Syntactic Theory and The Structure PDF
Syntactic Theory and The Structure PDF
Syntactic Theory and The Structure PDF
B. comrie Aspect
R. M. KEMPSON Semantic Theory
p. h. Matthews Syntax
A. RADFORD Transformational Syntax
S. C. LEVINSON Pragmatics
G. BROWN AND G. yule Discourse Analysis
r. lass Phonology
b. comrie Tense
A. (JRUTTENDEN Intonation
M. GARMAN Psycholinguistics
G. G. CORBETT Gender
ANDREW RADFORD
DEPARTMENT OF LANGUAGE AND LINGUISTICS, UNIVERSITY OF ESSEX
®w Cambridge
UNIVERSITY PRESS
PUBLISHED BY THE PRESS SYNDICATE OF THE UNIVERSITY OF CAMBRIDGE
The Pitt Building, Trumpington Street, Cambridge CB2 1RP, United Kingdom
A catalogue record for this book is available from the British Library
TAG
CONTENTS
Preface page xi
vii
Contents
3 Syntactic structure
3.1 Overview 86
3.2 Forming phrases 86
3.3 Specifiers 89
3.4 Forming sentences 94
3.5 Tree diagrams 97
3.6 Configurational relations 99
3.7 Testing structure 102
3.8 Additional tests 107
3.9 C-command 110
3.10 Summary 116
Workbook section 117
4 Empty categories
4.1 Overview 131
4.2 PRO subjects 131
4.3 Null auxiliaries 135
4.4 Unfilled INFL 137
4.5 Bare infinitives 144
4.6 Null complementizers 147
4.7 Null determiners 151
4.8 Pronouns 154
4.9 Attributive adjectives 157
4.10 Summary 159
Workbook section 159
5 Checking
5.1 Overview 170
5.2 Interpretable and uninterpretable features 170
5.3 Checking 174
5.4 Phrases 178
5.5 Percolation 184
5.6 Determiner phrases 187
5.7 PRO subjects 191
5.8 Objective subjects 193
5.9 Bare phrase structure 198
5.10 Summary 200
Workbook section 202
viii
Contents
6 Head movement
6.1 Overview 216
6.2 Auxiliary inversion 216
6.3 Traces 219
6.4 Verb movement 222
6.5 Strong and weak features 226
6.6 Negation 231
6.7 The syntax of have 235
6.8 Tense affix 240
6.9 Question affix 245
6.10 Summary 250
Workbook section 252
7 Operator movement
7.1 Overview 267
7.2 Wh-operators 267
7.3 Enlightened self-interest and shortest movement 271
7.4 Pied-piping 276
7.5 Embedded questions in Belfast English 282
7.6 Embedded questions in Standard English 286
7.7 Subject questions 292
7.8 Yes-no questions 294
7.9 Nonoperator questions 298
7.10 Summary 299
Workbook section 301
8 A movement
8.1 Overview 315
8.2 VP-intemal subject hypothesis 315
8.3 Evidence that subjects originate in spec-VP 318
8.4 Argument structure and theta-marking 324
8.5 Case-checking via head-adjunction and attraction 329
8.6 Raising predicates 334
8.7 Differences between raising and control predicates 337
8.8 Passivization 341
8.9 Explanation 345
8.10 Summary 349
Workbook section 351
IX
Contents
9 VP shells
9.1 Overview 367
9.2 Ergative predicates 367
9.3 Adverbs and prepositional particles 370
9.4 Ditransitive and resultative predicates 377
9.5 Three-place predicates with clausal complements 379
9.6 Object-control predicates 382
9.7 Monotransitive predicates 388
9.8 Unergative predicates 390
9.9 Unaccusative predicates 392
9.10 Summary 401
Workbook section 403
10 Agreement projections
10.1 Overview 422
10.2 Subject agreement projections 422
10.3 Evidence from other varieties of English 427
10.4 Object agreement projections 431
10.5 Exceptional case-marking 439
10.6 Indirect object agreement projections 444
10.7 Genitive DPs 448
10.8 For-infinitives and prepositional objects 449
10.9 Passives and unaccusatives reconsidered 454
10.10 Summary 459
Workbook section 461
x
PREFACE
The aim of this book is to provide an intensive introduction to recent work in syntactic
theory (more particularly, to key concepts which are presupposed in works written
within the broad framework of the minimalist program in the version outlined in
Chomsky 1995b). There are six main features which mark this book out as different
from other introductions to syntax currently available.
The first is that it does not presuppose any background knowledge of syntactic
theory: it is suitable for use with true beginners, and does not presuppose that students
have already done a course on generative syntax (though it is also suitable for false
beginners who have already taken a course on an earlier model of syntax, and want to
learn about more recent work).
The second is that it does not adopt a historical approach, or presuppose any historical
knowledge (for example, there is no discussion of earlier work in the 1980s government
and binding paradigm). Rather (for the most part), it deals directly with 1990s work
within the minimalist program.
The third is that cross-linguistic variation is illustrated in terms of different varieties
of English (e.g. Belfast English, Jamaican Creole, Child English, Shakespearean
English, etc.), rather than in terms of different languages. Hence it does not make the
unrealistic assumption that the reader knows (for example) Spanish, German, Arabic,
Chinese and Chuckchee.
The fourth is that the book contains a substantial workbook section at the end of each
chapter, containing exercise material designed to be used for class discussion, self-
study or coursework assignments, with fully worked out model answers provided for
key examples, and helpful hints for potentially problematic points. Where a particular
exercise presupposes understanding of key concepts introduced in particular sections
of the text, the relevant text sections are indicated in parentheses, after the exercise
number.
The fifth is that there is an extensive glossary at the end of the overall book,
intended to alleviate the terminological trauma of doing syntax.
The sixth is that the book is published alongside an abridged version called Syntax:
a Minimalist Introduction. The two books cover roughly the same range of topics: the
xi
Preface
I am grateful to Laura Rupp, Sam Featherston and Martin Atkinson (Essex), and Jon
Erickson (Cologne) for helpful comments on an earlier draft of this manuscript; and
above all to the series editor Neil Smith (University College London) for his patient
and good-humoured comments on numerous ((re-)re-)revised drafts of the manuscript.
This book is dedicated to my father and my sister (who both died before I had time to
thank them for all they did for me) and to my mother for battling so bravely against
bereavement and blindness over the past couple of years, before finally passing away
on 16 February 1997.
xii
1
Principles and parameters
1.1 Overview
The aim of this chapter is to outline contemporary ideas on the nature of
grammar and the acquisition of grammar. The approach adopted here is that associated
with the principles-and-parameters model developed by Noam Chomsky during the
1980s and 1990s, in works ranging from his 1981 book Lectures on Government and
Binding to his 1995c book The Minimalist Program.
1.2 Grammar
Grammar is traditionally subdivided into two different but inter-related
areas of study - morphology and syntax. Morphology is the study of how words are
formed out of smaller units (traditionally called morphemes), and so addresses ques¬
tions such as ‘What are the various component parts (= morphemes) of a word like
antidisestablishmentarianism, and what kinds of principles determine the ways in
which the parts are combined together to form the whole?’ Syntax is concerned with
the ways in which words can be combined together to form phrases and sentences, and
so addresses questions like ‘Why is it OK in English to say Who did you see Mary
with?, but not OK to say *Who did you see Mary and?' (A star in front of an expression
means that it’s ungrammatical.) ‘What kinds of principles determine the ways in which
we can and cannot combine words together to form phrases and sentences?’
However, grammar is traditionally concerned not just with the principles which
determine the formation of words, phrases and sentences, but also with the principles
which govern their interpretation - i.e. with the principles which tell us how to inter¬
pret (= assign meaning to) words, phrases and sentences. For example, any compre¬
hensive grammar of English will specify that compound words like man-eater and
man-made have very different interpretations: in compounds like man-eater, the word
man is traditionally said to have a patient interpretation, in the sense that man is the
patient/hapless victim on whom the act of eating is going to be performed; by contrast,
in compounds like man-made, the word man is said to have an agent interpretation, in
the sense that man is the agent responsible for the act of making. Thus, structural
aspects of meaning are traditionally said to be part of the domain of grammar. We
1
Principles and parameters
might therefore characterize grammar as the study of the principles which govern the
formation and interpretation of words, phrases and sentences. In terms of the tradi¬
tional division of grammar into morphology and syntax, we can say that morphology
studies the formation and interpretation of words, whereas syntax is concerned with the
formation and interpretation of phrases and sentences.
In a fairly obvious sense, any native speaker of a language can be said to know the
grammar of his or her native language. After all, native speakers clearly know how to
form and interpret words, phrases and sentences in their native language. For example,
any native speaker of English can tell you that the negative counterpart of I like syntax
is 1 don’t like syntax, and not e.g. */ no like syntax: thus, we might say that native
speakers know how to negate sentences in their language. However, it is important to
emphasize that this grammatical knowledge is tacit (i.e. subconscious) rather than
explicit (i.e. conscious): so, it’s no good asking a native speaker of English a question
such as ‘How do you form negative sentences in English?’, since human beings have
no conscious awareness of the psychological processes involved in speaking and
understanding a language. To introduce a technical term, we might say that native
speakers have grammatical competence in their native language: by this, we mean that
they have tacit knowledge of the grammar of their language - i.e. of how to form and
interpret words, phrases and sentences in the language.
In work dating back to the 1960s, Chomsky has drawn a distinction between compe¬
tence (the fluent native speaker’s tacit knowledge of his language) and performance
(what people actually say or understand by what someone else says on a given
occasion). Competence is ‘the speaker-hearer’s knowledge of his language’, while
performance is ‘the actual use of language in concrete situations’ (Chomsky 1965,
p. 4). Very often, performance is an imperfect reflection of competence: we all make
occasional slips of the tongue, or occasionally misinterpret what someone else says to
us. However, this doesn’t mean that we don’t know our native language, or don’t have
competence (i.e. fluency) in it. Misproductions and misinterpretations are performance
errors, attributable to a variety of performance factors like tiredness, boredom, drunk¬
enness, drugs, external distractions, and so forth. Grammars traditionally set out to
tell you what you need to know about a language in order to have native speaker com¬
petence in the language (i.e. to be able to speak the language like a native speaker):
hence, it is clear that grammar is concerned with competence rather than performance.
This is not to deny the interest of performance as a field of study, but merely to assert
that performance is more properly studied within the different - though related -
discipline of psycholinguistics, which studies the psychological processes underlying
speech production and comprehension. It seems reasonable to suppose that compe¬
tence will play an important part in the study of performance, since you have to
understand what native speakers tacitly know about their language before you can
study the effects of tiredness, drunkenness, etc. on this knowledge.
2
1.2 Grammar
If we say that grammar is the study of grammatical competence, then we are implic¬
itly taking a cognitive view of the nature of grammar. After all, if the term grammatical
competence is used to denote what native speakers tacitly know about the grammar of
their language, then grammar is part of the more general study of cognition (i.e. human
knowledge). In the terminology adopted by Chomsky (1986a, pp. 19-56), we’re
studying language as a cognitive system internalized within the human brain/mind; our
ultimate goal is to characterize the nature of the internalized linguistic system (or
I-language, as Chomsky terms it) which enables humans to speak and understand their
native language. Such a cognitive approach has obvious implications for the descrip¬
tive linguist who is interested in trying to describe the grammar of a particular
language like English. What it means is that in devising a grammar of English, we are
attempting to describe the grammatical knowledge possessed by the fluent native
speaker of English. However, clearly this competence is not directly accessible to us:
as noted above, you can’t ask native speakers to introspect about the nature of the
processes by which they produce and understand sentences in their native language,
since they have no conscious awareness of such processes. Hence, we have to seek to
study competence indirectly. But how?
Perhaps the richest vein of readily available evidence which we have about the
nature of grammatical competence lies in native speakers’ intuitions about the gram-
maticality and interpretation of words, phrases and sentences in their native language.
For example, preschool children often produce past tense forms like goed, corned,
seed, buyed, etc. and any adult native speaker of (Modem Standard) English will intu¬
itively know that such forms are ungrammatical in English, and will know that their
grammatical counterparts are went, came, saw and bought. Similarly, any native
speaker of English would intuitively recognize that sentences like (1 a) below are gram¬
matical in English, but that sentences like (lb) are ungrammatical:
(1) (a) If you don’t know the meaning of a word, look it up in a dictionary
(b) *If you don’t know the meaning of a word, look up it in a dictionary
3
Principles and parameters
sentence (e.g. (la) above) are grammatical, but also the ability to judge that others (e.g.
(lb) above) are ungrammatical. Indeed, much of contemporary work in syntax is con¬
cerned with trying to explain why certain types of structure are ungrammatical: it
would perhaps not be too much of an exaggeration to say that whereas traditional
grammars concentrate on grammaticality (i.e. on telling you how to form grammatical
phrases and sentences), work on grammar within the Chomskyan paradigm tends to
focus much more on explaining ungrammaticality (i.e. on explaining why certain types
of structures are ungrammatical).
A second source of introspective evidence about the nature of grammatical compe¬
tence relates to native speaker intuitions about the interpretation of words, phrases and
sentences in their native language. For example, any native speaker of English can tell
you that a sentence such as:
(3) (a) Sam loves you more than Jim loves you
(b) Sam loves you more than Sam loves Jim
So, it seems that the native speaker’s grammatical competence is reflected not only in
intuitions about grammaticality, but also in intuitions about interpretation.
4
1.3 Criteria of adequacy
questions like: ‘What are the inherent properties which natural language grammars do
and don’t possess?' Just as there are criteria of adequacy for grammars, so too there are
a number of criteria which any adequate theory of grammar must satisfy. One obvious
criterion is universality, in the sense that a theory of grammar should provide us with
the tools needed to describe the grammar of any natural language adequately; after all,
a theory of grammar would be of little interest if it enabled us to describe the grammar
of English and French, but not that of Swahili or Chinese. So, what we mean by saying
that universality is a criterion of adequacy for a theory of grammar is that a theory of
grammar must enable us to devise a descriptively adequate grammar for every natural
language: in other words, our ultimate goal is to develop a theory of Universal
Grammar. In the linguistic literature, it is a standard convention to abbreviate the term
Universal Grammar to UG. and hence to talk of devising a theory of UG.
However, since the ultimate goal of any theory is explanation, it is not enough for a
theory of Universal Grammar simply to list sets of universal properties of natural
language grammars; on the contrary, a theory of UG must seek to explain the relevant
properties. So, a key question for any adequate theory of UG to answer is: ‘ Why do nat¬
ural language grammars have the properties they do?’ The requirement that a theory
should explain why grammars have the properties they do is conventionally referred to
as the criterion of explanatory adequacy.
Since the theory of Universal Grammar is concerned with characterizing the proper¬
ties of natural (i.e. human) language grammars, an important question which we want
our theory of UG to answer is: ‘What are the essential defining characteristics of
natural languages which differentiate them from, for example, artificial languages like
those used in mathematics and computing (e.g. Basic, Prolog, etc.), or from animal
communication systems (e.g. the tail-wagging dance performed by bees to communi¬
cate the location of a food source to other bees)?’ It therefore follows that the descrip¬
tive apparatus which our theory of Universal Grammar allows us to make use of in
devising natural language grammars must not be so powerful that it can be used to
describe not only natural languages, but also computer languages or animal communi¬
cation systems (since any such excessively powerful theory wouldn’t be able to
pinpoint the criterial properties of natural languages which differentiate them from
other types of communication system). In other words, a third condition which we
have to impose on our theory of language is that it be maximally restrictive: that is, we
want our theory to provide us with technical devices which are so constrained in their
expressive power that they can only be used to describe natural languages, and are not
appropriate for the description of other communication systems. Any such restrictive
theory would then enable us to characterize the very essence of natural language.
The neurophysiological mechanisms which underlie linguistic competence make it
possible for young children to acquire language in a remarkably short period of time:
children generally start to form elementary two-word structures at around 18 months of
5
Principles and parameters
age, and by the age of 30 months have acquired a wide range of different grammatical
constructions and are able to produce sentences of considerable grammatical complex¬
ity. Accordingly, a fourth condition which any adequate linguistic theory must meet is
that of leamability: it must provide grammars which are leamable by young children in
a relatively short period of time.
A related requirement is that linguistic theory should provide grammars which make
use of the minimal theoretical apparatus required to provide a descriptively adequate
characterization of linguistic phenomena: in other words, grammars should be as sim¬
ple as possible. Much of the work in syntax in the 1980s involved the postulation of
ever more complex structures and principles: as a reaction to the excessive complexity
of this kind of work, Chomsky in the 1990s has made minimalism (i.e. the require¬
ment to minimize the theoretical and descriptive apparatus used to describe language)
the cornerstone of linguistic theory. The minimalist program for linguistic theory
which he has been developing (cf. Chomsky 1995c) is motivated to a large extent by
the desire to minimize the acquisition burden placed on the child, and thereby maxi¬
mize the leamability of natural language grammars.
6
1.4 Language faculty
Thus, the central phenomenon which any theory of language acquisition must seek
to explain is this: how is it that after a long drawn-out period of many months in which
there is no obvious sign of grammatical development, at around the age of 18 months
there is a sudden spurt as multiword speech starts to emerge, and a phenomenal growth
in grammatical development then takes place over the next twelve months? This uni¬
formity and (once the spurt has started) rapidity in the pattern of children's linguistic
development are the central facts which a theory of language acquisition must seek to
explain. But how?
Chomsky maintains that the most plausible explanation for the uniformity and rapid¬
ity of first language acquisition is to posit that the course of acquisition is determined
by a biologically endowed innate language faculty (or language acquisition program,
to borrow a computer software metaphor) within the brain, which provides children
with a (genetically transmitted) algorithm (i.e. set of procedures) for developing a
grammar, on the basis of their linguistic experience (i.e. on the basis of the speech
input they receive). The way in which Chomsky visualizes the acquisition process can
be represented schematically as in (4) below (where L is the language being acquired):
(4)
Language
Faculty
Children acquiring a language will observe people around them using the language,
and the set of expressions in the language which the child hears - and the contexts in
which they are used - in the course of acquiring the language constitute the child’s
linguistic experience of the language. This experience serves as input to the child’s lan¬
guage faculty, which provides the child with a procedure for (subconsciously)
analysing the experience in such a way as to devise a grammar of the language being
acquired. Thus, the input to the language faculty is the child’s experience, and the out¬
put of the language faculty is a grammar of the language being acquired.
The hypothesis that the course of language acquisition is determined by an innate
language faculty is known popularly as the innateness hypothesis. Chomsky maintains
that language acquisition is an activity unique to human beings, and different in kind
from any other type of learning which human beings experience, so that learning a lan¬
guage involves mental processes entirely distinct from those involved in e.g. learning
to play chess, or learning to ride a bicycle.
7
Principles and parameters
Whatever evidence we do have seems to me to support the view that the ability
to acquire and use language is a species-specific human capacity, that there are
very deep and restrictive principles that determine the nature of human
language and are rooted in the specific character of the human mind. (Chomsky
1972a, p. 102)
In addition, the apparent uniformity in the pattern of acquisition suggests that children
have genetic guidance in the task of constructing a grammar of their native language: cf.
We know that the grammars that are in fact constructed vary only slightly
among speakers of the same language, despite wide variations not only in
intelligence but also in the conditions under which language is acquired.
(Chomsky 1972a, p. 79)
Furthermore, the rapidity of acquisition (once the grammar spurt has started) also
points to genetic guidance in grammar construction:
(The sequence ‘under . . . data’ means simply ‘in so short a time, and on the basis of
such limited linguistic experience’.) What makes the uniformity and rapidity of acqui¬
sition even more remarkable is the fact that the child’s linguistic experience is often
degenerate (i.e. imperfect), since it is based on the linguistic performance of adult
speakers, and this may be a poor reflection of their competence: cf.
A good deal of normal speech consists of false starts, disconnected phrases, and
other deviations from idealized competence. (Chomsky 1972a, p. 158)
8
7.5 Creativity
Descartes asks: how is it when we see a sort of irregular figure drawn in front
of us we see it as a triangle? He observes, quite correctly, that there’s a
disparity between the data presented to us and the percept that we construct.
And he argues, I think quite plausibly, that we see the figure as a triangle
because there’s something about the nature of our minds which makes the
image of a triangle easily constructible by the mind. (Chomsky 1968, p. 687)
The obvious implication is that in much the same way as we are genetically predis¬
posed to analyse shapes (however irregular) as having specific geometrical properties,
so too we are genetically predisposed to analyse sentences (however ungrammatical)
as having specific grammatical properties. A further argument Chomsky uses in sup¬
port of the innateness hypothesis relates to the fact that language acquisition is an
entirely subconscious and involuntary activity (in the sense that you can’t consciously
choose whether or not to acquire your native language - though you can choose
whether or not you wish to learn chess); it is also an activity which is largely unguided
(in the sense that parents don’t teach children to talk): cf.
The implication is that we don’t learn to have a native language, any more than we
learn to have arms or legs; the ability to acquire a native language is part of our genetic
endowment - just like the ability to learn to walk.
1.5 Creativity
An essential tenet of Chomsky’s theory is that the nature of the language
acquisition process in humans is profoundly different in character from animal learning
processes with which we are familiar - for instance, the process by which animals learn to
respond to specific stimuli (e.g. learn to press buttons in specific combinations in order to
get food), and so develop a habit structure (i.e. a conditioned response to a specific type of
stimulus). If language acquisition involved this kind of conditioned response, we should
expect it to be a purely imitative process - i.e. a process in which children simply memorize
lists of sentences which they have heard people around them produce. However (Chomsky
argues), language acquisition is not a purely imitative process, but rather an inherently
creative process. Evidence which supports this conclusion is that when we acquire a lan¬
guage, we are capable of producing and understanding not only sentences we have previ¬
ously heard, but also novel sentences which we have never encountered before: cf.
9
Principles and parameters
The most striking aspect of linguistic competence is what we may call the
creativity of language, that is, the speaker’s ability to produce new sentences,
sentences that are immediately understood by other speakers although they bear
no physical resemblance to sentences which are familiar. (Chomsky 1966, p. 11)
(By resemblance here, Chomsky means identity: i.e. he is saying that we can produce
and understand sentences which are not exact word-for-word repetitions of any sen¬
tences we have ever heard before.) To cite one of Chomsky’s own examples (from
p. 132 of the published version of The Logical Structure of Linguistic Theory - written
in 1955 but published in 1975), you have probably never encountered any of the
following sentences before:
And yet - if you are a native speaker of English - you intuitively know that (5a) is
perfectly OK in English (though of course, the zoo-keepers among you might object that
elephants can’t be cross-eyed). By contrast, (5b) is decidedly odd, and (5c) seems
nonsensical. Any native speaker is capable of producing and understanding such novel
utterances, or making judgments about their acceptability. What is the significance of the
fact that all native speakers have the ability to produce, understand and make judgments
about sentences that they have not come across before? Chomsky argues that this essen¬
tial creativity of language shows that language can't simply be learned by imitation:
i.e. learning a language doesn’t simply involve rote-leaming a list of sentences produced
by others, and repeating them parrot-fashion. On the contrary’, as Chomsky notes:
The normal use of language is innovative in the sense that much of what we say
in the course of normal language use is entirely new, not a repetition of
anything that we have heard before, and not even similar in pattern - in any
useful sense of the terms ‘similar’ and ‘pattern’ - to sentences or discourse that
we have heard in the past. (Chomsky 1972a, p. 12)
The novelty of most sentences that we produce or hear provides a strong argument
against the claim made by behavioural psychologists that language-learning is a purely
imitative process which involves the acquisition of a set of linguistic habits.
Empirical studies of children acquiring their native languages would seem to bear
out Chomsky’s claim that language acquisition is not an imitative but a creative
process. Consider, for example, the process by which children acquire grammatical
inflections. On the imitative view, acquisition would involve memorization of a set of
inflected forms which children have previously encountered (so that e.g. children
would only be able to form the plural of a noun which they have previously heard used
in the plural). By contrast, on the creative view, acquisition would be a very different
10
1.5 Creativity
11
Principles and parameters
Clearly, no adult English speaker would produce sentences like (6), so that they cannot
be the product of simple imitation. How do the children come to produce them? The
obvious answer is that on the basis of hearing adult sentences such as:
the children create for themselves the generalization that ‘You form a question by
putting is as the first word in the sentence.’ Of course, the generalization is wrong (in
the sense that this isn’t how adults form questions) and will be corrected at a later stage
of acquisition by the children. But these novel structures produced by young children
provide evidence that acquiring a grammar involves formulating a set of generaliza¬
tions about how inflections are used, and how words are combined together to form
phrases and sentences. Thus, its inherently creative nature makes language acquisition
a qualitatively different process from animal learning behaviour, so reinforcing
Chomsky’s view that language acquisition is the product of a species-specific language
faculty.
1.6 Principles
If human beings do indeed have a genetically endowed language faculty,
an obvious question to ask is what are the defining characteristics of the language fac¬
ulty. An important point to note in this regard is that children can in principle acquire
any natural language as their native language (e.g. Bosnian orphans brought up by
English-speaking foster parents in an English-speaking community acquire English as
their first language). It therefore follows that the language faculty must incorporate a
set of principles of Universal Grammar/UG - in the sense that the language faculty
must be such as to allow the child to develop a grammar of any natural language on the
basis of suitable linguistic experience of the language (i.e. sufficient speech input).
Experience of a particular language L (examples of words, phrases and sentences in L
which the child hears produced by native speakers of L) serves as input to principles of
UG which are an inherent part of the child’s language faculty, and UG then provides
the child with an algorithm for developing a grammar of L.
If the acquisition of grammatical competence is indeed controlled by a genetically
endowed language faculty incorporating principles of Universal Grammar, then it
follows that certain aspects of child (or adult) competence are known without experi¬
ence, and hence must be part of the genetic blueprint for language with which we are
biologically endowed at birth. Such aspects of language would not have to be learned,
precisely because they form part of the child’s genetic inheritance. If we make the (not
unnatural) assumption that the language faculty does not vary significantly from one
12
1.6 Principles
(normal) human being to another, then it follows that those aspects of language which
are innately determined will also be universal. Thus, in seeking to determine the nature
of the language faculty, we are in effect looking for universal principles which deter¬
mine the very structure of language - principles which govern the kinds of grammati¬
cal operations which are (and are not) permitted in natural languages, principles of
such an abstract nature that they could not plausibly have been learned on the basis of
experience.
But how can we discover what these abstract universal principles are which con¬
strain the range of grammatical operations permitted in natural language grammars?
The answer is that since the relevant principles are posited to be universal, it follows
that they will constrain the application of every grammatical operation in every lan¬
guage. Thus, detailed observation of even one grammatical operation in one language
should reveal evidence of the operation of principles of UG (Universal Grammar). By
way of illustration, let’s consider how yes-no questions are formed in English (i.e.
questions to which ‘Yes’ and ‘No’ are appropriate as one-word answers). If we com¬
pare a simple declarative sentence (= statement) like (8a) below with its interrogative
(= question) counterpart (8b):
we see that the yes-no question in (8b) appears to be formed by preposing the word
will in front of the word memories (this particular grammatical operation is often
referred to as interrogative inversion, the word inversion here serving to indicate a
change of word order). We might therefore imagine that inversion involves some oper¬
ation such as the following:
(9) Move the second word in a sentence in front of the first word
However, it’s easy enough to show that (9) is descriptively inadequate. For example, it
would wrongly predict that the interrogative counterpart of (10a) below would be
(10b):
Why is (9) wrong? The most principled answer which we can hope to give is that gram¬
mars of natural languages simply don’t work like that. More specifically, we might
suppose that what’s wrong with an operation like (9) is that it is structure-independent,
in the sense that it operates independently of (i.e. makes no reference to) the grammati¬
cal structure of the sentence, so that you don’t need to know what the grammatical struc¬
ture of the sentence is (i.e. which words belong to which grammatical categories, or
which words modify which other words, etc.) in order to know how inversion works.
13
Principles and parameters
In fact, no linguist has ever claimed to have found any grammatical operation like
(9) which operates in a completely structure-independent fashion in any natural
language. It therefore seems reasonable to suppose that we would want to exclude
structure-independent operations like (9) from our theory of grammar. One way of
doing this would be to incorporate into our theory of Universal Grammar a principle
such as (11) below:
What this means is that grammatical operations only apply to certain types of gram¬
matical structure, not others. Let’s look briefly at what is meant by grammatical
structure in this context.
It is traditionally said that sentences are structured out of words, and phrases, each
of which belongs to a specific grammatical category and serves a specific grammat¬
ical function within the sentence containing it. (We shall turn to consider categories in
detail in the next two chapters, so for the time being don’t worry if you’re not familiar
with the terminology used here.) To see what we mean by this, consider the grammati¬
cal structure of our (pessimistic) sentence (10a) Memories of happiness will fade away.
In traditional grammatical analysis, each of the words in the sentence would be
assigned to a specific grammatical category, e.g. memories and happiness belong to
the category of noun .fade is a verb, will is an auxiliary (so called because it provides
additional - hence auxiliary - information about the action or process described by the
verb, in this case indicating that the process of fading away will take place in the
future), of is a preposition, and away is traditionally classed as an adverb (though
might be analysed as a prepositional particle). Certain of the words in the sentence
I
combine together to form phrases, and each of these phrases in turn belongs to a spe¬
cific category: for example, the string memories of happiness is a noun phrase, and the
string fade away is a verb phrase. These two phrases are joined together by
the auxiliary will, thereby forming the overall sentence Memories of happiness will
fade away. Each of the various constituents (i.e. component parts) of the sentence also
serves a specific grammatical function. For example, the noun phrase memories of
happiness serves the grammatical function of being the subject of the auxiliary will,
whereas the verb phrase fade away serves the function of being the complement of
will. (See the glossary if you are unfamiliar with these terms.)
Having looked briefly at some traditional assumptions about grammatical structure,
let’s now return to our earlier discussion and see how we might develop a structure-
dependent approach to the phenomenon of interrogative inversion. In keeping with
suggestions made in traditional grammars, we might suppose that inversion involves
an operation such as the following:
14
1.6 Principles
In (13b) we have the prepositional particle down inverted with the auxiliary will; and in
(14b) we have the noun expression John inverted with the verb received. Since in
neither case do we have inversion of a subject noun expression with an auxiliary, our
structure-dependent characterization of inversion correctly predicts that inversion
cannot take place here. By contrast, our structure-independent characterization of
inversion in (9) wrongly predicts that inversion should be possible in both cases.
So, the phenomenon of inversion in questions provides us with clear evidence in
support of incorporating into our theory of grammar the structure dependence princi¬
ple (11), which specifies that all grammatical operations are sensitive to the grammati¬
cal structure of the sentences they apply to. It seems reasonable to suppose that (11) is
a fundamental principle of Universal Grammar, so that (11) holds for all grammars of
all natural languages. If (11) is universal, and if we assume that abstract grammatical
principles which are universal are part of our biological endowment, then the natural
15
Principles and parameters
conclusion to reach is that (11) is a principle which is incorporated into the language
faculty, and which thus forms part of the child’s genetic blueprint for a grammar. If so,
the innate language faculty incorporates a set of universal grammatical principles (such
as the structure dependence principle): these are conventionally referred to as UG
principles or principles of UG. If the language faculty incorporates a theory of
Universal Grammar which provides the child with an algorithm (i.e. set of procedures)
for developing a grammar of any language, we can revise our earlier model of acquisi¬
tion given in (4) above as (15) below:
► UG
In this (revised) model, the child’s experience is processed by the UG module which is
an integral part of the language faculty.
A theory of grammar which posits that the internal grammatical structure of words,
phrases and sentences in natural language is determined by innate principles of UG
offers the important advantage that it minimizes the burden of grammatical learning
imposed on the child. This is a vital consideration, since we saw earlier that leamability
is a criterion of adequacy for any theory of grammar - i.e. any adequate theory of
grammar must be able to explain how children come to learn the grammar of their
native language(s) in such a rapid and uniform fashion. The UG theory developed by
Chomsky accounts for the rapidity of the child’s grammatical development by positing
that there is a universal set of innately endowed grammatical principles which deter¬
mine the nature of grammatical structure and the range of grammatical operations found
in natural language grammars. Since UG principles which are innately endowed do not
have to be learned by the child, the theory of UG minimizes the learning load placed on
the child, and thereby maximizes the leamability of natural language grammars.
1.7 Parameters
Thus far, we have argued that the language faculty incorporates a set of
universal grammatical principles (= UG principles) which are invariant across lan¬
guages, and which determine the nature and acquisition of grammatical structure.
However, it clearly cannot be that all aspects of the grammatical structure of languages
are determined by innate grammatical principles; if this were so, all languages would
have precisely the same grammatical structure, and there would then be no structural
learning involved in language acquisition (i.e. no need for children to learn anything
about the grammatical structure of sentences in the language they are acquiring), only
lexical learning (viz. learning about the lexis (= vocabulary) of the language - i.e. learn¬
ing the words in the language and their idiosyncratic linguistic properties, e.g. whether a
given item has an irregular plural form, etc.). However, it is quite clear that although
16
1.7 Parameters
there are universal principles which determine the broad outlines of the grammatical
structure of words, phrases and sentences in every natural language, there are also lan¬
guage-particular aspects of grammatical structure which children have to learn as part
of the task of acquiring their native language. Thus, language acquisition involves not
only lexical learning, but also some structural learning. Our main concern here is to
examine structural learning, and what it tells us about the language acquisition process.
Clearly, structural learning is not going to involve learning those aspects of structure
which are determined by universal (hence innate) grammatical principles. Rather,
structural learning will be limited to those parameters (i.e. ‘dimensions’ or ‘aspects’)
of grammatical structure which are subject to language-particular variation (i.e. which
vary from one language to another). In other words, structural learning will be limited
to parametrized aspects of structure (i.e. those aspects of structure which are subject to
parametric variation from one language to another). Thus, the obvious way to deter¬
mine just what aspects of the grammatical structure of their native language children
have to learn is to examine the range of parametric variation in grammatical structure
between different (adult) natural languages.
We can illustrate one type of parametric variation across languages in terms of the
following contrast between the Italian examples in (16 a-b) below, and their English
counterparts in (16 c-d):
As we see from (16a/c), in both Italian and English, finite verbs (i.e. verbs which carry
present/past etc. tense) like parla!speaks license (i.e. ‘can have’) an overt subject like
Maria; the two languages differ, however, in that finite verbs also license a null (i.e.
‘missing but understood’) subject in Italian (as we see from the fact that (16b) is gram¬
matical in Italian, and is understood as meaning 'He/she speaks French’), but not in
English (so that (16d) is ungrammatical in English). Thus, finite verbs in a language
like Italian license either overt or covert (= null) subjects, but in a different kind of
language like English, finite verbs license only overt subjects, not null subjects. We
might describe the differences between the two languages by saying that Italian is a
null subject language, whereas English is a non-null subject language. More gener¬
ally, there appears to be parametric variation between languages as to whether or not
they allow finite verbs to have null subjects. The relevant parameter (termed the null
subject parameter) would appear to be a binary one, with only two possible settings,
viz. ‘does/doesn’t allow finite verbs to have null subjects’. There appears to be no lan¬
guage which allows the subjects of some finite verbs to be null, but not others - e.g. no
language in which it is OK to say Drinks wine (meaning ‘He/she drinks wine’) but not
17
Principles and parameters
OK to say Eats pasta (meaning ‘He/she eats pasta’). The range of grammatical varia¬
tion found across languages appears to be strictly limited: there seem to be just two
possibilities - languages either do or don’t systematically allow finite verb? to have
null subjects.
A more familiar aspect of grammatical structure which is obviously parametrized
relates to word order, in that different types of language have different word orders in
specific types of construction. One type of word order variation can be illustrated in
relation to the following contrast between English and Chinese:
18
1.7 Parameters
noun phrase students of linguistics is said to be the complement of the noun students.
In much the same way, an expression such as in the kitchen is a prepositional phrase
which comprises the head preposition in and its complement the kitchen. Likewise, an
expression such as stay with me is a verb phrase which comprises the head verb stay
and its complement with me. And similarly, an expression such as fond of fast food is
an adjectival phrase formed by combining the head adjective fond with the comple¬
ment of fast food.
In English (and many other languages) head nouns, verbs, prepositions, adjectives,
etc. precede their complements; however, there are also languages like Korean in
which heads follow their complements. Thus, in informal terms, we might say that
English is a head-first language, whereas Korean is a head-last language. The differ¬
ences between the two languages can be illustrated by comparing the English examples
in (18) below with their Korean counterparts in (19):
In the English verb phrase close the door in (18a), the head verb close precedes its
complement the door, if we suppose that the door is a determiner phrase, then the head
of the phrase (= the determiner the) precedes its complement (= the noun door).
Likewise, in the English noun phrase desire for change in (18b), the head noun desire
precedes its complement for change; the complement for change is in turn a preposi¬
tional phrase in which the head preposition for likewise precedes its complement
change. Since English consistently positions heads before complements, it is a head¬
first language.
By contrast, we find precisely the opposite ordering in Korean. In the verb phrase
moonul dadala (literally ‘door close’) in (19a), the head verb dadala ‘close’ follows its
complement moonul ‘door’; likewise, in the noun phrase byunhwa-edaehan kalmang
(literally ‘change-for desire’) in (19b) the head noun kalmang ‘desire’ follows its com¬
plement byunhwa-edaehan ‘change-for’; the complement byunhwa-edaehan ‘change-
for’ is in turn a prepositional phrase whose head preposition edaehan ‘for’
follows its complement byunhwa ‘change’ (so that edaehan ‘for’ might more appropri¬
ately be called a postposition). Since Korean consistently positions heads after their
complements, it is a head-last language. Given that English is head-first and Korean
head-last, it is clear that the relative positioning of heads with respect to their comple¬
ments is one word-order parameter along which languages will differ; the relevant
parameter might be referred to as the head (position) parameter.
19
Principles and parameters
1.8 Parameter-setting
We might generalize our discussion at this point in the following terms. If
the head parameter reduces to a simple binary choice, and if the wh-parameter and
the null subject parameter also involve binary choices, it seems implausible that
binarity could be an accidental property of these particular parameters. Rather, it seems
much more likely that it is an inherent property of parameters that they constrain the
range of structural variation between languages, and limit it to a simple binary choice.
Generalizing still further, let us suppose that all grammatical variation between
languages can be characterized in terms of a set of parameters, and that for each para¬
meter, UG (= Universal Grammar) specifies a binary choice of possible values for the
parameter. If so, then the only structural learning which children face in acquiring
20
1.8 Parameter-setting
their native language is the task of determining the appropriate value for each of the
relevant structural parameters along which languages vary. (Of course, children also
face the formidable task of lexical learning - i.e. building up their vocabulary in the
relevant language.)
If our reasoning here is along the right lines, then it leads us to the following view of
the language acquisition process. The central task which the child faces in acquiring a
language is to construct a grammar of the language. The child’s language faculty incor¬
porates a theory of Universal Grammar which includes (i) a set of universal principles
of grammatical structure, and (ii) a set of structural parameters which impose severe
constraints on the range of structural variation permitted in natural languages (perhaps
limiting the range of variation to a series of binary choices). Since universal principles
of grammatical structure don’t have to be learned, the child’s structural learning task is
limited to that of parameter-setting (i.e. determining an appropriate setting for each of
the relevant structural parameters). For obvious reasons, the model outlined here has
become known as the principles-and-parameters theory (= PPT) of language.
The PPT model clearly has important implications for the nature of the language
acquisition process: more precisely, such a model would vastly reduce the complexity
of the acquisition task which children face. It would lead us to expect that those struc¬
tural principles of language which are invariant across languages will not have to be
learned by the child, since they will be part of the child’s genetic endowment: on the
contrary, all that the child has to learn are those grammatical properties which are sub¬
ject to parametric variation across languages. Moreover, the child’s learning task will
be further simplified if it turns out (as we have suggested here) that the values which a
parameter can have fall within a narrowly specified range, perhaps characterizable in
terms of a series of binary choices. This simplified parameter-setting conception of the
child’s acquisition task has given rise to a metaphorical acquisition model in which the
child is visualized as having to set a series of switches in one of two positions
(up/down) - each such switch representing a different structural parameter. In the case
of the head parameter which we discussed above, we might hypothesize that if
the switch is set in the up position, the language will be head-first, whereas if it is set
in the down position, the language will be head-last. Of course, an obvious implication
of the switch metaphor is that the switch must be set in either one position or the other,
and so cannot be set in both positions. (This would preclude the possibility of a language
having both head-first and head-last structures.)
The assumption that acquiring the syntax of a language involves the relatively sim¬
ple task of setting a number of structural parameters provides a natural way of account¬
ing for the fact that the acquisition of specific parameters appears to be a remarkably
rapid and error-free process in young children. For example, young children acquiring
English as their native language seem to set the head parameter at its appropriate
head-first setting from the very earliest multiword utterances they produce (at around
21
Principles and parameters
age 18 months), and seem to know (tacitly, not explicitly, of course) that English is a
head-first language. For example, the earliest verb phrases and prepositional phrases
produced by young children consistently show verbs and prepositions positioned
before their complements, as structures such as the following indicate (produced by a
young boy called Jem at age 20 months; head verbs or prepositions are italicized, and
their complements are in non-italic print):
(20) (a) Touch heads. Cuddle book. Want crayons. Want malteser. Open door.
Want biscuit. Bang bottom. See cats. Sit down
(b) On mummy. To lady. Without shoe. With potty. In keyhole. In school. On
carpet. On box. With crayons. To mummy
The obvious conclusion to be drawn is that children consistently position heads before
their complements from the very earliest multiword utterances which they produce.
They do not use different orders for different words of the same type (e.g. they don’t
position the verb see after its complement but the verb want before its complement), or
for different types of words (e.g. they don’t position verbs before and prepositions after
their complements).
A natural question to ask at this point is how we can provide a principled explana¬
tion for the fact that from the very onset of multiword speech we find heads correctly
positioned before their complements. The principles-and-parameters model of
acquisition enables us to provide a principled explanation for why children manage to
learn the relative ordering of heads and complements in such a rapid and error-free
fashion. The answer provided by the model is that learning this aspect of word order
involves the comparatively simple task of setting a binary parameter at its appropriate
value. This task will be a relatively straightforward one if UG tells the child that the
only possible choices are for a language to be uniformly head-first or uniformly head-
last. Given such an assumption, the child could set the parameter correctly on the basis
of minimal linguistic experience. For example, once the child is able to parse (i.e.
grammatically analyse) an adult sentence such as Help Daddy and knows that it
contains a verb phrase comprising the head verb help and its complement Daddy, then
(on the assumption that UG specifies that all heads behave uniformly with regard to
whether they are positioned before or after their complements), the child will automati¬
cally know that all heads in English are normally positioned before their complements.
1.9 Evidence
One of the questions posed by the parameter-setting model of acquisition
outlined in the previous section is just how children come to arrive at the appropriate
setting for a given parameter, and what kind(s) of evidence they make use of in setting
parameters. As Chomsky notes (1981, pp. 8-9), there are two types of evidence which
we might expect to be available to the language learner in principle - viz. positive and
22
1.9 Evidence
negative evidence. Positive evidence would comprise a set of observed sentences illus¬
trating a particular phenomenon: for example, if children’s experience (i.e. the speech
input they receive) is made up of structures in which heads precede their complements,
this provides them with positive evidence which enables them to set the head parame¬
ter appropriately. Negative evidence may be of two kinds - direct and indirect. Direct
negative evidence comes from the correction of the child’s errors by other speakers of
the language. However, correction (contrary to what is often imagined) plays a fairly
insignificant role in language acquisition, for two reasons. Firstly, correction is rela¬
tively infrequent: adults simply don’t correct all the errors children make (if they did,
children would soon become inhibited and discouraged from speaking); and secondly,
children are notoriously unresponsive to correction, as the following dialogue (from
McNeill 1966, p. 69) illustrates:
As you can see, when I attempted to correct Suzy, she responded by attempting to
correct me.
Direct negative evidence might also take the form of self-correction by other speak¬
ers. Such self-corrections tend to have a characteristic intonation and rhythm of their
own, and may be signalled by a variety of fillers (such as those italicized in (23)
below):
(23) (a) The picture was hanged ... or rather hung ... in the Tate Gallery
(b) The picture was hanged . . . sorry hung ... in the Tate Gallery
(c) The picture was hanged . . . I mean hung ... in the Tate Gallery
23
Principles and parameters
1.10 Summary
In this chapter, we have been concerned with the nature and acquisition of
grammar. We began in §1.2 by arguing that a grammar of a language is a model of the
grammatical competence of the fluent native speaker of the language, and that grammati-
24
Workbook section
cal competence is reflected in native speakers' intuitions about grammaticality and inter¬
pretation. We suggested in §1.3 that, correspondingly, the main criterion of adequacy for
grammars is that of descriptive adequacy (i.e. correctly accounting for grammaticality and
interpretation). We noted that a theory of grammar is concerned with characterizing the
general properties and organization of grammars of natural languages; we suggested that
any adequate theory of language should be universal, explanatory, restrictive, minimally
complex and should provide grammars which are leamable (in order to explain how chil¬
dren acquire grammars in such a rapid and error-free fashion). In §1.4 we outhned
Chomsky's innateness hypothesis, under which the course of language acquisition is
genetically predetermined by an innate language faculty. In §1.5 we argued that language
acquisition is a creative process (which involves the child in formulating generalizations
about how words, phrases and sentences are formed), not an imitative one. In §1.6 we
argued that the language faculty must be assumed to incorporate a set of principles of
Universal Gramtnar/UG (i.e. universal grammatical principles) if we are to account for
children’s ability to acquire the grammar of any language (on the basis of suitable expo¬
sure to the language). We argued that if principles of UG determine the nature of every
grammatical process and structure in every natural language, close examination of any
one type of grammatical structure or operation in any one language will reveal evidence
for the operation of UG principles - and we saw that the operation of inversion in English
provides evidence for postulating the structure dependence principle. In §1.7 we went
on to suggest that languages diff er in their structure along a range of different grammatical
parameters. We looked at three such parameters - the wh-parameter, the null subject
parameter and the head parameter. In §1.8 we argued that each of these parameters is
inherently binary in nature, and that consequently the structural learning which the child
faces involves parameter-setting — i.e. determining which of the two alternative settings
provided by UG is the appropriate one for each parameter in the language being acquired.
We further argued that if the only structural learning involved in language acquisition is
parameter-setting, we should expect to find evidence that children correctly set parameters
from the very onset of multiword speech: and we presented evidence to suggest that from
their very earliest multiword utterances, children correctly set the head parameter at the
head-first value appropriate for English. In § 1.9 we asked what kind(s) of evidence chil¬
dren use in setting parameters, and concluded that they use positive evidence from their
experience of the occurrence of some particular type of structure (e.g. head-first structures,
or null-subject structures, or wh-fronted structures).
Workbook section
Exercise I
Say how many morphemes (i.e. grammatical units) each of the following
words is composed of; identify the constituent morphemes, and say how you arrived at
your answer:
25
Principles and parameters
For the purposes of this exercise, assume that a word is a compound of two or more
morphemes if each of the potential constituent morphemes can occur with the same
form, function and meaning when used as (part of) a different word.
Exercise II
Discuss the interpretation(s) which the italicized expressions can have in
the following sentences, giving an appropriate paraphrase for each interpretation:
In particular, consider which word(s) the italicized expression is the subject or comple¬
ment of, and whether it occupies the canonical (i.e. normal) position associated with a
subject/complement. (See the entries for subject and complement in the glossary if
these terms are unfamiliar.)
26
Workbook section
The ambiguity here seems to arise through ellipsis (i.e. the omission of one or more
words which can be understood from the context): on interpretation (i), he and loves
undergo ellipsis (i.e. are omitted); on interpretation (ii), love and me undergo ellipsis. On
interpretation (i), you is interpreted as the complement of the ellipsed sequence he loves.
On interpretation (ii), you is understood as the subject of the ellipsed expression love me.
Exercise III
Say whether you think each of the following sentences is (or is not) gram¬
matical in (Modem Standard) English:
27
Principles and parameters
Helpfid hints
Don’t assume that only one of the sentences in each set is grammatical,
and that all the others are ungrammatical: bear in mind the possibility that all the sen¬
tences in each set may be grammatical (or indeed none); or, different sentences may be
grammatical in different styles of English. In some cases, you may find it helpful to
distinguish different degrees of grammaticality, e.g. using a prefixed ? to indicate a
sentence which is a bit forced, ?? to indicate a sentence which is unnatural, ?* to indi¬
cate a sentence which is pretty awful (while not being entirely ungrammatical), * to
indicate a sentence which is completely ungrammatical, and ** to indicate a sentence
which is doubly ungrammatical (etc.).
Exercise IV
Pedagogical grammarians (i.e. grammarians writing grammar books
designed for teaching in schools) often take a prescriptive approach to grammar, in that
they prescribe (i.e. lay down) grammatical norms for Standard English and brand all non¬
standard structures as deviations from supposedly correct usage. By contrast, contempo¬
rary linguists adopt a descriptive approach to language, and are concerned to describe
what people actually say (rather than prescribe what they ought to say). Compare and
contrast the prescriptive and descriptive approaches to the following set of sentences.
1 The mission of the USS Enterprise is to boldly go where no man has gone
before
2 The weather will hopefully clear up by tomorrow
3 It’s me that gets the blame for everything
4 John and Mary love one another
5 What are you up to?
6 You are taller than me
7 Everyone loses their cool now and then
8 Those kind of people get on my nerves
9 If I was you, I’d complain
10 You and me were made for each other
28
Workbook section
Exercise V
Nonstandard varieties of English often differ from Standard English (SE)
in respect of certain aspects of their grammar. Each of the sentences below is taken
from a different (nonstandard) variety of English; identify how the sentences differ
from SE in respect of their grammar. (The SE counterparts of the relevant construc¬
tions are given in parentheses where they are not immediately obvious.)
Model answer
Sentence 1 is characteristic of a variety of English widely referred to as
African American English (= AAE). One nonstandard grammatical feature here is the
omission of are. A well-known characteristic of AAE is the omission of are and is (but
not other forms of be) in contexts where speakers of Standard English (= SE) would
29
Principles and parameters
use the contracted forms ’re and’s: hence in place of SE He’s working we find AAE He
workin’. But where SE requires the full forms are and is, they cannot be omitted in
AAE (e.g. in structures like You sure are!). A second grammatical characteristic of 1 is
the use of the form real where SE would require the form really (cf. real pretty in place
of the standard form really pretty): in SE, regular adverbs differ from the corresponding
adjectives in that the adverbs end in +ly (hence we have the adjective real as in They
are real (people), and the adverb really as in It was really great): however, in AAE,
regular adverbs do not end in +ly, and have the same form as the corresponding adjec¬
tives. Sentence 1 also shows interesting lexical (i.e. vocabulary) characteristics, in the
nonstandard ways in which the words cool, dude and man are used.
Exercise VI
Sentences such as those given below are grammatical in one variety of
Northern Irish English spoken in Belfast (as described in two interesting studies by
Alison Henry, 1992, 1995), but are ungrammatical in standard varieties of English.
Give their Standard English counterparts, and discuss the differences between
Standard English and (the relevant variety of) Belfast English.
30
Workbook section
For fuller discussion, see Henry 1995; for unfamiliar grammatical terminology, see the
glossary.
31
Principles and parameters
we presuppose that the proposition introduced by the word that is true; hence if we
then go on to deny the truth of the proposition, we are implying that the proposition is
both true and false, and thereby expressing a contradiction.
The oddity of sentence 2 seems to be largely pragmatic: pragmatics is the study of
the relation between language and the world (and is concerned e.g. with the kinds of
situations which a given sentence could be used to describe, and with the role played
by nonlinguistic knowledge in our use of sentences). Thus, whether or not you find
expressions like My goldfish thinks that. . . well formed depends on whether or not
you believe that goldfish do (or might) possess powers of thought; a sentence like 2
implies that goldfish are capable of thought, and a person who rejects sentences like 2
is in effect rejecting the implied proposition that goldfish can think (because it conflicts
with their personal beliefs about the world).
The oddity of 3 seems to be stylistic. The form whom is used only in very formal
styles of English, whereas the colloquial expression get off with is used only in very
informal styles. Hence, there is a clash of styles here, leading to stylistic incongruity.
Exercise VIII
The following sentences are all taken from various plays written by
Shakespeare (around the year 1600):
5 Whether had you rather lead mine eyes or eye your master’s heels? (Mrs
Page, Merry Wives of Windsor, Ill.ii)
6 Hath he not a son? (Duke, Two Gentlemen of Verona, II.iv)
7 What an ass art thou! I understand thee not (Speed, Two Gentlemen of
Verona, II.v)
32
Workbook section
Give the counterparts of these sentences in Modem Standard English, and describe the
differences between the sentences above and their contemporary counterparts.
(iv) Fear not you that! (Mrs Page, Merry Wives of Windsor, IV.iv)
(v) Clamber not you up to the casements! (Shylock, Merchant of Venice, II.v)
(vi) Come not thou near me! (Phebe, As You Like It, III.v)
Exercise IX (§1.7)
In the text, we claimed that the head parameter always has a uniform
setting in a given language: either it is head-initial (so that all heads precede their com¬
plements), or it is head-final (so that all heads follow their complements). However,
although this would seem to be true of many languages (English included), there are
other languages which don’t show the same uniform setting for the head parameter.
In this respect, consider the German phrases and sentences given below:
33
Principles and parameters
In relation to these sentences, make the following assumptions about their structure. In 1
and 2 muss is a verb, Hans is its subject and stolz auf seine Mutter sein is its complement;
sein is a verb and stolz auf seine Mutter is its complement; stolz is an adjective, and auf
seine Mutter is its complement; auf is a preposition and seine Mutter is its complement;
seine is a determiner, and Mutter is its complement. In 3 geht is a verb, Hans is its subject
and den Fluss entlang is its complement; entlang is a preposition (or, more precisely, a
postposition) and den Fluss is its complement; den is a determiner and Fluss is its comple¬
ment. In 4 muss is a verb, Hans is its subject and die Aufgaben losen is its complement;
losen is a verb and die Aufgaben is its complement; die is a determiner and Aufgaben is
its complement. In 5 glaube is a verb, ich is its subject and dass Hans die Aufgaben losen
muss is its complement; dass is a complementizer (i.e. a complement-clause-introducing
particle) and Hans die Aufgaben losen muss is its complement; muss is a verb, Hans is its
subject, and die Aufgaben losen is its complement; losen is a verb and die Aufgaben is its
complement; die is a determiner and Aufgaben is its complement.
Helpful hints
Look at which categories are positioned before their complements and
which after. Which words always seem to occupy a fixed position in relation to their
complements, and which occupy a variable position? In the case of categories which
sometimes go before (and sometimes after) their complements, try and establish
whether the position of the head word is determined by structural factors (i.e. the type
of structure involved) or lexical factors (i.e. the choice of head word).
Exercise X (§§1.7-1.8)
Below are examples of utterances produced by a girl called Lucy at age
24 months. Comment on whether Lucy has correctly set the three parameters discussed
in the text (the head parameter, the wh-parameter and the null subject parameter).
Discuss the significance of the relevant examples for the parameter-setting model of
acquisition.
34
Workbook section
Helpful hints
If Lucy has correctly set the wh-parameter, we should expect to find that
she systematically preposes wh-expressions and positions them sentence-initially. If she
has correctly set the head parameter, we should expect to find (for example) that she cor¬
rectly positions the complement of a verb after the verb, and the complement of a preposi¬
tion after the preposition; however, where the complement is a wh-expression, we expect
to find that the complement is moved into sentence-initial position in order to satisfy the
requirements of the wh-parameter (so that in effect the wh-parameter over-rides the
head parameter). If Lucy has correctly set the null subject parameter, we should expect
to find that she does not use null subjects. However, the picture here is complicated by the
fact that young children often produce truncated sentence structures in which the first
word of the sentence is omitted (just as in diary styles, adults truncate sentences by omit¬
ting the subject when it is the first word in a sentence: cf. Went to a party. Had a great
time. Got totally sozzled). Hence, when a child sentence has a missing subject, it is impor¬
tant to determine whether we are dealing with a null subject (i.e. whether the child has
35
Principles and parameters
mis-set the null subject parameter), or a truncated subject. Since truncation occurs only
sentence-initially (as the first word in a sentence), but null subjects can occur in any sub¬
ject position in a sentence, one way of telling the difference between the two is to see
whether children omit subjects only when they are the first word in the sentence (which
would be the result of truncation), or whether they also omit subjects in non-initial posi¬
tions in the sentence (as is the case in a genuine null-subject language like Italian). Another
way of differentiating the two is that in null subject languages we find that overt pronoun
subjects are only used for emphasis, so that in an Italian sentence like L’ho fatto io (liter¬
ally ‘It have done I’) the subject pronoun io T has a contrastive interpretation, and the
relevant sentence is paraphraseable in English as ‘It was I who did it’: by contrast, in a
non-null-subject language like English, subject pronouns are not intrinsically emphatic -
e.g. he doesn’t have a contrastive interpretation in an English diary-style sentence such as
Went to see Jim. Thought he might help). A third way of differentiating between sentences
with null and truncated subjects is that in truncation structures we sometimes find that
expressions other than subjects can be truncated (e.g. preposed complements).
In relation to the sentences in 1-22, make the following assumptions (where I use the
informal term covert subject to mean ‘understood null or truncated subject’). In 1 making
is a verb which has the subject Daddy and the complement what, in 2 want is a verb which
has a covert subject and the complement bye-byes', in 3 go is a verb which has the subject
Mummy and the complement shops; in 4 have is a verb which has the subject me and the
complement yoghurt', in 5 doing is a verb which has the subject Daddy, and its complement
is a covert counterpart of what, in 6 think is a verb with a covert subject and its complement
is Teddy sleeping (with Teddy serving as the subject of the verb sleeping)', in 7, having is
a verb which has the subject me and the complement what', in 8 no is a negative particle
which has the complement me have fish (assume that no is the kind of word which
doesn’t have a subject), and have is a verb which has the subject me and the complement
fish', in 9 gone is a verb which has the subject Daddy and the complement where', in 10
gone is a verb which has a covert subject and the complement office', in 11 want is a verb
which has a covert subject and the complement bickies; in 12 have is a verb which has
the subject Teddy and the complement what', in 13 going is a verb which has the subject
Mummy and the complement where', in 14 go is a verb which has the subject me and the
complement shops', in 15 drinking is a verb which has the subject Daddy and the comple¬
ment coffee', in 16 eating is a verb which has the subject Nana and the complement what,
in 17 want is a verb which has a covert subject and the complement choc’ate; in 18 gone
is a verb which has the subject Dolly and its complement is a covert counterpart of where',
in 19 watch is a verb which has a covert subject and the complement te’vision', in 20 have
is a verb which has the subject me and the complement more', 21 is a phrase in which the
preposition in has the complement kitchen (assume that phrases don’t have subjects); and
in 22 play is a verb which has the subject me and the complement with Daddy (and in
turn Daddy is the complement of the preposition with).
36
2
Categories and features
2.1 Overview
In the previous chapter, we saw that the structure dependence principle
determines that all grammatical operations in natural language are category-based (so
that any word-based operation will apply to whole categories of words rather than to
specific individual words). In this chapter, we provide further evidence in support of
this conclusion, and argue that a principled description of the grammar of any language
(the language chosen for illustrative purposes being Modem Standard English)
requires us to recognize that all words in the language belong to a restricted set of
grammatical categories. We look at the main categories found in English and explore
their nature, arguing that categories are composite elements, built up of sets of gram¬
matical features.
For example, by saying that words like boy, cow, hand, idea, place, team, etc. belong to
the grammatical category noun, what we are saying is that they all share certain gram¬
matical properties in common: e.g. they have a plural form (ending in the suffix +s),
they can all be premodified by the, and so forth. Likewise, by saying that words such as
see, know, like, understand, write, appear, etc. belong to the grammatical category
verb, what we imply is that they too have certain grammatical properties in common
(e.g. they can take the progressive +ing suffix, they can occur after infinitival to, etc.).
Similarly, by saying that tall, hot, narrow, old, wise, sad, etc. belong to the grammatical
category adjective, we capture the fact that they share a number of grammatical proper¬
ties in common (e.g. they can take the comparative +er suffix). In much the same way,
by saying that words like quickly, cleverly, urgently, truly, greatly, completely, etc.
37
Categories and features
There are three complications which should be pointed out, however. One is the exis¬
tence of irregular nouns like sheep which are invariable and hence have a common
singular/plural form (cf. one sheep, two sheep). A second is that some nouns have no
plural by virtue of their meaning: only those nouns (generally called count nouns)
which denote entities which can be counted have a plural form (e.g. the noun chair, cf.
one chair, two chairs, etc.); some nouns (like furniture) denote an uncountable mass
(and for this reason are called mass nouns or noncount nouns) cind so Cciririot be
pluralized - hence the ungrammatically of *one furniture, *two furnitures. A third
complication is posed by noun expressions which contain more than one noun; only
the head noun in such expressions can be pluralized, not any preceding noun used as a
dependent or modifier of the head noun: thus, in expressions such as car doors, wheel
trims, policy decisions, skate boards, horse boxes, trouser presses, coat hangers,
etc. the second noun is the head noun and can be pluralized, whereas the first noun is a
nonhead (i.e. is a modifier/dependent of some kind) and cannot be pluralized.
In much the same way, we can identify verbs by their inflectional morphology in
English. In addition to their uninflected base form (= the form under which they are
listed in dictionaries) verbs typically have up to four different inflected forms, formed
38
2.2 Morphological evidence
by adding one of four inflections to the appropriate stem form: the relevant inflections
are the past/perfective participle suffix +n, the past tense suffix +d, the third person
singular present tense suffix +5, and the present/imperfective/progressive participle
suffix +ing (see the glossary if these terms are unfamiliar), giving the range of forms
illustrated in the table in (3) below:
Like most morphological criteria, however, this one is complicated by the irregularity
of English inflectional morphology; for example, many verbs have irregular past or
perfective forms, and in some cases either or both of these forms may not in fact be
distinct from the (uninflected) base form, so that a single form may serve two or three
functions (thereby neutralizing the relevant distinctions), as the table in (4) below
illustrates:
(In fact, the largest class of verbs in English are those which have the morphological
characteristics of wait, and thus form both their past and perfective forms by suffixing
+(e)d.) The picture becomes even more complicated if we take into account the verb
be, which has eight distinct forms (viz. the base form be, the perfective form been, the
imperfective form being, the past forms was/were, and the present forms am/are/is).
The most regular verb suffix in English is +ing, which can be attached to the base form
of almost any verb (though a handful of defective verbs like beware are exceptions).
The obvious implication of our discussion of nouns and verbs here is that it would
not be possible to provide a systematic account of English inflectional morphology
unless we were to posit that words belong to grammatical categories, and that a specific
type of inflection attaches only to a specific category of word. The same is also true if
39
Categories and features
we can use an (appropriate kind of) noun, but not a verb, preposition, adjective or
adverb, as we see from (6) below:
40
2.3 Syntactic evidence
Thus, using the relevant syntactic criterion, we might define the class of nouns as the
set of words which can terminate a sentence in the position marked_in (5).
Using the same type of syntactic evidence, we could argue that only a verb (in its
uninflected infinitive/base form) can occur in the position marked_in (7) below to
form a complete (non-elliptical) sentence:
And support for this claim comes from the contrasts in (8) below:
And the only category of word which can occur after very (in the sense of extremely) is
an adjective or adverb, as we see from (9) below:
(But note that very can only be used to modify adjectives/adverbs which by virtue of
their meaning are gradable and so can be qualified by words like very!rather!some¬
what etc; adjectives/adverbs which denote an absolute state are ungradable by virtue
of their meaning, and so cannot be qualified in the same way - hence the oddity of
!Fifteen students were very present, and five were very absent, where ! marks semantic
anomaly.) Moreover, we can differentiate adjectives from adverbs in syntactic terms.
For example, only adverbs can be used to end sentences such as He treats her_, She
behaved_, He worded the statement_: cf.
And since adjectives (but not adverbs) can serve as the complement of the verb be (i.e.
can be used after be), we can delimit the class of (gradable) adjectives uniquely by
saying that only adjectives can be used to complete a four-word sentence of the form
They are very_: cf.
41
Categories and features
Another way of differentiating between an adjective like real and an adverb like really
in syntactic terms is that adjectives are used to modify nouns, whereas adverbs are used
to modify other types of expression: cf.
Adjectives used to modify a following noun (like real in There is a real crisis) are
traditionally said to be attributive in function, whereas those which do not modify an
immediately following noun (like real in The crisis is real) are said to be predicative in
function.
As for the syntactic properties of prepositions, they alone can be intensified by right
in the sense of ‘completely’, or by straight in the sense of ‘directly’:
However, there are two minor caveats which should be noted here. Firstly, since
right!straight serve to intensify the meaning of a preposition, they can only be com¬
bined with those (uses of) prepositions which express the kind of meaning which can
be intensified in the appropriate way (hence they cannot be used to intensify ungrad-
able prepositions like oflwithlfor as they are used in phrases such as *a man
right! straight of great courage, *a house right/straight with shattered windows, *a pre¬
sent right/straight for Mary - though in other uses for can be intensified, cf. He made
right/straight for the exit). A second caveat to note is that in some varieties of English
(e.g. Northern British English) right can be used to intensify adjectives, adverbs and
nouns, so that (14b—d) are grammatical with right in these varieties, though not with
straight.
A further syntactic property of some prepositions (namely those which take a fol¬
lowing (pro)nominal complement (traditionally called transitive prepositions)) which
42
2.3 Syntactic evidence
they share in common with (transitive) verbs is the fact that they permit an immedi¬
ately following objective pronoun as their complement (i.e. a pronoun in its object
form, like me/us/him/them): cf.
Even though a preposition like with does not express the kind of meaning which allows
it to be intensified by right or straight, we know it is a (transitive) preposition by virtue
of the fact that it is invariable (so not a verb) and permits an objective pronoun as its
complement, e.g. in sentences such as He argued with meluslhimlthem.
Given that different categories have different morphological and syntactic proper¬
ties, it follows that we can use the morphological and syntactic properties of a word to
determine its categorization (i.e. what category it belongs to). The morphological prop¬
erties of a given word may provide an initial rough guide to its categorial status: in
order to determine the categorial status of an individual word, we can ask whether it
has the inflectional and derivational properties of a particular category of word. For
example, we can tell that happy is an adjective by virtue of the fact that it has the
derivational properties of typical adjectives: it can take the negative prefix un+ (giving
rise to the negative adjective unhappy), the adverbializing suffix +ly (giving rise to the
adverb happily), the comparative/superlative suffixes +erl+est (giving rise to the forms
happier!happiest), and the nominalizing suffix +ness (giving rise to the noun
happiness).
However, we cannot always rely entirely on morphological clues, owing to the fact
that inflectional morphology is sometimes irregular, and derivational morphology
often has limited productivity: this means that a word belonging to a given class may
have only some of the relevant morphological properties, or even (in the case of a
completely irregular item) none of them. For example, although the adjective fat has
comparative/superlative forms in +er/+est (cf. fatlfatterlfattest), it has no negative
un+ counterpart (cf. *unfat), no adverb counterpart in +ly (cf. *fatly), and (for many
speakers) no noun counterpart in +ness (cf. *fatness): even more exceptional is the
adjective little, which has no negative un+ derivative (cf. *unlittle), no adverb +ly
derivative (cf. *littlely/*littly), no productive noun derivative in +ness (cf. the awk¬
wardness of ?littleness), and no productive +er/+est derivatives (the forms littler!
littlest are not usual - at least, for me).
What makes morphological evidence even more problematic is the fact that many
morphemes may have more than one use. For example, we noted earlier that +n and
+ing are inflections which attach to verbs to give perfective or imperfective verb forms
43
Categories and features
we find that better can be replaced by a more + adjective sequence like more fluent in
(16a) but not (16b), and conversely that better can be replaced by a more + adverb
sequence like more fluently in (16b) but not in (16a): cf.
Thus, our substitution test provides us with syntactic evidence that better is an adjec¬
tive in (16a), but an adverb in (16b).
44
2.4 Functional categories
45
Categories and features
(Quantifying determiners are determiners like all/some which denote quantity, and are
sometimes said to belong to the subcategory quantifier, referential determiners are deter¬
miners like the/this/that/my which are used to introduce referring expressions - e.g. an
expression like the car in a sentence hke Shall we take the car? is a referring expression in
the sense that it refers to a specific car whose identity is assumed to be known to the
hearer.) Since determiners are positioned in front of nouns (cf. the boys), and adjectives
can similarly be posidoned prenominally (cf. tall boys), an obvious question to ask at this
point is why we couldn’t just say that the determiners in (18) have the categorial status of
adjectives. The answer we shall give is that any attempt to analyse determiners as adjec-
dves in English runs up against a number of serious descriptive problems. Let’s see why.
One reason for not subsuming determiners within the category of adjectives is that
adjectives and determiners are syntactically distinct in a variety of ways, in respect of
their distribution. For example, adjectives can be recursively (i.e. repeatedly) stacked
in front of the noun they modify (in that you can go on putting more and more adjec¬
tives in front of a given noun), whereas determiners cannot be stacked in this way, in
that you can generally only have one determiner (of a given type - e.g. one referential
determiner and one quantificational determiner) premodifying a noun:
(19) (a) adjectives: men; handsome men; dark handsome men; tall dark
handsome men; sensitive tall dark handsome men;
intelligent sensitive tall dark handsome men, etc.
(b) determiners; the car; *a my car; *that the car; *that his the car; *a
that car, etc.
Moreover, both determiners and adjectives can be used together to modify a noun, but
when they do so, any determiner modifying the noun has to precede any adjective(s)
modifying the noun: cf. e.g.
Thus, determiners seem to form a distinct distributional class (hence belong to a differ¬
ent category) from adjectives.
A further difference between determiners and adjectives can be illustrated by what
speaker B can - and cannot - reply in the following dialogue:
46
2.4 Functional categories
As already noted, nouns like chair have the property that they are countable (in
the sense that we can say one chair, two chairs, etc.), and in this respect differ from
nouns like furniture which are uncountable (hence we cannot say * one furniture, *two
furnitures, etc.). As we see from (21), a singular count noun like chair cannot stand on
its own as a complete noun expression, nor indeed can it function as such even if
premodified by an adjective like comfortable; rather, a singular count noun requires a
premodifying determiner like a/the/another!this!my etc. This provides us with clear
evidence that determiners in English belong to a different category from adjectives.
Indeed, a more general property which differentiates determiners from adjectives is
that determiners tend to be restricted to modifying nouns which have specific
number!countability properties. For example, the determiner a modifies a singular
count noun, much modifies a (singular) mass noun, several modifies a plural count
noun, more modifies either a plural count noun or a (singular) mass noun: cf.
By contrast, typical adjectives like nice, simple, comfortable, modem, etc. can gener¬
ally be used to modify all three types of nominal: cf.
(It should be noted, however, that a handful of determiners like the can also be used to
modify singular/plural count and noncount nouns alike.)
It seems reasonable to suppose that determiners constitute a functional category
(whereas adjectives are a lexical category). After all, there is an obvious sense in which
adjectives (e.g. thoughtful) have descriptive content but determiners do not - as we can
illustrate in terms of the following contrast (? and ! are used to denote increasing
degrees of semantic/pragmatic anomaly)
As (24a) illustrates, an adjective like thoughtful can only be used to modify certain
types of noun; this is because its descriptive content is such that it is only compatible
with (for example) an expression denoting a rational entity. By contrast, determiners
like those bold-printed in (24b) lack specific descriptive content, and hence can be used
to premodify any kind of noun (the only restrictions being grammatical in nature - cf.
e.g. the fact that a(n)/another can only be used to premodify a singular count noun).
47
Categories and features
Pronouns like he/him and she/her change their morphological form according to the
position which they occupy within the sentence, so that the nominative forms he!she are
required as the subject of a present-tense verb like admires (or a past-tense verb),
whereas the objective!accusative forms him/her are required e.g. when used after (more
precisely, as the complement of) a transitive verb or preposition: these variations are
said to reflect different case forms of the pronoun. By contrast, nouns such as John and
Mary don’t overtly inflect for nominative/objective case in Modem English, and hence
don’t change their fomi according to whether they are used as subjects or complements.
Personal pronouns might be argued to be functors by virtue of the fact that they lack
descriptive content: thus, whereas a noun like dogs denotes a specific type of animal, a
personal pronoun like they denotes no specific type of entity, but has to have its refer¬
ence determined from the linguistic or nonlinguistic context. Personal pronouns simply
encode sets of person, number, gender and case properties - as represented in the
table in (26) below:
(26) case
48
2.5 Auxiliaries and infinitival to
Although most such items have the same form when used as determiners or pronouns,
there are a few which have slightly different forms according to whether they are used
as (italicized) determiners or as (bold-printed) pronouns: cf.
(28) (a) No student has failed the test/None has failed the test
(b) My house is bigger than your house/Mine is bigger than yours
Thus, no is a determiner whose pronoun counterpart is none, and similarly my/your are
determiners whose pronoun counterparts are mine/yours. Consequently, one way of
trying to decide whether a word which has dual determiner/pronoun status is being
used as a determiner or as a pronoun in a given sentence is to use the familiar substitu¬
tion test and see whether the word in question can be replaced by items like no and/or
my (which can be used as determiners but not as pronouns), or by items like none
and/or mine (which can be used as pronouns but not as determiners). If we apply this
simple substitution test to sentences such as:
(29) (a) Neither of the candidates was suitable for the post
(b) Neither candidate was suitable for the post
we find that neither in (29a) can be replaced by none and hence is a pronoun, whereas
in (29b) neither can be replaced by no and so is a determiner.
49
Categories and features
Traditional grammarians posit that there is a special class of items which once func¬
tioned simply as verbs, but in the course of the evolution of the English language have
become sufficiently distinct from other verbs that they are now regarded as belonging
to a different category of auxiliary (conventionally abbreviated to AUX). Auxiliaries
differ from other verbs in a number of ways. Whereas a typical verb like want may take
a range of different types of complement (e.g. a subjectless infinitival fo-complement
as in I want [to go home], an infinitive with a (bold-printed) subject as in I want [you to
keep quiet], or a noun expression as in / want [lots of money]), by contrast auxiliaries
typically take a verb expression as their complement, and have the semantic function
of marking grammatical properties associated with the relevant verb, such as tense,
aspect, voice, mood or modality. (See the glossary if you are not familiar with these
terms.) The items italicized in (30) below (in the use illustrated there) are traditionally
categorized as auxiliaries taking a [bracketed] verbal complement:
By contrast, typical verbs do not themselves permit inversion, but rather require what
is traditionally called do-support (i.e. have inverted forms which require the use of the
dummy auxiliary do): cf.
50
2.5 Auxiliaries and infinitival to
A second difference between auxiliaries and verbs is that auxiliaries can generally be
directly negated by a following not (which can usually contract down onto the auxil¬
iary in the form of n Y): cf.
By contrast, verbs cannot themselves be directly negated by not/n’t, but require indirect
negation through the use of do-support: cf.
(Note that in structures such as John decided not to stay the negative particle not negates
the infinitive complement to stay rather than the verb decided.) And thirdly, auxiliaries
can appear in sentence-final tags, as illustrated by the examples below (where the part
of the sentence following the comma is traditionally referred to as a tag): cf.
In contrast, verbs can’t themselves be used in tags, but rather require the use of do-
tags: cf.
So, on the basis of these (and other) syntactic properties, it seems that we are justified
in positing that auxiliaries constitute a different category from verbs.
51
Categories and features
A second type of functor which is associated with verbs is the infinitive particle
to - so called because the only kind of complement it will allow is one containing a
verb in the infinitive form. (The infinitive form of the verb in English is its uninflected
base form - the form found in dictionary entries.) Typical uses of infinitival to are illus¬
trated in (37) below:
By contrast, the dummy functor infinitival to (because of its lack of descriptive con¬
tent) cannot be intensified by right!straight: cf.
52
2.5 Auxiliaries and infinitival to
Moreover, what makes the prepositional analysis of infinitival to even more problematic
is that infinitival to takes an entirely different range of complements from prepositional
to (and indeed different from the range of complements found with other prepositions in
general). For example, prepositional to (like many other prepositions) takes a noun
expression as its complement, whereas infinitival to requires a verbal complement — as
we see from examples such as those below:
By contrast, infinitival to can only take a verbal complement when the verb is in the
uninflected base (= infinitive) form, never when it is in the gerund form: cf.
Thus, there are compelling reasons for assuming that infinitival to belongs to a differ¬
ent category from prepositional to. But what category does infinitival to belong to?
In the late 1970s, Chomsky suggested that there are significant similarities between
infinitival to and a modal auxiliary like should. For example, they occupy the same
position within the clause: cf.
53
Categories and features
We see from (45) that to and should are both positioned between the subject John and
the verb show. Moreover, just as should requires after it a verb in the infinitive form
(cf. You should show/*should showing/*should shown more interest in syntax), so too
does infinitival to (cf. Try to show/*to showing/*to shown more interest in syntax).
Furthermore, if to is analysed as an infinitival auxiliary, we can account for the fact that
infinitival to, like auxiliaries (e.g. should) but unlike typical nonauxiliary verbs (e.g.
want), allows ellipsis of its complement: cf.
(46) (a) I don’t really want to go to the dentist’s, but I know I should
(b) I know I should go to the dentist’s, but I just don’t want to
(c) *1 know I should go to the dentist’s, but I just don’t want
The fact that to patterns like the auxiliary should in several respects strengthens the
case for regarding them as belonging to the same category. But what category?
Chomsky (1981, p. 18) suggested that the resulting category (comprising finite aux¬
iliaries and infinitival to) be labelled INFL or inflection, though (in accordance with
the standard practice of using single-letter symbols to designate categories) in later
work (1986b, p. 3) he replaced INFL by the single-letter symbol I. The general idea
behind this label is that finite auxiliaries inflect for tense/agreement, and infinitival to
serves much the same function in English as infinitive inflections in languages like
Italian which have overtly inflected infinitives (so that Italian canta+re = English to
sing). We can then say (for example) that an auxiliary like should is a finite I/INFL,
whereas the particle to is an infinitival I/INFL.
2.6 Complementizers
The last type of functional category which we shall look at is that of complementizer
(abbreviated to COMP in earlier work and to C in more recent work): this is a term
used to describe a special kind of (italicized) word which is used to introduce comple¬
ment clauses such as those bracketed below:
Each of the bracketed clauses in (47) is a complement clause, in that it functions as the
complement of the word immediately preceding it {think!doubt!anxious)-, the italicized
word which introduces each clause is known in recent work (since 1970) as a complemen¬
tizer (but would be known in more traditional work as a particular type of subordinating
conjunction). Complementizers are functors in the sense that they encode particular sets of
grammatical properties. For example, complementizers encode (non)finiteness by virtue
of the fact that they are intrinsically finite or nonfinite (see the glossary if these terms are
unfamiliar). Thus, the complementizers that and if are inherently finite in the sense that
54
2.6 Complementizers
they can only be used to introduce a finite clause (i.e. a clause containing a present- or
past-tense auxiliary or verb), and not e.g. an infinitival fo-clause; by contrast, for is an
inherently infinitival complementizer, and so can be used to introduce a clause containing
infinitival to, but not a finite clause containing a tensed (i.e. present/past-tense) auxiliary
like should; compare the examples in (47) above with those in (48) below:
Complementizers in structures like (47) serve three grammatical functions: firstly, they
mark the fact that the clause they introduce is the complement of some other word
(think/doubt!anxious)-, secondly, they serve to indicate whether the clause they introduce
is finite (i.e. contains a present/past-tense verb/auxiliary) or infinitival (i.e. contains
infinitival to); and thirdly, they mark the illocutionary force (i.e. semantic/pragmatic
function) of the clause they introduce (thus, if introduces an interrogative clause,
whereas that/for introduce other types of clause: e.g. that typically introduces a declara¬
tive/statement-making clause).
However, an important question to ask is whether we really need to assign words such
as forlthatlif (in the relevant function) to a new category of complementizer, or whether
we couldn't simply treat (for example)/or as a preposition, that as a determiner and if as
an adverb. The answer is ‘No’, because there are significant differences between comple¬
mentizers and other apparently similar words. For example, one difference between the
complementizer for and the preposition for is that the preposition for has intrinsic seman¬
tic content and so (in some but not all of its uses) can be intensified by straight/right,
whereas the complementizer for is a dummy functor and can never be so intensified: cf.
Moreover, the preposition for and the complementizer for also differ in their syntactic
behaviour. For example, a clause introduced by the complementizer for can be the sub¬
ject of an expression like would be unthinkable, whereas a phrase introduced by the
preposition for cannot: cf.
(50) (a) For you to go there on your own would be unthinkable (- for-clause)
(b) *For you would be unthinkable (=/or-phrase)
What makes it even more implausible to analyse infinitival for as a preposition is the
fact that prepositions in English aren’t generally followed by a [bracketed] infinitive
complement, as we see from the ungrammaticality of:
55
Categories and features
On the contrary, as examples such as (42) above illustrate, the only verbal comple¬
ments which can be used after prepositions are gerund structures containing a verb in
the +ing form.
A further difference between the two types of for is that if we replace a noun expression
following the preposition for by an appropriate interrogative expression like who?/what?/
which one?, the interrogative expression can be preposed to the front of the sentence (with
or without for) if for is a preposition, but not if for is a complementizer. For example, in
(52) below, for functions as a preposition and the (distinguished) nominal Senator
Megabucks functions as its complement, so that if we replace Senator Megabucks by which
senator?, the wh-expression can be preposed with or without for. cf.
However, in (53a) below, the bold-printed expression is not the complement of the
complementizer for (the complement of for here is the infinitival clause Senator
Megabucks to keep his cool), but rather is the subject of the expression to keep his
cool: hence, even if we replace Senator Megabucks by the interrogative wh-phrase
which senator?, the wh-phrase can’t be preposed:
(53) (a) They were anxious for Senator Megabucks to keep his cool
(b) *Which senator were they anxious for to keep his cool?
(c) *For which senator were they anxious to keep his cool?
56
2.6 Complementizers
Consider now the question of whether the complementizer that can be analysed as a
determmer. At first sight, it might seem as if such an analysis would provide a natural
way of capturing the apparent parallelism between the two uses of that in sentences
such as the following:
Given that the word that has the status of a prenominal determiner in sentences such as
(56a), we might suppose that it has the function of a preclausal determiner (i.e. a deter¬
miner introducing the italicized clause Randy Rabbit runs Benny’s Bunny Bar) in
sentences such as (56b).
However, there is strong empirical evidence against a determiner analysis of the
complementizer that. Part of the evidence is phonological in nature. In its use as a
complementizer (in sentences such as (56b) above), that typically has the vowel-
reduced form /Sot/, whereas in its use as a determiner (e.g. in sentences such as (56a)
above), that invariably has the unreduced form /Saet/: the phonological differences
between the two suggest that we are dealing with two different items here, one of
which functions as a complementizer and typically has a reduced vowel, and the other
of which functions as a determiner and always has an unreduced vowel.
Moreover, that in its use as a determiner (though not in its use as a complementizer)
can be substituted by another determiner (such as this/the):
Similarly, the determiner that can be used pronominally (without any complement),
whereas the complementizer that cannot: cf.
(58) (a) Nobody can blame you for that mistake (prenominal determiner)
(b) Nobody can blame you for that (pronominal determiner)
(59) (a) I’m sure that you are right (preclausal complementizer)
(b) *I’m sure that (pronominal complementizer)
The clear phonological and syntactic differences between the two uses of that argue
strongly that the particle that which serves to introduce complement clauses should not
be analysed as a determiner, but rather as a complementizer (hence assigned to the cat¬
egory C of complementizer).
The third item which we earlier suggested might function as a complementizer
in English is interrogative if However, at first sight, it might seem that there is a
potential parallelism between the use of if and interrogative wh-adverbs like
when/where/whether: cf.
57
Categories and features
Finally, whereas a wh-adverb can typically be coordinated with (e.g. joined by or to)
another similar adverb, this is not true of if: cf.
For reasons such as these, then, it seems more appropriate to categorize if as an interroga¬
tive complementizer and whether/where/when as interrogative adverbs. More generally,
our discussion highlights the need to posit an additional category C of complementizer,
to designate clause-introducing items such as if!that!for which serve the function of intro¬
ducing specific types of finite or infinitival clause.
2.7 Parsing
Having given an outline of the major lexical and functional categories
found in English, we are now in a position where we can start to parse (i.e. analyse the
grammatical structure of) phrases and sentences. The first step in parsing any expres¬
sion is to categorize each of the words in the expression. A conventional way of doing
this is to use the traditional system of labelled bracketing: each word is enclosed in a
pair of square brackets, and the lefthand member of each pair of brackets is given an
appropriate subscript category label to indicate what category the word belongs to. To
save space, it is conventional to use the following (bold-printed) capital-letter abbrevi¬
ations to represent categories:
58
2.8 Subcategorialfeatures
Adopting this notation, we can represent the categorial status of each of the words in a
sentence such as Any experienced journalist knows that he can sometimes manage to
lure the unsuspecting politician into a cunning trap as in (65) below:
What (65) tells us is that the words journalist/politician/trap belong to the category
N (= noun), he to the category PRN (= pronoun), anyithela to the category D (= deter¬
miner), experienced!unsuspecting!cunning to the category A (= adjective), sometimes
to the category ADV (= adverb), into to the category P (= preposition),
knows/manage/lure to the category V (= verb), can/to to the category I/INFL (since
can is a finite present-tense auxiliary and to an infinitive particle) and that to the
category C (= complementizer). It is important to note, however, that the category
labels used in (65) tell us only how the relevant words are being used in this particular
sentence. For example, the N label on trap in (65) tells us that the item in question
functions as a noun in this particular position in this particular sentence, but tells us
nothing about the function it may have in other sentences. So, for example, in a
sentence such as Greed can trap careless politicians, the word trap functions as a verb
- as represented in (66) below:
Thus, a labelled bracket round a particular word is used to indicate the grammatical
category which the word belongs to in the particular position which it occupies in the
phrase or sentence in question, so allowing for the possibility that the same word may
have a different categorial status in other positions in other structures.
59
Categories and features
to the same category share in common, it provides us with no way of describing impor¬
tant grammatical differences between words belonging to the same category. For exam¬
ple, traditional grammars draw a distinction between two different subclasses of noun,
traditionally called common nouns and proper nouns. The essential difference
between the two is that proper nouns typically denote names of people (e.g. Chomsky),
places (e.g. Cambridge), dates (e.g. Tuesday) or magazines (e.g. Cosmopolitan), and
the first letter of a proper noun is generally capitalized; from a semantic viewpoint,
proper nouns have the property of having unique reference. In terms of their syntactic
properties, what differentiates proper nouns from common nouns is that common
nouns can freely be modified by determiners like the, whereas proper nouns (perhaps
because of their unique reference) generally cannot - cf. contrasts such as:
60
2.8 Subcat egorial features
(1965, p. 82) suggests that the distinction should be handled in terms of a grammatical
feature such as [±count] - i.e. countable/uncountable. We could then say that a count
noun like chair and a mass noun like furniture have in common their categorial status
as N words, but differ in respect of the grammatical feature [±count] in that countable
nouns like chair are [+count], whereas mass nouns like furniture are [-count].
We have noted that one of the characteristic properties of regular count nouns is that
they have both singular and plural forms (cf. chair/chairs). However, our existing
model of grammar provides no way of representing the difference between singular
and plural nouns. Clearly, any proposal to handle the distinction in categorial terms
(e.g. by assigning singular nouns to the category SG, and plural nouns to the entirely
distinct category PL) would fail to capture the essential nominal properties which the
two types of noun share. Hence, it would seem preferable to handle number in terms of
a grammatical feature such as [±plural] (i.e. plural/nonplural), so that a plural noun
like scissors would carry the feature [+plural], whereas an inherently singular noun
like news would be specified as [-plural]; by contrast, the lexical entry for a noun like
book (which has both the singular form book and the plural form books) would assign
it the number feature [±plural] .
Generalizing our discussion at this point, we might suppose that the categorial prop¬
erties of words are specified by assigning each word to an appropriate category, and
that their subcategorial properties are specified in terms of a set of grammatical fea¬
tures. So, for example, the (relevant parts of) the lexical entries for news and scissors
might be as in (68) below:
The lexical entries in (68) say that news is a noncount common noun which is inher¬
ently singular (so has no plural form: cf. The news is/*are bad), and that scissors is a
noncount common noun which is inherently plural (so has no singular form: cf. These
scissors are blunt/*This scissor(s) is blunt). But note that two different technical
devices are used to describe the relevant properties: thus, the nounhood properties of
the two items are described in terms of the category label N, whereas their common¬
ness, countability and number properties are described in terms of sets of grammatical
features.
We might extend the use of features to handle the relation between quantifiers and
determiners. In our discussion of the examples in (18) above, we suggested that quanti¬
fiers are a subclass of determiner. One way of handling the relation between the two
would be in terms of a feature such as [±Q] (i.e. quantifying/nonquantifying), so that a
determiner such as many would carry the feature [+Q], whereas a referential determiner
such as this would carry the feature [-Q]. Similarly, we could use features as a way of
capturing the traditional insight that adverbs are a subclass of adjectives, typically
61
Categories and features
derived from adjectives by the addition of +ly, so that from the adjective quiet we form
the adverb quietly by the addition of the suffix +ly. Some evidence that adverbs are a
subclass of adjectives comes from the fact that the comparative form of an adverb like
quickly is derived by adding the comparative suffix +er to the corresponding adjective
stem quick (cf. He runs quicker than me, where quicker is an adverb paraphraseable as
‘more quickly’). Moreover, in some nonstandard varieties of English, we find adjective
forms used where Standard English requires an adverb form carrying the suffix +ly (e.g.
in nonstandard sentences such as Tex talks real slow, corresponding to Standard English
Tex talks really slowly). One way of capturing the inter-relation between adjectives and
adverbs would be to say that they share the categorial properties of adjectives, but differ
in respect of a subcategorial feature such as [±ADV], with adverbs like quickly being
[+ADV] and hence adverbial, and adjectives like quick being [-ADV] and hence nonad-
verbial.
Chomsky (1965) argues that we can achieve a more unitary theory of syntax if we
suppose that all grammatical properties of words are encoded as grammatical features.
If so, then category labels such as N will be replaced by categorial features such as
[+N], and the (relevant parts of the) lexical entries in (68) will be replaced by the sets
of grammatical features in (69) below:
Of course, the wider implication of our discussion here is that the categorial and
subcategorial properties of other categories of word can similarly be described in terms
of a feature matrix (i.e. a set of features). We can illustrate this by a brief discussion of
how we might differentiate the (various different uses) of verb forms such as
show!showing!shown!showed!shows. (The set of features used here is based loosely on
that proposed in Halle and Marantz 1993.)
Consider first of all how we might differentiate those (participle) verb forms which
are overtly inflected for aspect (shown!showing) from those which are not. An obvious
suggestion would be to handle this difference in terms of a feature such as [±aspect] or
alternatively [±participle], with participles like shown!showing being positively speci¬
fied for the relevant feature, and nonparticipial forms like show!shows!showed being
negatively specified. In order to differentiate past-tense finite verbs from the corre¬
sponding present-tense forms, and to differentiate past (perfective) participles from
present (imperfective) participles, we might make use of a feature such as [±past]. And
in order to differentiate between those third person singular present-tense forms like
shows which are overtly inflected for agreement from those like show which are not,
we might posit a feature such as [+AGR] - or perhaps [±3SG], The various uses of the
italicized verb forms in the examples below will then have the tense/aspect/agreement
feature specifications indicated:
62
2.9 Cross-categorialfeatures
Correspondingly, we can say that different inflectional affixes encode different sets of
tense/aspect/agreement features. For example, the +5 affix encodes the properties third
person singular present tense, and the affix +n encodes the properties past participle.
More extensive similarities between the two are found in Jamaican Creole, as we see
from the examples below (from Bailey 1966):
63
Categories and features
The examples in (72) show that both a verb like bait ‘bite’ and an adjective like swiit
‘sweet’ (but not a preposition or noun) can function as a predicate in an independent
sentence; those in (73) that both a verb like kom ‘come’ and an adjective like gud
‘good’ can be negated by no\ and those in (74) that both a verb like dringk ‘drink’ and
an adjective like sowa ‘sour’ can be the complement of a modal like kyaan ‘can’t’.
Similarly, nouns and adjectives share certain grammatical properties in common:
e.g. in Russian, nouns and adjectives inflect for grammatical case, but not verbs or
prepositions: cf.
Thus, the nouns and adjectives in (75) carry (italicized) case endings (+a is a nominative
case suffix, +u is an accusative/objective case suffix), but not the verb or preposition.
How can we account for the fact that adjectives seem to share properties in common with
verbs on the one hand, and with nouns on the other?
In work dating back to Chomsky 1970, Chomsky suggested that we can account for
cross-categorial properties such as these by analysing categories as composites of
binary grammatical features (with each feature value representing a set of shared gram¬
matical properties). More specifically, he suggested that the four primary lexical
categories noun, adjective, verb and preposition can be analysed as complexes of just
two binary grammatical features, namely [±N] (nominal/non-nominal) and [±V]
(verbal/non-verbal), and decomposed into feature matrices (= sets of features) in the
manner indicated in (76) below:
Within the feature-based analysis of categories outlined in (76) above, the fact that
verbs and adjectives form a natural class (as we see from the fact that both allow un-
prefixation) is not an accidental property, but rather follows from the assumption that
they form a supercategory, by virtue of sharing the feature [+V] (though of course they
differ in respect of the feature [±N]). We could then specify in relation to the paradigm
in (71) above that only [+V] words allow tm-prefixation, so that words belonging to the
64
2.9 Cross-categorialfeatures
[+V] categories verb!adjective allow an un+ prefix, but those belonging to the [-V]
categories noun!preposition do not; likewise, in relation to the examples in (72-4), we
could say that only [+V] categories in Jamaican Creole can function as predicates, can
be negated by no and can occur as the complement of a modal. This would mean that
the feature [±V] serves to cross-classify categories (in that it is a feature which cuts
across traditional category boundaries). Likewise, the feature [±N] serves to cross-
classify categories in a similar way, since it implies that the [+N] categories noun and
adjective share certain properties in common (and hence in effect form a
supercategory) which differentiate them from the [-N] categories verb and preposition.
Given the feature analysis in (76) we could account for the data in (75) by positing that
the property of inflecting for case in Russian is associated with the supercategory of
[+N] constituents (viz. adjectives and nouns).
We can exploit the use of grammatical features to provide an interesting account of
the inter-relations between lexical and functional categories. Each functional category
seems to be closely related to a corresponding lexical category: auxiliaries appear to be
related to verbs, pronouns to nouns, determiners to adjectives, and the complementizer
for and the infinitive particle to to the corresponding prepositions.
To make our discussion more concrete, let’s explore the relation between auxiliaries
and verbs. In §2.5, we argued that auxiliaries and verbs should be assigned to the two
separate categories auxiliary (= AUX) and verb (= V), on the grounds that the two
show systematic syntactic differences (e.g. with respect to their behaviour in negatives,
interrogatives and tags). However, any proposal that AUX and V should be analysed as
two different categories predicts that auxiliaries should be utterly distinct from verbs in
their morphosyntactic characteristics. And yet, this is not the case at all: on the
contrary, auxiliaries behave just like verbs in some respects - and this fact is implicitly
recognized in their classification as auxiliary verbs in traditional grammar. One such
common property which is shared by verbs and auxiliaries (illustrated in the table in
(77) below) is that both have potentially distinct past/present-tense forms:
Within a feature-based theory of syntax, a natural way of capturing the fact that auxil¬
iaries are similar to verbs in some respects (e.g. in inflecting for tense) but different in
others (e.g. in undergoing inversion) is to suppose that auxiliaries and verbs share the
core features [+V, -N], but differ in respect of some other feature. Bearing in mind that
65
Categories and features
auxiliaries axe functors and verbs are not (verbs are contentives), one way of capturing
the differences between them would be in terms of a functionality feature [±F]: auxil¬
iaries would be [+F] and hence functors, and verbs would be [-F] and hence
contentives. Given these twin assumptions, the respective feature specifications of
verbs and auxiliaries would be as in (78) below:
It would then follow that pronouns are functional nominals whereas nouns are lexical
nominals, determiners are functional adjectives whereas descriptive adjectives are lexi¬
cal adjectives, auxiliaries are functional verbs whereas nonauxiliary verbs are lexical
verbs, and that particles (e.g. the infinitive particle to and the prepositional comple¬
mentizer/or) are functional prepositions whereas contentive prepositions are lexical
prepositions.
A further relationship which is captured within the analysis in (79) is that between
infinitival to and auxiliaries. (Recall from §2.5 that these are different exponents of the
category INFL.) Under the analysis in (79), infinitival to has the feature specification
[-V, -N, +F], whereas auxiliaries carry the feature specification [+V, -N, +F]: this
means that the two share in common the features [-N, +F] (hence belong to the same
category INFL), but differ in respect of the feature [±V], in that auxiliaries are [+V]
66
2.9 Cross-categorialfeatures
(and hence carry verbal tense/agreement inflections), whereas infinitival to is [—V] (and
so is an uninflected tenseless and agreementless particle).
Our discussion here provides a clear illustration of how a feature-based theory
achieves a higher level of descriptive adequacy, in terms of capturing important
descriptive generalizations about relations between categories (e.g. between auxiliaries
and verbs). We have argued that the categorial and subcategorial properties of any
given word can be described in terms of a matrix of categorial features [±Fp ±F2, ±F3,
±F4 . . . ±FJ which defines its feature specification (and hence its categorial properties).
This being so, we can redefine the notion of grammatical category as follows:
It then follows that traditional capital-letter category labels are nothing more than
abbreviations for sets of features (in much the same way as a chemical symbol like H is
simply a convenient abbreviation for the atomic properties of hydrogen).
An interesting (though not immediately obvious) corollary of the feature-based
definition of categories given in (80) is that there is a variety of different category
labels which we might attach to a given item on the basis of its feature specification,
depending on how fine- or coarse-grained a system of category labels we choose: an
analysis based on a small number of shared feature values will give us a relatively
coarse categorization, whereas an analysis based on a much larger number of shared
feature values will give us a much finer categorization. We can illustrate this point by
considering what kind of category label we might give to have in its use as a perfec¬
tive auxiliary (e.g. in sentences like They have gone home, where have marks the
perfection - in the sense of ‘completion’ - of the activity of going home). Let’s
assume that perfective have is specified as [+V, -N, +F, +PERF], where the feature
[+PERF] marks perfectivity. If we adopt a relatively coarse system of category labels
based on the two features [±V] and [±N], then perfective have is labelled as a verb
(= V) by virtue of being [+V, -N]. If we adopt a rather less coarse system of category
labels based on the three features [±V], [±N] and [±F], then perfective have is catego¬
rized as an auxiliary verb (= AUX) by virtue of being [+V, -N, +F], But if we adopt a
much more finely differentiated set of category labels based on the four features [±V,
±N, ±F, ±PERF], perfective have will be assigned the category label PERF, by virtue
of being [+V, -N, +F, +PERF]. Thus, depending how coarse- or fine-grained a system
of category labelling we use, perfective have can variously be labelled as V, AUX or
PERF. Of course, within a category-based theory which regards categories as primi¬
tive (irreducible) elements, this would lead to an obvious contradiction: however,
there is no contradiction within a feature-based theory if we adopt the definition of
category given in (80) above, since each different category symbol (viz. V, AUX,
PERF) simply picks out a different subset of the grammatical features associated with
67
Categories and features
perfective have. For analogous reasons, the so-called progressive auxiliary be (as in
He may be working) could variously be labelled as V, AUX or PROG; and the passive
auxiliary be (as in He may be arrested) could variously be labelled as V, AUX or
PASS. And, in the same way, a quantifier such as many could be labelled either as
D (by virtue of being a determiner) or as Q (by virtue of being a quantifying deter¬
miner).
Since the feature-based analyses outlined in the last two sections are rather more
abstract than the purely categorial analyses in earlier sections, you may feel (if you are
a beginner) that you don’t have a good grasp of feature-based analyses at this stage. Since
the discussion in the next two chapters will be category-based, this doesn’t matter for
the time being. We return to consider grammatical features in chapter 5, and it may be a
good idea for you to reread sections §2.8 and §2.9 in conjunction with chapter 5.
2.10 Summary
In this chapter, we have looked at the nature of grammatical categories. In
§2.2 we defined a category as a class of expressions which share a common set of
grammatical properties; we argued that inflectional and derivational morphology
provide us with strong empirical evidence for categorizing words, in that certain types
of inflectional or derivational affix attach only to certain categories of word. In §2.3 we
argued that there is also syntactic evidence for categorization, in that different cate¬
gories of word occur in a different range of positions within the phrase or sentence. We
suggested that we can determine the categorial status of a word from its morphological
and syntactic properties, with substitution being used as a test in problematic cases. For
example, verbs (= V) have the morphological property that they can take a range of
inflectional suffixes (+s/+d/+n/+ing), and have the syntactic property that they can be
used as the complement of a word like can; nouns (= N) have the morphological
property that they typically inflect for number (cf. cat/cats), and the syntactic property
that they can be preceded by a/the\ adjectives (= A) have the morphological property
that they have +ly or +ness derivatives, and the distributional property that they can be
modified by words like very and occur after the verb be\ adverbs have the morphologi¬
cal property that they end in +ly, and the distributional property that they can follow a
verb like behave; and prepositions have the morphological property that they are
invariable, and the distributional property that they can be modified by right!straight.
In §2.4 we suggested that verbs, nouns, adjectives, adverbs and prepositions are lexical
categories, in that words belonging to these categories typically have descriptive con¬
tent. We noted that English also has a number of functional categories, whose mem¬
bers lack descriptive content and serve to mark grammatical properties (e.g. number,
person, tense, etc.). We argued that determiners (= D) constitute a functional category,
and that they differ from their lexical (adjective) counterparts in that they precede
(but don’t follow) adjectives, they can’t be stacked and they impose grammatical
68
2.10 Summary
restrictions on the types of expression they can modify (e.g. a can only modify a singu¬
lar count noun expression). We went on to look at the functional counterparts of nouns,
namely pronouns (= PRN); we argued that these differ from nouns in that they have no
descriptive content and simply serve to encode functional properties like person, num¬
ber, gender and case. In §2.5 we looked at the functional counterparts of verbs, namely
auxiliaries', we argued that these are functors in that (unlike verbs) they describe no
specific action or event, but rather encode verb-related grammatical properties such as
tense, mood and aspect; we noted that auxiliaries are syntactically distinct from verbs
in that (for example) they undergo inversion. We showed that infinitival to is distinct
from the preposition to, and shares a number of properties in common with finite auxil¬
iaries (e.g. auxiliaries and infinitival to allow ellipsis of their complements, but prepo¬
sitional to does not): we noted Chomsky’s suggestion that finite auxiliaries and infiniti¬
val to are different exponents of the same category I (or INFL). In §2.6 we argued that
complementizers (= C or COMP) like thatliflfor form a further category of functors
which mark the illocutionary force of a complement clause (e.g. indicate whether it is a
statement or question), and that (for example) if is, distinct from interrogative adverbs
like how/when/whether in that it can only introduce a finite clause, and cannot intro¬
duce a clause which is used as the complement of a preposition. In §2.7 we showed
how the labelled bracketing technique can be used to parse the words in a sentence. In
§2.8 we argued that categories are not primitive elements, but rather are composites of
grammatical features. We noted that in a category-based theory of grammar, we have
no straightforward way of accounting for subcategorial properties (e.g. the fact that
count nouns and mass nouns differ in certain respects), but that in a feature-based
grammar we can handle these differences in terms of a feature such as [±count]. In
§2.9 we suggested that cross-categorial properties (i.e. properties which are shared by
members of two or more different categories) can similarly be handled in a feature-
based grammar: for instance, if we suppose that the four primary categories (noun,
verb, adjective and preposition) are composites of two primitive features [±V] and
[±N], so that nouns are [+N, -V], adjectives are [+N, +V], verbs are [-N, +V] and
prepositions are [-N, -V], we can then say that only items which are specified as [+V]
allow im-prefixation. We further suggested that each major class of contentive category
has a functional counterpart, so that auxiliaries are the functional counterparts of verbs,
determiners of adjectives, pronouns of nouns, and particles (e.g. complementizer for
and infinitival to) of prepositions; and we suggested that the similarities between func¬
tors and their contentive counterparts are a consequence of their having shared values
for certain features, and that the differences between them could be handled in terms of
a functionality feature such as [±F], with functors being [+F] and contentives [—F]. We
noted that within a feature-based theory, a grammatical category is no longer a primi¬
tive construct, but rather is defined as a set of elements which have the same value(s)
for a given set of grammatical features. We pointed out that one consequence of this is
69
Categories and features
that there can be a variety of ways of categorizing a given item on the basis of its fea¬
ture specification, depending on how fine- or coarse-grained a categorization we
choose: thus, perfective have can be variously labelled as a verb (= V) by virtue of
being [+V, -N], or an auxiliary verb (= AUX) by virtue of being [+V, -N, +F], or a
perfective auxiliary verb (= PERF) by virtue of being [+V, -N, +F, +PERF].
Workbook section
Exercise I (§§2.2-2.3)
Each of the words below is polycategorial, in that it is a member of more
than one different grammatical category. What are the different categories that each
word can belong to? Provide a different example sentence to illustrate each different
use of each word, and give reasons in support of your proposed categorization.
Exercise II (§§2.2-2.3)
Below are some examples of Shakespearean English (dating back to
around the year 1600) containing words which are no longer used (or which have a
different use) in contemporary English:
70
Workbook section
7 What will you adventure to save this brat’s life? (Leontes, Winter’s Tale,
11.111)
8 He is not like to marry me well (Touchstone, As You Like It, DLiv)
9 I would fain prove so (Polonius, Hamlet, II.i)
10 An old black ram is tupping your white ewe (Iago, Othello, I.i)
11 I had as lief be wooed of a snail (Rosalind, As You Like It, IV.i)
12 I'll pheeze you (Sly, Taming of the Shrew, Induction)
13 I am no breeching scholar in the schools (Bianca, Taming of the Shrew,
Ill.i)
14 He that ears my land spares my team (Clown, All’s Well That Ends Well,
1.111)
15 Let thy tongue tang arguments of state (Malvolio, Twelfth Night, II.v)
Attempt to categorize each of the italicized words (saying how you arrived at your
analysis), and (if they are still in use in contemporary English), compare and contrast
their use in Shapespearean English with their use in contemporary English.
71
Categories and features
prevents the generation of a +ly adverb for adjectives which already have an adverb
counterpart listed in the lexicon (so that the existence of far an adverb prevents the
generation of *farly). A third class of adverbs are those like soon (cf. He ll arrive soon)
which have no adjective counterpart (cf. *his soon arrival): we might suppose that
these are listed in the lexicon as intrinsically adverbial.
For each of the words below (and any others which you care to consider), devise
examples of your own to illustrate whether or not the word in question can function as
an adjective (and if so whether it has an immediate adverb counterpart) or an adverb
(and if so whether it has an immediate adjective counterpart). Say whether any of the
relevant words pose problems for the tripartite classification of adverbs suggested here
- and if so, which and why.
Helpful hints
One problem which arises is how to deal with adjectives which have no
+ly counterpart. An ad hoc (lexical) solution to this problem would be simply to indicate
in the lexical entry of the relevant words that they are exceptions to the +ly formation
rule. However, given that linguistics is concerned with explanation, a more principled
approach would be to see whether we can explain exceptions in phonological terms (e.g.
in terms of avoiding certain sound sequences), or in morphological terms (e.g. avoiding
certain sequences of suffixes), or semantic terms (only words with a certain meaning
allow +ly suffixation), or in etymological terms (e.g. only words of Anglo-Saxon or
Latinate origin permit +ly suffixation), or in terms of blocking (e.g. +ly adverb forma¬
tion is blocked when the language already contains a related lexical item with the same
adverbial function and meaning). A second problem which arises is how to deal with +ly
adverbs which appear to be derived from a word other than an adjective. A third prob¬
lem relates to the fact that some words have more than one meaning/use, and may
behave differently in different uses: you may wonder whether to deal with each different
use separately (i.e. by having a separate lexical entry for each use).
Exercise IV (§2.5)
On the basis of their behaviour in negatives, interrogatives and tags,
determine whether each of the italicized items below (in each of its uses) can have the
categorial status of an auxiliary and/or a verb (bearing in mind the possibility that some
items in some uses may be polycategorial, and so may have dual auxiliary/verb status).
72
Workbook section
Give examples to illustrate the ways in which each item can and cannot be used, and
discuss any problematic cases. The italicized labels in parentheses are provided purely
as a way of helping you differentiate between different uses of a given item.
Helpful hint
When followed by a bare (fo-less) infinitival complement, need and dare
are restricted to occurring in a negative or interrogative structure - hence e.g. we don’t
say *He dare/need do it, though we do have I don't think he dare/need do anything and
I wonder whether he dare/need do anything.
73
Categories and features
Thus, transitive get patterns in all the relevant respects like a typical nonauxiliary verb,
and thus belongs to the category V/verb.
Exercise V (§§2.5-2.6)
Discuss the categorization of the italicized words in each of the following
examples, giving empirical arguments in support of your analysis.
(i) (a) For parents, it is important to spend time with their children
(b) For how many parents is it important to spend time with their children?
The alternative possibility that for might be used as a complementizer (with the infini¬
tival clause parents to spend time with their children serving as its complement) is sug¬
gested by the fact that the/or-clause here could be substituted by a that-c\mse, as in:
(ii) It is important that parents should spend time with their children
74
Workbook section
Executives like to drive straight to work, but not * Executives like straight to drive to
work). Moreover, the second to is a contentive preposition which has the antonym from
(cf. Executives like to drive from work), whereas the first has no obvious antonym since
it is a dummy infinitive particle (cf. * Executives like from drive/driving to work). In
addition, like a typical transitive preposition, the second to (but not the first) can be fol¬
lowed by an objective pronoun like them — cf. Executives think the only way of getting
to their offices is to drive to them. Conversely, the first (infinitival) to allows ellipsis of
its complement (cf. Executives like to), whereas the second (prepositional) to does not
(cf. *Executives like to drive to). Thus, in all relevant respects the first to behaves like a
dummy infinitive particle, whereas the second to behaves like a contentive preposition.
Exercise VI (§§2.2-2.6)
Some linguists have argued that the word of has two different uses, illus¬
trated in 1 and 2 below:
It has been suggested that of in expressions such as 1 belongs to the category P of prepo¬
sitions, and is a contentive which is often paraphraseable by with (and hence which has
the antonym without: cf. items with/without sentimental value). In this (contentive) use,
an o/-phrase can be modified by an appropriate adverb (cf. items purely of sentimental
value), and can be (noncontrastively) negated by not- cf. items not of much sentimental
value. By contrast, it is suggested that of in structures like 2 is a dummy functor belong¬
ing to the category K of case particle, and that its function is to link a noun (e.g. criti¬
cism) which cannot directly take a nominal or pronominal complement (cf *criticism
[the press/me]) to a following nominal or pronominal complement (cf. criticism of [the
press/me]). Of is required in structures like 2 because nominal/pronominal expressions
like the press!me can only serve as the complement of a transitive item, and of is transi¬
tive whereas nouns are intransitive. One reason for supposing that of in structures like
2 is a meaningless functor is that of is not used in corresponding verbal constructions
like He criticizes the press, suggesting that it has no intrinsic meaning. Moreover,
o/-phrases like those in 2 are not paraphraseable by wfi/z-phrases, cannot be modified by
adverbs (cf. *his criticism truly of the press), have no antonyms (e.g. whatever his
criticism without the press may mean, it is not the opposite of his criticism of the press)
and cannot be noncontrastively negated by not (cf. *his criticism not of the press: this is
only grammatical if we use contrastive negation, and continue with e.g. but of the
cinema). On the basis of considerations such as these, determine how of should be cate¬
gorized in each of the following:
75
Categories and features
76
Workbook section
9 Linguists don’t really know the extent to which peer group pressure
shapes linguistic behaviour patterns in very young children
10 You don’t seem to be too worried about the possibility that many of the
shareholders may now vote against your revised takeover bid
An issue of particular interest which arises here relates to the status of the words aver¬
age and morphology. Are these nouns or adjectives - and how can we tell? Since nouns
used to modify other nouns are invariable in English (e.g. we say skate boards, not
*skates boards), we can’t rely on morphological clues here. However, we can use
syntactic evidence. If (as we claim), the word average functions as an adjective in 1,
we should expect to find that it can be modified by an adverb like relatively which can
be used to modify adjectives (cf. relatively good)', by contrast, if morphology serves as
a noun in 1, we should expect to find that it can be modified by the kind of adjective
(e.g. inflectional) which can be used to modify such a noun. In the event, both predic¬
tions are correct, as we see from (i) below:
Some additional evidence that average can function as an adjective comes from the
fact that it has the +/y adverb derivative averagely, and (for some speakers at least) the
noun derivative averageness (cf. The very averageness of his intellect made him a
natural choice for prime minister).
Helpful hint
In structures such as morphology exercises, you will not always find it
easy to determine whether the first word (in this case, morphology) is a noun or an
adjective. As a rule of thumb, where the item concerned is clearly a noun in other uses,
assume that it has the categorial status of a noun in this type of structure as well, unless
(as in the case of average in the expression average grade) you have clear evidence
that it is an adjective.
and [±plural]. In the light of this, specify which subclass(es) of noun the following
words belong to, and what their feature composition would be in their various uses:
77
Categories and features
describe the grammatical properties of each word as a whole, rather than the properties
of the particular form of the word cited below.
In addition, say what subclasses of noun the following determiners can (and can’t) be
used to modify
However, this does not correspond to the kind of usage we find in linguistics textbooks.
The word datum is generally not used at all, and the word data is used primarily as a
singular mass noun, though can also be used as a plural count noun (but not as a singu¬
lar count noun): cf.
(Note also that (ii)(a) is more natural than (ii)(b) for many people.) When a singular
count expression is required, a phrase such as an interesting piece of data is used
(rather than the pedantic form a datum). Thus, in this variety of English, data carries
the two alternative feature specifications [+count, tplural], and [-count, -plural].
The word little has two different uses in English: it can function either as an adjec¬
tive meaning ‘small in size’, or as a quantifier meaning ‘a small quantity of’. In its
quantifier use, it can generally modify only a singular mass noun: cf.
78
Workbook section
(where little might seem to be quantifying the plural count noun people) involve little
used as an adjective meaning ‘small in size’, and so are not genuine counterexamples
to the claim that when used as a quantifier, little can only modify a singular mass noun.
Exercise IX (§2.8)
In the text, we suggested that different inflectional affixes carried by verb
forms encode different sets of tense/aspect/agreement features. Identify the affixes in
the italicized verb forms in the examples below, and discuss the grammatical features
which they encode.
What features are encoded by the present-tense forms love/loves in each of these vari¬
eties? Finally, provide a feature-based account of the difference between the standard
and nonstandard past-tense forms illustrated below (where % marks a nonstandard
form):
79
Categories and features
Helpful hints
Assume that apparently uninflected verb forms carry a null (i.e. inaudi¬
ble) inflection 0. Also, examine the possibility that where a single affix marks more
than one property, it may encode some single feature shared between the various
forms. (To take a nonoccurring example, suppose that the past- and present-participle
forms of some verb carried the same affix; we could then say that the relevant affix
simply encodes the feature [+participle].)
Exercise X (§2.8)
Young children might be said to acquire the grammar of words ‘one feature
at a time’. Two-year-olds have generally acquired some (but not all) of the grammatical
properties of words, so that (from an adult perspective) particular child words might be
said to be underspecified in respect of particular sets of grammatical features. On the
basis of what you are told below, try to develop a feature-based account of the child’s
acquisition of English verb forms, using the system of features introduced in the text.
At the very earliest stage of development (stage I), children typically use only unin¬
flected verb forms like go: these are used with any choice of subject (e.g. Me/Daddy go
home). At stage II, we see the emergence of +ing forms of verbs, so that at this point
we have a two-way contrast (cf. e.g. Daddy go/going home). At stage III, we find past-
participle forms ending in +n (e.g. gone) being acquired alongside past-tense forms
(e.g. went), so that we now have a four-way contrast (cf. Daddy go/going/gone/went
home). At stage IV, we see the emergence of agreement in present-tense forms, so that
there is a distinction between He/she goes there and Ilwelyoulthey go there: at this
point, we have a five-way contrast between go/going!gone/went/goes.
Exercise XI (§2.8)
English might be said to have the following system of personal pronouns
(note that although we analysed forms like my/our/your etc. in the text as determiners,
here we are following an alternative traditional analysis of them as genitive case forms
of pronouns):
80
Workbook section
These pronouns inflect for case (nominative, objective or genitive), number (SG =
singular, PL = plural), gender (M = masculine, F = feminine, N = neuter) and person
(1 = first person, 2 = second person, 3 = third person). It might be argued that the gram¬
matical differences between these various pronoun forms could be described in terms of
a system of binary person/number/animacy/gender/case features such as the following:
Using this feature system, discuss the feature composition of each of the eighteen dif¬
ferent pronoun forms in the table above.
Helpful hints
Make the following set of assumptions. The feature [+1 person] differen¬
tiates first person pronouns like I/we from others, and pronouns which are [-1 person]
pronoun like we (which includes the speaker and someone else) from its [-plural]
counterpart I (which includes only the speaker). The feature [±animate] would differ¬
entiate between pronouns like he/she which can be used to denote animate beings, and
pronouns like it which are used to denote inanimates; pronouns which carry the feature
[+animate] would be further differentiated into [±feminine] forms. Thus, she would be
[+animate, + feminine], he would be [-panimate, -feminine] and it would be simply
[-animate]. The case feature [±nominative] would be used to differentiate nominative
pronouns from others, and the feature [±genitive] would be used to differentiate geni¬
tive pronouns from others: since no pronoun can carry two positive case features,
pronouns cannot be both [tgenitive] and [tnominative]: hence, it follows that any
pronoun which is [-pnominative] will automatically be [-genitive] (so need not be
specified as nongenitive), and conversely that any pronoun which is [-pgenitive] will
automatically be [-nominative] (so again need not be specified as non-nominative).
81
Categories and features
Objective pronouns (by virtue of being neither nominative nor genitive) carry the
features [-nominative, -genitive]. Thus, they would be [+nominative], their would be
[+GENITIVE] and them would be [-nominative, -genitive]. Analyse pronoun forms like
you which serve two different nominative/objective functions as a single item (not as
two different items, one nominative and the other objective). Assume that some
pronouns are unspecified with respect to certain features. For example, the pronoun
you might be argued to be unspecified (i.e. not morphologically marked) for number,
gender or the nominative/non-nominative case distinction, so that - in terms of its case
properties - you is a nongenitive form.
82
Workbook section
In English, (transitive) verbs and prepositions can have a noun expression as their
complement (immediately following them), but not nouns or adjectives:
In Italian, nouns and adjectives are inflected for (masculine or feminine) gender, but
not verbs or prepositions: cf.
In English, adverbs like really can be used to modify adjectives, prepositions and
verbs, but not nouns (instead, the corresponding adjective real must be used to modify
a noun): cf.
83
Categories and features
Helpful hints
You should look at the question of whether a given set of expressions
which behave similarly share a given (positive or negative) feature specification:
e.g. whether all the words which behave in a certain way have the feature value [+F]
(or alternatively have the value [—F]) for some feature F. So, in 1, you look to see
whether (in terms of the feature analysis in (76) in the main text), prepositions and
nouns form a natural supercategory in that they share some feature specification in
common, in 2 whether verbs and prepositions share a common feature specification, in
3 whether nouns and adjectives constitute a supercategory, in 4 whether adjectives and
verbs have a common feature specification, in 5 whether adjectives, prepositions
and verbs have some common set of featural properties and in 6 whether nouns, prepo¬
sitions and adjectives form a natural supercategory.
Consider whether Grimshaw’s analysis (or that proposed in the text) could better
account for the following facts:
1 Items such as for, to, have, this and that have more than one use. (Are these
different uses of the same item related or unrelated in the two analyses?)
2 Verbs and auxiliaries inflect for tense/agreement in English, nouns and
(some) determiners inflect for (singular/plural) number (cf. this
book/these books), but prepositions and complementizers are invariable.
3 Prepositions in Irish inflect for person, number and gender (e.g. leis =
‘with him’, leithi = ‘with her’); finite complementizers in West Flemish
inflect for agreement with the subject of the auxiliary/verb following
them (e.g. Kpeinzen dan Valere en Pol morgen goan = ‘I think that+3PL
Valere and Pol will go tomorrow’, from Haegeman 1992).
84
Workbook section
85
3
Syntactic structure
3.1 Overview
In the previous chapter, we looked at the way in which the grammatical
properties of words can be described in terms of grammatical categories or grammati¬
cal features. In this chapter, we turn to address the rather different question of the ways
in which words can be combined together to form phrases and sentences, and we look
at how we can represent the structure of the phrases and sentences thereby formed.
As speaker B’s utterance illustrates, the simplest way of forming a phrase is by com¬
bining two words together: for example, by combining the word privatize with the
word hospitals in (1), we form the phrase privatize hospitals.
An important question to ask, however, is the following: 'When two words combine
together to form a phrase, what grammatical properties does the resulting phrase have, and
how are they determined?’ There is clear evidence that the grammatical properties of
phrases are determined by one of the two words in the phrase. For example, when we com¬
bine a verb like privatize with a noun like hospitals, the resulting phrase privatize hospitals
seems to have verbal (= verblike) rather than nominal (= nounlike) properties. This we can
see from the fact that the phrase privatize hospitals can occupy the same range of sentence
positions as a verb like reconsider, and hence e.g. occur after the infinitive particle to: cf.
By contrast, privatize hospitals cannot occupy the kind of position occupied by a plural
noun such as hospitals, as we see from (3) below:
86
3.2 Forming phrases
(3) (a) Hospitals are at the heart of the debate about policy
(b) *Privatize hospitals are at the heart of the debate about policy
So, it seems clear that the grammatical properties of a phrase like privatize hospitals
are determined by the verb privatize, and not by the noun hospitals. We might say that
the verb privatize is the head of the phrase privatize hospitals, and conversely that the
phrase privatize hospitals is a projection (i.e. ‘phrasal expansion’) of the verb
privatize. More specifically, we can say that a phrase like privatize hospitals is a verb
phrase: and in the same way that we abbreviate category labels like verb to V, we can
abbreviate the category label verb phrase to VP. If we use the traditional labelled
bracketing technique to represent the category of the overall phrase privatize hospitals
and of its component words privatize and hospitals, we can represent the structure of
the resulting phrase as in (4) below:
What (4) tells us is that the overall phrase privatize hospitals is a verb phrase/VP, and
that it comprises the verb/V privatize and the noun/N hospitals. The verb privatize is
the head of the overall phrase, and the noun hospitals is traditionally said to fulfil the
grammatical role of being the complement of the verb privatize.
Although we have used our familiar technique of labelled bracketing to represent the
structure of the verb phrase privatize hospitals in (4), an alternative way of represent¬
ing the structure of phrases is in terms of a labelled tree diagram such as (5) below
(you have to imagine that the tree has been uprooted in a storm, hence it is upside
down):
(5) VP
privatize hospitals
We might say that each of the category labels in a tree diagram is a different node in the
tree, and that the words are the leaves on the tree. The tree diagram in (5) is entirely
equivalent to the labelled bracketing in (4), in the sense that the two provide us with
precisely the same information about the structure of the phrase privatize hospitals: the
differences between the two are purely notational (e.g. each category is represented by a
single node in a tree diagram, but by a pair of brackets in a labelled bracketing), and not
of any theoretical significance. In general, we shall use tree diagrams to represent phrase
structure in our discussions below (though where we only want to indicate part of the
structure of a given phrase or sentence, we shall sometimes use labelled bracketings).
Given that our ultimate goal is to determine the principles of phrase formation, an
important question for us to ask at this juncture is whether our analysis of the verb
87
Syntactic structure
phrase privatize hospitals leads to any wider conclusions about the way in which phrases
are formed. We might generalize our discussion of (5) and hypothesize that all phrases
are formed by a process of merger whereby two categories are merged together to
form a new (phrasal) category. The phrase which is thereby formed is (to use a tradi¬
tional technical term) endocentric (i.e. headed), in that it is a projection of a head word.
In the case of (5), the resulting phrase is formed by combining two words together.
However, not all phrases are formed by combining two words - as we see if we look at
the structure of B’s utterance in (6) below:
The phrase in (6) would seem to be formed by combining the infinitive particle to with
the verb phrase privatize hospitals. What’s the head of the resulting phrase to privatize
hospitals? A reasonable guess would be that the head is the infinitive particle to, so that
the resulting string (i.e. sequence of words) to privatize hospitals is an infinitive
phrase. If this is so, we’d expect to find that infinitive phrases have a different distribu¬
tion from verb phrases; and indeed this does seem to be the case, as sentences such as
(7) and (8) below illustrate:
(9)
privatize hospitals
The structure (9) is formed by two different merger operations; one merging the verb
privatize with the noun hospitals to form the verb phrase/VP privatize hospitals, and
88
3.3 Specifiers
the other merging the infinitive particle to with the verb phrase privatize hospitals to
form the infinitive phrase/IP to privatize hospitals.
What is implicit in our discussion here is that we can build up complex structures in
a stepwise fashion by merging successive pairs of constituents together to form even
larger phrases. For example, by merging the infinitive phrase to privatize hospitals
with the noun plans, we can form the phrase plans to privatize hospitals, as in speaker
B’s reply in (10) below:
The resulting phrase would seem to be headed by the plural noun plans, and so
requires plural agreement if used as the subject of an auxiliary like be: cf.
So, it seems clear that plans is the head of the bracketed phrase in (10), and the infini¬
tive phrase to privatize hospitals is the complement of the noun plans. This being so,
the bracketed phrase in (11) has the structure (12) below:
to V N
privatize hospitals
The structure (12) is the product of three different merger operations - one merging the
verb privatize with the noun hospitals to form the verb phrase privatize hospitals: another
merging the infinitive particle to with the verb phrase privatize hospitals to form the
infinitive phrase to privatize hospitals', and a third merging the noun plans with the infini¬
tive phrase to privatize hospitals to form the noun phrase plans to privatize hospitals.
3.3 Specifiers
So far, all examples of the merger operation which we have looked at
have involved merging a head word with a complement (the complement being either a
word, or a phrase which is itself formed by one or more merger operations). This might
lead us to wonder whether phrases are always of the form head + complement, with the
complement being positioned after the head (because English has a head-first setting
for the head parameter, as noted in §1.7). However, this seems unlikely, as we can
show in relation to phrases such as that produced by speaker B in (13) below:
89
Syntactic structure
If (contrary to what we are suggesting here) all phrases were formed by combining a
head with a following complement, then we would have to posit that the italicized
phrase in (13) was a noun phrase formed by merging the head noun government with
the noun phrase complement plans to privatize hospitals. But this can’t be right
because the overall phrase is clearly plural in number, as we see from the pattern of
agreement in (14) below:
If the singular noun government were the head of the bracketed phrase in (14), we’d
wrongly predict that the singular auxiliary is would be required. The fact that the plural
form are is required makes it clear that the head of the overall phrase is the noun plans.
Since this noun has a complement of its own (namely the infinitive phrase to privatize
hospitals) and since in any case complements follow their heads in English, the noun
government which precedes the head plans cannot be the complement of plans. So
what grammatical function does it serve? Introducing a new term at this point, let’s say
that the noun government functions as the specifier of the noun plans (since it specifies
who has devised the plans in question).
The obvious question to ask at this point is ‘What is a specifier, and how can we tell
whether a given expression functions as a specifier or not?’ This is not a straightfor¬
ward question to answer, since although the notion specifier is central to much
contemporary work in syntax, it is hard to identify any common set of properties which
all specifiers share (and indeed there is sometimes disagreement between linguists
about whether one expression does or doesn’t function as the specifier of another). For
the time being, rather than attempt to give a unitary definition of specifier, we shall
simply illustrate some of the ways in which the term is used in the contemporary
linguistic literature. In this connection, consider the syntax of the following phrases
and sentences:
One way of analysing the relevant expressions is as follows. In (15a) the preposition
to is the head of the relevant structure, the noun bed is the complement of to and
the adverb straight is the specifier of to. In (15b), the determiner a is the head of the
structure, the noun pity is its complement and the degree adjective such is its specifier.
90
3.3 Specifiers
In (15c), the verb teasing is the head of the structure, the expression the other is its
complement and the quantifier each is its specifier. In (15d), the auxiliary have is the
head of the structure, the verb finished is the complement of have and the pronoun they
is the specifier of have. And in (15e), the inverted auxiliary are is the head of the struc¬
ture, the expression we waiting is the complement of are and the adverb why is
the specifier of are. If these assumptions are correct, each of the structures in (15) is of
the form specifier + head + complement, suggesting that heads are generally positioned
before complements in English, and specifiers positioned before heads (so that English
might be said to be not only a head-first language because it positions heads before
complements, but also a specifier-first language because it positions specifiers
before heads). We should underline once again, however, that there is by no means uni¬
versal agreement among linguists as to the nature and function of specifiers (e.g. the
analysis of (15d-e) presented here is generally accepted, but the analysis of (15a-c) is
more controversial). Since there is no universally agreed definition of what a specifier
is, it is only to be expected that you yourself won’t at first be very clear about whether
this or that expression should be analysed as a specifier or not. Hopefully, it
will become clearer as our exposition unfolds just how the term is used in this book;
and the justification for analysing particular expressions as specifiers is ultimately
empirical - i.e. it resides in the fact that if we posit that a particular expression serves
as a specifier in a given construction, we can begin to make sense of various aspects of
the syntax of the construction (in ways which will gradually become clearer in subse¬
quent chapters).
Returning now to our discussion of the bracketed expression government plans
to privatize hospitals in (14), let’s assume that the noun government here serves as
the specifier of the noun plans (setting aside any unclarity about what the term
specifier means). If this is so, the relevant expression will be a noun phrase/NP
built up out of three different constituents - namely the head noun plans, its com¬
plement to privatize hospitals and its specifier government. Now, if we assume that
the merger operation which forms phrases combines pairs of constituents into
successively larger phrases, then it follows that the overall noun phrase govern¬
ment plans to privatize hospitals must be formed by two different merger opera¬
tions. Reasoning along these lines, we might suppose that the head noun plans
combines with its infinitive phrase complement to privatize hospitals to form an
intermediate nominal expression which (following the terminology introduced in
Chomsky’s Remarks on Nominalization paper in 1970) we might call an N (also
written as N' or as N-bar, but in all three cases pronounced en-bar)\ let’s also sup¬
pose that the resulting N-bar is then merged with the noun government which
serves as its specifier, so forming the full noun phrase/NP constituent government
plans to privatize hospitals. Given these assumptions, the resulting NP will have
91
Syntactic structure
privatize hospitals
What (16) tells us is that the head noun plans merges with its IP complement to priva¬
tize hospitals to form the N-bar plans to privatize hospitals, and the resulting N-bar
then merges with its N/noun specifier government to form the overall NP/noun phrase
government plans to privatize hospitals.
It’s important to be clear about the system of category labels which we are employ¬
ing at this point. We are using the category label N-bar to designate an intermediate
nominal projection - i.e. to designate a nominal expression which is larger than a noun,
but which itself projects into an even larger nominal structure (this larger nominal
structure being designated as NP). By contrast, the category label NP denotes a nomi¬
nal expression which is larger than a noun, but which doesn’t project into an even
larger type of nominal expression. Thus, the expression plans to privatize hospitals is
an NP constituent in (10) (since it doesn’t project into a larger nominal structure), but is
an N-bar in (13) (since it projects into the larger nominal expression government plans
to privatize hospitals).
Given the assumptions we have made so far, there are two different types of phrase:
on the one hand, there are phrases of the schematic form (17a) below which comprise
just a head X (where X represents a word category of some kind - e.g. a noun, or verb,
or preposition, etc.) and its complement; while on the other hand, there are more com¬
plex phrases of the schematic form (17b) below which comprise a specifier, a head (=
X) and a complement:
X complement
head
X complement
head
On this view, specifiers differ from complements not only in respect of the fact that
they are positioned to the left (rather than the right) of the head, but also in respect of
the structural position they occupy (specifiers are sisters to an intermediate (X-bar)
projection, whereas complements are sisters to a head (as we shall see shortly)).
Having looked at the internal structure of the noun phrase government plans to pri¬
vatize hospitals, let’s now consider the structure of the more complex phrase produced
by speaker B in the dialogue below:
92
3.3 Specifiers
The italicized structure in (18) would seem to be formed by merging the preposition
against with the noun phrase (16), so forming the prepositional phrase (= PP) con¬
stituent (19) below:
(19)
privatize hospitals
The resulting structure (19) is formed by five separate merger operations, viz. one
merging the V privatize with the N hospitals to form the VP privatize hospitals', a
second merging the I to with the VP privatize hospitals to form the IP to privatize hos¬
pitals; a third merging the N plans with the IP to privatize hospitals to form the N-bar
plans to privatize hospitals', a fourth merging the N government with the N-bar plans to
privatize hospitals to form the NP government plans to privatize hospitals', and a fifth
merging the preposition against with the NP government plans to privatize hospitals to
form the PP against government plans to privatize hospitals.
Now consider speaker B’s response in (20) below:
The italicized structure in (20) is arguably a verb phrase/VP, formed by merging the
verb vote with the prepositional phrase against government plans to privatize hospitals
in (19), and so has the structure (21) below:
(21)
government
plans
to N
privatize hospitals
hospitals
What’s the structure of the sentence produced by speaker B in (22)? Let’s make
the (unifying) assumption that sentences are formed by the same merger operation
as phrases (merging categories together in a pairwise fashion to form larger cate¬
gories). This being so, we might suppose that speaker B’s reply in (22) is formed by
merging the auxiliary will with the verb phrase vote against government plans to priva¬
tize hospitals, and then subsequently merging the pronoun they with the resulting
structure. Since the sentence thereby formed is finite, it seems reasonable to take the
finite auxiliary will to be the head of the clause. Moreover, given that complements
follow heads and specifiers precede them, we might suggest that the VP vote against
government plans to privatize hospitals is the complement of will, and that the pronoun
they is its specifier. If we adopt the suggestion made in §2.5 that finite auxiliaries
are I/INFL constituents (by virtue of being inflected for finiteness), we can offer the
following account of how the italicized sentence in (22) is formed. The I constituent
will is merged with its VP complement vote against government plans to privatize
hospitals to form the intermediate projection! (= I' = I-bar, pronounced eye-bar) will
vote against government plans to privatize hospitals. The resulting I-bar is then
merged with the pronoun they which serves as its specifier (and subject) to form
the full IP (= inflected auxiliary phrase) They will vote against government plans to
privatize hospitals. Given these assumptions, the italicized sentence in (22) will have
the structure (23) below:
94
3.4 Forming sentences
(23)
prn'
They
will
vote
against N
government
plans
to N
privatize hospitals
The overall sentence is headed by the auxiliary (= I constituent) will, and thus has the
status of an auxiliary phrase (= IP).
But now consider the reply given by speaker B in the following dialogue:
In speaker B's reply in (24), it would seem that the auxiliary phrase (IP) they will vote
against government plans to privatize hospitals has been merged with the complemen¬
tizer that to form an even larger type of structure, which (in work since 1970) has been
termed a complementizer phrase (= CP). Given this assumption, speaker B's reply in
(24) will have the structure (25) below:
(25) CP
privatize hospitals
95
Syntactic structure
Following this line of reasoning, we might suggest that clauses introduced by comple¬
mentizers like that! if I for are CP constituents (i.e. complementizer phrases). This analy¬
sis (which takes the complementizer to be the head of the structure containing it)
implies that complementizers determine the nature of the clause they introduce; and
this is in keeping with the observation we made in §2.6 that a complementizer like that
indicates that the clause it introduces is a finite declarative (i.e. statement-making)
clause.
So far, all the examples of merger operations which we have discussed have
involved merging a word with another word, or merging a word with a phrase.
However, if we take the simplest characterization of merger as an operation which
forms phrases by combining pairs of categories, we should expect to find that merger
can also involve combining two projections (i.e. two expressions larger than a word).
In this connection, compare (26a) and (26b) below:
Given the assumptions we have been making, (26a) is formed by merging the preposition
against with the noun privatization to form the prepositional phrase against
privatization; the resulting PP is then merged with the verb vote to form the verb phrase
vote against privatization; this VP is then merged with the I constituent will to form the
I-bar will vote against privatization-, and the I-bar is then merged with its specifier/subject
(the pronoun they) to form the IP They will vote against privatization.
However, in the case of (26b), the subject which merges with I-bar is not the
pronoun they, but rather the phrase the opposition. This is formed by merging the
determiner the with the noun opposition to form a type of constituent which is termed a
determiner phrase (= DP) in recent work; the resulting DP has the structure (27)
below:
the opposition
This being so, the overall clause (26b) is formed by merging the I-bar will vote against
privatization with a phrase which has itself been formed by another merger operation -
namely the determiner phrase the opposition. Thus, the resulting IP is formed by
merging the DP subject the opposition with the I-bar will vote against privatization,
and so has the structure (28) below:
96
3.5 Tree diagrams
vote p N
I I
against privatization
So, (28) illustrates the possibility of forming a phrase by merging one projection (i.e.
expression larger than a word) with another (in this case, merging DP with I-bar).
97
Syntactic structure
tions combines two categories together, and never combines e.g. three, four or five cat¬
egories together at one go. This amounts to positing the following constituent structure
principle:
Principle (29) claims that all nonterminal nodes branch into two and only two immedi¬
ate constituents, never more than two, and never fewer than two. This is a controversial
assumption (in that not all syntacticians accept it), though it is widely adopted in much
recent work (and will be assumed throughout the rest of this book).
Given a tree diagram such as (28), there is a relatively informal procedure which we
can use in order to determine whether a given string (i.e. sequence of words) does or
doesn’t form a constituent of the relevant structure. Imagine that you’re the kind of
anti-social and environmentally unfriendly individual who gets their kicks out of
breaking branches off trees. Let’s also suppose that nodes are the weak points of trees,
and that you can only break branches off at nodes. (Recall that a node is a point of a
tree which carries a category label.) We might then say that any part of a tree (any
substructure) which comes away when you break the tree off at a particular node is a
constituent (of the structure containing it), and that the category of the relevant
substructure is represented by the category label carried by the root (= topmost) node
in the structure. For example, if we break the tree off at the I-bar node in (28), the part
of the tree that will come away is:
against privatization
What this act of environmental vandalism tells us is that the string will vote against pri¬
vatization is a constituent of the structure (28), and that it belongs to the category I-bar
(and so is an intermediate auxiliary projection), since the topmost node in the substruc¬
ture (30) carries the category label I-bar. Of course, if we had been in a rather less
destructive mood, we could have chosen simply to break off a branch at the V node in
(28); all we would have pulled away would have been the poor twig in (31) below:
(31) V
vote
What we learn from this vandal-like break-off-a-branch game is that every part of a
tree structure which you can break off at a node is a constituent of the overall structure.
98
3.6 Configurational relations
The obvious conclusion is thus that there are as many constituents in a tree structure as
there are nodes carrying category labels. Hence, there are eleven different constituents
in (28) - namely [N privatization], [p against], [pp against privatization], [v vote],
[vp vote against privatization], [(will], [y will vote against privatization], [N opposition],
[D the], [Dp the opposition] and [Ip the opposition will vote against privatization],
H J
The dominance relations between nodes (relating to whether one node occurs higher
up in the structure than another) are indicated by the branches (i.e. solid lines) in the
tree; the precedence relations between nodes (i.e. whether one node occurs to the left
of another or not) are indicated by the relative left-to-right ordering of nodes on
the printed page. (We can say that one node X precedes another node Y if X occurs to the
left of Y, and if neither node dominates the other.) A simple way of understanding the
relation dominates is to think of tree diagrams like (32) as train networks, with each of
99
Syntactic structure
the labelled nodes being different stations in the network, and with the solid lines repre¬
senting the tracks linking the various stations in the network. We can then say that X
dominates Y if it is possible to get from X to Y on a southbound train. So, for example,
A dominates H in (32) because it is possible to get from A to H on a southbound train
(via E and G). By contrast, B does not dominate H because it is not possible to get from
B to H on a southbound train: on the contrary, to get from B to H, you first have to
catch a northbound train to A, and then from A catch a southbound train to H (via E
and G). As you should be able to work out for yourself, node A in (32) dominates nodes
B, C, D, E, F, G, El and J; node B dominates only nodes C and D; node E dominates
nodes F, G, El and J; node G dominates nodes H and J; and nodes C, D, F, H and J are
terminal nodes and so dominate no other nodes (i.e. in train terms, they represent the
terminus at the end of each line).
We can make use of dominance relations to provide a configurational definition
of the term constituent as in (33) below (i.e. a definition couched in terms of tree con¬
figurations, viz. the relative positions occupied by different nodes in a tree):
Let’s see how this definition can be applied in relation to a structure such as (34)
below:
(34) ^.Ip
PRN
They
can o'
4 I
both V
1I |1
speak French
The quantifier both (in this kind of use) is referred to as a floating quantifier because
although it quantifies they, it is floating in a position where it is separate from they. We
have assumed in (34) that the verb speak combines with its noun complement French
to form a V (= V' = Y-bar, in each case pronounced vee-bar) constituent speak French,
and that (as suggested in recent work - e.g. Sportiche 1988) the floating quantifier both
is the specifier of this V-bar, and combines with it to form the VP both speak French.
Given these assumptions, it follows from the definition in (33) that the string (of
words) speak French is a V-bar constituent, since [v speak] and [N French] are both
dominated by V-bar, and are the only terminal nodes dominated by V-bar. For analo¬
gous reasons, both speak French is a VP constituent, can both speak French is an I-bar
constituent, and They can both speak French is an IP constituent. By contrast, the
100
3.6 Configurational relations
string They speak French is not a constituent of the structure (34), since although
[prn they], [v speak] and [N French] are all dominated by the IP node, they are not the
only terminal nodes dominated by IP: after all, [j can] and [q both] are also terminal
nodes dominated by IP. Similarly, the string They both speak French is not a con¬
stituent of the structure (34), nor are strings such as can speak French, They both
speak, both French, etc.
Given the obvious resemblance between the tree diagrams like (34) above used to
represent the syntactic structure of sentences, and family tree diagrams (representing
who is related to whom in a given family), it has become a standard convention in the
linguistic literature to use kinship terminology to denote relationships between con¬
stituents within a given tree structure. We can illustrate how this terminology is used in
terms of an abstract tree diagram of the following form (where A, B, C, D and E are
labelled nodes):
(35) ^A
B "c
X \
D E
A here is said to be the mother of B and C (by virtue of being the next node up in the
tree from B and C); conversely, B and C are the two daughters of A; and B and C are
sisters. Likewise, C is the mother of D and E; in addition, D and E are the daughters
of C; and similarly D and E are sisters. (Interestingly, the corresponding tree relations
in Italian are father, son and brother. I leave the sociologists among you to ponder on
the hidden significance of this transcultural terminological transvestism.)
We can use these family relations to define the property of complementhood as
follows:
Returning now to our tree diagram (34), it follows from (36) that [N French] is the
complement of [v speak], since the relevant V and N nodes are sisters, and the mother
of N is a V-bar constituent speak French which is a projection of the verb speak; and,
by extension, we might say that French is also the complement of the V-bar speak
French, and of the VP both speak French. Similarly, [vp both speak French] is the com¬
plement of [, can] in (34), since the two are sisters, and the mother of VP is an I-bar can
both speak French which is a projection of the I constituent can; by extension, we
might say that the relevant VP is also the complement of the I-bar can both speak
French, and of the IP They can both speak French. By contrast, [y speak French] is
not the complement of [Q both] in (34) because although the two are sisters, the mother
101
Syntactic structure
In terms of the definition in (37), we can say that [q both] is the specifier of [v speak] in
(34), since [Q both] is the sister of [v speak French] and the daughter of [vp both speak
French]', and by extension, we can say that [Q both] is also the specifier of [y speak
French] and of [vp both speak French]. In much the same way, [pRN they] is the specifier
of [j can], since [pRN they] is the sister of [y can both speak French] and the daughter of
[IP they can both speak French]; and by extension, [pRN they] is also the specifier of
[y can both speak French] and of [ip They can both speak French], By contrast, [j can] is
not the specifier of [Q both] since can is not the sister or daughter of a projection of both;
and similarly, [v speak] is not the specifier of [N French], since speak is not the sister or
daughter of a projection of French.
102
3.7 Testing structure
theoretical apparatus which linguists find they need to make use of in order to explain
certain facts about language (just as molecules, atoms and subatomic particles are con¬
structs which physicists find they need to make use of in order to explain the nature of
matter in the universe). It is no more reasonable to rely wholly on intuition to deter¬
mine syntactic structure than it would be to rely on intuition to determine molecular
structure. Inevitably, then, much of the evidence for syntactic structure is of an essen¬
tially empirical character (i.e. evidence based on the observed grammatical properties
of words, phrases and sentences). The evidence typically takes the form ‘Unless we
posit that such-and-such a sentence has such-and-such a constituent structure, we shall
be unable to provide a principled account of the observed grammatical properties of
the sentence.' Thus, structural representations ultimately have to be justified in empiri¬
cal terms, i.e. in terms of whether or not they provide a principled account of the gram¬
matical properties of words, phrases and sentences.
To put our discussion on a more concrete footing, let’s consider the structure of a
sentence such as:
Example (38) contains the lexical verb watching and three auxiliary verbs, viz. the
modal auxiliary might, the perfective auxiliary have and the progressive auxiliary
been. What we shall claim here is that each of these four (auxiliary or nonauxiliary)
verbs heads a separate phrasal projection, with watching projecting into the phrase
watching television, been projecting into the phrase been watching television, have
projecting into the phrase have been watching television, and might projecting into the
overall phrase/sentence He might have been watching television. More specifically, we
shall claim that (38) has a structure along the lines of (39) below:
(39) IP
PRN I
I
He I PERFP
have PROG VP
I
been V N
| I
watching television
The label PERF here is used to denote a perfective auxiliary, and PERFP to denote a
perfective phrase - i.e. a phrase headed by a perfective auxiliary. Similarly, PROG
denotes a progressive auxiliary, and PROGP denotes a progressive phrase. Of course,
given our observation in §2.9 that (at a relatively coarse level of categorization) auxil¬
iaries like have/be can be classified as verbs, an alternative labelling for the relevant
103
Syntactic structure
constituents in (39) would be to suppose that been and have belong to the category V
(verb), and that correspondingly their phrasal projections been watching television and
have been watching television belong to the category VP (verb phrase). We assume that
the modal auxiliary might occupies the head INFL position of IP because it is finite
(hence inflected for tense, as we can see from the fact that it carries the past-tense suf¬
fix +f); the auxiliaries have and been do not occupy an INFL position here because they
are not finite (have is an infinitive form and been a perfective participle form).
Our structural representation (39) has the status of a hypothesis (i.e. untested and
unproven assumption) about the structure of the corresponding sentence (38). So how
can we test our hypothesis and determine whether (39) is or isn’t an appropriate repre¬
sentation of the structure of (38)? The answer is that there are a number of standard
heuristics (i.e. ‘tests’) which we can use to determine structure. One such test relates to
the phenomenon of coordination. English and other languages have a variety of coor¬
dinating conjunctions (which we might designate by the category label CONJ or J)
like and/but/or which can be used to coordinate (= conjoin = join together) expressions
such as those bracketed below:
In each of the phrases in (40), a coordinating conjunction has been used to conjoin the
bracketed pair of expressions. Clearly, any adequate grammar of English will have to
provide a principled answer to the question: ‘What kinds of strings (i.e. sequences of
words) can and cannot be coordinated?’
Now, it turns out that we can’t just coordinate any random set of strings, as we see
by comparing the grammatical reply produced by speaker B in (41) below:
Why should it be possible to coordinate the string up the hill with the string up the
mountain in (41), but not possible to coordinate the string up his mother with the string
up his sister in (42)? We might seek to provide a principled answer to this question in
terms of constituent structure. More specifically, we might argue that the italicized
string up the hill in (41) is a constituent of the phrase run up the hill (up the hill is a
prepositional phrase, in fact), and so can be coordinated with another similar type of
prepositional phrase (e.g. a PP such as up the mountain, or down the hill, or along the
104
3.7 Testing structure
path, etc.). Conversely, however, we might claim that the string up his mother in (42) is
not a constituent of the phrase ring up his mother, and so cannot be coordinated with
another similar string. (Traditional grammarians say that up is associated with ring,
and that the expression ring up forms a complex verb which carries the sense of ‘tele¬
phone’.) On the basis of contrasts such as these, we might suggest that:
A constraint (i.e. structural restriction) along the lines of (43) is assumed in most work
in traditional grammar.
Having established the constraint (43), we can now make use of this constraint as a
heuristic (i.e. diagnostic test) for evaluating the structural representation (i.e. testing the tree
diagram) in (39) above. A crucial claim made in (39) is that the strings watching television,
been watching television, have been watching television and might have been watching
television are all constituents (of various different types). Some empirical evidence for this
assumption comes from coordination facts in relation to paradigms (i.e. sets of examples)
such as:
Given the crucial premise (43) that only strings of like constituents can be conjoined,
example (44a) provides empirical support for analysing watching television as a VP,
since it can be conjoined with another verb phrase like playing pool, likewise, (44b)
provides empirical evidence for analysing been watching television as a PROGP, (44c)
for analysing have been watching television as a PERFP, and (44d) for analysing might
have been watching television as an I-bar. In other words, the analysis in (39) correctly
predicts that all of the sentences in (44) are grammatical (though for pragmatic reasons
they become increasingly more unwieldy as more and more information is repeated,
since the longer sentences like (44d) violate Grice’s (1975) conversational maxim ‘Be
concise!’ by virtue of repeating redundant words which could have been omitted).
The type of coordination discussed above might be called simple coordination.
However, there is a second type of coordination test which we can use, and this relates
to coordination structures such as that produced by speaker B in the following dia¬
logue:
105
Syntactic structure
Schema (46) is intended to show that we have coordinated two fry-phrase^ and that
the bold-printed string for clemency is the shared complement both of the verb appeal
and of the verb beg. We might suppose that when a string is shared between two or
more conjuncts in this way, it need be mentioned only once - hence the fact that it
occurs only once in speaker B’s utterance in (45). For obvious reasons, we might refer
to this type of coordination as shared string coordination, and to the bold-printed
sequence in (46) as the shared string.
However, there are interesting restrictions on the kinds of sequences which can
function as the shared string in such cases: although the reply given by speaker B in
(45) above is grammatical, the reply in (47) below is not: cf.
Why should the sequence for clemency be able to function as the shared string in (45), but
not the sequence up his mother in (47)? A natural answer would be to suppose that this is
because the string/or clemency is a constituent of the verb phrases appeal for clemency/beg
for clemency (it is a prepositional phrase, in fact), whereas the sequence up his mother is not
a constituent of the verb phrase ring up his mother, and we might further suppose that:
(48) Only a constituent can function as the shared string in the relevant type
of coordinate structure.
Given (48), it follows that speaker B’s reply in (45) is grammatical because the itali¬
cized shared string is a constituent, whereas speaker B's reply in (47) is ungrammatical
because the shared string is a nonconstituent sequence.
In the light of (48), we might suggest that one way of testing whether a given
expression is or is not a constituent of a given structure is by seeing whether or not the
relevant expression can function as the shared string in elliptical coordinate structures.
In this connection, consider the following:
(49) (a) He might have been (or might not have been) watching television
(b) He might have (or might not have) been watching television
106
3.8 Additional tests
Given the premise (48) that the shared string in this type of coordinate structure must
be a constituent, the fact that watching television in (49a) can function as a shared
string is consistent with the analysis of it as a verb phrase in (39); similarly, the fact
that been watching television is the shared string in (49b) is consistent with the claim in
(39) that it is a progressive phrase constituent; likewise, the fact that have been watch¬
ing television is the shared string in (49c) lends some empirical support to the claim in
(39) that it is a perfective phrase constituent; and finally, the fact that might have been
watching television is the shared string in (49d) is consistent with the assumption in
(39) that it is an I-bar constituent.
Why should it be that to my girlfriend can serve as a sentence fragment in (50), but not
up my girlfriend in (51)? The traditional answer is (52) below:
Given the postulate in (52), we can account for the contrast between (50) and (51) in a
straightforward fashion if we maintain that to my girlfriend is a constituent of the verb
phrase talking to my girlfriend, whereas up my girlfriend isn’t a constituent of the verb
phrase ringing up my girlfriend.
In the light of our premise (52), consider the following mini-dialogue (paying partic¬
ular attention to the various replies which speaker b can give):
(53) speaker a: What d’you think she might have been doing?
speaker b: Watching television!Been watching television!Have been
watching television
107
Syntactic structure
(Or course, given Grice’s conciseness maxim, the longer the replies in (53), the more
cumbersome they are, since they repeat redundant information: but all of the replies in
(53) are grammatical.) In the light of our premise in (52) that only constituents can be
used as sentence fragments, speaker B’s replies in (53) are consistent with the analysis
in (39), in that when B replies watching television he is using a VP, when he replies
been watching television he is using a PROGP, and when he replies have been watch¬
ing television, he is using a PERFP. Thus, the analysis in (39) correctly predicts that all
three replies produced by speaker B in (53) are grammatical.
But wait! Doesn’t the analysis in (39) also predict that we should be able to use the
I-bar (= auxiliary expression) might have been watching television as a sentence frag¬
ment? However, this is not the case, as we see from the ungrammaticality of (54)
below:
(54) speaker a: What d’you think she might have been doing?
speaker b: * Might have been watching television
Surely (54) is a counterexample to the analysis in (39), since (39) specifies that might
have been watching television is an I-bar constituent, and a core assumption of the sen¬
tence fragment test is that constituents can serve as sentence fragments? The answer is
‘No’, since the claim in (52) that only constituents can serve as fragments does
not entail that all constituents can serve as fragments. More specifically, we might
hypothesize that:
Following Chomsky (1995a, p. 396) we might define the term maximal projection
informally as follows:
(56) A category that does not project any further is a maximal projection.
(Note that further here means ‘into a larger structure with the same head’.) We could
then say that might have been watching television can’t be used as a sentence fragment
in (54) because it is an intermediate projection: by this we mean that although it is a
projection of the I constituent might, it isn’t the maximal (i.e. largest) projection of I,
since I projects further into the IP He might have been watching television. Hence, if
speaker B in (54) wants to reply with a structure headed by might, he has to use the full
IP He might have been watching television.
Our revised premise (56) has interesting implications for how we analyse speaker
B’s reply in (57) below:
(57) speaker a: What do you think she might have been watching?
speaker b: Television
Given the definition of maximal projection in (56), the noun television is a maximal
projection in the structure (39) above, precisely because it does not project into a larger
108
3.8 Additional tests
NP constituent. Thus a word which has no complement or specifier of its own func¬
tions as a maximal projection - and hence can serve as a sentence fragment.
A fourth structural diagnostic which we can use to test structures like (39) relates to
the syntax of a class of items which are sometimes referred to collectively as
proforms. These are items which can be used to replace (or refer back to) a constituent
of an appropriate kind, as in (58) below:
(58) (a) The Vice-President is giving away freebies to anyone promising to vote
for him
(b) They say that he is guilty, though I personally don't think so
(c) If John does quit his job (which he might), who will replace him?
In (58a), the proform him is used to replace the determiner phrase the vice-president, in
order to avoid repetition (equivalently, we might say that the vice-president is the
antecedent of him, i.e. the expression that him refers back to); in (58b) the proform so
replaces/refers back to the complementizer phrase that he is guilty, and in (58c) the pro¬
form which replaces/refers back to quit his job.
Of course, if different kinds of proform can be used to replace (or serve as the
antecedent of) different kinds of constituent, another way in which we can test the struc¬
ture in (39) is to see whether the various constituents in (39) can be replaced by (or
serve as the antecedent of) an appropriate type of proform. In this connection, consider
the use of so in (59) below:
(59) He suggested that she might have been watching television, and . . .
(a) so she might have been
(b) so she might have
(c) so she might
(d) *so she
The proform so replaces (or has as its antecedent) the verb phrase watching television
in (59a), the progressive phrase been watching television in (59b) and the perfective
phrase have been watching television in (59c). Given that (as noted in §2.9) lexical
verbs like watch and auxiliary verbs like have/be share in common the features
[+V -N], we could say that so can replace a projection of a [+V, -N] head (i.e. a projec¬
tion of an auxiliary or nonauxiliary verb). However, what remains to be accounted
for is why (59d) should be ungrammatical - i.e. why so can’t replace the I-bar might
have been watching television. One reason might be that proforms can only be used
to replace maximal projections, and I-bar is an intermediate projection. An alternative
possibility might be to suppose that so has to be used as the complement of an appro¬
priate auxiliary or nonauxiliary verb and there is no such verb here. Thus, proform
facts work out much as we expect, and lend further empirical support to our postulated
structure (39).
109
Syntactic structure
The fifth and final constituent structure test which we’ll make use of here relates to a
phenomenon traditionally referred to as ellipsis - a term used to designate the process by
which redundant information in a sentence is ellipsed (i.e. omitted) if it can be inferred
from the context (e.g. if it has been previously mentioned in the preceding discourse).
There are two rather different kinds of ellipsis - one involving the ellipsis of head words,
and the other of projections (i.e. of expressions comprising more than just a head word).
We can see the difference by comparing the two sentences in (60a-b) below:
(60) (a) He can speak French better than she can German
(b) He can speak French better than she can
(61) (a) She might have been watching television more often than he might have
been
(b) She might have been watching television more often than he might have
(c) She might have been watching television more often than he might
(d) She might have been watching television more often than he
(Of course, given Grice’s conciseness maxim, the longer sentences like (61a) which repeat
redundant words are more cumbersome than the corresponding shorter sentences - but
they are none the less grammatical.) An analysis such as (39) above can provide a straight¬
forward account of the ellipsis facts in (61), since what has been ellipsed is the VP watch¬
ing television in (61a), the PROGP been watching television in (61b), the PERFP have
been watching television in (61c) and the I-bar might have been watching television in
(6 Id). Thus, ellipsis facts provide further empirical support for the analysis in (39).
3.9 C-command
We have argued at some length here that heuristics such as our two coor¬
dination tests, the sentence fragment test, the proform test and the ellipsis test provide
110
3.9 C-command
empirical evidence in support of our claim that any phrase or sentence in a language
has a complex hierarchical categorial constituent structure which can be represented
in terms of a labelled tree diagram such as (39). In this section, we look at a further
type of evidence, based on the assumption that certain syntactic constraints (i.e.
restrictions on the use of particular kinds of constituents) can be given a straightfor¬
ward characterization in structural terms. As a case in point, consider the syntax of the
quantifier any (and its compounds like anyone, anything, anywhere, etc.). This has two
uses. One is as a universal quantifier with a meaning similar to that of every!all, as in
Any policeman can tell you the time. The second use (and the one which concerns us
here) is as an existential (or partitive) quantifier with a meaning similar to that of some,
e.g. in sentences such as Is there any coffee left? (where the speaker is asking a ques¬
tion about the existence of coffee). In its existential use, any (and its compounds) can
occur in negatives like (62a) below, interrogatives like (62b), conditionals like (62c)
and degree structures like (62d) - but not (for example) in ordinary declaratives like
(62e):
Edward Klima (1964, p. 313) conjectured that negative words like nobody, interroga¬
tives like whether, conditionals hke if and degree expressions like too share ‘a
common grammatico-semantic feature to be referred to as affective': in his terms,
nobody/whether/if/too are affective constituents. (By this, he seems to mean that they
are all non-assertive.) Given this terminology, we can say that existential any is
restricted to occurring in a structure containing an affective constituent. It turns out that
numerous other expressions (e.g. ever, care a damn, lift a finger, etc.) are similarly
restricted to occurring in affective contexts: cf.
(64) (a) He’s too selfish to care a damn about what happens to you
(Note that the asterisk on (65b) indicates that it is not grammatical on the intended
idiomatic interpretation of ‘He would do something to help you.’) Expressions which
are restricted in this way are referred to as polarity expressions, because they seem to
111
Syntactic structure
have an inherent affective polarity (in the sense that they are restricted to occurring in
affective contexts).
However, it’s not enough simply to say that a polarity expression must occur in a
phrase or sentence containing an affective constituent, as contrasts such as the follow¬
ing illustrate:
(66) (a) The fact that he has resigned won’t change anything
(b) *The fact that he hasn’t resigned will change anything
In both cases, the polarity item anything occurs after a negative auxiliary (after won’t in
(66a), and after hasn’t in (66b)), and yet only the first of these two examples is gram¬
matical. Why? The traditional answer is that anything falls within the scope (i.e.
‘sphere of influence’) of the negative auxiliary won’t in (66a), but does not fall within
the scope of the negative auxiliary hasn’t in (66b). Implicit in this account is the
assumption that a polarity item must fall within the scope of an affective constituent;
and it has been suggested that the relative scope of constituents can be defined in terms
of the structural relation c-command (a conventional abbreviation of ‘constituent-
command’). Given these assumptions, the restriction on the distribution of polarity
items can be given the following structural characterization:
The relation c-command can be defined configurationally (i.e. in terms of the relative
positions occupied by the constituents concerned within a given tree) in the following
manner:
(If you prefer to think of structural relations in terms of networks of train stations, (68)
amounts to claiming that X c-commands Y if you can get from X to Y by catching a
northbound train, getting off at the first station and catching a southbound train on a
different line - i.e. you can’t travel south on the line you travelled north on.)
To see how the c-command condition in (67) works, consider the structure of (66a)
above given in (69) below:
112
3.9 C-command
The crucial question here is whether the pronoun anything is c-commanded by the
negative auxiliary won't. Since the mother of the I node containing won’t is the encir¬
cled I-bar node containing won’t change anything, and since the relevant I-bar node
dominates the PRN node containing anything (and the relevant I and PRN nodes are
disconnected), it follows that won’t does indeed c-command anything in (69). (You can
reach the same conclusion if you think of (69) as a network of train stations: you
can get from the I-node containing won’t to the PRN node containing anything by
travelling north, getting off at the station represented by the I-bar node containing
won't change anything, then catching a southbound train to the station represented by
the PRN node containing anything.) Since anything is c-commanded by won’t in (69),
the c-command condition (67) is satisfied, and sentence (66a) is correctly predicted to
be grammatical.
Now consider why, by contrast, (66b) is ungrammatical. Example (66b) has the
structure (70) below:
hC | I
hasn’t resigned
113
Syntactic structure
Since the only affective constituent in (70) is the negative auxiliary hasn’t, the question
which we need to ask here is whether the polarity expression anything is c-commanded
by hasn’t. The mother of the I-node containing hasn't is the encircled I-bar node con¬
taining hasn’t resigned: since the relevant I-bar node does not dominate the PRN node
containing anything, hasn’t doesn’t c-command anything. But since there is no other
affective constituent in the sentence, this in turn means that anything does not fall
within the scope of an affective constituent, thereby violating the c-command condi¬
tion in (67); hence, we correctly predict that (66b) is ungrammatical.
The overall conclusion which we reach is that the restricted distribution of polarity
expressions can be given a configurational characterization in terms of the relation
c-command. Of course, it goes without saying that such an account is only tenable if
we suppose that sentences have a hierarchical categorial constituent structure: in other
words, the fact that the restriction on polarity items can be given a straightforward
characterization in structural terms lends empirical support to the more general claim
that sentences have an abstract syntactic structure which can be represented in the form
of tree diagrams like (69) and (70).
A second class of expressions whose distribution can be accounted for in structural
terms are so-called anaphors. These include reflexives (i.e. +self forms) like
myself/yourselflthemselves etc., and reciprocals like each other and one another. Such
anaphors have the property that they cannot be used to refer directly to an entity in the
outside world, but rather must be bound by (i.e. take their reference from) an
antecedent elsewhere in the same phrase or sentence. Where an anaphor has no (suit¬
able) antecedent to bind it, the resulting structure is ungrammatical - as we see from
contrasts such as that in (71) below:
In (71a), the third person masculine singular anaphor himself is bound by a suitable
third person masculine singular antecedent (he), with the result that (71a) is grammati¬
cal. But in (71b), himself has no suitable antecedent (the feminine pronoun she is
obviously not a suitable antecedent for the masculine anaphor himself), and so is
unbound (with the result that (71b) is ungrammatical). In (71c), there is no potential
antecedent of any kind for the anaphor himself, with the result that the anaphor is again
unbound.
There seem to be structural restrictions on the binding of anaphors by their
antecedents, as we can illustrate in terms of the following contrasts:
114
3.9 C-command
As a third person masculine singular anaphor, himself must be bound by a third person
masculine singular antecedent like the president, similarly, as a plural anaphor, each
other must be bound by a plural antecedent like they/them. However, it would seem
from the contrasts above that the antecedent must occupy the right kind of position
within the structure in order to bind the anaphor, or else the resulting sentence will be
ungrammatical. The question of what is the right position for the antecedent can be
defined in terms of the following structural condition:
The relevant bound constituent is the reflexive anaphor himself in (72), and its
antecedent is the president, the bound constituent in (73) is the reciprocal anaphor each
other, and its antecedent is they/them. Let’s also suppose that (72a) has the structure
(75) below, and that the DP the president is the antecedent of the anaphoric PRN himself
(as is indicated below by the fact that the two carry the same subscript letter index i):
of D N congratulate himself
the president
115
Syntactic structure
The answer is that the DP node containing the president doesn’t c-command the PRN
node containing himself, because the mother of DP is the PP node, and PP doesn t
dominate PRN. Since there is no other potential antecedent for himself within the
sentence (e.g. although the NP supporters of the president c-commands himself it is
not a suitable antecedent because it is a plural expression, and himself requires a singu¬
lar antecedent), the anaphor himself remains unbound - in violation of the c-command
requirement. Thus, (72b) is correctly predicted to be ungrammatical. However, the
binding condition which the sentence violates is one which crucially involves the
structural relation c-command - and this relation is defined in terms of the relative
structural positions occupied by the anaphor and its antecedent. The fact that the
relevant restriction on the binding of anaphors can be given a straightforward charac¬
terization in structural terms provides further (indirect) support for our claim that
sentences have a hierarchical constituent structure. There’s much more to be said about
binding, though we shan’t pursue the relevant issues here. (For a recent discussion, see
Roberts 1996, chapter 3: don’t attempt to read it until you have finished this book.)
3.10 Summary
In this chapter, we have looked at how words are combined together to
form phrases and sentences, and at how we can represent the structure of the phrases
and sentences thereby formed. In §3.2 we hypothesized that phrases and sentences are
formed by a binary merger operation which combines pairs of constituents to form
larger and larger structures. We suggested that one way of forming a larger structure is
to project (i.e. expand) a head word into a phrase by combining it with a following
complement, so that e.g. by combining the verb privatize with the complement noun
hospitals, we form the verb phrase privatize hospitals, and by combining this verb
phrase with the infinitive particle to we in turn form the infinitive phrase to privatize
hospitals. In §3.3 we argued that some phrases have a more complex structure and
contain a specifier as well as a head and a complement: we suggested that in a noun
phrase such as government plans to privatize hospitals, the head noun plans merges
with the infinitive phrase to privatize hospitals to form the intermediate nominal con¬
stituent (N-bar) plans to privatize hospitals, and that the resulting N-bar then merges
with its specifier (the noun government) to form the full noun phrase government plans
to privatize hospitals. In §3.4 we suggested that sentences are formed by the same
binary merger operation as phrases: for example, a sentence such as The government
might privatize hospitals is formed by merging the past-tense-inflected auxiliary (= I)
might with the verb phrase complement privatize hospitals to form the intermediate
auxiliary projection (= I-bar) might privatize hospitals', the resulting I-bar is then pro¬
jected into an IP (i.e. complete auxiliary phrase) by merging it with the determiner
phrase the government which serves as its specifier (or, more traditionally, its subject),
thereby forming the clause or sentence The government might privatize hospitals. We
116
Workbook section
noted that the resulting IP can in turn be merged with a complementizer like that to
form the complementizer phrase (CP) That the government might privatize hospitals
(which could be used e.g. as a reply to a sentence such as What’s the latest rumour?).
In §3.5 we looked at the structure of tree diagrams, and presented a simple (break-off-
a-branch) procedure for determining which strings of words in a tree do and don’t form
constituents. In §3.6 we looked rather more closely at the formal properties of syntactic
trees, suggesting that they are sets of nodes interconnected by precedence and domi¬
nance relations. We showed how terms such as complement and specifier can be
defined configurationally (i.e. in relation to the position occupied by a given expression
in a given structure). In §3.7 we noted that tree diagrams embody specific claims about
syntactic structure which must be tested empirically; and we argued that simple coordi¬
nation and shared string coordination could be used as ways of testing structure. In
§3.8, we presented further constituent structure tests relating to sentence fragments,
proforms and ellipsis. In §3.9 we saw that the distribution of affective polarity items
and anaphors could be characterized in structural terms (via the relation c-command),
so providing further evidence for positing that sentences have a hierarchical categorial
constituent structure.
Workbook section
Exercise I (§§3.2-3.8)
In the text, we argued that the terms constituent, specifier and comple¬
ment can be defined configurationally. In the light of the text definitions, look at the
structure immediately below:
DP
/ \
D N
The Dean
hold D N
a meeting
and answer the following questions about this structure (giving reasons for your
answer):
1 Is a meeting a constituent, and if so of what kind?
2 Is hold a meeting a constituent, and if so of what kind?
117
Syntactic structure
118
Workbook section
Exercise II (§§3.2-3.8)
Discuss the derivation of the following sentences, showing how their
structure is built up in a pairwise fashion by successive merger operations, and show¬
ing how constituent structure tests can be used to support your analysis.
Helpful hint
Assume that the sentences are derived by first merging the last two words
in the sentence to form a constituent, then merging the constituent thereby formed with
the third-from-last word to form an even larger constituent, then merging this even
larger constituent with the fourth-from-last word . . . and so on. (It should be noted, how¬
ever, that while this simple parsing procedure will work with the sentences in this exer¬
cise, it requires modification to handle more complex sentences.)
P PRN
| I
to her
Merging the PP in (i) with the verb writing immediately to its left in turn forms the
verb phrase (ii) below:
119
Syntactic structure
(ii)
I / \
writing P PRN
to her
Merging the VP in (ii) with the progressive auxiliary been derives the progressive aux¬
iliary phrase (iii) below:
(iii) PROGP
been V PP
I / \
writing P PRN
I I
to her
Merging the PROGP in (iii) with the finite (third person singular present tense) auxil¬
iary has forms the incomplete intermediate projection has been writing to her (which is
incomplete in the sense that it cannot stand on its own as a sentence fragment, but
rather requires a subject like he). If finite auxiliaries are I/INFL constituents, and inter¬
mediate auxiliary projections have the status of I-bar constituents, merging has with
the VP in (iii) will derive the I-bar (iv) below:
(iv)
PP
\
writing P PRN
I I
to her
Merging the I-bar in (iv) with the subject pronoun he will in turn derive the IP (i.e.
inflected auxiliary phrase) (v) below:
(v)
PP
/ \
writing P PRN
to her
120
Workbook section
On this view, syntactic structures are derived in a bottom-up fashion, i.e. by building
up trees in successive layers from bottom to top.
Evidence in support of positing that the PP, VP, PROGP and I-bar in (v) are con¬
stituents comes from simple coordination facts in relation to sentences such as (vi)
below:
Moreover (for reasons which should be evident to you), the fact that the italicized
sequences in (vii) below can serve as the shared string in the appropriate kind of coor¬
dinate structure provides further empirical evidence in support of the same conclusion
(bold print marks contrastive stress):
Furthermore, as we can see from (viii) below, the PP, VP and PROGP constituents (by
virtue of being maximal projections) can serve as sentence fragments:
(though (viii)(b) is more concise than (viii)(c), and so preferred). In addition, VP and
PROGP can serve as the antecedent of a pronoun like which: cf.
(ix) (a) If he has been writing to her (which he has), why doesn’t she write
back?
(b) If he has been writing to her (which he has been), why doesn’t she write
back?
and VP and PROGP (by virtue of being the complements of auxiliaries) can undergo
ellipsis, as we see from (x) below:
(x) (a) I wonder why he has been writing to her as often as he has been
(b) I wonder why he has been writing to her as often as he has
121
Syntactic structure
9 NP 10 PRNP
jS \
/ \
N IP PRN IP
I / I / X
John I VP They I QP^
11 NP 12 PRNP
/ \
N IP. PRN IP
I
Congress 1 VP. si
She r/
I X-
y
\ i / \
DP must v ap
expecting D NP be A PP
I / \ | / \
the N IP keen P PRNP
president I VP
I / \
for PRN IP
I / \ I / \
to V P him I V
I I I I
back down to win
122
Workbook section
13 DP 14 IP
D NP PRN I-
I
The N IP You VP
I / \
bullet VP might V PP
I
has AD VP fall P
I
gone ADV PP out P DP
I I
right DP of D N
I I
through D N the window
I
his skull
Secondly, 9 provides no principled account of the fact that has agrees with John, since
the two are contained within different projections (John within NP and has within IP),
and agreement typically involves a local relation between the head and specifier of the
same projection (typically, IP). Thirdly, 9 wrongly predicts that the string taking drugs
should not be able to serve as a sentence fragment, since it is not a maximal projection
(it is a V-bar, not a VP); but this prediction is wrong, as we see from the grammaticality
of speaker B’s reply in (iii) below:
(iv)
taking drugs
The analysis in (iv) correctly predicts that (as a finite IP), the relevant structure can be
used as the complement of a finite complementizer like if or that: cf.
123
Syntactic structure
Moreover, since has is the head and John the specifier of IP in (iv), we correctly predict
that the two enter into a local specifier-head agreement relation. And since taking drugs
is a maximal projection (more specifically, a VP) in (iv), we correctly predict that it can
serve as a sentence fragment, as in (iii).
Exercise IV (§§3.2-3.8)
Below are a number of alternative analyses of the structure of the sen¬
tence He might have been watching television which have been proposed in the lin¬
guistic literature over the past four decades. (Some of the category labels have been
adapted to make them compatible with those used here.)
-S-
PRN VERB NP
AUX' 1
He "'V N
^ 1 ^\ |
1
1
PERF PROG watching television
|1 |
1 1
might have been
-s
np
|
r PREDP
NP AUX VP
|
PRN ASP V NP
1 1 / \ 1 1
He might PERF PROG watching N
11 1i 1
have been television
124
Workbook section
NP
I
PRN
I
He have been watching N
television
NP
I
PRN PROGP
I
He
NP
I I
watching N
television
(Some of the less familiar abbreviations used above are as follows: S = sentence/clause;
M = modal auxiliary; ASP = aspectual auxiliary; PREDP = predicate phrase.) Comment
on significant structural differences between the analysis for the relevant sentence
proposed in (39) in the text, and those in 1-5 above. In addition, for each of the five
analyses outlined above, say whether it would account for the (un)grammaticality of the
sentences in (44), (49), (53), (54), (59) and (61) in the text.
125
Syntactic structure
all maximal projections were phrases, and hence that (for example) a noun or pronoun
could not serve as a maximal projection unless it projected into NP).
The exercise analysis in 1 above correctly predicts that sentences like (44d) in the
text are grammatical (since the string might have been watching television is analysed
as a VP which can therefore be coordinated with another similar VP), but wrongly pre¬
dicts that (44a-c) are ungrammatical by virtue of the fact that the strings watching tele¬
vision, been watching television and have been watching television are not constituents
according to the exercise analysis 1. In much the same way, the exercise analysis in 1
correctly predicts that (49d) in the text is grammatical (since the shared string might
have been watching television is analysed as a VP constituent), but wrongly predicts
that (49a-c) are all ungrammatical (because the shared string in each case is a noncon¬
stituent sequence). Moreover, the exercise analysis 1 wrongly predicts that watching
television, been watching television and have been watching television cannot be used
as sentence fragments in (53) in the text (since they are not constituents), and
conversely wrongly predicts that the string might have been watching television should
be able to function as a sentence fragment in (54) in the text, since it is a VP constituent
and hence a maximal projection. In much the same way, the exercise analysis 1
wrongly predicts that might have been watching television should be able to function
as the antecedent of so in (59) in the text (since it is a VP), and also wrongly predicts
that the strings watching television, been watching television and have been watching
television should not (since they are nonconstituents). Finally, the exercise analysis 1
correctly predicts that might have been watching television can undergo ellipsis in
(61) in the text, but wrongly predicts that watching television, been watching television
and have been watching television cannot. Overall, then, the exercise analysis 1 is
descriptively inadequate, since it makes grossly inaccurate predictions about the
(un)grammaticality of the relevant sentences.
Exercise V (§§3.2-3.9)
As noted in the text, a number of words in English are said to be polarity
items, in the sense that they seem to have an inherently affective polarity, and so are
restricted to occurring in a position where they are c-commanded by an affective (e.g.
negative or interrogative) constituent. One such polarity item (as we saw in the text) is
existential any (and its compounds like anyone!any thing/anywhere). Two other polar¬
ity items are need and dare, when used with a bare (to-less) infinitive complement, as
in I don t think he dare/need say anything. (Note that when need/dare are used with a to
complement as in He needs to see a specialist, they are not polarity items.) Show how
the c-command condition would account for the (un)grammaticality of the following:
126
Workbook section
Helpful hints
Assume that need/dare (when they take a bare infinitive complement) are
modal auxiliaries which occupy the head I position of IP, and that they take a VP com¬
plement. Assume that pronominal quantifiers like any one/anything/anywhere/nobody
belong to the category PRN. Likewise, assume that no in 8 is a quantifier (= Q) which
heads a quantifier phrase (= QP) constituent which has a noun phrase (= NP) comple¬
ment. Finally, assume that mustn't/wouldn’t/don’t!couldn’t are negative auxiliaries
occupying the head I position of IP.
I / \
talk P PRN
I I
to anyone
The I node containing the negative auxiliary mustn’t here c-commands the PRN node
containing the polarity item anyone, because the two are disconnected, and the mother
of the I node is I-bar, and I-bar dominates [pRN anyone]. Thus, since the polarity item
anyone is c-commanded by the negative auxiliary mustn’t, the c-command condition on
the use of polarity items is satisfied, so sentence 1 is grammatical.
Exercise VI
A distinction which we overlooked in the text is that between phrases and
compound words. It is traditionally claimed that an expression such as prone to acci¬
dents is an adjectival phrase with the structure la below, whereas an expression such as
accident prone is a compound adjective with the structure lb:
127
Syntactic structure
to accidents
Let’s say that in both cases, prone is the head of the phrase or (compound) word con¬
taining it (which is why both expressions are adjectival in nature), and that accident(s)
is its complement noun. (Strictly speaking, we should say that the complement of
prone in lb is the prepositional phrase to accidents.) Try and identify the phonological,
morphological and syntactic differences between compounds and phrases illustrated
in the examples below (note that the capital letters in 2 mark syllables which carry
primary stress):
In addition, draw tree diagrams representing the structure of the two expressions in
(7a-b) below:
7a manufacturers of pianos
b piano manufacturers
and highlight phonological, morphological and syntactic differences between the two.
Helpful hints
Look at the stress properties of the head and its complement; at the num¬
ber (singular/plural) properties of the complement noun; at whether the relevant struc¬
tures are right- or left-headed (i.e. whether the head word is positioned on the left or
right of the structure); at whether the complement can be a pronoun rather than a noun;
and at whether the complement noun can be extracted (i.e. moved) out of its containing
expression. Don’t attempt to draw tree diagrams representing the structure of sentences
(2-6), since some parts of these sentences involve constructions we haven’t yet looked
at: focus on the adjectival structures we are concerned with.
128
Workbook section
In the text, we noted that the syntactic structure of any phrase or sentence
can be represented either by a labelled tree diagram or by a labelled bracketing; each
constituent is represented by a single node in a tree diagram, but by a pair of brackets
(the lefthand one of which carries the relevant category label) in a labelled bracketing.
The labelled bracketings given below represent the structure of a number of different
sentences. Label each of the righthand brackets in the structure, and then convert each
of the labelled bracketings into an equivalent tree diagram.
(Note that the dots (...) in 2-4 are simply a graphic device to indicate that the structure
continues on (or from) the following line: thus, each structure in 2-4 should be treated
as if it were on one continuous line.)
The lefthand bracket round we marks the beginning of the pronoun constituent we, and
the righthand bracket marks the end of the same PRN constituent. Proceeding from left
to right, we now look for the next unlabelled righthand bracket (= the bracket immedi¬
ately following have) and we pair it with the nearest unpaired lefthand bracket to its
left (= the I bracket immediately preceding have) - as in (ii) below:
129
Syntactic structure
We then find the next unlabelled righthand bracket (= that immediately following
made), and we pair it with the nearest unpaired lefthand bracket to its left (= the V
The next unlabelled righthand bracket in the structure (iii) is that immediately follow¬
ing the word progress, which we pair with the nearest unpaired lefthand bracket to its
left (= the N bracket to the immediate left of progress), so resulting in:
We now find the next unlabelled righthand bracket, which is the third bracket from the
end of the overall structure. We then pair it with the nearest unpaired lefthand bracket
to its left (i.e. that labelled VP), as in (v):
(v) [jp [pRN We PRfj] [y [| have j] [yp [y made y] progress j^] yP]]]
The next unlabelled righthand bracket in (v) is the last-but-one bracket: the nearest
unpaired bracket to its left is the I-bar bracket preceding have, so the two are paired as
in (vi) below:
(vi) [jP [pRjsj We pRj^] [y [j have j] [yp [y made y] progress j^] yp] y]]
The one remaining unlabelled righthand bracket (= the rightmost bracket) is then
paired with the one remaining unpaired lefthand bracket (= the leftmost IP bracket), as
in (vii) below:
(vh) tip [prn We prn^ tf li have J [yp [y made y] progress ^] yp] y] jP]
What (vii) tells us is (for example) that the VP comprises all the constituents which are
contained within the lefthand and righthand VP brackets - i.e. the verb made and the
noun progress. The resulting structure (vii) can equivalently be represented in terms of
the tree diagram in (viii) below:
(viii)
PRN
I
We
made progress
I leave it to you to devise an algorithm (i.e. set of procedures) of your own for convert¬
ing a labelled bracketing into a labelled tree diagram, and conversely for converting a
labelled tree diagram into a corresponding labelled bracketing: this is an interesting
exercise in itself.
130
4
Empty categories
4.1 Overview
So far, our discussion of syntactic structure has tacitly assumed that all
constituents in a given structure are overt. However, we now turn to argue that syntac¬
tic structures may also contain empty (= covert = null) categories - i.e. categories
which have no overt phonetic form, and hence which are inaudible or silent. As we
shall see, empty categories play a central role in the theory of grammar which we are
outlining here.
Each of the bracketed infinitive complement clauses in the (a) examples in (1-3)
contains an overt (italicized) subject. By contrast, the bracketed complement clauses in
the (b) examples appear to be subjectless. However, we shall argue that apparently
subjectless infinitive clauses contain an understood null subject. (By saying that a con¬
stituent is null or empty or covert, we mean that it has no overt phonetic form and so is
silent.) The kind of null subject found in the bracketed clauses in the (b) examples
has much the same grammatical and referential properties as pronouns, and hence is
conventionally designated as PRO. (The fact that English allows infinitives to have
null subjects does not mean that English is a null subject language, since null subject
languages are languages which allow finite verbs to have a null subject.)
131
Empty categories
Given this assumption, sentences such as (la-b) have essentially the same structure,
except that the bracketed IP has an overt pronoun you as its subject in (la), but a covert
pronoun subject PRO in (lb) - as represented in (4a-b) below:
(4)
Using the relevant technical terminology, we can say that the null subject PRO in (4b)
is controlled by (i.e. refers back to) the subject we of the matrix (= containing = next
highest) clause - or, equivalently, that we is the controller or antecedent of PRO.
Verbs (such as like) which allow an infinitive complement with a PRO subject are said
to function (in the relevant use) as control predicates. An important corollary of the
analysis in (4b) is that apparently subjectless infinitive clauses have essentially the
same structure as infinitive clauses with overt subjects, in that both are IPs: the two
types of infinitive clause differ only in respect of whether the infinitive subject is an
overt or covert pronoun.
An obvious question to ask at this juncture is why we should posit that apparently
subjectless infinitive complements like those bracketed in (1—3b) above have a null
PRO subject. Part of the motivation for positing PRO is semantic in nature. In tradi¬
tional grammar it is claimed that subjectless infinitive clauses have an understood or
implicit subject - and positing a PRO subject in such clauses is one way of capturing
the relevant intuition. The implicit subject becomes explicit if the relevant clauses are
paraphrased by a finite clause, as we see from the paraphrases for the (a) examples
given in the (b) examples below:
The fact that the bracketed clauses in the (b) examples contain an overt (italicized) sub¬
ject makes it plausible to suppose that the bracketed clauses in the synonymous (a)
examples have a covert PRO subject.
132
4.2 PRO subjects
In the case of structures like (8), a local antecedent means ‘an antecedent contained
within the same [bracketed] clause as the reflexive’. Example (8a) is grammatical
because it satisfies this locality requirement: the antecedent of the reflexive himself is
the noun John, and John is contained within the same [bracketed] /ie/p-clause as him¬
self. By contrast, (8b) is ungrammatical because the reflexive themselves does not have
a local antecedent (i.e. it does not have an antecedent within the bracketed clause
containing it); its antecedent is the pronoun they, and they is an immediate constituent
of the want clause, not of the [bracketed] help-clause. In the light of the requirement
for reflexives to have a local antecedent, consider now how we account for the gram-
maticality of the following:
Given the requirement for reflexives to have a local antecedent, it follows that the
reflexive himself must have an antecedent within its own [bracketed] clause. This
requirement is satisfied in (9) if we assume that the bracketed complement clause has a
PRO subject, and that PRO is the antecedent of himself. Since PRO in turn is
controlled by John (i.e. John is the antecedent of PRO), this means that himself is
coreferential to (i.e. refers to the same person as) John.
We can formulate a further argument in support of positing a PRO subject in appar¬
ently subjectless infinitive clauses in relation to the syntax of predicate nominals:
these are nominal (i.e. noun-containing) expressions used as the complement of a
copular (i.e. linking) verb such as be, become, remain (etc.) in expressions such as
John was/became/remained my best friend, where the property of ‘being my best
friend’ is predicated of John. Predicate nominals in copular constructions have to agree
with the subject of their own clause, as we see from examples such as the following:
As examples like (10) illustrate, the italicized predicate nominal has to agree with the
(bold-printed) subject of its own [bracketed] become-clause, and cannot agree with the
subject of the want-clause. In the light of this clause-internal agreement requirement,
consider now how we account for the agreement pattern in (11) below:
133
Empty categories
If we posit that the become clause has a PRO subject which is controlled by the subject
of the want clause, the relevant agreement facts can be accounted for straightfor¬
wardly: we simply posit that the predicate nominal (a) millionaire(s) agrees with PRO
(since PRO is the subject of the become clause), and that PRO in (11a) is plural
because its controller/antecedent is the plural pronoun they, and conversely that PRO
in (1 lb) is singular because its antecedent/controller is the singular pronoun he. It goes
without saying that it is far from obvious how we would handle the relevant agreement
facts if we didn’t posit a PRO subject for the bracketed infinitive complements in
sentences such as (11).
There is a range of additional empirical evidence which supports the claim that
apparently subjectless infinitive clauses have an understood PRO subject. Some of this
evidence is given in the paradigms (12-15) below:
The examples in (12a-b) suggest that in expressions like one’s own, the possessive
one’s requires a local antecedent: this requirement will only be met in (12c) if the
bracketed infinitive complement has a PRO subject which is controlled by John. The
examples in (13a—b) show that in expressions such as lose one’s cool, the possessor
one's must agree with the subject of lose-, this requirement will only be met in (13c) if
we assume that the bracketed lose clause contains a PRO subject controlled by I. The
examples in (14a—b) and (15a—b) show that expressions like each other and together
require a local plural antecedent; this requirement will only be met in (14c) and (15c) if
we posit that the bracketed complement clause has a PRO subject which is plural by
virtue of the fact that it is controlled by the plural expressions the congressmen/they.
Examples such as (5-15) above provide empirical evidence in support of positing
that apparently subjectless infinitive clauses have a PRO subject. In examples such as
134
4.3 Null auxiliaries
those discussed above, PRO has a controller in a higher clause (generally the subject of
the higher clause). However, in examples such as (16) below, PRO has no obvious con¬
troller within the sentence containing it:
(16) (a) [Why PRO kid myselffl I know she doesn’t love me any more
(b) [PRO kill himself]?! He hasn’t got the guts!
(c) It’s important [PRO to rid ourselves of private prejudices]
(d) speaker a: What is their greatest ambition?
speaker b: [PRO to see themselves on television]
Clearly we need to posit a PRO subject for the bracketed clauses in (16) in order to sat¬
isfy the requirement that each of the italicized reflexives have a local antecedent. In
examples like (16), PRO has no antecedent within the sentence containing it; instead,
the reference of PRO seems to be discourse-determined (i.e. PRO seems to denote
some specific individual or set of individuals who is talking, being talked to or being
talked about in the relevant discourse). By contrast, in sentences such as (17) below:
(17) It’s wrong [PRO to blame oneself for other people’s misfortune]
PRO seems to denote an arbitrary individual (i.e. it means ‘any arbitrary person you
care to choose’), and thus might be said to have arbitrary reference. There is a great
deal more that could be said about the referential properties of PRO, but we won’t
pursue the matter further here.
Both clauses here (viz. the /e/f-clause and the stayed-clause) appear to be finite, since
both have nominative subjects (he/she). If all finite clauses contain an IP projection
headed by a finite INFL, it follows that both clauses in (18) must be IPs containing a
finite I/INFL constituent. This is clearly true of the left-clause, which contains the finite
modal auxiliary could; however, the stayed-clause doesn’t seem to contain any finite
auxiliary constituent, since have is an infinitive form (the corresponding finite form
being has). How can we analyse finite clauses as projections of an INFL constituent
when clauses like that bracketed in (18) contain no finite auxiliary?
An intuitively plausible answer is to suppose that the string she have stayed in (18)
is an elliptical variant of she could have stayed, and that the I constituent could under-
135
Empty categories
goes head-ellipsis (alias gapping) in the second clause. If this is so, then the second
clause will have the structure (19) below (where e marks the ellipsed auxiliary):
(19)
PRN
she
The head I position of IP would then be filled by an ellipsed auxiliary. We can think of
ellipsis as a process by which a constituent (in this case, could) is given a null phonetic
form, but retains its grammatical and semantic properties (so that e in (19) is a silent
counterpart of could). The null auxiliary analysis in (19) provides a principled account
of three sets of facts. Firstly, the bracketed clause in (18) is interpreted as an elliptical
form of she could have stayed: this can be straightforwardly accounted for under the
analysis in (19), given that e is an elliptical form of could. Secondly, the subject is in
the nominative case form she: this can be attributed to the fact that the I position in (19)
is filled by a null counterpart of could and so is finite (and thereby requires a nomina¬
tive specifier). Thirdly, the perfective auxiliary have is in the uninflected infinitive
form: this is because e (being an elliptical form of could) has the same grammatical
properties (hence the same complement-selection properties) as could, so that e (like
could) requires a complement headed by a word (like have) in the infinitive form.
A further argument in support of the analysis in (19) comes from facts relating to
cliticization (a process by which one word attaches itself in a leechlike fashion to
another). The perfective auxiliary form have has a range of different allomorphs (i.e.
variant forms) in the spoken language. When unstressed, it loses its initial /h/ segment
and has its vowel reduced to schwa hi, and so is pronounced as hvl e.g. in sentences
such as You should have been there. (Because of is also pronounced hvl when
unstressed, some people write this as You should of been there - not you, of course!)
However, when have is used with a subject ending in a vowel or diphthong (e.g. a sub¬
ject pronoun like I/we/you/they), it can lose its vowel entirely and be contracted down
to /v/; in this form, it is phonetically too insubstantial to survive as an independent
word, and cliticizes (i.e. attaches itself) to its subject, e.g. in structures such as:
136
4.4 Unfilled INFL
Why should cliticization of have onto she be blocked here? The null auxiliary analysis
in (19) provides us with an obvious answer, if we make the reasonable assumption that
have can only cliticize onto an immediately preceding word ending in a vowel or
diphthong. We can then say that have is blocked from cliticizing onto she in (19) by the
presence of the null auxiliary which intervenes between have and she. Thus, the null
auxiliary analysis seems entirely appropriate for elliptical finite clauses like the stayed-
clause in (18) - i.e. clauses which contain an ellipsed auxiliary.
A rather different kind of null auxiliary constituent is found in African American
English (AAE), in sentences such as the following (from Labov 1969, p. 717):
In AAE, specific forms of be have null variants (so that we find null allomorphs of are
and is in contexts where Standard English (SE) would require the contracted forms ’s
and ’re). Hence, in place of SE He’s getting crippled we find AAE He gettin cripple
(with a null counterpart of ’s). Evidence in support of the assumption that AAE
sentences like (22) incorporate a null variant of is comes from the fact that the missing
auxiliary is may surface in a tag, as in sentences such as the following (where the
sequence following the comma is the tag):
Given the general assumption that the auxiliary found in the tag is a copy of the
auxiliary used in the main clause, it follows that the main gonna-clause in (23) must
contain a null variant of is. Interestingly, the form am (contracted to ’m) has no null
counterpart in AAE, nor do the past-tense forms was/were. It would seem, therefore,
that the only finite forms of be which have a null counterpart in AAE are the specific
auxiliary forms are and is. No less interestingly, Wolfram (1971, p. 149) reports that in
nonstandard southern white English the use of null auxiliaries is even more restricted,
and that the only form of be with a null counterpart is are, not is\ cf. the parallel obser¬
vation by Fasold (1980, p. 30) that ‘There are many southern whites who delete only
are.'
Why not treat auxiliariless finite clauses in much the same way as we analysed struc¬
tures like (19) which contain an ellipsed auxiliary - namely as IP structures headed by
137
Empty categories
a null I constituent? More specifically, let’s suppose that auxiliaries finite clauses like
those in (24) above are IP constituents which have the structure (25) below:
(25)
The fact that the INFL node in (25) contains no leaf (not even an invisible empty one)
is meant to represent the assumption that the head I position of IP is simply unfilled;
that it is to say, the INFL position in (25) is neither occupied by an overt item, nor
occupied by a covert item - it is simply unoccupied.
However, one apparent problem posed by the analysis in (25) is that of how we
account for the fact that he agrees with hates. After all, agreement (e.g. between he and
has in He has gone) typically involves a local (phrase-internal) spec(ifier)-head
relation between INFL and its specifier: since the verb hates is an immediate con¬
stituent of VP and he is an immediate constituent of IP, the two are clearly not in a local
specifier-head relation, so would not be expected to agree. How can we deal with this
problem? One possible solution would be to suppose that when INFL is unfilled, the
tense and agreement properties of the head V of VP percolate up to INFL. Since hates
is a third person singular present-tense verb form, the features of INFL (inherited from
hates) will indicate that it is a third person singular present-tense form, and so will
require it to have a third person singular nominative subject. Thus, the fact that the
subject in (25) surfaces in the nominative form he and agrees with the verb hates is
unproblematic.
‘But why on earth would we want to pretend that clauses which obviously don’t
contain an auxiliary actually contain an unfilled auxiliary position?’ you might wonder.
Well, from a theoretical point of view, an obvious advantage of the unfilled INFL
analysis is that it provides a unitary characterization of the syntax of clauses, since it
allows us to say that all clauses contain an IP projection, and that the subject of a clause
is always in spec-IP (i.e. always occupies the specifier position within IP), that INFL
in a finite clause always has a nominative subject, and always agrees with its subject.
Lending further weight to theory-internal considerations such as these is a substantial
body of empirical evidence. A direct consequence of the unfilled INFL analysis (25) of
auxiliaries finite clauses is that finite auxiliaries and finite verbs occupy different
positions within the clause: finite auxiliaries (being functors) occupy the head I posi¬
tion of IP, whereas finite nonauxiliary verbs (being contentives) occupy the head V
position of VP. An interesting way of testing this hypothesis is in relation to the behav-
138
4.4 Unfilled INFL
iour ot polycategorial items which have the status of auxiliary verbs in some uses, but
of nonauxiliary verbs in others. One such word is have. In the kind of use illustrated in
(26) below, it functions as a perfective auxiliary:
However, in the uses illustrated in (27) below, have functions as a causative or experi¬
ential verb (i.e. a verb carrying much the same meaning as ‘cause’ or ‘experience’):
How can we account for this contrast? If we assume that perfective have in (28a) is a
finite auxiliary which occupies the head I position of IP, but that causative have in
(28b) and experiential had in (28c) are nonauxiliary verbs occupying the head V posi¬
tion of a VP complement of an unfilled INFL, then prior to cliticization the three
clauses will have the respective (simplified) structures indicated by the labelled brack¬
etings in (29) below (where f,_] denotes an INFL position which is unfilled):
139
Empty categories
(In labelled bracketings like (29) which give a partial representation of structure, we
omit intermediate projections like I-bar and V-bar in order to simplify exposition: of
course, if we wanted to be more precise, we should indicate that e.g. seen a ghost is a
V-bar in (29a), and have seen a ghost is an I-bar.) If /zuve-cliticization is subject to an
adjacency condition (to the effect that cliticization is only possible when have immedi¬
ately follows the expression to which it cliticizes and is blocked by the presence of an
intervening constituent), it should be obvious why have can cliticize onto they in (29a)
but not have!had onto they!she in (29b-c): after all, have is immediately adjacent to
they in (29a), but have/had is separated from they/she by a null INFL constituent in
(29b-c). A crucial premise of this account is the assumption that (in its finite forms)
have is positioned in the head I of IP in its perfective use, but in the head V of a VP
complement of an IP headed by an unfilled I in its causative and experiential uses.
Thus, have cliticization facts suggest that finite clauses which lack a finite auxiliary are
IPs headed by an unfilled I constituent.
A further piece of empirical evidence in support of the IP analysis comes from tag
questions. As we see from the examples below, sentences containing (a finite form of)
perfective have are tagged by have, whereas sentences containing (a finite form of)
causative have are tagged by do:
Given the I-analysis of perfective have and the V-analysis of causative have and the
assumption that all clauses are IP constituents, the main clauses in (30a-b) will have
the respective (simplified) structures indicated in (31a-b) below (where_denotes an
unfilled position):
If we assume that the I constituent which appears in the tag must carry the same semantic
and grammatical properties as the I constituent in the main clause, the contrast in (30) can
be accounted for in a principled fashion. In (31a), the head I position of IP is filled by the
perfective auxiliary has, and so the tag contains a copy of this auxiliary. In (31b),
however, the head I position of IP is unfilled (and hence contains no meaning-bearing
constituent), and so can only be tagged by the dummy auxiliary does (which carries the
same present-tense feature as the unfilled I constituent in the main clause). Note, inciden¬
tally, that the assumption that auxiliaries in tags carry the same grammatical properties as
the I constituent in the main clause provides us with evidence for positing that the
unfilled I constituent in a structure such as (31b) carries tense-features (since these are
copied in the auxiliary in the tag); the relevant tense-features percolate up from V to I
when I is unfilled, so that I in (31b) inherits the present-tense feature carried by has.
140
4.4 Unfilled INFL
Thus, the different behaviour of perfective and causative have provides us with
strong empirical evidence in support of the IP analysis of auxiliariless finite clauses.
We can adduce a further piece of evidence in support of the IP analysis from facts relat¬
ing to the syntax of floating quantifiers in sentences such as:
Although the quantifier all here is interpreted as quantifying the students, it seems clear
that it is a separate constituent from the students. One reason for thinking this is the
fact that a string such as the students all cannot serve as a sentence fragment - as we
see from the ungrammaticality of speaker B’s reply in (33) below:
Moreover, all can be separated from the students by (for example) an intervening
adverb such as definitely, cf.
But if all is separate from the students (and is floating somewhere internally within the
structure of the sentence - hence its designation as a floating quantifier), where is all
positioned? At this point, it is timely to recall our suggestion in §3.6 that floating quan¬
tifiers occupy the specifier position within VP in sentences such as They can both
speak French. It would seem natural to extend that suggestion to sentences such as:
and maintain that the floating quantifier all occupies spec-VP (i.e. the specifier posi¬
tion within VP). If we extend the spec-VP analysis of floating quantifiers to sentences
such as (32), we might suggest that (32) has a structure along the lines of (36) below:
(36)
D N
The students
hate syntax
The IP analysis in (36) enables us to say that the students occupies spec-IP, whereas the
floating quantifier all occupies spec-VP. We can then say that floating quantifiers and
the expressions they quantify have in common the fact that they both occupy specifier
positions within the clause. (We shall return to consider the syntax of floating quanti-
141
Empty categories
fiers in more detail in chapter 8.) However, a crucial premise of the analysis in (36) is
that auxiliariless finite clauses are headed by an unfilled INFL constituent.
We can adduce a further piece of evidence in support of the IP analysis from facts
relating to the syntax of adjuncts. A traditional distinction is drawn between argu¬
ments (which are expressions which typically denote the participants in the activity or
event described by a verb) and adjuncts (which are expressions providing additional
information about the relevant activity/event, e.g. its location, the time at which it took
place, the manner in which it took place, etc.). Thus, in sentences such as:
the bracketed subject they and the bracketed complement her are arguments of the verb
ignored, while the italicized expressions are adjuncts. (The term adjunct describes the
function of the italicized expressions, not their categorial status, since completely is an
adverb, and on the boat and during the visit are prepositional phrases.)
On distributional grounds, we might suggest that there are (at least) two different
kinds of adverbial adjunct; the difference between the two can be illustrated by exam¬
ples such as the following:
In (38), the adverb certainly seems to be associated with the IP headed by have, since it
can be positioned between the specifier (= they) and head (= have) of the relevant IP;
hence, we might say that certainly is an EP-adverb (i.e. an adverb which is positioned
internally within IP). Conversely, the adverb completely seems to be associated with
the bracketed VP headed by ignored, and can be positioned at the beginning or end of
the bracketed VP; thus, completely seems to function as a VP-adverb (i.e. as an adverb
positioned internally within VP). Evidence that completely is a constituent of the
bracketed VP in (38) comes from sentences such as the following:
(39) It seems incredible that they should have ignored her completely, but. . .
(a) [ignored her completely], they certainly have
(b) they certainly have
The fact that the adverb completely is fronted when the bracketed VP is fronted in
(39a) and is ellipsed (i.e. omitted) when the VP is ellipsed in (39b) suggests that com¬
pletely is a constituent of the bracketed VP. The assumption that certainly is an IP
142
4.4 Unfilled INFL
The IP-adverb certainly in (40) is positioned between the head I constituent have of IP
and its specifier they, likewise, the VP-adverb completely is positioned between the
head V ignored of VP and its specifier both. If we assume that all projections are
binary-branching and that all intermediate projections are single-bar constituents, (40)
will have the structure (41) below:
completely V PRN
I I
ignored her
The VP-adverb completely in (41) attaches to the V-bar ignored her and expands it into
the extended V-bar completely ignored her, we might therefore say that completely is a
V-bar adjunct. Similarly, the IP-adverb certainly in (41) expands the I-bar have both
completely ignored her into the extended I-bar certainly have both completely ignored
her, and hence is an I-bar adjunct. Example (41) illustrates two properties of the
relevant kind of adverbial adjuncts: firstly, they merge with intermediate projections
(like V-bar and I-bar); and secondly, they serve to expand a category of a given type
into an extended category of the same type (e.g. completely expands V-bar into an
extended V-bar, and certainly expands I-bar into an extended I-bar).
In the light of our discussion of adverbial adjuncts, consider now how we account
for examples such as the following:
Given that certainly is an I-bar adjunct, it follows that (42) must be an IP; and since the
relevant IP contains no overt I constituent, it must contain a covert I. In other words,
(42) must have the structure (43) below:
143
Empty categories
(43)
prn'
I
They
The analysis in (43) enables us to maintain that certainly is the kind of adverb which
merges with an I-bar to form an extended I-bar. But a crucial premise of the analysis in
(43) is the assumption that auxiliariless finite clauses are IPs headed by an unfilled I
constituent.
We can formulate a further empirical argument in support of the IP analysis in
relation to facts about sentence fragments. In this connection, consider the two alterna¬
tive replies which might be given by speaker B in the dialogue in (44) below:
Under the IP analysis which we are arguing for here, speaker B’s reply in (44a) will
have the structure (45) below:
(45) IP
PRN
I
He I
Recall that in §3.8, we noted that maximal projections can serve as sentence fragments:
since the VP blows his nose is the maximal projection of the V blows in (45), and the
IP He blows his nose is the maximal projection of the empty INFL constituent, the IP
analysis in (45) correctly predicts that both can be used as sentence fragments -
precisely as we find in (44).
144
4.5 Bare infinitives
auxiliary are also IPs, the natural conclusion to draw is that all finite clauses are IPs.
Since to infinitive clauses are also IPs, can we generalize still further and say that all
finite and infinitival clauses are IPs? Well, this depends on how we analyse a class of
bare (i.e. fo-less) infinitive complement clauses such as those bracketed below:
If (as we are suggesting) all finite and infinitival clauses are indeed IPs, bare infinitive
clauses like those in (46) will be IPs headed by an abstract INFL constituent. More
specifically, we might suppose that the relevant clauses are IPs headed by a null coun¬
terpart of infinitival to (below denoted as 0), so that (e.g.) the bracketed infinitive in
(46a) would have the structure (47) below:
(47) IP
N
I
Tom r VP,
V PRN
I
criticize him
We have assumed in (47) that rather than being unfilled, INFL is filled by a null variant
of infinitival to. One reason for thinking this is that the head V in a bare infinitive
complement is always in the infinitive form - i.e. the same form as is required after
infinitival to: we can account for this if we assume that the null infinitive particle 0 has
the same complement-selection properties as infinitival to, and thus selects a comple¬
ment headed by a verb like criticize in the infinitive form. We could then say that verbs
like know/hear/let/watchlsee as used in (46) above select an IP complement headed by
the null infinitive particle 0, whereas verbs like expect, judge, report, consider, want,
etc. select an IP complement headed by to, as in (48) below:
It would then follow that all infinitive clauses contain an IP headed by to or by its
covert counterpart 0.
145
Empty categories
What lends plausibility to the suggestion that bare infinitive complements are
headed by a null counterpart of to is the fact that many bare infinitive complements
have to infinitive counterparts - cf.
(52) (a) I saw [her coral lips to move] (Lucentio, Taming of the Shrew, I.i)
(b) My lord your son made [me to think of this] (Helena, All!s Well That
Ends Well, I.iii)
(c) What would you have [me to do]? (Lafeu, All’s Well That Ends Well,
V.ii)
(d) I had rather hear [you to solicit that] (Olivia, All’s Well That Ends Well,
Ill.i)
The fact that bare infinitives sometimes have to infinitive counterparts (in particular
types of construction) makes it all the more plausible to suppose that they are headed
by a covert counterpart of to, and hence are IPs.
Further evidence in support of the IP analysis comes from cliticization facts in rela¬
tion to sentences such as the following (% = acceptable in some varieties, e.g. mine):
we can account for the fact that have cannot cliticize onto you by positing that the
presence of the null infinitive particle 0 between you and have blocks cliticization.
Another argument leading to the same conclusion comes from structures like:
146
4.6 Null complementizers
It has been argued by Safir (1993) that the pronoun there (in this use as a contentless
dummy pronoun) is restricted to occurring in the specifier position within IP. Such a
restriction would account for contrasts such as:
(56) (a) I don’t consider [there to be any good reason why I should do it]
(b) *1 don’t consider [there any good reason why I should do it]
since the first bracketed complement is an IP headed by infinitival to, and the second is
a type of verbless clause sometimes referred to as a small clause which appears not to
be headed by INFL (since it contains no auxiliary or infinitival to, and no VP). If
dummy there can only occur as the specifier of an IP, it follows that the bracketed com¬
plements in (55) must be IPs (headed by the null infinitive particle 0).
Our discussion here leads us to the wider conclusion that both to infinitive clauses
and bare infinitive clauses are IP (infinitive phrase) constituents headed by an infiniti¬
val INFL; INFL is filled by to in structures like (48), but is filled by the null infinitive
particle 0 in bare infinitive complements like those bracketed in (46). Given that we
have also argued that all finite clauses contain an IP projection (headed by an INFL
which is either filled or unfilled), the overall conclusion which we reach is that all finite
and infinitival clauses contain an IP, and that INFL is filled in clauses containing a
finite auxiliary or infinitival to/0, but unfilled elsewhere. One obvious advantage of this
analysis is that it enables us to attain a uniform characterization of the syntax of (finite
and infinitival) clauses as structures which contain an IP with a VP complement.
(57) (a) Phyllis Stein thinks [that money can buy happiness]
(b) Phyllis Stein thinks [money can buy happiness]
(58) (a) We didn’t intend [for that to happen]
(b) We didn’t intend [that to happen]
The question we are asking is whether all four bracketed complement clauses in (57-8)
are CPs, or whether the (a) sentences containing an overt complementizer are CPs, and
the complementizerless (b) sentences are IPs. In more concrete terms, this amounts to
asking (for example) whether the complement clause in (57b) is a CP with the structure
(59a) below headed by a null declarative complementizer 0, or simply an IP with the
structure (59b) below:
147
Empty categories
buy happiness
What kind of theoretical or empirical considerations might lead us to favour one analy¬
sis over the other?
At first sight, it might seem as if semantic considerations favour the null complemen¬
tizer analysis (59a). After all, the bracketed that-clause in (57a) is declarative in illocution¬
ary force (i.e. is used to make a statement), and we might argue that its declarative force is
attributable to the presence of the declarative complementizer that. More generally, we
might suppose that complementizers serve the function of marking the illocutionary force
of clauses. We could go on from there to argue that because the bracketed complement
clause in (57b) is also interpreted as declarative in force, the relevant clause must be a CP
headed by a null (declarative) complementizer 0. This would then enable us to arrive at a
structurally uniform characterization of all clauses as CPs in which the illocutionary force
of the clause is indicated by the choice of (overt or covert) complementizer heading CP:
for obvious reasons, we might refer to the claim that all clauses are CPs as the structural
uniformity hypothesis.
Unfortunately, however, this type of illocutionary force argument carries little
conviction. After all, we might equally account for the declarative force of the brack¬
eted complement clause in (57b) by supposing that clauses which do not contain an
interrogative head or specifier are interpreted as declarative by default, there would
then be no reason not to analyse the bracketed complement clause in (57b) as a simple
IP with the structure (59b), interpreted as declarative by default (e.g. by virtue of not
containing an interrogative constituent). Some evidence in favour of the default analy¬
sis comes from nonstandard structures such as:
found in many varieties of English (e.g. Belfast English). Since the bracketed CP in
(60) is interrogative in force, the complementizer that clearly cannot be declarative in
force and cannot determine the illocutionary force of the bracketed CP (which is in fact
determined by the interrogative expression which street?), but rather must simply serve
to mark the CP containing it as finite. So, the fact that the bracketed CP in (57a) is
declarative in force might not be attributable to the presence of that but rather to the
absence of an interrogative constituent preceding that.
148
4.6 Null complementizers
(where minimal means ‘as economical/short as possible’). Peggy Speas (1995) has
proposed a related principle to the effect that UG does not license contentless projections
(i.e. projections whose head and specifier have no independent content of their own).
Since CP in structures such as (59a) has no specifier and contains a head with no pho¬
netic, semantic or grammatical properties, it follows that CP is a contentless projection,
and so violates the no-contentless-projections constraint. So, economy considerations
favour analysing the (a) sentences in (57-8) as CPs, and the (b) sentences as IPs.
There seems to be some supporting empirical evidence in favour of the economy
(IP) analysis of clauses such as those in (57—8b). One such piece of evidence comes
from coordination facts. Boskovic (1994) argues that complementizerless clauses such
as those italicized in (62) below cannot (idiomatically) be coordinated with clauses
containing an overt complementizer such as that. cf.
(62) (a) *John said [Peter left] and [that Bill kissed Mary]
(b) *John reckoned [Peter left] and [that Bill kissed Mary]
If this is so (and the grammaticality judgments are very subtle), why should it be? If we
suppose that the italicized that-less complement clauses are IPs but that the bold-printed
t/jm-clauses are CPs, we can attribute the ungrammaticality of sentences like (62) to the
fact that we have coordinated an IP with a CP, thereby violating the condition that we
can only (idiomatically) conjoin constituents belonging to the same category. By con¬
trast, if we claim that italicized complementizerless clauses in (62) are CPs, we would
have no obvious account of why they can’t be coordinated with the bold-printed CPs.
A further piece of evidence leading to the same conclusion comes from facts relating
to preposing. As we see from the examples below, we can prepose the subject of a
complementizerless clause (to emphasize it in some way by moving it into a more
prominent position at the front of the sentence), but not the subject of a clause intro¬
duced by a complementizer like that4.
149
Empty categories
If we adopt the economy analysis and assume that the bracketed clauses in (63) are IPs
but those in (64) are CPs, we can provide a (somewhat simplified: cf. Culicover 1992)
account of the facts if we posit a constraint (i.e. structural restriction) to the effect that
the subject cannot be extracted out of a CP Example (64b) will then be ungrammatical
because the subject money has been extracted out of the CP that money can buy happi¬
ness; but (63b) will correctly be predicted to be grammatical because the subject
money has been extracted out of the IP money can buy happiness, not out of a CP
However, if we (wrongly) suppose that the bracketed clause in (63b) is a CP, we
wrongly predict that (63b) is ungrammatical (because under the CP analysis, (63b)
would involve extracting a subject out of a CP). The provisional conclusion which our
discussion leads us to is that complementizerless clauses are IPs - not CPs headed by a
covert complementizer. Thus, all clauses are IPs, but only some clauses (viz. those
containing a complementizer like thatliflfor) project further into CP.
A further argument leading us to the same conclusion comes from facts relating to
the cliticization of is (in the guise of its variant ’s) onto the preceding verb {think) in
structures such as:
Under the IP analysis of complementizerless clauses, (65a) would have the structure
(66a) below, whereas under the CP analysis it would have the structure (66b):
However, given that we earlier saw in relation to sentences such as (21), (28b-c) and
(53b) above that an intervening null head blocks cliticization, we should expect the
intervening null complementizer 0 in (66b) to block cliticization of is onto think. The
fact that cliticization is indeed possible in (65b) suggests that the CP analysis in (66b)
can’t be right, and that the more economical IP analysis in (66a) is more plausible.
We can provide a parallel cliticization argument against the claim made in earlier
work (e.g. Radford 1988) that infinitive complements with null PRO subjects are CPs
headed by a null complementizer. If this were so, it would be difficult to account for
the fact that to can cliticize onto want (forming wanta/wanna) in structures such as:
150
4.7 Null determiners
since we would expect that the presence of a complementizer between want and to
would prevent to from cliticizing onto want. After all, cliticization of to onto want is
not possible in varieties of English (like Belfast or Ozark English) which allow for to
infinitives: cf.
and it seems reasonable to suppose that this is because to-cliticization is blocked by the
presence of an intervening complementizer (in this case, for). But if this is so, the fact
that cliticization is possible in sentences like (67b) I wanna go home would argue that
there cannot be a complementizer intervening between want and to in sentences like
(67) - hence that the infinitive complement to go home has the status of an IP rather
than a CP.
So, the overall conclusion which our discussion of examples such as (57-68) leads
us to is that all clauses are IPs, but that only clauses which contain an overt comple¬
mentizer (like that/if/for) project further into CP. For additional evidence pointing to
this conclusion, see Boskovic 1995, chapter 2.
Given the assumptions we have made so far, (69) will have the structure (70) below:
to V DP
I / \
enjoy D N
the course
However, (70) presupposes a curiously asymmetric analysis of the status of nominals (i.e.
of expressions containing nouns or pronouns). Some (like we) have the status of PRN;
151
Empty categories
others (like students) have the status of N; and others (like the course) are DPs. Given our
earlier arguments that clauses have a uniform status as IP constituents, it would seem nat¬
ural to ask whether we can attain a uniform characterization of the syntax of nominals. But
how? What we shall suggest here (following ideas developed by Abney 1987 and
Longobardi 1994) is that all nominals are projections of a head determiner constituent.
What this implies in the case of bare nominals (i.e. noun expressions used without
any modifying determiner) is that such nominals are DPs headed by a null determiner
(below symbolized as 0). This means (for example) that the bare noun students in (69)
is not simply an N, but rather a DP like (71) below:
(71) DP
/ \
D N
I I
0 students
The assumption that bare nominals contain a null determiner is a traditional one - for
example, Chomsky (1965, p. 108) suggests that the noun sincerity in a sentence such
as Sincerity may frighten the boy is premodified by a null determiner. If this is so, then
empty categories play just as central a role in the syntax of nominals as they do in the
syntax of clauses: in the same way as auxiliariless finite clauses are IPs headed by an
empty INFL, so too bare nominals are DPs headed by an empty determiner.
However, an important question to ask about the empty determiner analysis of bare
nominals is whether it is consistent with a minimalist approach to syntax which posits
an economy principle like (61) which prohibits superfluous projections. More specifi¬
cally, we might ask whether a DP analysis of bare nominals is consistent with the
no-contentless-projections constraint proposed by Peggy Speas (1995). The answer
is that it is, by virtue of the fact that the null determiner 0 in structures like (70) -
although lacking phonetic content - has clear semantic and grammatical properties of
its own, and thus has intrinsic content.
Let’s look first at the semantic properties of the null determiner 0. In this connection,
consider the interpretation of the italicized bare nominals in sentences such as:
The nouns eggs/bacon in (72) have a generic interpretation, and hence are interpreted
as meaning ‘eggs/bacon in general’. In (73) they have an existential (= partitive) inter¬
pretation, roughly paraphraseable as ‘some eggs/bacon'. If we say that bare nominals
are DPs headed by a null generic/existential determiner 0, we can say that the semantic
properties of 0 determine that bare nominals will be interpreted as generically or exis¬
tentially quantified.
152
4.7 Null determiners
Moreover, there is evidence to suggest that the null determiner 0 carries person
properties - and in particular, is a third person determiner. In this respect, consider sen¬
tences such as:
The examples in (74a) show that a first person expression such as we syntacticians can
only bind a first person reflexive like ourselves, and can only be tagged by a first per¬
son pronoun like we. The examples in (74b) show that a second person expression like
you syntacticians can only bind a second person reflexive like yourselves, and can only
be tagged by a second person pronoun like you. The examples in (74c) show that a bare
nominal like syntacticians can only bind a third person reflexive like themselves and
can only be tagged by a third person pronoun like they. One way of accounting for the
relevant facts is to suppose that the nominals we syntacticians/you syntacticians/
syntacticians in (74) are DPs with the structure (75) below:
(75)
D' N
I
we syntacticians
you syntacticians
0 syntacticians
and that the person properties of the DP are determined by those of its head determiner.
If we suppose that we is a first person determiner, you is a second person determiner
and 0 is a third person determiner, the facts in (74) above are precisely as the analysis
in (75) would lead us to expect.
In addition to having quantificational properties and person properties, the null
determiner 0 also has specific complement-selection properties - as can be illustrated
by the following set of examples:
If we suppose that each of the italicized bare nouns in (76) is the complement of the
null determiner 0, the restrictions in (76) would seem to suggest that 0 can select as its
complement an expression headed by a plural count noun like poems, or a singular
mass noun like poetry - but not by a singular count noun like poem. The complement-
153
Empty categories
selection properties of the null determiner 0 would seem to be parallel to those of the
overt determiner enough: cf.
The fact that the null generic/existential determiner 0 has much the same quantifica-
tional, person and complement-selection properties as a typical overt determiner such
as enough strengthens the case for positing the existence of a null determiner 0, and
for analysing bare nominals as DPs headed by a null determiner.
4.8 Pronouns
The conclusion we have arrived at so far is that both nominals containing
an overt determiner and bare nominals are DPs headed by a determiner which is overt in
the former case and covert in the latter. This leads us towards the conclusion that all nom¬
inal arguments are projections of a head D constituent (a conclusion argued for at length
by Longobardi 1994). But can we extend this analysis to pronouns - and if so, how?
The answer is that there is some evidence (from studies by Postal 1966, Abney 1987,
Longobardi 1994 and Uriagereka 1995) to suggest that pronouns have the categorial
status of determiners. We have already implicitly suggested in (75) that in structures
such as we/you syntacticians, the so-called pronouns we and you function as determin¬
ers which take the noun syntacticians as their complement. It therefore seems natural
to suppose that simple pronouns like we and you could be analysed as determiners used
without any noun complement. If this is so, sentences such as:
(79) (a)
\
trust D N
I I
you linguists
(b)
don’t V D
trust you
154
4.8 Pronouns
The analysis in (79) would enable us to provide a unitary account of the syntax of nom¬
inal, and to suppose that they are all projections of a head determiner constituent. In
structures such as (79a), the determiner we is used prenominally (with a following
noun as its complement), whereas in structures such as (79b) it is used pronominally
(i.e. on its own, without any following noun complement). The determiner analysis of
pronouns would also provide us with a straightforward answer to the question of why
(as noted in §2.4) most determiners have pronoun counterparts.
The determiner analysis of pronouns might also help us understand why two-year-
old children sometimes produce structures such as the following (reported by McNeill
1970, p. 28):
One answer might be that children analyse it as a determiner, and wrongly assume that
(like most determiners) it can be used not only pronominally, but also prenominally;
this would mean that it ladder in (80) is a DP with the structure (81) below:
D N
it ladder
The analysis in (81) assumes that the child uses the definite pronoun it in (80) in much
the same way as an adult would use the definite prenominal determiner the.
However, the analysis in (81) raises the interesting question of why it can be used
pronominally but not prenominally in adult English, and conversely why the can be
used prenominally but not pronominally - in other words, how we account for
contrasts such as (82) below:
155
Empty categories
are DPs, bare nominals are DPs headed by a null determiner, and pronouns are deter¬
miners used without a complement. We can then conclude that all nominal and
pronominal arguments are projections of an (overt or covert) D constituent, and so
arrive at a uniform characterization of nominals as D-projections. Given this assump¬
tion, our earlier sentence (69) above will no longer have the asymmetric structure (70)
above, but rather the more uniform structure (83) below:
I / \
enjoy D N
l I
the course
in which the three nominal arguments we, students and the course are all analysed as
D-projections.
The conclusion to be drawn from our discussion here is that all nominal and pronomi¬
nal expressions are D-projections. However, there are reasons for thinking that this is
only true of nominal expressions used as arguments (i.e. as subjects or complements),
not of nonargument nominals (e.g. nominals which have a vocative, predicative or
exclamative use). Examples such as those in (84) below would suggest that non¬
argument nominals can be N-projections lacking a determiner:
The italicized nominal expression serves a vocative function (i.e. is used to address
someone) in (84a), a predicative function in (84b) (in that the property of being head
of department is predicated of the unfortunate Dick Head) and an exclamative func¬
tion in (84c). Each of the italicized nominals in (84) is headed by a singular count
noun (doctor/head/fool): in spite of the fact that such nouns require an overt deter¬
miner when used as arguments, here they function as nonarguments and are used
without any determiner. This suggests (as noted by Longobardi 1994) that non¬
argument nominals can be N-projections, whereas argument nominals are always
D-projections.
156
4.9 Attributive adjectives
The analysis we have presented here points to significant potential parallels between
the internal structure of clauses and that of nominal arguments. Just as clauses are
projections of an overt or covert I constituent, so too nominal arguments are projec¬
tions of an overt or covert D constituent. Using the terminology suggested by Jane
Grimshaw (1991), we can say that V has an extended projection into IP in the same
way as N has an extended projection into DP (so that IP is an extended projection of V,
and DP is an extended projection of N). The parallels between IP and DP may go even
further - as we can illustrate in terms of the following examples:
(85) (a) We can arrange [for the accountants to audit the books]
(b) We can arrange [for an audit of the books]
In (85a), the verb audit has a direct projection into the VP audit the books, an extended
projection into the IP the accountants to audit the books, and might be argued to
have a further extended projection into the CP (complementizer phrase) for the
accountants to audit the books. In (85b) the noun audit has a direct projection into the
NP audit of the books, an extended projection into the DP an audit of the books, and a
further extended projection into the PP (prepositional phrase) for an audit of the books.
Thus, CP and PP might be analysed as (secondary) extended projections of V and N
respectively.
attempt:
157
Empty categories
(87) DP
the A
various
government N' IP
attempts D
We could then suppose (cf. our earlier discussion of (25) above) that the plural number
feature of the noun attempts percolates up from N to F, and there is checked against the
number feature of the inherently plural adjective various.
Further evidence that the grammatical properties of nouns percolate from N to F
comes from phrases such as that bracketed in (88) below:
(89)
chair
158
Workbook section
4.10 Summary
In this chapter, we have seen that empty categories (i.e. categories which
have no overt phonetic form) play a central role in syntactic theory. In §4.2 we argued that
apparently subjectless io-infinitive clauses have an empty PRO subject which is typically
controlled by an antecedent in a higher clause. In §4.3 we aigued that elliptical clauses hike
that bracketed in He could have left and [she have stayed] are IPs headed by a null
(ellipsed) finite auxiliary. In §4.4 we argued that clauses like He hates syntax containing a
finite nonauxiliary verb are IPs headed by a null ENFL constituent, and that the
tense/agreement properties of the verb hates percolate from V to INFL. In §4.5 we
claimed that bare infinitive clauses are also IPs, headed by a null counterpart of infinitival
to. In §4.6 we considered the possibility of adopting a symmetrical analysis of clause
structure under which all clauses would be analysed as CPs, with complementizerless
clauses (i.e. clauses which do not contain an overt complementizer) being analysed as CPs
headed by a null complementizer whose properties determine the illocutionary force of the
clause. However, we rejected the CP analysis of complementizerless clauses in favour of a
simpler IP analysis, supporting the IP analysis on the basis of considerations relating to
economy, coordination, subject extraction and cliticization. In §4.7 we turned to look at
the syntax of nominal arguments, suggesting that bare nominals should be analysed as
DPs headed by a null determiner 0 which has a generic or existential interpretation, and
which has the complement-selection property that it can only be used to quantify a singu¬
lar mass noun or plural count noun. In §4.8 we went on to argue that pronouns (e.g. per¬
sonal pronouns like we/you/they etc.) are simply determiners used without a complement.
We concluded from this that all nominal arguments are D-projections: e.g. in a sentence
such as We don’t expect students to enjoy the course, we is a pronominal (i.e. complement¬
less) determiner, students is a projection of the null determiner 0, and the course is a
projection of the overt determiner the. We suggested that there are significant parallels
between the syntax of clauses and that of nominal arguments, in that just as verbs have an
extended projection into IP (and in some structures may have a further extended projection
into CP) so too nouns have an extended projection into DP (and in some structures may
have a further extended projection into PP). In §4.9 we briefly discussed the syntax of
attributive adjectives, suggesting that these serve as the specifiers of an empty functional
category F which is positioned between D and N.
Workbook section
Exercise I (§4.2)
In the text, we made no mention of imperative clauses (i.e. clauses used
to issue a command). In English, imperatives can have overt (usually second person)
subjects such as those italicized in (1) below:
159
Empty categories
2a Do be careful!
b Mind the step!
Model answer
Empirical evidence in support of positing a null second person subject in
imperatives comes from tag sentences such as the following:
Tags typically contain a pronominal copy of the subject of the tagged clause (cf. Harry
likes pasta, doesn’t he?). The fact that imperative sentences always carry a you tag
suggests that imperatives have an understood null second person (singular or plural)
subject. Further support for positing a null second person subject in imperatives comes
from evidence of essentially the same type as that which we adduced in support of
positing a PRO subject in apparently subjectless infinitives in §4.2. (You take over at
this point, and devise relevant examples and supporting arguments.)
Exercise II (§§4.3-4.4)
In the stylized construction illustrated by the examples in (1) below:
the italicized items are traditionally said to be finite verbs in their subjunctive form. If
we assume that finite auxiliaries occupy the head I position of IP, then this might sug¬
gest that the bracketed complement clause in la has the structure 2 below:
160
Workbook section
However, an alternative possibility which has been suggested in the relevant literature
is that the IP constituent in subjunctive clauses is headed by a null modal (e.g. a null
counterpart of should). If this were so, the bracketed complement clause in la would
have the alternative structure 3 below:
3 .CP
C' IP.
I
that D
I
he PASSP
I
0 PASS
(where 0 denotes a null modal, PASS denotes a passive auxiliary and PASSP denotes a
passive phrase - i.e. a phrase headed by a passive auxiliary). Discuss whether either or
both of the analyses could provide a principled account of relevant aspects of the
morphosyntax of subjunctive clauses.
Helpful hints
Some points which you might like to look at include how to account for
the fact that subjunctive clauses require nominative subjects like he, and for the fact
that so-called subjunctive verb forms are homophonous with the corresponding infini¬
tive forms. You might also try and account for why have cannot cliticize onto he in lb
- as we see from the ungrammaticality of *The major ordered that he’ve completed the
operation by midnight. Also look at how subjunctive clauses are negated: ordinary
finite clauses are negated by positioning not after the finite auxiliary (cf. He was not
arrested), whereas subjunctive clauses seem to require preauxiliary negation (cf. The
defence counsel asked that he not be banned from driving).
161
Empty categories
Which adverbs are VP-adverbs, which are IP-adverbs - and which can serve both func¬
tions? Draw tree diagrams for each of the sentences in 1-5 to illustrate your answer,
and give arguments in support of your analysis.
(i)
all ADV
severely D
him
The analysis in (i) assumes that severely is a V-bar adjunct, and hence merges with the
V-bar censure him to form the extended V-bar severely censure him, which then merges
with the floating quantifier all (which we assume to occupy spec-VP, i.e. the specifier
position within VP) to form the maximal projection (VP) all severely censure him.
Some evidence that this string is indeed a maximal projection comes from the fact that
it can serve as a sentence fragment, e.g. in a discourse such as:
Exercise IV(§4.7)
Discuss the syntax of the italicized nominal expressions in the following
Jamaican Creole sentences (from Bailey 1966), commenting on differences with
Modem Standard English.
1 Manggo swiit
mango sweet
‘Mangoes are sweet’
2 Di manggo swiit
the mango sweet
‘The mango is sweet’
162
Workbook section
3 Di manggo-dem swiit
the mango-them sweet
‘The mangoes are sweet’
4 Jan mash di eg-dem
John smash the egg+them
‘John broke the eggs’
5 Jien-dem naa kom
Jane+them no+are come
‘Jane and her friends aren’t coming’
6 Wan daag bait mi
one dog bite me
‘A dog bit me’
7 Dat a plom
that are plum
‘That is a plum’
8 Mi a tiicha
me are teacher
‘I am a teacher’
9 Tiicha no kom yet
teacher no come yet
‘The teacher hasn’t come yet’
10 Fait brok out
fight broke out
‘A fight broke out’
11 Disya bwai niem Piita
this+here boy name Peter
‘This boy is called Peter’
12 Dis bwai-ya niem Piita
this boy+here name Peter
‘This boy is called Peter’
13 Dem-de manggo swiit
them+there mango sweet
‘Those mangoes are sweet’
14 Dem manggo-de swiit
them mango+there sweet
‘Those mangoes are sweet’
15 Im waip im mout
him wipe him mouth
‘He wiped his mouth’
163
Empty categories
(i)
D N
di manggo-dem
In order to account for the fact that a plural definite noun like manggo-dem can only be
modified by a definite determiner like di (which can be used in a plural function - as
well as having a singular use), not (for example) by an indefinite singular determiner
like wan ‘one’ (as we see from the ungrammatically of *wan manggo-dem ‘one
mangoes’), we clearly have to suppose that the determiner must agree with its noun
complement in number and definiteness. One way of handling agreement between the
noun and its determiner (which we shall discuss in more detail in the next chapter) is to
assume that the number/definiteness features of the noun percolate from N to D, and
are thereby checked against the features of the determiner.
Helpful hints
The nominals in sentences 15-20 all involve prenominal possessive struc¬
tures not discussed so far in the text. There are a number of alternative analyses of
prenominal possessives in English. However, for the purposes of this exercise, assume
164
Workbook section
the following analysis. Prenominal possessives (like the president’s or my) are DP/D con¬
stituents which function as specifiers of a null determiner 0 (cf. Abney 1987), and carry
genitive case (marked by the suffix ’s on nominal expressions like the president, and by
using the genitive form in pronouns like Itmetmy which inflect for case). Thus, expres¬
sions like the president’s car and my car have the respective structures indicated below:
car
where the null determiner 0 has a genitive specifier (in much the same way as a finite
auxiliary has a nominative specifier). Assume that prenominal possessives in Jamaican
Creole occupy the same spec-DP position as their English counterparts - though don’t
assume that they necessarily have the same case properties.
Exercise V (§4.8)
Discuss the syntax of the italicized items as they are used in the bracketed
nominal expressions below:
In addition, say what problems (if any) are posed for the analysis in the text by the ital¬
icized items in the bracketed expressions in the examples below (some of which are
taken from nonstandard varieties of English):
165
Empty categories
Where variety-specific pronouns are being used in a way which is different from their
use in Standard English, specify the nature of the differences.
None is the pronominal counterpart of no, and correspondingly can only be used on its
own, without a following nominal complement. Thus, both no and none are determin¬
ers, but they differ in their complement-selection properties: no has the property that it
requires a nominal complement, while none has the property that it cannot have a nom¬
inal complement (though it can have a complement headed by of - cf. none of us).
Given these assumptions, no student and none will have the respective structures indi¬
cated in (ii) below:
However, an alternative possibility (not discussed in the text) would be to assume that
none is a determiner which takes an empty N or empty NP as its complement, so that
none has the structure (iii) below (where Nmax denotes a maximal projection of N -
i.e. an N on its own, or an NP - and 0 is an empty category):
D Nmax
I '
none
The problem of interpreting none would then amount to that of determining what none
quantifies - i.e. determining the antecedent of 0. In sentences such as (iv) below:
(iv) I was looking for [a chair!some chairs/some furniture ], but found none
166
Workbook section
has a regular noun plural form in +s (cf. one/ones). Secondly, it has the countability
property of many nouns, and so cannot have a mass noun as its antecedent: cf.
Moreover, like a typical noun such as fish (but unlike a typical pronoun such as him),
one can be modified by a preceding adjective: cf.
If one is indeed a pronominal noun, it follows that it cannot be the case that all pro¬
nouns are pronominal determiners (though a determiner analysis might be argued to be
appropriate for most pronouns).
Exercise VI (§§4.2-4.8)
Draw tree diagrams representing the structure of the following sentences,
giving arguments in support of your analysis, and commenting on points of interest.
Helpful hints
Assume that wouldn't is a single word, with much the same syntax as
would. Try and account for why he had can’t become he’d in 2, and why me have can't
become me’ve in 7.
167
Empty categories
become N PP
I / \
chairman P DP
I / \
of D N
the board
A key assumption made in 1 is that the become clause has a null PRO subject. This can
be justified in terms of agreement facts. As we see from examples like (10) in the text,
predicate nominals have to agree with the subject of their own clause (i.e. with the
subject of their own IP). Only if we assume that the become clause has a null PRO sub¬
ject which is controlled by I and so first person singular can we account for why the
predicate nominal chairman of the board has to be singular in 1 - cf. the ungrammati-
cality of:
Given the PRO analysis in (i), we can say that chairman of the board agrees with the
PRO subject of the IP containing it, and so has to be singular because PRO is singular
(since PRO is controlled by /).
In the text, it was claimed that argument nominals are D-projections (i.e. expressions
headed by a determiner), and that pronouns are determiners. If this is so, then both the
overt pronoun / and the covert pronoun PRO will be determiners. We can account for
the fact that neither can have a noun complement after it (cf. */ deputy) by positing that
it is a complement-selection property of inherently pronominal determiners that they
can’t have a noun complement.
Although argument nominals are D-projections, nonargument nominals (used in a
vocative, predicative or exclamative function) can simply be N-projections. In this
respect, it is interesting to note that the nominal chairman of the board is headed by
the singular count noun chairman: when used as arguments, singular count noun
expressions must contain an overt determiner in English. The fact that there is no deter¬
miner preceding chairman of the board in 1 suggests that it is here being used in a
nonargument function: in traditional terms, chairman of the board is a predicate nomi¬
nal (by virtue of being used as the complement of a predicative verb such as be,
168
Workbook section
become, stay, remain etc.). Since predicative nominals are not arguments, and non¬
arguments are not required to contain a D-projection, chairman of the board can be
analysed as a bare NP.
In the text, it is claimed that all clauses contain an I-projection. The complement
clause PRO to become chairman of the board is clearly an IP headed by infinitival to.
Some evidence that it isn’t a CP headed by a null complementizer comes from wanna-
contraction facts in relation to sentences such as (iii) below, where to cliticizes onto
want:
If (contrary to what we are suggesting here) the to-clause were a CP, (iii) would have
the structure (iv) below prior to cliticization of to onto want:
(iv) I want [cp COMP [Ip PRO to become chairman of the board]]
Such an analysis would wrongly predict that to cannot cliticize onto want, because the
intervening null COMP would block cliticization (in the same way as the intervening
null INFL blocks cliticization in sentences such as (21), (28b-c) and (53b) in the main
text).
The analysis in (i) also assumes that the wont-clause is an IP, but one headed by an
unfilled I constituent (with the first person singular present-tense features carried by
want percolating up to I). Some empirical support for the claim that there is a null pre¬
sent-tense I constituent in (i) comes from the fact that the sentence is tagged by a present-
tense form of the dummy auxiliary do in sentences such as:
169
5
Checking
5.1 Overview
In chapters 3 and 4, we discussed the ways in which words are projected
into syntactic structures. In this chapter, we shall be concerned with the principles
which determine the morphological form of words. The kinds of question we shall ask
here include (for example) why we say We are winning not *Us are winning, or *We is
winning, or *We are win. Why is it that we require we rather than us here, are rather
than is, and winning rather than win? We shall suggest that the morphological proper¬
ties of words can be characterized in terms of sets of grammatical features, and
features must be checked in an appropriate manner: this chapter thus provides an intro¬
duction to the concept of checking. It should be noted that work on checking theory is
as yet in its infancy, so that many of the ideas and descriptive details in this chapter are
inevitably somewhat sketchy and speculative.
170
5.2 Interpretable and uninterpretable features
171
Checking
(1)
phrase LF operations
formation
operations -► 1 spellout ]
(selection
and merger) PF operations
Spellout is the point at which the phrase structures generated by the processes of selec¬
tion and merger feed into two different components - a PF component which processes
their phonetic features, and an LF component which processes their grammatical and
semantic features. We shall have little to say about PF representations here, and will
focus mainly on how LF representations are computed.
The assumption that PF representations contain only phonetically interpretable fea¬
tures and LF representations only semantically interpretable features raises interesting
questions about the status of grammatical features (which Chomsky 1995b terms
formal features, since they determine the morphological form of items). The first ques¬
tion to ask is: ‘What are grammatical features7 Informally, we can define them as
features which play a role in grammatical (i.e. morphological or syntactic) processes.
So, grammatical features include number (singular/plural) features, since these play an
obvious role in the syntax of agreement (cf. these/*this books). They also include
gender (masculine/feminine/inanimate) features, since these play a role in the syntax of
reflexive anaphors (cf. He/*She/*It turned himself into a giant). Likewise, they include
person features, which play a role in the syntax of subject-verb agreement (cf.
He/*I/*You likes syntax). They also include features which determine the morphologi¬
cal form of items - for example, the case features of pronouns (cf. He/*Him likes
me/*l), or the inflectional features of verbs (cf. He has gone/*go/*going). But they do
not include features which have no morphological or syntactic correlate: for example,
although words such as calf foal, kitten, puppy, etc. share the feature [young], this is a
purely semantic feature which plays no role in any grammatical process, and so is not a
grammatical/formal feature. (By convention, features are enclosed in square brackets,
and are often written in capital letters.)
However, the distinction between grammatical and semantic features is not always
clearcut, since many grammatical features have clear semantic content, as we can illus¬
trate in terms of the following sentence:
The grammatical features of she indicate that it is third person feminine singular nomi¬
native determiner, those of has indicate that it is third person singular present-tense aux¬
iliary, and those of gone indicate that it is an ^-participle. Some of these grammatical
features are interpretable at LF (in the sense that they have semantic content and so
contribute to determining meaning), whereas others are uninterpretable at LF (in that
172
5.2 Interpretable and uninterpretable features
they have no semantic content and so make no contribution to meaning). For example,
the fact that she is a third person singular expression plays a role at LF (since it tells us
e.g. that she can refer to the girl next door but not to the curtains next door). By con¬
trast, the fact that she is nominative does not - as we can see from sentences such as the
following:
The fact that the italicized subject of the bracketed complement clause plays the same
semantic role in both sentences (as the subject of the win-clause), even though it has
the nominative form she in (3a) and the objective form her in (3b), suggests that case is
an uninterpretable feature. In much the same way, the fact that has is a present-tense
auxiliary has a role to play at LF (since a sentence like She had gone containing the
past-tense auxiliary had has a different interpretation), but the fact that has is third
person singular seems to play no role at LF (it is simply a consequence of the grammat¬
ical requirement for has to agree with its subject he). Likewise, the fact that gone is a
participle seems to have no role to play at LF (but rather is simply a consequence of the
fact that have requires a complement headed by a verb in the +n participle form). It
would seem that the case-features of pronouns and the agreement/participial inflections
of verbs have no role to play in semantic interpretation, and thus are purely formal
features.
So, the problem we face is that some of the grammatical features of words are inter¬
pretable at LF (i.e. contribute to meaning), but others are uninterpretable at LF (i.e. do
not contribute to meaning). If we assume that (in consequence of the principle of full
interpretation) LF representations contain only semantically interpretable features, it
follows that uninterpretable features must somehow be eliminated in the course of
deriving an LF representation, in order to ensure that the derivation converges at LF.
Since phonetic features are processed by the PF component (and hence do not input
into the LF component), the problem we face is how to eliminate uninterpretable gram¬
matical features from LF representations.
Before we try and answer this question, let’s consider a second problem posed by
grammatical features - namely that of how to deal with grammatical restrictions which
heads impose on their choice of specifier and complement. In this connection, consider
the restrictions on the choice of specifier (i.e. subject) and complement for has illus¬
trated in the following sentences:
The examples in (4a) show that has can have a third person singular nominative pro¬
noun like he as its specifier, but not a third person singular objective pronoun like him.
173
Checking
or a third person plural nominative pronoun like they: in other words, a finite auxiliary
like has imposes person/number/case restrictions on its specifier/subject. The examples
in (4b) show that has allows as its complement a verb in the n-participle form, but not a
verb in the mg-participle form, or a verb in its infinitive form: in other words, has
imposes morphological restrictions on its complement.
5.3 Checking
Thus, grammatical features pose two sets of problems: one is how to
eliminate uninterpretable features; the other is how to ensure that specifiers and com¬
plements carry the appropriate features for a given type of head. We can deal with both
types of problem in a unified way within the framework of (feature-) checking theory.
Let us suppose that items carry three different sets of grammatical features: head-
features (which determine their intrinsic grammatical properties), specifier-features
(which determine the kinds of specifier which they allow), and complement-features
(which determine the kinds of complement which they can take). Let us lurther assume
(following Chomsky 1995b) that all uninterpretable features must be checked in an
appropriate checking configuration within an appropriate checking domain, and that
checked uninterpretable features are erased (in much the same way as when checking a
shopping list, you cross off the items which you have bought). I shall assume (follow¬
ing Bobaljik 1995) that a head checks features of its specifier and its complement. I
shall further assume that all specifier- and complement-features are uninterpretable, as
are purely formal head-features (i.e. head-features with no intrinsic semantic content).
We can assume that UG specifies which head-features are interpretable and which are
not - e.g. that number-features are interpretable, but case-features are not.
To put our discussion on a concrete footing, let’s look at how checking works in the
case of our earlier sentence (2) She has gone. Consider first the grammatical features
carried by each of the three words in the sentence. (Throughout our discussion in this
chapter, we shall only be concerned with grammatical features: hence, phrase structure
trees are simplified by ignoring phonetic and semantic features, and LF representations
are simplified by ignoring semantic features.) The head-features of has indicate its
categorial properties (i.e. that it is an auxiliary verb) and its tense properties (i.e. that it
is present tense): its specifier-features tell us that it requires a third person singular
nominative subject; and its complement-features indicate that it requires a complement
headed by a verb in the n-participle form. The head-features of she tell us that it is third
person feminine singular determiner: and the head-features of gone indicate that it is an
n-participle form. Since neither she nor gone has a specifier or complement here, we
can assume that in the relevant uses their head- and complement-features specify that
they can be used without any specifier or complement. Given these assumptions, (2)
She has gone will have the grammatical structure (5) below (as noted above, we only
show grammatical/formal features):
174
5.3 Checking
(5) IP
I V
(3 = third person, F = feminine, S = singular, Nom = nominative case, Pres = present tense,
+n = n-participle; the blank entries for the specifier- and complement-features of she and
gone mean that (in this use) they don’t have a specifier or complement.) The information
within the square brackets in (5) provides an informal indication of the fact that she carries
the head-feature ‘third person feminine singular nominative’; has carries the head-feature
‘present tense’, the specifier-feature ‘requires third person singular nominative subject’
and the complement-features ‘takes an ^-participle complement’; and gone carries the
head-feature ‘n-participle’. The head-features of each item also include its categorial
features (which tell us that she is a determiner, has is an auxiliary verb, and gone is a
nonauxiliary verb), but these are represented by the category-labels (D, I and V) carried by
the relevant nodes in the tree.
Now consider how the process of (feature-) checking works in (5). For concreteness,
let’s make the following assumptions about checking:
175
Checking
Now consider what happens when the complement-features of has are checked
against the head-features of gone. These match exactly, since the [+n] complement-
feature of have tells us that it requires a complement headed by an rc-participle, and the
[+n] head-feature of gone tells us that it is an n-participle. If we assume that the inflec¬
tional properties of nonfinite verbs (like the case properties of pronouns) play no role
in semantic interpretation, both [+n] features will be erased.
So, checking in (5) will mean that the only grammatical features which survive in
the LF representation are those in (7) below:
(7) IP
All specifier- and complement-features have been erased (because they play no role in
semantic interpretation). Likewise, purely formal head-features (e.g. the nominative
case-feature of she and the [+n] inflectional feature of gone) have also been erased,
because they too are uninterpretable. The only grammatical features which survive at
LF are interpretable head-features. Since all the features in (7) are interpretable, (7)
satisfies the principle of full interpretation, and so the derivation converges at LF.
Of course, (7) is only a partial LF-representation: because we are concerned here
with how grammatical features are checked, we have not included purely semantic fea¬
tures in (7); a full LF representation for She has gone would clearly have to include the
semantic features of has (e.g. the fact that it is a perfective auxiliary) and gone (e.g. the
fact that it is a verb of motion). Hence, the orthographic representations of the words
she, has and gone in (7) should be understood as a shorthand abbreviation for the
semantic features of the relevant items: we could make this explicit e.g. by using capi¬
tal letters for the words which appear in LF representations (so that e.g. gone denotes
the semantic features carried by the word gone) - but rather than do this, we shall
simply take it as implicit in what follows (so that e.g. gone in (7) should be understood
as representing the set of semantic features which characterize its meaning).
Now let’s turn to consider how checking breaks down in an ungrammatical structure
such as:
(8)
r V
1
1
Them has go
head-features [3PObj] [Pres] [Inf]
specifier-features [3SNom]
complement-features [+n]
176
5.3 Checking
(9)
I)
r
1
1
Th em has go
head-features [3PObj] [Pres] [Inf]
specifier-features [SNom]
complement-features [+n]
But since numerous (bold-printed) uninterpretable features remain in (9) (viz. the [Obj]
case-feature of them, the [Inf] inflectional feature of go, the [SNom] specifier-features of
has, and the [+n] complement-feature of has), the resulting LF representation violates
PFI by virtue of containing uninterpretable features and so crashes at LF.
As we see in relation to structures such as (5) and (7-9), one set of features earned
by a finite auxiliary like has which have to be checked are its (third person singular)
agreement-features. However, since modal auxiliaries (such as can/could, may/might,
will/would, shall/should and must) never take the third person singular +5 inflection in
their present-tense forms, an interesting question which arises is what kind of specifier-
features modals carry. To make our discussion more concrete, consider the specifier-
It is clear that can imposes case restrictions on its choice of subject, since it requires a
nominative subject like they, not an objective subject like them or a genitive subject
like their. However, can doesn’t seem to impose any person/number restrictions on its
choice of subject, since it allows (for example) a first person singular/plural subject
like I/we, a second person subject like you or a third person singular/plural subject like
he/they. We shall therefore make the simplifying assumption that the only specifier-
177
Checking
feature carried by a modal like can is the case-feature [Nom], indicating that it requires
a nominative subject. This being so, a sentence like (10) will have the grammatical
(11)
D
(3 = third person; P = plural; Nom = nominative case; Pres = present tense; Inf = infini¬
tive.) The [Nom] specifier-feature of can will be checked against the [Nom] head-
feature of they, and both [Nom] case-features erased thereby (since case-features are
uninterpretable). The [Inf] complement-feature of can will be checked against the [Inf]
head-feature of swim, and both [Inf] features thereby erased (on the assumption that
inflectional features of nonfinite verbs are uninterpretable). Thus, the only features
which survive at LF are the interpretable [3P] third person plural head-features carried
by they, and the [Pres] present-tense head-feature carried by can.
5.4 Phrases
So far, we have only looked at elementary structures in which comple¬
ments and specifiers consist of single words rather than phrases. But consider how
checking works in more complex sentences such as (12) below, where the complement
of are is a [bracketed] phrase:
Let’s assume that (12) has the grammatical structure (13) below:
(13)
The head-features of the relevant items here indicate that they is a third person plural
nominative determiner (recall that the head-features of items include their categorial
features, indicated by the category-label they carry), are is a present-tense auxiliary,
178
5.4 Phrases
getting is a verb in its +ing form, and old is an adjective. The [2/PNom] specifier-
features of are tell us that it requires a second person or plural nominative specifier; the
[+ing] complement-feature of are specifies that it requires an +ing complement; and
the [A] complement-feature of get specifies that it requires an adjectival complement.
Consider now how the relevant complement-features are checked.
The [A] complement-feature of getting (requiring it to have an adjectival complement)
can be checked (and erased) straightforwardly, since the complement of getting is old,
and its head-features (which include its categorial features, represented by the category
label A in the tree) indicate that it is an adjective. But now consider how the [+ing]
complement-feature of the auxiliary are is checked. We earlier noted in (6) above that the
complement-features of a head are checked against the head-features of its complement.
The complement of are in (12) is the verb phrase getting old, which is a projection of the
head getting. Since getting is an ing-head, we might suppose that its projection getting
old is correspondingly an mg-projection, and hence can check the [+ing] complement-
feature of are (requiring are to have an ing-complement), so that both f+ing] features can
be erased. What we are tacitly assuming here is that it is in the very nature of the merger
operation by which phrases are formed that the head-features of a head are shared by all
projections of the head (e.g. a phrase headed by an infinitive verb form is an infinitival
verb phrase). If so, what a complement-feature such as [+ing] effectively means is
‘requires a complement headed by an +mg-participle form’.
We can treat phrasal specifiers in a similar fashion. In this connection, consider the
(14) fP
dp:^ ^Is.
D N
1I 1|
We students are protesting
head-features [3PNom] [P] [Pres] [+ing]
specifier-features [2/PNom]
complement-features [PN] [+ing]
179
Checking
We suggested in (6) above that the specifier-features of a head are checked against the
head-features of its specifier. Since the specifier of are is the DP we students, and this is
a projection of the first person plural nominative determiner we, it follows (from our
assumption that phrases share the same head-features as their heads) that the DP we
students will also be a first person plural nominative expression; as such, it can check
(and erase) the [2/PNom] specifier-features of are (with the uninterpretable nominative
case-feature carried by we (students) being erased at the same time). Thus, what a
specifier-feature like [2/PNom] in effect means is ‘requires a specifier with a second
person or plural nominative head’.
Having looked at how specifier- and complement-features are checked in simple
phrases, let’s consider how checking works in more complex structures such as:
PERF PROGP
V D
11
He might have been helping them
head-features [3MSNom] [Past] [Inf] [+n] [+ing] [3PObj]
specifier-features [Nom]
complement-features [Inf] [+n] [+ing] [Obj]
The (intrinsic) head-features in (16) (which include the categorial features represented
by the category labels) tell us that he is a third person masculine singular nominative
determiner, might is a past-tense auxiliary, have is a perfective auxiliary in its infinitive
form, been is a progressive auxiliary in its n-participle form, helping is a verb in its
mg-participle form and them is a third person plural objective determiner. But let’s
concentrate on the specifier- and complement-features of the relevant items, and how
they are checked.
180
5.4 Phrases
All purely formal features (e.g. case-features on pronouns and inflectional features on
nonfinite verbs) have been erased, so that the only grammatical features which remain
are interpretable person/number/gender/tense-features and interpretable categorial fea¬
tures such as PERF (indicating perfective aspect) and PROG (indicating progressive
aspect). (As already noted, we are concerned here only with grammatical/formal fea¬
tures and so omit purely semantic features from partial LF representations like (17)
above.)
Now let’s consider the corresponding ungrammatical example (15b) *Him might
having be helped they, and see what goes wrong there. We can assume that (15b) has
the (partial) grammatical structure (18) below (simplified by ignoring phonetic and
semantic features):
181
Checking
(18)
The head-features in (18) tell us that him is a third person masculine singular objective
determiner, might is a past-tense auxiliary, having is an ing-participle, be is an (irregular)
infinitive form, helped is an ^-participle (it could of course also be a past-tense form) and
they is a third person plural nominative determiner. The specifier- and complement-
features of the relevant items are as in our earlier discussion of (16) above.
Now consider how the various specifier- and complement-features in (18) are
checked. In fact, none of the specifier- and complement-features can be erased. Thus,
the [Nom] specifier-feature of might (requiring it to have a nominative subject) is
incompatible with the [Obj] head-feature carried by the objective pronoun him,
so neither case-feature can be erased. Likewise, the [Inf] complement-feature of
might (requiring an infinitive complement) is incompatible with the [+ing] head-
feature of the phrase headed by having; the [+n] complement-feature of having is
incompatible with the [Inf] head-feature of the phrase headed by be\ the [+ing]
complement-feature of be is incompatible with the [+n] head-feature of the phrase
headed by helped; and the [Obj] objective-case complement-feature of helped is
incompatible with the [Nom] head-feature of the nominative pronoun they. Since all of
these unerased features are uninterpretable case or inflectional features, the resulting
LF representation violates the principle of full interpretation, so causing the deriva¬
tion to crash at LF.
The distinction between interpretable and uninterpretable features has an interesting
correlate in the case of creole (i.e. hybrid) languages. It is often said that English-based
creoles have a simplified morphology compared to that of English. But simplified in
what ways? One way is by eliminating uninterpretable head-features, thereby simplify¬
ing the process of checking. For example, in Jamaican Creole (as described in Bailey
1966) personal pronouns inflect for person, number and (in the third person singular
only) gender, but not for case, so that pronouns have the following head-features {1/2/3
denote person, S/P denote singular/plural number, and In denotes inanimate gender;
English counterparts are given in inverted commas):
182
5.4 Phrases
The features which have been lost (compared to English) are uninterpretable case-
features; those which have been retained are interpretable person-, number- and
gender-features. We find much the same picture in relation to verb morphology, with
the loss of uninterpretable agreement inflections. For example, in a sentence such as:
the auxiliary a is invariable, and so doesn’t inflect for agreement at all. (I have glossed
a here by English are'. it should be noted, however, that Rickford 1986 and Harris 1986
posit that a is a reduced form of does - though this seems implausible for sentences
like Mi a tiicha, ‘I am (a) teacher’.) It would seem that in Jamaican Creole, only inter¬
pretable head-features survive.
The distinction between interpretable and uninterpretable features also has an inter¬
esting developmental correlate. We should expect that interpretable features (which
have semantic content) would be acquired more easily than uninterpretable features
(which are purely formal and have no intrinsic semantic content). In this connection, it
is interesting to note that young children acquiring English have problems in mastering
the case properties of pronouns and the agreement properties of finite verbs - as the
following child utterances illustrate:
In (21a) we find the objective pronoun me used where an adult requires the nominative
pronoun /; and in (21b) we find the uninflected verb form want where an adult requires
the agreement-inflected form wants. It seems to be uninterpretable features which pose
particular acquisition problems, in that (for example) the interpretable singular/plural
number features in nouns like dog/dogs are typically acquired before the uninter¬
pretable case-features of pronouns like I/me, and the interpretable past-tense feature in
verbs like went is typically acquired before the uninterpretable third person singular
agreement-features in forms like goes. There are interesting potential parallels between
Child English and Creole English, in that in both cases uninterpretable features are
‘lost’.
183
Checking
5.5 Percolation
After a brief look at Jamaican Creole and Child English, let's now return
to Standard English. Thus far, we have considered how checking works in clauses
headed by a finite auxiliary. But what about auxiliariless finite clauses such as the
following?
In the previous chapter, we argued that sentences like (22) are IPs headed by an
unfilled INFL constituent. If we further assume that the head-features of hates tell us
that it is a present-tense form, its complement-features tell us that (in this transitive
use) it requires an objective complement like him (cf. *She hates he), and its specifier-
features tell us that it requires a third person singular nominative subject (cf.
*Her/*They hates him), (22) will have the grammatical structure (23) below:
V D
il 11
S ie hates him
head-features [3FSNom] [Pres] [3MSObj]
specifier-features [3SNom]
complement-features [Obj]
It is clear that the objective complement-feature of hates can be checked against the
objective head-feature of him here (with both objective features thereby being erased,
because case-features are uninterpretable at LF), since him is the complement of hates.
But if we assume (as in (6) above) that the specifier-features of a head are checked
against the head-features of its specifier, we cannot check the specifier-features of
hates against the head-features of she here - for the obvious reason that the two are
contained within different phrases, she being the specifier of IP, and hates being the
head of VP (so that she is not the specifier of hates). However, if the relevant features
remain unchecked, the derivation will crash at LF, so wrongly predicting that sentences
like (22) She hates him are ungrammatical. How are we to overcome this problem?
In §4.4, we outlined a possible solution. We suggested that in auxiliariless finite
clauses, the grammatical features carried by the verb percolate from V to I. However,
clearly we don't want to assume that the complement-features of the verb hates (which
tell us that it requires an objective complement) percolate to I, since these are checked
internally within VP. So, let's assume that the complement-features of hates are first
checked internally within VP and erased, and then the remaining unchecked features of
hates (i.e. its head- and specifier-features) percolate from V to INFL, as shown by the
arrows in (24) below (where we assume that the objective complement-feature of hates
184
5.5 Percolation
and the objective head-feature of him have already been checked and erased internally
within VP):
(24)
We can suppose that the [Pres] present-tense head-feature of hates percolates from V to
INFL in order to ensure LF convergence: if we make the reasonable assumption that
INFL is only interpretable at LF if it carries a tense-feature, percolation of the [Pres]
feature from V to I provides a way of ensuring that INFL acquires a tense-feature and
can therefore be interpreted at LF. The assumption that the present-tense head-feature
of hates percolates up to INFL accounts for the fact that the corresponding tag-
sentence She hates him, doesn’t she? contains a present-tense auxiliary {doesn't), since
the auxiliary in a tag generally carries the same tense-features as the head INFL con¬
stituent of the clause to which the tag is attached (cf. e.g. She has finished, hasn’t
she?).
In much the same way, we might suppose that the [3SNom] specifier-features of
hates also percolate to INFL in order to ensure LF convergence: since specifier-fea¬
tures are uninterpretable, they must be checked and erased in the course of deriving the
relevant LF representation. But since hates and she are contained within different
phrases (hates is the head V of VP, and she is the specifier of IP), the only way for the
specifier-features of hates to be checked is for them to percolate up to INFL, where
they can then be checked against the corresponding [3SNom] head-features of she (by
virtue of being in the same IP-projection as she): checking will result in erasure of the
uninterpretable [3SNom] specifier-features of hates, and of the uninterpretable [Nom]
head-feature of she.
Erasing the relevant uninterpretable checked features in (24) yields the (partial) LF
representation (25) below (all surviving features are head-features):
(25)
She him
[3FS] [Pres] [3MS]
185
Checking
And (25) is essentially the same LF representation (in relevant respects) as is required
for a sentence such as:
(27)
Erasure (via checking) of the specifier- and complement-features in (27) along with
nominative/objective-case head-features of she/him and the infinitival head-feature of
the verb hate derives the LF representation (28) below:
And (28) is the same in the relevant respects as (26) - the difference in semantic inter¬
pretation between the two lying in whatever (contrastive or emphatic) content does has.
An interesting question which arises from our discussion here is why sentences such
as (29) below should be ungrammatical:
Let’s assume that (29) has the morphosyntactic structure (30) below:
(30)
186
5.6 Determiner phrases
One factor which causes the derivation to crash at LF is the fact that the [Inf]
complement-feature of does (requiring a complement headed by a verb in the infinitive
form) cannot be checked, since the [Pres] head-feature of hates indicates that it is a
present-tense form, hence not an infinitive. Another is that the specifier-features of does
and hates require them each to have a (third person singular) nominative subject, and
the single subject pronoun she can’t serve as the subject of both verbs. Why not? Because
if the nominative-case specifier-feature of does is checked against the nominative-case
head-feature of she, the nominative case-feature of she will thereby be erased (because
it is uninterpretable); and this will mean that the nominative specifier-feature of hates
cannot be erased (even if it percolates from V to INFL), because the nominative head-
feature of she has already been erased.
We can summarize our discussion so far in the following terms. The morphological
properties of lexical items can be described in terms of sets of head-, complement- and
specifier-features: uninterpretable features need to be checked in an appropriate domain.
Checking involves either a local specifier-head relation whereby the specifier-features of a
head are checked against the head-features of its specifier, or a local head-complement
relation whereby the complement-features of a head are checked against the head-features
of its complement. Where a feature carried by the head word of a phrase cannot be
checked within its containing phrase it may percolate up to the next highest head position
in the structure for checking purposes. Checking erases features which are uninterpretable
- hence, it erases all specifier- and complement-features, along with those head-features
(e.g. case-features of pronouns and inflectional features of nonfinite verbs) which play no
role in semantic interpretation. Any uninterpretable features which are not erased by
checking cause a derivation to crash.
Since the specifier-features of are (like those of all finite auxiliary and nonauxiliary
verbs) include the fact that are requires a nominative specifier, it is clear that the DP the
students must be nominative. Moreover, since it contains the plural noun students, it is
clearly plural (as we also see from the fact that it is used as the subject of are, and are
requires a plural or second person subject like we/you/they etc.). The DP the students
must also be a third person expression, since a DP headed by the determiner the can
only bind a third person reflexive anaphor like themselves, not a first or second person
reflexive: cf.
187
Checking
So, the DP the students in (31) must carry the features [3PNom], indicating that it is a
third person plural nominative expression. Since phrasal categories are projections of
the head-features of their heads, and since the DP the students is a projection of the
determiner the, it follows that the must carry the head-features [3PNom], and so be
third person plural nominative.
While it is plausible to suppose (on the basis of sentences like (32) above) that the is
intrinsically third person, we clearly can’t claim that it is intrinsically plural (since it is
singular in sentences such as The coursebook is boring), or that it is intrinsically nomi¬
native (since it is objective in sentences like / hate the coursebook). One solution
would be to suppose that the has variable number and case properties, and hence can be
either singular or plural in number, and either nominative or objective in case. This
seems a reasonable assumption: after all, determiners like this/these carry overt num¬
ber properties {this being singular and these plural), and pronominal determiners like
I/me, he/him, etc. carry overt case properties (/ being nominative and me objective):
hence, it is by no means implausible to suppose that determiners like the carry covert
(variable) number and case properties, since this enables us to arrive at a unitary
characterization of the feature specification of determiners as constituents which
carry overt or covert person/number/case properties.
One way in which we might think of variable feature-values is in the following
terms. Let’s suppose that when an item which has variable values for one or more
features is selected, it is pulled out of the lexicon with one fixed value for each such
variable feature. So, for example, if we use the as a nominative plural determiner in a
sentence like (31) above and are as a plural (rather than second person) auxiliary, (31)
will have the grammatical structure (33) below:
(33) IP
DP^
/ \
D N
i I
The students are complaining
head-features [3NomP] [P] [Pres] [+ing]
specifier-features [NomP]
complement-features [PN] t+ing]
The [NomP] specifier-features of are indicate that (in this particular use), it requires a
nominative plural subject. The subject/specifier of are is the DP the students: since this
is a projection of the determiner the, and since the [3NomP] head-features of the
indicate that (in this particular use) it is a third person nominative plural determiner, it
follows that the specifier-features of are can be checked and erased, along with the
uninterpretable nominative case-feature carried by the (students). It goes without say-
188
5.6 Determiner phrases
ing that if (for example) we used the as an objective singular determiner in (33), the
relevant case- and agreement-features carried by the students and are would be incom¬
patible, so could not be checked. (An interesting question of detail arising from the
analysis in (33) is whether we need to assume that the plural head-feature carried by
students is erased prior to LF, since it is redundant if the carries a plural head-feature:
an alternative way of dealing with this problem would be to suppose that the carries no
intrinsic number feature at all, and that the plural number-feature of students percolates
from N to D.)
It should be evident that our discussion here can be generalized from overt determiners
like the to covert determiners like the null determiner quantifying students in sentences
such as:
Given our arguments in §4.7 that bare nominal arguments like students are DPs headed
by a null determiner 0 with a generic or partitive interpretation, it follows from our
assumption that all determiners in English have person/number/case properties that the
null determiner 0 must have person/number/case properties of its own. It seems clear
that the relevant null determiner must be intrinsically third person, since a bare nomi¬
nal like students can only bind a third person reflexive like themselves - as we see from
sentences like:
It is also clear that the null determiner must be nominative plural in (34), since other¬
wise the DP 0 students would not be able to satisfy the requirements for are to have a
nominative plural subject. Thus, (34) has essentially the same structure as (33) above,
save for the fact that the head D position of DP is occupied by the null determiner 0
rather than by the overt determiner the.
One matter which we haven’t yet discussed concerns the checking of features internally
within determiner phrases. In this connection, consider the syntax of the bracketed subject
DP in:
189
Checking
allegations of incompetence
Let’s consider first the number features of relevant constituents of the DP in (37). It is
clear that the noun allegations is plural in number, since it carries the plural suffix +s.
Moreover, the adjective numerous is intrinsically plural (by virtue of its meaning). In
addition, the determiner the (as used here) must be plural, in order that the overall
phrase which it heads {the numerous opposition allegations of incompetence) should
be plural and so agree with the (third person plural) auxiliary are (cf. our earlier discus¬
sion of (31) above). The fact that the, numerous and allegations are all plural would
suggest that there is an agreement relation between them (hence the resulting sentence
becomes ungrammatical if the plural noun form allegations is replaced by its singular
counterpart allegation).
Consider first how we can account for the fact that the adjective numerous agrees in
number with the plural noun allegations. One answer (suggested in §4.9) is to suppose
that the plural number-feature carried by allegations percolates up the tree to the head
F position of FP, as indicated informally in (38) below (where the arrow indicates fea¬
ture percolation):
The [P] plural number-feature which the functional head F acquires via percolation can
then be checked against the plural number-feature carried by numerous (checking here
involving a spec-head relation between F and its specifier). Thus, percolation ensures
that allegations agrees with numerous.
But of course we also need to ensure that allegations agrees with the (and, more gen¬
erally, that determiners agree with the nouns they modify - cf. this/*these ridiculous
190
5.7 PRO subjects
allegation, where the determiner this/these has to agree with allegation). One way of
doing this would be to extend the percolation analysis and to suppose that the plural
number-feature of the noun allegations percolates first from N to F (as in (38) above),
and then from F to D - as in (39) below:
Once the plural number-feature of the noun allegations has percolated from N (through
F) to D, it can be checked against the plural number-feature carried by the determiner the
(as used here), and thereby erased. (An alternative to the analysis sketched here would be
to suppose that the has no number-feature of its own, and acquires its plural number-
feature by percolation from the noun allegations.) It will be apparent from (39) that per¬
colation (like movement) always proceeds in an upwards direction, never downwards.
It goes without saying that many details of this account of the internal syntax of DPs
remain to be filled in. For example, while it is clear that the determiner the carries (covert)
nominative case in sentences such as (36), it is less clear whether the noun allegations and
the adjective numerous also carry nominative case. In languages (like Latin or Russian)
with a much richer case morphology than English, nouns, attributive adjectives and deter¬
miners all inflect overtly for case: hence, we might conclude (on universalist grounds) that
the determiner the, the adjective numerous and the noun allegations all carry covert nomi¬
native case in (36), and that they agree with each other in case as well as number (with
case agreement being treated via percolation, as in (39) above). On the other hand, since
no nouns or adjectives overtly inflect for case in English, it might be argued to be against
the spirit of minimalism to posit that nouns and adjectives carry case properties which are
never overtly marked: on this alternative view, we might say that the noun allegations and
the adjective numerous carry (plural) number-features, but no case-features (case being a
property carried only by determiners in English). For obvious reasons, it is hard to find
compelling empirical evidence that nouns and adjectives do (or don’t) have covert case
properties in English: hence, we shall leave the issue open here.
191
Checking
interesting question of what features are carried by the null PRO subject in control
structures such as (40) below (where PRO is controlled by they), and how they are
checked:
If PRO is a null pronoun, and if pronouns carry case properties, an obvious conclusion
to draw is that PRO must carry case. Chomsky and Lasnik (1995, pp. 119-20) suggest
that just as PRO has a null phonological form, so too it carries null case. If so, how is
the null case carried by PRO checked? Chomsky and Lasnik suggest that it is checked
by infinitival to. What this implies is that infinitival to carries a null-case specifier-fea¬
ture which is checked against the null-case head-feature of PRO. So, for example, we
might suppose that the bracketed infinitive complement in (40) has the (simplified)
morphosyntactic structure (41) below:
(41)
PRO to escape
head-features [Null] [Inf]
specifier-features [Null]
complement-features [Inf]
The [Null] specifier-feature of to indicates that it requires a specifier with null case;
and the corresponding [Null] head-feature of PRO indicates that it carries null case.
The two [Null] case-features are checked against each other and thereby erased, since
case-features are uninterpretable.
There are potential parallels here with how nominative case is checked in finite-
clause structures such as (5) above, repeated as (42) below:
(42)
In (42), the [Norn] nominative-case specifier-feature carried by the finite INFL con¬
stituent has is checked against the corresponding [Norn] head-feature carried by its sub¬
ject she. Likewise, in (41) the [Null] null-case specifier-feature carried by the nonfinite
INFL constituent to is checked against the corresponding [Null] head-feature carried by
its PRO subject. In both types of structure, checking involves a spec-head relation
192
5.8 Objective subjects
(more specifically, a relation between INFL and its specifier), in that the case of the sub¬
ject is checked by the INFL constituent heading the IP containing the subject. We might
suppose that a finite INFL constituent checks nominative case (and so requires a nomi¬
native subject), whereas an infinitival INFL checks null case (and so requires a PRO
subject).
However, the overall picture is more complicated than this, since some verbs (like
those bold-printed in (43-4) below) which take a to infinitive complement don’t allow
their complement to have a PRO subject, and instead require it to have an objective
subject like him: cf.
One way of handling the relevant facts is to suppose that the infinitive particle to has two
different uses. In its use in control structures like (40) above, infinitival to checks null
case, and so requires a PRO subject with null case. But in its very different use in struc¬
tures like (43—4), to does not check null case and so does not allow a PRO subject. We
might further suppose that it is a lexical (i.e. word-specific) property of verbs like tty that
they select (i.e. ‘take’) an IP complement headed by null-case to (i.e. the kind of to which
can check null case), and conversely a lexical property of verbs like believe that they do
not. Thus, (40) is grammatical because try is a control verb (i.e. the kind of verb which
selects an BP complement headed by null-case to), and (43b) is ungrammatical because
believe is not a control verb (and so does not select an IP complement headed by null-case
to). On this account, verbs like try and believe differ in their complement-features (more
specifically, in respect of whether they do or don’t select an infinitive complement
headed by null-case to). Of course, a second premise of the analysis is that there are two
different uses of infinitival to - one on which it checks null case, and one on which it
does not.
193
Checking
An interesting property of IPs with objective subjects is that they occur only as the
complement of a transitive verb (like expect/consider/like), or of the transitive comple¬
mentizer/or. Since transitive verbs and complementizers have the property that they
check objective case, it seems reasonable to suppose (for example) that the objective
case of us is checked by the transitive verb expected in (45a), and likewise that the
objective case of them in (45d) is checked by the transitive complementizer for. Such
an analysis would correctly predict that sentences such as (46) below are ungrammati¬
cal:
This is because in none of the relevant examples is the bracketed infinitive clause the com¬
plement of a transitive verb or complementizer: in (46a), the bracketed IP is the subject of
would; in (46b) it is the complement of the adjective anxious (adjectives are intransitive);
and in (46c) it is the complement of the noun order (nouns too are intransitive).
If what we are suggesting here is along the right lines, it follows that the case of a
null PRO subject is checked in a different way from the case of an objective subject.
More specifically, the null case of a PRO subject is internally checked (from within IP)
by the null-case infinitive particle to\ but the case of an objective subject is externally
checked (from outside IP) by a transitive matrix verb or complementizer (i.e. by a tran¬
sitive verb or complementizer which takes the relevant IP as its complement). Since it
is exceptional for a subject to have its case externally checked from outside its contain¬
ing IP, the relevant phenomenon is widely referred to as exceptional case-marking
(conventionally abbreviated to ECM): similarly, an infinitive complement with an
objective subject is referred to as an ECM complement; and a verb which selects an
infinitive complement with an objective subject is referred to as an ECM verb. As we
shall see, the different ways in which the case properties of null and objective subjects
are checked are reflected in systematic asymmetries between control infinitives with
PRO subjects and ECM infinitives with objective subjects.
One such asymmetry relates to the behaviour of the relevant complements in active
and passive structures. A verb like decide (when used as a control verb taking an IP
complement headed by null-case to) allows a PRO subject irrespective of whether (as in
(47a) below) it is used as an active verb or (as in (47b) below) as a passive participle:
By contrast, a verb like believe can function as an ECM verb taking an infinitive com¬
plement with an objective subject only when used actively (as in (48a) below), not
when used passively (as in (48b) below):
194
5.8 Objective subjects
Why should this be? If (as we suggest) the null case of PRO in control structures like
(47) is checked by infinitival to, it makes no difference whether the matrix verb decide
is used in an active or passive form. By contrast, if (as we also suggest) the objective
case earned by the subject of an ECM infinitive clause is externally checked by a
transitive matrix verb (or complementizer), we correctly predict that only when the
matrix verb is used in an active form (and hence is transitive) will it allow an ECM
complement with an objective subject — not when used in a passive form (given the
traditional assumption that passive participles are intransitive).
A second asymmetry between control infinitives and ECM infinitives relates to
adverb position. An adverb modifying a control verb can be positioned between the
control verb and its IP complement, as we see from (49) below (where the adverb hard
modifies the verb tried):
How come? If we suppose that the case of PRO in control structures like (49) is
checked by infinitival to, we correctly predict that an adverb positioned between the
matrix verb and its IP complement will not prevent the case of PRO from being
checked. By contrast, if we posit that the case of him in (50) is checked by the transi¬
tive verb believes, we could conjecture that the presence of an intervening adverb
between the matrix verb and the objective subject will prevent the objective case of
him from being checked by the transitive verb believes. Since the transitive verb
believes in (50) does not immediately precede the pronoun him, believes cannot check
the objective case of him (with the consequence that the resulting sentence is ungram¬
matical by virtue of containing an unchecked case-feature).
Independent evidence that intervening adverbs block a verb from checking the
objective case of its complement (so that objective case can only be checked by an
immediately preceding transitive head) comes from contrasts such as the following:
In (51a), the transitive verb plays immediately precedes (and hence can check the
objective case carried by) its complement chess (which is a DP headed by the null
determiner 0); however, in (51b), the adverb well intervenes between the transitive
195
Checking
verb plays and its complement chess, so that the strict adjacency requirement for
checking objective case is not met, and case-checking cannot take place (so causing the
derivation to crash, since case-features are uninterpretable at LF).
We thus have a substantial body of empirical evidence that the objective case carried
by the subject of an ECM infinitive is externally checked from outside the infinitive
complement by an immediately preceding transitive matrix verb or complementizer.
One consequence of this is that infinitival to in ECM structures does not check the
objective case carried by the infinitive subject, but rather is a caseless particle. If to in
ECM structures is indeed caseless (and cannot check the case of its subject), it follows
that the objective case of an ECM infinitive subject cannot be checked from inside IP by
infinitival to, and hence (exceptionally) has to be checked from outside IP by an imme¬
diately preceding transitive matrix verb or complementizer. By contrast, in control
structures (as we saw in the previous section) to carries a null-case specifier-feature (and
hence requires a PRO subject); but in all other uses (e.g. in ECM structures) infinitival
to functions as a caseless particle which cannot check the case of its subject.
In the light of these observations, let’s now consider how we handle the syntax of
three different types of verb which take infinitival IP complements. Consider first verbs
like hope which allow a to complement with a PRO subject, but not one with an objec¬
tive subject like him: cf.
One way of describing the data in (52) is to specify in the lexical entry (i.e. dictionary
entry) for hope that it is an intransitive verb which selects an infinitival IP headed by
null-case to\ since null-case to requires a PRO subject, this accounts for the pattern in
(52) . An alternative possibility would be to suppose that hope selects an IP headed by
either type of to, but that (by virtue of being intransitive) hope can’t check the case of
an objective subject like him in sentences like (52b).
Now consider what we say about verbs like believe, which allow an IP complement
with an objective subject like him, but not one with a PRO subject: cf.
One way of handling the relevant data is to suppose that believe (in this use) is a transi¬
tive verb which selects an IP complement headed by caseless to: hence, since it doesn't
allow an infinitive complement headed by null-case to, it doesn’t allow a complement
with a null-case PRO subject. An alternative possibility would be to posit that believe
can select an IP complement headed by either kind of to, but that it is always transitive
when it takes an infinitival IP complement: this will mean that believe has to check
objective case, and this it clearly cannot do in a structure such as (53b) - at any rate, if
196
5.8 Objective subjects
we make the plausible assumption that neither infinitival to nor PRO can carry objec¬
tive case. (Of course, if the objective case-checking feature carried by believe is not
checked and erased, it will cause the derivation to crash at LF, since all case-features
are uninterpretable.)
Now consider a third class of verbs like expect which allow an infinitival IP comple¬
ment with either a PRO subject or an objective subject: cf.
One way of handling the data would be to suppose that expect has a dual use: in
sentences like (54a) it is an intransitive verb selecting an IP complement headed by
null-case to, whereas in sentences such as (54b) it is a transitive verb selecting an IP
complement headed by caseless to. Another possibility would be simply to say that
expect can select a complement headed by either kind of to, and can be either transitive
(and hence carry an objective case-checking feature) or intransitive.
Thus, the analysis of infinitival IPs presented here makes four crucial assumptions:
firstly, that there are two different kinds of infinitival to (one which checks null case,
and another which has no case-checking properties); secondly, that different items
select different kinds of infinitive complement (some select an IP headed by null-case
to, others an IP headed by caseless to)', thirdly, that some items are transitive (and in the
relevant use have to check objective case), others are intransitive, and yet others can be
either transitive or intransitive; and fourthly, that the case of an infinitive subject is
checked externally (by an immediately preceding transitive matrix verb or comple¬
mentizer) in structures where it cannot be checked internally.
If there are indeed two different uses of infinitival to (one which checks null case, and
another which is caseless), we should expect to find that they differ not only in their
case-checking properties, but also in other respects. A number of linguists have
suggested that the two different uses of infinitival to differ in their tense properties (cf.
Stowed 1982, Martin 1992, Watanabe 1993, Boskovic 1995 and Felser 1995). In this
connection, compare the bracketed control infinitive in (55a) below with the bracketed
ECM infinitive in (55b):
In (55a), the to-clause seems to carry different temporal properties from the hopes-
clause. (The hopes-clause refers to the present whereas the fo-clause refers to the future
- hence (55a) is paraphraseable as ‘Mary hopes she will be an actress’.) By contrast,
the to-clause in (55b) seems to have the same temporal properties as the believes-
clause (both have present time-reference, as we see from the fact that (55b) can be
paraphrased as ‘He believes that she is an actress’). More generally, we might suggest
197
Checking
that to in control infinitives has tense properties independent of those in the matrix
clause (Stowell 1982 notes that to in control infinitives often has future time-refer¬
ence), whereas to in ECM infinitives has tense properties determined by (and typically
simultaneous with) those of the matrix clause. If this is so, there is a correlation
between the tense properties of infinitival to and its case-checking properties, in that to
only checks null case in (control) structures where it has independent tense properties.
However, the proposed correlation between the tense and case properties of to is far
from unproblematic (e.g. it is not clear that to has different tense properties in (54a)
from those which it has in (54b)), so we shall not pursue this matter any further here.
To summarize: we have suggested in this section that infinitive complements with
objective subjects are IPs headed by caseless to, and that consequently the objective
case of the subject can only be checked from outside IP by an immediately preceding
transitive matrix verb or complementizer. We shall return in chapter 10 to consider the
precise way in which the exceptional objective case carried by the subject of an ECM
infinitive is checked, so for the time being we shall have no more to say about it here.
(56)
r)
""v
i |1
s: ie has gone
head-features [3FSNom] [Pres] [+n]
specifier-features [3SNom]
complement-features [+n]
However, there is a potential notational inconsistency here, in that the categorial prop¬
erties of words are represented by category labels attached to terminal nodes, whereas
other grammatical properties of words are represented by sets of features attached to
the words themselves. For example, the fact that she has the categorial status of a
determiner is indicated by the label D attached to the terminal node carrying the word
she, whereas the fact that she is third person feminine singular nominative is indicated
by the bracketed [3FSNom] head-features carried by she.
One way of eliminating this notational inconsistency would be to incorporate the
categorial properties of words into the head-features of the relevant items. This would
mean that terminal nodes would no longer carry category labels. In the true minimalist
spirit (seeking to get rid of all redundant apparatus), we might then go further and ask
whether the category labels carried by nonterminal nodes should also be eliminated.
198
5.9 Bare phrase structure
Perhaps (as suggested by Chomsky 1995a, 1995b) the only information which we need
about the constituents represented by nonterminal nodes is what is the head of each
such constituent. One way of providing this information would be for nonterminal
nodes to carry an indication of which item is the head of the relevant constituent - as in
(57) below (where the [D] head-feature of she indicates its categorial status as a deter¬
miner, the [I] head-feature of has indicates its categorial status as an auxiliary and the
[V] head-feature of gone indicates its categorial status as a verb):
(57) has
head-features
specifier-features [3SNom]
complement-features [+n]
A structure like (57) tells us that has gone and she has gone are both projections of has.
What our discussion here illustrates is that the economy principle (which proscribes
the use of superfluous symbols in structural representations) leads us to question
whether category labels play an essential role in syntactic representations, or whether
they are superfluous because the information which they encode is redundant. Of
course, we might go even further than in (57) and question whether nonterminal nodes
need to carry any labels at all, or whether structure can’t simply be represented in terms
of unlabelled tree diagrams such as (58) below:
(58)
head-features
specifier-features [3SNom]
complement-features [+n]
We might argue that the fact that has gone is a projection of has (not of gone) is pre¬
dictable from the lexical entries for the relevant items: that for has tells us that it takes
an n-participle complement, and that for gone does not allow for it to take an auxiliary
as its specifier. Similarly, we might argue that the fact that she has gone is not a projec¬
tion of she follows from the fact that she does not allow a complement of any kind.
An interesting question which arises in the wake of the bare phrase structure model
proposed by Chomsky relates to the question of whether (if nodes no longer carry
labels such as X/X/XP) the distinction between an XP (i.e. a phrase) and an X (i.e. a
head/word) can be defined in purely structural terms or not: in this connection, it is
199
Checking
interesting to note that Camie (1995, chapter 6) argues that indefinite complex nominal
predicates in Irish (e.g. the Irish counterpart of the bold-printed string in John is a big
man) behave like heads rather than phrases in respect of their syntax (though their
English counterparts behave like phrases rather than heads).
As is all too apparent from our discussion here, many key theoretical questions
remain unresolved for the moment; and indeed it must be apparent that many of the
descriptive assumptions we have made in our exposition here are also questionable
(e.g. the assumption that the features carried by n/mg-participles are purely formal and
play no role in semantic interpretation). To a large extent, this reflects the fact that
research in checking theory is only in its infancy. The initial success of any new
research paradigm should be judged in terms of whether it raises interesting questions,
rather than whether it provides definitive answers; and it is only to be expected that the
answers to many descriptive and theoretical questions must await the outcome of
future research. Given that the eventual outcome of questions about node labels is
uncertain, we shall continue to use traditional trees (in which each node carries an
appropriate category label) throughout the rest of the book - but the potential redun¬
dancy embodied in such conventional tree diagrams should be apparent.
5.10 Summary
In this chapter, we have discussed how the grammatical features carried by
words are checked. In §5.2 we suggested that the phonetic, grammatical and semantic
properties of words can be described in terms of sets of features. We argued that grammars
generate two types of structural representation for sentences - a PF representation (a
representation of their phonetic form) and an LF representation (a representation of their
logical form). We noted that the principle of full interpretation requires that PF repre¬
sentations should contain only phonetic features, and that LF representations should
contain only semantic features; a derivation which satisfies this requirement converges,
whereas one which does not crashes. We noted that some grammatical features (e.g. the
person/number/gender-features of pronouns like she) have semantic content and so are
interpretable (at LF), whereas others (e.g. the case-features of pronouns and the inflec¬
tional features of nonfinite verbs) are uninterpretable and so must be erased in the course
of the derivation (in order to ensure that they do not appear in LF representations). In §5.3
we argued that words carry three sets of grammatical features: head-features (which deter¬
mine their intrinsic grammatical properties), specifier-features (which determine the kinds
of specifier which they allow) and complement-features (which determine the kinds of
complement they can take): so, for example, the head-features of perfective has indicate
that it is a present-tense form, its specifier-features indicate that it requires a third person
singular nominative subject and its complement-features indicate that it requires an
^-participle complement. We suggested that specifier- and complement-features (as well
as those head-features which are purely formal and hence have no semantic content) are
200
5.10 Summary
201
Checking
Workbook section
Exercise I (§§5.2-5.4)
Draw tree diagrams representing the grammatical structure of the follow¬
ing sentences, and say how the features carried by each of the words in each sentence
are checked.
1 I helped him
2 She helps them
3 It may help him
4 They had helped us
5 You were helping me
6 We have been helping her
7 This could have helped you
8 He might be helping them
9 Someone must have been helping her
10 She might have been being blackmailed
(i)
V D
I helped him
head-features [ISNom] [Past] [3MSObj]
specifier-features [Nom]
complement-features [Obj]
202
Workbook section
V D
I J
I helped him
[IS] [Past] [3MS]
Since all features in (ii) are interpretable, the derivation satisfies the principle of full
interpretation and converges at LF.
Exercise II (§§5.2-5.4)
The sentences below (based on examples given in Miller 1989 and
Brown 1991) illustrate some of the ways in which modals can be used in Scottish
English:
Comment on the ways in which the morphosyntax of Scottish modals differs from that
of their Standard English counterparts.
203
Checking
0) IP
ModP
Mod
I 1
1 1
They might can come
head-features [3PNom] [Past] [Inf] [Inf]
specifier-features [Nom]
complement-features [Inf] [Inf]
1 Dem en si we
them been see we
‘They saw us’
204
Workbook section
2 Im a kom
him are come
‘He/she/it is coming’
3 Dem ena kos mi {ena = en+a = ‘been + are’)
them been+are curse me
‘They were cursing me’
4 Yu shudn en tel dem
you shouldn’t been tell them
‘You shouldn’t have told them’
5 Im shuda en a ron
him should been are run
‘He should have been running'
6 Im wuda mos hafi priti {hafi = ha+fi = ‘have + for’)
him would must have+for pretty
‘She would have to (be) pretty’
7 Mi en nuo se im wudn kom
me been know say him wouldn’t come
‘I knew that he wouldn’t come’
8 Mi miin se yu fi go
me mean say you for go
‘I mean that you should go’
9 Mi waan yu fi sel i
me want you for sell it
‘I want you to sell it’
10 Im waan fi haid dem
him want for hide them
‘He wants to hide them’
11 Yu fi tikya dem
you for take-care them
‘You should be wary of them’
205
Checking
their subject (i.e. a D or DP) rather than a nominative subject, and transitive verbs
require a D-projection as their complement rather than an objective complement. On
this (minimalist) view, 1 would have the morphosyntactic structure (i) below:
V D
| |
1 1
Dem en si wi
head-features [3P] [Past] [IP]
specifier-features [D]
complement-features [V] [D]
The bracketed head-features indicate that dem is third person plural, en is past tense
and wi is first person plural. (The categorial head-features of the words are indicated by
their category labels.) The [D] specifier-feature of en (indicating that it requires a sub¬
ject which is a projection of the category D) can be checked and erased, because its
subject dem belongs to the category D. The [V] complement-feature of en (requiring it
to have a complement which is a projection of the category V) can be checked and
erased because its complement is the VP [si wi], and this is headed by the V si. The [D]
complement-feature of see (requiring it to have a co.,iplem .. wnich is headed by the
category D) can be checked and erased because its complement is the D wi. So, check¬
ing yields the LF representation (ii) below:
(ii)
VP
Dem en wi
[3P] [Past] [IP]
Since the only surviving grammatical features are interpretable head-features, (ii) satis¬
fies the principle of full interpretation, so the derivation converges at LF.
If the analysis in (i—ii ) above is along the right lines, it suggests that all head-features
in JC are interpretable (e.g. there are no uninterpretable case-features or agreement-
features), and that all specifier- and complement-features are purely categorial in
nature. However, this analysis may be oversimplistic in that it ignores the distinction
between overt pronouns like mi ‘me’ and the covert pronoun PRO. It may be that rather
than say that JC has no case-system, we should instead say that it has a two-way dis¬
tinction between overt case and covert/null case, so (for example) mi ‘me’ carries overt
case, and PRO carries null case. Since the distinction between overt and covert forms
206
Workbook section
is one which is visible in the PF component (because overt items have phonetic fea¬
tures, but covert items do not), we might argue that the overt/covert distinction is the
minimal case distinction we should expect to find in any language.
Helpful hints
The form ena glossed as ‘were’ is composed of the past-tense auxiliary en
‘been’ and the progressive auxiliary a ‘are’; treat them as separate words heading sepa¬
rate projections in the syntax. (They subsequently fuse together to become a single
word in the PF component, we might suppose.) Although fi is cognate with English for,
assume that in JC its function is more akin to that of the English infinitive particle to:
likewise, assume that the form hafi is composed of the verb ha ‘have’ and the infinitive
particle/z ‘for’, and that they are separate words heading separate projections in the
syntax (subsequently fusing together to form a single word in the PF component).
Recall from §2.9 that adjectives have verblike properties in JC (e.g. they can serve as
predicates, occur as the complement of a modal, etc.). The word se (cognate with
English say) is generally assumed to serve as a complementizer (corresponding
roughly to English that) in JC.
Exercise IV (§§5.2-5.5)
The utterances below (edited in minor ways) were produced by children
acquiring English as their native language (the names of the children producing the
examples are given in parentheses, together with their age in years;months at the point
when the examples were produced). Comment on the nature of errors made by the chil¬
dren in the way they use auxiliaries.
207
Checking
(here used in its negative imperative form don't!) has the complement-feature [Inf],
and so requires an infinitive complement (i.e. a complement headed by a verb which
carries the head-feature [Inf]). However, the complement of don’t in 1 is a verb (fight¬
ing) in the mg-participle form, suggesting that Adam has not mastered the complement-
features of auxiliaries as yet. This is consistent with the suggestion made in the text
that children have problems in acquiring uninterpretable features, and that the
inflectional features of nonfinite verbs are purely formal (hence uninterpretable, and
therefore difficult to acquire). In this particular case, it may be that Adam has confused
the negative imperative don’t! with its positive imperative synonym stop! (cf. Stop
fighting me!).
Exercise V (§5.7)
Discuss the derivation of the following sentences, drawing tree diagrams
to illustrate your answers.
208
Workbook section
(i) IP
In the text, we suggested that infinitival to can function either as a null-case particle
or as a caseless particle, and that some verbs select an IP complement headed by null-
case to, and others select a complement headed by caseless to. On the basis of data like
1, we might suppose that it is a lexical (= word-specific) property of the verb plan
(when used with an infinitive complement) that it is an intransitive verb which selects
an IP headed by null-case to, but not an IP headed by caseless to. If so, the null-case
head-feature carried by PRO will be checked against the null-case specifier-feature of
to in (i)(a), with both case-features thereby being erased. Example (i)(b) will be
ungrammatical because the null-case specifier-feature of to is incompatible with the
objective-case head-feature of him, with the result that both case-features remain
unchecked and cause the derivation to crash at LF (since case-features are uninter¬
pretable).
Evidence that the null case carried by PRO is checked by the infinitive particle to
(rather than by the verb planned) comes from the fact that even when used as a passive
participle, the verb planned still allows an IP complement with a PRO subject: cf.
Moreover, an adverb modifying plan can freely be positioned between plan and its
complement: cf.
In these respects, plan behaves like a typical control verb, and unlike an ECM verb.
209
Checking
Exercise VI (§5.7)
The following examples (cf. Henry 1995) illustrate infinitive construc¬
tions with for found in (one variety of) Belfast English:
Discuss the syntax of the bracketed infinitive complement in each of these examples,
paying particular attention to the morphosyntax of its subject.
him I V be amazing
I I
to win
We might suppose that (as in Standard English) for takes an IP complement headed by
caseless to, and that the objective-case head-feature of him is checked by the immedi¬
ately preceding transitive complementizer for.
At first sight, it would seem as if 2 has much the same structure as (i), save that the
infinitive subject is PRO rather than him. If this were so, 2 would have the structure (ii)
below:
210
Workbook section
(ii)
VP
V
/ \A
I I
be amazing
We could then say that for in Belfast English (unlike Standard English) can serve as an
intransitive head taking an IP complement headed by null-case to, and that the null
case of PRO is checked by infinitival to.
However, an alternative possibility is to suppose that (via a process of cliticization
similar to that by which want to becomes wanna) forto has become a compound infini¬
tive particle in Belfast English, with a syntax similar to that of Standard English to. On
this alternative analysis, 2 would have the very different structure (iii) below:
(iii)
VP
would V
I
forto win be amazing
If we assume that forto in (iii) has essentially the same case properties as Standard
English null-case to, it can check the case of a null PRO subject. One potential compli¬
cation with this analysis is that we need to account for the fact that forto isn’t used in
infinitive structures like 1 which are introduced by the infinitival complementizer for,
as we see from the ungrammaticality of:
Helpful hint
Consider the possibility that whether functions as an interrogative com¬
plementizer in Belfast English (even though it isn’t a complementizer in Standard
English).
211
Checking
1 I be a lion
2 I got my coke
3 I got you coke (=‘I got your coke’)
4 She drink she bottle
5 She don’t pick she food (= ‘She shouldn’t pick at her food’)
6 We got we drink
7 Him don’t want it
8 Him need sit down (= ‘He needs to sit down’)
9 Him hasn’t got him supper, he haven’t
10 Me can mix it
11 She said she want the net
12 I don’t want she help me
Helpful hints
In the text, we didn’t discuss how genitive case is checked. For the pur¬
pose of this exercise, make the following assumptions. Just as nominative case is asso¬
ciated with the specifier of a functional head (INFL) which is often null, so too genitive
case (in adult English) is associated with the specifier of a null determiner constituent.
More concretely, assume that the bracketed DPs in sentences such as:
(i) (a) It may be faster than [my car], but [mine] has more class
(b) [The queen’s car] is faster than [the king Y]
have the respective structures indicated in (ii) below (where e is an empty category
whose antecedent is the noun car):
Assume that the null determiner 0 has the specifier-feature [Gen], and so requires a
specifier with genitive case - i.e. a genitive determiner like my/mine, or a genitive DP
like the queen’s/the king’s (carrying the genitive suffix’s).
212
Workbook section
agreement-features of finite verbs and auxiliaries). The fact that Knox uses him as the
subject of hasn’t in 9 suggests that he hasn’t yet fully acquired the specifier-features of
finite verbs: he seems to think that finite verbs allow either a nominative or an objec¬
tive subject (and some children also use genitive subjects: cf. My did get my leg dry,
produced by a girl called Betty at age 2;6). The fact that Knox uses he as the subject of
haven’t may suggest that he analyses forms like haven’t as imposing no person/number
restrictions on their choice of specifier (like adult modals); by contrast, in adult
English, have(n’t) allows only a first person, second person or plural subject). It may
well be that ^-inflected forms are correctly specified for agreement (e.g. forms like has,
wants, etc. are only used with third person singular subjects), but that uninflected forms
like have, want, etc. are unspecified for agreement (i.e. they carry no agreement
features, and so impose no person/number restrictions on their choice of subject).
The use of the objective pronoun him in him supper (where adult English would
require the genitive form his) might be accounted for along similar lines. Assuming the
analysis in the helpful hints, him supper will have the structure (i) below:
him D N
0 supper
We could then say that Knox hasn’t yet mastered the specifier-features of the null
possessive determiner 0, since he allows it to have an objective specifier like him
(whereas adults require a genitive specifier like his). It may be that Knox assumes that
0 has the categorial specifier-feature [D], and so allows a specifier which is a
D-projection, irrespective of whether it has nominative, objective, genitive or even null
case: this would lead us to expect that Knox alternates between e.g. Teddy eat he din¬
ner, Teddy eat him dinner, Teddy eat his dinner and perhaps even Teddy eat PRO
dinner.
At any rate, what our analysis here suggests is that children like Knox have prob¬
lems in acquiring the specifier-features of items (viz. the case and agreement restric¬
tions which finite verbs and auxiliaries impose on their choice of specifier, and the case
restrictions which the possessive determiner 0 imposes on its choice of specifier). More
generally, it would seem that such items are underspecified with respect to their speci¬
fier-features at the relevant stage of development.
213
Checking
Helpful hints
See those for the previous exercise.
(i)
The head-features in (i) indicate that him is a third person masculine singular objective
pronoun, want is a present-tense verb and some is an objective pronominal quantifier
214
Workbook section
(as used here). The [Obj] specifier-feature of want indicates that it allows an objective
subject; its [Obj] complement-feature indicates that it takes an objective complement.
The [Obj] complement-feature of want can be checked against the [Obj] head-feature
of some, with both case-features being erased in consequence. The head- and specifier-
features of want percolate up from V to I, as indicated by the arrows in (i). Percolation
of the [Pres] head-feature of want from V to INFL ensures that INFL carries a
tense-feature, and so is interpretable at LF. Once the [Obj] specifier-feature of want
percolates to INFL, it can be checked against the [Obj] objective-case head-feature of
him, resulting in the erasure of both case-features. Thus, the only grammatical features
which survive at LF are those indicated below:
(ii)
Q
Him want some
[3MS] [Pres]
Since the surviving grammatical features are all interpretable, (ii) converges at LF.
On this view, the major deficit in the SLI child’s grammar in respect of sentences
like 1 relates to the specifier-features of finite verbs like want. In English, a present-
tense form like want requires a nominative subject which is first person, second person,
or plural (i.e. which is not third person singular). The child has thus not yet mastered
the case/person/number specifier-features of uninflected finite verb forms. It may be
that the child’s initial hypothesis is that only verbs which carry overt agreement inflec¬
tions impose person/number restrictions on their choice of subject: by contrast, in
Standard English, the present-tense form want contains a null agreement inflection
which requires a first or second person or plural nominative subject. Since specifier-
features are uninterpretable, the more general conclusion we reach is that SLI children
have problems in acquiring uninterpretable features.
215
6
Head movement
6.1 Overview
So far, the analysis we have presented has assumed that syntactic struc¬
tures are derived by a series of binary merger operations. We now go on to argue that
derivations may involve not only merger, but also movement operations. In this chap¬
ter, we look at two similar types of movement operation, one which affects auxiliaries
in Modem English, and another which affected verbs in earlier stages of English. We
shall argue that both types of movement involve essentially the same head movement
operation, involving movement from one head position to another. We begin by look¬
ing at the syntax of so-called auxiliary inversion in English.
216
6.2 Auxiliary inversion
What's the structure of the two bold(-printed) proposals which speaker A makes in (1)?
The answer is straightforward enough in the case of If you will marry me: it’s a clause
introduced by the complementizer if, and so is a CP (complementizer phrase) con¬
stituent, with the structure (2) below:
(2)
IP
will V D
marry me
But now consider the more problematic question of the structure of the second pro¬
posal, Will you marry me? Here we have an auxiliary inversion structure, in which the
auxiliary will appears in front of the subject you. What position is being occupied by
the inverted auxiliary will! Since will appears to occupy the same presubject position
that the complementizer if occupies in (2), a natural suggestion to make is that the
inverted auxiliary actually occupies the head C position of CP. If this is so, then we’d
expect to find that will and if are mutually exclusive (on the assumption that we can
only insert one word in a given head position like C, not two words): in other words, if
both complementizers and inverted auxiliaries occupy the head C position of CP, we’d
expect to find that a clause can be introduced either by a complementizer or by a
preposed auxiliary - but not by the two together. This is indeed the case, as we see
from the ungrammaticality of speaker B's reply in (3) below:
The fact that no clause can contain both a complementizer and an inverted auxiliary
provides us with strong empirical evidence that inverted auxiliaries occupy the same
structural position as complementizers - i.e. that both occupy the head COMP position
in CP.
But how can it be that a finite auxiliary (which normally occupies the head INFL
position within IP) comes to be positioned in the head COMP position of CP? The answer
provided by descriptive grammarians is that auxiliaries move out of their normal postsub¬
ject position into presubject position, by an operation traditionally referred to as inversion.
In terms of the framework being used here, this would mean that an inverted auxiliary
moves from the head I position in IP into the head C position in CP, as in (4) below:
217
Head movement
will V D
marry me
This type of inversion operation involves movement of a word from the head position
in one phrase into the head position in another phrase (in this case, from the head I
position of IP into the head C position of CP), and so is known more generally as head-
to-head movement (or head movement).
An obvious question which is begged by the head movement analysis is why auxil¬
iaries should undergo inversion in questions. Using a metaphor developed by Chomsky
(1995c), let’s say that COMP in an interrogative clause is a strong head, and that a
strong head has to be filled. In a complement-clause yes-no question like that brack¬
eted below, COMP can be filled by the complementizer if.
(Speaker A’s first proposal in (1) can be regarded as an elliptical form of I wanted to
ask you if you will marry me, with if introducing the bold-printed complement clause.)
However, complementizers can’t be used to introduce main clauses in English, so
some other way has to be found of filling the strong COMP node in main-clause
questions. A strong COMP node has the power to lure an auxiliary from INFL to
COMP (as in (4) above), thereby satisfying the requirement for a strong COMP to be
filled.
The assumption that some categories are strong (and conversely others are weak)
provides us with an interesting account of an otherwise puzzling property of questions
in English - namely the fact that the question counterpart of a statement which con¬
tains no auxiliary requires the use of the (dummy or expletive) auxiliary do, as we can
see from sentences such as the following:
Why should this be? One answer would be to suppose that an interrogative COMP is
strong in present-day English (and so has to be filled) whereas INFL is weak (and so
doesn’t have to be filled). Since INFL is weak, it can be left empty in sentences such as
(6a), which has the structure (7) below:
218
6.3 Traces
know him
However, if we assume that interrogative clauses are CPs headed by a strong C, and that
complementizers can’t be used to fill COMP in main clauses, the only way of filling
COMP is to resort to generating the auxiliary do in INFL and then raising it from INFL
to COMP (to satisfy the requirement for a strong COMP to be filled) as in (8) below:
(8)
V D
know him
6.3 Traces
An interesting question which arises from the assumption that auxiliaries in
questions move from I to C is what happens to the head I position in IP once it is vacated
by movement of the inverted auxiliary into C. What we shall argue here is that the head I
position of IP remains in place, but in the form of an empty category. What properties
does this empty category have? It seems clear that the empty category left behind in the
head I position of IP in a structure such as (4) must have the same head-features as will,
since (like will in (2) above) it occupies the head I position of IP. Likewise, the empty I in
(4) must have the same specifier-features as will, since (like will) it requires a nominative
219
Head movement
specifier (cf. Will she/*her marry me?). Moreover, the empty I must have the same
complement-features as will, since the head V constituent of the VP has to be in the
infinitive form marry (and cannot e.g. be in the +ing form, cf. * Will you marrying me?).
So, it would seem that the empty category left behind in I by movement of will from
I to C in (4) has the same head-features, specifier-features and complement-features as
will. The empty category would therefore seem to be a silent copy of will - i.e. a
constituent which has the same grammatical properties as will, and which differs from
will only in that it has no phonetic content. To use the relevant terminology, we can say
that when the auxiliary will moves, it leaves behind (in the position out of which it
moves) an empty trace of itself, and that this trace (by virtue of being a silent copy of
will) has precisely the same grammatical features as will. (The romantics among you
can think of traces as being like the footprints you leave behind in the sand when you
walk along the beach in Mallorca or Malibu.) Given this assumption, our earlier
question Will you marry me? has the superficial structure (9) below (simplified e.g. by
not representing features):
t V D
marry me
in which the head I position of IP is filled by an abstract trace t of the moved auxiliary
will. The assumption that moved constituents leave behind a (silent) trace of them¬
selves is the cornerstone of what became known in the 1970s as trace theory. Using
the relevant terminology, we might say that a moved constituent is the antecedent of
its trace, and that the antecedent of an empty trace serves to bind the trace - i.e. to
determine its properties. (This involves a slightly extended use of the terms
antecedent/bind compared with our earlier use of them in §3.9.) We might further posit
that a moved constituent and its trace together form a (movement) chain, so that C and
I form a chain in (9). Extending the chain analogy still further (to gratify the bikers
among you), let’s say that the moved auxiliary will in C and its empty trace t in I are
the two different links of the relevant movement chain. Let’s also say that the moved
constituent is the head of the associated movement chain, and the trace is the foot of
the chain. In our ensuing discussion, we shall adopt the convention of using the same
type-face to denote a trace and its antecedent (hence both the trace t and its antecedent
will are printed in italics in (9) above): the general convention used in the literature is
to mark the binding relation between a trace and its antecedent by attaching identical
subscript letters - called indices - to them (e.g. wili and t).
220
6.3 Traces
‘Why on earth should we want to assume that moved constituents leave behind an
invisible empty category trace!' you might wonder at this point. There is both theoreti¬
cal and empirical evidence in support of this assumption. The relevant theoretical
evidence comes from the fact that trace theory enables us to explain an otherwise puzz¬
ling property of movement operations. It will become more evident as our exposition
unfolds that moved constituents always move from a lower to a higher position in any
given structure, never from a higher to a lower position: for example, the moved
auxiliary will in (9) moves from the head I position in IP into the head C position in CP,
and thereby moves from a lower to a higher position in the structure. Why should
movement always be from a lower to a higher position? Trace theory provides us with
a natural explanation for the fact that movement is always upwards, never downwards.
If we assume that a moved constituent leaves behind a trace which it binds, then the
upward nature of movement is a direct consequence of the c-command condition on
binding which we posited in our brief discussion of anaphors in §3.9 (where we saw
that a reflexive anaphor like himself must be c-commanded by its antecedent). If a
moved constituent has to bind its trace, and if a bound constituent has to be
c-commanded by its antecedent, it follows that a moved constituent must always move
into a position where it c-commands (and hence occurs higher up in the structure than)
its trace: hence, movement will always be in an upwards direction. For example, in
(9) above the moved auxiliary will in C c-commands its trace in I by virtue of the fact
that (using our train analogy) if you travel one stop on a northbound train from C you
arrive at CP, and from there you can catch a southbound train to I (via IP and I). So, one
theoretical argument in support of trace theory is that it explains the upward nature of
movement.
A second theoretical reason for positing that a moved constituent leaves behind an
empty category trace (and hence e.g. that there is an empty I constituent in inversion
structures such as (9) above) relates to the headedness property of projections. If all
phrases and clauses are projections of a head word category (as we argued in chapters
3 and 4), then IP must be headed by an I constituent; and if there is no overt I con¬
stituent in (9), there must be a covert one.
In addition to theory-internal considerations such as these, there is also empirical evi¬
dence for claiming that a moved constituent (e.g. the moved auxiliary will in (9) above)
leaves behind an empty category trace. Part of this evidence comes from familiar facts
about /iave-cliticization. In this connection, note that have cannot cliticize onto the imme¬
diately preceding pronoun welllyoulthey in inversion structures such as the following:
221
Head movement
(’ve represents the vowel-less clitic form /v/ here.) The sequence we’ve in (10a) does¬
n’t rhyme with weave (in careful speech styles), since we have can be reduced to /wiov/
but not /wiv/: similarly, I’ve doesn’t rhyme with hive in (10b), nor you’ve with groove
in (10c), nor they’ve with grave in (lOd). Why should cliticization of have onto the
pronoun be blocked here? We can give a straightforward answer to this question if we
posit that inverted auxiliaries move from I to C, and leave behind an empty category
trace t in the I position out of which they move. Given this assumption, a sentence such
as (10a) will have the structure (11) below. (To simplify exposition, details not of direct
relevance to the discussion at hand are omitted - e.g. all features have been omitted,
and the internal structure of the VP finished the rehearsal is not shown.)
(11) cp
^)IP
1
1
Will I
|
1
we Y' PERFP
1I
t PERF ^.VP
|
1
have finished the rehearsal
It would seem natural to suggest that the fact that there is an empty trace t intervening
between have and we prevents have from cliticizing onto we in structures such as (11),
since cliticization is subject to an adjacency condition (in that the clitic must be imme¬
diately adjacent to its host in order for cliticization to be possible). More generally,
cliticization facts lend empirical support to the claim that auxiliary inversion results in
an IP headed by an I constituent which is filled by a trace of the moved auxiliary.
Our discussion of auxiliary inversion here has interesting implications for the
derivation of sentences (i.e. the way in which they are formed). More specifically, it
implies that derivations may involve both merger and movement operations. For
example, our earlier sentence Will you marry me? is formed by merging marry with me
to form the VP marry me\ merging this VP with the auxiliary will to form the I-bar will
marry me\ merging the resulting I-bar with you to form the IP you will marry me; and
finally merging this IP with a C constituent into which the auxiliary will moves,
forming the CP Will you marry me? (and leaving a trace behind in I, as in (9) above).
222
6.4 Verb movement
longer productive in Modem Standard English (= MSE). Since part of the evidence for
V movement involves negative sentences, we begin with a brief look at the syntax of
negation in EME.
In Shakespearean English, clauses containing a finite auxiliary are typically negated
by positioning not between the auxiliary and the verb: cf.
(12) (a) You may not deny it (Princess, Love’s Labour’s Lost, V.ii)
(b) I would not lose you (Portia, Merchant of Venice, Ill.ii)
(c) Thou shalt not die for lack of a dinner (Orlando, As You Like It, II.vi)
(d) I will not hear thy vain excuse (Duke, Two Gentlemen of Verona, Ill.i)
It would seem plausible to suppose that not in EME is an adverb which is positioned
internally within VP, to the left of the verb. If so, (12a) will have a structure along the
lines of (13) below:
(13) IP
may ADV V
I / \
not V D
I I
deny it
An analysis such as (13) accounts for the position which not occupies in front of the
verb deny. It also enables us to provide a straightforward account of interrogatives such
as:
(14) (a) Didst thou not hear somebody? (Borachio, Much Ado About Nothing, LH.iii)
(b) Will you not dance? (King, Love’s Labour’s Lost, V.ii)
(c) Have I not heard the sea rage like an angry boar? (Petruchio, Taming of
the Shrew, I.ii)
If interrogatives involve movement from INFL to COMP, then a sentence such as (14a)
will have the derivation (15) below:
not V D
I
hear somebody
223
Head movement
As (15) shows, the auxiliary didst originates in I and moves to C, leaving behind a
trace t in the position out of which it moves. The assumption that not is contained
within VP provides a straightforward account of the fact that not remains positioned in
front of the verb hear when didst is preposed.
However, what is particularly interesting about negative sentences in Shakespearean
English is that in auxiliariless finite clauses, the main verb is positioned in front of the
negative not: cf.
(16) (a) He heard not that (Julia, Two Gentlemen of Verona, IV.ii)
(b) I care not for her (Thurio, Two Gentlemen of Verona, V.iv)
(c) My master seeks not me (Speed, Two Gentlemen of Verona, I.i)
(d) I know not where to hide my head (Trinculo, The Tempest, Il.ii)
(e) Thou thinkest not of this now (Launce, Two Gentlemen of Verona, IV.iv)
(f) She lov’d not the savour of tar (Stephano, The Tempest, Il.ii)
(g) My charms crack not (Prospero, The Tempest, V.i)
(h) Demetrius loves her and he loves not you (Lysander, Midsummer Night’s
Dream, 3.ii)
If we assume that not in EME occupied a preverbal position internally within VP, how
can we account for the fact that the verb (which would otherwise be expected to follow
the negative not) ends up positioned in front of not in sentences like (16)? An obvious
answer is to suggest that when INFL is not filled by an auxiliary, the verb moves out of
the head V position in VP into the head I position in IP. If this is so, (16a) He heard not
that will have the derivation (17) below:
(17)
D
I
He
Thus, the verb heard originates in the head V position within VP, and then moves
across not into the head I position in IP, so accounting for the fact that heard is posi¬
tioned in front of not.
There would seem to be significant parallels between V-to-I movement in (17) and
I-to-C movement in (15). For one thing, in both cases movement is from one head
position to another, and so is an instance of the more general operation of head
movement. Moreover, in both cases movement is from a lower to a higher position,
and so results in a structure in which the moved head c-commands its trace (thereby
satisfying the c-command condition on binding). Furthermore, in both cases move-
224
6.4 Verb movement
ment is local, with the moved head being moved into the head position in the next
highest phrase within the structure; and in both cases, the moved head can move across
an intervening nonhead constituent. Thus, didst in (15) moves from the head I position
in IP into the head position within the next highest phrase in the structure (into the head
C position in CP), and in doing so moves across its specifier thou; likewise, heard in
(17) moves from the head V position in VP into the head position in the next highest
phrase in the structure (= the head I position of IP), and in doing so moves across the
adverb not. Since both V-to-I movement and I-to-C movement are local operations, it
seems reasonable to suppose that their locality property is not accidental, but rather
reflects the operation of some principle of Universal Grammar. Lisa Travis (1984)
suggested that the relevant principle is a head movement constraint (= HMC) to the
effect that a head can only move from the head position in one phrase to the head
position in the immediately containing (i.e. next highest) phrase in the structure.
As we see from the examples in (14) above, questions in EME (= Early Modem
English) involved the same inversion operation as in MSE (= Modem Standard
English). Given our assumption that inversion involves movement from I to C, an
obvious prediction made by the assumption that verbs move from V to I in EME is that
they can subsequently move from I to C, so resulting in sentences such as:
(18) (a) Saw you my master? (Speed, Two Gentlemen of Verona, I.i)
(b) Heard you this, Gonzalo? (Alonso, The Tempest, II.i)
(c) Speakest thou in sober meanings? (Orlando, As You Like It, V.ii)
(d) Call you this gamut? (Bianca, Taming of the Shrew, Ill.i)
(e) Came you from the church? (Tranio, Taming of the Shrew, ULii)
(f) Know you not the cause? (Tranio, Taming of the Shrew, IV.ii)
(g) Spake you not these words plain? (Grumio, Taming of the Shrew, I.ii)
If so, a typical Early Modem English question such as (18f) Know you not the cause?
would be derived in the manner represented in (19) below:
not V DP
i -(1)-!i D
/ \N
the cause
The fact that the verb know is positioned to the left of the subject you suggests that it is
raised first from V to I and then from I to C by two successive applications of head move-
225
Head movement
ment (numbered (1) and (2) respectively). In structures like (19), head movement is said
to apply in a successive cyclic fashion, moving the verb know (in successive steps) first
into INFL, then into COMP. Each separate movement operation is local (in that it moves
know only into the head position in the next highest phrase containing it in the tree).
(20) (a) Thou seest how diligent I am (Petruchio, Taming of the Shrew, IV.iii)
(b) Thou say.vt true (Petruchio, Taming of the Shrew, IV.iii)
(c) The sight of love feeder/? those in love (Rosalind, As You Like It, III.v)
(d) She taker/? most delight in music, instruments and poetry (Baptista,
Taming of the Shrew, I.i)
(e) Winter tames man, woman and beast (Grumio, Taming of the Shrew, IV.i)
(f) It looks ill, it eats drily (Parolles, All’s Well That Ends Well, I.i)
226
6.5 Strong and weak features
Accordingly, we might argue that finite verbs have strong agreement features in EME
by virtue of the relatively rich system of agreement inflections they carry; and
conversely that finite verbs have weak agreement features in MSE by virtue of their
relatively impoverished agreement morphology in present-day English. (See
Rohrbacher 1994 and Vikner 1995 for interesting attempts to explore the correlation
between the strength of agreement-features and the relative richness of agreement
inflections in a range of different languages.)
The different strength of the agreement-features carried by finite verbs in EME on
the one hand and MSE on the other is reflected in a further syntactic difference
between them. Early Modem English was a null subject language, as we see from
sentences such as the following:
(21) (a) Hast any more of this? (Trinculo, The Tempest, Il.ii)
(b) Sufficeth, I am come to keep my word (Petruchio, Taming of the Shrew,
IILii)
(c) Would you would bear your fortunes like a man (Iago, Othello, IV.i)
(d) Lives, sir (Iago, Othello, IY.i, in reply to ‘How does Lieutenant Cassio?’)
Since the null subject in sentences like (21) occurs in a nominative position (by virtue
of being the subject of a finite clause), it has different case properties from the PRO
subject of infinitives (which has null case), and hence is generally taken to be a differ¬
ent kind of null subject conventionally designated as pro (affectionately known as little
pro, whereas its big brother is affectionately known as big PRO). By contrast, MSE is a
non-null subject language, as we see from the fact that the present-day counterparts
of (21) require (italicized) overt subjects: cf.
It would seem, therefore, that finite verbs can have a null pro subject in a language like
EME where they carry strong agreement features, but not in a language like MSE
where they carry weak agreement-features. Why should this be? An obvious sugges¬
tion is that in a language with a rich system of agreement inflections, the agreement
inflections on the verb serve to identify the null subject (e.g. the +st inflection on hast
in (21a) is a second person singular inflection, and hence allows us to identify the null
subject as a second person singular subject with the same properties as thou). But in a
weak agreement language like MSE, agreement morphology is too impoverished to
allow identification of a null pro subject (e.g. if we asked *Can help?, we’d have no
way of telling from the agreementless form can whether the missing subject is I, you,
he, they or whatever).
227
Head movement
What our discussion here suggests is that there is parametric variation across languages
in respect of whether finite verbs carry strong or weak agreement features, and that the rel¬
ative strength of these features determines whether nonauxiliary verbs can raise to INFL,
and whether null subjects are permitted or not. However, this still poses the question of
why finite verbs should raise out of V into I in languages like EME where they carry
strong agreement-features. One answer to this question is provided by checking
theory: let us suppose that movement is a last resort mechanism by which heads can
check features which would otherwise remain unchecked. As we have seen, finite verbs in
EME carry strong agreement-features; hence a finite verb raises to INFL in order to check
its strong agreement-features (i.e. its person/number specifier-features) against those of the
subject occupying the specifier position within IP. To see how this might work, consider
the syntax of a sentence such as Thou thinkest not of this (cf. (16e) above). The verb think-
est originates in the head V position of VP, and (because it contains strong agreement-
features) then raises to INFL as in (23) below (the [2SNom] features of thou mark the
second person singular nominative head-features of thou, the [Pres] feature of thinkest
marks its present-tense head-feature, and the [2SNom] features carried by thinkest are
specifier-features which mark the fact that it requires a second person singular nominative
subject as its specifier; all other features are omitted, to simplify exposition):
Since subject-verb agreement involves a local checking relation between INFL and its
specifier, an obvious consequence of moving thinkest from V into I in (23) is that it
enables the specifier-features of thinkest to be checked: this is because thinkest ends up
in INFL, and from there can check its [2SNom] specifier-features against the corre¬
sponding [2SNom] head-features of thou. Since the two sets of features match, the
specifier-features of thinkest are erased along with the nominative case-feature of thou
(because the relevant features are uninterpretable), so ensuring that the derivation does
not crash. Using a metaphor developed by Chomsky, we can say that movement from
V to INFL is motivated by considerations of greed - i.e. by the selfish desire of the
verb to check its own morphological features. Movement of thinkest to INFL also
ensures that INFL carries a tense-feature (viz. the present-tense head-feature of
thinkest), and so is interpretable at LF (i.e. at the level of logical form).
228
6.5 Strong and weak features
As we have seen, the agreement properties of finite nonauxiliary verbs in EME are
checked by moving the verb into INFL, so that the verb is in a local spec-head relation
with its subject, and its person/number/case specifier-features can be checked. But
recall (from our discussion in the previous chapter) that the specifier-features of finite
nonauxiliary verbs are checked in a rather different way in MSE - as we can illustrate
in relation to a simple sentence such as:
Example (24) is an IP headed by an empty INEL constituent, with the verb mistrusts
occupying the head V position of VP, and the subject she occupying the specifier posi¬
tion within IP. We suggested in the previous chapter that the head- and specifier-fea¬
tures of finite nonauxiliary verbs percolate from V to INFL in MSE, to satisfy the
requirement that INFL carry a tense-feature (in order to be interpretable at LF), and to
enable the specifier-features of the verb to be checked. In the case of (24), percolation
will work in the manner indicated in (25) below (to simplify exposition, we show only
the features directly relevant to our discussion here):
(25) IP
D
I
She
[3FSNom]
229
Head movement
way they are, an important question for us to ask is why strong features should be
checked by movement, and weak features by attraction. The answer to this question
may lie in the nature of the two different operations. Attraction involves movement of
a set of grammatical features carried by a head on their own (without movement of the
corresponding phonetic features) - more specifically, movement of those grammatical
features which could not be checked otherwise. For example, in (25) above, the
tense/agreement-features of mistrusts are attracted to INFL, but its phonetic features
(represented by the spelling form mistrusts) remain attached to the V-node. By con¬
trast, head movement involves movement of the phonetic and grammatical features of
a head together (our discussion here is simplified by leaving semantic features to one
side): thus, in (23) above not only do the tense/agreement-features of thinkest move to
INFL, but so too do its phonetic features (represented by the spelling form thinkest).
We might conjecture that what it means for a word like thinkest in (23) to carry strong
agreement-features is that these strong agreement-features cannot be separated from
the phonetic features carried by the relevant word: hence, the only way of checking
the strong agreement-features of thinkest is to move the whole word (i.e. the whole set
of phonetic and grammatical features carried by the word). Conversely, what it means
to say that a verb like mistrusts in (25) has weak grammatical features is that the
relevant grammatical features can move to INFL on their own, with the phonetic
features carried by the word being stranded (i.e. left behind) in the head V position of
YP.
We might further suppose that in consequence of the economy principle, only the
minimal set of features needed to satisfy some grammatical requirement undergo
movement in a given structure. Thus, because INFL requires a tense-feature and the
[3SNom] specifier-features of mistrusts cannot be checked unless they move to INFL,
the relevant tense/agreement-features percolate to INFL in (25); because these features
are weak, they can be separated from the phonetic features carried by mistrusts, and
hence the economy principle requires that only these tense/agreement-features should
move. By contrast, the verb thinkest in (23) carries strong [2SNom] agreement fea¬
tures, with the result that not just these features but also the phonetic features carried
by thinkest move to INFL (so that in effect the whole word thinkest moves to INFL).
Attraction is more economical than movement, since movement affects both the pho¬
netic and the grammatical features carried by a word, whereas attraction involves
movement of grammatical features alone: hence, the economy principle will ensure
that attraction will be preferred to movement wherever possible (e.g. in structures like
(25)), with movement only being forced where the relevant features being checked are
strong (e.g. in (23)).
230
6.6 Negation
6.6 Negation
An unexpected complication is raised by contrasts such as the following:
It seems reasonable to suppose (on the basis of examples such as (27) below) that
never can function as a VP adverb:
This assumption provides a straightforward account of the fact that never is positioned
in front of the verb admit in (27a), and of the fact that the bracketed string never admit
defeat behaves like a typical VP in being able to occur as a sentence fragment in (27b),
and as the complement of be in (27c).
If we assume that never is a VP-adverb, (26a) can be derived straightforwardly as in
(28) below, with the tense- and agreement-features of trusts being attracted to INFL (to
simplify exposition we show only features directly relevant to the discussion at hand):
a never V D
trusts him
231
Head movement
category NEG(ation), and functions as a syntactic head which projects into a NEGP
(negation phrase). Given this assumption, (26b) will have the structure (29) below (as
before, we show only the head-features of she and the head- and specifier-features of
trusts):
trusts him
[Pres
3SNom]
232
6.6 Negation
But this isn’t quite the end of the story, since the ungrammatical (26b) *She not
trusts him has the grammatical counterpart (30) below involving the use of do-support:
(31)
Since the NEGP not trust him is a projection of NEG, it follows that the NEGP string
not trust him is a negative infinitive phrase, and hence can check the [Inf] complement-
feature of does (which requires does to have an infinitival complement), leading to era¬
sure of both [Inf] features. We might suppose that not (by virtue of its traditional status
as a preverbal particle) can attract the head-features of the head verb of its complement,
(though by virtue of being an A-bar head, it cannot attract the specifier-features of a
finite verb). Since the verb trust here is nonfinite, it has no specifier-features to percolate
up to NEG. Instead, the choice of subject is dictated by the finite auxiliary does', the
[3SNom] specifier-features of does are checked for compatibility with the [3FSNom]
head-features of she and thereby erased (along with the nominative case-feature of she).
In addition, the objective complement-feature of trust is checked for compatibility with
the objective head-feature of him, and both objective case-features erased. So, after
checking, the only grammatical features which survive at LF are the [3FS] head-
features of she, the [Pres] head-feature of does and the [3MS] head-features of him. (Of
course, we are overlooking purely semantic features here - e.g. the negation feature
carried by not.) Since no uninterpretable features remain after checking, the derivation
satisfies the principle of full interpretation and converges at LF. Thus, the use of do-
support provides a way of rescuing a derivation which would otherwise crash if do were
not used. Our analysis thus provides a principled account of the observation made by
233
Head movement
Williams (1994) that not can modify a nonfinite verb, but not a finite verb. What still
remains to be accounted for, however, is why do can’t be used in sentences such as:
The answer is that (as we saw in relation to our earlier discussion of the examples in (6)
above) do is used only as a last resort - i.e. only in cases where some morphological
feature would remain unchecked if do were not used. There is simply no need to use do
with never, since sentences such as She never trusts him are grammatical, for reasons
given in our earlier discussion of (28). So, do is only used where it has to be, and is
avoided otherwise. The last resort condition on the use of do can be argued to be a con¬
sequence of the more general economy principle, which rules out the use of superflu¬
ous constituents: since dummy do has no intrinsic semantic content, it follows that its
use must be avoided wherever possible. At any rate, the last resort condition (together
with our earlier assumption that never is a VP-adverb but not is a NEG constituent
which heads a NEGP) enables us to account for the data in (26), (30) and (32) above.
Consider now how we deal with contracted negative structures such as:
We might suppose that (33) has essentially the same structure as (31) above, except
that the head NEG position in NEGP is occupied by n't rather than not. Since n’t is a
suffix (and cannot be used as an independent word standing on its own), n’t cliticizes to
the preceding auxiliary does, adjoining to does in the manner represented in (34)
below:
does n’t f V D
A , |
- trust him
What (34) suggests is that when n’t is attached to the auxiliary does, it forms a complex
head comprising the auxiliary does and the negative suffix n’t (thereby forming a nega¬
tive auxiliary). The resulting negative auxiliary behaves like any other auxiliary in
being able to undergo inversion, e.g. in questions such as:
234
6.7 The syntax of have
Such data suggest that n’t can be suffixed to some items, but not others. How can we
account for this? One way would be to invoke the traditional morphological distinction
between free and bound forms - i.e. forms which can and can’t stand on their own as
independent words. (In these terms, not would be a free form, and n’t a bound form.)
We could then say that may can only serve as a free form, and so does not allow n’t to
be suffixed to it; by contrast, might can be used either as a free form (cf. She might not
come), or as a bound form (cf. She mightn’t come). In order to account for an item like
will (which has the irregular contracted negative form won’t), we might suppose that its
lexical entry indicates that it has the free form will and the bound form wo+ (the +
symbol here indicating that a suffix like +n’t needs to be added to it - and the + symbol
on +n’t indicating that it needs to be suffixed to a bound finite verb form).
To summarize: a sentence such as She trusts him cannot be negated by not alone (cf.
*She not trusts him), because the presence of not prevents the agreement features of
trusts percolating from V to I, so leaving the agreement features of the verb and the
nominative case-feature of its subject unchecked. Instead, such sentences have to be
negated by the use of Jo-support (cf. She does not trust him), which provides a way of
ensuring that the nominative case-feature carried by she can be checked. Thus, do is
used because it is needed in order to check the case properties of the subject. Since
does can function either as a free form or as a bound form to which n’t can be suffixed,
an alternative possibility is for n’t to be used in place of not - in which case, n’t is
adjoined to does to form the negative auxiliary doesn’t (which can undergo inversion
like any other auxiliary).
- a type of structure which is still current in some (e.g. British) varieties of English, but
which has fallen out of use in other (e.g. American) varieties. In (37), have behaves
like an auxiliary in respect of undergoing inversion. And yet have in this use otherwise
seems to have the properties of a typical transitive verb - e.g. it has much the same
235
Head movement
meaning as other transitive verbs like own or possess, and (like them) takes a DP com¬
plement (any wool). It seems plausible, then, to suppose that have (in this possessive
use) is a transitive verb which originates in the head V position of VP, and which
(being transitive) selects a DP complement. Such an assumption will, in any case, be
required for structures such as:
where it is clear that have occupies the head V position within the bracketed VP.
But if transitive have originates in the head V position of VP, how does it end up in
the head C position of CP in sentences like (37) Have you any wool? Given that head
movement is a strictly local operation, the answer must be that have in (37) moves first
from V to I, then from I to C in a successive cyclic fashion, as represented by the
arrows in (39) below:
t D N
I I
any wool
Movement of the transitive verb have from V to INFL (and from there into COMP) in
Baa Baa Black Sheep varieties of English would then be a last vestige of a once pro¬
ductive V-to-I movement process (whereby all finite verbs moved from V to INFL in
EME). In many (e.g. American) varieties of English this type of movement of have
from V to I is no longer possible, and (38) is grammatical in such varieties, but (37) is
not (I leave the sociologists among you to ponder on whether this might be because
politically correct parents no longer sing Baa Baa Black Sheep to their children in the
relevant varieties, because of its racist and animalist overtones).
Independent evidence for claiming that possessive have can move out of V into I
comes from the fact that have can cliticize onto its subject in structures such as:
If (as we claim) have moves from V to I in such structures, (40) will have the deriva¬
tion (41) below:
236
6.7 The syntax of have
I / \
have V DP
* | / \
1-t no wool
and we correctly predict that have (in the guise of its clitic variant ’ve) can cliticize
onto they, since the two are immediately adjacent. By contrast, if have always remains
in the head V position of VP, we wrongly predict that cliticization of have onto they
will be blocked by the intervening INFL constituent, since have and they will never be
immediately adjacent. (Recall that in §4.4, we noted that cliticization requires adja¬
cency, and is blocked by an intervening empty category.)
However, it should be noted that movement of possessive have from V to I is
optional in Baa Baa Black Sheep varieties, as indicated by the syntax of the floating
quantifier both in sentences such as the following:
If we assume (as in §3.6) that floating quantifiers are in spec-VP, we can account for
the data in (42) by supposing that in (42a) the verb have remains in V (and so follows
both which is in spec-VP), whereas in (42b) the verb have moves from V into I
(thereby moving across both), and (once in I) can then cliticize onto the subject they in
spec-IP. (For obvious reasons, sentences like (42b) are only grammatical in Baa Baa
Black Sheep varieties of English - not in varieties in which transitive have always
remains in V.)
Now consider a second (at first sight, rather puzzling) aspect of the syntax of posses¬
sive have. As the examples below illustrate, possessive have can be directly negated by
the contracted negative n’t, but not by the full negative not: cf.
How can we account for the puzzling contrast in (43)? The answer to this question
turns out to provide us with a potential clue to the nature of head movement.
Consider first how we account for (43a) We haven’t any wool. If we assume (as
before) that n’t originates as the head NEG of NEGP, (43a) will have the structure (44)
below, if have remains in situ in the head V position in VP:
237
Head movement
Let’s assume that have adjoins to n’t, since n’t is a suffix, have will adjoin to the left of
n’t, as in (45) below:
NEG VP
/ \ / \
V NEG V DP
II I / \
have n’t t any wool
A
l )
This process of adjoining one head to another leads to the formation of a complex head
which comprises both have and n’t, and which combines properties of both heads (e.g.
have is a finite verb, n't is negative, so haven’t is a negative finite verb). The next step
in the derivation is for the resulting complex head haven’t to move into INFL (so that
the agreement features of have can be checked against those of we). Once haven’t is in
INFL, it can move from there to COMP, e.g. in questions such as:
Thus, the grammaticality of sentences like (43a) can be accounted for in a principled
fashion.
But now consider how we account for the ungrammaticality of (43b), *VTe have not
any wool (with the independent negative not rather than the suffixal negative n’t). We
might expect this to be derived in the same way as (43a), with have first adjoining to
not as in (45), and the resulting compound head have + not then moving to INFL. But
since the resulting sentence (43b) is ungrammatical, something must prevent such a
derivation. What could it be? The obvious answer is that the first step in the derivation
(adjunction of have to not) is blocked because not is a free form (not a bound form),
and so doesn’t allow another word to adjoin to it. So, while adjoining have to NEG is
possible where NEG contains the bound form n’t (as in (45) above), it is not possible
where NEG contains the free form not. For this reason, it is not possible to derive (43b)
238
6.7 The syntax of have
*We have not any wool by first adjoining have to not, and then adjoining have + not to
INFL.
But isn’t there a more direct way of generating (43b) - namely by moving have
directly into INFL, as in (47) below?
not V DP
I / \
-have any wool
If the derivation in (47) were possible, we would wrongly expect (43b) *We have not
any wool to be grammatical. The fact that (43b) is ungrammatical means that some
principle of Universal Grammar must rule out the possibility of have skipping across
NEG in (47). But what principle?
The answer is provided by the familiar head movement constraint (= HMC) pro¬
posed by Lisa Travis (1984) to the effect that a head can only move to the head position
within the immediately containing (i.e. ‘next highest’) phrase in the tree: since the next
highest phrase above the VP containing have in (47) is NEGP (not IP), this means that
have can only adjoin to NEG, not to INFL. Such a constraint would ensure that head
movement can only apply in a strictly local fashion - in precisely the same way as we
earlier saw that attraction (i.e. movement of grammatical features) is local.
Given that HMC determines that head movement can only apply in a strictly local
fashion, direct movement of have to INFL in (47) is ruled out. Since successive cyclic
movement (adjoining have first to not then to INFL) is also ruled out (because not is a
free form), both derivations crash. The only way of negating the sentence with not is to
use do-support as a last resort, yielding the sentence:
And if the bound form +n’t is used together with the bound form do+ (which rhymes
with doe/dough), +n’t will adjoin to do+, giving:
Thus, we can provide a principled account of why in Baa Baa Black Sheep varieties of
English, we find (48), (49) and (43a) - but not (43b).
Our discussion of negation here suggests that not and n’t complement each other: not
is a free form which can be used to negate a following nonfinite verb (in structures
such as (31) above) but not a finite verb (as we see from the ungrammaticality of struc-
239
Head movement
tures such as (29) above). By contrast, n’t is a bound form which must be suffixed to a
finite verb, and which cannot be suffixed to a nonfinite verb (perhaps because nonfinite
verbs are free forms), as we see from the ungrammaticality of (50b) below (where have
is an infinitive form):
Thus, we might characterize not informally as a free nonfinite form, and n’t as a bound
finite form.
(51)
D"
I
They
240
6.8 Tense affix
Movement of have to I would then satisfy the requirement for the bound form have to
be attached to an affix (though, of course, the present-tense affix it is attaching to here
is null, and so has no overt phonetic form). One consequence of raising have from V to
INFL is that the present-tense head-feature carried by have can thereby be checked
against that of the present-tense affix T. (Let’s assume that as a result of checking, the
tense-feature of have is erased, but that of the T affix in INFL is not erased, since INFL
must carry tense in order to be interpretable at LF.) Once have adjoins to INFL, it can
then undergo further movement to COMP (e.g. in sentences like (37) Have you any
wool?), or can cliticize onto they (as in (40) They’ve no wool).
However, there is an interesting hidden complication which arises out of the analysis
in (51) above. It has standardly been assumed (in work over the past three decades) that
adjoining one head X to another head Y produces a structure of the form (52) below:
(52) Y
X xY
(53)
/ \
V T
! I
have Pres
The resulting string have+Pres thus has the status of a T constituent - not (as we had
hitherto assumed) of an INFL constituent.
But this assumption in turn has even further ramifications. Since the head of the
overall clause They have no wool is the T constituent have+Pres in (53), this means
that the string have no wool must have the status of a T-bar constituent (not an I-bar),
and likewise that the overall clause They have no wool must have the status of a TP
constituent (not an IP). Thus, the overall sentence They have no wool will not be an IP
with the structure (51) above, but rather a TP with the structure (54) below:
241
Head movement
More generally, the assumption that finite clauses are headed by an abstract tense affix
leads us to the conclusion that they have the status of TP (rather than IP) constituents.
So far, we have considered how to derive Baa Baa Black Sheep sentences in which
possessive have moves out of the head V position in VP, and adjoins to an abstract
tense affix (perhaps, as we suggested, because have is a bound form in this kind of
use). But now consider how we account for the fact that all other lexical verbs in pre¬
sent-day English remain in the head V position of VP. If we suppose that clauses are
headed by an abstract tense affix T, it follows that T must have some element affixed to
it. Exploiting Chomsky’s strength metaphor, let’s suppose that the relevant tense affix
is a weak affix in present-day English, and so is only strong enough to attract grammat¬
ical features, not phonetic features (i.e. not strong enough to trigger movement of the
whole item). What this means (e.g. in the case of our earlier sentence (24) She mis¬
trusts him) is that the head- and specifier-features of the verb (i.e. the tense/agreement-
features of mistrusts) will adjoin to the tense affix T, as in (55) below, but its phonetic
features (represented by the spelling form mistrusts) remain in V:
(55)
/ \
[Pres T
3SNom] I
* Pres mistrusts him
242
6.8 Tense affix
there to COMP in questions), thereby forming questions and negatives without Jo-
support (as in (57a-b) below): cf.
In its use as a lexical verb requiring Jo-support, possessive have behaves like any other
typical lexical verb (e.g. mistrusts in (55) above) in remaining in situ in the head V
position of VP, and so is unproblematic. More puzzling is how we account for the fact
that possessive have can adjoin to T as in (54) above. Why should this be? After all, we
have said that T is weak in present-day English, and hence is only strong enough to
trigger attraction of grammatical features, not strong enough to trigger movement of
the whole set of phonetic and grammatical features carried by a head.
One answer which we could explore is along the following lines. Let's assume that
possessive have in Baa Baa Black Sheep varieties has the idiosyncratic morphological
property that when its tense/agreement-features adjoin to T, they can drag along with
them all the other (e.g. phonetic and semantic) features carried by have. In terms of the
metaphor used by Chomsky (1995c), we might say that the phonetic features of have
are pied-piped along with its tense/agreement-features. (If you were brought up on a
diet of nursery stories rather than virtual-reality videos, you may remember that the
Pied Piper in the village of Hamelin induced the rats to follow him out of the village by
playing his pipe.) We can then say that even though the weak tense affix T is only
strong enough to attract the tense/agreement-features of verbs in present-day English,
have has the peculiar property (in the relevant use) that its grammatical features are
inseparable from the other features it carries, so if the tense/agreement-features of have
are adjoined to T for checking purposes, its phonetic features are adjoined as well
(hence, in effect, the whole word have adjoins to T). In terms of our discussion in §6.5,
this amounts to claiming that have (in the relevant use) carries strong tense/agreement-
features. (The relevant tense/agreement-features are strong in the sense that they are
strongly bound to and hence inseparable from the phonetic features carried by have.)
Of course, since have can also behave like a typical lexical verb (as in (56a-b) above),
we have to suppose that the tense/agreement-features carried by have can either be
weak (as for other lexical verbs in English), or (exceptionally) strong. Since have is the
only verb which allows pied-piping in this way, it is clearly anomalous in the relevant
respect: hence it comes as no surprise that this use of have is no longer found in most
varieties of English, and is dying out even in those varieties of British English in which
it survived longest. (In Radford 1992,1 reported that preschool British children invari¬
ably use Jo-support with possessive have, hence treating it like a typical lexical verb
which carries weak tense/agreement-features.)
243
Head movement
Let’s suppose that (58a) is formed by merging the auxiliary has with a present-tense
affix Pres to form a T constituent has+Pres which in turn merges with the VP seen it to
form the T-bar has+Pres seen it, and that this T-bar in turn merges with she to form the
TP in (59a) below. Similarly, let’s suppose that (58b) is derived by merging the auxil¬
iary could with a past-tense suffix Past to form the T constituent could+Past which in
turn merges with the VP see it to form the T-bar could+Past see it, and that this T-bar
then merges with she to form the TP (59b) below:
244
6.9 Question affix
She T VP She T ^ VP
/ \ / x / \ / \
Perf T V D Mod T V D
11 11 11 11 I i 1
1 1 1 1
has Pres seen it could Past see it
The requirement for the weak tense affixes Pres/Past to have a set of tense/agreement
features adjoined to them is then met by virtue of the fact that has (which is a third per¬
son singular present-tense perfective auxiliary) merges with Pres in (59a), and could
(which is a past-tense modal auxiliary) merges with Past in (59b).
Although we shall not enter into relevant details here, it may be that we can extend
the analysis in (59) from finite clauses headed by auxiliaries to infinitival clauses
headed by to (especially in view of Stowell’s 1982 claim that to carries tense proper¬
ties). If so, an infinitival to-clause would have essentially the same TP structure as
(59b) above, save that in place of could we find to.
Now consider how the requirements of the strong tense affix heading finite clauses
in Early Modem English can be satisfied (recall that we are assuming that a strong affix
must have an overt item attached to it). One possibility is for an auxiliary to merge with
the tense affix (much as in (59a-b) above). A second possibility is for a nonauxiliary
verb to raise from V to adjoin to T, e.g. as in (60) below:
Once the verb came adjoins to T in (60), it can move from there to COMP, so deriving
the structure associated with the question (61) below:
(61) Came you from the church? (Tranio, Taming of the Shrew, Ill.ii)
245
Head movement
structures like (54) involves adjunction of (the set of phonetic and grammatical fea¬
tures carried by) a verb to a null tense affix; and that attraction in structures like (55)
involves adjunction of a set of grammatical features to a tense affix. The wider general¬
ization which this would seem to suggest is that movement and attraction both involve
adjunction of a set of features to an affix.
However, if we conclude that all head movement is adjunction, we clearly have
to revise our earlier analysis of inversion in questions. Rather than moving into
COMP, it must be the case that an inverted auxiliary adjoins to an affix in COMP. If we
assume (as earlier) that affixes have specific semantic and categorial properties (e.g. T
has the semantic property that it denotes tense, and the categorial property that it is a
verbal affix), we might suppose that the affix in COMP in questions has the semantic
property that it marks the illocutionary force of the sentence (i.e. marks it as interroga¬
tive rather than declarative or imperative), and has the categorial property that it is a
verbal affix. Given these assumptions, COMP will be filled by a question affix in inter¬
rogative sentences, and inversion will involve movement of a verb from T to adjoin to
the question affix in COMP. The assumption that inversion in questions involves
adjunction to a question affix gains cross-linguistic plausibility from Latin examples
such as:
where it would seem that the verb necavit ‘killed’ has been adjoined to the overt ques¬
tion suffix +ne in COMP. (Verbs are normally positioned at the end of clauses in Latin.)
An obvious way of providing a unitary account of the two types of head-movement
operation found in English and Latin questions is to assume that both types of structure
involve adjunction of the preposed (auxiliary or nonauxiliary) verb to an (overt or
covert) affixal question-particle in COMP. For concreteness, let’s use the symbol Q to
denote the null question affix found in English questions (following work in an earlier
framework by Katz and Postal 1964 and Baker 1970). You can think of Q as the syn¬
tactic counterpart of the question-mark ? used in the spelling system to indicate that a
sentence is a question.
In the light of this assumption, consider now how we derive a question such as:
The modal auxiliary can merges with a present-tense affix Pres to form the T con¬
stituent can+Pres\ this in turn merges with the VP help me to form the T-bar can+Pres
help me\ the resulting T-bar merges with you to form the TP You can+Pres help me.
This TP is then merged with a C constituent containing the strong question affix Q, so
forming the CP (64) below:
246
6.9 Question affix
(64)
VP
help me
However, Q is a strong affix, and so needs an overt head (i.e. a non-null head with
overt phonetic features) attached to it. The economy principle requires us to move the
minimal constituent which will satisfy the requirement for Q to be attached to an overt
head (and any other relevant grammatical requirements). The minimal overt head
which we can adjoin to Q is the modal can\ hence, we might suppose that the modal
can adjoins to Q as in (65) below:
(65)
VP
help me
In (65), the auxiliary can is said to excorporate out of the T constituent containing it,
thereby detaching itself from the tense affix Pres (see Roberts 1991 for a technical
discussion of excorporation).
However, an apparent problem posed by the excorporation analysis is that it
presupposes that the tense affix Pres can be stranded by movement of the modal can
on its own (without Pres). This might be argued to be potentially problematic because
affixes can’t generally be stranded (cf. Lasnik 1981), as we see from the fact that the
negative affix n’t attached to couldn’t in (66a) below can’t be stranded by inversion of
could:
When could undergoes inversion, the negative affix n’t must be inverted along with
could (as in (66b)), and can’t be left behind and thereby stranded - hence the ungram¬
matically of (66c). So since Pres is a tense affix, and affixes can’t be stranded, the
derivation in (65) will crash.
247
Head movement
How can we ensure that the tense affix Pres isn’t stranded in (65)? One answer is to
posit that it is pied-piped along with can, so that the whole T constituent can+Pres
adjoins to the question affix Q, as in (67) below:
help me
Since Pres remains adjoined to can, there is no violation of the constraint against affix-
stranding in (67). Moreover, since (the T constituent containing) can is adjoined to (the
C constituent containing) Q, the requirement for the strong affix Q to be attached to a
set of phonetic features (i.e. to an overt item like can) is satisfied in (67).
So, the assumption that affixes can’t be stranded would seem to favour the pied-
piping analysis of inversion in (67) (whereby the tense affix Pres is pied-piped along
with can, so that the whole T constituent can+Pres is moved) over the excorporation
analysis in (65) (whereby only the modal auxiliary can is moved): accordingly, the T
movement analysis in (67) has become the standard analysis of inversion (in one form
or another) in the linguistic literature over the past two decades.
However, the conclusion that the pied-piping analysis is superior to the excorpora¬
tion analysis is only as strong as the assumption on which it is based - namely that
affixes can never be stranded. But is this necessarily so? After all, we earlier drew a
distinction between strong and weak affixes, to the effect that a strong affix must be
bound to a set of phonetic features (i.e. affixed to an overt item), whereas a weak affix
need not be. (A weak affix serves simply to check an appropriate set of grammatical
features.) In these terms, (66c) *Could he n’t find it? is ungrammatical because n’t is a
strong affix which needs to be attached to an overt finite form (like could). However,
since we earlier argued that the T constituent heading finite clauses in Modem
Standard English contains a weak tense affix, it follows that there is little reason to
assume that the weak affix Pres cannot be separated from (phonetic features of) the
modal can in (65). After all, if can raises to T in order to check its present-tense head-
feature and its nominative specifier-feature, there would seem to be no obvious reason
why can should not subsequently raise on its own to adjoin to Q once the relevant
head- and specifier-features carried by can have been checked (and erased). Indeed, if
we posit that the tense affix Pres remains in T (as in (65) above) after adjunction to Q
takes place, this will facilitate LF convergence (if we assume that T is the locus of
tense at LF). Thus, the choice between the pied-piping analysis in (67) and the excor¬
poration analysis in (65) is far from straightforward (and depends heavily on theory-
248
6.9 Question affix
internal assumptions): as already noted, the pied-piping analysis is the one standardly
adopted in the relevant literature (though it is usually assumed rather than argued for).
Our assumption that interrogative clauses contain a Q affix raises obvious questions
about the syntax of the interrogative complementizer if m a complement clause such as
that bracketed in (68) below:
If all interrogative clauses contain a Q affix, it seems plausible to suppose that the
complementizer if is merged with the affix Q, so that the bracketed clause in (68) has
the structure (69) below (where the symbol Comp is used to categorize the comple¬
mentizer if):
(69)
VP
can Pres
This implies that the requirement for the strong interrogative affix Q to be attached to
an overt item can be satisfied either by merger (i.e. by merging Q with a complemen¬
tizer like if as in (69) above) or by movement (by adjoining an auxiliary/T-constituent
to Q as in (65/67) above). One minor problem posed by the analysis in (69) is that it
would seem to call in question our earlier claim that Q is a purely verbal affix, since the
complementizer //doesn’t seem to be a verbal head.
Since we have seen that there are movement operations in which the grammatical
and phonetic features carried by a head are adjoined to another head, and also attrac¬
tion operations which affect grammatical (but not phonetic) features, an obvious
question to ask is whether there is a third type of movement operation by which pho¬
netic features alone are adjoined to another head (leaving behind the corresponding
grammatical features). This seems a plausible assumption in relation to the kind of
cliticization phenomena illustrated below:
At first sight, it might seem plausible to suppose that cliticization here involves a
movement operation adjoining (the grammatical and phonetic features of) one word to
the other (as in the case of n’r-suffixation). But such a movement analysis would
249
Head movement
wrongly predict that a compound head is thereby formed which can itself undergo
movement as a single syntactic unit. However, this assumption is false - as we see
from examples such as (71) below:
6.10 Summary
In this chapter, we have been concerned with the syntax of head move¬
ment. In §6.2 we looked at auxiliary inversion in questions in English, arguing that this
involves an I movement operation whereby an auxiliary moves from INFL to COMP.
We suggested that an interrogative COMP is strong (and so has to be filled), and that
moving an auxiliary from INFL into COMP in questions serves to fill COMP. In §6.3,
we argued that an inverted auxiliary leaves behind a trace (i.e. a silent copy of itself) in
the INFL position out of which it moves when it moves to COMP. In §6.4 we saw that
finite verbs in Early Modem English (EME) could move from V to INFL by an opera¬
tion of V movement (as is shown by word-order facts in negative sentences like I care
not for her), but that this kind of movement is generally no longer possible in Modem
Standard English (MSE). We argued that I movement and V movement are two
250
6.10 Summary
different reflexes of a more general head movement operation, and that head move¬
ment is subject to a strict locality constraint (the head movement constraint) which
requires it to apply in a successive cyclic fashion. In §6.5 we suggested that verbs in
EME had strong agreement-features (by virtue of the relatively rich agreement inflec¬
tions they carried) and consequently allowed a null pro subject, whereas their counter¬
parts in MSE have weak agreement-features (by virtue of their impoverished agree¬
ment morphology) and so do not license a pro subject. We noted that the strong
agreement features of finite verbs in EME were checked by movement of the verb
(along with its features) from V to INFL, whereas the weak agreement features of finite
verbs in MSE are checked by attraction (i.e. percolation) of the relevant agreement
features from V to INFL (with the verb itself remaining in situ in the head V position of
VP). We went on to argue that movement and attraction are two different reflexes of a
common feature movement operation which moves grammatical features from one
constituent to another, and which in the case of movement also has the effect of pied-
piping the phonetic features of the head along with its grammatical features. In §6.6 we
looked at the syntax of negation in MSE, arguing that the negative particle not func¬
tions as a syntactic head NEG which projects into a NEGP constituent, and suggested
that not can attract the head-features of (the head verb of) its complement, but not the
specifier-features of a finite verb; we also suggested that n’t differs from not in that not
is a free form whereas n’t is a bound form which can only attach to other bound forms.
In §6.7 we looked at the syntax of possessive have in those (Baa Baa Black Sheep)
varieties of English in which it can behave like an auxiliary. We argued that possessive
have originates in the head V position of VP and from there can raise to INFL. We
noted that possessive have can be directly negated by n’t, but not by not. We suggested
that this is because have can serve as a bound form, and hence can adjoin to the bound
negative n’t (but not to the free form not). More generally, we suggested that adjunc¬
tion can only adjoin one bound form to another, and cannot target a free form like not.
In §6.8 we went on to speculate that all head movement may involve adjunction to an
affix, so that V-to-I movement involves adjunction of V to an abstract tense affix T: we
noted that one consequence of this is that clauses are projections of a tense affix T, and
hence have the status of TP (= tense phrase) constituents. We argued that T is a weak
affix in MSE, but was a strong affix in EME; and we suggested that weak affixes have
the property that they check grammatical features, whereas strong affixes must be
attached to a set of phonetic features. We noted that the requirements of the weak T in
MSE are generally satisfied either by merging an auxiliary with T, or by attraction of
the tense/agreement-features of a nonauxiliary verb to T. By contrast, the requirements
of the strong T in EME are satisfied either by merger with an auxiliary, or by movement
of a nonauxiliary verb to adjoin to T. In §6.9 we suggested that questions are CP
constituents headed by a C node containing a strong Q affix; we noted that the require¬
ment for Q to be attached to a set of phonetic features is satisfied in main clauses by
251
Head movement
Workbook section
Exercise 1 (§§6.2-6.3)
Discuss the derivation of the following sentences, saying why each
derivation crashes or converges.
(i) VP
us
Since us is an objective pronoun, it satisfies the requirement for the verb help (when
used transitively, as here) to have an objective complement.
The VP in (i) in turn merges with the auxiliary (I constituent) can to form the I-bar
(ii) below:
(ii)
can V D
I
help us
252
Workbook section
Since the VP help us is headed by the infinitive verb help, it satisfies the requirement
for can to have an infinitive complement.
The pronoun he in turn merges with the I-bar in (ii), forming the IP in (iii) below:
(iii) IP
D'
He I VP
I
can V D
I
help us
The nominative pronoun he satisfies the requirement for can to have a nominative
subject.
The IP in (iii) is then merged with a C constituent into which the auxiliary can
moves (leaving behind a trace in the position out of which it moves), so forming the
CP(iv) below:
(where t is a trace of the auxiliary can which moves from INFL to COMP).
Exercise II (§§6.4-6.5)
Discuss the syntax of the following Early Modem English (EME) sen¬
tences taken from various plays by Shakespeare:
253
Head movement
(i) VP
V
I
marvell’st DP
my words
The VP in (i) then merges with an abstract INFL constituent to form an I-bar con¬
stituent which in turn merges with the pronoun thou to form an IP. The verb marvel! ’st
raises from V to INFL as in (ii) below:
-1 P DP
at D' N
I I
my words
Why should finite verbs like marvell ’st raise from V to INFL in EME? The answer sug¬
gested in §6.5 is that finite verbs in EME carry strong tense/agreement-features - i.e.
features which are strongly bound to (and hence inseparable from) the corresponding
phonetic features carried by the word. The [Pres] present-tense head-feature of marvel¬
l’st needs to raise to INFL in order to ensure that INFL is interpretable at LF (by virtue
of carrying a tense-feature); the [2SNom] specifier-features of marvell’st need to raise
to INFL in order to check whether the requirement for marvell’st to have a second
person singular nominative specifier is satisfied by thou. (Checking erases the speci¬
fier-features of marvell’st and the nominative-case head-feature of thou.) Since the
tense/agreement-features of marvell’st are strong, they cannot raise to INFL on their
own by attraction; instead, they carry the phonetic features of marvell’st along with
them, so that the whole word marveU’st undergoes head-movement and thereby
moves into INFL - as in (ii) above.
Given the assumptions made in §§6.4—6.5, sentence 2 will be derived as follows.
The verb come merges with the auxiliary/INFL constituent doth to form an I-bar which
254
Workbook section
in turn merges with the subject Macbeth to form the structure (iii) below (doth is a
regional variant of does):
doth come
(We are assuming here, on the basis of the arguments presented in §4.7, that Macbeth
is a DP headed by a null determiner; this is not implausible, given that we find overt
determiners used with proper names in many languages - cf. Italian la Callas, literally
‘the Callas’.)
A potential problem for the analysis in the text posed by sentences like 2 is that they
might seem to call into question the assumption that dummy verbs like do are used
purely as a last resort. We might suppose that if do-support were purely a last resort,
sentences like 2 would be ungrammatical, since they have the do-less counterpart
Macbeth cometh (which involves raising cometh from V to INFL). However, one way
of maintaining the last resort account would be to suppose that do in EME was not a
dummy auxiliary at all, but rather had independent semantic content of some kind. One
possibility is that do in EME served the function of marking aspect. After all, there are
varieties of present-day English in which do is an aspectual auxiliary: for example, in
Caribbean creoles (according to Rickford 1986 and Harris 1986) we find the auxiliary
do being used in sentences like:
to mark habitual aspect (so that (iv) has much the same meaning as He is usually
sick). Likewise, do functions as a habitual aspect marker in Irish English (cf. Guilfoyle
1983, Harris 1986) and in south-western varieties of British English (cf. Wakelin 1977,
pp. 120-1). However, since sentence 2 Macbeth doth come doesn’t have the habitual
sense of Macbeth usually comes, it’s implausible that doth functions as an aspectual
auxiliary here. An alternative possibility is that do in EME had a performative
sense, and thus meant something like perform the action of. . . This performative
use of do in EME may be connected to its use in present-day British English in
sentences such as:
where do might be glossed as perform the relevant action (i.e. of reading books). If do
had independent semantic content in EME, it would clearly not be subject to the last
resort condition (since the latter applies only to dummy or expletive items with no
semantic content of any kind).
255
Head movement
Since the second line must contain four syllables and end in a word rhyming with
drum, and since the three-syllable utterance Macbeth cometh satisfies neither require¬
ment, the four-syllable utterance Macbeth doth come (which satisfies both) is used
instead.
Overall, the precise function and content of do in EME is anything but clear. As
noted by Tieken-Boon van Ostade (1988, p. 1) the origin of do ‘may truly be called one
of the great riddles of English linguistic history’.
1 He smokes
2 *Smokes he?
3 Does he smoke?
4 *He not smokes
5 *He smokesn’t
6 He does not smoke
7 Does he not smoke?
8 *Does not he smoke?
9 He doesn’t smoke
10 Doesn’t he smoke?
Given the assumptions made in the text, 1 will be derived as in (i) below
(3 = third person; S = singular; M = masculine; Nom = nominative; Pres = present
tense):
He T V
[3SMNom] / \ |
[Pres T smokes
3SNom] | [_]
t Pres
256
Workbook section
The head of the clause is a weak tense affix T. Since weak affixes trigger attraction but
not movement, the (bold-printed) head- and specifier-features of smokes (its head-
features indicating that it is present tense, its specifier-features that it requires a third
person singular nominative specifier) are adjoined to present-tense affix T, as indicated
by the arrow in (i) - though the verb smokes remains in V (more specifically, its pho¬
netic features remain in V). The tense properties of smokes can then be checked against
those of T, and its agreement properties (requiring it to have a third person singular
nominative specifier) checked against those of he. Checking erases all grammatical
features except the [3SM] head-features of he and the [Pres] present-tense head-feature
ofT.
One question which arises in relation to (i) is why it is not possible to check the
agreement features of smokes by movement - i.e. by adjoining (the grammatical and
phonetic features of) smokes to the tense affix T. The answer is provided by the econ¬
omy principle: attraction is more economical than movement (since attraction
involves movement of grammatical features alone, whereas movement involves pied-
piping of phonetic features along with grammatical features), so attraction applies
unless movement is forced by morphological considerations. Since T is a weak affix
(which therefore only serves to check grammatical features), there is no morphological
requirement for movement here, hence (by the economy principle) no possibility of
movement.
A second question which arises in relation to (i) is why (io-support is not required
here. The answer is that the derivation in (i) converges without the use of rfo-support,
so that the economy principle (in the guise of the last resort condition) determines
that do cannot be used because it is not required. In this respect, it is interesting to note
the following contrast (where does denotes a form carrying contrastive stress, and d’s
denotes the unstressed form /doz/):
The last resort account correctly predicts that the unstressed form d’s cannot be used
here. But how are we to account for the fact that the stressed form does can be used in
(ii)(a)? One possible answer is to suppose that stressed forms of do have an emphatic
or contrastive function, and so have intrinsic semantic content - and that (by virtue of
their semantic content) they are not subject to the last resort condition on the use of
contentless dummy items.
Helpful hints
Consider two alternative derivations for 2 - one in which smokes raises
from V to T to C in successive cyclic fashion, and the other in which smokes raises
directly from V to C: say why both derivations crash.
257
Head movement
Exercise IV (§§6.6-6.9)
The following data (kindly provided by Larry Lamb) illustrate three dif¬
ferent uses of have in a particular Baa Baa Black Sheep variety of English:
The use of have illustrated in 1-14 might be called its possessive use (since have in
this use can roughly be paraphrased by ‘possess’ or ‘own’), that in 15-24 its activity
use (since have in this use can be paraphrased by the activity verb ‘take’), and that
in 25-32 its perfective use. (Note that the grammaticality judgments in 15-24 are
258
Workbook section
relative to the activity use of have where it is paraphraseable by ‘take’, not to the
different possessive use of have where it is paraphraseable by ‘own’.) Discuss the
derivation of each of these sentences, and say why each derivation converges or
crashes.
The phonetic features of has remain in the head V position of VP There is no movement
of has to T, since T is a weak affix (i.e. the kind of affix which requires grammatical
rather than phonetic features to be attached to it), and hence the economy principle
dictates that attraction is preferred to movement. Since T is empty here, we expect the
relevant structure to be tagged by do, as in:
However, it is a peculiarity of Baa Baa Black Sheep varieties that possessive have
also has a second use, on which it appears to behave like an auxiliary (in not requiring
Jo-support in questions and negatives). This cannot be because T can (e.g. optionally)
be a strong affix in the relevant varieties, since otherwise all other lexical verbs
would behave like have. Rather, it must be an idiosyncrasy of possessive have. The
nature of this idiosyncrasy would appear to be that possessive have has the property
that its phonetic features can optionally be pied-piped along with its tense/agreement-
features. Movement of the tense/agreement-features of has alone will give rise to the
derivation (i) above; pied-piping of other features along with the tense/agreement-
features will cause the whole (set of features carried by the) item has to adjoin to T as
in (iii) below:
259
Head movement
(iii)
D
He
Since T is occupied by has in (iii), we correctly predict that the resulting structure will
be tagged by has in sentences such as:
We also correctly predict that (because the two are immediately adjacent) has can cliti-
cize to he in (iii), giving:
Helpful hints
Consider the possibility that activity have is a transitive verb which is
directly generated in V and cannot adjoin to T (why?), and that perfective have is
merged with T.
Exercise V (§§6.6-6.9)
Discuss the derivation of the following sentences, saying what determines whether
they are grammatical or not.
1 He cannot do anything
2 *Cannot he do anything?
3 Can he not do anything?
4 He has not returned
5 He’s not returned
6 He hasn’t returned
7 *He’sn’t returned
8 He will not survive
9 *He willn’t survive
10 He won’t survive
11 Will he not survive?
260
Workbook section
0)
D
(ii)
NEGP
VP
Since can remains a separate word from not in the syntax, we correctly predict that, in
questions, only can will be inverted (i.e. move from T to C), not cannot (since the latter
remains two separate words in the syntax): this accounts for why the negative counter-
261
Head movement
part of 1 is 3, not 2. However, because not can serve as a PF clitic and can is able to
serve as host for a PF clitic, not cliticizes to can in the PF component (with the
phonetic features of not adjoining to those of can), so forming cannot (which therefore
represents one word at PF, and two words in the syntax and at LF).
Exercise VI (§§6.6-6.9)
The sentences below illustrate the use of be as a copular verb linking a sub¬
ject like he to a nonverbal (i.e. adjectival, nominal, or - as in this case - prepositional)
predicate like in the army:
Helpful hints
A traditional analysis of copular be (dating back in spirit to Klima 1964)
is that it originates in the head V of VP, and (in finite indicative forms) raises from V to
T (but in nonfinite or subjunctive forms, remains in V). An alternative minimalist
analysis would be to suppose that copular be is a dummy or expletive item with no
semantic content, which serves purely to encode morphological features: for example,
we might suppose that been is required in 1 because perfective have requires a comple¬
ment headed by a perfective participle verb form. Say how the sentences above would
be analysed on these two accounts, and whether either (or both) analyses run into prob¬
lems in accounting for any of the examples. Note that be in 10 is a subjunctive form
(cf. the discussion of such forms in exercise II of chapter 4).
262
Workbook section
V T V pp
II I / \
isPres t P DP
* i \ / \
1-1 in the army
(ii)
PP
P DP
.1 / \
in the army
Why should is be required here? One answer could be that T is a weak verbal affix, and
so must either be merged with an auxiliary (like is), or associated with a set of verbal
features which are attracted to T and thereby percolate up from the head V in VP: but
since there is no VP (and hence no V) in (ii), the only possibility is to merge the affix
with an auxiliary. On this account, be is a dummy auxiliary used only as a last resort
(to satisfy the categorial requirements of the tense affix T - i.e. the requirement that it
be attached to a verbal stem): it is used in structures with a nonverbal predicate
(whereas dummy do is used as a last resort in structures with a verbal predicate). An
alternative possibility would be to suppose that tense-features must be visible at PF -
which amounts to claiming that every finite clause in English must contain an overt
tensed auxiliary or nonauxiliary verb (though clearly this claim would have to be mod¬
ified in relation to clauses where a finite verb undergoes ellipsis/gapping).
263
Head movement
It might be argued that the economy principle favours the minimalist analysis in (ii)
over the bc-raising account in (i), since (ii) involves a more economical derivation (one
with no VP projection and no movement).
1 Rastus be happy
2 Rastus don’t be happy
3 *Rastus ben’t happy
4 Do Rastus be happy?
5 *Be Rastus happy?
(0
[be happy
i )
The head-feature [Pres] of be tells us that it is a present-tense form, and its [Nom]
specifier-feature tells us that it requires a nominative subject: its tense-feature is
checked against that of T, and its specifier-feature against those of the DP Rastus.
(Given the claim in §4.7 that all argument nominals are DPs, Rastus will be a DP
headed by a null nominative determiner.)
264
Workbook section
by children aged 2-4 years are given below (the name of the child producing each
utterance is given in parentheses):
Akmajian and Heny (1975, p. 17) report a three-year-old girl producing interrogative
structures such as:
9 Is I can do that?
10 Is you should eat the apple?
11 Is Ben did go?
12 Is the apple juice won’t spill?
A similar pattern is noted by Davis (1987), who reports the following examples of
child yes-no questions:
Discuss how do/be appear to be used by the children concerned in each of the exam¬
ples, highlighting differences with the corresponding adult utterances.
Helpful hints
You might find it useful to bear in mind the helpful hints/model answers
given in relation to the two previous exercises.
265
Head movement
Here, was is a copular (i.e. ‘linking’) verb which moves from T to adjoin to the strong
question affix Q which occupies the head C position of CP. Since adult be can itself
adjoin to Q in this way, do would be superfluous in be questions, and so cannot be used
(since the economy principle bars the use of superfluous elements in syntactic repre¬
sentations).
The fact that Sarah uses Jo in (1) might suggest that she misanalyses be (in the rele¬
vant use) as a lexical verb with intrinsic semantic content (so that e.g. be in 1 is synony¬
mous with ‘behave like’); if so, it will originate (and remain) in the head V position of
VP, and cannot adjoin to T (and thence to Q) because T is a weak affix in English. If be is
prevented from moving out of the head V position of VP, it follows that (like typical
nonauxiliary verbs) be (in the relevant use) will require Jo-support in interrogative sen¬
tences: thus, in 1, Jo will be merged with a past-tense affix T, and will then adjoin to Q
(perhaps pied-piping the past-tense affix along with it, so that it is the whole T constituent
rather than just do which adjoins to Q: see the discussion of excorporation in §6.9).
266
7
Operator movement
7.1 Overview
In the previous chapter, we looked at a particular kind of movement oper¬
ation found in English, called head movement. In this chapter, we look at a very dif¬
ferent kind of movement operation, known as operator movement because it applies
to expressions which contain an (e.g. negative or interrogative) operator of some kind.
Our discussion in the text will be concerned solely with interrogative operators (though
we shall look at other types of operator construction in the exercises).
7.2 Wh-operators
To get our discussion underway, consider the syntax of the following
wh-questions (i.e. questions containing an interrogative wh-word):
Each of the sentences in (1) contains an (italicized) inverted auxiliary occupying the
head C position of CP, preceded by a (bold-printed) interrogative expression. Each of
the bold-printed pre-auxiliary phrases contains an interrogative operator (viz. what?,
which?, where? or who?). Expressions containing an operator are - for obvious
reasons - called operator expressions', hence what languages?, which road?, where?
and who? are all (interrogative) operator expressions in (la-d).
It seems clear that each of the operator expressions in (1) functions as the comple¬
ment of the verb at the end of the sentence. One piece of evidence leading to this con¬
clusion comes from the fact that each of the examples in (1) has a paraphrase in which
the operator expression occupies the canonical (i.e. ‘typical’) complement position
after the relevant verb: cf.
267
Operator movement
Structures like those in (2) are sometimes referred to as wh-in-situ questions, since the
italicized wh-expression does not get preposed, but rather remains in situ (i.e. ‘in
place’) in the canonical position associated with its grammatical function (e.g. what
languages in (2a) is the complement of speak, and complements are canonically posi¬
tioned after their verbs, so what languages is positioned after the verb speak).
Structures like (2) are used primarily as echo questions, to echo and question some¬
thing previously said by someone else (e.g. if a friend boasts ‘I just met Nim
Chimpsky’, you could reply - with an air of incredulity - ‘You just met who?’).
Sentences such as (2) make it seem plausible to suppose that the italicized operator
expressions in (1) originate as complements of the relevant verbs, and subsequently
get moved to the front of the overall sentence. But what position do they get moved
into?
The answer is obviously that they are moved into some position preceding the
inverted auxiliary. Now, if inverted auxiliaries occupy the head C position of CP
(adjoined to a question affix Q), we might suppose that preposed operator expressions
are moved into some prehead position within CP. Given that specifiers are canonically
positioned before heads, a natural suggestion to make is that preposed operator expres¬
sions occupy the specifier position within CP (= spec-CP). If this is so, a sentence such
as (la) will be derived as in (3) below (simplified by ignoring the possibility that can
originates adjoined to a tense affix in T):
speak t
Two different kinds of movement operation (indicated by the arrows) are involved in
(3): (1) involves movement of can from T to adjoin to the C position occupied by the
question affix Q, and is an instance of head movement. By contrast, (2) involves
movement of an operator expression (viz. the determiner phrase what languages? con¬
taining the interrogative operator what?) from complement position within VP into
specifier position within CP: since this second type of operation involves movement of
268
7.2 Wh-operators
In nonsloppy speech styles, however, the sequence want to cannot contract to wanna in
a sentence such as (5a) below, as we see from the ungrammaticality of (5b):
Why should this be? Well, let's assume that who originates as the subject (and speci¬
fier) of the IP (infinitive phrase) to help you - as seems plausible in view of the fact that
(5a) has the echo question counterpart:
Let’s also assume that the wh-pronoun who (being an interrogative operator) undergoes
movement from its underlying spec-IP position to the spec-CP position at the front of
the clause, and that it leaves behind a trace in spec-IP. Assuming that auxiliary inver¬
sion also takes place, the resulting structure will then be along the lines of (7) below
(where t is the trace of who\ we omit the trace of would for simplicity):
Why should wanna contraction then be blocked in a structure such as (7)? Since we have
already seen in earlier chapters that cliticization is subject to an adjacency condition (and
hence is only possible when the two words involved are immediately adjacent), the pres¬
ence of the intervening trace t between to and want will prevent to from cliticizing onto
want, and hence blocks wanna-contraction. (We might suppose that in sloppy speech
styles, intervening traces are ignored, and so do not suffice to block contraction.)
A similar kind of argument in support of claiming that moved operator expressions
leave behind a trace comes from facts about /jave-cliticization. The perfective auxiliary
form have has the clitic variant Ivl and can cliticize to an immediately preceding word
269
Operator movement
(8) (a) Which students would you say have got most out of the course?
(b) * Which students would you say’ve got most out of the course?
(Hence the sequence say’ve is not homophonous with save.) Why should have be pre¬
vented from cliticizing onto say here? We might assume that prior to being moved to
the front of the sentence by operator movement, the operator phrase which students
was the subject of have - as in the echo question counterpart (9) below:
(9) You would say which students have got most out of the course?
If we also assume that when the phrase which students is fronted, it leaves behind a
trace (= t) in the position out of which it moves, then the superficial structure of (8b)
will be (10) below (simplified by omitting the trace of would):
(10) Which students would you say t have got most out of the course?
This being so, we can say that have cannot cliticize onto say in (10) because it is not
immediately adjacent to say, the two words being separated by the intervening trace t
in (10) - so accounting for the ungrammatically of (8b).
The analysis presented here correctly predicts that while have cliticization is not
possible in structures like (8b) above, it is indeed possible in structures like (lib)
below:
Given our twin assumptions that preposed operators move into spec-CP and inverted
auxiliaries adjoin to a question affix Q in COMP, (lib) will have the derivation (12)
below (where movement (1) is head movement, and (2) is operator movement):
(12)
D'
Who
A
270
7.3 Enlightened self-interest
(Note that (12) is simplified by ignoring the possibility that have may be merged with a
tense affix.) Since who is immediately adjacent to have here, our analysis correctly pre¬
dicts that PF cliticization should be possible: the fact that the resulting sentence (lib)
is grammatical makes it all the more plausible to posit that spec-CP is the landing-site
for preposed operator expressions (i.e. spec-CP is the position into which they move).
Additional support for the claim that operator expressions move into spec-CP comes
from the fact that in some varieties of English, preposed wh-expressions can precede a
complementizer like that. This is true, for example, of Belfast English structures such
as the following (from Henry 1995, p. 107):
If we assume that the complementizer that occupies the head C position of CP, it seems
reasonable to suppose that the wh-expressions which dish/which model that precede
the complementizer that are in spec-CP (i.e. occupy the specifier position within CP) -
and this is what Alison Henry argues.
What originates as the complement of doing-, merging doing with what forms the VP
(15) below:
(15) VP
V
I
doing what
271
Operator movement
The progressive auxiliary was is merged with a past-tense affix T as in (16) below:
Prog T
was Past
The resulting T constituent in (16) is then merged with the VP in (15) to form the T-bar
(17) below:
The subject he is then merged with the overall T-bar constituent in (17) to form the TP
(18) below:
he T VP
7 \ / \
Pr°g T y D
I I ! I
was Past doing what
(19)
doing what
Since Q is a strong affix, the T constituent was+Past adjoins to the C-node containing
Q, as in (20) below:
t i I I
doing what
272
7.3 Enlightened self-interest
(In the context of our discussion in §6.9, note that we are assuming here that the whole
T constituent adjoins to C, rather than just progressive was - though this is a question
of detail of no direct relevance to our discussion of operator movement here.) Since Q
also carries a [wh] specifier-feature, what? (which is an interrogative determiner con¬
taining the head-feature [wh]) raises to become the specifier of C-bar, as indicated by
the arrow in (21) below:
; Prog T Q he T VP
: I I ! / \
! was Past t V D
' !
doing t
t 1
The [wh] specifier-feature of Q can then be checked by the [wh] head-feature carried
by what, and is thereby erased, since specifier-features are uninterpretable at LF.
(However, the [wh] head-feature carried by what is not erased, since it plays a role at
LF in identifying what as an interrogative operator.) Checking here will involve a local
spec-head relation (between the head and specifier of CP) if we assume that the [wh]
feature of Q is carried by both C segments of the split-segment C constituent. On this
account, movement of what to spec-CP is motivated by altruism (i.e. enlightened self-
interest), in that this enables the specifier-feature of Q to be checked. (We shall con¬
sider why the relevant feature must be checked by movement rather than attraction in
§7.7.)
The assumption that what moves to spec-CP for altruistic reasons (to check and
erase the [wh] specifier-feature of Q) provides an interesting account of why in mul¬
tiple wh-questions (i.e. questions containing more than one wh-operator) such as (22)
below, only one wh-operator can be preposed, not both:
Example (22a) has the derivation (23) below. (Here and throughout the rest of this
chapter, we simplify the internal structure of T and C constituents for ease of exposi¬
tion; fuller structures for T and C are given in (18) and (21) above.)
273
Operator movement
The [wh] specifier-feature of the C-node (which is associated with the question affix Q
contained in C) is erased once it has been checked by movement of who into spec-CP.
If we assume that operator movement is motivated purely by the altruistic desire to
erase the [wh] specifier-feature of the C-node containing Q, it is clear that once this
[wh] feature has been erased by movement of who to spec-CP, there is no motivation
for the second wh-operator to be preposed, since the [wh] specifier-feature of Q has
already been checked and erased. Thus, the principle of enlightened self-interest pro¬
vides us with a neat account of the fact that only one wh-operator is preposed in mul¬
tiple wh-questions in English.
However, a question not answered by our existing account is why it should be pos¬
sible to prepose who on its own in a multiple wh-question like (22a), but not what - i.e.
why we can’t say:
Example (24) would have the derivation (25) below (simplified in respect of the inter¬
nal structure of T and C):
274
7.3 Enlightened self-interest
(25)
think T
who VP
say t
Since movement of what into spec-CP would enable the [wh] specifier-feature of Q to
be checked, why does the derivation in (25) crash? Why should who be able to move to
spec-CP in (23), but not what in (25)?
A natural suggestion to make is that this is because who is closer to Q than what, and
the economy principle favours shorter movements over longer ones. So, Q lures the
nearest wh-operator into spec-CP: since who is nearer to Q in (23/25) than what, it is
who which must move to spec-CP, not what. The relevant condition (favouring shorter
moves over longer ones) is known as the shortest movement principle (cf. Chomsky
1995b). Since the effect of the principle is to favour the formation of movement chains
with minimal (i.e. the smallest/shortest possible) links, it is also referred to as the
minimal link condition (or the minimality condition). It should be apparent that the
head movement constraint discussed in the previous chapter (which holds that head
movement is only possible from a complement head into a matrix head position) can
be subsumed under the shortest movement principle (or the equivalent minimal
link/minimality condition).
275
Operator movement
7.4 Pied-piping
Thus far, most of the instances of wh-movement which we have looked at
have involved movement of a single wh-determiner (a pronominal wh-determiner such
as who? or what?) into spec-DP. However, in sentences such as (26-8) below, we find
movement of a whole DP headed by a wh-determiner into spec-CP, with movement of
the determiner alone leading to ungrammaticality:
To use the familiar pied-piping metaphor, it would seem that when the wh-determiner
which?/what? moves to spec-CP in (26-8), its complement (film/of them/reason) has to
be pied-piped along with the moved wh-operator. Why should this be?
Consider first why moving the wh-determiner without its complement leads to
ungrammaticality. Sentence (26b) will have the simplified derivation (29) below:
Why should the derivation in (29) crash? One possibility is that movement of which on
its own into spec-CP results in the formation of a chain which . . . t which is nonuni¬
form in that which and its trace differ in respect of their phrase structure status. To see
why, consider the immediate structures containing which and its trace, given in (30a-b)
below:
In (30a) the D-node containing which is a maximal projection (i.e. it is the largest
category headed by which), whereas in (30b) the D-node containing its trace t is not a
276
7.4 Pied-piping
maximal projection (since the maximal projection of the D-node containing t is the DP
t film). If (following Chomsky 1995a, p. 406) we assume the following principle gov¬
erning the well-formedness of chains:
the ungrammaticality of the (b) examples in (26-8) above can be accounted for in a
principled fashion, since the (b) examples all result in a nonuniform chain whose head
(= the preposed wh-word which? Iwhat?) is a maximal projection, but whose foot
(= the trace of the preposed wh-word) is not.
But now consider what happens if we prepose the whole DP which film? in (26a), in
the manner indicated in (32) below:
see t
i
Here the resulting chain is uniform, since its head is the DP which film (which is a max¬
imal projection) and its foot is a DP trace (hence also a maximal projection). Thus, it
would seem that operator movement moves a D alone wherever possible (e.g. in struc¬
tures like (25) above), but moves a DP when movement of D alone would violate the
chain uniformity principle (e.g. in structures like (32) above). The more general
principle which this suggests is that movement operations move only the minimal con¬
stituent required to satisfy UG principles: this can be argued to be a particular instance
of the more general economy principle, which in effect tells us to move features rather
than constituents whenever possible, and (when moving constituents) to move the
smallest constituent possible the smallest distance possible.
We can extend the account given here from contrasts such as (26-8) above to those
such as (33) below:
277
Operator movement
who D N
’s car
Each of the sentences in (35) is derived by preposing a wh-expression into the spec-CP
position indicated by the question-mark in (36) below (and moving the auxiliary were
from T to adjoin to Q):
(36)
7
A
PP
talking D
who
278
7.4 Pied-piping
will also be true if the PP to who moves into spec-CP, since the result will be a uniform
chain whose head and foot are both maximal (PP) projections; and the same is likewise
true if the VP talking to who moves into spec-CP, since this will again result in a uni¬
form chain whose head and foot are both maximal (VP) projections. So, since prepos¬
ing the D who, the PP to who and the VP talking to who all result in the formation of a
uniform chain, why should preposing D be preferred to preposing PP, or preposing
VP? The obvious answer is that the D who is smaller than the PP to who, which in turn
is smaller than the VP talking to who. And if we assume that (in consequence of the
economy principle) operator movement moves only the minimal constituent needed
to check the [wh] specifier-feature of Q, the relevant facts fall out much as we expect.
However, our discussion here overlooks an important stylistic variable. In formal
styles of English (i.e. in styles where the objective form of the relevant pronoun is
whom rather than who), things are rather different, as we see from the sentences below:
As these examples show, in formal English, the wh-pronoun whom cannot be preposed
on its own; rather, the preposition to must be pied-piped along with whom (though the
verb talking cannot be pied-piped in the same way). What is going on here?
Consider first why (37a) is ungrammatical. What’s happened here is that the preposi¬
tion to has been stranded or orphaned (i.e. separated from its complement whom)-, the
fact that (37a) is ungrammatical suggests that there is a constraint against preposition-
stranding in formal styles of English - a constraint rather inaccurately encapsulated in
the traditional rule of prescriptive grammar: ‘Never end a sentence with a preposition.’
The fact that prepositions can be stranded in colloquial English but not in formal
English would seem to correlate with the fact that we find the overtly case-inflected
form whom in formal English (where +m marks objective case), but the uninflected
form who in colloquial English. How might these two sets of facts be related?
Speculating at this point, let’s conjecture that in formal styles of English, the overtly
case-marked form whom checks its case in sentences like (37) above by attraction to
the transitive preposition to. If we suppose that attraction of overt case-features takes
place in the overt syntax (prior to spellout), and triggers pied-piping of other
unchecked grammatical features along with the attracted case-feature, it follows that
not only will the objective case-feature of whom percolate to to, but so too will the
wh-feature carried by who (with the result that the wh-feature is attached to to). The
preposition to cannot be preposed on its own (since this would result in violation of the
chain uniformity principle), so its complement whom is pied-piped along with it, and
consequently the whole PP to whom moves to spec-CP in order to check the wh-feature
of COMP, as in (37b) above.
279
Operator movement
But what of colloquial English? Here, we might suppose that (since it does not
overtly inflect for objective case) who carries covert case, and that its case properties
are checked after spellout in the covert syntax (i.e. in the LF component). Because
there is no attraction of the case properties of who to to in the overt syntax, there is no
pied-piping of the wh-feature carried by who, with the result that only who is moved to
spec-CP to check the wh-feature carried by COMP. (A more radical and heretical
variant of this analysis is to suppose that the colloquial form who has no case proper¬
ties to be checked: for a different view of preposition pied-piping and stranding, see
Salles 1995.)
Given the preposition-stranding constraint, we can begin to make sense of the
data in (37). Movement of whom alone will violate the constraint, and so leads to
ungrammaticality in (37a). The minimal constituent which can be preposed without
violating the constraint is the PP to whom, in which to is pied-piped along with whom-.
hence, (37b) is grammatical. In (37c), the whole VP talking to whom has been
preposed, with the verb talking being pied-piped as well; but this violates the economy
principle (which requires that only the minimal constituent required to ensure conver¬
gence should be preposed). So, the contrasting patterns of grammaticality found in
colloquial English sentences like (35) on the one hand and formal English sentences
like (37) turn out to be reducible to the fact that the constraint against preposition-
stranding holds in formal English (and many other languages - e.g. French, Italian,
Spanish, etc.), but not in colloquial English. Everything else about the relevant data
follows from UG principles.
Our discussion here suggests that pied-piping of additional material along with wh-
operators occurs only when it is forced by the need to ensure convergence. This
assumption offers us an interesting account of the otherwise puzzling set of data in (38)
below (cf. Chomsky 1995b):
(38) (a) * Whose did you think pictures of mother were on the mantelpiece?
(b) * Whose mother did you think pictures of were on the mantelpiece?
(c) * Of whose mother did you think pictures were on the mantelpiece?
(d) Pictures of whose mother did you think were on the mantelpiece?
280
7.4 Pied-piping
D were on the
I mantlepiece
0
pictures P DP
I / \_
of D D
I / \
-who D N
’s mother
Whose cannot be preposed on its own, for reasons already familiar from our earlier
discussion of (33) above - i.e. it is not a constituent, given the analysis in (39). More
puzzling is why the whole DP whose mother can’t be extracted in (38b). This is all the
more puzzling in view of the fact that whose mother can indeed be extracted in sen¬
tences such as:
How come whose mother can be extracted in (40), but not in (39)?
The answer would seem to lie in the different positions occupied by the DP pictures
of whose mother prior to extraction. In (40), it is the complement of the verb
taking!exhibiting!collecting-, but in (39) it is the subject of were. It would seem that (to
use a colourful metaphor suggested by Ross 1967) subjects are islands - i.e. they are
constituents which nothing can be extracted out of. Or, to put matters the other way
round, we might say - following the condition on extraction domains (= CED) sug¬
gested by Huang 1982 - that complements are the only types of constituent which
allow extraction out of them. In the light of this constraint, we can begin to make sense
of the data in (38). Moving the DP whose mother in (38b) or the PP of whose mother in
281
Operator movement
(38c) will lead to violation of the CED, since the DP/PP concerned are being moved
out of the larger containing subject expression pictures of whose mother. The minimal
structure which can be moved without violating CED is the whole subject DP pictures
of whose mother. Thus, pied-piping of the rest of the DP along with the wh-pronoun
who is a last resort, forced by the need to avoid violation of CED. Once again, we find
that pied-piping is used only as a last resort, when needed to ensure convergence (i.e.
to ensure a grammatical outcome).
We can summarize the assumptions we have made about wh-movement (i.e. move¬
ment of wh-operator expressions) in the following terms. Wh-questions are CPs
headed by a C constituent containing an interrogative affix Q which has a [wh] speci¬
fier-feature checked by movement of a wh-word to spec-CP. Other constituents are
pied-piped along with the wh-word only if this is required for convergence - i.e. only if
the derivation would otherwise crash.
It seems natural to suppose that we should treat embedded wh-questions such as those
bracketed in (41) in precisely the same way as we treated their root-clause counter¬
parts. More concretely, we might suppose that the embedded clause who had / seen in
(41a) is a CP whose head C position contains the question morpheme Q, and that Q is a
strong (affixal) head with a [wh] specifier-feature, so that had adjoins to Q and who
282
7.5 Embedded questions in BE
moves to spec-CP, as in (42) below. (Here and subsequently, we simplify the internal
structure of C, and show only those aspects of the derivation which are of immediate
concern, and not e.g. the fact that the head- and specifier-features of asked are attracted
toT.)
seen t
Such an analysis would offer the obvious advantage that it requires us to posit no addi¬
tional descriptive apparatus or theoretical assumptions over and above what we already
need to handle root-clause wh-questions.
Moreover, the analysis can be extended in a fairly straightforward fashion to deal with
wh+that questions in Belfast English such as those in (13) above, repeated in (43) below:
Given that the greed of a strong affix (i.e. its need to attach to an overt head) can be satis¬
fied either by movement or by merger, an obvious suggestion to make about structures
such as (43) is that they involve merger of the complementizer that with the interrogative
affix Q, together with movement of the wh-DP which dish from VP-complement position
to CP-specifier position, as in (44) below (simplified by showing only movement, not
feature attraction):
283
Operator movement
(44) TP
which dish C TP
A \
that Q D
they T VP
/ \
Past V DP
picked
We could then say that the requirement for the strong affix Q to be bound to an overt
head is satisfied in structures like (42) by movement (i.e. adjunction of had to Q), and
in structures like (44) by merger (of that with Q); in both cases, the [wh] specifier-
feature of Q is checked by movement of wholwhich dish to spec-CP. Since it is a lexi¬
cal property of overt complementizers like that!if,'/for that they occur only in embedded
clauses and not in root clauses, it follows that C in root-clause questions will always be
filled by movement (i.e. adjunction of an auxiliary to Q), not by merger of Q with a
complementizer like that (cf. the ungrammaticality of * Which dish that they picked? as
a root clause).
Interestingly, we find that inversion is also permitted (in the relevant variety of
Belfast English) in noninterrogative clauses from which an italicized interrogative
operator has been extracted, such as those bracketed in (45) below (cf. Henry 1995,
p. 108):
It is clear that the bracketed complement-clauses are declarative rather than interroga¬
tive, since hope/claim don’t select interrogative complements (as we see from the
ungrammaticality of sentences such as */ hoped/claimed whether he was coming). Thus,
the bracketed complement-clauses in (45) must be CPs headed by a null declarative
COMP. Since declarative COMP constituents generally have weak head-features in all
varieties of present-day English (and so don’t trigger inversion), we would therefore not
expect to find inversion in declarative complement clauses like those bracketed in (45).
So what triggers inversion in such cases?
Since inversion in declarative clauses is only found in clauses out of which an inter¬
rogative operator has been extracted, an obvious suggestion to make is that the extrac-
284
7.5 Embedded questions in BE
tion of the operator somehow triggers inversion. But how? Well, since inversion is
found in CPs with an interrogative specifier, let’s suppose that when the interrogative
operator is extracted out of the bracketed complement clauses in (45), it moves first to
the front of the complement clause (thereby triggering inversion in the complement
clause), and then to the front of the root clause (thereby triggering inversion in the root
clause). This would mean that a sentence such as (45a) would have the (simplified)
derivation (46) below:
he T VP
t V CP
he T VP
i1 i- I 1 V
/ \D
; I
i see t
;-.
What is being claimed in (46) is that the wh-determiner who moves to the front of the
overall sentence in two successive steps, numbered (1) and (2). On its first movement,
who moves into the spec-CP position within the would clause; and on its second move¬
ment, who moves into spec-CP in the did clause. Thus, operator movement applies in a
successive cyclic fashion, moving the operator to the front of the sentence one clause at
a time. We might suppose that successive cyclicity is forced by the shortest move¬
ment principle, since this will require that a moved operator expression should move
into the nearest spec-CP position above it (each time it moves).
However, this still leaves the question of why did and would undergo inversion in
(46), and thereby move to COMP. In the case of did, the answer is straightforward,
since the root (i.e. topmost) clause is interrogative, and so the head C position within
CP contains the strong interrogative affix Q: thus, inversion of did is simply adjunction
of did to Q. But inversion of would is more problematic, since the won/J-clause is
declarative, and a declarative C is normally weak (and so cannot trigger movement of
an auxiliary to COMP). What is going on here? One answer is to suggest that any C
285
Operator movement
which has an interrogative specifier always has strong head-features (and so behaves
like an affix, triggering movement of an auxiliary into C). This means that although the
embedded COMP filled by would in (46) is noninterrogative, it is strong by virtue of
the fact that it comes to have an interrogative specifier at the point in the derivation at
which who moves into the specifier position in the embedded clause (i.e. immediately
after movement (1) takes place). Thus, a declarative COMP is generally weak in
Belfast English, but is strong where it has an interrogative specifier. It may be that the
auxiliary would in (46) adjoins to an abstract declarative affix Dec (which would be the
declarative counterpart of the question affix Q found in interrogatives).
286
7.6 Embedded questions in SE
features in SE (so doesn’t allow auxiliary inversion). However, this apparently simple
solution is fraught with hidden complications. If embedded questions contain the same
Q morpheme as root questions, how come Q isn’t strong (since it is strong in root
clauses, where it triggers inversion)? We could try saying that Q can be either strong or
weak, and that verbs like ask/wonder which select an interrogative complement always
select a complement headed by a weak Q. But this then poses the problem of how we
account for why Q is always strong (and triggers inversion) in root clauses. Clearly, we
can’t say that principles of UG determine that Q universally has strong head-features in
root clauses with an interrogative operator in spec-CP, since (for example) in Jamaican
Creole we find root wh-questions with no inversion - as the following examples (from
Bailey 1966) illustrate:
Why is inversion not possible here? The answer is that the morphological requirements
of the affix Q are satisfied by merger of that with Q, so that there is no necessity for
(and hence no possibility of) adjoining the auxiliary to Q. Suppose we further posit that
Standard English has a null complementizer 0 which can adjoin to Q in complement-
clause questions (in much the same way as that adjoins to Q in Belfast English) -
though not in root questions (since it is a lexical property of complementizers like that
and 0 that they only occur in complement questions). It follows that a Standard English
sentence such as:
287
Operator movement
The fact that the null complementizer 0 merges with Q satisfies the requirement for Q
to be affixed to an appropriate host, and thereby blocks any form of auxiliary inversion
- hence, Jo-support is neither necessary nor possible in (52). A minor complication
posed by the analysis is that we have to assume that a null constituent (in this case, the
null complementizer 0) can satisfy the morphological requirements of a strong affix,
whereas in the previous chapter we assumed that only an overt constituent could do so.
This means we have to posit that the defining characteristic of strong affixes is that
they require an (overt or covert) lexical item to be attached to them.
In order for the null complementizer account to work, however, we have to posit that
all embedded CPs in Standard English must be headed by an (overt or covert) comple¬
mentizer: in a finite declarative CP in SE, the complementizer will be the overt
complementizer that; in a wh-question complement clause in SE, it will be the covert
complementizer 0. It may be (as suggested by Kayne 1982) that the head C of an
embedded CP is always nonverbal in Standard English (as is suggested by the fact that
for is a prepositional complementizer), and hence does not allow a verb to be adjoined
to it: by contrast, a root-clause C is not restricted in the same way, and hence allows an
inverted auxiliary to be adjoined to it. The obvious drawback to this analysis is that it
is stipulative rather than explanatory (i.e. it doesn’t offer us any account of why C
should be nonverbal in embedded clauses in Standard English, though not in Belfast
English).
288
7.6 Embedded questions in SE
We can assume that the Q affix to which the null complementizer 0 adjoins has the
interrogative specifier-feature [wh]: in structures like (52), the [wh] specifier-feature of
Q is checked by movement of which dish into spec-CP. However, in embedded
w/iet/ier-questions such as:
there seems little reason to suppose that whether moves to spec-CP from somewhere
lower down in the sentence. Rather, it seems more likely that whether is directly gener¬
ated in spec-CP, and thus merges with C-bar, so that the bracketed complement clause
in (53) has the structure (54) below:
(54) tp
I was leaving
Thus, spec-CP may be filled either by merger (as in (54)), or by movement (as in
(52)).
We argued earlier in relation to the derivation outlined in (46) above that operator
movement applies in a successive cyclic fashion in Belfast English. If we assume that
this is in consequence of a principle of Universal Grammar (e.g. Chomsky’s 1995b
shortest movement principle or Rizzi’s 1990 relativized minimality principle), it
follows that the same must also be true in other varieties of English (and indeed in all
natural languages). This in turn means that in a Standard English question such as:
operator movement applies in a successive cyclic fashion, moving how first into the
spec-CP position in the that-clause, and then into spec-CP in the Jo-clause, as indi¬
cated by the (simplified) derivation in (56) below:
289
Operator movement
How C TP
feeling t
'-(1)-'
(We have assumed here that how is an adjectival wh-proform, since it is the comple¬
ment of feel, and feel typically selects an adjectival complement, like fine in I feel fine:
although how doesn’t begin with wh, it is standardly classified as a wh-word because it
exhibits the same syntactic behaviour as other wh-words.) Thus, how first becomes the
specifier of the COMP containing that, and then moves on to become the specifier of
the COMP node containing do.
At first sight, a successive cyclic derivation such as (56) might seem to be poten¬
tially problematic, in that it requires us to posit an intermediate stage of derivation
(immediately after movement (1) has applied) in which the complementizer that has
the wh-specifier how. What might seem to be problematic about this is that Standard
English doesn’t allow wh+that structures, as we see from the ungrammaticality of sen¬
tences such as:
But this objection turns out to have little force. After all, one might counter that what
the ungrammaticality of (57) actually illustrates is not that the complementizer that can
never have an interrogative specifier at any stage of derivation in Standard English, but
rather that it can’t be the head of an interrogative clause (i.e. can’t attach to Q) - as we
noted earlier in relation to our discussion of (44/47) above. So, since that heads a
declarative clause in (56), the relevant derivation converges.
The assumption that operator movement applies in a successive cyclic fashion pro¬
vides us with a principled account of why a sentence such as (58a) below doesn’t have
a grammatical interrogative counterpart like (58b):
290
7.6 Embedded questions in SE
If (as we claim) whether occupies the specifier position within the bracketed comple¬
ment clause, it is clear that we can’t derive (58b) by moving how first into the specifier
position within the bracketed CP, and then into the specifier position within the main
clause (for the obvious reason that the specifier position within the bracketed CP is
already occupied by whether). So, the only way of deriving (58b) is by long movement
of how directly into the main clause spec-CP position, as in (59) below:
Why should the resulting sentence (58b) be ungrammatical? The answer is that the
shortest movement principle prohibits long movement of how directly into the main
clause spec-CP position, since this isn’t the nearest spec-CP position above the (VP-
complement) position in which how originates (rather, the complement-clause spec-CP
position occupied by whether is the nearest spec-CP position). So, since successive
cyclic movement is blocked by the fact that the complement-clause spec-CP position is
filled by whether, and the long movement in (59) is blocked because it violates the
shortest movement principle, there is no way of deriving a sentence like (58b). More
generally, the ungrammaticality of sentences like (58b) provides empirical support for
the claim that operator movement applies in a successive cyclic fashion, moving a
moved operator into the next highest spec-CP position within the structure containing
it. (For an interesting minimalist account of locality constraints on operator movement,
see Manzini 1994.)
291
Operator movement
If we assume that all wh-questions (excluding whether questions) are CPs containing a
wh-operator which moves into spec-CP, (60) will involve movement of who from its
underlying position in spec-TP into its superficial position in spec-CP, and so will have
the (simplified) derivation (61) below:
(61) CP
Who C TP
Past V D
helped you
But what is problematic about the derivation in (61) is that Q is a strong affix in
English, and so triggers auxiliary inversion in root clauses. If (61) were the right analy¬
sis for (60), we would therefore expect to find that do-support is required as a last
resort here, to provide a verbal host for the strong Q affix to attach to. The absence of
any inverted auxiliary in (60) calls into question the assumption that the subject who
moves into spec-CP in (61).
If who doesn’t move to spec-CP, it must remain in situ in spec-IP. If we continue to
assume that all questions are CPs headed by a question affix Q, the structure of (60)
must be along the lines of (62) below:
(62) CP
helped you
But an analysis like (62) poses two problems: (i) how is the wh-feature of Q checked;
and (ii) why is there no do-support and no auxiliary inversion?
292
7.7 Subject questions
Since there are three different types of operation by which grammatical features can
be checked (viz. merger, movement and attraction) and since who doesn’t merge with
Q in (43) and doesn’t move to spec-CP, we might follow Chomsky (1995b, p. 293) in
supposing that in this type of structure the [wh] feature of Q here is checked by attrac¬
tion, with the [wh] head-feature of who adjoining to the C node containing Q, thereby
enabling the [wh] feature of Q to be checked and erased. Thus, the [wh] feature of Q in
an interrogative clause with an interrogative subject is checked by wh-attraction, not
by wh-movement. Feature attraction is preferred to movement in such structures
because (as noted in the previous chapter) attraction is more economical than move¬
ment. If we suppose that attraction is a strictly local operation whereby a head can only
attract the features of the head or specifier of its complement, it follows that only the
subject of an interrogative clause can check the strong [wh] feature of Q by attraction:
other types of constituent can only check the strong [wh] feature of Q by moving to
spec-CP.
So, our analysis provides a principled account of why we don’t find wh-movement
in interrogative clauses with interrogative subjects. But why don’t we find auxiliary
inversion in such sentences? One possible answer is to suppose that Q is only a strong
affix when the CP headed by Q has an interrogative operator-specifier of its own. Since
the CP headed by Q in (62) contains no specifier, it follows that Q in (62) is weak, and
so does not trigger auxiliary inversion/do-support. An alternative possibility is to posit
that T in interrogative clauses carries a [wh] specifier-feature; it follows that T would
move to C (for example) in complement questions like (3) above (where what lan¬
guages is in spec-CP); but in subject questions like (62), T could check its interrogative
specifier-feature by remaining in situ, since TP has an interrogative specifier of its own
(namely who).
Thus far, we have assumed (following Chomsky 1995b) that all interrogative
clauses are CPs headed by an abstract question affix Q. But now let’s consider an alter¬
native possibility - namely that the defining characteristic of an interrogative clause is
that it must contain an interrogative specifier in order to be interpretable as a question
at LF. Let’s also assume that COMP is strong in a CP with an interrogative specifier in
Modem Standard English, but INFL is always weak. This alternative (interrogative
specifier) analysis of questions offers a subtly different account of the syntax of wh-
questions. In the case of sentences such as (la) What languages can you speak?, we
can say that the wh-DP what languages? raises to spec-CP in order to ensure LF con¬
vergence (i.e. to ensure that the sentence is interpretable as a question at LF by virtue
of having an interrogative specifier). In multiple wh-questions like (22), only one of
the wh-operator expressions will raise to spec-CP (as in (23) above), since the econ¬
omy principle specifies that we move the minimal material required in order to ensure
convergence. (The requirement for an interrogative clause to have an interrogative
specifier can be met by moving only one of the wh-constituents to spec-CP.) But where
293
Operator movement
the interrogative specifier analysis differs most from Chomsky s interrogative head
analysis is in the analysis of subject questions like (60) Who helped you? Under
Chomsky’s interrogative head analysis, this is a CP with the structure (62) above. But
under the alternative interrogative specifier analysis, (60) is instead a simple TP, with
the structure (63) below:
(63) TP
T^ ^ VP
1
Past V
/ \D
1 1
helped you
Since who occupies spec-TP, the resulting TP satisfies the requirement for a question to
contain an interrogative specifier; hence there is no need to project the structure further
into CP, and therefore (by the economy principle) no possibility of doing so. (For a TP
analysis of subject questions, see Grimshaw 1993a, 1993b.)
(64) Is it raining?
(65)
Op
raining
294
7.8 Yes-no questions
(66) (a) Whether had you rather lead mine eyes or eye your master’s heels? (Mrs
Page, Merry Wives of Windsor, Ill.ii)
(b) Whether dost thou profess thyself a knave or a fool? (Lafeu, All’s Well
That Ends Well, IV.v)
Given the null-operator analysis, we could posit that root yes-no questions have essen¬
tially the same syntax in present-day English as in Early Modem English, save that
they could be introduced by the overt operator whether in Early Modem English, but
are introduced by a covert operator Op in present-day English.
A further piece of evidence in support of assuming that inversion questions contain a
null counterpart of whether comes from the fact that root yes-no questions can be
introduced by whether when they are transposed into reported speech (and so occur in
a complement clause): cf.
(68) (a) When he asked whether we voted for Larry Loudmouth, I said ‘Yes’ and
you said ‘No’
(b) When he asked ‘Did you vote for Larry Loudmouth?’, I said ‘Yes’ and
you said ‘No’
Moreover, from a theoretical point of view, the null-operator analysis offers the advan¬
tage that it allows us to arrive at a unitary analysis of the syntax of questions as clauses
which contain an (overt or covert) interrogative operator as their specifier.
A natural question to ask at this juncture is how we deal with embedded //questions
such as that bracketed in (70) below:
295
Operator movement
If we suppose that it is a lexical property of if that it only licenses the null yes-no ques¬
tion operator Op as its specifier, we can account for the fact that if is never used in
wh-questions (as we see from the ungrammaticality of sentences like */ wonder how if
he’s feeling).
Empirical evidence in support of claiming that (/-questions contain a null interroga¬
tive operator comes from data relating to operator movement. If we replace the adjec¬
tive better in (71) by the corresponding interrogative operator how, we find that how
cannot be moved to the front of the overall sentence - as we see from the ungrammati¬
cality of:
(though note that it’s OK to ask How do you think that he’s feeling?). Why should it be
ungrammatical to extract a wh-operator out of an (/-clause? If we suppose that (/-ques¬
tions have an abstract yes-no question operator (a null counterpart of whether) in
spec-CP, we can give precisely the same answer as we earlier gave in relation to the
impossibility of extracting a wh-operator out of a whether-clause in sentences like
(58b) *How do you wonder whether he is feeling? That is, a successive cyclic deriva¬
tion like (73) below:
is ruled out by the fact that how can't move into the specifier position within the
embedded (righthand) CP because this position is already occupied by the null operator
Op Moreover, (single-step) long movement of how directly into the main-clause (left-
hand) CP as in (74) below:
is also ruled out, since it would lead to violation of the shortest movement principle
(which requires how to move into the spec-CP position immediately above it). As
should be apparent, a crucial assumption underpinning the argumentation here is that
the (/-clause contains a null operator-specifier which prevents successive cyclic move¬
ment.
On the analysis presented here, English has two interrogative complementizers (if
and 0). They differ in their specifier-features, in that the overt complementizer if
requires a null yes-no question operator as its specifier, and the covert complementizer
0 requires an overt interrogative specifier. They also differ in their selectional proper¬
ties, as we see from contrasts such as the following:
(77) (a) She was curious about [whether 0 I was wearing one]
(b) *She was curious about [Op if I was wearing one]
(78) (a) [Whether 0 007 was a double-agent] will remain a secret
(b) *[Op if 007 was a double-agent] will remain a secret
The bracketed embedded clauses in (77) occur as the complement of the transitive
preposition about, those in (78) occur as the subject of the finite auxiliary will. To put
matters rather differently: the bracketed clauses occupy an objective case position in
(77), and a nominative case position in (78). It would seem from the contrast between
the (a) and (b) examples in (77-8) that a CP headed by the null complementizer 0 can
occur in a (nominative or objective) case position, whereas a CP headed by if cannot.
To use the terminology of Stowell 1981, if is a case-resistant complementizer, whereas
0 is not. Since it is typically a property of D-constituents (e.g. pronouns like you) that
they carry case (and so occupy case positions), what this might suggest is that the null
complementizer 0 is actually a null determiner - and that the structure into which it
projects is a DP rather than a CP. However, this can’t be the whole story, since
wh-question clauses (headed by 0) can also occur (as can (/'-clauses) as the complement
of an intransitive adjective like sure: cf.
297
Operator movement
There’s clearly a lot more to be said about the different properties of the overt interrog¬
ative complementizer if and its covert counterpart 0, but we shan’t attempt to probe the
relevant issues any further here.
Sentences such as (80) differ from typical operator questions in a number of ways - for
example, they don’t show auxiliary inversion or wh-movement. Moreover (unlike a
typical auxiliary-inversion operator question such as (81a) below), they can’t contain
polarity items such as any - as we see from the ungrammatically of any in (81b) below:
If we assume that root yes-no questions with auxiliary inversion contain an abstract
yes-no question operator Op, (81a) will have the (simplified) derivation (82) below:
(82) CP
ordered D N
the/any drinks
But why should any be ungrammatical in (81b)? Since they don’t show auxiliary
inversion and don’t license polarity items, let’s suppose that such questions don’t con-
7.10 Summary
tain the interrogative operator Op (and thus are nonoperator questions), so that (81b)
is simply a TP with the (simplified) structure (83) below (cf. Grimshaw 1993a, 1993b):
(S3)
We could then say that any is ruled out by virtue of the fact that (83) is a nonoperator
question, and thus does not contain an interrogative operator c-commanding any.
However, what a TP analysis such as (83) fails to account for is the fact that sen¬
tences like (80) are questions, as is shown orthographically by the question-mark at the
end of the sentence, lexically by the use of the interrogative pronoun what in (80b) and
phonetically by their intonation. One way of marking them as questions is to suppose
that the head T constituent of TP carries an interrogative feature of some kind in the
relevant structures. An alternative possibility (in keeping with Chomsky’s assumption
that all questions are CP constituents headed by a question affix Q) would be to sup¬
pose that they are CPs headed by Q, so that (81b) would have the structure (84) below:
If we suppose that Q is only a strong affix when it has an (overt or covert) interrogative
operator-specifier (as we earlier assumed in relation to subject-question structures like
(62) above), it follows that Q in (84) will be weak (since (84) is a nonoperator ques¬
tion), and so will not trigger auxiliary inversion. Since (84) contains no interrogative
operator, we correctly predict that the polarity item any is ungrammatical.
7.10 Summary
In this chapter, we have looked at the syntax of interrogative operators.
In §7.2 we argued that wh-questions like What did you say? involve moving an inter¬
rogative wh-operator like what? into the specifier position within CP, and argued that
the moved wh-operator leaves behind a trace in the position it moves out of. In §7.3
we suggested that movement of operators into spec-CP is motivated by Lasnik’s princi¬
ple of enlightened self-interest, and that operator movement serves to check the
[wh] specifier-feature carried by the question affix Q which heads interrogative
clauses. We noted that such an analysis predicts that in multiple wh-questions, only one
299
Operator movement
of the wh-operator expressions will move to spec-CP, and that the shortest movement
principle determines that it is the nearest wh-operator expression which moves.
In §7.4 we argued that (in consequence of the economy principle) movement opera¬
tions affect the minimal constituent which will suffice to check the features triggering
their movement. We noted that in simple questions like What did he say? the wh-
operator what is moved on its own, but that movement of what alone in sentences like
* What did he give reason? would violate Chomsky’s chain uniformity principle, and
that in consequence the complement of what (i.e. the noun reason) has to be pied-piped
along with what, so deriving What reason did he give? We also noted that Huang’s
condition on extraction domains (which allows extraction only out of complements)
prevents what from being extracted out of a subject expression in sentences like
*Which country did you say the capital of was bombed during the war?, and requires
the whole subject constituent to be pied-piped along with which, so deriving The capi¬
tal of which country did you say was bombed during the war? In §7.5 we looked at the
syntax of embedded questions in Belfast English, noting that these allow the head
C position of CP to be filled either by an inverted auxiliary or by the complementizer
that, so deriving sentences like I asked which street did he live in and / asked
which street that he lived in. We also noted that inversion takes place in noninterroga-
tive clauses from which an interrogative operator has been extracted, e.g. in sentences
like Who did he hope would he see? We suggested that this could be accounted for
by positing that operator movement applies in a successive cyclic fashion (moving
an operator expression into the nearest spec-CP position above it on each of its move¬
ments), and by positing that a C which has an interrogative specifier is a strong (affixal)
head, and hence can trigger auxiliary inversion. In §7.6 we turned to look at Standard
English, and argued that C is always filled by a null interrogative complementizer 0
in Standard English complement-clause questions, with spec-CP being filled either by
merging the adverb whether with C-bar, or by moving an operator expression from
some position within TP into spec-CP. We argued that operator movement applies
in a successive cyclic fashion, and that this accounts for why wh-operators can’t be
extracted out of wh-questions (i.e. why sentences like *How do you wonder whether he
is feeling? are ungrammatical), since successive cyclic movement is blocked by
the fact that whether occupies spec-CP in the embedded clauses, and long (one-step)
movement violates the shortest movement principle. In §7.7 we looked at subject
questions like Who helped him? We noted Chomsky’s claim that such sentences
are CPs headed by the question-affix Q, and that the subject who is in spec-IP, with the
wh-feature of who being attracted to Q. We saw that Chomsky’s CP analysis would
enable us to provide a unitary characterization of questions as CPs with an interroga¬
tive Q head. However, we suggested an alternative characterization of questions
as clauses with an interrogative specifier, noting that this would allow us to posit that
subject questions like Who helped him? are TPs which are questions by virtue of the
300
Workbook section
fact that they have an interrogative specifier (= who)-, in questions such as What did
you say?, movement of what into spec-CP would be motivated by the need to generate
a structure with an interrogative specifier. In §7.8 we noted that the interrogative
specifier analysis would entail positing that yes-no questions like Is it raining? contain
an abstract yes-no question operator Op in spec-CP, and we saw that there were
significant parallels between the syntax of such questions and whether questions which
made it plausible to posit that they contain a null-operator counterpart of whether.
We went on to suggest that embedded yes-no questions headed by the complementizer
if also have a null yes-no question operator-specifier, and we argued that this assump¬
tion accounts for the fact that operators can’t be extracted out of (/-question clauses
(cf. *How do you wonder if he is feeling?), since successive cyclic movement is
blocked by the null operator and long movement is blocked by the shortest movement
principle. In §7.9 we argued that noninversion questions like You’re leaving? and
He said what? do not contain an interrogative operator, and hence are nonoperator
questions. We suggested that they could be analysed as TPs headed by an interrogative
T, or perhaps as CPs headed by a question particle Q which is weak by virtue of
lacking an interrogative operator in spec-CP.
Workbook section
Exercise I (§§7.2-7.4)
Discuss the syntax of the following sentences in Standard English:
301
Operator movement
: PRO T VP
; I / \
: to v d
: l I
'
l
do t
1 __)
The wh-operator what moves to spec-CP to check the [wh] specifier-feature carried by
Q (movement of what being motivated by enlightened self-interest); the auxiliary did
adjoins to Q because Q is a strong affix.
What remains to be accounted for in (i) is how to can cliticize onto want, ultimately
giving rise to wanna. This is not an entirely straightforward question to answer, since
the two are separated by the intervening empty category PRO, and we have seen on
numerous occasions that an intervening trace or empty INFL constituent blocks cliti-
cization. For the time being, let’s simply stipulate that an intervening PRO doesn’t
block cliticization but other empty categories do. (We shall provide a principled
answer to the question of why PRO doesn’t block cliticization in the next chapter.)
Exercise II (§7.5)
Discuss the syntax of the following sentences in (one variety of) Belfast
English (cf. Henry 1995):
Why do you think Alison Henry (1995) argues that whether functions as a complemen¬
tizer in Belfast English?
302
Workbook section
Given the assumptions made in the text, la will have the (simplified)
derivation (i) below:
303
Operator movement
upset you
The [wh] feature of Q can be checked by wh-attraction (i.e. by percolation of the [wh]
feature of who from D to C), since who is sufficiently close to Q to satisfy the locality
restriction on attraction (to the effect that a head can only attract features of the
head/specifier of its complement). Because it is more economical, wh-attraction is
preferred to wh-movement - hence there is no movement of who into spec-CP. If we
suppose that Q is only a strong affix when it has an operator-specifier in spec-CP, Q
will be weak here (because there is no operator-specifier in spec-CP), and so not trigger
Jo-support and auxiliary inversion.
On the other hand, if we adopt the alternative interrogative specifier analysis of
questions outlined in §7.7, sentence 1 will simply be a TP of the form (ii) below:
(ii) TP
Who T VP
I /
Past V D
upset you
304
Workbook section
The resulting structure will be interpretable as a question at LF by virtue of the fact that
it contains an interrogative operator-specifier (since who occupies the specifier position
within TP). Since T is a weak affix in English, the past-tense head-features of upset are
checked by attraction to T, not by movement.
Helpful hints
In 8, assume that the to clause has a PRO subject; in 9, ask yourself why
for can’t be used to head an interrogative CP (bearing in mind the text discussion of
wh+that structures), and why the case of us can’t be checked if for is not used.
Similarly, in 11 say why the case of what can’t be checked. Consider also whether
any of these examples might lead you to modify some of the assumptions made in the
text.
Exercise IV (§§7.6-7.7)
Clauses like those bracketed in the examples below are traditionally
referred to as relative clauses, because they contain a (bold-printed) relative pronoun
which relates to (i.e. has as its antecedent) an (italicized) expression outside the brack¬
eted relative clause:
Relative pronouns (like their interrogative counterparts) usually begin with wh (cf.
who/which/where/when). However, English also has two other types of relative clause.
The first are structures like (2) below which contain no wh-pronoun, but contain that:
The second are relative clauses which contain neither a relative pronoun nor that. cf.
305
Operator movement
null relative operator (a silent counterpart of wholwhich etc.) which originates internally
within the sentence and is then moved to the spec-CP position at the front of the brack¬
eted relative clause. Given these assumptions, discuss how the bracketed relative
clauses in the examples below might be derived, saying why they are (un)grammatical.
Finally, say what’s interesting about the bracketed Shakespearean relative clauses in 11
below, and how and why they differ from their contemporary counterparts:
11a I have an answer [will serve all men] (Clown, All’s Well That Ends Well,
Il.ii)
b There’s little [can be said in’t] (Parolles, All’s Well That Ends Well, Li)
c You are the man [must stead us all] (Tranio, Taming of the Shrew, I.ii)
d I am he [am bom to tame you] (Petruchio, Taming of the Shrew, II.i)
e Youth’s a stuff [will not endure] (Clown, Twelfth Night, Il.iii)
306
Workbook section
(0 CP
D'
Op C
A /
Comp
I
that
l_ _ j
If we suppose that the null relative pronoun-operator Op who carries the head-feature
[rel] (indicating its status as a relative operator) and that the null relative-clause head R
carries the specifier-feature [rel] (indicating that it requires a relative pronoun as its
specifier), we can say that movement of who into spec-CP is motivated by enlightened
self-interest - i.e. by the requirement to check the [rel] specifier-feature of R (which
cannot be checked by attraction, because Op is not the specifier of the TP complement
of R). Thus, movement of the null relative operator Op to spec-CP is used as a last
resort, as the only way of checking the [rel] specifier-feature of R. In relation to the
cliticization of to onto want (ultimately forming wanna), see the model answer for
exercise I.
Helpful hints
In 7a, consider whether the subject moves at all. In relation to the contrast
between 7d and the sentences in 11, consider whether (and if so, how and why) this
might be related to the fact that Early Modern English was a null subject language
whereas present-day English is not. Are there puzzling differences between finite rela¬
tive clauses like those in 4-7 and infinitival relative clauses like those in 8-10?
307
Operator movement
Exercise V (§§7.6-7.7)
Discuss the syntax of the following Early Modem English (EME) sen¬
tences taken from various plays by Shakespeare:
The TP in (i) then merges with the strong question affix Q, and the T constituent con¬
taining the verb sayst raises to adjoin to Q as in (ii) below:
II I / \
sayst Pres t V D
: !ii
1_l t what
308
Workbook section
V T Q thou T VP
: I l I / \
sayst Pres t V D
j I I
t t
1_ j
If we assume that T and Q are strong affixes in EME, movement of sayst from V to
T to C will be motivated by enlightened self-interest, driven by the desire to satisfy
the requirement for the strong affixes T and Q to be attached to a lexical item. Since
T is a strong affix, we might assume that it cannot be separated from sayst, so that
when sayst- moves from T to C as in (ii) above, the tense affix T is pied-piped along
with it.
The movement of what into spec-CP might similarly be said to be motivated by
enlightened self-interest, since it enables the [wh] specifier-feature carried by Q to be
checked (and erased) by the [wh] head-feature carried by what. It is not possible for the
relevant [wh] feature to be checked by attraction here, since what is too far away from
Q to satisfy the locality restriction on attraction (which specifies that a head can only
attract features from the head or specifier of its complement): hence, movement of
what to spec-CP is a last resort strategy.
However, an alternative account of why what moves into spec-CP would be to sup¬
pose that this is in order to ensure LF convergence - i.e. to ensure that the resulting
structure contains an interrogative operator-specifier, and hence can be interpreted as
an (operator) question at LF.
Helpful hints
Sentences like those in 6-7 are traditionally referred to as exclamatives: 8
might be called a topic sentence, in that the constituent seawater has been made into
the topic of the sentence by being moved to the front of the sentence; sentence 10 is an
imperative. Assume for the purposes of the exercise that all the sentences in 1-10 are
CP structures in which the initial (auxiliary/nonauxiliary) verb in the sentence is
adjoined to (an exclamative/topic/imperative affix in) C.
Exercise VI (§§.6-7.8)
Discuss the syntax of the following questions in Jamaican Creole (in sen¬
tences 1-3, the b sentence is intended to be synonymous with the corresponding a sen¬
tence):
309
Operator movement
I
gat D
/ \N
I I
eni fish
Since eni ‘any’ is a polarity item (and so must be c-commanded by an affective opera¬
tor), we might argue that eni here is licensed by the null yes-no question operator Op
(a null counterpart of English whether), since Op c-commands eni (by virtue of the fact
that the mother of Op is CP, and CP dominates eni). The fact that there is no auxiliary
310
Workbook section
6 I don’t even know whether if they’ll give me some credits for the courses
(but don’t try and analyse the internal structure of the italicized VP headed by give,
since this presupposes familiarity with assumptions about structure not introduced
until chapter 9).
Since the partitive/existential quantifier any (and its compounds like anyone/anything)
is a polarity item, it must be c-commanded by a (negative, interrogative or conditional)
operator. This condition is met in (i) in that the yes-no question operator Op c-commands
311
Operator movement
anyone and anything by virtue of the fact that the mother of Op is CP, and CP
dominates anyone!anything.
Exercise VIII
Discuss the syntax of the following topic structures in present-day
Standard English:
(i)
we T VP
/ \
cannot V DP
I I
tolerate t
l
A second (similar but subtly different) possibility is to suppose that the topicalized
expression moves into the specifier position within a TopP (= topic phrase) constituent
headed by a null Top particle, as in (ii) below:
312
Workbook section
A third possibility (argued for by Jane Grimshaw, 1993a, 1993b) is to suppose that top-
icalization involves adjunction of the topicalized constituent to TP, thereby expanding
TP into an extended TP constituent, as in (iii) below:
TP
This kind of Q
behaviour
+ we T^ ^ VP
I / \
I
I cannot V DP
I | |
1 1 1
l
I tolerate t
i
1
Exercise IX
Discuss the derivation of the following sentences, accounting for the (un)grammatical-
ity of each of the examples (note that % = ‘grammatical in some varieties of English,
but not in Standard English’):
1 He said nothing
2 Nothing did he say
3 Nobody said anything
313
Operator movement
Helpful hints
Expressions such as nobody and nothing incorporate the negative opera¬
tor no, and thus might be said to be negative operator expressions. Consider possible
similarities and differences between the syntax of negative and interrogative operators.
In considering the landing-site for preposed negative operators, bear in mind two pos¬
sibilities: firstly, that preposed negative operators move to spec-CP; and secondly, that
they move to the specifier position in a different kind of projection (e.g. a NEGP
headed by a strong NEG affix).
314
8
A movement
8.1 Overview
In this chapter, we take a close look at the syntax of subjects. So far, we
have assumed that subjects occupy the specifier position within TP and remain in situ
(except where the subject is an interrogative operator which undergoes operator move¬
ment, e.g. in sentences like Who did he say was coming?). However, in this chapter we
argue that subjects originate internally within VP, and subsequently move to spec-TP
for checking purposes (an assumption known as the VP-internal subject hypothesis).
We look at the syntax of so-called raising predicates like seem, and examine how (and
why) they differ from control predicates like try. In addition, we look at the syntax of
subjects in passive sentences. Finally, we look at the nature of the A movement opera¬
tion by which subjects are raised up (in a successive cyclic fashion) into the spec-TP
position which they occupy in the superficial syntactic structure of the sentence.
Sentence (la) contains two different occurrences of there. The second (bold-printed)
there is a locative pronoun paraphraseable as 'in that place’, and contains the diphthong
/eo/; the first (italicized) there is an expletive (i.e. dummy or pleonastic) constituent
which contains the unstressed vowel /o/ and does not have a locative interpretation (i.e.
it is not paraphraseable as ‘in that place’), but rather has no intrinsic reference (as we see
from the fact that its reference can’t be questioned - hence the ungrammatically of
*Where is nobody living there?). Expletive there seems to have the categorial status of a
pronominal determiner, since (like other pronominal determiners - e.g. personal pro¬
nouns such as he/she/it/they) it can occur in sentence tags, as we see from examples like
those in (2) below (where the part of the sentence following the comma is the tag):
315
A movement
It seems clear that the pronoun there in sentences such as (1) occupies the specifier
position within TP. Some evidence in support of this claim comes from the fact that the
auxiliary is/are can be moved in front of it (into COMP) in yes-no question structures
such as (3) below:
where is/are originates in the head T position of TP and moves across expletive there
(which occupies the specifier position in TP) into the head C position of CP. Given that
auxiliary inversion typically moves an auxiliary across a subject in spec-TP, the fact
that the auxiliary is moves across the expletive pronoun there in getting from its posi¬
tion in (1) into its position in (3) suggests that there must be in spec-TP.
Moreover, given that auxiliaries select a VP complement, it seems likely that
the complement of the auxiliary is/were in (1) is a verb phrase. If this is so, then (lb)
will be a TP with the simplified structure (4) below. (Here and throughout the rest
of this chapter, we simplify tree diagrams by not showing the internal structure of T -
e.g. we don’t show the abstract present-tense affix which is merges with in (4)
below.)
An analysis such as (4) would enable us to account for the fact that the verb knocking
surfaces in the +ing form, and not (for example) in the infinitive form knock: given the
structure in (4), we should expect that the head T constituent (= is) of TP will determine
the morphological form of the head V constituent (= knocking) of the complement VP:
since the progressive auxiliary be selects a complement headed by a verb in the +ing
form, the fact that knocking is in the +ing form can be accounted for straightforwardly.
316
8.2 VP-internal subject hypothesis
But what is the internal structure of the VP constituent in (4)? It seems clear that the
V knocking is the head of the VP, and that at the door is its complement. But what is
the role of the pronoun someone (which is sometimes referred to as the associate of
expletive there)! A natural suggestion to make is that someone is the subject of knock¬
ing at the door, if we assume that the canonical position for the subject of a particular
constituent is the specifier position within the relevant projection, we might suggest
that the expletive associate someone in (4) occupies spec-VP (i.e. the specifier position
within VP). If so, (4) will have the simplified structure (5) below:
someone V DP
! / \
knocking at the door
What (5) claims is that the head V constituent (= knocking) of the VP merges with its
PP complement (= at the door) to form a verbal expression which is conventionally
termed a V (= V' = V-bar, in each case pronounced vee-bar), and the resulting V-bar
constituent knocking at the door is then predicated of the pronominal quantifier/deter¬
miner someone, so forming the overall VP someone knocking at the door.
An interesting variant of a sentence like (lb) is the type of sentence illustrated in (6)
below:
Here, someone is clearly the subject of the auxiliary is, and hence would appear to be
positioned in spec-TP. And yet, in our earlier structure (5), someone was instead the
subject of knocking, and occupied the specifier position in VP. The obvious question to
ask is how we can account for the fact that subjects like someone are in spec-VP in
expletive structures like (5), but in spec-TP in nonexpletive structures like (6). In this
connection, recall our suggestion in chapters 6 and 7 that constituents can move from a
lower to a higher position within the sentence containing them (e.g. auxiliaries can
move from T to C, and operator expressions can move e.g. from VP-complement posi¬
tion to CP-specifier position). This opens up the possibility of a movement account of
the dual position of someone in sentences like (lb) and (6). More specifically, let us
suppose that someone originates in spec-VP and remains in situ in expletive structures
such as (5), but is raised into spec-TP in nonexpletive structures such as (6). If this is
so (and if we assume that a moved subject leaves behind a trace in the position out of
which it moves), then (6) will be derived as in (7) below:
317
A movement
Since the movement operation which applies in (7) has the effect of raising someone
from being the subject of VP to becoming the subject of TP, it is traditionally referred
to as subject-to-subject raising (usually abbreviated to subject raising, or more
simply raising). The more general claim underlying the analysis in (7) (known as the
VP-internal subject hypothesis) is that subjects originate in spec-VP, and are subse¬
quently raised into spec-TP in all but a few constructions - e.g. not in expletive
structures like (5) above in which the spec-TP position is filled by a dummy or expletive
constituent. Substantial empirical evidence in support of the hypothesis from a variety
of languages is presented in papers by Kitagawa 1986, Speas 1986, Contreras 1987,
Zagona 1987, Kuroda 1988, Sportiche 1988, Rosen 1990, Ernst 1991, Koopman and
Sportiche 1991, Woolford 1991, Burton and Grimshaw 1992, Guilfoyle, Hung and
Travis 1992, McNally 1992, and Huang 1993; in the next few sections, we review a
tiny fragment of the relevant evidence.
318
8.3 Evidence
(9) TP
I) I
We T VP
PRO T VP
I / \
to help you
But then we should expect that the intervening empty category PRO would prevent to
from cliticizing onto want, thereby wrongly predicting that (8b) is ungrammatical.
What’s gone wrong here? Baltin (1995, p. 244) suggests that the VP-intemal subject
hypothesis provides us with an answer. If we assume that PRO subjects originate (and
remain) in spec-VP, then (8a) will no longer have the structure (9) above, but rather
that in (10) below:
PRO V D
help you
Since there is no (overt or covert) category intervening between want and to in (10), we
correctly predict that to can cliticize onto want, forming (8b) We wanna help you. But
note that a crucial premise in the argument is that PRO is positioned in spec-VP, not in
spec-IP. (Of course, Baltin’s analysis raises questions about how the null case carried
by PRO is checked: if it is checked by to rather than by help, we are going to have to
say that the case of PRO is checked from outside its containing VP by infinitival to, in
much the same way as, in exceptional case-marking structures like We expect him to
resign, the case of him is checked from outside its containing phrase by the verb
expect. We return to this issue in §8.5 below.)
319
A movement
The core assumption of the VP-internal subject hypothesis is that subjects origi¬
nate in spec-VP: in control structures like (10) and in expletive structures like (5), the
subject remains in spec-VP; but in other structures, the subject raises from spec-VP
into spec-TP. Given that a moved constituent leaves behind a trace in any position out
of which it moves, it follows that subjects which move from spec-VP to spec-TP will
leave behind a trace in the spec-VP position out of which they move. Empirical moti¬
vation for positing that moved subjects leave behind traces in spec-VP comes from evi¬
dence which is essentially similar in character to that which we used to support the
postulation of a PRO subject in control structures in chapter 4. Consider, for example,
how we account for the syntax of the italicized reflexive pronouns contained within the
bracketed verb phrases in the sentences below:
Reflexives generally require a local c-commanding antecedent within the phrase con¬
taining them. This being so, it follows that the reflexives in (11) must have an
antecedent within the bracketed verb phrase containing them. This will obviously be
the case if we assume that a sentence such as (1 la) is derived in the manner indicated
informally in (12) below:
The derivation in (12) claims that the subject he originates in spec-VP as the subject of
compromised, and is then raised into spec-TP, where it becomes the subject of has -
leaving a trace t behind in the spec-VP position which it vacates. The trace t of the moved
subject he provides an appropriate phrase-internal antecedent for himself in (12), since
both himself and the trace are immediate constituents of the bracketed VP, and since both
are third person masculine singular (the trace carrying the same grammatical properties
as its antecedent he). By contrast, themselves could not be used in place of himself in (12)
because it would lack an antecedent within the bracketed VP (the trace t cannot be its
antecedent since the trace is third person masculine singular, and themselves is third
person plural). If we suppose (following Chomsky 1995b) that a trace is a silent copy
of the relevant moved constituent, it follows that traces will have the same syntactic
and semantic properties as their antecedents, and will differ from their antecedents only
in that they have no overt phonetic form: this assumption accounts for the fact that the
trace t in (12) has the same third person masculine singular features as its antecedent he.
We can derive the structure associated with (lib) if we prepose the VP complement
following has in (12), and move it to the relevant position in front of the overall TP -
as in (13) below (where_marks the position out of which the bracketed preposed
verb phrase moves, and t is the trace of he):
320
8.3 Evidence
Since the subject of the preposed bracketed VP is the trace t of the moved (third person
masculine singular) subject he, the subject trace inside the bracketed VP can only bind
(i.e. serve as the antecedent of) the third person masculine singular reflexive himself,
hence, replacing himself by the third person plural reflexive themselves in (13) leads to
ungrammatically.
We can construct essentially parallel arguments in support of the claim that
apparently subjectless VPs contain a trace subject in spec-VP in relation to structures
such as the following (where t denotes a trace of the italicized moved subject, and
where_denotes the position out of which the preposed verb phrase moves):
If we posit a trace subject occupying the specifier position within the bracketed VP, we
can account for the fact that the predicate nominal in (14) has to be in the plural form
millionaires, since it agrees with the trace subject t in spec-VP (which is plural because
it is the trace of the moved plural subject they). In much the same way, we can posit
that his in (15) agrees with the trace of John, and that your in (16) agrees with the trace
of you. Similarly, we can claim that each other (which requires a local plural
antecedent) is bound by the trace of we in (17), and that together (which similarly
requires a local plural antecedent) is bound by the trace of they in (18). The logic of the
argumentation should be clear: in each case we have an expression which requires a
local antecedent, and which will only have a local antecedent if we assume that
subjects originate in spec-VP and thence raise to spec-TP, leaving behind a trace in
spec-VP which can bind the expression requiring a local antecedent.
A rather different kind of argument in support of the VP-intemal subject hypothesis
can be formulated in relation to the syntax of quantifiers (cf. Sportiche 1988). In
sentences such as (19) below:
321
A movement
the italicized quantifiers both/all/each are separated from the bold-printed subjects
theylwelyou which they quantify. In this use, they are referred to as floating quantifiers
(or stranded quantifiers), for obvious reasons. How can we account for the fact that
(for example) in (19a) both quantifies they, and yet the two clearly occupy different
positions? The VP-intemal subject hypothesis provides us with an answer. Let us
suppose that the pronoun they in (19a) originates as the complement of both (in much
the same way as of them seems to function as the complement of both in an expression
such as both of them). Let’s also assume that the relevant QP (quantifier phrase) both
they (which has essentially the same interpretation as both of them) originates as the
subject of the VP headed by helping, and that the pronoun they is subsequently raised
up to become the subject of are, as in (20) below:
(20)
D'
I
They
*
The quantifier both thereby ends up modifying the trace of the moved pronoun they.
Movement of they leaves the quantifier both stranded within QP, separated from the
pronoun they which it quantifies. The analysis in (20) correctly specifies that both
superficially occupies an intermediate position between are and helping (the QP both t
occupying spec-VP and so serving as the subject of helping).
Thus the assumption that subjects originate in spec-VP and raise to spec-TP
provides us with an interesting account of how quantifiers come to be separated from
the subject expressions which they quantify. A parallel separation argument can be for¬
mulated in relation to the syntax of idioms. We can define idioms as expressions (such
as those italicized below) which have an idiosyncratic meaning that is not a purely
componential function of the meaning of their individual parts:
There seems to be a constraint that only a string of words which forms a unitary con¬
stituent can be an idiom. Thus, while we find idioms like those in (21) which are of the
form verb + complement (but where the subject isn’t part of the idiom), we don’t find
322
8.3 Evidence
idioms of the form subject + verb where the verb has a complement which isn’t part of
the idiom: this is because in subject + verb + complement structures, the verb and its
complement form a unitary constituent (a V-bar), whereas the subject and the verb do
not (and only unitary constituents can be idioms).
In the light of the constraint that an idiom is a string of words which forms a unitary
constituent with an idiosyncratic interpretation, consider now more restrictive idioms
such as the following:
In idioms like those in (22), not only is the choice of verb and complement fixed, but so
too is the choice of subject. In such idioms, we can’t replace the subject, verb or com¬
plement by near synonyms, so that sentences like (23) below are ungrammatical (on
the intended idiomatic interpretation):
Hence, since the choice of all three constituents (subject, verb and complement) in
clauses like (22) is fixed, we might refer to such idioms as clausal idioms.
However, what is puzzling about clausal idioms like those in (22) is that auxiliaries
can freely be inserted between the subject and verb: cf.
If (as suggested earlier) only a string of words which form a unitary constituent can
constitute an idiom, how can we account for the fact that (for example) the idiom all
hell. . . break loose is not a unitary constituent in any of the sentences in (24), since the
subject all hell and the predicate break loose are separated by the auxiliaries
will/has/could havel To put the question another way: how can we account for the fact
that although the choice of subject, verb and complement is fixed, the choice of auxil¬
iary is not?
323
A movement
We can then say that (in the relevant idiom) break loose must be predicated of all hell,
and that this condition will be met only if all hell originates in spec-VP as the subject
of break loose. We can account for how the subject all hell comes to be separated from
its predicate break loose by positing that subjects raise from spec-VP to spec-TP across
an intervening auxiliary like will, so that the subject and predicate thereby come to be
separated from each other.
324
8.4 Argument structure and theta-marking
the italicized verbs are predicates, and the bracketed expressions represent their argu¬
ments: each of the verbs in (27) has a single argument, and so is said to function as a
one-place predicate (in the use illustrated); each of the verbs in (28) has two
arguments, and so is said to function as a two-place predicate. Using more familiar
grammatical terminology, we can say that the arguments of a verb are its subject and
complement. Since (according to the analysis we have assumed so far) the comple¬
ments of verbs are positioned inside V-bar whereas their subjects are positioned out¬
side V-bar (they originate in spec-VP and typically raise to spec-TP), complements are
said to be internal arguments and subjects external arguments. Thus, in a sentence
such as (28a) The police arrested the suspects, the DP the suspects is the complement
and hence the internal argument of the predicate arrested, and the DP the police is the
subject and hence the external argument of arrested. Using this terminology, we might
say that the argument structure of the predicate arrest specifies that it is a two-place
predicate which takes a DP as its internal argument and another DP as its external
argument.
Not all of the expressions which are associated with a verb function as arguments of
the verb; for example, in sentences like those in (29) below:
it is clear that the bracketed DPs the police and the suspects are participants in the act
of arrest, and so are arguments of the verb arrested. However, there’s no sense in
which we can say that the bold-printed prepositional phrases represent participants in
the arrest. On the contrary, they simply serve to provide additional information about
the event; thus, on Saturday specifies the day on which the arrest took place, in Beverly
Hills specifies the place in which it took place and with minimum use of force specifies
the manner in which the arrest was effected (e.g. the suspects were forced to lie face
down in the mud and kicked into submission). Expressions such as those bold-printed
in (29) are adjuncts.
However, simply saying that a verb like arrest takes two DP arguments (one as its
subject, the other as its complement) fails to account for the fact that the two arguments
play very different semantic roles in relation to the act of arrest - i.e. it fails to account
for the fact that the policeman is the person who performs the act (and hence gets to
verbally and physically abuse the suspect), and that the suspect is the person who suf¬
fers the consequences of the act (viz. being handcuffed, thrown into the back of a win-
325
A movement
dowless vehicle and beaten up). Hence, any adequate account of argument structure
should provide a proper description of the semantic role which each argument plays
with respect to its predicate.
In research over the past three decades - beginning with the pioneering work of
Gruber (1965), Fillmore (1968) and Jackendoff (1972) - linguists have attempted to
devise a universal typology of the semantic roles played by arguments in relation to
their predicates. In (30) below are listed some of the terms traditionally used to describe
a range of different roles, and for each such role an informal gloss is given, together
with an illustrative example (in which the italicized expression has the semantic role
specified):
(30) theme (or patient) = entity undergoing the effect of some action
(Mary fell over)
agent/causer = instigator of some action
(John killed Harry)
experiencer = entity experiencing some psychological state
(John felt happy)
recipient/possessor = entity receiving/possessing some entity
(John got Mary a present)
goal = entity towards which something moves
(John went home)
We can illustrate how the terminology in (30) might be used to describe the semantic
role fulfilled by arguments in terms of the following examples:
Given that - as we see from these examples - the theme role is a central one, it has
become customary over the past two decades to refer to the relevant semantic roles as
thematic roles; and since the Greek letter 0 (= theta) corresponds to th in English and
the word thematic begins with th, it has also become standard practice to abbreviate the
expression thematic role to 0-role (pronounced theeta role by some and thayta-role
by others). Using this terminology, we can say (for example) that in (31a) the FBI is
the agent argument of the predicate arrested, and that Larry Luckless is the
theme/patient argument of arrested.
326
8.4 Argument structure and theta-marking
The thematic role played by a given argument in relation to its predicate determines
the range of expressions which can fulfil the relevant argument function, as we see
from examples such as (32) below (from Lakoff 1971, p. 332: ? and ! represent
increasing degrees of pragmatic anomaly):
The nature of the relevant restrictions depends on the semantic properties of the predicate
on the one hand and on the semantic (= thematic) role played by the argument on the
other. As sentences such as (32) illustrate, the experiencer argument (i.e. subject) of a
cognitive predicate like realize has to be an expression denoting a rational entity (i.e. an
entity capable of rational thought - hence e.g. not an expression denoting a politician).
A central theoretical question which arises is how theta-roles are assigned to argu¬
ments. It seems clear that in V-bar constituents of the form verb + complement, the
thematic role of the complement is determined by the semantic properties of the verb.
As examples like (31 a—d) illustrate, the canonical (i.e. typical) 9-role associated with
DP complements is that of theme. However, the question of how subjects are assigned
theta-roles is more complex. Marantz (1984, pp. 23ff.) and Chomsky (1986a,
pp. 59-60) argue that although verbs directly assign theta-roles to their internal argu¬
ments (i.e. complements), it is not the verb but rather the whole V-bar constituent (i.e.
verb + complement string) which determines the theta-role assigned to its external
argument (i.e. subject). The evidence they adduce in support of this conclusion comes
from sentences such as:
Although the subject of the verb threw in both (33a) and (33b), John plays a different
thematic role in the two sentences - that of agent in the case of threw a balk but that of
experiencer in threw a fit. Likewise, although the subject of the verb broke in both (34a)
and (34b), John plays the role of agent in (34a) but that of experiencer on the most
natural (accidental) interpretation of (34b) (though if he deliberately broke his own arm to
get out of taking a syntax exam, John would have the role of agent). From examples such
as these, Marantz and Chomsky conclude that the thematic role of the subject is not
determined by the verb alone, but rather is compositionally determined by the whole
verb+complement structure - i.e. by V-bar.
327
A movement
In a nutshell, what is being claimed is that a verb assigns a theta-role directly to its
internal argument (i.e. complement), but only indirectly (= compositionally, i.e. as a
compositional function of the semantic properties of the overall V-bar) to its external
argument (= subject). To use the relevant technical terminology, we might say that
verbs directly 0-mark their complements, but indirectly 0-mark their subjects.
A related observation here is that auxiliaries seem to play no part in determining the
assignment of theta-roles to subjects. For example, in sentences such as:
the thematic role of the subject he is determined purely by the choice of V-bar con¬
stituent (i.e. whether it is throw the ball or throw a fit), and is not affected in any way
by the choice of auxiliary. Clearly, any explanatory theory of 0-marking (i.e. 0-role
assignment) should offer us a principled account of how thematic roles are assigned,
and why some constituents (e.g. auxiliaries) play no part in this process.
One way of resolving the various puzzles surrounding 0-marking would be along
the following lines. Since auxiliaries are functional categories and play no role in theta¬
marking, let us assume that theta-roles are assigned only by lexical categories (i.e.
contentive categories), not by functional categories. More specifically, let’s assume
that 0-roles are assigned to arguments via the process of merger with a lexical cate¬
gory. Given these assumptions, a sentence such as (35a) He will throw the ball will be
derived as follows.
The verb throw will be merged with the DP the ball (itself formed by merging the
with ball) to form the V-bar throw the ball. As a corollary of this merger operation,
the DP the ball is assigned the 0-role of theme argument of throw. The V-bar throw the
ball is in turn merged with pronominal determiner he; as a corollary of the merger
operation, the subject he is assigned the role of agent argument of throw the ball.
Thus, the relevant VP will have the simplified structure (36) below (where arrows
indicate the assignment of thematic roles):
(36) VP^^
D AGENT-^ V
I
he THEME- --> DP
I / \
throw the ball
328
8.5 Case-checking
(37)
D
I
He
A
329
A movement
indeed be. In this connection, consider how the case of the italicized subject pronouns is
checked in auxiliary contraction structures such as those below:
Auxiliaries such as have, had, would, will and am have two unstressed forms: one
comprises the reduced vowel schwa and a consonant, so that have reduces to /ov/,
would!had to /ad/, will to fall and am to /am/ (I shall refer to these as reduced forms);
the other is a nonvocalic form consisting of a single consonant segment, whereby have
contracts down to /v/, would/had to IdJ, will to f\J and am to lm! (I shall refer to these as
contracted forms, since they involve complete loss of the vowel nucleus). Contracted
forms (i.e. vowel-less forms which comprise only a single consonant) must attach to an
immediately preceding word ending in a vocalic segment (i.e. a vowel or diphthong),
for phonotactic reasons relating to syllable structure. Each of the italicized auxiliaries
in (38) is a nonsyllabic contracted form, comprising only the single consonants /v, d, 1,
ml. (Note that for this and other reasons, the’s variant of has and is - which can imme¬
diately follow a word ending in a consonant as in Tom’s unhappy - are excluded from
our discussion of contracted auxiliary forms here; they are not subject to the same
constraints on their use as the contracted auxiliary forms discussed in relation to (40)
and (41) below.)
Returning now to our earlier discussion of case, consider how the case of the itali¬
cized nominative pronouns is checked in auxiliary contraction structures such as (38).
The conventional answer is that case is checked by raising the subject pronoun to
spec-TP, and that (for phonological reasons) the contracted auxiliary subsequently
attaches to the pronoun in the PF component. However, an alternative (more economi¬
cal) analysis would be the following. Let us suppose that contracted auxiliaries are
affixal heads which are only legitimate PF objects if another head attaches to them. Let
us further suppose that in structures such as (38), the italicized subject pronouns (by
virtue of being heads) check their nominative case by adjoining directly to T. In con¬
crete terms, this would mean that (38a) would be derived by raising the pronoun you
out of spec-VP to adjoin directly to the auxiliary 've in T as in (39) below (simplified
by ignoring the possibility that’ve is itself adjoined to a tense affix):
330
8.5 Case-checking
upset her
As we see from (40a-d), the contracted form /v/ cannot attach to a preceding auxiliary,
infinitive particle, verb or adverb ending in a vowel or diphthong; nor can it attach to
the pronouns me/we in (40e-f). If we assume that auxiliary contraction is only possible
as part of a checking via head-adjunction process which (in structures like (39) above)
is designed to check the case-feature of you and the agreement-features of have, the
data in (40) are relatively straightforward to account for: in (40a-d) have contracts
onto a non-nominal constituent which cannot in principle check the case/agreement
properties of have: and in (40e-f), have contracts onto a pronoun (me/we) which has its
own case/agreement properties, but which is not checked by have (the case of me is
checked by let, that of we by should).
The checking analysis of auxiliary contraction also provides us with a principled
account of the ungrammaticality of contraction in sentences such as the following:
331
A movement
(Note that the asterisk here indicates that the contracted forms /v/ and /d/ are ungram¬
matical; the weak forms hvl and /ad/ are grammatical, but this is obviously of no
relevance to our discussion of the syntax of contraction.) If auxiliary contraction
involves checking via head-adjunction, it is clear why it is blocked in (41): structures
like John and you and the Masai are clearly phrases (i.e. full DPs) and not heads, and
hence cannot undergo adjunction to T, but rather must move to a phrasal position in
spec-TP (thereby blocking contraction, which is only possible in structures where one
head is adjoined to another for checking purposes). Similarly, if (as in §4.7) we argue
that an expression like Di is a DP headed by a null determiner, it follows that Di is also
phrasal and so cannot be adjoined to T (with the result that Di cannot serve as the
subject of the contracted form’d of would).
The analysis of auxiliary contraction as a head-adjunction checking operation might
be extended to cover contrasts such as the following:
Have can be reduced to /ov/, but not contracted to N! in (42b).) A possibility which we
did not consider in the previous chapter is that rather than being moved to spec-CP,
wh-pronouns like who in structures like (42a) (where C contains a contracted auxil¬
iary) actually adjoin to the C node which contains the question particle Q, as shown
informally in (43) below (where the middle trace is the trace of they):
(43)
've Q they T
A
t chosen t
l I
Adjunction of who to the C node containing Q here is driven by the need to check the
[wh] feature of Q. The wh-adjunction analysis in (43) would predict that inverted have
can only contract down to hi when it has a wh-head ending in a vowel or diphthong
adjoined to it for checking purposes. This means that contraction will be blocked where
the wh-constituent is a full DP such as which one of you? (since phrases cannot adjoin to
heads); and this is indeed the case, as we see from the ungrammaticality of (42b) above.
If the analysis suggested above is along the right lines, it follows that the case prop¬
erties of subjects can be checked either via a spec-head relation or by a head-head
(adjunction) relation. But are there also structures in which the case properties of
subjects are checked by attraction (i.e. adjunction of a feature to a higher head)? There
are three types of subject for which this is a plausible assumption. One is the null-case
332
8.5 Case-checking
subject PRO. As we noted in §8.3 above, if we follow Baltin (1995, p. 244) in positing
that PRO remains in spec-VP in sentences such as:
(so that PRO does not block cliticization of to onto want in I wanna be free), we have to
assume that PRO does not check its case by raising to spec-TP, but rather by attraction
(i.e. by percolation of the relevant null case-feature from the PRO subject in spec-VPto
the infinitive particle to which heads TP). A natural inference would be that the null case
earned by PRO is associated with a weak case-feature (which is therefore checked by
attraction), whereas (for example) nominative case is associated with a strong case-
feature which has to be checked by movement of a nominative DP to spec-TP (or by
adjunction of a nominative pronoun to an affixal T).
A second type of construction in which the case (and indeed agreement) properties
of subjects would seem to be checked by attraction are existential structures with
expletive there subjects. In this connection, consider contrasts such as:
Such examples would seem to suggest that when the subject of waiting for you is an
indefinite nominal like students, the subject remains in spec-VP as the subject of wait¬
ing, and the expletive pronoun there is used to fill spec-TP; but when the subject is a
definite nominal like the students, it cannot remain in spec-VP but rather must raise
into spec-TP. How can we account for this definiteness effectl
One possibility (adapting ideas suggested by Belletti 1988 and Lasnik 1992, 1995) is
that the indefinite nominal students carries a different case from the definite nominal the
students. Let’s assume that the students carries nominative case, and so must raise to
spec-TP to check its strong nominative case-feature (hence the grammaticality of (46a)
and the ungrammatically of (46b)). But let’s also assume that the indefinite nominal stu¬
dents in (45) carries partitive case (as indefinite nominals do in languages like Finnish
which have a morphologically distinct partitive case), and that partitive case is a weak
case which is checked by attraction (so that the relevant partitive case-feature percolates
up from the indefinite nominal students in spec-VP to the auxiliary are). If the agree¬
ment-features of students also percolate up to T in the same way, we can then account for
the fact that are in (45a) agrees with the associate of expletive there - i.e. with students.
Moreover, if we assume (following Groat 1995) that expletive there is a defective pro¬
noun which carries case-features but not (number/person/gender) agreement-features, we
can say that there in (45a) carries nominative case and is inserted in spec-TP in order to
satisfy the strong nominative-case specifier-feature carried by a finite T-node. Finally, if
333
A movement
only a handful of verbs like be can check partitive case (as argued by Lasnik 1992 and
Boskovic 1995), we can account for the ungrammaticality of sentences such as:
Sentences like (47) would then be ungrammatical because be is the only auxiliary
which can check partitive case. (An alternative account modelled on Chomsky 1995b
is that the expletive associate students carries a weak nominative case-feature which is
checked by attraction, and that there is inserted in spec-TP to satisfy the strong [D]
specifier-feature of T - i.e. to satisfy the need for T to have a D(P) specifier; this in turn
may be a consequence of the predication principle posited by Rothstein 1995, requir¬
ing syntactic predicates like V-bar and T-bar to have syntactic subjects.) Numerous
questions of detail and potential problems arise, but we shall not attempt to delve into
these here. (For a technical discussion of expletive structures, see Authier 1991, Lasnik
1992, 1995, Chomsky 1995b, Groat 1995, Rothstein 1995 and Vikner 1995.)
A third type of construction in which the case properties of subjects may be checked
by attraction are ECM structures such as:
Since we argued in §5.8 that the objective case carried by the subject (him) of the
[bracketed] infinitive complement in such structures is checked by the immediately
preceding transitive head (expect/for), it would seem reasonable to suppose that here
too the case of the infinitive subject is checked by attraction (the objective case head-
feature of him being attracted to the immediately preceding transitive head expect/for).
334
8.6 Raising predicates
The bold-printed verbs in these examples have a that-clmse complement in the (a) exam¬
ples, and an infinitive complement in the (b) examples. But what is puzzling about
sentences like (49-52) is that the italicized expression which functions as the subject of
the bracketed complement clause in the (a) examples surfaces as the subject of the matrix
clause (i.e. the clause containing the complement clause) in the (b) examples: for exam¬
ple, he is the subject of understands in (49a), but the subject of seems in (49b). Moreover,
the bracketed infinitive complements in the (b) examples appear to have no subject.
So, sentences like (49-52) raise two related questions: how does the complement-
clause subject in the (a) examples come to be the matrix-clause subject in the (b) exam¬
ples, and how does the complement clause in the (b) examples end up seemingly
subjectless? A unitary answer to both questions is to suppose that the italicized nominal
originates as the subject of the complement clause and is then raised up to become the
matrix-clause subject by application of (subject-to-subject) raising, leaving behind an
empty category trace as the subject of the complement clause. In other words, he in a
sentence like (49b) originates as the subject of understands, but is subsequently raised
up to become the subject of seems. If this is so, an obvious question to ask is how the
subject comes to be raised.
In this connection, consider the following set of sentences:
Given our claim in §8.3 that floating quantifiers modify the traces of moved subjects,
the fact that the quantifier phrase headed by all in (53a-c) can serve as the subject of
understand the situation, or to understand the situation, or seem to understand the situ¬
ation suggests that the subject DP the men is raised in a successive cyclic fashion. In
the italicized subject DP the men originates as the subject of understand, then becomes the
subject of to, then becomes the subject of seem, and finally becomes the subject of do. In
case that was too fast for you, let’s look at a slow-motion replay of what’s going on here.
The verb understand merges with its DP complement the situation to form the V-bar
understand the situation', this in turn merges with the DP the men to form the VP (55)
below:
(55)
335
A movement
The VP (55) then merges with the infinitive particle to to form a T-bar; the DP the men
is then raised to become the subject of this T-bar, as in (56) below:
(56)
The resulting TP (56) in turn merges with the verb seem to form a V-bar; the DP the
men raises to become the subject of this V-bar, as in (57) below:
(57)
The VP thereby formed then merges with the auxiliary do to form a T-bar; the DP the
men then raises to become the subject of this T-bar, as in (58) below:
The men T VP
and (58) is the structure associated with (54) The men do seem to understand the situa¬
tion. What (58) claims is that the DP the men originates as the subject of understand
the situation, then raises in a successive cyclic fashion to become first the subject of to
understand the situation, then the subject of seem to understand the situation, and
336
8.7 Raising and control predicates
finally the subject of do seem to understand the situation. In each case, the movement
operation which applies is (subject-to-subject) raising; assuming that each separate
application of raising leaves behind a separate trace, there will be three (italicized)
traces of the moved subject the men in (58) - one serving as the subject of understand
the situation, another as the subject of to understand the situation, and the third as the
subject of seem to understand the situation. If we further assume (as in §8.3) that float¬
ing quantifiers modify traces of moved subjects, we can account for sentences like
those in (53) by supposing (in relation to (58) above) that all modifies the rightmost
trace of the men in (53c) The men do seem to all understand the situation, the middle
trace in (53b) The men do seem all to understand the situation, and the leftmost trace in
(53a) The men do all seem to understand the situation. (What remains to be accounted
for is why (53a) is better than (53b), and (53b) better than (53c). Perhaps floating
quantifiers become more awkward the further away they are from the expression they
quantify ; perhaps they are only fully grammatical as the specifier of the complement of
a finite T; and perhaps the awkwardness of (53c) is in part attributable to violation of
the traditional prohibition against split-infinitive structures in which to is separated
from its dependent verb. However, we shall not pursue these possibilities any further
here.)
where PRO has a controller (= she) in the will try clause. An important question to ask,
therefore, is how we can tell whether a given verb which selects an apparently subject¬
less infinitive complement is a raising predicate or a control predicate, and how we can
explain the differences between the two classes of predicate.
One difference between the two is that raising verbs like seem allow a dummy sub¬
ject like expletive there, whereas control verbs like try do not: cf.
337
A movement
subject of to, then the subject of seem and finally the subject of the abstract T constituent
preceding seem. More puzzling is the question of why a control verb like try doesn’t
allow expletive there as its subject — e.g. why we can’t have a structure such as (61)
below in which there is the controller of PRO:
One reason is that PRO is a referential pronoun which takes its reference from its
controller; hence, the controller of PRO must be a referential expression. But since
expletive there is a nonreferential pronoun (as we see from the fact that we can’t
question its reference - cf. * Where is someone living there?), it cannot serve as the
controller of PRO, and hence cannot serve as the subject of a control predicate like try
which selects an infinitive complement with a PRO subject. From this, it follows that
the subject of a control predicate like try must always be a referential expression.
A second reason why structures like (61) are ill formed relates to thematic consider¬
ations. It is a thematic property of the verb try that it assigns the 9-role agent to its
subject: hence, it requires as its subject an expression denoting a rational being. Since
there is a nonreferential dummy pronoun, the requirement for try to have an agent
subject is clearly not met in sentences like (60-1) where try has expletive there as its
subject. Conversely, the fact that seem allows expletive there as its subject in sentences
like (60) suggests that it does not 0-mark its subject (e.g. it does not require an agent
or experiencer subject), and hence allows a nonreferential subject like expletive there.
Our assumption that try 9-marks its subject whereas seem does not accounts for a
further difference between the two. By virtue of requiring an agent subject, the verb
try can only have as its subject an expression denoting an entity capable of rational
thought, as we see from examples such as the following (where ? and ! indicate
increasing degrees of anomaly):
By contrast, the verb seem (by virtue of the fact that it doesn’t 0-mark its subject)
imposes no such restrictions on its choice of subject, as we can see by comparing (62)
above with (63) below:
338
8.7 Raising and control predicates
We noted in relation to the examples in (32) above that verbs which 0-mark their sub¬
jects impose restrictions on their choice of subject: hence, the fact that try constrains its
choice of subject in (62) whereas seem imposes no restrictions on its choice of subject
in (63) suggests that try 0-marks its subject whereas seem does not.
What our discussion here suggests is that the essential difference between control and
raising predicates is that control predicates (like try) 0-mark their subjects, whereas raising
predicates (like seem) do not. This core thematic difference between the two can be argued
to determine why seem can function as a raising predicate but not as a control predicate,
and conversely why try can function as a control predicate but not as a raising predicate.
A crucial premise of the argumentation is that the way in which 0-roles are assigned is
constrained by the following UG principle (called the 0-criterion because it is a criterion
which an LF representation must meet in order to be well formed):
(64) ©-CRITERION
Each argument bears one and only one 0-role, and each 0-role is
assigned to one and only one argument. (Chomsky 1981, p. 36)
We can illustrate how this constraint works in terms of contrasts such as the following:
In (65a), Percy Peabrain is the experiencer argument of admire and himself the theme
argument; thus, each separate 0-role is associated with a separate argument expression
(and conversely), so the 0-criterion is satisfied. But what goes wrong in (65b)? One pos¬
sibility is that Percy Peabrain is assigned the role of experiencer argument of admire,
but that the theme role associated with admire is unassigned: however, this would violate
the 0-criterion requirement that each 0-role associated with a predicate must be
assigned to some argument. An alternative possibility is that both the experiencer and
theme roles of admire are assigned to Percy Peabrain, thereby wrongly predicting that
(65b) can have the same interpretation as (65a): however, this is ruled out by the
0-criterion requirement that each argument can carry only a single 0-role.
Consider now the role of the 0-criterion in determining that verbs (like try) which
0-mark their subjects can serve as control (but not raising) predicates, whereas con¬
versely verbs (like seem) which don’t 0-mark their subjects can serve as raising (but
not control) predicates. More concretely, consider how we derive sentence pairs such
as the following:
If seem is a raising predicate, he will originate as the subject of enjoy syntax in (66a)
and be assigned an appropriate 0-role (that of experiencer argument of enjoy).
339
A movement
Given the assumption that auxiliaries, infinitival to and raising predicates like seem do
not 0-mark their subjects, the only 0-role assigned to the pronoun he will be that of
experiencer argument of enjoy (thereby satisfying the 0-criterion requirement that
‘Each argument bears one and only one 0-role’). We might assume that he has to raise
to become the subject of does for reasons of greed (i.e. to check its nominative case),
and that it has to raise in a successive cyclic fashion in order to satisfy the shortest
movement principle. But now consider what would happen if we tried to use seem as
a control predicate. The pronoun he would originate as the subject of seem (raising to
become subject of does), and would serve as the controller of the PRO subject of enjoy
syntax, as in (68) below:
(68) [jp He does [yp t seem [jp to [yp PRO enjoy syntax]]]]
i_l
argument of enjoy syntax. However, the pronoun he would be assigned no 0-role at all,
since it originates as the subject of the raising verb seem and then raises to become the
subject of does (and neither raising verbs nor auxiliaries 0-mark their subjects). Thus,
a derivation such as (68) would violate the 0-criterion requirement that ‘Each
argument bears one and only one 0-role’, and hence crash at LF.
Consider now the syntax of (66b) He does try to enjoy syntax. If we suppose that try
is a control predicate, the pronoun he will originate as the subject of try (subsequently
raising to become subject of does) and will control the PRO subject of enjoy syntax, as
in (69) below:
(69) [-pp He does [yp t try fpp to [yp PRO enjoy syntax]]]]
l l
The null pronoun PRO will be assigned the 0-role of experiencer argument of enjoy,
and the nominative pronoun he will be assigned the 0-role of agent argument of try
(by virtue of originating as the subject of try). Since PRO and he are each assigned a
0-role of their own, the ©-criterion is satisfied.
340
8.8 Passivization
However, the picture is very different if we use try as a raising predicate, with he
originating as the subject of enjoy syntax and then being raised up successive-cyclically
to become subject of to enjoy syntax, try to enjoy syntax and does try to enjoy syntax, as
in (70) below:
The problem here is that he ends up with two different 9-roles, viz. as experience!?
argument of enjoy (at the stage of derivation where it is subject of enjoy syntax) and
agent argument of try (at the point where it becomes subject of try to enjoy syntax).
This leads to obvious violation of the 9-criterion requirement that ‘Each argument
bears one and only one 9-role.’
8.8 Passivization
Having looked briefly at the syntax of raising predicates (and how they
differ from control predicates), we now turn to look at the syntax of passive
predicates. Traditional grammarians maintain that the bold-printed verbs in sentences
such as the (a) examples below are in the active voice, whereas the italicized verbs in
the corresponding (b) sentences are in the passive voice:
There are four main properties which differentiate passive sentences from their active
counterparts - as the examples in (71-3) illustrate. One is that passive (though not
active) sentences generally require the auxiliary be. Another is that the lexical verb in
passive sentences is in the n-participle form (cf. seen!stolen!taken), known in this use as
the passive participle form. A third is that passive sentences may (though need not) con¬
tain a fey-phrase in which the complement of by seems to play essentially the same the¬
matic role as the subject in the corresponding active sentence: for example, hundreds of
passers-by in the active structure (71a) serves as the subject of saw the attack, whereas
in the passive structure (71b) it serves as the complement of the preposition fey (though
in both cases seems to have the role of experiencer). The fourth difference is that the
expression which serves as the complement of an active verb surfaces as the subject in
the corresponding passive construction: for example, the attack is the complement of
saw in the active structure (71a), but is the subject of was seen by hundreds ofpassers-
341
A movement
by in the passive structure (71b). Here, we focus on the syntax of the subjects of passive
sentences (setting aside the derivation of fry-phrases).
Evidence that passive subjects do indeed play the same thematic role as active com¬
plements comes from the fact that the two are subject to the same restrictions on the
choice of expression which can fulfil the relevant argument function, as we see from
sentences such as the following (where ?, ?! and ! mark increasing degrees of prag¬
matic anomaly):
How can we account for this fact? If we assume that principles of UG correlate
thematic structure with syntactic structure in a uniform fashion, then it follows that two
arguments which fulfil the same thematic function with respect to a given type of
predicate must occupy the same underlying position in the syntax. The assumption that
there is a uniform mapping between thematic structure and syntactic structure is
embodied in the uniform theta assignment hypothesis/UTAH argued for at length in
Baker 1988. If we adopt UTAH, it follows that passive subjects must originate in the
same position as active complements. Since the passive subject the students in (74a)
The students were arrested bears that theme/patient role which is normally assigned to
the complement of arrest (so that the students are the ones taken away for question¬
ing), a natural suggestion to make is that the students originates as the complement of
the verb arrested. But if this is so, how does the students come to be subject of the aux¬
iliary were?
The answer we shall suggest here is that the DP the students is raised in a successive
cyclic fashion to become first the subject of the passive participle arrested and then the
subject of the auxiliary were - as in (75) below:
arrested t
Thus, the students originates as the complement of arrested, then undergoes passiviza-
tion (i.e. movement from being complement of the passive participle arrested to
becoming its subject), and finally undergoes raising (i.e. movement from being subject
of arrested to becoming subject of were).
342
8.8 Passivization
Some evidence in support of the assumption that the students becomes the subject of
the passive participle arrested before becoming subject of the auxiliary were comes
from facts relating to floating quantifiers in sentences such as:
If we assume (as in §8.3 above) that a floating quantifier modifies the trace of a moved
subject, we can say that the quantifier all in (76) modifies the trace in spec-VP of the
moved DP the students in (75).
Additional evidence in support of the claim that passivized arguments first become
the subject of the passive participle before moving on to become the subject of the
passive auxiliary comes from expletive structures such as:
The quantifier phrase several students in (77) originates as the complement of the verb
arrested. But where does it end up? Since it is positioned immediately in front of the
passive participle arrested, it seems likely that it moves into (and remains in) spec-VP,
and hence serves as the subject of arrested; this would mean that (77) has the (simpli¬
fied) derivation (78) below:
(78)
There
several students QP
♦
arrested t
)
passivization seems to have the effect of raising the italicized expression out of subject
position in the bracketed complement clause into subject position within the main
343
A movement
Let’s try and work out what’s going on here, by looking at how (79a) He is thought to
admire her is derived.
The passivized pronoun he originates as the subject of admire her. Some empirical
evidence in support of this claim comes from the fact that the restrictions imposed on
the choice of subject in (79a) above mirror those found in a simple sentence such as He
admires her, as we see from (81) below:
Thus, it seems reasonable to suppose that the pronoun he in (79a) originates as the
subject of admire her and is then raised up (in a successive cyclic fashion) to become
first the subject of to admire her then the subject of thought to admire her, and finally
the subject of is thought to admire her. This being so, (79a) will be derived in the
manner indicated in (82) below:
(82) TP
He T VP
admire her
Step (1) in the derivation involves raising he to become the subject of the TP headed by
infinitival to\ step (2) involves passivizing he, i.e. moving it into a position where it
344
8.9 Explanation
becomes the subject of the VP headed by the passive participle thought; step (3)
involves raising he to become the subject of the TP headed by the auxiliary is.
Although in describing the travels of the subject in (82), we used the terms raising
and passivization as if they denoted different processes, closer reflection suggests that
there is no meaningful distinction between the two. In relation to our earlier discussion
of structures such as (75) above, we implicitly thought of raising as movement from
one subject position to another and passivization as movement from complement to
subject position. But the fact that passivization in structures such as (82) involves
subject-to-subject movement undermines this distinction. In reality, passivization and
raising are different manifestations of a single argument-movement operation (conven¬
tionally termed A movement) which has the effect of moving a constituent from one
argument position into another (more specifically, from a subject or complement
position into a subject position). Extending this terminology in a conventional way,
we might say that operations which move maximal projections into a nonsubject posi¬
tion are instances of A movement (where the bar here is a negation operator, so that
A-bar movement is movement to a nonargument position): so, for example, operator
movement is one type of A-bar movement operation, since it moves wh-operators into
spec-CP (and spec-CP is a nonargument position by virtue of the fact that nonargu¬
ments like whether or why can occupy spec-CP).
8.9 Explanation
Since the ultimate goal of any theory is to explain the phenomena which
fall within its domain, an important question for us to ask is why arguments should
undergo A movement in the way that they do (e.g. why he should move from being the
subject of admire to becoming the subject of is in (82) above). The most principled
answer which we can give to this question is that the syntax of A movement is entirely
determined by principles of UG. Let’s look at some of the principles which are
involved.
Consider first the question of why A movement should involve movement from a
lower to a higher position in structures such as (56-8), (75), (78) and (82). This follows
from principles of trace theory (cf. §6.3) which require that a moved constituent
leave behind a trace which must be bound by its antecedent, and from the c-command
condition on binding (cf. §3.9) which requires a bound constituent to be c-com-
manded by its binder (i.e. its antecedent). This latter condition is satisfied e.g. in (82)
by virtue of the fact that the trace subject of admire is c-commanded and bound by the
trace subject of to, which in turn is c-commanded/bound by the trace subject of
thought, which in turn is c-commanded/bound by the he subject of is.
Now let’s look at the question of why A movement should involve movement to a
higher specifier position, never to a higher complement position (e.g. the DP the
students in (75) moves first to become subject and specifier of arrested and then to
345
A movement
become subject/specifier of were). The answer is that within the theory assumed here, a
complement position can only be created by merger with a head (not by movement),
whereas a specifier position can be created either by merger or by movement.
A third aspect of the syntax of A movement which we need to explain is why it
should apply in a successive cyclic fashion: e.g. why should he in (82) first become
subject of to admire her and then subject of thought to admire her before finally
becoming subject of is thought to admire herl Again, the answer comes from
principles of UG: for example, the shortest movement principle requires that each
application of A movement should move the relevant constituent into the next highest
subject position in its containing structure. Alternatively, it may be that Rothstein’s
(1995) predication principle (requiring syntactic predicates to have a subject) will
ensure successive cyclicity, if we assume that V-bar and T-bar are syntactic predicates
(and hence require a subject at some stage of derivation).
A further principle which constrains the operation of A movement is the 0-criterion
(64). This plays an important role not only in the syntax of raising (as we saw earlier),
but also in the syntax of passivization. In this connection, consider how we account for
why the complement of a passive verb can be passivized, but not the complement of an
active verb - e.g. why the jewels can become the subject of the passive sentence (83a)
below, but not the subject of its active counterpart (83b):
The derivation of (83b) on its intended interpretation would be as follows. The DP the
jewels originates as the complement of the verb stole, and is then raised in successive
cyclic fashion to become first the subject of stole and then the subject of the abstract T
constituent heading TP, as in (84) below:
(84) TP
t V DP
A I I
stole t
346
8.9 Explanation
spec-VP and becoming the subject of stole, the jewels is also assigned the 0-role of
agent argument of stole. However, since the 0-criterion specifies that no argument
can carry more than one 0-role, the resulting derivation is correctly ruled out as
ungrammatical. In other words, the 0-criterion correctly predicts that active verbs
don’t allow passivization.
But now consider the derivation of the corresponding passive sentence (83a) The
jewels were stolen. If we assume that the DP the jewels originates as the theme com¬
plement of stolen and is then raised up successive-cyclically to become first the subject
of the passive participle stolen and then the subject of the auxiliary were, (83a) will
have the derivation (85) below:
t V DP
A I I
stolen t
At first sight, it might seem as if the derivation (85) violates the 0-criterion, by virtue
of the fact that the DP the jewels goes from being theme complement of stolen to
becoming agent subject of stolen (before eventually becoming subject of were).
However, let us suppose (following Chomsky 1981, pp. 124-7) that passive participles
0-mark their complements but not their subjects: to use a traditional metaphor, we
might say that the passive participle suffix +n absorbs the subject 0-role, thereby
dethematizing the passive subject position (i.e. removing the ability of the passive
participle to 0-mark its subject). If passive participles don’t 0-mark their subjects, it
follows that the 0-criterion will not prevent the DP the jewels from moving through
spec-VP into spec-TP as in (85) above: the only 0-role which the jewels receives is
that of theme argument of stolen. (For a more detailed discussion of the thematic prop¬
erties of passive verbs, see Roberts 1986, and Baker, Johnson and Roberts 1989.)
Thus, we see that passivization is made possible by the fact that passive participles
do not 0-mark their subjects. However, an important question for us to ask is why it
isn’t just possible but rather necessary for a passivized argument to move. One answer
is provided by the principle of greed - i.e. by the need for constituents to selfishly sat¬
isfy their own morphological requirements. We might suppose that (pro)nominal argu¬
ments have intrinsic case properties (e.g. he is nominative and him objective) which
have to be checked in the course of a derivation (and this will only be possible if e.g. a
nominative pronoun occupies a nominative position at some stage in the course of a
347
A movement
derivation). Let us make the traditional assumption that passive participles are inher¬
ently intransitive: if this is so, then we can provide a straightforward account of why
transitive verbs allow objective complements when used actively, but not passively —
cf. contrasts such as:
The objective case carried by him can be checked when him is used as the complement
of the active verb saw (since this is transitive), but not when used as the complement of
the passive participle seen (since passive participles are intransitive). To use our earlier
absorption metaphor, we might say that the passive participle +n inflection on seen
absorbs the ability of a transitive verb like see to check objective case, and thereby
detransitivizes the verb. Since the case-features carried by him cannot be checked in
(86b), the resulting derivation crashes, so that such impersonal passive structures are
ungrammatical in English.
But now suppose that in place of the objective pronoun him, we use the nominative
pronoun he. We cannot use he as the complement of a passive participle like seen, as
we see from the fact that impersonal passives such as (87) below are ungrammatical:
The reason why (87) is ungrammatical is that a nominative pronoun cannot check its
case if used as the complement of a passive participle like seen. But now suppose that
he undergoes passivization, becoming first the subject of seen and then the subject of
was - as in (88) below:
(88)
seen t
I
348
8.10 Summary
greed analysis is that a constituent will move no further than it needs to in order to
satisfy its morphological requirements. In this connection, consider how we account
for contrasts such as the following:
How come we can passivize the subject of an infinitive clause like that bracketed in
(89a), but not the subject of a finite clause like that bracketed in (89b)? Greed provides
us with the answer. If the subject he does not passivize but remains in situ within the
bracketed clause, the result will be:
The pronoun he cannot remain in situ in (90a), since if it does it will be unable to check
its nominative case-feature; hence, morphological (case-checking) requirements force
it to passivize, and move into the nominative position it occupies in (89a). But the
converse is the case in (90b): here he already occupies a nominative position (as
subject of the finite auxiliary has), so that its nominative case-feature can be checked
(and erased) in situ\ hence, he in (90b) need not (and therefore, by the economy
principle, cannot) move into another nominative position, so accounting for the
ungrammaticality of passivization in (89b).
8.10 Summary
In this chapter, we have been concerned with the syntax of subjects. In
§8.2 we looked at the syntax of expletive sentences such as There is someone waiting
for you, and argued that the expletive pronoun there serves as the specifier of the TP
headed by is, and that someone is the specifier of the VP headed by waiting. We then
went on to compare this kind of structure with its nonexpletive counterpart Someone is
waiting for you, suggesting that we could achieve a unitary account of both types of
structure if we posited that someone originates in spec-VP in both cases, and remains in
situ in the expletive structure (where spec-TP is filled by there), but raises to spec-TP
in the corresponding nonexpletive structure. We noted that the assumption that subjects
originate internally within VP is known as the VP-internal subject hypothesis. In
§8.3 we presented a range of empirical evidence in support of the hypothesis, from
data relating (for example) to the syntax of PRO, reflexives, floating quantifiers and
idioms. In §8.4 we argued that the VP-intemal subject hypothesis enables us to develop
a principled theory of 0-marking, in which arguments are assigned a 0-role by
merger with a lexical category: hence, the complement of a verb is 0-marked directly
by the verb, and its subject is 0-marked by V-bar. We suggested that nominative sub¬
jects raise to spec-TP in order to check their (strong) case-feature. In §8.5 we argued
349
A movement
that nominative pronouns can check their case by adjoining to a T constituent which
contains an affixal (i.e. contracted) auxiliary, e.g. in sentences like We’ll help you
out. We suggested that the null case carried by PRO is checked by attraction to a
nonfinite T constituent (containing e.g. infinitival to), and that the (partitive?) case of
expletive associates (like someone in There is someone waiting for you) may be
checked by attraction to the verb be. We also suggested that the objective case carried
by the subject of an ECM infinitive complement may be checked by attraction to a
matrix transitive head (e.g. the objective case-feature carried by him in They expect
him to win is attracted to expect). In §8.6 we went on to argue that a handful of
nonauxiliary verbs (like seem) resemble auxiliaries in that they do not 9-mark their
subjects; in consequence, such verbs (traditionally known as raising predicates) allow
the subject of an embedded complement to be raised up to become the subject of the
raising verb. In §8.7 we contrasted raising predicates with control predicates, noting
that control predicates 0-mark their subjects and take a complement clause with a
PRO subject which is 0-marked by the predicate in the complement. We looked at syn¬
tactic differences between the two types of predicate, noting (for example) that raising
predicates like seem allow an expletive there subject whereas control predicates like
try don’t. In §8.8 we examined the syntax of passivized arguments, claiming that
(in simple cases) they originate as complements of passive participles and are raised
up (in a successive cyclic fashion) to become first the subject of the passive participle,
and then the subject of the passive auxiliary be. We argued that passive participles have
two distinctive characteristics which differentiate them from the corresponding active
verb forms - namely that they do not 0-mark their subject, and do not check the
case of their complements. We noted that in infinitival passives such as He is thought
to admire her, the passive subject he originates as the subject of the verb admire in
the infinitive complement, and is raised up (by successive movement operations) to
become first subject of to, then subject of thought and finally subject of is. We argued
that raising and passivization are two different manifestations of a single A movement
operation whereby an argument moves from a subject or complement position into
a higher subject position. In §8.9 we argued that the operation of A movement is
constrained by a number of UG principles, including the c-command condition on
binding, the shortest movement principle, the 0-criterion and the principle of
greed.
350
Workbook section
Workbook section
Exercise I (§§8.2-8.4)
Discuss the derivation of the following sentences, paying particular atten¬
tion to the syntax of subjects:
followed her
The resulting VP is merged with the perfective auxiliary have-, the pronoun they raises
from being subject of followed in (i) above to becoming subject of have in (ii) below:
(ii) PERFP
t followed her
351
A movement
Finally, the PERFP in (ii) is merged with a T constituent which is itself formed by
merging the modal must with a present-tense affix, and the subject they raises up to
spec-TP to check its nominative case, as in (iii) below:
followed her
Everything about the syntax of the subject they follows from UG principles. They must
originate in spec-VP in order to receive its agent 0-role (from the V-bar followed her)
and thereby satisfy the 9-criterion (which requires that every argument be assigned a
0-role). It moves to spec-TP for reasons of greed, in order to check its strong nomina¬
tive case-feature. The requirement for they to move in a successive cyclic fashion
(becoming subject of have before becoming the subject of must) is imposed by the
shortest movement principle, which requires that an argument which undergoes
raising should move into the nearest subject position above it. Alternatively, we might
suppose that they has to become the specifier of have (before becoming the specifier of
must) because verbs and auxiliaries require a subject: this might either be for checking
purposes (e.g. verbs and auxiliaries have some strong specifier feature which they need
to check), or to satisfy some constraint such as Rothstein’s (1995) predication princi¬
ple (which requires that syntactic predicates should have subjects).
Given the assumption that floating quantifiers can modify traces of moved con¬
stituents, an obvious prediction made by the successive cyclic analysis is that each of
the two traces of they in (iii) can be modified by a floating quantifier like all/both. In
the event, this prediction is correct: cf.
For reasons which are not entirely clear, floating quantifiers seem to give better results
the nearer they are to the expression which they quantify: hence (iv)(a) is better than
(iv)(b) because both is closer to they in (iv)(a) than in (iv)(b).
352
Workbook section
Exercise II (§8.3)
Discuss the syntax of the quantifiers both/all in the following sentences:
Helpful hints
Make the following set of assumptions. Quantifiers like all or both have a
variety of uses. They can function as the specifier of a DP headed by a determiner like
the; but they cannot function as the specifier of a personal pronoun like them because
personal pronouns are normally unprojectable and so allow neither specifier nor
complement. Quantifiers like all/both can also function as heads taking a complement,
but in this use are intransitive and so cannot directly assign case to their complement:
this means that the complement must either be introduced by a dummy transitive
preposition like of, or must move to a case position. Consider the possibility that in 6a
the phrase them all is a QP headed by all, and that the pronoun them originates as the
complement of all, and then adjoins to the left of the quantifier. Why might it be that
the pronoun moves in this way - and why can’t a DP like the chocolates move in much
the same way?
(i) DP
/ \
both D N
the men
353
A movement
Given the VP-intemal subject hypothesis, the DP both the men will originate in spec-VP
as the subject of cheating, and will then raise into spec-TP to become the subject of were,
as in (ii) below (to simplify exposition, we ignore the past-tense affix merged with were):
(ii)
The DP both the men originates in spec-VP, where it receives its agent 0-role by
merger with the V-bar cheating at cards: it raises to spec-TP for reasons of greed, to
check the strong nominative case-features which it carries.
At first sight, it might seem plausible to derive lb in essentially the same way, and to
argue that both the men has the structure (i) and originates in spec-VP, with the men
being preposed on its own, leaving the quantifier both stranded in spec-VP, as in (iii)
below:
(iii)
However, any movement such as that arrowed in (iii) would violate the chain unifor¬
mity principle, since the constituent at the head of the chain would be a maximal
projection (viz. the DP the men), whereas the trace at the foot of the chain would be
nonmaximal (since it is a D-bar which is nonmaximal because it projects into a larger
DP containing both).
An alternative possibility is to suppose that in lb, the phrase both the men which
originates as the subject of cheating has the structure (iv) below:
(iv) QP
Q DP
I
both D N
I I
the men
354
Workbook section
If we assume that both is intransitive, it follows that the DP the men occupies a caseless
position in (iv), and so must move to a case position in order to check its case. It there¬
fore moves into spec-TP (which is a nominative position), as in (v) below:
In this position, the DP can check its case and agreement properties against the speci¬
fier properties of were (which requires a plural or second person nominative subject).
The resulting chain the men ... r is uniform, since both the DP the men at the head of
the chain and its trace t at the foot are maximal (DP) projections. Movement of the DP
the men is motivated by greed - i.e. the need to move the men into a case position (= a
position where it can check its nominative case). An interesting problem not discussed
here relates to the question of what case (if any) the floating quantifier both carries in
(v), and how it is checked.
355
A movement
Helpful hints
See those for exercise V, chapter 7.
I T VP
* I /
t V D
love t
Exercise IV (§§8.2-8.4)
Discuss the syntax of the following child sentences produced by a young
girl called Iris at around three years of age, concentrating particularly on the mor-
356
Workbook section
The analysis in (i) assumes continuity between the structure of clauses in adult and
child grammars, since both are assumed to have the status of TPs. We might suppose
that the null case carried by PRO is checked by attraction to the null infinitive particle
0. But how is the case of me checked? One possibility is that it is checked by attraction
to T (which may carry no tense in this use and so be nonfinite). Another is that the case
of me is not checked at all, and that objective case is the default case form in English
(i.e. the one case which does not need to be checked). A default analysis of objective
case might account for why in (adult) dialogues such as:
357
A movement
speaker B’s utterance (in spite of being interpreted as an elliptical form of I failed syn¬
tax) carries objective case in a position where it is the only word in the sentence, so
there is no other item to check its case. Default case would be earned by a (pro)nomi-
nal argument which occupies a position where it is unable to check any other case.
A radically minimalist analysis of child sentences like 5 (proposed in Radford 1990)
would be to suppose that since they contain no evidence of tense-marking or infinitival
to, the relevant clauses are simply VPs, so that 5 has the structure (iii) below:
We might then suppose that case in children’s earliest clause structures is checked (or
assigned) by merger, and that a subject which merges with a nonfinite (tenseless) V-bar
is thereby assigned either objective or null case. (Of course, we could alternatively
claim that the subject me in (iii) is assigned objective case by default.)
Helpful hints
Since it is clear that sentences like 1 (by virtue of containing the finite
auxiliary might) are TPs, the overall issue which arises in relation to the sentences in
this exercise is whether (at the relevant stage of development) all Iris’s clauses are TPs,
or whether some are TPs and others are VPs - or whether we simply can’t tell. Assume
in relation to 6 and 7 that no is a negative particle occupying the head NEG position of
a NEGP which has a VP complement.
Exercise V (§8.5)
Discuss the syntax of the following sentences. (The judgments given
below in relation to the apostrophized auxiliaries relate to their use as contracted forms
which comprise a single consonant segment and no vowel.)
1 They’ve eaten it
2 Who’d been helping her?
3 Who’ve they been helping?
4 Where’ve you been?
5 *Could they’ve made a mistake?
6 *Dave may’ve gone to a rave
7 *He would like to’ve met her
8 *The guy’d lied to her
9 *He wouldn't let you’ve done it
358
Workbook section
Helpful hints
Bear in mind that the ’s variant of is/has may have a different status from
contracted auxiliary forms like’ve and ’d - e.g. ’s may be a PF affix.
eaten it
Thus, the pronoun they originates in spec-VP as the agent argument (and subject) of
eaten it, and checks its nominative case by adjoining to the T constituent containing the
contracted (affixal) auxiliary form’ve. On this view, auxiliary contraction is an integral
part of case-checking, and not (except in the case of the ’s forms mentioned in the
helpful hints) the result of a PF attachment process. Evidence against the PF attach¬
ment analysis of contracted auxiliaries is presented in the main text, in relation to
sentences such as (40-1).
Exercise VI (§§8.6-8.7)
Say whether the italicized verbs as used in the type of construction illus¬
trated in the examples below function as raising and/or control predicates, and discuss
the derivation of each sentence, giving arguments in support of your answer.
359
A movement
The fact that tend can have a nonreferential subject like expletive there and imposes no
restrictions on the type of referential subject it allows is consistent with the assumption
that it does not 0-mark its subject, and hence can function as a raising predicate (for
the reasons set out in the text in relation to seem).
A further piece of evidence (not noted in the text) in support of the same conclusion
comes from the fact that tend can have an idiomatic subject: cf.
Given that all hell can serve only as the subject of break loose in the relevant idiom, it
is clear that we could not analyse tend as a control predicate in (iii)(a) and claim that
360
Workbook section
all hell originates as the subject of tends and PRO as the subject of break loose, since
this would violate the requirement that all hell can occur only as the subject of break
loose (in the relevant idiom). By contrast, if tend is a raising predicate, we can claim
that all hell originates as the subject of break loose and then raises up in a successive
cyclic fashion to become first the subject of to break loose, then the subject of tends to
break loose and finally the subject of the abstract T constituent preceding tends.
Given the assumption that tend is a raising predicate, sentence 1 will have the
simplified derivation (iv) below:
(1V) [pp Power T [yp t [y tends] [-pp t [p to] [yp i [y corrupt] people ]]]]
A i A A
That is, power will originate as the subject of corrupt people, and will then be raised
(in a successive cyclic fashion) first to become the subject of to corrupt people, then
to become the subject of tends to corrupt people, and finally to become the subject
of the abstract T constituent which heads the overall TP. The relevant derivation
satisfies the 9-criterion by virtue of the fact that power is assigned only a single 0-role
(by the V-bar corrupt people)-, it also satisfies the shortest movement principle
because movement applies in a successive cyclic fashion; and it satisfies the principle
of greed in that the subject moves only the minimal distance required in order to
get into a position (as the specifier of the T constituent in the main clause) where
it can check its nominative case (the present-tense head-features and third person
nominative singular specifier-features of tends are attracted to the main-clause T con¬
stituent).
361
A movement
hates syntax
The pronoun he originates in spec-VP and raises to spec-TP; the tense- and agreement-
features of the finite verb hates are attracted to T. In consequence, the case- and
agreement-features of he and hates can be checked, and also the strong specifier-
feature of T. In a similar fashion, we might suppose that the impersonal it subject of the
overall sentence originates in spec-VP and raises to spec-TP (at any rate, if we assume
that in consequence of Rothstein’s predication principle, syntactic predicates like
T-bar and V-bar require a subject).
Sentence lb is generated by allowing he to raise further, first to become the subject
of seems, and then the subject of the T constituent above seems, as in (ii) below:
hates syntax
However, the illicit movements (2) and (3) are blocked by the principle of greed: since
the pronoun he has its nominative case-feature checked and erased in the lower spec-
362
Workbook section
TP position (after movement (1) has taken place), the principle of greed dictates that it
cannot move any further; in particular, it cannot move to the higher spec-TP position,
since if it does so it cannot satisfy the requirement for the finite T constituent heading
the overall sentence to have a nominative subject, because the nominative case-feature
of he has already been checked and erased.
Sentence lc is derived by moving the DP syntax (which is headed by a null deter¬
miner) to become the subject of the overall sentence; it may be that it first becomes the
subject of seems and then the subject of the T constituent above seems, as in (iii)
below:
hates t
'- (2) -1
However, such a derivation violates at least two UG principles. One is the principle of
greed, which inter alia forbids movement from one case position to another: since the
DP syntax can check its case by remaining in situ (since it is the complement of the
transitive verb hates), it cannot undergo movement to a further case position - e.g. to
the nominative root specifier position at the top of the tree. A second principle which is
violated in (iii) is the principle of shortest movement, which is violated because
syntax does not move into the nearest subject position above the position in which it
originates (the next-highest subject position would be subject/specifier position in the
VP headed by hates - a position which is occupied by the trace of he).
363
A movement
get D V
I X
t V D
* I I
arrested t
1 )
The derivation in (i) satisfies the c-command condition on binding by virtue of the fact
that the rightmost trace is c-commanded (and bound) by the middle trace, which in turn is
c-commanded (and bound) by the leftmost trace, which itself is c-commanded (and
bound) by the moved pronoun she. The derivation (i) also satisfies the shortest move¬
ment principle by virtue of the fact that each movement of she takes it into the next-
highest subject position in the structure. The principle of greed is also satisfied, since
she moves from (and through) a series of positions in which it can’t check its nominative
case, into its ultimate position as subject of may (where it can check its case against the
nominative specifier-feature of may). Since the resulting sentence 1 is grammatical, it fol¬
lows that the derivation (i) must also satisfy the 0-criterion. This means that neither
arrested nor get nor may 0-marks its subject: arrested because it is a passive participle
(and passive participles have dethematized subjects), get because it is a raising predicate
in this use (and raising predicates don’t 0-mark their subjects), and may because it is an
auxiliary (and functional categories aren’t 0-markers). Thus, the only 0-role which she
is assigned is that of patient argument of arrested (thereby satisfying the 0-criterion
requirement that each argument is assigned one and only one 0-role).
364
Workbook section
Exercise IX (§§8.8-8.9)
Discuss the syntax of the following passive sentences:
(Examples 8a-b are from student exam papers; example 8c was produced by the
deputy commissioner of police for south-west London in a BBC radio interview;
examples 8d-e were produced by BBC radio reporters - complaints about falling stan¬
dards should be addressed directly to the Director General of the BBC, Bush House,
London.) Discuss the derivation of the relevant sentences and say what’s odd about
(the people who produced) them.
365
A movement
t V D
arrested t
At first sight, it might seem as if lb can be derived in the same way. However, there is
an important difference between be and get - namely that be is an auxiliary, whereas
get is a nonauxiliary verb - as we see from the fact that be can undergo auxiliary inver¬
sion whereas get cannot: cf.
Accordingly, it would seem likely that lb has the more complex derivation (iii) below:
It follows that the auxiliary were in (i) can undergo inversion because it is positioned in
T, but the verb got in (iii) can’t undergo inversion because it is positioned in V. What
lends plausibility to the assumption that the passivized subject they becomes subject of
the passive participle before becoming subject of get is the existence of sentences such
as:
where the passivized pronoun you appears to be the subject of the passive participle
arrested.
366
9
VP shells
9.1 Overview
In this chapter, we take a close look at the internal constituent structure of
verb phrases. We shall argue that VPs have a complex structure, comprising an inner
VP core and an outer vp shell, and that some (e.g. agent) arguments originate within
the outer vp shell, while other (e.g. theme) arguments originate within the inner VP
core.
If we make the conventional assumption that complements are sisters to heads, it fol¬
lows that the V-bar constituent headed by rolled in (la) will have the structure (2)
below:
However, a structure such as (2) is problematic within the framework adopted here.
After all, it is a ternary-branching structure (Vbranches out into three separate con¬
stituents, namely the V rolled, the DP the ball and the PP down the hill), and this poses
367
VP shells
an obvious problem within a framework which assumes that the merger operation
which forms phrases is an inherently binary operation which can only combine con¬
stituents in a pairwise fashion. Moreover, a ternary-branching structure such as (2)
would wrongly predict that the string following the verb rolled does not form a con¬
stituent, and so cannot be coordinated with another similar string (given the
traditional assumption that only constituents can be conjoined); yet this prediction is
falsified by sentences such as:
(3) He rolled the ball down the hill and the acorn up the mountain
(Verbs which can be used in this way either as three-place or as two-place predicates
are sometimes referred to as ergative predicates.) Moreover, the italicized DP seems
to play the same thematic role with respect to the bold-printed verb in each pair of
examples: for example, the ball is the theme argument of roll (i.e. the entity which
undergoes a rolling motion) both in (4a) We rolled the ball down the hill and in (4b)
The ball rolled down the hill. Evidence in support of the claim that the ball plays the
same semantic role in both sentences comes from the fact that the italicized argument
is subject to the same restrictions on the choice of expression which can fulfil the
relevant argument function in each type of sentence: cf.
368
9.2 Ergative predicates
(10) (a) The ball/the rock/!the theory/.'sincerity rolled down the hill
(b) John rolled the ball/the rock//the theory//sincerity down the hill
In the case of (4b), the resulting VP will then be merged with a null tense affix to form
a T-bar constituent; the subject the ball will then be raised to spec-TP (by A
movement), as in (12) below (simplified e.g. by ignoring attraction of the tense/agree¬
ment-features of rolled to T):
(12) .TP
369
VP shells
Subsequently, the vp in (13) merges with an abstract tense affix to form a T-bar, and the
subject we raises into spec-TP to check its nominative case, as in (14) below:
The objective case carried by the DP the ball is checked by the transitive verb rolled
(or perhaps by the light verb 0).
370
9.3 Adverbs and prepositional particles
Let us make the traditional assumption that adverbs like gently merge with intermedi¬
ate verbal projections like V-bar and v-bar. Let’s also assume that such adverbs are
adjuncts which have the property that when they merge with a given category, they
form an expanded category of the same type (so that an adverb merged with V-bar
forms an expanded V-bar, and an adverb merged with v-bar forms an expanded v-bar).
Given these assumptions and the light-verb analysis in (14), we could then propose the
following derivations for (15a-b).
In (15a), the verb rolled merges with the PP down the hill to form the V-bar rolled
down the hill, and this V-bar in turn merges with the DP the ball to form the VP the ball
rolled down the hill, as in (11) above. This VP then merges with a causative light verb 0
to which the verb rolled adjoins, forming the v-bar rolled the ball down the hill. The
resulting v-bar merges with the adverb gently to form the expanded v-bar gently rolled
the ball down the hill; and this v-bar in turn merges with the subject we to form the vp
we gently rolled the ball down the hill. The vp thereby formed merges with an abstract
tense affix, forming a T-bar; the subject we raises to spec-TP forming the TP (15a) We
gently rolled the ball down the hill. Thus (15a) has the derivation (16) below:
Now consider how (15b) We rolled the ball gently down the hill is derived. As
before, the verb rolled merges with the PP down the hill, forming the V-bar rolled down
the hill. The adverb gently then merges with this V-bar to form the expanded V-bar
gently rolled down the hill. This V-bar in turn merges with the DP the ball to form the
VP the ball gently rolled down the hill. The resulting VP is merged with a causative
light verb 0 to which the verb rolled adjoins, so forming the v-bar rolled the ball gently
down the hill. This v-bar is then merged with the subject we to form the vp we rolled
the ball gently down the hill. The vp thereby formed merges with an abstract tense
affix, forming a T-bar; the subject we raises to spec-TP forming the TP (15b) We rolled
the ball gently down the hill, which has the derivation (17) below:
371
VP shells
(17) TP
n
The different positions occupied by the adverb gently in (16) and (17) reflect a subtle
meaning difference between (15a) and (15b): (15a) means that the action which initi¬
ated the rolling motion was gentle, whereas (15b) means that the rolling motion itself
was gentle.
A light-verb analysis such as that sketched above also offers us an interesting
account of adverb position in sentences like:
(18) (a) He had deliberately rolled the ball gently down the hill
(b) *He had gently rolled the ball deliberately down the hill
It seems reasonable to suppose that deliberately (by virtue of its meaning) can only be
an adjunct to a projection of an agentive verb (i.e. a verb whose specifier has the the¬
matic role of agent). If we suppose (as earlier) that the light verb 0 is a causative verb
with an agentive subject, the contrast in (18) can be accounted for straightforwardly: in
(18a) deliberately is contained within a vp headed by the agentive causative light verb
0; but in (18b) it is contained within a VP headed by the nonagentive verb roll (roll is a
nonagentive verb because its specifier has the 0-role theme, not agent). We can then
say that adverbs like deliberately are strictly vp adverbs.
This in turn might lead us to expect to find a corresponding class of VP adverbs. In
this connection, consider the following contrasts (adapted from Bowers 1993, p. 609):
(19) (a) Mary jumped the horse perfectly over the last fence
(b) *Mary perfectly jumped the horse over the last fence
Given the assumptions made here, the derivation of (19a) would be parallel to that in
(17), while the derivation of (19b) would be parallel to that in (16). If we assume that
perfectly (in the relevant use) can function only as a VP-adverb, the contrast between
(19a) and (19b) can be accounted for straightforwardly: in (19a), perfectly is merged
with a V-bar (consistent with its status as a VP adverb), whereas in (19b) it is merged
with a v-bar (in violation of the requirement that it can only serve as a VP adverb).
372
9.3 Adverbs and prepositional particles
We might suppose that (20a) is derived as follows. The DP the store originates as the
subject of close down as in (21) below:
the store V P
close down
The VP in (21) is then merged with a causative light verb 0 (whose agent subject is
they), and the verb close adjoins to the light verb as in (22) below:
t down
The vp in (22) then merges with a T constituent containing may (and an abstract tense
affix, though this is a detail which we overlook below); the pronoun they raises to spec-
TP to become the subject of may as in (23) below:
And (23) is the structure associated with (20a) They may close the store down.
Now consider how we derive (20b) They may close down the store. Let’s suppose
that close merges with down to form the V-bar (24) below:
373
VP shells
close down
Let’s further suppose that the particle down (optionally) adjoins to the verb close,
forming the complex verb close down, as in (25) below:
P
/ \ I
V P t
I I ;
close down
! :
The resulting V-bar merges with its DP subject the store to form the VP (26) below:
V P t
close down
The VP in (26) is then merged with a causative light verb 0 whose agent subject is
they, the complex verb close down adjoins to this light verb, as in (27) below:
Subsequently, the vp in (27) merges with a T constituent containing may, the subject
they raises to spec-TP to become the subject of may, as in (28) below:
374
9.3 Adverbs and prepositional particles
And (28) is the structure of (20b) They may close down the store. Thus, we see that a
VP shell analysis enables us to provide a principled account of the two positions which
can be occupied by prepositional particles in sentences such as (20a-b).
As we have seen, the VP shell analysis outlined here provides an interesting solution
to the problems posed by three-place ergative predicates which appear to take two
complements. However, the problems posed by predicates which take two comple¬
ments arise not only with transitive verbs (like those in (4-9) above) which have
intransitive counterparts, but also with verbs such as those italicized in (29) below (the
complements of the verbs are bracketed):
Verbs like those in (29) cannot be used intransitively, as we see from the ungrammati-
cality of sentences such as (30) below:
375
VP shells
However, it is interesting to note that in structures like (29) too we find that VP adverbs
can be positioned either before the verb or between its two complements: cf.
This suggests that (in spite of the fact that the relevant verbs have no intransitive coun¬
terpart) a VP shell analysis is appropriate for structures like (29) too. This would mean
(for example) that a sentence such as (29a) would have the derivation (32) below:
We could then say that the adverb carefully adjoins to v-bar in (31a), and to V-bar in
(31b). If we suppose that verbs like load are essentially affixal in nature (and so must
adjoin to the agentive light verb 0) we can account for the ungrammaticality of intran¬
sitive structures such as (30a) *The truck loaded with hay. Alternatively, we might
conjecture that load is an obligatorily transitive verb (and so has to check objective
case), and hence can only occur in structures like (32) where it is raised into a position
where it can check the objective case of the immediately following DP the truck. A
third possibility would be to posit that verbs like load carry a strong causative feature
[+caus] which has to be checked by adjunction to an abstract causative light verb.
Moreover, sentences like (29f-g) show the same dual position of prepositional
particles as sentences such as (20): cf.
This suggests that they should be given essentially the same kind of VP shell analysis
as sentences like (20a-b).
376
9.4 Ditransitive and resultative predicates
Since complement DPs are referred to as objects in traditional grammar, structures like
(35) are said to be instances of the double-object construction. (The first bracketed
DP in each case is said to be the indirect object of the italicized verb and the second its
direct object.) More specifically, we might suggest that (35a) is derived in the follow¬
ing manner. The verb get merges with the DP a present (assigning it the 0-role theme)
get a present
This in turn merges with the DP the teacher (which is assigned the 0-role recipient) to
form the VP (37) below:
the teacher V DP
I / \
get a present
The resulting VP (37) merges with the abstract causative light verb 0 (whose agent
subject is they), and the verb get adjoins to this light verb as in (38) below:
t a present
The vp thereby formed then merges with a T constituent containing will, and the sub¬
ject they raises to spec-TP to become the subject of will, as in (39) below:
377
VP shells
t a present
And (39) is the structure associated with sentence (35a) They will get the teacher a pre¬
sent. (For a range of alternative analyses of the double-object construction, see Larson
1988, 1990, Johnson 1991, Bowers 1993 and Pesetsky 1995.)
The VP shell analysis outlined above can further be extended to so-called resulta-
tive predicates like those italicized below:
(The verb paint as used in structures like (40b) is called a resultative predicate because
the result of the painting process in that the house becomes pink.) Resultative predi¬
cates pose much the same problem as the three-place predicates which we looked at
earlier - namely that the resultative verb in each case appears to have two comple¬
ments, and thus poses a potential problem for a binary-branching framework. However
(as we shall see), we can overcome these problems if we adopt a light-verb analysis of
resultatives.
More concretely, we might suppose that (40a) is derived as follows. The verb turned
originates in the head V position of VP, with the DP the litmus paper as its subject and
the adjective red as its complement, as in (41) below:
the litmus V A
paper I I
turned red
The VP in (41) then merges with the causative light verb 0 (whose subject is the DP the
acid), and the verb turned raises to adjoin to 0 as in (42) below:
378
9.5 Three-place predicates
(42)
A
I
red
Subsequently, the vp in (42) merges with an abstract tense affix, and the DP the acid
raises to spec-TP, so deriving (40a) The acid turned the litmus paper red. (For very dif¬
ferent analyses of resultative sentences like (40), see Carrier and Randall 1992 and
Keyser and Roeper 1992.)
(43) (a) He had remarked [to her] [that Senator Scumme-Bagge was a fraud]
(b) She suggested [to him] [that they should try safe syntax]
(c) I concluded [from his tight fist] [that Dougal Frugal was Scots]
(d) He learned [from Superman] [that Lois Lane was a pain]
(e) He agreed [with her] [that Senator Doleful was soulful]
If we make the traditional assumption that the two bracketed expressions in each
example in (43) are complements of the italicized verbs and that complements are sis¬
ters to their heads, the V-bar headed by remarked in (43a) will have a structure along
the lines of (44) below:
But the problem with (44) is that V-bar is a ternary-branching category (since it has
three immediate constituents, namely V, PP and CP); hence, the traditional analysis in
(44) is incompatible with our key assumption that syntax is binary-branching.
379
VP shells
The VP in (45) then merges with an agentive light verb whose subject is he; the verb
remarked adjoins to the light verb as in (46) below:
Subsequently, the vp in (46) merges with a T constituent containing the auxiliary had,
and he raises to spec-TP to become the subject of had, so deriving (43a) He had
remarked to her that Senator Scumme-Bagge was a fraud.
We might propose a similar derivation for structures like those below which have so-
called ‘impersonal it’ as their subject:
(47) (a) It seems [to me] [that Cindy Suquet was taking him for a ride]
(b) It occurs [to me] [that your analysis needs more light verbs]
(c) It appears [to me] [that there is evidence of a cover-up]
The main difference from (46) would be that the subject of the light verb in (47) would
be impersonal it rather than an agent argument like he in (46).
A second class of three-place predicates with clausal complements are verbs like
those italicized below, which take bracketed D(P) and CP complements:
(48) (a) You must satisfy [the jury] [that you are innocent]
(b) She reminded [him] [that the FBI were tailing him]
380
9.5 Three-place predicates
Once again, if we make the traditional assumption that complements are sisters to their
heads, the V-bar headed by satisfy in (48a) will be analysed as in (49) below:
The VP in (50) would then merge with a causative light verb whose subject is you, with
the verb satisfy raising to adjoin to the light verb as in (51) below:
(51)
The resulting vp would then merge with a T constituent containing the modal auxiliary
must, and the subject you would raise to spec-TP to become the subject of must, so
deriving (48a) You must satisfy the jury that you are innocent. We might propose a sim¬
ilar derivation for impersonal sentences like (52) below:
381
VP shells
save that the subject of the light verb in (52) is impersonal it rather than an agent argu¬
ment like you in (51).
Each of the verbs in (53) takes a D(P) object and an infinitive complement with a PRO
subject: thus, (53a) has the fuller structure (54) below:
Since in each of the sentences in (53) the PRO subject of the infinitive is controlled by
the D(P) object of the italicized verb, the relevant predicates (in this kind of use) are
known as object-control predicates.
Some evidence in support of positing that the verb decide in sentences like (53a) is
indeed a three-place object-control predicate, and that you is the object of decided
(rather than the subject of to take syntax) comes from the fact that (53a) can be para¬
phrased (for some speakers, albeit rather awkwardly) as in (55) below:
where the first you corresponds to the object you in (54) and the second you corre¬
sponds to PRO in (54). Moreover, the verb decide imposes restrictions on the choice of
argument expression following it (which must be a rational entity - not an irrational
entity like the exam)'.
The obvious conclusion to draw from facts such as these is that the (pro)nominal fol¬
lowing decide is an (object) argument of decide in sentences such as (53a), and serves
as the controller of a PRO subject in the following ro-infinitive.
382
9.6 Object-control predicates
This means that decide has two complements in structures such as (54) - the pro¬
noun you and the infinitive to take syntax. If we make the traditional assumption that
complements are sisters to the verb which 0-marks them, this would seem to lead us to
the conclusion that the V-bar headed by decided in (54) has the structure (58) below:
(58)
More specifically, we might suppose that (53a) is derived as follows. The verb decided
merges with the infinitival TP to PRO take syntax, so forming a V-bar which in turn
merges with the pronoun you to form the VP (60) below:
The resulting VP then merges with an abstract causative light verb 0 (a null counterpart
of made in (59) above) whose subject is whaf, the verb decide adjoins to the light verb
0, as in (61) below:
(a structure paraphraseable as ‘What made + decide you to take syntax?’). The objective
case carried by the pronoun you is checked by the immediately preceding transitive light
verb 0 (or by the verb decided, which is transitive in this use). The resulting vp is then
merged with an abstract T constituent, and the pronoun what raises to spec-TP to check
its nominative case, as in (62) below:
383
VP shells
The light-verb analysis in (62) offers two main advantages over the traditional analysis
in (58). Firstly, (62) is consistent with the view that the merger operation by which
phrases are formed is binary; and secondly, (62) enables us to attain a more unitary the¬
ory of control under which the controller of PRO is always a subject, never an object
(since PRO in (62) is controlled by you, and you is the subject of the VP headed by the
verb decided). This second result is a welcome one, since the verb decide clearly func¬
tions as a subject-control verb in structures such as:
where the PRO subject of take syntax is controlled by the he subject of decided.
Object-control verbs like decide in sentences like (53a) What decided you to take
syntax? occur in structures in which they are followed by an objective argument and a
fo-infinitive: using traditional grammatical terminology, we might say that verbs like
decide allow an accusative and infinitive string after them (the traditional term
accusative denotes objective case). In this respect, they resemble ECM (exceptional
case-marking) infinitive structures like (64) below:
(65)
VP
t take syntax
384
9.6 Object-control predicates
with me originating in spec-VP as the subject of take syntax, and raising to spec-TP in
order to get into a position where it can have its objective case checked by the immedi¬
ately preceding transitive verb expected. Since some accusative and infinitive verbs
(e.g. decide) are object-control predicates and others (e.g. expect) are ECM predicates,
an obvious question to ask is how we determine whether a given verb which allows the
accusative and infinitive construction is a control predicate or an ECM predicate.
Work in the 1960s (e.g. Chomsky 1965 and Rosenbaum 1967) uncovered numerous
differences between ECM predicates like expect and object-control predicates like
persuade (the verb persuade was argued to be a typical object-control predicate). For
one thing, the two have different semantic properties: in structures like persuade some¬
one to ... , someone is an argument of persuade, whereas in expect someone to ... ,
someone is not an argument of expect, as we can see from the two different finite-
clause paraphrases for the relevant structures given in (66) and (67) below:
The fact that me in (66a) is an argument of persuade is suggested by the fact that its
italicized counterpart me in (66b) is positioned outside the bracketed fhat-clause.
Conversely, the fact that me in (67a) is the subject of the infinitive to take syntax is
suggested by the fact that its italicized counterpart / in (67b) is positioned inside the
bracketed that-clmse, and carries the nominative case typical of the subject of a finite
clause.
Further evidence that the nominal following persuade is an argument of persuade
(whereas that following expect is not an argument of expect) comes from the fact that
the verb persuade imposes restrictions on the choice of nominal following it (requiring
it to be a rational entity, not an irrational entity like the exam), whereas the verb expect
does not: cf.
Similarly, expect (but not persuade) can be immediately followed by an idiomatic sub¬
ject (i.e. by a subject which is part of a clausal idiom): cf.
Likewise, expect (but not persuade) can be followed by expletive there: cf.
385
VP shells
The relevant differences follow from the assumption that expect is an ECM predicate,
whereas persuade is an object-control predicate: the nominal following persuade is an
experiencer argument of persuade, and so must be a rational entity (hence cannot e.g.
be an irrational entity like the exam or a dummy pronoun like there)-, but the nominal
following expect is not an argument of expect, and so is not constrained by expect.
Given the assumption that persuade is an object-control predicate in sentences like
(66a), we might propose to analyse it in essentially the same way as we earlier
analysed the verb decide in (62) above. This would mean that the nominal expression
following persuade would originate as the subject of persuade, and the verb persuade
would itself originate as the complement of an agentive light verb 0 with a causative
sense (so that (66a) He persuaded me to take syntax would be analysed as being similar
in certain respects to He made me decide to take syntax). On this view, persuade would
merge with its infinitival TP complement to form a V-bar, and the resulting V-bar
would then merge with its experiencer subject me to form the structure (71) below
(where persuade has much the same sense as decide):
^ VP'^\
me V TP
persuade T VP
N
/
XQ
l>
to
1 / \
PRO take syntax
Fillmore (1972, p. 10) suggested that personally is typically associated with experi¬
encer subjects, and this suggestion would appear to be borne out by the contrast in
(73) below:
where personally can be associated with I in the first sentence but not the second - and
this would seem to correlate with the fact that I is an experiencer subject in (73a), but
an agent subject in (73b). Thus, the fact that personally can be construed with me in
(72) lends empirical support to the analysis of me as the experiencer subject of
persuade in (71).
386
9.6 Object-control predicates
The VP in (71) is subsequently merged with the agentive light verb 0, and persuade
adjoins to the light verb, so forming a v-bar which itself merges with its agent subject
he, as in (74) below:
The agentive light verb 0 assigns the 0-role agent to its subject he: the objective case
carried by me is checked by the immediately preceding transitive verb persuaded (or,
perhaps, by the light verb to which it adjoins). Subsequently, the vp in (74) merges
with an abstract tense affix, so forming a T-bar; the subject he then raises to spec-TP to
check its nominative case. (See Larson 1991 and Bowers 1993 for a similar analysis of
object-control verbs like persuade.)
Some supporting empirical evidence in favour of the light-verb analysis in (74)
comes from the fact that an adverb like eventually can be positioned either before
persuaded or after me: cf.
How can we account for this difference? One possibility is to assume that persuade is
an inherently affixal (or causative) verb which can therefore only be used in structures
like (74) which contain a causative light verb; another is to make the traditional
assumption that persuade (in active uses) is an obligatorily transitive verb, and hence
can only occur in structures like (74) (where it can check the objective case of the
immediately following objective pronoun me).
387
VP shells
In the (a) examples in (77-85), the bold-printed verb has both an (italicized) indirect
object (= me) and a direct object (= the DP following me), and thus serves as a ditransi-
tive predicate; but in the corresponding (b) examples, the verb has a direct object but
no indirect object, and thus serves as a monotransitive predicate (i.e. a predicate which
has only one object). How are we to analyse monotransitive predicates?
To make our discussion more concrete, consider how we derive a sentence such as
(85b) He sent a card. If we adopt the uniform theta assignment hypothesis, we
clearly want to arrive at an analysis of (85b) which maximizes structural symmetry
with (85a) He sent me a card. So, let’s first consider how we might analyse (85a).
Given our earlier analysis of ditransitive verbs in §9.4, (85a) will be derived as follows.
The verb sent merges with its theme argument a card to form the V-bar sent a card;
this V-bar in turn merges with its recipient argument me to form the VP (86) below:
(86) vp
me V DP
I / \
sent a card
388
9.7 Monotransitive predicates
The VP in (86) then merges with an abstract agentive light verb whose agent subject is
he, and the verb sent adjoins to the light verb as in (87) below:
(87)
/ \
a card
The case of the indirect object me is checked by the immediately preceding transitive
verb sent (or by the agentive light verb 0), and the case of he is checked by raising he
to spec-TP.
Now consider how we derive the corresponding monotransitive structure (85b) He
sent a card. Since the DP a card plays the same thematic role of theme argument and
complement of sent as in (85a) He sent me a card, let’s assume (in conformity with the
uniform theta assignment hypothesis/UTAH) that it again serves as the complement
of sent', but since there is no overt recipient argument in (85b), let’s also assume that
the VP headed by sent simply comprises the verb sent and its theme argument a card,
as in (88) below:
(88)
DP
/ \
sent a card
(An alternative possibility would be to assume that there is an implicit recipient argument
in (88), represented by an empty category of some kind.) Since he is an agent argument in
(85b) He sent a card, and since we have analysed agent arguments as specifiers of an
agentive light verb, let’s assume that the VP in (88) merges with an agentive fight verb
whose agent subject is he, and the verb sent adjoins to the fight verb as in (89) below:
II I / \
sent 0 t a card
389
VP shells
This would mean that him would originate as the complement of a VP headed by hit,
and she as the subject of an agentive light verb to which hit adjoins, as in (91) below:
hit 0 t him
(Subsequently he raises to spec-TP for case-checking.) The agentive light verb 0 has a
performative sense, so that (91) can be paraphrased as ‘She performed the act of hitting
him.’ It may be that the (italicized) verb do found in so-called pseudo-cleft sentences
such as:
is an overt reflex of the kind of abstract agentive light verb found in (91). (A very dif¬
ferent analysis of the complements of monotransitive verbs as VP-specifiers is offered
in Stroik 1990 and Bowers 1993.)
Verbs like those italicized in (93) which have agent subjects but no direct object com¬
plement are referred to in the relevant technical literature as unergative predicates.
Some of these verbs also allow fo-phrase PP complements, as we see from structures
such as (94) below:
390
9.8 Unergative predicates
In accordance with the uniform theta assignment hypothesis, we might assume that
the to-phrase in structures like (94) has the same syntax as the to-phrase in structures
like (45) above, and thus serves as the specifier of a VP headed by a lexical verb.
However, this assumption poses something of a problem, in that the verbs in (94) have
no complement; hence, the inner VP core of a sentence like (94a) would appear to have
a structure along the lines of (95) below:
(95) VP
pp" V
/ \
to her V
I
lie
But the problem posed by (95) is that it is not a binary-branching structure, since V-bar
is a unary-branching constituent (its sole immediate constituent being the V lie). Thus,
sentences like (94) pose a further potential problem for our binary-branching frame¬
work. How can we overcome this problem?
A clue comes from the fact that unergative verbs like those in (93) generally have
verb + noun paraphrases: cf.
This might lead us to follow Baker 1988 and Hale and Keyser 1993 in positing that
unergative verbs like those in (93) are denominal predicates (i.e. verbs derived from
nouns) which are formed by incorporation of a noun complement into a verb. This
would mean (for example) that the verb lie in a sentence such as (94a) He may lie to
you is an implicitly transitive verb, formed by merging the noun lie with an abstract
verb 0 (with much the same sense as ‘tell’) as in (97a) below, and then incorporating
the noun into (i.e. adjoining it to) the verb as in (97b), so forming the lexical verb lie:
(97) (a) ,V
V "n
I I
0 lie
391
VP shells
The resulting V-bar in (97b) is then merged with the PP to you, forming the VP in (98)
below:
(It goes without saying that if we take the abstract verb 0 to be a light verb, it will pro¬
ject into v and vp rather than V and VP; however, if categories are sets of features, the
labels they carry are little more than labels of convenience, so little hangs on this
essentially notational issue.) The VP in (98) in turn merges with an agentive light verb
0 whose subject is he, and the verb lie raises to adjoin to the light verb as in (99) below:
lie 0 t t
A
Subsequently, the vp in (99) merges with a T constituent containing the auxiliary may,
and he raises to spec-TP to become the subject of may, so deriving (94a) He may lie to
you. The crucial assumption underlying the analysis in (99) is that verbs like lie are
implicitly transitive.
392
9.9 Unaccusative predicates
In some respects, the italicized arguments seem to behave like complements - for exam¬
ple, they occupy the postverbal position canonically associated with complements.
However, in other ways, they seem to behave like subjects: for instance, the italicized
argument agrees with the verb preceding it, as we see (for example) from the fact that
stands in (lOOe) is a singular form which agrees with the singular nominal a statue of
General Ghouly, so that we require the plural form stand if we are unfortunate enough
to have several statues of General Ghouly. Moreover, the postverbal argument arguably
carries the nominative case associated with subjects, not the objective/accusative case
associated with complements. This is clearer in languages where nouns carry overt
case-marking (cf. Vikner 1995), but is also suggested by (somewhat archaic) structures
such as:
(It should be noted, however, that Belletti 1988 suggests that unaccusative subjects
carry partitive case.)
Only certain types of verb seem to allow postverbal subjects, as we see from the fact
that structures such as those in (102) below are ungrammatical:
(102) (a) *When the British Rail snail arrived five hours late, there complained
many passengers
(b) *In the dentist’s surgery, there groaned a toothless patient
(c) *Every time General Wynott Nukem goes past, there salutes a guard at
the gate
(d) *There waved Wee Willie Widget at the window
(e) *There has apologized Major Muddle for his minor indiscretions
We might refer to verbs like those in (100) which can have postverbal subjects as
unaccusative predicates. By contrast, verbs with agent subjects but no overt object
like those in (102) are known as unergative predicates (as noted in §9.8).
In addition to the contrast illustrated in (100/102) above, there are a number of other
important syntactic differences between unaccusative verbs and other types of verb
(e.g. unergative verbs or transitive verbs). For example, Alison Henry (1995) notes that
in one dialect of Belfast English (which she refers to as dialect A) unaccusative verbs
can be used with (italicized) postverbal subjects in imperative structures like (103)
below:
By contrast, other (unergative or transitive) verbs don’t allow postverbal imperative sub¬
jects, so that imperatives such as (104) below are ungrammatical in the relevant dialect:
393
VP shells
There are also developmental differences between unaccusative verbs and others.
Amy Pierce (1992) observes that although children around two years of age correctly
position subjects in front of transitive or unergative verbs, they sometimes position
subjects after unaccusative verbs like come, go and fall - as we see from examples
such as the following (the information in parentheses indicates the names and ages in
years;months of the children producing the relevant utterances):
(g) Come Mommy. Come airplane. Here come tickle. Come Cromerl
(Adam 2;3-2;5)
(Young children often seem to use allgone as a variant of gone.) Why should it be that
they position unaccusative arguments postverbally? If we assume that the nature of
children’s initial grammars is determined by UG principles, it may be that children
assume that in consequence of the uniform theta assignment hypothesis/UTAH,
theme arguments are projected as complements - hence they are positioned postver¬
bally.
Additional evidence for positing that unaccusative verbs are syntactically distinct
from other verbs comes from auxiliary selection facts in relation to earlier stages of
English when there were two perfective auxiliaries, have and be, each taking a comple¬
ment headed by a specific kind of verb. The sentences in (106) below (taken from vari¬
ous plays by Shakespeare) give examples of verbs which could be used with the per¬
fective auxiliary be in Early Modem English:
(106) (a) Mistress Page is come with me (Mrs Ford, Merry Wives of Windsor, V.v)
(b) Is the duke gone? Then is your cause gone too (Duke, Measure for
Measure, V.i)
(c) How chance thou art returned so soon? (Antipholus, Comedy of Errors,
I.ii)
(d) I am arriv’d for fruitful Lombardy (Lucentio, Taming of the Shrew, I.i)
(e) You shall hear I am run away (Countess, All’s Well That Ends Well, Ill.ii)
(f) The heedful slave is wandered forth (Antipholus, Comedy of Errors, H.i)
394
9.9 Unaccusative predicates
(g) Don Pedro is approached (Messenger, Much Ado About Nothing, I.i)
(h) Did he not say my brother was fledl (Don Pedro, Much Ado About
Nothing, V.i)
(i) Thou told’st me they were stol’n into this wood (Demetrius, Midsummer
Night’s Dream, Il.i)
(j) She is fallen into a pit of ink (Leonato, Much Ado About Nothing, IV.i)
(k) And now is he become a man (Margaret, Much Ado About Nothing, IILiv)
(l) Why are you grown so rude? (Hermia, Midsummer Night ’s Dream, Ill.ii)
(107) (a) The train arrived at platform 4 is the 8.28 for London Euston
(b) They arrested a business man recently returned from Thailand
(c) Several facts recently come to light point to his guilt
(d) A number of objects gone from the church were found in his room
(e) OJ is something of a. fallen hero
In this respect, unaccusative verbs resemble passive participles, which can also be used
adjectivally (cf. a changed man, a battered wife, a woman arrested for shoplifting,
etc.).
We thus have a considerable body of empirical evidence that unaccusative subjects
behave like complements in certain respects. (Supporting evidence is available from
other languages: cf. e.g. Burzio 1986 on the syntax of ne ‘of it/of them’ in Italian, and
Contreras 1986 on bare nominals in Spanish.) We might therefore suppose (following
Burzio 1986) that unaccusative structures with postverbal arguments involve leaving
the relevant argument in situ in VP-complement position. Thus, we might posit that in
unaccusative expletive structures such as (100) above, in Belfast English unaccusative
imperatives such as (103) above and in Child English unaccusative declaratives/
395
VP shells
interrogatives such as (105) above, the italicized argument remains in situ in VP-
complement position. On this view, a sentence such as (lOOf) There have arisen several
problems would have a derivation along the lines of (109) below (in which several
(109)
(111)
\
Go you to school
396
9.9 Unaccusative predicates
If we suppose that causative have requires a complement whose subject has volitional
control (e.g. is an agent), then a natural conclusion to draw is that unaccusative verbs
But the obvious problem posed by a structure like (113) is that it provides us with no
way of accounting for the fact that unaccusative subjects surface postverbally in struc¬
tures such as (110) above. How can we overcome this problem?
397
VP shells
One suggestion might be the following. Let us suppose that unaccusative VPs like
(113) (i.e. VPs headed by an unaccusative verb) are embedded as the complement of an
outer vp shell headed by a strong v, and that the unaccusative verb raises to v in the
manner indicated by the arrow in (114) below:
t to school
(It may be that v is strong because it contains an affixal eventive light verb - i.e. a light
verb denoting an event - which has much the same sense as happen.) If we assume (as
Alison Henry argues) that subjects remain in situ in imperatives in dialect A of Belfast
English, the postverbal position of unaccusative subjects in sentences such as (110) can
be accounted for straightforwardly. And the vp shell analysis is consistent with the
assumption that the merger operation by which phrases are formed is intrinsically
binary.
Moreover, the vp shell analysis in (114) enables us to provide an interesting account
of the position of VP adverbs like quickly in unaccusative imperatives (in dialect A of
Belfast English) such as:
If we suppose that VP adverbs like quickly are adjuncts which merge with an interme¬
diate verbal projection (i.e. a single-bar projection comprising a verb and its comple¬
ment), we can say that quickly in (115) is adjoined to the V-bar go to school in (113).
What remains to be accounted for (in relation to the syntax of imperative subjects in
dialect A of Belfast English) is the fact that subjects of transitive and unergative verbs
occur in preverbal (not postverbal) position: cf.
(116) (a) You read that book! (b) *Read you that book!
(b) You protest! (b) *Protestyou!
Why should this be? If we assume (as in §9.7) that monotransitive verbs with agent
398
9.9 Unaccusative predicates
(117)
VP
DP
/ \
that book
The agent subject you will originate in spec-vp, as the subject of the agentive light
verb 0. Even after the verb read adjoins to the light verb 0, the subject you will still be
preverbal.
Given these assumptions, we could then say that the difference between unaccusative
subjects and transitive/unergative subjects is that unaccusative subjects originate in
spec-VP (as the subject of a lexical verb), whereas transitive/unergative subjects origi¬
nate in spec-vp (as the subject of an agentive light verb). If we assume that unaccusative
structures contain an outer vp shell headed by a strong v (e.g. an eventive light verb) and
an inner VP core headed by a lexical verb, and that the lexical verb raises from V to v,
the postverbal position of unaccusative subjects can be accounted for by positing that
the subject remains in situ in such structures.
The light verb analysis sketched here also offers us a way of accounting for the fact
that in Early Modem English, the perfective auxiliary used with unaccusative verbs
was be (as we see from the examples in (106) above), whereas that used with transitive
and unergative verbs was have. We could account for this by positing that the perfec¬
tive auxiliary have in EME selected a vp complement headed by an agentive light verb
with a thematic subject (much as causative have does in present-day English: cf. our
discussion of (112) above), whereas the perfective auxiliary be in EME selected a
complement headed by an eventive light verb which lacked a thematic subject. The
distinction has been lost in Modem English, with perfective have being used with
either type of vp complement (though sentences such as They are gone are a last
vestige of the earlier use of be as a perfective auxiliary).
The analysis of unaccusative verbs outlined here also enables us to account for
expletive unaccusative structures such as the following:
We might suppose that a cry of anguish in (118) is in spec-VP, that came originates in
V and raises to v, and that there originates in the nonthematic spec-vp position and
from there raises to spec-TP, as in (119) below:
399
VP shells
(120) (a) Every so often, there comes/*come a cry of anguish from inside the house
(b) Every so often, there come/*comes cries of anguish from inside the house
For alternative analyses of expletive structures, see Authier 1991, Lasnik 1992, 1995,
Chomsky 1995b, Groat 1995, Rothstein 1995 and Vikner 1995.
Alongside expletive unaccusative structures such as (118) we also find nonexpletive
structures such as:
In this type of sentence, the need for T to have a subject is satisfied not by the use of
expletive there but rather by raising the subject a cry of anguish from spec-VP through
spec-vp into spec-TP, as in (122) below:
(122) TP
400
9.10 Summary
Thus, the subject a cry of anguish remains in situ in expletive structures such as (119),
but raises to spec-TP in structures such as (122).
9.10 Summary
We began this chapter in §9.2 by looking at the syntax of ergative verbs
like roll which can be used both transitively and intransitively. We suggested that in
intransitive structures like The ball rolled down the hill, the theme argument of roll
(= the DP the ball) originates as the subject/specifier of a VP headed by the lexical verb
roll, and then raises up into spec-TP for case-checking purposes. We noted that in tran¬
sitive structures such as We rolled the ball down the hill, the verb roll appears to have
two complements (namely the DP the ball and the PP down the hill), and that this poses
problems for the assumption that syntactic structures are binary-branching. We
suggested that these problems could be solved if we assume that the ball (in transitive
structures like We rolled the ball down the hill) originates as the subject/specifier of a
VP headed by the lexical verb roll, and that this VP serves as the complement of a null
causative light verb 0 which has an agent subject (we), with the verb roll raising to
adjoin to the light verb (deriving a structure paraphraseable as ‘We made + roll the ball
down the hill’). We noted that such an analysis would be compatible with Baker’s
uniformity of theta assignment hypothesis, since both in transitive structures like We
rolled the ball down the hill and in intransitive structures like The ball rolled down the
hill the theme argument (the ball) would originate as the subject and specifier of a VP
headed by the lexical verb roll. In §9.3 we argued that the light-verb analysis would
also enable us to provide a straightforward account of the dual position occupied by the
adverb gently in sentences such as We gently rolled the ball down the hill and We rolled
the ball gently down the hill, with gently modifying either a V-bar headed by a lexical
verb, or a v-bar headed by a light verb. We also suggested that a light-verb analysis
would provide a principled account of the dual position occupied by the prepositional
particle down in sentences such as They closed the store down and They closed down
the store, if we assumed that either the simple verb closed or the complex verb closed
down could adjoin to an abstract causative light verb. In §9.4 we looked at the syntax
of ditransitive verbs like get as they are used in double-object constructions such as
They will get the teacher a present. We suggested that the teacher originates as the sub¬
ject of a VP headed by get (precisely as in The teacher got a present), and that the
resulting VP serves as the complement of a causative light verb with an agent subject
(they), with the verb get raising to adjoin to the causative light verb, and the subject
(we) of the light verb raising to become the subject of will (so deriving They will get
the teacher a present). We then examined the syntax of resultative structures such as
The acid turned the litmus paper red. We argued that the DP the litmus paper originates
as the subject of the V-bar turned red, and that the resulting VP the litmus paper turned
red merges with a causative light verb 0 (whose subject is the acid), with the verb
401
VP shells
turned adjoining to the causative light verb, so deriving a structure which can be para¬
phrased as ‘The acid made + turn the litmus paper red.’ (The DP the acid subsequently
raises to spec-TP.) In §9.5 we discussed the syntax of three-place predicates which
have a clausal complement. We argued that in a sentence like He had remarked to her
that Senator Scumme-Bagge was a fraud the that-clause serves as the complement of a
VP headed by the lexical verb remarked, and the ta-phrase serves as its specifier; we
hypothesized that the VP serves as the complement of an agentive light verb whose
subject is he, and that the verb remarked adjoins to this light verb thereby forming a
structure which can loosely be paraphrased as ‘He made + remark to her that Senator
Scumme-Bagge was a fraud.’ Subsequently, the pronoun he raises to spec-TP to
become the subject of the auxiliary had, so deriving He had remarked to her that
Senator Scumme-Bagge was a fraud. We suggested a similar derivation for sentences
such as You must satisfy the jury that you are innocent, in which the r/zat-clause serves
as the complement of a VP headed by the lexical verb satisfy, and the DP the jury
serves as its specifier, with the subject you originating as the subject of an agentive
light verb (to which satisfy adjoins), and then being raised up to become the subject of
the auxiliary must. In §9.6 we examined the syntax of object-control structures such as
What decided you to take syntax?, noting that under the traditional analysis the verb
decided has two complements, the object pronoun you and the infinitive complement to
take syntax. We proposed an alternative analysis under which you originates as the
subject of decided to take syntax, and the resulting VP you decided to take syntax
serves as the complement of a causative light verb 0 whose subject is what, with
decided adjoining to the light verb (and what raising to spec-TP) to derive What
decided you to take syntax? We went on to examine differences between object-control
predicates like persuade and ECM (exceptional case-marking) predicates like expect,
noting that they differ in the type of r/iat-clause paraphrase they allow (He persuaded
me to take syntax is paraphraseable as ‘He persuaded me that I should take syntax’,
whereas He expected me to take syntax is paraphraseable as ‘He expected that I would
take syntax’), in whether they constrain the choice of DP following persuade!expect
(cf. He expected! ^persuaded the exam to be difficult), and in whether they permit an
expletive + infinitive structure (cf. He expected!*persuaded there to be someone wait¬
ing for him). In §9.7 we outlined an analysis of monotransitive verbs (like sent in He
sent a card) in which the theme argument (a card) originates as the complement of a
VP headed by a lexical verb (sent), and the subject (he) originates as the agent argu¬
ment (and specifier) of a vp headed by an agentive light verb. In §9.8 we suggested that
unergative verbs like lie are derived via incorporation of a noun into an abstract verb.
In §9.9 we turned to look at a special class of intransitive verbs called unaccusative
predicates (like come, go, occur, arise, appear, etc.). We noted that the subjects of such
verbs behave in certain respects like complements: for example, they are positioned
after the verb in expletive structures (cf. There arose an unfortunate misunderstand-
402
Workbook section
ing), and in imperative structures in dialect A of Belfast English (cf. Leave you now!).
We outlined the traditional analysis of unaccusative structures, under which the ‘sub¬
ject’ originates as the complement of the unaccusative verb, and remains in situ in
expletive structures such as There will come a time when you are sorry. We noted,
however, that such an analysis faces obvious problems in relation to unaccusative
predicates which have two arguments (e.g. go in John went to school), since the com¬
plement analysis would require us to suppose that such verbs have two complements
(i.e. John and to school in the case of John went to school) - an assumption which is
incompatible with our hypothesis that constituent structure is binary-branching. We
therefore outlined an alternative light-verb analysis of unaccusative predicates under
which unaccusative verbs and their arguments originate within a VP which is embed¬
ded as the complement of an eventive light verb (with much the same sense as
‘happen’), and the unaccusative verb raises up to adjoin to the light verb (thereby mov¬
ing into a position where it precedes the subject in spec-VP). We noted that such an
analysis would provide an account of the fact that unaccusative verbs are positioned in
front of both their arguments e.g. in Belfast English imperatives such as Go you to
school! (if we assume, following Alison Henry, that imperative subjects remain in situ
Workbook section
Exercise I (§§9.2-9.5)
Discuss the derivation of the following sentences, saying why each
derivation crashes or converges:
1a It kept warm
b She kept it warm
2a I’m going crazy
b You’re driving me crazy
3a He deliberately spilled the milk over the floor
b *The milk deliberately spilled over the floor
4a He carefully reversed the car into the garage
b He reversed the car carefully into the garage
5a He drives the car well round bends
b *He well drives the car round bends
6a The building blew up
b They blew the building up
7a Taxes went up
b They put up taxes
8a He turned the key slowly round
b He slowly turned round the key
403
VP shells
it V A
kept warm
The resulting VP merges with a tense affix T to form a T-bar; the subject it then raises
to spec-TP to check its nominative case, as in (ii) below:
t V A
kept warm
The subject it carries nominative case by virtue of being the subject of a finite T con¬
stituent.
Sentence lb has the following derivation. As before, the verb kept merges with
the adjective warm to form the V-bar kept warm', the subject it merges with V-bar to
form the VP (i) above. The resulting VP then merges with a causative light verb 0
with an agent subject (she), and the verb kept adjoins to the light verb, as in (iii)
below:
t warm
404
Workbook section
The objective case of it is checked by the immediately preceding transitive light verb 0
(or by the verb kept, which is transitive in this use). The vp in (iii) then merges with an
abstract tense constituent to form a T-bar, and the pronoun she raises to spec-TP to
check its nominative case, as in (iv) below:
kept 0 it V A
t warm
An analysis such as that outlined here is consistent with the uniform theta assignment
hypothesis in that in both la and lb the theme argument of kept (= it) originates as the
subject/specifier of a VP headed by the lexical verb kept.
Exercise II (§9.6)
Discuss the syntax of the following infinitive structures:
405
VP shells
Moreover, if we have an adverb like desperately following want, we can use the com¬
plementizer/or, and the pronoun you follows for (and thus is clearly the subject of the
[bracketed] complement clause introduced by for, not the object of want): cf.
(iii) (a) She wanted her uncle to shave his beard off
(b) She wanted her goldfish to swim faster
(c) She wanted her pet amoeba to reproduce
(d) She wanted the kettle to boil
(e) She wanted her theory to be foolproof
Since control verbs impose restrictions on the choice of postverbal expression follow¬
ing them (but ECM verbs do not), it seems likely that want is an ECM verb.
If this is so, sentence 1 will be derived as follows (if we ignore the later discussion
of monotransitive verbs in §9.7). The pronoun you originates as the subject of read it,
as in (vi) below:
406
Workbook section
(vii) TP
t V D
I
read it
The TP in (vii) is merged with the verb wanted, forming a V-bar which is further pro¬
jected into the VP in (viii) below by merger with the subject she'.
(viii)
read it
The objective case carried by you is checked by the immediately preceding transitive
verb wanted. Subsequently, the VP in (viii) merges with an abstract T constituent, and
the subject she raises to spec-TP to check its nominative case, as in (ix) below:
407
VP shells
t V D
read it
1 It would surprise me
2 He sold the television
3 She received a letter
4 The rhino may charge
5 He wants her to slave for him
6 She may fear him
7 They could fine him
8 He should shave
9 He might appeal to the Supreme Court
10 They disgust me
11 It could snow
12 You can’t/oo/me
13 She should kick him
14 He can cash the cheque
15 They should fire him
16 They may jail him
17 It will anger him
18 The government should tax the opposition
19 l pity you
20 He could poison the wine
Note that there are a number of different predicate types represented above; different
analyses will be appropriate for different types of predicate. Which examples are the
most problematic, and why?
408
Workbook section
If personally modifies experiencer subjects, this suggests that me must originate as the
subject of the verb surprise. But if subjects are specifiers which merge with a V-bar con¬
stituent, it follows that the verb surprise must be a V-bar constituent with a ‘hidden’ com¬
plement of some kind. How can this be? One possibility is that the verb surprise is derived
by incorporating the noun surprise into an abstract experiential verb 0, and so in effect has
much the same meaning as ‘experience a feeling of surprise’. If this is so, then 1 has a
derivation which is similar in certain respects to that of It would make me feel surprise.
If we follow this line of reasoning, 1 will be derived as follows. The experiential
verb 0 is merged with the noun surprise to form the V-bar (ii) below (which has much
the same meaning as ‘experience surprise’):
V N
0 surprise
(Of course, if we treat 0 as a light verb, it will have the status of v rather than V,
and will project into v rather than V; however, since category labels are abbreviations
for sets of grammatical features, little of any consequence hangs on the category labels
we use here.) The noun surprise is then incorporated into (i.e. adjoined to the left of)
the affixal verb 0 as in (iii) below:
(iii) V
The resulting V-bar is then merged with the subject me as in (iv) below, and the subject
is indirectly assigned the 0-role of experiencer by the experiential verb 0:
409
VP shells
The VP thereby formed is then merged with a causative light verb 0 (whose subject is
it), and the (complex) verb surprised is then adjoined to the higher (causative) light
surprise 0 t t
* >
The vp in (v) is then merged with a T constituent containing the auxiliary would (and a
past-tense affix, which we ignore below), and the subject it raises to spec-TP to check
its nominative case, as in (vi) below:
The nominative case carried by the pronoun it is checked by the tense morpheme T,
and the objective case carried by me is checked by the transitive agentive light verb 0.
(Note that an issue not addressed above is how we account for the fact that the verb
surprise as used in 1 is an infinitive form.)
Exercise IV (§9.8)
Discuss the syntax of the italicized verbs as they are used in the sentences
below:
410
Workbook section
(i) V
/
V
I
0 blind
(Of course, if we analyse 0 as an inchoative light verb, it will carry the label v rather
than V, and project into v-bar rather than V-bar.) The adjective blind then incorporates
into (i.e. adjoins to) the abstract inchoative verb 0, and you merges with V as in (ii)
below:
(ii)
(a structure paraphraseable as ‘you become blind’). The VP in (ii) is then merged with
a causative light verb 0 whose subject it has the thematic role of causer; the deadjecti¬
val verb blind+Q then adjoins to the causative light-verb, as in (iii) below:
411
VP shells
A V 0 you V A
blind 0 t /
▲ 1
(where the verb blind is paraphraseable as ‘make become blind ). The vp in (iii) then
merges with a T constituent containing the modal auxiliary can, and the subject it
raises to spec-TP to become the subject of can, as in (iv) below:
(iv) tp
blind 0 t t
The nominative case carried by it is checked by T, and the objective case carried by
you is checked by the immediately preceding transitive causative light verb 0.
Exercise V (§9.8)
Discuss the syntax of verbs formed with the suffixes listed below:
412
Workbook section
It cannot, however, be used with a direct object DP, as we see from the ungrammatical-
ity of (ii) below:
One reason why apologize cannot be used with a direct object DP may be that it is
itself a denominal verb, formed by incorporating the noun object apology into the suf-
fixal verb +ize. If we assume (as in (45) in the main text) that the fo-phrase in (i) is the
specifier of the VP headed by apologize, sentence (i) will be derived as follows.
The suffixal verb +iz,e is merged with the noun apology, forming the V-bar in (iii)
below (paraphraseable as ‘offer an apology’)
(iii)
V N
I I
+ize apology
The noun apology incorporates into (i.e. is adjoined to) the suffixal verb +ize, and in so
doing loses its final y, as in (iv) below:
(iv) .Vx
,Vs N
I
N t
I
apolog
apolog ize
The VP in (v) merges with an agentive light verb 0, whose subject is he\ the verb apol¬
ogize raises to adjoin to this light verb as in (vi) below:
413
VP shells
The vp in (vi) then merges with a T constituent containing the modal auxiliary should,
and he raises to spec-TP to become the subject of should as in (vii) below:
V v PP^
I I / \ / \
apoIog+ize 0 to her V N
II
t t
(To simplify exposition, we ignore the possibility that T also contains an abstract past-
tense morpheme; we also leave aside the question of how we account for the fact that
apologize as used here is an infinitive form.)
The expression horrify you is paraphraseable as ‘make you feel horror’, so that horrify
can plausibly be taken to be a denominal verb. One way of deriving (viii) is to suppose
that the suffixal verb +ify is used as a causative experiential verb (i.e. a verb para¬
phraseable as ‘cause to experience’) which merges with the noun horror to form the
V-bar(ix) below:
V N
+ify horror
The noun horror then incorporates into the verb +ify (in the process losing its final or
segments), as in (x) below:
horr ify
414
Workbook section
The V-bar in (xi) merges with its experiencer subject you, so forming the VP (xi)
below:
This VP in turn merges with an abstract causative light verb 0 whose subject it has the the¬
matic role of causer, and the verb horrify raises to adjoin to the light verb as in (xii) below:
horr ify t t
A
Finally, the vp in (xii) merges with a T constituent containing the modal auxiliary will,
and the pronoun it raises to spec-TP to become the subject of will as in (xiii) below:
N V 0 you V N
horr ify t t
415
VP shells
Since thicken (in this use) is paraphraseable as ‘become thick’, we might suppose that
+en functions as a suffixal inchoative verb (i.e. a verb which indicates the beginning of
a state or action) paraphraseable as ‘begin to be’, ‘come to be’ or ‘become’. This would
mean that (xiv) is derived as follows. The inchoative verb +en originates in the head V
position of VP, and merges with the adjective thick as in (xv) below:
(xv) -VV
A
I I
+en thick
The resulting V-bar is paraphraseable as ‘become thick’. The adjective thick then
adjoins to the affixal verb +en, as in (xvi) below:
(xvi) vx
A V t
thick +en
A
The V-bar in (xvi) merges with the DP the soup, so forming the VP (xvii) below:
thick +en
The resulting VP then merges with a T constituent containing will (and an abstract
tense affix, which we ignore below); the DP the soup raises to spec-TP to become the
subject of may as in (xviii) below:
thick +en
416
Workbook section
One salient feature of the three model answers sketched out above is the assumption
that the suffixes +ize, +ify and +en originate as independent items from the nouns or
adjectives which incorporate into them (e.g. horrify is treated as composed of two
separate words, a noun horr(or) and a verb +ify). An alternative possibility would be to
treat a word like horrify as a unitary lexical item (i.e. as a single word) which origi¬
nates as an A, then raises to adjoin to a null experiential verb and subsequently raises to
adjoin to a null causative light verb. The other two model answers presented here could
be revised along similar lines.
Exercise VI (§9.9)
Discuss the derivation of the following sentences, saying why each
derivation crashes or converges. (The sentences in 3 are from dialect A of Belfast
English; those in 4 were produced by two-year-old children; those in 5 are from plays
by Shakespeare.)
And in Shakespearean English, we find it used with the auxiliary be: cf.
(iv) The time is come even now (Duke, Measure for Measure, IV.i)
417
VP shells
the rain V p
came down
The VP in (v) then merges with an abstract eventive light verb 0; the verb came raises
to adjoin to the light verb, as in (vi) below:
The prepositional particle down then moves into the vacant (nonthematic) spec-vp
position (perhaps to satisfy a requirement that vp have a specifier) as in (vii) below:
Subsequently, the vp in (vii) merges with an abstract tense morpheme, and the pre¬
positional particle down moves from spec-vp into spec-TP as in (viii) below, thereby
satisfying the requirement for spec-TP to be filled:
418
Workbook section
t t
Although the prepositional particle down occupies spec-TP, it is clearly not the subject
of the overall clause. For one thing, it does not permit subject-auxiliary inversion, as
we see from the ungrammatically of (ix) below:
From this, we can conclude that there is no specific requirement for TP (or vp) to have
a subject, but merely a requirement for TP (and vp) to have a specifier. The shortest
movement principle will determine that the constituent which moves into spec-TP is
that which occupies spec-vp. With agentive verbs, spec-vp is always occupied by a the¬
matic (agent) subject, and so it is this subject which moves into spec-TP; but with
unaccusative verbs, spec-vp is a 0-bar position and so can be filled by a nonthematic
element like expletive there or (as in this case) a preposed prepositional particle.
Needless to say, there are numerous questions of detail which remain to be worked out
in relation to the derivation in (viii). One relates to the case carried by the DP subject the
rain: does it carry partitive case (checked by the immediately preceding unaccusative
verb came), or nominative case (checked by attraction of the relevant case-feature to T)?
419
VP shells
Another relates to whether the preposed prepositional particle down actually ends up in
spec-TP (as claimed above), or in spec-CP. The fact that sentence la can itself serve as
the complement of a complementizer like that, e.g. in a sentence such as:
(xi) He was upset to find that [down came the rain], washing away the seeds
he had just planted
makes it less likely that down occupies spec-CP, and more likely that it occupies spec-
TP.
Discuss the derivation of the relevant sentences, and the nature of the errors made by
the children.
(i) VP
420
Workbook section
Let’s further assume that the VP in (i) is merged with a strong causative light verb 0
(whose agent subject is she), and that came raises to adjoin to 0 as in (ii) below:
came 0 it V PP
t It / \
Qver ^gj-g
Subsequently, the subject she raises to spec-TP in order to check its nominative case, as
in (iii) below:
(iii) XP
came 0 it V PP
I X \
t over there
However, since the corresponding sentence She came it over there is ungrammatical in
adult English (instead, we say She took it over there), an important question to ask is
‘What’s wrong with sentences like 1 in adult English?’
One minor error in 1 relates to the use of the locative adverb there in place of here.
(Generally, we use here with the verb come and there with the verb go: cf. Come here!
Go there!) However, the error which is our main concern here is employing the motion
verb come in a causative use. Why can’t come be used causatively in adult English (so
deriving sentences like *Come it here!)? One answer might be that come is a nonaffixal
verb, and hence cannot be adjoined to a causative light verb like 0: an alternative possi¬
bility would be to suppose that come lacks the causative feature which would enable it
to be adjoined to a causative light verb; a third (more traditional analysis) is to say that
come is intransitive (and so could not check the objective case of it if used in causative
structures such as (iii) above). On the first view, the child’s error lies in not having
learned which verbs are (and which aren’t) affixal stems; on the second, it lies in not
having identified which verbs do (or don’t) carry the relevant causative feature; on the
third, it lies in not having identified which verbs are transitive (and so carry an uninter¬
pretable objective case-feature which needs to be checked) and which are intransitive.
421
10
Agreement projections
10.1 Overview
In this chapter, we take a closer look at the internal structure of clauses,
examining the range of projections which they contain. We shall argue that clauses
have a much more richly articulated constituent structure than we have hitherto sup¬
posed, and that they contain subject and object agreement projections.
At first sight, (1) seems relatively unproblematic. After all, we might suppose that all is
a floating quantifier stranded in spec-vp (by movement of they into spec-TP), and prob¬
ably is an adverb adjoined to vp, so that (1) has the simplified derivation (2) below.
(Throughout this chapter, the internal structure of complex heads is sometimes simpli¬
fied by omitting the category label of one or more of the heads, where this is self-
evident: hence have is not labelled PERF here.)
i_i
Of course, the analysis in (2) requires us to posit that adverbs like probably adjoin to
maximal projections rather than (as assumed in earlier chapters) to intermediate pro-
422
10.2 Subject agreement projections
jections. However, there are theoretical reasons for supposing that this is a plausible
assumption.
The relevant considerations relate to the traditional assumption that languages
contain movement rules which adjoin one constituent to another (for obvious reasons,
the relevant type of movement operation is referred to as adjunction). For example, a
traditional way of handling the relation between sentence pairs such as (3a-b) below:
is to suppose that in (3b) the word not (in the guise of its contracted form n’t) adjoins to
the auxiliary have, so forming haven't (which behaves like a single word e.g. in respect
of undergoing inversion in sentences like Haven’t they understood anything?).
Similarly, one way of describing what happens when the italicized phrase such behav¬
iour is preposed (in order to focus it in some way) in a sentence pair such as:
(4) (a) You must know that we cannot tolerate such behaviour
(b) You must know that such behaviour we cannot tolerate
is to say that the DP such behaviour adjoins to the TP headed by cannot (cf. Grimshaw
1993a). We might then conclude (as Chomsky does in his Barriers monograph, 1986b)
that the only type of adjunction operations which can result from movement are
adjunction of one head to another or of one maximal projection to another. It would
therefore seem natural to suppose that the same is true of the merger operations which
attach adverbial adjuncts to the expressions they modify: and this in turn would rule
out the possibility of merging an adverb with an intermediate projection, but would
allow for structures like (2) in which probably (which is itself a maximal projection in
(2) above) is an adjunct to the maximal projection vp.
However, the analysis in (2) proves to be descriptively inadequate for a number of
reasons. For one thing, it assumes that adverbs like probably can serve as vp-adjuncts.
Yet if this were so, we’d expect that probably could be positioned before or after
another vp-adjunct like completely, given the traditional assumption that adjuncts of
the same kind can be freely ordered with respect to one another. However, this is not
so, as we see from sentences such as:
The fact that probably must be positioned to the left of the vp-adverb completely sug¬
gests that it is a different kind of adverb altogether. More concretely, it seems plausible
to posit that probably is a TP-adverb - i.e. an adverb which merges with a TP to form
an extended TP. But, of course, if probably is not a vp-adverb, the analysis in (2) can¬
not be right.
423
Agreement projections
Further support for the claim that probably is not a vp-adverb comes from sentences
like (6) below:
Given the traditional assumption that the NEGP constituent containing not is positioned
between TP and vp, it follows that probably cannot be adjoined to vp in sentences like
(6).
The problems are compounded when we come to consider how to deal with sentences
such as:
It is less than obvious how we might deal with sentences like (7) if we continue to
assume that auxiliaries like have are generated in T (merged with a tense morpheme),
and that floating quantifiers are stranded in subject QPs: after all, how can they and the
QP containing all both be subjects of the same auxiliary have?
Given the conventional assumption that each auxiliary permits only one subject, the
answer is that they can’t. So, an alternative possibility which we might pursue is that
there are in fact two different functional projections between CP and vp, with they
serving as the subject of one of them, and the QP containing the stranded quantifier all
as the subject of the other (a claim which amounts to positing that there are two differ¬
ent auxiliary positions in clauses). Since auxiliaries like have/be typically inflect for
tense and agreement, and since we have already suggested that tense heads its own
projection into a tense phrase (TP), an obvious suggestion to make (following Pollock
1989, Belletti 1990 and Chomsky 1993) is that finite clauses also contain an abstract
agreement morpheme which projects into an agreement phrase. Let us also suppose
(following Belletti and Chomsky) that the relevant agreement head occupies a higher
position than T, and that auxiliaries are generated in T and from there can raise to
adjoin to the separate agreement head, and that subjects raise from spec-vp to the spec¬
ifier position in the agreement phrase, to check their case and agreement features.
Since the agreement relation in question involves subjects, it has become conventional
in the relevant literature to denote the relevant subject agreement constituent as AgrS
(and to its maximal projection as AgrSP). If we go down this road, our earlier INFL
head will in effect be split into two different heads - a T head and an AgrS head: hence,
for obvious reasons, this assumption has become known as the split-INFL hypothesis.
Within the split-INFL framework, a sentence such as (1) They have probably all
given up smoking might be derived as follows. The QP all they originates in spec-vp
424
10.2 Subject agreement projections
(as the subject of given up smoking), and from there raises up into spec-TP; the adverb
probably merges with TP to form an extended TP, as in (8) below:
(8)
given up smoking
TP is then merged with an AgrS (= subject agreement) head which projects into AgrSP
(= subject agreement phrase); the pronoun they raises to spec-AgrSP (i.e. into the spec¬
ifier position within the subject agreement phrase) to check its nominative case, and the
auxiliary have raises from T to adjoin to AgrS (thereby enabling it to check its agree¬
ment properties) as in (9) below (where the righthand trace is left behind by movement
of the QP all they from spec-vp to spec-TP):
(9) AgrSP
given up smoking
Since there are two functional projections above vp in (9) (AgrSP and TP), it follows
that there are two different subject positions outside vp, one of which (= spec-AgrSP)
houses the subject pronoun they, and the other of which (= spec-TP) houses the QP
containing the stranded quantifier all.
We might propose to derive (7) They probably all have given up smoking in essentially
the same way, except that the auxiliary have remains in the head T position of TP, and does
not adjoin to AgrS. If this is so, (7) will have the (simplified) derivation (10) below (where
the righthand trace is left behind by movement of the QP all they from spec-vp to spec-TP):
425
Agreement projections
(10) Agr SP
D" AgrSP
I
They AgrS
A
ADV
I
probably
t given up smoking
l
The fact that the auxiliary have is positioned after the TP-adverb probably in (10) but
before it in (9) would suggest that have occupies the head T position of TP in (10), and
moves from there to occupy the head AgrS position of AgrSP in (9) (as argued by
Koizumi 1995, p. 41). We might follow Boskovic (1995, p. 22) in supposing that finite
auxiliaries in English can have either strong or weak agreement-features, and raise
to AgrS when they have strong agreement-features, but remain in T when they have
weak agreement-features. If we assume that AgrS has strong specifier-features in
English, it follows that subjects will always raise to spec-AgrSP to check their case-
and agreement-features.
A further argument in support of the split-INFL analysis comes from facts relating to
the relative scope of adverbs and modal auxiliaries. Ernst (1991, p. 754) notes that in a
sentence such as:
the adverb apparently has scope over the modal can, as we see from the fact that (11)
can be paraphrased as ‘It is apparent that Gary can lift 100 pounds.’ If we make the
traditional assumption that scope relations are defined in terms of the relation c-
command (so that X has scope over Y only if X c-commands Y), it is difficult to account
for the fact that can falls within the scope of apparently in (11), since apparently does
not c-command can. How can we resolve this problem? One answer is to suppose that
can originates in T and from there raises to AgrS across the TP adverb apparently, as in
(12) below (where each italicized t denotes a trace of the moved subject he):
(12) AgrSP
426
10.3 Other varieties of English
If we follow Ernst (1991, p. 753) in positing that scope is determined by the following
scope principle (the term operator here is used to denote a constituent with scope
properties):
then we can say that apparently has scope over can in (12) by virtue of the fact that
apparently c-commands the trace of can which occupies the head T position of TP.
(14) (a) Thy physic I will try (King, All’s Well That Ends Well, Il.i)
(b) She may more suitors have (Tranio, The Taming of the Shrew, I.ii)
(c) The king your mote did see (Boyet, Love’s Labour’s Lost, IV.i)
427
Agreement projections
(15) AgrSP
AgrSP
The king
A
Of course, given the assumption that scrambling in (15) involves adjunction of the
scrambled DP to TP, we might propose to analyse (14b) in a parallel fashion, with the
auxiliary may moving from T to AgrS (hence preceding the scrambled nominal more
suitors).
Alison Henry (1995) presents an interesting argument from the syntax of Belfast
English (BE) in support of the AgrSP/TP analysis of clause structure presented above.
She notes that in BE, we find sentences such as (16) below in which (what appears to
be) a singular verb form is used with a plural objective subject:
She argues that the +s inflection on verb forms like is/was/has/does/goes in BE struc¬
tures like (16) marks present tense (as indeed is the case in many other varieties of
English - e.g. in South-Western British English, where we find paradigms such as
I/we/you/he/she/they hates syntax), and that consequently there is no morphological
marking of agreement in structures like (16). She notes that absence of agreement¬
marking correlates with the assignment of objective case to the subject of the verb.
(Where a nominative subject is used, the verb must obligatorily agree with the subject,
as in They are/*is working hard.) She also notes that agreementless finite verb forms
like those in (16) cannot undergo auxiliary inversion - as we see from the ungrammati-
cality of questions such as:
428
10.3 Other varieties of English
t annoying youse
Consider now why the interrogative counterpart (17a) *Is themuns annoying youse? is
ungrammatical. If we suppose that questions are CP/AgrSP/TP/vp/VP structures and
involve adjunction of an auxiliary to an abstract Q morpheme in the head C position of
CP, there are two ways in which we might seek to move the auxiliary is from T to C (to
adjoin to Q). One is in two short steps, adjoining is first to AgrS (as in movement (1)
below) and then to Q (as in movement (2) below):
t annoying youse
(As before, the italicized t is the trace of the subject themuns which moves from spec-
vp to spec-TP; Op is the abstract yes-no question operator which we posited in §7.8.)
429
Agreement projections
However, the successive cyclic derivation in (19) is ruled out by virtue of the fact that
AgrS is weak in agreementless finite clauses, so step (1) of the derivation violates the
principle of greed (which licenses movement only as a way of checking strong fea¬
tures).
A second way in which we might seek to derive questions like (17a) is to move is
directly from T to adjoin to Q, as in (20) below:
t annoying youse
If we suppose that only items which carry agreement properties license nominative
subjects, and can raise to AgrS (and thence to C in inversion structures), it follows that
we must assume that modals like can have covert agreement properties in MSE. If (like
auxiliaries in Belfast English) they were agreementless forms, we should (wrongly)
predict that they don’t allow nominative subjects, and can’t undergo inversion. (Recall
that Boskovic 1995 assumes that finite auxiliaries in MSE may have either strong or
weak agreement-features, and so can either stay in T or move to AgrS.) Rather than say
430
10.4 Object agreement projections
that modals like can have no (person/number) agreement properties, we might there¬
fore conclude that they have the variable agreement properties [aPerson, aNumber],
and that these variable person/number-features can be checked by a first, second or
third person subject which is singular or plural in number. (See Rooryck 1994 for dis¬
cussion of the use of feature variables in syntax and phonology.)
(22) (a) He reported to the police that there had been a robbery
(b) He reported the robbery to the police
(23) (a) He admitted to her that he was guilty
(b) He admitted his guilt to her
(24) (a) He announced to the press that he was retiring
(b) He announced his retirement to the press
(25) (a) He mentioned to his boss that the miners were on strike
(b) He mentioned the strike to his boss
(26) (a) He recommended to her that she should consult Cy Coe
(b) He recommended an analyst to her
(27) (a) He whispered to her that he loved her
(b) He whispered sweet nothings to her
431
Agreement projections
If that-clause complements don’t carry case, it seems reasonable to assume that they
don’t move for case-checking purposes, but rather remain in situ. This being so, con¬
sider how we account for the clause-final position of the that-clause in a sentence such
as (22a) He reported to the police that there had been a robbery.
Given the VP shell analysis outlined in the previous chapter, (22a) will be derived as
follows. The verb reported merges with its theme argument (the CP that there had
been a robbery) to form a V-bar; the resulting V-bar reported that there had been a rob¬
bery merges with its recipient argument (the PP to the police) to form the VP (29)
below:
The VP in (29) then merges with a performative light verb 0 whose agent subject is
he, and the verb reported adjoins to the light verb as in (30) below:
(30)
Subsequently, the subject he raises through spec-TP to spec-AgrSP to check its nomi¬
native case. A crucial assumption underlying the derivation in (30) is that the CP that
there had been a robbery remains in situ throughout.
432
10.4 Object agreement projections
If we adopt the uniform theta assignment hypothesis, it follows that the DP com¬
plement the robbery in (22b) He reported the robbery to the police must originate in
the same position as the CP complement that there had been a robbery in (30), since
the relevant DP and CP constituents play the same thematic role as the theme argument
of reported. But since the DP the robbery ends up in a position between the verb
reported and the PP to the police, it must subsequently move to some higher position
between the vp containing the verb reported in (30) and the VP containing the PP to the
police. By hypothesis, the higher position which the robbery moves to is the specifier
position within an AgrOP projection positioned between vp and VP. If this is so, (22b)
will have the following derivation.
The verb reported merges with its DP complement the robbery to form a V-bar; the
resulting V-bar reported the robbery then merges with the PP to the police to form the
VP below:
(31)
<
*0
\
<\
PP
/
^ DP
-<
to the police
(32) Agr OP
DP
the robbery
A
reported DP
A
I
t
Since the DP the robbery and the AgrO constituent containing the transitive verb
reported are in a spec-head relation in (32), the objective case-feature carried by each
can be checked against that of the other, and erased. Subsequently, the AgrOP con¬
stituent in (32) merges with a performative light verb 0 (whose agent subject is he)
and the verb reported adjoins to this light verb as in (33) below:
433
Agreement projections
AgrOP.,
AgrO
t to the police V DP
I
t t
(An alternative possibility is that AgrO is pied-piped along with the verb, so that the
whole V+AgrO constituent adjoins to the light verb 0, rather than just the V reported.)
Subsequently, the subject he raises through spec-TP to spec-AgrSP to check its nomi¬
native case.
An analysis along the lines sketched out above has both empirical and theoretical
merits. From an empirical point of view, its main merit is that it accounts for why DP
objects are postverbal, whereas CP complements are clause-final - i.e. it accounts for
the word-order contrasts illustrated in (22-7) above. On the analysis presented here,
this follows from the fact that DP complements carry objective case and hence have to
raise to spec-AgrOP for case-checking; by contrast, CP complements are caseless and
so need not (and therefore, by the economy principle, cannot) raise to spec-AgrOP.
(We might also assume that the economy principle determines that clauses which have
no DP object do not contain an AgrOP projection at all - as indeed we tacitly assumed
earlier in relation to the derivation sketched out in (30) above.)
From a theoretical point of view, the main merit of the AgrOP analysis in (33) is that
it enables us to arrive at a more unitary theory of checking. Until now, we had assumed
that three main types of relation were involved in checking, viz. a relation between a
head and another head which is adjoined to it, a relation between a head and its speci¬
fier, or a relation between a head and its complement. This led to obvious asymmetries,
in that (for example) checking the case of a nominative DP involved a specifier-head
relation, whereas checking the case of an objective DP involved a head-complement
relation (or, in ECM structures, a relation between a matrix head and a complement
specifier). But suppose we now eliminate the possibility that head-complement rela¬
tions are involved in checking, and argue instead that a head can only check its features
against those of its specifier, or against those of another head (or feature) which is
adjoined to it. If we further suppose that nominative and objective DPs in English carry
strong case-features, it follows that the only way in which an objective DP can check
its case-feature is by raising into a position in which it can enter into a spec-head rela¬
tion with a transitive verb; the required spec-head configuration will obviously come
about if the object raises to spec-AgrOP and the transitive verb raises to AgrO (perhaps
434
10.4 Object agreement projections
with the case-features of the verb and the object both being copied onto AgrO in order
to be checked, as suggested by Koizumi 1995, p. 40).
We can formulate a different kind of argument in support of the claim that DP
objects move to spec-AgrOP for case-checking purposes in relation to the position of
adverbs. In this connection, consider how we account for the following contrasts:
(35) VP
ADV
well V DP
I / \
plays 0 chess
The VP in (35) is then merged with an AgrO morpheme (which projects into AgrOP); the
verb plays adjoins to AgrO, and the DP 0 chess raises to spec-AgrOP as in (36) below:
(36) Agr OP
DP AgrO
/ \
AgrO VP
d chess ' \
A / \ /
plays Agr O ADV VP
* I
well DP
I I
t
The objective case-feature carried by the DP 0 chess (cf. He plays lots of games, and
plays them well) is checked at this point, since the DP 0 chess is in a spec-head rela¬
tion with the transitive verb plays (by virtue of the fact that plays is in AgrO, and 0
chess is in spec-AgrOP: Koizumi 1995, p. 40, suggests that checking involves copying
the objective case-feature of both the verb and its specifier onto AgrO, and erasing both
if they match).
The AgrOP constituent in (36) is then merged with a performative light verb 0
whose subject is he\ the verb plays adjoins to the light verb, as below:
435
Agreement projections
(37) .vp
The pronoun he then raises through spec-TP to spec-AgrSP, so deriving (34a) He plays
chess well. The assumption that both verbs and objective DPs raise across VP-adverbs
correctly predicts that sentences like (34b-c) are ungrammatical.
An analysis along the lines sketched out above accounts for the fact that verbs in
English are immediately adjacent to their objects (as illustrated in (34) above) and can¬
not be separated from them by intervening adverbials. On this account, the only way in
which a transitive verb could be separated from its object would be for an adverb to be
adjoined to AgrOP. But if we make the traditional assumption that adverbs modify pro¬
jections whose heads have specific semantic content, it follows that we can have
adverbs adjoined to projections of V, v or T (since V contains a lexical verb with its
own semantic properties, v contains an abstract light verb which has a specific - e.g.
causative or performative - sense and T has temporal properties), but that we can’t
adjoin adverbs to projections of agreement heads, since these have a purely formal
function (viz. that of checking case/agreement properties).
The assumption that objective DPs raise to spec-AgrOP offers us an interesting
account of the syntax of particles like out in sentences such as the following:
poured out
436
10.4 Object agreement projections
The VP in (39) is then merged with an AgrO constituent; the verb poured adjoins to
AgrO, and its DP specifier the whisky raises to spec-AgrOP: cf.
(40)
poured slowly DP
A
I
—<i— -t P
I
t out
The objective case-features carried by the DP the whisky (cf. He poured them slowly
out) and by the verb poured (which is transitive as used in (38) above, though is intran¬
sitive in other uses) are checked at this point; by hypothesis, objective-case checking
involves a spec-head relation between an AgrO containing a transitive verb and an
objective DP in spec-AgrOP (here, between an AgrO containing the transitive verb
poured and its DP specifier the whisky).
The next stage of derivation is for AgrOP to be merged with an abstract performative
light verb 0 whose agent subject is he; since 0 is a strong head, the verb poured
adjoins to it, as in (41) below:
slowly
Subsequently, the subject pronoun he raises to spec-TP and from there to spec-AgrSP
to check its case and agreement properties, so deriving (38a) He poured the whisky
slowly out.
Now consider how we derive (38b) He poured the whisky out slowly. Let us suppose
that (as before) the verb pour projects into a VP of the form (39) above. But let’s also
suppose that the particle out incorporates into the verb poured as in (42) below, so
forming the complex verb poured out'.
437
Agreement projections
V P t
poured out
A
Let’s further suppose that the whole complex verb poured out adjoins to AgrO (and
that the DP the whisky raises to spec-AgrOP to check its objective case-feature) as in
(43) below (where the rightmost trace is the trace of the particle out which has incorpo¬
rated into the verb):
(43) Agr OP
AgrOP then merges with the performative light verb 0; if the verb poured excorporates
out of the V complex poured out and raises on its own to adjoin to 0, the result will be
(44) below:
t out
438
10.5 Exceptional case-marking
But now suppose that instead of the simple verb poured adjoining to the light verb 0
on its own in (44), the prepositional particle out is pied-piped along with it, so that the
whole verbal complex poured out adjoins to 0. The result will be (45) below:
The pronoun he then raises through spec-TP into spec-AgrSP, thereby deriving (38c)
He poured out the whisky slowly. Thus, our assumption that (in transitive structures)
objective DPs move to spec-AgrOP and verbs move through AgrO to adjoin to v (and
can strand a particle in AgrO) provides an interesting account of the syntax of verb +
particle structures like (38).
(46) (a) The DA proved [the witness conclusively to have lied] (adapted from
Bowers 1993, p. 632)
(b) I suspect [him strongly to be a liar] (Authier 1991, p. 729)
(c) I’ve believed [Gary for a long time now to be a fool] (Kayne 1984, p. 114)
(d) I have found [Bob recently to be morose] (Postal 1974, p. 146)
439
Agreement projections
form the VP proved the witness to have lied, and that the adverb conclusively adjoins to
this VP to form the VP (47) below. (Here I am assuming, as in relation to examples
such as (9) and (10) above, that adverbs adjoin to maximal projections; for familiarity,
I have labelled infinitive phrases as IPs, though they are arguably AgrSPs with much
the same internal structure as finite clauses.)
(47)
Let’s further suppose that the VP in (47) then merges with an AgrO constituent to which
the verb proved adjoins, and that the DP the witness raises to spec-AgrOP, as in (48)
below. (Here and elsewhere, in accordance with the convention noted at the beginning
of the chapter, I simplify the internal structure of complex head-adjunction structures by
omitting the category labels of some of the heads, where these are self-evident; hence
proved is not labelled V.)
(48) AgrOP
DP
/ \
the witness AgrO
A / x
proved AgrO ADV
A
conclusively V IP.
I
t D I-
i
t to have lied
The objective case-features carried by the witness (cf. The DA proved him conclusively
to have lied) and by the transitive verb proved are checked at this point, via a
spec-head relation between the verb proved in AgrO and its specifier the witness in
spec-AgrOP.
The AgrOP in (48) is subsequently merged with an abstract performative light verb 0
whose agent subject/specifier is the nominative DP the DA: because 0 is a strong head,
the verb proved raises to adjoin to it, as in (49) below. (The subject the DA subse-
440
10.5 Exceptional case-marking
quently raises from spec-vp through spec-TP into spec-AgrSP, but we do not show
these later stages of derivation here.)
(49) vp
DP
/ \
the DA v AgrOP
/ \
proved 0 DP AgrO
conclusively V IP
t DP
t to have lied
As a result of movement of the DP the witness to spec-AgrOP and of the verb proved to
the head v position of vp, both end up positioned in front of the adverb conclusively -
precisely as we find in (46a). Thus, in effect what happens in ECM structures is that the
subject of the infinitive raises up to become the object of the main-clause verb (as was
suggested in an earlier framework by Paul Postal in his 1974 book On Raising).
Boskovic (1995, p. 176) argues that an AgrOP analysis provides a straightforward
account of the syntax of floating quantifiers in ECM structures such as (50) below:
We might derive (50) as follows. Let’s assume that (at some point in the derivation),
the verb proved is merged with an infinitive phrase (= IP) complement whose subject is
the QP all the defendants, as in (51) below:
(51)
V "
1
proved QP.
The VP in (51) is then merged with AgrO, the verb proved adjoins to AgrO and the DP
the defendants moves to spec-AgrOP as in (52) below, stranding the quantifier all in
spec-IP:
441
Agreement projections
Q DP to be lying
all t
The objective case carried by the defendants (cf. The DA proved them all to be lying) is
checked at this point; case-checking involves a spec-head relation between the transi¬
tive verb proved in AgrO and the objective DP the defendants in spec-AgrOP.
The V proved then adjoins to the performative light verb 0, deriving:
(53) vp
DP
the DA v AgrOP
' \
proved 0 DP. ' AgrO
t AgrO V IP
I
QP ,1
Q DP to be lying
I l
all t
The subject the DA subsequently raises through spec-TP into spec-AgrSP, so deriving
(50) The DA proved the defendants all to be lying. Thus, facts from floating quantifiers
lend further support to the claim that the objective infinitive subject in ECM structures
raises overtly to spec-AgrOP.
The raising analysis of ECM structures, taken together with our earlier analysis of
verb + particle structures, enables us to provide a principled account of the syntax of
ECM particle structures such as the following:
Let’s suppose that made out is (at some stage of derivation) a complex verb in which
the particle out is adjoined to the verb made. (This does not preclude the possibility
that out may originate as an independent head, and incorporate into the verb made.)
442
10.5 Exceptional case-marking
Let’s also suppose that the complex verb made out merges with the infinitive comple¬
ment the defendants to be lying to form the VP (55) below:
I I \ X \
made out the defendants to be lying
The VP in (55) is then merged with AgrO; the DP the defendants raises to spec-AgrOP,
and the complex V made out raises to adjoin to AgrO, as in (56) below:
Both movements are motivated by greed - i.e. by the need for the objective case car¬
ried by the defendants to be checked in a spec-head relation within AgrOP.
AgrOP is then merged with an abstract light verb 0 whose subject is the DA.
Because 0 is a verbal affix, a V constituent must adjoin to it. However, there are two V
constituents contained within AgrO in (56): one is the simple verb made, the other is
the complex verb made out. If the complex V made out adjoins to the light verb 0, the
result will be (57) below:
/ \ i / \
V P 0 the defendants AgrO VP
II / \ Z' \
made out V AgrO V IP
: I I / \
: / t DP I
i / \
t to be lying
443
Agreement projections
Here, the verb made has excorporated out of the complex verb made out, so leaving the
particle out stranded between the defendants and to be lying. The DA raises from spec-
vp through spec-TP to spec-AgrSP, so deriving (54b) The DA made the defendants out
to be lying. (For further arguments that ECM subjects raise to spec-AgrOP, see Authier
1991, Johnson 1991, Ura 1993, Koizumi 1995 and Boskovic 1995; for further discus¬
sion of the syntax of particle constructions, see Kayne 1984, Gueron 1990, Johnson
1991 and the references cited there.)
(59) (a) The crew handed back the passengers their passports
(b) The crew handed the passengers back their passports
(c) The crew handed the passengers their passports back
444
10.6 Indirect object agreement projections
(60)
V p their passports
I I
handed back
The direct object DP their passports in (60) has the 0-role theme, and the indirect
object DP the passengers has the 0-role of recipient. We can assume that the case car¬
ried by internal arguments such as these is determined by their thematic function, and
that dative is the case canonically associated with recipient DPs, and objective the
case canonically associated with theme DPs. In languages like German or Romanian
which have a relatively rich nominal morphology, dative and objective DPs are
morphologically distinct; but in languages like English which have an impoverished
nominal morphology, the two have the same morphological form (e.g. them is an
objective pronoun in I gave them to Mary and a dative pronoun in I gave them a
present).
The VP in (60) merges with AgrO, and the complex verb handed back adjoins to
AgrO and the objective DP their passports raises to spec-AgrOP as in (61) below:
V P the passengers V DP
handed back ■* t t
The objective case of the DP their passports in spec-AgrOP can then be checked under
a spec-head relation with the ditransitive verb handed (back) in AgrO (which, by
virtue of being ditransitive, can check both objective and dative case).
445
Agreement projections
AgrOP
The dative case carried by the DP the passengers in spec-AgrIOP is then checked
under a spec-head relation with the ditransitive verb handed (back) adjoined to AgrIO.
Finally, the complex verb handed back raises to adjoin to a performative light verb 0
which heads vp (and whose agent subject is the nominative DP the crew), so forming
the vp (63) below:
AgrIOP
DP AgrIO
The DP the crew then raises through spec-TP into spec-AgrSP to check its nominative
case, so deriving (59a) The crew handed back the passengers their passports.
Now consider how we derive (59b) The crew handed the passengers back their
passports. Assume that the derivation proceeds essentially as for (59a), until we reach
the stage of derivation represented in (62) above. At that point, the verb handed excor¬
porates out of AgrIO and adjoins to the light verb 0 on its own, as in (64) below:
446
10.6 Indirect object agreement projections
(64)
DP'"
/ \
the crew _____ AgrIOP
/ \
V v DP"" AgrIO
1 1
handed 0 the passengers AgrIO ^ AgrOP
A / \
V AgrIO AgrO
/ \
V
1
P
|
their passports / \
t back
The result will be that the particle back is stranded in AgrIO, between the indirect
object the passengers and the direct object their passports. Raising the DP the crew
through spec-TP to spec-AgrSP will then derive (59b) The crew handed the passengers
back their passports.
Finally, consider how we derive (59c) The crew handed the passengers their pass¬
ports back. Let’s assume that the derivation proceeds as for (59a) until we reach the
stage represented in (61) above. At that point, the verb handed excorporates out of
AgrO and adjoins to AgrIO on its own, with the indirect object the passengers raising
to spec-AgrIOP to check its case, as in (65) below:
(65) AgrIOP^
AgrIO.
Subsequently, the verb handed adjoins to the strong light verb 0 heading vp, as in (66)
below:
447
Agreement projections
(66) -vp.
DP
/ \
the crew v' AgrIOP
/ \
V v DP AgrIO
I
handed 0 the passengers AgrIO AgrOP
\
V AgrIO DP AgrO
I \ / \
t their passports AgrO VP
/ ' / \
V AgrO
/ \
V P
I I
t back
The DP the crew then raises through spec-TP to spec-AgrSP, thereby deriving (59c)
The crew handed the passengers their passports back.
So, we see that the assumption that indirect object DPs raise to spec-AgrIOP in order
to check their dative case (taken together with the assumption that when the verb in a
verb + particle complex raises, the particle can either be pied-piped along with the
verb or stranded) provides us with a principled account of the syntax of double-object
particle structures such as (59a-c) above. From a theoretical standpoint, the assump¬
tion that dative DPs raise to spec-AgrIOP to check their case offers the obvious advan¬
tage of providing us with a unitary account of case-checking under which all DPs
check their case by raising to the specifier position within an appropriate agreement
phrase (nominative DPs raising to spec-AgrSP, objectives to spec-AgrOP and datives
to spec-AgrIOP).
Genitive gerund structures like that bracketed in (67) have the twin properties that they
occupy typical DP positions (and hence can function as the complement of a preposi¬
tion, for example), yet clearly contain a vp structure internally within them. One way of
accounting for the dual DP/vp status of gerunds is to suppose that they are DPs headed
by a null determiner which has a vp complement, and that the subject of the gerund (viz.
the genitive nominal the UN’s) moves from spec-vp to spec-DP as in (68) below:
448
10.8 Vox-infinitives
(A variant of this analysis is to suppose that’s originates as the head D of DP, and sub¬
sequently attaches to the DP the UN.) An analysis such as (68) would lead us to expect
that the vp withdrawing troops from Utopia can be modified by an adverb - and this
expectation is borne out by sentences such as:
If we make the traditional assumption that the DP the UN’s carries genitive case in
(67/69), we can maintain that movement of the subject DP the UN’s from spec-vp to
spec-DP is motivated by greed - i.e. by the need to check the genitive case carried by
the UN’s. More generally, we can maintain that DPs check their case properties by
moving into the specifier position within an appropriate functional projection. If (as
Abney 1987 argues) the head D in DPs with a genitive specifier has abstract agreement
properties (and so is in effect an abstract Agr head), we can maintain the even stronger
position that case-checking always involves an (overt or covert) agreement relation
between an agreement head and its specifier. There is much more to be said about geni¬
tive case-marking, but we shall not pursue the relevant issues here. The main issue
from our point of view is that there is no reason to think that genitive DPs provide any
challenge to the claim that DPs are case-checked by raising to the specifier position
within an appropriate functional projection.
At the end of §8.5, we suggested that the objective case of the (italicized) infinitive
subject in such structures is checked by attraction, the objective case-feature carried by
him percolating up to the c-commanding transitive complementizer for. However, any
such analysis is inconsistent with the view expressed in this chapter that objective DPs
in English carry a strong case-feature, and raise to spec-AgrOP in order to check their
case. One way of overcoming this problem would be to treat structures like (70) in a
way which is more directly comparable to the way that we earlier treated subjects in
449
Agreement projections
ECM structures like We believe him to be innocent. One possibility which we might
explore along these lines would be to propose a split-CP analysis of complementizer
phrases parallel to the split-VP analysis of verb phrases outlined in the text. This might
mean (for example) that CP serves as the complement of an AgrOP constituent, with
the complementizer adjoining to AgrO and the infinitive subject raising to spec-AgrOP
as in (71) below:
We could then hypothesize that checking the case of the infinitive subject involves a
spec-head relation between the transitive preposition for in AgrO and the objective
pronoun him in spec-CP. The complementizer for would subsequently raise to adjoin to
the (strong) head c constituent of a superordinate cp as in (72) below:
t to succeed
(73) I can assure you [that never again will I take a syntax course]
Given the assumption made in chapter 6 that inverted auxiliaries adjoin to a null
COMP, a sentence such as (73) is problematic in that it contains the overt complemen¬
tizer that in addition to the inverted auxiliary will: moreover, since the two are sepa¬
rated by the adverbial phrase never again, it is clearly implausible to claim that will has
450
10.8 For-infinitives
adjoined to the complementizer node containing that. So, how are we to analyse such
sentences? The split-CP hypothesis offers us one possible answer: perhaps the auxil¬
iary adjoins to the head C of CP (with the negative adverbial never again occupying
spec-CP) and the complementizer that merges with the head c constituent of a higher
cp projection, as in (74) below:
(It goes without saying that (74) is simplified in a number of respects, e.g. by omitting
traces of moved constituents.) What is less than immediately self-evident, however, is
the extent to which relatively plausible structures like (74) justify postulating rather
more abstract structures like (72).
Returning now to the problems posed by the assumption that all DPs check their
case by raising to the specifier position within an agreement projection, further poten¬
tial problems are posed by prepositional objects (i.e. objective (pro)nominals which
serve as the complements of prepositions) such as me in structures like with me. If we
are to maintain that all objective D(P)s in English check their case by raising to spec-
AgrOP, we have to propose a parallel split-PP analysis for prepositional phrases. This
would mean that a PP like with me would be derived as follows. The preposition with
merges with the pronoun me to form the PP with me. The resulting PP merges with an
AgrO constituent; the preposition with adjoins to AgrO and its complement me raises
to spec-AgrOP (in order to check its objective case in a spec-head relation with the
transitive preposition with) as in (75) below:
(75) AgrOP
D AgrO
I
me AgrO PP
A / \
P AgrO D
I I
with t
A
The intermediate stage of derivation in (75) may provide us with a clue to the structure
of interrogative phrases such as that italicized in (76) below:
(76) I know he’s going out with someone, but I’m not sure who with
451
Agreement projections
where the complement who precedes the preposition with', it might also give us some
insight into the syntax of postpositional structures like Latin mecum (literally ‘me +
with’).
Subsequently, AgrOP merges with a light preposition 0, and the preposition with
raises to adjoin to the light preposition as in (77) below:
(77)
PP
\
D
I
t
The assumption in (75/77) that the preposition originates in P and raises to p may pro¬
vide us with an insight into the structure of Spanish PPs such as conmigo ‘with me’
(which is paraphraseable as ‘with + me + with’).
We might extend the analysis still further to account for compound prepositional
phrases such as those italicized below (% indicates a structure found only in some vari¬
eties of English):
We could derive a structure such as out of the window as follows. Let’s suppose that
out (in this use) is an intransitive preposition which merges with the DP the window to
form the PP out the window. This in turn merges with an AgrO containing the dummy
transitive preposition of, out adjoins to the AgrO constituent containing of and him
raises to spec-AgrOP as in (79) below. (It may be that of is a preposition which itself
merges with an abstract AgrO constituent, but this is a question of detail which we
shall not pursue here.)
(79) AgrOP
DP AgrO
/ \
the window AgrO PP
A / \ / \
P AgrO P DP
l I l
out of t t
A
452
10.8 For-infinitives
The objective case of the DP the window (cf. She fell out of them) is checked via the
spec-head relation it enters into with the dummy transitive preposition of in AgrO.
AgrOP then merges with a light preposition 0 to which the compound preposition out
of adjoins, as in (80) below:
(80)
AgrOP
AgrO p AgrO
/ ' I / \
P AgrO 0 the window AgrO PP
I I I / \
out of t P DP
I I
t t
In the (informal-style) sentence (81a), the DP which window has been preposed: in the
(formal-style) sentence (81b), the pp out of which window has been preposed; in the
ungrammatical (81c) the nonconstituent string of which window has been preposed
(and we know that only unitary constituents can be preposed). One question of detail
which remains to be resolved is why out cannot be adjoined to the light preposition 0 in
(80) without of, perhaps this is because the one cliticizes onto the other, or perhaps it is
because of (by virtue of being a preposition) must adjoin to the light preposition 0. At
any rate, the split-PP analysis of compound prepositions sketched out above seems
preferable to the traditional analysis of dummy of as a genitive-case particle, since
what has always remained a mystery under the genitive analysis is why of takes an
objective complement (cf. The windows were open and she fell out of them). The
analysis of dummy of as a prepositional AgrO constituent (or a transitive preposition
merged with an AgrO constituent) can be extended to adjectival structures such as fond
of him and nominal structures such as loss of income in ways which we shall not
explore here.
In more general terms, what our discussion here illustrates is that a uniform
spec-head agreement theory of case-checking (under which all case-marked DPs
check their case by raising to the specifier position within a higher agreement phrase)
can only be bought at the price of considerable abstraction - e.g. positing split cp/CP
and pp/PP projections, positing a different type of AgrP projection for each different
type of DP (e.g. for subjects, direct objects and indirect objects), positing syntactically
distinct cases which are not morphologically distinct in English (e.g. dative and objec-
453
Agreement projections
tive) - and so on. If we reject the level of abstraction involved and return to the more
traditional position that the case of some constituents (e.g. subjects) is checked via
spec-head agreement whereas the case of others (e.g. objects of prepositions) is
checked via attraction, the price we pay is that we end up with a nonunitary theory of
case-checking. The alternatives are clear - though for the moment we don’t have suffi¬
cient empirical evidence to make a principled choice between them: this will have to
await the outcome of future research.
In the active sentence (82a), the DP the agreements is the complement of the verb
signed, and raises to spec-AgrOP in order to check its objective case (cf. The president
signed them). But in (82b) the same DP raises to spec-AgrSP in order to check its nom¬
inative case (cf. They were signed by the president). A question we might ask is
whether the DP the agreements in the passive sentence (82b) moves through spec-
AgrOP before moving into spec-AgrSP, or whether (on the contrary) passive sentences
contain no AgrOP projection at all.
An interesting argument that passivized complements move through spec-AgrOP on
their way to spec-AgrSP is provided by Kayne (1989); he notes that passive participles
in French agree with their superficial subjects, e.g. in sentences such as the following
(FS = feminine singular):
If we suppose that the DP la decision moves to spec-AgrOP before moving into spec-
AgrSP, (83) will involve an intermediate stage of derivation at which la decision has
moved into spec-AgrOP and the participle prise has adjoined to AgrO, as in (84) below
(we ignore par le senat ‘by the senate’ here):
(84)
\
prise AgrO V DP
A
454
10.9 Passives and unaccusatives
Since the DP la decision and the passive participle prise will then be in a spec-head
agreement relation, we can account for the fact that the two agree in number and gen¬
der (as a reflex of spec-head agreement).
Interestingly, we find a similar pattern of agreement with perfective participles of
unaccusative (though not transitive or unergative) verbs in French, as can be seen from
the example below (FP = feminine plural):
In this connection, it is instructive to recall the observation by Luigi Burzio (1986) that
unaccusative subjects seem to share many properties in common with objects. One
way of accounting for this would be to suppose that (at an intermediate stage of deriva¬
tion) unaccusative subjects raise to spec-AgrOP, so that the derivation of (85) would
involve the intermediate stage represented in (86) below (if we suppose that theme
arguments are canonically projected as complements, and hence the subject of the sen¬
tence originates as the complement of the unaccusative verb):
parties AgrO PP V
♦ / \ / \
en Belgique V DP
! I i
.. t t
We could then say that the agreement between the DP les ecolieres ‘the schoolgirls’
and the unaccusative participle parties ‘departed’ is a reflex of a spec-head agreement
relation within AgrOP. If we were to extend this analysis to unaccusative verbs in
English, we could further say that in sentences such as:
In our treatment of such structures in §9.9, we hypothesized that you originates as the
specifier of a VP headed by go, and to school as its complement, and that you remains
455
Agreement projections
in situ in spec-VP, while the verb go raises to adjoin to v. However, this assumption is
at variance with the analysis of structures like (31) adopted in this chapter, under which
internal D(P) arguments are canonically projected as complements, and PP arguments
as specifiers. An analysis of unaccusative imperatives like (88) more in keeping with
the assumptions made in this chapter would be the following. The verb go merges with
its pronoun complement you to form the V-bar go you\ the resulting V-bar then merges
with its PP specifier to school to form the VP (89) below:
(89)
VP then merges with an AgrO constituent; the verb go adjoins to AgrO and the comple¬
ment you raises to spec-AgrOP, as in (90) below:
(90) AgrOP
go to school V D
* I I
t t
AgrOP then merges with an abstract light verb, to which the verb go adjoins, as in (91)
below:
(91)
AgrO
go 0 you AgrO
V AgrO PP
t
/
to school
\ \
D
I
t t
If we assume that the subject and verb subsequently remain in the positions they
occupy in (91), we can account for the fact that the verb precedes the subject in unac¬
cusative imperatives (though the converse word order holds e.g. in transitive impera¬
tives, since transitive subjects originate in spec-vp). An analysis such as (91) in effect
incorporates Burzio s intuition that unaccusative subjects are like complements in
456
10.9 Passives and unaccusatives
respect of their syntax. Of course, questions of detail remain to be worked out - e.g.
‘What case does the subject you carry, and how is it checked?’ If you carries partitive
case, it may well be that this is checked via a spec-head relation between the verb
come in AgrO and the pronoun you in spec-AgrOP.
In much the same way as we might argue that postverbal unaccusative subjects are
in spec-AgrOP, so too we might argue that the italicized postverbal subject in a passive
structure such as (92) below
(92) Below are listed the names of students who failed syntax
(in which the italicized argument follows the passive participle listed) is a further
instance of a complement DP raising to spec-AgrOP and then remaining there (with
the verb raising above it to adjoin to v). As should be self-evident, numerous questions
are left unanswered by such an analysis (e.g. what case does the italicized postverbal
nominal carry and how is it checked; and how do we account for agreement between
the postverbal nominal and the plural auxiliary are/werel).
If passivized arguments raise to spec-AgrOP in French structures like (83—4) and in
English sentences such as (92), the most general conclusion we can arrive at is that
passivized arguments raise through spec-AgrOP on their way to spec-AgrSP. However,
if we assume that all movement is motivated by greed, this raises the question of why
passivized objects should move into spec-AgrOP in English. An interesting answer is
suggested by French passives such as (83-4), where the passive participle adjoins to
AgrO and the object moves to spec-AgrOP in order to check that the passive participle
agrees with its object. Just as a French passive participle like prise ‘taken’ carries an
object agreement suffix (marked by the final +e in the spelling), so too we might sup¬
pose that the suffix +n (and its variants +d/+t) carried by passive participles like
seen/heard/sent etc. in English is also an object-agreement inflection (one which has
variable agreement properties, and hence can agree with an object carrying any set of
person/number/gender-features - in much the same way as an auxiliary like can has
variable subject agreement properties). It would then follow that the object agreement
properties of passive participles can only be checked if the passive participle adjoins to
AgrO and its complement moves into spec-AgrOP. This in turn correctly predicts that
only transitive verbs can passivize, since only if a verb has an object which can move
into spec-AgrOP can the object agreement properties of the passive participle be
checked.
If our reasoning here is along the right lines, it follows that verbs raise to AgrO and
their objects to spec-AgrOP either to check agreement or to check case. Active transi¬
tive verbs and their objects in English raise to appropriate positions within AgrOP in
order to check the case of the object (but do not check agreement, since active verbs
carry no object agreement inflections in English); passive verbs and their objects raise
to appropriate positions within AgrOP in order to check the agreement properties of the
457
Agreement projections
passive participle, but do not check the case of the object (with the result that the pas¬
sivized DP has to raise to spec-AgrSP in order to check nominative case). In other
words, active transitive verbs in English check the case properties (but not the agree¬
ment properties) of their DP complements, whereas passive participles check the
agreement properties (but not the case properties) of their DP complements. It would
seem that (for languages like English), the following generalization (relating to the
case and agreement properties of verbs and their objects) holds:
(93) A verb doesn’t case-check an object DP it agrees with, and doesn’t agree
with an object DP which it case-checks.
While I shall not be concerned here with the question of the extent to which (93) holds
cross-linguistically, it is interesting to note that data from Ukrainian (reported in Sobin
1985) seem to lend some support to (93). Ukrainian has two passive constructions: in
one, the passivized argument carries nominative case, and the passive verb form agrees
with it in gender and number (i.e. the verb is agreeing with, but not case-checking, its
object): in the other (so-called ‘impersonal passive’) structure, the passivized argument
carries objective case, but the passive verb form does not agree with it (i.e. the verb is
case-checking, but not agreeing with, its object).
Consider what all of this means for the derivation of a passive sentence such as:
(95) VP
V DP
I / \
stolen the jewels
The VP in (95) is then merged with AgrO; the passive participle stolen adjoins to AgrO
and the DP the jewels raises to spec-AgrOP as in (96) below:
(96) AgrOP
DP AgrO
/ \
the jewels AgrO VP
A
/ \
V AgrO V DP
I I I
stolen <. •t t
At this point, the participle stolen is checked for agreement with its specifier the jewels:
since stolen has variable agreement properties (i.e. it can agree with any object DP),
458
10.10 Summary
this requirement is satisfied in (96). However, the case properties of the jewels are not
checked at this point, given the generalization (93). Subsequently, the passive partici¬
ple stolen raises up to adjoin to the strong light verb heading vp as in (97) below:
(97)
AgrOP
AgrO
AgrO VP
/ \
V AgrO V DP
I I
t
The DP the jewels is then raised up in a successive cyclic fashion to become first the
subject of vp, then the subject of the PROGP headed by been, then the subject of TP
and then the subject of the AgrSP headed by have (checking its nominative case in
spec-AgrSP). If such an analysis of passives can be maintained, it follows that the
‘object’ of both an active and a passive verb will raise to spec-AgrOP, in active sen¬
tences for case-checking purposes, and in passive sentences for agreement-checking
purposes. Needless to say, numerous details of the syntax of passive sentences remain
to be worked out (particularly the syntax of fey-phrase agents and the syntax of the
implicit agent arguments in fey-less passives), but the motivation for assuming that
passivized objects move to spec-AgrOP at an intermediate stage of derivation should
be clear.
10.10 Summary
In this chapter, we have argued that clauses contain three different types
of agreement phrase. We began by outlining the split-INFL hypothesis in §10.2, argu¬
ing that we need to posit the existence of a subject agreement phrase (AgrSP) in addi¬
tion to a tense phrase (TP) in order to account for the fact that adverbs like probably
and floating quantifiers like all can be positioned either before or after a finite auxiliary.
We noted that the split-INFL analysis would also enable us to account for the fact that
an adverb positioned after an auxiliary can have scope over the auxiliary, e.g. in a
sentence such as Gary can apparently lift 100 pounds. In §10.3 we saw that the split-
INFL analysis would provide a straightforward account of scrambling in Early Modem
English, and would enable us to say that in sentences like The king your mote did see,
the DP your mote has adjoined to TP, while the subject the king is in spec-AgrSP. In
addition, we noted Alison Henry’s arguments that the split-INFL analysis enables us to
provide an insightful description of the syntax of the case and agreement properties of
subjects and auxiliaries in Belfast English, under which nominative subjects are in
spec-AgrSP, and any auxiliary which checks the agreement properties of the subject is
in AgrS (and so can move to C in questions), whereas objective subjects are in spec-TP
459
Agreement projections
and any associated auxiliary is in T, and (by virtue of lacking agreement properties)
can’t move through AgrS to C in questions.
In §10.4 we outlined the split-VP hypothesis, under which clauses headed by transi¬
tive verbs contain an object agreement phrase (AgrOP) positioned between vp and VP:
we argued that a direct object DP raises to spec-AgrOP and its associated verb adjoins
to AgrO in order to check objective case (the verb subsequently raising still further to
adjoin to v). We noted that such an analysis would provide a principled account of why
object DPs occupy a different position from object CPs (e.g. in sentences such as He
reported to the police that there had been a robbery and He reported the robbery to the
police), why DP objects are positioned in front of VP adverbs (e.g. in sentences such
as He plays chess well), and why prepositional particles can occupy three different
positions in sentences such as He poured the whisky slowly out/He poured the whisky
out slowly I He poured out the whisky slowly. In §10.5 we argued that the split-VP
hypothesis provides an interesting account of a range of aspects of the syntax of ECM
structures, including the position of the adverb conclusively in sentences such as The
DA proved the witness conclusively to have lied, the position of the quantifier all in
sentences like The DA proved the defendants all to be lying, and the position of the
particle out in sentences such as The DA made the defendants out to be lying.
In §10.6 we argued that indirect object DPs carry dative case, and check their case
by raising to the specifier position within an indirect object agreement projection
(AgrIOP) which is positioned immediately above AgrOP (with the verb raising to
adjoin first to AgrO, then to AgrIO and finally to v). We saw that such an analysis
would allow us to claim that in sentences such as The crew handed the passengers back
their passports, the particle back has been stranded in AgrIO. More generally, we
concluded that an analysis in which a nominative DP checks its case by raising to spec-
AgrS, an objective DP by raising to spec-AgrOP, and a dative DP by raising to
spec-AgrIOP would enable us to maintain the position that case-checking in English
canonically involves a spec-head relation between a functional (agreement) head and
its specifier. In §10.7 we briefly looked at how such an analysis might be extended to
handle DPs which carry the genitive case-suffix ’s; and in §10.8 we suggested a
split-CP analysis of/or-infinitive structures, and a split-PP analysis of transitive prepo¬
sitional phrases.
Finally, in §10.9 we went on to consider whether passivized direct object DPs move
through spec-AgrOP on their way to spec-AgrSP. We saw that there was some evi¬
dence in support of this suggestion from the fact that passive participles agree with
their (underlying) objects in French. And we noted that if the passive participle suffix
+n is taken to be an agreement marker in English, we could claim that in passive sen¬
tences such as The jewels have been stolen the passivized object DP the jewels moves
into spec-AgrOP at an intermediate stage of derivation (to check agreement with the
passive participle stolen), before moving on to check its nominative case-feature in
460
Workbook section
spec-AgrSP. We also suggested that unaccusative subjects might also move into spec-
AgrOP at an intermediate stage of derivation.
It need scarcely be pointed out that the analyses we have presented in this chapter
are not fully worked out (and doubtless flawed) in a number of respects. The reason is
simple: the ideas explored in this chapter (inspired by Chomsky’s minimalist program
in the 1990s) are part of an ongoing research programme which is leading different
researchers in different directions. Many important questions of detail and principle
remain unresolved for the present. For example, there is disagreement on whether all
objective DPs (or some, or none) in English check their case by raising overtly to spec-
AgrOP. Koizumi (1995) maintains that all objective DPs check their case in this way
(though does not discuss CPs or PPs with a transitive head): Boskovic (1995) main¬
tains that while it is plausible to analyse objective subjects in ECM structures as raising
to spec-AgrOP in English, it is not plausible to extend the raising-to-spec-AgrOP
analysis to other direct (or indirect) object DPs. Bobaljik (1995) goes even further and
rejects the claim that any objective DPs raise to spec-AgrOP in English, preferring to
maintain that objective case-checking canonically involves a head-complement rela¬
tion (much as we assumed in chapter 5). It might seem ironic that linguists can’t agree
about the syntax of agreement; but it should be emphasized that work in this domain is
only in its infancy, and the first step in understanding any problem is to identify the
nature of the problem, to outline possible solutions, and to be aware of the hidden costs
associated with such solutions.
While it is true that there are a wide range of alternative analyses of particular struc¬
tures (e.g. passives or double-object constructions) found in the contemporary linguis¬
tic literature, it is equally true that a considerable amount of the relevant research work
presupposes some variant of the richly articulated AgrSP/TP/vp/AgrIOP/AgrOP/VP
clause structure that we have argued for in this chapter - and in particular, assumes the
existence of subject agreement and object agreement projections (and perhaps also
indirect object agreement projections); and indeed, some work goes still further and
posits the existence of aspect phrase, voice phrase and modal phrase projections as
well (cf. e.g. Ouhalla 1991 and Cinque 1995). Moreover, even work which questions
the empirical and theoretical motivation for positing agreement projections (e.g.
Bobaljik 1995 and Chomsky 1995b) presupposes familiarity with work on the syntax
of AgrP constituents. So, for both AgrOphiles and AgrOphobes alike, the syntax of
agreement remains at the very heart of contemporary debates about the nature of syn¬
tactic structure.
Workbook section
(Note that some of the exercises below are rather more challenging than
those in previous chapters, and invite you to consider alternatives to the analyses sug¬
gested in the main text.)
461
Agreement projections
Exercise I (§10.2)
Discuss the syntax of the adverbs and floating quantifiers in the following
sentences, saying which of the sentences pose problems for the analysis proposed in
the text (and why), and suggesting alternative analyses.
If we assume that apparently is a TP adverb, and that adverbs adjoin to maximal pro¬
jections, apparently will originate as an adjunct to the TP in which have originates, as
shown informally in (ii) below (where SP = AgrSP and S = AgrS):
(ii) [sp They [s have] [Tp apparently [Tp t [T t] [vp t given up smoking]]]]
On this view, have would originate in T and adjoin to AgrS, and they would originate in
spec-vp and move through spec-TP into spec-AgrSP; apparently would adjoin to TP to
form a split-segment TP category.
The problem with the analysis in (ii), however, is that it provides no account of the
position of the floating quantifier all in la. If floating quantifiers are stranded internally
within subject QPs, we’d expect the quantifier all to be stranded within the QP subject
of TP, to the right of the TP adjunct apparently - as in:
Of course, (iii) is perfectly grammatical - but so is la, and the crucial point is that
analysis (ii) provides no way of accounting for the grammaticality of la.
In order to account for the grammaticality of la, we need to modify one (or more) of
the assumptions implicit in analysis (ii). These assumptions are spelled out in (iv)
below:
462
Workbook section
One way of handling la is to reject assumption (iv)(a), while retaining the other
assumptions in (iv). We might suppose, for example, that the perfective auxiliary have
originates as the head Aux(iliary) constituent of an AuxP (= auxiliary phrase) projec¬
tion which is positioned between TP and VP (with have raising to adjoin first to T and
then to AgrS), and that apparently is an adjunct to AuxP, as shown informally in the
partial structure (v) below:
(v) [SP They [s have] [w [Qp all t] [T t] [AuxP apparently [AuxP t [Aux t] [vp ... ]]]]]
This would mean that the QP all they originates in spec-vp, and then raises through
spec-AuxP into spec-TP; subsequently, the pronoun they raises to spec-AgrSP, strand¬
ing the quantifier all inside a QP in spec-TP; the adverb apparently is an adjunct to
AuxP. Such an analysis would correctly predict that a further adverb can be adjoined to
TP, as in (vi) below:
The core assumption underlying the analysis is that there are three different (auxiliary-
like) functional projections between CP and vp, so that INFL would be split up into
three distinct heads.
We reach a similar conclusion if we reject assumption (iv)(b) in favour of the sug¬
gestion in Cinque (1995) that adverbs are not adjuncts, but rather are specifiers - and
that each different type of adverb serves as the specifier of a different kind of head. If
(as Cinque assumes) QPs containing floating quantifiers are also specifiers, we might
arrive at an analysis of la along the lines of (vii) below:
(vii) [SP They [s have] [TP [Qp all t] [T t] [AuxP apparently [Aux t] [vp . . . ]]]]
(viii) [SP They [s have] [TP [Qp all t) [f apparently [f [T t] [vp . . . ]]]]]
463
Agreement projections
extended T-bar), and that the QP containing all is a specifier for TP (expanding T-bar
into TP). An analysis along the lines of (viii) would enable us to avoid splitting INFL
into three different heads, since we would no longer need to posit an AuxP projection
in sentences like la. Of course, it might be argued that the analysis in (viii) amounts to
returning to our assumption in earlier chapters that adverbial adjuncts can adjoin to
intermediate projections (especially if we follow Kayne 1994 in maintaining that spec¬
ifiers are adjuncts).
Yet a further possibility would be to abandon Sportiche’s assumption (iv)(c) that
floating quantifiers are stranded via movement of D(P)s out of subject positions, and
instead to assume (as in Bowers 1993 and Baltin 1995) that floating quantifiers are
directly generated as adjuncts (rather like adverbs). After all, given that we saw in §7.4
that subjects are islands (i.e. no constituent can generally be extracted out of a subject),
it might seem implausible to suppose that a pronoun like they can ‘escape’ out of a sub¬
ject QP, in violation of the condition on extraction domains. More specifically, if
(like Bowers) we suppose that floating quantifiers are merged with maximal clausal
projections, both all and apparently in la would be adjuncts merged with TP, and
hence (if we make the traditional assumption that multiple adjuncts of the same kind
can be freely ordered with respect to each other) we should expect both structures in
(ix)(a-b) to be possible (where t is the trace of have and t is the trace of they):
(ix) (a) [sp They [s have] all [Tp apparently t [T t] [vp ... ]]]]]
(b) [sp They [s have] [Tp apparently [Tp all t [T t] [vp ... ]]]]]
Of course, if (like Baltin) we assume that floating quantifiers can merge with interme¬
diate projections, additional possibilities arise. But I’ll leave you to ponder over these
(and the problems posed by the other examples in the exercise) for yourself.
Exercise II (§10.4)
464
Workbook section
(i)
to her V DP
I / \
explained the problem
The VP in (i) will merge with an AgrO constituent; the verb explained will adjoin to
AgrO, and the DP the problem will raise to spec-AgrOP (to check its objective case-
feature), as in (ii) below:
465
Agreement projections
AgrOP
(ii)
the problem
A
ADV
explained carefully
A
to her V DP
1 I
.. t t
AgrOP then merges with a performative light verb 0 whose subject is he, and the verb
explained adjoins to the light verb as in (iii) below:
V AgrO ADV VP
! I I /
. . t carefully PP V
/ \ / \
to her V DP
I I
t t
(Alternatively, we might assume that AgrO is pied-piped along with the moved verb
explained.) The subject he subsequently raises through spec-TP to spec-AgrSP to
check its nominative case, so deriving 1 a He explained the problem carefully to her.
It seems likely that the derivation of 1 b He carefully explained the problem to her is
similar in most respects, save that the adverb carefully is an adjunct to vp rather than
VP.
More problematic is the derivation of lc He explained the problem to her carefully.
Given what we have said so far in this model answer, we should expect carefully to end
up either to the immediate left of to her as in la, or to the immediate left of explained
as in lb. How come it ends up at the end of the overall clause? One (traditional) possi¬
bility would be to suppose that (at least some) adjuncts can be positioned either to the
left or to the right of the constituent they merge with. Reasoning along these lines, we
might suppose that the VP headed by explained could either have the adverb carefully
positioned initially (as the first constituent of the VP) as in (i) above, or finally (as the
last constituent of VP) as in (iv) below:
466
Workbook section
(iv)
V DP
I / \
explained the problem
The remaining stages of derivation would then be as outlined above for sentence la,
and the result would be a structure in which carefully occupies the final position in the
sentence, precisely as we find in lc He explained the problem to her carefully.
One potential problem posed by the analysis in (iv) is that it allows for the possibility
that an adverb can be positioned either before or after the constituent it modifies.
However, English is a language in which modifiers generally precede the expressions
they modify (e.g. the modifying adjective safe precedes the noun syntax in an expression
such as safe syntax), so we might want to exclude any possibility of generating the
modifying adverb carefully in a position where it follows the VP it modifies in (iv). In
other words, all things being equal, we would prefer a derivation in which the adverb
carefully is generated in a position where it precedes the VP it modifies.
One way of attaining this goal would be to propose an even more complex model of
clause structure in which each separate argument of a verb is contained within a sepa¬
rate projection, and in which each projection can be modified by an appropriate kind of
adverb. Given this possibility, we could derive lc as follows. The verb explained
merges with its DP complement the problem to form a VP to which the adverb care¬
fully adjoins, as in (v) below:
carefully V ^ DP
(Note that carefully is positioned to the left of the VP explained the problem which it
modifies.) The resulting VP merges with a VP headed by a light verb 0 whose specifier
is the PP to her, and the verb explained adjoins to this light verb as in (vi) below:
467
Agreement projections
The vp in (vi) then merges with an AgrO constituent; the verb explained adjoins to
AgrO, and the DP the problem moves to spec-AgrOP as in (vii) below:
(vii)
explained
k
II 1 / \
t 0 carefully V DP
t t
Subsequently, AgrOP merges with a performative light verb whose subject is he, and
the verb explained raises to adjoin to the light verb as in (viii) below:
I | 1 / \
t 0 carefully v DP
I I
t t
The subject he then raises through spec-TP into spec-AgrSP to check its nominative
case, so deriving lc He explained the problem to her carefully.
The ungrammaticality of Id *He explained carefully the problem to her can be
accounted for rather more straightforwardly. As we can see from (viii) above, the only
way of deriving this sentence would be by merging the adverbial adjunct carefully with
AgrOP; but given the constraint noted in the text that adverbs can only modify projec¬
tions which have clear semantic content, and given that AgrO is a morpheme with no
intrinsic semantic content (its sole function is to check objective case), this possibility
is ruled out.
468
Workbook section
and then raising want to adjoin to a superordinate light verb as in (ii) below:
(ii)
wanted 0 desperately CP
A
469
Agreement projections
Raising the subject she through spec-TP into spec-AgrSP to check its nominative case
will derive la She wanted desperately for them to start a family. (We shall not be con¬
cerned here with the internal syntax of the CP for them to start a family, for one sug¬
gestion, see the discussion of/or-infinitives in §10.8.) Since the adverb desperately is
an adjunct to the VP originally headed by wanted, it follows that it cannot be posi¬
tioned internally within CP - hence the ungrammaticality of lb *She wanted for them
desperately to start a family.
More interesting is the sentence in lc. For almost three decades (in work dating from
Bresnan 1972 to Boskovic 1995) it has been argued that objective-subject infinitive
complements of want-class verbs are CPs headed by a null complementizer, while
those of believe-dass verbs are simple IPs. However, the CP analysis provides no
account of the fact that desperately can be positioned between them and to in lc, since
this is not possible in the clearcut CP structure la. Hence, it would seem more plausible
to treat the infinitive complement of wanted in lc as an IP. We could then say that
wanted originates as the head V of the VP in (iii) below:
(iii)
desperately
The transitive verb wanted subsequently adjoins to AgrO and the pronoun them moves
to spec-AgrOP to check its objective case, as in (iv) below:
(iv) AgrOP
th
desperately V
t to start a family
The verb wanted then raises to adjoin to the strong light verb 0 occupying the head v
position of vp as in (v) below:
Workbook section
(v) . vp.
D
I
she AgrOP
wanted 0 D AgrO
4 |
them AgrO VP
/ \
t AgrO ADV VP
I
desperately V IP.
I
t D I-
I
t to start a family
Exercise IV (§10.6)
Discuss the syntax of the following sentences, saying which sentences
prove problematic to derive, and why:
471
Agreement projections
(i)
the CIA V DP
l / \
handed the tapes
The VP in (i) will then merge with an AgrO constituent, the V handed will adjoin to
AgrO and the DP the tapes will raise to spec-AgrOP to check its objective case, as in
(ii) below:
(ii) AgrOP
/ \
the tapes AgrO
A / \
V AgrO V
I / \ \
handed the CIA DP
A
I
t
AgrOP merges with an AgrIO constituent, the V handed adjoins to AgrIO and the DP
the CIA raises to spec-AgrIOP in order to check its dative case, as in (iii) below:
(iii) AgrIOP
DP AgrIO
/ \
the CIA AgrIO AgrOP
/ \
V AgrIO D ' AgrO
i / \
handed the tapes AgrO VP
/ \
V AgrO DP V
I I
t i V DP
I I
t t
AgrIOP then merges with an agentive light verb (whose agent subject is the FBI), and
the V handed adjoins to the performative verb, as in (iv) below:
472
Workbook section
(iv)
AgrIO
Subsequently, the DP the FBI raises through spec-TP into spec-AgrSP to check its
nominative case, so deriving the structure associated with sentence la The FBI handed
the CIA the tapes.
Consider now why the corresponding sentence lb *The FBI handed the tapes the
CIA is ungrammatical. Here, the objective theme argument the tapes is positioned in
front of the dative recipient argument the CIA. There are a number of suggestions we
can make about why the derivation of lb crashes. One might be that case is misas-
signed, with the recipient DP the CIA being wrongly assigned objective case (so
raising to spec-AgrOP), and the theme DP the tapes being wrongly assigned dative
case (so raising to spec-AgrIOP, and ending up in front of the CIA): however, this pos¬
sibility would be ruled out if we assume that UG principles specify that objective is the
case canonically associated with theme arguments, and dative with recipient argu¬
ments. An alternative possibility is that case is correctly assigned, but AgrOP is
generated above AgrIOP (so that the objective DP the tapes raises to a higher position
than the dative DP the CIA). After all, nothing in what we have said in the text guaran¬
tees that AgrIOP will be generated above AgrOP: we have simply stipulated this in our
description thus far. And indeed, ordering AgrOP above AgrIOP might be appropriate
for those (%) varieties of English in which direct objects can precede indirect objects,
e.g. in sentences such as:
(v) %Givedme!
We might derive sentence lc The FBI handed the tapes to the CIA as follows. If we
assume (as in the text) that theme arguments are canonically projected as complements
and recipient arguments as specifiers, the theme DP the tapes will originate as the
complement of handed, and the recipient PP to the CIA as tne specifier of handed, as
in (vi) below:
(vi)
to the CIA DP
/ \
the tapes
473
Agreement projections
The VP in (vi) will then merge with an AgrO constituent, the V handed will adjoin to
AgrO and the DP the tapes will raise to spec-AgrOP to check its objective case, as in
(vii) below:
(vii) AgrOP
DP ^ AgrO
/ \
the tapes AgrO VP
A / \
V AgrO DP V
1 / \ /
handed to the CIA V
A I |
. . t I
AgrOP then merges with an agentive light verb whose agent subject is the FBI; the
verb hand adjoins to the light verb as in (viii) below:
(viii)
V AgrO
1
t to the CIA V DP
t t
Subsequently, the DP the FBI raises through spec-TP into spec-AgrSP to check its
nominative case, so deriving the structure associated with sentence 1 c The FBI handed
the tapes to the CIA.
Exercise V (§10.8)
Discuss the derivation of the infinitive structures in the sentences below
(the Belfast English examples in 1 are from Henry 1995, and the Jamaican Creole
examples in 2 from Bailey 1966):
474
Workbook section
Helpful hints
The Jamaican Creole word se (although cognate with the English verb
say) is generally taken to have the status of a declarative complementizer (rather like
English that). Pronouns in Jamaican Creole are not overtly inflected for case.
(i) ^ CP
^IP
for DP
/ \
Jimmy to come with me
The CP in (i) would then merge with a V headed by the verb wanted, so forming the
VP (ii) below:
(ii) VP
V^
11 ^ CP\^
wanted C"'
for DP " ( 1 .
/ \
Jimmy to come with me
If we suppose that want (in this use) is a transitive verb, the VP in (ii) will merge with
an AgrO constituent to which wanted adjoins, with the DP Jimmy raising to spec-
AgrOP in order to check its objective case, as in (iii) below:
475
Agreement projections
(iii) Agr°p
to come with me
AgrOP then merges with an experiential light verb 0 (whose subject is the experiencer
argument /), and the verb wanted adjoins to the light verb as in (iv) below:
t to come with me
Subsequently, the subject 1 raises through spec-TP to spec-AgrSP to check its nomina¬
tive case, so deriving the structure associated with sentence la / wanted Jimmy for to
come with me. A crucial assumption underlying the derivation sketched out above is
that complementizers like for project into the minimal structure needed to satisfy
grammatical and lexical requirements: hence, since for is not transitive in this use,
there is no need to project a cp/AgrOP structure on top of the CP headed by for.
Exercise VI (§10.9)
Discuss the derivation of the following passive sentences:
476
Workbook section
(i) VP.
DP :v
the FBI V DP
/ \ / \
V P the tapes
given back
The VP in (i) then merges with AgrO; the transitive V given raises to AgrO, and the DP
the tapes raises to spec-AgrOP to check its objective case (cf. The FBI were given them
(ii) AgrOP
/ \
the tapes
A / \
given AgrO
A
t back
All
Agreement projections
AgrOP in (ii) is then merged with AgrIO: the verb given adjoins to AgrIO and its indi¬
rect object the FBI raises to spec-AgrIOP, as in (iii) below:
AgrIOP
(iii)
DP AgrIO
/ \
the FBI AgrIO AgrOP
A
/ \
given AgrIO DP AgrO
A / \
the tapes AgrO VP.
AgrO DP V
i
.. t V DP
V
I I
At this point, the (variable) agreement properties of the passive participle given are
checked against those of its specifier the FBI; this in turn means that the case of the
FBI cannot be checked (since verbs do not case-check object DPs with which they
enter into an agreement-checking relation in English). Subsequently, the verb given
adjoins to the head v of vp, as in (iv) below:
AgrO
/ \
t AgrO DP
I
The indirect object DP then ‘passivizes’ - i.e. it moves in a successive cyclic fashion
into spec-AgrSP, where it checks its nominative case. Whether or not it moves into
spec-vp prior to moving into spec-TP and thence to spec-AgrSP is a matter for you to
determine empirically - on the basis of the potential evidence provided by some of the
example sentences in this exercise.
478
Workbook section
Exercise VII
As the examples below illustrate, there is evidence from verb + particle
structures that weak object pronouns behave differently from other objective nominals
- i.e. from contrastively stressed objective pronouns and from ordinary nominals like
Mary, the president, etc. On the basis of the examples below, say which of the various
analyses of weak pronouns proposed in the model answer provides the most adequate
characterization of the syntax of weak object pronouns:
(Note that the adjective weak here is used to refer to the phonetic properties of pro¬
nouns, indicating lack of stress and possible reduction in the form of the pronoun e.g.
from them to ’em, and is not to be confused with the very different use of the word
weak in checking theory.)
479
Agreement projections
tence is grammatical, as in lb He turned, them off', but where the pronoun and verb are
separated by the particle, the sentence is ungrammatical, as in Id *He turned off them.
But the question we have to ask is why the pronoun should have to be immediately
adjacent to the verb.
One answer might be (as suggested by Postal 1974, p. 102) that weak object pro¬
nouns are phonetic clitics. We might interpret this as meaning that they must cliticize
to an appropriate host in the PF component: if we suppose that verbs can host clitics
but particles can’t, and that the presence of an intervening constituent between a clitic
and its host blocks cliticization, the behaviour of the pronouns in 1 can be accounted
for. However, such an analysis begs two very important questions, namely ‘Why are
verbs so hospitable and particles so inhospitable?’ and ‘What is the precise nature of
PF cliticization?’
One way of accounting for why weak pronouns attach to verbs rather than particles
would be to correlate this with the fact that it is the verb which case-checks the pro¬
noun, not the particle. Since pronouns attach to the verb which case-checks them, it
would seem natural to suppose that attachment is an integral part of the case-checking
process (and hence is a syntactic rather than a purely phonetic phenomenon). Bearing
in mind that a head can check its features by adjoining to another (functional) head and
that pronouns (by virtue of being D constituents) are heads, we might suggest that
weak pronouns check their case by adjoining directly to the relevant agreement head.
Such an analysis seems plausible for (say) weak subject pronouns in French like je ‘F,
tu ‘you’, il ‘he’, etc. which arguably adjoin directly to the same functional head as
finite verbs in French (perhaps AgrS), and hence cannot be separated from their depen¬
dent verbs by intervening adverbs etc. We might therefore extend it to weak object
pronouns in English, and argue that they check their case by adjoining directly to
AgrO. (Of course, the verb which case-checks any such pronoun will also have to
adjoin to AgrO.) This would mean that objective DPs and weak objective pronouns
check their case in different ways: objective DPs by raising to spec-AgrOP, and weak
objective pronouns by adjoining to AgrO.
Given this assumption, lb might be derived as follows. The verb turn projects into a
VP of the form:
(i)
DP
i
them
VP is then merged with AgrO; both the transitive verb turned and the objective pro¬
noun them adjoin to AgrO, as in (ii) below:
480
Workbook section
(ii)
turned D AgrO t V P
I
them -* t off
Since the verb turned and the objective pronoun them are both adjoined to AgrO at this
point, their objective case properties can be checked (via a head-head relation).
Subsequently, the verb turn raises up to adjoin to the strong light verb 0 which heads
vp, as in (iii) below:
(iii)
turned 0 V AgrO D
■ 1 / \ I \
D AgrO t P
I
them off
(An alternative possibility is that the whole AgrO constituent is adjoined to 0, so that
the pronoun is pied-piped along with the verb turned and adjoined to 0, perhaps
because a weak pronoun cannot be separated from a verb which case-checks it.)
Subsequently, he raises through spec-TP into spec-AgrSP, so deriving lb He turned
them off.
What remains to be accounted for is the ungrammaticality of Id *He turned off
them. Here, we might conjecture that the particle off adjoins to the verb turned as in
(iv) below:
(iv)
V p t
I I
turned off<
Both the pronoun them and the complex verb turned off then adjoin to AgrO, as in (v)
below (where the righthand trace is the trace of the particle off which has incorporated
into the verb):
481
Agreement projections
(v) AgrOP
AgrO
V
\
V P P
I I I
turned ojf t
The verb turned then excorporates out of AgrO, adjoining to the strong light verb 0
heading vp, as in (vi) below:
(Alternatively, it might be that the complex verb turned ojf or even the whole AgrO
constituent turned ojf them is adjoined to 0.) The subject he then raises through spec-
TP into spec-AgrSP, so deriving Id *He turned off them. However, the fact that the
resulting sentence Id is ungrammatical means that one or more steps in the derivation
(iv/v/vi) must be impermissible. But which?
A likely candidate is step (v). We might rule out (v) by a constraint to the effect that
a verb cannot be part of a head which has more than one other overt constituent
adjoined to it: this would rule out (v) because AgrO contains not only the verb turned
but also the particle off and the pronoun them. However, the one-on-one constraint
seems arbitrary, and clearly has no explanatory force whatever; moreover, passives
such as:
482
Workbook section
would seem to falsify it, if we make the traditional assumption that (for example) in
(viii)(a) the noun advantage and the preposition of (which are clearly different kinds of
head) are both adjoined to the verb taken.
A third way of ruling out (v) would be in terms of a structural constraint to the effect
that a verb can only case-check an incorporated pronoun which it head-commands. We
can define the term head-command as follows:
(x) AgrO
[V] AgrO
/ \ / \
V P D AgrO
I I
turned off them
The head which immediately dominates the V turned here is the bracketed [V] node,
and since this does not dominate the D-node containing them, the verb turned cannot
check the objective case of the pronoun them. Since case-features are uninterpretable at
LF, the relevant derivation crashes at LF. Of course, the weakness of this account is
that we have given no independent motivation for the head-command condition
(though command constraints of various kinds seem to be required elsewhere in the
grammar - e.g. in relation to the syntax of scope). {You take over at this point, and
discuss the syntax of the pronouns in the remaining examples. Try and establish
whether some form of case-checking account can be made to work for more complex
examples, or whether we have to fall back on a PF cliticization account. Be warned
that this is a difficult exercise which you need to think through very carefully, taking a
range of alternative possibilities into account.)
Exercise VIII
The sentences below illustrate a phenomenon which is sometimes
referred to as pronoun shift which affects (phonetically) weak object pronouns (the
relevant weak pronouns are italicized in the examples below). The examples in (1-3)
are imperative structures found in what Alison Henry (1995) terms ‘dialect B’ of
Belfast English (here abbreviated to BEB), while those in (4-5) are from Early Modem
English (= EME) and come from various plays by Shakespeare:
483
Agreement projections
(I am grateful to Alison Henry for providing examples lc and 3a-d; the other examples
in 1 and 2 are from Henry 1995.) Discuss the syntax of weak object pronouns in these
two varieties of English.
Helpful hints
Make the following assumptions about BEB (adapted from Henry 1995):
(i) imperative verbs in BEB adjoin to an Imp morpheme in C in order to check their
imperative morphology; (ii) always in the relevant sentences is adjoined to the left of
vp; (iii) subjects either remain in situ (in spec-vp) or raise to spec-AgrSP; (iv) impera¬
tive clauses contain no TP projection, since imperative verbs have no tense inflections
to check; (v) AgrIOP and AgrOP are positioned between AgrSP and vp; (vi) only weak
direct and indirect object pronouns check their case by raising overtly to an appropriate
position within AgrOP/AgrIOP; other direct and indirect object D(P)s remain in situ
within VP.
484
Workbook section
Let’s also suppose (contrary to what was assumed in the text) that AgrOP is positioned
above vp. Henry (1995, p. 75) suggests that weak object pronouns move to spec-
AgrOP to check their case; however, this seems unlikely, since it would mean that
AgrO has strong specifier-features, and we would therefore expect that strong (e.g.
contrastively stressed) object pronouns and objective DPs like the book would also
raise in the same way - yet this is not the case, as we see from the ungrammaticality of
If. Since only weak object pronouns can raise, and weak elements are arguably affixal
in nature, it seems more likely that weak object pronouns adjoin to AgrO (together with
the verb), as in (ii) below:
The objective case-features carried by the pronoun it and the transitive verb read can
then be checked and erased (checking here involving a relation between two con¬
stituents adjoined to the same functional head, AgrO). Subsequently, the verb read
excorporates out of AgrO and adjoins to AgrS (recall that Henry argues that impera¬
tives contain no TP projection); in addition, the subject you raises to spec-AgrSP
(where the case and agreement properties of you are checked against those of the
imperative verb read: note that imperative verbs license only a restricted choice of sub¬
ject — cf. *Read we/he the book!), as in (iii) below:
485
Agreement projections
(iii) AgrSP
AgrOP
... to me
Finally, the verb read raises to adjoin to the strong Imp morpheme in C, as in (iv)
below:
(iv)
/ \ /
t AgrO ADV vp
' \ I / \
it AgrO always ... to me
The analysis presented here is in effect a hybrid, grafting the vp/VP analysis of clause
structure outlined in chapter 9 onto Henry’s assumptions that AgrOP is positioned out¬
side the projection in which subjects originate (in our terms, vp), and that only weak
pronouns raise to a position within AgrOP
This exercise serves to illustrate the diversity of views in current work on what kinds
of objects check their case by raising to an appropriate position in AgrOP (only weak
pronouns in specific varieties of English, or all objective D(P)s in all varieties?), and
where AgrOP is positioned (above or below vp?). Anyone who wants to maintain that
(in BEB and EME) all direct and indirect object DPs raise to the specifier position
within an appropriate agreement phrase which is positioned below vp will clearly have
to maintain that weak object pronouns in BEB and EME raise to adjoin to some
higher functional head F positioned somewhere above vp: in this connection, it is inter¬
esting to note Uriagereka’s (1995) claim that, in Spanish, weak object pronouns
raise out of vp to adjoin to a higher functional head F which heads a separate FP pro¬
jection. Of course, given the assumption that constituents only move to check gram¬
matical features, we would then have to ask what kind of feature is checked by
adjoining weak pronouns to F. (Not case, on this account, since this would already
have been checked in AgrOP; perhaps some discourse feature, relating to the fact that
486
Workbook section
weak pronouns - like topics - typically represent old information which is familiar
from the discourse.)
Exercise IX
Discuss the syntax of the Early Modem English sentences below (from
various plays by Shakespeare). Which sentences prove problematic to derive, and
why?
from her V D
stole it
487
Agreement projections
VP then merges with an AgrO constituent; the transitive V stole adjoins to AgrO and
the D it raises to spec-VP to check its objective case as in (ii) below:
(ii) AgrOP
V AgrO PP V
| / \ / \
stole from her V D
* I I
: . . t t
AgrOP then merges with a light verb which has a null (second person singular) pro
subject, and the V stole adjoins to the light verb as in (iii) below:
t from her V D
t t
The vp constituent in (iii) then merges with a T constituent containing the perfective
auxiliary hast, and pro raises to spec-TP to check its nominative case (and to check that
hast agrees with it), as in the simplified structure (iv) below:
(iv) TP
488
Workbook section
And (v) is a simplified representation of the structure of sentence 1 Hast stole it from
her?
489
GLOSSARY AND LIST OF
ABBREVIATIONS
(Abbreviations used here are: ch. - chapter; ex. = exercise; §6.2 = chapter 6, section 2; bold-
printed terms within glosses denote technical terms, and generally cross-refer to related entries
elsewhere in the glossary.)
A See adjective.
A-bar movement An A-bar movement operation is one which moves a maximal projection
into an A-bar position (i.e. a nonargument position, or more specifically, a posi¬
tion which can be occupied by expressions which are not arguments). So, oper¬
ator movement, scrambling and the kind of adjunction operation whereby
this kind of behaviour is adjoined to the clause containing it in a sentence such
as This kind of behaviour we cannot tolerate are all specific types of A-bar
movement operation.
Absorption A passive participle is said to absorb the theta-role which a verb would otherwise
assign to its subject (thereby dethematizing the subject) and to absorb the objec¬
tive case which a transitive verb would otherwise assign to its object (thereby
detransitivizing the verb, so that passive participles are intransitive). See §8.9.
Acceptability Native speakers have the ability to judge sentences to be acceptable or unac¬
ceptable. However, the fact that a sentence is regarded as unacceptable doesn’t
mean that it is necessarily ungrammatical (i.e. syntactically ill formed), since
(for example) a sentence might be regarded as unacceptable because it is seman¬
tically incoherent, stylistically incongruous, pragmatically infelicitous, etc. See
ch. 1, ex. VII.
Accusative See case.
Acquisition The study of the way in which children acquire their first language (i.e. their
mother tongue) - referred to more precisely as LI acquisition, to differentiate it
from L2 acquisition, the acquisition of a second language (i.e. a language which
is not one’s mother tongue).
Active A contrast is traditionally drawn between sentence pairs such as (i) and (ii) below:
Example (i) is said to be an active clause (or sentence), and (ii) to be its passive
490
Glossary and abbreviations
counterpart; similarly, the verb stole is said to be an active verb (or a verb in the
active voice) in (i), whereas the verb stolen is said to be a passive verb (or a verb
in the passive voice - more specifically, a passive participle) in (ii); likewise, the
auxiliary were in (ii) is said to be a passive auxiliary.
Adequacy, criteria of These are the criteria which an adequate grammar or linguistic theory
must meet. See §1.3.
Adjacency condition A condition requiring that two expressions must be immediately adja-
cent (i.e. one must immediately follow the other) in order for some operation to
apply. For example, to can only contract onto want (forming wanna) if the two
are immediately adjacent.
Adjective A category of word which often denotes states (e.g. happy, sad), which typically
has an adverb counterpart in +ly (cf. sad!sadly), which typically has compara¬
tive/superlative forms in +er/+est (cf. sadder/saddest), which can often take the
prefix +un (cf. unhappy), and which can often form a noun by the addition of
+ness (cf. sadness). See §§2.2 and 2.3.
Adjoin See adjunction.
Adjunct One way in which this term is used is to denote an optional constituent typically
used to specify e.g. the time, location or manner in which an event takes place
(e.g. in the pub is an adjunct in a sentence such as We had a drink in the pub).
Another way in which it is used is to denote a constituent which has been
adjoined to another to form an extended constituent. (See adjunction.)
Adjunction A process by which one word is adjoined (= attached) to another to form a larger
word, or one phrase is adjoined to another phrase to form a larger phrase. For
example, we might say that in a sentence such as He shouldn’t go, not (in the
guise of its contracted form n’t) has been adjoined to the auxiliary should to
form the negative auxiliary shouldn't. Likewise, in a sentence such as You know
that such behaviour we cannot tolerate, we might argue that such behaviour has
been adjoined to the wc-clause. See §10.2.
Adposition A cover term for preposition and postposition. For example, the English word in
is a preposition since it is positioned before its complement (cf. in Tokyo),
whereas its Japanese counterpart ni is a postposition because it is positioned
after its complement (cf. Tokyo ni). Both words are adpositions.
ADV/Adverb A category of word which typically indicates manner (e.g. wait patiently) or
degree (e.g. exceedingly patient). In English, most (but not all) adverbs end in
+ly (cf. quickly - but also almost). See §2.2 and 2.3.
Adverbializing suffix A derivational suffix which converts another class of word into an
adverb. For example +ly is an adverbializing suffix which converts an adjective
like sad into the corresponding adverb sadly.
Affective An affective constituent is a (negative, interrogative or conditional) expression
which can have a polarity expression like (partitive) any in its scope. So, for
example, interrogative whether is an affective constituent as we see from the
fact that a whether-clause can contain partitive any in a sentence such as He
asked me whether I had any news about Jim.
491
Glossary and abbreviations
492
Glossary and abbreviations
*Himselfis waiting, since the anaphor himself here has no antecedent. Anaphora
is the phenomenon whereby an anaphor refers back to an antecedent.
Animate A term used to describe the property of denoting a living being (e.g. a human being
or animal); the corresponding term inanimate is used in relation to an expression
which denotes lifeless entities. For example, the pronoun he is often said to be ani¬
mate in gender, and the pronoun it to be inanimate, since he can refer to a male
human being or male animal, whereas it is typically used to refer to a thing.
Antecedent An expression which is referred to by a pronoun or anaphor of some kind. For
example, in John cut himself shaving, John is the antecedent of the anaphor
himself, since himself refers back to John. In a sentence such as He is someone
whom we respect, the antecedent of the pronoun whom is someone.
Antonym A term used to denote an expression which has the opposite meaning to another
expression: e.g. tall is the antonym of short (and conversely) - hence tall and
short are antonyms.
AP Adjectival phrase - i.e. a phrase headed by an adjective - e.g. fond of chocolate,
keen on sport, good at syntax, etc.
A position A position which can be occupied by an argument, but not by a nonargument
expression (e.g. not by an adjunct) - e.g. a subject position, or a position as the
complement of a verb, adjective or noun. See ch. 8.
Arbitrary When we say that an expression has arbitrary reference, we mean that it can
denote an unspecified set of individuals, and hence have much the same mean¬
ing as English one/people, French on, German Man, etc. In a sentence such as It
is difficult [to learn Japanese], the bracketed clause is said to have an abstract
pronoun subject PRO which has arbitrary reference, so that the sentence is para-
phraseable as ‘It’s difficult for people to learn Japanese.’ See §4.2.
Argument This is a term borrowed by linguists from philosophy (more specifically, from
predicate calculus) to describe the role played by particular types of expression
in the semantic structure of sentences. In a sentence such as John hit Fred, the
overall sentence is said to be a proposition (a term used to describe the semantic
content of a clause), and to consist of the predicate hit and its two arguments
John and Fred. The two arguments represent the two participants in the act of
hitting, and the predicate is the expression (in this case the verb hit) which
describes the activity in which they are engaged. By extension, in a sentence
such as John says he hates syntax the predicate is the verb says, and its two
arguments are John and the clause he hates syntax; the second argument he
hates syntax is in turn a proposition whose predicate is hates, and whose two
arguments are he and syntax. Since the complement of a verb is positioned inter¬
nally within V-bar (in terms of the analysis in ch. 8) whereas the subject of a
verb is positioned outside V-bar, complements are also referred to as internal
arguments, and subjects as external arguments. Expressions which do not func¬
tion as arguments are nonarguments. The argument structure of a predicate pro¬
vides a description of the set of arguments associated with the predicate, and the
thematic role which each fulfils in relation to the predicate. See §8.4.
493
Glossary and abbreviations
the auxiliary has is said to be an auxiliary which marks perfective aspect, in that
it marks the perfection (in the sense of ‘completion’ or ‘termination’) of the
activity of taking the medicine; for analogous reasons, taken is said to be a per¬
fective (participle) verb form in (i) (though is referred to in traditional grammars
as a past participle). Similarly, is is said to be an auxiliary which marks imper-
fective or progressive aspect in (ii), because it relates to an activity which is not
yet perfected (i.e. ‘completed’) and hence which is ongoing or in progress (for
this reason, is in (ii) is also referred to as a progressive or imperfective auxil¬
iary); in the same way, the verb taking in (ii) is said to be the imperfective or
progressive (participle) form of the verb (though is known in traditional gram¬
mars as a present participle).
Aspectual auxiliaries Auxiliaries which mark aspect - e.g. perfective have and progressive
be. See aspect.
Associate An expression which represents the thematic subject in an expletive construc¬
tion, and which is associated with the expletive subject there: e.g. someone in
There's someone waiting for you.
Attraction An operation by which features carried by one constituent percolate up to (and
are inherited by) another. See §6.5.
Attributive adjectives These are adjectives which are used to modify a following noun
expression - e.g. red in John has a red Ferrari, where red attributes the property
of being red to the noun Ferrari. Attributive adjectives contrast with predicative
adjectives, which are adjectives used in structures such as The house was red.
They painted the house red, etc. (where the property of being red is said to be
predicated of the expression the house).
AUX/Auxiliary A term used to describe items such as willlwouldlcantcould!shall!should/
may/might/must/ought and some uses of have/be/do/need/dare. Such items
differ from typical lexical verbs e.g. in that they undergo inversion (cf. Can I
help you?). See §2.5.
Auxiliary inversion See inversion.
Auxiliary phrase A phrase/clause/sentence headed by an auxiliary. For example, a sentence
such as He will help is an auxiliary phrase headed by the modal auxiliary will
(hence also a modal phrase, or will-phrase).
Auxiliary selection This term relates to the type of verb which a given auxiliary selects as its
complement: e.g. in many languages (the counterpart of) be when used as a per¬
fective auxiliary selects only a complement headed by an unaccusative verb
494
Glossary and abbreviations
(like come, go, etc.), whereas (the counterpart of) have selects a complement
headed by other types of verb. See §9.9.
Baa Baa Black Sheep variety A variety of English which allows the verb have (in its posses-
sive use, where it has much the same meaning as possess) to be used as an auxil¬
iary, and hence to undergo inversion in sentences like Have you any wool? (the
second line from a well-known nursery rhyme called Baa Baa Black Sheep).
-bar An X-bar/X constituent is an intermediate projection of some head X - i.e. a
projection which is larger than X but smaller than XP (see projection). In
another use of the term (in which the -bar suffix has much the same function as
the prefix non-), mA-bar/A position is a nonargument position. See argument.
Bare A bare infinitive clause is a clause which contains a verb in the infinitive form,
but does not contain the infinitive particle to - e.g. the bracketed clause in He
won't let [me help him], A bare noun is a noun used without any determiner to
modify it (e.g. fish in Fish is smelly). Bare phrase structure is a system of repre¬
senting syntactic structure in terms of unlabelled tree diagrams (i.e. tree dia¬
grams in which nodes do not carry category labels).
Base form The base form of a verb is the simplest, uninflected form of the verb (the form
under which the relevant verb would be listed in an English dictionary) - hence
forms like go/be/have/see/want/love are the base forms of the relevant verbs.
Binarity principle A principle of Universal Grammar specifying that all nonterminal nodes
in tree diagrams are binary-branching. See §3.5.
Binary A term relating to a two-valued property or relation. For example, number is a
binary property in English, in that we have a two-way contrast between singular
forms like cat and plural forms like cats. It is widely assumed that parameters
have binary settings, that features have binary values, and that all branching in
syntactic structure is binary.
Binary-branching A tree diagram in which every nonterminal node (i.e. every node not at
the very bottom of the tree) branches down into two other nodes is binary¬
branching.
Bind/Binder To say that one constituent * binds (or serves as the binder for) another con-
stituent y (and conversely that y is bound by x) is to say that x determines the
semantic (and grammatical) properties of y. For example, in John wants to PRO
leave, John binds (and is the binder for) PRO, and PRO is bound by John.
Bound In one use of this term, a bound form is one which cannot stand alone and be
used as an independent word, but rather must be attached to some other mor¬
pheme (e.g. negative n't, which has to attach to some auxiliary such as could). In
a completely different use of the term, a bound constituent is one which has a
binder (i.e. antecedent) within the structure containing it (see bind).
Bracketing A technique for representing the categorial status of an expression, whereby the
expression is enclosed in square brackets, and the lefthand bracket is labelled
with an appropriate category symbol - e.g. [D the]. See §2.7.
Branch A term used to represent a solid line linking a pair of nodes in a tree diagram,
marking a mother/daughter relation between them.
495
Glossary and abbreviations
One problem posed by this (traditional) classification relates to the fact that so-
called ‘genitive’ pronouns like my/your/their etc. have the variant forms
mine/yours/theirs - and it is not obvious why such items should have two differ¬
ent genitive case forms. An alternative analysis of items like these is to say that
they are not genitive pronouns but rather are determiners, which have the form
my/your/their when used prenominally, and the form mine/yours/theirs when
used pronominally. Similarly, some linguists have claimed that the so-called gen¬
itive ’s inflection is actually a determiner rather than a case inflection, so that an
expression such as John's father is a determiner phrase whose head determiner is
’s, whose specifier is John and whose complement is father: if so, then this raises
the possibility that English simply has no genitive case at all. However, it has
been suggested by others that of serves as a genitive case marker (perhaps
belonging to the category K of case particle) in expressions such as fond of
whisky, loss of face, etc. (since in languages with richer case morphology than
English, the relevant o/-phrases correspond to a nominal which is morphologi¬
cally marked for genitive case: see ch. 2, ex. VI). In chapter 10, we suggest that
indirect objects in English (e.g. him in I gave him some) carry dative case, even
though dative forms are not morphologically distinct from objective forms. In
some works (e.g. Chomsky 1980) it is claimed that transitive verbs assign objec¬
tive case to their complements, whereas transitive prepositions assign oblique
case to their complements (although the two cases are not morphologically dis¬
tinct in English): under this analysis, him in Fetch him! would carry objective
case, whereas him in Fetch one for him would carry oblique case. Belletti (1988)
suggests that in sentences like There could have been an accident, the expression
an accident carries partitive case (Finnish has overt marking of partitive case
forms). In Chomsky and Lasnik 1995, it is suggested that the null subject PRO
found in (control) infinitive constructions carries null case.
Caseless A caseless constituent is one which has no case properties (i.e. which neither
496
Glossary and abbreviations
carries a case of its own nor checks the case of another constituent). A caseless
position is a position in which no case can be checked (hence which cannot be
occupied by a constituent carrying case).
Case particle See case.
Case position A position in which some case is checked.
Categorization Assigning an expression to a (grammatical) category.
Category A term used to denote a set of expressions which share a common set of linguis¬
tic properties. In syntax, the term is used for expressions which share a common
set of grammatical (i.e. morphological/syntactic) properties. For example, boy
and girl belong to the (grammatical) category noun because they both inflect for
plural number (cf. boys/girls), and can both terminate a sentence such as The
police haven't yet found the missing_. See ch. 2.
Causative verb A verb which has much the same sense as ‘cause’. For example, the verb
have in sentences such as He had them expelled or He had them review the case
could be said to be causative in sense (hence to be a causative verb).
C-command A structural relation between two categories. To say that one category X c-com-
mands another category Y is (informally) to say that X is no lower than Y in the
structure (i.e. either X is higher up in the structure than T, or the two are at the
same height). If you think of X and Y as different stations in a train network, we
can say that X c-commands Y if you can get from X to Y by taking a northbound
train from X, getting off at the first stop, and then taking a southbound train to Y
(on a different line). More formally, X c-commands Y if the mother of X domi¬
nates Y, and X and Y are disconnected (i.e. X * Y, and neither dominates the
other). The c-command condition on binding is a condition to the effect that a
bound constituent (e.g. a reflexive anaphor like himself or the trace of a moved
constituent) must be c-commanded by its antecedent (i.e. by the expression
which binds it). This amounts to claiming that the antecedent must be higher up
in the structure than the anaphor/trace which it binds. See §3.9.
CED See condition on extraction domains.
Chain A set of one or more constituents comprising an expression and any traces asso¬
ciated with it.
Chain uniformity principle/condition A principle of Universal Grammar requiring that a
chain should be uniform with regard to its phrase structure status. See §7.4.
Checked/Checker In a sentence such as He has left, the auxiliary has checks the nominative
case-feature carried by he: accordingly, has is said to be the checker (for the rel¬
evant nominative case-feature), and he the checked.
Checking (domain/theory) In Chomsky’s checking theory, words carry grammatical features
which have to be checked in the course of a derivation. For example, a nomina¬
tive pronoun like / must have its nominative case checked, which means that it
must occupy a nominative position (as the subject of the kind of constituent
which allows a nominative subject, e.g. a finite auxiliary) at some point in the
derivation. When a feature has been checked, it is erased if it is uninterpretable
(i.e. if it is a purely formal feature with no semantic content). Any uninterpretable
497
Glossary and abbreviations
features which remain unchecked (and hence which have not been erased) at the
level of logical form will cause the derivation to crash (i.e. to be ungrammatical).
See ch. 5. For checking domain, see domain.
Clause A clause is defined in traditional grammar as an expression which contains a
subject and a predicate, and which may contain other types of expression as
well (e.g. a complement and an adjunct). In most cases, the predicate in a clause
is a lexical (= nonauxiliary) verb, so that there will be as many different clauses
in a sentence as there are different lexical verbs. For example, in a sentence such
as She may think that you are cheating on her, there are two lexical verbs (think
and cheating), and hence two clauses. The cheating-clause is that you are cheat¬
ing on her, and the think-clause is She may think that you are cheating on her, so
that the cheating-clause is one of the constituents of the think-clause. More
specifically, the cheating-clause is the complement of the think-clause, and so is
said to function as a complement clause in this type of sentence.
Clitic(ization) The term clitic denotes an item which resembles a word but which has the
property that it must cliticize (i.e. attach itself) to another word. For example,
we could say that the contracted negative particle n’t is a clitic which attaches
itself to a finite auxiliary verb, so giving rise to forms like isn’t, shouldn’t, might¬
n’t, etc. Likewise, we might say that ’ve is a clitic form of have which attaches
itself to (for example) a pronoun ending in a vowel or diphthong, so giving rise
to forms like we've, you’ve, they’ve, etc.
Cognition/Cognitive (Relating to) the study of human knowledge.
Common Noun In traditional grammar, common nouns are contrasted with proper nouns.
Proper nouns are names of individual people (e.g. Chomsky), places (e.g.
Colchester, Essex, England), dates (e.g. Tuesday, February, Easter), magazines
(e.g. Cosmopolitan), etc., whereas common nouns (e.g. boy, table, syntax, etc.)
are nouns denoting general (nonindividual) entities. Proper nouns have the
semantic property of having unique reference, and the syntactic property that
they generally can't be modified by a determiner (cf. *the London).
COMP See complementizer.
Comparative The comparative form of an adjective or adverb is the +er form used when
comparing two individuals or properties: cf. John is taller than Mary, where
taller is the comparative form of tall.
Competence A term used to represent fluent native speakers' knowledge of the grammar of
their mother tongue(s). See ch. 1.
Complement This is a term used to denote a specific grammatical function (in the same way
that the term subject denotes a specific grammatical function). A complement is
an expression which combines with a head word to project the head into a larger
structure of essentially the same kind. In close the door, the door is the comple¬
ment of close; in after dinner, dinner is the complement of after, in good at
physics, at physics is the complement of good', in loss of face, of face is the
complement of loss. As these examples illustrate, complements typically follow
their heads in English. The choice of complement (and the morphological form
498
Glossary and abbreviations
499
Glossary and abbreviations
tired but otherwise all right, the two conjuncts (i.e. expressions which have been
conjoined) are rather tired and otherwise all right.
Conjunction A word which is used to join two or more expressions together. For example, in
a sentence such as John was tired but happy, the word but serves the function of
being a coordinating conjunction because it coordinates (i.e. joins together)
the adjectives tired and happy. In John felt angry and Mary felt bitter, the con¬
junction and is used to coordinate the two clauses John felt angry and Mary felt
bitter. In traditional grammar, complementizers like that!fori if axe categorized
as (one particular type of) subordinating conjunction.
Constituent A structural unit - i.e. an expression which is one of the components out of
which a phrase or sentence is built up. For example, the various constituents of a
prepositional phrase (= PP) such as straight into touch (e.g. as a reply to Where
did the ball go?) are the preposition into, the noun touch, the adverb straight and
the intermediate projection (P-bar) into touch.
Constituent structure The constituent structure (or phrase structure, or syntactic structure)
of an expression is (a representation of) the set of constituents which the expres¬
sion contains. Constituent structure is usually represented in terms of a labelled
bracketing or a tree diagram.
Constrained See restrictive.
Constraint A principle of Universal Grammar which prohibits certain types of grammatical
operation from applying to certain types of structure.
Content This term is generally used to refer to the semantic content (i.e. meaning) of an
expression. Flowever, it can also be used in a more general way to refer to other
linguistic properties of an expression: e.g. the expression phonetic content is
used to refer to the phonetic form of an expression: hence, we might say that
PRO is a pronoun which has no/null phonetic content (meaning that it is a silent
pronoun with no audible form).
Contentives/content words Words which have intrinsic descriptive content (as opposed to
functors, i.e. words which serve essentially to mark particular grammatical
functions). Nouns, verbs, adjectives and (most) prepositions are traditionally
classified as contentives, while pronouns, auxiliaries, determiners, complemen¬
tizers and particles of various kinds (e.g. infinitival to) are classified as functors.
See §2.4.
Contracted form/Contraction Contraction is a term used in §8.5 to indicate a process by
which a word like have comes to be reduced to the contracted form /v/, and
thereby loses its vowel nucleus (along with the consonant /hi). Compare
reduced form.
Control(ler)/Control predicate In infinitive structures with a PRO subject which has an
antecedent, the antecedent is said to be the controller of PRO (or to control
PRO), and conversely PRO is said to be controlled by its antecedent; and the
relevant kind of structure is called a control structure. So, in a structure like
John decided to PRO quit, John is the controller of PRO, and conversely PRO is
500
Glossary and abbreviations
controlled by John. The term control predicate denotes a word like try which
takes an infinitive complement with a (controlled) PRO subject. Verbs like try
which takes a complement containing a PRO subject controlled by the subject of
try are called subject-control predicates (see §4.2): verbs like persuade in sen¬
tences like I persuaded him to take syntax which take an infinitive complement
whose PRO subject is controlled by the object of the main verb (here, the him
object of persuade) are called object-control predicates (see §9.6).
Convergence A derivation converges if the resulting PF representation contains only pho¬
netically interpretable features, and the associated LF representation contains
only semantically interpretable features.
Coordinate/Coordination A coordinate structure is a structure containing two or more
expressions joined together by a coordinating conjunction such as and/but/or/
nor (e.g. John and Mary is a coordinate structure.). Coordination is the process
by which two or more expressions are joined together by a coordinating con¬
junction.
Copula/Copular verb A verb used to link a subject with a verbless predicate. The main
copular verb in English is be (though verbs like become, remain, stay, etc. also
have the same copular - i.e. linking - function). In sentences such as They are
lazy. They are fools and They are outside, the verb are is said to be a copula in
that it links the subject they to the adjective predicate lazy, or the noun predicate
fools, or the prepositional predicate outside.
Coreferential Two expressions are coreferential if they refer to the same entity. For exam¬
ple, in John cut himself while shaving, himself dead John are cofererential in the
sense that they refer to the same individual.
Countability The countability properties of a noun determine whether the relevant item is a
count noun or not.
Counterexample An example which falsifies a particular hypothesis. For example, an auxil¬
iary like ought would be a counterexample to any claim that auxiliaries in
English never take an infinitive complement introduced by to (cf. You ought to
tell them).
Count noun A noun which can be counted. Flence, a noun such as chair is a count noun
since we can say one chair, two chairs, three chairs, etc.; but a noun such as fur¬
niture is a noncount noun or mass noun since we cannot say *one furniture,
*two furnitures, * three furnitures, etc.
Covert A covert expression is one which has no phonetic content (i.e. which is empty of
phonetic content and so is inaudible). For example, in a structure such as He
may try [to PRO escape from prison], the bracketed clause has a covert (= null =
empty = silent) subject pronoun PRO. A covert feature (or property) is one
which has no overt morphological realization: for example, we might say that an
invariable noun like sheep has covert (singular/plural) number properties, since
(unlike cat/cats) the word sheep doesn’t add +s in the plural (cf. one sheep/two
sheep).
501
Glossary and abbreviations
502
Glossary and abbreviations
503
Glossary and abbreviations
504
Glossary and abbreviations
Dummy A type of word which has no intrinsic semantic content, but which is used
simply to satisfy a structural requirement that a certain position in a structure be
filled. For example, the auxiliary do in a sentence such as Does he like pasta? is
said to be a dummy, satisfying the need for COMP to be filled in questions.
Likewise, the first occurrence of the pronoun there in a sentence like There is
nobody there is a dummy (i.e. contentless) pronoun, since it cannot have its ref¬
erence questioned (cf. * Where is someone there?).
Early Modern English The type of English found in the early seventeenth century (at
around the time Shakespeare wrote most of his plays, between 1590 and 1620).
In the text, all examples of Early Modem English are taken from various plays
by Shakespeare. It should perhaps be noted that some linguists have suggested
that Shakespeare’s English is rather conservative, and hence reflects a slightly
earlier stage of English.
Echo question A type of sentence used to question something which someone else has just
said (often with an air of incredulity), repeating all or most of what they have just
said. For example, if I say I’ve just met Nim Chimpsky and you don’t believe me
(or don’t know who I’m talking about), you could reply with an echo question
such as You 've just met who ?
ECM See exceptional case-marking.
Economy principle A principle which requires that (all other things being equal) syntactic
representations should contain as few constituents and syntactic derivations
involve as few grammatical operations as possible.
Ellipsis/Elliptical Ellipsis is a process by which an expression is omitted in order to avoid
repetition. For example, in a sentence such as / will do it if you will do it, we can
ellipse (i.e. omit) the second occurrence of do it to avoid repetition, and hence
say / will do it if you will: the resulting sentence is an elliptical structure (i.e. a
structure from which something has been omitted).
Embedded clause An embedded clause is a clause which is positioned internally within some
other phrase or clause. For example, in a sentence such as He may suspect that /
hid them, the hid-clause (= that I hid them) is embedded within the suspect clause.
EME See Early Modern English.
Empirical evidence Evidence based on observed linguistic phenomena. In syntax, the term
empirical evidence usually means ‘evidence based on grammaticality judgments
by native speakers’. For example, the fact that sentences like *Himself likes you
are judged ungrammatical by native speakers of Standard English provides us
with empirical evidence that anaphors like himself can’t be used without an
appropriate antecedent (i.e. an expression which they refer back to).
Empty category A category which is covert (i.e. which is silent or null and hence has no
overt phonetic form). Empty categories include traces, the null pronouns PRO
and pro, the null generic/partitive determiner 0, etc. See ch. 4.
Endocentric Headed: for example, an expression such as fond of pasta is an adjectival
phrase whose head is the adjective fond; hence, the overall expression is endo¬
centric (i.e. headed).
505
Glossary and abbreviations
506
Glossary and abbreviations
Extended projection In a 1991 paper, Jane Grimshaw suggested that IP and CP are extended
projections of V, and that DP and PP are extended projections of N. See §4.8.
External argument Subject (see argument).
Extraction An operation by which one constituent is moved out of another. For example, in
a structure such as Who do you think [he saw —] ? the pronoun who has been
extracted out of the position marked — in the bracketed clause, and moved to
the front of the overall sentence. The extraction site for a moved constituent is
the position out of which it is extracted/moved (marked by — in the example
above).
F This symbol is used as a category label to denote an abstract functional head of
some kind (see §4.9). It is also used (in a different way) to represent the gender-
feature [feminine].
Feature A device used to describe a particular linguistic property (e.g. we might use a
feature such as [Nom] to denote the nominative case-feature carried by pro¬
nouns such as he). By convention, features are normally enclosed in square
brackets, and semantic features written in capital letters. The head-features of an
item describe its intrinsic grammatical properties; the complement-features of an
item determine the range of complements which it allows; the specifier-features
of an item determine the range of specifiers which it allows (see §5.3). \ feature
matrix is a set of features describing the grammatical properties of some particu¬
lar category or expression. The feature specification of a word is (a representa¬
tion of) the set of features which characterize the idiosyncratic properties of the
word. See also strong features and weak features.
Feminine This term is used in discussion of grammatical gender to denote pronouns like
she/her/hers which refer to female entities.
Finite The term finite verb!clause denotes an auxiliary or nonauxiliary verb or clause
which can have a subject with nominative case like I/we/he/she/they. Thus, if
we compare the two bracketed clauses in:
we find that the bracketed clause and the verb annoy in (i) are finite because
in place of the subject people we can have a nominative pronoun like they, by
contrast, the bracketed clause and the verb annoy are nonfinite in (ii) because
people cannot be replaced by a nominative pronoun like they (only by an objec¬
tive pronoun like them): cf.
By contrast, a verb or clause which has a subject with objective or null case is
nonfinite; hence the bracketed clauses and bold-printed verbs are nonfinite in the
examples below:
507
Glossary and abbreviations
508
Glossary and abbreviations
509
Glossary and abbreviations
510
Glossary and abbreviations
Illocutionary force The illocutionary force of a sentence (or clause) describes the kind of
speech act which it is used to perform (e.g. a sentence is declarative in force if
used to make a statement, interrogative in force if used to ask a question,
imperative in force if used to issue an order, exclamative in force if used to
exclaim surprise, etc.). See Grice 1975.
Immediate constituent The immediate constituents of a given phrase XP are those con¬
stituents which are contained within XP, but not within any other phrase which
is itself contained within XP. In practice, the immediate constituents of XP are
the head X of XP, the complement of X, the specifier of X and any adjunct
attached to a projection of X.
I(-to-C) movement Movement of a verb out of the head I position in IP into the head C
position in CP. See §6.2, and inversion.
Imp A symbol used to designate an (affixal) imperative morpheme which occupies
the head C position of CP in an imperative sentence.
Imperative A term used to classify a type of sentence used to issue an order (e.g. Be quiet!,
Don't say anything!), and also to classify the type of verb form used in an imper-
ative sentence (e.g. be is an imperative verb form in Be quiet!).
Imperfective See aspect.
Impersonal The pronoun it is said to be impersonal when it has no thematic role (and
doesn’t refer to any external entity outside the sentence), e.g. in sentences such
as It is rumoured that he is unhappy, or It is unlikely that he 'll come back.
Impoverished See rich.
Inanimate See animate.
Inchoative An inchoative verb is one which marks the beginning of an action or state. For
example, we might say that become is an inchoative verb in a sentence such as
He didn't become famous until he released his first album, since it can be para¬
phrased as ‘start to be’.
Indefinite See definite.
Indicative Indicative (auxiliary and nonauxiliary) verb forms are finite forms which are
used (inter alia) in declarative and interrogative clauses (i.e. statements and
questions). Thus, the bold-printed items are indicative forms in the following
sentences: He is teasing you, Can he speak French?, He had been smoking. He
loves chocolate, He hated syntax. An indicative clause is a clause which con¬
tains an indicative (auxiliary or nonauxiliary) verb. See mood.
Indices Subscript letters attached to sets of constituents to indicate whether or not there
is a binding relation between them. For example, in a structure such as Johni
thinks that Harry. is deceiving himself., the indices indicate that himself is bound
by (i.e. interpreted as referring to) Harry, not by John.
Indirect object See object.
Indirect question An interrogative complement clause (i.e. an interrogative clause used as
the complement of a word like ask, unsure, question, etc.) such as those brack¬
eted ml wonder [what he will do] and I don’t know [if he will turn up].
Indirect theta-marking See theta-marking.
511
Glossary and abbreviations
512
Glossary and abbreviations
phonetic feature like [nasal] is uninterpretable at LF, and so too are purely gram¬
matical/formal features (e.g. case-features). See §5.2.
Interpretation To say that an expression has a particular interpretation is to say that it
expresses a particular set of semantic relations. So, for example, we might say
that a sentence such as He loves you more than Sam has two different interpreta¬
tions - one on which Sam has a subject interpretation and is implicitly under¬
stood as the subject of loves you, and a second on which Sam has an object
interpretation and is implicitly understood as the object of he loves. The first
interpretation can be paraphrased as ‘He loves you more than Sam loves you’,
and the second as ‘He loves you more than he loves Sam.’
Interrogative An interrogative clause or sentence is one which is used to ask a question.
For example, the overall sentence is interrogative in Is it raining?, and the
bracketed complement clause is interrogative in I wonder [if it is raining].
Intransitive See transitive.
Inversion A term used to denote a movement process by which the relative order of two
expressions is reversed. It is most frequently used in relation to the more specific
operation by which an auxiliary (and, in earlier stages of English, nonauxiliary)
verb comes to be positioned before its subject, e.g. in questions such as Can you
speak Swahili?, where can is positioned in front of its subject you. See §6.2.
Inverted auxiliary/verb An auxiliary/verb which is positioned in front of its subject (e.g.
will in Will I pass the syntax exam?). See §6.2.
IP Inflection phrase, i.e. a phrase/clause which is a projection of INFL. Thus, a
sentence such as It might rain is an IP - more specifically, a projection of the
INFL constituent might.
IP adverb An adverb (like certainly) which is positioned internally within IP (i.e. which is
adjoined to some projection of INFL).
Island A structure which does not allow any constituent to be extracted from it (e.g.
anJ-structures are islands - hence we can’t say *John I admire Harry and, mov¬
ing John out of the and-structure Harry and John).
K Case particle. See case.
Labelled bracketing See bracketing.
Landing-site The landing-site for a moved constituent is the position it ends up in after it has
been moved (e.g. the specifier position within CP is the landing-site for a moved
operator expression).
Language faculty Chomsky has argued that human beings have an innate language faculty
(i.e. brain module) which provides them with an algorithm (i.e. set of procedures
or program) for developing a grammar of their native language(s). See §1.4.
Last resort principle The principle that grammatical operations do not apply unless they have
to as the only way of satisfying some grammatical requirement: for example,
</o-support is used in questions only as a last resort, i.e. if there is no other auxil¬
iary in the structure which can undergo inversion. See §6.2.
Leaf In tree diagrams, the term leaf refers to a word attached to a terminal node in a
tree.
513
Glossary and abbreviations
Learnability A criterion of adequacy for linguistic theory. An adequate theory must explain
how children come to learn the grammar of their native languages in such a
short period of time, and hence must provide for grammars of languages which
are easily leamable by children. See § 1.3.
Least effort principle See economy principle.
Level (of representation) A level of representation (of the structure of a sentence) is a stage (in
a derivation) at which representations comprise only features of a single type.
There are two different levels of representation in a grammar, LF and PF. LF/log-
ical form is the level at which representations include only semantic features;
PF/phonetic form is the level at which representations include only phonetic fea¬
tures. By contrast, the grammatical structures produced by merger and movement
operations do not constitute a separate level of representation, since they contain
three different sets of features (phonetic, grammatical and semantic).
Lexical/Lexicon The word lexical is used in a number of different ways. Since a lexicon is a
dictionary (i.e. a list of all the words in a language and their idiosyncratic lin¬
guistic properties), the expression lexical item means ‘word’, the expression lex¬
ical entry means ‘the entry in the dictionary for a particular word’, the term lexi¬
cal property means ‘property associated with some individual word’ and the
term lexical learning means ‘learning words and their idiosyncratic properties’.
However, the word lexical is also used in a second sense, in which it is con¬
trasted with functional (and hence means ‘nonfunctional’). In this second sense,
a lexical category is a category whose members are contentives (i.e. items with
idiosyncratic descriptive content): hence, categories such as noun, verb, adjec¬
tive or preposition are lexical categories in this sense. So, for example, the term
lexical verb means ‘nonauxiliary verb’ (i.e. a verb like go, find, hate, want, etc.).
LF Logical form. An LF representation (for an expression) is a representation of the
logical form of the expression (see representation). The LF component of a
grammar is the component which converts the syntactic structures produced by
merger and movement operations into LF representations.
LF movement A movement operation which applies in the LF component, and hence which
does not affect the phonetic form of a given sentence.
License To say that a head licenses a certain type of specifier/complement is to say that it
can have such a specifier/complement. For example, a finite auxiliary licenses a
nominative subject (but since this is the only type of subject licensed by a finite
auxiliary, this in effect means that a finite auxiliary must have a nominative sub¬
ject).
Light verb An affixal verb (often with a causative sense like that of make) to which a noun,
adjective or verb adjoins. For example, it might be claimed that the suffix +en in
a verb like sadden is an affixal light verb which can combine with an adjective
like sad to form the causative verb sadden (meaning ‘make sad’, ‘cause to
become sad’). This type of analysis could be extended to verbs like roll as they
are used in sentences like He rolled the ball down the hill, where we could sug¬
gest that roll is used causatively (in the sense of ‘make roll' or ‘cause to roll’).
514
Glossary and abbreviations
and hence involves adjunction of the verb roll to an abstract light verb (i.e. to a
null verbal counterpart of +en). See ch. 9.
Link A constituent (or position) which is part of a movement chain.
Local An operation is local only if it operates within a highly restricted domain (e.g.
internally within a phrase, or across no more than one intervening phrasal
boundary). For example, agreement typically involves a local relation between
the head and specifier of a given type of phrase (e.g. in a sentence like He has
gone, between the INFL constituent has and its specifier/subject he). Similarly,
anaphors like himself typically require a local antecedent (i.e. an antecedent
within the phrase containing them). A movement operation like head move¬
ment is local in the sense that a head can only move into the next-highest head
position within the structure (and so can cross only one intervening phrase
boundary containing it).
Locative A locative expression is one which denotes place. So, for example, there/where
are locative pronouns in sentences such as Are you going there? or Where are
you going?
Main clause See root.
Map To say that one structure is mapped into another is to say that it is ‘transformed’
or ‘converted’ or ‘changed’ into the other structure by a grammatical operation
of some kind.
Masculine A term used in discussions of grammatical gender to denote pronouns like
he/him/his which refer to male entities.
Mass noun See count noun.
Matrix In a sentence like I think [you are right], to say that the think-clause is the matrix
clause for the bracketed complement clause is to say that it is the clause which
immediately contains the bracketed clause (hence that the bracketed clause is
embedded within the think-clause). For a different use of the term, see feature
(matrix).
Maximal projection See projection.
Merger An operation by which two categories are combined to form another category.
See ch. 3.
Minimal domain See domain.
Minimalism/Minimalist program A theory of grammar (outlined in Chomsky 1995b)
whose core assumption is that grammars should be described in terms of the
minimal set of theoretical and descriptive apparatus necessary.
Minimal link condition/Minimality condition A principle of grammar requiring that the
links in movement chains should be as short as possible (hence that constituents
should move from one position to another in the shortest possible steps). See
also shortest movement principle.
Minimal projection See projection.
MIT The Massachusetts Institute of Technology (located in Cambridge,
Massachusetts), where Chomsky has worked for the past four decades.
Mod(al)/Modality A modal auxiliary is an auxiliary which expresses modality (i.e. a notion
515
Glossary and abbreviations
516
Glossary and abbreviations
Multiple wh-questions Questions containing more than one wh-word. See §7.3.
Multiword speech A stage in child language acquisition at which (one-year-old) children
first begin combining words together into two- and three- (etc.) word utterances.
N/N/N-bar/NP N represents the category of noun', N/N-bar is an intermediate projection
headed by N; and NP (noun phrase) is a maximal projection headed by N. See
noun, noun phrase and projection; see also ch. 4.
+n The inflection used to form the perfective/passive participle form of a verb (see
aspect, active). For example, shown is a perfective participle in The referee has
already shown him the yellow card once, but is a passive participle in He has
already been shown the yellow card once. The term n-participle refers to the
perfective/passive participle form of a verb (for some verbs, this may end in
+d/+t: cf. He has tried. She has bought one).
Natural language Human language; more specifically, a language acquired in a natural set¬
ting by human beings (hence, excluding e.g. computer languages, animal com¬
munication systems, etc.).
NEG/NEGP A negative constituent which heads a projection into a NEGP (negative
phrase): see §6.6.
Negation A process or construction in which some proposition is said to be false.
Negation involves the use of some negative item such as not, n’t, nobody, noth¬
ing, never, etc. - though most discussions of negation tend to be about not/n’t.
Negative evidence In the context of discussions of child language acquisition, this term
relates to evidence based on the nonoccurrence of certain structures in the
child’s speech input. See §1.9.
Negative particle This informal term typically denotes not/n't.
Neuter See gender.
Neutralization When a morphological contrast (e.g. that between a singular noun like cat
and a plural noun like cats) is not marked in some expression (e.g. the
singular/plural noun form sheep), the contrast is said to have been neutralized
(in the relevant item).
No-contentless-projections constraint The informal name given in the text for a constraint
proposed in Speas 1995 to the effect that projections cannot have heads and
specifiers which lack content. This amounts to a requirement to the effect that
either the head of a projection must have content, or its specifier must - or both
of them.
Node A term used to denote a point in a tree diagram which carries a category label.
Nom An informal abbreviation for the case-feature [nominative]. See case.
Nominal This is the adjective associated with noun, so that in principle a nominal or a
nominal constituent is an expression headed by a noun. However, the term is
often extended to mean ‘expression which is a projection or extended projection
of a noun or pronoun’. In current work, a phrase like a supporter of monetarism
would be analysed as a determiner phrase, and hence is not a nominal, if by
nominal we mean ‘expression headed by a noun’. However, if we say that the
DP (= determiner phrase) here is an extended projection of the noun supporter.
517
Glossary and abbreviations
we can none the less continue to say that it is a nominal. Further confusion is
caused by the fact that in earlier work, what are now analysed as determiner
phrases would then have been analysed as noun phrases.
Nominalization/Nominalizing affix Nominalization is a process by which some other type
of expression is converted into a nominal (i.e. noun expression). For example,
+ness is a nominalizing affix in that if we suffix +ness to an adjective like sad,
we form the noun sadness.
Nominative See case.
Nonargument See argument.
Nonauxiliary verb A lexical verb (like want, try, hate, smell, buy, etc.) which requires do-
support to form questions, negatives and tags.
Nonconstituent string/sequence A string/sequence of words which do not together form a
constituent.
Noncount noun See count noun.
No-negative-evidence hypothesis The hypothesis that children acquire their native lan¬
guage^) on the basis of positive evidence alone, and do not make use of nega¬
tive evidence. See § 1.9.
Nonfinite See finite.
Nonoperator question A question which does not contain an interrogative operator. See
§7.9.
Nonterminal See terminal node.
Nonuniform chain See chain uniformity principle.
Nonvocalic Not containing a vowel or diphthong.
Noun A category of word (whose members include items such as boyIfriendlthought!
sadness/computer) which typically denotes an entity of some kind. See §2.2.
Noun phrase A phrase whose head is a noun. Thus, the expression lovers of opera is a noun
phrase, since its head is the noun lovers. In earlier work, determiners were
thought to be the specifiers of noun phrases, so that an expression such as a fan
of Juventus would have been analysed as a noun phrase (though in more recent
work, it would be analysed as an expression headed by the determiner a, and
hence as a determiner phrase, DP). See chs. 3 and 4.
NP See noun phrase.
Null See covert.
Null case The case carried by PRO (see case).
Null operator See Op.
Null subject A subject which has grammatical/semantic properties but no overt phonetic
form. More specifically, this term usually denotes the null pro subject found in
finite declarative or interrogative clauses in languages like Italian or Early
Modem English, and not the covert subject found in imperative clauses like Shut
the door! or the covert PRO subject found in control structures like The prison¬
ers tried to PRO escape from jail. Accordingly, a null subject language is a lan¬
guage which allows finite declarative or interrogative clauses to have a null pro
subject. For example, Italian is a null subject language and so allows us to say Sei
518
Glossary and abbreviations
simpatica (literally ‘Are nice’, meaning ‘You are nice’); by contrast, English is
not a null subject language, and so doesn’t allow the subject to be omitted in this
type of structure (hence *Are nice is ungrammatical in English). The null subject
parameter is a dimension of variation between languages according to whether
finite (declarative and interrogative) verbs allow null pro subjects or not.
Number A term used to denote the contrast between singular and plural forms. In
English, we find number contrasts in nouns (cf. one dog, two dogs), in some
determiners (cf. this book, these books), in pronouns (cf. he/they) and in finite
verbs (cf. He smells. They smell).
Obj An informal abbreviation for the case-feature [objective]. See case.
Object The complement of a transitive item (e.g. in Help me!, me is the object of the
transitive verb help', and in for me, me is the object of the transitive preposition
for). The term object is generally restricted to complements which carry objec¬
tive case - i.e. to nominal or pronominal complements: hence, nothing would be
the object (and complement) of said in He said nothing, but the that-clause
would be the complement (but not the object) of said in He said [that he was
tired] - though some traditional grammars extend the term object to cover
clausal complements as well as (pro)nominal complements. In sentences such as
She gave him them, the verb give is traditionally said to have two objects,
namely him and them: the first object him (representing the recipient) is termed
the indirect object, and the second object them (representing the gift) is termed
the direct object', the relevant construction is known as the double-object con¬
struction. Where a verb has a single object (e.g. nothing in He said nothing), this
is the direct object of the relevant verb.
Object-control predicate See control.
Objective See case.
Oblique See case.
O movement See operator movement.
One-place predicate A predicate which has only one argument (e.g. yawn in John yawned,
where John is the sole argument of the predicate yawn).
Op A symbol used to denote the empty question operator found in yes-no questions
like Do you enjoy syntax? It can be thought of as a counterpart of the question-
mark used in the spelling system to indicate that a sentence is a question, or as
the counterpart of whether in I wonder whether you enjoy syntax. See §7.8.
Operator This term is used in syntax to denote (for example) interrogative and negative
expressions which have the syntactic properties that they trigger auxiliary inver¬
sion (cf. What have you done?, Nothing would I ever do to upset anyone).
Operator movement Movement of an operator expression into spec-CP (i.e. into the speci¬
fier position within CP). See ch. 7.
Orphaned See stranded.
Overt An expression is overt if it has phonetic content, but covert if it lacks phonetic
content. Thus, him is an overt pronoun, but PRO is a covert (or null. or empty or
silent) pronoun.
519
Glossary and abbreviations
520
Glossary and abbreviations
Performance A term which denotes observed language behaviour, e.g. the kind of things
people actually say when they speak a language, and what meanings they assign
to sentences produced by themselves or other people. Performance can be
impaired by factors such as tiredness, drunkenness, etc. Performance is con¬
trasted with competence (which denotes the fluent native speakers’ knowledge
of the grammar of their native language). See §1.2.
PERFP A phrase headed by a perfective auxiliary.
Person In traditional grammar, English is said to have three grammatical persons', a first
person expression (e.g. Ilwe) is one whose reference includes the speaker(s); a
second person expression (e.g. you) is one which excludes the speaker(s) but
includes the addressee(s) (i.e. the person or people being spoken to); a third per¬
son expression (e.g. he!she lit/they) is one whose reference excludes both the
speaker(s) and the addressee(s) - i.e. an expression which refers to someone or
something other than the speaker(s) or addressee(s).
Personal pronouns These are pronouns which carry inherent person properties - i.e. first
person pronouns such as I/we, second person pronouns such as you and third
person pronouns such as he/she/it/they.
PF (representation) (A representation of the) phonetic form (of an expression). See repre¬
sentation. The PF component of a grammar is the component which converts
the syntactic structures produced by merger and movement operations into PF
representations. A PF clitic is a clitic which attaches to another item in the PF
component (not in the syntax), so that the two form a single phonetic word, but
not a single syntactic constituent. See §6.9, and ch. 6, ex. V.
PFI The principle of full interpretation (see full interpretation).
PF movement An operation which moves the phonetic features of a constituent but not its
other (e.g. grammatical) features. See §6.9.
Phonetic representation See representation.
Phonological features Features used to describe sound properties. For example, the differ¬
ence between nasal and oral sounds might be described in terms of the feature
[nasal].
Phrase The term phrase is used to denote an expression larger than a word which is a
maximal projection', see projection. In traditional grammar, the term refers
strictly to nonclausal expressions. (Hence, reading a book is a phrase, but He is
reading a book is a clause, not a phrase.) However, in more recent work, clauses
are analysed as types of phrases: e.g. He will resign is an auxiliary phrase (IP),
and That he will resign is a complementizer phrase (CP).
Phrase-marker A tree diagram used to represent the syntactic structure of a phrase or sen¬
tence. See §3.5.
Phrase structure See constituent structure.
Pied-piping A process by which a moved constituent (or set of features) drags one or more
other constituents (or sets of features) along with it when it moves. For example,
if we compare a sentence like Who were you talking to? with To whom were you
talking?, we might say that in both cases the pronoun who(m) is moved to the
521
Glossary and abbreviations
front of the sentence, but that in the second sentence the preposition to is pied-
piped along with whom. See §7.4.
Plural A plural expression is one which denotes more than one entity (e.g. these cars is
a plural expression, whereas this car is a singular expression).
P-marker See phrase-marker.
Polarity expression A word or phrase (e.g. a word like ever or a phrase like at all or care a
damn) which has an inherent affective polarity, and hence is restricted to occur¬
ring within the scope of an affective (e.g. negative, interrogative or conditional)
constituent.
Polycategorial word A word which belongs to more than one category.
Positive evidence In discussions of child language acquisition, this expression denotes evi-
dence based on the occurrence of certain types of structure in the child’s experi¬
ence. See §1.9.
Possessive A possessive structure is one which indicates possession: the term is most com¬
monly used in relation to structures like John’s book (where ’s is said to be a
possessive morpheme) or his book (where his is said to be a possessive pronoun
or possessive determiner). The expression ‘possessive have’ relates to a particu¬
lar use of the verb have to express possession (e.g. John has a house on the
Costa Brava).
Postmodify To say that nice in an expression such as someone nice postmodifies someone is
to say that nice modifies and follows someone.
Postposition A type of adposition which is the counterpart of a preposition in languages
which position prepositions after their complements.
Postulate A postulate is a theoretical assumption or hypothesis; to postulate is to assume
or hypothesize.
PP See prepositional phrase.
PPT See principles-and-parameters theory.
Pragmatics The study of the role played by nonlinguistic knowledge in our use of language.
Precede(nce) To say that one constituent precedes another is to say that it is positioned to its
left (on the printed page) and that neither constituent dominates the other.
Preclausal A preclausal expression is one which is positioned in front of a clause.
Predicate On predicate, see argument. A predicate nominal is a nominal expression used as
a predicate - e.g. the bold-printed expressions in John is a fool, I consider them
fools (where the expressions a fool/fools are said to be predicated of John/them).
Predication The process by which a predicate is combined with a subject in order to form a
proposition (see argument). For example, in a sentence such as Boris likes
vodka, the property of liking vodka is said to be predicated of Boris.
Predication principle A principle suggested by Rothstein (1995) to the effect that a syntac¬
tic predicate (e.g. an I-bar or V-bar constituent) requires a subject.
Predicative An adjective which is used as a predicate is said to be predicative in the rele¬
vant use - e.g. the bold-printed adjectives in John is drunk, I consider your
behaviour unforgiveable. Likewise, the nominal fools is said to be used predica-
tively in sentences such as They are fools, I consider them fools.
522
Glossary and abbreviations
523
Glossary and abbreviations
524
Glossary and abbreviations
other than John.) Pronouns differ from nouns in that they have no intrinsic
descriptive content, and so are functors (see §2.4). In much recent work, most
types of pronoun are analysed as determiners: for example, this is said to be a
prenominal determiner in a sentence such as / don’t like this idea (since it modi¬
fies the following noun idea), but a pronominal determiner (i.e. a determiner
used without any following noun expression) in a sentence such as I don’t like
this (see §4.8).
Proper noun See common noun.
Proposition A term used to describe the semantic content (i.e. meaning) of a sentence. For
example, we might say that the sentence Does John smoke? questions the truth
of the proposition that ‘John smokes’.
Pseudo-cleft Pseudo-cleft sentences are sentences such as What he bought was a car. What I said
was that I was tired. What I'm going to do is sell my car, where the constituent fol¬
lowing the verb be (viz. a car, that I was tired, sell my car) is said to have under¬
gone pseudo-clefting. The different term cleft sentence is traditionally used for sen¬
tences like It was a car that he bought. It was on Friday that I last saw her (where
the constituents a car and on Friday are said to have undergone clefting).
Psycholinguistics The study of the psychological processes by which we produce and under¬
stand speech.
Q In one use, an abbreviation for quantifier; in another use, an abbreviation for
question affix. (On the possibility that COMP in questions may contain an
abstract question affix Q, see §6.9.)
Quantifier A quantifier is a special type of determiner used to denote quantity. Typical
quantifiers include the universal quantifiers all/both, the distributive quantifiers
each!every, the partitive quantifiers sometany, etc.
Quantifier floating See floating quantifier.
QP/Quantifier phrase A phrase whose head is a quantifier - e.g. an expression such as many
people, few of the students, etc.
Question This refers to a type of sentence which is used to ask whether something is true, or
to ask about the identity of some entity. See yes-no question and wh-question.
Question operator See Op.
Raising (predicate) The term raising is used in two senses. On the one hand, it is used in a
general sense to denote any movement operation which involves moving some
word or phrase from a lower to a higher position in a structure. On the other
hand, it can also be used with the more specific sense of a subject-to-subject
raising operation by which an expression is moved from one subject position to
another (e.g. from being the subject of VP to being the subject of IP). The term
raising predicate denotes a word like seem whose subject is raised out of subject
position in a complement clause to become subject of the seem clause. See ch. 8.
RECIPIENT The name of the thematic role borne by the entity which receives (or comes to
possess) something - e.g. Mary in John gave Mary a present. See §8.4.
Reciprocal Expressions like each other and one another (in sentences like They hate each
other/one another) are traditionally classified as reciprocal anaphors.
525
Glossary and abbreviations
Recursive A recursive operation is one which can be repeated any number of times. For
example, the process by which an adjective comes to modify a noun might be
said to be recursive in that we can position any number of adjectives in front of a
noun (e.g. a tall, dark, handsome stranger).
Reduced form A form of a word which has lost one or more of its segments (i.e. vowels/
consonants), and/or which contains a vowel which loses its defining characteris¬
tics and is realized as the neutral vowel schwa hi. For example, the auxiliary
have has the full (unreduced) form /haev/ when stressed, but has the various
reduced forms /hov/, hvl (and has the contracted form /v/). See §8.5.
Reference/Referential The reference of an expression is the entity (e.g. object, concept,
state of affairs) in the external world to which it refers. A referential expression
is one which refers to such an entity; conversely, a nonreferential expression is
one which does not refer to any such entity. For example, the second there in a
sentence such as There was nobody there is referential (it can be paraphrased as
‘in that place’), whereas the first there is nonreferential and so cannot have its
reference questioned by where? (cf. * Where was nobody there?).
Reflexive A reflexive is a +self/+selves form such as myself, himself, ourselves,
themselves, etc. See also anaphor.
Relative pronoun/relative clause In a sentence such as He's someone [who you can trust],
the bracketed clause is said to be a relative clause because it ‘relates to’ (i.e.
modifies, or restricts the reference of) the pronoun someone. The pronoun who
which introduces the clause is said to be a relative pronoun, since it ‘relates to’
the expression someone (in the sense that someone is the antecedent of who).
Relativized minimality principle A principle proposed by Rizzi (1990) which amounts to
the claim that a moved constituent moves to the nearest appropriate position
(where what is an appropriate position is relative to the type of constituent
being moved: e.g. a moved head will move to the next-highest head position, an
argument will move to the next-highest A position, an operator will move to the
next-highest operator position, etc.).
Representation A syntactic representation (or structural representation) is a notation/
device used to represent the syntactic structure of an expression (typically, a tree
diagram or labelled bracketing): an LF representation is a representation of the
logical form of an expression; a PF representation is a representation of the
phonetic form of an expression.
Restrictive A restrictive theory is one which imposes strong constraints on the types of
structures and operations found in natural language grammars. See §1.3.
Resultative A verb such as paint in a sentence such as John painted his house pink is said to
be a resultative verb in that the result of the action of painting is that the house
becomes pink.
Rich To say that a language has a rich system of verb inflections (of a given type) is
to say that it has a large number of inflectional affixes (of the relevant type)
which attach to verbs; to say that a language has an impoverished system of verb
inflections is to say that it has only a small number of verb inflections.
526
Glossary and abbreviations
Root The root of a tree diagram is the topmost node in the tree. Hence, a root clause
is a free-standing clause, i.e. a clause which is not contained within any other
expression. In traditional grammar, a root clause is termed a ‘principal clause’,
‘independent clause’ or 'main clause’. By contrast, an embedded clause is a
clause which is contained within some larger expression and a complement
clause is an (embedded) clause which is used as the complement of some item.
So, in a sentence such as I think he loves you, the think-clause (i.e. the expres¬
sion I think he loves you) is a root clause, whereas the /oves-clause (i.e. the
expression he loves you) is an embedded clause. Moreover, the Zoves-clause is
also a complement clause, since it serves as the complement of the verb think.
S In one use, this symbol represents the category label sentence', in another use, it
represents the number-feature [singular].
+s In one use, this denotes the plural suffix found in plural nouns such as dog+s; in
another use, it denotes the third person singular present tense suffix found in
verbs such as adore+s.
Schwa The neutral vowel [a] - e.g. the vowel corresponding to the bold-printed letters
in words like about, affair, potato, etc.
Scope The scope of an expression is the range of constituents which it ‘modifies’ or
which fall within (what we might call informally) its ‘sphere of influence’.
Scrambling A process which reorders maximal projections internally within clauses, moving
them further to the front of the clause. For example, in ar Bany Modem English
sentence like The king your mote did see (Boyet, Love ■ Labour's Lost, IV.i), we
might say that your mote has been scrambled out of its normal postverbal posi¬
tion (after see) into a position in front of did (see § 10.3).
SE Standard English
Second person See person.
Segment In phonology, to say that a word like man comprises the three segments /m/, Ixl
and Ini is to say that these are the three sound units which determine the pho¬
netic form of the word. In syntax, to say that a category comprises two segments
(and so is a split-segment category) is to say that it comprises a constituent to
which some other constituent has been adjoined, so forming an even larger con¬
stituent of the same type (e.g. if we adjoin n’t to an auxiliary like could we form
the even larger auxiliary couldn’t, and this contains two auxiliary segments since
both couldn't and could are auxiliaries).
Select(ion) When a word can have a particular type of complement, we say that it selects
(i.e. ‘takes’) the relevant type of complement (and the relevant property is
referred to as complement-selection). So, for example, we can say that it is a
complement selection property of the verb want that it selects (i.e. ‘can take’) a
fo-infinitive complement (as in I want to help you).
Semantics The study of linguistic aspects of meaning.
Sentence This term denotes a free-standing clause which is not contained within some larger
expression. In terms of the conventions of the English spelling system, a sentence
might be defined (rather inaccurately) as a string of words which starts with a word
527
Glossary and abbreviations
beginning with a capital letter and which ends with a word immediately followed
by a full-stop, so that this entry for sentence contains two sentences.
Sentence fragment See fragment.
Shared string coordination This refers to coordinate structures such as John is buying and
Mary is selling a house, where the expression a house at the end of the sentence
functions both as the complement of buying and as the complement of selling,
and so is in some sense ‘shared’ between the two verbs (so that a house can be
said to function as the shared string in the coordinate structure).
Shell This term is used in connection with the idea (discussed in chapter 9) that verb
phrases comprise two different projections, an outer vp shell headed by a light
verb, and an inner VP core headed by a lexical verb.
Shortest movement principle A principle of grammar requiring that a constituent should
move the shortest distance possible in any single movement operation.
Silent See covert.
Simple coordination This refers to simple coordinate structures such as John and Mary in
sentences like John and Mary are expecting you.
Singular A singular expression is one which denotes a single entity (e.g. this car is a sin¬
gular expression, whereas these cars is a plural expression).
Sister Two nodes are sisters if they have the same mother. See §3.6.
Small clause See clause.
Spec See specifier.
Spec-CP/Spec-IP/Spec-VP (etc.) The specifier position within CP/IP/VP.
Spec-head A spec-head relation is a relation between a head and its specifier. For exam¬
ple, we might say that subject-auxiliary agreement involves a spec-head rela¬
tion in sentences like He has gone, since has is the head of the clause and agrees
with its specifier he.
Specification See feature.
Specifier The grammatical function fulfilled by certain types of constituent which (in
English) precede the head of their containing phrase. For example, in a sentence
such as John is working, John is the specifier (and subject) of is working (see
§3.3). In a sentence such as What did John do?, what is the specifier of the CP
headed by the inverted auxiliary did. See ch. 7.
Specifier-features Features which determine the kind of specifier which a given type of head
can have. For example, the specifier-features of the auxiliary has are [3SNom],
and these tell us that it requires a third person singular nominative subject like
he/she/it. See §5.3.
Specifier-firstAlast A specifier-first language is one which normally positions specifiers
before their heads; a specifier-last language is one which normally positions
specifiers after their heads.
Spellout The point in a derivation at which phonetic and semantic features are processed
by separate components of the grammar (the PF component and the LF com¬
ponent respectively). See §5.2.
Split infinitive A structure in which the infinitive particle to is separated from the verb with
528
Glossary and abbreviations
which it is associated: a sentence such as It’s important to really try hard con¬
tains an example of a split infinitive, since the particle to has been separated
from the verb try by the intervening adverb really.
Split-INFL hypothesis The hypothesis that there is not just one auxiliary position in clauses
(= INFL), but rather two (T and AgrS). See §10.2 and §10.3.
Split-segment category See segment.
Split-VP hypothesis The hypothesis that verb phrases have a complex internal structure
comprising an outer vp shell headed by a light verb, AgrOP/AgrIOP projec¬
tions, and an inner VP core headed by a lexical verb. See § 10.4.
Stack(ing) To say (for example) that prenominal adjectives can be stacked in front of a
noun is to say that we can have an indefinitely large number of adjectives posi-
tioned in front of a noun (e.g. a big. red, juicy, ripe apple).
Standard language The variety of language used by central government administration.
taught in schools and used e.g. by the presenters of (nonregional) radio/televi¬
sion news bulletins.
Star An asterisk (*) used in front of an expression to indicate that the expression is
ungrammatical.
Stem A morpheme which contains no inflectional or derivational affixes. For exam¬
ple, the stem form of the verb going is go.
Stranded A stranded (or orphaned) preposition is one which has been separated from its
complement (by movement of the complement). For example, in a sentence
such as Who were you talking to?, to is a stranded/orphaned preposition by
virtue of the fact that it is separated from its complement who (which has been
moved to the front of the sentence). By extension, in a sentence such as They
have all left, the quantifier all could be said to have been stranded, since it is
separated from the pronoun they which it quantifies.
String A continuous sequence of words contained within the same phrase or sentence.
For example, in the phrase a couple of drinks, the sequences a couple, couple of,
of drinks, a couple of and couple of drinks are all strings, whereas the sequences
a of, a drinks, a couple drinks, a of drinks and couple drinks are not. Note that a
string need not be a constituent.
Strong feature A strong feature is one which can trigger movement; a weak feature is one
which cannot trigger movement. For example, finite verbs carry strong agree¬
ment-features in Early Modem English, and so raise to INFL; but finite verbs
carry weak agreement-features in Modem Standard English, and so cannot
move to INFL but rather remain in situ. See ch. 5.
Structural learning This term denotes what children leam about grammatical structure in
the language they are acquiring. See § 1.8.
Structural representation See representation.
Structural uniformity hypothesis The hypothesis that all clauses have a uniform CP/I PA'P
structure. See §4.6.
Structure See constituent structure.
Structure dependence principle A principle which states that grammatical operations are
529
Glossary and abbreviations
530
Glossary and abbreviations
531
Glossary and abbreviations
532
Glossary and abbreviations
533
Glossary and abbreviations
position in the relevant phrase/sentence (e.g. the objective subject him is unli¬
censed in a sentence such as *Him is kind, since it occupies a position in which
only a nominative pronoun like he is licensed (i.e. permitted)).
Unreduced See reduced form.
Unspecified Containing no specification of its value with respect to a given feature. For
example the determiner the might be claimed to be unspecified with respect to
number (i.e. whether it is singular or plural) - though in §5.6 we suggested that
the carries a variable number-feature (and hence can be either singular or
plural).
UTAH See uniform theta assignment hypothesis.
V/V/V-bar/VP V is a lexical verb; V/V-bar is an intermediate projection headed by a lexical
verb; VP is a maximal projection headed by a lexical verb. See lexical and pro¬
jection.
v/v/v-bar/vp v is a light verb; v/v-bar is an intermediate projection headed by a light verb;
vp is a maximal projection headed by a light verb. See light verb and projec-
tion.
Variable The category variable X is used to denote ‘any category of head which you care
to choose’ (and similarly X denotes ‘any type of intermediate projection’, and
XP denotes ‘any type of phrase’). The feature-variable [a] is used to represent
‘any value for the relevant feature which you care to choose’ (so that e.g.
[aNumber] in effect means ‘singular or plural in number’, and [aPerson] means
‘first, second or third person’).
Variation A term used to describe differences between languages or between different
varieties of the same language.
Variety A particular (e.g. geographical, social or stylistic) form of a language.
Verb A category of word which has the morphological property that it can carry a
range of inflections including past tense +d, third person singular present tense
-hs, perfective +n and progressive +ing (cf. show/shows/showed!shown!show¬
ing), and the syntactic property that it can head the complement of infinitival to
(cf. Do you want to show me?). See §2.2.
Verb movement See V movement.
Verb phrase A phrase/maximal projection which is headed by a verb - e.g. the bracketed
phrase in They will [help you] (see ch. 3). In terms of the shell analysis pre¬
sented in ch. 9, there are two different types of verb phrase - a VP headed by a
lexical verb, and a vp headed by a light verb.
V(-to-I) movement An operation by which a finite verb moves from V to INFL (e.g. in Early
Modem English). See §6.4.
Vocalic Containing a vowel or diphthong.
Vocative A vocative expression is one which is used to address one or more individuals,
and which is set off in a separate tone-group at the beginning or end of the
sentence (separated from the rest of the sentence by a comma). So, for example,
Fred is a vocative expression in Fred, can you give me a hand? and similarly
you two is a vocative expression in Come here, you two!
534
Glossary and abbreviations
535
REFERENCES
Abney, S.P. (1987) The English noun phrase in its sentential aspect’, PhD diss., MIT.
Akmajian, A. and F. Heny (1975) An Introduction to the Principles of Transformational
Syntax, MIT Press, Cambridge, Mass.
Aronoff, M. (1976) Word Formation in Generative Grammar, MIT Press, Cambridge, Mass.
Authier, J.-M. (1991) ‘V-govemed expletives, case theory and the projection principle’,
Linguistic Inquiry 22: 721 —4-0.
Bailey, B.L. (1966) Jamaican Creole Syntax: a Transformational Approach, Cambridge
University Press.
Baker, C.L. (1970) ‘Notes on the description of English questions: the role of an abstract ques¬
tion morpheme’. Foundations of Language, 6: 197-219.
Baker, M. (1988) Incorporation, University of Chicago Press.
Baker, M., K. Johnson and I. Roberts (1989) ‘Passive arguments raised’. Linguistic Inquiry 20:
219-51.
Baltin, M. (1995) ‘Floating quantifiers, PRO and predication’, Linguistic Inquiry 26:
199-248.
Belletti, A. (1988) ‘The case of unaccusatives’, Linguistic Inquiry 19: 1-34.
(1990) Generalized Verb Movement, Rosenberg & Sellier, Turin.
Berko, J. (1958) ‘The child’s learning of English morphology’, Word 14: 150-77.
Berman, R. (1988) ‘Word class distinctions in developing grammars’, in Y. Levy, I. M.
Schlesinger and M. D. S. Braine (eds.) Categories and Processes in Language
Acquisition, Erlbaum, London, pp. 45-72.
Bishop, D.V.M. (1994) ‘Grammatical errors in specific language impairment: competence or
performance limitations?’, Applied Psycholinguistics 15: 507-50.
Bobaljik, J.D. (1995) ‘Morphosyntax: the syntax of verbal inflection’, PhD diss., MIT.
Bolinger, D and D.A. Sears (1981, 3rd edn) Aspects of Language, Harcourt Brace Jovanovich,
New York.
Boskovic, Z. (1994) ‘Categorial status of null operator relatives and finite declarative comple¬
ments’, Language Research 30: 387-412.
(1995) ‘Principles of economy in nonfinite complementation’, PhD diss.. University of
Connecticut.
Bowerman, M. (1995) ‘Don’t giggle me!’, talk presented at the University of Essex, 30th
November 1995.
536
References
537
References
Contreras, J. (1987) ‘Small clauses in Spanish and English’, Natural Language and Linguistic
Theory, 5: 225-44.
Culicover, P.W. (1992) ‘Evidence against ECP accounts of the that-t effect’. Linguistic Inquiry
23:557-61.
Davis, H. (1987). ‘The acquisition of the English auxiliary system and its relation to linguistic
theory’, PhD dissertation. University of British Columbia.
Emonds, J.E. (1976) A Transformational Approach to English Syntax, Academic Press, New
York.
Ernst, T. (1991) ‘On the scope principle’. Linguistic Inquiry 22: 750-6.
Fabb, N. (1988) ‘English suffixation is constrained only by selectional restrictions’. Natural
Language and Linguistic Theory 6: 527-39.
Fasold, R. (1980) ‘The relation between black and white speech in the south’, mimeo. School
of Languages and Linguistics, Georgetown University, Washington, D.C.
Felser, C. (1995) ‘The syntax of verbal complements: a study of perception verbs in English’,
PhD diss., University of Gottingen.
Fillmore, C.J. (1968) ‘The case for case’, in E. Bach & R.T. Harms (eds.) Universals in
Linguistic Theory, Holt Rinehart & Winston, New York, pp. 1-88.
(1972) ‘Subjects, speakers and roles’, in D. Davidson & G. Harman (eds.) Semantics of
Natural Language, Reidel, Dordrecht.
Gopnik, M. & M. Crago, (1991) ‘Familial aggregation of a developmental language disorder’.
Cognition 39: 1-50.
Grice, H.P. (1975) ‘Logic and conversation’ in P. Cole & J. Morgan (eds.) Syntax and
Semantics 3: Speech Acts, Academic Press, New York, pp. 41-58.
Grimshaw, J. (1991) ‘Extended projection’, draft manuscript. Brandeis University, July 1991.
(1993a) ‘Minimal projection, heads, and optimality’, draft manuscript, Rutgers University,
June 1993.
(1993b) ‘Minimal projection, heads, and optimality’, conference handout, Rutgers
University, November 1993 [revised version of Grimshaw 1993a],
Groat, E. (1995) ‘English expletives: a minimalist approach’. Linguistic Inquiry 26: 354—65.
Gruber, J.S. (1965) ‘Studies in lexical relations’, PhD diss., MIT.
Gueron, J. (1990) ‘Particles, prepositions and verbs’, in J. Mascara and M. Nespor (eds.)
Grammar in Progress, Foris, Dordrecht.
Guilfoyle, E. (1983) ‘Habitual aspect in Hiberno-English’, McGill Working Papers in
Linguistics 1: 22-32.
Guilfoyle, E., H. Hung and L. Travis (1992) ‘Spec of IP and spec of VP: two subjects in
Austronesian languages’, Natural Language and Linguistic Theory 10: 375—414.
Haegeman, L. (1992, 1st edn) Introduction to Government and Binding Theory, Blackwell,
Oxford.
(1994, 2nd edn) Introduction to Government and Binding Theory, Blackwell, Oxford.
Hale, K. & S.J. Keyser (1991) ‘On the syntax of argument structure’, Lexicon Project Working
Papers, MIT, Center for Cognitive Science, Cambridge, Mass.
(1993) ‘On argument structure and the lexical expression of semantic relations’, in K. Hale &
S. J. Keyser (eds.) The View from Building 20, MIT Press, Cambridge, Mass., pp. 53-109.
538
References
539
References
540
References
Rickford, J.R. (1986) ‘Social contact and linguistic diffusion: Hiberno-English and New World
Black English’, Language 62: 245-89.
Ritter, E. and S.T. Rosen (1993) ‘Deriving causation’. Natural Language and Linguistic
Theory 11: 519-55.
Rizzi, L. (1990) Relativized Minimality, MIT Press, Cambridge, Mass.
Roberts, I. (1986) The Representation of Implicit and Dethematized Subjects, Foris, Dordrecht.
(1991) ‘Excorporation and minimality’, Linguistic Inquiry 22: 209-18.
(1993) Verbs and Diachronic Syntax, Kluwer, Dordrecht.
(1996) Comparative Syntax, Arnold, London.
Rohrbacher, B. (1994) The Germanic VO Languages and the Full Paradigm: a Theory of V to I
Raising, GLSA Publications, Amherst.
Rooryck, J. (1994) ‘On 0- and a-underspecification in syntax and phonology’, MIT Working
Papers in Linguistics 22: 197-216.
Rosen, S.T. (1990) Argument Structure and Complex Predicates, Garland, New York.
Rosenbaum, PS. (1967) The Grammar of English Predicate Complement Constructions, MIT
Press, Cambridge Mass.
Ross, J.R. (1967) ‘Constraints on variables in syntax’, PhD diss., MIT (published as Infinite
Syntax! by Ablex Publishing Corporation, Norwood, N.J., 1986).
Rothstein, S.D. (1995) ‘Pleonastics and the interpretation of pronouns’, Linguistic Inquiry 26:
499-529.
Safir, K. (1993) ‘Perception, selection and structural economy’. Natural Language Semantics
2: 47-70.
Salles, H. M. P. (1995) ‘Preposition pied-piping and preposition stranding; a minimalist
approach’, Bangor Research Papers in Linguistics 6: 97-123.
Schaeffer, J. (1994) ‘The split INFL hypothesis: evidence from language-impaired children’,
talk presented to the Child Language Seminar, Bangor.
Sobin, N. (1985) ‘Case assignment in Ukrainian morphological passive constructions’,
Linguistic Inquiry 16: 649-62.
Speas, M. (1986) ‘Adjunction and projections in syntax’, PhD diss., MIT.
(1995) ‘Economy, agreement and the representation of null arguments’, mimeo, University
of Massachusetts.
Spencer, A.J. (1991) Morphological Theory, Blackwell, Oxford.
Sportiche, D. (1988) ‘A theory of floating quantifiers and its corollaries for constituent struc¬
ture’, Linguistic Inquiry 19: 425-49.
Stowell, T. (1981) ‘Origins of phrase structure’, PhD diss., MIT.
(1982) ‘The tense of infinitives’. Linguistic Inquiry 13: 561-70.
Stroik, T. (1990) ‘Adverbs as V-sisters’, Linguistic Inquiry 21: 654-61.
Tieken-Boon van Ostade, I. (1988) ‘The origins and development of periphrastic auxiliary do:
a case of destigmatisation’, Dutch Working Papers in English Language and Linguistics
3: 1-30.
Travis, L. (1984) ‘Parameters and effects of word order variation’, PhD diss., MIT.
Trudgill, P. & J.K. Chambers (1991) (eds.) Dialects of English: Studies in Grammatical
Variation, Longman, London.
541
References
Ura, H. (1993) ‘On feature-checking for wh-traces’, MIT Working Papers in Linguistics 18:
243-80.
Uriagereka, J. (1995) ‘Aspects of the syntax of clitic placement in Western Romance’,
Linguistic Inquiry 26: 79-123.
Vainikka, A. (1994) ‘Case in the development of English syntax’, Language Acquisition 3:
257-325.
Vikner, S. (1995) Verb Movement and Expletive Subjects in Germanic Languages, Oxford
University Press.
Wakelin, M.F. (1977, 2nd edn) English Dialects: an Introduction, Athlone, London.
Watanabe, A. (1993) ‘Agr-based theory and its interaction with the A-bar system’, PhD diss.,
MIT.
Williams, E. (1994) ‘A reinterpretation of evidence for verb movement in French’, in D.
Lightfoot & N. Homstein (eds.) Verb Movement, Cambridge University Press, pp. 189-205.
Wolfram, W. (1971) ‘Black-white speech differences revisited’, in W. Wolfram & N. H. Clark
(eds.) Black-White Speech Relationships, Center for Applied Linguistics, Washington,
D.C., pp. 139-61.
Woolford, E. (1991) ‘VP-intemal subjects in VSO and nonconfigurational languages’,
Linguistic Inquiry 22: 503—40.
Zagona, K. (1987) Verb Phrase Syntax, Kluwer, Dordrecht.
542
INDEX
543
Index
544
Index
545
Index
546
Index
547
Index
548
Index
549
Index
550
Index
551
Index
552
Index
553
Index
554
Index
555
Index
556
Index
557
Index
558