TABLE OF CONTENTS

PROLOGUE 1

PART I. ELEMENTS OF SYNTAX

CHAPTER 1. ELEMENTS OF SYNTAX 11

1.1 Instructions 11

1.2 Merge: nests 16

1.3 Merge: internal and external 22

1.4 Onset 27



PART II. PATTERNS

CHAPTER 2. THE RELATIONSHIP BETWEEN COMPLEMENTIZERS
AND INFLECTIONAL HEADS 31

2.1 The C-Infl link 31

2.1.1-features 32

2.1.2 Tense 36

2.1.3 Mood 40

2.1.4 Modality 44

2.1.5 Negation 48

2.2 Conclusion 55

THE EMERGENCE OF ORDER IN SYNTAX vi
CHAPTER 3. DISCONTINUOUS SYNTACTIC PATTERNS 57

3.1 The source of the C-Infl link 57

3.2 Probe-goal relations in a phase 57

3.2.1 Subextraction (I): subject islands and -features 58

3.2.2 The Generalized Feature Inheritance Theory 61

3.3 Revising the Generalized Feature Inheritance Theory 65

3.3.1 The status of the phase head C and of Infl heads 66

3.3.1.1 Infl 66

3.3.1.2 C 67

3.3.2 Revising the Phase-Impenetrability Condition 67

3.3.2.1 Cyclicity 67

3.3.2.1.1 The classical empirical argument for strict cyclicity 69

3.3.2.1.2 Successive cyclicity 76

3.4 Discontinuities 83

3.5 Subextraction (II): relativized opacity for probe-goal relations 89



CHAPTER 4. ANALYTIC SYNTACTIC PATTERNS 95

4.1 On cartographies 95

4.2 The nature of ordering restrictions 99

4.2.1 Some allegedly non-primitive order restrictions 100

4.2.1.1 Epistemic modality and tense 100

4.2.1.2 Tense and aspect 103
TABLE OF CONTENTS vii
4.2.2 Some allegedly primitive order restrictions 105

4.3 Toward a principled account for some order restrictions 106

4.3.1 Voice and tense 107

4.3.2 Quantificational aspectual adverbs and completive/prospective
aspectual adverbs 109

4.3.3 Evidential mood and epistemic modality 110

4.4 The Full Interpretation Principle 112

4.5 Cartographic effects 113



CHAPTER 5. SYNCRETIC SYNTACTIC PATTERNS 115

5.1 On structural minimization 115

5.2 Why and where V moves 119

5.3 A side-effect of V-to-T movement 134

5.3.1 SPEC-T becomes an A-position when V has moved to T 135

5.3.1.1 Preverbal subjects in NSLs 135

5.3.1.2 Postverbal subjects in NSLs 136

5.3.1.3 Clitic Left dislocated elements in NSLs 138

5.3.1.4 Both preverbal subjects and CLLD elements match a-features
in SPEC-T in NSLs 140

5.3.2 The source of a-features associated with SPEC-T when V instantiates
-features on T 142

5.3.2.1 Subextraction (III): phase sliding 142

5.3.2.2 Maximize Matching Effects Principle 157

THE EMERGENCE OF ORDER IN SYNTAX viii
5.3.3 When SPEC-T is created in NSLs 158

5.4 Other conundrums to be solved on the basis of structural minimization 160

5.4.1 Double agreement dialects 160

5.4.2 From the Vacuous Movement Hypothesis to syntactic syncretism 170

5.5 Contraction 179

5.6 Conclusion 182



PART III. CONCLUSION

CHAPTER 6. ON THE EMERGENCE OF ORDER IN SYNTAX 189

REFERENCES 195

SUBJECT INDEX 208

LANGUAGE INDEX 211







PROLOGUE


The main idea of this study can be expressed in a few words: the syntactic
component of the faculty of language is responsible for ordering categories and for
ordering categories only. This would be a completely uninteresting thought, a
truism, if one did not attempt to account for how and why the attested patterns
emerge from the external requirements that the syntactic component has to satisfy.
Although saying that syntax is responsible for order does no more than to
express the etymological meaning of the word syntax, the use of the term order
in this study may deserve some attention, since it does not subscribe to the com-
mon use in grammatical studies. Throughout the text, the term order does not
exclusively refer to the literal precedence relation among terminals (this is the
common use of the term in grammar), but rather to the hierarchical properties that
are attributed to syntactic representations, as will later become clear. Literal pre-
cedence is mapped from hierarchy. Thus, the object of inquiry of the discipline
called syntax is how categories are ordered or how hierarchies are generated.
More precisely, this study poses two questions: what are the basic elements of
the syntactic component? and why do syntactic patterns have the shape they seem
to have? The first question is addressed in Part I and the second question in Part II.
Part III summarizes the conclusions of the preceding two parts and discusses the
possibility that Universal Grammar (Chomskys Factor II) is a rewiring of
elements that are in place independently (Chomskys Factor III).
Chapter 1 suggests that the basic elements of the syntactic component are
features and a combinatorial operation known as Merge. A feature is defined as an
instruction for a particular level of interpretation of the faculty of language and
Merge is defined as an operation that takes as input two categories or sets included
in an alphabet and yields as output the union of these two sets. The specific in-
structions that functional categories provide for the several levels of in-terpretation
are described. It is argued that the hierarchical properties of syntactic objects
derive from a derivational record, a set K (a nest) where the outputs of successive
Merge operations are linearly ordered by the strict inclusion relation. Conse-
quently, Kaynes Linear Correspondence Axiom is no longer an axiom one needs
to postulate to account for the X-theory; hierarchy is a product of creating struc-
THE EMERGENCE OF ORDER IN SYNTAX 2
ture successively and keeping the derivational information in a record. Appli-
cations of both internal and external Merge have been argued to be triggered by
the requirement of matching [+type] categories and [+token] categories, without
postulating any special device for the property of displacement; it has also been
argued that suicidal greed, a device postulated to account for movement speci-
fically, shows up three problems: the problem of generality, the problem of deter-
minacy and the problem of consistency.
Part II (chapters 2, 3, 4 and 5) discusses the particular shape of syntactic
objects by considering how syntactic derivations carried out by Merge are trig-
gered by the morpholexical characterization of lexical items (clusters of features)
and constrained by the Full Interpretation semantic legibility condition and by the
Maximize Matching Effects principle of structural minimization.
I argue that there are three types of syntactic patterns: discontinuous, analytic
and syncretic. That is, it is possible for a feature to be assigned to two projections,
for there to be a one-to-one relation between features and projections, and for more
than one feature to be assigned to one projection. Thus the terms discontinuous,
analytic and syncretic are relative to how features are associated with pro-
jections, or to how features are coded in the spine of a tree. It may be worth
stressing that the use of the terms analytic and syncretic in this study differs
from the use of the terms analytic and synthetic in morphological typology,
which are relative to how features are associated with words. This is also true for
the term discontinuous, commonly relative to the morpholexical level. In a very
illuminating work I became aware of a few days before writing this prologue,
Huang (2005, 2006) explores the idea that multiple parameters such as the head-
parameter, the pro-drop parameter, the wh-parameter, the telicity parameter and
the configurationality parameter may derive from the distinction between analytic
and synthetic languages in the abovementioned traditional sense, a distinction that
Huang conceives in terms of an analytic vs. synthetic macro-parameter.
Chapter 2 illustrates the relationship between C and Infl. The so-called C-Infl
link can be materialized in the following semantic and morphosyntactic properties
and generalized to several Infl-like features such as -features, tense, mood, mo-
dality and negation.

Property (I): complementizers can replicate Infl-like features

Property (II): there is a correlation between the characteristics of features sur-
facing on Infl and the choice of C

Property (III): Infl-like features are involved in triggering V-to-C movement.
PROLOGUE 3
Chapter 3 discusses the source of the C-Infl link and concludes that C and Infl
are polarities of the same feature ([clause typing]) and that Infl-like features are
assigned to both polarities, directly accounting for the three properties of the C-
Infl link. It is argued that the reason for inserting Infl-like features to both
polarities is that the semantic instruction provided by Infl-like features (to trigger
referential displacement) is orthogonal to the [clause typing] distinction. The -V
link and the P-K link are also analyzed as two syntactic discontinuities. The pro-
posal that C and Infl constitute a discontinuous syntactic object is preceded by a
relatively intricate evaluation of a plausibility argument: Chomskys (2005b)
Feature Inheritance Theory, a subcomponent of the theory of phases, which can be
generalized to Infl-like features. Consider the Phase-Impenetrability Condition.

Phase-Impenetrability Condition (PIC)
Consider a Phase PH = [o, [H, ]], H being the phase head
Call o and H the edge of PH, and the domain of H
The domain of H is not accessible to syntactic operations beyond PH, only the
edge, {o, H}, since Transfer sends to C-I and A-P once H has terminated its
work (adapted from Chomsky 2001b: 5-6).

Briefly, if the characteristics of Infl-like features depend on the choice of C
(semantic property II) and Infl-like features can surface and be syntactically
operative not only in Infl but also in C (morphosyntactic properties I and III), and
it can be independently argued that there is an important asymmetry as to the role
that C and T/Infl play in syntactic derivations, then it becomes interesting to con-
sider the possibility that C, the head responsible for defining the relatively com-
plete fragments of structure to be transferred from the syntactic workspace to the
external systems, is precisely the locus where Infl-like features are base-generated,
appearing only derivatively on Infl heads due to inheritance mecha-nisms. Con-
sider the following sketch of the Generalized Feature Inheritance Theory, which is
based on Chomskys recent insights.

Generalized Feature Inheritance Theory (GFIT)
(i) C has a central role in the generation of syntactic objects
(ii) Infl does not have a central role in the generation of syntactic objects
(iii) Infl-like features are generated on C
(iv) In the lexicon, Infl lacks Infl-like features
(v) Infl-like features surface on Infl only derivatively
THE EMERGENCE OF ORDER IN SYNTAX 4
(vi) The feature inheritance mechanism has the function of bringing semantic
distinctions into a syntactic representation

As will be shown below, the classical empirical argument for strict cyclicity is
no longer valid, due to parallel probing at the phase level. And after thoroughly
revising the Phase-Impenetrability Condition, it must be concluded that this con-
dition is both too weak and too strong. It is too weak because penetrating into the
edge of H is as problematic as penetrating into the domain of H (as Chomskys
refinement of Huangs subject islands shows) and because not only A-movement
but also A-movement is successive cyclic. It is too strong because probing into the
domain of H is possible when there is no intervention effect, as in experiencer
constructions in languages like Icelandic and Catalan. In order to account for sub-
ject islands, wh-islands and the requirement that both A-movement and A-mo-
vement are successive cyclic, the Phase-Impenetrability Condition has been re-
placed by the Relativized Opacity Principle.

Relativized Opacity Principle
In a syntactic object [
o1P
[o
1
[
o2P
[o
2
] ] ] ],
where: (i) o
1
and o
2
are two probes of the same type o each projecting an
aP
(ii) is SPEC-o
2
and is o
2
-COMPL, and
(iii) : is a constituent of and y: y is a constituent of ,
o
1
can probe or if they provide a suitable token for o
1
, but it cannot readily
probe y or

The factor that determines whether o
1
can readily probe a goal is the relative
depth of such a goal in the o
2
-projection whose label is of the same type as the
searching probe o
1
. o
2
P does not render the complete and opaque to o
1
, but
search in them becomes difficult. For this reason, long A-movement must use in-
termediate SPEC-T
def
as an escape hatch and long A-movement must use inter-
mediate SPEC-Cs and SPEC-s to attain its final position. The source of subject
islands and wh-islands is the same: a probe o fails to search too deep into the
SPEC or the COMPL of a lower o-projection.
Therefore, it seems that there is no clear asymmetry between C and T/Infl, and
thus the Generalized Feature Inheritance Theory loses plausibility. Unless further
arguments are provided, the simplest account for the C-Infl link is that Infl-like
PROLOGUE 5
features are merged in both poles of the C-Infl discontinuity without lowering or
raising mechanisms.
Chapter 4 considers the idea that natural languages privilege invariable and
richly articulated representations with semantically devoted positions (carto-
graphies or analytic syntactic patterns). I begin by revising Cinques (1999a) pre-
liminary position on how cartographies are related to semantic constraints. Al-
though Cinques arguments in this respect may be scarce and too weak, it is clear
that the empirical results of his work on clausal structure pose an interesting
problem for the syntax-semantic interface. I propose that several distributional res-
trictions derive from the Prohibition Against Tangled Structures (relative to dis-
continuous patterns) and the Prohibition Against Vacuous Quantification.

Prohibition Against Tangled Structures (PATS)
Two discontinuous syntactic templates [A
1
A
2
], [B
1
B
2
] cannot be broken
off by yielding the tangled object [A
1
[B
1
A
1
] B
2
]

Prohibition Against Vacuous Quantification (PAVQ)
For every occurrence of a variable x there must be a quantifier Q binding x,
and for every quantifier Q there must be a variable x such that Q binds x

A syntactic object that violates the PAVQ contains elements that are not useful
to the Conceptual-Intensional system, namely a free variable or a quantifier that
does not bind any variable. Similarly, a discontinuous template that is broken off
automatically becomes useless. Thus, the PATS and the PAVQ are two instances
of the more general Full Interpretation Condition.

Full Interpretation Condition
An object generated by a grammatical component must be constituted only of
useful features for a particular level of interpretation when it attains such a
level of interpretation

The partial considerations that have just been summarized suggest the possi-
bility that cartographies do not constitute a primitive component of Universal
Grammar but a cluster of properties emerging from legibility conditions.
I note that discontinuous syntactic objects do not contradict the idea that the
C-I system requires articulated syntactic representations with semantically devoted
positions, but rather they adhere to it: when an Infl-like feature is introduced into
THE EMERGENCE OF ORDER IN SYNTAX 6
the two poles of the C-Infl discontinuity, both occurrences of this Infl-like feature
are minimally distinguished, and thus they are not the same element in two posi-
tions but two co-variations. Discontinuous syntactic objects are a subtype of ana-
lytic syntactic objects, or cartographic effects.
Chapter 5 argues that several intricate phenomena can be better understood if a
less rigid view on how features are coded in a syntactic representation is adopted.
More precisely, it has been affirmed that syncretic syntactic patterns (or anticarto-
graphic effects) also exist: they are favored by the Maximize Matching Effects
Principle, a principle of structural minimization.

Maximize Matching Effects Principle (MMEP)
Match as many features as possible using the smallest span of structure

The MMEP does not apply arbitrarily, but under special circumstances favor-
ing structural condensation to the detriment of the C-I legibility requirement of
creating analytic syntactic patterns and discontinuous syntactic patterns. Consider
the following paragraphs as a simplified outline of the rather intricate course de-
veloped in this chapter.
If a richly inflected verb matches -features on T, there is no reason to claim
that A-movement is necessary. If it is not necessary, it is impossible, under effi-
ciency considerations. In this circumstance, SPEC-T is available to match other
features, and it should not be used to match -features. This accounts for the A-
status of preverbal subjects in Null Subject Languages: matching -features with
V-movement renders SPEC-T as an A-position, to be filled by topical/referential
subjects or by (topical/referential) non-subject CLLD elements (Sol 1992).
There is no reason to think that preverbal subjects in V2 languages have
moved to a peripheral position. Thus, when the subject is the first constituent, the
inflected V should be on T, not on C. This is substantiated by Zwarts (1993) evi-
dence from double agreement dialects: V shows up with the agreement mark cha-
racteristic of complementizers in inversion contexts and a different mark when the
first constituent is a subject; the mark in this second case is the same that V shows
up with when it remains inside the VP (that is, in embedded clauses headed by a
complementizer). Call the former type of agreement C-agreement and the latter
Infl-agreement. Thus, the question is why V does not raise to C to check tense
when the first constituent is a subject, yielding an inversion pattern. The proposed
answer is, again, expressed in terms of the MMEP: syntax proceeds in an op-
portunistic way taking advantage of the discontinuous nature of tense; if tense is
matched in TP as -features, all the computations take place in the same projection
PROLOGUE 7
and CP is not initiated to match a feature that, due to its discontinuous nature, can
be indistinctively matched in C or in T.
In English, wh-island effects are weaker when the embedded wh-item is nomi-
native (Chomsky 1986b, Chung & McCloskey 1983). This observation suggests
that the nominative interrogative is in SPEC-T and that the SPEC-C of the embed-
ded clause is available as an escape hatch. Similarly, in Dutch, focal and topical
2SG subjects are followed by a V with Infl-agreement, not with C-agreement;
therefore, V never raises to C in Dutch when the first constituent is a 2SG subject,
regardless of whether such a subject has peripheral features or not; this suggests
that focal/topical subjects in Dutch are also in SPEC-T, as English nominative
interrogatives. In these cases, SPEC-T becomes a mixed A/A-position. The
special circumstance favoring the application of the MMEP is that both -features
and a-features (wh-feature and topic/focus-feature) are matched by the same
category.
This chapter closes with a discussion of comp-trace effects in English. If
SPEC-T is the position where wh-features and -features are matched by a nomi-
native interrogative, then SPEC-T may also be the position where the defective
peripheral features triggering movement to intermediate positions are located. If
both peripheral features and non-peripheral features are located in TP, there is no
reason anymore to maintain the C-T distinction: TP has become a [clause typing]
category. Therefore, the complementizer cannot be inserted when the wh-phrase
undergoing long-movement is nominative because there is no structural space.
This reasoning also accounts for the more complicated patterns of that-trace ef-
fects discovered by Bresnan (1977) and also discussed by Rizzi (1997), among
others.
Part III (chapter 6) provides a summary of the general line of argumentation
and brings Chomskys (2005a) three factors into the discussion. The idea put for-
ward is that, with regard to the syntactic component, UG is a rewiring of elements
that must be in place independently: features, Merge and principles of analysis and
data processing.
I would like to express my gratitude to Vctor Acedo, Noam Chomsky, Bernat
Corominas, Txuss Martn, Josep Quer, Joana Rossell, Jaume Sol and Jeroni
Tutusaus, who have been an endless source of inspiration. I would also like to
thank the members of my dissertation committee, Itziar Laka, Carme Picallo,
Gemma Rigau, Joana Rossell and Jan-Wouter Zwart, for contributing to the im-
provement of this work with their comments. I am also indebted to Cedric Boeckx,
for his support and insistence on making this text more readable, and to Emily
Toder for her invaluable help in proofreading the manuscript. All the shortcomings
THE EMERGENCE OF ORDER IN SYNTAX 8
in the conception, development and expression of the ideas are mine alone and
reflect the limits of my current possibilities.
PART I. ELEMENTS OF SYNTAX





CHAPTER 1

ELEMENTS OF SYNTAX

This chapter discusses the most basic elements of the syntactic component of the faculty of
language: features, a combinatorial operation called Merge. Features are defined as in-
structions, provided by the lexicon, for a level of interpretation of the faculty (1.1). Merge
is defined as an operation that takes as input two categories A and B, viewed as sets, and
generates a new set C that is the union of A and B. The emergence of hierarchical pro-
perties derives from Merge (or union-formation) and from the existence of a derivational
record, a set K (a nest) where the outputs of Merge are sets linearly ordered by the strict
inclusion relation (1.2). Internal and external applications of Merge are argued to be trig-
gered by the necessity of matching [+type] categories and [+token] categories, with no
interpretable/uninterpretable dichotomy (1.3).


1.1 Instructions
An elementary metatheoretical notion relative to grammar and to any other
analytical discipline is that of level of representation or level of analysis,
understood as a set of concepts and devices used to construct a model for a
phenomenon. Since any object of inquiry is too complex to be analyzed as a
whole, one is forced to categorize the observed properties in several levels and
subsequently study how they interact. It is well known that grammatical phe-
nomena are classified at least as semantic, syntactic, morphological or pho-
netic. It is also a truism that, although such splits are methodologically necessary,
there are no facts that belong to a unique level of representation, and researchers
are forced to investigate a particular phenomenon by relating properties that are
relative to different levels of representation.
A feature is a property of an object. A linguistic feature is particularly a
property of a linguistic object, usually called a linguistic expression. If linguistic
properties or features are classified into levels of representation or analysis, then a
feature is relative to a level of representation or analysis. This means that a
feature cannot exist unless there is a level of representation or analysis where
such a feature can be represented or analyzed. Thus, we may say that there are
features that are relative to the levels of representation of phonetics, such as the
THE EMERGENCE OF ORDER IN SYNTAX 12
[plosive] feature or the [voiced] feature, and others that are relative to the
level of semantics, such as the [singular] feature or the [negative] feature.
It may be interesting to shift from the metatheoretical notion of level of
representation (a set of concepts and devices used to construct a grammatical
model) to what could be called an empirical notion of level of interpretation.
According to this empirical notion, the levels of interpretation are components of
the faculty of language, viewed as a set of emergent properties of the brain, that
interpret the features of a linguistic object as instructions. This implies that the sort
of theory we shall be seeking is not only a model for the shape of linguistic
products, but a model for the linguistic knowledge that generates those products.
Such a move, inherent to the biolinguistic-generative grammar amalgam, has the
virtue of reframing the usual purely intuitive considerations of elegance and
parsimony relative to the model in construction in terms of empirical questions
about the Faculty that try to discover to what extent it is an optimal solution to
multiple usability conditions, such as information coding and productive thinking.
Therefore, instead of saying this grammatical model is better than that one
because the principles it needs are fewer and more general, we will say the
faculty of language must be like this, and not like that, because this is all we need
to satisfy usability conditions. This approach to Language is what Noam
Chomsky calls the Strongest Minimalist Thesis (SMT): Language is an optimal
solution to legibility conditions.
According to the empirical notion of level of interpretation, linguistic
features are useful instructions for a particular level of interpretation of the faculty
of language.
A rather common working hypothesis in grammatical studies is that linguistic
objects are comprised of categories, which are clusters of features. It is also
customary to distinguish between substantive categories and functional categories,
the former constituting an open class and the latter a closed class; accordingly,
nouns and verbs would be considered substantive categories, and, for instance,
person-number marks would be considered functional categories. However, it
must be noted that the outlined distinction does not seem to be accurate, since the
so-called substantive categories are not atomic, but made of functional categories
and roots; if so, the right distinction may be in terms of functional categories and
roots.
Lets begin by considering functional categories, which will be central to our
concerns. A first basic question is: for which levels of interpretation of the faculty
of language do they contain features or instructions? Since several subdivisions
may be necessary for the so-called semantic and phonetic levels of interpretation, I
ELEMENTS OF SYNTAX 13
shall often refer to them as the Conceptual-Intentional (C-I) system and the
Articulatory-Perceptual (A-P) system, adopting two familiar terms.
Functional categories contain instructions for the C-I system. For instance,
conjunctions or clause typing categories are standardly considered to specify the
force of a proposition ([declarative], [exclamative], etc.) and the [embedded]
distinction. A common semantic instruction of Inflection-like functional categories
such as tense, mood, modality, negation and the person-number conglomerate is
that they trigger referential displacement in different dimensions (temporal, modal,
or individual).
More concretely, modal categories provide a relationship that determines the
force of a conclusion with respect to an evaluation parameter provided by a
conversational background (see Kratzer 1981, 1986), indicating whether a pro-
position holds in some possible world ([+existential]) or in all possible worlds
([+universal]). Mood categories define an embedded proposition as [actual] (see
Quer 1998, 2003 for discussion and a new dynamic characterization of the
mapping of mood morphemes in terms of model shift) and tense categories
locate the time of the event in an actual time ([+present]) or in a non-actual time
([+past] or [+future]). As recently argued by Sigursson (2003, 2004) following a
rather traditional view, the person feature relates the individuals of the event (the
arguments) to the participants of the speech act, allowing reference to an in-
dividual present in the speech act ([+logophoric agent, -logophoric patient], [-logo-
phoric agent, +logophoric patient]) or to an individual absent in the speech act [-
logophoric agent, -logophoric patient]. Hence, the values for the person feature of
an individual i can be defined as in (1):

(1) a. Person of i: [1
st
] iff i = [+logophoric agent, -logophoric patient]
b. Person of i: [2
nd
] iff i = [-logophoric agent, +logophoric patient]
c. Person of i: [3
rd
] iff i = [-logophoric agent, -logophoric patient]

It is noteworthy that the values for the number feature (singular, plural, dual, etc.)
do not strictly reflect properties of the logophoric participants. The [+plural] spe-
cification of the 1
st
and 2
nd
person pronouns in (2) does not necessarily convey that
the logophoric agent or the logophoric patient are plural.

(2) a. We are smoking
b. You are smoking

This is also noted by Sigursson (2004: 19), among others (see references cited
therein):
THE EMERGENCE OF ORDER IN SYNTAX 14
We, for instance, is obviously not a plural of I in the sense that it denotes more than one
speaker (except perhaps under extremely rare and special circumstances). However, it is
the plural of I in the sense that it denotes more than one potential linguistically active
selves, where a linguistically active self is a speaking or a thinking (experiencing, feel-
ing) actor: I and others that could be in my footsteps as speakers/thinkers/
experiencers.

Hence, the instruction provided by the values of the number feature relative to an
individual i belonging to the set I of individuals that are involved in an event , is
to define the cardinality of I (symbolically,| I |).

(3) a. Number of i (i I): [singular] if | I | = 1
b. Number of i (i I): [plural] if | I | _ 2
c. Number of i (i I): [dual] if | I | = 2

Finally, if negation in natural languages is not essentially different from the
Boolean connective , the semantic instruction it provides must be to shift from
the truth value of , to the truth value of , in such a way that , is false if and
only if , is true, and , is true if and only if , is false.
In sum, the ability to trigger referential displacement seems to be a general
feature or instruction provided by functional inflectional categories to the semantic
level of interpretation: the semantic feature provided by modal categories, mood
categories, tense categories and the person-number conglomerate is to shift from
the actual to the non-actual, and the semantic feature provided by negation is to
shift from the truth , of to the falsity of ,, and from the falsity of , to the truth of
,.
The morphological level of interpretation also requires features from
functional categories. For instance, a tense category may be morphologically
realized as a free particle, as an attached head or as a maximal projection (an
adverbial phrase). Furthermore, it may be the case that two different categories are
assigned to two different slots (analytic pattern), that two different categories are
assigned to a single slot (syncretic pattern) or finally, that a single category is
assigned to two different slots (discontinuous pattern). Thus, a functional category
provides templatic features such as [free], [analytic], [syncretic], [dis-con-
tinuous].
Functional categories also provide features to the A-P system. For instance, in
English, an embedded finite assertive conjunction is spelled out as [et] and an
indirect partial question conjunction as [If] or [wce].
If there exists a grammatical component that orders categories, and this
component is called syntax, one of the most elementary questions about syntax is
ELEMENTS OF SYNTAX 15
where the necessary instructions for ordering pieces come from. In ordering
pieces, syntax associates a category that displays a token of a particular type of
feature with a position semantically specialized for that type of feature. This
observation can be implemented in the theory of the syntactic component by
postulating two sorts of features: types and tokens. If so, the syntactic level of
interpretation interprets certain categories as types, which indicate which category
can appear in a particular position, and others as tokens, which instantiate a
particular type. According to this reasoning, syntactic operations are driven by the
necessity of matching type categories with token categories.
It may well be that categories are not grammatical primitives but grammatical
by-products, or in other words, that there is no such a component as a unified
lexicon with categories (clusters of features or instructions) listed, but a multi-
plicity of lexical components, each providing a particular subset of features. Since
this issue is immaterial for our concerns, I shall adopt the handiest option, i.e., that
phonetic features, morphological templatic features, the type/token syntactic
feature and semantic features are coded in lexical entries dubbed categories. The
lexicon, the locus of lexical entries, contains the instructions that activate gram-
matical computations.
The various grammatical components of the faculty of language (phonetics,
morphology, syntax and semantics) are at the same time levels of interpretation
and computational devices following the interpreted instructions. Syntax interprets
objects selected from the lexicon and submits them to its own computations,
constructing greater objects that are interpreted at the Logical Form (LF) level and
at the morphological level. LF is not a final product, but the interface between
syntax and the C-I system; similarly, morphology is also an interface between
syntax and further devices of the A-P system. Indeed, syntax is also an interface,
which relates the lexicon to morphology and LF. Both LF and morphology submit
syntactic objects to its own computations creating new outputs, which will be
repeatedly interpreted and computed until two final representations are attained,
one (SEM) usable by thought systems and the other (PHON) by externalization
systems. Whereas the lexicon is the first motor of grammatical computations, SEM
and PHON are their products.
It is important to express these preliminary considerations here, since our
efforts will be concretely guided to define the design properties of the syntactic
component of the faculty of language as an optimal interface between the lexicon,
the morphological component and the semantic component.



THE EMERGENCE OF ORDER IN SYNTAX 16
1.2 Merge: nests
If there is a continuous set of branches going down a phrase marker like (4)
from a higher category to a lower category, it is said that the higher category
dominates the lower one. Thus, in (4) A dominates the rest of categories, B
dominates D and E, C dominates F and G, and D, E, F and G dominate no
category. The relation of term-of is the inverse relation of dominance; hence, in (4)
D and E are terms of B, F and G are terms of C, B and C are terms of A, all the
categories are terms of A and A is not a term of any category.

(4) A

B C

D E F G

On the basis of the dominance relation, two further structural relations have
been defined: c-command and asymmetric c-command.

c-command
A category X c-commands another category Y iff X does not dominate Y and
every category dominating X dominates Y.

asymmetric c-command
A category X asymmetrically c-commands another category Y iff X c-com-
mands Y and Y does not c-command X.

A very important proposal developed by Richard Kayne (Kayne 1994)
suggests that the hierarchical nature of syntactic objects is closely related to the
defining properties of the linear order relation. More precisely, Kayne proposes
that the terminals of a syntactic object can be ordered if and only if the non-ter-
minals that dominate them follow an asymmetric c-command relation. This is
expressed by the Linear Correspondence Axiom:

Linear Correspondence Axiom (LCA)
Given a set T of terminals of a phrase marker P and a total asymmetric c-
command relation A among the non-terminals of P, the dominance relation
ELEMENTS OF SYNTAX 17
from non-terminals to terminals d(A) yields as an image a linear ordering of T
(adapted from Kayne 1994: 3-6).

In the phrase marker (5), A = {<J, M>, <J, N>, <J, P>, <M, P>}. By virtue of
the LCA, T = <j, m>, <m, p> and <j, p>. However, in the phrase marker (6), the
terminals m and p cannot be ordered, since neither M asymmetrically c-commands
P nor P asymmetrically c-commands M. The tree-diagram (5) conforms to the X-
theory and provides a strict linear ordering relation on the terminals; diagram (6)
does not.

(5) K (6) L

J L M P

j M N m p

m P

p

The so-called X-theory was an early stage in the effort to resolve the tension
between explanatory and descriptive adequacy (Chomsky 1995: 241) that at-
tempted to reduce multiple phrase structure rules belonging to particular languages
and redundant in its form (Chomsky 1965, among many others) to the following
unified syntactic template attributed to Universal Grammar (Chomsky 1981,
Chomsky & Lasnik 1993, among many others):

(7) a. X ZX
b. X XY
c. X X
0
W

The lexical item inserted in X
0
determines the type or label of X and X and
restricts the choices of W and Z, called respectively complement and specifier.
Kaynes general conclusion is that the X-theory is not a primitive of Universal
Grammar (UG), but a construct derived from the necessity of assigning a strict
linear order to the terminals of a phrase marker.
It is crucial for Kaynes proposal to assume a substantive (not merely no-
tational) distinction between maximal, intermediate and minimal categories; how-
ever, as Chomsky has argued (1994a, 1995), these elements are stipulated con-
THE EMERGENCE OF ORDER IN SYNTAX 18
ventions of the X-theory: there is no independent motivation to think that the
computations taking place at the C-I system or at the A-P system must be sensitive
to whether a category is maximal, intermediate or minimal, a distinction that
furthermore violates the inclusiveness condition (any structure formed by the
computation is constituted of elements already present in the lexical items
Chomsky 1995: 228). Consequently, Chomsky proposes substituting the X-theory
with a bare phrase theory, where the unique legitimate objects are terminal ca-
tegories or lexical items (clusters of features) and sets of lexical items formed by
syntactic operations.
Therefore, the LCA is based on stipulated conventions that seem to lack an
independent motivation. As a matter of fact, the LCA cannot be formulated in a
bare phrase structure, which dispenses with the distinction between maximal, in-
termediate and minimal categories.
In what follows I shall develop a reformulation of the LCA such that it is
compatible with bare phrase structure, postulating no entities but terminals and
sets.
The simplest possible way of defining a combinatorial system is to postulate
the operation Merge as a basic element. Merge takes n objects and constructs a
new object. The objects taken as input may be syntactic atoms (namely, lexical
items) or objects previously constructed by Merge. Thus, Merge is a recursive op-
eration: it can take objects that result from previous applications of itself as input.
The output of Merge is an object that displays the property of discrete infinity: it
is discrete because it is ultimately made of atomic elements and infinite be-
cause the number of applications of merge operations is potentially boundless.
We shall provide a more precise definition of Merge. Given an alphabet S, let
Merge be a successive set-formation operation that takes as input two sets A, B
included in S and yields as output a set C that contains all the elements of A and all
the elements of B and that does not contain any other elements. Thus, Merge of A
and B forms the union of A and B (Merge (A, B) A B). For Merge to be suc-
cessive, A or B (or both) must be the output of an immediately preceding ap-
plication of Merge. Lets examine the consequences of this definition.
Given the alphabet S, consider the following derivation by successive binary
Merge. Merge
1
is the first application of Merge, which means that is the output
of the immediately preceding application of Merge. The first application of Merge
will thus yield a singleton as output. If the input of Merge
1
were {a} and {b},
instead of {a} and , Merge
1
would not be successive.



ELEMENTS OF SYNTAX 19
Alphabet
S = {a, b, c, d, e}

Derivation
Merge
1
(, {a}) {a}
Merge
2
({a}, {b}) {a, b}
Merge
3
({a, b}, {c}) {a, b, c}
Merge
4
({a, b, c}, {d}) {a, b, c, d}
Merge
5
({a, b, c, d}, {e}) {a, b, c, d, e}

The theory of grammar must ensure not only that smaller categories are
combined into greater categories, but also that the derivational history is not lost.
Let the representation generated by the five applications of successive Merge be
the derivational record of the five outputs of successive Merge, i.e., the set K
where the five outputs are sets linearly ordered by the strict inclusion relation. We
shall call a set or family of sets linearly ordered by inclusion a nest.

Derivational record
K = {{a}, {a, b}, {a, b, c}, {a, b, c, d}, {a, b, c, d, e}}

The sets that belong to K can be identified with the syntactic nodes of a tree-
diagram and the elements of these sets as the terminals dominated by the syntactic
nodes.
As is well-known, a linear order can be derived from the basic terms of set
theory, i.e., from sets and elements. Particularly, Kuratowski (1921), who sim-
plified previous works by Hessenberg and Hartogs, demonstrated that, given a set
S, a family of sets F linearly orders S if:

(i) the elements of F are subsets of S
(ii) F is a nest: for any X, Y that belong to F, either X Y or Y X
(iii) F is saturated as to the property of being a nest of S: F is not a sub-
family of a family F that satisfies (i) and (ii)

For an alphabet S = {a, b, c, d, e}, the derivational record K = {{a}, {a, b}, {a,
b, c}, {a, b, c, d}, {a, b, c, d, e}} is a family of sets that satisfies conditions (i)-
(iii). Thus, we can conclude that the derivational record K provides a linear order
of the terminals of S. This is a very intuitive result: observe that a belongs to all
THE EMERGENCE OF ORDER IN SYNTAX 20
the subsets of K and that e belongs to solely one subset of K. b belongs to four
subsets of K, c belongs to three subsets of K and d belongs to two subsets of K.
Thus, the five elements of S are linearly ordered by the number of sets they belong
to or by being more or less embedded in K.
Without moving into Kuratowskis detailed demonstration that F gives a
transitive, asymmetric and total relation among the elements of S, it is easy to
conclude that nests provide a linear order.

The essential property of tuples is that:

(<a
1
, a
2
, , a
n
> = <b
1
, b
2
, , b
n
>) (a
1
= b
1
a
2
= b
2
a
n
= b
n
)

Thus, all we need to show is that:

({{a
1
}, {a
1
, a
2
}, {a
1
, a
2
, }, {a
1
, a
2
, a
n
}} = {{b
1
}, {b
1
, b
2
}, {b
1
, b
2
, },
{b
1
, b
2
, bn}}) (a
1
= b
1
a
2
= b
2
a
n
= b
n
).

Assume the truth of the precedent. First, observe that a
1
= b
1
,

since {a
1
} is the
sole singleton of the first nest and {b
2
} is the sole singleton of the second nest.
Second, a
2
= b
2
, since {a
1
, a
2
} is the sole pair of the first nest and {b
1
, b
2
} is the
sole pair of the second nest, and we know that a
1
= b
1
. If we continue, we finally
observe that {a
1
, a
2
, a
n
} = {b
1
, b
2
, bn}), since {a
1
, a
2
, a
n
} is the sole set of
n elements of the first nest and {b
1
, b
2
, bn} is the sole set of n elements of the
second nest, and we know that a
1
= b
1
a
2
= b
2
. Therefore, we derive the truth of
the consequent from the truth of the precedent.
Consider the step Merge
3
in the Derivation 1, where the input of Merge is two
objects previously constructed by Merge.

Derivation 1
Merge
1
(, {novels}) {novels}
Merge
2
({novels}, {reads}) {novels, reads}
Merge
3
({novels, reads}, {the, gardener}) {novels, reads, the, gardener}

K
1
= {{novels}, {novels, reads}, {novels, reads, the, gardener}}

The derivational record K
1
is a nest (its subsets are linearly ordered by the
inclusion relation), but a linear order among terminals is not obtained unless the
derivational record K
2
is accessed: novels is more embedded than reads, both
ELEMENTS OF SYNTAX 21
novels and reads are more embedded than the and gardener, but no ordering exists
in K
1
between the and gardener.

Derivation 2
Merge
1
(, {gardener}) {gardener}
Merge
2
({gardener}, {the}) {gardener, the}

K
2
= {{gardener}, {gardener, the}}

In the X-theoretic jargon, all this means that a complement is more deeply
embedded than a head, and that they are both more deeply embedded than a
specifier, whose terminals are independently ordered because they are merged in a
different (sub-)derivation. Thus, specifiers and non-branching complements are no
longer a problem for linearization or hierarchization. Note that, if Merge were
defined as set-formation (as Chomsky 2005b suggests), instead of union-
formation, specifiers would still pose a problem for linearization: {{the,
{gardener}}, {read, {novels}}}.
Based on different assumpitions, Zwart (2007) has independently suggested
that Merge must yield order by assigning elements from an alphabet (or nume-
ration) to a derivation. This intuition is now accurately expressed in set-theoretical
terms. Thus, we are finally in a good position for reformulating the LCA in bare
phrase structure terms, a reformulation that is no more than Kuratowskis (1921:
164) Definition 1.

LCA
A family of sets F linearly orders the terminals of an alphabet S if F is
saturated as to the property of being a nest of S

However, the above sentence is not an axiom. It is not a proposition one needs
to take for granted to derive other propositions, namely the X-theory. Indeed, all
we need in the theory of syntax to account for hierarchy is a successive operation
of union-formation and the existence of derivational records with the form of
nests. In this case, the history of the theory of phrase structure is no more than an
instance of how the construction of a model for a particular phenomenon is guided
by the desiderata of minimal design.



THE EMERGENCE OF ORDER IN SYNTAX 22
1.3 Merge: internal and external
In the development of the minimalist inquiries, it has been proposed that the
following two properties of language are related:

(I) Functional features are interpretable on some lexical items and uninter-
pretable on others

(II) The surface phonetic position of a category is dissociated from the deep se-
mantic position

The person-number conglomerate (referred to as -features, as it is customary in
the minimalist tradition) supports property (I). As illustrated in the following
Catalan paradigm, -features may surface on nominals (nouns, pronouns and de-
terminers), but also on verbs and adjectives, although the semantic instruction that
each occurrence provides is relative to nominals exclusively.

(8) a. vindr jo
come.FUT.1SING 1SING
I will come

a. vindrem nosaltres
come.FUT.1PLUR 1PLUR
We will come

b. vindrs tu
come.FUT.2SING 2SING
You will come

b. vindreu vosaltres
come.FUT.2PLUR 2PLUR
You will come

c. vindr lestudiant
come.FUT.3SING the.SING student.SING
The student will come

c. vindran els estudiants
come.FUT.3PLUR the.3PLUR estudiants.PLUR
The students will come
ELEMENTS OF SYNTAX 23
(9) a. un cotxe rpid
a.SING car.SING fast.SING
A fast car

a. uns cotxes rpids
a.PLUR car.PLUR fast.PLUR
Some fast cars

An occurrence of -features on a nominal is said to be [+interpretable] because
it provides semantic instructions relative to such a nominal, whereas an occurrence
of -features on a verb or an adjective is said to be [-interpretable] because it does
not provide instructions relative to events, but rather reflects semantic properties of
the nominal with which it agrees.
Property (II), usually conceived of as a displacement operation of a category
from one position to another, is broadly attested in natural languages and displays
a relatively rich phenomenology: there is movement to case positions and to
peripheral positions, movement of heads and movement of projections, apparently
long-movement and local movement, movement with pied-piping and movement
with stranding.
Both properties (I) and (II) have been viewed as striking, or in fact as
imperfections: why should lexical items contain uninterpretable features?; why
should a category be pronounced in a position where it is not semantically inter-
preted?, or in other words, why is it that the occurrences phonetically interpreted at
the A-P system and semantically interpreted at the C-I system are not the same,
and hence PHON (the linguistic object to be used by externalization systems)
diverges from SEM (the linguistic object to be used by thought systems)?
In Chomsky (2001a, 2001b), and related work by many others, the working
hypothesis that these two apparent imperfections of narrow syntax may be related
is investigated. According to this view, uninterpretable features are the trigger of
displacement. Consider the following prototypical examples of raising cons-
tructions.

(10) a. [A proof] is likely [to have been discovered [a proof]]
b. [A proof] seems [to have been discovered [a proof]]

It is claimed that matrix T has a complete set of uninterpretable -features;
embedded T has a defective set of -features which cannot match the complete set
of -features of a referential DP like [a proof]. T matrix forces the DP to move to
its SPEC, causing [a proof] to be pronounced in the clause of be-likely or
THE EMERGENCE OF ORDER IN SYNTAX 24
seem, and thematically interpreted in the VP of discover. According to this
analysis, the traditional case filter effects are not attributed to the necessity for a
DP to receive case, but rather to the necessity for the uninterpretable -features of
matrix T to get deleted. Thus the source of movement, property (II), is precisely
property (I): uninterpretable features trigger movement of a category with in-
terpretable features in order to satisfy their need to get deleted. Call this need, the
device that triggers displacement, suicidal greed. This deletion operation must be
carried out before the syntactic object arrives at the C-I system; otherwise, the de-
rivation will yield a gibberish object, containing categories without useful in-
structions for the C-I system and thus violating the Full Interpretation Condition:

Full Interpretation Condition
An object generated by a grammatical component must be constituted only of
useful elements for a particular level of interpretation when it attains such a
level of interpretation

Recently, Chomsky has observed that the existence of movement in natural
languages is not a striking property requiring special motivation:

For over 40 years, there have been efforts to motivate displacement. That seems to have
been a mistake. Recourse to any device to account for the displacement phenomena also is
mistaken, unless it is independently motivated (Chomsky 2001b: 8, note 29).

The argument is simple: Merge, the recursive operation responsible for putting
pieces together by creating greater syntactic objects, comes for free in a com-
binatorial system like the syntactic component. Merge takes as input A and B. If
these two objects are separate objects, Merge is external, whereas, if A is a term
of B, Merge is internal; in the second case, A is taken from B and re-merged to B,
yielding the property of displacement. Accordingly, displacement is not an im-
perfection of language; its absence would be an imperfection (Chomsky 2001b:
8).
Property (I) must be studied more thoroughly. It is currently understood that
displacement, property (II), cannot be seen as an imperfection; however, it seems
that it does not apply vacuously, since all instances of displacement yield semantic
effects of multiple types. One might consider whether suicidal greed is the device
responsible for constraining or triggering the applications of internal Merge: an
uninterpretable feature located in a position of the hierarchy needs to be deleted,
and hence triggers movement of a category with a matching interpretable feature.
Accordingly, a category would be associated with multiple interpretations due to
ELEMENTS OF SYNTAX 25
uninterpretable features that would be present in different positions of a hierarchy
just to be deleted.
Such a theoretical device presents some problems. I shall point out three of
them, which suffice to show the inadequacy of suicidal greed as the trigger of
displacement. The first one may be viewed as a problem of generality, the second
one as a problem of determinacy, and the third one as a problem of consistency.
As for the problem of generality, there are many functional features triggering
displacement. Thus, if uninterpretability is the real trigger of displacement, it is
expected that the uninterpretable status of triggering features can be determined
not only for -features but for a plentitude of functional features. But once we
move to functional features other than -features of T, there are no convincing
reasons for uninterpretability. For example, it has always been intuitively unclear
that features triggering A-movement are uninterpretable, since the claim that
peripheral features (force features or topic/focus features) are uninterpretable on its
locus, C, is entirely stipulative. Indeed, a-features (peripheral features) on C are
more naturally characterized as positional indications for where a a-constituent
must be merged in order to be properly interpreted. This is usually expressed in
terms of Rizzis (1991, 1997) or Haegemans (1995) criteria, but these re-
presentational conditions cannot be the derivational trigger of syntactic operations.
More plausibly, as argued above, force features are not the only type of triggering
features that are clearly interpretable: the class of inflectional-like features (mood,
tense, -features and negation) provide the semantic instruction of triggering re-
ferential displacement and also trigger movement operations. Hence, suicidal
greed does not seem to be the general source of displacement. This problem of
generality casts some doubt on the adequacy of the suicidal greed device for the
particular case of -features: a model that accounted for all instances of movement
with a unique device is theoretically superior to a model with a plurality of de-
vices.
Lets now shift to the problem of determinacy. We must more closely examine
whether the occurrences of -features triggering A-movement are uninterpretable.
Assume that -features on a verb are [-interpretable], for they do not provide se-
mantic instructions relative to V but rather relative to a nominal. Note that the
(un)interpretable status of -features on T does not derive from the premise that
they are uninterpretable on V. The interpretability of a feature is relative to a ca-
tegory, and thus one cannot determine the uninterpretable status of -features on T
from the uninterpretable status of -features on V. In the absence of specific ar-
guments for the uninterpretability of -features on T, the neutral hypothesis is that
they are interpretable, as a-features of C. In fact, we may extend Rizzis view on
THE EMERGENCE OF ORDER IN SYNTAX 26
a-features to -features of T: -features of T are responsible for determining the
binder of a subject-oriented anaphor, as a-features of C are responsible for deter-
mining the scope and the quantificational force of a a-phrase. Again, this suggests
that -features of T, the trigger of A-movement, are as interpretable as a-features
of C, the trigger of A-movement.
Finally, we address the problem of consistency. Although there is no reason to
think that the uninterpretability of -features on V entails the uninterpretability of
-features on T (the determinacy problem), which would indeed be an exceptional
case among triggering features (the problem of generality), continue assuming that
-features on both V and T are uninterpretable. As will be argued in detail in
chapter 5, V moves to T precisely in those languages where V shows up with an
overt occurrence of uninterpretable -features; if V does not show up with an
overt occurrence of uninterpretable -features, it remains inside the P. This pres-
ents an inconsistency for the hypothetical suicidal greed device, since un-
interpretable features on V delete uninterpretable features on T. It is also important
to note that Null Subject Languages display rich inflectional mor-phology on V,
which suggests that an interpretable instance of -feature (a DP) and an unin-
terpretable instance of -features (an inflected V) compete to delete an unin-
terpretable feature on T (see chapter 5); on these grounds, it could be stipu-lated
that -features on T provide useful semantic instructions (i.e., they are interpre-
table) when V is inflected for (uninterpretable) -features, whereas they are unin-
terpretable when V is not inflected for -features, in which case a DP with inter-
pretable -features must move to delete the uninterpretable -features of T. It is
not worth considering such an inelegant possibility, whereby the interpretability of
features is decided in a purely stipulative way.
The problems of generality, determinacy and consistency strongly push the
device of suicidal greed into an unacceptable situation and hence manifest its
inadequacy. If this is so, what is the trigger of internal Merge of a category? Since
the property of displacement dissolves into the basic operation of Merge, there can
be only one answer: the same that the triggers external Merge of a category. Any
device specially constructed for displacement is suspicious.
Indeed, an accurate way of expressing the asymmetry between hosting
categories and hosted categories has already been proposed in this study. A func-
tional category X can enter the derivation specified as a type or as a token. If X is a
type, it is interpreted as a hosting category in syntax. There are two possible stra-
tegies to instantiate a [+type] feature with a suitable [+token] feature: by external
Merge or by internal Merge. Thus, both internal and external Merge of categories
with [+token] features are triggered by the necessity of matching [+type] features
ELEMENTS OF SYNTAX 27
present in a syntactic hierarchy and [+token] features, which yields a positional
system where particular semantic instructions are associated with particular po-
sitions.

1.4 Onset
Therefore, the most basic elements of the syntactic component are features
provided by categories or lexical units and a combinatorial operation called Merge
that is responsible for recursively creating hierarchically structured objects.
In the next part, I shall examine how syntactic derivations by successive (in-
ternal and external) Merge are driven by the morpholexical characterization of lex-
ical items and constrained by semantic legibility conditions that apply as well-
formedness conditions and by general efficiency principles that reduce the neces-
sary span of structure to match types and tokens. Or in other words, I shall attempt
to understand why syntactic objects have the shape they seem to have by ins-
pecting how the syntactic component interfaces the morphological component and
the semantic component in an efficient way.


PART II. PATTERNS







CHAPTER 2

THE RELATIONSHIP BETWEEN COMPLEMENTIZERS AND
INFLECTIONAL CATEGORIES

This chapter argues (i) that complementizers can replicate Infl-like features as -fea-
tures, tense, mood, modality and negation, (ii) that there is a correlation between the cha-
racteristics of Infl-like features and the choice of C, and (iii) that Infl-like features can
trigger movement both to Infl and C. These three properties materialize the link between C
and Infl.


2.1 The C-Infl link
There is a strong correlation between complementizers (or [+clause typing]
categories) and inflectional categories (or [-clause typing] categories), a traditional
observation adhered to in several classical works in generative grammar
(Rosenbaum 1965, Bresnan 1972, den Besten 1977/1981, Koster 1975, 2003,
Stowell 1981). This point is also noted in cartographic studies as Rizzi (1997) or
Cinque (1999a: note 53 and relative text in p. 18; note 20 and relative text in p.
84). For example, Rizzi (1997: 283) observes that:

It is a traditional observation that the choice of the complementizer reflects certain
properties of the verbal system, an observation formalized by, e.g., agreement rules
between C and I, responsible for the co-occurrence of that and a tensed verb, or for
and an infinitive in English (Chomsky and Lasnik 1977), etc. A straightforward manner to
account for these dependencies would be to assume that C contains a tense specification
which matches the one expressed on the lower inflectional system (an idea that goes back
at least to den Besten (1977).

I shall discuss a sample of phenomena that substantiate the idea that several
features generally associated with [-clause typing] positions are linked to [+clause
typing] positions, a topic profusely studied in grammar. I shall use the label Infl
instead of the currently more common T, since it would be confusing to label
[-clause typing] occurrences of features such as negation mood or modality as T.
Chapter 4 shall consider the source of the generalized C-Infl link.

THE EMERGENCE OF ORDER IN SYNTAX 32
2.1.1 -features. In several languages, not only verbs but complementizers as well
are inflected for -features. Certain Germanic languages offer well-known
examples of this morphosyntactic phenomenon (see, among others, Bayer 1984,
Haegeman 1992, Hoekstra & Marcz 1989 and Zwart 1993). Consider the fol-
lowing paradigm from West Flemish (Haegeman 1992), where com-plementizers
show up with two different subject agreement marks successively attached:

(1) a. da-n-k ik komen
that-1SG-I I come-1SG
b. da--j gie komt
that-2SG-you you come-2SG
c. da-t-j ij komt
that-3SG-he he come-3SG
d. da--se sij komt [ < t]
that-3SG-she she come-3SG
e. da--me wunder komen [ < n]
that-1PL-we we come-1PL
f. da--j gunder komt [ < t]
that-2PL-you you come-2PL
g. da-n-ze zunder komen
that-3PL-they they come-3PL

It could be argued that -features are base-generated in the Infl-area, but
languages may vary with regard to replicating these instructions in the C-area. In
that case, there would be a rather arbitrary morphological device for replicating
functional features where they are not interpreted. Such a process may be op-
erative in natural languages, as in multiple auxiliaries that agree with a nominal in
Swahili (2.a) or Arabic (2.b), multiple determiners or adjectival categories that
agree with N in Catalan (3), or multiple elements of the clause that may show an
optional plural mark when the subject is plural in Korean (4), a phenomenon
known as spurious agreement.

(2) a. Juma a-li-kuwa a-ngali a-ki-fanya kazi
Juma SA-PST-be SA-still SA-PROG-do work
Juma was still working




THE RELATIONSHIP BETWEEN COMPLEMENTIZERS AND
INFLECTIONAL CATEGORIES
33
b. al-bint-aani kaan-ataa ta-ktub-aani darsa-humaa
the-girls(F)-3D be-PST-3FD 3f-write-D lesson-FD
The two girls were writing their lesson
(Carstens 2001: 395)

(3) a. La meva casa verda
the.FEM.SING my.FEM.SING house.FEM.SING green.FEM.SING
My green house

b. El meu cotxe verd
the.MASC.SING my.MASC.SING car.MASC.SING green.MASC.SING
My green car

(4) ai-tul-i sensayngnim-kkey-tul yelsimhi-tul cilmun-ul
child.PL.NOM teacher.DAT.PL intently.PL questions.ACC
ha-ko-tul iss-ta
do-comp-PL be
The children are asking questions to the teacher intently
(Kim 1994: 303)

In (4) plural features of N spread not only to the verb, but also to an indirect
object and an adverb. Interestingly, as Kim (1994) illustrates, they may also spread
to the complex [V-complementizer] of an embedded clause, but not to arguments
or adjuncts of such an embedded clause. In (5) the plural feature of the matrix
subject may spread to the embedded [V-complementizer], but not to the locative
adjunct of the embedded clause, whose subject is singular.

(5) a. sensayngnim-tul-un [kim-I emma-lul hakkyo-ey(*-tul)
teacher-PL-TOP Kim-NOM mother-ACC school-LOC-PL
mosie olila-ko]-tul sayngkak hayessta
bring-COMP-PL thought
Teachers thought that Kim would bring his mother to the school

But the sort of correlation between [+clause typing] positions and [-clause
typing] positions seems to be deeper than this morphological process of spreading
features to several categories, since there is a meaningful relationship between the
value of -features showing up on Infl and the choice of C. The most basic
observation is that the finiteness of a clause is coded not only on Infl but also on C
in such a way that the Infl head of an embedded clause in several languages has a
THE EMERGENCE OF ORDER IN SYNTAX 34
complete set of -features (V is finite and a full DP can be licensed) only if the
embedded C is finite; in Catalan, for instance, the complementizer de correlates
with non-finite embedded verbs, whereas the complementizer que correlates with
finite embedded verbs, and only in finite clauses can a referential subject DP be
licensed.

(6) a. Diu [que/*de vinguin]
say.3SG C[+FIN]/[-FIN] come.SUBJ.3PLUR

b. Diu [*que/de venir]
say.3SG C[+FIN]/[-FIN] come[-FIN]

(7) a. Diu [que vinguin els estudiants]
say.3SG C[+FIN] come.SUBJ.3PLUR the students

b. *Diu [de venir els estudiants]
say.3SG C[-FIN] come[-FIN] the students

In fact, the [fin] contrast in languages like Catalan is essentially the same as
the richer paradigm of West Flemish: in Catalan, complementizers display [fin]
distinctions, whereas in West Flemish, complementizers also display person and
number distinctions.
Chomsky (2000, 2001a, 2001b) has argued that T has a complete set of -
features associated with nominative case only if it is selected by C, as V has a
complete set of -features associated with accusative case when it is selected by
* (this is essentially Burzios 1986 generalization). If T is selected by V (instead
of C) and V by (instead of *), T and V lack -features. Thus, there is a one-to-
one relationship between -completeness and the presence of a phase head, C and
*.

(8) a. T is [-complete]

if selected by C
a. T is [-defective]

if selected by V

(C = )
b. V is [-complete] if selected by *
b. V is [-defective] if selected by

Note that there is a suspicious asymmetry between the phase head C,
associated with the properties of a proposition, and the phase head *, associated
with argument structure. * is one of the possible values for , the selector of the
THE RELATIONSHIP BETWEEN COMPLEMENTIZERS AND
INFLECTIONAL CATEGORIES
35
root and the locus of the agentive/non-agentive distinction. Consequently, when
the phase head * is not present, the archicategory must be present to define its
complement as verbal. However, when there is no phase head C, there is no archi-
category to define the clause type of proposition or to encode the [ embedded]
distinction.
In order to overcome this vagueness for propositions, I shall go a step further
in the analogy between the two phases, by expressing that, as is always present
containing argument structure information, C is always present containing clause
typing information. Both and C can have several values; more concretely, they
can be * and C*, in which case they select a full set of -features, or C
def
and
def
.
Hence, * is a specific value for that selects an external argument and a com-
plete set of -features that offers an A-position for accusative DPs, as C* is a par-
ticular value for C that selects a complete set of -features that offers an A-po-
sition for nominative DPs.

(9) C* *
C
C
def

def


T is [-complete] if selected by C*


T is [-defective] if selected by C
def



V is [-complete] if selected by *
V is [-defective] if selected by
def


The notion of defectiveness is very vague. What does it mean for a set of -
features to be defective or complete? Let us say that -features of an embedded T
can be free or bound by the occurrence of -features of the matrix T. If they are
bound, they must have the same value as the matrix occurrence, and hence the
subject of both sentences must be correferential. As is well known, referential DPs
must be free, and thus a set of bound -features cannot match the unbounded -
features of a DP. Thus, the property of defectiveness/completeness of -features of
T can be properly understood in terms of binding. If this is correct, the set of -
features of control and raising infinitivals are identical; the difference lies in how
the complete set of -features of matrix T is matched: by external Merge in control
structures and by internal Merge in raising structures.

(10) a. T is [-free] if selected by C* ([+fin]): [T
i
[C
fin
[T
j
]]]

THE EMERGENCE OF ORDER IN SYNTAX 36

b. T is [-bound] if selected by C ([-fin]): [T
i
[C
-FIN
[T
i
]]]


Therefore, -features can surface on C in the form of finiteness distinctions or
in the form of richer paradigms of person-number distinctions, as in the case of
West Flemish. Moreover, there is a meaningful relation between the type of -
features active on T and the choice of C: if C is [+fin] -features are free, while if
C is [-fin], -features must be bound by a higher occurrence of -features. If these
two properties are gathered, it becomes clear that the morphosyntactic phenomena
match the semantic relation, or in other words, that -features appear on both C
and T, and thus the inflectional properties of V and the ability to license a subject
are related to the properties of C. That is, the semantic relation between C and T is
no more than an agreement relation between two occurrences of -features, which
may or may not become morphologically overt or syntactically active:
[[]
C
[]
T
].

2.1.2 Tense. In the generative tradition, den Besten (1977/1981) might have been
the first to consider the C position as a [+tense] position. In his revision of the
standard analysis of root-phenomena in the 70s, he argued that all Complementizer
Attraction Transformations were structure-preserving rules of the following type:

(11) X - [+F
i
] - Y- [
C
+F
i
] - Z
1 2 3 4 5
1 4 3 e 5

where C is some constituent and
F
i
is some morphosyntactic feature

Den Besten states that the presentential domain must contain features other
than [+wh]. Consider the following base rule (12), where [+wh] defines the
landing site for interrogatives, according to the scheme rule (11).

(12) S [+wh] [+T] S

In this rule, den Besten argues, COMP is rendered as [+T], i.e. as [+Tense].
[+T] lexicalizes as dat that or of whether, if, and [-T] as om for. These
features are not counterintuitive because in a sense there is agreement between om
and a te-infinitival and between dat, of and the finite verb.
THE RELATIONSHIP BETWEEN COMPLEMENTIZERS AND
INFLECTIONAL CATEGORIES
37
Viewing COMP as [+T] led to a welcome consequence: it could obviate the
obligatory rule of Complementizer Deletion when V was fronted, by respecting the
general rule scheme (11); in the words of Besten:

The position [+T] allows for another instantiation of rule scheme [11]. The resulting rule is
Move Tense, i.e., Verb Fronting (Verb Second). This new formalization of the rule of Verb
Fronting predicts that there will be Verb Fronting only if the corresponding lexical comple-
mentizer (dat or of) is absent since the preposed finite verb occupies the complementizer
position. This way, the obligatory rule of complementizer deletion which was necessary
under the formalization for Verb fronting [] is obviated.

As illustrated in (2.1.1), complementizers can display not only [fin] dis-
tinctions, but also more fine-grained person-number distinctions, arguing against
the possibility that the occurrence of -features on C is essentially poorer than the
one surfacing on Infl. As with the complementizer agreement paradigms, Irish
offers a rather rich system of complementizers, marking a clause not only as sub-
ordinate, relative or interrogative, but also as being [past] (see Chung &
McCloskey 1987 and Hendrick 2000).

(13) Nonpast Past
Subordinating go gur
Direct relative a a
Indirect relative a ar
Interrogative an ar
Matrix negative n nor
Embedded negative nach nr
(Chung & McCloskey 1987: 218)

The characteristics of -features tend to correlate with the characteristics of
tense: usually -features are complete (they can license a referential DP) only
when tense is complete. Both features depend on finiteness, and therefore on the
choice of C (see Bianchi 2001a, b, Holmberg & Platzack 1995 and Zagona 1998
on the role of finiteness in defining the characteristics of -features and tense).

(14) a. -features and tense are complete if they are selected by C* ([+fin])
b. -features and tense are defective if they are selected by C
def
([-fin])

There are several well-known counterexamples to this tendency, which sug-
gests that, although the characterization of tense and -features tend to correlate,
there is no necessary one-to-one relation between them. Rigau (1995), for instance,
THE EMERGENCE OF ORDER IN SYNTAX 38
has argued that the temporal characterization of prepositional adjunct infinitival
clauses in Catalan and Spanish is defective, although they display a complete set
of -features licensing a full DP. Interestingly, -features are morphologically
realized in European Portuguese adjunct infinitivals (Raposo 1987). See also
Rigau (1998) for the analyses of several issues related to these constructions.
Examples in (15) are extracted from Rigau (1995).

(15) a. Tothom va aplaudir en acabar el concert (Catalan)
everybody applauded in to.finish the concert
Everybody applauded when the concert was finished

b. Todo el mundo aplaudi al acabar el concierto (Spanish)
everybody applauded in.the to.finish the concert
Everybody applauded when the concert was finished

c. Ao entrares tu, a Maria saiu (European Portuguese)
in.the to.enter-2SG you Maria left
When you went in, Maria left

Since [fin] Cs tend to correlate not only with the characteristics of tense but
also with those of -features, it may be generally immaterial whether the former or
the latter is the trigger of V-to-C movement in constructions like (16). That being
the case, we shall consider for the moment that the trigger of such an operation is a
featural conglomerate made of -features and tense, i.e., finiteness, as argued in
Holmberg & Platzack (1995). See section 5.4.1 of chapter 5 for a more detailed
discussion on the trigger of V-movement in Germanic languages.

(16) a. Diesen Buch las ich letztes Jahr
this book read.PAST I last year
I read this book last year

b. Diesen Buch habe ich letztes Jahr gelesen
this book have.PRES I last year read.PART
I have read this book last year

In (2.1.1), the inability of -features to host a full DP is associated with
binding relations; concretely, a set of -features selected by C
def
needs to be bound
by a higher set of -features. In both raising and control constructions C is [-fin]
THE RELATIONSHIP BETWEEN COMPLEMENTIZERS AND
INFLECTIONAL CATEGORIES
39
and -features are anaphoric. It can also be argued that tense is anaphoric when C
is [-fin]. Consider the interpretation of the following control infinitivals, where the
value of the embedded tense TENSE
E
is defined by the matrix tense TENSE
M
.

(17) a. John remembered [C
def
to lock the door]
TENSE
M
: PAST TENSE
E
: PAST

a. John remembers [C
def
to lock the door]
TENSE
M
: PRESENT TENSE
E
: PRESENT

b. Peter tried [C
def
to give up smoking]
TENSE
M
: PAST TENSE
E
: PAST

b. Peter tries [C
def
to give up smoking]
TENSE
M
: PRESENT TENSE
E
: PRESENT

If the event of remembering occurs prior to the speech time (TENSE
M
: PAST),
then the event of locking the door must also occur prior to the speech time
(TENSE
E
: PAST). The reading of John locking the door simultaneously or after the
speech time is not available when TENSE
M
: PAST. When the matrix event is present,
the embedded event can occur neither before nor after the speech time. Note that a
habitual reading is prominent in (17.a) (It is habitual that John remembers to lock
the door), unless an adverb like now or right now is present (John remembers
right now to lock the door).
Control infinitival clauses display an unrealized reading, as noted by Stowell
(1981), following Bresnans observations (1972). That is, in a sentence like (17.a),
the event of locking the door is not realized when the event of remembering takes
place. Observe that sentences (17.a) and (17.b) coincide in that they are unrealized
when the matrix event takes place, but differ in that they are realized at the speech
time: when (17.a) is uttered, the embedded proposition must be true; this is not the
case when (17.b) is uttered. This contrast must be derived from semantic factors,
specifically, from the lexical meaning of the selecting superordinate category, i.e.,
the matrix verb.
The interpretation of tense in the non-finite embedded clause of a raising
construction is also dependent upon the tense feature of the finite matrix clause.
Compare the temporal interpretation of the non-finite embedded sentences in (18)
with that of the finite embedded sentences in (19).

(18) a. Peter seems to be intelligent
THE EMERGENCE OF ORDER IN SYNTAX 40
b. Peter seemed to be intelligent

(19) a. It seems that Peter is intelligent
a. It seems that Peter was intelligent
a. It seems that Peter will be intelligent
b. It seemed that Peter was intelligent

Finite embedded sentences in (19) show that the tense of the embedded clause
selected by seems can differ from the tense of the matrix clause, both
morphologically and semantically. However, the temporal interpretation of the
proposition Peter-intelligent in (18.a) is determined by PRESENT
M
, whereas in
(18.b) it is determined by PAST
M
. In other words, the value of TENSE
E
must be
present in (18.a) and past in (18.b). Observe that (18.a) can only be equivalent to
(19.a) and (18.b) to (19.b), in accordance with the idea that tense is bound by a
higher tense in raising constructions.
Therefore, in [-fin] contexts, tense is anaphoric and must be bound by a higher
occurrence of tense, whereas in [+Fin] contexts can be free.

(20) a. [tense
i
[C* [tense
j/i
]]]
b. [tense
i
[C
def
[tense
*j/i]
]]

2.1.3 Mood. Languages may express the mood feature as a morpheme attached to
the verb (21.a), a free inflectional particle (21.b), or a clause typing category
(21.c). These examples are extracted from Quer (in press).

(21) a. Volen que dimiteixi (Catalan)
want.3PL that resign.SUB.3SG
They want her/him to resign

b. lun na pareti (Greek)
want.3PL SUB resign.3SG
They want her/him to resign

c. ja zhelaju cto-by ona ushla (Russian)
I wish.1SG that-BY she go.PAST
I wish that she should go

In other languages mood inflectional particles co-occur with mood com-
plementizers (Quer in press).
THE RELATIONSHIP BETWEEN COMPLEMENTIZERS AND
INFLECTIONAL CATEGORIES
41

(22) a. Ion vrea ca sa manince (Romanian)
Ion want.3SG that.SUB SUB eat.3SG
Ion wants him/her to eat
b. Jani do q t haj (Albanian)
Jani want.3SG that.SUB SUB eat.3SG
Jani wants him/her to eat

So there is no doubt that, like -features and tense features, mood features can
be marked on both C and Infl.
The classic Latin system of complementizers offers a rather clear and rich
instance of the correlation between the choice of complementizer and the verbal
mood ending. See table (23), extracted from Valent i Fiol (1992: 156), an ele-
mentary introduction into Latin syntax where the correlation between com-
plementizers and verbal mood endings is very explicitly drawn. Examples (24-27)
are also extracted from this work. The English translations are mine.

(23)

Infinitival
phrases



Introduced
by
conjunctions
By quod: indicative verbal mood
By ut, ne, quin, quominus, quod:
subjunctive verbal mood




Completive
clauses




Assertive






Indirect
questions

Introduced by interrogative pronouns or adverbials
(quis/quid, ut, ubi, etc): subjunctive verbal mood

In Latin, indirect questions always display subjunctive mood.

(24) a. Ego instare [sic] ut mihi responderet quis esset
I insist.INF that I.DAT told.SUBJ who.NOM was.SUBJ
I insisted that he told me who he was
(Cic., Verr., 2, 77)



THE EMERGENCE OF ORDER IN SYNTAX 42
b. Non properaui scire quis illis esset
NEG hasten.PAST know.INF what.NOM them.ABL was.SUBJ
rerum mearum status, quid afferent
wells.GEN my.GEN state.NOM what.ACC produced.SUBJ
I did not hasten to know what the state of my wells was in them,
nor what they produced
(Sen., Lucil, 77, 3)

Assertive embedded clauses may display indicative or subjunctive mood
depending on the selecting category. Interestingly, the choice of complementizer
tends to correlate with the value of the mood ending; generally, quod correlates
with the indicative mood (25) and ut/ne, quin, quominus with the subjunctive
mood (26).

(25) Eumeni inter Macedonas viventi multum
Eumenius.DAT among Macedonians.ACC living much
detraxit quod alienae erat civitatis
harmed that foreigner.GEN was.IND city.GEN
That he was from a foreigner city harmed Eumenius very much
while he was living among Macedonians
(Nep., 18, 1)

(26) a. Peto igitur ab te ut omnibus rebus quod
beg.1SG thus from you.ABL that all.ABL things.ABL that
sine molestia tua facere possis ei
without.ABL nuisance.ABL your.ABL do.INF can.SUBJ he.DAT
commodes
please.SUBJ
I thus beg you to please him in all the things that you can without
[causing] your nuisance
(Cic., Fam., 13, 35)

b. Quibus non humana ulla neque divina obstant quin
them.DAT not human any nor divine avoid.3PL that
socios, amicos, procul, iuxta sitos, inopes
allies-ACC friends.ACC distant close positions, weak
potentisque trahant, excidant
strong.AND push.SUBJ destroy.SUBJ
THE RELATIONSHIP BETWEEN COMPLEMENTIZERS AND
INFLECTIONAL CATEGORIES
43
No human or divine law prevents them from pushing and
destroying their allies or their friends close or distant, weak or
strong
(Sall., Mithr., 17)

c. Neque impedio quominus susceptum negotium
Not oppose.1S that assumed.ACC obligation.ACC
gerere possis
carry out.INF can.SUBJ
I am not against that you may carry out the assumed obligations
(Cic., Fam., 13, 5)

However, this correlation is not perfect: although ut, ne, quin, and quo-
minus never co-occur with indicative mood, quod can co-occur with both sub-
junctive and indicative verbal endings in certain cases, especially when the embed-
ded clause is selected by affective verbs like laudo (27.a vs. 27.b). This is usually
attributed to the subjective status of the predicate embedded under the affective
verb.

(27) a. Laudat Panaetius Africanum quod fuerit abstinens
laud.3SG P.NOM A.ACC that was.SUBJ neutral
Panaetius lauds the African for keeping neutral
(Cic., Off., 2, 22)

b. Quod viris fortibus honos habitus est, laudo
that men-ABL strong.ABL honor customary is.IND laud
That honor is customary in strong men, I laud
(Cic., Rosc. Am., 47)

Despite the apparent exceptions, the correlation between the choice of com-
plementizer and the value of mood is prevalent:

(28) C = [+ embedded, -assertive] mood: [subjunctive]
C = [+embedded, + assertive]:
(a) ut/ne, quin, quominus mood: [subjunctive]
(b) quod mood: [indicative]*

*quod can select subjunctive mood in some special cases, as when
it is selected by an affective verb
THE EMERGENCE OF ORDER IN SYNTAX 44

Finally, some cases of V-to-C movement are sensitive to mood distinctions;
particularly, in formal English and formal French, V is fronted when the con-
ditional complementizer is absent and the verb shows up with non-factual overt
morphology. In formal English, for instance, had and should, but not other aux-
iliaries like have or is, can be fronted in the absence of a complementizer.

(29) a. Had I known you were coming, I would have stayed longer
a. If (*had) I (had) known you were coming, I would have stayed
longer
b. Should you see him, tell him that his letter has already arrived
b. If (*should) you (should) see him, tell him that his letter has already
arrived

(30) a. *Have you failed the exam, you will be punished
b. *Is it raining, lets stay at home

Similarly, in formal French, according to Rizzi & Roberts (1989: 4), a con-
ditional clause can be introduced either by a complementizer or by a verb in the
conditional mood.

(31) a. Si tu avais fait cela,
If you had done that,
b. Aurais-tu fait cela,
Had you done that,
c. *Si aurais/avais-tu fait cela,
If had you done that,

From this summary we may conclude that mood distinctions may be displayed
on complementizers and also as free inflectional particles and verbal endings; the
value of mood seems to correlate with the choice of C, as the example of Latin
clearly illustrates, and mood features can be syntactically active on C triggering V-
to-C movement.

2.1.4 Modality. To the best of my knowledge, modal distinctions (epistemic
necessity, deontic possibility, etc.), unlike -features, tense and mood, are not
marked on complementizers and there are no instances of V-to-C movement
triggered by modal features. However, before arriving at any conclusion, we must
THE RELATIONSHIP BETWEEN COMPLEMENTIZERS AND
INFLECTIONAL CATEGORIES
45
consider whether there is a semantic relation between the choice of clause typing
features and the modal value of a proposition.
The idea that modal features are related to the C-system is not new in the
literature. Here I shall focus on two possible pieces of evidence, those that seem to
me more compelling. Different analytical arguments for this idea are given in
Drubig (2001) and Rochette (1988), who attribute to C the emergence of opacity
effects in certain subjunctive embedded clauses in certain Romance languages
when they contain a modal verb (Picallo 1985), assuming that C is the site where
sentential operators must appear at LF. The emergence of opacity effects does not
necessarily indicate that the opacity trigger is associated with C (although this is a
rather common view), but I prefer to leave it as an open question.
One may consider the possibility that the complementizer of conditional
sentences (if in English) encodes a modal feature. A possible analysis might
consist in viewing if as a modal quantifier; accordingly, in if his light is on, he is
awake, the semantic role of if is to define the set of possible worlds where the
proposition his light is on holds as a subset of the set of possible worlds where
the proposition he is awake holds. Or, equivalently, the clause introduced by if
(his light is on) would introduce the restriction for if and the matrix clause (he
is awake) the nuclear scope. Generally, if p, q would mean:

{w: p(w)} {w: q(w)}, or equivalently, w [p(w)]: q(w)

In this case, the only difference between the meaning of if and the universal
epistemic quantifier must would be that the restriction of the former is overt:
p(w). However, this treatment seems inadequate when dealing with the interaction
of if-clauses and quantificational adverbs (always if, rarely if, often
if), as observed by Lewis (1975), who argues that, in these particular cases,
if-clauses bear the restriction for the quantificational adverb. This insight is
extended to if-clauses in general by Kratzer (1986) in her claim that the history
of the conditional is the story of a syntactic mistake. There is no two-place ifthen
connective in the logical forms for natural languages. If-clauses are devices for
restricting the domains of various operators (p. 656). Adopting Kratzers and
Lewis ideas, the (simplified) LF representations of conditional sentences should
be as follows:

(32) a. Sometimes if his light is on he is awake (quantificational adverb)



THE EMERGENCE OF ORDER IN SYNTAX 46
w
i
(his light is on in w
i
): he is awake in w
i

w
i
(if-his light is on in w
i
) he is awake in w

w [if-his light is on in w]

[if]
C
his light is on in w

b. If his light is on he is awake (covert universal epistemic quantifier)

w
i
(his light is on in w
i
): he is awake in w
i


w
i
(his light is on in w
i
) [he is awake in w]

w [if-his light is on in w]

[if]
C
his light is on

c. If his light is on, he must be awake (overt universal ep. quantifier)

w
i
(his light is on in w
i
): he is awake in w
i

w
i
(his light is on in w
i
) [he is awake in w]

w [if-his light is on in w]

[if]
C
[his light is on]

An if-clause is not strictly a modal proposition, but it serves as a restriction
for a modal operator. Kratzers analysis suggests that previous analyses of the
meaning of if were indeed dealing with the meaning of the operator that the if-
clause restricts. Consequently, we may define if as a [+clause typing] category
responsible for defining a proposition as an overt restriction that contains a var-
iable ranging over possible worlds, a useful instruction for the levels of interpre-
tation of the C-I system.
There is another case where the characterization of C determines the modal
interpretation of a clause. Non-finite questions and non-finite relatives have an
THE RELATIONSHIP BETWEEN COMPLEMENTIZERS AND
INFLECTIONAL CATEGORIES
47
obligatory modal reading although there is no overt modality feature. Unlike con-
ditionals, these infinitival clauses are not interpreted as restrictions for modal op-
erators, but exactly as modal propositions.

Infinitival questions
(33) a. She knows [
CP
where to go]
b. I finally found out [
CP
what to do]
c. [
CP
How to learn French in five days]
Infinitival relatives
(34) a. The people [
CP
to talk to at the party]
b. The book [
CP
to read for tomorrow]
c. The man [to fix the sink] (from Bhatt 2006)

The presence of a covert modal semantic instruction depends on the choice of
C, since modal readings are not obtained in all non-finite contexts. For instance,
they are absent in the complement clauses of raising and control verbs.

(35) a. He prefers [
CP
to talk at the party]
b. He seems [
CP
to talk at the party]

This suggests that the covert modal feature is derivative from C. If C = [-FIN,
+question] or if C = [-FIN, +relative], the proposition must be interpreted modally.

(36) C [-fin, +qu/+rel] C = [+modal]

The idea that the source of covert modality in infinitive clauses is C is
developed in detail in Bhatts book (Bhatt 2006). Bhatt (2006: chapter 3), observes
that infinitival relatives with an object gap are interpreted necessarily with a modal
value, whereas infinitival relatives with a subject gap need not be interpreted mo-
dally under special licensing conditions, e.g., under the scope of modifiers such as
superlatives, ordinals or the nominal only.

Infinitival relatives with an object gap (Bhatt 2006: 11)
(37) a. a book [Op
i
[PRO to read t
i
]]
b. a knife [Op
i
[PRO to cut the bread [with t
i
]]]
c. the person [Op
i
[PRO to believe [t
i
to be innocent]]]

THE EMERGENCE OF ORDER IN SYNTAX 48
Infinitival relatives with a subject gap (Bhatt 2006: 45)
(38) a. This is the best argument to be considered by the committee
( which has been considered) (non-modal)
( which is to be considered ) (modal)

b. Joseph was the first/only/tallest man to walk on the moon last year
Joseph was the first/only/tallest man who walked on the moon last
year (non-modal)
Joseph was the first/only/tallest man who was supposed to walk
on the moon last year (modal)

In light of the asymmetry between object and subject gaps in infinitival rel-
atives, Bhatt (p. 45) proposes that:

in general the infinitival clause that appears in subject infinitival relatives is ambiguous
between an infinitival clause that is interpreted modally and an infinitival that is inter-
preted non-modally. The infinitival clause that is interpreted modally does not have spe-
cial licensing conditions and can appear wherever subject infinitival relatives appear. [].
The infinitival clause that is interpreted non-modally, however, needs to be licensed and
can therefore appear only in certain environments.

Note that covert modality in non-finite questions and relatives (37-38) cannot
be epistemic. Thus, the modal structure of finite clauses is more full-fledged than
the modal structure of non-finite clauses. This indicates that modality is derivative
from C not only in non-finite clauses but also in finite clauses.

(39) a. C [+fin] modality is overt and can be epistemic or deontic
b. C [-fin, +qu/rel] modality is covert and cannot be epistemic

In sum, a modal feature is encoded in the complementizer if of conditional
clauses, which introduces a restriction of possible worlds for a quantifier, and the
characteristics of modality are related to the choice of C. Concretely, if C = [+fin],
modality is full-fledged, and if it is [-fin, question/relative], it cannot be epistemic;
the modal interpretation is forced in non-subject infinitival relatives and optional
in subject infinitival relatives when a special licenser appears.

2.1.5 Negation. Complementizers are known to encode negative features in several
languages. For instance, in Latin the subordinator ne is the negative counterpart
of ut.
THE RELATIONSHIP BETWEEN COMPLEMENTIZERS AND
INFLECTIONAL CATEGORIES
49

(40) a. Imperat ut mihi respondeas
He/she orders that you reply me

b. Imperat ne mihi respondeas
He/she orders that you do not reply me

Celtic languages have negative particles in both main and embedded sentences,
which are usually analyzed as clause typing categories. See the table in (13), and
the following constructions, from Hendrick (2000: 22-23). Note that Breton (44)
presents two negative markers: a preverbal one, typing a clause as [embedded],
and a postverbal one. This is similar to the constructions in (22), where inflectional
mood particles co-occur with subjunctive complementizers.

(41) a. N heo an bord (Irish)
NEG here it the table
This here is not the table

b. Deir s [nach eo an bord]
say he NEG here it the table
He says this here is not the table

(42) a. Nach robh cta aice (Gaelic)
NEG was coat at.3SF
She didnt have a coat

b. Bha Ceiteag ag rdh nach robh cta aice
was kaitei prog say NEG was coat at3SF
Katie was saying that she didnt have a coat

(43) a. N welodd y bachgen y dyn (Welsh)
NEG saw the boy the man
The boy did not see the man

b. Dywedodd nad oedd y bachgen yn dod
say NEG was the boy PROG coming
He said the boy was not coming


THE EMERGENCE OF ORDER IN SYNTAX 50
(44) a. Ne lennan ket al lizher (Breton)
NEG read.1S NEG the letter
I dont read the letter

b. Gouzout a ran mat ne lennan ket al lizher
know PRT do.1S well NEG read.1S NEG the letter
I know I dont read the letter

As shown above, the properties of -features, tense and modality depend on
the choice of C; in a sense these features are shrunk in non-finite contexts: -
features and tense must be anaphoric and modality must be non-epistemic when
C = [-fin, qu/rel]. Negation differs from these Infl-like features in that it is not
shrunk in infinitival contexts (45).

(45) a. He prefers not to go out at night
b. He told me where not to buy cheese
c. He seemed not to be able to manage that problem
d. Hafdis knows where not to get gas (Bhatt 2006: 145)

Clearly, negation does not appear universally on the C-area: like other Infl-like
categories, it can also show up on the Infl-area. The chameleonic status of
sentential negation is again present in grammatical studies of negation. On the one
hand, several comparative studies, such as Pollock (1989) and Zanuttini (1997),
have focused on the position of negation in the Infl-area, and on the other hand,
Klima (1964) has attempted to substantiate the idea that negation is in a pres-
entential position. In fact, Lasnik (1972) and Laka (1990) have each defended, on
different grounds, that negation is generated both on the C-area and on the Infl-
area. Here I shall review a sample of syntactic properties of sentential negation
that demonstrate how this feature shares properties with wh-features. See Horn
(1989) and Horn & Cato (2000) for a review of the syntactical and logical analyses
of negation.
In English, negative phrases can move to the C-area and trigger subject-
auxiliary inversion, or Aux-to-C movement. This syntactic operation, which shows
that the type feature that defines the landing site of negative phrases operates from
a position of the C-area, is not general among nominal quantifiers. Observe the
following minimal pair, extracted from Rizzi (1990: 53).

(46) a. With no job would Bill be happy
b. *With some job would Bill be happy
THE RELATIONSHIP BETWEEN COMPLEMENTIZERS AND
INFLECTIONAL CATEGORIES
51

As observed by Rizzi (1990: 53), who follows Klima (1964), this is not a
characteristic property of negative phrases, but rather of the so-called affective
quantifiers (47), which are defined by the ability of licensing negative polarity
items (48).

(47) a. Few people did anything
b. *A few people did anything

(48) a. With few jobs would Bill be happy
b. *With a few jobs would Bill be happy

These syntactic patterns suggest that negation is a value of a broader feature
which also includes quantifiers such as few, and not an isolated feature. This
feature, which we call affectiveness, seems to be active on C in English, as wh-
features.
Chomskys (1957) analysis of the insertion of not/nt into the auxiliary verb
phrase in English suggests that negation is C-related. Chomsky claims that if
negation is not included in the kernel sentence but is a transformation that applies
to it, all instances of auxiliary support are accounted for in a very simple and
uniform way. Chomsky (1957) postulates that a transformation T
not
is responsible
for negation and that a transformation T
A
is responsible for emphatic affirmation.
The first transformation denies the kernel sentence (49) and the second
emphatically asserts it and requires an extra heavy stress (50).

(49) a. John doesnt arrive
b. John cant arrive
c. John hasnt arrived

(50) a. John DOES arrive
b. John CAN arrive
c. John HAVE arrived

Laka (1992) has shown that very similar patterns are observed in Basque,
where sentential negation and emphatic affirmation trigger auxiliary-movement
(51). Laka postulates that a more abstract feature is responsible for affirmation and
denial, which she terms , a label that attempts to suggest the notion of speech act.
heads the projection where V moves to to match sentential negation or emphatic
affirmation.
THE EMERGENCE OF ORDER IN SYNTAX 52

(51) a. Mari joan da (declarative sentence)
Mary left has
Mary has left

b. Mari ez da joan (negative sentence)
Mary NEG has left
Mary hasnt left

c. *Mari da joan (declarative sentence)
Mary has left
Mary has left

d. Mari da joan (emphatic affirmative sentence)
Mary has left
Mary has left

As Chomsky shows, the advantage of the transformational analysis becomes
clear once we observe that other instances of transformations that apply to kernel
sentences trigger auxiliary insertion. Transformations T
q
and T
w
, responsible for
total (52) and partial (53) questions, require auxiliary insertion and were con-
sidered to apply to kernel sentences.

(52) a. Do they arrive?
b. Can they arrive?
c. Have they arrived?
d. Are they arriving?

(53) a. What did John ate?
b. Where can they arrive?
c. When have they arrived?
d. Why are they coming?

If the particle so is the affirmative counterpart of neither, then emphatic
affirmation, one of the two possible values of , can also surface on C, like
affective-features, and trigger auxiliary insertion and auxiliary-subject inversion.
This possibility is also considered by Chomsky (1957), who postulates a T
so
,
analogous to T
neg
, T
A
, T
q
, and T
w
. Chomsky does not explicitly treat the particle
so as an affirmation particle, but he postulates a special transformation T
so
.
THE RELATIONSHIP BETWEEN COMPLEMENTIZERS AND
INFLECTIONAL CATEGORIES
53
Bearing in mind the opposition so/neither, it seems that we can account for the
syntax of these particles simply in terms of T
neg
/T
A,
with no empirical loss and a
gain in simplicity.



(54) a. John arrives and so do I

b. John can arrive and so can I
c. John has arrived and so have I

(55) a. John doesnt arrive and neither do I
b. John cant arrive and neither can I
c. John hasnt arrived and neither have I

These observations reveal that affirmation, negation and other affective
categories indeed constitute a sole feature, which may be called , adopting Lakas
terminology. can appear in a [+clause typing] position or in a [-clause typing]
position, and can trigger movement of -phrases and of auxiliaries to the C-area,
as well auxiliary insertion in the Infl-area.
It could be argued that Chomskys analysis of auxiliary insertion is no longer
valid, as the distinction between obligatory and optional transformations has been
long abandoned. In the model where the notion of kernel sentence was for-
mulated (see Chomsky (1955/75-1957), the proposed architecture of grammar
consisted of (a) a context-free rewriting base which generated phrase markers, and
(b) a transformational component that took phrase-markers as input (generated by
the base) and yielded new phrase markers. In this framework, a kernel sentence
was understood as a structure generated solely by obligatory transformations; op-
tional transformations took kernel sentences as input. A mono-clausal affirmative
active sentence was a kernel sentence, and the transformations of Passive and
Question were instances of optional transformations. Putting aside the accuracy
of the technical distinction between optional and obligatory transformations, and
the reasons for which the sketched model was abandoned, it seems that the syn-
tactic similarities between -features and wh-features are real, or even that these
two classes of instructions may belong to a broader class.
One way to conceive of the similarity between -features and wh-features is to
view denial, (emphatic) affirmation and query as three different values of a force
feature relative to a proposition. Thus, the illocutionary force of a proposition can
be emphatically affirmative, negative or interrogative. This allows us to account
for the syntactic similarities between -features and wh-features without needing
to resort to the dichotomy between optional and transformational operations.
THE EMERGENCE OF ORDER IN SYNTAX 54
Another legitimate qualm about the proposed version of Chomskys (1955/75-
1957) analysis is that negation and emphatic affirmation trigger auxiliary insertion
(47-48) into the Infl-area, whereas wh-features trigger auxiliary insertion into the
C-area, an areal distinction not present is Chomskys (1955/75-1957) framework.
This is not a real problem, since negation and affirmation, like wh-features (52,
53), can surface on C (40-44) and can trigger auxiliary raising to C (46, 51.b, 54,
55).
One might ask why in English the negative marker cannot appear in the C-
area, unlike negative phrases, if it conveys a force indication. This may be due to a
cross-linguistic parameter relative to the characterization of negative markers: (a)
certain languages, such as Celtic languages, present a negative marker lexically
defined as [+embedded] and another one as [-embedded], and thus negative
markers surface on C in both embedded and matrix clauses; (b) other languages,
suh as Latin, display negative markers specified as [+embedded], but not as
[-embedded]; in this case, a negative marker surfaces on C only in embedded
clauses (ne in Latin); (c) in other languages, such as Romance languages or Eng-
lish, negative markers are not specified as [embedded], and thus they are forced
to occur in Infl positions; and finally, (d) other languages display a negative mar-
ker defined as [+embedded], a negative marker defined as [-embedded] and a neg-
ative marker not defined as [embedded]; such is the case of Breton. As expected,
syntactic variation is reduced to the morpholexical specification of functional
categories.
From this we may conclude that the English negative marker has a hybrid
status: it is a head that conveys a value of the illocutionary force feature but does
not convey the [embedded] distinction. This second property forbids the negative
marker from appearing on C (only heads specified as [embedded] can surface as
complementizers), and thus forces the type feature responsible for hosting the neg-
ative marker to show up in an Infl-position in English. As argued above, this mor-
pholexical property distinguishes English negative markers from Celtic negative
markers. The first property accounts for auxiliary insertion, a syntactic operation
triggered by wh-features and -features, two subtypes of force features.
In sum, complementizers may display negative or affirmative forms, and
negation, along with other affective categories, can trigger V-to-C movement. If -
features (affectiveness, emphatic affirmation and negation) are viewed as a class of
illocutionary force features, along with wh-features, all instances of auxiliary in-
sertion can be attributed to a homogeneous class of features.



THE RELATIONSHIP BETWEEN COMPLEMENTIZERS AND
INFLECTIONAL CATEGORIES
55
2.2 Conclusion
The so called C-Infl link may be materialized in the following three properties
and generalized to various Infl-like feature such as -features, tense, mood, mo-
dality and negation:

Property (I): complementizers can replicate Infl-like features

-features, tense, mood and negation can surface on Infl-positions or on C-
positions. The complementizer of conditional clauses (if) has been argued to
encode a modal value for the proposition it introduces. Concretely, this comple-
mentizer marks the clause it heads as a restriction for a higher quantifier.

Property (II): there is a correlation between the characteristics of features that
surface on Infl and the choice of C

The characteristics of -features, tense and modality correlate with the choice
of C:

If C = [-fin], -features and tense are anaphoric
If C = [-fin, qu/rel], modality is not epistemic
If C = [+fin], -features and tense are free, and modality is full-fledged.

The classic Latin system of complementation has been argued to display a
clear and rich correlation between the mood of verbal endings and the choice of C.

Property (III): Infl-like features are involved in triggering V-to-C movement

Tense, mood and negation are involved in triggering V-to-C movement. It has
been argued that negation is not an isolated feature, but rather a value of a broader
head , along with affective categories and emphatic affirmation. Furthermore, -
features have been associated with wh-features through the claim that they both
convey illocutionary force distinctions. Consequently, all operations of auxiliary
insertion in English are triggered by force features. No evidence has been found
for the possibility that modality distinctions trigger V-to-C movement.







CHAPTER 3

DISCONTINUOUS SYNTACTIC PATTERNS

This chapter begins by considering the possibility pointed out by Chomsky (2005b) that
Infl-like features are base-generated on C and inherited by T/Infl; accordingly, the C-Infl
link would derive from inheritance mechanisms. This is called the Generalized Feature
Inheritance Theory, whose plausibility depends on the adequacy of the theory of phases:
there is an important asymmetry regarding the role that C and T/Infl play in syntactic
derivations; more precisely, as the Phase-Impenetrability Condition states, the former
head, but not the latter, is a phase head and defines its complement as inaccessible to
further probes. After reviewing the Phase-Impenetrability Condition, I conclude that there
is no clear reason for such an asymmetry, and thus the Generalized Feature Inheritance
Theory loses plausibility. The Phase-Impenetrability Condition is replaced by the Rela-
tivized Opacity Condition to account for subject islands, wh-islands and the requirement
whereby both A-movement and A-movement are successive cyclic. The C-Infl link is de-
rived from the assertion that C and Infl constitute a discontinuous syntactic pattern: C and
Infl display different polarities of the same lexico-semantic feature ([clause typing]). The
semantic instruction provided by Infl-like features (to trigger referential displacement) is
orthogonal to the [clause typing] distinction, and thus they are externally merged in both
poles of the discontinuity, without lowering or raising mechanisms. This reasoning is ex-
tended to account for the -V link and the P-K link.


3.1 The source of the C-Infl link
In the preceding chapter, I addressed the illustration of a traditional topic in
grammar: the relationship between complementizers and inflectional heads as to
-features, tense, mood, modality and negation. The principal objective of the
present chapter is to define the source of the C-Infl link. I shall take as a starting
point Chomskys feature inheritance theory.

3.2 Probe-goal relations in a phase
Chomsky (2005b) has recently provided interesting evidence for the idea that
C and T/Infl constitute a unit concerned with the timing of A- and A-movement in
the C phase. One of the many implications of Chomskys findings is that a phase
not only defines the syntactic domain that becomes impenetrable (due to transfer
THE EMERGENCE OF ORDER IN SYNTAX 58
operations), but also the level at which syntactic operations apply, since both -
features and peripheral features (a-features) of a phase probe their goal in parallel.

3.2.1 Subextraction (I): subject islands and -features. Chomskys reasoning be-
gins by addressing subject islands; as observed by Huang (1982), subextraction of
a wh-phrase from a subject DP contrasts with the subextraction of a wh-phrase
from an object DP.

(1) a. *about who did [stories about who] terrify John
b. about who did John read [stories about who]
a. *of who did [a critic of who] see you
b. of who did you see [a critic of who]

Chomskys new observation is that the surface subject position is not the
island, but the base-position SPEC-*, since island effects are attested solely with
agentive subjects, and not with subjects of passive or unaccusative verbs. That is,
subextraction of a PP from a non-agentive subject (3) patterns like subextraction
from an object (2), and not like subextraction from an agentive subject (4).

Subextraction from an object (Chomsky 2005b: 14)
(2) a. it was the CAR (not the TRUCK) of which they found the (driver,
picture)
b. of which car did they find the (driver, picture)

Subextraction from the subject of a passive verb (Chomsky 2005b: 14)
(3) a. it was the CAR (not the TRUCK) of which the (driver, picture) was
found
b. of which car was the (driver, picture) awarded a prize

Subextraction from an agentive subject (Chomsky 2005b: 14)
(4) a. *it was the CAR (not the TRUCK) of which the (driver, picture)
caused a scandal
b. *of which car did the (driver, picture) cause a scandal

The inactivity of SPEC-T is not a sufficient condition to account for these
patterns; if the key were the inactivity condition, subextraction from a subject
would always be unavailable, regardless of its base-position. Indeed, if a-features
DISCONTINUOUS SYNTACTIC PATTERNS 59
subextracted the PP from the DP when it is in SPEC-T, the contrast between
agentive and non-agentive subjects should be neutralized. To account for the
contrast between subextraction from agentive and from non agentive subjects, -
features and a-features must try to extract their respective tokens in parallel and
when they are in the base-position.
More precisely, * can subextract a wh-phrase (by pied-piping, a PP) from an
object DP because such a DP is in the search domain of *; however, a wh-phrase
cannot be subextracted from an external argument by * because the external
argument is outside the search domain of *. Consequently, when a type wh-
feature of C searches a suitable token, it can see a wh-phrase at the edge of *
(previously subextracted from an object DP by *), but not a wh-phrase that is a
constituent of an external argument when the phase head C is active. The work
done by the * phase head is thus crucial to explain the asymmetry between
subextraction from accusative DPs and nominative agentive DPs.
Subextraction from a non-agentive subject (3) may be accounted for in two
different ways:

(i) C can subextract the wh-phrase without problems from the domain of
def

because
def
does not constitute a phase boundary, and hence, intermediate
movement to the edge of
def
P is not necessary

(ii) subextracts the wh-phrase from the DP and relocates it in an edge
position, and thus C can have access to it.

Chomsky tends to adopt option (i) in his work, keeping as close as possible to
the ideal that those units with a complete set of -features are the units transferred
to the outer systems. However, the existence of an edge of
def
Ps has been
independently motivated by Legate (2003), who observes that passive and un-
accusative Ps offer reconstruction sites for wh-phrases. This being the case, we
must conclude that
def
subextracts a wh-phrase from its search domain and
relocates it in the edge of
def
P, exactly as * does.
Interestingly, a wh-phrase can be subextracted from an agentive subject in
raising and ECM constructions. This reveals that a probe can penetrate into a
SPEC that is in its search domain. However, a probe cannot penetrate into a SPEC
that belongs to the edge of a passed phase (SPEC-*). Thus, the source of subject
islands is search that goes too deeply into a phase already passed, not the
difference between base and surface position (Chomsky 2005b: 20).

THE EMERGENCE OF ORDER IN SYNTAX 60
(5) a. it is the CAR (not the TRUCK) of which [the (driver, picture) is likely
[t to cause a scandal]]]
b. of which car is [the (driver, picture) likely [t to [cause a scandal]]]
c. of which car did they believe the (driver, picture) to have caused a
scandal

In the derivation behind sentence (5), matrix C detects which car when the
DP the (driver, picture) of which car is in SPEC-T
def
. Since no phase is pen-
etrated, C can subextract the wh-phrase.
According to the above considerations, subextraction from SPEC-C should be
degraded. This prediction is verified by the following English and Catalan con-
structions. See Gallego (2006) and references cited therein for further evidence.

(6) a. *[
CP
[which car] did you ask [
CP
[which driver of which car] Peter
saw]]
b. *[
CP
[which director] did you ask [
CP
[which film of which director]
Peter bought]]
c. *[
CP
[who] did Mary ask [
CP
[which novel of who] you read]]

(7) a. *[
CP
[de quin cotxe] em vas demanar [
CP
[quin conductor de quin
cotxe] ha vist en Pere]]
b. *[
CP
[de quin director] em vas demanar [
CP
[quina pellicula de quin
director] ha comprat en Pere]]
c. *[
CP
[de qui] em va demanar la Maria [
CP
[quina novella de qui]
vas llegir]]

We ought to note that there is a potential problem for the idea that a
subconstituent at the edge of C cannot be readily probed: subextraction of a wh-
phrase
2
from a wh-phrase
1
is less degraded if wh-phrase
1
is a subject, a phe-
nomenon observed by Torrego (1985) founded on Spanish data. See also Chomsky
(1986b).

(8) a.
??
/*esta es la autora de la que varias traducciones han ganado
premios internacionales
This is the author by whom several translations have won
international awards


DISCONTINUOUS SYNTACTIC PATTERNS 61
b. de que autora no sabes qu traducciones han ganado premios
internacionales
By what author dont you know what translations have won
international awards

This issue has recently been reconsidered by Uriagereka (2004) and Gallego
(2006), extending Torregos original contrast and bringing several intricate factors
into consideration; Uriagereka argues that subextraction in (8.b or 9) is possible
because the embedded wh-phrase, being topicalized and having a de re reading, is
not a SPEC but an adjunct, which, in Uriagerekas model, allows for subex-
traction; if a de dicto reading is forced, as in (9), subextraction is degraded.
Uriagerekas suggestion that topics are adjuncts, and not SPECs, and that adjuncts
allow subextraction is not uncontroversial.

(9) de que autora no sabes qu traducciones estn a la venta
of what author not know.2SG what translations are.3PL at the sale
Of what author dont you know that what translations are on sale

# y de hecho dudas que haya ninguna
and of fact doubt that there is.SUBJ.3SG none
and actually you doubt that there is any?

Gallego, however, argues that the key factor is not the topical status of the de
re embedded wh-phrase, but the presence of the negative marker, which forces a
pressupositional/D-linked/de re interpretation. According to Gallego, if the nega-
tive marker of (10) is eliminated, subextraction is degraded.

(10) De que autora
??/
*(no) sabes qu traducciones estn a la venta?

It is worth stressing that Torregos original contrast, or its Catalan or English
counterparts, is quite weak, with some disagreement among speakers. The same
consideration applies to (9-10): although it may be true that in general the de re
reading of the embedded wh-phrase makes subextraction easier or that the pres-
ence of negation tends to facilitate the de re reading, I would resist the claim that
they are necessary conditions that determine whether or not extraction from a do-
main is possible.

3.2.2 The Generalized Feature Inheritance Theory. Chomsky (2005b) interprets
the parallel probing effects and the observation that the characterization of -fea-
THE EMERGENCE OF ORDER IN SYNTAX 62
tures on T correlates with C through what may be called a Feature Inheritance
Theory (FIT), which can be summarized as follows:

Feature Inheritance Theory
(i) both -features and a-features are generated on C
(ii) both -features and a-features operate in parallel
(iii) -features surface on T only derivatively
(iv) The feature inheritance mechanism has the function of bringing semantic
distinctions (arguments vs. operator-variable constructions) into a syn-
tactic representation
(v) C, as a phase head, has a central role in the generation of syntactic ob-
jects
(vi) In the lexicon, T lacks -features

Note that the conclusion that -features and a-features of a phase work in
parallel (ii) does not necessarily lead to the idea that -features are generated on C;
the attested patterns of subextraction do not have any implication on where fea-
tures are generated but only on the ordering of probing operations. The possibility
that -features were generated on T would be compatible with the evidence of
subextraction provided above, with the proviso that -features operated in parallel
with a-features, since they belong to the same domain where syntactic operations
apply, i.e., to the same phase. In fact, the FIT is a consequence of the theory of
phases. If C, and not T, has a central role in the generation of syntactic objects (v),
and the value of T as to -features correlates with the choice of C, the simplest
explanation is that -features, like a-features, are base-generated on C (i). Accord-
ingly, the probes of both A- and A-movement are generated in the same category,
the phase head, and -features surface only derivatively on T (iii), T being a lex-
ical item generated without -features (vi); the derivational device responsible for
spreading -features from C to T is an inheritance mechanism, which yields se-
mantically specialized positions: according to Chomsky, SPEC-T becomes spe-
cialized for arguments and SPEC-C for operators (iv). Thus, the source of feature
spreading is a C-I requirement, in the spirit of the SMT.
There are several other conceivable implementations for the C-T link, such as
selection rules or coindexing algorithms; however, if there is independent reason
for the idea that C, and not T, has a special generative status, these alternatives are
just unnecessarily complicated ways of saying that -features are generated on C
and spread to T in order to yield semantically dedicated positions.
DISCONTINUOUS SYNTACTIC PATTERNS 63
The -V system appears to have the same properties as the C-T system.
Ideally, -features inheritance should be a property of phase heads, and thus *
should also be generated with -features, which should be inherited by V. If
SPEC-V, and not SPEC-*, offers an A-position for an accusative DP, as T offers
an A-position for a nominative DP, then, as Chomsky (2005b) observes, the
intriguing but puzzling conclusions about raising of object to SPEC-V (p. 15) can
be accounted for automatically. Chomsky refers to an array of observations by
Postal (1974) and further discussed by Lasnik & Saito (1991) and Lasnik (2001,
2002), whereby a verb precedes an object which in turn c-commands an adjunct.
Patterns in (26, 27), taken from Lasnik (2001), illustrate that ECM constructions
provide a DP with a position high enough to bind an anaphor and to license an NPI
into a VP adjunct; as expected, when the relevant DP is in SPEC-T of the
embedded finite sentence, it fails to bind an anaphor or to license an NPI into the
adjunct of the upper VP. These contrasts imply that the relevant DP moves in
ECM constructions to a position higher than the VP adjunct of the matrix clause.

Condition A satisfaction
(11) a. The DA proved [two men to have been at the scene of the crime]
during each otherss trials (ECM)

b.
?
*The DA proved [that two men were at the scene of the crime]
during each others trial (finite embedded sentence)

NPI licensing
(12) a. The DA proved [noone to have been at the scene] during any of the
trials
b.
?
*The DA proved [that noone was guilty] during any of the trials

Note that DP objects of transitive verbs, which do not involve raising from the
embedded clause to the matrix clause, can also bind an anaphor or license an NPI
inside a VP adjunct.

(13) a. The DA accused two men during each others trials
b. The DA cross-examined none of the witnesses during any of the
trials

THE EMERGENCE OF ORDER IN SYNTAX 64
This phenomenon supports the idea that the verbal phrase contains a position
for case licensing. This position would host the object DP of a transitive predicate
as well as the DP moving from the non-finite embedded clause, which can be
agentive, in an ECM construction. This accounts for the observation that these DPs
occur higher than a VP adjunct.
The exact position that objects occupy has not yet been defined. The two
possibilities are an extra SPEC-* and SPEC-V, both higher than the VP adjunct.
Since a verb precedes the accusative DP and it cannot be as high as T, the most
reasonable account is that V raises to * and the DP to SPEC-V, in virtue of the
inheritance mechanisms previously formulated for the C-T system, which are
responsible for offering an A-position for a DP. Thus, V inherits -features from
* as T inherits -features from C*.

(14) *P

[V
i
-*] VP

DP
j
VP

Adjunct VP

V
i
DP
j

This analysis is in line with Koizumis VP-split (Koizumi 1995), according to
which the AGR-Obj projection is tucked into two VPs, a higher one, which hosts
the external argument, and a lower one, in which internal arguments are base-
generated.
The FIT developed by Chomsky (2005b) for -features can be extended to
several Infl-like features, as Chomsky himself points out, since the value of several
Infl-like features correlates with the choice of C, as illustrated in chapter 2. This
leads to what we may call the Generalized Feature Inheritance Theory:

(15) Generalized Feature Inheritance Theory (GFIT)
(i) C has a central role in the generation of syntactic objects
(ii) Infl does not have a central role in the generation of syntactic objects
(iii) Infl-like features are generated on C
(iv) In the lexicon, Infl lacks Infl-like features

DISCONTINUOUS SYNTACTIC PATTERNS 65
(v) Infl-like features surface on Infl only derivatively
(vi) The feature inheritance mechanism has the function of bringing semantic
distinctions into a syntactic representation

The GFIT, like the FIT, is a sub-theory of the broader theory of phases: if there
are independent reasons to think that C, and not Infl, defines a phase boundary (i-
ii), as it can be demonstrated that C, and not Infl, has a central role in the gen-
eration of syntactic objects (the essence of the theory of phases), and it is observed
that Infl-like features are linked to C (they can surface on C and on Infl, their char-
acteristics correlate with the choice of C, and they can trigger V-to-C and V-to-Infl
movement), the simplest account for the C-Infl link is that the head argued to have
a central role in the generation of syntactic objects is also the locus where Infl-like
features are generated (iii). If this is the case, two syntactic properties, namely the
locality of syntactic operations and the C-Infl link, can be attributed to the phase
head, which defines both the syntactic domains to be transferred and the inter-
mediate steps in long movement, as well as the locus where functional features are
generated. Thus, Infl-like features are inherent to C and appear only derivatively
on Infl heads (v), which lack them in the lexicon (iv). Following the ideal of the
SMT, the inheritance mechanism responsible for spreading Infl-like features from
the phase head C to an Infl head should be related to the legibility requirement of
bringing semantic distinctions into the syntactic representation (vi).

3.3 Revising the Generalized Feature Inheritance Theory
In order to determine whether or not the C-Infl discontinuity is the result of
inheritance mechanisms that take place from C to Infl, I shall revise first the status
of T/Infl in the GFIT, and second, the status of C in the theory of phases. These are
the propositions of the GFIT to be revised:

(16) (i) C has a central role in the generation of syntactic objects
(ii) Infl does not have a central role in the generation of syntactic
objects
(iii) Infl-like features are generated on C
(iv) In the lexicon, Infl lacks Infl-like features

Propositions (16.iii) and (16.iv) are the core of the GFIT, whereas propositions
(16.i) and (16.ii) are the core of the theory of phases. In section 3.3.1.1, I shall
address proposition (16.iv). In section 3.3.1.2 I shall examine proposition (16.iii),
whose plausibility depends on the accuracy of the theory of phases. Consequently,
propositions (16.i) and (16.ii) will be investigated in section 3.3.2.
THE EMERGENCE OF ORDER IN SYNTAX 66

3.3.1 The status of the phase head C and of Infl heads

3.3.1.1 Infl. There are several open questions concerning the lexicon. For example,
it is unclear whether the lexicon must contain redundant information which must
be independently present in the outer systems (as selectional restrictions or cate-
gorial specification for substantive categories), or if, instead, it must be a unique or
distributed component (see Bonet 1991, Hale & Marantz 1993 and Marantz 1997,
among others). However, a very common view which seems to be the minimal
assumption, is that the lexicon must contain at least what cannot be derived from
principles of the syntactic component or interface conditions, i.e., idiosyncrasies,
as argued by Chomsky (1995).
At first glance, this line of argumentation casts some doubt on the legitimacy
of postulating a category like T, or more generally Infl, since it would not satisfy
the minimal requirement of providing genuine instructions: the features Infl shows
up with are not carried from the lexicon, but are inherited from the phase head C
during syntactic derivations. The semantic instructions that appear on Infl are at-
tributable to the existence of a device of the syntactic component, the inheritance
mechanisms, and not to the alleged lexical item Infl. Thus, if the features that ap-
pear on Infl do not come from the lexicon but are due to the syntactic mechanisms
of inheritance, it can be concluded that Infl is not a proper lexical item.
One might argue that Infl is a legitimate category in the lexicon in a vacuous
sense: Infl is not originated with semantic instructions, but serves to inherit fea-
tures and bring semantic distinctions into syntactic representations. Thus, it would
be a continent category. But all this is just an unnecessary complicated way of say-
ing that inflectional categories are syntactic elements, not symbols of the lexicon.
Even if Infl heads are not originated with Infl-like features such as -features,
tense, mood, modality or negation, a more careful inspection reveals that they pro-
vide a useful instruction for the C-I system. As has argued repeatedly in this study,
C can be lexically defined as [+clause typing] and bearing the [embedded] dis-
tinction, whereas Infl can be lexically defined as [-clause typing] and lacking the
[embedded] instruction. Consequently, Infl heads lack Infl-like features in the
lexicon (16.iv), but provide inherent semantic instructions different from those
provided by C. Therefore, the GFIT should be slightly modified, reformulating
proposition (16.iv) as (16.iv).

(16') (iv) In the lexicon, Infl lacks Infl-like features, but it is defined as [-
clause typing]

DISCONTINUOUS SYNTACTIC PATTERNS 67
This modification has a certain propaedeutic value, for it helps to understand
why inheritance mechanisms should apply not only to -features, but to Infl-like
features in general. Recall that, in the case of -features, Chomsky (2005b) argues
that feature inheritance is related to the semantic requirement of distinguishing
between a position suitable for an operator to bind a variable (an A-position) and
a position suitable for a DP to instantiate -features (an A-position). In the general
case of Infl-like features, an Infl-like feature should be inherited from C to Infl in
order to associate the relevant Infl-like feature with a [-clause typing] position
unspecified for the [embedded] distinction. Keeping to the SMT, the mechanisms
of feature inheritance are a solution to the C-I requirement by providing Infl-like
features with a [-clause typing] position.

3.3.1.2 C. The empirical basis of the theory of phases is that two alleged properties
of the syntactic component, cyclicity and successive cyclicity, are relativized to C
and , the so-called phase heads. Accordingly, C, and not Infl, will play a central
role in the derivation of syntactic objects if the former, and not the latter, is
responsible for defining the domains to be mapped from the syntactic component
to the outer systems. In section 3.3.2 I shall discuss the general motivation for
postulating syntactic cycles or phases in the theory of the syntactic component,
and more particularly, on the motivation for claiming that C, and not Infl, defines
syntactic cycles or phases, as the Phase-Impenetrability Condition states.

3.3.2 Revising the Phase-Impenetrability Condition

3.3.2.1 Cyclicity. The intuition under the property of cyclicity is that syntactic re-
lations hold first between the most deeply nested element and its domain, then
between the next least nested element and its domain, and so on. This derivational
constraint is introduced into syntactic theory by Chomsky (1965) through the
Transformational cycle:

(17) Transformational Cycle
In a derivation, for all syntactic domains o in a phrase marker, a linear
sequence of transformations applies to a domain o
i
before applying to o
j
,
where o
j
contains o
i
.

Accordingly, derivations should proceed in a bottom-up direction, from the
most deeply nested domain to the least deeply nested one. Transformational rules
THE EMERGENCE OF ORDER IN SYNTAX 68
cannot apply to a domain o
j
unless they have applied to a domain o
i
embedded un-
der o
j
in a previous derivational step.
Note that the Transformational Cycle does not predict if a domain or cycle can
be tampered once it has been passed. Suppose that in derivation (18) probe y has
extracted a constituent
1
from its domain yielding the derivational stage
1
, and
that probe has subsequently extracted
1
from its domain mapping
1
onto
2
.
When the derivation is at
2
, the domain has already been passed. The question
is, thus, whether the derivation can go back to
1
so that the probe y extracts the
element
2
from and creates
3
.

(18)
1
= [
1
[y, [

1
,
2
]]]

2
= [
1
[ [

1
[y, [

1
,

2
]]]]]

3
= [
1
[ [

1,

2
[y, [

1
,

2
]]]]]

Such a derivation is prohibited in Chomsky (1973: 243) under the Strict Cycle
Condition, which sharpens the Transformational Cycle.

(19) Strict Cycle Condition (SCC)
No rule can apply to a domain dominated by a cyclic node A in such a
way as to affect solely a proper subdomain of A dominated by a node B
which is also a cyclic node

The important empirical question was (and still is) to determine which nodes
define a cycle. First, C (or S, in previous models) was responsible for defining the
syntactic cycles, but for several reasons it was soon suggested that N (see, for
instance, Akmajian 1975) or P (Riemsdijk 1978a, b) might also define syntactic
cycles. According to the theory of phases, C* but not Infl, and * but not V, are
the cyclic nodes that define its domain as inaccessible once it has been passed.
According to the theory of phases, this is not due to the Strict Cycle Condition, but
rather to the existence of cyclic mappings from the syntactic component to the
external systems, which eliminate a domain or render it invisible to further
syntactic computations.
The intuition has been that cyclicity, under any of its formulations, is a rather
plausible principle of efficiency that a computational device as the syntactic com-
ponent should satisfy. Under the more precise version of the Strict Cycle Con-
dition, both the number of possible derivations and the size of the structural span
that must be active in the syntactic workspace are minimized, since a syntactic
domain is forgotten once it has been passed. It is also worh noting that there are
DISCONTINUOUS SYNTACTIC PATTERNS 69
good reasons for thinking of cyclicity as an important principle for phonological
computations, for which it was originally postulated. As Lasnik (in press) observes
in his revision of the several conceptions of the cycle (see also Kean 1974 and
Mascar 1976 for early discussion on the phonological cycle):

The cyclic principle, along with analysis into distinctive features, was one of two major
syntactic innovations of the mid 1960s borrowed from phonology. The principle was first
formulated in Chomsky et al. (1956), and applied in the phonological analysis of a variety
of languages from the early 1960s on. Chomsky & Halle (1968), the monumental ex-
amination of the phonology of English, made crucial use of the cycle in virtually every
analysis.

But as Freidin (1978) argued in a very important paper, the empirical support
for the existence of syntactic cyclicity can be accounted for through other condi-
tions independently motivated. This issue is more recently reconsidered by Freidin
(1999) within a minimalist framework and by Lasnik (in press). In 3.3.1.1 I shall
revise Freidins (1978) observations about the classical empirical argument for
strict cyclicity, and in 3.3.1.2 I shall focus on successive cyclicity, another partic-
ular facet of cyclicity.

3.3.2.1.1 The classical empirical argument for strict cyclicity. Ill-formed syntactic
outputs that adhere to the following derivational scheme (adapted from Freidin
1978: 531) provide empirical support for the Strict Cycle Condition. X and Y are
categories and o
1
and o
2
consecutive cycles.

(20)
1
= [
o1
[X] [Y] ]

2
= [
o2
[X] [
o1
[ ] [Y] ] ]

3
= [
o2
[X] [
o1
[Y] [ ] ] ]

The derivational step mapping
1
onto
2
unproblematically displaces X from a
position in o
1
to a position in o
2
. The derivational step that moves from
2
to
3

violates the SCC, since an operation affects only the syntactic subdomain o
1
of o
2
.
Let X and Y be two wh-phrases base-generated in the same embedded C*P, and
consider the ill-formed sentence (21) and the partially detailed derivation (22),
using for the moment traces and indexes instead of copies to keep track of
Freidins arguments.

(21) *What did she wonder where John put?

(22)
1
= [
*P
John put what where]
THE EMERGENCE OF ORDER IN SYNTAX 70

2
= [
C*P
what
i
[
*P
John put [t
i
] where]]

3
= [
C*P
what
i
did she wonder [
C*P
t
i
John [
*P
t
i
put where]]]

4
= [
C*P
what
i
did she wonder [
C*P
where
j
John [
*P
t
i
put t
j
]]]

The derivation (22) violates the SCC, which arguably causes the ill-formed-
ness of sentence (21). More concretely, the derivational step that maps
3
onto

4
is
acyclic, since the operation of moving where to the embedded SPEC-C* only af-
fects a proper subdomain, the embedded C*P, dominated by the matrix C*P. Ac-
cording to this analysis, C* heads are cyclic nodes. Note that this class of ill-
formed sentences cannot be banned by the weaker Transformational Cycle.
Freidins principal observation is that the class of derivations that violate the
SCC can be banned by other principles that have independent empirical moti-
vation. Specifically, two opacity conditions for anaphora binding, the Tensed-S
Condition and the Specified Subject Condition, serve this purpose.

(23) Tensed-S Condition (TSC)
In a structure
X [
o
Y ]
where o is a tensed sentence, X may not properly bind Y (Freidin 1978:
527)

(24) Specified Subject Condition (SSC)
In a structure
X [
o
Y ]
where o contains a subject distinct from Y and not controlled by X, X may
not properly bind Y (Freidin 1978: 528)

The independent motivation of these conditions is, respectively, the inability of
a subject DP to penetrate an embedded tensed sentence to bind a subject anaphor
or an empty category, or to cross over a distinct and not controlled DP to bind an
object anaphor or an empty category.

TSC violations
(25) a. *John
i
thought that himself
i
was clever
b. *They
i
expected that each other
i
would win
c. *Jill
i
was reported that e
i
had won a prize

DISCONTINUOUS SYNTACTIC PATTERNS 71
SSC violations
(26) a. *John
i
believed Mary to like himself
i

b. *We
i
expected John to consult each other
i

c. *Jill
i
was reported Jack to have insulted e
i

In the light of these two conditions, consider the contrast given in (27), from
Freidin (1978: 529):

(27) a. [
C*P
who
i
John said [
C*P
t
i
Fred saw t
i
]]
b. [
C*P
who
j
John knew [
C*P
what
k
[t
k
frightened t
j
]]]

Structure (27.a) is an acceptable syntactic output, whereas structure (27.b) is
not. According to Freidins reasoning, the binding relation between who
j
and its
trace [in the ill-formed (27.b)] violates the TSC and the SSC, whereas in the well-
formed (27.a) there is a trace t
i
in the embedded SPEC-C* that suffices to bind the
trace in the base-position.
Freidin extends this line of reasoning to other ill-formed syntactic outputs
which were attributable to the SCC and concludes that they all can be accounted
for in terms of the TSC and the SSC, or rather, in terms of a more general Opacity
Principle, defined as follows.

(28) Opacity Principle
In a structure
[X [
u
Y ]
where: a. o = S (or NP), and
b. Y is not bound to any c-commanding category in o (linking
convention)
if Y is in the domain of
i. a finite clause (TSC), or
ii. a subject not controlled by X (SSC)
then X may not properly bind Y (Freidin 1978: 530)

This principle treats S (and N) not as a cyclic node that plays a crucial role in
the definition of the syntactic material available to syntactic operations, but as an
opacity inducer that defines its domain as non-transparent to a category c-com-
manding an anaphora. Such opacity impedes a DP from binding an anaphor or an
empty category and a wh-phrase from binding its trace in the base-position across
a S.
THE EMERGENCE OF ORDER IN SYNTAX 72
In sum, as the SCC has no independent empirical motivation, there is no
reason to stipulate it as a part of the theory of grammar (Freidin 1978: 539). A
theory of grammar that avoided the generation of sentences like (21) would be
excessively precise since these syntactic objects can be banned by representational
conditions or, in minimalist terms, by legibility conditions. This leads Freidin to
argue that derivational restrictions such as the SCC are redundant and that con-
ditions on representations are the decisive factor.
Reconsider the derivational step that maps
3
onto
4
in the derivation (22),
repeated in (29).

(29)
1
= [
*P
John put what where]

2
= [
C*P
what
i
[
*P
John put t
i
where]]

3
= [
C*P
what
i
did she wonder [
C*P
t
i
John [
*P
t
i
put where]]]

4
= [
C*P
what
i
did she wonder [
C*P
where
j
John [
*P
t
i
put t
j
]]]

It is crucial for Freidins analysis that in
4
the wh-phrase where is relocated
in the position where the wh-phrase what occurred in
2
(embedded SPEC-C*),
which results in the erasure of the bound trace t
i
. It is precisely the erasure of such
an intermediate trace that disallows the wh-phrase what to bind its trace in the
base-position in
4
. Here the assumption is that C*, and any other head category,
can have no more than one SPEC. This has been the general view for a long time,
and the keystone for the analysis of wh-island effects, but indeed, as Chomsky has
observed, there is no reason to think that a recursive operation like Merge should
be restricted in such a way that it applies only twice to a sole target, which would
create syntactic objects as (30.a), but successive applications of different items to a
sole target being prohibited by stipulation (30.b).

(30) a. [
HP
Y [
HP
H X]]
b. [
HP
Z [
HP
[
HP
Y [
HP
H X]]]]

Nonetheless, since the internal or external applications of Merge are linked to
properties of heads, it is expected that, in actual derivations, they become re-
stricted. A further expectation is that languages vary in the number of SPEC
positions, i.e., that different heads allow for different numbers of SPECs in dif-
ferent languages. Both expectations seem to be corroborated empirically: firstly,
several syntactic phenomena suggest the possibility that multiple features are
instantiated in different SPEC positions of a sole head in the relevant languages
(Ura 2000), and secondly, several languages display overt fronting of multiple wh-
DISCONTINUOUS SYNTACTIC PATTERNS 73
phrases to multiple SPEC positions of C*, repeatedly instantiating a sole wh-
feature (see Richards 1997, 1999, Boskovic1999).
The second abovementioned case of multiple SPECs is particularly relevant to
the study of wh-islands: a rather sensible prediction is that languages with multiple
wh-fronting display weaker wh-islands effects or lack them altogether. This seems
to be confirmed by some evidence from Bulgarian, a language with wh-fronting,
provided by Richards (1997: 40).

(31) a. Koja kniga te popita ucitelja kogo ubedi Ivan da
which book you asked teacher who convinced Ivan to
publikuva?
publish
Which book did the teacher asked you who Ivan convinced to
publish?

b. Koj izdatel te popita ucitelja kakvo ubedi Ivan da
which publisher you asked teacher what convinced Ivan to
publikuva?
publish
Which publisher did the teacher ask you what Ivan convinced to
publish?

These well-formed sentences do not violate Freidins Opacity Principle, as the
embedded C* in Bulgarian provides enough structural space for the two wh-
phrases (one SPEC for each wh-phrase), which avoids the erasure of a bound trace.
Therefore, the ability of a [wh] type features to trigger fronting of multiple wh-
phrases is related to the obviation of the wh-island effects.
One might consider the possibility that, in languages like Bulgarian, multiple
wh-fronting is a syntactic correlate of the semantic operation of absorption
(Higginbotham & May 1981). Thus, the [wh] type feature in Bulgarian would be
specified as [+absorbent] and the [wh] type feature of English as [-absorbent], with
the caveat that the [absorbent] distinction that refers to the [wh] type feature does
not reflect the absence or presence of a semantic operation but rather of its
syntactic correlate. Note that, apart from the adequacy of this suggestion, it would
be a mistake to attribute the absence of wh-island effects particularly to the
[+absorbent] specification of embedded wh-features, since the wh-phrase that
moves to matrix SPEC-C* does not undergo absorption in the embedded C*P, as it
is illocutionarily interpreted in the matrix C*P. This leads to the triggering
problem of intermediate copies, which will be discussed in 3.3.2.1.2. For the
THE EMERGENCE OF ORDER IN SYNTAX 74
moment, let us say that the C* head allows multiple SPECs in Bulgarian, but not
in English, a conclusion we cannot easily avoid once we keep in mind that the
former language, but not the latter, displays multiple wh-fronting constructions and
lacks wh-island effects.
There is a further assumption in Freidins revision of the SCC, namely, that a
wh-phrase
2
moves to the embedded SPEC-C* after a wh-phrase
1
has raised from
the embedded SPEC-C* to the matrix SPEC-C*, resulting in the erasure of an
intermediate trace or in a counter-cyclic operation. But recall that the patterns of
subextraction from a subject DP recently discovered by Chomsky lead to the con-
clusion that the a-features that trigger A-movement to SPEC-C* work in parallel
with the -features that trigger A-movement to SPEC-T; if these two operations
did not take place in parallel, we could not account for the observation that a
subject DP is an island for wh-subextraction only if the SPEC of a *P is pen-
etrated. Thus, there is reason to think that two probing relations of the same phase
take place during the same derivational step, driven by the phase head, according
to Chomskys implementation. If A-movement and A-movement take place in
parallel at the same phase, it seems reasonable, or even obligatory, to assume that
two A-movement operations at the same phase take place also in parallel. Hence,
wh-phrase
1
and wh-phrase
2
should move in parallel to two different SPEC posi-
tions of the embedded C*, just as a subject DP moves to SPEC-T and a a-phrase to
SPEC-C* in parallel.
Accordingly, the ill-formedness of English sentences like *what do you
wonder where John put? cannot be attributed to the possibility that where de-
letes the trace or copy of what of the embedded SPEC-C*, but rather to the
inability of (embedded) C* to attract more than one wh-phrase in English. Ac-
cordingly, in English, the syntactic properties of C* are the source of wh-islands,
or in other words,
3
is ill-formed in derivation (32) because in English C* cannot
map
1
onto
2
. Bulgarian C*, arguably, displays the ability to trigger multiple
parallel wh-movement.

(32)
1
= [
*P
John put what where]

2
= [
C*P
[what] [where] [
*P
John put [what] [where]]

3
= [
C*P
[what] did she wonder [
C*P
[what] [where] [
*P
John put [what]
[where]]

Recall that Freidins observation was that the classical empirical evidence for
the SCC, a constraint on derivations, could be accounted for in terms of an Opacity
Principle relative to representations, which had independent motivation. Now, it
seems that the ill-formed sentences that constitute the classical empirical evidence
DISCONTINUOUS SYNTACTIC PATTERNS 75
for the SCC do not involve any derivational step that violates the SCC, due to
parallel probing at the phase level. The real source of wh-island effects is the in-
ability of C* to provide a SPEC position for each of the two wh-phrases mapping

1
onto
2
.
From the preceding considerations we may conclude that the SCC, a particular
formulation of the syntactic cycle, has no empirical foundation. The classical e-
vidence for such a derivational principle is not valid because it does not adhere to
the hypothetical derivational scheme (20), repeated in (33).

(33)
1
= [
o1
[X] [Y] ]

2
= [
o2
[X] [
o1
[ ] [Y] ] ]

3
= [
o2
[X] [
o1
[Y] [ ] ] ]

As probe-goal relations take place in parallel at the phase level, X and Y would
be attracted during the same derivational step.
For our concerns, it is important to stress that there seems to be an asymmetry
in the role that C and Infl/T play in the derivation of syntactic objects: C, and not
Infl, defines its domain as inaccessible to further probes. Firstly, we have evidence
to conclude that the domain of T/Infl is not a cycle or a phase domain, since C*
extracts its goals from SPEC-, which is in the domain of Infl; and secondly, a
matrix C* can probe a wh-phrase only if it is in a SPEC of the embedded C*P, but
not if it is in the domain of the embedded C*P.
The relevant derivational condition that expresses the asymmetry between C
and T/Infl is the Phase-Impenetrability Condition:

(34) Phase-Impenetrability Condition (PIC)
Consider a Phase PH = [o, [H, ]], H being the phase head
Call o and H the edge of PH, and the domain of H

The domain of H is not accessible to syntactic operations beyond PH,
only the edge, {o, H}, since Transfer sends to C-I and A-P once H has
terminated its work (adapted from Chomsky 2001b: 5-6).

Therefore, the PIC defines an asymmetry between C* and T/Infl in the role
they play in the generation of syntactic objects: C*, but not T/Infl, defines its
domain as impenetrable, since C is a phase head.

THE EMERGENCE OF ORDER IN SYNTAX 76
3.3.2.1.2 Successive cyclicity. An important property of internal Merge is that a
moving goal reaches its final landing site by means of shorter steps, successively
passing through intermediate positions, which work as escape hatches, and leaving
a copy in them. This raises the question of whether these intermediate positions for
successive movement are provided by the phase head C and not by T/Infl. If that
were the case, the categories that defined its domain as inaccessible to further
probes would be the categories that provide escape hatches, a natural expectation,
since escape hatches allow a category to move towards a further landing site.
The evidence for successive A-movement is compelling, and will not be
revised here. Both internal and external considerations strongly suggest that each
C offers an intermediate SPEC position for a wh-phrase that undergoes apparently
long movement. With respect to internal considerations, recall that a wh-phrase
cannot move from a CP whose SPEC is filled by another wh-phrase in languages
that do not allow for multiple wh-fronting. With respect to external considerations,
intermediate copies are active at SEM, as binding effects reflect (Barss 2001), and
they can have a morpho-phonological correlate in several languages (Felser 2004).
It is also worth noting that intermediate A-copies are spelled out in early child
English, although the external evidence that children are exposed to does not
overtly display intermediate copies (De Villiers et al. 1990, McDaniel et al. 1995).
Interestingly, there is reason to think that successive cyclic movement of a wh-
phrase must leave a copy not only at every CP phase but at every P phase as well.
These copies display binding effects (see Fox 2000 and Legates 2003 subsequent
extensions to non-agentive predicates) and morpho-phonological effects (Cole &
Hermon 2000 and Saddy 1991).
There are three types of phenomena that argue that A-movement is comprised
of shorter steps: quantifier stranding, subextraction from subjects and binding
effects.
In Standard English, the quantifier all can either be pied-piped to the matrix
SPEC-T (35.a) or stranded in different positions; in (35.a) and (35.b), all is
stranded in positions contiguous with embedded non-finite verbs, arguably in
SPEC-T
def
, and in (35.c), it is stranded in a postverbal position, inside the P. This
is proposed by Boeckx (2003), in his elaboration of McCloskeys work on quan-
tifier float in successive A-movement based on West Ulster English (McCloskey
2000).

(35) a. All the boys seem to appear to like ice cream
b. The boys seem all to appear to like ice cream
c. The boys seem to appear all to like ice cream
d. The boys seem to appear to all like ice cream
DISCONTINUOUS SYNTACTIC PATTERNS 77
In accordance with Sportiches (1988) analysis of quantifier float, this pattern
suggests that the complete quantificational phrase all the boys moved from the
base-position to its final landing site by passing through several SPEC-T
def
po-
sitions, where the quantifier can be stranded.
Recall that, as observed in Chomsky (2005b), subextraction of a wh-phrase
from an agentive subject is possible in raising constructions.

(36) a. *it was the CAR (not the TRUCK) of which [the (driver, picture)
[t caused a scandal]]
b. *of which car did the driver cause a scandal

(37) a. it is the CAR (not the TRUCK) of which [the (driver, picture) is likely
[t to cause a scandal]]
b. of which car is [the (driver, picture) likely [t to [cause a scandal]]]

To account for the observation that the PP of which can be extracted from the
external argument the (driver, picture) of which in (37) but not in (36) it is nec-
essary for the probing C* in (37) to detect the PP when the external argument is
not in the SPEC-*. There are two possible positions: embedded SPEC-T
def
and
matrix SPEC-T. There is reason to think that matrix

SPEC-T cannot be probed by
matrix C*: if it were, the thematic characterization of the subject would not be a
relevant factor in determining subextractability patterns where no SPEC-T
def
is in-
volved. Thus it seems that patterns (37) require that A-movement be successive
cyclic.
Standard conceptions of binding principle A also suggest the necessity of
successive cyclic A-movement. Consider the following paradigm, noted by Danny
Fox, which plays a decisive role in the discussion of the successive cyclic status of
A-movement.

(38) a. John
i
seems to Mary to appear to himself
i
to be happy
b. *Mary seems to John
i
to appear to himself
i
to be happy

An important factor in the analysis of (38) is that the experiencer NP
embedded inside the PP has the ability to bind an anaphor not dominated by the PP
(see Boeckx 1999 and references cited therein); note that, in (39.a), the NP John
cannot be correferential to the experiencer him, and that, in (39.b), the anaphor
himself is bound by the experiencer John, a further argument for the necessity
of A-reconstruction.

THE EMERGENCE OF ORDER IN SYNTAX 78
(39) a. They
i
seem to him
j
[to like John
k/*j
]
b. [Pictures of himself
i
] seem to John
i
[[pictures of himself
i
] to be
ugly]

By this reasoning, A-reconstruction in (38) is necessary in order to allow that
the experiencer Mary does not trigger intervention effects in (38.a) and to dis-
allow that the experiencer John in (38.b) binds the anaphor in the embedded
sentence.
Therefore, both internal and external considerations suggest the necessity of
successive cyclic A-movement. For internal considerations, intermediate positions
of A-movement are necessary to account for the observed patterns of subextraction
from agentive subjects in raising constructions (36-37), and for external con-
siderations, a part of the intermediate copy is spelled out in the case of quantifier
float (35) and the binding effects illustrated in (38-39) require the matrix subject to
bind the anaphor from the SPEC-T
def
position.
This leads us to the conclusion that the PIC is not a necessary condition for
successive cyclicity: T
def
, a category argued not to be a phase head, and whose
domain should be accessible to further operations, forces a moving DP to pass
through its SPEC.
One could claim that the successive cyclic nature of A-movement and of A-
movement derive from independent sources: the former is a consequence of the
PIC, and every intermediate step is triggered at the level of every phase, whereas
the latter is a consequence of a condition on chain formation that requires steps to
be as local as possible, in a Takahashi/Boeckx-style (Takahashi 1994 and Boeckx
2003): a category X undergoes movement when its final landing site is present in
the structure, by passing through all intermediate positions of type X to minimize
the chain links.
But there is no reason to stipulate that successive cyclic A- and A-movement
are caused by two different economy conditions, a uniform account being the
preferable one. Optimally, intermediate links are triggered either in a stepwise
manner or in Takahashi/Boeckx-style for A-movement and for A-movement
alike. In this sense, note that the subextractability from an agentive subject in rais-
ing constructions is unexpected under the Takahashi/Boeckx-approach: if a PP
remains inside the DP in the base-position and undergoes movement only when
(matrix) C is merged, sentences (36) and sentences (37) should be equally de-
graded.
The best way to account for successive cyclicity arises from the following two
observations (see Rizzi 2002, 2004 for very similar ideas):

DISCONTINUOUS SYNTACTIC PATTERNS 79
(40) a. a moving category with a [+token] feature X passes through
intermediate positions offered by categories with a [+type] feature
X

b. the [+type] feature X that offers the intermediate position does not
suffice to host the category with a [+token] feature X

Lets take a closer look at observation (a). A phrase with a marked value for a-
features (a wh-phrase, for instance) passes through SPEC positions of C and on
its way to the SPEC-C position where it is illocutionarily interpreted. The fact that
specifically C and are the categories that offer an escape hatch for successive A-
movement is rather natural when we bear in mind that SPEC-C and extra SPEC-
are associated with peripheral features. It is uncontroversial that C offers the do-
main for identifying peripheral features. Recently, Jayaseelan (2001) and Belletti
(2004, 2001, 1999) argued that there is an InflP-internal periphery above the P,
which offers the landing site for topical and focal phrases. This introduces an
interesting parallelism between pre-sentential and pre-predicate positions, as both
would provide suitable positions for a-phrases and intermediate positions for suc-
cessive A-movement. It is also clear that a DP with a token of -features uses
SPEC-T specifically, the position where -features associated with nominative
case are prototypically identified, as an escape hatch on its way to the final landing
site. Thus, the choice of intermediate positions of successive A-movement and
successive A-movement is determined by the type of the token feature: a-phrases
move through SPEC-C and SPEC-, and nominative DPs through SPEC-T.
Although there is a type agreement between intermediate copies and the
hosting head (observation (a)), the [+type] feature does not provide a suitable po-
sition for the [+token] category to be interpreted (observation (b)), and thus the
category with the [+token] feature must still be remerged to another position. The
anaphoric -features of T
def
cannot offer an appropriate position for a DP with a
complete set of -features; thus, in a raising construction or in an ECM con-
struction, a DP moves from the SPEC of the head T
def
, whose set of -features is
anaphoric, to the SPEC of a higher occurrence of -features until it arrives at a
SPEC-T
compl
in raising constructions and at a SPEC-V
compl
in ECM constructions.
When the DP is in SPEC-T
def
it is still active, as it cannot be properly interpreted
there. Similarly, a moving wh-phrase cannot be properly interpreted in inter-
mediate A-positions. In (41), for instance, the type feature of force of the em-
bedded C is [+assertive], whereas who provides a [+interrogative] token feature;
THE EMERGENCE OF ORDER IN SYNTAX 80
consequently, the embedded C cannot offer an appropriate position for the wh-
phrase.

(41) Who do you think [who [that]
C[+assert]
Peter saw]

Both wh-phrases and -phrases (or DPs) are active when they are not in a
semantically adequate position (SPEC-T
def
in the case of DPs and SPEC-C
[+assert]
in
the case of wh-phrases, respectively) and frozen in place when they are in a
semantically adequate position.
Recall that languages vary as to the number of SPECs provided by C, a factor
that determines the presence or absence of wh-island effects. Bulgarian, a language
that allows multiple wh-fronting, obviates wh-island effects. As noted above, the
intermediate copy of a wh-phrase in a SPEC of an embedded interrogative C
cannot be analyzed in terms of absorption, since such a wh-phrase is not
illocutionarily interpreted with respect to the embedded interrogative C but ruther
with respect to a higher interrogative C. This situation seems to suggest the
necessity for the embedded interrogative Cs to contain a force feature specified as
[+wh], which is instantiated by the wh-phrase that is spelled out and illocutionary
interpreted in embedded SPEC-C, as well as an extra peripheral feature, which
does not suffice to provide the semantically appropriate position for a wh-phrase
illocutionarily related to a higher C. In Bulgarian, but not in English, an embedded
C can have both an interrogative force feature and an extra peripheral feature that
triggers intermediate steps of A-movement.
In sum, successive cyclicity is not an effect of the PIC, since both C/ and T
provide intermediate positions for A- and A-movement respectively, but rather a
consequence of the presence of non-interrogative a-features in C and that fail to
provide an appropriate interpretive positions for a wh-phrase and of anaphoric -
features of T that fail to provide an appropriate interpretive position for a DP with
a complete set of -features. Wh-island effects emerge when a C specified as
[+wh] fails to contain an extra a-feature, as in English embedded questions.
This line of reasoning forces us to revise the conclusion we arrived at in
3.3.2.1.1: there is an asymmetry between C and T in the role these categories play
in the generation of syntactic objects, since C, but not T, defines its domain as
inaccessible. Recall that the premise for such a conclusion was that the domain of
T is accessible to C, whereas a superordinate C could attract a wh-phrase only if
such a wh-phrase was at the edge of the immediately subordinate CP, that is, if A-
movement proceeded successive-cyclically. According to the arguments given
above, successive cyclicity is attributed to the presence of defective features, and it
cannot be attributed to the PIC: anaphoric -features in successive A-movement
DISCONTINUOUS SYNTACTIC PATTERNS 81
and non-interrogative a-features in successive A-movement. This accounts for the
existence of successive cyclicity, but does not account for the idea that a [wh] type
feature fails to penetrate the domain of an embedded CP to probe a wh-phrase, as
in English. However, the aforementioned asymmetry between the so-called phase
heads C/ and Infl-heads is quite a natural one: if, for instance, a a-feature located
on penetrated the domain of an immediately lower C to probe a wh-phrase, it
would penetrate the domain of a head containing a-features, but if a a-feature
located on C penetrated the domain of an immediately lower Infl head to probe a
wh-phrase at the edge of P, then the penetrated domain would not be the domain
of a head containing a-feature. In other words, the asymmetry between C/ and
Infl-heads does not derive from a special status of the so-called phase heads in the
generation of syntactic objects, but from a relativized opacity factor relative to
probe-goal relation, which may be defined as follows:

Relativized Opacity Principle (ROP) (non-final definition)
In a syntactic object [o
1
[ [o
2
y] ] ] ,
where: (i) o
1
and o
2
are two probes of the same type o,
(ii) y is in the domain of a
2
, and
(iii) y is a goal with a token o,
o
1
cannot readily probe y

When probes a wh-phrase that belongs to the domain of a lower C, it violates
the ROP, but when C probes a wh-phrase into the domain of Infl, it does not.
It can thus be concluded that no deep asymmetry exists between C and Infl:
successive cyclicity derives from a type-agreement relation between a type feature
and a token feature that does not yield a satisfactory matching relation that
provides an appropriate instance for the type feature and a semantically suitable
position for the moving category, which remains accessible to further probes, and
the observation that C can penetrate the domain of Infl but C cannot penetrate the
domain of a lower /C derives from the ROP. Consequently, neither successive
cyclicity nor penetrability restrictions provide empirical motivation for the PIC.
It is clear that the ROP formulated above is not incompatible at all with the
existence of cyclic mappings to the external systems, but it does not require them.
Although the PIC or the existence of cyclic mappings to SEM and PHON seems
unnecessary to account for impenetrability, it remains unresolved whether cyclic
mappings to SEM and PHON are necessary for external requirements and whether
they are defined by C/, and not by Infl/V. Indeed, the possibility that cyclic
mappings are relevant to minimize the computations that take place in the external
THE EMERGENCE OF ORDER IN SYNTAX 82
systems, but not necessarily in narrow syntax, is not very far from Chomskys
position. Consider these two quotations from two recent papers:

PIC sharply restricts search and memory for [the phonological component], and thus
plausibly falls within the range of principled explanation []. It could be that PIC extends
to NS [Narrow Syntax] as well, restricting search in computation to the next lower phase
(Chomsky 2001b: 6).

Note that for narrow syntax, probe into an earlier phase will almost always be blocked by
intervention effects. One illustration is agreement into a lower phase without intervention
in experiencer constructions in which the subject is raised (voiding the intervention effect)
and agreement holds with the nominative object of the lower phase (Icelandic). It may be,
then, that PIC holds only for the mappings to the interface, with the effects for narrow
syntax automatic (Chomsky 2005b: 9-10).

There is something odd in this position. Firstly, the PIC, as defined, is a
condition that reduces the search domain of a phase head: the edge of a lower
phase is accessible, but its domain is not; prototypically, probe-goal relations take
place at narrow syntax. Secondly, in the absence of an intervening category, a
probe can detect a goal located in the domain of a lower phase, which argues that
the relevant constraint on probe-goal relations is not the PIC but rather a
minimality principle; this seems to be the case in experiencer constructions of
several languages like Icelandic or Catalan, in which V agrees with a nominative
object.

(42) Li agraden les pellcules
3SG.DAT like.3PL the films
He likes films

And thirdly, the PIC is argued to hold only for the mappings to the interface,
restricting search and memory for . What seems strange here is that the effects
of the PIC, a condition that restricts the probe-goal relations that take place at
narrow syntax, are claimed to hold only for the mappings to the interface,
restrictions on probe-goal relations being derived from other conditions as min-
imality.
In sum, the PIC is neither a necessary condition in order to account for
successive cyclicity (there exists both A- and A-movement are successive) nor a
sufficient condition in order to account for impenetrability (when there is no inter-
vener, the domain of a phase can be accessed by a higher probe).


DISCONTINUOUS SYNTACTIC PATTERNS 83
3.4 Discontinuities
The plausibility argument we address in this section is as follows: if it can be
motivated on independent grounds that C and [but not Infl] are the labels
driving internal operations and relevant to external Merge, and are also the points
of feature-valuation and transfer (Chomsky 2005a), then the C-Infl link might
derive from inheritance mechanisms. This would raise the possibility that the
labels defining the domains accessible to syntactic computations offer the locus
where functional features are base-generated, and that more elaborate structures
revealed by the cartographic inquiries are based on linearization of features in
these labels, and possibly labels closely linked to them (as in the C-T connection),
as Chomsky (2005a: 18) speculates.
In order to decide if internal operations are driven by the phase head and if the
C-Infl connection is due to inheritance mechanisms, it is crucial to find out
whether there is an asymmetry between C and Infl with regard to the role they play
in the generation of syntactic objects. This question has been examined in the
context of derivational restrictions for extraction; since, firstly, there is no
empirical reason to think that C defines strict cycles and, secondly, both C and Infl
are relevant in the determination of successive cycles, it has been concluded that
there is no such asymmetry. Consequently, the PIC (which assumes an asymmetry
between C and Infl) has been replaced by the ROP.
It should not be forgotten that, in the theory of phases, the categories that
define the fragments of structure to be transferred (and thus that become invisible)
are not stipulated or defined on the basis of impenetrability effects, but rather they
are independently determined on the basis of both internal and external con-
siderations. The way phases are defined varies from the way barriers were defined.
The main internal consideration relevant to the idea that C* and * are the
phase heads is that Infl and V display a complete set of -features that can host a
DP with a full set of -features if and only if C* and * are present. The analysis
of this observation states that -features are inherent to C*/* and derivative to
Infl/V. However, this is not a necessary conclusion; it might be the case, for
instance, that C and Infl or and V simply had to agree. The virtue of in-
heritance mechanisms is dependent on whether it can be independently argued that
there is an asymmetry between C*/* and Infl/V with respect to the role they play
in the derivation of syntactic objects. For the same reason, the condition that
probe-goal relations take place in parallel in the C-Infl system does not necessarily
imply that C is the label driving both A-movement and A-movement.
The external A-P considerations of which I am aware (see Chomsky 2000) do
not provide a very strong argument, since the points of feature-valuation do not
coincide with the fragments that display distributional freedom: whereas an em-
THE EMERGENCE OF ORDER IN SYNTAX 84
bedded clause of a raising verb cannot be dislocated (43), a non-agentive P can
be (44). The observation that control structures can be clefted or dislocated is pro-
vided by Rizzi (1982a).

(43) a. To stay at home I think he prefers
b. *To stay at home I think he seems

(44) a. John wants to read the book, and read the book he will
b. John wants to stay at home, and stay at home he will

In fact external A-P considerations conflict with internal considerations, since
it is not clear at all whether clauses embedded under control predicates display a
complete set of -features: although they can be dislocated as finite clauses, their
set of -features is anaphoric, like that of raising and ECM constructions (see
above 2.1.1 and also 2.1.2). Thus, external A-P considerations are not compelling.
As to external C-I considerations, it might well be that C, and not Infl, is the
locus of force features, and that *, and not V, is the locus of agentivity; thus, in
some sense, C and * are responsible for closing the proposition and the predicate,
respectively, and thus constitute objects with a relatively semantic autonomy. But
again, this does not necessarily imply any significant asymmetry between C/ and
Infl/V, as the semantic instructions provided by Infl-like features are as necessary
as those provided by C/ in defining the properties of the proposition or the
predicate.
These reasons lead me to work out an alternative to the GFIT in order to
account for the C-Infl link, materialized in the following three properties:

(I) Complementizers can replicate Infl-like features

(II) There is a correlation between the characteristics of features surfacing on
Infl and the choice of C

(III) Infl-like features are involved in triggering V-to-C movement

Let us use the observation made in section 3.3.1.1 as a stating point. There we
saw that the lexico-semantic feature with which C heads and Infl heads are origi-
nated is the same, [clause typing], but with a different polarity: C is specified for
a positive polarity ([+clause typing]) and Infl for a negative polarity ([-clause
typing]). An issue that seems to me undecidable is whether such a distinction is
DISCONTINUOUS SYNTACTIC PATTERNS 85
inherited from the lexicon (in which case Infl and C would be lexical primitives)
or whether it is a syntactic product that derives from splitting a hypothetical lexical
unit [clause typing] into two positions, one for each polarity. Be as it may, what
is crucial for our concerns is that C and Infl turn out to be two different positions
with a different polarity of the same feature, and in that sense a discontinuous
featural pattern.

(45) [+clause typing]
C
[-clause typing]
Infl


Observe that Infl-like features introduce semantic instructions that are
orthogonal to the [clause typing] distinction; for instance the instruction of being
3
rd
person singular referred to an individual or of being negative referred to a
proposition is independent of either the [-clause typing] instruction provided by
Infl nodes or the [+clause typing] instruction provided by C nodes. This ortho-
gonality with respect to both values of the [clause typing] feature may force Infl-
like features, when they are to be introduced into the discontinuous template
[+clause typing]
C
[-clause typing]
Infl
, to be merged in both poles, and then to
split into a [+clause typing] occurrence and a [-clause typing] occurrence. Thus, an
Infl-like feature o is associated with two positions of a discontinuous syntactic
template, and is therefore a discontinuous feature in syntax.

(46) [u]
+C


[u]
C
[u]
+Infl


[u]
Infl


This would straightforwardly account for properties (I) and (II): the agreement
relation observed between C and Infl is no more than a manifestation of inserting
Infl-like features in both poles of the [+clause typing] [-clause typing] dis-
continuity. Naturally, languages vary with respect to the morphological material-
ization given to such a deep link; as described in chapter 2, certain languages dis-
play a relatively rich inflectional paradigm of -features on C, while others only a
[fin] distinction, certain languages mark subjunctive distinctions on C, while
others use verbal endings and others free particles, certain languages have negative
markers encoding the [embedded] distinction, while others do not, and so on. All
this falls under the indisputable degree of morpholexical variation.
THE EMERGENCE OF ORDER IN SYNTAX 86
The idea that the C-Infl relationship should be understood as a discontinuous
unit is not new in the generative literature; as far as I know, at least Stowell (1981)
as well as Rochette (1988) and Drubig (2001) have resorted precisely to this
possibility at some point within their argumentations, Stowell to account for cer-
tain properties of tense and Rochette and Drubig to account for certain properties
of modality. In Stowells words (1981: 241):

we might adopt a suggestion of Y. Aoun (personal communication) to the effect that the
complementizer and Infl form a discontinuous element. The matching between com-
plementizers and Infl would then follow from the fact that the two actually form a single
unit at some level, so that selection for one implies selection for the other.

My proposal is that this intuition, thus far not seriously considered nor
systematically researched, may turn out to be the appropriate analysis of the con-
nection between complementizers and inflectional heads.
Property (III), or the observation that Infl-like features can trigger V-
movement both to C and to Infl, can be argued to be just a side-effect of properties
(I) and (II). If Infl-like features are inserted both in C and in Infl, it is expected that
they can work as probes from one position or the other. This raises the very com-
plex question of which factors determine which of the two occurrences of Infl-like
features is active in which language and in which construction. This issue will be
partially explored in the next chapter.
First we shall tentatively explore the possibility of the existence of other
discontinuous syntactic patterns beside C-Infl. If this is so, the possibility for a
feature to be associated during syntactic computations with two positions would
not be an idiosyncrasy of Infl-like features but a more general property of the
syntax of functional features. I shall focus on two possible cases: the connection
between prepositions and the case endings of nominals and the connection
between the functional head and V.
It is also a traditional observation that P assigns case to a nominal, or, in
other words, that the value of K (case) correlates with the choice of P as the value
of several Infl-like features correlates with the choice of C. As a matter of fact, the
problem of precisely describing the correlation between the mood endings and the
choice of complementizer in a language like classic Latin is analogous to that of
precisely describing the correlation between the case endings of nominals and the
choice of preposition. This close relationship suggests that P and K constitute a
sole unit split into two positions, just as C and Infl do.
Again languages vary in how they materialize the P-K link with morpholexical
material, resorting only to prepositions, only to nominal endings (which are often
DISCONTINUOUS SYNTACTIC PATTERNS 87
grammaticalized postpositions in the case of oblique cases), to both or to none of
them.
The functional similarity between C and P is quite clear. As Rizzi (1997: 283)
observes, We can think of the complementizer system as the interface between a
propositional content (expressed by the IP) and the superordinate structure (a
higher clause, or possibly, the articulation of discourse, if we consider a root
clause); analogously, we might add that the prepositional system can be thought
of as an interface between an argumental content (expressed by the DP), and an
embedding structure, a higher VP or DP, or perhaps discourse if we consider
nominal fragments.
As Emonds (1985) argued in a work with the very explicit title Sas Pand
COMP as P, it may be that P and C are not two distinct categories but rather
instantiations of the same archicategory appearing in different syntactic frames.
Emonds particular position was that complementizers are a subcategory of
prepositions appearing in the frame S (where S is equivalent to Infl), and thus
that a CP is a subtype of PP (in Emonds terms, Swould be a subtype of P).
Although I sympathize with Emonds methodological intuition that it is desirable
to reduce the number of primitive grammatical categories of UG, if empirically
supported, I must adopt a different technical implementation. The principal reason
for this is that, in the model we are developing but not in Emonds, there is no
such a thing as a frame S that defines P as C, for Infl is neither an independent
element of C nor an element pre-existing C, as C and Infl constitute a sole unit
spread into two positions. In other words, claiming that P becomes C in the frame
Infl would be a circular argument as the frame itself constitutes a unit with C.

Since the subtypes C and P cannot emerge contextually as in the course
developed by Emonds, I propose that both C and P are instances of a more abstract
category that cannot be directly identified with C or P. If we recall that both C and
P connect an internal structure (IP, DP) to an external one, it seems reasonable to
define such an archicategory with a [+connective] feature.

(47) C
[+connective]
P

It is as desirable for the theory of UG to find an accurate characterization for
the common substrate of C and P as to define the source of divergence between
them. The simplest way to achieve this second objective, keeping to the idea that
THE EMERGENCE OF ORDER IN SYNTAX 88
the C-Infl discontinuity is defined by the [clause typing] feature, is to define a
distinctive feature for the P-K discontinuity.
The semantic instruction provided by prepositions seems to be a topological
relation (direction, origin, location, etc.) between arguments. The choice of P
correlates with the value of K in a relatively systematic way; for example, in Latin,
where both P and K can be overt, a preposition like ad, which conveys a relation
of direction, correlates with the accusative ending [-em] in ad civitatem
(to/toward the city), and a preposition like e/ex, which conveys a relation of
origin, correlates with ablative ending [-e] in ex civitate. Therefore C can be
characterized as a [+clause typing] connector and P as a [+topological typing]
connector, and Infl and K as [-clause typing] and [-topological typing],
respectively.

C: [+clause typing]
Infl: [-clause typing]
P: [+topological typing]
K: [-topological typing]

As observed by Kratzer (1996), there must be a connection between the
thematic role of the external argument and the Aktionsart properties of V, which
are argued to originate from selectional restrictions for the event argument
(actions, states, events proper and so on). Kratzer attributes the link between the
external argument and the event argument to the functional head voice (or ),
which is responsible for conjoining the external argument and the VP. Thus,
strictly speaking, the external argument is not selected by V but chained to it by
virtue of the functional head voice. Consider the following informal sketch from
Kratzer (1996: 123):

Suppose there are two kinds of voice heads in English: active and non-active. Active voice
heads add external arguments and assign (check) accusative Case. Non-active voice heads
do not add external argument and do not assign (check) accusative case. []. Suppose
furthermore that the repertoire of (basic) active voice heads is very limited. Maybe there
are just two of them, one adding an agent argument to an active verb, and the other one
adding the holder to the stative verb.

Like C and P, connects an internal structure (a VP) to an external one (an
external argument) ensuring that their semantic properties are compatible. This
raises the question of whether the node , which selects the external argument (,
agent or holder, following Kratzers approximation), and the node V (non-active,
active agentive, active stative) constitute a discontinuous template, which would
DISCONTINUOUS SYNTACTIC PATTERNS 89
straightforwardly account for the one-to-one relation that exists between the choice
of and the properties of V: a particular definition of the properties of the external
argument automatically imply a particular definition of the Aktionsart properties
of V simply because they are the same (discontinuous) unit.
The head may still connect the internal properties of V to another kind of
external element, manner adverbs, an observation that goes back to Chomsky
(1957, 1965) and Lees (1960): The Verbs that do not take Manner Adverbials
freely Lees has called middle verbs (Lees 1960: 8), and he has also observed that
these are, characteristically, the Verbs with following NPs that do not undergo the
passive transformation (Chomsky 1965: 103). Therefore, if -V = [+middle], then
can neither become passive nor conjoin a manner adverb with V.

(48) a. *John married Mary well
a. *Mary was married by John
b. *John resembled Mary well
b. *Mary was resembled by John
c. *John had a car well
c. *A car was had by John

Cinque (1999a: 101-103) resorts to this observation and some cross-linguistic
evidence to argue that manner adverbs appear precisely in the SPEC position of a
head he terms voice.
Following Kratzers spirit, the two nodes of the -V discontinous syntactic
pattern can be characterized as two polarities of the same feature: is [+voice
typing] and V is [-voice typing].
In sum, after revising some aspects of the theory of phases, it has been
concluded that there is no asymmetry between C and Infl in the generation of
syntactic objects. Consequently, the GFIT, a sub-theory of the theory of phases,
has been discarded: the C-Infl connection is not due to feature inheritance
mechanisms that spread features inherent to C to Infl, but to the discontinuous
status of C and Infl. Similarly, the condition that A-movement and A-movement
are triggered in parallel does not imply that C is the label that triggers internal
merge operations. It has been pointed out that, beside the C-Infl connection, the P-
K connection and the -V connection might also be analyzed as discontinuous
syntactic patterns.

3.5 Subextraction: (II): relativized opacity for probe-goal relations
Before closing this chapter, I must turn to an issue related to the theory of
phases that has not been satisfactorily analyzed in our previous discussion.
THE EMERGENCE OF ORDER IN SYNTAX 90
A very important observation from Chomsky (2005b) is that a a-probe can
subextract a a-phrase from the SPEC of a lower head if such a head is T
def
but not
or C. Putting some potential empirical complications aside (see Uriagereka 2004,
Gallego 2006 and section 5.3.2.1 of this study), Chomskys refinement of Huangs
subject islands seems to be solid. However, it still remains to be explained why
subextraction from SPEC-C or SPEC- is problematic; in Chomskys words
(2005b: 14):

It remains to explain why the probe for wh-movement cannot readily access the wh-phrase
within the external argument of o [o = the *P phase]. That could reduce to a locality
condition: which in o is embedded in the lower phase, which has already been passed in
the derivation. We know that the external argument itself can be accessed in the next
higher phase, but there is a cost to extracting something embedded in it.

This possibility is unclear: to be precise, the lower phase has not been
transferred, but only its domain; consequently, according to the PIC, we would
expect its edge and the constituents of its edge to be accessible to a further phase
head.

On this point I shall attempt to argue that the asymmetry attested between
subextracting a a-phrase from SPEC-*/C* and from SPEC-T
(def)
can be derived
from a redefinition of the Relativized Opacity Principle. Consider, as a point of
departure, the first approach of the ROP given above and repeated here:

Relativized Opacity Principle (ROP) (non-final definition)
In a syntactic object [o
1
[ [o
2
y] ] ] ,
where: (i) o
1
and o
2
are two probes of the same type o,
(ii) y is in the domain of a
2
, and
(iii) y is a goal with a token o,
o
1
cannot readily probe y

As argued, such a definition accounts for the observation that a a-phrase
cannot be extracted from the domain of a lower C (wh-islands effects), whereas it
can be extracted from the domain of a lower T (i.e., from a SPEC-), since C, but
not Infl, contains a-features, and thus induces relativized opacity effects for A-
movement. In other words, a a-probe cannot penetrate into the domain of a lower
a-probe, but it can detect its SPEC.
DISCONTINUOUS SYNTACTIC PATTERNS 91
However Chomskys observations reveal that, although a higher probe can
detect and extract a lower phrase at a SPEC of a a-projection, it cannot subextract
a constituent from a SPEC of a a-projection: a C head can subextract a a-phrase
from SPEC-T
(def)
, but not from SPEC-C/. This leads to the conclusion that, for a
higher probe, there is no distinction between penetrating into a complement or into
a SPEC: a wh-probe can neither penetrate into the complement of a lower a-head
(wh-island effects) nor into the SPEC of a lower a-head (subextraction patterns),
but it can penetrate into the complement of T (extraction of a a-phrase from SPEC-
to SPEC-C of the same clause) or into SPEC-T
def
(subextraction of a a-phrase
from an agentive subject of a predicate embedded into a raising or ECM
construction).
Recall that -features and a-features must probe in parallel if they belong to
the same discontinuous pattern (or to the same phase, in Chomskys terms);
accordingly, the factor that prevents C from subextracting a a-phrase from SPEC-
T when C and T constitute a discontinuous pattern is not the ROP (T is not a a-
probe), but rather a condition that requires operations to take place in parallel at
the same discontinuous pattern (or phase). In the case of raising/ECM con-
structions, it is a superordinate a-head that penetrates embedded SPEC-T
def
, and
thus C and T
def
do not constitute a discontinuous domain and do not probe in
parallel. Note that it is crucial that the embedded clause lacks a CP projection (as
Chomsky assumes) or that its C head is defective: it provides the [+embedded]
semantic instruction (as proposed above), but lacks a-features. If it had a-features,
we would expect subextraction from an agentive subject in raising/ECM con-
structions to be problematic. It is also important to note that a contradiction occurs
when two criteria for defining phases are applied to raising constructions: on the
one hand, T
def
P should not be a phase because the -features and tense features of
T
def
are anaphoric, but on the other hand, T
def
P should be a phase because T
def
must
probe its goal before matrix C, and not in parallel. This suggests that the syntactic
unit where probe-goal relations take place in parallel is a syntactic discontinuity,
which also includes embedded clauses of raising constructions, C
def
-T
def
.
According to the above reasoning, penetrating into a SPEC of a a-projection is
as troublesome as penetrating into the complement of a a-projection for a higher a-
probe. To keep the parallelism as close as possible, extracting the whole
complement of a a-projection should be as easy as extracting the whole SPEC.
This means that moving an InflP and a VP, with C and stranding, should be as
easy as extracting a wh-phrase from SPEC-C/. Although this seems to be false, it
does not necessarily challenge the proposed analysis, since the static nature of
InflP and VP may be due to other factors; for instance, it could be that InflP and
VP did not manifest, contrary to arguments or CPs, a suitable token to instantiate a
THE EMERGENCE OF ORDER IN SYNTAX 92
type a-feature (i.e., they would not be born with a wh-feature or a topic/focus-
feature), or that syntactic operations had problems in affecting only one sub-ele-
ment of a discontinuous element. I shall continue to assume that the whole InflP
and VP are visible to a further probe, as SPEC-C and SPEC- are, its apparent un-
extractability being related to other factors.
Plainly, the ROP, as defined above, cannot account for subextraction patterns,
but a minimal modification would automatically account not only for wh-islands
effects and for the local nature of movement, but also for the asymmetry between
subextracting from SPEC-C/ and from SPEC-T
def
. This slight modification is to
consider that the opacity inducer is not the head o but the label of the a-projection.

Relativized Opacity Principle (ROP) (final definition)
In a syntactic object [
o1P
[o
1
[
o2P
[o
2
] ] ] ],
where: (i) o
1
and o
2
are two probes of the same type o each projecting an
aP
(ii) is SPEC-o
2
and is o
2
-COMPL, and
(iii) : is a constituent of and y: y is a constituent of ,

o
1
can probe or if they provide a suitable token for o
1
, but it cannot
readily probe y or

According to this definition, the factor that decides whether a feature o
1
of type
o can readily probe a goal is not being inside or outside the search domain of o
2
,
but their relative depth in the a
2
-projection whose label is of the same type as the
searching probe a
1
. o
2
P does not render the complete and opaque to o
1
, but
search in them becomes difficult. In other words, o
1
manages to detect and
but not and y because and are located on a more superficial layer than and
y.
As desired, the definition of the ROP given above allows a a-probe to search
into a SPEC and a complement of a non-a-projection. Crucially, can penetrate its
complement VP to probe an object wh-phrase or an object DP and it can also
penetrate such an object DP to subextract a wh-phrase; C can also penetrate into its
complement TP to extract a wh-phrase which is at SPEC-, but it cannot penetrate
the -COMPL or the SPEC-, since P is a a-projection; and finally, a super-
ordinate C can penetrate the subordinate SPEC-T
def
without penetrating any a-
projection (because there is no embedded CP between the matrix C* and the
DISCONTINUOUS SYNTACTIC PATTERNS 93
embedded SPEC-T
def
or because there is a defective CP lacking non-interrogative
a-features).
It is worth noting that, if the conclusions on locality that this chapter has drawm
are on the right track, the principle governing locality is no more than a variation
of the first locality principle formulated in syntax, the A-over-A principle
(Chomsky 1964a, b). See Ross (1967) classical revision of the A-over-A prin-
ciple, which gave raise to a great deal of empirical generalizations and ideas that
still keep on feeding syntactic theory.

A-over-A principle
If a phrase X of category A is embedded within a larger phrase ZXW which is
also of category A, then no rule applying to the category A applies to X but
only to ZXW (Chomsky 1964a: 931).

The question whether the ROP is the best definition of the metrics for internal
Merge requires further research.







CHAPTER 4

ANALYTIC SYNTACTIC PATTERNS

This chapter considers the possibility that the Conceptual-Intentional system generally
favors analytic representations with a one-to-one relation between positions and features.
It is argued that several distributional restrictions observed by Cinque (1999a) derive
from the Prohibition Against Tangled Structures and the Prohibition Against Vacuous
Quantification, two instances of a general legibility condition, the Full Interpretation
Condition. Accordingly, the ordering of functional categories observed by the cartogra-
phic project does not lead to a primitive syntactic component, but rather to the study of the
C-I system. Finally, it is observed that discontinuous syntactic patterns are no more than a
subtype of analytic syntactic patterns.


4.1 On cartographies
The general observation of the cartographic project is that languages do not
seem to differ as to the ordering relations between functional categories they
allow. If this is correct, the combinatorial possibilities (roughly, literal precedence
and scope) of the multiple functional categories are very limited; indeed, it is
argued that, for any three different categories A, B and C, it is always the case that
they follow a unique strict linear order. Since allegedly there is no flexibility as to
the linearization of functional categories, a very precise map of the functional
structure where every category occupies its position can be drawn. Belletti & Rizzi
(2002: 123) express the basis of the Cartographic Project very clearly in an inter-
view with Noam Chomsky:

If it is true that a constitutive characteristic feature of natural languages is to privilege
representations with many dedicated positions, each with simple interpretive properties, it
becomes important to draw a map as precise and fine-grained as possible of this complex-
positional system. This is the rationale behind the so-called cartographic studies, which
are pursued intensely in some research centers in Italy and elsewhere.

Hence, the main issue of the cartographic project concerns the relation between
functional categories and positions, and the conclusion the cartographic studies
yield is that UG favors a strict linear order of functional categories or analytic syn-
tactic representations.
THE EMERGENCE OF ORDER IN SYNTAX 96
As argued in the preceding chapter, there is a very general distinction between
[+clause typing] categories and [-clause typing] categories, and it is the case that
the former are always in a higher (or less nested) position than the latter. If we
recall that C links a proposition (say, Infl) to a superordinate structure (see above
3.4), then it becomes entirely natural that C is higher than Infl. But if the
cartographic observations are correct, the [clause typing] distinction is just the
starting point, and both C and Infl must split into many semantically specialized
positions, the C-area and the Infl-area.
According to Rizzi (1997), the map of the C-area is composed of two sub-
areas, the Force-Finiteness sub-area and the Topic-Focus sub-area. Rizzi asserts
that the Force-Finiteness is the essential part of the C system that connects the
superordinate structure and the propositional content, the Infl-area. Comple-
mentizers express the clause type or the force of a proposition, whether it be a
question, an assertion or an exclamation. This instruction is encoded in the head of
the highest projection of Rizzis hierarchy, Force. The head of the lowest pro-
jection, Fin, contains the core IP-related characteristics that the C system ex-
presses. Rizzi postulates that Fin accounts for the link between the C-system nd
the Infl-system.
It is claimed that the Topic-Focus area is present in the structure only if
required, i.e., if there are phrases with a topic-focus interpretation. When the topic-
focus field is active Rizzi argues, it will be inevitably sandwiched in between
force and finiteness, as these two specifications must terminate the C-system
upward and downward [] (p. 288). Rizzi identifies three possible positions: a
Focus position surrounded by two Topic positions. This would be the cartography
for the full-fledged C-area, which would limit with the upper superordinate
structure and the lower Infl-area:

(1) [ [Force [Topic [Focus [Topic [Fin [ ] ] ] ] ] ] ]

The Infl-area appears to contain a great deal of functional categories that are
rigidly and uniformly ordered across languages. Indeed, as Cinque (1999a: 127)
asserts, [L]anguages do not seem to differ as to whether they have aspectual
projections higher or lower than mood projections, epistemic modality higher or
lower than root modality, etc. Therefore, UG would provide a very fine-grained
structure where every inflectional category would occupy its position, without any
deep difference across languages as to the type, number and order of functional
categories. According to Cinques results, the Infl-area would be composed of
around thirty functional projections (Cinque 1999a: 106). For further refinements,
ANALYTIC SYNTACTIC PATTERNS 97
see Cinque (1999b, 2000), where ordering restrictions among restructuring verbs
are taken into account.

(2) [frankly Mood
speech act
[fortunately Mood
evaluative
[allegedly Mood
evidential

[probably Mod
epistemic
[once T(Past) [then T(Future) [perhaps Mood
irrealis

[necessarily Mod
necessity
[possibly Mod
possibility
[usually Asp
habitual
[again
Asp
repetitive(I)
[often Asp
frequentative(I)
[intentionally Mod
volitional
[quickly
Asp
celerative(I)
[already T(Anterior) [no longer Asp
terminative
[still Asp
continuative
[always Asp
perfect(?)
[just Asp
retrospective
[soon Asp
proximative
[briefly Asp
durative

[characteristically(?) Asp
generic/progressive
[almost Asp
prospective
[completely
Asp
SgCompletive(I)
[tutto Asp
PlCompletive
[well Voice [fast/early Asp
celerative(II)

[again Asp
repetitive(II)
[often Asp
frequentative(II)
[completely Asp
SgCompletive(II)


In his attempt to disprove the assumption that languages may vary in the
number and type of the functional projections that they admit and/or in their
relative order (p. iii), Cinque constructs a plausibility argument that can be
formulated as follows. If attached heads, free heads and maximal categories
(adverbials) follow the same strict linear order, languages can vary as to in-
stantiating a particular functional feature as a free head, an attached head or a max-
imal projection, but all languages have the same set of functional features which
are uniformly ordered in the same hierarchy. Thus, adverbs are selected categories
that enter into a SPEC-H relation with different heads of the universal hierarchy.
The entirety of Cinques book is devoted to giving maximum plausibility to this
argument, by resorting to extensive illustrations from a wide range of languages.
There is a crucial issue that must be noted here: if adverbs appear in SPEC
positions of several silent heads, then a verbal head may appear between two
adverbs (see Cinques chapter 2, A Case for Adverb Phrases in Spec). Cinque il-
lustrates this phenomenon with sentences like (3), where an active past participle
can be found preceding, or following, each AdvP in the sequence [] (p. 45):

(3) a. Da allora non hanno rimesso di solito mica pi sempre
completamente tutto bene in ordine
b. Da allora non hanno di solito rimesso mica pi sempre
completamente tutto bene in ordine
c. Da allora non hanno di solito mica rimesso pi sempre
completamente tutto bene in ordine
d. Da allora non hanno di solito mica pi rimesso sempre
completamente tutto bene in ordine
THE EMERGENCE OF ORDER IN SYNTAX 98
e. Da allora non hanno di solito mica pi sempre rimesso
completamente tutto bene in ordine
f. Da allora non hanno di solito mica pi sempre completamente
rimesso tutto bene in ordine
Since then, they havent usually not any longer always put
everything well in order

In trying to support such a plausibility argument, Cinques work on the
structure of the Infl-area may have brought one of the most careful, subtle and
exhaustive syntactic descriptions ever attempted. Such a descriptive milestone
raises some important theoretical questions. For instance, in his chapter 4 (Some
Implications and Residual Remarks), Cinque poses the following two questions:

(4) (a) does the C-I system require full-fledged structures containing
specialized features with a default value when there is no overt
category with a marked value?

(b) can combinatorial restrictions and hence, cartographies be reduc-
ed to C-I conditions?

I shall not attempt to make any direct contribution as to how or whether
morphologically absent categories are coded (4.a); as far as I can see, such a
question cannot be empirically substantiated, and one is forced to give purely
conceptual arguments, which too often can be indistinctively constructed either for
one choice or the other. Hence I shall not attempt to argue for any of the three
conceivable possibilities outlined below.

(5) a. A functional feature with a [-marked] value is always present in the
syntactic representation and projects its own projection (Cinques
view)

b. A functional feature with a [-marked] value is present in the syn-
tactic representation structure, but syncretized with a [+marked]
feature in a sole position (Giorgi & Pianessis 1996, 1997 view)

c. Functional features are only present in the structure when they are
[+marked], in which case they host morpholexical material (Rizzis
1997 view on the particular case of the topic/focus field)
ANALYTIC SYNTACTIC PATTERNS 99
But I shall directly face question (4.b). Although this question may be premature, as
much work is still needed in order to understand the nature of cartographies, it is clear that
cartographies lead us to inquire into the way syntactic representations provide useful
instructions to the C-I system, an extremely difficult enterprise.
Let us thus consider whether cartographies belong to the set of unexplained
properties of the faculty of language, or in other words, to the set of properties
specific to the faculty of language.

4.2 The nature of ordering restrictions
The general skeleton of the clause is one of the most solid syntactic properties.
To the best of my knowledge, the observation that tense is higher than aspect or
voice in a mono-clausal structure is undisputed; similarly, it is also unquestioned
that aspectual markers are taken to be systematically lower than epistemic markers
and mood markers. It is generally accepted that deontic modals are lower than
epistemic modals, though some cases where deontic modals scope over epistemic
modals have been given (Kratzer 1976 and Cormack & Smith 2002). However,
there is still no satisfactory account for why the general skeleton of the clause has
the form it seems to have.
If Cinque is correct, each of these categories must split into several specialized
projections, again, rigidly ordered; accordingly, there would be a universal pro-
jection for proximative/retrospective aspect, which would be universally lower
than perfective aspect and higher than durative aspect, and a universal evaluative
mood, which would be universally lower than speech act mood and higher than
evidential mood, and so forth. Languages would vary only as to instantiating these
type features of a mono-clausal Infl-area as free heads, verbal periphrasis, affixes
or adverbials occurring in a SPEC position of the relevant head.
Cinques empirical results are preliminary, and therefore may be (partially)
inaccurate; for instance, as Cinque acknowledges, It is not entirely clear whether
retrospective and proximative are two values of one and the same aspect. As
seen with terminative and continuative, their appearing opposite, and com-
plementary, dimensions might be a consequence of their semantics (and of their
contiguity). The same reason may also apply in epistemic necessity and epistemic
possibility and virtually in other cases. It is also true that the above cartography
may have to be enlarged to include several specialized projections for aspect and
root modality, if Romance functional verbs can be analyzed as heads of a mono-
clausal structure, as Cinque (1999b, 2000) argues (but see Sol 2002, where the
mono-clausal/bi-clausal debate is revised and arguments for analyzing clitic climb-
ing as restructuring in a reduced bi-clausal structure without a CP layer in the
embedded clause are given).
THE EMERGENCE OF ORDER IN SYNTAX 100
As will become clear, it is not a minor task checking whether an apparent
ordering restriction is real or not. For this reason, the empirical scope of this
endeavor will be relatively narrow: I shall focus on the general skeleton of the Infl-
area and some of the more fine-grained ordering restrictions, but I shall ignore the
cartographic studies concerning other categorial areas (see for instance Koopmans
1993 and Riemsdijk 1978a, b, 1990, 1996 works on prepositional phrases,
Cinques 1994, 1996, 2005 works on determiner phrases and Corver 1997a,b and
Zamparellis 1995 work on adjective phrases). Similarly I shall neglect to discuss
several studies concerning the clausal structure (as Zanuttinis 1991, 1997 work on
negation, Beghelli & Stowells 1997 work on nominal quantifiers and Cormack &
Smiths 2002 work on modals).

4.2.1 Some allegedly non-primitive order restrictions. Cinques 1999a: 135)
position on the relation between ordering restrictions and semantic constrains is
that:

while the relative order of some such notions may indeed reflect intrinsic logical relations
among them, the hierarchy is only indirectly related to such semantic, or logical, prop-
erties. For instance, certain possibilities which in terms of logical relative scope would be
expected are not found, or are downright impossible, thus suggesting that the hierarchy is
a construct of the computational system of language, not completely reducible to other
components.

Lets begin by considering those restrictions that are allegedly related to C-I
constraints (4.2.1.1 and 4.2.1.2).

4.2.1.1 Epistemic modality and tense. Cinque (1999a: 135) asserts that:

the fact seen in chapters 3 and 4 that epistemic modality is higher than (takes scope over)
Past, or Future, Tense (see also Bybee 1985, 119ff) appears to reflect the intrinsic relative
scope of the two types of operators. Epistemic modality, as noted, expresses the degree of
the speakers commitment to the truth of the proposition expressed by the sentence. But
the truth of the sentence can only be evaluated if the proposition is located in a precise
moment of time. Consequently, epistemic modality presupposes an operand which is al-
ready tensed (and thus its scope is external).

The claim that epistemic modality requires a temporal indication in its operand
does not ban a temporal operator from taking scope over a proposition whose
epistemic modal force has already been calculated, which would create a recursive
structure [TP
2
[MODP [TP
1
]]]. Consider the statement given in (6), where
the plausibility of the proposition Great Britain avoid- war with Germany is
ANALYTIC SYNTACTIC PATTERNS 101
evaluated with respect to a set of premises that some presupposed epistemic agents
shared in 1937, a past time. Such a proposition was at that time evaluated as
compatible with the relevant premises, although the evidence was not conclusive,
and thus an existential epistemic modal auxiliary is selected instead of a universal
one. Note that here we are concerned only with the epistemic interpretation of
could, and not with the deontic interpretation.

(6) The general opinion in 1937 was that Great Britain could/might avoid war
with Germany, but in 1939 nobody believed that

Thus, (6) presents two tenses, one in the matrix, which scopes over the
epistemic modal operator (could/might), and one in the embedded clause,
within the scope of the epistemic modal operator. The former locates the time
where the epistemic agents evaluate the proposition, a past tense specified by a
temporal PP (in 1937), and the latter locates the proposition that is evaluated in a
time prior to the speech time but posterior to the matrix tense. Note that the time of
evaluation can be shifted, for instance, to 1939, where the epistemic agents have
access to a new set of premises with which the proposition under evaluation is
incompatible.
The formula (Tense
2
(Epist-Mod (Tense
1
()))) does not display any
conceptual deviance: a proposition that takes place in a time
1
is evaluated with
respect to a time
2
(in (6), time
2
< time
1
< speech time), and not with respect to the
utterance time. The relevant issue to decide if the alleged hierarchy of inflectional
functional features is hardwired in such a way that tense cannot recursively scope
over epistemic modals is whether the scope relation observed in the bi-clausal
structure (6) can be expressed in a mono-clausal structure such as (7).

(7) In 1937 Great Britain could/might avoid war with Germany, but not in
1939

Meaning A: Nowadays it is thought that it was possible that Great Britain
avoided war in 1937; but (nowadays) it is thought that it was impossible
that Great Britain avoided war in 1939

Meaning B: In 1937 it was thought that it was possible that Great Britain
avoided war in a time t such as 1937 t < utterance time; but in 1939 it
was thought that it was impossible that Great Britain avoided war in a time

t such as 1939 t < utterance time

THE EMERGENCE OF ORDER IN SYNTAX 102
Sentence (7), but obviously not sentence (6), can express the meaning A,
where the PP in 1937 does not define an outer past tense where the proposition
Great Britain avoid- war with Germany is epistemically evaluated, but instead, it
is associated with the inner tense under the scope of the epistemic operator. In the
interpretation A, the evaluation time is, by defect, the utterance time (or rather a
present time that includes the utterance time nowadays). Meaning A is notably
more prominent than meaning B, but nonetheless it is not impossible for meaning
B to emerge from the mono-clausal structure (7). Consider the following passage,
in which such an interpretation is contextually salient:

From 1937 to 1939, there were no objections to German expansion: British Prime Minister
seemed to be able to do everything to warrant peace and stability and, although France
feared German hegemony, it was safer not to undergo any active protest, provided that
Maginot line was a secure defense. For these reasons, according to most analysts, {in 1937
Great Britain could avoid war with Germany/in 1937 it was possible that Great Britain
avoided war with Germany}, but peace expectations dimmed in April of 1939, when
Britain engaged to keep Polands security signing an assistance pact with Poland.

The modal could of the underlined sentence can be epistemically interpreted
under the scope of the outer past tense defined by the PP. Thus, one of the
interpretations of the mono-clausal structure (7) can be paraphrased as that Great
Britain avoided war with Germany was compatible with what most analysts knew
in 1937, an interpretation equivalent to the bi-clausal structure (6). Note that it is
not discussed whether Great Britain had the ability to avoid war in 1937 but not in
1939, but rather whether the possibility that Great Britain avoided war was
compatible with what some epistemic agents knew about the political context.
Mono-clausal structure (7) is therefore ambiguous, although one of the two pos-
sible readings is less prominent unless the context makes it salient, as in the above
passage.
The scope relation between nominal quantifiers and epistemic auxiliaries
constitutes a similar case: although nominal quantifiers appear more naturally un-
der the scope of epistemic modals, it is not impossible for them to scope over epis-
temic modals. At least in Catalan, the following reasoning does not introduce any
logical incongruence and the nominal quantifier must scope over the epistemic
modal (but see Iatridou and von Fintel 2003, who assert that nominal quantifiers
cannot scope over epistemic auxiliaries in English).

(8) (i) Alguns llibres sn a can Joan
(ii) Altres a la biblioteca
(iii) No sabem on s cada llibre
ANALYTIC SYNTACTIC PATTERNS 103
(iv) Per tant, pel que sabem, tots els llibres poden ser a can Joan, tot i
que no s possible que tots hi siguin

(i) Some books are at Johns house; (ii) others are at the library. (iii) We
do not know where each book is. (iv) Thus, as far as we know, all the
books may be at Peters house, although it is not possible that they are all
there

In sum, it remains unclear whether the Infl-area is rigidly defined to avoid
tense from scoping over an epistemic auxiliary, a scope relation that seems to be
possible although it is not generally prominent. In fact, if tense could not scope
over an epistemic modal, Cinques semantic argument (epistemic modality
presupposes an operand which is already tensed and thus its scope is external)
would be insufficient: it could not avoid an outer tense to scope over a modal
operator if such a modal operator scoped over an inner tense.

4.2.1.2 Tense and aspect. The second, and final, ordering restriction that may
reflect semantic constraints taken under consideration by Cinque is that tense must
appear higher in the hierarchy than grammatical aspect. The semantic reason given
for this is that aspect is more closely related to the predicate (it expresses different
ways of viewing the event expressed by the predicate) than tense (which locates
the time of the event whatever its aspectual make up with respect to the speech
time (p. 135), a broadly extended standpoint that Cinque attributes to Foley &
Van Valin (1984: 209ff). It could be that, informally speaking, the internal
structure of an event must be defined before the event is located in time, and hence
that an aspectual marker merged higher than tense could not be associated with
any event, and therefore the created object would be semantically vacuous.
But quantificational aspectual adverbs such as already, still, always, usu-
ally and often are expected, on purely conceptual grounds, to be able to scope
over epistemic adverbs or auxiliaries, as the bi-clausal structures in (9) suggest.

(9) a. Ja s el cas que probablement s a casa seva
It is already the case that he is probably at home

b. Ja s el cas que pot ser a casa seva
It is already the case that he can be at home

The interesting issue here is whether a quantificational aspectual marker can
scope over an epistemic marker when they both are at the same Infl-area, a scopal
THE EMERGENCE OF ORDER IN SYNTAX 104
relation that is conceptually acceptable. Although this is not the most prominent
reading and sounds quite unnatural to me, I think that it cannot be regarded as
impossible. Thus, the Catalan sentence ja hi pot haver vida a Mart (there
already can be life on Mars), an odd way of saying ja s possible que hi hagi
vida a Mart (it is already possible that there is life on Mars), can be a felicitous
conclusion to the following reasoning:

(10) Fins i tot fa noms un any, cap cientfic assenyat hauria pensat que hi
podia haver vida a Mart; per ara, grcies al descobriment tan important
que acabem desmentar, hem de concloure que {ja hi pot haver vida a
Mart/ja s possible que hi hagi vida a Mart}

Even only a year ago, no sensible scientist would have thought that there
could be life on Mars; but now, thanks to the very important discovery
aforementioned, it must be concluded that {there already can be life on
Mars/it is already possible that there is life on Mars}

This context eliminates the deontic reading of the modal: it is not discussed
whether scientists permit life to exist on Mars, but the compatibility of the pro-
position There is life on Mars with what is known about the conditions of the
planet and what is known about the conditions where life may emerge. It also
eliminates the interpretation whereby ja is under the scope of the epistemic mo-
dal: life was already existing before the discovery. Thus, the reasoning in (10)
forces the underlined mono-clausal structure to be interpreted as that life exists on
Mars is already compatible with our current epistemic model.
In sum, the two hypothetical ordering restrictions that were attributed to C-I
considerations have been revised, and we have concluded that:

(a) tense can be coerced to scope over epistemic modality, a relation that is not
prominent (or, according to Cinque, impossible),

(b) aspect appears lower than tense because it is more closely related to the
event; however, quantificational aspectual adverbs cannot readily scope
over tense, contrary to what would be expected on purely conceptual
grounds; nonetheless, it seems that quantificational aspect can be coerced
to scope over tense,

ANALYTIC SYNTACTIC PATTERNS 105
(c) the situation in which tense could not scope over epistemic modality and
quantificational aspectual adverbs could not scope over tense would not
reflect purely conceptual restrictions.

4.2.2 Some allegedly primitive order restrictions. Cinque (1999a: 135-136) pro-
vides three cases of ordering restrictions that allegedly do not derive from con-
ceptual restrictions, which will be briefly illustrated in this section and considered
in detail in section 2.3. Cinques method for proving that an ordering restriction
between two projections of a mono-clausal structure cannot be derived from
conceptual constraints involves constructing a bi-clausal structure where the order-
ing relation forbidden in the mono-clausal structure becomes acceptable and does
not introduce any logical incongruity.

Evidential mood, instantiated by affixes, particles or adverbials, is higher than
epistemic modality

(11) a. Allegedly John will probably give up (Italian)
b. *Probably John will allegedly give up

(12) a. (?)Evidentemente Gianni ha probabilmente lasciato lalbergo
Evidently G. has probably left the hotel

b. *Probabilmente Gianni ha evidentemente lasciato lalbergo
Probably G. has evidently left the hotel

(13) probabile che sia evidente que lui il colpevole
It is probable that it is evident that he is the guilty one

Retrospective/proximative aspect (soon) are higher than prospective aspect
(almost/imminently)

(14) a. He will soon almost be there
b. *He will almost soon be here

(15) a. He is about to soon be admitted to hospital




THE EMERGENCE OF ORDER IN SYNTAX 106
Habitual aspect (di solito) is higher than terminative aspect (pi)

(16) a. Dopo le 10, Gianni non beve di solito pi niente
After 10 oclock, G. drinks usually no longer anything

b. *Dopo le 10, Gianni non beve pi di solito niente

(17) Gianni ha smesso di andare di solito a trovare suo padre la sera
G. stopped usually going to visit his father in the evening

It remains unclear whether the so-called terminative aspect must occur lower
than the habitual aspect. As Svenonius (2002: 213-214) has observed, in English,
both precedence relations and both scope relations are possible. The static char-
acter of the terminative aspect in Italian, and its relatively low position, may be a
peculiarity of pi.

(18) a. After 10, John usually no longer drinks anything
It is usually the case that John no longer drinks anything

b. After 10, John no longer usually drinks anything
It is usually the case that John no longer drinks anything

Although Cinques arguments for the inherently grammatical status of the Infl-
hierarchy are scarce and too weak to arrive at a well-grounded conclusion, Cinque
may be correct in claiming that the set of combinations in a mono-clausal structure
allowed by UG is smaller than the set of combinations that are legitimate on mere-
ly conceptual grounds, a possibility that merits some attention.

4.3 Toward a principled account for some order restrictions
Interestingly, auxiliaries must be combined in a rigid order in a language like
English in such a way that a modal (should) or a temporal auxiliary (will) must
precede aspectual auxiliaries. In fact, if a perfective periphrasis (have + past
participle) and a continuative auxiliary (being) appear in the same sentence, the
perfective periphrasis must precede the continuative auxiliary, and they both must
precede the lexical verb with a voice mark [-d] (produce-d), which appears in the
final position of the clauses in (19). If a complementizer (that) is present, it must
precede all aforementioned categories.

ANALYTIC SYNTACTIC PATTERNS 107
(19) a. [
C
That (this model) [
Mod
should [
Asp-Perf
have been [
Asp-Cont
being
[
voice
produced]]]]]
b. [
C
That (this model) [
Fut-T
will [
Asp-Perf
have been [
Asp-Cont
being [
voice

produced]]]]]
[C [Mod [T [Asp-Perf [Asp-Cont [-V [V]]]]]]]

These facts do not seem to be an accident of English grammar, since ordering
restrictions among grammatical categories (verbal auxiliaries, affixes, free par-
ticles or adverbials) do not vary cross-linguistically. As important and pressing as
this question is, I cannot attempt to construct a good answer for it. However, I
hope to show that there is no reason to think that hierarchies are hardwired in a
clumsy fashion.

4.3.1 Voice and tense. Lets begin with the ordering between and tense. If, as
argued in section 3.4, forms a discontinuous element with V and Tense (an in-
flectional head for the Infl-like feature of tense) forms a discontinuous element
with C, then it is quite natural that cannot occur higher in the structure than
Tense; if it did, the two discontinuous syntactic templates would be tangled, and
thus the C-Infl connection and the -V connection would be broken off. Therefore,
some functional features would be required to be ordered in such a way that two
discontinuous units do not get tangled, a requirement that may be expressed in
terms of the following Prohibition Against Tangled Structures.

Prohibition Against Tangled Structures (PATS)
Two discontinuous syntactic templates [A
1
A
2
], [B
1
B
2
] cannot be broken
off by yielding the tangled object [A
1
[B
1
A
1
] B
2
]

It would be more illuminating to define the semantic effects of tangled struc-
tures. On the particular case of the ordering between tense and voice, Kratzer
(1996) observes that if the external argument is not strictly an argument of V, then
nothing prevents voice from carrying out temporal or aspectual information, or in
fact, from occurring higher than temporal or aspectual projections. Kratzer ex-
plores a way of locating the voice projection lower than tense based on the order in
which two semantic operations, Event Identification and Existential Quantifica-
tion, apply. Kratzer proposes that, whereas the operation of Existential
Quantification is responsible for providing a truth-value (p. 122), Event Identifi-
cation is one of several admissible conjunction operations responsible for chain-
ing together various conditions for the event described by a sentence; more
THE EMERGENCE OF ORDER IN SYNTAX 108
specifically, the addition of the external argument might proceed via the operation
of Event Identification, yielding the connection between the Aktionsart of a verb
and the thematic role of the external argument. Kratzers operation of Event Iden-
tification is a post-syntactic algorithm that checks the compatibility of the prop-
erties of two predicates, the external argument and the verb.
Kratzer asserts that voice can appear anywhere in the hierarchy of a verbs
inflectional heads as long as the event argument is not existentially quantified (p.
125), since the operation of Event Identification cannot apply once a truth value
has been obtained. If we have more than one inflectional head in addition to voice
(possibly all of Tense, Agr, Mood and Aspect), then we have to find out which
head does the existential quantification (p. 126). For the time being, let us tenta-
tively assume that Voice is located directly above the VP, but stay open to the
possibility that it may turn out that there are intervening functional heads after all.
Thus, the semantic algorithms of Event Identification and Existential Quanti-
fication over events cannot apply in the right order in the tangled structure [C [ T]
V]: the external argument and V can be properly linked at the right stage before
the event variable is bound.
An alternative way of grasping Kratzers intuition could be in terms of a
condition on representations that avoided vacuous quantification. Consider the
following definition. See Kratzer (1995) and Chomsky (1982, 1991) on the prohi-
bition against vacuous quantification in natural language.

Prohibition Against Vacuous Quantification (PAVQ)
For every occurrence of a variable x there must be a quantifier Q binding x,
and for every quantifier Q there must be a variable x such that Q binds x

If voice and its argument (the so-called external argument) occur higher than
tense (e), we obtain the formula (20.a), whereas if they occur lower than tense,
we obtain (21.b).

(20) Peter bought the house
a. e(e: past) [Agent(Peter)(e) & bought(the house)(e)]
b. Agent(Peter)(e) & [e (e: past) [bought(the house)(e)]]

In (20.a) the external argument and the verb are unproblematically chained
before tense has provided a truth-value and before all occurrences of the eventive
variable are bound by the existential quantifier; but in (20.b) Event Identification
fails to link the properties of the event denoted by the external argument and the
ANALYTIC SYNTACTIC PATTERNS 109
verb (because a truth-value has been obtained via Existential Quantification), and
the occurrence of the event variable of the external argument (Agent(Peter)(e)) is
free. Thus, the PAVQ is violated if the voice projection (the voice head and its
argument) occurs higher than the tense projection.
The PAVQ may be useful in understanding why some combinations that do
not seem to involve tangled structures and that seem to be legitimate on pure
conceptual grounds are not allowed. The general expectation is that a conceptually
possible scope relation is banned because it involves a free variable or a quantifier
with no variable to bind. Lets examine, in the light of the PAVQ, the two ordering
restrictions illustrated in 4.2.2 that seemed to be irreducible to purely conceptual
restrictions.

4.3.2 Quantificational aspectual adverbs and completive/prospective aspectual
adverbs. Aspectual adverbs such as usually, often, always or soon can be re-
garded as quantifiers over events, whereas others, such as completely or partial-
ly, can be regarded as predicates that must be conjoined to the verb. Accordingly,
if predicate adverbs were to occur higher than quantificational adverbs, the re-
sulting expression would be equivalent to that of voice occurring higher than
tense: the eventive variable associated with the predicate adverb would be free,
and thereby would violate the PAVQ.

(21) a. Peter will have soon completely eaten the sandwiches
soon(e): [ (completely)(e) & Agent(Peter)(e) & eat(sandwiches)(e)]

a. *Peter will have completely soon eaten the sandwiches
(completely)(e) & [soon(e): [Agent(Peter)(e) & eat(sandwiches)(e)]]

b. Peter usually completely eats the sandwiches
usually(e): [ (completely)(e) & Agent(Peter)(e) & eat(sandwiches)(e)]

b. *Peter completely usually eats the sandwiches
(completely)(e) & [usually(e): [Agent(Peter)(e) & eat(sandwiches)(e)]]

Note that a prospective adverb like almost cannot readily be combined with a
predicate that does not involve a gradation:
#
he is almost tall vs. he is almost 2
meters tall;
#
he is almost sitting down vs. he is almost at home; in a closed
interval [0, 1] of a property p where 0 is the minimal degree of p and 1 the
maximal degree of p, and for all x such as 0 x K, almost(p) denotes an x
prominently close to 1. This suggests that there must be a semantic conjunction
THE EMERGENCE OF ORDER IN SYNTAX 110
operation (Kratzers Event Identification) that checks whether the properties of
almost are compatible with those of the verb. If almost is merged higher than
the existential quantifier, then the eventive variable of the adverb cannot be bound,
which yields a violation of the PAVQ condition.

(22) a. Peter will soon almost be at home
soon(e): [(Almost)(e) & be (Peter, at home)(e)]

a. *Peter will almost soon be here
(Almost)(e) & [soon(e): [be (Peter, at home)(e)]]

b. Peter often is almost on time
often(e): [(Almost)(e) & be (Peter, on time)(e)]

b. *He almost is often on time
(Almost)(e) & [often(e): [be (Peter, on time)(e)]]

Thus we account for Cinques observation that prospective adverbs cannot
occur higher than proximative adverbs (22), a scope relation that cannot be banned
by purely conceptual restrictions, along with the more general observation (21)
that predicate aspectual adverbs (completely, partially) cannot occur higher
than quantificational aspectual adverbs (always, often, soon, usually).

4.3.3 Evidential mood and epistemic modality. There is a clear distinction between
evidential categories such as the adverb evidently or the adjective evident and
epistemic categories such as the adverb possibly or the adjective possible: the
former are veridical operators, while the latter are non-veridical operators.

(23) Given an epistemic model M, an operator Op is
a. veridical iff [[Op(x)]] = 1[ [[x]] = 1 in M
b. non-veridical iff [[Op(x)]] = 1 [ [[x]] = 1 in M
c. anti-veridical iff [Op(x)]] = 1 [ [[x]] = 0 in M

Thus, when the sentence it is evident that he has left the hotel is uttered, the
proposition under the scope of evident, he has left the hotel, must be true in the
relevant M. When the sentence it is possible that he has left the hotel is uttered,
the proposition he has left the hotel is neither true nor false in M.

ANALYTIC SYNTACTIC PATTERNS 111
An epistemic marker scopes over a proposition that does not yet have a truth-
value specified in M because its variable over worlds has not been bound, whereas
an evidential marker takes a proposition with a truth-value in M because its
variable over worlds has already been bound by an overt or covert modal quanti-
fier. If epistemic modals are merged lower than evidential markers, they provide
an existentially quantified proposition for the evidential marker. However, when
epistemic modals are introduced higher than an evidential marker, problems arise,
suggesting that there is no variable over worlds to be bound by the epistemic
modal.
More precisely, the source of the violation of the PAVQ in (24.a) lies in the
conjunction of (i) an opacity factor and (ii) a poverty factor: (i) the evidential
mood projection prevents the higher epistemic modal from binding a lower varia-
ble over worlds and (ii) the evidential mood projection fails to introduce a suitable
variable over worlds for the higher epistemic modal. If the epistemic modal scopes
over a full CP containing an evidential projection, the scope relation is legitimate
(24.b): the matrix clause where probabile is merged contains a variable over
worlds to be bound with no intervening category.

(24) a. *Probabilmente Gianni ha evidentemente lasciato lalbergo
Probably G. has evidently left the hotel

b. probabile che sia evidente que lui il colpevole
It is probable that it is evident that he is the guilty one

This reasoning can be applied to other familiar cases. If Cinque (1999a) and
Iatridou & von Fintel (2003) were correct in claiming, respectively, that quanti-
ficational aspectual adverbs and tense cannot scope over epistemic modality and
that nominal quantifiers cannot scope over epistemic modals, the source of such
restrictions could be the conjunction of the opacity factor in (i) and the poverty
factor in (ii):

(i) that tense, quantificational aspectual adverbs and nominal quantifiers fail to
bind its variable, which is arguably in the VP, across epistemic modality
(this possibility is in fact explored by Iatridou & von Fintel 2003 in the
case of epistemic modality and nominal quantifiers), and

(ii) that the projection of epistemic modality fails to introduce a suitable varia-
ble.

THE EMERGENCE OF ORDER IN SYNTAX 112
Nonetheless, discourse may favor these non-prominent readings and coerce
tense, aspectual quantifiers and nominal quantifiers into penetrating the lower
epistemic projection or the epistemic modality projection into introducing a new
eventive variable for those quantifiers.

4.4 The Full Interpretation Principle
The main idea I have tried to pursue in this chapter is that the Infl-hierarchy is
not an arbitrary solution, but satisfies conditions imposed by the C-I system;
concretely, it must satisfy the PATS and the PAVQ. A syntactic object that vio-
lates the PAVQ contains elements that are not useful to the Conceptual-Intensional
system, such as a free variable or a quantifier that does not bind any variable.
Similarly, a discontinuous template that is broken off automatically becomes a
useless element: C and would fail to connect Infl and V to the superordinate
structure/discourse and to the external argument/manner adverb. Thus, the PATS
and the PAVQ are two instances of the more general Full Interpretation Condition.

Full Interpretation Condition
An object generated by a grammatical component must be constituted only of
useful features for a particular level of interpretation when it attains such a
level of interpretation

Consequently, the empirical results of the cartographic project do not lead us
to a primitive element of the syntactic component, but rather to the study of the C-I
system. Although we are far from a good understanding of the restrictions the C-I
system imposes on the combination of functional categories, there is no compel-
ling reason to think that cartographies belong to the set of primitive elements of
the faculty of language, as there is no compelling reason to think that a deeper
understanding of a phenomena cannot be attained by reducing an apparent prop-
erty of it to higher order principles.
Devices such as the PATS and the PAVQ are clearly relative to the levels of
interpretation of the C-I system, and hence one may be skeptical about coding
them in the theory of grammar in the form of derivational devices or in the form of
universal hierarchies. As Chomsky (1991: 151) observes, when he deals with the
prohibition of vacuous quantification, an aspect where language differs from
typical formal systems that permit vacuous quantification freely:

[I]f some theory of grammar stipulates specific devices and rules to bar such constructions
and interpretations [see below (28) and (29)], we conclude that it is the wrong theory: it is
generating expressions and structures too accurately and is therefore incorrect. There is
ANALYTIC SYNTACTIC PATTERNS 113
nothing paradoxical about this conclusion. The unwanted constructions are excluded on
general grounds, in terms of the overarching condition FI [Full Interpretation]; there is no
reason to suppose that the mechanisms of language include superfluous devices and rules
to achieve, redundantly, the same result in special cases. Similarly, the phonological com-
ponent contains no rules to express special cases of general properties of universal phonet-
ics or of phonetic representations.

(25) a. who John saw Bill
b. who did John see Bill
c. every some person

(26) a. who did Mary see him
b. the man that Mary saw him

The theory of the syntactic component of the faculty of language need not
and therefore must not encode devices that translate the kind of external
requirements to be satisfied: the syntactic component

does not have to be defined
to avoid the generation of (25)-(26); similarly, if cartographies derive from the
Full Interpretation condition, cartographies do not reflect the knowledge of gram-
mar. To be accurate, there is no overgeneration: there are no generated expressions
that cannot be used at the performance systems, but a multitude of effects that ex-
pressions may have; if the devices of the levels of interpretation of the C-I system
that are independently in place are provided with the information to discriminate
the multiple type of deviance effects an empirical requirement that the theory
must satisfy, overgeneration disappears. The syntactic component is thus a com-
putational mechanism that arranges and re-arranges features or in-structions, with
a wide range of external effects, and with no real distinction between grammatical
and ungrammatical expressions (see Chomsky 2005a and references cited therein,
especially Borer 2004.a, b).

4.5 Cartographic effects
It is important to note that discontinuous syntactic objects (27) can coexist with
analytic syntactic objects (28).

(27) [uP]
+C
(28) uP

u [uP]
+Infl
u P


u ...

THE EMERGENCE OF ORDER IN SYNTAX 114
Discontinuous syntactic objects do not contradict the idea that the C-I system
requires articulated syntactic representations with semantically devoted positions,
but rather they adhere to it: according to the line of argumentation of this study,
the insertion of Infl-like features into C and Infl is the way to provide Infl-like
features with a [+clause typing] occurrence and a [-clause typing] occurrence, the
semantic instruction provided by Infl-like features being orthogonal to the [clause
typing] distinction. Thus, when an Infl-like feature such as tense is introduced into
the C-Infl discontinuous syntactic template, it is simultaneously associated with a
[+clause typing] position and with a [-clause typing] position. Both occurrences of
tense are minimally distinguished, and therefore, to be accurate, they are not the
same element in two positions but two co-variations. Thus, the creation of [clause
typing] variations (or discontinuous syntactic objects) is at the service of the crea-
tion of syntactic representations with semantically devoted positions (or analytic
syntactic objects). Discontinuous syntactic objects are a subtype of analytic syn-
tactic objects.






CHAPTER 5

SYNCRETIC SYNTACTIC PATTERN

This chapter argues that syncretic syntactic patterns, where more than one feature is
matched in one projection, are favored, under special syntactic circumstances, by a
principle of structural minimization, the Maximize Matching Effects Principle. I begin by
considering the nature of V-movement and verbal inflection, concluding that (i) move-
ment of both maximal and minimal categories takes place in narrow syntax, (ii) func-
tional features attached on V convert V into a multicategorial word, which does neither
mean that they are uninterpretable on V nor that they must be deleted, (iii) the position of
V is determined by the presence (and not the richness) of inflectional features attached on
V, and (iv) the so-called Null Subject Parameter derives from the richness (and not the
presence) of verbal inflection. I propose three instances of syntactic syncretic patterns: (i)
in the TP of Null Subject Languages, -features are matched by a finite V in T and a-
features such as topicality and referentiality are matched in SPEC-T by a preverbal sub-
ject or a CLLD element (Sol 1992); (ii) in the TP of V2 languages, -features are
matched by A-movement to SPEC-T and tense is matched in T (and not in C) only when
the first constituent is a subject (Zwart 1993), which avoids initiating the CP to match
only tense, a feature that can be matched either in T or in C, due to its discontinuous
character; (iii) in English, nominative interrogatives match -features and wh-features in
SPEC-T, which becomes a mixed A-/A-position (similar patterns are obtained in Dutch
with topical/focal 2SG subjects). Finally, I raise the possibility that C and T may contract
when both -features and a-features are matched in T in certain languages, which would
yield comp-trace effects.


5.1 On structural minimization
All syntactic models implicitly or explicitly assume a constraint that bans
superfluous structure. Consider, as a starting point, the following very rough for-
mulation of this constraint in terms of an Elemental Economy Principle.

Elemental Economy Principle
A syntactic object cannot contain superfluous elements

Since the quality of superfluity referred to a syntactic object is not inherent to
the object but relative to the requirements imposed precisely by a level of inter-
THE EMERGENCE OF ORDER IN SYNTAX 116
pretation, what bans syntactic objects with superfluous elements is not an econ-
omy or minimization constraint that favors simplicity in the creation of structure,
such as the Elemental Economy Principle, but rather a general convergence condi-
tion such as the Full Interpretation Condition.

Full Interpretation Condition
An object generated by a grammatical component must be constituted of only
useful elements for a particular level of interpretation when it attains such a
level of interpretation

Thus, a syntactic object containing elements that do not provide useful
instructions for a particular level is not an excessively sophisticated object but an
object crashing at that particular level of interpretation; such an object should be
filtered by a convergence condition, such as the Full Interpretation Condition,
which applies at the relevant level of interpretation, not by an economy principle.
Convergence conditions and principles of structural minimization are not to be
confused.
The conception of the Full Interpretation Condition in Chomsky & Lasnik
(1993) and in the foundational texts of Minimalism (Chomsky 1991, 1993, 1995)
has an unclear status; for instance, Chomsky & Lasnik (1993: 27) claim that
[T]here are also certain general ideas that appear to have wide acceptability,
among them, principles of economy stating that there can be no superfluous steps
in derivations (Chomsky 1986a, chapters 2-4 of [Chomsky 1995] and Chomsky
(1991: 130) argues that [S]ome of these guidelines have a kind of least effort
flavor to them, in the sense that they legislate against superfluous elements in
representations and derivations. Thus the notion of Full Interpretation (FI) re-
quires that representations be minimal in a certain sense. However, to be ac-
curate, the Full Interpretation Condition does not require representations to be
minimal but rather to provide objects that converge with the relevant legibility
requirements.
In chapter 3 I revised two plausible principles of derivational minimization,
the Strict Cycle Condition, a particular version of the Transformational Cycle, and
the Phase-Impenetrability Condition, and I concluded that it is unclear whether
they play any role in narrowing the set of conceivable syntactic derivations: there
is no compelling argument for strict cyclicity, especially once parallel probing at
the phase level (or equivalently, at the level of discontinuous units) is taken into
account, and successive cyclicity is not relative to the so-called phase heads. Note
that the Relativized Opacity Principle, proposed as an alternative to account for
wh-islands effects and subextraction patterns, can be indistinctly viewed either as
SYNCRETIC SYNTACTIC PATTERNS 117
a constraint reducing the search domain of a probe and the number of possible
derivations, or as a sign that the searching power of a probe is not unbounded.
Thus, it is unclear whether the Relativized Opacity Principle is an efficiency
factor or an effect of the feebleness of probes.
This chapter explores whether genuine and unambiguous factors of structural
minimization matter for the recursive generative procedure of the syntactic com-
ponent. The candidate that will be examined is the Maximize Matching Effects
Principle, understood as follows:

Maximize Matching Effects Principle (MMEP)
Match as many features as possible using the smallest span of structure

The most obvious difficulty in exploring the MMEP is how to measure the
size of a span of structure where a set of type features are instantiated by external
or internal merge of token features. Take a syntactic derivation that must match
two different type features F
typ1
and F
typ2
with two different token features F
tok1

and F
tok2
. In assigning each type feature a particular position in the syntactic
hierarchy of functional type features, viewed as a strict order relation O, the syn-
tactic component may connect F
typ1
and F
typ2
to the rest of elements and to each
other (1.a) or it may not connect them to each other although it connects them to
the rest of type features (1.b).

(1) a. O
1
= {<F
typ3
, F
typ1
>, <F
typ1
, F
typ2
>, <F
typ2
, F
typ4
>, <F
typ3
, F
typ2
>,
<F
typ3
, F
typ4
>, <F
typ1
, F
typ4
>}

b. O
2
= {<F
typ3
, F
typ1
>, <F
typ3
, F
typ2
>, <F
typ2
, F
typ4
>, <F
typ3
, F
typ4
>,
<F
typ1
, F
typ4
>}

In O
2
, F
typ1
and F
typ2
are not ordered with respect to each other (<F
typ1
, F
typ2
>
O
2
and <F
typ2
, F
typ1
> O
2
), and they appear in the same minimal context (their
immediate precedent is F
typ3
and their immediate consequent is F
typ4
).
In standard arborian representations, it could be claimed that two different
type features occur in the same minimal context when they are contained in the
same head of the same projection. Thus, the syntactic hierarchies of functional
features given in (1) can be expressed as in (2).


THE EMERGENCE OF ORDER IN SYNTAX 118
(2) a. F
typ3
b. F
typ3


F
typ3
F
typ1
F
typ3
[F
typ1,
F
typ2
]

F
typ1
F
typ2
[F
typ1
, F
typ2
]

F
typ4




F
typ2
F
typ4
F
typ4
...

F
typ4
...

(2.a) is an analytic syntactic representation where all type features are strictly
ordered, whereas in (2.b) there are two type features that are not ordered with
respect to each other, F
typ1
and F
typ2.
The tree diagram (2b) contains what can be
called a syncretic syntactic pattern, since two different type features appear in the
same minimal context, i.e., in the same head ([F
typ1,
F
typ2
]). If it can be argued that
two different type features can be instantiated in the same head or projection, then
there is reason to think that the MMEP is operative in the syntactic component, as
the span of structure used to instantiate two type features in the syncretic syntactic
pattern (2.b) is smaller than the one used in the analytic syntactic object (2.a): in
the former case, a sole projection is used, while in the latter two different pro-
jections are used.
Whereas the preceding chapter argues that cartographic effects (analytic
syntactic patterns and discontinuous syntactic patterns) derive from the C-I
requirement of bringing semantic distinctions into the syntactic representation,
this chapter states that a principle of structural minimization, the MMEP, is the
source of anticartographic effects (syncretic syntactic patterns). Note that the
MMEP is not concerned with banning useless elements or crashing objects but
with condensing the structural span where useful features are coded and matched;
in this sense, it is a genuine principle of structural minimization.
Accordingly, the way to explore whether the MMEP plays any role in the
computations of the syntactic component is to determine whether syncretic
syntactic patterns exist and define the special syntactic circumstance that favors a
principle of structural minimization to the detriment of the C-I legibility condition
that requires semantically devoted positions or a strict linear order of type fea-
tures. Several controversial issues such as the role of verb movement in the com-
putations of the syntactic component, the nature of inflectional verbal morpholo-
gy, the inexistence of vacuous movement, the existence of mixed A-/A-positions
and the possibility that structure contracts will be studied carefully leading to an
array of delicate conclusions on the mechanics of creating structure.
SYNCRETIC SYNTACTIC PATTERNS 119

5.2 Why and where V moves
Consider the following four propositions on verb movement (henceforth, V-
movement) as a starting point. Although they are not uncontroversial, they reflect
predominant views.

I. V-movement does not take place at narrow syntax but at the PF-
branch

II. Infl-morphology on V is uninterpretable

III. V moves from the P if and only if it bears rich Infl-morphology

IV. The Null-Subject Parameter derives from V-to-T movement, and
hence from the existence of a rich Infl-morphological paradigm in
a particular language

Whereas propositions I and II are characteristic of the main stream of
minimalism, and they are not present, as far as I know, in the pre-minimalist
period of the Principles and Parameters era, propositions III and IV reflect two
important generalizations formulated in the 80s that allowed for certain progress
in deriving the degree of cross-linguistic variation from morphological properties:
if V shows up rich Infl-morphology in a language L, L displays movement of V
outside the P (see, among others, Pollock 1989) and allows for postverbal sub-
jects and null subjects (see, especially, Taraldsen 1980, Chomsky 1981 and Rizzi
1982b, 1986). In spite of certain empirical complications, propositions III and IV
have not been rejected, although there has not been much progress in under-
standing them. I find that the belief that SPEC-T must be universally filled (the
Extended Projection Principle translated in current terminology as the EPP feature
of T) and the minimalist tenets conveyed in propositions I and II have not been a
good guide.

Proposition I. V-movement does not take place at narrow syntax but at the
PF-branch

It may be worth considering two arguments given for the idea that V-move-
ment does not take place at narrow syntax (see Acedo & Fortuny 2006 for a more
general discussion of operations that do not seem to have semantic effects, such as
scrambling and clitic climbing).
THE EMERGENCE OF ORDER IN SYNTAX 120
Firstly, Chomsky (1995) observes that, whereas movement of maximal
categories follows the Extension Condition as it always targets the root, head
movement never does, since it is a case of syntactic infixation. For several reasons
given throughout this study, Chomskys observation is no longer empirically
valid: at the phase level, the Extension Condition is not operative, since move-
ment of maximal categories takes place in parallel; crucially, because movement
to SPEC-T and movement to SPEC-C occur during the same derivational step,
only the operation that creates SPEC-C extends the root, and any operation of A-
movement is a case of syntactic infixation, triggered by the phase head C in
Chomskys (2005b) analysis. Indeed, if one should cast doubt on any of the prem-
ises of the argument at hand, this is the Extension Condition, an even stricter de-
rivational constraint than the Strict Cycle Condition: if operations take place in
parallel at the phase level, the Extension Condition is neglected at the phase level
and the classical empirical evidence given for the Strict Cycle Condition (that is,
for an extension condition at an inter-phase level) is not relevant, as argued in
chapter 3.
Secondly, whereas movement of maximal categories has clear semantic ef-
fects (it provides peripheral positions, scope positions for quantifiers and binding
positions for nominative and object DPs), head movement is usually argued to be
semantically vacuous (Chomsky 2000; but see Acedo & Fortuny 2006, and refer-
ences cited therein, for cases in which clitic climbing reflects semantic changes in
the eventive structure). On the one hand, the observation that V-movement does
not have the type of semantic effects that are attested in movement of maximal
projections cannot be taken as a valid premise to conclude that it should be con-
fined to the PF branch: V cannot have the same semantic contribution as a quanti-
fier because it is not a quantifier and it cannot be interpreted correferentially to an
anaphor because it is not an argument. On the other hand, it should not be ignored
that, as discussed below, V-movement interacts with movement of maximal cate-
gories in such a way that a DP moves to SPEC-T only when V remains inside the
P, an observation that can be more easily accounted for if both V-movement and
DP-movement take place at the same computational level.
Thus far there is no reason to assume that movement of maximal categories
and movement of minimal categories take place at different computational levels,
say, the former at narrow syntax and the latter at the PF branch. They might both
take place at narrow syntax or at the PF branch. In fact, the only requirement at
the C-I system is for token features and type features to be matched; whether or
not probe-goal relations become spatially material by means of internal merge
could be a matter of parameterization linked to the morpho-phonological com-
ponent.
SYNCRETIC SYNTACTIC PATTERNS 121
However, some well-known patterns in English discussed by Laka (1990)
suggest that V-movement takes place at narrow syntax, since it is relevant for
operations that hold at the C-I system. As illustrated in (3.a), a negative polarity
item occurring at SPEC-T cannot be licensed by a lower negative marker in an
assertive matrix sentence; if V raises to C carrying the attached negative marker
as a free-rider, then the negative polarity item can be licensed (3.b).

(3) a. *Anybody didnt come
b. Didnt anybody come?

For these reasons, I shall keep to the idea that both movements of minimal and
maximal categories take place at narrow syntax, a position that will be further
supported as the argumentation proceeds (see Zwart 2001 for a broader discussion
of the nature of V- movement).
Another interesting issue observed by Laka (1990) is that an embedded C
selected by a negative matrix verb has the ability to license a negative polarity
item in SPEC-T. In (4.a), anybody is unlicensed, because it is not in the domain
of the embedded negative marker, whereas in (4.b) it is licensed. The contrast
between (4.b) and (4.c) shows that it is not the negative characterization of matrix
V that directly licenses the embedded negative polarity item, but the embedded C:
if the embedded CP layer is absent, the negative polarity item is unlicensed
(Progovac 1988).

(4) a. *I say that anybody didnt leave
b. I deny that anybody left
c. I deny *anybody/everything

This suggests, as argued by Laka (1990), that a matrix negative verb selects an
embedded C specified as [+neg], just as a matrix interrogative verb selects an
embedded C specified as [+wh] (I wonder what he saw), a further piece of
evidence for the idea that negation can be operative on C.

Proposition II. Infl-morphology on V is uninterpretable

In section 1.3, I argued on the basis of the problem of generality, the problem
of determinacy and the problem of consistency that the source of movement was
not the existence of uninterpretable functional features located in the syntactic
hierarchy of features just to be deleted, a device called suicidal greed. -features
of T, like the rest of Infl-like features located in the hierarchy (tense, mood,
THE EMERGENCE OF ORDER IN SYNTAX 122
modality and negation), were claimed to provide useful instructions for the C-I
system since they are responsible for referential displacement. Since the property
of movement dissolves into the basic operation of merge, the trigger of internal
merge has to be the same as that of external merge: a type feature present in the
syntactic hierarchy must be instantiated by a suitable token feature and a token
feature needs to be hosted in a suitable position in order to be properly interpreted;
when a type feature is instantiated by a token feature present in its search domain,
a probe-goal relation takes place, which may then become spatially material.
A partially different question is whether the tokens of inflectional features
attached to V are uninterpretable, as stated in proposition II, and consequently,
whether the source of V-movement lies in deleting the uninterpretable features on
V before the derivation arrives at the C-I system. If that were the case, un-
interpretable features of V would instantiate a type Infl-like feature and be deleted
from the syntactic object when moving toward the C-I system.
It could be argued that -features are uninterpretable on V because V denotes
properties of an event and person-number distinctions are relative to arguments of
that event, to the nominative DP in the case of subject agreement, to the
accusative DP in the case of object agreement or to the dative DP in the case of
dative agreement. This reasoning can be applied to other functional features that
appear attached on V, such as tense and mood, as they do not reflect internal
properties denoted by V such as telicity or duration, but rather an ordering relation
between the time where the event takes place and a reference time in the case of
tense and the [realis] characterization of the proposition or the source of
evidence for a statement in the case of mood.
Let us assume that the reasoning outlined above is correct, and that therefore
-features, tense and mood are uninterpretable on V. As will be argued below,
type -features can be alternatively instantiated through movement of a V inflec-
ted for -features or by a DP; similarly, it is a strong cross-linguistic tendency for
languages to instantiate tense or mood type features (and generally, functional
features) through V-movement or by inserting a maximal category (an adverbial)
or a minimal category (a free particle). This means that matching Infl-like type
features could either provide a suitable position for the relevant token to be inter-
preted or a suitable position for a token to be deleted in the sole case that such a
token is attached to V. All this seems to be an unnecessary complication: it re-
mains unclear why the morpho-phonological property of being attached or free
should be relevant in determining whether the matching type feature offers a
position for the token to be deleted or to be interpreted. In fact, to say that -
features, tense and mood are uninterpretable on V seems to be a misleading way
of expressing that V contains (in some languages) not only inherent features (say,
SYNCRETIC SYNTACTIC PATTERNS 123
thematic-features and Aktionsart features), but also non-inherent features that are
acquired through the morphological process of attachment.
There is still a further problem for the assumption that attached features are
uninterpretable and must be deleted before the object attains the C-I system. V
undergoes (overt) movement only when it displays rich inflectional paradigms;
otherwise, it remains inside the P (proposition III). If the need for uninterpretable
attached features to be deleted were the source of V-movement, rich inflection
would never be overt: when inflectional paradigms were pre-syntactically rich, V
would undergo movement and its attached features would be deleted during syn-
tactic computations. Accordingly, moved verbs would be expected to be bare
verbal roots and in situ verbs, poorly inflected verbs or bare verbal roots. How-
ever, we can see that displaced Vs do not get their inflection deleted from the
object to which morpho-phonological form is assigned.
Stipulating that the deletion of attached features takes place during the C-I
branch after the syntactic object has been sent to the A-P system is futile: if V has
undergone movement, a probe-goal relation must have taken place; in this case, a
type feature is instantiated and a token feature is hosted or deleted (in the sole
case that it is attached). On the one hand, it is unclear why instantiation of a type
and deletion of a token should have different timings, the former holding before
the syntactic object is sent to the A-P system, and the latter during the C-I branch
after the syntactic object is sent to the A-P system; optimally, both instantiation
and deletion should be two immediate effects of the same matching operation, and
therefore should take place during the same derivational step. On the other hand,
if both typing and deletion took place at the C-I branch, after the syntactic object
was sent to the A-P system, then there would be no reason for V to overtly move.
The most straightforward account is that there is no deletion operation and no
uninterpretable features to be deleted, all token features being hosted by a suitable
type feature at the same step at which the type feature is automatically instanti-
ated.
These conceptual considerations solely serve to show, along with the problem
of generality, the problem of determinacy and the problem of consistency exposed
in section 1.3, that the interpretable/uninterpretable dichotomy can neither be a
powerful analytical tool nor provide a deeper understanding of the relation bet-
ween displacement patterns and inflectional paradigms. Here I shall keep to the
idea that any device especially constructed for displacement is suspicious. There
are thus four valid alternatives strategies for a token feature to instantiate a type
feature, with no interpretable/uninterpretable dichotomy: (a) external merge of a
minimal category, (b) internal merge of a minimal category, (c) external merge of
a maximal category and (d) internal merge of a maximal category.
THE EMERGENCE OF ORDER IN SYNTAX 124

Proposition III. V moves outside the P if and only if it bears rich Infl-
morphology

As argued by Pollock in a very important paper (Pollock 1989), which in fact
set the stage for cartographic studies, there is a systematic difference as to the
position at which a lexical verb surfaces in French and in English: in French, but
not in English, a lexical verb can move to an Infl-position, preceding the negative
marker pas (5), a frequentative aspectual adverb (6), or a stranded quantifier (7)
argued to be, la Kayne (1975), in an adverbial position, or to a C position, which
gives rise to auxiliary-subject inversion (8).

(5) a. *[
InflP
John [
InflP
likes [
NegP
not [
NegP
likes [
P
likes Mary]]]]
b. [
InflP
Jean [
InflP
(n)aime [
NegP
pas [
NegP
(n)aime [
P
aime Marie]]]]]

(6) a. *[
InflP
John [
InflP
kisses [
AdvFreqP
often [
P
kisses Mary]]]]
b. [
InflP
Jean [
InflP
embrasse [
AdvFreqP
souvent [
P
kisses Marie]]]]
c. [
InflP
John [
InflP
[
AdvFreqP
often [
P
kisses Mary]]]]
d. *[
InflP
Jean [
InflP
[
AdvFreqP
souvent [
P
embrasse Marie]]]]

(7) a. *[
InfP
My friends [
InflP
love [
ADVP
all [love Mary]]]]
b. [
InflP
Mes amies [
InflP
aiment [
ADVP
tous [P aiment Marie]]]]
c. [
InflP
My friends [
InflP
[
ADVP
all [
P
love Mary]]]
d. *[
InflP
Mes amies [
InflP
[
ADVP
tous [
P
aiment Marie]]]]

(8) a. *[
CP
Likes [
InflP
he [
P
likes Mary]]]?
b. [
CP
Aime-t-il [
InfP
[
P
aime-t-il Marie]]]?

(Adapted from Pollock 1989: 367)

In (6.b), Pollock assumed that the optional preclitic negative marker n-
(<ne) had moved pied-piped by V from the head of NegP, where pas is spelled
out as a SPEC, to the head Infl via successive head movement, thereby satisfying
Emonds Head Movement Constraint (Emonds 1978). Another possibility is that
ne and pas occur in different negative projections (see Zanuttini 1997).
It is tempting to connect, as Pollock did, these syntactic minimal pairs, which
reflect that lexical verbs must move outside the P in French and cannot move
outside the P in English, to the different degree of robustness or richness of in-
SYNCRETIC SYNTACTIC PATTERNS 125
flectional paradigms in the two languages. Clearly, French inflectional paradigms
are richer than English ones: whereas French inflectional paradigms display dis-
tinctions of tense (present, past and future), mood (indicative and subjunctive) and
person-number features (1
st
[singular], 2
nd
[singular] and 3
rd
[singular]), Eng-
lish inflectional paradigms do not display any productive subjunctive/indicative
distinction and future is not expressed by inflectional marks but by means of
verbal periphrasis (will kiss, is going to kiss, etc.); apparently, through inflec-
tional means English expresses only a temporal present/past distinction and a 3SG
mark for present.

(9) Present Past
1SG. kiss kiss-ed
2SG. kiss kiss-ed
3SG. kiss-(e)s kiss-ed
1PL. kiss kiss-ed
2PL. kiss kiss-ed
3PL. kiss kiss-ed

Thus, a richly inflected V leaves the P in French and a poorly inflected V
does not leave the P in English.
According to the reasoning above, the richness of inflectional paradigms is
assumed to be determined by the number of overt morpho-phonological dis-
tinctions, an observable property, although the morphological analysis of inflec-
tional forms is rarely transparent, especially in syncretic morphological patterns.
Since languages vary as to the number of overt morpho-phonological distinctions,
languages vary as to the relative richness of their inflectional paradigms. Conse-
quently, one should ideally be able to predict whether in a language L V leaves the
P simply on the basis of the number of overt morpho-phonological distinctions
of the inflectional verbal paradigm in L. Or equivalently, one can test whether
proposition III is accurate simply by correlating the position of V in different
languages and the number of morpho-phonological distinctions in the inflectional
verbal paradigms in each language; if, all things being equal, the verbal inflec-
tional paradigms of a language L
1
display a smaller or equal number of morpho-
phonological distinctions than the verbal inflectional paradigms of a language L
2

and V leaves the P in L
1
but not in L
2
, then proposition III is contradicted. See
Platzack & Holmberg (1989), Roberts (1993) and Vikner (1994) for two attempts
to decide the critical number of substantial distinctions for V to move to Infl.

THE EMERGENCE OF ORDER IN SYNTAX 126
However, there are both empirical and conceptual reasons to suspect that
inflectional richness is not the real trigger of V-movement.
The best-known and most compelling empirical complication is provided by
languages like Swedish, Norwegian and Danish. In Swedish, for instance, mood
distinctions are conveyed exclusively through verbal periphrasis (ma + in-
finitive), as future tense (skulle + infinitive). Inflectional verbal paradigms
only distinguish past from present tense, with no person-number opposition.

(10) Present Past
1SG. kyss-er kyss-te
2SG. kyss-er kyss-te
3SG. kyss-er kyss-te
1PL. kyss-er kyss-te
2PL. kyss-er kyss-te
3PL. kyss-er kyss-te

Therefore, inflectional verbal morphology is slightly poorer in Swedish than
in English: apparently, in both languages V displays a past-present distinction, but
only in English V displays a 3SG marking. Since a finite lexical V leaves the P in
Swedish but not in English, our starting supposition that morphological richness
determines whether V leaves the P should be brought into question.
Proposition III reflects a conceptual oddity that deserves some attention. It
would be quite natural if Infl type features did not require a lexical V to move to T
if inflectional features attached on V did not display a relatively high number of
distinctions: type features would not bother to host an inflected V unless its
inflection was robust enough to satisfactorily instantiate them. However, verbal
inflectional features need to undergo movement to be assigned a suitable position
to be interpreted; therefore, unless type Infl features did not unselfishly host a
poorly inflected V, the derivation would crash, because it would contain tokens
that could not be appropriately interpreted. Thus, for convergence reasons, any V
should move to the C-T discontinuity if it bears attached inflectional features, as
relatively poor or relatively rich as they may be, and a V can remain inside the P
only if it completely lacks attached inflectional features. This is clear for French
and Mainland Scandinavian lexical Vs, which display Infl-like features, but it
seems problematic for the standard analysis of English lexical verbs, according to
which they manifest tense features and a person mark for present 3SG.
The idea that the trigger of V-movement is not the relative richness of
inflectional paradigms or the strength of features (in the sense of Chomsky 1993,
1995) but the mere presence of inflectional features attached on V has already
SYNCRETIC SYNTACTIC PATTERNS 127
been developed in some detail by Sol (1996). In this study, Sol addresses the
problem posed by English lexical verbs, and he concludes that they can be viewed
as non-finite forms, concretely, as participial forms. The core argument goes as
follows (see Sol 1996 for a detailed argumentation). A very solid observation is
that the past form of regular verbs in Modern English has become homophonous
with the participial form (work-ed). This means that, at some historical stage of
English, the past form may have been reanalyzed as a past participle. If this is so,
the lexical verb worked appears in the same position in both John has worked
and John worked. Such a participle is claimed to bear an aspectual feature:
[-progressive]. Moreover, a present morpheme Sol (1996: 233-234) argues
cannot exist if it does not stand in opposition to a past morpheme (just like
singular cannot exist without a plural). It should then also have been reanalyzed as
a participle. However, works and worked do not contrast in aspect, as they
both are [-progressive]. Sol solves this potential problem by arguing that they are
selected by null tense morphemes: when [-Past], it would select arrives, when
[+Past], it would select arrived. In other words, temporal features are not glued
in V but covertly present in the syntactic hierarchy of functional features as null
morphemes. English lexical verbs would not directly convey temporal features,
but aspectual features, and the morpho-phonological contrast between arrives
and arrived derives from the value of the selecting tense; hence, the commonly
called inflectional morphology of present -/-s and the commonly called inflec-
tional morphology of past are suppletive forms of a [-progressive] morpheme.
Finally, the morpheme s is analyzed as just number agreement (see also Kayne
1989), and the observation that number agreement is confined to the 3
rd
person is
related to the participial status of the verbal form.
Sols reasoning implies that English lexical verbs are not spelled out in their
base position, since their aspectual [-progressive] features must be hosted by an
appropriate aspectual type feature. It is known that the position where V is spelled
out must be lower than negation and quantificational adverbs (5-8) but, as argued
above, it must be higher than the A-position where an accusative DP occurs,
which has a VP-adjunct in its domain, as it is able to bind an anaphor and to
license an NPI into such a VP-adjunct. The following constructions, adapted from
Lasnik (2001), are repeated from section 3.2.2.

Condition A satisfaction
(11) a. The DA proved [two men to have been at the scene of the crime]
during each others trials (ECM)
THE EMERGENCE OF ORDER IN SYNTAX 128
b.
?/
*The DA proved [that two men were at the scene of the crime]
during each others trial (finite embedded sentence)
c. The DA accused two men during each others trials

NPI licensing
(12) a. The DA proved [no one to have been at the scene] during any of
the trials
b.
?
*The DA proved [that no one was guilty] during any of the trials
c. The DA cross-examined none of the witnesses during any of the
trials

These surprising patterns were accounted for following Chomskys (2005b)
idea that an accusative DP is attracted by * to SPEC-V, just as a nominative DP
is attracted by C to SPEC-T, and V moves to a higher functional head, *, or per-
haps an aspectual head.
I thus adhere to Sols main idea that there is good reason to keep to the
simplest possibility, namely that V-movement is triggered by the presence of
inflectional features, and not by the relative richness of inflectional paradigms.
The counterexample provided by English verbs may be just apparent, as a finer
morphological analysis of English lexical verbs, along the course developed by
Sol, should make clear. This opens a very interesting and subtle path for cross-
linguistic research: the cross-linguistically variant distribution of V must be deter-
mined by the cross-linguistically variant number and type of of non-inherent
functional features (voice, aspect, tense/mood/-features, force) glued on V.

Proposition IV. The Null-Subject Parameter derives from V-to-T movement,
and hence from the existence of a rich Infl-morphological paradigm in a
particular language

Null Subject Languages (NSLs) are defined by their ability to allow not only
preverbal subjects (13) but also postverbal subjects with no overt preverbal exple-
tive (14) and omitted subjects (15). Accordingly, Catalan is a NSL whereas Eng-
lish is a non-NSL.

Preverbal subjects
(13) a. En Pere va treballar
b. Peter worked

SYNCRETIC SYNTACTIC PATTERNS 129
Postverbal subjects
(14) a. Va treballar en Pere
b. *Worked Peter

Omitted subjects
(15) a. Va treballar
b. *Worked

In the wake of studies by Taraldsen (1980), Chomsky (1981) and Rizzi
(1982b, 1986), it has been generally understood that the Null-Subject Parameter is
an epiphenomenon derived from morphological properties, more particularly,
from the richness of verbal inflectional paradigms. This is indeed a very solid em-
pirical generalization that offers an interesting framework for the study of para-
meter setting and morpho-syntactic diachronic process.
Note that, if both the Null-Subject Parameter and the observation that V leaves
the P were to be derived from verbal inflectional richness, the rate of inflectional
richness trigging V-movement would be different from the rate of inflectional
richness that yields the Null-Subject Parameter, as Modern French, a language
displaying V-to-T/-C movement (16.a) is not parameterized as a NSL (16.b-c).

(16) a. Marie travaille souvent
b. *Travaille souvent Marie
c. *Travaille souvent

A more straightforward account can be attained if, as argued above, the trigger
of V-to-T/C movement is not the inflectional richness, but rather the presence of
Infl-features on V. The T-head in a language like Modern French would have to
unselfishly trigger V-to-T movement in order to warrant convergence at the
external systems (i.e., to warrant that inflectional features attached on V were in-
terpreted at the appropriate position); however, such a token would not be rich
enough to instantiate type -features in a language like Modern French, and thus a
DP must be merged to SPEC-T to avoid the derivation to crash (i.e., to avoid a
type feature to arrive at the C-I system without being satisfactorily instantiated).
Therefore, EPP-effects appear when V is [-fin], as in English, but also when V has
undergone V-to-T movement and is poorly inflected, as in French. They do not
appear when V is richly inflected, in which case a V inflected for Infl-features is
hosted by an Infl type feature in an appropriate position, and this Infl type feature
is satisfactorily instantiated by the richly inflected V.
THE EMERGENCE OF ORDER IN SYNTAX 130
The idea that it is precisely the inflectional richness, and not the presence of
-features on V, that defines a language as a NSL is strongly supported by cross-
linguistic and diachronic considerations. Old French, a language with verbal in-
flectional paradigms richer than Modern French, was a NSL (see Adams 1987 and
Roberts 1993). Similarly, several dialects of Modern Brazilian Portuguese mani-
fest simpler inflectional verbal paradigms than do Traditional Brazilian Portu-
guese or Modern European Portuguese: the 2SG form tu falas and 2PL form vs
falais have been replaced, respectively, by the old address form voc fala and
vocs falam. In addition, as Kato observes (Kato 1999: 4), the first person
plural ns is being replaced by the nominal a gente (the folks = we folks),
which triggers 3
rd
person singular agreement, thus reducing the paradigm to three
distinctive forms.

(17) Present indicative of falar (speak)

Traditional Brazilian Portuguese Modern Brazilian Portuguese
Modern European Portuguese

1SG eu fal-o fal-o fal-o
2SG direct tu fal-as --- ---
2SG indirect voc fal-a fal-a fal-a
3SG ele/-a fal-a fal-a fal-a
1PL ns fal-amos fal-amos (a gente) fal-a
2PL direct vs fal-ais --- ---
2PL indirect vocs fal-am fal-am fal-am
3PL eles/-as fal-am fal-am fal-am
(6/8) (4/8) (3/8)

Whereas Traditional Brazilian Portuguese and Modern European Portuguese
are NSLs (and the inflectional richness of present indicative paradigms is the
same as two other NSLs, Standard Spanish yo habl-o, tu habl-as, usted habl-a, l
habl-a, nosotros habl-amos, vostros habl-ais, ustedes habl-an, ellos hablan (6/8)
and Standard Catalan jo parl-o, tu parl-es, vost parl-a, ell parl-a, vosaltres
parl-eu, nosaltres parl-em, vosts parl-en, ells parl-en (6/8)), the dialects of
Modern Brasilian Portuguese we are concerned with do not omit referential
subjects (21), although they (still) manifest null expletives (18), omission of
arbitrary subjects (19), null bound pronouns (20.a) and anaphoric null subjects
(20.b). See Duarte (1993) and Kato (1999) for further discussion. The following
illustrations are taken from Kato (1999: 5).
SYNCRETIC SYNTACTIC PATTERNS 131

(18) a. T chovendo
It is raining
b. Tem novidade
There is news
c. Parece que vai chover
It seems that it is going to rain

(19) a. Aqui pode fumar
You/one can smoke
b. Aqui conserta sapatos
One repairs shoes

(20) a. Ningum acha que estpido
Nobody
i
thinks that he
i
is stupid

b. O Joo disse que comprou um carro
John
i
said that he
i
has bought a car

(21) a. *(Eu) como pizza
I eat pizza
b. *(Voc) come pizza
You eat pizza
c. *(Ele) come pizza
He eats pizza
d. *(A gente) come pizza
The folks (we folks) eat pizza

As argued by Sigursson (1993), Old Icelandic allowed not only null agreeing
subjects (22) but also null non-agreeing verbal objects (23) and null non-agreeing
prepositional objects (24).

(22) ok kom hann
i
angat, ok var Hoskuldr uti, er
and came he there and was H. outdoors when

i
rei i tn
rode into field
And he came there, and Hoskuldr was outdoors when (he) rode
into the field

THE EMERGENCE OF ORDER IN SYNTAX 132
(23) dvergrinn mlti, at sa baugr
i
skyldi vera hverjum hofusbani,

the dwarf said that that ring should be to-anybody a headbane
er atti
i

that possessed
The dwarf said that that ring should bring death to anybody
possessed (it)

(24) tla ek, at ntir eigi boga minn
i
ottu
believe I that you (can-)use not bow my even-if-you
spyrnir ftum i
i

push with-feet in
I believe that you cannot use my bow even if you push with your
feet in (it)
(Sigursson 1993: 248)

Thus, the grammatical strategies that allow null subjects in languages like Old
Icelandic and Catalan must be different: in the former, null subjects are part of
more general coindexing algorithms that do not depend on verbal inflection and
hence extend to non-agreeing objects, whereas in the latter, the existence of null
subjects derives from inflectional richness, which is not involved in licensing null
objects (25). In simpler terms, Old Icelandic was not a Null-Subject Language but
a Null Argument Language, and Catalan is a Null-Subject Language (NSL) but
not a Null Argument Language (NAL).

(25) []
subject
va donar el llibre a tots els estudiants que *(el) van
demanar
He/she gave the book to all the students that asked for it

According to Sigurssons analysis, null subjects and objects in Old Icelandic
were identified under free coindexing with an NP in preceding discourse (much
like Old Icelandic null-topics, which could, however, also be identified under free
coindexing with a construed discourse topic). The claim that it is not rich verbal
inflection that identifies type -features but coindexing with an NP present in
preceding discourse is strongly supported by the observation that in Old Icelandic
prose null arguments never initiate discourse, since a preceding linguistic context
is required. This is a clear difference with respect to null subjects in NLSs: the
subject is omitted in (26.b) although it cannot be co-indexed with any NP present
in preceding discourse.

SYNCRETIC SYNTACTIC PATTERNS 133
(26) a. Qu fa en Joan?
what do.3SG ART Joan?
What is Joan doing?

b. (
#
jo) no ho s
I no it-ACC know.1SG
I dont know it

Interestingly, Modern Icelandic has lost both null referential subjects and null
referential objects without any concomitant or previous reduction of rich verbal
inflectional, as illustrated below with the present indicative of leita search,
segja say, and sj see, borrowed from Sigursson (1993: 249). Non-referential
subjects can still be null in Modern Icelandic, as in the dialects of Modern
Brazilian Portuguese considered above, which suggests that abstract -features do
not need to be instantiated by rich inflectional morphology.

(27) Old Icelandic Modern Icelandic
1SG. leita segi s leita segi s
2SG. leitar segir sr leitar segir sr
3SG. leitar segir sr leitar segir sr
1PL. leitum segjum sjum leitum segjum sjum
2PL. leitit segit sjit leiti segi sji
3PL. leita segja sj leita segja sj

Sigurssons results, which are very important for the understanding of
discourse indexing mechanisms, lead to the conclusion that, even if in Old
Icelandic the inflected V obligatorily moves to T, it does not instantiate type
features on T, which are instantiated by co-indexation to an NP in preceding
discourse. Note that Sigursson does not reject the idea that rich verbal inflection
is the decisive factor in defining a language as a NSL, but he convincingly shows
that verbal inflection plays no role in defining a language as a Null Argument
Language, in which free discourse indexing (and not verbal inflection) plays a
crucial role in instantiating Infl-features. With this idea in mind, it is not surpris-
ing that Chinese allows subjects to be omitted: although it lacks verbal inflection
whatsoever, it also allows objects to be omitted (Huang 1989).
From this revision I conclude that the four propositions under discussion must
be replaced or reformulated in the following terms:

THE EMERGENCE OF ORDER IN SYNTAX 134
I. Both movement of maximal and minimal categories take place during
syntactic computations; thus both V-movement to T/C and DP-movement
to T take place during syntactic computations.

II. Inflected verbs contain not only inherent features (thematic-features and
Aktionsart features), but also non-inherent functional features (tense, -
features, mood, aspect, etc.). Functional features attached on V convert V
into a multicategorial word, which means neither that they are uninter-
pretable on V nor that they must be deleted.

III. The position of V is determined by the non-inherent functional features
attached on V; particularly, V leaves the P moving up to C or to T if it is
finite.

IV. A language L is a NSL only if verbal inflection is rich in L; L is a NAL if
and only if L displays free discourse indexing mechanisms, independently
of verbal inflection.

5.3 A side-effect of V-to-T movement
In this section I shall introduce the first possible case of syntactic syncretism,
and the least controversial one, for the analysis I shall pursue is somehow present
in the literature on NSLs: since richly inflected Vs instantiate -features on T,
SPEC-T is available to match other features.
I shall begin by providing in 5.3.1 some evidence for the idea that SPEC-T is
not a position where -features are matched in NSL (see Sol 1992, among
others, for a broader discussion): SPEC-T is both too restrictive (only a subset of
subjects can occur there) and too permissive (non-subject Clitic Left Dislocated
(CLLD) elements do also occur there). On this basis, I shall argue that a-features,
namely, topic and referential features, are matched in SPEC-T, but not -features.
In 5.3.2 I shall discuss the source of a-features matched in SPEC-T; I shall
first focus on a very recent proposal by Gallego (Gallego 2006): in NSLs, the
phase head moves along with V to T, a syntactic operation that has the effect of
transforming T into a phase head (5.3.2.1); thus a-features on T in NSLs would be
provided by . After showing some technical and empirical problems for
Gallegos phase sliding, I shall propose that a-features surfacing on T are inher-
ent to the C-T discontinuity, not inherited from the lower -V discontinuity. Ac-
cordingly, [-V]-to-T movement does not provide T with a-features, but creates a
syntactic circumstance that favors structural minimization to the detriment of the
SYNCRETIC SYNTACTIC PATTERNS 135
C-I legibility requirement of creating semantically devoted projections: as an
inflected V is responsible for matching -features, the MMEP favors SPEC-T to
be used to match a-features, and not to redundantly match -features. If this is so,
in a single projection, both -features and a-features are matched, thereby mini-
mizing the span of structure used for syntactic com-putations.
In 5.3.3 I shall close the discussion of NSLs by noting the problematic status
of null expletives (Rossell 2000, among others) and the inadequacy of postu-
lating covert movement of postverbal subjects to SPEC-T.

5.3.1 SPEC-T becomes an A-position when V instantiates -features on Infl.
Consider the following contrasts between English and Catalan, which illustrate a
very well known generalization: any kind of subject occurs preverbally in a non-
NSL like English, whereas, in a NSL like Catalan, only a subset of subjects does.

5.3.1.1 Definite subjects in NSLs. The constructions in (28) illustrate that a
definite subject that bears new information cannot occur preverbally in Catalan
(28.a), unless it bears focal stress and undergoes focus fronting, in which case it
arguably occurs at Rizzis FocP of the C-area (28.b); but subjects that bear new
information can occur postverbally (28.c), and the verb, which conveys old
information present in discourse, is optionally omitted (32.d).

Definite subjects bringing new information
Qui va venir?
who PAST come.INF
Who came?

(28) a.
#
En Pere va venir
ART Pere PAST come.INF
Pere came
b. EN PERE va venir
c. Va venir en Pere
d. En Pere

The constructions in (29) illustrate that, if a definite subject expresses old
information present in discourse, it can occur preverbally (29.a), and optionally,
an intonational pause can appear between the topical preverbal subject and the
verb (29.b), in which case the preverbal subject may be left-dislocated to a TopP
THE EMERGENCE OF ORDER IN SYNTAX 136
of the C-area. Definite subjects that bring old information cannot occur post-
verbally (29.c) and can optionally be omitted (29.d).

Definite subjects bringing old information
Qu fa en Pere?
what does ART Pere
What is Pere doing?

(29) a. En Pere llegeix
ART Pere reads
Pere is reading
b. En Pere(,) llegeiex
c.
#
Llegeix en Pere
d. Llegeix

Crucially, English referential preverbal subjects are not informationally
restricted: referential subjects can (indeed, must) appear preverbally, indepen-
dently of whether they bring new (30) or old (31) information.

Definite subjects bringing new information
Who came?

(30) a. Peter came
b. *Came Peter

Definite subjects bringing old information
Where did Peter study?

(31) a. Peter studied in Barcelona
b. *Studied Peter in Barcelona

5.3.1.2 Non-definite subjects in NSLs. Non-definite preverbal subjects must satisfy
further constraints; as argued by Sol (1992), bare indefinite subject NPs cannot
occur preverbally in Catalan and Italian, while they must occur preverbally in
English (32). Negative subject quantifiers and counting subject quantifiers too
cannot occur preverbally in NSLs (34-35), but must occur preverbally in non-
NSLs (33). The distribution of these three kinds of non-definite quantifiers in
NSLs and non-NSLs is thus complementary. All these contrasts are adapted from
SYNCRETIC SYNTACTIC PATTERNS 137
Sol (1992, chapter 5), where the very intricate restrictions that govern preverbal
subjects are delineated in great detail. My sole intention here is to express in a
clear way Sols general results, which will be useful for research on syntactic
syncretism.

(32) a. Students have arrived (English)
a. *Have arrived students

b. *Estudiants han arrivat (Catalan)
b. Han arrivat estudiants

c. *Studenti sono arrivati (Italian)
c. Sono arrivati studenti

(33) a. Nothing has happened
a. *Has happened nothing
b. Few students have come
b. *Have come few students

(34) a. *Res ha passat
a. No ha passat res
b. *Pocs estudiants han vingut
b. Han vingut pocs estudinats

(35) a. *Niente successo
a. Non successo niente
b. *Pocchi studenti sono arrivati
b. Sono arrivati pocchi studenti

Indefinite subjects can appear preverbally or postverbally in Catalan; if they
appear preverbally they strongly tend to be interpreted as referential expressions,
whereas if they appear postverbally the pure existential reading is much more
prominent (37). In English, indefinite subjects obviously must occur preverbally,
and admit both interpretations (36).





THE EMERGENCE OF ORDER IN SYNTAX 138
Indefinite subjects in English: a stone
(36) a. A stone has fallen down the hill
One of the stones has fallen down the hill (referential reading)
An indefinite stone has fallen down (pure existential reading)

Indefinite subjects in Catalan: un roc (a stone)
(37) a. Un roc ha caigut de la muntanya (referential interpretation of un
roc, a stone)
One of the stones has fallen down the hill (referential reading)
An indefinite stone has fallen down the hill (pure existential
reading)

a. Ha caigut un roc de la muntanya
An indefinite stone has fallen down the hill (pure existential
reading)
One of the stones has fallen down the hill (referential reading)

Therefore, in NSLs, definite subjects can occur preverbally only if they are
topical; bare NPs, negative quantifiers and counting quantifiers cannot occur
preverbally at all, and indefinite quantifiers can naturally occur preverbally only if
they are referentially interpreted. If the preverbal position is so restricted for
subjects in NSLs and so unrestricted for subjects in non-NSLs, it is reasonable to
conclude that preverbal subjects in non-NSLs match only -features; as to pre-
verbal subjects in NSLs, there are two possibilities: they match (a) both -features
and topicality/referentiality features or (b) only topicality/referentiality features,
which are indeed those that constrain the set of legitimate preverbal subjects in
NSLs. As will be illustrated below following Sols arguments, there is reason to
think that option (b) is the correct one.

5.3.1.3 Clitic-Left Dislocated elements in NSLs. Non-definite non-subject
quantifiers can be resumed by a clitic only if they can be interpreted referentially;
thus, bare NPs, negative quantifiers such as nothing and counting quantifiers
such as few cannot be CLLD elements (38.a-c), whereas universal quantifiers
and existential quantifiers can be (38.d-f), in which case they are unambiguously
referential. English translations do not reflect resumption, for obvious reasons.


SYNCRETIC SYNTACTIC PATTERNS 139
(38) a. *Estudiants
i
els
i
he vist (Catalan)
students them.have.1SG seen
I have seen the students

b. *Res ho he vist
nothing it.have.1SG seen
He/she has done nothing

c. *Pocs estudiants
i
els
i
he convidat
few students them.have.1SG invited
He/she has invited few students

d. A tots aquests alumnes
i
els
i
veur
to all these pupils them.will.see.1SG
I will see all the pupils

e. A un meu amic
i
el
i
van arrestar
to one my friend him.arrested.3PL
They arrested one of my friends; or rather, one of my friends
was arrested

f. A dos amics meus
i
els
i
han arrestat
to two friends mine them.have.3PL arrested;
They have arrested two friends of mine; or rather two friends of
mine have been arrested

Definite DPs can be CLLD elements only if they are topical (39a vs. 39.b),
in which case they involve a pause only optionally (39.a).

(39) a. Has vists els estudiants?
Have.2SG seen.PL the students
Have you seen the students?

a. Els estudiants
i
(,) no els
i
he vist
the students not them.have.3SG seen

b. A qui has vist?
to whom have.2SG seen?
Who have you seen?
THE EMERGENCE OF ORDER IN SYNTAX 140
b.
#
Els estudiants
i
els
i
he vist
the students them.have.1SG seen
I havent seen the students

Therefore, the restrictions imposed on preverbal subjects and resumed
elements are the same (if they are definite, they must be topical (28, 39), if they
are non-definite, they must be referential (37, 38)). A further question is whether
it can be decided whether the position where both preverbal subjects and CLLD
elements occur is the same, as would be expected: if they satisfy the same seman-
tic requirements, they should instantiate the same type features.

5.3.1.4 Both preverbal subjects and CLLD elements match a-features in SPEC-T
in NSLs. It seems that the more a quantifier is likely to be interpreted as D-linked
or referential, the more it is acceptable as dislocated (Sol 1992). Firstly, a
[-human] negative quantifier cannot appear preverbally and with a resumptive
clitic, as illustrated in (40.b), whereas a [+human] negative quantifier (ning
nobody) or a universal quantifier (tothom everybody) can appear preverbally
and with a resumptive clitic (in the case they are non-subjects). Although ning
and tothom are referential enough to be resumed (contrary to res nothing), they
cannot be dislocated; but other quantifiers that have a more prominent referential
reading, such as dos amics two friends and alguns amics some friends can be
resumed by a clitic (if they are non-subjects) and moreover dislocated. Examples
are again from Catalan.

Dislocation
(40) a. A ning
i
(*
/#
,) l
i
han acceptat (resumed non-subject)
to nobody him.have.3PL accepted

a. Ning(*
/#
,) ha aprovat lexamen (subject)
nobody has passed the exam

b. A tothom
i
(*
/#
,) l
i
han acceptat (resumed non-subject)
to everybody him.have.3PL accepted

b. Tothom(*
/#
,) ha aprovat lexamen (subject)
everybody has passed the exam

c. A dos amics meus
i
(,) els
i
han arrestat (res. non-subject)
to two friends of mine them.have.3PL arrested
SYNCRETIC SYNTACTIC PATTERNS 141
c. Dos amics meus(,) han tingut un accident (subject)
two friends of mine have.3PL had an accident

d. Alguns amics
i
(,) els
i
han arrestat (res. non-subject)
some friends them.have.3PL arrested

d. Alguns amics(,) no hi vindran (subjects)
some friends not there.come.3PL

Crucially, those quantifiers that cannot be dislocated must occur in a position
adjacent to the inflected verb both if they are subjects and non-subjects, which
strongly argues that both subjects and resumed non-subjects appear in the same
position, SPEC-T. Those quantifiers that can be dislocated, because they are more
prominently referential, have allegedly moved further, to SPEC-Top of the C-
area, independently of whether they are subjects or non-subjects.

Adjacency effects
(41)
a. A ning
i
(*aquesta vegada) l
i
han acceptat (resumed non-subject)
to nobody (this time) him.have.3PL accepted

a. Ning (*
/?
aquesta vegada) ha aprovat lexamen (subject)
nobody (this time) has passed the exam

b. A tothom
i
(*
/?
aquesta vegada) l
i
han acceptat (res. non-subj.)
to everybody (this time) him.have.3PL accepted

b. Tothom (*
/?
aquesta vegada) ha aprovat lexamen (subject)
everybody (this time) has passed the exam

c. A dos amics meus
i
(avui) els
i
han arrestat (res. non-subj.)
to two friends of mine (today) them.have.3PL arrested

c. Dos amics meus (avui) han tingut un accident (subject)
two friends of mine (today) have.3PL had an accident

d. Alguns amics
i
(avui) els
i
han arrestat (res. non-subject)
some friends (today) them.have.3PL arrested

THE EMERGENCE OF ORDER IN SYNTAX 142
d. Alguns amics (al casament) no hi vindran (subjects)
some friends (to the wedding) not there.come.3PL

It is thus fair to conclude that preverbal subjects in Catalan, a NSL, do not
match -features but only a-features such as topicality and referentiality; since
(non-subject) resumed elements instantiate the same a-features and, as preverbal
subjects, appear adjacent to V unless they can be dislocated to a TopP of the C-
area, both CLLD elements and preverbal subjects occur in SPEC-T, a position
used for matching topic and referential features, instead of -features, unlike in
English. The ability of both preverbal subjects and resumed elements to be dis-
located to SPEC-Top depends on their relative degree of referentiality.

5.3.2 The source of a-features associated with SPEC-T when V instantiates -
features on T

5.3.2.1 Subextraction (III): phase sliding. Gallego (2006) considers the possibility
that, although C and are universal inherent phase heads, the fact that moves
along with V to T (instantiating Infl-features) in NSLs endows the TP with the
phase properties characteristics of and frees the P from assuming its phase
responsibilities. A crucial point for Gallegos analysis is that of subject islands in
NSLs: if preverbal subjects are in SPEC-T and postverbal subjects are in SPEC-,
and subextraction from preverbal agentive subjects is more problematic than
subextraction from postverbal agentive subjects, then there is good reason to think
that, in NSLs, TP behaves, derivatively, as a phase, while *P has lost its phase-
hood; in Gallegos terms, a phase sliding would have taken place. This obvious-
ly raises the possibility, as Gallego argues, that the A-status of SPEC-T in NSLs
(its topicality/referentiality features) is introduced by the head by means of [V-
]-to-T movement.
The cases of subextraction that Gallego considers can be classified in two
types. Though he does not mention this distinction, it is rather instrumental in
understanding subextraction patterns in NSLs:

(a) The subextracted PP and the DP from where the PP has been
subextracted are spelled out in the same clause: [
CP1
[
PP
]
i
C
1
[
DP

t
i
]]

(b) The subextracted PP and the DP from where the PP has been
subextracted are spelled out in different clauses, the former in a
SYNCRETIC SYNTACTIC PATTERNS 143
superordinate one and the latter in a subordinated one: [
CP1
[
PP
]
i
C
1

[
CP2
C
2
[
DP
t
i
]]]

In (a), neither subextraction from what Gallego considers an agentive
preverbal subject nor from what he considers a postverbal agentive subject leads
to an entirely acceptable output (42.a/b). Note that subextraction from a
postverbal non-agentive subject is unproblematic (42.c), as expected. These
Spanish examples are borrowed from Gallego (2006: 84); example (42.c) is from
Uriagereka (2004: 9). Direct questions in written Spanish are always preceded by
a question mark (Qu dices? what are you saying?), which is not to be
confused with the superindex
?
, which indicates acceptability judgments.

(42) a.
??
De qu artistas han herido tu sensibilidad las obras?
of what artists have.3PL hurt your sensitivity the works
Which artists have the works by hurt your sensitivity?
[
CP
De qu artistas
i
han herido
z
[
TP
t
z
[
v*P
tu sensibilidad
j
[
v*P
[las
obras t
i
] t
z
t
j
]]]]

b. *De qu artistas han herido las obras tu sensibilidad?
of what artists have.3PL hurt the works your sensitivity
Which artists have the works by hurt your sensitivity?
[
CP
De qu artistas
i
han herido
z
[
TP
[las obras t
i
]
j
t
z
[
v*P
t
j
t
z
tu
sensibilidad]]]

c. De que artistas han llegado ya las obras?
of what artists have.3PL arrived already the works
Which artists have the works of arrived?
[
CP
De qu artistas
i
han llegado
z
[
TP
ya t
z
[
vP
t
z
[las obras t
i
]]]

Thus, both subextraction from a preverbal and a postverbal agentive subject
seems to yield degraded outputs, which means that the meager contrasts provided
by type (a) does not provide a solid argument for phase sliding.
Observe in the above representations that Gallego assumes that V has moved
from T to C in questions across the adverb ya (already); thus the position of
the subject is determined with respect to the object (tu sensibilidad): when it
precedes the object, it is considered to appear in SPEC-T, following V in C, and
when it follows the object, it is considered to be in its base-position, SPEC-*,
with the object shifted. However, it is unclear that V moves from T to C in (42):
THE EMERGENCE OF ORDER IN SYNTAX 144
the position of V relative to aspectual adverbs such as ya does not indicate
whether V is in T or in C, for V can follow a subject and precede such an adverb
also in an assertive clause in Spanish (43.b). In either of these two constructions,
no pause intonation is needed. Note also that the verb can follow ya in questions
(43.c).

(43) a. Las obras ya han llegado
the works already have.3PL arrived
The works have already arrived

b. Las obras han llegado ya
the works have.3PL arrived already
The works have already arrived

c. De qu artistas ya han llegado las obras?
of what artists already have.3PL arrived the works
Which artists have the works of arrived?

Whether V ever raises to C in Catalan and Spanish is a topic of debate; see
Suer (1994) and Ordez (1998) on the claim that V-to-C never occurs in
Spanish and Catalan, and Gallego (2004) on the possibility that it does. However
the case may be, the adverb ya is not a good positional reference.
One might also be skeptical about Gallegos test to decide whether a subject is
in SPEC-T or in SPEC-* in questions: if it follows V and precedes the object it is
in SPEC-T, and if it follows the shifted object it is in SPEC-*; again, in an
assertive clause, the subject can follow V and at the same time precede the object
(44.a). Since claiming that V raises to C in Spanish assertive clauses would be
entirely unmotivated, the subject las obras de Pedro may be in a position lower
than T even if it precedes the object tu sensibilidad (44.a).

(44) a. Han herido las obras de Pedro tu sensibilidad
have hurt the works of Pedro your sensibility
The works by Pedro have hurt your sensibility

b. Han herido tu sensibilidad las obras de Pedro
have hurt your sensibility the works of Pedro
The works by Pedro have hurt your sensibility

SYNCRETIC SYNTACTIC PATTERNS 145
It would be more reasonable to conclude that the subject is in SPEC-* in
both (44.a) and (44.b); if this is the case, what would be variable is the position of
the object: it remains in its base-position in (44.a) and it is shifted to a higher
position in (44.b).

(44) a. [
TP
[Han herido]
i
-T [
*P
[las obras de Pedro] t
i
[
VP
tu sensibilidad]]]
b. [
TP
[Han herido]
i
-T [
*P
[tu sensibilidad]
j
[
*P
[las obras de Pedro]
t
i
[
VP


t
j
]]]]

Consider the following contrast, also from Gallego (2006: 86), which seems to
be stronger than the one provided in (42). The analysis in brackets of these
constructions is mine. For abovementioned reasons, I am not following Gallegos
positional analysis.

(45) a.
??
De qu universidad causaron los estudiantes problemas?
of what university caused.3PL.PAST the students problems
Of what university did the students cause problems?
[
CP
[De qu universidad]
j

[
TP

causaron
i
[
P*
[los estudiantes [t
j
]]

[
VP
t
i
problemas]]]]

b. De qu universidad causaron problemas los estudiantes?
of what university cause.3PL.PAST problems the students
Of what university did the students cause problems?
[
CP
[De qu universidad]
j
[
TP
causaron [
v*P
problemas
i
[
v*P
[los
estudiantes [t
j
]] [
VP
t
i
]]]]]

If the PP de qu universidad is replaced by the PP de qu coche (of which
car), and the DP los estudiantes by the DP el conductor (the driver), then
there is no strong acceptability contrast: both (46.a) and (46.a) are degraded
outputs, as in (42.a/b), repeated in (46.b/b). A further factor to consider is the
presence of an article: if problemas is replaced by un problema, subextraction
yields an even less acceptable output (46.a).

(46) a.
??
De qu coche caus problemas el conductor?
of what car cause.3SG.PAST problems the driver
Of which car did the driver cause problems?


THE EMERGENCE OF ORDER IN SYNTAX 146
a. *De qu coche caus el conductor problemas?
of what car cause.3SG.PAST the driver problems
Of which car did the driver cause problems?

a.
???
De qu coche caus un problema el conductor?
of what car cause.3SG.PAST a problem the driver
Of which car did the driver cause a problem?

b.
??
De qu artistas han herido tu sensibilidad las obras?
of what artists have.3PL hurt your sensitivity the works
Which artists have the works by hurt your sensitivity?

b. *De qu artistas han herido las obras tu sensibilidad?
of what artists have.3PL hurt the works your sensitivity
Which artists have the works by hurt your sensitivity?

There are indeed several intricate issues involved in subextraction. For
instance, Chomsky (2005b, note 38) observes that examples with the expression
the picture-PP introduce extraneous issues because of the ambiguity of the
phrase, which can be understood with the PP interpreted not as a complement of
picture but rather as a reduced relative clause (roughly, I have a picture which
is of Boston contrary to *I saw a driver who is of the car, *I saw an author who
is of the book). The same caution that led Chomsky to avoid constructions such
as Mary, of whom a picture hit me on the head should lead us to distrust
expressions such as the students of which university, which can be interpreted as
I saw the students which are of University of Barcelona, contrary to expressions
such as the driver of that car (*the driver who is of that car).
Thus, the empirical generalization is that subextraction of a PP from a DP
located in SPEC-* yields a degraded output when the PP is generated as a
complement of the DP ([
DP
[
PP
]]), as in (46); when the DP and the PP can be
interpreted as forming a reduced relative clause [DP PP] located in SPEC-,
extraction of PP is unproblematic (45.b). This is not at all surprising, as the PP is
more embedded in the nested structure [
DP
[
PP
]] than in the symmetric structure
[DP PP], and indeed, it could be accounted for by the Relativized Opacity
Principle, reproduced below.




SYNCRETIC SYNTACTIC PATTERNS 147
Relativized Opacity Principle (ROP)
In a syntactic object [
o1P
[o
1
[
o2P
[o
2
] ] ] ],
where: (i) o
1
and o
2
are two probes of the same type o each projecting an
aP
(ii) is SPEC-o
2
and is o
2
-COMPL, and
(iii) : is a constituent of and y: y is a constituent of ,

o
1
can probe or if they provide a suitable token for o
1
, but it
cannot readily probe y or

When the nested structure [
DP
[
PP
]] is agentive, it is base-generated in the
SPEC position of a a-projection, the *P. Thus, the immediately higher C, a a-
probe, can probe , i.e., the whole [
DP
[
PP
]], but not a constituent of , i.e., the
PP. If reduced clauses are symmetric structures, with no projecting head, no
complement and no specifier, then there is no projection , and C can detect the
whole [DP PP] but also the PP, which is not too deeply nested.
It is not entirely clear to me why extraction of PP from [DP PP] should be less
degraded when the object moves to an extra-SPEC-* preceding the agentive DP
located in SPEC-* (45.b) than when it follows the agentive DP (45.a). As a
matter of fact, a similar contrast is attested in (45), where the wh-phrase is not
extracted from a subject.

(45) a.
?
Cundo causaron los estudiantes problemas?
when cause.PAST the students problems

b. Cundo causaron problemas los estudiantes?
when cause.PAST problems the students
When did the students cause problems?

If so, the contrast in (45) may not be related to subextraction, but rather to
ordering restrictions between objects and subjects in Spanish questions. This
reasoning may apply to (42) as well: both (42.a) and (42.b) are degraded, but the
least degraded one is that in which the object moves over the subject.
In sum, on the one hand, the contrast illustrated in (42.a/b) is too weak to be
considered as proof of phase sliding, and on the other hand, it is doubtful that
such a slight contrast indicates that subextraction from SPEC-* is less problem-
atic than subextraction from SPEC-T in NSLs, since there is no reason to think
THE EMERGENCE OF ORDER IN SYNTAX 148
that the subject is spelled out in SPEC-T in the slightly more degraded (42.a): the
subject can precede the object and follow the verb not only in questions (46.a) but
also in assertions (42.a), which indicates that neither in (42.a) nor in (42.a) is V in
C nor the subject in SPEC-T. A PP can be more easily extracted from a reduced
relative clause [DP PP] located in SPEC-* (45) than from a nested structure [
DP

[
PP
]], because in the former the PP is less embedded than in the latter.
Examples of type (b) given by Gallego seem to provide, initially, a stronger
contrast between subextraction from a preverbal agentive subject and a postverbal
agentive subject. Example (47) is from Uriagereka (2004:10).

(47) a.
??/
*De qu conferenciantes te parece que las propuestas
Of what speakers you.seem.3SG that the proposals
me van a impresionar?
me.go to impress
Which speakers does it seem to you that the proposals by will
impress me?
[De qu conferenciantes
i
te parece [
CP
que [
TP
[las propuestas t
i
]
j
me
van a impresionar [
v*P
t
j
]]]]

b. De qu conferenciantes te parece que me van a
Of what speakers you.seem.3SG that me.go to
impresionar las propuestas
impress the proposals
Which speakers does it seem to you that the proposals by will
impress me?
[De qu conferenciantes
i
te parece [
CP
que [
TP
me van a impresionar
[
v*P
[las propuestas t
i
]]]]]

(48) a.
??
De qu escritor dice Maria que muchas novelas
of what writer say.3SG Mara that many novels
la han impresionado?
her.have.3PL impressed
Of what writer does Mara say that many novels have impressed
her?
[De qu escritor
i
dice Mara [
CP
que [
TP
[muchas novelas t
i
]
j
la han
impresionado [
v*P
t
j
]]]]

SYNCRETIC SYNTACTIC PATTERNS 149
b. De qu escritor dice Mara que la han impresionado
of what writer say.3SG Mara that her.have.3PL impressed
muchas novelas?
many novels
Of what writer does Mara say that many novels have impressed
her?
[De qu escritor
i
dice Mara [
CP
que [
TP
la han impresionado [
v*P
[muchas novelas t
i
]]]]]

(49) a.
??
De qu universidad dice la prensa que muchos estudiantes
of what university says the press that many students
han protestado?
have protested
Of what university do the media say that many students have
protested?
[De qu universidad
i
dice la prensa [
CP
que [
TP
[muchos estudiantes
t
i
]
j
han protestado [
v*P
t
j
]]]]

b. D que universidad dice la prensa que han protestado
of what university say.3SG the press that have.3PL protested
muchos estudinates?
many students
Of what university do the media say that many students have
protested?
[De qu universidad
i
dice la prensa [
CP
que [
TP
han protestado [
v*P
[muchos estudiantes t
i
]]]]]

There are at least two misleading factors that may easily go unnoticed in
examples (47-49).
Firstly, the examples of subextraction from agentive subject involve psych-
verbs like impress (47, 48) and unergative verbs like protest (49); as Gallego
warns (note 113), I will avoid introducing verbs with explicit direct objects as
much as I can, since those may add additional problems. Clearly, once verbs
with explicit objects are used, subextraction from a postverbal agentive subject is
degraded, and thus yields the same result as in English.

(50) a.
??
De qu coche caus problemas el conductor?
of what car caused problems the driver
THE EMERGENCE OF ORDER IN SYNTAX 150
b. *De qu coche caus el conductor problemas?
of what car caused the driver problems
Of which car did the driver cause problems?

Gallego does not specify what kind of problems may be added by explicit
objects, but they may be not additional. A conceivable additional problem could
be that an explicit object yielded minimality effects; but, significantly, what we
observe are anti-minimality effects: in (45.a), the explicit object precedes the
subject, in (45.b), the object follows the subject, and nonetheless, subextraction
yields a degraded output in (45.b) and not in (45.a).
It could be simply that different -grids correspond to different P-structures,
and that the acceptability of subextraction from an argument depended on the
exact position where such an argument is base-generated. This is envisioned by
Chomsky, as Gallego acknowledges (2006: note 113):

Choice of * may have an effect. Perhaps, of which books did the author receive the
prize is more acceptable than of which car did the driver cause a scandal. If so,
difference among theta roles might be relevant, perhaps requiring a deeper analysis of
base structures (Chomsky 2005b: note 39)

In other words, the contrast of acceptability between (51.a/a) and (51.b) may
not reflect that objects add additional problems, but that transitive predicates and
psych-predicates have different argument structures.

(51) a.
??
De qu coche caus problemas el conductor?
of what car caused problems the driver

a.
??
De qu coche caus el conductor problemas?
of what car caused the driver problems
Of which car did the driver cause problems?

b. De qu conferenciantes me van a impresionar las propuestas?
Of what speakers me.GO.PL to impress the proposals
Of which speakers are the proposals going to impress me?

In (51.a/a) the base position of the agentive DP el conductor de qu coche
(the driver of which car) is outside the search domain of * (it is an external
argument of a transitive verb), whereas the DP las propuestas de qu conferen-
ciantes, which is not an external argument or an agent of the psych-verb, is base-
SYNCRETIC SYNTACTIC PATTERNS 151
generated in the search domain of . Accordingly, the P structure of the pre-
dicates in (51) would be like these.

(51) a. [
*P
[el conductor de qu coche] [
*P
*
[+, +ag]
[
VP
V problemas]]]
(cause)
b. [
P

[-, -ag]
[
VP
[las propuestas de qu conferenciantes] [
VP
V
GOAL]]] (impress)

And secondly, in (49), if the PP de qu universidad is replaced by the PP de
qu coche and the DP los estudiantes by the DP el conductor, which cannot be
interpreted as constructing a small clause [DP PP], then subextraction of PP is
degraded both when the subject follows the object and when it precedes the
object, as noted above with respect to (45-46), repeated in (52.b/b). Accordingly,
subextraction from the subject of unergative verbs and subextraction from the
subject of a transitive verb, both external arguments, yield the same result.

(52) a. *De qu coche

dice la prensa

que el conductor ha protestado?
Of which car does the media say that the driver has protested?

a.
??
De qu coche dice la prensa que ha protestado el conductor?
Of which cars does the media say that many drivers have
protested?

b.
??
De qu coche caus problemas el conductor?
Of which car did the driver cause problems?

b. *De qu coche caus el conductor problemas?
Of which car did the driver cause problems?

If the tests are defined by avoiding misleading factors such as psych-verbs
(which lack an agent and hence a thematic SPEC-*) or small relative clauses [DP
PP] appearing in SPEC-*), we can conclude that in Catalan and Spanish, two
NSLs, C cannot subextract a PP from SPEC-*.
Firstly, both subextraction from the external argument of a transitive verb (53)
and an unergative verb (54) yield degraded outputs; (53) illustrates that sub-
extraction from the agentive subject is problematic either when the subject raises
to an extra SPEC-* preceding the object (53.a) as well as when it remains in situ
(53.b). Secondly, subextraction from non-agentive subjects is unproblematic (55,
THE EMERGENCE OF ORDER IN SYNTAX 152
56, 57), as is subextraction from verbal objects (58): the head probes the PP
when it is in V-COMPL and relocates it in SPEC-, a position in which C can
unproblematically access the PP. The examples of subextraction from non-
agentive subjects involve passive subjects (55), subjects of impersonal construc-
tions (66), and unaccusative subjects (57), all base-generated in V-COMPL. This
paradigm can be extended to expressions such as the architect of which build-
ing, the writer of which book, etc.

(53) Subextraction from an agentive subject of a transitive verb
Catalan
a. *[De quina pellcula]
i
va provocar [el director t
i
] un escndol?
of which film PAST.SG cause the director a scandal

a. *[De quina pellcula]
i
va provocar un escndol [el director t
i
]?
of which film PAST.SG cause a scandal the director
Of which film did the director cause a scandal?

Spanish
b. *[De qu pelcula]
i
provoc [el director t
i
] un escndalo?
b. *[De qu pelcula]
i
provoc un escndalo [el director t
i
]?

Relativization test
c. *The director who is of that film

(54) Subextraction from an agentive subject of an unergative verb
Catalan
a. *[De quina pellcula]
i
va treballar [el director t
i
]?
of which film PAST.SG work the director
Of which film did the director work?

Spanish
b. *[De qu pelcula]
i
trabaj [el director t
i
]?

(55) Subextraction from a non-agentive subject of a passive verb


SYNCRETIC SYNTACTIC PATTERNS 153
Catalan
a. [De quina pellcula]
i
va ser aplaudit [el director t
i
]?
of which film PAST.SG be applauded the director
Of which film was the director applauded?

Spanish
b. [De qu pelcula]
i
fue aplaudido [el director t
i
]?

(56) Subextraction from a non-agentive subject of impersonal se-
constructions

Catalan
a. [De quines pellcules]
i
shan aplaudit [els directors t
i
]
of which films SE-have.3PL applauded the directors
Of which films were the directors applauded?

Spanish
b. [De qu pelculas]
i
se aplaudiron [los directores t
i
]?

(57) Subextraction from an unaccusative subject
Catalan
a. [De quina pellcula]
i
ha arribat [el director t
i
]?
of which film have.3PL arrived the director
Of which film has the writer arrived?

Spanish
b. De qu pelcula ha llegado el director?

(58) Subextraction from a verbal object
Catalan
a. [De quina pellcula]
i
va aplaudir [el director t
i
] en Pere?
of which film PAST applaud the director ART Pere
Of which film did Pere applaud the director?

Spanish
b. [De qu pelcula]
i
aplaudi Pedro [al director t
i
]?

THE EMERGENCE OF ORDER IN SYNTAX 154
Although a PP cannot readily be subextracted from SPEC-* in NSLs, as in
English, there remains an important question to consider. The principal difference
between type (a) and type (b) of Gallegos examples of subextraction is that in
type (b) we can observe when the DP from which the PP is subextracted appears
in a preverbal or in a postverbal position. Recall examples (47), (48) and (49).

(59) a.
??/
*De qu conferenciantes te parece que las propuestas me van a
impresionar?
Which speakers does it seem to you that the proposals by will
impress me?

b. De qu conferenciantes

te parece que me van a impresionar las
propuestas?
Which speakers does it seem to you that the proposals by will
impress me?

(60) a.
??
De qu escritor

dice Mara que muchas novelas la han
impresionado?
Of what writer does Mara say that many novels have impressed
her?

b. De qu escritor

dice Mara que la han impresionado muchas
novelas?
Of what writer does Mara say that many novels have impressed
her?

(61) a.
??
De qu universidad

dice la prensa que muchos estudiantes

han
protestado?
Of what university does the media say that many students have
protested?

b. De qu universidad

dice la prensa que han protestado muchos
estudiantes?
Of what university does the media say that many students have
protested?

In these cases, when the DP is postverbal, subextraction of a PP is possible
because, in (59) and (60), the DP is not an external argument but a theme or
source, and in (61), the DP and the PP form a small clause [DP PP]. But why is
SYNCRETIC SYNTACTIC PATTERNS 155
subextraction more degraded when the DP is spelled out in SPEC-T of the
embedded clause? Spanish and Catalan SPEC-* and English SPEC-* do not
differ, but why do Spanish (and Catalan) SPEC-T differs from English SPEC-T,
as illustrated in (59-61)?
A possible answer could be as follows. Gallego may be correct in claiming
that [V-]-to-T movement gives SPEC-T the same status in NSLs as SPEC-C and
SPEC-*, since it provides a-features to TP. Thus, although *P is still a phase in
NSLs, V-movement creates a new phase by inheritance; like in the frame of
barriers, there would be inherent phases and derived phases. Accordingly, there
would not be, strictly, phase sliding, but rather the creation of new phases as part
of syntactic computations. Indeed, the idea that movement creates new effects is
uncontroversial, but the hypothesis that it eliminates previous effects from
syntactic representations is problematic. Why should [V-]-to-T movement free
the *P from its phasehood? Or why should [V-*]-to-T movement prevent V
from assigning a -role to its complement or to its SPEC? If movement could
eliminate previous effects from syntactic representations, it would destroy
semantic relations, and there would be no way of accounting for the observation
that movement brings semantic richness by associating a category to different po-
sitions (predicate positions -area, Infl-positions Infl-area and illocutionary
positions C-area).
The notion of phase disappears at the moment in which phases are claimed
to be created derivationally. The notion of phase is not equivalent to the notion
of barrier. Barriers were defined on the basis of syntactic opacity phenomena,
but phases are defined on the basis of completeness at the external systems: a
syntactic object is a phase if it can be argued to be complete at the external
systems C-I and A-P. Syntactic opacity is an effect of multiple transference
operations that cyclically map those autonomous units from the syntactic
workspace to the extenal systems, and thereby rendering them invisible at syntax.
Since -V and C-T constitute interpretive units (roughly, C-T conveys propo-
sitional features and -V predicate features) or discontinuous elements both in
NSLs and in English, there is no reason to think that [V-]-to-T movement
redefines TP as a phase, although it may still be accurate to consider that [V-]-
to-T movement redefines TP as a syntactic barrier or as an opacity inducer for
syntax.
Let us finally consider the source of a-features that surface in TP when an
inflected V has instantiated -features on T. Reconsider Gallegos intuition: [V-
]-to-T movement provides T with a-features. We know from Chomskys work
on English subject islands that -features of T and a-features of C work in par-
THE EMERGENCE OF ORDER IN SYNTAX 156
allel; as far as I see, this is a conclusion one cannot easily discard. Thus, in Eng-
lish, -features trigger movement of a DP to SPEC-T and a-features trigger
movement of a wh-phrase to SPEC-C. In Catalan, as argued above, -features are
instantiated by a finite V in T, not by movement of a DP to SPEC-T. Consider
now, for instance, the sentence given in (59.a). a-features of the embedded C
would probe a wh-phrase at the same time that -features probe the finite V (
1
in
(62)). In a further step (
2
), once T had been provided with topic-features by V-
movement, the topical DP would move to SPEC-T. And this is the derivational
paradox: SPEC-C would be created by remerging de qu conferenciantes before
SPEC-T was created by remerging las propuestas, and nonetheless, it is the
presence of las propuestas in SPEC-T that complicates the way C subextracts the
PP de qu conferenciantes.

(62)