-command its trace.
CATEGORIAL
ROOT
ATTACHMENT ROOT
. Constituents
So far, we have concentrated on showing that although dominance was useful
for the purposes of identifying the root of a structure, it conflated two notions of root which the proposed c-command-based view distinguishes (the
site of cyclic attachment and the categorial root). We conclude that there
is no need to refer to the dominance relation for the purposes of root de-
The question we will address in the next few sections is what becomes of the
concept of constituent if we re-interpret it in terms of c-command. We will
see that we can maintain a close-to-traditional view of constituency, while at
the same time providing insight into certain phenomena that have remained
puzzling under the traditional view.
Consider the subtree in (24). We want to be sure that whatever our new
interpretation of constituency is, it gives us the constituents listed in (25).
(24)
AP
A
BP
B
CP
(AP)
(A)
(BP)
(B)
(CP)
A natural place to begin is to carry over the notion of less embedded, used
in the preceding sections to approximate dominance for the purposes of determining roots. The idea would be to define a constituent something like
in (26). For reasons we will turn to directly, however, (26) is insufficient as a
definition of constituent.
(26) Revised definition of constituent (first attempt):
For a node r, Constituent(r) = { m | r is less embedded than m }
For the structure in (24), the definition of constituent given in (26) gets the correct results. The nodes picked out by AP are {AP, A, BP, B, CP}, those picked
out by BP are {BP, B, CP}, that picked out by B is {B}, and so forth. The defini-
tion in (26) runs into problems when we consider adjunction structures, such
as (27) below.
(27)
ZP
Z
Y
WP
Z
Recall from earlier discussion that Y is not less embedded than Z, nor vice
versa. As a consequence of this, there is no node r that will pick out exactly the
nodes {Y, Z} under the definition in (26). Rather, we find that the constituent
picked out by Z is {Z} and that picked out by Y is {Y}. This runs counter what
we know about the structure of complex heads, however. Specifically, the members of a complex head move together in iterated head movement, so they must
form a constituent (assuming that movement can only involve constituents).
The facts lead us to expect Z to pick out the constituent {Y, Z}, and the failure
of (26) to provide this result is fatal.
(28)
XP
X
Y
Z
YP
X
ZP
t
WP
We presented (26) as a possibility because it is a natural extension of the preceding discussion, but thinking about constituency in terms of c-command,
there is an equally natural alternative conception of constituent: A constituent
is the collection of nodes that are picked out by some specific node by virtue
of being c-commanded by that node. That is, as in (29) below, A picks out the
constituent {B, C, D} because these are the nodes that A c-commands. Moreover, no node picks out exactly the set of nodes {A, C, D}, which is the desired
result since {A, C, D} is not a constituent.
(29)
F
A
B
C
This definition is nearly correct, but there is one case that requires consideration, illustrated in (30). Were we to be faced with a ternary branching structure,
such a definition would incorrectly allow two branches together to count as a
constituent, picked out via c-command by the third.
(30)
*
A
F
B
A slight refinement solves this problem; we need only require that the root
of the constituent (C, above) does not c-command any of the elements of the
constituent (specifically, excluding B from the constituent). The final definition
of constituent we will adopt is given below in (31).
(31) Revised definition of Constituent (final version):
constituent(r)= {n | g C n and r C n} for a node g
such that r C g and g C r.
Walking through (31), it says that the constituent with root r includes
those nodes which are c-commanded (or picked out) by a certain node g
and are not c-commanded by r, where g is in a mutual c-command relation
with the root r.
Also, notice what this node g in (31) is, structurally. It is, essentially, the
governor of the constituent. We will return to the significance of this shortly,
but first let us try the definition (31) on the adjunction structure that caused
trouble for the previous potential definition back in (26). The final definition
of constituent yields the constituents listed in (32) for (27).
(27) (repeated)
ZP
Z
Y
(32) constituent()
ZP
Y
Z
WP
WP
Z
is
{Y}
{Y, Z}
{WP}
governor
WP
WP
Y, Z
YP
WP
XP
YP
WP
(34) constituent()
YP
WP
XP
Y
ZP
ZP
is
{XP, WP, YP, Y, ZP}
{XP, WP}
{XP}
{Y}
{ZP}
governor
A
A
A
ZP
Y
The constituency in (34) is consistent with (16) (i.e. (33)) but not with (15).7
In the remainder of this section, we consider the restrictions on certain subcases of movement, but the explanations of each will conform to a common
theme: where movement of a certain part of a tree is not allowed, it is because
that part of the tree does not constitute a constituent.
. Head movement
The first case we will consider is that of head movement. Recall from the list in
(32) that the X0 -level constituents are {Z, Y} and {Y}, but not {Z}.
(27) (repeated)
ZP
Z
WP
YP
Y
XP
WP
(X')
XP
X
ZP
What would it mean to be able to move X (the lower XP) in the structure
above? It would mean that there would have to be a constituent which contains XP, X, and ZP but to the exclusion of WP. There is, however, no such
whose features project the (categorial) root of S . We might expect that F (or
more properly its projection) should then Merge with S . If, however, we require that this S be identified as a constituent, such an instance of Merge
will be impossible, as there is no governor for S .10
As things stand so far, we also predict that XP-adjuncts and specifiers
should be constituents and thus should be capable of movement quite generally. For example, in (36), constituent(WP) = {WP} with governor Y, analogous to excorporation from a complex head as discussed above. However, we
know empirically that movement of specifiers and adjuncts is much more limited than movement of complements. This is an issue which we will discuss in
some depth after summarizing the basic proposal with respect to constituency.
. Conclusion
In the preceding sections, we have proposed that the constituent status of each
of the following entities is as listed below:
An X0 and everything adjoined to it (X0max , in the
terminology of Chomsky 1995) is a constituent.
b. A maximal projection is a constituent
(except at the root).
c. A non-maximal projection is not a constituent.
(37) a.
(38)
YP
Y
XP
WP
XP
X
ZP
We will also include in our set of potential governors certain functional heads,
e.g. empty complementizer and T, so as to allow for extraction of subjects (from
that-less CPs) and adjuncts, respectively. Under this view, the source of that-t
effects is the inability of that to define a constituent, leaving the DP in (43)
as a non-constituent.
(43)
CP
TP
C
that
DP
who
TP
T
VP
Notice that this also implies that the TP complement here is not itself a constituent either, since the C node that would also play a crucial role in defining the TP constituent, if it existed. However, there is at least some evidence to
support the hypothesis such TP complements are not constituents.
First, as seen in (44), note that TP, when the complement of that, cannot be
extracted. Second, such a TP cannot be elided, as shown in (45) (cf. Lobecks
1995 proposal that elided constituents must be governed; under our proposal
this would simplify to only constituents can be elided).
(44) *John left, I was told that t.
(45) *Even though Mary hopes that e,
she doubts that Bill is coming to the party.
Similar behavior is observed with the wh-complementizer if, which also fails to
license extraction from the specifier of its TP complement:
(46) *John finished his novel, I asked Mary if t.
(47) *Even though Mary asked if e,
she is pretty sure that Bill wont be coming.
In the presence of empty complementizers, which license extraction of embedded subjects, both the extraction and elision of TP are possible:11
(48) John left, I was told [CP [C ] t ].
(49) Even though Mary asked who [CP [C ] e],
she is sure that Bill is bringing Louise.
One potential problem with the hypothesis that TP is not a constituent is the
fact that TPs appear to be conjoinable (50).
(50) I would never have believed that
VP
V
CP
WH
CP
C
TP
Syntactic heads have selectional properties which specify what type of complement they require. When considering the complement of V in (51), what
property does it have that satisfies the selectional requirements of V? Certainly,
selection must attend to the categorial root, which determines that <CP>
has the categorial properties of a CP. Suppose, however, that there are two kinds
of selection. The first dictates properties of the root(s) of a proper complement, the second dictates properties of the categorial root (only) of a proper
complement.
Consider, as an example, the difference between say (a bridge verb) and
whisper (a non-bridge verb). If we suppose that say has the property that its
complement must be a CP (interpreted as meaning that the categorial root
of its complement must have the categorial features of a CP), whereas whisper
has the (stricter) property that (all of) the root(s) of its complement must
have the categorial features of a CP (52), we derive the result that whisper does
not tolerate successive-cyclic extraction through the specifier of its CP complement. Specifically, if a wh-word left an intermediate trace in the specifier of a
As a second example of how this distinction between selection for a root vs.
selection for a categorial root, consider the difference between raising and
control verbs.
(54) Selectional properties of raising verbs vs. control verbs
a. seem categorial root: TP
b. try
root: TP
YP
Y
XP
WP
XP
X
ZP
X
(57)
XP
CATEGORIAL ROOT
YP
WP
YP
WP
ZP
Y
ATTACHMENT ROOT of
XP
of <YP>
YP
YP
<YP>
Notice in particular that WP does not c-command X, which means that X cannot serve as a governor to define WP as a constituent. That is, WP is not a
constituent in (57). This is a step toward the solution of our problem (that
is, how to restrict the movement of specifiers), since specifiers and complements are no longer symmetrical; specifiers are not necessarily constituents,
whereas complements are. If only constituents can move, then we are not
guaranteed to be able to move a specifier.
Of course, we must not stop here, since it is certainly not the case (empirically) that specifiers can never move. Given that specifiers can sometimes move,
and continuing to assume that only constituents can move, it must be the
case that WP can become a constituent in certain circumstances. For WP to
become a constituent means essentially that WP must come to c-command
X. The question is: How might this come about?
Our suggestion is (somewhat paradoxically) movement. That is, as illustrated in (58), if WP were moved higher, the WP would c-command X. At this
point, WP would be a constituent (with governor X), since WP is in a mutual c-command relation with X. Put another way, movement is allowed if in
the end what moved is a constituent.
(58)
XP
XP
WP
X
WP X
YP
WP
YP
Lastly, what is the status of the principle that says only constituents can
move in light of the hypothesis that something can become a constituent
via movement? One possibility is that the notion of constituent is not a fundamental part of syntax, but rather an important requirement on the part
of pronunciation. That is, perhaps constituency is verified at and useful for
PF, while syntax (and even perhaps LF) are not sensitive to issues of constituency. If this speculation is on the right track, we would expect to find
that post-Spell-Out movement (of specifiers, for example) would be unconstrained compared to overt movement (since it would no longer be important to move only constituents. This also raises an interesting possible take on
why the phase would be the relevant locality domain as well. Suppose that
constituency is necessary for pronunciation (e.g., to determine the location of
prosodic boundaries), and suppose that the linguistic structure is handed off
to the PF interface at every phase. We might then conclude that phase-internal
movement of an XP is required for pronunciation if the XP in its base position
is not itself a constituent; this predicts a difference between direct objects
and other arguments (which are presumed to be introduced into the structure in specifier position) in whether they must undergo phase-internal movement. Notice also that this need not be movement to the edge of the phase;
it need only be movement to a position c-commanding the would-be governor, but within the phase. This makes a slightly different set of predictions
about the impact of phases on derivations than the view proposed in Chomsky
(2000) (for example, it predicts that direct objects need not move successivecyclically, while subjects, adjuncts, and indirect objects must, and it predicts
that movement to a c-commanding position further inside a phase boundary is possible for those things which must move successive-cyclically). Further pursuit of this and other implications of these suggestions are left for future study.
Notes
. Note that we represent adjunction structures as is standard, involving two segments of the
single category to which adjunction takes place. In primitive c-command terms, however,
there is no corresponding distinction between segment and category. Rather, it is simply the
case that the adjunct (asymmetrically) c-commands the node to which it is adjoined.
. We will use the notation <X> to label subtrees in the text to distinguish them from node
labels. <X> is a collection of nodes and c-command relations, whose categorial root is X.
. For Chomsky (2000), the operation involved in substitution is what he calls set merge,
an operation that puts two structures into a single set. This is a notational variant of our
proposal that substitution involves the assertion of mutual c-command. One sees these two
variants in different formalizations of undirected graphs in mathematics: they may either be
characterized by a symmetric relation on the set of nodes (a set S of pairs such that (x,y) S
iff (y,x) S) or by a set of 2-subsets of the set of nodes. In the case of adjunction, Chomskys
formulation of the operation as pair merge is identical to ours, as he takes it to involve
the creation of an ordered pair. This is nothing but the addition of a single assertion to a
(c-command) relation.
. In Section 9, we show how the notion of constituent that we develop derives the conclusion that the structure that is derived from successive adjunction is not indeterminate
between attachment at any of the roots, but instead involves attachment at the categorial
root. In other words, while (15) and (16) are nondistinct, constituency tells us that (16) is
the correct way to draw the structure.
. Example (i) below shows that [Neg] on ne need not be checked via overt movement, and
is therefore not a strong/selectional feature. This eliminates the possibility of an alternative
derivation for (21) in which personne moves first to SpecNegP, checking [Neg] on the Neg0
head, after which personne moves on to SpecTP and Neg0 moves on to head-adjoin to T0 .
Such a derivation would not make our point, since there is a stage in the derivation where
Neg0 and personne would be in a direct Spec-head relation.
(i)
Je nai vu personne.
I neg-have seen noone
I did not see anyone.
. Given the formulation of adjunction given in the text, this result will require the sort
of interarboreal derivation discussed in Bobaljik and Brown (1997). An alternative to this
approach might invoke a transitivity condition on the c-command relation of the sort in (i):
(i)
For all nodes x, y, z, if xCy and yCz and not xz, then xCz.
. This has the further implication that in cases we may have thought were multiple adjunctions to XP, such as multiple adverbial phrases, are not. That is, although we can draw a tree
like (15), the constituents will always come out as in (16). This result meshes with Kaynes
(1995) proposal, under which multiple adjunction to XP is impossible, and forces us to a
view along the lines of that proposed by Cinque (1999) for multiple adverbs.
. Whether excorporation is allowed at all or in only these cases is under debate; see Roberts
(1991) for discussion and an opposing view.
. As we tentatively suggest in Section 11.3, the need for identifying a moveable constituent may only pertain to structures that need to be pronounced, however.
. An immediate question raised by this proposal is how expletive there is able to check
Ts EPP features under pure Merge. One possibility might build on the idea that there is
a radically impoverished lexical item, being the realization of the single categorial feature
D. Perhaps as a consequence the derivation can avoid the step of identifying a containing
constituent, as the bare feature itself is moveable. We would expect, then, that all such
cases of feature checking under Merge will involve such radically impoverished lexical items.
. To accept that (48)(49) are the correct structures for these sentences requires that the
empty complementizer is not being moved/elided in these cases. One argument, admittedly indirect, for the structure in (48) might center on Stowells (1981) proposal that empty
complementizers must occur in governed positions (to explain the fact that we get them
in object CP complements, but not with sentential subjects). If this is right, the empty C
couldnt front. Note, however, that this notion of needing a governor differs from the notion of needing a governor to be a constituent; here, it is something like a selectional
requirement, determining whether the empty complementizer can appear in the structure.
Concerning (49), note that, although (i) is possible, (ii) is unexpectedly ill-formed. This indicates that something more complicated is going on in elision of this kind, but we have
nothing further to offer here.
(i)
(ii) *Even though Mary hopes [CP [C ] e ], she doubts that Max is coming to the party.
. There is a loophole, however. If the phrase which occupies SpecCP in the complement of
a non-bridge verb like whisper is itself a CP, then the selectional property of the verb could
be satisfied despite the fact that its complement CP has a specifier. Sufficiently developed,
this might make interesting predictions for languages like Basque (Ortiz de Urbina 1989),
or Quechua (Cole 1982), in which clausal pied piping is allowed in wh-movement. For
example, in Basque, wh-arguments can in general either move to the matrix SpecCP alone
(i-a), or take the embedded clause along (i-b). We might expect to find that a version of (i)
with a non-bridge verb would show obligatory clausal pied-piping, because its complement
cannot have a specifier, yet without clausal pied-piping the wh-argument would have to land
in the specifier of the embedded clause on its way to its matrix position. More interesting
would be a case structured like (ii); here we would predict (given certain assumptions) that
(ii-a) would be ill-formed because the DP must land in the specifier of the complement of
whisper whereas in (ii-b) the entire subordinate clause stopped there. In (ii-b), whispers selectional requirements would be met, since what is in SpecCP of the complement of whisper
is itself a CP. However, not only do we not have the relevant Basque data, it is also not completely clear that the assumptions necessary for these predictions to fall out are warranted;
more research must be done before any solid conclusions can be reached. For example, it is
not clear whether wh-words must move internal to a pied-piped CP (Ortiz de Urbina 1990)
or not (Echepare 1995), nor is it clear that wh-words and pied-piped CPs share the same
landing site (Echepare 1995). Development of this area must await further research.
(i)
a.
b.
References
Ackema, P. & A. Neeleman (1998). Optimal questions. Natural Language and Linguistic
Theory, 16, 443490.
Bobaljik, J. D. & S. Brown (1997). Interarboreal Operations: Head movement and the
extension requirement. Linguistic Inquiry, 28, 345356.
Chomsky, N. (1995). The Minimalist Program. Cambridge, MA: The MIT Press.
Chomsky, N. (2000). Minimalist Inquiries: The framework. In R. Martin, D. Michaels, and J.
Uriagereha (Eds.), Step by Step, Essays in Minimalist Syntax in Honor of Howard Lasnik.
Cambridge, MA: The MIT Press.
Chung, S. (1994). Wh-agreement and Referentiality in Chamorro. Linguistic Inquiry, 25,
144.
Cinque, G. (1999). Adverbs and Functional Heads. Oxford: OUP.
Cole, P. (1982). Imbabura Quechua. The Hague: North Holland.
Echepare, R. (1995). A Case for Two Types of Focus in Basque. In E. Benedicto, M. Romero
and S. Tomioka (Eds.), Proceedings of Workshop on Focus [UMOP 21]. Amherst, MA:
GLSA.
Fox, D. (1999). Reconstruction, Binding Theory, and the Interpretation of Chains. Linguistic
Inquiry, 30, 157196.
Frank, R. & K. Vijay-Shanker (2001). Primitive c-command. Syntax, 4(3), 164204.
Grewendorf, G. & J. Sabel (1996). Multiple Specifiers and the Theory of Adjunction: On
scrambling in German and Japanese. Sprachwissenschaft in Frankfurt Arbeitspapier 16.
Johann Wolfgang Goethe-Universitt, Frankfurt am Main.
Higginbotham, J. & R. May (1981). Questions, Quantifiers and Crossing. The Linguistic
Review, 1, 4180.
Hornstein, N. (1998). Movement and Chains. Syntax, 1, 99127.
Kayne, R. (1995). The Antisymmetry of Syntax. Cambridge, MA: The MIT Press.
Lobeck, A. (1995). Ellipsis. Oxford: OUP.
May, R. (1985). Logical Form: Its structure and derivation. Cambridge, MA: The MIT Press.
Nissenbaum, J. (1998). Movement and Derived Predicates: Evidence from parasitic gaps. In
U. Sauerland and O. Percus (Eds.), The Interpretive Tract [MITWPL 25]. Cambridge,
MA: MIT Working Papers in Linguistics.
Ortiz de Urbina, J. (1989). Parameters in the Grammar of Basque. Dordrecht: Foris.
Ortiz de Urbina, J. (1990). Operator Feature Percolation and Clausal Pied-piping. In L.
Cheng and H. Demirdash (Eds.), Papers on Wh-movement [MITWPL 13]. Cambridge,
MA: MIT Working Papers in Linguistics.
Pollock, J.-Y. (1989). Verb Movement, Universal Grammar, and the Structure of IP. Linguistic
Inquiry, 20, 365424.
Rizzi, L. (1990). Relativized Minimality. Cambridge, MA: The MIT Press.
Roberts, I. (1991). Excorporation and Minimality. Linguistic Inquiry, 22(1), 209218.
Rudin, C. (1988). On Multiple Questions and Multiple WH Fronting. Natural Language and
Linguistic Theory, 6, 445501.
Wilder, C. (1994). Coordination, ATB and ellipsis. Ms., ZAS, Max-Planck-Gesellschaft,
Berlin.
Introduction
Murat Kural
b. Unergative verbs:
VP
VP
DP
e
V
DP
V
V
The key distinction between (1a) and (1b) is that the argument of an unaccusative verb in (1a) is generated as a complement as if it were the object
of a transitive verb, but the argument of an unergative verb is generated as its
specifier in the same position as the subject of a transitive verb. The following table contains some of the better-known unaccusativity tests that have been
developed over the years:
(2)
there-insertion (English)
Locative inversion (Chichewa)
Subject case (Basque, Hindi)
Agreement (Creek)
Cognate objects (English)
Resultatives (English)
way-construction (English)
Unaccusatives Unergatives
X
Absolutive
Ergative
Object
Subject
These tests and various issues regarding them have been discussed thoroughly
by Burzio (1986), Bresnan and Kanerva (1989), Bresnan (1994), Laka (1992),
Mahajan (1990), Martin (1991), Hale and Keyser (1993), Jackendoff (1992),
Levin (1993), Levin and Rappaport (1995) among many others.
The two-way distinction outlined above has been quite influential in the
current theories and has been adopted almost universally. This dichotomy has a
solid foundation since each test does in fact differentiate two classes of monadic
verbs across a clear line of separation, although the classes overlap significantly
in some cases. With the exception of some well-defined classes, such as motion
verbs, verbs typically display either the set of properties associated with unaccusative verbs or the ones associated with unergative verbs, as stated in (2).
What verbs do not do is to have some properties from the first column and
others from the second in a random manner, which is what would be expected
if these properties were not a reflection of some deeper structural organization
of monadic verbs.
On the other hand, what makes this dichotomy a questionable classification is the fact that it is not a perfect division. Not all verbs align in exactly
the same way described above, and perhaps more to the point, the verbs that
stand out of this classification usually form semantically coherent classes, and
the way in which a given verb class deviates from the pattern in (2) is not random either. An example would be verbs of motion, which allow there-insertion
and locative inversion in (3ab), which are unaccusative properties, but they
also allow cognate objects and resultatives in (3cd), which are expected of
unergative verbs.
(3) a.
b.
c.
d.
It is also the case that verbs that disallow cognate objects do not always behave
the same way. A verb like break allows the transitivity alternation, as in (4b),
but not appear, as in (5b) below.
(4) a. *The vase broke a great break
b. The magician broke the vase
(5) a. *The rabbit appeared a quick appearance
b. *The magician appeared the rabbit
On the other hand, there are some well-known unaccusative verbs that occasionally display some of the unergative properties, such as cognate objects with
die, as in (6), and grow, shrink, and sink allow bare measure phrases, as in (7).
(6) The rabbit died a horrible death
(7) a. The tree grew two inches
b. The shirt shrank two sizes
c. The ship sank a thousand feet
One must bear in mind when evaluating the results of the tests above that despite the fact that each one effectively points at some two-way contrast within
monadic verbs, the arguments presented thus far do not establish that all these
tests actually draw the same type of distinction. It is entirely possible that at
least some of these tests are sensitive to different properties, such that even
though they all define two different classes of monadic verbs, they actually define different types of classes, setting up different contrasts. What lies at the
heart of the matter is the question of what exactly each of these tests is sup-
Murat Kural
. An alternative approach
A major dividing line between the tests listed in (2) has to do with the position that is being tested: there-insertion, locative inversion, the case of the subject, and the subject-verb agreement patterns refer to some VP-external position, and more specifically, to the subject position within the inflectional field,
whether it is the tense or agreement projection. The remaining three, cognate
objects, resultatives, and the way-construction all relate to the VP-internal base
position of the object, as they all seem to be localized at the internal argument
position within the VP, i.e., what may be regarded as the complement of the
verb. When evaluated from this perspective, some of the discrepancies sketched
out in (3) through (7), and others that will be discussed below, become more
than just some quirky properties of each test or verb type. By shedding some
light into the internal organization of various VPs, these tests effectively serve
as diagnostic tools that identify the fault lines that separate monadic verbs.
Given this interpretation, the distribution of the classic unaccusativity tests
across various monadic verb types points at a four-way distinction in terms
of the differences in VP architecture, where each class can be defined as a
semantically coherent group. These groups are as follows:
1. Verbs of being are verbs that indicate that their subject comes to be in some
fashion, as in appear, arise, arrive, emerge, ensue, exist, lapse, and occur.
Although the verbs in this class seemingly cover a wide range situation
types, what they all have in common is the notion of becoming present in
some way.
2. Change of state verbs indicate that their subject has undergone a change of
state, as in the case of break, burn, change, fold, grow, heat, heal, melt, shrink,
and sink. Specifically, the change of state comes in the form of the end state
of the subject being stated in the verb. In other words, with these verbs,
a DP V-s can be paraphrased as DP becomes V-en, e.g., The window
breaks is equivalent to The window becomes broken.
3. Change of location verbs indicate the motion of the subject, as in fall, jump,
march, roll, run, skip, slide, swing, turn, and walk. The crucial concept here
is that the verb provides the manner in which the subject moves. If there
were a way to paraphrase DP V-s in this class, it would be DP moves by
V-ing, e.g., though admittedly not a perfect sentence, something like The
boy fell is truth-functionally equivalent to The boy moved by falling.
4. Verbs of creation are verbs indicate that the subject has produced an often
abstract, though sometimes concrete but intangible product, as in cough,
dance, dream, laugh, sing, sleep, smile, speak, and think. With these verbs, a
DP V-s can be perhaps quite awkwardly paraphrased as DP produces
an NV , where NV refers to the nominal version of the verb, e.g., The girl
coughs is equivalent to The girl produces a cough, cf. Hale and Keyser (1993).
The type of VP architecture that will be proposed for these verb classes is
sketched out in (8) below. Note that the internal organization of verbs of
being in (8a) is the same as Burzios (1986) classic unaccusative VP structure, whereas the change of location verbs in (8c) have the same VP design as
Burzios unergative verbs. By contrast, both the change of state verbs and verbs
of creation in (8b) and (8d) have complex, multi-layered VP architecture, that
involves inchoative and causative predicates respectively (the diacritics used
below will be discussed later on).
(8) a. Verbs of being:
c. Change of location:
VP
e
VP
V
DP
DP
appear
Bill
Bill V
walk
XP
Murat Kural
b. Change of state:
d. Verbs of creation:
VP
VP
V
DPi
DP
the vase V
INCH
VP
Bill
DP
PROi V
S
VP
CAUSE
XP
break
DP
(a dance) V
dance
XP
-1
As an aside, consider the VP structure for verbs of creation in (8d) in the context of the small v analysis of Kratzer (1994) and Chomsky (1995).2 A small
v is, by definition, a light verb whose function is to introduce the external argument, which naturally leads to the question of how v and cause relate to
one another. If they are assumed to be different verbs, it would not be clear
how the external argument of cause is introduced into the structure. Since v
is posited as the source of all external arguments, it must be provided by v.
However, this would suggest that cause is a monadic predicate, though it is
not obvious what a monadic cause might mean. In some sense, it would have
a meaning close to happen, but that would imply that the causative meaning
itself comes from v. This would, in turn, suggest that v actually has semantic
content that is the equivalent of cause. However, if indeed v has the semantics of cause, that would have to mean that it occurs only with verbs that have
causative meaning, excluding verbs that lack this sense altogether, e.g., verbs
that depict physical contact, such as kiss and touch, or verbs that do not entail
that their object undergoes any change of state, as in watch, read, and mention.
Finally, it must be pointed out that what this paper proposes is a way to
draw finer distinctions within the classic unaccusative/unergative dichotomy.
Although it presents an alternative that has a four-way distinction, this does not
mean that there can be no additional monadic verb categories. As our diagnostic skills sharpen in the future, there may be more than four classes that need
to be accounted for. In this respect one should follow the discussion provided
below as a potential starting point to a finer grained monadic verb typology.
. Verbs of being
As shown in (8a), verbs of being, such as appear, arise, arrive, emerge, ensue,
exist, lapse, and occur, are basically handled in this work as single-layered null
specifier verbs. The verb projects a single VP and its sole thematic argument is
generated as its complement.
(9)
VP
e
V
V
DP
appear
Bill
The structure given above is basically the same structure that has been assumed
for unaccusative verbs as first presented in Burzio (1986). As will be argued
below, most of the properties of this verb class follow directly from this classic
VP architecture shown in (9).
The discussion will concentrate on four fundamental properties of verbs
of being (as well as the other classes):
i.
First, they allow postverbal subjects in the there-insertion, (10), and locative inversion constructions in English, (11).
(10) a. There appeared a huge rabbit on the stage
b. There exist various possibilities
(11) a. On the stage appeared a huge rabbit
b. In this forest exist many magical beings
ii. Second, they do not transitivize with null morphology. The examples in
(12) are not possible even though the pragmatics of both situations are
controlled in a way that would allow the causer to be able to cause the
event of appearing, (12a), and the state of existing, (12b).
(12) a. *The magician appeared the rabbit out of the hat
b. *God existed the universe
A magician can make a rabbit appear out of a hat, and God is presumed to
have the power to bring the universe into existence. However, neither situation can be expressed using the type of null causatives that would otherwise
yield the transitivity alternation.
Murat Kural
iii. Third, these verbs do not allow the type of non-thematic complements
(NTCs) mentioned above, i.e., cognate objects, as in (13), resultatives, the
way-construction, (14a), and bare measure phrases (BMPs), (14b).
(13) a. *The rabbit appeared a quick appearance
b. *?Those people exist a strange existence
(14) a. *Cockroaches will exist their way into dominance on the planet
b. *The toothpaste appeared three inches
Note that Turkish and German allow verbs of this class to appear with the
passive morphology, albeit with the semantics of the impersonal construction (see Maling 1993 and Kural 1996).
The ability to license there-insertion and to invert the locative around the subject are both properties that involve a specifier position. Given the VP architecture in (9), and that no other verb class seems to allow either process (with the
notable exception of directionals, which will be discussed below), it would be
entirely plausible if the locus of both properties is the specifier inside the VP
rather than a specifier higher up. Based on this, we may conjecture that there
is licensed at the [Spec, VP] position, and that [Spec, VP] is a crucial intermediary position in the movement of the locative phrase to its surface position.
In cases where the [Spec, VP] is occupied by an argument, which is true for
all other verb classes, there-insertion or locative inversion would be blocked.
In a language like Dutch where the equivalent of there-insertion is available
with other verb classes would then suggest that Dutch allows er to be generated higher up, perhaps at the [Spec, TP] position. The contrast between the
two language types in terms of the availability of [Spec, TP] for there may be a
function of overt verb movement as suggested by Holmberg (1986), i.e., correlating with the fact that an inflected verb stays relatively low in English, below
the T, but much higher in Dutch, at T or above.
In order to understand why verbs of being are not compatible with transitivization through what presumably is null causativization, one needs to look
at the VP architecture of causatives with a structure like the one in (9) and more
traditional VP structures, both shown in (16). A key background assumption
in this account is that cause is a diadic verb with a Patient role that the causee
must associate with it. When there is an external argument in the [Spec, VP],
the Patient role associates with that argument as in (16a), but when the only
possible candidate lies further below, the association of the Patient role with the
target argument is blocked by the intervening thematic head, i.e., the root verb.
(16) a. Plain VP under cause:
VP
DP
DP
V
V
CAUSE
VP
V
DP
V
CAUSE
XP
VP
V
e
V
DP
The effect intended in (16b) is similar to the minimality effects of Rizzi (1990).
The correlation would not be an unusual one under Stowells (1981) conception of -role association as an instance of coindexation between the predicate
and its arguments. Under this view, the minimality violation in (16b) would be
handled as a case of binding violation, cf. Aouns (1985) Generalized Binding.
The VP architecture in (9) also derives the inability of verbs to license
NTCs: since the sole argument of the verb takes up the complement position,
either by being generated there or binding an empty category in that position,
there is no room available for NTCs. The fact that these verbs can generate only
one argument also provides the grounds for a straightforward explanation as to
why this verb class is cross-linguistically so resistant to passivization. The only
argument that these verbs can generate gets demoted in the passive construction. According to the Extended Projection Principle, the EPP (Chomsky 1982,
1995), the subject position must be occupied by some constituent. An expletive
can fulfill this requirement at the surface, but things get further complicated at
LF because of the principle of Full Interpretation, FI (Chomsky 1986), which
requires that LF representations contain all and only the elements that can re-
Murat Kural
VP
DP
Bill V
XP
walk
i.
These verbs allow there-insertion and locative inversion only in the presence of a directional PP. There is allowed in (18) and the locatives are inverted in (19), where the sentences contain a directional PP. By contrast,
both there-insertion and locative inversion are blocked in (20), where the
PPs are locational.
(18) a. There ran three people away from the crime scene
b. There walked a woman into the room
(19) a. Away from the crime scene ran three people
b. Into the room walked a woman
(20) a. *There walked a woman in the garden
b. *In the room walked a woman
iii. Change of location verbs license various NTCs, such as BMPs, (22a), cognate objects, (22b), resultatives, and the way-construction.
(22) a. Bill ran five miles in the race
b. Mary walked a quiet walk in the woods
iv. Finally, these verbs can passivize with much ease, although with some
cross-linguistic variation: In a language like English, they passivize in the
presence of an NTC, as is the case with the BMP in (23a) and the cognate
object in (23b), although no such NTC is needed in Dutch.
(23) a. Five miles were run by Bill in the race
b. ?A long walk was walked in the woods
If the assumption made above is correct, and both there-insertion and locative
inversion require a vacant [Spec, VP], one would expect the VP architecture
in (17) to exclude both constructions. However, it is also the case that thereinsertion and locative inversion are licensed in the presence of a directional
phrase. Given that both constructions key in on a vacant VP level specifier, one
can maintain a consistent analysis by positing a directionality phrase above
the VP in these instances, whose specifier position would be available for generating there and moving the locative phrase, much like the [Spec, VP] with
verbs of being.4
(24)
DirP
e
(there)
(into the room)
Dir
Dir
VP
V
DP
a woman
PP
walk
Murat Kural
location verb and cause. The situation would more or less the same in the case
of directionals because Dir does not have a thematic grid associated with it.
The complement position of a change of location verb is available for various NTCs, see (17), which is a property these verbs display in abundance, some
of which is exemplified in (22). These verbs are also quite agreeable with passivization, and this is expected since any NTC that is primarily a noun phrase
would be able to move up to the subject position once the external argument of
the verb is demoted to an explicit or implicit by-phrase. To see that some of the
NTCs are truly objects that can be passivized, observe the following paradigm
in Turkish. A BMP, such as bes mil five miles appears in the accusative case
depending on the specificity of the distance, (25a), and the same expression
becomes the subject of the corresponding passive, (25b).
(25) a.
Ahmet o
bes mil-i
yarsta kostu
A.-nom that five mile-acc race-loc run-past-3sg
Ahmet ran those five miles in the race
b. Yarsta (Ahmet tarafndan) bes mil kos-ul-du
race-loc A.
by
five mile run-pass-past-3sg
Five miles were run in race
Based on the interpretive requirement that the demoted subject must be animate, Maling (1993) argues that the impersonal passives in Dutch contain a
pro-arb subject at LF, even though the surface subject is the expletive er. What
separates Dutch from English is the ability of the former to license an expletive
at a higher position, i.e., the specifier of an inflectional head. Once a language
allows expletives to be generated in cases where there is no NTC, it would not
require presence of an NTC in the passivized form of a change of state verb. Although the mechanism that introduces the pro-arb is unclear at the moment, it
is likely to be independent of the ability to license expletives above the VP level.
(26)
VP
V
DPi
the vase V
VP
INCH
DP
PROi V
XP
S
break
A couple of points need to be noted here: First, a change of state verb indicates
that the denotation of its argument undergoes some change as a result of the act
denoted by the verb.5 Second, inch refers to the inception of a transformation,
i.e., the change of state, not its endpoint, e.g., The boat is sinking is true in case
the ship starts sinking and only a small part of it is under water, and it does
not require that the whole ship to be submerged. Third, the head that inch
combines with, which is the root verb, is a stative verb that designates the end
state of the transformation, i.e., The ship is sinking means The ship has begin to
be in a sunk state. This is indicated in the present work with the diacritic VS ,
where a VS is a defined as a verb that must incorporate into an inch to be a
legitimate verb. Also note in passing that the relation between the specifier of
inch and the specifier of the root verb in (26) is one of control rather than
raising. This distinction is important for the internal logic of what is being
proposed here, which will become more apparent below.
Change of state verbs have the following properties:
i.
Murat Kural
iii. They allow only limited types of NTCs: they take BMPs, (30a), and resultatives, (30b), but not cognate objects, (31a), or the way-construction,
(31b).
(30) a. The ship sank a thousand feet
b. The house burned to a crisp
(31) a. *The boat sank a complete sink
b. *The house burned a spectacular burn
iv. They also do not passivize, even with NTCs that are otherwise allowed in
the active form.
(32) a. *A thousand feet were sunk by the ship
b. *A crisp was burned by the house
Given the scheme of things argued thus far, the fact that there-insertion or locative inversion is not allowed with change of state verbs suggests that the VP
architecture of these verbs does not allow for a vacant VP-level specifier. This
problem is avoided by assuming that inch is a control predicate rather than
a raising predicate. Having the thematic structure of this verb class separated
into two VPs as in (26) has the added advantage of ensuring that the argument that undergoes the change is uniformly associated with the Patient role.6
However, this does not mean that all cases of Patient are necessarily introduced
through inch. The object of the causative predicate cause is arguably a Patient
argument since it is acted on by the causer, yet being the Patient-of-cause does
not entail any discernible change of state.7
The ability of change of state verbs to transitivize under null causatives
strongly suggests that the relevant thematic argument of the inch-breakS combination is not generated lower than a thematic role providing predicate. The
predicate in question in these cases is inch, which is a thematically active
predicate and is located as the higher predicate in the structure. Thus, for the
biphrasal scheme to work in these circumstances, the argument of a change of
state verb must be generated as the specifier of inch. Doing so, however, would
deprive the root verb breakS of an argument, which would violate -criterion,
or any principle that is meant to regulate the bijection between thematic roles
and arguments. This is where the concept of control comes into play. As can
be seen in (26), both inch and breakS generate their own arguments, and their
identity is ensured by the control relationship that holds between the two.8 A
consequence of the VP architecture given for this verb class is that an event
like the intransitive breaking is treated as a composite event that contains the
endstate component represented with the root verb VS , i.e., breakS , and the
inception component that is represented with the inch.
Change of state verbs allow only a subset of all NTCs, and disallow the
others. It was argued in Kural (1996, 1998) that what regulates the distribution
of NTCs with this class of verbs is their requirement that their complements be
secondary predicates. While the BMPs measure out the extent of the motion
with change of location verb, they primarily predicate on the state of the subject
with change of state verbs, e.g., *a five-mile run athlete versus a thousand-foot
sunk ship. Note that BMPs cannot bear accusative case in Turkish, even if one
construed the BMP as a predesignated specific distance that is being measured,
cf. change of location verbs in (25).
(33) a.
Bot
bes metre(*yi) batt
boat-nom five meter-acc sink-past-3sg
The boat sank five meters
b. Hava
onbes derece(*yi) snd
air-nom fifteen degree-acc warm-past-3sg
It got fifteen degrees warmer
. Verbs of creation
Semantically what verbs of creation, such as cough, dance, dream, laugh, sing,
sleep, smile, speak, and think have in common is that they entail the creation of
the nominal equivalent of the verb, e.g., cough can be paraphrased as produce
a cough, and dance as produce a dance. This verb class is treated in this work as
causative-layered verbs that appear in a double-layered structure reminiscent
of Hale and Keysers work (1993).
Murat Kural
(34)
VP
DP
Bill
V
V
CAUSE
VP
V
DP
(a dance) V
dance
XP
-1
The lower VP contains the root form of the verb of creation and the cognate
object that denotes the entity that is being created either in the implicit or in
the explicit form. Note that much like a VS , a V-1 is also an incomplete predicate
as is. The diacritic of a raised -1 indicates that the verb needs one more argument to be satisfied, but it cannot provide that argument thematically. This
argument is supplied by the null causative predicate cause, which is an elementary predicate like inch: it lacks a phonetic content and it does not provide the core meaning of the verb. The source of the creation sense one finds in
this verb class is the combination of the causative layer and the often implicit
cognate object.10
This verb class has the following properties:
i.
ii. They do not transitivize with null (causative) morphology, a fact that has
been commented on by Hale and Keyser (1993):
(37) a. *The DJ/music danced the people
b. *The clown laughed the children
iii. They allow the whole range of cognate objects, resultatives, BMPs, the wayconstruction. The following set of sentences contain a specific class of cognate objects that are very common with verbs of creation: nouns that refer
to a subclass of what the true cognate object would refer to, e.g., the rela-
tion between tango and dance, as seen in (38a). The example (38b) shows
one of the better known instances of the resultative construction with a
verbs of creation.
(38) a. The couple danced a tango during the reception
b. His friends laughed Bill out of the room
iv. These verbs passivize across languages with a type of variation that is familiar from change of location verbs: In a language like English they passivize
in the presence of an overt NTC, as in (39), while in a language like Dutch,
they passivize without one, as in (40).
(39) a. A tango was danced during the reception
b. Bill was laughed out of his room by his friends.
(40) Er wordt hier veel gedanst
There is danced a lot here
The there-insertion and locative inversion facts follow from the internal organization of the VP structure in (34), which provides no vacant specifier at the
higher levels of the VP complex to host there or become the landing site for the
inverted locative. On the other hand, the base position of the thematic argument is perfectly compatible with the null causative construction that would
transitivize a verb of creation. In fact, the inability of these verbs to transitivize in the same manner as change of location and change of state verbs is
due to an independent constraint on null causatives. Unlike overt causatives,
null causatives cannot be iterated, which can be seen in the following example that starts out with a change of location verb, run, in (41a), transitivizes it
once in (41b), and then attempts to add one more layer of causation in (41c).
Compare (41c) with (42) and the corresponding example in the periphrastic
version in (43c).
(41) a. The horses ran around the barn
b. John ran the horses around the barn
c. *Sue ran John the horses around the barn
(42) Sue made John run the horses around the barn
(43) a. The horses ran around the barn
b. John made the horses run around the barn
c. Sue made John make the horses run around the barn
Murat Kural
further observed in (42) that there is nothing wrong in principle about embedding a transitivized change of state verb under another causativized verb. What
goes wrong in (41c) is that a verb that was already incorporated into a null
cause attempts to be further incorporated into another null cause. Given the
VP architecture of creation verbs in (34), where a root V-1 necessarily incorporates into a null cause, embedding a verb of creation under a null causative
layer would be the exact equivalent of the successive incorporation into null
causative seen in (41c) above. Thus one may plausibly argue that whatever rules
out cases like (41c) would also be behind the ungrammaticality of (37).11
In terms of the VP-internal characteristics of verbs of creation, it may be
sufficient to note that the VP structure in (34) allows enough room for all the
NTCs that have been discussed in this work. A similar statement can be made
with respect to the passivization facts. Assuming that the language is like Dutch
and can generate an expletive at a position that is relatively higher up, the EPP
would be satisfied by moving the covert cognate object in the lower specifier
to the subject position at LF. In a language like English, where expletives can
only be generated low, this verb class provides no vacant specifier to generate an
expletive, which means that a verb of creation can passivize only in the presence
of an NTC in the overt syntax.
kuttf
ne bhfkaa
dogs(plur) erg barked(masc-sing)
The dogs barked
b. Sitaa
(*ne) aayii
S.(fem) erg came(fem)
Sitaa came
(44) a.
(Hindi)
(45) a.
(Basque)
The issue regarding the case of the subject is one that recalls Fillmores (1968)
classic account of way in which the thematic licensing of an argument reflects
on the types of prepositional phrase that each will appear later on at the surface. The most obvious correlation between the paradigm above and Fillmores
theory is his conjecture that all agents are licensed as by-phrases at the deep
structure, and they remain as such in the passive construction, but in the active structures where they become subjects at the surface, the subjectivization
transformation deletes the preposition at some point in the derivation.
Taking this correlation as more than a mere coincidence, suppose that any
argument that is thematically licensed by the predicate cause is licensed as a
PP, which is either realized as a by-phrase, or as an ergative case, depending on
the language.
(46)
VP
V
PP
P
DP V
CAUSE
YP
XP
V
V
YP
The point that this is not such a far-fetched idea can be made by showing that
an argument that is both a causee and a causer in a multiple causative construction appears as a by-phrase in Turkish if both the accusative and the dative cases
are taken up by other arguments. This can be seen in the following example
where Ayses reading is ultimately caused by the speaker but indirectly through
Murat Kural
Ahmet. The causal interaction is between Ahmet and Ayse, where Ahmet acts
under the speakers directive:
(47) pro [Ahmet tarafndan] [[bu kitab] Ayseye
1.sg A.
by
this book-acc A.-dat
oku-t-tur-du-m
read-caus-caus-past-1sg
I made Ahmet make Ayse read this book
The fact that the intermediate causer appears as a by-phrase would have a
straightforward account if one assumed that all causers start out their lives
as by-phrases. The arguments that eventually surface as accusative, dative, or
nominative expressions have their prepositions incorporated either into the
verb directly, or into some functional/inflectional projection that mediates in
the P-incorporation.
With this correlation in the background, consider Mahajans (1994) claim
that deep down, split ergativity in Hindi is the same procedure as auxiliary
selection in the perfective in Romance and Germanic languages. Kaynes (1993)
work establishes the key connection between the two phenomena by arguing
in effect that have is derived from be by incorporating a preposition into it in
the manner that has been sketched out below.
(48)
Aux
Aux
Aux
VP
P
PP
P
t
DP V
CAUSE
YP
XP
V
V
YP
In other words, he argues that have is be-P.12 Mahajan takes this idea further by
first observing that be is preserved in Hindi in the perfective tense, where the
subject is in the ergative case.
(49) raam-ne vah kitaabe
par
th
R.-erg those book-plur read-perf-fem-plur be-fem-plur-past
Ram had read those books
Aux
DirP
Aux
e
Dir
Dir
Dir
VP
PP
DP
a woman
walk
XP
Murat Kural
. Conclusion
A system in which monadic verbs are differentiated following a four-way classification provides a more effective and efficient layout that allows the apparent
discrepancies between verb classes emerge as lines of demarcation between a
much richer array of monadic verb typology. The preceding discussion demonstrates that once we move away from the preconceived idea of a two-way classification of monadic verbs, we find not only that the syntactic behavior of these
verbs becomes more sensible, but we also realize that the classification of these
verbs along syntactic lines fully coincides with their broad semantic properties
such as denoting a change of location or creation of an abstract entity.
Notes
. I am leaving out ne-cliticization facts from this paper.
. Arad (this volume) extends the classic v theory of tansitivity by positing a stative v, a
causative v, and an inchoative v, which is closer to the spirit of what I am arguing for. Although we agree on the compositional nature of semantic verb classes, there are also clear
differences between the two proposals. First, the basic criticism of the v theory I present in
the text holds for her system as well. Second, the predicates I am using have the ability to
combine with one another, while Arads light verbs do not. An alternative view of some of
the phenomena she discusses can be found in Kural (1996).
. Such strategies are arguably in place in languages that allow verbs of being to passivize.
In the case of German and Turkish, these passives are the equivalent of the impersonal construction, and as suggested by Maling (1993), the EPP appears to be satisfied in these cases by
a pro-arb. On the other hand, the equivalent construction behaves more like a true passive
in Lithuanian, where one can argue that the by-phrases are moved to the subject position
(Kural 1996).
. There is an obvious parallelism between this view and Mateus and Rigaus work in
this volume, which provides a structural frame in which Talmys (1987) and by extension,
Grubers (1965) work can be interpreted.
. This is a rather crude definition that does not account for negation (The window broke
versus The window did not break) and other modalities. Strictly speaking, one should be
talking about hypotheticals. These issues are handled in detail in Kural (1996).
. Stripped out of its Patient component, the (PRO) arguments of the stative root predicates
sinkS (be sunk) or melt S (be molten) are more likely to be a non-Patient, perhaps similar to a
Theme or Rozwadowskas (1988) Neutral.
. It is argued in Kural (1996) that the interpretation of the causative structure with causer
acting on the causee is one of two possible interpretations. The other one, where the causer
manipulates the circumstances to bring about an event are derived by having the Patient-ofcause associate with the entire VP, which stands for the whole proposition.
. The theory of PRO has been in a state of flux since the end of Chomskys (1980) classic theory that restricted PRO only to ungoverned domains. This concept was based on the
complementary distribution between anaphors and pronouns in the same binding domain,
which came to an end first, with Huang (1983), then Chomsky (1986), when different types
of binding domains were established for anaphors and pronouns, making them no longer
incompatible in every context. Various proposals have been made to fill this vacuum, including Chomskys (1995) null case and Hornsteins (1999) raising, none of which necessarily duplicate the ungoveredness requirement, which is being violated in the representation
in (26).
. As an amusing aside, note that the equivalents of the verb die behave like they belong in
different verb classes; a change of location verb in English, John died a horrible death, but a
change of state verb in French, Jean est mort *(d)une mort horrible.
. This covert cognate object in (34) serves the function that the trace of the cognate object
does in Hale and Keysers (1993) system, where an unergative verb is derived by incorporating a cognate object into a light verb. An outstanding problem with this approach is that
these verbs can cooccur with the cognate object, as in danced a great dance or even related
objects as in danced a great waltz. This is no longer a problem if we assume that the verb
binds a nominal expression at all times, covert or otherwise, instead of a trace as in Hale and
Keyser (1993).
. It is argued in Kural (1996) that the movement of a verb into successive null predicates
of the same type creates a problem in index transferal and leads to a violation of chain
formation.
. Kayne actually assumes that the preposition starts out as the head of the complement of
the auxiliary verb, but Mahajan turns this around and has the P start out as the preposition
of the VP-internal subject, much along the lines of Fillmores theory of agents.
. Note that it is not clear how this would relate to the subject versus object agreement
Martin (1991) reports for Creek.
References
Aoun, J. (1985). Generalized Binding. Dordrecht: Foris Publications.
Bresnan, J. (1994). Locative Inversion and the Architecture of UG. Language, 70, 72131.
Bresnan, J. & J. Kanerva (1989). Locative Inversion in Chichewa: A case study of factorization
in grammar. Linguistic Inquiry, 20, 150.
Burzio, L. (1986). Italian Syntax. Dordrecht: Reidel Publishing.
Chomsky, N. (1982). Some Concepts and Consequences of the Theory of Government and
Binding [LI Monographs]. Cambridge, MA: The MIT Press.
Chomsky, N. (1986). Knowledge of Language: Its nature, origin, and use. New York: Praeger.
Chomsky, N. (1995). The Minimalist Program. Cambridge, MA: The MIT Press.
Murat Kural
Fillmore, C. (1968). The Case for Case. In E. Bach and R. T. Harms (Eds.), Universals in the
Linguistic Theory. New York: Holt, Reinhart, and Winston.
Gruber, J. (1965). Studies in Lexical Relations. Doctoral dissertation, MIT.
Hale, K. & S. J. Keyser (1993). Argument Structure and the Lexical Expression of Syntactic
Relations. In K. Hale and S. J. Keyser (Eds.), The View from Building 20. Cambridge,
MA: The MIT Press.
Holmberg, A. (1986). Word Order and Syntactic Features in the Scandinavian Languages
and English. Doctoral dissertation, University of Stockholm.
Hornstein, N. (1999). Movement and Control. Linguistic Inquiry, 30, 6996.
Huang, C.-T. James (1983). A Note on the Binding Theory. Linguistic Inquiry, 14, 554561.
Jackendoff, R. (1992). Babe Ruth Homered His Way to the Hearts of America. In T. Stowell
and E. Wehrli (Eds.), Syntax and Semantics, Vol 26: Syntax and the Lexicon. San Diego:
Academic Press.
Kayne, R. (1993). Towards a Modular Theory of Auxiliary Selection. Studia Linguistica, 47,
331.
Koopman, H. & D. Sportiche (1991). The Position of Subjects. Lingua, 85, 211258.
Kratzer, A. (1994). The Event Argument and the Semantics of Voice. Unpublished
manuscript, University of Massachusetts, Amherst.
Kural, M. (1996). Verb Incorporation and Elementary Predicates. Doctoral dissertation,
UCLA.
Kural, M. (1998). Two Types of Bare Measure Phrases. In Proceedings of WECOL 96, 177
187. Department of Linguistics, California State University, Fresno.
Laka, I. (1992). Ergative for Unergatives? Talk presented at UCLA.
Levin, B. (1993). English Verb Classes and Alternations. Chicago, IL: The University of
Chicago Press.
Levin, B. & M. R. Rappaport (1995). Unaccusativity: At the syntax-lexical semantics interface
[LI Monographs]. Cambridge, MA: The MIT Press.
Mahajan, A. (1990). The A/A-bar Distinction and Movement Theory. Doctoral dissertation,
MIT.
Mahajan, A. (1994). The Ergativity Parameter: Have-Be alternation, word order, and
split ergativity. In M. Gonzales (Ed.), Proceedings of the Twenty-Fourth Meeting
of the Northeastern Linguistic Society. University of Massachusetts, Amherst: GSLI
Publications.
Maling, J. (1993). Unpassives of Unaccusatives. Unpublished manuscript, Brandeis
University.
Martin, J. (1991). The Determination of Grammatical Relations in Syntax. Doctoral
dissertation, UCLA.
Perlmutter, D. (1978). Impersonal Passives and the Unaccusative Hypothesis. In Proceedings
of the Fourth Annual Meeting of the Berkeley Linguistic Society. UC Berkeley. Berkeley.
Rizzi, L. (1990). Relativized Minimality [LI Monographs]. Cambridge, MA: The MIT Press.
Rozwadowska, B. (1988). Thematic Restrictions on Derived Nominals. In W. Wilkins (Ed.),
Syntax and Semantics, Vol 21: Thematic Relations. San Diego: Academic Press.
Stowell, T. (1981). The Origins of Phrase Structure. Doctoral dissertation, MIT.
On Agreement
Locality and feature valuation
Luis Lpez
University of Illinois-Chicago
I propose a new look at the operations Agree and Move, taking many of the
concepts in Chomsky (1998, 1999) as starting point but departing from these
papers in important respects. First, I argue that the operation Agree is strictly
local. Second, I argue that the operation Move is triggered by the instability
created in the system by unvalued features (following similar ideas in
Frampton & Gutmann 1999). Third, the concept of co-valued features is
introduced: two terms with unvalued features of the same type that are
related by the operation Agree must have their features valued in tandem.
It is shown that feature co-valuation underlies expletive-associate and
movement chains. These alterations are shown to have healthy consequences,
since they allow us to revisit and eliminate some unnecessary assumptions.
Introduction
The goal of this paper is to advance the Minimalist Program. Some of its assumptions are scrutinized and rejected and a more streamlined framework is
proposed that also has some empirical advantages. Concretely, I propose a new
look at the operations Agree and Move, taking many of the concepts in Chomsky (1998, 1999) as starting point but departing from these papers in important
respects. First, I argue that the operation Agree is strictly local and long distance
agreement, as proposed by Chomsky, does not exist. Second, I argue that the
operation Move is triggered by the instability created in the system by unvalued
features (following similar ideas in Frampton & Gutmann 1999). Third, the
concept of co-valued features is introduced: two terms with unvalued features
of the same type that are related by the operation Agree must have their features valued in tandem.1 It is argued that the notion of feature co-valuation
makes explicit some implicit assumptions concerning the expletive-associate
Luis Lpez
relation as well as the locality that restricts the connection between the links of
a chain. On the conceptual side, these alterations will be shown to have healthy
consequences, since they allow us to revisit and eliminate some unnecessary
assumptions. Additionally, this paper presents novel analyses of structural case
assignment and of the typological differences in expletive constructions, amply
demonstrating the empirical advantages of this approach.
As Chomsky defines it, Agree is an operation that involves two terms,
a probe and a goal. Take a functional category that has a set of unvalued/uninterpretable features that need to be valued and deleted from narrow
syntax. In order to value its features, the functional category must probe in its
c-command domain until it finds a goal: a set of matching features valued
features of the same type that the probe can agree with. If the functional category has an additional EPP feature, the goal will be pied-piped into the spec
position of the probe. For instance, assume that Tense (henceforth T) bears
unvalued -features. In order to value and delete its features, T probes in its
c-command domain for a goal with matching features of the same type. The
subject in Spec,v has valued -features, therefore it can be a goal for Ts features. The subjects features value those of Tense, which delete. Additionally,
T selects for a D it has an EPP feature a requirement that is satisfied by
pied-piping the subject to Spec,T.
An important distinction that Chomsky makes is that matching and agreeing are not the same thing in his system. If a probe encounters a DP with matching features whose structural case has already been assigned and deleted, the
probe will not be able to agree with this DP and, moreover, it will not be able to
keep probing. In other words, agreement can only take place with a constituent
that has an unvalued case feature. Chomsky refers to this as a freezing effect
and it provides an account of Relativized Minimality effects. Consider (1). In
(1a), the matrix T cant agree with John because there is an intervening DP, it,
with matching features. Can the matrix T agree with it? No, because it agrees
with the subordinate T and, as a consequence, it has had its case deleted. Once
case is deleted, agreement is not possible. Since T cant agree with the intervening DP, it cant pied-pipe it either. Finally, since the features of T and it match,
T cannot go on probing, hence the ungrammaticality of (1b):
(1) a. T seem that it is likely to John win.
b. *John seems that it is likely to win.
A second restriction on Agree is the Phase Impenetrability Condition (henceforth PIC). A phase is a structure created by a sub-numeration headed by
transitive v (an abstract light verb that introduces external arguments, see Sec-
On Agreement
The PIC prevents the head of a phase to probe into another phase, except its
edge for our purposes, the edge of a phase is the outer specifier. Thus in (2)
H1 can probe XP, but cant go further down. On the other hand H2, which
is not the head of a phase, can probe XP and YP. In Chomsky (1999), H1 = C,
H2 = T and H3 = v, so T can probe its c-command domain indefinitely whereas
C can only probe Spec,T and Spec,v.
The validity of operations is evaluated at the phase level. So, turning back
to example (2), and assuming that H3 = v is transitive, any Agree or Move operations involving the edge of v, XP and H2 = T will be evaluated once H1 = C
is merged and the phase finished. If H3 = unaccusative/passive v, operations
involving H3 and YP will also be evaluated when H1 is merged and the strong
phase completed.
A third property of Agree that I discuss in this paper is that of completeness. This issue appears in connection with raising and ECM predicates (3a and 3b respectively) and in participle agreement (3c, a Spanish sentence):
(3) a. Several prizes are likely to be awarded.
b. We expect several prizes to be awarded.
c.
(Chomsky 1999: 5)
Las mujeres fueron
vistas
en la tienda.
The women were.3rd.pl seen.fem.pl in the store
Luis Lpez
probe a DP but, since they are not -complete, the Case of the DP is not deleted
and the DP can be probed again from higher up. Under one version of the story,
Tdef has an EPP feature that attracts the DP to its spec, from where it can be
probed from above. Under another version also considered by Chomsky, Tdef
does not even have an EPP, but the higher probe can simply by-pass it to reach
the DP.
To sum up: Chomsky allows for long distance agreement between a probe
and a goal, provided that freezing effects and the PIC are respected. Probes
can by-pass other probes if the latter are not -complete.
I think it is fair to say that some of the assumptions that ground this framework for analysis are stipulatory and a system that could dispense with them
would be preferable. First of all, the distinction between matching and agreeing is fairly artificial and does not follow from any principles: why should a
DP without Case be more able to agree than a DP with Case? Notice that the
-features of the DP do not delete, since they are not interpretable apparently, Case is necessary to make the -features of a DP visible to a probe, but
we do not know why these -features cant stand on their own and be accessed
by a probe without the intermediary of a Case feature. Moreover, the requirement that the goal has Case leaves out all the concord phenomena that Carstens
(2000) discusses (i.e. sharing of gender and number features between a noun
and its modifiers). Second, the assumptions that make up the PIC could also be
questioned turning back to (2), why can H2 probe YP but H1 cant? It would
be preferrable to have the same locality requirement for all instances of Agree.
Finally, how is -completeness to be defined? In example (3c), the participle
has number and gender features and T has number and person features. Since
the DP has number, gender and person features, it would seem that neither T
nor the participle has a complete set of -features. Are we going to assume that
T has unexpressed gender features in Spanish? On what grounds?
Consideration of multiple agreement in Bantu provides further fuel against
the notion of -completeness. Consider the Kiswahili sentence in (4), cited
from Carstens and Kinyalolo (1989):
(4) a.
Juma a-ta-pika
chakula
Juma agr.fut.cook food
Juma will cook food
On Agreement
In Kiswahili, a verbal root cant support both tense and aspect morphology.
So, the strategy resorted to by this language is to have the aspect morphology
attached to the stem and the tense morphology attached to an auxiliary verb,
glossed as be. The agreement marker a- indicates subject agreement, third person singular and first noun class (where noun class is taken to be an expression
of gender, see Carstens 1991). Notice that subject agreement is repeated on
the main verb and on the auxiliary. Lets consider the stage in the derivation in
which the aspect marker has just been merged and the subject is still in its basegenerated position. Further, I assume that assembling of the aspect marker and
the verb is a PF process:
(5) [AspP a-me
[vP Juma pika . . . ]]
agr.perf
Juma cook
Obviously, -me- is an agreement probe, as shown by the overt agreement morpheme. Notice that the probe is -complete or, at least, it cant be said that
the Aspect head is any less complete than T. So, matching of features between
-me- and the subject should lead to deletion of the Case feature of the subject.
At this point, the subject is inactive and cant be probed by T, even if -me- had
pied-piped the subject to its spec. So Chomsky predicts that T cant agree with
Juma, contrary to fact. A consequence is that we need Case features associated
with some heads, independent of -completeness.
Alternatively, we could assume that the Case feature of Juma is deleted but
still present in the computation until the phase is completed. Thus, the Case
feature is still accessible to T. However, notice that the same reasoning can apply
to the examples in (3): we could say that Tdef does delete the Case of the DP
but it survives until the phase is completed, so it is still accessible from the
higher probe. But if we adopt this way of thinking, we simply do not need
-completeness at all, only the stipulation that T is not the head of a phase.
It seems safe to conclude that a system in which the operation Agree were
not concerned about -completeness would be preferable.2
I propose that the operation Agree is more restricted than what Chomsky
claims: I argue that there is no such thing as long distance agreement. Take the
structure (6). Chomsky would allow for Agreement to take place between H1
and XP across H2P, unless H1 is head of a phase. Instead, I propose that the
Luis Lpez
H1P
H1
H2P
H2
XP
Further, I argue that this intervention effect is what triggers Move. In Chomskys model, once a probe has agreed with a goal the latter can be pied-piped to
the spec position of the probe in order to satisfy an EPP feature of the probe.
Instead, I argue that XP in (6) needs to move if it has unvalued features that
cant be probed. Movement is triggered by unvalued features of the term that
moves. We will see below how this can be conceptualized without falling into
the problems that Chomskys (1993) Greed created and that led to its abandon.
The obvious conceptual advantage of my proposal is that, by disallowing
any sort of long distance agreement, I reduce drastically the search space for
an agreement probe, thus reducing the computational complexity involved in
the operation. As will be shown, this proposal makes the PIC unnecessary and
provides a good motivation for freezing effects, including superraising, so
they do not need to be stipulated. Finally, it allows us to dispense with Tdef and
-completeness.
In Section 2, I present my assumptions concerning Agree and Move and the
notion of feature co-valuation. In Section 3, I discuss the mechanisms of Agree,
Case assignment and Move within the assumptions laid out in Section 2.
In Section 4, I explore another aspect of the theory of agreement, namely,
the possibility that two sets of unvalued features can be co-valued. I analyze expletive constructions and show that Locality of Agreement+Feature covaluation provide some new insight into the properties of this construction in
German, French, English and Icelandic. It is shown that my assumptions provide analyses for a variety of phenomena that cant be explained under Chomskys. A crucial ingredient of my analyses is that the expletive-associate relation
is an Agree relation established in a strictly local configuration. In Section 5 I
discuss Superraising, and show how my proposal to enforce locality between
two agreeing terms can be extended naturally to enforce locality between the
links of a chain. The final section summarizes my main conclusions.
Although not every aspect of the theory of Case and Agreement is discussed
and some knots are still tied (or uncut), I believe enough is accomplished to
conclude that the direction taken here is promising.
On Agreement
. Framework
. Transitivity
I adopt Chomskys (1995) proposal that external arguments are introduced by
a functional category, a light verb represented as v (for similar ideas, see Kratzer
1996, among others who develop ideas ultimately rooted in Larson 1988).
What does v select for as a complement? The traditional assumption is that
it should be a VP. However, it is worth considering Marantzs (1997) recent
proposals concerning the morphology-syntax interface.
Marantz argues vigorously against the Lexicalist Hypothesis or more appropriately, he complains that Lexicalism died a while ago but most of us did
not read the obituary and missed the funeral. Two lexicalist assumptions that
Marantz rejects are crucial for our purposes. The first is that the lexicon is
a computational space, separate from syntax, in which words are formed by
putting together different bits and pieces, including roots with an inherent category label. The second assumption is that syntax does not see these bits and
pieces, only the resulting lexical item with the category label attached to it. Instead, Marantz proposes a narrow lexicon composed of roots and bundles of
grammatical features. The roots enter the computational system there is only
one for morphology and syntax without a category label and take the complements that they select. Then they are themselves selected by a functional
category. If the functional category is a v, the resulting structure will be a verbal phrase. If the functional category is a D, the result will be a nominal phrase.
Henceforth, I represent a label-less root as an X and its projection as an XP.
(7) a.
b.
vP
v
XP
X
DP
D
YP
XP
X
YP
Luis Lpez
does not assign a -role at all but may in some languages assign partitive Case
(Belletti 1988; Lasnik 1992). Call it v().4 Further, I propose that the selectional properties of v in any of its versions are invariant: it selects an XP as a
complement and a DP as a specifier, even if it cant assign it a -role or, to
say the same thing in different words, v has an obligatory EPP feature that is
satisfied by the external argument if we have a v(AG) or by a raising DP or an
expletive if we have a v() (more on this in Sections 3.2 and 4).
. Agree
Following Chomsky (1998), Agree is an operation between two items, a probe
and a goal (p, g) whose objective is to provide a value for features that are
introduced into the computational system without a value. Henceforth, I represent unvalued features as a variable so if I want to say that x has unvalued
-features I simply write that x has []. Thus, a functional category with []
can probe within its c-command domain until it finds a DP, which has a set
of valued -features, []. As a result of the probe, the [] of the probe can
be valued. Following a long tradition I assume that v, finite T and C have sets
of [] in need of valuation (see Haegeman 1992 and Zwart 1997 for evidence
that C can agree).
We can understand Agree (p, g) as an operation that co-values two sets of
features. If one of the two sets is already valued, this value is simply copied on
the other set and the [] symbol is removed. This is represented in (8a), where
X is a probe and Y is a goal. If both probe and goal have unvalued features
of the same type, they will remain unvalued, with a twist: since they are now
involved in the Agree relation, these features will be co-valued. In other words,
two unvalued but agreeing features cant vary freely, eventually they must end
up with the same value. I represent this with a subscript number in (8b):
(8) a. Agree (X[], Y[1]) X[1], Y[1]
b. Agree (X[], Y[]) X[1 ], Y[1 ]
On Agreement
but looks for a specific instantiation of a feature. For all this, I prefer to leave
EPP out of the Agree (p, g) system, as a residue of the -assignment system.
. Case assignment as Agree
In the Principles and Parameters tradition, certain heads had the stipulated
property of assigning Case, namely P, T and V. This view did not change radically until the most recent developments of the Minimalist Program. As I mentioned in the introduction, in Chomsky (1999: 4), [assign x Case] is not a feature of the probe there is no matching relationship between probe and goal
in this respect. Rather, when agreement between probe and goal takes place,
and the probe is -complete, the unvalued Case of the nominal is valued and
deleted. As we saw in my discussion of the example (4), in Swahili we find
that both the T head and Aspect co-occur in the same sentence and appear
to be -complete. As a consequence, the aspect head probes the DP and since
the probe is -complete, the unvalued Case of the DP should be valued and
deleted, effectively freezing the DP in place before T can probe it. This seems to
be an undesirable result. Instead, it seems we should retain from earlier frameworks the idea that some heads are responsible for Case licensing of DPs and
others are not.
Lets then assume that some functional categories do have a Case feature
and, further, that Case is one of the features that can enter an Agree (p, g)
relation notice that I am talking about Case features now like [nominative]
or [accusative], not Case assigning features. This feature may be inherent or
may be added freely as the lexical item is drawn from the lexicon to form a
numeration.5 This Case feature is valued from the onset but uninterpretable.
Since this feature is valued, it can be copied on an agreeing goal that has an
unvalued Case. Since it is uninterpretable, it will be deleted as soon as the phase
is spelled-out. For example, v(AG) has an [accusative] feature, so when it agrees
with an object with a unvalued Case, the feature [accusative] can be copied on
the feature matrix of the object. After the vP phase is completed and spelledout, both accusative features are deleted. Case assignment is therefore only a
variant of the agreement relation, as represented in (9). In (9a) I represent the
situation in which one of two items involved in Agree has a Case feature. What
if two terms X, Y are involved in the Agree relation but both of them have
unvalued Case features? I posit that the Agree relation forces them to co-value
their Case variables (9b), following the general pattern in (8):
Luis Lpez
Co-valued Case features are going to be at work in my analyses of the connection between expletive and associate as well as in movement chains.
The final question to be decided is which heads bear a Case feature. I
assume that v may bear a Case feature. As already mentioned, v(AG) bears
accusative Case (so that the effects of Burzios generalization are captured),
whereas v() may have no Case or may have so called partitive Case (after
Belletti 1988; Lasnik 1992). Finite T does have [], at least in English, but,
contrary to standard assumptions, I claim that T does not have a Case feature.
Instead, finite C bears [nominative] and non-finite C [null].6 Nominals have
valued -features but unvalued Case, which I represent as unvalued Case.
It may seem somewhat exotic to have C as a Case assigner (but see Platzack
1986; Vikner 1995; and Chomsky 1999, fn 17), but clearly this assumption leads
to a simplification of the theory. Currently, Chomsky (1998) must assume that
there are two types of infinitival heads. On the one hand, Tdef does not assign
Case, has only an [person] feature and is selected by a V. On the other, ordinary infinitival T in control and PROarb constructions has [person] and
[number], assigns null Case and is selected by C. Therefore, Chomsky makes
two sets of assumptions: (i) there are two types of nonfinite T, (ii) C and V have
different selectional properties because V cannot select non defective T and C
cannot select defective T. Instead, I propose that there is only one type of infinitival T, which neither has -features nor assigns Case (see also Romero, this
volume). Non-finite T can be freely selected by a non-finite C that bears null
Case, by a prepositional complementizer (like for) that bears accusative, or by
a lexical root that is selected by v (giving rise to raising/ECM constructions).
If infinitival T does not have a set of [] features, one of the motivations for
-completeness disappears (as I will show in detail in Section 3.2). Another advantage is that disociating Case assignment from agreement with T turns out
to be a necessary step in order to account for the properties of expletive constructions (Section 4) for instance, the associate of the expletive in English
agrees with T but appears in non-nominative (accusative or partitive) Case.
. -features and spell-out
In Chomskys system, Full Interpretation forces features without semantic content uninterpretable to be deleted by LF after being checked. The question
is whether Full Interpretation also plays a role in PF. Lets assume it does. The
On Agreement
Luis Lpez
Hann
telur
sig
vera [ t sterkan].
He.nom believes himself.acc to be strong.acc
b. Hann
tel-st
vera [ t sterkur].
He.nom believes.refl to be strong.nom
He believes himself to be strong
On Agreement
(see Branchadell 1992 and references therein). When this happens, both the
moved DP and the stranded quantifier show up in dative Case, spelled-out as
a (see Demonte 1995 for arguments that a is a spell-out of dative Case and
not a preposition). Additionally, notice the morphological change in the dative
clitic.:
(14) a.
Le
di
a
cada uno de los hombres un libro.
cl.3rd.sg gave dat each one of the men
a book
b. Les
di
a
los hombres un libro a
cada uno.
cl.3rd.pl gave dat the men
a book dat each one
I gave each of the men a book
In (14a) the clitic agrees with the quantifier, which is singular, whereas in (14b)
it agrees with the determiner, which is plural. How can this be?
I suggest that the puzzle of having two dative constituents in the same
sentence can be analyzed in terms of feature co-valuation. Lets go step by step.
First, assume that dative shift is an instance of Case-driven movement
(as in Collins & Thrinsson 1996, among others). It has been noted before
(Demonte 1994) that the dative clitic can augment the Case valence of a verb:
(15) a.
It seems natural to conclude that the clitic itself is involved in assigning Case.
Following suggestions in Ura (1996), we can assume that it is associated with a
light verb v head. Thus, the clitic has the features [dative] and [].
Take sentence (13a). The QP has dative shifted to a position where it can
be probed by the clitic. The unvalued Case of the QP is valued as [dative] and
the [] of the clitic are valued as [3rd.sg] because these are the -features of
the quanti fier cada.
Take now (13b). The crucial assumption here is that the DP starts out as the
complement of the quantifier. In this configuration, the quantifier can probe
the DP and, as a result, they covalue their Case features, as in (16a). The DP
raises, stranding the quantifier, to a position where it can be probed by the
clitic. The [dative] Case of CL is copied onto the feature structure of the DP. By
feature co-valuation, the QP also ends up with the feature [dative]:
Luis Lpez
XP
ZP
X
X
YP
I assume that specs create adjunction structures, so that XP and X are not
categories but segments of a category (see May 1985 and Chomsky 1986 on
the notion of segment). As a consequence, ZP is not dominated by an Xmax ,
but only by a segment. Thus, ZP can be probed from outside XP and can also
probe YP and Spec,Y. However, an item outside of XP cant probe YP.8
Locality of Agreement subsumes the PIC. The purpose of PIC is to ensure
that a probe cant look into a finished phase, except at the edge (spec) of this
phase. (17) ensures that a probe never looks at anything but its own complement and the spec position of its complement. Locality of Agreement derives
Chomskys freezing effects (Section 5.1) and renders -completeness useless
(Section 3.2). The difficulty and the interest is to see whether this more restric-
On Agreement
tive condition can give us empirical advantages, which happens to be the Case,
as I will show.
. Move
Another point in which I distance myself from Chomsky (1998) concerns
Move. It might be useful here to briefly review Chomskys (1993, 1995) motivations for Move or Attract to provide some background to my own solution.9
In Chomsky (1993), constituents moved to satisfy their own formal requirements to get Case, in other words within the principle called Greed.
Greed rules out (19), because the moved item does not achieve anything by
moving to the subject position of the matrix clause:
(19) *John seems t is happy.
Luis Lpez
seen as the combination of Agree (p, g) and Merge of the goal in the spec position of the probe. The idea is that the (p, g) relation is sufficient to delete the
uninterpretable features of probe and goal, except for one, the selectional EPP
feature of the probe. In order to delete the EPP feature, the goal is pied-piped
into the spec position of the probe. Thus, the application of Move presupposes
a previous Agree operation and is triggered by a selectional requirement of the
probe. Within this model, it still seems that successive cyclic movement must
be stipulated in some form. Assuming phases and the PIC, an account of (20a)
requires stipulating optional EPP features in Spec,v and Spec,C, which are licensed if they have an effect on outcome (more on this in Section 5). It is easy
to see how this takes us back to the Global Economy problems that Collins discussed. The ungrammaticality of (19) must also be stipulated: if the head triggers movement, there is no principled reason why John cant raise upstairs.
Chomsky proposes the above-mentioned freezing effect.
To sum up, neither Move nor Agree+pied pipe integrate successive cyclic
movement without stipulation and without assuming Global Economy. The
challenge that I undertake here is to conceptualize displacement in such a way
that it satisfies these theoretical desiderata.
Let me introduce my approach to this problem by means of two metaphors.
The first one is the tension metaphor, which I borrow from Frampton and
Gutmann (1999). The presence of an [] feature creates a tension in the structure built by Merge and the computational system tries to release it before
proceeding to the next phase. Tension is released by valuing the [] feature.
The tension metaphor needs to be complemented with a reaction
metaphor. We can compare syntactic movement to a chemical reaction: putting
two substances together may unchain a reaction which sometimes gives as a
result a product that is not stable yet and requires further reaction until a final stable product is obtained. As in syntactic movement, the reacting substances do not know that the product of their initial reaction is not the final
stage, but the initial reaction takes place regardless. I propose to view Successive Cyclic Movement as intermediate reactions. Regarding movement as a
reaction may help us abandon teleological metaphors of movement, with all
their concomitant problems, at the same time sticking to the fundamental idea
that all syntactic operations must be motivated.
In Chomsky (1999), movement is a consequence of agreement. Contrariwise, I want to propose that movement takes place when Agree fails. Concretely,
Move is triggered by the necessity of creating a (p, g) relation to value and delete
[] features of items that might be too distant for probing, given the strict locality requirement imposed by (17). Thus, the ultimate reason why there is move-
On Agreement
Luis Lpez
. Analyses
. Accusative Case and transitivity
The purpose of this section is to show how Locality of Agreement and reactive Move, together with some current assumptions concerning the role of v
in assigning accusative Case can account for some data that have remained
recalcitrant after several years of intensive investigation. At the same time,
this analysis will be taken as a starting point for an analysis of expletives in
Section 4.
Consider (21) and (22), which represent, respectively, the steps involved in
building a regular transitive vP and the resulting structure (see also Romero,
this volume). In derivations I omit category labels on phrases:
(21) 1.
2.
3.
4.
5.
6.
7.
8.
(22)
SU
v
v
XP
OB
X
X
t(OB)
Lets decode the information contained in (21). In line 1, the root merges with
object, which bears valued -features and an unvalued Case feature. The unvalued Case feature forces it to move, as represented in lines 2 and 3. Notice that I
have not subscripted any features on the trace of object: this is incorrect, I will
On Agreement
fix it in Section 5. In line 4, the light verb merges, with unvalued -features
and a valued Case feature. The light verb and object may agree at once, with
the result that their unvalued features are now valued, as shown in line 5. Since
now we have a structure without -features, it can be spelled-out at once (line
6) and the uninterpretable features can delete (line 7) deleted features are
represented with strike-thru. Finally, in line 8, the subject can merge. Subject
merges with its own valued -features and unvalued Case (not represented in
21) and satisfies the selectional feature of v (EPP).
This structure preserves the assumptions in Arad (1998, this volume),
Chomsky (1995, 1998) and Kratzer (1996) that the same head that formally
licenses the object is involved in introducing the external argument. However,
Chomsky (1995) and the others assume that object raises to Spec,v and checks
its accusative Case in that position, as in structure (23):
(23)
vP
OB
v
SU
v
v
VP
V
t(OB)
Lpez (2001) points out that it is problematic to have the subject -role assigned in the same domain where accusative Case is checked and by the same
head. Take a situation in which subject has been merged with v and received a
-role and v has an accusative Case to assign. What is to prevent v from assigning accusative Case to subject instead of object? All Chomsky (1995: 311312,
356) does is to stipulate that arguments can only check features if they form
non-trivial chains. Or take another situation: the object has raised overtly to
Spec,v before subject has merged (as briefly suggested by Chomsky, and developed by Ura 1996). In this situation, what prevents the object from assuming
a second -role? Nothing does, unless we reintroduce the -Criterion as a theoretical primitive, an undesirable move. Given these problems, it seems clear
that it is not a good idea to have a configuration like (23) if v assigns a -role
to subject and Case to object. This is avoided by leaving object in Spec,X and
ensuring that Agree only involves the c-command domain of the probing head
(and notice that there is nothing wrong wih structure (23) if v does not assign
Case to object in Spec,v).
Luis Lpez
Word order data favors my approach against both Chomsky (1995) and
Chomsky (1998). Consider the following sentences (that I cite from Johnson 1991):
(24) a. *Chris ate slowly the meat.
b. Chris talked slowly to Gary.
On Agreement
e.
Juan le
hizo
a
Pedro reparar el coche.
Juan cl.3rd.sg make-past.3rd.sg dat Pedro repair the car
Juan made Pedro fix the car
Lpez (2001) argues that A-chains whose head has accusative or dative Case
can have two -roles, one assigned by the subordinate predicate, the other by
the matrix predicate. This is not accounted for if ECM SUs and causees move
to matrix v, but it follows if they move to Spec,X, which becomes a position
that is both - and Case-related.
. Nominative Case and participle agreement
Lets proceed with the derivation of an ordinary transitive sentence. After a vP
structure like (22) is completed, the Case feature of subject is still unvalued.
T merges and probes subject so it can value its own -features. However, T
cannot value subjects Case, and this forces subject to raise to Spec,T. Then C
merges with TP. With subject in Spec,T, C probes subject and values its own
-features while it values the subjects Case feature. The CP can now spell-out
and all the uninterpretable features can delete:
(26)
CP
C
TP
SU
T
T
vP
t(SU)
1. Agree(T[ ] ,SU [ ][ C] )
T[ ] . . . SU[ ][ C]
2. Agree(C[nom][ ] ,SU[ ][ C] ) C[nom][ ] . . . SU[ ][nom] . . .T[ ]
This is the structure of the solution for the simplest Case. Let us now consider Cases in which the DP that is generated in complement position raises to
Spec,T:
(27) a. A book is on the table.
b. A lady passed away.
c. The body was discovered by a lady.
Luis Lpez
In these Cases, we have the v() variant: v selects for a D in its spec although
it does not have a -role to assign. By assumption, v() in these instances does
not assign Case either, because the DPs generated as complements of the main
predicate show up in Spec,T. The passive is the most complicated example, as
shown in (28).
(28) [C [5 T[4 be [3 participle [2 v [1 discover- the body ]]]]]]
The bare root discover is selected by a light verb, so the structure becomes verbal and can be selected by participle morphology. The object is merged with a
unvalued Case that forces it to raise to 1. In 1, object can be probed by v, which
satisfies vs unvalued -features but object still has its unvalued Case feature.
So, object raises again to 2. In 2, object can be probed by the participle, valuing
the latters -features, and satisfies vs selectional feature (EPP). Additionally,
if the participle has a Case feature, as in Icelandic, both the participles and
objects Case features are co-valued at this point. Object continues raising successively through 3 and 4 and in the latter it can be probed by T, thus valuing
the [] of T. Finally, object stops at 5, where it can value its Case with C (and
so does the participles, if it has one, by co-valuation).
Kaynes (1989) facts about participle agreement in French follow naturally.
Kayne showed that a participle in this language agrees with an object provided
that the latter has raised, either as an instance of wh-movement, clitic movement or, as in this Case, A-movement to subject position. In my terms, there
is participle agreement with an object that stops in 2, but not if it stays in situ,
because of Locality of Agreement. If position 2 is not filled because nothing
raises into it, we get a default form of the participle, without a set of -features.
As Chomsky (1999) himself notes, Kaynes findings have no account in his system. Since he allows the participle to probe object in its c-command domain
regardless of the intervening distance, he cannot explain why agreement shows
up only when object raises.
. Eliminating -completeness and Tdef
Recall that Chomsky suggests that two probes are not -complete: participles
and non-finite T in Raising constructions. Recall also that I argued that there
does not seem to be a coherent way to decide which probes are complete and
which are not.
Lets start with participles. In Chomsky (1999), T can probe the DP across
the participle because the participle is not -complete. In his framework, if
the participle were -complete, it would freeze the DP in place and would not
On Agreement
John has been merged within the lowest structure, which I call aP, in parallelism with vP. In many languages, adjectives and nouns agree. This is analyzed here as taking the the light head that selects the root to be a probe. After
movement to 1, John can be probed and agrees with the adjective but it still
has a unvalued Case that needs to be valued and deleted. Recall that the adjective/participle has Case in Icelandic. As a result of Agree (a/ptc,DP), their
unvalued Case are covalued. The unvalued Case forces the DP to raise to the
next available specifier, which is number 2 here. In 2, the DP cant be probed
by a Case assigning head (by hypothesis, be does not assign Case here), so
the DP has to continue raising. Since T is not a Case assigner and there is no
C in the subordinate sentence, raising of the DP only stops at 6, where it can
be probed by v and its Case is valued as [accusative]. By covaluation, the adjective/participles Case is also valued as [accusative]. All Case values can now
delete. The DP is now stable and does not (cannot) raise anymore.
Chomskys (1999) analysis of (28b) is very different from mine: either Tdef
pied pipes John to Spec,T, so its EPP feature is deleted or the upstairs probe v
by-passes Tdef . In any Case, the Case feature of the DP is not valued and deleted
by Tdef because Tdef does not have a complete set of -features. Later, the matrix
v probes in its c-command domain until it reaches John in Spec,T or in situ.
v can then agree with and assign Case to the DP. Finally, v pied-pipes the DP to
Spec,v to satisfy an EPP feature of v.
Among the advantages of my analysis: (i) no need for -completeness; (ii)
no need for two types of non-finite T, one of which is, by stipulation, selected
by V while the other is selected by C; (iii) as discussed above, word order facts
Luis Lpez
The initial numeration includes both the expletive and the DP a man. Since
Merge is simpler than Move, there is merged in Spec,T rather than have the
DP raise. There only has the feature [person], which is enough to satisfy the
EPP feature of T, but leaves the -features of T intact because the expletive is incomplete. Thus, T can probe and agree with the associate. The -features of
T are valued and deleted and the Case of the associate is valued as nominative
and deleted. Notice that, as a result, the expletive and the DP do not enter any
sort of relation there is no expletive-associate relation, contrary to what had
been commonly assumed up to this point.
This analysis makes some specific predictions, some correct, others not.
Among the incorrect ones, in English the associate shows up in accusative Case,
not nominative (as Chomsky himself recognizes). Importantly, if the expletive
merges with Spec,T, as universally assumed, one would expect to find expletive constructions with all kinds of predicates; however, in many languages including English expletive constructions are limited to unaccusative and passive verbs or, in other words, to predicates without an external argument. Even
On Agreement
when expletives are grammatical with transitive verbs (Icelandic), the resulting
construction has a somewhat different form, as I will show in a minute.
Unlike Chomsky (1998, 1999), I do not claim that Tense agrees directly
with the associate of the expletive which would violate Locality of Agreement instead, the expletive enters agreement relationships with the associate
and with Tense. Moreover, I claim that in the general Case, expletives merge in
Spec,v, which explains why only unaccusatives and passives accept expletives.
With some plausible assumptions concerning the properties of functional categories and expletives, I account for the range of variation found in German,
French and Icelandic, namely:
1. The Case of the associate is non-nominative in English and French, nominative in German.
2. The associate agrees with Tense in English and German, but not in French.
3. There are Transitive Expletive Constructions (TEC) in Icelandic but not in
English or French.12
. Parameters of variation in expletive constructions
My view of expletives of the there type assigns them a richer feature structure
than is standard so far. Expletives are of category D and as such they have a
set of -features and Case (as for there having Case, see Groat 1999; Lasnik
1995). Their Case feature is unvalued, like that of the other nominals. What
makes expletives different from other nominals is that they are not referential,
which entails that they do not have inherent -features either. At this point,
languages have two choices. The expletive can come from the lexicon with a
fixed default value, say [3rd person singular], as is the Case in French. Or the features may simply be unvalued, as in English or German: thus, there ends up
being both [] and unvalued Case (although expletives never seem to exhibit
explicit number agreement, for reasons that I do not know).
If my assumptions concerning the feature bundle of the English and German expletive are right, valuing the features of the expletive is somewhat more
complicated than those of a functional category or an ordinary DP: its unvalued Case forces it to establish an Agree(p, g) relation with a v or C but its []
forces it to do so with a DP establishing the expletive-associate relation.
The other parameter of variation in constructions with expletives concerns
v(). v() may bear partitive Case or may not bear Case at all (on partitive
Case: Belletti 1988; Lasnik 1995). In German, we only seem to find the non
Case bearing v() and in French we only see the Case bearing v(). English
Luis Lpez
seems to have a mixed system, where be bears partitive Case (as argued by
Lasnik 1995) but other unaccusatives do not.
Furthermore, I assume that the expletive is merged in Spec,v, (except in
TECs, Section 4.6) (see Groat 1999 and references therein):
(32)
vP
EXPL
v
v
XP
OB
X
X
t(OB)
(32) also reflects object raising to Spec,X, pushed by its own unvalued Case.13
Notice that no Xmax is interposed between the expletive and object, so they can
enter an agree relation without violating Locality of Agreement.
One could consider the possibility of retaining the traditional notion that
expletives are merged with TP. However, merging the expletive with vP seems
preferable conceptually. We know that v(AG) selects for an external argument.
We can simply assume that v always selects for a D (Section 2.1), the only difference between v(AG) and v() is that the latter has had its -assigning property bleached, but otherwise they have identical syntactic properties. In other
words, the D selecting feature of v does not need to be stipulated. However,
if the expletive is merged with Tense, we do have to stipulate that Tense has
an additional D selecting feature, somewhat arbitrarily, since T never assigns
a -role. Notice that we do not need an EPP feature on T to force movement
of a DP to Spec,T because the need of DPs to value their Case is sufficient to
trigger raising. Therefore, I stick to the idea that the expletive merges with vP,
obtaining the bonus point that we explain why expletive constructions occur
only with unaccusative and passive verbs.
Additionally, word order suggests that object cant have moved higher than
Spec,X in English, which within my assumptions suggests that Spec,v must be
filled with an expletive:
(33) There arrived three men. / *There three men arrived.
Lasnik (1995) shows that object must be adjacent to the verb be or the unaccusative in English:
(34) a. *There will be usually a man here / There will be a man usually here.
On Agreement
b. *There will arrive usually a man here / There will arrive a man usually
here.
As we saw before, the fact that an adverb cant stand between the object and the
verb is taken as a sign that object must have raised to Spec,X.14
. German
In German, as in English, we can have expletive constructions with unaccusative and passive verbs. The associate of the expletive is in nominative Case
and it agrees with T (examples provided by Susanne Winkler, p.c.):
(35) a.
Luis Lpez
C is now merged. C has [] and [nominative]. It can probe the expletive in Spec,T, valuing its own -features. Moreover, the Case of the expletive
is valued as [nominative] and, automatically, so does the value of the associate, by virtue of the Agree relation that co-values their Case features. There
are no unvalued features left, so the structure is ready for spell-out and all
uninterpretable features can delete at this point, as shown in (36.3).
(36) 1. Agree(Da[ ][ C] ,NP[ ][ C] ) Da[ ][ 1C] . . . NP[ ][ 1C]
2. Agree(T[ ] ,Da[ ][ 1C] )
T[ ] . . .Da[ ][ 1C] . . . NP[ ][ 1C]
3. Agree(C[nom][ ] ,Da[ ][ 1C] )
C[nom][ ] . . . Da[nom][ ] . . . T[ ] . . . NP[nom][ ]
Thus, my analysis of expletive constructions attributes a different role to expletives than that found in Chomskys recent work. For Chomsky, there satisfies
the EPP of T and T probes and agrees with the object in situ. In my account, T
cant value its -features with object because of Locality of Agreement, unless
either object is in Spec,v or an expletive in Spec,v can act as a mediator between
T and object. The expletive is close to T, so it can be probed and it is also close
enough to object to probe it. The expletive-associate relation that has been so
widely discussed turns out to be an Agree(p, g) relation.
So far, there is no empirical difference between Chomskys approach and
mine. Testing them against more complicated Cases will show our different
predictions.
. French
In French, Tense does not agree with the associate of the expletive, which receives partitive Case (Belletti 1988), not nominative, as can be seen in examples (37):
(37) a.
Il y
a
trois hommes sur la table.
it there have three men
on the table
There are three men on the table.
b. Il est arriv trois femmes.
it is arrived three women
There arrived three women.
c. Il en est arriv trois.
it part is arrived three
Three of them arrived
On Agreement
object raises to 1, where it gets partitive Case from v(). Since vP does not
include any unvalued features, it can spell-out and delete the uninterpretable
features, as shown in (39.1). The expletive is merged with vP with features [3rd
person singular] and unvalued Case. unvalued Case should force il to probe.
However, there is nothing to probe, because the Case feature of object has already deleted. Notice the difference with the German Case, where vP could
not spell-out because of the remnant unvalued Case feature on object, leaving
objects features available for probing by the expletive.
If this reasoning is correct, when il is ready to probe, it will not find a Case
value available for it and there is no co-valuation between the expletive and
object. After T is merged, it can probe the expletive, valuing its own -features
as [3rd person singular], as in (39.2). Since il still has an unvalued Case feature,
it raises to Spec,T. Then C is merged. Finally, il is assigned nominative Case by
C and values the -features of C (39.3). The valued uninterpretable features of
C, T and il can now be deleted:
(39) 1. Agree(v[part][ ] ,NP[ ][ C] ) v[part][ ] . . . NP[part][ ]
2. Agree(T[ ] ,il[3rd.sg][ C] ) T[3rd.sg] . . . il[3rd.sg][ C]
3. Agree(C[nom][ ] ,il[3rd.sg][ C] ) C[nom][3rd.sg] . . .il[3rd.sg][nom]. . .T[3rd.sg]
Luis Lpez
On Agreement
raises to Spec,T where it values its own unvalued Case as nominative. All the
uninterpretable features of the expletive, C and T can now be spelled-out and
deleted. Thus, we have obtained (i) agreement between T and object (through
there), (ii) non-nominative Case for object.
(45) 1.
2.
3.
4.
Lets now see if this analysis can be extended to unaccusatives and passives. It
seems not. Consider sentences (46):
(46) a. *It is/has arrived three men.
b. *It was found a body.
c. It is a pity that a body was found in Pearl Street.
(46c) reminds the reader that English it can be an expletive with the verb
be, so the question is why it cant be one in (46a, b), in the French manner.
It seems that the solution must be found in the absence of partitive Case in
English unaccusatives. If unaccusatives could assign Case in English (as Lasnik argue), (46a, b) should be grammatical: three men would get Case from
v() and the expletive would get Case from C. It seems that English unaccusatives differ from be in that only the latter can assign partitive Case for
reasons that I do not know. The derivation of English expletive constructions
with unaccusatives and passives must be more in the German way: the expletive and its associate have their Case features co-valued. The sentences in (47)
provide evidence that this analysis is correct. The grammaticality judgments
are delicate because pronouns do not like presentation focus, which is forced
in expletive constructions. However, there seems to be a clear preferrence for
the nominative pronoun:
(47) a. ?There arrived she, with a big, mysterious trunk.
b. *There arrived her, with a big mysterious trunk.
The expletive will raise through the spec numbered 1 to spec 2, where it can
receive accusative Case from the matrix v.17 Notice that this analysis predicts
Luis Lpez
agreement between the matrix v and the associate of the expletive. Given the
poverty of English inflection, this prediction cannot be checked.
. Transitive Expletive Constructions
Finally, lets explore why there are TECs in Icelandic but not in English. In (50)
is an example:18
(50) Thadh klarudhu margar mys ostinn
alveg
there finished many mice the.cheese completely
Many mice completely finished the cheese.
(Bobaljik and Jonas 1996: 217)
The immediate solution would be simply to say that Icelandic T has an EPP
feature and the expletive is merged in its spec position. Subject is in Spec,v and
object is in Spec,X and the adverb adjoined to a lower projection of X essentially the structure of transitive predicates in English except for the presence of
the expletive in Spec,T.
However, matters do not seem to be so simple. Throughout the 1990s,
evidence has accumulated that languages with TECs have a clausal structure
more complex than those languages without TECs, at least in the realm of the
Germanic languages. Bobaljik and Thrinsson (1998) have argued that TECs
correlate with other phenomena 2 subject and 2 object positions, multiple
inflectional morphemes in one stem, obligatory V raising which confirm the
complex structure hypothesis.
I assume therefore that the existence of TECs depends on the existence of
a functional category that merges with TP. Following a long tradition, I call it
AgrP.19 Agr and v both select for a DP in their specs, where an expletive can be
merged, which makes me think that they probably are really the same category.
Consider the following structure.
(51) C [thadh Agr [1 T [SU v [ 2 Adv X OB ]]]
object raises to Spec,X (and I abstract away from the option of object shift and
its complications). The bare root adjoins to v and the latter raises to T. The
subject raises to Spec,T. From here, there are two possibilities. One possibility
assumes that Agr assigns Case to a probed DP and the expletive gets it from C.20
The second possibility takes Agr to not be a Case assigner. In that Case,
subject would still raise to Spec,T, where its Case feature would be co-valued
with that of the expletive. The expletive could get Case from C and subject
On Agreement
would receive the same Case by co-valuation. I cant see how both possibilities
can be teased apart at this point.
Interestingly, TECs provide another piece of evidence in favor of locality of
agreement and against long distance agreement. Sigurhsson (1991), Frampton (1997) and Vikner (1995) have shown that in TECs subject raises to a fairly
high position, to the right of the highest modal:
(52) a.
theta epli
this apple
theta epli
this apple
(Vikner 1995: 191)
This can be accounted for within my assumptions, but not within Chomskys
(or Framptons, who leaves this as an unresolved problem). But before I present
my analysis, notice that matters become even more puzzling when we consider
that in expletive constructions with unaccusatives and passives the word order
is very different (McGinnis 1998: 152):
(53) Thadh hafa aldrei fari-st sjmann.
there have never died sailors
Sailors have never died
The question now is why the associate must raise in TECs but not in unaccusatives. Consider the following compressed representations for (52a)
and (53):
(54) a. [AgrP thadh mundi [TP einhver strakur t [ . . . ]]]
b. [CP C [TP thadh hafa [vP t farist [XP sjmann t ]]]]
Take (54a) first and assume that the highest modal is in Agr, where it has raised
from T (Bobaljik and Jonas 1996). My assumptions force subject to raise to
Spec,T, where it can values its Case feature, hence the high position of subject.
But Chomsky would predict that (52b) should be grammatical: the expletive in
Spec,Agr satisfies the EPP, T would probe and find subject in its c-command
domain. Long distance agreement is possible so subject would have no need to
raise.
Take (54b) now. At this point, all we have to assume is that expletive constructions with unaccusatives may have the basic structure in (32), with the
expletive merged in Spec,v. Thus, (i) object raises to Spec,X, (ii) the root ad-
Luis Lpez
joins to v and (iii) the expletive raises to Spec,T, as in the previous analyses of
German, French and English. The resulting word order is as expected.
On Agreement
had to say was that a DP with all its features satisfied has no motivation to
move.
In a similar manner, within my framework, the ungrammaticality of (55)
comes about because several people is protected by a maximal category and
consequently is protected from probes. Consider the phrase structure of (55),
represented in (56):
(56)
TP
T
vP
v
there
v
XP
seem
CP
C
IP
several people I
are
According to Chomsky, the matrix T probes into the subordinate clause until it
finds the goal it, which has matching features. Although T and it cant agree,
the probe is interrupted and can never reach John. Hence, from (57b) we cant
derive (57a).
How would Superraising be accounted for within my framework without
stipulating Relativized Minimality?
Luis Lpez
First, we will have to start by looking at the theory of chains and traces.
Chomsky (1993) proposes the copy theory of movement i.e., the hypothesis
that what we had been calling trace is actually a copy of the moved item. He
argues that this is conceptually superior to our previous conception of a trace
as an independent syntactic object because the copy theory allows us to maintain Inclusiveness. Lets assume that traces are copies of their antecedents. A
question has sometimes been asked that has not, to the best of my knowledge,
received a careful answer, namely, what is the connection between the copies
that form a chain. As pointed out by Roberts (1998), once the uninterpretable
features of the head of a chain are deleted, we need to know what we are going
to do with their copies in the foot of the chain. Presumably, they should delete
too, the question is what mechanism we are going to use for this purpose.
I propose that Feature Co-valuation is exactly what we need. Assume structure (58), in which XP is a constituent that has moved leaving a copy behind,
which also has a copy of the unvalued features of the head of the chain:
(58) XP[] . . . copy(XP[])
The unvalued features of XP and its copy are co-valued and this is what makes
XP and its copy form a chain. This co-valuation of features should be uncontroversial: if XP and its copy did not have their features co-valued, they would
not be copies of each other. Therefore, the relation between the links of a chain
is no different from the relation between expletive and associate (pursuing intuitions here that hark back to Chomsky 1986), and both are simply an agreement relation involving co-valued features. When the features of XP are valued
by a probe, those of its copy are too and can delete, as was the Case with the
expletive-associate relation.
This entails, of course, that Locality of Agreement also holds of the links of
a chain, as much as any other agreement relation. However, the formulation in
(17) is limited to agreement relations that involve a probe and a goal, and here
we have a different instance of feature co-valuation, one which does not involve
a probe. We need a definition of Locality broader than (17), so that it embraces
also agreement without a probe. I propose the following informal definition,
which subsumes Locality of Agreement:
(59) Locality of Co-Valued Features (LCVF)
Two features and of the same type can be co-valued if there is no Xmax
such that Xmax dominates one of them but does not dominate the other.
Lets now return to Superraising in light of LCVF. The links of a chain cant
skip over a spec position because an Xmax acts as a barrier and disrupts the
On Agreement
connection between co-valued features. If a moving term skips a spec, a maximal category now is going to interpose between two links of a chain, disrupting
the connection between the features. Thus, even if the head of the chain had
its features valued and deleted, the foot of the chain would not, and this would
lead to a crashed derivation at PF. Consider the structure in (60), under the
crucial assumption that the head H only accepts one spec:
(60) XP . . .[HP YP H [. . . copy(XP) ]]
XP had to skip Spec,H, which is taken by YP. Now XP and its copy cant Agree
without violating LCVF because of the intervening Spec,H. As a result, the unvalued features of copy(XP) are never valued and deleted and the derivation
cant spell-out at PF.
(60) raises the issue of whether a head should allow for multiple specifiers. I agree with Chomsky (1995) that the theory of bare phrase structure
does not include any such restriction. Additionally, it seems likely that C accepts more than one spec of the A variety, given the existence of multiple whmovement in some languages. On the other hand, there does not seem to be
any reason why particular languages could not restrict the number of specs of
a given functional head to one as a lexical property of that head which can be
parametrized.
Lets assume that in some languages, T allows for only one A-spec, giving rise to superraising violations; other languages would not impose any restrictions and could consequently tolerate superraising violations. This correlation between multiple specifiers and grammaticality of superraising is exactly the conclusion that Ura (1994, 1996) arrived at after investigating a wide
range of languages. Notice that Chomskys freezing cannot accommodate
Uras findings, because Chomskys approach, in which a head attracts/piedpipes a DP unless there is an intervening DP, would predict that no language
would ever allow for superraising. On the other hand, a theory that allows for
movement to be triggered by the features of the item that moves, and not by
the features of the attractor, can incorporate Uras findings if coupled with a
parametrization of the multiple specs option. Therefore, Uras proposal can
fit into LCVFs unproblematically: if a language allows for multiple specs, an
XP can leapfrog (to use McGinnis graphic metaphor) another constituent
without violating (59).
Thus, Relativized Minimality does not need to be stipulated in my framework since it derives from (59), an overall restriction on co-valued features which is independently needed. Additionally, it allows us to accommo-
Luis Lpez
On Agreement
Theory was simply a theoretical primitive. It was a formal system built into
the CHL that we needed to assume in order to account for some pesky data (ie:
raising of internal arguments to Spec,T in unaccusative and passive constructions). This kept the community of syntacticians more or less satisfied until the
arrival of the Minimalist Program started to demand more challenging standards of explanation and propose bolder hypotheses about the constitution of
CHL . The boldest and most challenging hypothesis, without doubt, is the suggestion that CHL contains no imperfections, that the system is designed as if
a super-engineer had planned it. Within this point of view, Case features, and
uninterpretable features more generally, are problematic: why should Tense, or
a participle, show agreement with a DP? Why should DPs have Case? Chomsky
(1998) proposes that uninterpretable features are there to provoke displacement, and the latter is meant to give rise to differences in surface semantic
interpretation. Uninterpretable features are not an imperfection, they exist
for a reason.
Lets very briefly summarize how he analyzes Scandinavian Object Shift.
As is well known, Scandinavian languages have the option of moving an object to a position that we can identify as Spec,v. An object in Spec,v is interpreted as specific/presupositional while a lower object is interpreted as nonspecific/existential. According to Chomsky, what pied-pipes an object to Spec,v
is an optional EPP feature of v. This EPP feature is licensed because it has an effect on outcome, namely, the specific interpretation. If there is no EPP in v, the
object simply stays in situ, where it can enter an Agree relation with v and receive a focus interpretation (and here I abstract away from the tight restrictions
on OS summarized under the rubric Holmbergs Generalization).
Let me suggest an alternative to Chomskys analysis within the confines of my framework which makes no use of EPP features to provoke displacement. As I suggested in Section 2, we can assume that a Case feature can be added to some functional heads as they are drawn from the
lexicon. Thus, v can be [accusative] or [partitive], according to the version of v. Assume another functional head, call it F, that selects for vP.
Assume further that F can also bear a Case feature. Then, we can have
the situation in which the feature [accusative] is added to F and not to
v. Following my assumptions concerning Move, object will have to raise
to spec,v to value its unvalued Case against F. The interpretation rule for
this object works exactly as in Chomskys system and object in Spec,v becomes specific.
Luis Lpez
focused OB
specific OB
This, of course does not amount to an analysis (but see Lpez 2000 where the
details of this approach are developed). Rather, I only want to show that within
my framework it is possible to explore Chomskys hypothesis that the Case
system is indeed built in for surface semantic interpretation.
Notes
. Frampton and Guttman (2000) hit on a similar idea independently, which they call
feature coalescence. I find this coincidence deeply reassuring.
. The Swahili example suggests that subject movement stops at every available step before
reaching its final destination. We find evidence in many languages, including English, for
successive cyclic A-movement of subject as it goes to Spec,T, as Sportiche (1988) pointed
out. Consider sentence (i):
(i)
a.
b.
The fact that the floating quantifier can appear in several places between the final and the
initial positions of the subject would suggest successive cyclic movement. However, Chomskys assumptions do not predict this, rather, they would predict just one movement from
Spec,v to Spec,T triggered by the EPP feature of T. The only way he could open the door for
this sort of successive movement is by stipulating that each modal/aux has a separate EPP
feature and only T is -complete, so it can assign Case.
Baltin (1995) and Bobaljik (1995) cast some doubt on the assumption that floating quantifiers indicate the path taken by the suject as it raises, proposing instead that the original idea
that floating quantifiers are adverbs was right. However, floating quantifiers agree with the
displaced constituents in languages with rich agreement morphology, a most unusual behavior for an adverb. Alternatively, it could be said that they are secondary predicates, which
may also display overt agreement, but the distribution of floating quantifiers does not match
that of secondary predicates:
(i)
On Agreement
complement DP (which needs Case) or a complement CP/PP (which does not), a situation
that can only occur if [accusative] is added freely.
. In a more articulated CP structure, like the one in Rizzi (1997), the Case assigning head
could be Finite0 .
. Chomsky (1999) analyses (12) as an instance of long distance agreement in which the
participle, which is not -complete, can be by-passed by the upstairs probe. A criticism of
-completeness is in Section 1.
. The idea that a spec position could be governed from outside was debated frequently during the GB era, generally agreeing that specs are accessible to outside governors, although
some of the empirical reasons we might not find as compelling nowadays. In Kaynes (1994)
system, specs are adjunctions, so ZP in (18) would also not be dominated by XP. In Chomskys (1995) Bare Phrase Structure, adjunction to Xmax is banned because the resulting structure of segments, it is said, would be uninterpretable. However, Chomskys recent notion of
edge of a phase that can be probed from outside again opens the door for a porous Xmax .
. Several interesting proposals are not going to be discussed explicitly in this section,
among them Collins (1997) Last Resort and Lasniks (1995) Enlightened Self-Interest.
. Attract Feature was very hard to define and raised difficult problems. For instance,
as Frampton and Gutmann (1999) point out, it is unclear what kind of chain results from
feature movement. Moreover, if a head detects a feature that matches, thus establishing a
relation between attractor and attractee, it is unclear why attraction is further required.
. If movement is seen as expulsion from a configuration and remerge in another there is
no room for a DP to get trapped in a loop, as it merges endlessly with the same XP, each
time creating a new spec. Each of this remerges would be to the same configuration. Once
an item has been expelled from Spec,X, it cant go back to it.
. As for German, some speakers that I consulted accept (i), but others do not:
(i)
Given this uncertainty, I will leave German aside when discussing TECs.
. In the analysis of German, English and French expletives I adopt the non-trivial assumption that in these languages v does not accept more than one A-spec. This prevents
object from raising higher than Spec,X. This is an issue that we are going to encounter again
later on.
. Lasniks (1995) conclusion is different: he claims that adjacency is evidence for Case
assignment by v (or equivalent). In my terms, adjacency simply indicates that object moved
to Spec,X (remaining neutral as to whether it gets Case from v or not).
. Examples like the ones in (i) could cast some doubt on the assertion that transitive verbs
do not accept expletives:
(i)
a.
b.
Luis Lpez
However, notice that the verbs in these sentences are conceptually unaccusative (hit means
arrive in (i-a)) except that they can assign Case. Phrase-structurally, the correct analysis
for this type of sentence would probably have both DPs as arguments of the root, leaving Spec,v free for the expletive. Case assignment is a challenge for the v-as-Case-assigner
theories generally.
. Examples (41b) and (41c) are taken from Alice Munros short story Meneseutung, in
the volume Friend of my Youth, published in 1990. Other examples tend to sound awkward,
although it is hard to tell if the reason is the syntax or the information structure. Sentence
(i) is considered ungrammatical by Chomsky (1999: 15), but (ii) sounds much better with
only a small change in word order:
(i) *There came several angry men into the room.
(ii) ?Into the room there came several angry men.
In (ii), the DP is focused, whereas in (i) it is not. It seems correct to assume that expletive
constructions fulfill the information structural function of focusing the DP. The contrast between (i) and (ii) arises because the DP can be more naturally focused at the end of the sentence. This seems to support Chomskys claim that the ungrammaticality of (i) shows that
English object must be dislocated in unaccusative/passive constructions (see also examples
in the previous footnote).
. Or it may adjoin to v, as proposed by Boskovic (1997).
. Due to the inadequacy of my word processor I represent the Icelandic interdental fricatives as -th- and -dh-.
. Chomsky (1995) eliminates AgrPs because they have no effect on either interface. However, AgrPs may play a role in the Chomsky (1999) model. Assume AgrPs are there to assign
Case features. Further, assume that checking Case features is connected with surface interpretability (following Chomskys (1998, 1999) hypothesis that imperfections are not real
but part of an optimal design, see Section 7 of this paper). Then AgrPs do have an effect on
outcome, so they are licensed to live.
. This analysis entails that there are two sources for nominative Case in Icelandic.
Uriagereka (1995: 167) reaches the same conclusion for constructions usually referred to
as nominativus pendens in Romance languages. The following is his example from Galician:
(i)
In (i), both meu fillo and a fame are in nominative Case. In our terms, meu fillo would get it
from C and a fame from Agr.
On Agreement
References
Arad, M. (1998). VP-Structure and the Syntax-Lexicon Interface. Doctoral dissertation,
University College London.
Arad, M. (This volume). On Little v.
Baltin, M. (1995). Floating Quantifiers, PRO and Predication. Linguistic Inquiry, 26, 199
248.
Bejar, S. & D. Massam (1999). Multiple Case Checking. Syntax, 2, 6579.
Belletti, A. (1988). The Case of Unaccusatives. Linguistic Inquiry, 19, 134.
Bobaljik, J. (1995). Morphosyntax: The syntax of verbal inflection. Doctoral dissertation,
MIT.
Bobaljik, J. & D. Jonas (1996). Subject Positions and the Roles of TP. Linguistic Inquiry, 27,
195236.
Bobaljik, J. & H. Thrinsson (1998). Two Heads Arent Always Better than One. Syntax, 1,
3771.
Boskovic, Z. (1997). The Syntax of Non-finite Complementation. Cambridge, MA: The MIT
Press.
Branchadell, A. (1992). A Study of Lexical and Non-lexical Datives. Doctoral dissertation,
Universitat Autnoma de Barcelona, Spain.
Carstens, V. (2000). Concord in Minimalist Theory. Linguistic Inquiry, 31, 319356.
Carstens, V. & K. Kinyalolo (1989). On IP-structure: Tense, aspect and agreement.
Unpublished manuscript, Cornell University and U.C.L.A.
Chomsky, N. (1981). Lectures on Government and Binding. Dordrecht: Foris.
Chomsky, N. (1986). Barriers. Cambridge, MA: The MIT Press.
Chomsky, N. (1993). A Minimalist Program for Linguistic Theory. In K. Hale and J. Keyser
(Eds.), The View from Building 20 (pp. 145). Cambridge, MA: The MIT Press.
Chomsky, N. (1995). Categories and Transformations. In The Minimalist Program, 219394.
Cambridge, MA: The MIT Press.
Chomsky, N. (1998). The Minimalist Program: The framework [MIT Occasional Papers in
Linguistics 15]. Department of Linguistics and Philosophy, MIT, Cambridge, MA.
Chomsky, N. (1999). Derivation by Phase [MIT Occasional Papers in Linguistics 18].
Department of Linguistics and Philosophy, MIT, Cambridge, MA.
Collins, C. (1997). Local Economy. Cambridge, MA: The MIT Press.
Collins, C. & H. Thrinsson (1996). VP-Internal Structure and Object Shift in Icelandic.
Linguistic Inquiry, 19, 131174.
Demonte, V. (1995). Dative Alternation in Spanish. Probus, 7, 530.
Diesing, M. (1992). Indefinites. Cambridge, MA: The MIT Press.
Emonds, J. (1976). A Transformational Approach to English Syntax. New York: Academic
Press.
Frampton, J. (1997). Expletive Insertion. In C. Wilder, H.-M. Grtner and M. Bierwisch
(Eds.), The Role of Economy in Linguistic Theory (pp. 3657). Berlin: Akademie Verlag.
Frampton, J. & S. Gutmann (1999). Cyclic Computation, a Computationally Efficient
Minimalist Syntax. Syntax, 2, 127.
Frampton, P. & S. Guttman (2000). Agreement is Feature Sharing. Unpublished manuscript,
Northeastern University.
Luis Lpez
Groat, E. (1999). Raising the Case of Expletives. In S. Epstein and N. Hornstein (Eds.),
Working Minimalism (pp. 2743). Cambridge, MA: The MIT Press.
Grohmann, K. (To appear). A Movement Approach to Contrastive Left Dislocation. Rivista
di Gramatica Generativa.
Haegeman, L. (1992). Theory and Description in Generative Grammar: A Case Study in West
Flemish. Cambridge: CUP.
Halle, M. & A. Marantz (1993). Distributed Morphology and the Pieces of Inflection. In K.
Hale and J. Keyser (Eds.), The View from Building 20 (pp. 111176). Cambridge, MA:
The MIT Press.
Holmberg, A. (1999). Remarks on Holmbergs Generalization. Studia Linguistica, 53, 139.
Johnson, D. & S. Lappin (1999). Local Constraints vs. Economy. Stanford, CA: CSLI
publications.
Johnson, K. (1991). Object Positions. Natural Language and Linguistic Theory, 9, 577636.
Kayne, R. (1989). Facets of Romance Past Participle Agreement. In P. Beninc (Ed.), Dialect
Variation and the Theory of Grammar (pp. 85103). Dordrecht: Foris.
Kayne, R. (1994). The Antisymetry of Syntax. Cambridge, MA: The MIT Press.
Koopman, H. & D. Sportiche (1991). The Position of Subjects. Lingua, 85, 211258.
Kratzer, A. (1996). Severing the External Argument from its Verb. In J. Rooryck and
L. Zaring (Eds.), Phrase Structure and the Lexicon (pp. 109138). Dordrecht: Kluwer.
Larson, R. (1988). On the Double Object Construction. Linguistic Inquiry, 19, 335391.
Lasnik, H. (1992). Case and Expletives: Notes toward a parametric account. Linguistic
Inquiry, 23, 381405.
Lasnik, H. (1995). Case and Expletives Revisited: On greed and other human failings.
Linguistic Inquiry, 26, 615634.
Longobardi, G. (1994). Reference and Proper Names: A theory of N-movement in syntax
and logical form. Linguistic Inquiry, 25, 609665.
Lpez, L. (2000). The Edges of Catalan. Ms., University of Illinois-Chicago.
Lpez, L. (2001). On the (Non)complementarity of -theory and Checking Theory.
Linguistic Inquiry, 32(4).
Marantz, A. (1997). No Escape from Syntax: Dont try morphological analysis in the privacy
of your own lexicon. In A. Dimitriadis (Eds.), Proceedings of the 21st Annual Penn
Linguistics Colloquium [University of Pennsylvania Working Papers in Linguistics 4.2]
(pp. 201225).
Masullo, P. (1992). Incorporation and Case Theory in Spanish. A Crosslinguistic Perspective.
Unpublished dissertation, University of Washington.
May, R. (1985). Logical Form. Cambridge, MA: The MIT Press.
McGinnis, M. J. (1998). Locality of A-Movement. Doctoral dissertation, MIT.
Platzack, C. (1986). COMP, INFL and Germanic Word Order. In L. Hellan and K. Andersen
(Eds.), Topics in Scandinavian Syntax (pp. 185234). Reidel: Dordrecht.
Pollock, J.-I. (1989). Verb Movement, UG and the Structure of IP. Linguistic Inquiry, 20,
365424.
Postal, P. (1974). On Raising. Cambridge, MA: The MIT Press.
Rizzi, L. (1990). Relativized Minimality. Cambridge, MA: The MIT Press.
Rizzi, L. (1997). The Fine Structure of the Left-periphery. In L. Haegeman (Ed.), Elements
of Grammar (pp. 281337). Dordrecht: Kluwer.
On Agreement
Roberts, I. (1998). Have/Be Raising, Move F and Procrastinate. Linguistic Inquiry, 29, 113
126.
Romero, J. (This volume). Morphological Constraints on Syntactic Derivations.
Ross, J.-R. (1983). Inner Islands. Unpublished manuscript, MIT.
Selkirk, E. (1994). Sentence Prosody, Intonation, Stress and Phrasing. In J. Goldsmith (Ed.),
The Handbook of Phonological Theory (pp. 550569). New York: Blackwell.
Sigurhsson, H. (1991). Icelandic Case Marked Pro and the Licensing of Lexical Arguments.
Natural Language and Linguistic Theory, 9, 327363.
Sportiche, D. (1988). A Theory of Floating Quantifiers and its Corollaries for Constituent
Structure. Linguistic Inquiry, 19, 425449.
Stowell, T. (1981). Origins of Phrase Structure. Doctoral dissertation, MIT.
Suer, M. (1988). The Role of Agreement in Clitic-Doubled Constructions. Natural
Language and Linguistic Theory, 6, 391434.
Ura, H. (1994). Varieties of Raising and the Feature-Based Bare Phrase Structure Theory
[Occasional Papers in Linguistics 7]. Cambridge, MA: The MIT Press.
Ura, H. (1996). Multiple Feature-Checking: A Theory of Grammatical Function Splitting.
Doctoral dissertation, MIT.
Uriagereka, J. (1995). An F position in Western Romance. In K. Kiss (Ed.), Discourse
Configurational Languages (pp. 153175). New York: OUP.
Vickner, S. (1995). Verb Movement and Expletive Subjects in the Germanic Languages. New
York: OUP.
Yang, C. (1999). Unordered Merge and Its Linearization. Syntax, 2, 3864.
Zagona, K. (1982). Government and Proper Government of Verbal Projections. Doctoral
dissertation, University of Washington.
Zwart, J.-W. (1997). The Morphosyntax of Verb Movement: A minimalist approach to the
syntax of Dutch. Dordrecht: Kluwer.
A minimalist account
of conflation processes
Parametric variation at the
lexicon-syntax interface*
Jaume Mateu and Gemma Rigau
Introduction
The main purpose of this paper is to show that the conflation processes involved in so-called lexicalization patterns (see Talmy 1985) can receive an adequate explanation when translated into syntactic terms. An analysis of these
conflation processes in purely semantic terms like that put forward by Talmy
(1985) can be said to be descriptively adequate, but the parametric variation
to be found in such processes will be seen to crucially involve morphosyntax,
not only semantics (see Snyder 1995).
First of all, it will be necessary to review some of the main insights of
Talmys work. As is well-known, this cognitive linguist claims that languages
can be classified according to how semantic components like Figure, Motion,
Path, Manner, or Cause are conflated into the verb. For example, conflation of
motion with path is argued to be typical of Romances languages like Spanish
(see (1)), whereas conflation of motion with manner is typical of English (see
(2)). The examples in (1) and (2) are all drawn from Talmy (1985: 69f).
(1) a.
La
the
b. La
the
c. El
the
botella entr
bottle went+into
botella sali
bottle went+out
globo subi
balloon went+up
a la cueva flotando.
(Spanish)
to the cave floating
de la cueva flotando.
of the cave floating
por
la chimenea flotando.
through the chimney floating
d. El
the
e. La
the
(2) a.
b.
c.
d.
e.
globo baj
por
balloon went+down through
botella se alej
de
la
bottle went+away from the
la chimenea flotando.
the chimney floating
orilla flotando.
bank floating
In fact, Spanish and English can be regarded as two poles of a typological dichotomy that Talmy (1991) characterized as verb-framed languages versus
satellite-framed languages. Given this distinction, there are languages encoding the path element into the verb: for example, consider the Spanish path verbs
entrar go in(to), salir go out, subir go up, etc. By contrast, other languages
do not incorporate the path into the verb but leave it as a satellite around the
verb. According to Talmy, the latter option is typically found in the majority
of Indo-European languages (Romance being excluded). When the path remains as a satellite, one option becomes available: the manner component (for
example, floating in the examples in (2)) can be encoded into the verb.
The well-known elasticity of the verb meaning in English (cf. Rappaport
Hovav & Levin 1998) can be exemplified with data involving not only conflation of motion with manner (see (2)), but also conflation of causation with
manner (see the examples in (3), drawn from Levin & Rapoport 1988: 279).
The fact that the directionality or path component remains as a satellite in English allows the manner component (e.g., brushing) to be conflated into the
causative verb in (3). As expected, the lexicalization pattern corresponding to
the Romance languages (i.e., the path incorporates into the verb, saturating it
lexically) prevents them from having the kind of verbal elasticity in (3), the
manner component being then forced to be expressed as an adjunct if necessary: e.g., cf. Sp. ella quit las hilas con un cepillo/cepillando (lit.: she took+out
the lint with a brush/brushing).
(3) a.
b.
c.
d.
e.
f.
Notice that it is precisely the conflation of the motion or causation verb with
manner what accounts for those cases where the construction rather than the
verb has been argued to determine the argument structure (see Jackendoff
1990, 1997 or Goldberg 1995). As shown in Jackendoff (1990, 1997), constructions like those in (4) through (6) have syntactic and semantic restrictions
of their own and, in this sense, it is indisputable that each of them deserves
the status of constructional idioms. Moreover, Jackendoff (1997: 554f) noted
that these constructions can be considered instances of a more general abstract
construction, the verb subordination archi-construction in (7).
(4) Ones way construction:
e.g., He moaned his way out of the room.
a. [VP V [bound pronoun]s way PP]
b. go PP (by) V-ing
(5) Resultative construction:
e.g., He wiped the table clean.
a. [VP V NP {AP/PP}]
b. cause NP to become AP/go PP by V-ing (it)
(6) Time-away construction:
a. [VP V NP away]
b. waste [Time NP] V-ing
Although we do not have any problem in attributing the status of constructional idioms to the constructions in (4)(6) in the sense that each of them has
its own set of syntactic and semantic peculiarities, we want to show that Jackendoff s (1997) Verb Subordination Archi-construction in (7), as it stands,
can be regarded as an epiphenomenon, once a principled account of the parametric variation in the lexicon-syntax interface is taken into account.1
Quite importantly, we claim that the relevant explanation of the parametric
issue concerning the existence of (3)(6) in English, but not in Romance, cannot be formulated in purely semantic or aspectual terms, since it can be argued
to have nothing to do with the positive or negative application of some ad hoc
operations over the Lexical Conceptual Structure (LCS) (Levin & Rapoport
1988), the Aspectual Structure (Tenny 1994), or the Event Structure (Pustejovsky 1991), but with one empirical fact: i.e., the syntactic properties associ-
ated with the lexical element encoding directionality are not the same in English as in Romance (cf. Snyder 1995 and Klipple 1997 for two proposals in
tune with our syntactic account).
Semanticocentric analyses run into problems when language variation is
taken into account, since no principled explanation can be given to why some
languages (e.g., Romance) appear to lack the relevant LCS operation, the aspectual operation or the event-type shift strategy involved in the conflation processes in (2) and (3). Accordingly, we will take pains to show that the solution
of such a problem cannot be stated in purely semantic or aspectual terms.
lexicon
computational system
spell-out
PF
sensoriomotor systems
systems of thought
Let us begin with the semantic properties that must be optimally coded into the
lexical entry. We will assume that the semantic information to be located in the
lexicon is the optimal information required by the computational system. It
is widely acknowledged that lexical entries include semantic features entailing
their corresponding categorial features.2 These lexically encoded semantic features will have to be interpreted at the interface level LF. Two classes of semantic
features can be distinguished: non-relational features vs. relational features. The
former entail the syntactic category Noun (N), whereas the latter entail the
categories considered as syntactic predicates: Verb (V) and Particle (P).
For our present purposes, P is to be regarded as a cover birelational term
for Adposition, Adjective, and Adverb. So-called Adpositions are pure Particles, whereas Adjectives and Adverbs can be seen as complex Particles that incorporate a non-relational element.3 This proposal nicely captures the argument structure similarities of sentences like those in (9). All of them turn to
share the same syntactic structure, that in (10) (where functional categories
have been omitted).
(9) a. The cat is in the room.
b. The cat is happy.
c. The cat is here.
P
N
P
P
dance
a.
phonological properties
b.
c.
V (<causal relation)
v[N V]
What is meant by (11b) is that the categorial property V is entailed by the semantic feature, i.e., the causal relational feature. The fact that dance has tense
and phi-features will not be indicated in the lexical entry, since that much
is determined by its category V (presumably by UG), as noted by Chomsky
(1995: 238). Finally, what is meant by (11c) is that a N must be incorporated
into the verb dance (see Hale & Keyser 1993, 1998). This information is clearly
idiosyncratic, and hence it must be encoded into the lexical entry. As we will
see below, the information optimally encoded into lexical entries will be argued
to be crucially relevant when dealing with the crosslinguistic variation involved
in Talmys conflation processes.
We can now concentrate on the semantic properties that must be located in
the output conditions on LF, the interface linguistic level related to systems of
thought. It is clear that LF has to meet certain legibility conditions in order for
systems of thought to access this interface level (Chomsky 1998: 7). According
to Chomsky (1998: 27f), bare output conditions on LF include Binding conditions, the Case Theory, the Chain condition, the Projection Principle, etc. The
legibility conditions we are interested in at present are those concerning the
Projection Principle. We will assume that the Projection Principle conditions
govern the relation among those three basic syntactic objects depicted in (12)
(where the X in (12a) is to be regarded as a variable: it is N in unergative structures, and P in transitive and unaccusative structures). We will also assume that
they govern the syntax-semantics associations depicted in (13). As a result, notice that there appears to be a strong homomorphism between the syntax and
semantics of argument structure at LF.6
(12) a. V
V
b.
X
c. N
P
N
P
P
(13) a.
V is to be associated to an eventive relation: if there is an external argument, it is interpreted as a causal relation; otherwise it is interpreted
as a transitional relation.
b. P is to be associated to a non-eventive relation.
c. N is to be associated to a non-relational element.
pattern corresponding to English (i.e., conflation of manner into the {motion/causation}verb). Secondly, we will show why this lexicalization pattern
does not hold for Romance languages, where the relevant lexicalization pattern
involves conflation of path into the verb.
Consider the examples in (14):
(14) a. Sue danced.
b. Sue danced across the room.
c. John danced Sue across the room.
It has often been noted in the literature that unergative verbs in English can
be unaccusativized when a directional PP is added (see Hoekstra 1984; Levin
& Rappaport Hovav 1995; and Ritter & Rosen 1998, among others). One interesting question to be solved is why the so-called unaccusativization process
involved in (14b) does not take place in some languages, e.g., in Spanish. As we
will see below, our proposal is that the solution is to be found in the different
syntactic properties associated to the lexical element encoding directionality in
English vs. Spanish.
In order to get the syntactic derivation involved in (14b), it is required
that the lexical subarray contain the substantive categories in (15), where their
corresponding lexical entries can be argued to include those three kinds of information which we have exemplified with the unergative verb dance in (11).
We put functional categories aside here.9
(15)
dance
go
across
Sue
room
phonological
properties
V (<causal
V (<transitional P (<spatial
N (<non-relational N (<non-relational
element)
relation)
relation)
element)
relation)
v[N V]
We assume that the lexicon of satellite-framed (i.e., non-verb-framed) languages like English has a phonologically null verb expressing dynamic transition, besides its phonetically realized correspondent. We represent this empty
unaccusative verb in boldface: go. By virtue of expressing a dynamic transition,
this unaccusative verb subcategorizes for a PP denoting a directional spatial relation, which in turn relates two non-relational elements: Sue (i.e., the Figure),
and room (i.e., the Ground). The syntactic object in (16) is the result of merging
go with the birelational P headed by across:
(16)
V
V
go
P
N P
Sue
P N
across room
V
V N
go room
However, as it stands, the syntactic computation of (16) would not be convergent at PF, because the verb go, being devoid of phonological properties, would
not be interpretable or legible at the interface level with sensoriomotor systems.
In order to avoid its crashing at PF, it is required that the empty verb be conflated with another element with phonological properties. The unergative verb
dance represented in the numeration in (15) turns out then to be adjoined
to the phonologically null unaccusative verb by means of Merge. As a result,
the conflation of dance with go will be spelled out as dance. Its corresponding
syntactic representation is given in (18) (functional categories omitted).
(18)
V
V
V V N P
dance go Sue
P N
across room
Given (18), our claim is that the generalized transformation used by Hale &
Keyser (1997) in their account of sentences like (14b) is not but an instantiation of Merge. By using this operation, we provide the empty unaccusative
verb with the phonological features needed for it to be legible at the interface
level with sensoriomotor systems. This is in accordance with Chomskys claim
that syntactic operations can be argued to be used in order to satisfy external
conditions.
Let us now analyze (14c) John danced Sue across the room. As noted by
Ritter and Rosen (1998: 140141, 157158), a sentence like (14c) does not alternate with (14a) or (14b), but with John danced. That is to say, according to
them, (14c) is not an example of the well-known causative-inchoative alternation. In their event-based approach, a delimiting predicate (e.g., across the
room) is posited to be added to the activity verb dance in sentences like (14c),
the former predicate licensing then a delimiting argument (e.g., Sue). Accordingly, Ritter and Rosen point out that the fact that John is the subject of the
verb dance and Sue an argument of the secondary predicate across the room,
explains why the object must not necessarily be engaged in the action denoted
by the verb: for example, they note that John must be dancing in (14c) but Sue
could be a doll John is holding as he dances.
Although we assume Ritter & Rosens empirical considerations, we have
our qualms on their event-based analysis, according to which a delimiter
phrase is said to be added to the activity verb, this secondary predicate licensing then a delimiting object. Rather our syntactic analysis of the conflation
process involved in (14c) takes the configuration in (19) as the main or basic
one, i.e., that formed by merging a PP headed by across with the lexical item in
(20), which represents a phonologically null causative verb. As above, we express the non-verb-framed nature of English path constructions by positing a
null verb like that in (20).
(19)
V
V
cause
P
N P
Sue
P N
across room
(20)
cause
no phonological properties
V (<causal relation)
V
V
V
V N P
dance cause Sue
P N
across room
On the other hand, note that the same syntactic conflation process represented
in (21) appears to be involved in examples like those in (22), where the path
represented by the P(article), which is a complex one in (22ac) (cf. (9bc)),
is not incorporated into the verb, this requiring the conflation of the empty
causative predicate with an unergative verb by means of Merge.10
(22) a.
b.
c.
d.
e.
f.
Wechsler (1995)
Er
He
b. Er
He
schwamm aus
dem Gefngnis.
swam
out.of the prison
hat sich aus
dem Gefngnis geschwommen.
has refl out.of the prison
swum
(German)
According to Seibert (1992: 66), the adverbial <in (23b)> does not denote a
place the subject reaches as a natural result of swimming, i.e., the person might
have been swimming in a completely different place, or the person may have
never left the prison while actually swimming. By contrast, the adverbial out
of the prison in (23a) does denote a place the subject reaches as a natural result
of swimming. Seiberts comments on (23) can be explained on the basis of our
conflation analysis in quite an elegant way. While (23a) involves merging the
verb schwimmen (swim) with the null verbal element corresponding to the
transition (i.e., go), (23b) involves merging schwimmen with the null verbal element corresponding to the causation (i.e., cause), this being in full accordance
with the interpretive effects noted by Seibert. That is to say, (23a) can be analyzed as (24), whereas (23b) can be analyzed as (25) (As above, we assume that
the external argument in (23b) (i.e., Er he) is to be introduced by the relevant
functional projection omitted here).
(24)
V
schwimmen
V N
go Er
P
P N
aus Gefngnis
(25)
V
V
V
V N P
schwimmen cause sich
P N
aus Gefngnis
So far our syntactic analysis of the lexicalization pattern typical of English, that
involving conflation of manner into the verb.11 Let us now deal with the lexicalization pattern corresponding to conflation of path into the verb, which has
been argued to be characteristic of Romance languages.12 As noted by Talmy,
in Spanish the (telic) directional or path element is conflated into the motion
verb. To put it in our present terms, the lexical entry of Spanish path verbs like
entrar (to go in) contains the information that a telic path particle is incorporated into the verb. Since the verb-framed nature of Spanish path constructions
is a fossilized property (this being due to the diachronic evolution of this language), it is clear that each verbal lexical entry affected by such a fossilization
will have to reflect it. For example, the lexical entry of entrar (to go in) will
have to contain information like that depicted in (26):
(26)
entrar
phonological properties
V (<transitional rel.)
v [P V]
Given (26), we can now explain why Spanish lacks constructions like (27b):
(27) a.
Sue bail.
Sue danced
b. *Sue bail a la habitacin.
Sue danced to the room
American Spanish
Sue entr
a
Sue went+into (in)to
b. European Spanish
Sue entr
en
Sue went+into in(to)
la habitacin (bailando).
the room
(dancing)
la habitacin (bailando).
the room
(dancing)
The syntactic stucture associated to (28a) is that in (29), the adjunct bailando
(dancing) being omitted from the syntactic argument structure:
(29)
V
V
entr
P
N
Sue
P
P
a
N
habitacin
The examples in (28) could be argued to pose a potential problem for our analysis of lexical incorporation:13 why is it the case that the lexically incorporated
prepositional complement of the unaccusative verb can reappear again in syntax (cf. a/en la habitacin into the room)? We think that the fossilized kind of
incorporation of P into the verb entrar is crucial in order to understand why the
prepositional complement reappears: P is always projected in the syntax, this
being a copy of the P incorporated into the verb. Otherwise, note that there
would not be any internal specifier position available for the subject of the unaccusative sentence. This copy can be pronominal, as in the Catalan example
(30a), or phonologically null when recovered via deixis (see 30b).14
(30) a.
La Sue
Sue
b. La Sue
Sue
hi
entr.
loc.clitic went+into
entr.
went+into
(Catalan)
So far we have been dealing with cases where those two lexicalization patterns
concerning conflation of motion with manner and that of motion with path
do not coincide in a unique language. This is the case in English and Spanish:
recall Talmys proposal that while English usually lacks conflation of path into
the verb, Spanish lacks conflation of manner into the verb.
However, it is interesting to notice that there are some languages that appear to combine both options, as shown by the Dutch data in (31), drawn from
van Hout (1996), and by the Russian data in (32), drawn from Spencer & Zaretskaya (1998). Nevertheless, as we will see below, the incorporation of the Particle into the verb in the following data must not be analyzed in the same way
as in the Spanish examples we have just dealt with above.
(31) a.
John is weg-gelopen.
John is away-walked
John walked away.
(Dutch)
(Russian)
It has often been said in the literature that there is an unaccusativization process
involved in (31)(32). The basic verb is said to be unergative, but the syntactic
construction where it appears turns out to be unaccusative when a directional
element is present. For example, note that the auxiliary selected in the Dutch
data is zijn (be) (see Hoekstra 1984).
Our analysis of (31a) to be presented below can be argued to hold for the
rest of the data in (31b) and (32). As assumed above when dealing with English
data like (14b) Sue danced across the room, we want to propose that the lexicon of both Dutch and Russian contains a phonologically null verb denoting
dynamic transition, besides its phonetically realized correspondent. Following
our present convention, we represent this phonologically null unaccusative verbal element in boldface: for expository reasons, let us call it go once again. As
noted, it is required that the empty verb be conflated with another element
with full phonological properties in order for the former to be interpretable
or legible at the interface level with sensoriomotor systems. To avoid its crashing at PF, the unergative verb (ge)lopen (walk(ed)), which has also been selected from the lexical subarray, turns to be adjoined to the phonologically null
unaccusative verb by means of Merge. As a result, the conflation of (ge)lopen
with go will be spelled out as (ge)lopen. Its corresponding simplified syntactic
configuration is given in (33).
(33)
V
V
V
V N P
(ge)lopen go John weg[affix]
the preverbs.15 Consequently, this P will have to move to the superior verbal
head, adjoining to it. In this case, Move is clearly justified because of the affixal
status of P.
On the other hand, it should be clear that in (31) and (32), the incorporation of the P(article) into the verb is not a fossilized process, as it is in Spanish.
Crucially, notice that the morphological analysis of the verbs in (31) and (32)
is quite transparent: the prefix corresponding to the P(article) can be easily
identified. By contrast, Spanish path verbs like entrar (to go in), bajar (to go
down), subir (to go up), etc., constitute morphophonological atoms (that is,
what corresponds to the P(article) and what corresponds to the verb in such
verbs cannot be distinguished synchronically any longer), this being due to the
above-mentioned fossilization process. Accordingly, it should not be surprising
that the fossilized status of the incorporation of the telic P into the transition
verbal head prevents Spanish from merging an unergative verbal head expressing what Talmy refers to as manner into a null unaccusative verb expressing
dynamic transition. By contrast, the non-fossilized nature of the incorporation of P into the verb in (31)(32) allows a subordinate unergative verb to be
merged into the null unaccusative verb of the path of motion construction.
Spanish
En este rbol anidan cigeas.
In this tree nest-3pl storks
There are some storks nesting in this tree.
b. Central Catalan
(En aquest esbart,) hi
ballaran
adolescents.
In this
group, loc.cl. will-dance-3pl teenagers
There will be some teenagers dancing here (in this group).
c. N
hi
ballaran molts.
part.cl. loc.cl. will-dance many
Many of them will dance there.
Spanish
En este rbol hay
cigeas {anidando/que anidan}.
in this tree has-loc.cl. storks
{nesting/that nest-3pl}
There are some storks nesting in this tree.
b. Central Catalan
En aquest esbart, hi hauran adolescents {ballant/que ballaran}.
in this group, loc.cl. will-have-3pl teenagers {dancing/that will
dance}
There will be some teenagers dancing here (in this group).
Sardinian
Jones (1993: 105)
B
at ballatu tres pitzinnas.
loc. cl. has danced three girls
Three girls danced.
b. Northwestern Catalan
En aquest esbart, hi
balla adolescents.
in this
group, loc.cl. dances teenagers
There are some teenagers dancing here (in this group).
Crucially, notice that the construction is ungrammatical when there is a definite DP:
(38) a.
Sardinian
Jones (1993: 195)
*B
at ballatu cussos pitzinnas.
loc. cl. has danced these girls
b. Northwestern Catalan
*En aquest esbart, hi
balla els adolescents.
in this
group, loc.cl. dances the teenagers
On the other hand, the Sardinian sentence in (37a) shows that the auxiliary selected by the impersonal existential constructions under study is ere (have),
this auxiliary also being selected by the existential verb ere. By contrast, the
existential or locative verb ssere (be) selects the auxiliary ssere. See the Sardinian examples in (39):18
(39) a.
B
at ppitu metas problemas.
loc.cl. has had many problems
There has been many problems.
b. Bi
sun/*at istatus issos.
loc.cl. are/*has been they
They were there.
As an idiosyncratic property of (40), an abstract central coincidence preposition (P) is incorporated into the verb expressing static or negative transition
(see Freeze 1992; Hale & Keyser 1993; Kayne 1993, among others). This P selects a locative determiner <Cat. hi / Sard. bi> as its specifier/subject, and an
NP as a complement.19 Following Rigau (1997), we assume that the incorporation of P into the existential verb allows this verb to assign partitive case, an
instance of the inherent case that P is able to assign when incorporated into the
host light verb.20
Let us now deal with the syntactic representation of the conflation process involved in constructions like those in (34) and (37). We want to argue
that there is an empty unaccusative verb expressing static or negative transition (i.e., be), which in turn selects a phonologically null central coincidence P.
This null P incorporates into the V in the syntax in order to satisfy the Full Interpretation Principle at PF. Since both V and P are phonologically null heads,
the derivation will crash unless the empty complex verbal head is merged with
the unergative verb ballar (to dance) present in the numeration: Functional
categories omitted, the syntactic structure corresponding to both (34b) and
(37b) is represented in (41). The PP en aquest esbart (in this group) has been
set aside, since it is an adjunct.
(41)
V
V
V
balla
V
be
P
N
hi
P
N
central adolescents
coincidence
Our proposal is that the phonologically null verb be does not exist in an otherwise non-verb-framed language like English. As seen in Section 3, its satelliteframed nature has been argued to be restricted to path constructions (pace
Talmy 1985, 1991). Recall that we have accounted for its non-verb-framed nature by positing two empty verbal heads: go and cause. However, as a result of
its lacking a null verbal head be, merging an unergative verb like dance into a
static or negative transition verb is not a real possibility in this language, as it
is in Romance languages like Spanish or Catalan, which are otherwise typically
verb-framed: they lack those two empty verbal heads go and cause involved in
complex path constructions.
Moreover, according to our present argumentation, it is plausible to assume that the absence of an empty verbal head be from English is related to
the fact that this language has no existential verb equivalent to the Romance
impersonal existential verb: e.g., cf. Cat. haver-hi.
To conclude, we have shown that the so-called unaccusativization process
involved in both (41) and (18) (Sue danced across the room) can be analyzed in
a uniform way. Quite interestingly, we have not made use of different mechanisms or strategies when dealing with each of these constructions. Rather the
conflation processes under study have been defined as an instance of the Merge
operation that combines two different verbs present in the numeration, the
phonologically null one being the main verb, while the full verb being the subordinate one: it is the former verb that will determine the {cause/motion/state}
meaning of the construction, the latter verb expressing what Talmy refers to as
the manner component.
. Concluding remarks
The most general conclusion to be drawn from our study is fully coherent
with Chomskys (1995: 8) claim that the apparent richness and diversity of
linguistic phenomena is illusory and epiphenomenal, the result of interaction
of fixed principles under slightly varying conditions. Adopting such a perspective, we have shown that Talmys (1985, 1991) descriptive analysis of the conflation processes involved in the constructions under study can be explained
within a minimalist conception. In particular, we have shown that the distinction between satellite-framed and verb-framed constructions correlates with
the (un)availability of the relevant empty heads (cause/go/be), whose licensing
involves appealing to Merge to avoid crashings at PF.
On the other hand, the data we have analyzed here do not appear to affect
functional aspects of the lexicon (Borer 1984; Chomsky 1995). Accordingly, we
must conclude that parametrized variation is not to be confined to inflectional
systems. This conclusion has been independently reached by Hale & Keyser
(1998), Juffs (1996), and Snyder (1995), among others.
Notes
* We are grateful to our colleagues of the Grup de Gramtica Terica of the Universitat
Autnoma de Barcelona, especially to Laia Amadas and Carme Picallo for their careful critical readings. Thanks to Gretel de Cuyper and Michael Kennedy for the Dutch and English
data, respectively. Laia Amadas also deserves a special mention for her support and willingness to help us with the preparation of the manuscript. We are also grateful to the audience at
the 1999 GLOW: Universals (ZAS, Berlin) for insightful comments and suggestions. Research
for this paper has been supported the Ministerio de Ciencia y Tecnologa through project
BFF2000-0403-C02-01/02, and by the Generalitat de Catalunya through project 2001 SGR
00150.
. We do not intend to reduce the importance of semantics by adopting a syntactic approach. Our syntactic account should not be regarded as incompatible with Jackendoff s
(1990) or Goldbergs (1995) works on the semantic restrictions concerning constructional
idioms. We have put them aside in the present paper, because what we are mostly concerned
with here is how these constructions can be dealt with from a syntactic perspective.
. Accordingly, we assume the epistemological priority of semantic selection over categorial selection (see Grimshaw 1979 and Chomsky 1995, among others).
. It is important to note that N, V, and P must be regarded as the syntactic categories derivable from their associated semantic features, not as their corresponding language-specific
morphosyntactic realizations (see Hale & Keyser 1997, 1998). For example, it should be
clear that we are not positing that happy or here are morphosyntactic Ps that turn out to
incorporate a N as a morphosyntactic category. Rather what we are positing is that both
involve the incorporation of a non-relational element into a relational one.
. See Hale (1986) for an in-depth analysis of these semantic relations.
. According to Chomsky (1995: 238), <the> lexical entry represents in the optimal way
the instructions for the phonological component and for the interpretation of the LF representation: a phonological matrix, and some array of semantic properties. It must also
contain whatever information is provided by the verb itself for the operations of CHL
(= computational system).
. Some relevant remarks are in order here: as noted by Juffs (1996), it should be clear
that the distinction between satellite-framed languages and verb-framed ones must not be
drawn across the board, but rather it depends on the lexical-semantic domains analyzed.
For example, English can be typically analyzed as satellite-framed with regard to physical
motion. This notwithstanding, concerning abstract motion, it is both satellite-framed (e.g.,
cf. the adjectival resultative construction in (22bc)) and verb-framed (cf. the huge number
of change of state verbs in English (cf. Levin 1993). That is, it appears to be more appropriate
to speak of satellite- and verb-framed constructions rather than satellite- vs. verb-framed
languages.
. This notwithstanding, it is important to keep in mind the following remarks found in
Talmy (1985: 72): English does have a certain number of verbs that genuinely incorporate
Path, as in the Spanish conflation type, for example: enter, exit, pass, rise, descend, return,
circle, cross, separate, join (...). But these verbs are not the most characteristic of English. In
fact, the majority (here all except rise) are not original English forms but rather borrowings
from Romance, where they are the native type.
. One caveat is in order here: although the PP a/en la habitacin (into the room) can be
omitted, it is not an adjunct: see Tortora (1998), where it is argued for the argumental status
of these dispensable elements. According to Tortora (1998: 344), PPs like those in (28) do
not occupy a VP-external position; rather they are part of the core eventuality of the VP, just
like English resultative adjectival phrases in the river froze solid or the window broke open.
. In (30b) the P lexically incorporated into V allows the phonologically null P to be
properly interpreted, since the former ensures the recoverability of the latter.
. We assume that a non-relational element (i.e, an abstract Ground) is lexically incorporated into the relational element involved in weg (away). Here we will not comment on how
this information is to be encoded into the lexical entry of weg: See Hale & Keyser (2000) for
the proposal that some intransitive particles incorporate a non-relational element. Here we
will assume that particles and prefixes do not essentially differ with respect to argument
structure: both involve the birelational element P(article) (see Section 2).
. Torrego (1989) and Rigau (1997) relate the construction in (34) to the so-called locative
inversion. Nevertheless, it must be noted that the latter construction appears to have different properties: for example, those discourse conditions governing locative inversion are
not the same as those governing (34). See Levin & Rappaport Hovav (1995) for arguments
against taking the locative inversion construction as an unaccusative diagnostic. In order to
distinguish those unaccusative constructions in (34) from the locative inversion constructions which can be unaccusative or unergative, we will refer to the former as existential
locative constructions (see Footnote 21).
. Following Longa, Lorenzo & Rigau (1998), we assume that Spanish has a phonologically null locative determiner represented as <HI>. See this article for motivation of this
assumption.
. As noted by Jones (1993: 113f), the clitic bi is obligatory in (39a), but optative in (39b).
In the latter sentence the clitic could be replaced by a locative PP or adverbial phrase, this
showing that the clitic bi is the true predicate, and not a subject clitic as in (39a). Accordingly,
the subject issos (they) in (39b) has nominative case.
. In impersonal deontic existential constructions like those in (i), the specifier selected by
P is a dative or locative clitic determiner, that is, a quirky case clitic (see Rigau 1999).
(i)
a.
b.
c.
d.
e.
Hi
cal
tres ous.
loc.cl. is-necessary three eggs
Three eggs are necessary.
Mos cal
un mili de francs.
to-us is-necessary a million of francs
We need a million of francs.
Bi
keret
tres ovos.
loc.cl is-necessary three eggs
Three eggs are necessary.
Nos keret
unu milione de francos.
to-us is-necessary a
million of francs
We need a million of francs.
I
cau
tres cagires.
loc.cl. is-necessary three chairs
Three chairs are necessary.
Northwestern Catalan
Aranese Occitan
f.
Mos cau
tres cagires.
to-us is-necessary three chairs
We need three chairs.
. Transitive verbs are associated with accusative case, not partitive case. Consequently, the
clitic en in (i-b) is the genitive case that an overt or covert quantifier assigns to the N (see
Rigau 1997):
(i)
a.
b.
. This example is commented on by Kurat (this volume). See this work for an alternative
explanation of the ungrammaticality of (42) (his (36a)).
Once again one caveat is in order here: as noted above (cf. fn. 16), the possibility for an
unergative verb like dance to enter into the so-called locative inversion is not what is at issue
here (cf. Levin & Rappaport Hovav 1995: 285). That is, here we are not referring to locative
inversion cases like (36a), which is not an unaccusative construction, but to impersonal
existential constructions like those in (34), which have been shown to be unaccusative (see
Footnote 16).
References
Arad, M. (This volume). Universal Features and Language-particular Morphemes.
Baker, M. (1997). Thematic Roles and Syntactic Structure. In L. Haegeman (Ed.), Elements
of Grammar. Dordrecht: Kluwer.
Borer, H. (1984). Parametric Syntax. Dordrecht: Reidel.
Borer, H. (1994). The Projection of Arguments. In E. Benedicto and J. Runner (Eds.),
Functional Projections [UMOP 17]. Amherst: University of Massachusetts.
Bouchard, D. (1995). The Semantics of Syntax. Chicago and London: The University of
Chicago Press.
Chomsky, N. (1975). Reflections on language. New York: Pantheon.
Chomsky, N. (1995). The Minimalist Program. Cambridge, MA: The MIT Press.
Chomsky, N. (1998). Minimalist Inquiries: The framework. Ms., MIT. (Published as
Chomsky, N. (2000). In R. Martin et al. (Eds.), Step by Step: Essays on Minimalist Syntax
in Honor of Howard Lasnik. Cambridge, MA: The MIT Press).
Dowty, D. (1991). Thematic Proto-Roles and Argument Selection. Language, 67, 547619.
Freeze, R. (1992). Existential and Other Locatives. Language, 68, 553595.
Goldberg, A. (1995). Constructions: A constructional approach to argument structure.
Chicago: The University of Chicago Press.
Grimshaw, J. (1979). Complement Selection and the Lexicon. Linguistic Inquiry, 10, 279
325.
Hale, K. (1986). Notes on World View and Semantic Categories: Some Warlpiri examples.
In P. Muysken and H. van Riemsdijk (Eds.), Features and Projections. Dordrecht: Foris.
Hale, K. & S. J. Keyser (1993). On Argument Structure and the Lexical Expression of
Syntactic Relations. In K. Hale and S. J. Keyser (Eds.), The View from Building 20: Essays
in honor of Sylvain Bromberger. Cambridge, MA: The MIT Press.
Hale, K. & S. J. Keyser (1997). The Limits of Argument Structure. In A. Mendikoetxea and M.
Uribe-Etxebarra (Eds.), Theoretical Issues at the Morphology-Syntax Interface. Bizcaia:
Servicio Editorial de la UPV.
Hale, K. & S. J. Keyser (1998). The Basic Elements of Argument Structure. In H. Harley
(Ed.), Papers from the UPenn/MIT Roundtable on Argument Structure [MIT Working
Papers in Linguistics 32] (pp. 73118).
Hale, K. & S. J. Keyser (2000). On the Time of Merge. Ms., MIT.
Harley, H. (1995). Subjects, Events, and Licensing. Doctoral dissertation, MIT.
Hoekstra, T. (1984). Transitivity. Dordrecht: Foris.
Hoekstra, T. (1992). Aspect and Theta-theory. In I.M. Roca (Ed.), Thematic Structure: Its
role in grammar. Berlin: Mouton de Gruyter.
van Hout, A. (1996). Event Semantics of Verb Frame Alternations: A case study of Dutch
and its acquisition. Doctoral dissertation, TILDIL dissertation series.
Jackendoff, R. (1990). Semantic Structures. Cambridge, MA: The MIT Press.
Jackendoff, R. (1997). Twistin the night away. Language, 73, 534559.
Jones, M. A. (1993). Sardinian Syntax. London: Routledge.
Juffs, A. (1995). Learnability and the Lexicon. Amsterdam: John Benjamins.
Kayne, R. (1993). Toward a Modular Theory of Auxiliary Selection. Studia Linguistica, 47,
331.
Klipple, E. (1997). Prepositions and Variation. In A.-M. Di Sciullo (Ed.), Projections and
Interface Conditions. New York: OUP.
Kratzer, A. (1996). Severing the External Argument from its Verb. In J. Rooryck and
L. Zaring (Eds.), Phrase Structure and the Lexicon. Dordrecht: Kluwer.
Kural, M. (This volume). A Four-way Classification of Monadic Verbs.
Levin, B. (1993). English Verb Classes and Alternations: A preliminary investigation. Chicago:
The University of Chicago Press.
Levin, B. & T. Rapoport (1988). Lexical Subordination. Chicago Linguistics Society, 24, 275
289.
Levin, B. & M. Rappaport Hovav (1995). Unaccusativity: At the syntax-lexical semantics
interface. Cambridge, MA: The MIT Press.
Longa, V., G. Lorenzo & G. Rigau (1998). Subject Clitics and Clitic Recycling: Locative
sentences in some Iberian Romance languages. Journal of Linguistics, 34, 125164.
Marantz, A. (1997). No Escape from Syntax: Dont try morphological analysis in the privacy
of your own lexicon. In A. Dimitriadis et al. (Eds.), Penn Working Papers in Linguistics
4 (2) (pp. 201225).
Mateu, J. (1999). Universals of Semantic Construal for Lexical Syntactic Relations. Paper
presented at the 1999 GLOW Workshop: Sources of universals, Potsdam. Distributed as
GGT-99-4 Research Report, Universitat Autnoma de Barcelona (http://ggt.uab.es).
Rappaport Hovav, M. & B. Levin (1998). Building Verb Meanings. In M. Butt and W. Geuder
(Eds.), The Projection of Arguments: Lexical and compositional factors. Stanford, CA:
CSLI Publications.
Morphological constraints on
syntactic derivations*
Juan Romero
Universidad de Alcal de Henares/Universidad Autnoma de Madrid
The aim of this paper is to sketch a model for linguistic variation within a minimalist framework. The changes the minimalist program (MP) has introduced
in the syntactic explanation make, from my point of view, untenable the Principles and Parameters (PP) postulates. The proposal I will argue for is based on
the idea that languages differ in the formal features they encode, i.e., there is no
a universal catalogue of shared formal features. Each language determines independently (although not arbitrarily) its own formal features from the universal
set of features F available for the faculty of language (Chomsky 1998, 1999).
The basic assumption in the PP model is that purely grammatical properties are encoded as principles. Principle P expresses a requirement that must
be satisfied by every derivation in every language. P can be satisfied in different ways within the range specified by the parametric options of P. Therefore, cross-linguistic variation is encoded as parameters over principles. Consider Case theory. Every NP is roughly subject to the following condition: it
must receive/check Case. We can express this requirement as a principle, and
then we can define parametric options such as right or left assignment, or
overt/covert checking. These options delimit the range of variation UG allows
for Case Theory.
In the government and binding model, the modular architecture of the
computational component provides a well-defined locus for the formulation of
each principle. Some of them, like government, affect the whole system; others,
like Case Filter, are module-specific. The properties of each module are satisfied in a certain level of representation (D Structure, S Structure, etc.). As a
Juan Romero
Strong features imply the existence of two different cycles: the strong/overt
cycle, and the weak/covert cycle. In a derivational approach this is clearly an
unexpected imperfection that would require good evidence to be sustained
(see Kayne 1998; Chomsky 1998, 1999). At this point, brand new tools are required to account for linguistic variation. Chomsky (1999) proposes that the
P&P approach can be reduced to a kind of uniformity approach:
(1) Uniformity Principle
In the absence of compelling evidence to the contrary, assume languages
to be uniform, with variety restricted to easily detectable properties of
utterances
The first question is what can be considered uniform. There at least two kind
of candidates. First, those properties that constitute an inherent part of the
computational component. And second, those properties that due to its interpretability properties are motivated at the interfaces. Some candidates are the
following:
Thematic relations. Following Hale and Keysers hypothesis, thematic relations are configurational in nature. In the MP, this can be understood
as Merge by s-selection (Romero 1997). If a verb V selects a theme, in the
computational component V takes a complement that will be thematically
interpreted. Note that although these relations are usually considered to be
evaluated at LF, it is also possible to understand them other way. Specifically, this can be a subcase of a more general procedure: the I/C interpretation of syntactic structures (Spechead, and headcomplement relations). Thematic interpretation would be the particular instance in which
syntactic structures involve relations between lexical heads.
Basic sentence structure. The sequence [COMP TENSE v V] seems
to be universal, although not all the functional heads are required in every sentence (see Chomsky 1999). (See Uriagereka 1998 for a possible I/C
motivation for this sequence.)
Computational operations. The expressions in every language are composed by means of the same operations: Merge, Move, Agree, etc. These
are not substantive universals, but purely computational devices, possibly
the core of the faculty of language.
What is crucial about all of these candidates (that do not exhaust the list), is
that no variation at all is allowed for them. There is no language in which the
theme relation is expressed by means of a Spechead relation; or in which
variation is allowed for the inner works of Merge, Agree, etc.
Juan Romero
The second question about (1) is what means easily detectable properties of
utterances. Chomsky (1999) following ideas developed by Vergnaud and Borer,
suggests two properties that could satisfy this condition: . . . parametric variation is restricted to the lexicon, and insofar as syntactic computation is concerned, to a narrow category of morphological properties, primarily inflectional
(p. 2). Interestingly, in the very same paragraph Chomsky states that . . . basic
inflectional properties are universal though phonetically manifested in various
ways. . .. From this point of view, syntactic variation is basically expressed by
means of the phonetic properties of the formal features.3 However, under this
view, it is not clear what means that syntactic variation is just a diverse phonetic
manifestation of universal properties. Consider the analysis of the Thematization/Extraposition English rule in Chomsky (1999). The application of this
rule has some syntactic effects, but crucially no feature checking relation or
other syntactic operations seem to be involved in it. We can build a parallel
case for agreement. Consider the sentence in (2):
(2) a. *whom did you give t the car
b. who gave whom the car
Romero (1999) argues that agreement in English (but not, for instance, in
Spanish) is morphologically configurational, therefore, in the morphological
component, agreement relations must be expressed by means of Spechead relations. Furthermore, Romero (1999) argues that dative shift involves an agreement relation. As a consequence, if at Spell Out the dative has moved away from
the position where this relation is established (2a), the derivation will crash. On
the other hand, if the dative remains in situ (2b), the agreement relation is morphologically evaluated and the derivation succeeds. Therefore, it seems that this
line of reasoning has some initial plausibility. The question is if it is enough
to account for linguistic variation. Chomskys hypothesis is based on the idea
that every language has the very same set of interpretable/uninterpretable features (its basic inflectional properties). Therefore, every language has to express, for instance, the same agreement or Case relations. Throughout this paper I will argue that this approach is theoretically questionable and empirically
insufficient, and that, as a consequence, the answer to this question, whether
the phonetic properties of the formal features are enough to capture syntactic
variation, is no.
Syntactic derivations act primarily on formal features. Its role on the faculty of language is defined precisely on their ability to enter into syntactic computations. At the other side of the interfaces only those features that can receive
either a phonetic or a semantic interpretation can survive. Where do formal
features come from? Suppose a universal set of features F available for the faculty of language.4 Due to the bare output conditions we know that this set
contains phonetic and semantic features: those are the features used by the A/P
and I/C components respectively. Formal features are split into two groups:
interpretable and uninterpretable. Interpretable means interpretable at the interfaces (Chomsky 1995), so they are either phonetic or semantic in nature. As
a consequence, I will assume that interpretable features are phonetic or semantic features that have been formalized, and when they reach the interfaces they
recover their original phonetic or semantic status, i.e., they are interpretable.
Regarding uninterpretable features two options come to mind. On the one
hand, they can belong to the computational component. This approach seems
to be the one assumed in Chomsky (1999):
. . . The relation Agree and uninterpretable features are prime facie imperfections.
In MI and earlier work it is suggested that both may be part of an optimal solution
to minimal design specifications by virtue of their role in establishing the property
of displacement, which has (at least plausible) external motivation in terms of
distinct kinds of semantic interpretation and perhaps processing. If so, displacement is only an apparent imperfection of natural language, as are the devices that
implement it (p. 3).
Under this view, uninterpretable features form part of the mechanics of the
computational component, in the same way that Merge or Move do. On the
other hand, it can be considered that uninterpretable features are interpretable
features in the wrong place (Romero 1999). From this perspective, they are also
formalized features, but due to the fact that they are attached to a category that
cannot interpret them, they do not receive an interpretation at the interfaces.
There is a clear difference between these two approaches: under the first one (to
which I will refer as Visible Feature Uninterpretability Hypothesis (VFUH)),
being uninterpretable is an active property in the computational component;
under the second one (the Formal Features Hypothesis (FFH)), it is not, if
an uninterpretable feature remains unchecked at the interfaces, the derivation crashes. The interfaces are the only places where being (un)interpretable
becomes relevant.
As said, Chomsky argues that uninterpretable inflectional features are the
devices that implement displacement. There are three different kinds of uninterpretable features:
Juan Romero
Juan Romero
In this dialect, the clitic 3rd person animate masculine objects has the form
le, and the clitic for the 3rd person non animate masculine objects has the
form lo. The contrast between (4a) and (4b) shows that the PCC affects
also 3rd person arguments.
iii. Its nature is not morphological, but syntactic. Consider the following example from Haitian Creole (M. Degraff, p.c.):
(5) a. *mwen pral bay Jan/li Mary/l
I will give Jan/him Mary/her
b. mwen pral bay Jan/li yon menai
I will give Jan/him a girlfriend
In Haitian Creole, as in Spanish, the PCC ranges from 1st and 2nd person
to proper names. Interestingly, Haitian Creole lacks either clitics or object
agreement.
iv. It shows up in Double Object Constructions (DOC). Following an analysis
pursued by different authors in the last years (Uriagereka 1988; Demonte
1995; Romero 1997; Anagnostopoulou 1998; etc.), I will assume that all the
b-pairs in (6)(10) are instances of the same construction, the DOC, which
includes the constructions traditionally labeled with this name in English
(6), and the applicative constructions in Bantu languages (7), as well as
dative-clitic or dative-agreement constructions in Romance (8), Basque (9)
or Southern Tiwa (10):
(6) a. I gave a book to Mary
b. I gave Mary a book
(7) a.
(Baker 1988)
(10) a.
bi-musa-wia-ban uide-ay
agr-cat-give-past child-to
I gave the cats to the child
b. uide tam-musa-wia-ban
child agr-cat-give-past
I gave the cat to the child
(Rosen 1990)
Although for the purposes of this paper I will simply assume that all the sentences in (6)(10) involve the same structure, note that one of the main reasons
to think that the members of the pairs in (6)(10) represent the same construction is the fact that all of these languages are subject to the PCC precisely in
this alternation. Furthermore, all these pairs also share identical c-command
asymmetries (Barss & Lasnik 1986), and the so-called possession or animacy
restriction (Green 1974) exemplified in (11).
(11) a. I sent a letter to Paris
b. *I sent Paris a letter
Although the existence of this restriction has been attested in more than 300
languages (see Albizu 1997), there are reasons to think that it is not universal.
Specifically, it seems that this constraint does not show up at least in Japanese,
Turkish, and Basque infinitive DOCs. Due to space reasons in Section 2.1 I
will only consider Basque infinitives (but see Romero 1999 for a more detailed
description including both Japanese and Turkish DOCs). In Section 2.2 I will
propose that the lack of formal features in the languages that are no subject to
the PCC can also be related to the way they drop arguments.
. PCC in Basque infinitives
Consider data in (12) from Basque (Laka 1993).
(12) a. *zuk ni etsaiari saldu na-i-o-zu
you.erg me.nom enemy.dat sell 1sO-aux-3sD-2sS
You sold me to the enemy
b. gaizki iruditzen zait [zuk ni etsaiari saltzea]
wrong seem aux [you.erg I.nom enemy.dat selling]
Your selling me to the enemy seems wrong to me
Juan Romero
eskutitza Parise-ra
bidali du-te
letter
Paris-allative send aux.3sO-3pS
They sent the letter to Paris
b. *eskutitza Paris-i
bidali diote
letter
Paris-dative send aux.3sO-3sD-3pS
As a consequence, I will assume that (13a) is the Basque counterpart of the English DOC,6 and (13b) to the to-construction. This analysis is appealing since
Basque uses the allative construction exemplified in (13b) precisely in the same
contexts where the to-construction is mandatory in ditransitive constructions
in English (11a)/(14a). As shown in (12b), Basque infinitives do not show PCC
effects. Note that the indirect object in (12b) is dative marked, and therefore
patterns with (13a), that is, (12b) is an infinitive DOC. However, contrary
to what happens in (12a), although a violation of the PCC is expected, the
resulting sentence is grammatical. Finally, Basque infinitives clearly show that
what is important for the computational component are the specifications of
the attracting head ([+-features] on tensed sentences, [-features] on infinitive sentences), and not those of the shifted NP, which are the same in both
structures.
. Agreement restrictions
An obvious way to account for the asymmetry between (12a) and (12b) is to relate it to the fact that non tensed clauses lack agreement morphology (see Laka
1993). Specifically I propose that Basque infinitives lack the formal features that
encode agreement, and, as a consequence, no agreement relation is established
between the verb and its internal arguments. If this proposal is on the right
track, the presence of features on a certain language (for a certain relation)
can be tested by the effects of agreement restrictions such as the PCC. The contrasts previously exemplified denote a small amount of linguistic variation
that can be made dependent on this property. However, Romero (1999) argues
that actually two former parameters can be reduced to the presence/absence of
agreement features: the null topic parameter (Huang 1984), and the pro-drop
parameter. Consider the sentences in (15)(17), where different possibilities of
dropping in Spanish, English, and Japanese are exemplified.
(15) a.
Mara compr
el billete
Mara bought.3sS the ticket
b. Mara lo compr
Mara it bought.3sS
Mara bought it
c. compr
el billete
bought.3sS the ticket
He/she bought the ticket
Each language displays a different pattern. In Spanish (15), NP-dropping requires an agreement marker or a clitic. In English (16) it requires pronouns.
However, in Japanese (17) the arguments simply disappear in the phonetic output (actually, there are almost no 3rd person pronouns in this language). Since
the seminal paper by Taraldsen (1978), the pro-drop parameter has been related to the presence of strong agreement, arguing that only strong agreement
licenses an empty category. The intuition is that an argument can be dropped
whenever its information is somehow recoverable from the agreement morphology (recoverability principle). However, languages as Japanese or Turk-
Juan Romero
ish, or Basque infinitives, can freely drop arguments without resorting to any
morphological agreement (Huang 1984).
The hypothesis I propose is that thematic interpretation license NPdropping. I assume that thematic relations are subject to some general principle of the kind argued for in Bakers UTAH or Hale & Keysers recent hypothesis on argument structure. Under this view, the presence of an argument is
fully predictable, and NPs can be freely drop whenever the pragmatic conditions that govern dropping are met (see, for instance, Li & Thompson 1979). If
this is correct, the question is what syntactic conditions, if any, apply to dropping. These restrictions, the licensing of empty categories, were a major topic
in the Government and Binding framework. Since appealing to the phonetic
content of a category (leaving aside traces) does not seem to be allowed within
the MP, it is not clear how all of these proposals can be expressed in minimalist
terms. As a point of departure I will adopt a very naive approach to this problem: let us say that in the numeration it is only inserted the material required
for convergence. From this perspective, Japanese or Turkish constitute the unmarked case: whenever the appropriate conditions are met, an empty category
with the minimum possible amount of information (maybe a variable, as suggested by Huang) is inserted in the numeration. However, if we try to do the
same in Spanish or English, the result is ungrammatical. This can be explained
by the fact that these languages must satisfy an agreement relation. Therefore,
some extra material must be inserted (a pro, a clitic, or a pronoun). Specifically,
the formal features required to delete the uninterpretable features encoded in
the verb (v, T). In this sense, Spanish and English are subject essentially to the
same constraint.
This idea basically restates Huang (1984) proposal for empty categories in
object position. He argues that there are two different parameters: null topic
parameter and pro-drop parameter. Null topic languages introduce an null operator in topic position that can license a variable. Chinese, Japanese or Korean belong to this group. In these languages strong agreement is not required
for dropping an argument, since they have an alternative way for licensing an
empty category: an operator-variable chain.7 The pro-drop parameter splits
non null topic languages into two groups: those than can license a pro by means
of strong agreement (Spanish, Basque), and those that do not have an agreement strong enough to license a pro (English, French). In the present proposal, null topic languages are those that do not have formal agreement features, and non null topic languages are those that must satisfy agreement relations. If the numeration is subject to an economy of insertion principle, when
possible, the first option is always going to be preferred. Actually, Campos
(1986) argues that in Spanish, under certain conditions, the null topic strategy is available. Finally, this system straightforwardly accounts for the question of the parameter hierarchy posed by Huang (1984). Under his view, no
[+null topic]/[+pro-drop] language is allowed, [pro-drop] languages constitute a sub-set of [null topic] languages. This implies the existence of a hierarchy of parameters, and, as a consequence, a hierarchy of principles. No such a
problem arises in the model sketched.
In brief, the proposed system relies on the hypothesis that thematic projection is uniform and that it is thematic interpretation what allows to recover
the information missing in the phonetic output. Therefore, if no agreement relation needs to be satisfied, argument dropping is free. However, if a language
has encoded formal agreement features, argument dropping will be restricted
by the necessity of satisfying agreement relations. Furthermore, different languages express agreement relations in different ways that can impose additional
restrictions on syntactic derivations, such as those found between English and
Spanish.
Summarizing, linguistic variation can be explained in terms of the formal
features encoded in each language. The presence of these features is available
for the learner in the primary linguistic data. Furthermore, the relations established by these features may receive different morphosyntactic representations
that can also affect syntactic computation.
The landing site for object movement is (Spec, V) (see also Romero 1997):
(18) [vP SUBJECT [ v [VP OBJECT [ V t OBJECT ]]]]
Juan Romero
ii. The argument in (Spec, V) measures out the event in the sense of Tenny
(1987), this explains why the argument in (Spec, V) must be specific
(within a totally different framework, Koizumi 1993 makes a similar proposal);
iii. If more than one argument appears in (Spec, V), i.e. if there are multiple
specifiers, the derivation crashes for I/C reasons: the aspectual interpretation of the sentence can only interpret one argument. Somehow this can
be considered the aspectual counterpart of the thematic criterion.
Assuming these analysis, the nature of the EPP feature can be characterized
according to the following properties: (i) it is a selectional interpretable feature; in the case at stake it is not lexical, but functional, and the projected
structure enters into an aspectual interpretation; and (ii) its presence is independent of agreement features. This property is clearly shown in infinitives
(Chomsky 1995). In the case of object movement, the mere existence of DOCs
in Japanese or Basque infinitives also supports this conclusion. However, there
must be something more to be said, since from these properties, no agreement restrictions are expected. Recall that the PCC shows up if and only if
the following conditions are met: (i) the verb encodes object agreement; (ii)
object agreement has the value [+person] (animate/specific); and (iii) there is
dative shift.
This description leads us to the following question: why do not nonanimate object agreement trigger PCC effects? To answer this question first
I will analyze object movement in transitive structures. It will be shown that, as
it is commonly assumed, an object does not require to be animate to undergo
object movement. However, if it is animate, object movement becomes obligatory (when possible). As a consequence, I will argue that in the same way as
languages can determine whether verb agreement encodes person plus number, but not, for instance, gender; languages can also associate person and EPP.
Note that this is not I/C (or A/P) motivated, it is just the way formal features
are organized.
. Scandinavian object shift
Descriptively speaking, object shift (OS) in Icelandic and other Scandinavian
languages is testable with respect to the position of adverbs. In (19a), the object appears to the left of the adverb. However, as shown in (19b), this movement is not always available. Specifically, it has been argued that OS is depen-
dent on verb movement. This constitutes the core of the so-called Holmbergs
Generalization (HG) (examples from Holmberg 1999).
(19) a.
Jag
I
b. *Jag
I
This operation has several interesting properties. First, if the object is pronominal, OS is mandatory, at least in some languages (Icelandic, Danish, and some
varieties of Norwegian). Therefore, in these languages, the in-situ counterpart
of (19a) is ungrammatical (2). This property is clearly reminiscent of the PCC:
whenever the object is pronominal, it has to undergo movement.
(20) *Jag kysste inte henne
I kissed not her
Second, if the NP is ambiguous between a specific and a non specific interpretation, the specific interpretation is obtained in the raised position (21a),
and the non specific is obtained in situ (21b) (examples from Bobaljik &
Thrinsson 1998). This fact is related to the event measurement properties:
non-specific arguments cannot bound the event (see, for instance, Jackendoff
1995). Examples in (21) also show that object movement is not restricted to
non animate arguments.
(21) a.
An intriguing property of this operation is found in verb topicalization environments (Holmberg 1999). In this construction, the verb is displaced to initial
position, and it is contrastively interpreted. Interestingly, although there is an
auxiliar, OS is allowed in this case:
(22) Kysst har jag henne not (bara hAllit henne i handen)
kissed have I her not (only held her by the hand)
In the early nineties, HG was used in conjunction with the notion of equidistance: the object could only raise to (Spec, AgrO) if the verb moved first to
AgrO, making the subject and the object equidistant to OS position. However,
Juan Romero
according to recent analysis, the object moves to (Spec, v) for agreement and
Case checking. Therefore, if negation (or VP adverbs) are generated in a position above VP (or adjoined to VP), as it is commonly assumed, this movement
is not enough to account for the properties of OS.
(23) [vP Adv [vP Object [ Subject [ v [. . . t Object... ]]]]]
Therefore, OS implies an additional movement to some other position. Holmberg (1999) treats OS as a phonological operation triggered by a [Focus] feature, and sensible to the phonetic context. Holmberg notes that not just an
unmoved verb, but any phonologically visible category inside VP preceding the
object position blocks OS. The range of phonologically visible categories in
Swedish include prepositions (24a), indirect objects (24b), and particles (24c).
(24) a. *jag talade henne inte med t-henne
I spoke her
not with
b. *jag gav den inte Elsa t-den
I gave it not Elsa
c. *dom kastade mej inte tu t-mej
they threw me not out
Chomsky argues that object shift languages such as Icelandic have a rule DISL,
triggered by interpretive reasons, that moves the object to its final position.
Since I assume that object shift moves the object to (Spec, V), here I am not
going to discuss the specific details of these proposals. Contrary to Chomsky,
I propose that it is the first step the one that is responsible for the interpretive
properties of the construction. Recall that assuming the properties assigned to
the PCC, to obtain a DOC, it is necessary to move the indirect object to (Spec,
V). It is at this point where the PCC arises, and therefore it is this position the
one that has the interpretive properties mentioned: the argument in (Spec, V)
(i) measures out the event and (ii) is specific. Consider the sentences in (26):
(26) a.
Elsa den
Elsa it
bkasafninu aldrei bkunum
the.library never the.books
Swedish
Icelandic
Icelandic
It is known that in this kind of sentences, the dative is in subject position (see
Boeckx 1998 and references therein). Therefore, although the experiencer triggers agreement with the verb, it is not either raised to subject position, or to
object position. However, since it has the relevant properties, it can be affected
by the same rule that dislocates objects to the left of the negation. If this line
of reasoning is correct, OS can be interpreted as a kind of clitic movement,
possibly triggered by prosodic reasons.
Juan Romero
. EPP
In the previous section I have argued that an EPP feature can project a specifier
in (Spec, V) in transitive clauses and attract a non animate object. Interestingly,
its presence seems to be obligatory whenever the object is animate. In other
analysis, this property may be made dependent on the phonological features
of the argument. However, under this two step analysis, the first movement
is purely syntactic. This fact correlates with many properties found in different
languages. Consider Spanish. In Spanish, clitic doubling in general is barred for
objects (29a). However, when the object is a pronominal, it is not only possible,
but obligatory (29b).
(29) a.
(*lo) vi el coche
cl.3sO saw the car
I saw the car
b. *(lo)
vi a l
cl.3sO saw him
I saw him
If both sets are [+person], two EPP features project specifiers. Since the
aspectual interpretation only tolerates one, the derivation crashes.
ii. If only one set is [+person], the indirect object raises to (Spec, V) and the
derivation converges. The other argument agrees in situ with the verb.
In languages with no object agreement, the aspectual properties of the verb
encode an EPP feature that forces object movement to (Spec, V). Two important conclusions arise from this characterization: first, EPP is always an interpretable feature. As a consequence, its interpretive nature may impose conditions at the interface. Note that this analysis does not imply that multiple
specifiers are always barred. Multiple specifiers are only barred whenever a configurational interpretation is assigned: thematic relations, aspectual relations,
or temporal relations. Possibly this argument does not apply when multiple
specifiers serve as landing site for scope reasons.
The second conclusion is related to the interpretable/uninterpretable properties of formal features. If the EPP is interpretable, as proposed, interpretable
features can be involved in movement operations. Therefore, it is its formal nature, and not its interpretability status, what makes formal features to be active
in the computational component.
I think this is a good result for at least the following reasons:
i.
No looking-forward property needs to be invoked to trigger computational operations. If being (un)interpretable is an interface property, then
it cannot be relevant troughout the derivation.
ii. It is not necessary to stipulate that every language shares the same set of formal features. Linguistic variation can be this way linked to the formal features each different language encodes and clearly express. Formal features
are the best candidates to be easily detectable properties of utterances.
iii. The computational component can be simply understood as a formal device that manipulates formal objects to create new objects for the interfaces. It is not required to encode substantive properties such as formal
features from the scractch.
Notes
* This material was presented at the 22nd GLOW Colloquium at Berlin, and to the Linguistic Seminar at Instituto Universitario Ortega y Gasset (Madrid). I am very grateful to these
audiences for helpful comments and discussion. I am also grateful to Violeta Demonte, Olga
Fernndez Soriano, Javier Ormazabal, and Miriam Uribe Etxebarria. This research has been
partly supported by a Comunidad de Madrid Postdoctoral Fellowship 144/2001.
Juan Romero
. If this were the case, the derivation should be able to look forward, a very problematic
move (Johnson and Lappin 1997; Collins 1997; Chomsky 1998).
. The introduction of phases has slightly changed this picture. Chomsky (1999) argues that
there is an evaluation process at strong phases, at the point where representations are handed
over to the interfaces (see Section 3).
. Romero (1999), in a somewhat different framework, proposes an analysis of the parameter of pro-drop based on the phonetic properties of agreement features in different
languages.
. In the case of phonetic features, this set has been pretty well studied, and there are reasons
to believe that something similar can be postulated for semantic features (see Uriagereka
1997).
. I am assuming with Hale and Keyser (1997) that by definition nominal items do not
trigger Merge (see also Romero 1997). However, under a phase derivational system, uninterpretable feature deletion would be evaluated at the next phase (Chomsky 1999), therefore, an uninterpretable feature may remain unchecked until the end of the phase, where its
presence would lead to a crashed derivation.
. Itziar Laka (p.c.) has reported to me that Montoya (1988) reaches the same conclusion.
Montoya shows that the dative c-commands the absolutive, but the absolutive c-commands
the allative, etc. Unfortunately I have not had access to this paper.
. Actually, parameter setting is necessary to exclude [pro-drop] in Chinese, and Huang
argues that it lacks agreement. I will come back to this in Section 4.
References
Albizu, P. (1997). The Syntax of Person Agreement. Doctoral dissertation, USC.
Anagnostopoulou, E. (1998). Dative Argument and Clitic Doubling. Ms., MIT.
Anagnostopoulou, E. (to appear). The syntax of ditransitives: Evidence from clitics. Berlin:
Mouton de Gruyter.
Barss, A. & H. Lasnik (1986). A Note on Anaphora and Double Objects. Linguistic Inquiry,
17, 347354.
Bobaljik, J. D. & H. Thrinsson (1998). Two Heads arent Always Better Than One. Syntax,
1 (1), 3771.
Boeckx, C. (1998). Agreement Constraints in Icelandic and Elsewhere. Working Papers in
Scandinavian Syntax, 62 (2), 135.
Bonet, E. (1991). Morphology after Syntax: Pronominal clitics in romance. Doctoral
dissertation, MIT.
Bonet, E. (1994). The Person-Case Constraint: A morphological approach. MIT Working
Papers in Linguistics, 22, 3352.
Campos, H. (1986). Indefinite Object Drop. Linguistic Inquiry, 12 (2), 354359.
Chomsky, N. (1995). The Minimalist Program. Cambridge, MA: The MIT Press.
Chomsky, N. (1998). Minimalist Inquiries: The framework. MIT Occasional Papers in
Linguistics, 15.
Juan Romero
Introduction*
Joachim Sabel
The anaphor in (3a) violates Principle A of the Binding Theory since the potential antecedent is not located in the embedded clause which constitutes its relevant Binding Domain. If, as in (3b), the NP containing the anaphor is moved
to Spec CP i.e., into a position closer to its antecedent, Principle A is fulfilled and the sentence becomes grammatical. Turning now to (3c) we observe
that the anaphor in the matrix Spec CP position should also violate Principle
A because it is not c-commanded by its antecedent. Given that this sentence
is nevertheless grammatical it can be concluded that binding of an anaphor is
possible if there is a movement site to which the anaphor may be reconstructed
and in which it may be bound in accordance with Principle A. The relevant
reconstruction site in (3c) is the position of the intermediate trace t i which is
located in the intermediate Spec CP position. In this position the anaphor may
be bound in accordance with Principle A, as can be seen in (3b).
Joachim Sabel
As with (3a), the anaphoric expression in (4a) may not be bound in its base
position. Importantly, in contrast to (3bc) it also cannot be bound in the intermediate Spec CP position, which is occupied by where. The only account
for the slightly deviant character of (4b) i.e., for the possibility for John to act
as a binder for himself, seems to be that there is a trace (t i ) in a VP-adjoined
position of the matrix clause from which binding can be achieved. (4) provides
evidence for intermediate adjunction of wh-phrases to VP.
A similar argument for intermediate adjunction can be constructed in connection with scrambling. In analyses that treat scrambling as a movement phenomenon, scrambling is traditionally analyzed as Chomsky-adjunction to a
maximal projection (Saito 1985, 1992). As can be seen from (5a), the embedded subject is the only possible antecedent for otagai each other.3 In (5b) the
embedded object containing the reflexive is scrambled in front of the embedded subject. In this position the reflexive otagai may be bound by the matrix
subject. Now consider (5c), where the NP containing the anaphor is scrambled out of the embedded clause in front of the matrix subject. In this case
the matrix subject can also be co-referent with the anaphor. (5c) provides evidence that the scrambled element has moved through an IP-adjunction site
in the embedded clause, which was its ultimate landing site in (5b) (Nemoto
1993: 93):4
(5) a.
Hence adopting the intermediate adjunction hypothesis provides a unified account for reconstruction phenomena associated with binding properties in
connection with wh-movement and scrambling.
. Weak crossover
The absence of weak crossover effects in scrambling languages has been taken
as evidence for intermediate adjunction. Several authors have argued that XPmovement proceeds via intermediate adjunction to IP in scrambling languages
like German (Grewendorf 1988; Bayer 1993; Richards 1997; among others).
This assumption was used to explain that (for many speakers) weak crossover
effects are absent in contexts where they appear in non-scrambling languages
like English. The contrast is illustrated in (6) vs. (7):
(6) [CP Weni hat [IP t i [IP seinei Mutter [VP immer t i gekt]]]]?
whoacc has
his
mother
always
kissed
Who did his mother always kiss?
(7) *[CP Whoi does [IP hisi mother often kiss t i ]]?
The intermediate trace t i in (6) is argued to be located in an IP-adjoined position with A-properties, which prevents both the pronoun and the variable t i
from being locally A -bound by the operator in Spec CP. Assuming intermediate adjunction therefore provides an answer to the question of why the weak
constraint crossover (see Koopman & Sportiche 1982) is violated in (7) but not
in (6).
. Locality effects
The preceding sections have shown that weak crossover and certain binding
phenomena are easily accounted for if the concept of intermediate adjunction is adopted. A further argument for the intermediate adjunction hypothesis comes from locality effects. Since Chomsky (1986a) it has been widely accepted that locality effects in connection with movement out of islands receive
a straightforward explanation if intermediate adjunction is adopted. Before I
discuss the relevant examples I will shortly review some technical notions.
Recall that one theoretical motivation for assuming intermediate adjunction of wh-phrases in the Barriers framework has to do with the aim of reduc-
Joachim Sabel
ing the ECP to antecedent government.5 Consider for example the case of =
VP in (8).
(8) a. What do you [ like t]?
b. What do you [ t [ like t]]?
both examples. Following Chomsky and Lasnik, we can assume that this trace
gets *-marked (the analogue of [--marked], i.e., that the wh-phrase enters the
numeration bearing a *-feature that becomes visible on one of its copies in
(9ac)). Importantly, *t remains at LF only in (9b), whereas it must be deleted
in (9a), yielding the operator-variable pair (What, t). Therefore, the ECP is
violated in (9b) but not in (9a). Let us now turn to (9c). In contrast to (9a),
the variable t in (9c) is not antecedent governed by an intermediate trace (IPadjunction being excluded by assumption) and is *-marked. Hence, because t
is a variable it may not be deleted and violates the ECP.
If intermediate adjunction to VP in the sentences of (9) were impossible, complement extraction across a wh-island should be as ungrammatical
as adjunct- and subject-extraction, because the undeletable initial traces in all
these examples would not be antecedent governed and would be *-marked.
Hence, the concept of intermediate adjunction provides an explanation for the
asymmetries found in (9).
Another strategy to account for the wh-island effects is pursued in Chomsky and Lasnik (1993). Adopting the basic idea of Rizzis (1990a) Relativized
Minimality, no appeal to barrier theory is made for the explanation of (9). The
data in (9) are then accounted for because movement of the wh-phrase does not
proceed in a successive-cyclic way via Spec CP. The long-extracted wh-phrases
fail to make the shortest move because they all skip Spec CP, which is a potential landing site. Hence, this movement violates the condition Minimize Chain
Links (MCL), a derivational version of Relativized Minimality.7 Under these assumptions, the trace in Spec IP in (9c) is *-marked because the wh-phrase fails
to make the shortest move (again, IP-adjunction is excluded by assumption).
As with the former account in terms of barrier theory, this trace remains at LF
because it represents the variable. In (9b) a uniform chain is created with one
*-marked trace. Again, deletion of the *-marked trace may not apply. In (9a),
on the other hand, a trace in a VP-adjoined position ensures that the variable
is not *-marked. This trace itself is *-marked but deleted at LF. As with the
barrier-theoretic analysis, intermediate adjunction to VP is necessary to account for the data in (9). Furthermore, given that MCL forces chain links to be
be minimal in length, VP-adjunction is obligatory under the assumption that
VP represents a possible landing site.
To sum up, an account for the asymmetries with respect to wh-islands
found in (9) can be given if one relies on the intermediate adjunction hypothesis in conjunction with the mechanism of intermediate trace deletion.
Before I turn to the next piece of evidence for the intermediate adjunction
hypothesis, I would like to clear up a potential objection that could be raised
Joachim Sabel
against this account of locality effects which relies on the deletion of intermediate traces. At first sight, this mechanism seems to raise a problem if at the
same time it is assumed that the Binding Theory applies at LF as assumed in
Chomsky (1995, Chapter 4). Consider again an example such as (3c); repeated
here as (10).
(10) [CP [Which friends of each other]i did they say [CP ti that [IP I should talk
to ti ]]]?
Recall that the anaphor inside the wh-element is not in a structurally adequate
position to fulfill Principle A of the Binding Theory; however, being located
in the intermediate Spec CP position, the anaphor may be bound by its antecedent. As pointed out in Chomsky (1993, 1995), adopting the copy theory
of movement, has the advantage that one need not assume literal reconstruction for the purposes of the Binding Theory i.e., that the wh-element in (10) is
lowered into the intermediate Spec CP position. Assuming that a copy of the
wh-phrase is located in the intermediate Spec CP, Chomsky (1993, 1995) assumes (following the analysis in Lebeaux 1983, 1985) that in (10), the anaphor
inside the intermediate copy (t i ) fulfills the Binding Theory at LF. Note that we
now seem to face the problem that at LF the anaphor inside the intermediate
trace (copy) in (10) can fulfill the Binding Theory only if intermediate traces
(or copies) are not deleted at LF. On the other hand, the Uniformity Condition
on Chains forces intermediate trace deletion in (10) (see Note 6).
A potential solution to this paradox is to reject the assumption that the
Binding Theory applies only at LF and to assume that Principle A of the Binding Theory can be stated in derivational terms (11) (as has been argued by several authors, cf. Belletti and Rizzi 1988; Uriagereka 1988; Lebeaux 1991; Sabel
1996; among others).
(11) Principle A of the Binding Theory can be fulfilled at any stage of the
derivation.
To sum up, in this section I have shown that an account of locality effects
with respect to wh-islands can be given if one relies on the intermediate adjunction hypothesis in conjunction with the mechanism of intermediate trace
deletion.
. Scope reconstruction
Further evidence for intermediate adjunction to VP can be gained from whquantifier interactions and scope reconstruction facts. The following sentences
(12) are not ambiguous. A wh-phrase has crossed a weak island and cannot
be interpreted as being in the scope of the universal quantifier (the examples (12) and (14) are from John Frampton, mentioned in Cheng 1991: 185,
see also Longobardi 1987, Cinque 1990, Saito 1994c, and Frampton 1999 for
discussion):9
(12) a. Which books do you wonder whether every student read?
b. Which books dont you know that every student read?
c. Which book didnt every student think that his teacher wrote?
Following Rizzi (1992), the Spec position of NegP can be analyzed as an A position filled with a sentential negation operator (that is phonetically unrealized in English). Let us assume that a wh-phrase that crosses Spec NegP violates MCL. Assuming intermediate adjunction, movement from the matrix VP
position to Spec CP in (12bc) leaves a *-marked trace, as shown in (13bc):
Which books do you [VP t wonder [CP whether every student [VP *t
read t]]]?
b. Which books dont you [VP *t know [CP t that every student [VP t
read t]]]?
c. Which book didnt every student [VP *t think [CP t that his teacher
[VP t wrote t]]]?
(13) a.
The data in (13) can then be explained if we assume that a *-marked trace
blocks scope reconstruction in the sense that a wh-phrase may not be reconstructed into a *-marked trace position or into a trace position that is/was ccommanded by a coindexed *-marked trace. Therefore, the wh-phrases in (13)
may not be interpreted as being in the scope of the universal quantifier. Now
consider the examples in (14):
(14) a.
book did every student [VP t wonder [CP whether his teacher
[VP *t wrote t]]]?
??Which
Joachim Sabel
b. Which book did every student [VP t think [CP t that his teacher didnt
[VP *t write t]]]?
In contrast to (13), the examples in (14) are ambiguous. Nothing excludes the
possibility that the wh-phrase in (14a) could be interpreted in the scope of
the universal quantifier in the matrix clause, given that t is neither *-marked
nor c-commanded by a *-marked trace. Although (14b) can be accounted for
in a different way i.e., by assuming that Spec CP is the relevant reconstruction site for the wh-phrase, (14a) is an example that might be taken as further
evidence for the claim that (long) wh-movement proceeds via adjunction to
VP. Again, assuming a derivation which contains an intermediate trace in an
adjoined position provides an account of the asymmetry found in (13)(14).
A further argument for intermediate traces in VP-adjoined positions
comes from scope reconstruction facts in connection with variable binding. It
is well-known that the interpretation of a pronoun as a bound variable is only
possible if the pronoun is bound by a quantificational antecedent. Fox (1999)
assumes that variable binding has to be synactically encoded at LF and that
it (potentially) triggers reconstruction. He argues that the contrasts between
the examples in (15) provide an argument for intermediate VP-adjunction of
wh-phrases. In (15a), reconstruction of the wh-phrase into the trace position
is possible, licensing the bound variable reading of the pronoun he. A different
situation arises in (15b). Reconstruction leads to a configuration in which the
referential expression Ms. Brown is bound by the pronoun she, giving rise to a
violation of Principle C of the Binding Theory. The crucial example is (15c).
As evident from (15b), the position *t in (15c) is not a possible reconstruction
site. If reconstruction took place into such a position this would incorrectly
predict (15c) to yield a Principle C violation like (15b). Therefore, in order to
account for the well-formedness of (15c) it seems to be necessary to postulate
an intermediate trace t in a VP-adjoined position that represents a possible
reconstruction site.
(15) a.
[Which of the books that he1 asked her2 for] did Ms. Brown2 [VP give
every student1 t]?
b. *[Which of the books that he1 asked Ms. Brown2 for] did she2 [VP give
every student1 *t]?
c. [Which of the books that he1 asked Ms. Brown2 for] did every student1
[VP t get from her2 *t]?
Joachim Sabel
(16) *Where did they [VP t [VP leave Boston [PP t [PP before [CP (t ) PRO [VP
t [VP meeting John t]]]]]]]?
In order to rescue the intermediate adjunction hypothesis one might argue that
(16)(17) could be ruled out by another independently motivated constraint
i.e., a constraint that excludes movement from Spec CP to an adjoined position
as a case of Improper movement (Hoekstra & Bennis 1989; Mller & Sternefeld
1993; Grewendorf & Sabel 1994). Note that this constraint could be argued
to be motivated anyway given that movement from Spec CP to an A-position
must be similarily ruled out (Chomsky 1981). For an illustration of this fact,
consider the following cases of illicit long A-movement:
(18) a.
b. i. *John was expected [CP (t ) that [IP it was told t [that Mary is
beautiful]]].
ii. *Who was expected [CP t that [IP it was told t [that Mary is beautiful]]]?
Chomsky (1995: 326) excludes (18a-i) by arguing that the complement of decide requires a PRO subject. Given that John cannot bear Null Case, a feature
mismatch arises, which cancels the derivation (Chomsky 1995: 309). The same
explanation rules out the derivation (18a-ii). On the other hand, example (18b-
i) is ruled out by the Minimal Link Condition (MLC) (see Chomsky 1995: 296,
311, 358 for discussion):
(19) a.
Joachim Sabel
.. Intermediate adjunction to IP
In adopting the intermediate adjunction hypothesis, it has been stipulated that
adjoining wh-phrases to IP is forbidden, in contrast to other operators (Chomsky 1986a: 5, 32). In order to see why this constraint is needed, let us consider
the case of wh- and subject-islands:
(21) *Who do you [VP t wonder [CP how [IP t [IP t could fix the car]]]]?
(22) *Who do you t think [CP (t ) that [IP t [IP [NP pictures of t] are on sale]]]?
(23) a.
Joachim Sabel
(26) a.
Co komu
Monika daa t t?
what to-whom Monica gave
What did Monica give to whom?
b. [CP Co [IP komu [IP Monika . . .]]]
Given that only one wh-phrase may move to Spec CP in Polish and that further
wh-elements in a sentence containing multiple wh-words obligatorily front
to IP-adjoined positions, as in (26), the ban on intermediate adjunction predicts that long wh-movement of a wh-phrase located in an adjoined position
should be impossible. This is in fact the case, as can be seen from (27) (Rudin
1988: 454). As shown in (27a), a subjunctive clause permits long extraction in
Polish, but only one wh-element may be long fronted (27b). Given that one of
the wh-elements in (27) is moved to the embedded Spec CP position and the
other is adjoined to the embedded IP, only the former wh-phrase may move
from the embedded to the matrix Spec CP:
Janek kupi t t]?
Co Maria chce [CP z eby komu
what Maria wants that to whom Janek buy
b. *Co komu
Maria chce [CP z eby Janek kupi t t]?
what to whom Maria wants that Janek buy
What does Maria want Janek to buy for whom?
(27) a.
If intermediate adjunction to IP is prohibited, it is expected that the second whelement is not allowed to move successive cyclically from the embedded clause
into the matrix clause. Hence, wh-movement in Polish provides independent
empirical evidence for the ban on intermediate adjunction of wh-elements
to IP.16
.. Copy movement and intermediate adjunction to VP
Let us now turn to intermediate VP-adjunction. If intermediate adjunction is
generally impossible, traces in VP-adjoined positions cannot be generated. This
conclusion solves a long standing empirical problem in languages such as German and Afrikaans (du Plessis 1977) which allow for the spelling-out of traces
of successive-cyclic wh-movement (see (28a) vs. (28b)). The fact that these
copies may only be spelled-out in Spec CP (28b) but never in VP-adjoined
positions, as shown in (29), is explained if intermediate adjunction to VP is
impossible.17
(28) a.
(29) *[CP Wen glaubst du [CP wen [IP Hans wen meinte [CP wen
who
Hans who thought
who
who believe you
[IP das Argument (wen) berzeugt]]]]]?
the argument who convinces
Who do you believe that Hans thought that the argument convinces?
Furthermore, dispensing with VP-adjunction has advantages from a theoretical point of view if we give up the barrier theory that made VP-adjunction
necessary.18 Thus, for example, the adoption of the condition Minimize Chain
Links in Chomsky and Lasnik (1993: 540ff.) to account for wh-island violations
makes the notion of a barrier by inheritance superfluous and gives the possibility of simplifying the theory of barriers along the lines of Huangs (1982)
Condition on Extraction Domains (CED). I will adopt this idea and assume
in the following that every non-complement is a barrier (cf. also Cinque 1990;
Sabel 2002). This revised notion of barrier makes VP-adjunction superfluous
for elementary cases of extraction such as object extraction in (28) and it correctly predicts that adjuncts and subjects are islands for extraction (as long as
they do not represent intermediate adjunction sites).
.. Pronoun binding and intermediate adjunction to VP
If VP-adjunction is excluded, the question remains of how to account for the
data in Section 2.4 involving reconstruction and variable binding. As argued
in Fox (1999), *t in (15b) and (15c), repeated here as (30a) and (30b), is not a
possible reconstruction site.
(30) a. *[Which of the books that he1 asked Ms. Brown2 for] did she2 [VP give
every student1 *t]?
b. [Which of the books that he1 asked Ms. Brown2 for] did every student1
[VP t get from her2 *t]?
Joachim Sabel
To account for the grammaticality of (30b), he assumes the presence of an intermediate trace t in VP-adjoined position that serves as a reconstruction site
yielding the bound variable reading of the pronoun inside the wh-phrase.
There are two reasons to reject this argument for intermediate VPadjunction. Firstly, the argument depends on the claim that *t in (30b) like
*t in (30a) is not a possible reconstruction site. But, as can be seen from (30 b),
*t in (30b) is a possible reconstruction site.
(30 ) a. *Did she2 give every student1 [the books that he1 asked Ms. Brown2
for]?
b. Did every student1 get from her2 [the books that he1 asked Ms.
Brown2 for]?
Secondly, (30 a) and (30a) are equally bad, although the wh-phrase in (30 a)
does not contain a pronoun. Therefore reconstruction should not be forced
in (30 a). This shows that the Principle C effect in (30a) does not result from
reconstruction. I conclude that the distribution of Principle C violations in
(30) is not due to reconstruction of the wh-phrase (i.e., either into the position
t or into an intermediate VP-adjoined position t ).
(30 ) a. *[Which of the books that John asked Ms. Brown1 for] did she1 [VP give
every student t]?
b. [Which of the books that John asked Ms. Brown1 for] did every student
[VP get from her1 t]?
This provides an alternative account for the analysis of the examples (15a) and
(30). Given that the pronouns in these examples are all A-bound before whmovement takes place, reconstruction is not necessary.19
The discussion so far has shown that the ban on intermediate adjunction
to adjuncts, arguments, VP, and IP should be subsumed under a more general constraint, according to which intermediate adjunction in general is impossible. This conclusion is empirically supported and represents a welcome
conceptual simplification. I will therefore assume that the following constraint
holds:
(31) Constraint on Adjunction Movement (CAM)
Movement may not proceed via intermediate adjunction.
(31) ensures that movement can never go into an intermediate adjoined position, but go only into an adjoined position which represents a goal position
(a final destination for movement). According to the CAM, elements that are
base-generated in adjoined positions, may be long extracted via successivecyclic movement as long as no intermediate trace in an adjoined position is
created, as it is the case, for example, with extraction of base-generated adjuncts and inverted subjects in null subject languages. The CAM predicts that
an element may undergo successive-cyclic movement only via specifier positions. Alternatively, may be moved if it is adjoined to an element , which
itself undergoes movement. The latter option represents a typical case of head
movement or multiple fronting (see Note 38).
However, the data discussed in Section 2, which were taken as evidence
for intermediate adjunction to VP, have been only partially explained in a way
compatible with the CAM. In the following, it will be shown that an alternative
explanation may also yield a coherent picture for the other phenomena discussed in Section 2 i.e., the Principle A reconstruction effects and extraction
asymmetries that arise from wh-movement across wh-islands.
.. Wh-islands and intermediate adjunction to VP
Consider again, the well-known complement/non-complement asymmetry, already discussed in Section 2.3, and repeated here in (32), without intermediate
traces:
(32) a. ??What do you wonder [CP how John could [VP fix t]]?
b. *How do you wonder [CP what John could [VP fix t]]?
c. *Who do you wonder [CP how [IP t could fix the car]]?
Joachim Sabel
Chomsky and Lasnik (1993) assume that movement that violates MCL leaves
a *-marked trace. If there is no intermediate trace in the embedded sentences
of (32), complement extraction across a wh-island as in (32a) should be as
ungrammatical as adjunct- and subject-extraction (32bc), because the initial
traces in these examples are always *-marked. Hence, if there is no intermediate trace in the embedded sentences in (32), why is there a difference in grammaticality between complement extraction, on the one hand, and adjunct and
subject extraction, on the other? The question of how to explain this extraction asymmetry is also left unanswered in Chomskys (1995: 295) discussion
of wh-islands. Recall that Chomsky (1995) assumes that there is a distinction
between +Interpretable and Interpretable features. The [wh] features on whphrases are +Interpretable. Importantly, these features on XPs remain accessible for the computational system after checking, ensuring that one and the
same wh-phrase may undergo successive-cyclic movement or attraction.20 It
follows that wh-island violations never arise. An embedded [wh] C0 as in (33a)
always attracts the closest wh-phrase according to the MLC (see (19)). This
wh-phrase moves to Spec CP and is then attracted again by the [wh] matrix
C0 , as in (33b), because when located in the embedded Spec CP, it is closer to
the matrix C0 than the wh-phrase in situ. The [wh] feature of the embedded
C0 is checked in (33a). Therefore, it cannot attract what at a later step of the
derivation (33b) (yielding *Who you wonder what could solve?). Furthermore,
such a movement would be counter-cyclic. Chomsky notes that the example
(33b) converges with all relevant features checked, yielding gibberish since the
structure cannot be interpreted adequately.21
(33) a. . . . wonder [CP who [IP t could solve what]]
b. [CP Who do you wonder [CP t [IP t could solve what]]]?
The observation that extraction out of wh-islands is only possible from positions extends to the analysis of similar examples in pro-drop languages.
As argued in Rizzi (1986), the subject position in languages such as Italian
and Spanish may be occupied with an expletive pro and the inverted subject
in -position may be Case marked in a way other than via specifier head agreement with Infl, granted that the inverted subject position behaves like a Case
marked A-position. Chomsky (1995, Section 4.5), who assumes that covert
movement is in fact feature-movement, reaches a similar conclusion. If we accept that the base position of the inverted subject is a -position, then in Spanish (and Italian; see Rizzi 1982: 51 for the corresponding examples in Italian),
nothing blocks movement of the subject (35c), in contrast to English (32c), or
German (34c):
(35) a. *Qu no sabes
quin compr t ?
what not know-you who bought
What dont you know who bought?
b. *Por qu no sabes
qu comprar t ?
why
not know-you what to-buy
Why dont you know what to-buy?
Joachim Sabel
c.
?Quin
no sabes
qu compr t ?
who not know-you what bought
Who dont you know what bought?
(Jaeggli 1988)
Let us turn to the question of how the derivation of wh-island violations proceeds, in light of the prohibition of intermediate adjunction and the MLC. I
assume that the embedded C-System in these cases may contain multiple landing positions for wh-phrases i.e., multiple specifiers (see also Reinhart 1981;
Comorovski 1986, 1989; Mulders 1997; Richards 1997). I follow an idea of
Koizumi (1994) who assumes a multiple specifier analysis for topicalization
in English, embedded verb second in Yiddish and the Scandinavian languages,
and multiple wh-fronting in the Slavic languages. Koizumi (1994) presents evidence that the head of a phrase with multiple specifiers must contain hierarchically ordered features which have to be checked in a certain order.23 Adopting
the main idea of Koizumis analysis, I assume that the selected C-head in indirect questions may bear more than one [wh] feature. The [wh] features in this
head are hierachically ordered ([wh1] > [wh2]) and thus have to be checked in
different specifier positions of CP by different wh-phrases.24 If this is true, the
embedded Spec CP positions in the examples above have the following structure, with t 2 either as an A -position (in the case of adjunct extraction) or a
base/derived A-position (in the case of argument extraction):
(36)
Both wh-phrases move to the intermediate specifier positions. The intermediate trace t 2 is located in Spec2 , whereas Wh1 has moved to Spec1 . In (36a)
and (36b), different wh-phrases occupy the specifier positions, because they
differ with respect to the [wh] feature they bear. In both cases, Wh2 is closer to
the attracting matrix C-head. The idea that the wh-phrases bear different [wh]
features allows us to motivate movement of both wh-phrases without violating
the MLC.25 Even more important, the embedded C-head bears [wh] features
that require a wh-operator in both Spec positions. Note that the intermediate
trace of the wh-phrase Wh2 is not an operator. Hence, this trace is *-marked,
after the matrix C-head attracts Wh2 .26 Let us now adopt a proposal made
in Chomsky (1995: 388, Fn. 75), according to which, besides L-relatedness, positions are relevant for the Uniformity Condition on Chains (37). Recall that
operator-variable chains are the only chains in which intermediate trace deletion takes place. No intermediate trace deletion applies in uniform chains. If
we take L-relatedness and -positions to be the relevant property P, then we
can reformulate the condition for intermediate trace deletion. Let us assume
that intermediate trace deletion may only apply if a chain fulfills both conditions in (38).
(37) Uniformity Condition on Chains
A chain C is uniform with respect to P (UN[P]) if each i has property P
or each i has property non-P.
(38) a. A . . . (A ) . . . A
b. . . . ( ) . . .
(operator-variable construction)
Now we are able to explain the examples (32) and (34)(35). Let us assume
that these examples are derived as shown in (36ab). In (32bc), showing subject and adjunct extraction in English, the intermediate trace *t 2 (with respect
to (36ab)) may not be deleted since the chain (Wh2 , *t 2 , t 2 ) is uniform i.e.,
each member of the chain is located in a non--position. The initial trace t 2
in (32b) marks the base-position of the adjunct whereas t 2 (with respect to
(36a)) in (32c) is located in Spec IP. Therefore, the intermediate traces may not
be deleted. In contrast, the chain (Wh2 , *t 2 , t 2 ) in (32a) is not uniform because t 2 is located in a -position. In this example the intermediate trace must
be deleted, and at LF we get the operator-variable pair (Wh2 , t 2 ). The slightly
deviant character of this sentence is due to the fact that a *-marked trace was
created during the derivation. The explanation for (32bc) extends to the German examples (34bc). In contrast to (32a), object extraction out of a whisland is ungrammatical in German since t 2 is located in a non--position i.e.,
the position in which structural Case is assigned to the extracted object. Hence
the relevant chains in (34) are all uniform, and consequently the intermediate *-marked trace cannot be deleted. The explanation for the Spanish (and
corresponding Italian) cases (35ab) is the same as for German. In contrast to
these languages, subjects may be extracted out of wh-islands, as in (35c), since
extraction takes place from a (Case-marked) -position. Again deletion of the
intermediate trace *t 2 is forced to create an operator variable pair.27, 28
Note that this analysis makes the prediction that in languages with object
shift, wh-questioning of objects across wh-islands should be impossible. In fact,
languages with obligatory object movement into a Case position such as Icelandic do not allow for object extraction out of wh-islands (Maling 1979), see
Joachim Sabel
(40) a.
CP
In addition, this analysis shows that contrary to what has been assumed
the reconstruction data discussed in Section 2.1 do not provide evidence for
intermediate adjunction. Consider again the examples (4) (=(41)) and the
examples (42):
(41) a. *John wonders [where [Mary bought the pictures of himself ]].
b. ??[Which pictures of himself ]i does John ti wonder [where [Mary ti
bought ti ]]?
(42) a. *John asked Mary [where [Paul bought the pictures of herself ]].
b. ??[Which pictures of herself ]i did John ask Mary [where [Paul ti bought
ti ]]?
The fact that the anaphoric expression may not be bound in its base position (41a), (42a) provides no reason to conclude that there is a trace in a VPadjoined position of the matrix clause. Instead, the long-moved wh-phrase in
(41b), (42b) is extracted via Spec2 of the embedded CP, and in this position it
may be bound in accordance with (11) by the antecedent in the matrix clause
i.e., by the matrix subject in (41b) and by the matrix object in (42b).
Summarizing the discussion on wh-islands, we can conclude that extraction out of wh-islands does not provide evidence for intermediate traces in
If it is assumed that the (potential) intervening barriers in (43a) and (44a) are
neutralized by successive-cyclic adjunction, it is unclear why the same derivation does not provide an acceptable result in (43b), (44b). The assumption that
successive-cyclic adjunction may apply in one example but not in the other
Joachim Sabel
(45) a.
(46) a.
Consider the possibility that (45b), (46b) are derived via intermediate adjunction. The objects in these examples will be adjoined to the embedded VP, leaving a trace t . Assuming the Lasnik and Saito mechanism, the trace t prevents
the initial trace t from being *-marked. Given that intermediate argument
traces in non-A-positions must delete, the resulting chain, (NP, t), should be
a legitimate object. This is clearly the wrong prediction. The empirical problem that (45b) and (46b) provide for intermediate adjunction does not only
hold for Russian. Scrambling out of adjuncts is impossible in all scrambling
languages (Sabel 1996).
Furthermore, I assume that the examples (45a), (46a) show that adjunction to a sentence-level is possible. This provides an additional problem for an
analysis of scrambling derived by successive-cyclic adjunction. If the grammar
allows for successive-cyclic adjunction, then (45a), (46a) represent intermediate steps of the derivations (45b), (46b). Therefore it must be explained why
a derivation of (45b), (46b) containing intermediate traces in adjoined positions is impossible, while its underlying partial derivations are possible. If we
assume that scrambling applies in accordance with (31), this question does
not arise. Hence, scrambling provides further evidence for the constraint on
adjunction.29
Joachim Sabel
Chomsky (1994, 1995, Chapter 4) and Ura (1994), assuming the Merger theory of phrase-structure building in Chomsky (1994, 1995), have claimed that
the sentence structure in Japanese may contain multiple specifiers, which provide multiple L-related positions. According to Chomsky (1994, 1995: 286)
this option may be due to parameterized properties of Agreement (or Case)
i.e., one and the same head may check Agreement (or Case) more than one
time. If this is true, then the higher subject in (49) and the scrambled NP
in (47b) are located in a Spec2 position, which is an L-related position with
A-properties (50a).
(50) a. [IP NPacc [I NPnom [I [VP . . . t. . .]]]]
b. [IP NPacc [IP NPnom [I [VP . . . t. . .]]]]
Multiple L-related specifiers are not available in German, where the corresponding feature-checker can check only once. The scrambled NP in German
therefore is located in an adjoined position (48b), which is associated with A properties (50b). Consequently, only in (47b) can the scrambled object act as
an A-binder for the anaphor. In (48b) the scrambled object is located in an
IP-adjoined position with A -properties. Therefore it cannot bind the anaphor
in subject position. Now, given the CAM, we predict that an element that is
moved to an adjunction site inside an embedded clause may not move further
into the matrix clause. This is the case in German (44b), whereas scrambling
in Japanese may proceed in a successive-cyclic manner via an embedded Spec2
IP position (44a) (cf. Ura 1994; Grewendorf and Sabel 1999 for more details of
this analysis). Therefore scrambling out of finite clauses is possible in Japanese
but not in German. This analysis provides a uniform account of the different
A-/A -properties and locality restrictions of scrambling in languages such as
German and Japanese in that it predicts that a scrambling language allows for
scrambling out of finite clauses and short A-scrambling if multiple A-specifiers are
licensed in this language.31
Note also that from this analysis an explanation for the reconstruction
properties with long scrambling arises i.e., one that does not rely on intermediate adjunction. Recall the discussion of the examples (5) in Section 2.1,
repeated as (51) below. In (51a) the embedded subject is the only possible antecedent for otagai each other. If the embedded object containing the reflexive
is short scrambled in front of the embedded subject, the reflexive otagai may
be bound by the matrix subject (51b). In (51c) the NP containing the anaphor
is scrambled out of the embedded clause in front of the matrix subject; also
in this case the matrix subject can be co-referent with the anaphor. Given the
analysis of scrambling in Japanese mentioned above, the embedded object in
(51c) is long scrambled via the embedded Spec2 IP position and not via an adjoined position as indicated in (51c). Then (51c) is not derived via successivecyclic adjunction and does not provide evidence for an intermediate trace in
an adjoined position inside the embedded clause.
(51) a.
Joachim Sabel
c.
The proposed account for the different options of long scrambling in Japanese
and German (44) as well as for its reconstruction properties (51c) relies on the
assumption that a scrambled element obligatorily moves through an intermediate landing site in the embedded clause. Why is an alternative derivation of
the examples (44) and (51c) impossible in which scrambling applies in one fell
swoop i.e., does not proceed in a successive-cyclic manner?
One possible answer is that (long) scrambling, like wh-movement, is an
obligatory movement operation that is driven by a feature i.e., a scrambling
feature [] (see also Collins 1995; Miyagawa 1997; Grewendorf & Sabel 1999;
Chomsky 2000 for this assumption). The scrambling feature is associated with
Agrs-features in Infl (or ) which triggers scrambling to IP (or P). The fact
that the so-called scrambling languages are all pro-drop languages (for example, SOV-languages such as German, Hindi, Japanese, Korean, Modern Persian,
Turkish and SVO-languages such as Polish and Russian all license argumental
or non-argumental pro), suggests that the language-specific ability of Agrs (or
Infl) to license pro is a necessary (although not a sufficient) condition for a
language to have scrambling i.e., for Agrs (or Infl) to bear the scrambling feature (for this generalization see also Koster 1986; Reuland & Kosmeijer 1988;
Tonoike 1997; among others).32 Then, in a simple sentence with short scrambling to IP, Infl0 and the constituent to be scrambled contain the scrambling
feature. For example, the scrambling feature in (47b) and (48b) is realized on
Infl0 . Given Chomskys (1993) definition of Checking Domain33 this feature
is checked via substitution into Spec2 of IP in Japanese (47b) or via adjunction
to IP in German (43a), (48b).
Applying the idea of feature-driven movement to long scrambling in (44)
and (51c), let us assume that assignment of the scrambling feature to Agrs in
I0 implies assignment of a scrambling feature to each intermediate I0 . Consequently, in sentences such as (44) or (51c) displaying long scrambling out of
a finite clause to IP, the scrambling feature is located in both I0 of the matrix
as well as I0 of the embedded clause and the scrambled element has to check
both scrambling-features. The scrambling-feature can be checked via adjunction to IP in German or via substitution into Spec2 of IP in Japanese. Now,
given the assumption that successive-cyclic adjunction is generally impossible,
elements in German may not be long scrambled because a scrambled element
that is moved to an adjunction site inside the embedded clause i.e., IP in (44b),
may not move further into the matrix clause. On the other hand, scrambling
in Japanese may indeed proceed in a successive-cyclic manner via the embedded Spec-(IP) position as in (44a) and (51c) i.e., not via XP-adjunction (see
Grewendorf & Sabel 1999 for more empirical motivation for this analysis). In
this way we can derive a uniform account of the different A-/A -properties and
locality restrictions of scrambling in languages such as German and Japanese.
. Quantifier raising
Let us briefly consider one more type of adjunction movement in light of the
CAM: quantifier raising (QR). Given that QR, like scrambling in German, is
adjunction movement, we expect that it may not apply successive-cyclically.
Combining the proposal of Chomsky (1995), that quantifier raising is triggered
by some feature with the suggestion of Kiss (1987: 249) that scrambling in the
overt syntax parallels QR at LF we could say that the feature triggering quantifier raising is located in the same functional heads as the scrambling feature.34
Then, given that QR is constrained by the CAM; we can explain its clausebound character i.e., the non-ambiguity of (52) (see Hornstein 1984; Williams
1986; Mahajan 1990; Cheng 1991; Lasnik & Saito 1992; among others):
(52) Someone thinks that everyone saw you at the rally.
Recall that May (1985) and Chomsky (1986a: 7) argue that the distributive
reading in examples such as (53) is a consequence of adjunction movement
of the subject to IP at LF, as in (53 ), where IP does not include everyone. Instead, who and everyone are dominated by exactly the same maximal projections (CP):
(53) [CP Who does [IP everyone like t]]?
(53 ) [CP Who1 does [IP everyone2 [IP t 2 like t 1 ]]]?
Joachim Sabel
Which book did every student wonder [CP whether [his teacher [VP
wrote t]]]?
b. Which book did every student think [CP t that [his teacher didnt [VP
write t]]]?
The wh-phrases can be interpreted in the matrix clause i.e., they yield a distributive reading. (54b), for example, can have a family of answers: Paul thinks
that his teacher didnt write Barriers, John thinks that his teacher didnt write Anarchy, State and Utopia. The distributive reading in (54a) was argued to be a
consequence of an intermediate trace in VP-adjoined position. However, the
distributive reading in (54a) (and also (54b)) can be due to the fact that after
quantifier raising the quantifier every student has scope over the entire question
a clause-bound process that, in addition, is blocked by so-called inner islands
such as negation as in (55c) (=(12c)) (see for example Beck 1996).
(55) c.
Which book didnt every student think [CP t that [his teacher [VP wrote
t]]]?
Hence, the scope reconstruction facts do not provide an argument for intermediate adjunction (to VP).
Let us sum up the empirical results so far. The CAM was shown to hold for
different types of XP-movement i.e., for wh-movement and scrambling (including covert quantifier movement). Furthermore, we have seen that we can
give an alternative account for the scope reconstruction facts in connection
with extraction out of wh-islands, i.e., one that does not rely on the intermediate adjunction hypothesis. Let us now look at the problematic consequences of
intermediate adjunction for empty operator movement.
. Movement of empty operators and further XP-movements
In his discussion of parasitic gap constructions, Chomsky (1986a: 65f.) assumes
that NPs may be adjoined to PP-adjuncts as in (56):
(56) Which booki did [IP you [VP return ti [PP Oi [PP without [reading ei ]]]]]?
For the parasitic gap construction to be licensed, it is assumed that one composed chain has to be constructed out of the two A -chains in (56). This (composed) chain can only be constructed if the empty operator is 0-subjacent to
the real trace t i . Hence, t i and the empty operator may not be separated by
one barrier or more. In order to compose the relevant chain, operator movement in the infinitival has to take place and the operator has to adjoin to PP.
Adjoined to PP it is 0-subjacent to the trace of the wh-phrase, according to the
framework of Chomsky (1986a).
Given the assumption that the PP-adjunct is base-generated in a VPinternal position and not adjoined to VP or IP, the empty operator in (56)
is not 0-subjacent to the structural position of the subject. This predicts the
ungrammaticality of sentences like (57):
(57) *Which paperi [IP ti [VP disappeared [PP Oi [PP before you could [VP read ei
]]]]]?
If the operator in PP-adjoined position (56) were permitted to leave an intermediate trace, a derivation like (57 ) would be possible:
(57 ) *Which paperi [IP ti [VP Oi [VP disappeared [PP ei [PP before you could [VP
ei [VP read ei ]]]]]]]?
Given (57 ), the sentence is predicted to be grammatical because the empty operator and the trace in subject position are no longer separated by a barrier (cf.
also Browning 1987: 201f.).36 Hence, the concept of intermediate adjunction
undermines the account of the parasitic gap constructions (56)(57). On the
other hand, if we dispense with intermediate adjunction, the derivation (57 )
is excluded.
To complete the discussion of the CAM in connection with movement of
XPs, it should be noted that further construction types have led several authors to the conclusion that intermediate adjunction may not be allowed. For
example, it must also be excluded that A-movement may apply via intermediate adjunction (cf. Chomskys 1986a: 74, 1995: 326 discussions of Improper
movement in different theoretical frameworks).
(58) a. *John seems that [IP t [IP it was told t [ that Mary is intelligent]]].
b. *John seems that [IP t [IP it is considered [ t to be intelligent]]].
Joachim Sabel
Li 1990). However, this account of (58) is in conflict with the Uniformity Condition on Chains. As already pointed out (cf. Section 2.3), only uniform chains
are legitimate LF objects in the framework of Chomsky and Lasnik (1993) and
Chomsky (1995). For illustration, consider (59)(61):
(59) A . . . A
. . . A (A-/L-related chain)
. . . A (A -/non-L-related chain, X0 -chain)
(60) A . . . A
(61) A . . . (*A ) . . . A (operator-variable construction)
(63) a.
Joachim Sabel
If te and lo form a complex head in (62c) and (63a) that is formed via adjunction, the CAM predicts that (62c) cannot be the input for a structure in which
lo moves on into the highest clause, leaving an intermediate trace in the position occupied by lo in (62c). This prediction is in fact borne out, as can be seen
in (63b). On the other hand, if (63b) is derived from (62a) and lo moves in one
fell swoop into the highest clause, clitic climbing violates MCL (see Sabel 2001
for more discussion).
To sum up, the impossibility of successive-cyclic adjunction in cases of verb
movement and clitic climbing (63b) follows from the CAM and requires no
further stipulation. Given the constraint on adjunction movement, derivations
including intermediate traces in adjoined positions cannot be generated. An
element dominated by only one segment of a complex head may not move further on its own because this movement creates a trace located in an adjoined
position.
At this point of the discussion, one might wonder whether the CAM is
too restrictive. Two cases have to be considered. First, the CAM does not exclude a head formed by adjunction from moving as a whole, as is the case with
clitic climbing (63a) (or as is the case with V-to-I-to-C in general). Here, a
new head is formed after each movement step, so that every head was adjoined
only once.38
Secondly, it does not exclude the possibility that the head of a complex
head may move and strand the segment that was created by adjunction to
itself. Derivations of this type must be allowed. For example, they are attested
in connection with verb-movement in Dutch (see Koster 1987).
As can be seen from the Dutch examples (64ab), it is impossible for the
infinitival complement to remain in its base position, nor is extraposition allowed. The only possibility is that incorporation creates a complex verb (64c).
(64d) shows that the matrix verb moves to C0 . This is necessary because of
the verb-second character of Dutch. Under the assumption that the matrix
and embedded verbs must form a single head (64c), a trace inside this verbal
complex must be licensed (64d).
(64) a. *dat zij [ het boek te lezen] schijnt
that she the book to read seems
She seems to read the book.
b. *dat zij ti schijnt [ het boek te lezen]i
seems
the book to read
that she
c.
te lezenv
to read
lezenv .
read
From this I conclude that the CAM is not too restrictive. It represents a correct
generalization and restricts the set of potential derivations. Adjunction movement can never go into an intermediate adjoined position, but only into an
adjoined position that represents a goal position. Successive-cyclic movement
of an element may only proceed via Spec positions.
Let us summarize the results of the discussion in the preceding sections. In
Section 2, I have presented evidence for the intermediate adjunction hypothesis from reconstruction, weak crossover, and locality effects. In Section 3, I
have tried to show that, on the other hand, a number of theoretical and empirical arguments argue against the intermediate adjunction hypothesis. Data
and considerations concerning a wide range of movement types, such as whmovement, A-movement, QR, extraposition, scrambling, and head movement,
motivated the postulation of the constraint on adjunction movement (CAM),
which generally excludes intermediate traces in adjoined positions:
(65) a. Substitution
b. XP-Adjunction
WP
YP
WP
YP
W'
WP
XP
t'
XP
XP
t'
XP
Joachim Sabel
c. X0 - Adjunction
WP
WP
Y
XP
W
X'
X
t'
YP
X
(Head-Movement)
In (66), who has to check nominative case, in addition to a [wh] feature. The
wh-element is first attracted by an operator feature of the embedded C0 . In
. Conclusion
In this article, I have argued that movement is restricted by the Constraint on
Adjunction Movement (CAM), a constraint that forbids traces in intermedi-
Joachim Sabel
ate adjoined positions. The CAM predicts that the only existing intermediate
traces of a moved element are traces in specifier positions. The empirical evidence against intermediate adjunction was formulated with respect to different movement types, such as wh-movement, empty-operator-movement, Amovement, extraposition, quantifier raising, scrambling, and head movement.
Furthermore, those data that were traditionally used as providing evidence for
intermediate adjunction have been explained as involving movement via a second specifier position. For example, the analysis of scrambling in German and
Japanese presented in Section 3 rests on the assumption that Japanese allows
for multiple A-specifiers whereas German does not. On the other hand, multiple (CP) A -specifiers seem to be the unmarked case in languages as argued in
connection with the proposed analysis of extraction out of wh-islands.
Notes
* For helpful comments I would like to thank Noam Chomsky, Chris Collins, John Frampton, Sam Epstein, Gisbert Fanselow, Danny Fox, Hans-Martin Grtner, Gnther Grewendorf, Eric Groat, Shin-Sook Kim, Shigeru Miyagawa, Norvin Richards, Jeff Runner, Mamoru
Saito, Gert Webelhuth, and Ede Zimmermann.
. In most recent analyses of movement phenomena that are within the Minimalist Program, the idea that successive-cyclic movement proceeds via intermediate adjunction plays
an integral part (see Chomsky & Lasnik 1993; Saito 1994b; Takahashi 1994; Agbayani 1997;
Takeda 1997; Fukui & Saito 1998; among others). For example, with respect to reconstruction properties discussed in the following section, the analysis in Takahashi (1994) is settled
in the Minimalist framework (Chomsky 1993; Chomsky & Lasnik 1993) and rests on the
intermediate adjunction hypothesis as the analysis in Barss (1986, 1988), who assumes the
Barriers framework. Furthermore, Richards (1997), working in the Minimalist framework,
in his analysis of the absence of weak crossover effects in several languages, addresses the
correlation with intermediate IP adjunction, like the authors mentioned in Section 2.2 who
present an analysis in the Barriers framework. In addition, concerning the discussion of locality effects of movement in Section 2.3, the idea that every non-complement is a barrier,
which is a variant of the notion of L-marking barrier in Chomsky (1986a), can be found in
Chomsky and Lasnik (1993) and in Chomsky (2000). This analysis in terms of barrierhood
relies on the distinction between the notion of segment and categories already found
in Chomsky (1986a) and adopted in the Minimalist framework (see Chomsky 1993: 11,
1995: 177; Fukui & Saito 1998; among others). Thus, the following arguments presented
in favor of the intermediate adjunction analysis carry over to the different frameworks.
. There has been some debate on whether a structural difference between specifier and
adjunction positions exists or whether all sisters to the projections of a head-complement
structure have to be analyzed as specifiers (see Fukui 1986; Kuroda 1992; Kayne 1994; Ura
1994; Fukui & Saito 1998; among others). In contrast to my analysis in the text, it might
Joachim Sabel
juncts and heads are non-L-related elements. They only move to non-L-related positions,
creating legitimate objects, i.e. uniform chains where every member occupies a non-Lrelated position. A-chains with each element in an A- or L-related position are also uniform. Hence, deletion of traces does not apply in these uniform chains. In the case of long
wh-movement of arguments it is important that only operator-variable pairs are licensed at
LF, therefore intermediate traces are deleted from A - or non-L-related positions as a Last
Resort operation, yielding legitimate LF-objects of the form (Wh, t), where t represents the
Case-marked position, i.e. the variable.
. Chomsky (1995) accounts for wh-island violations in terms of the Minimal Link Condition (MLC). This analysis will be discussed in Section 3.1.5.
. Furthermore, if it is assumed that reconstruction is impossible for A-moved elements
(Chomsky 1995) or that A-movement does not leave copies (Hornstein 1998; Lasnik 1998,
1999; Saito & Hoshi 2000), the examples in (i) are no longer problematic. Anaphors which
are contained in elements that are A-moved out of the c-command domain of their antecedents do not violate Principle A. The dependent elements in (i) act as if they were
contained in A -moved phrases (10) (cf. Barss 1986: 108; Belletti & Rizzi 1988; Johnson
1985: 41ff., 1987, 1992; Pesetsky 1987; among others).
(i)
a.
b.
(11) accounts for the well-formedness of examples such as (10) and (i) without referring to
the question of whether or not the anaphor (or the element that contains the anaphor) has
undergone A- or A -movement (see also Note 9).
. Cinque (1990, Section 1.4.2) notes that scope reconstruction of wh-phrases across weak
islands is much more restricted than reconstruction for the purposes of the Binding Theory.
An explanation for this asymmetry could make use of the idea that in line with what I have
argued for in the preceding section i.e., that reconstruction at LF for Binding Theory simply
does not exist.
. A further argument which was proposed in the literature in favor of the intermediate adjunction hypothesis concerns the explanation of Superiority effects in terms of the
ECP. Assuming that traces have to be antecedent governed, Chomsky (1991, Footnote 24
and 34) assumes that traces are licensed (i.e. +-marked) at intermediate stages of a derivation (see Barss 1986: 447; Saito 1994a: 209f.). In (i) the trace of the raised subject satisfies
the ECP. Hence, the contrast in (i) is due to the fact that LF wh-movement of the adjunct
violates the ECP, in contrast to covert movement of the object, as can be seen from the LFrepresentations of (i) in (ii). Given that the object adjoins to VP, as shown in (ii-a), the initial
trace t j is antecedent-governed and therefore not *-marked. Although further movement of
the wh-phrase to Spec CP leaves an intermediate trace which is not c-commanded and gets
therefore *-marked, this trace is deleted and the resulting operator-variable pair does not
cause ungrammaticality. On the other hand, deletion of an intermediate trace is impossible
in the case of the LF-moved adjunct in (ii-b). Therefore (i-b) violates the ECP.
(i)
Joachim Sabel
relation, weak crossover effects are found with wh-movement, whereas they do not arise
with overt operator-movement in SOV languages like German, Japanese, and Lakhota.
. In my analysis of the Polish data, I have followed the suggestion made by Toman
(1981), Lasnik and Saito (1984), Rudin (1988), Cheng (1991), Billings and Rudin (1994),
Koizumi (1995), Bokovic (1996), Richards (1997), Stepanov (1998) among others that
the IP-adjoined position is an operator-position in Polish. Several further languages provide empirical evidence for the fact that the IP-adjoined position may be an operator position for wh-elements. For example, Raghibdoust (1994) makes the same claim for Persian,
and further evidence for an IP-adjoined operator-position is provided by wh-scrambling
in Japanese (see Nishiyama et al. 1996; Grewendorf & Sabel 1999; among others). Mahajan (1990) in his scrambling analysis of wh-movement in Hindi also argues for the IPadjunction analysis of wh-movement. Obviously, from a typologically point of view languages must be divided according to their use of different destinations for wh-elements in
the overt syntax. We find languages in which wh-elements end up in Spec CP (as in English,
German), in a position adjoined to IP (as in Hindi, Japanese, and Persian) or in both positions (as in Polish). A potential explanation could make use of an idea, entertained by Rizzi
(1990b, 1996) that (in addition to C0 ) Infl0 can also be base-generated with [wh] features.
. See McDaniel (1989: 585f.) for the same problem with intermediate adjunction to VP
imposed by partial wh-movement constructions.
. Note also that the following derivation with intermediate VP-adjunction is problematic
for an account of wh-island violations in the Barriers framework of Chomsky (1986a):
(i) *How do you t wonder [CP what [John could [VP t [VP t [VP fix t]]]]]?
The adjunct has adjoined two times to the embedded VP. We have to exclude that t and
t antecedent govern each other, which is the case if the notion m-command (cf. Chomsky
1986a) is relevant for antecedent government. The problem that is caused by a derivation
like (i) (which also arises in connection with the that-t-effect, as pointed out by Chomsky
1986a, see also Takahashi 1994: 117f. for relevant discussion) disappears if c-command is
the notion relevant for antecedent government. Then although no barrier intervenes between the trace t and t t does not c-command t , hence t is not antecedent governed
and violates the ECP at LF. Note, however, that the notions inclusion and exclusion are
still needed in frameworks which express structural relations in terms of strict c-command
(defined in terms of the first branching node (Reinhart 1983)), in order to guarantee that
head movement such as V-to-I or N-to-V leaves a commanded trace t v (see Baker 1988: 449,
Footnote 10; Chomsky & Lasnik 1993: 518, 522). Defining therefore c-command relations
as, c-commands if does not dominate (include) and every that dominates dominates , the problem posed by derivations such as (i) needs to be solved with an independent
constraint such as the one which prohibits intermediate adjunction making derivations such
as (i) impossible.
. That licensing of the bound variable reading of a pronoun is subject to similar constraints as anaphoric binding can also be seen from the fact that it is independent of the
A- or A -movement properties of the constituent containing the pronoun (cf. Note 8). As
shown in (i-b), in contrast to (30) and (i-a), the bound variable reading of a pronoun can
also be licensed when the pronoun has left the c-command domain of its binder as a result
of A-movement (Engdahl 1986; Koizumi 1992; Abe 1993: 311):
(i)
a.
b.
Which of hisi parents do you believe that every mani likes [t best]?
Itsi nose seems to every intelligent roboti [t to be ugly].
. The corresponding [wh] feature in the head of the attractor is also +Interpretable and
strong in English, triggering overt movement. Furthermore, weak +Interpretable features
need not be checked. Therefore wh-phrases in situ (for example in multiple questions in
English) or their [wh] features are not moved at LF. The properties of +Interpretable features correlate with the assumption that +Interpretable features are legitimate LF objects
that enter into interpretation. In contrast, Interpretable features (such as Case) need to be
checked in any event, and hence are eliminated at LF. Interpretable features on XPs immediately disappear after checking. This prohibits, for example, an NP from checking one and
the same feature more than one time.
. Note that Chomsky (1995, Chapter 4) allows for feature-checking via Merge in non-positions. This is relevant for expletive constructions and for the analogue of (32) with
whether. Like who in (33a) whether in (i) checks the [wh] feature of the embedded C0 . At a
later stage of the derivation whether is attracted by the matrix C-head (i). However, a trace
inside the embedded Spec CP position is not licensed (see the discussion in Note 26 below).
Again, the derivation in (ii) is not possible.
(i) *Whether you wonder t whether John could fix what?
(ii) *What do you wonder whether John could fix t ?
. Certain binding phenomena seem to provide evidence for object shift in German (see
Wyngaerd 1989; Mahajan 1990; Sabel 1996; among others).
. As discussed in Section 3.5 below the idea of hierarchically ordered features can possibly
be derived from the assumption that lexical elements are taken to be sequences of features
which have to be checked in a certain order, as suggested in Chomsky (1995: 195).
. Koizumi assumes that hierarchically ordered [Top] and [Neg] features ([Top] > [Neg])
in one and the same functional head trigger checking of two different elements (Topics and
Negative Phrases) in different specifier positions of one and the same projection.
. This also rules out the possibility that Wh1 moves from Spec1 to Spec2 . See also Reinhart
(1981) for suggestions on how to regulate the filling of multiple landing positions in Spec CP.
. It is commonly assumed that intermediate traces are [wh] elements i.e., non-operators.
For example, this assumption automatically excludes examples such as *Who do you wonder [t [t won the race]] in which the strong [wh] C0 head of the embedded C0 needs a
[wh] element in its specifier. However, given that intermediate traces are [wh] elements
the ungrammaticality of this sentence is expected. An additional empirical argument for the
assumption that traces are [wh] elements can be gained from sentences such as Who knows
[who [John saw t]]. As can be seen from this example the embedded [wh] C0 is checked by
a wh-element. However, this example cannot be understood as a matrix double question
which means that this wh-element cannot be interpreted in the matrix Spec CP position.
Joachim Sabel
Again, this results in a mismatch since the fact that a [wh] element occupies an embedded
Spec CP with a [wh] C0 head (for further discussion see Lasnik & Saito 1992; Rizzi 1996).
. The analysis outlined in the text provides the basis for an account for the fact that the
sentential complements of factive verbs are syntactic islands. One possibility is to adopt the
analysis in Melvold (1991) who argues that extraction from factive islands is blocked by
an empty operator in the Spec CP position of the factive complement (see also Watanabe
1993). According to Melvold (1991) island effects with respect to wh-extraction from factive
complements have to be explained in analogy to wh-island violations. A different possibility
arises from the analysis of factive islands in Cinque (1990: 30) who, in contrast to Melvold,
assumes that these complements are not generated as sisters of V. Then they are barriers
for extraction according to the idea that every non-complement is a barrier (Huang 1982;
Chomsky & Lasnik 1993; among others). Both analyses are compatible with the analysis
presented here (see Sabel 2002 for discussion).
. Violations of the CNPC such as (ii) provide a weaker island effect than complement
extraction from relative clauses as in (i), which show a very strong violation that is equal to
an ECP effect.
(i) *Which book did John have [NP a friend [CP to whom [to read *t]]]?
(ii) ?*Which book did you hear [NP a rumor [CP *t that John had read t ]]?
In contrast to (ii) no intermediate trace can be generated in (i) because the intermediate
Spec CP position is occupied by the relative pronoun. Note that this explanation is not undermined by the multiple specifier analysis proposed in this section. It is plausible to assume
that in contrast to embedded interrogative complements, the C-head of relative clauses may
not project a second specifier position in which the long-moved wh-element can create an
intermediate trace. The reason for this may well lie in different feature specifications of complementizer types of embedded [wh] C0 -heads and the C0 -heads of relative clauses (see, for
example, Rizzi 1990a, Chapter 2). Given the asymmetries between relativization and whmovement, the idea that relative clauses and wh-questions may have different C-systems is
independently justified. For example, we find multiple wh-movement but never multiple relativization. Relative pronouns may never occur in-situ in contrast to wh-elements; relative
pronouns in English license resumptive pronouns in contrast to interrogatives wh-phrases
(Safir 1986), that-t and weak crossover effects are absent in relativization, in contrast to
wh-movement (Chomsky 1981, 1982). Finally, Horvath (1986: 48ff.) compares both movement types in Hungarian, showing that the landing sites of relative pronouns differ from the
landing sites of interrogative wh-elements. This asymmetry is also found in Italian, where
according to Rizzi (1995) relative operators and wh-operators occupy different positions; i.e.
the former must precede topics, in contrast to question operators (see Brandon & Seki 1981;
Tajima 1987; Tajima & Arimura 1988; and Mller & Sternefeld 1993 for further differences
between wh-movement and relativization). To conclude, the explanation for the strong ungrammaticality of (i) relies on the idea that in contrast to (ii) no intermediate trace is created
in an A -position and the *-marked trace may not be deleted.
. The same conclusion is drawn in Hoekstra and Bennis (1989), discussing examples like
(44b) from Dutch, where scrambling out of finite clauses in contrast to sentence-internal
scrambling is impossible, as in German.
(OM=Object Marker)
Finally, (iii-b) derived from (iii-a) shows that Persian allows for long distance scrambling.
(iii) a.
b.
In Hindi, similar correlations are found, although it is controversial whether Hindi has multiple subject constructions of the sort found in Japanese and Persian (Mahajan 1990, p.c.).
. Several authors have argued that in contrast to full pro-drop languages such as Italian
and Spanish, languages such as Dutch and German represent semi pro-drop languages.
The latter do not allow referential pro-subjects but only empty expletive pronominal prosubjects. According to this view, the subject position in impersonal passive constructions
and in constructions with VP-internal subjects is occupied with an expletive pro that satisfies
the EPP (Extended Projection Principle). For more details of the analysis of semi pro-drop
see McKay (1985); Platzack (1985); Safir (1985a, 1985b); Koster (1986); Grewendorf (1989).
. According to Chomsky (1993), an element is in the checking domain of a head (X) if
(i) it is in a Spec head relation with X, or (ii) it is in a position adjoined to the head X, or
(iii) it is adjoined to the maximal projection of X, or (iv) it is adjoined to the Spec of X.
. A similar treatment of scrambling and quantifier raising gains support from the observation that scrambling traces and traces generated by quantifier raising behave in the same
way: they are not subject to principle C (cf. see Hornstein 1984; Aoun & Hornstein 1985;
Aoun & Li 1990, 1993; Nemoto 1993: 25; Saito 1994b).
. Note that an alternative derivation of example (52) is impossible according to which long
quantifier raising applies without intermediate adjunction. Assuming that scrambling in the
overt syntax parallels QR at LF, the features triggering scrambling and quantifier raising are
Joachim Sabel
associated with Agrs-features in the functional heads Infl0 (or v0 ) in the embedded and
matrix Infl0 (or v0 ), triggering obligatorily successive-cyclic movement.
. Given that every non complement is a barrier, we can translate the constraints on
parasitic gaps in terms of intervention of maximal projections.
. Fukui (1993) argues that the same problem arises with respect to the examples in (i).
After the intermediate trace is deleted, no Principle C violation should occur.
(i)
However, as already mentioned in Section 3.1.1, the examples in (i) are ruled for independent reasons i.e., because the infinitival requires a PRO subject which checks Null Case.
Given that John cannot bear Null Case, a feature mismatch arises (see Chomsky 1995: 326
for further discussion).
. A similar construction occurs as well in the case of multiple XP-movement. As is wellknown, multiple wh-fronting languages such as Bulgarian and Romanian allow for long
distance fronting of multiple wh-elements as in (i) from Romanian (Comorovski 1986):
(i)
promis
t2 t1?
ce1 ziceai
ca t 3 i-a
Cine3 cui2
to-him has-promised
who to-whom what you-were-saying that
Who did you say promised what to whom?
At first sight, (i) seems to pose a problem for the assumption that movement may not
proceed via intermediate adjunction. As noted in Rudin (1988), to derive a sentence like
(i) without a violation of subjacency it is necessary for more than one wh-phrase to pass
through the embedded Spec CP position, which means that Bulgarian and Romanian
must allow multiple (intermediate) wh-traces to be adjoined to Spec CP as in (ii) (Rudin
1988: 455):
(ii) [CP Whi Whj . . . [CP [SpecCP t i [t j ] ] . . . t i . . . t j . . .]]?
As argued in Ackema and Neeleman (1998), Sabel (1998), Grewendorf and Sabel (1999),
Sabel (2001), multiple wh-elements in Bulgarian and Romanian move as one single constituent successive-cyclically from Spec CP to Spec CP (leaving only one intermediate trace
in the embedded Spec CP position). See also Richards (1997) for an alternative analysis of
multiple fronting phenomena in terms of multiple specifiers.
References
Abe, J. (1993). Binding Conditions and Scrambling without A/A Distinction. Doctoral
dissertation, University of Connecticut.
Ackema, P. & A. Neeleman (1998). Optimal Questions. Natural Language and Linguistic
Theory, 16, 443490.
Agbayani, B. (1997). Category Raising, Adjunction and Minimality. UCI Working Papers in
Linguistics, 3, 125.
Joachim Sabel
Fukui, N. & M. Saito (1998). Order in Phrase Structure and Movement. Linguistic Inquiry,
29, 439474.
Georgopoulous, C. (1991). Canonical Government and the Spec Parameter: An ECP
account of weak crossover. Natural Language and Linguistic Theory, 9, 146.
Grewendorf, G. (1989). Ergativity in German. Dordrecht: Foris.
Grewendorf, G. & J. Sabel (1994). Long Scrambling and Incorporation. Linguistic Inquiry,
25, 263308.
Grewendorf, G. & J. Sabel (1999). Scrambling in German and Japanese: Adjunction versus
Multiple Specifiers. Natural Language and Linguistic Theory, 17, 165.
Grimshaw, J. (1991). Extended Projection. Ms., Brandeis University, Waltham MA.
Guron, J. & R. May (1984). Extraposition and Logical Form. Linguistic Inquiry, 15, 131.
Hendrick, R. & M. Rochemont (1988). Complementation, Multiple wh, and Echo
Questions. Toronto Working Papers in Linguistics, 9.
Hestvik, A. (1990). LF-Movement of Pronouns. Doctoral dissertation, Brandeis University.
Hoekstra, T. & H. Bennis (1989). A Representational Theory of Empty Categories. In
H. Bennis and A. van Kemenade (Eds.), Linguistics in the Netherlands 1989 (pp. 9199).
Dordrecht: Foris.
Hornstein, N. (1984). Logic as Grammar. Cambridge, MA: The MIT Press.
Hornstein, N. (1998). Movement and Chains. Syntax, 1, 99127.
Horvath, J. (1986). Focus in the Theory of Grammar and the Syntax of Hungarian. Dordrecht:
Foris.
Huang, C.-T. J. (1982). Logical Relations in Chinese and the Theory of Grammar. Doctoral
dissertation, MIT.
Huang, C.-T. J. (1993). Reconstruction and the Structure of VP: Some theoretical
consequences. Linguistic Inquiry, 24, 103138.
Jaeggli, O. (1988). ECP Effects at LF in Spanish. In D. Birdsong and J.-P. Montreuil (Eds.),
Advances in Linguistics (pp. 113149). Dordrecht: Foris.
Johnson, K. (1985). A Case for Movement. Doctoral dissertation, MIT.
Johnson, K. (1987). Against the Notion SUBJECT. Linguistic Inquiry, 18, 354361.
Johnson, K. (1988). Clausal Gerunds, the ECP, and Government. Linguistic Inquiry, 19,
583609.
Johnson, K. (1992). Scope and Binding Theory: Comments on Zubizarreta. In T. Stowell
and E. Wehrli (Eds.), Syntax and the Lexicon. Syntax and Semantics 26 (pp. 259275).
New York: Academic Press.
Kayne, R. (1989). Facets of Romance Past Participle Agreement. In P. Beninc (Ed.), Dialect
Variation and the Theory of Grammar (pp. 85103). Dordrecht: Foris.
Kayne, R. (1994). The Antisymmetry of Syntax. Cambridge, MA: The MIT Press.
Kiss, K. (1987). Configurationality in Hungarian. Budapest: Akademiai Kiado; Dordrecht:
Reidel.
Koizumi, M. (1992). Copy and Reconstruction Effects. Ms., MIT.
Koizumi, M. (1994). Layered Specifiers. Proceedings of the North Eastern Linguistics Society
(Nels), 24, 255269.
Koizumi, M. (1995). Phrase Structure in Minimalist Syntax. Doctoral dissertation, MIT.
Koopman, H. & D. Sportiche (1982). Variables and the Bijection Principle. The Linguistic
Review, 2, 139160.
Joachim Sabel
Koopman, H. & D. Sportiche (1985). -Theory and Extraction. Glow Newsletter, 14, 5758.
Koopman, H. & D. Sportiche (1986). A Note on Long Extraction in Vata and the ECP.
Natural Language and Linguistic Theory, 4, 357374.
Koster, J. (1986). The Relation between pro-drop, Scrambling, and Verb Movements. Ms.,
Rijksuniversiteit Groningen.
Koster, J. (1987). Domains and Dynasties. The radical autonomy of syntax. Dordrecht: Foris.
Kroch, A. & A. K. Joshi (1987). Analyzing Extraposition in a Tree Adjoining Grammar. In
G. Huck and A. Ojeda (Eds.), Discontinuous Constituents. Syntax and Semantics, 20
(pp. 107149). New York: Academic Press.
Kuroda, S.-Y. (1992). Whether We Agree or Not: A comparative syntax of English
and Japanese. In S.-Y. Kuroda (Ed.), Japanese Syntax and Semantics (pp. 215257).
Dordrecht: Kluwer.
Lasnik, H. (1995). Verbal Morphology: Syntactic Structures meets the Minimalist Program.
In H. Campos and P. Kempchinsky (Eds.), Evolution and Revolution in Linguistic Theory
(pp. 251275). Washington DC: Georgetown, University Press.
Lasnik, H. (1998). Some Reconstruction Riddles. Proceedings of the 22nd Annual Penn
Linguistics Colloquium. Volume, 5 (1), 8398.
Lasnik, H. (1999). Chains of Arguments. In S. D. Epstein and N. Hornstein (Eds.), Working
Minimalism (pp. 188215). Cambridge, MA: The MIT Press.
Lasnik, H. & M. Saito (1984). On the Nature of Proper Government. Linguistic Inquiry, 15,
235289.
Lasnik, H. & M. Saito (1992). Move . Cambridge, MA: The MIT Press.
Lebeaux, D. (1983). A Distributional Difference between Reciprocals and Reflexives.
Linguistic Inquiry, 14, 723730.
Lebeaux, D. (1985). Locality and Anaphoric Binding. Linguistic Inquiry, 16, 343363.
Lebeaux, D. (1991). Relative Clauses, Licensing, and the Nature of the Derivation. In S. D.
Rothstein (Ed.), Perspectives on Phrase Structure. Syntax and Semantics, 25 (pp. 209
239). New York: Academic Press.
Lee, E.-J. (1993). Superiority Effects and Adjunct Traces. Linguistic Inquiry, 24, 177183.
Lee, R. K. (1994). Economy of Representation. Doctoral dissertation, University of
Connecticut.
Li, Y.-H. A. (1990). X -Binding and Verb Incorporation. Linguistic Inquiry, 3, 399426.
Lightfoot, D. & Weinberg, A. (1988). Barriers (A Review). Language, 64, 366383.
Longobardi, G. (1987). In Defense of the Correspondence Hypothesis. Ms.
Mahajan, A. (1990). The A/A-bar Distinction and Movement Theory. Doctoral dissertation,
MIT.
Maling, J. (1979). An Asymmetry with Respect to Wh-Islands. Linguistic Inquiry, 9, 7589.
Manzini, R. (1998). A Minimalist Theory of Weak Islands. In P. W. Culicover and L. McNally
(Eds.), The Limits of Syntax [Syntax and Semantics 29] (pp. 185209). New York:
Academic Press.
May, R. (1977). The Grammar of Quantification. Doctoral dissertation, MIT.
May, R. (1985). Logical Form. Cambridge, MA: The MIT Press.
McDaniel D. (1989). Partial and Multiple Wh-movement. Natural Language and Linguistic
Theory, 7, 565604.
McKay, T. (1985). Infinitival Complements in German. Cambridge: CUP.
Melvold, J. (1991). Factivity and Definiteness. MIT Working Papers in Linguistics, 15, 97117.
Miyagawa, S. (1997). Against Optional Scrambling. Linguistic Inquiry, 28, 125.
Mulders, I. (1997). Mirrored Specifiers. Linguistics in the Netherlands (pp. 135146).
Amsterdam: John Benjamins.
Mller, G. & W. Sternefeld (1993). Improper Movement and Unambiguous Binding.
Linguistic Inquiry, 24, 461507.
Nakajima, H. (1989). Bounding of Rightwards Movements. Linguistic Inquiry, 20, 328334.
Nemoto, N. (1993). Chains and Case Positions: A study from scrambling in Japanese.
Doctoral dissertation, University of Connecticut.
Nishiyama K. et al. (1996). Syntactic Movement of Overt Wh-Phrases in Japanese and
Korean. Japanese/Korean Linguistics, 5, 337351.
Pesetsky, D. (1982). Paths and categories. Doctoral dissertation, MIT.
Pesetsky, D. (1987). Binding Problems with Experiencer Verbs. Linguistic Inquiry, 18, 126
140.
Platzack, C. (1985). The Scandinavian Languages and the Null Subject Parameter. Working
Papers in Scandinavian Syntax, 20.
Plessis, H. du (1977). Wh-Movement in Afrikaans. Linguistic Inquiry, 8, 723726.
Raghibdoust, S. (1994). Multiple Wh-Fronting in Persian. Cahiers Linguistiques DOttawa,
21, 2758.
Raposo, E. (1987). Romance Inversion, the Minimality Condition and the ECP. Proceedings
of the North East Linguistic Society (Nels), 18, 357374.
Reinhart, T. (1981). A Second Comp Position. In A. Belletti, L. Brandi, L. Rizzi (Eds.), Theory
of Markedness in Generative Grammar (pp. 517557). Pisa: Scuola Normale Superiore
di Pisa.
Reinhart, T. (1983). Anaphora and Semantic Interpretation. London: Croom Helm.
Reuland, E. & W. Kosmeijer (1988). Projecting Inflected Verbs. Groninger Arbeiten zur
Germanistischen Linguistik GAGL, 29, 88113.
Richards, N. (1997). What moves where when in which language? Doctoral dissertation,
MIT.
Rizzi, L. (1982). Issues in Italian Syntax. Dordrecht: Foris.
Rizzi, L. (1986). Null Objects in Italian and the Theory of pro. Linguistic Inquiry, 17, 501
557.
Rizzi, L. (1990a). Relativized Minimality. Cambridge, MA: The MIT Press.
Rizzi, L. (1990b). Speculations on Verb Second. In J. Mascar and M. Nespor (Eds.),
Grammar in Progress (pp. 2532). Dordrecht: Foris.
Rizzi, L. (1992). Argument/Adjunct (A)symmetries. Proceedings of the North East Linguistic
Society (Nels), 22, 365381.
Rizzi, L. (1995). The Fine Structure of the Left Periphery. Ms., Universit de Genve.
Rizzi, L. (1996). Residual Verb Second and the Wh-Criterion. In A. Belletti and L. Rizzi
(Eds.), Parameters and Functional Heads. Essays in comparative syntax (pp. 6390).
Oxford: OUP.
Rudin, C. (1988). On Multiple Questions and Multiple Wh-Fronting. Natural Language and
Linguistic Theory, 6, 445501.
Joachim Sabel
Sabel, J. (1995). On Parallels and Differences between Clitic Climbing and Long Scrambling
& the Economy of Derivations. Proceedings of the North East Linguistic Society (Nels),
25, 405423.
Sabel, J. (1996). Restrukturierung und Lokalitt. Universelle Beschrnkungen fr Wortstellungsvarianten. Berlin: Akademie-Verlag.
Sabel, J. (1998). Principles and Parameters of Wh-Movement. Habilitations-Thesis,
Universitt Frankfurt/Main.
Sabel, J. (1999). Das Passiv im Deutcshen. Derivationale konomie vs. optionale Bewegung.
Linguistische Berichte, 177, 87112.
Sabel, J. (2000). Partial Wh-Movement and Typology of Wh-Questions. In U. Lutz et al.
(Eds.), Wh-Scope Marking (pp. 409446). Amsterdam: John Benjamins.
Sabel, J. (2001). Deriving Multiple Head and Phrasal Movement: The Cluster Hypothesis.
Linguistic Inquiry, 32 (3), 532547.
Sabel, J. (2002). A Minimalist Analysis of Syntactic Islands. The Linguistic Review, 19.
Safir, K. (1985a). Syntactic Chains. Cambridge: CUP.
Safir, K. (1985b). Missing Subjects in German. In J. Toman (Ed.), Studies on German
Grammar (pp. 193229). Dordrecht: Foris.
Safir, K. (1986). Relative Clauses in a Theory of Binding and Levels. Linguistic Inquiry, 17,
663689.
Saito, M. (1985). Some Asymmetries in Japanese and Their Theoretical Implications.
Doctoral dissertation, MIT.
Saito, M. (1992). Long Distance Scrambling in Japanese. Journal of East Asian Linguistics, 1,
69118.
Saito, M. (1994a). Additional-wh Effects and the Adjunction Site Theory. Journal of East
Asian Linguistics, 3, 195240.
Saito, M. (1994b). Improper Adjunction. In M. Koizumi and H. Ura (Eds.), Formal
Approaches to Japanese Linguistics I: MIT Working Papers in Linguisitcs 24 (pp. 263293).
Department of Linguistics and Philosophy, MIT, Cambridge, MA.
Saito, M. (1994c). Scrambling and the Functional Interpretation of Wh-Phrases. Ms.,
University of Connecticut.
Saito, M. & Hoshi, H. (2000). The Japanese Light Verb Construction and The Minimalist
Program. In R. Martin et al. (Eds.), Step by Step. Essays on Minimalist Syntax in Honor
of Howard Lasnik (pp. 261295). Cambridge, MA: MIT Press.
Sakai, H. (1994). Derivational Economy in Long Distance Scrambling. In M. Koizumi
and H. Ura (Eds.), Formal Approaches to Japanese Linguistics I: MIT Working Papers
in Linguisitcs, 24 (pp. 295314). Department of Linguistics and Philosophy, MIT,
Cambridge, MA.
Stepanov, A. (1998). On Wh-Fronting in Russian. Proceedings of the North East Linguistic
Society (Nels), 28.
Tajima, K. (1987). Wh-Q/Wh-Rel Asymmetries and Conditions on A -Chains. Proceedings
of CLS, 23, 336349.
Tajima, K. & Arimura, K. (1988). Two Types of Variables: A D-Structure Adjunction
Approach to Null Operator Constructions. Proceedings of CLS, 24, 362376.
Takahashi, D. (1994). Minimality of Movement. Dissertation, University of Connecticut.
Storrs, Connecticut.
Index
A
absorption 117
adjunction 111f.
with multiple roots 113f., 116, 118
Agree 48, 50, 54, 56, 165, 166, 168, 172,
241
antifreezing 86f.
Attract 179
C
case 167, 169, 173, 175, 182, 185, 237f.,
242f.
categorial identity condition 114
change of location verbs 143, 148f.
change of state verbs 142, 150f.
conflation 211ff.
constituent 120f., 126, 134
c-command 109f.
Constraint on Adjunction Movement
260, 277
copy and deletion 82f.
co-valued features 165, 172, 188f., 193f.,
198f.
locality of 200, 201
D
Distributed Morphology 15, 25
dominance 110
E
EPP 168, 186, 187, 203, 249f., 254f., 305
existential locative constructions 226f.
expletive constructions 188f.
F
floating quantifiers 176ff.
G
Global economy 180f.
L
Languages
Basque 157, 244, 245f.
Catalan 224, 227
Chamorro 1331
Croatian 66ff., 72ff., 80ff., 99f.
Dutch 224, 294f.
Finnish 34
French 34, 135, 192f.
Georgian 35
German 65ff., 73ff., 79ff., 90ff.,
191f., 263, 275, 279, 283, 285
Haitian Creole 244
Hebrew 18, 32
Hindi 35, 157
Icelandic 7f., 176, 196f., 251, 282
Irish 35
Italian 21, 22, 23, 27, 29, 30, 31ff.,
35, 279
Japanese 77, 247, 262, 284f.
Kiswahili 168f.
Polish 66, 273f.
Russian 34, 225, 284
Sardinian 228
Scottish Gaelic 35
Southern Tiwa 245
Spanish 167, 177f., 212ff., 243, 247,
254, 279, 293f.
Swedish 252f.
Turkish 150, 153, 158
locality of Agreement 178
long scrambling 283
Index
M
Minimal Link Condition 271
monadic verbs 140
unaccusative vs. unergative
Move 179, 181
Move F 46, 47, 49
139f.
O
object experiencer verbs 19f., 29f.
stative and non-stative reading
21, 22, 23
20,
P
Parallel Movement Constraint 89
Phase Impenetrability Condition 167,
202, 259
Person-Case Constraint 243f.
Principle A reconstruction effects 261f.
Proper Binding Condition 42, 45
Q
quantifier raising 42, 289f.
R
Relativized Minimality 41, 51f., 60, 198,
265
remnant movement 41, 42, 71
root node 111, 112f.
roots in distributed morphology 15ff.
S
Scandinavian object shift 250f.
scope reconstruction 267f.
Single-Output syntax 48
Specific Subject Condition 73
Split ergativity 156f.
subject experiencer verbs 33, 34, 35, 36
substitution with multiple roots 113,
115
superraising 199
T
transitivity alternation 141
U
Unambiguous Domination 76
Uniformity Condition on Chains 281,
292
Uniformity Principle 239
V
verbs of being 142, 145f.
verbs of creation 143, 153f.
W
weak crossover 263
X
XP-split construction 66f.
27. RUZICKA,
Rudolf: Control in Grammar and Pragmatics. A cross-linguistic study.
1999.
28. HERMANS, Ben and Marc van OOSTENDORP (eds): The Derivational Residue in
Phonological Optimality Theory. 1999.
29. MIYAMOTO, Tadao: The Light Verb Construction in Japanese. The role of the verbal
noun. 1999.
30. BEUKEMA, Frits and Marcel den DIKKEN (eds): Clitic Phenomena in European
Languages. 2000.
31. SVENONIUS, Peter (ed.): The Derivation of VO and OV. 2000.
32. ALEXIADOU, Artemis, Paul LAW, Andr MEINUNGER and Chris WILDER (eds):
The Syntax of Relative Clauses. 2000.
33. PUSKS, Genoveva: Word Order in Hungarian. The syntax of -positions. 2000.
34. REULAND, Eric (ed.): Arguments and Case. Explaining Burzios Generalization.
2000.
35. HRARSDTTIR, Thorbjrg. Word Order Change in Icelandic. From OV to VO.
2000.
36. GERLACH, Birgit and Janet GRIJZENHOUT (eds): Clitics in Phonology, Morphology
and Syntax. 2000.
37. LUTZ, Uli, Gereon MLLER and Arnim von STECHOW (eds): Wh-Scope Marking.
2000.
38. MEINUNGER, Andr: Syntactic Aspects of Topic and Comment. 2000.
39. GELDEREN, Elly van: A History of English Reflexive Pronouns. Person, Self, and
Interpretability. 2000.
40. HOEKSEMA, Jack, Hotze RULLMANN, Victor SANCHEZ-VALENCIA and Ton
van der WOUDEN (eds): Perspectives on Negation and Polarity Items. 2001.
41. ZELLER, Jochen : Particle Verbs and Local Domains. 2001.
42. ALEXIADOU, Artemis : Functional Structure in Nominals. Nominalization and
ergativity. 2001.
43. FEATHERSTON, Sam: Empty Categories in Sentence Processing. 2001.
44. TAYLAN, Eser E. (ed.): The Verb in Turkish. 2002.
45. ABRAHAM, Werner and C. Jan-Wouter ZWART (eds): Issues in Formal German(ic)
Typology. 2002.
46. PANAGIOTIDIS, Phoevos: Pronouns, Clitics and Empty Nouns. Pronominality and
licensing in syntax. 2002.
47. BARBIERS, Sjef, Frits BEUKEMA and Wim van der WURFF (eds): Modality and its
Interaction with the Verbal System. 2002.
48. ALEXIADOU, Artemis, Elena ANAGNOSTOPOULOU, Sjef BARBIERS and Hans
Martin GAERTNER (eds): Dimensions of Movement. From features to remnants.
n.y.p.
49. ALEXIADOU, Artemis (ed.): Theoretical Approaches to Universals. 2002.
50. STEINBACH, Markus: Middle Voice. A comparative study in the syntax-semantics
interface of German. 2002.
51. GERLACH, Birgit: Clitics between Syntax and Lexicon. n.y.p.
52. SIMON, Horst J. and Heike WIESE (eds): Pronouns. Grammar and representation.
n.y.p.