Professional Documents
Culture Documents
Small Clauses in English - The Nonverbal Types
Small Clauses in English - The Nonverbal Types
Editors
Jan Svartvik
Herman Wekker
Mouton de Gruyter
Berlin · New York
Small Clauses in English
The Nonverbal Types
Bas Aarts
Mouton de Gruyter
Berlin · New York 1992
Mouton de Gruyter (formerly Mouton, The Hague)
is a Division of Walter de Gruyter & Co., Berlin.
Aarts, Bas:
Small clauses in English : the nonverbal types / Bas Aarts. —
Berlin ; New York : Mouton de Gruyter, 1992
(Topics in English linguistics ; 8)
ISBN 3-11-013487-X
NE: GT
Preface vii
Chapter 1 Introduction 1
1.2. Methodology 2
2.1. Introduction 9
2.2. The Standard Theory and the Extended Standard Theory 9
2.3. Government-Binding Theory 12
2.3.1. Outline 12
2.3.2. Small Clause Theory 21
2.3.3. Predication Theory 25
2.3.4. Recent developments in GB-theory: the Barriers model 30
2.4. Descriptive treatments 35
Chapter 3 Small Clauses as constituents 37
3.1. Introduction 37
3.2. Evidence for the existence of Small Clauses 37
3.3. Small Clauses as instances of dependent and controlled predication 48
3.4. Clauses with and without to be 68
3.5. Conclusion 71
4.1. Introduction 73
4.2. Verb-preposition constructions and Small Clauses 74
4.2.1. Introduction 74
4.2.2. Background 75
4.2.3. The data 77
4.2.4. The analysis 80
4.2.5. Conclusion 89
4.3. Verbs of negative causation 90
4.3.1. Introduction 90
4.3.2. The status of the postverbal NP 91
4.3.3. The element from 95
4.3.4. With and without from 102
4.3.5. Passive pre\eat-constructions 108
χ Contents
Appendix 195
Notes 197
References 209
Index 221
Chapter 1
Introduction
The aim of this monograph will be to investigate in detail the properties of the
constructions shown in (1):
(1) V NP XP
where X = N, A or Ρ
I will be interested in those structures where the phrase following the postverbal NP
(which I will call the intervening NP, following Huddleston 1984) is in a subject-
predicate relation with this same NP. The constructions in which the XP is a VP
are excluded from the discussion.
There are two basic dependency relations. In the first there is a relation between
the main verb and the intervening NP, as well as between the intervening NP and
the following XP which is predicated of iL This can be represented as follows:
Verb NP Predicative XP
t it t
To give an example: in the sentence they appointed her head of the department, her
is the object of the verb appoint, as well as the subject of the predicative relation
holding between this element and the following NP head of the department.
In the second basic dependency relation there is again a subject-predicate relation
between the intervening NP and the following XP, but there is also a verb-object
relation between, on the one hand, the main verb and, on the other hand, the
intervening NP + XP taken together:
Verb NP Predicative XP
An example of this pattern is the sentence I wanted her happy. Here her is the
subject of happy, rather than the object of the main verb. In recent generative
literature the NP and XP taken together have been given the syntactic label Small
Clause (SC). One of the major concerns of this book is to investigate whether
Small Clauses exist as syntactic entities.
2 Introduction
1.2. Methodology
[w]hatever the future may hold, there is no question that generativist theory has
already demonstrated that it has implications for, and applications to, a number
of important language-related problems, from second language learning to
natural language processing to the inequal social status of speakers of
nonstandard dialects. (Newmeyer 1983: 157)
The opposition between these two approaches in linguistics has been long and
acrimonious; neither side concedes the other's reality. It has also affected the
other disciplines related to linguistics. The study of language acquisition is
divided between those who look at interaction and communicative function and
those who look for rules and principles; language teachers can be divided into
those who advocate Ε-language methods that stress communication and
behaviour and I-language methods that stress language knowledge, though the
former are more in fashion at present; computational linguists roughly divide
into those who analyse large stretches of text and those who write rules. An E-
linguist collects samples of actual speech or actual behaviour; evidence is
concrete physical manifestation. An I-linguist invents possible and impossible
sentences; evidence is whether speakers know if they are grammatical. The E-
linguist despises the I-Iinguist for not looking at "real" facts; the I-linguist
derides the Ε-linguist for looking at trivia. The I-language versus E-language
distinction is as much a difference of research methods and of admissible
evidence as it is of long-term goals. (Cook 1988: 13)
In essence this is a reasonable rendering of the current state of affairs regarding the
relation between traditional grammar and generative linguistics. Transformational
Grammar and Modem Descriptive Grammar should also be viewed as
complementary frameworks. However, we might tighten the bond between the two
4 Introduction
In the a-sentence, which corresponds to the first meaning above, the trace occurs
at the end of the sentence. In the b-sentence, which corresponds to the second
meaning, it occurs between want and to. In this position it blocks contraction. This
Methodology 5
explains why who do you wanna succeed can only have the first meaning given
above. The account has been simplified somewhat, but the point I want to make
here is that these sentences provide evidence for the fact that there are such entities
as traces whose presence has overt syntactic and phonological effects. A descriptive
linguist might consider to incorporate the notion trace in his description of a
particular language.
Despite the fact that some ideas from theoretical linguistics have been adopted
in descriptive grammars, very few of the insights of the Extended Standard Theory
(EST) and of GB-theory have been considered. This means that a wealth of ideas
developed since 1970 (the year of publication of Chomsky's paper "Remarks on
nominalisation") has not been taken into account. In view of what I have said
above, I do not believe that descriptive linguists are disinclined in principle to take
an interest in the recent theory, but for some reason they do not always seem to be
prepared to adopt some of the current analyses. Take, for example, Rodney
Huddleston's analysis of the NP a Ministry of Defence Official (1984: 111—112).
The analysis of the substring Ministry of Defence is a problem for him because it
is clearly a constituent, but not an NP: it does not occur in typical NP positions, nor
can it take determiners (cf. *a the Ministry of Defence Official)· It is observed that
the overall NP can be analysed either as in (3a) or as in (3b):
(3) a. NP
b.
NP
Stuurman (1990).
Let us now return to the constructions under (1). An analysis of these
constructions will be set up on the basis of the following guiding principles:
Ideally, the description is maximally constrained. There are two reasons for
this. The first is that on grounds of elegance a theory which covers/explains
the facts making only a limited number of statements is to be preferred to a
theory which makes an unlimited number of statements. This is the principle
of Occam's razor. Secondly, only if we have a constrained theory can we
explain the logical problem of language acquisition. I will have more to say
about language acquisition in chapter two.
The second principle is strongly linked to the first one in that a matching of syntax
and semantics is a means of achieving a constrained description.3
In my research I have made some use of the computerised corpus which has been
compiled at the Survey of English Usage (SEU) at University College London.4
The SEU corpus is a carefully compiled collection of several textual dimensions in
both spoken and written form. It will be used with the following purposes in mind:
(a) Finding syntactic evidence for the analyses which are proposed.
(b) Considering as wide a range as possible of the construction types given in (1)
above.
(c) Investigating hew these structures are used by educated native speakers of
English.5
The remainder of this book is structured as follows. In the next chapter I will
give an historiographical account of the various approaches to the structures under
investigation in the Standard and Extended Standard Theories of generative
grammar (1957—1980), in Government-Binding (GB) Theory (1980— ) and in
descriptive work. In chapter three I will discuss what evidence there is for claiming
that Small Clauses actually exist. A distinction will be drawn there between two
types of SC-construction. The last three chapters will be devoted to special Small
Methodology 7
2.1. Introduction
(4) [V NP XP]
Section 2.2 discusses work produced within the Standard Theory and Extended
Standard Theory frameworks. Section 2.3 has a dual function: on the one hand it
is intended as a concise introduction to the concepts and terminology of current
Government-Binding Theory, while on the other hand it deals with the analyses of
the constructions in (4) within that theory. In subsections 2.3.1—2.3.3 I will be
describing the theory as expounded in Chomsky's 1981 book Lectures on
government and binding (LGB) and related work, turning my attention to recent
modifications in section 2.3.4. The reason why the earlier model is discussed is that
most of the generative literature I will be referring to was written within the LGB
framework. Finally, in section 2.41 take a brief look at analyses put forward within
some of the descriptive linguistic frameworks.
Within the frameworks of the Standard Theory (roughly from 1957—1970) and of
the Extended Standard Theory (1970—1980) respectively, very little attention was
given to the constructions shown under (4) in the previous section. There is some
discussion in Chomsky's earliest work. In Chomsky (1975a [1955]: 479ff, 1957:
76—77 and 1964: 238) it is suggested that in, for example, [V NP AP] strings the
verb and adjective in some sense form a complex unit. A sentence like (5a) would
then be represented as in (5b):
The idea is picked up in later work but not developed. See Chomsky (1981: 31 Iff,
10 Previous approaches
The earliest proposals regarding the analysis of sentences like this date back to
Rosenbaum (1967). He proposed a rule which raised the postverbal NP from the
underlying subject position in the lower clause at Deep Structure to the object
position in the higher clause in surface structure, roughly as in (7):
A-verbs are outside the scope of this study. I will deal with N-verbs in section 4.3.
From the point of view of this book the Β and W-verbs are of greatest interest as
many of these also occur in the [V NP XP] constructions we are discussing here.
The Standard Theory and the Extended Standard Theory 11
Postal does not deal with constructions of this form in depth, but he does refer to
them. In an interesting section on 7oug/i-Movement he discusses the sentences
below:
He observes that the d-sentences in (8) and (9) are a problem for Chomsky's 1973
framework. In "Conditions" it is claimed that only direct object NPs can be fronted
under Γοκ^Λ-Movement, as in e.g. (10):
One of the main implications of the 1973 paper is that Raising-to-Object does not
exist and that in a sentence like (8a) Jones is the subject of a subordinate clause
which complements the verb consider. According to Chomsky, (8c) and (9c) are
ungrammatical because in these sentences subjects, not objects, are fronted under
7bug/i-Movement. But what about (8d) and (9d)? As noted, these were a problem
for the "Conditions" approach. Notice that the semantic relation between the verbs
consider and prove to the NPs Jones and Melvin in (8a)/(9a) is the same as the
relation between these verbs and these NPs in (8b)/(9b). That is, the presence of to
be does not alter the thematic relations holding in (8a)/(9a) and (8b)/(9b). Following
the "Conditions" line of reasoning we would therefore have to say that in (8b) and
(9b) Jones and Melvin are subjects as in (8a)/(9a). But if they are, why can they be
fronted under 7o«^/i-Movement, as in (8d) and (9d)? One of Chomsky's aides-de-
camp in the Raising debate was David Lightfoot. In a review of Postal (1974) he
claimed that in (8b) and (9b) Jones and Melvin are direct objects (see Lightfoot
1976: 276—277), a position which, in view of his later writings, he would certainly
not want to defend today. Taking the view that Jones and Melvin are direct objects
in (8b) and (9b) means losing the obviously correct insight noted above that
regardless of the presence of to be the thematic relations between the verb and the
following NP are the same in (8a)/(9a) and (8b)/(9b). In sum, at the time Postal
pointed out a serious problem for the "Conditions" framework. The issue is,
however, not pertinent in GB-theory any more, as it is no longer assumed that the
NPs Jones and Melvin are fronted under 7oug/i-Movement in the example sentences
above. For more recent treatments of 7oug/z-Movement constructions, see Chomsky
12 Previous approaches
(1977, 1981: 308ff, 1982: 30f, 1986a: 108f). For a discourse-based approach, see
Mair (1990). The identical thematic relations between the various elements in the
a and b-sentences in (8) and (9) have led some linguists to suppose that there exists
a rule of To i>e-Deletion which would derive (8b)/(9b) from (8a)/(9a). See e.g.
Stockwell—Schachter—Partee (1973) and Borkin (1973). I discuss these proposals
in more detail in section 3.4.
Returning to Raising-to-Object, in "Conditions" Chomsky does not discuss this
rule. He only mentions it once or twice, mostly in notes. Fairly recently he has said
that he became aware of the transformation only at a very late stage. In
Huybregts—van Riemsdijk he remarks "in the early fifties, I was working
extensively on grammar, virtually nothing else... Plenty of people thought of other
things later. For example, I didn't even notice Raising. I do not think it was until
Rosenbaum, or anyway, years later, that it became obvious." (Huybregts—van
Riemsdijk 1982: 81)
It is not until the eighties that [V NP XP] constructions are dealt with in any
great depth. The analytical proposals made within what has become known as
Government-Binding Theory will be discussed in sections 2.3.2/2.3.3. But first there
follows a concise outline of that theory.
2.3.1. Outline
I The speech a child hears in its environment is defective in that there are
many ill-formed sentences, slips of the tongue etc.
II The data a child hears in its environment are finite; nevertheless it will
eventually be able to deal with an unrestricted number of sentences.
III Certain aspects of language are known for which no direct evidence is
available in the data, (adapted from Lightfoot 1982: 17)
Government-Binding Theory 13
I do not mean by this simply that the concepts expressed and distinctions
developed in normal language use give us insight into the patterns of thought
and the world of "common sense" constructed by the human mind. More
intriguing, to me at least, is the possibility that by studying language we may
discover abstract principles that govern its structure and use, principles that are
universal by biological necessity and not mere historical accident, that derive
from natural characteristics of the species. (Chomsky 1975b: 4)
Chomsky is thus making the very strong claim that the analyses arrived at are
somehow encoded in the brain. Other linguists have often been critical of this
assumption, even within generative grammar; see e.g. Gazdar et al. (1985: 5). It
should be realised that GB-theory can also be regarded purely as a theory of
language without making any assumptions about innateness. This approach has been
referred to as instrumentalism.
Directing our attention at UG, we can distinguish two sets of subsystems as
expounded in Chomsky (1981: 5): the so-called subcomponents of the rule system
of grammar, in (11), and the system of subtheories, in (12):
(11) Lexicon
-Base
- a. categorial component —
Syntax -
-b. transformational component
PF component
LF component
Viewed from a slightly different perspective we can also represent the GB-model
14 Previous approaches
of grammar as in (13). The principles of the subtheories under (12) feed into this
representation in ways to be explicated below.
Move (a)
S-Structure8
The components of (11), (12) and (13) are all part of GB-theory. The subtheories
in (12) are also referred to as the modules of GB-theory. The theoretical approach,
then, is a modular approach.
I will not be discussing the generative framework in extenso, but I will give a
brief explanation of what is shown in (11)—(13). Before doing so, let us first
examine in some detail a prototypical phrase marker (a representation of sentence
structure in the form of a tree diagram) in the 1981 version of the theory. The
Phrase Structure (PS) rules in (14) generate φ ε tree in (15):
(15) S'
COMP
NP INFL
[±tense]
[AGR] V
Government-Binding Theory 15
In (14) COMP is the complementiser position where both clause introducers like
that and Wh-phrases may be positioned. Thus, in a sentence like (16) the
complementiser that occupies the lower of the two COMP positions, as is shown
in (17); the higher COMP slot remains empty:
(17)
COMP S
NP INFL
[+tns][AGR]
[-past][+sing] V
In (18) the Wh-element what occurs in COMP, as can be seen in the phrase marker
in (19):
(19)
COMP S
NP INFL VP
[+tns][AGR]
[+past][+sing] V NP
what, you
relations between subjects and verbs in sentences through the component AGR. The
structure of INFL looks like this:
(20) INFL
[± tense] [AGR]
[ a past]9 I
number features
gender features
person features
We will see in the next section that the PS rules have been abandoned in favour of
category neutral rule templates. In section 2.3.4 we will turn to recent modifications
of the basic phrase marker configuration shown in (15).
Let us now turn to a discussion of what is shown in (11)—(13), starting with the
syntactic component in (11).
X'-theory is a theory of syntax. Its basic principles state that all maximal phrasal
categories are of the form X".10 This category is expanded by the following Phrase
Structure rules which are part of the categorial component:
(21) XP — S p e c X'
X' — - X Complement(s)
(22) NP
lexical items. D-structures are generated in the Base by inserting these lexical items
into the structures generated by the categorial component, which is subject to some
version of X'-theory.11 It has been suggested that the PS component of the
grammar is in fact redundant, because the principles of subcategorisation, and hence
phrase structure, are determined in the lexicon and are then "projected" from there
to all other levels of the grammar (see (13) above) in accordance with what has
been called the Projection Principle (Chomsky 1981: 29ff), later modified to
incorporate the claim that all sentences have subjects (Chomsky 1982: 10). Thus,
for example, in the lexicon it is specified that a transitive verb like kick takes one
postverbal direct object argument. Such a specification obviates the need for a
phrase structure rule of the form VP—-V NP.
The transformational component in GB-theory now contains only one movement
rule, called Move (a), or sometimes Affect Ca). This rule permits anything to be
moved anywhere subject to the constraints of the modules in (12). Some of these
constraints will be discussed below. S-structure results after the application of Move
(a).
S-structures are assigned PF and LF representations by their respective
components. PF assigns a phonetic representation to sentences and is the level at
which stylistic operations are thought to take place. LF specifies an impoverished
semantic structure which serves as the input to a more comprehensive semantic
representation.
Let us now turn to a description of the principles of the subtheories in (12).
Bounding Theory is the subcomponent in GB-theory which is concerned with
constraining the possibilities of movement. It is assumed that unbounded movement
is not possible for most elements. The principle of Subjacency states that an
element may move across only one Bounding Node (NP or S in English). We can
formulate the principle as follows:
(23) Subjacency
(25) ls. [Coup whoJ [s did you think [s. [COMP e, J [s Bill saw ej)]]
t It I
On the first cycle the Wh-phrase moves to the lower COMP position from where
it is subsequently moved on the second cycle to the higher COMP slot.
Government Theory (the most important subcomponent of GB-theory) specifies
the relations that hold between, usually, the heads of phrases and their dependent
categories. Consider the following prototypical configurations:
(26) a. VP b. PP
V NP Ρ PP
In both (26a) and (26b) the head of the phrase governs its complement There are
a number of definitions of Government. In the one which I will generally use here,
an element α governs an element β in case α c-commands (constituent commands)
β, and every maximal projection which dominates β also dominates α. α is a
lexical category or one of its projections or the AGR element of INFL (Chomsky
1986a: 162). C-command is construed as follows:
Also: if a governs a maximal projection β, it also governs the Specifier and head
of ß. In the 1981 theory maximal projections, such as S', are barriers to
government. A special kind of government, proper government, is required for
traces left behind by moved elements. The Empty Category Principle (ECP) states
that all traces must be properly governed. Proper government is construed as
follows:
Theta Theory deals with the assignment of thematic roles (θ-roles), such as
agent-of-action, goal-of-action etc. to arguments. Theta role assignment usually
takes place under Government. Thus, in (26) above the heads V and Ρ assign a Θ-
role to their complements. Subjects, except subjects of predicates which are realised
by dummy elements such as nonreferential it or existential there, are assigned a Θ-
role by VP.12 The theory predicts that a string which is assigned a θ-role by a
Government-Binding Theory 19
particular lexical head is a complement of that head. Thus a verbal head can assign
a θ-role to an NP, as in (26a), or to a clause as in / believe that Jo is a teacher,
where the ito-clause is assigned a θ-role. To take another example, in I believe Jo
to be a teacher the postverbal NP is not assigned a θ-role because it is not an
argument of the verb believe. Rather, this verb assigns a θ-role to the proposition
Jo to be a teacher as a whole. The Theta Criterion states that each argument must
be assigned one and only one θ-role. Each θ-role may be assigned only once.
Binding Theory deals with the principles that regulate the distribution of
anaphors, pronouns, names and variables in relation to their antecedents. We can
summarise the 1981 version of these principles as follows:
In (28a) the anaphor himself is bound in its governing category to the NP argument
Lawrence (this is indicated by the identical subscript indices), hence the sentence
is grammatical. (28b) is grammatical only if himself is construed as being
preferential with Peter, not with Lawrence. (29a) can only be interpreted correctly
if the pronoun is not bound to the argument Lawrence, but to some other argument
outside the governing category. Similarly, (29b) is grammatical only if him is not
bound to Peter. It may, but need not, be bound to the NP Lawrence. (30a) is
straightforward: the R-Expression Lawrence may not be bound to any other
argument. It may at first sight seem that sentence (30b) violates Principle C of the
Binding Theory, as the trace (t) of the Wh-element, which is a variable, appears to
be bound by its moved antecedent. However, the Wh-phrase has moved to COMP
(see above), and as COMP is not an argument position, the sentence is well-formed.
20 Previous approaches
[+tns][AGR]
-Case J
[-tns] [AGR] to
t— Case —
Believe is a so-called Exceptional Case Marking verb (ECM verb) which has
deleted its complement S' in order to assign Case to the NP Frank. S'-deletion is
necessary because Government is not normally possible across this node, as we
have seen.
Finally, Control Theory deals with the referential properties of PRO. This is a
phonetically null pronominal element which appears, though not exclusively, in
positions where the now generally abandoned rule of Equi-NP-Deletion applied. We
distinguish Object Control in (33) from Subject Control in (34):
Small Clauses (SCs) are structures which have clausal characteristics in that they
contain a subject phrase and a predicate phrase. They are, however, generally
believed not to contain a complementiser position or an INFL-node (though see
chapter six for discussion). The bracketed sequences in the S-Structures (35)—(41)
are Small Clauses:
Not everyone who advocates the existence of Small Clauses agrees that all of these
sentences involve SCs. The analysis of (35) and (39)—(41) especially is
controversial.
A structure such as (36a), most centrally an SC construction, is analysed as
indicated on all levels of grammar on the basis of two important assumptions (cf.
Chomsky 1981: 106):
What the first clause means is that a verb like consider always shows the same
subcategorisation properties; that is, the lexicon specifies that consider occurs in
either of the following frames:
[— , NP]
[— , clause]
This must be so because this verb assigns a θ-role either to a single NP argument
(as in / considered his proposal) or to a propositional (clausal) argument (as in I
considered that he should tell her), but never to two arguments. In sentences like
(36a) what Mike is considering is not Sue, but rather the proposition that she is
intelligent. For this reason a subcategorisation frame of the shape [ — , NP AP] is
not available for consider. Because the two elements following the verb are in a
subject-predicate relation with each other the string Sue intelligent must be a clause.
The second assumption is less straightforward, and requires a more elaborate
explanation. It is the outcome of a line of reasoning in which it is argued that to a
large extent subcategorisation can be dealt with in the lexicon. In older versions of
the theory there was a rich structure in the categorial component with individual
rewrite rules for all the phrase types. This conception of the PS component has
changed radically over the last few years, as we saw above. Stowell (1981)
proposes to eliminate the categorial component altogether, retaining perhaps only
category neutral rules, such as, for example
XP — Spec X'
X' — X Complement(s) (=(21))
which specify that a phrasal category always has a Specifier and an X'-level, and
that X' always contains the head X of XP and possible complements. Inevitably,
this category-neutral rule template will overgenerate, but it is argued that the
ungrammatical structures are ruled out by the interacting principles of (12) in
Government-Binding Theory 23
section 2.3.1. For example, Case Theory is crucially involved in ruling out (42):
The Case Filter, it will be recalled, stipulates that overt lexical NPs which are not
assigned abstract Case are ruled out. In (42) the building is adjacent to a non-Case
assigning noun. Inserting the Case assigning preposition of renders the sentence
grammatical. The linear order of constituents in syntactic structures is also
determined by the principles in (12). Thus, in (43) the object NP a novel
necessarily occurs to the right of the verb because the NP needs to be assigned
Case. In English verbs are instrumental in assigning objective Case to the right.
s-selection
This idea was first proposed in standard theoretical woik discussed above (see
section 2.2). It is not developed further. As we will see below, it has also been
claimed that in Small Clause contructions there is a close syntactic relationship
between the matrix verb and the SC-predicate.
So far we have seen that in the GB-framework the XP in [V NP XP] structures
can be an AP, an NP, a PP, a bare infinitive, a present participle or a past participle
((35)—(41)). What about sentences in which the intervening NP is followed by an
XP in the shape of a ίο-infinitive? Such constructions are not thought to contain
Small Clauses, but are instead analysed as involving clauses which contain an
24 Previous approaches
INFL-node, headed by to, with the feature [-tense]. 7o-infinitive structures are
always given a clausal analysis; i.e. there is always a subject, although it need not
always be phonetically realised. Let us consider the prototypical verbs believe and
want. The analysis of the sentences in (44) and (45) is as in (46) and (47):
It may be asked how James in (46) receives Case. In this sentence there is no
complementiser, and Case is assigned by the matrix verb. As we have seen in the
previous section, believe and verbs like it are Exceptional Case Marking verbs,
which allow the S'-node, a barrier to Case assignment, to be deleted. What is
important is that a verb like believe is treated as a dyadic predicate, i.e. a predicate
taking two arguments: a subject argument and a clausal object argument.
As for (47), essentially following Bresnan (1970, 1972), a/or-complementiser is
posited at D-Structure for wanf-type verbs. In (47a) this prepositional
complementiser, which may be deleted at a later stage, is necessary in order for the
NP Maria to receive Case. The matrix verb cannot perform this task because Case,
which is assigned under under Government, cannot be assigned across an S'-node,
which is a barrier to Government, as we have seen. (47b) is a control structure,
where PRO is an empty pronominal element, the reference of which is controlled
by the matrix clause subject, as indicated by the indices. There is no
/or-complementiser here, because PRO may not receive Case. If it did it would be
governed, and the theory disallows that, as we have seen.
Let us now return to the sentences in (35)—(41). What is the categorial status of
the Small Clauses? Important proposals originate from studies by Tim Stowell
(1981, 1983). In these works it is proposed that all maximal projections (i.e.
categories of the form XP) may have subjects. It is also claimed that verbs that
subcategorise for Small Clauses are sensitive to what elements occur within these
SCs. In so far as he agrees that they contain SCs Stowell would analyse (35)—(41)
as follows:
In each case an NP (or an NP-trace) occurs in the Specifier position of the XP-
complement of the matrix verb, and this NP is the subject of the Small Clause.
According to Stowell the matrix verb is not "indifferent", to use his term, to the
categorial status of the predicate inside the clausal structure. He cites the sentences
below to show this:
It is argued that the Small Clause constituent cannot be S, because the verb would
then subcategorise for an element within an argument, namely for the predicate.
Stowell remarks that:
Thus, for Stowell the phrases AP, NP, PP and VP in (48)—(54) are semantic
arguments of the matrix verb and the subject of these phrases is a Specifier.
Contreras (1987) supports Stowell's analysis. Chomsky (1981) also adopts Stowell's
analysis with the proviso that SCs cannot be maximal projections. For theory-
internal reasons the lexical NP subject cannot be assigned Case if the SC is an XP.
More recently Chomsky has taken a different perspective of the categorial status of
SCs. In Chomsky (1986b) it is assumed that SCs can be maximal projections for
reasons which will be explained in chapter six. In that chapter a number of further
proposals regarding the categorial status of Small Clauses are described and
assessed. Chapter six also presents my own analysis of the internal structure of SCs.
In this section I will discuss the most influential theory of predication that has
appeared in the literature so far, namely that of Williams (1980). In this work
Williams establishes an additional level of representation: Predicate Structure (PS).
PS is derived from surface structure by applying the Rule of Predication:
26 Previous approaches
Thus, two elements NP and X which are in a predication relation with each other
are assigned a common index. Predication is subject to the condition in (59):
It is not entirely clear whether Williams's surface structure, from which the level
of PS is derived, is in fact the same as S-Structure. From the following
representation of his model (Wilüams 1980: 216) it would appear that these levels
are identical:
One might, however, also interpret the model as having done away with S-structure.
Wilüams does not discuss the matter. In any case, the exact status of surface
structure in Williams's approach is not crucial.
Any category, including S and S', can be a predicate. We can thus have the
sentences in (60a—e), taken from Williams (1980), in which the predicate is an AP,
an NP, a PP, a VP and an S respectively:17
Notice that sentence (60d) differs from (60 a—c). This is a case of what Wilüams
terms grammatically governed predication. Although this notion is not defined, we
may take it to refer to cases where there is a predication relation between a matrix
clause subject NP and another element. Other examples are John is sick and John
left singing. The sentences in (60 a—c) are examples of thematically governed
predication where the predicates, all occurring in VP, enter into a relation with the
Government-Binding Theory 27
theme of the verb, i.e. one of its arguments, usually the direct object
What is important for our purposes is that at the surface structure level the NP in
the [V NP XP] string is a direct object At the level of PS the NP has the status of
subject of a predication relation. It will be clear that this notion subject is not
defined structurally as [NP, S], i.e. as the first NP immediately dominated by S.
Instead, in Williams's framework subjects are regarded as external arguments of
maximal projections:
As for the naked (or bare) infinitive in (63b), Williams (1983: 302—303) adduces
three reasons against a Small Clause analysis of these infinitives following verbs
of causation and perception. The first reason is that the analysis of (66a) at the level
of Logical Form should be as shown in (66b), not as in (66c):
28 Previous approaches
The subject of the naked infinitive cannot have narrow scope. Williams argues that
"this would follow if the naked infinitives were not small clauses but Predicative
Phrases (PXPs)" (Wilhams 1983: 302).20 In an earlier section of his paper
Williams shows that for PXPs lowering of the existential quantifier, as in (67c), is
not possible:
Under SC-theory, the c-reading is possible, because (67) has the following structure
(cf. (35) and (48) of section 2.3.2):
To return to bare infinitives, these should be analysed as PXPs, not as SCs, because
in the sentences in which they occur it is also not possible for quantifier-lowering
to take place (cf. (66c)). According to Williams the right predictions concerning the
analysis of sentences such as (66a) are made at the level of Logical Form if PS
representations like (64b) are adopted.
The second reason adduced by Williams why bare infinitives should not be
Government-Binding Theory 29
analysed as SCs is that they never show what he calls thematic independence in
argument position, whereas "normal" infinitives do:21
In (71) jaw and leave do not assign separate θ-roles to John, therefore they are not
thematically independent predicates. For the same reason made and leave in (72)
are not thematically independent. What this means is that bare infinitives cannot
"stand on their own" with an empty subject, and must be related to a separate
argument expression. In (71) and (72) this is not the case, therefore these sentences
are ungrammatical.
The third and final reason for not analysing bare infinitives as SCs is that they
never occur in sentences like (74) with subjects of arbitrary reference:
Cf. It is fun [to leave]. Williams concludes that "[i]n all respects...naked infinitives
are like other naked predicates [such as PXPs], and not like infinitives." (1983:
302). The following structures are assigned to constructions containing perception
and causative verbs:
Williams remarks that "[t]he SC theory would, of course, assign clausal structures
to all of these, with all the problems that we have seen attend such assignments."
(1983: 303). Bare infinitives are seen to constitute an additional argument against
the SC-framework. However, it should be remembered that no strong claim has
been made by SC-proponents that constructions involving bare infinitives should be
analysed as containing Small Clauses (cf. Radford 1988b: 519 and Chomsky 1981:
45, 141, footnote 36). Furthermore, on a more general note, Stowell observes that
"the scope facts bear directly only on the LF representation of small clause
structures, and this leaves open the possibility that the LF structure of small clauses
differs from the S-structure and D-structure representations." (Stowell 1988: 30).
The difference between Small Clause Theory and Predication Theory is explained
by Williams as follows:
terms of clause, and predicate perhaps not at all. The Predication theory takes
the subject-predicate relation as primitive, and clause is then simply a phrase
that instantiates the subject-predicate relation. (Williams 1983: 307)
It is important to realise that for Williams the relation between subject and predicate
holds at his newly created level of Predicate Structure. At Surface Structure in [V
NP XP] configurations the NP subject of the predication relation is syntactically a
direct object. In SC-theory this NP is syntactically a subject at all levels of
representation.
In addition to Williams a number of other linguists have argued in favour of
some form of Predication Theory, notably Schein (1982); Rothstein (1983, 1991);
Culicover—Wilkins (1986) and Napoli (1989). The work of these linguists will be
discussed at various points in the next chapter. Small Clause Theory and
Predication Theory have provided the two dominant analyses in GB-theory of
predicative [NP XP] strings in [V NP XP] constructions. In this book I will adopt
SC-theory. Arguments in favour of positing the existence of Small Clauses, other
than those given in section 2.3.2, will be presented in the next chapter.
The intuitive idea is that certain categories in certain configurations are barriers
to government and to movement (application of the general rule Move-α). A
natrral speculation would be that the same categories are barriers in the two
cases...One of the questions I want to consider is whether there is a reasonable
notion of barrier that has these properties. (Chomsky 1986b: 1)
Barriers is a highly technical work, and the definition of the notion barrier is rather
complex. In simplified terms we can define barriers as follows:
This phrase marker shows that canonical clauses have the status of CP, i.e. they are
headed by the element "C", the complementiser. The lower clause (=IP) is regarded
as the complement of the complementiser. The subject of this lower clause is
positioned in the Specifier position of IP.24 The head of the lower IP clause is "I"
which contains tense and agreement assigners, as we have seen. The complement
of "I" is VP.
Consider now sentence (78) and its representation in (79) on page 32:
(79)
Spec
When there is subject-auxiliary inversion the modal in a sentence like (80) moves
to the complementiser position "C", as is shown in (81):
(81)
Spec
This process of modal fronting has been called I-Movement. See Radford (1988b:
411—420) for a fuller treatment. As the modal has acquired the (present) tense
features, the verb see does not move to "I".
Finally, Wh-elements, which were previously moved into the COMP (= "C")
position when they were fronted (see section 2.3.1), are now moved into the
Specifier position of CP. The tree diagram in (83) shows the derivation of the
sentence in (82), which involves both Wh-Movement and I-Movement:
Government-Binding Theory 33
(83) CP
Spec C
C IP
Spec r
VP
Spec V'
V NP
movement 2
Jtmovement 1
In this sentence the Wh-phrase what moves from the complement position in VP
34 Previous approaches
(88) *When did you see the book which Laura bought.
This sentence is ungrammatical under the reading in which when is construed with
the subordinate clause verb buy. (89) shows its S-Structure:
(89) [CP fspsc wheiij ] [ c did] [,ρ you [yp tj [ w see the book
• movement 4 -
[a-^sp-JjtochJ tip Laura
movement 2
movement ,J
movement 3
The Wh-element which moves from the complement position in the lower VP to
the Specifier position of the subordinate CP via the adjoined position in VP
(movements 1 and 2). Apart from crossing a VP which also crosses an IP.
However, this category is not an inherent barrier, nor does it inherit barrierhood
from VP. Let's now turn to the adjunct when. In (89) it is assumed that this phrase
originates in a position inside the lower IP, but outside the lower VP. It first moves
out of its containing IP (movement 3) into an adjoined position in the matrix VP.
It cannot move into the Specifier position of the lower clause as that is already
filled by which. The movement across the lower IP is not problematic because this
category is not an inherent barrier. However, when also moves out of the lower CP
(the relative clause). This category is a barrier because it is not L-marked: the noun
book does not assign a theta role to the relative clause. Barrierhood is inherited by
the NP which dominates iL The adjunct thus crosses two barriers. This results in
Descriptive treatments 35
ungrammaticality. Finally, when moves from the adjoined position in the matrix VP
to the Specifier position of the matrix clause (movement 4). This movement does
not cross any barriers. Notice that (88) also violates the ECP, as the lowest trace
of the adjunct cannot be properly governed by its antecedent governor (the higher
φ because two barriers (NP and CP) intervene.
It is predicted in the Barriers model that extraction out of complements is
unproblematic because these are L-marked, by definition. Consider (90) below
which exemplifies extraction out of a gerundival NP:
3.1. Introduction
In this chapter the central question is: do Small Clauses exist? What evidence can
be brought to bear on this issue? I will argue that SCs do exist and will present
several arguments supporting the claim that in sentences like (91) the verb consider
subcategorises for a Small Clause complement.
The argumentation will run as follows: if we can show that the [NP XP] string
forms a unit in sentences like (91), where there is a subject-predicate relation
between the postverbal NP and XP, then that unit is necessarily a clause. Thus, the
semantic fact that there is a subject-predicate relation combined with the syntactic
fact (to be demonstrated below) that we are dealing with an [NP XP] constituent
points to a clausal analysis of the [NP XPj string. The semantic and syntactic facts
on their own are not sufficient evidence for a Small Clause analysis, but combined
they are. There are at least five types of evidence supporting the claim that in (91)
consider subcategorises semantically and syntactically for a proposition rather than
for two separate arguments. In this chapter I will be looking mainly at nominal and
adjectival SC-constructions, i.e. constructions in which the XP is an NP or an AP.
It should be noted that all the evidence presented here in favour of Small Clauses
also constitutes evidence against theories of predication.
First, notice that the string this man an idiot in sentence (91) of the previous section
can be coordinated with a similar string:
(93) Scotland Yard last night justified the armed police operation which left an
armed robber dead on a London street and a second wounded.25
It could be objected that the NPs that man and an idiot in (96) are syntactically
unrelated separate constituents. This is unlikely, however, in view of the fact that
the string that man an idiot is pronounced without an intonational break. Radford,
Evidence for the existence of Small Clauses 39
although he recognises the fact that the highlighted [NP NP] sequence in (96) must
be a Small Clause, believes that the Mad Magazine phenomenon "is clearly a
marked construction in a fairly obvious sense: thus, Mad Magazine Sentences seem
to be limited to "echo contexts", and cannot be used to initiate a conversation"
(1988b: 330) In Radford (1988a: 42, note 4) it is claimed that these sentences are
probably outside core grammar. It is not entirely clear that we are dealing with a
marked phenomenon here, as Small Clauses do seem to readily occur on their own
and not only in Mad Magazine contexts. This is certainly the case in languages
other than English, such as Russian,27 but also in English. Consider, for example,
the adjectival Small Clauses in (97) and (98), which, for lack of a better term, we
might call announcement SCs:
I believe it is not controversial to claim that the APs in (97) and (98) are predicates.
If they are, they require a subject, as most linguists would agree (cf. Rothstein
1983: 2, 1991: 5; Napoli 1989: 230- In (97) this requirement is satisfied by the NP
doors, in (98) we must posit a PRO as subject, the reference of which is determined
contextually.28 Consider also the following example quoted in Jespersen: John had
seen Glory on the racecourse in Drake's company: he proud and triumphant, she
bright and gay and happy (1909—1949, part V: 47). Other examples of
independent Small Clauses in English are book titles like Women as traindrivers,
or indeed the title of this chapter. Small Clauses containing the element as will be
discussed in chapter four. In that chapter we will see that prepositional SCs can also
occur on their own.
Small Clauses may also occur as adjuncts. A sentence of the type we have in (99)
is fairly common in the blurb of book covers:
In that case we would still be dealing with an adjunct SC, the only difference being
that it would be part of the subject of the sentence. The appositive analysis of (99)
is perhaps more appropriate because we cannot move the string a journalist by
profession anywhere else other than to a position following the subject (cf. *Mr
Cosmos has written an excellent book...a journalist by profession). A further
example of a construction containing an appositive SC which follows the subject
NP is given in (100) below:
Presumably appositive SCs are adjoined to the X"-level of their host In both (99')
and (100) the Small Clause subject is the empty pronominal element PRO which
is coindexed with the NP subject of the matrix clause.
The Small Clause can also appear right at the end of a sentence as in (101):
(101) Ben Johnson arrived home in Canada last night a broken man,...29
(102) Ben Johnson, arrived home in Canada last night [ s c PRO, a broken man],...
Here again what appears to be a simple NP is in fact best analysed as a Small
Clause of the form [NP NP], where the first (subject) NP is the empty pronominal
element PRO. The highlighted strings in (99') and (102) have been called free
adjuncts in the literature. This term as Beukema (1984: 56) informs us, was coined
by the Dutch grammarian Kruisinga.
The phenomenon of two juxtaposed NPs occurring in a subject-predicate
relationship with each other is also fairly common in coordinative structures:
(103) Apart from the subjective work, we cannot say that Jesus is the Lord, or [His
death a death for sin], or [His resurrection the first-fruits of the new
creation of God]. (W.9.1.152—2)30
(104) and as they line up # Northern Circuit has to be # on the far side # drawn
one # two will be Princess Shumar # [three Marche d'Or] # [four Courts
of Love] # [five # Canute's Courtier] # [six # Sea Shanty] # and [seven # the
top weight # Non Proven] #... (S.10.4b.20)
(105) Hugh had been 8 and [he 10] when they had first realized there was
something a bit different about them. (W. 16.7.28—2)
Evidence for the existence of Small Clauses 41
(106) It is for the same reason that Frege maintains that an A4 can never be a B4
or vice versa, than an object can never be a concept or [a concept an
object]. (W.9.2.141—2)
All linguists would agree that in each of these sentences the bracketed strings are
clauses. They would not, however, necessarily agree that they are Small Clauses.
It could be argued that in each case we have ellipsis or gapping. For (103) a Small
Clause analysis of the higlighted sequences poses no problem because here only the
verb be and the feature Opresent tense] need to be inferred. In (104)—(106),
however, the situation is a little more complex: in (104) apart from the verb be we
also have to infer the modal auxiliary will. In (105) in order to interpret he 10 we
have to infer from the previous clause (Hugh had been 10) the verb be as well as
the features [+past] and [+perfect]. Finally, in (106) the element never of the clause
than an object can never be a concept also has scope over the clause a concept an
object. Furthermore, in this sentence the verb be and a modal also need to be
inferred. Because of all of this and especially also in view of sentences like (107),
where there is certainly no predication relation between the NPs that one and a
palace, an ellipsis or gapping analysis of (103)—(106) is more plausible.
(108) She was born in 1896, daughter of a well-to-do and rigidly conventional
family, her father a doctor.31
(109) The Red Cross workers entered the besieged building this morning, their flag
their only protection.
To interpret her father a doctor in (108) the reader need not rely on anything that
precedes this string, certainly not on the presence of the verb be in she was born
in 1896. If the sentence had read as in (108'):
(108') She died in 1896, daughter of a well-to-do and rigidly conventional family,
her father a doctor.
where there is no verb be, we would still interpret her father as a subject and a
doctor as a predicate. So, in (108) we again have an adjunct SC, this time with a
lexical subject. (109) also contains an adjunct SC with a lexical subject.32
Not only [NP NP] Small Clauses occur as adjuncts, [NP AP] strings do so as
well. The following example is from Beukema (1984: 58):
42 Small Clauses as constituents
(113) Its hands free, Moscow could turn to the great regional problem that had
long defied solution: the Arab-Israel conflict and the Palestinian issue.33
In (110) and (113) the SC-adjuncts are arguably sentential adverbials, whereas the
Small Clause in (112) is more plausibly analysed as an NP adjunct. The exact
structural position of these SCs is not directly relevant to the discussion, however,
so I will not go into this matter here. The syntactic position of sentential adjuncts
will be discussed below and the Case-theoretical properties of the Small Clause
constructions in (108), (109) and (112) will be discussed in chapter six.
Further examples of predicative [NP XP] strings occurring other than as
complements to transitive verbs are given in (114):
(114) a. With Peter the referee we might as well not play the match.
b. With Laura the principal actress, how could the Duras play go wrong?
Here nominal Small Clauses occur as the complement of a preposition (mostly with
or without in English, though see below) in the absolute construction. To satisfy the
Case Filter with assigns Case to the NPs Peter and Laura. Adjectival and
prepositional SCs can also appear in this construction:
(115) With Jeremy happy we can now go and see that film.
(116) With the boys in bed we can have a quiet cup of coffee.
Persuasive evidence based on Dutch data that the [NP XP] string following the
preposition is indeed clausal is given in Beukema—Hoekstra (1983,1984). In Dutch
Evidence for the existence of Small Clauses 43
non-human pronouns which are complements of prepositions are changed into so-
called R-forms, specifically the element er which then precedes the preposition. The
examples below are from Beukema—Hoekstra (1984: 692):
(117) a. *Jan keek naar het paard en Piet keek ook naar het.
Jan looked at the horse and Piet looked too at iL
'Jan looked at the horse and Piet looked at it too'
Thus, naar het (P + pronoun) is changed into er naar (R-pronoun + P). The crucial
point now is that non-human pronouns following the met (=with) of absolute
constructions are not subject to the R-suppletion rule described above:
(118) a. Met dat nog allemaal voor de boeg, wilde Jan niet weggaan.
with this still all ahead of him wanted Jan not leave
'With all this still ahead of him, Jan did not want to leave'
In (118) the pronoun dat is not changed into its R-form counterpart daar. This
shows that dat in (118a) is not the complement of the preposition met. As there is
a subject-predicate relation between dat (nog) allemaal and voor de boeg the two
phrases together should be taken to form a clausal constituent.
Small Clauses can also be preceded by other prepositions. Thus, we can have She
was distressed at the thought of [him alone] (from Kayne 1984a: 161, note 2835)
or All works, each in their degree, are the making of [madness sane] (from
Jespersen 1909—1949, part V: 29, quoting Carlyle). Jespersen also quotes the
following example from Defoe: I would have given half I had in the world for [him
back again], but believes, mistakenly I feel, that this sentence is no longer possible.
Finally, SCs may occur after the preposition than, as sentence (126) below shows.
However, it seems that in modern English only wir/i-absolutives are fully acceptable
and that other Preposition+SC constructions are slowly becoming obsolete, with the
exception perhaps of than+SC. Absolute constructions are problematic for Napoli's
theory of predication. See Napoli (1989: 278—279).
An additional set of facts confirming that Small Clauses can occur in positions
other than as complements to verbs was noted in Safir (1983). He observes that SCs
44 Small Clauses as constituents
can be what he calls honorary NPs (HNPs); these are maximal projections of some
type which he claims may occur as subjects only in copular or Raising
constructions (cf. (119a) and (119b)), but not in other structures (cf. (119c)):
In (119) the HNP is a Prepositional Phrase. Safir shows that SCs can also act as
HNPs, again, so he claims, only in copular or Raising environments:
(120) a. Workers angry about the pay is just the sort of situation that the ad
campaign was designed to avoid. (Safir's (12a))
b. Workers angry about the pay does indeed seem to be just the sort of
situation that the ad campaign was designed to avoid. (Safir's (12b))
c. ^Workers angry about the pay pleases Maybelle immensely.
(Safir's (12c))
These sentences are entirely acceptable when the Small Clause contains as:
A plural SC subject with a singular matrix verb seems much less acceptable though,
even with as:
Safir's claim that SCs may occur in subject position only in copular or Raising
environments is questionable.36 Consider the sentences in (124) which contain
Evidence for the existence of Small Clauses 45
(124) a. ?George Bush President distresses large groups of people all over the
world.
b. ?Margaret Thatcher Queen would prove quite ludicrous.
When the Small Clause subjects in (124) contain as the results are again better:
(125) a. George Bush as President distresses large groups of people all over the
world.
b. Margaret Thatcher as Queen would prove quite ludicrous.
However, that the element as need not be present is shown in (126) where we have
an attested instance of an adjectival SC without as occurring in subject position,
again in a non-copular, non-Raising environment:
(126) Will President Botha set him free to allow him to take part in that process?
Or does he fear that Mandela free would pose a greater threat than Mandela
behind bars.37
I will have more to say about Small Clauses containing as in the next chapter. The
only general restriction on SCs occurring as subjects seems to be that they must
express a situation. In view of this, Safir's sentence (120c) seems at least
marginally acceptable. The reason why it seems odd is because it expresses a
strange proposition: workers angry about pay is not something that normally pleases
people. An alternative to (120c) is the following: Workers angry about the pay
displeases the government which is fine. As noted above, sentence (126) also
demonstrates the possibility of an SC occurring after the preposition than (namely
Mandela behind bars)?*
As a fourth piece of evidence in favour of assigning a clausal status to [NP XP]
strings in which the NP and XP are in a subject-predicate relation, consider the
following sentences:
(127) I thought [it perhaps a pity] at the time, but his motivation was pessimism
(unnecessary in his case) about academic job prospects. (W.17.2.74)
(128) I must admit that I have found [these summer international schools probably
the most rewarding part of my work], (W.7.6.82)
What is interesting about these sentences is that the phrases perhaps and probably
occur in between the NPs and XPs which are in a predicative relationship with each
other. These elements are sentence adverbials which have scope over a string of
words which syntactically form a clause. In this connexion note that we can
paraphrase the first part of (127) as in (129a), (129b) or (129c), but not as in
46 Small Clauses as constituents
(129) a. I thought that [it was perhaps a pity] at the time, but his motivation was
pessimism...
b. I thought that [perhaps it was a pity] at the time, but his motivation was
pessimism...
c. I thought that [it was a pity perhaps] at the time, but his motivation was
pessimism...
(130) a. Perhaps I thought [it a pity] at the time, but his motivation was
pessimism...
b. Perhaps I thought that [it was a pity] at the time, but his motivation was
pessimism...
c. I thought perhaps that [it was a pity] at the time, but his motivation was
pessimism...
(131) a. / must admit that I have found that [these summer international schools
are probably the most rewarding part of my work],
b. I must admit that I have found that [probably these summer international
schools are the most rewarding part of my work],
(132) a. I must admit that probably I have found [these summer international
schools the most rewarding part of my work],
b. I must admit that I have probably found [these summer international
schools the most rewarding part of my work].
Thus, the distribution of the sentence adverbials perhaps and probably in (127) and
(128) strongly suggests that the bracketed strings are syntactically clauses.39 40
On adverbials in SCs see also Stowell (1988: 14—15) and chapter six.
A fifth and final way of proving the clausal status of predicative [NP XP]
sequences in sentences like I consider this man an idiot is to show that the
postverbal NP is a syntactically a subject. This is a strategy followed in Radford
(1988b). Let us look at his arguments. Radford (1988b: 324f) cites the following
sentences from Napoli (1989: 319):
It is concluded on the basis of data like this that what is called a floating emphatic
reflexive can only be related to a subject expression.41 If so, then the highlighted
NP in (134) is necessarily a subject:
Evidence for the existence of Small Clauses 47
If the NP the prime minister herself is a subject then the bracketed sequence must
be a clause.
Radford's other two arguments centre around Postal's noi-initial (1974: 94—99)
and alone-final (1974: 99—102) tests: nof-initial and α/one-final NPs are claimed
to occur only in (derived) subject position (cf. [Not many houses] were built here/*I
like [not many houses]; [That house alone] was paintedJ*I ate [that cake alone]).
If Postal is right then the higlighted NPs in (135) must be subjects:
(135) a. The head of department considers /not many students good PhD students],
b. The head of department considers [MA students alone good PhD
students].
Again, if not many students and MA students alone are subjects, then the bracketed
strings must be clauses.
The not-initial and abne-fmsi arguments are less convincing than the floating
emphatic reflexive argument We can have, for example, (136):
Kayne assumes that only binary branching is allowed in phrase markers and that
nothing can be extracted from left branches. If we combine Kayne's hypotheses
with the fact that subjects are on left branches, then we must assume that the father
of Kathy in (137) is a subject because we cannot extract from it:
Hoekstra (1984: 253f) argues that recognising the existence of Small Clauses (in
addition to making a few further assumptions) is necessary in order to give an
explanatory account of verbal compounds, and Stowell (1988: 15—17) argues that
an optimal account of Binding facts in English must make use of the notion Small
Clause.43
I have shown in the previous section that the NP this man is not a direct object
argument of the verb consider, but rather the subject of the Small Clause [this man
an idiot]. The analysis of (140) is shown in (141) (only the VP is shown):
The verb consider subcategorises a clausal complement the subject of which is this
man and the predicate of which is an idiot. The internal structure of the Small
Clause has been simplified here and will be discussed in more detail in chapter six.
The analysis in (141) is not appropriate for all constructions containing Small
Clauses: in some cases the NP of a [V NP XP] string is clearly a direct object.
Consider (142):
In this sentence her is an argument of the verb appoint. This is so because (142)
entails (143):
Also, notice that appoint and verbs like it cannot be followed by semantically
empty elements such as nonreferential it or existential there. We must conclude,
Small Clauses as instances of dependent controlled predication 49
then, that in (142) appoint assigns a θ-role to the postverbal NP. However, notice
that in (142) there is also clearly a subject-predicate relation between the object NP
and the NP professor of logic. The distinction between the construction we have in
(142) and that in (140) is not always made by grammarians.
There are a number of possible ways of analysing (142). A first possibility is to
argue that this sentence involves a complex verb appoint-professor of logic which
takes an NP argument her. For a proposal along these lines see e.g. Dowty (1976:
219f, 1979: 303). In deriving the S-Structure there would then have to be obligatory
movement of some sort. We cannot take such an analysis seriously, however, as it
would result in an unacceptably large lexicon which would contain countless strings
such as appoint-Head of Department, appoini-senior lecturer etc. Stowell (1988)
argues for a restructuring at LF of sequences like hammer NP flat as hammer-flat
NP. In the syntax, however, there is no complex verb of any sort. This view is
more plausible.
Another possible analysis, which I would like to discuss in some detail, is
presented in Hornstein—Lightfoot (1987: 27). They argue for the analysis of (144a)
as in (144b):
Eat in (144a) is like appoint, in that this verb too is followed by a direct object NP
and a predicate phrase (raw). This predicate phrase is analysed in (144b) as taking
a PRO subject and forms a Small Clause with iL Small Clauses are analysed as Ss
taking a zero INFL node. The particulars of Hornstein—Lightfoot's analysis need
not concern us here (see chapter six for further discussion); what is important is
that in (144b) the SC [PRO INFL0 raw] is a sister to V'.44 It is argued that this
adjectival SC is a VP-adverbial. This would seem uncontroversial.
The structure of (144a) is different from that in (145a) which is analysed as in
(145b):
(146) appears to indicate that (144a) and (145a), examples of what we might call
the meat-eating construction, are indeed syntactically different In the present study
50 Small Clauses as constituents
(147) John INFL [VP [VP ate the meat] [sc PRO raw]]
That is, the SC [PRO raw] is adjoined to VP, rather than being a sister to V'. The
reason for this is that I follow Chomsky (1986b) in allowing adjunction only to
maximal projections. Such a limitation results in a more restrictive and hence more
learnable grammar. Also, only if the SC is adjoined to VP does its PRO subject
remain ungoverned as is required by the PRO-theorem (see section 2.3.1). The
question now arises how to analyse (145a). If [PRO raw] in (144a) is adjoined to
VP, where do we adjoin [PRO naked/? Before answering that question a few
observations are in order. I would like to suggest that the sentence John ate the
meat naked is in fact ambiguous. Under one reading naked is a sentential adjunct.
The NP the meat and the AP naked are then separated by an intonational break,
represented by a comma in writing: John ate the meat, naked. This last sentence
then alternates with (146b). Under another reading naked is a VP-adjunct. The
following data, taken (slightly adapted) from Andrews (1982: 313) indicate that the
AP naked (or SC [PRO naked] in terms of the present study) can indeed be part
of VP. See also Roberts (1988); Napoli (1989) and Malldn (1991).
(148) a. John said he would eat the meat naked, and eat the meat naked he did.
b. Eat the meat naked though John did, nobody thought he was crazy.
c. What John did was eat the meat naked.
Her claim is that (149b) makes two assertions, namely that the ambassador arrived
and that he was nude, whereas (149a) makes only one assertion, namely that when
the ambassador arrived he was nude. In Napoli's words "[nudeness] characterises
the arrival" (Napoli 1989: 154). The different structures of (149a) and (149b) are
responsible for the differences in interpretation.
Roberts (1988) adduces two further arguments for taking subject-related AP-
predicates to be part of VP. Consider first the sentences in (150) below:
Small Clauses as instances of dependent controlled predication 51
and claims that this sentence cannot be construed in such a way that the scope of
negation extends over the main predicate alone, excluding the string angry at John.
In his own words: "[151] completely lacks the reading where Bill, angry at John,
didn't leave" (1988: 707). Thus, (151) can only have the readings (i) not [Bill left
angry at John] and (ii) Bill left [not [angry at John]], but not (iii) [not [Bill left]]
angry at John. Under reading (i) do is stressed in (151), under reading (ii) John is
stressed. In his paper Roberts shows that VP-adjuncts must be inside the scope of
negation, and the fact that (151) can only have the readings in (i) and (ii) confirms
that the AP angry at John must be a VP-adjunct. However, what about the sentence
Angry at John, Bill didn't leave? We will have to analyse this as a variant, not of
(151), but of (151'):
where there is an intonational break between leave and the AP angry at John
(Roberts contrasts John didn't kiss his wife deliberately with John didn't kiss his
wife, deliberately where deliberately is a VP-internal adverb in the first sentence,
but a VP-external adverb in the second sentence).
It now appears that a sentence like John ate the meat naked should be assigned
two distinct syntactic structures. Let us consider in some more detail the reading
under which naked is construed as a sentential adjunct. As we saw above, under
this interpretation there is an intonational break between the meat and naked. We
should now ask what is the exact position of the SC [PRO naked] in John ate the
meat, naked and in Naked, John ate the meat (=(146b))? I want to suggest that
when naked is construed as a sentential adjunct it is adjoined to CP. How can we
show this? With regard to sentences like (146b) Napoli (1989: 114) argues that the
adjunct naked (which for her is not an SC) is Chomsky-adjoined to CP. The
evidence for this is that adjuncts of this type can precede a matrix clause with a
filled Spec-of-CP position, as in Napoli's sentence Penniless, [Spec who] can we
52 Small Clauses as constituents
trust? I would like to extend Napoli's analysis for sentences like (146b) to John ate
the meat, naked where there is an intonational break between the meat and naked.
The sentences under discussion are then analysed as in (152a) and (152b)
respectively :
Thus the adjunct Small Clause [PRO naked] is adjoined to CP in both cases. These
analyses directly capture the scopal properties of [PRO naked] as a sentential
adverbial in both sentences. The representations above are each other's mirror
images, and we can straightforwardly account for the relation between them by
assuming that we can swivel the phrase markers of (152a) and (152b) around the
vertical axis that we can draw through the higher CP:
CP CP
(153) a. ?*John said he would eat the meat nude, and eat the meat he did nude.
b. ??Eat the meat though John did nude, nobody thought he was crazy.
c. What John did nude was eat the meat.
(154) a. ??John said he would eat the meat in the nude, and eat the meat he did
in the nude.
b. Eat the meat though John did in the nude, nobody thought he was crazy.
c. What John did in the nude was eat the meat.
These data seem to contradict, at least in part, those in (148) in that it looks as
though the AP nude can (marginally) appear outside VP. However, to the extent
that (153)—(154) are acceptable, they can be taken to support the CP-adjunction
analysis. In other words, the AP-predicates that trail behind in these sentences
([PRO nude]/[PRO in the nude] in the present framework) have sentential scope
and are adjoined to CP, and hence are not affected by VP movement processes.
Not all authors make mention of an alternation between John ate the meat naked
(without comma intonation) and Naked, John ate the meat, but those who do
implicitly or explicitly structurally relate these two sentences (thus
Hornstein—Lightfoot talk of "transportability"). They do not, however, spell out
exactly how the first sentence is syntactically related to the second. In view of what
I have said above, I would argue that there is no syntactic relationship between
Small Clauses as instances of dependent controlled predication 53
in her framework both (144a) and (145a) are analysed as involving "flat" VP
structures: [VP VNP XP], She disallows adjunction, as in (147). A further problem
for Napoli is noted by Malten (1991: 386—387):
For Napoli both the PP complement and the subject-related predicate are sisters of
the head verb. Under that analysis (156) ought to be possible. For the present
framework (156) is not problematic as on the table is analysed as a complement of
put, and hence as its sister, whereas happy (an SC with a PRO subject) is not a
sister of V, but adjoined to VP, and therefore higher up in the syntactic tree
representation than the PP.
Let us now consider some further ways of analysing (142). Six different possible
structures are shown in (157)—(162) below (recall that I am omitting Spec-of-VP
wherever it is irrelevant):
Small Clauses as instances of dependent controlled predication 55
In (157) her is a direct object NP, and the string professor of logic is analysed as
a second (predicative) sister category of the verb appoint. This kind of analysis is
found in theories of predication, notably in the work of Williams (1980) (see
chapter two); Rothstein (1983); Culicover—Wilkins (1986) and Napoli (1989). See
also Roberts (1988) and Malldn (1991). Green (1970, 1973) and Emonds (1985)
have similar analyses which are the same even for constructions containing verbs
like consider.
The representation in (157) has to be ruled out as it stands for two reasons. First
of all it violates the Theta Criterion as defined in chapter two because the NP her
is assigned a θ-role both by appoint and by the NP professor of logic. Secondly,
it does not take account of the fact that there is a subject-predicate relationship
between the NPs her and professor of logic. We could of course modify the Theta
Criterion, as has been proposed in the work of Schein (1982: 1 and 11), Rothstein
(1983: 93—95), Chomsky (1986a: 96—97) and Napoli (1989: 86), and we could
then make explicit the fact that there is a subject-predicate relationship between her
and professor of logic in (157) by positing predication coindexing rules which
coindex subjects and predicates. Examples of such rules are Williams' (1980) Rule
of Predication, which coindexes a subject and a predicate only when they c-
command each other (see chapter two), Rothstein's (1983) Predicate-Linking Rule
which also operates under mutual c-command, Culicover—Wilkins' (1986) Coindex
Rule and Napoli's (1989) Predication Coindexing Principles. With regard to the
Theta Criterion, we will see below why a revision is not desirable. As for
predication theoretical coindexing rules, as was said at the beginning of chapter
three (and as Hoekstra 1984: 239 has observed regarding Williams' work) all the
arguments given in favour of positing Small Clauses are also arguments against
such rules.
Let us take a closer look at some of the proposals made by a proponent of one
version of Predication Theory, Susan Rothstein, regarding constructions such as
those in (142) and (144a), repeated here as (163) and (164):
According to Rothstein these sentences are structurally distinct. She discusses the
similar sentences (165) and (166), and analyses (165a) as in (165b) and (166a) as
in (166b):
The AP predicates red and raw are linked to their subjects the car and the carrots
by the Predicate-Linking Rule (Rothstein 1983: 11). It is claimed that the verb paint
in (165) has two sister categories: an NP argument phrase and an AP predicate
phrase (cf. (157)), whereas in (166) eat takes only an NP as its sister. The AP
predicate in this sentence is adjoined to VP. In the present study I have also
analysed (166a) as in (166b). Rothstein terms the AP predicate in (165) resultative,
and that in (166) depictive, adopting terminology introduced in Halliday (1967:
63).45 The syntactic difference between the resultative and depictive constructions
in (165) and (166), she claims, is that in (165) there is a connexion between the
verb and the predicate, whereas there is no such connexion in (166): "Resultative
predicates, like thematic arguments of a verb [emphasis mine, BA] are selected by
the semantic nature of the verb and are represented at S-structure as immediate
sisters of the verb." (1983: 17)46 Also: "[W]ithin the lexicon, selectional demands
are for semantically appropriate predicates, and are stated in category-neutral
terms." (1983: 83) Rothstein claims that her analysis makes a number of desirable
predictions. One of these is that only one resultative can appear in a sentence, not
two:
That is, a verb tolerates only one resultative predicate as a sister. There is no
restriction on the number of depictives that can occur:
(168) They eat meat raw, tender. (=Rothstein's (30), quoting Simpson 1982)
The depictive predicate tender is Chomsky-adjoined to VP' (eat the meat raw) here.
Another prediction which Rothstein claims her analysis makes is that if a
sentence contains both a resultative and a depictive, the resultative must precede the
depictive. This is because the depictive is higher up in the tree (cf. (165)/(166)).
Emonds (1976: 109) makes the same point as Rothstein regarding cooccurring
resultatives and depictives by contrasting the a and b-sentences in (172) and (173):
The distinction between resultatives and depletives is clearly a valid one. Thus, in
(163) the person designated as her is a professor as a result of the appointing.
Similarly, the car in (165a) is red as a result of the painting process. By contrast,
in (166) the predicate raw does not express a result, but rather a circumstance. This
sentence can be paraphrased as 'John ate the meat while it was raw'. Within the
classes of resultative and depictive predicates further semantic differences can be
distinguished which I will not discuss here. I refer the reader to Green (1970,
1973). Returning now to Rothstein's claims, I agree with her that (167) is bad.
However, (168) does not seem to be any better. I would say that sentences (167)
and (168) are either equally good or equally bad. To me it seems to be the case that
a direct object cannot occur with two resultatives nor with two depictives.47 The
difference in judgements between (170) and (171), and between the a and b-
sentences in (172) and (173) is likewise very subtle, but here (170) is better than
(171), and in (172) and (173) the a-sentences are better than the b-sentences.
Rothstein's account thus seems to be making a correct prediction. I will return to
these sentences below.
I am not very happy with Rothstein's lexical selection analysis of resultative
constructions. Are they really to be analysed as in (174), as she suggests, or perhaps
as in (175), which is the analysis I assigned to depictive constructions such as he
ate the meat raw?
(174) VP
V NP XP
(175) VP
V NP
The answer to this question lies in the nature of the relationship, or lack of a
relationship, between the head verb and the resultative predicate: if such a relation
obtains (174) would seem to be more appropriate, if it does not obtain (175) is to
be preferred. Data which are relevant to this issue are given in (176)—(179):
r beautiful^
(176) *John hammered the metal 1 safe >
L tubular J
60 Small Clauses as constituents
damp Λ
{ dirty >
stained J
lame
( paranoid
(Rothstein's (17a))
and remarks that "unread is predicated of the patient NP the book, but drunk can
only be predicated of John. Despite the fact that configurational restrictions are
met, Mary is the goal NP and cannot be subject of drunk." (1983: 85)
(182) *He hammered the reflection of City Hall off of the fender.
Here "being apart" is not a property of the boards themselves, and McCawley's
hypothesis would predict this sentence to be unacceptable.
A different attempt at formulating the semantic restrictions holding between
resultative predicates and their subjects in sentences like (176)—(179) is made in
Fabb (1984). He discusses the sentences in (184), among others
and remarks that "these grammaticality differences are due, not to lexical selection
of specific adjectives, but to a semantic restriction on resultative predicates, that
they tend to express extreme resulting states, often final states (so result APs often
express the destruction or exhaustion of the NP)." (1984: 106) He furnishes
evidence for this by citing sentences like those in (185) (1984: 227):
in which the predicates clearly express an extreme/final state and yet the sentences
in which they occur are bad.
What is clear from all this is that if we do have semantic restrictions holding
between resultative predicates and their subjects they are very difficult to state.
Furthermore, notice that it seems to be the case that not all of the sentences in
(176)—(179) are equally bad. Thus, in (179) selecting lame as a predicate seems
to result in a better sentence than selecting paranoid. This suggests that the
observed restrictions should perhaps not be treated in terms of semantic principles,
which belong to the grammar, but in terms of extra-grammatical principles whose
application is context-driven. Dowty, in discussing (176)—(179), suggests the
possibility that
the rule responsible for producing this type of sentence is to be unrestricted and
that sentences like [(176)—(179)] are to be excluded by pragmatic or stylistic
considerations, if at all. I believe the careful observer will find that native
Small Clauses as instances of dependent controlled predication 63
(187) A I was surprised to see Clara clean the kitchen table this morning.
Β Yes, I asked her to tidy up, but she didn't do a very good job. At one point
she was wiping the tables, but rather than wiping them clean, she was
wiping them dirty by using that filthy rag which is normally used for the
loo!
(188) A I hear that loony shepard was shooting at people again the other day.
Β Yes, that's right, but this time things got out of hand: he shot Tim in the
arm and Jeff is now lame.
A He shot Jeff lame? That man should be put behind bars!
It would seem, then, that the optimal treatment for the so-called hammer flat-
construction is one in which the grammar freely generates resultative [V NP XP]
strings, some of which are filtered out by pragmatic principles (which I will not
attempt to formulate here). We can extend this pragmatic account to depictive
constructions such as those in (181), so that Rothstein's conditions (a) and (b) are
in fact pragmatic, rather than semantic, in nature.
Similar principles would also explain the contrast we get between (189), which
contains a fronted depictive predicate, and (190), which has fronted resultative
predicates:
These sentences are from Chomsky (1986b: 82). He tentatively suggests that the
difference in grammaticality judgements between (189) and (190) may be due to
the fact that there is a relation of some sort between the matrix verb and the Wh-
predicates in (190), but not in (189). He does not discuss the nature of this
relationship in any detail. I argued above against a syntactic distinction between
depictive and resultative constructions in terms of lexical selection of the matrix
verb. We can explain the contrast between (189) and (190) in pragmatic terms by
observing that depictive predicates tend to be non-gradable, at least in the contexts
under discussion, and hence also resist the type of Wh-movement we have in (189).
This observation about gradability is supported by the oddness of the sentences in
(191), all of which contain depictive predicates:
64 Small Clauses as constituents
See also Fabb's sentence (185b) above. Notice that we cannot form Wh-echo
questions from the sentences in (191):
Clearly, as Chomsky would agree, the oddness of (191) and (192) cannot be
explained in terms of lexical selection because we are dealing with adjuncts. (191)
and (192) contrast with (193), the Wh-echo versions of (190):
My point, then, is that the contrast between (189) and (190) comes about as the
result of a pragmatic constraint, and no reference needs to be made to the presence
or absence of a putative lexical selection relationship between the matrix verbs and
the predicates.
In the preceding paragraphs I have argued against analysing the VP constituent
of (142) (=We appointed her professor of logic) as in (157) (= [VP [v appointed]
[NP her] [NP professor of logic]J). In other words, I have argued against taking
predicates like professor of logic to be sisters of the head verb inside VP. I have
claimed that neither resultative nor depictive predicates are selected by the matrix
verb and that constructions containing such predicates should therefore be assigned
the structure in (175). Sentence (142) is then analysed as in (158). It might be
asked how constructions involving object-related resultative or depictive predicates
are interpreted differently if they are indeed syntactically the same. A very
straightforward solution would be to assume that the matrix verbs of the sentences
that contain them are marked in the lexicon with the feature [± resultative].
Let us now briefly return to sentences (170)/(171), (172) and (173), repeated here
for convenience:
is also suggested by the observation that we are always dealing with what is
traditionally called an affected objecL..even though [verbs like paint] can
normally take an effected object as their complements. For example, the verb
paint can take either an affected or an effected object in a simple V NP
construction (cf. the ambiguity of "John paints a house"), but in the SC
complement construction the postverbal NP is necessarily interpreted as an
affected object, i.e. an expression referring to an entity which exists
independently from the action mentioned by the verb rather than coming into
existence through the action. (Hoekstra 1988: 117)
Hoekstra argues that the fact that 'the door is painted' is an implication of the
sentence they painted the door green. This implication, so he claims, is a result of
66 Small Clauses as constituents
what he calls a shadow interpretation. He observes that "[o]n some occasions these
implications can be cancelled without any evidence as to a difference in structure
(e.g. I have painted my fingers black and blue when I painted the walls in this
room)." (1988: 117) I do not find Hoekstra's argumentation particularly convincing
as the relation between they painted the door green and they painted the door is
stronger than mere implication. As we saw above, we are dealing with entailment
here. That is, the first sentence entails the second. This can easily be demonstrated
by pointing out that it would be a contradiction to say "I painted the door green, but
I didn't paint it" (keeping the meaning of paint constant). All this means that the
postverbal NP in (160) is assigned a θ-role by appoint and should be analysed as
a direct object argument of that verb.
In the foregoing discussion I have analysed (142) as in (158). We should ask,
however, why the SC is adjoined to VP, rather than dangling from IP, as in (161),
or being adjoined to IP, as in (162). To see why (158) is correct consider the
following sentences:
(194) a. I said I would appoint her professor of logic, and appoint her professor
of logic I did
b. */ said I would appoint her professor of logic, and appoint her I did
professor of logic.
VP-fronting, which has applied in (194a), is a well-known test for VP-hood (see
e.g. Andrews 1982; Roberts 1988). The fact that (194a) is grammatical and (194b)
is not strongly suggests that the SC [PRO professor of logic] is part of VP, as in
(158). We can draw the same conclusion on the basis of the sentences in (195) and
(196):
(195) a. Appoint her professor of logic though I did, she made it known that she
wasn't interested in the chair.
b. ?* Appoint her though I did professor of logic, she made it known that she
wasn't interested in the chair.
(196) a. What I did was appoint her professor of logic,
b. *What I did professor of logic was appoint her.
constructions containing a predicative [NP XP] string can belong to either of two
classes, namely those given in (197) and (198) below:
It is appropriate at this point to introduce some new terminology. I will refer to the
canonical example of the subject-predicate relation, the sentence, as an instance of
independent predication. In constructions containing this type of predication the
subject is in the Spec-of-IP position, and the predicate is the VP (see chapter two).
The type of predication we have in (197) will be referred to as dependent
predication because the Small Clause is a complement of the main verb, and is
assigned a θ-role by that verb. ECM constructions, and indeed all constructions
involving clausal arguments, are also instances of dependent predication. In (198)
the predicative [NP XP] string is an example of what I will call controlled
predication: the subject of the adjunct Small Clause is controlled by the matrix
clause object which is assigned a θ-role by the main verb. Sentences like I ate the
meat naked are also instances of controlled predication, except that in these cases
the controller is the matrix clause subject. Constructions which contain persuade-
type verbs, i.e. verbs followed by an argument NP and a ίο-infinitive clause with
a PRO subject (e.g. I persuaded [him] [PRO to leave]) are not instances of
controlled predication, because the ίο-infinitival clause is a complement of the
matrix verb. They are a subclass of the dependent predication structures and we
might refer to them as instances of controlled dependent predication. Notice that
Small Clauses can occur in all types of predication structures. An example of a
controlled dependent predication construction involving an SC is the sentence /
described him as a complete idiot. See (393)/(394) in section 4.4.2.2 below for
further discussion. We should perhaps recognise a fourth type of predicative
construction, rather problematic for SC-theory, which we might term exceptional
predication. It occurs in NPs and is discussed at great length in Napoli (1989) (see
especially chapters three and four). Examples are given in (199) and (200) below:
to the type of Small Clause they subcategorise. Verbs conforming to pattern (198)
will be referred to as appoint-type verbs. -constructions involve what
Rothstein (1983) has called secondary predicates. Secondary predicates are XPs
which are predicated of an argument which is assigned a θ-role by some lexical
head. In Rothstein's framework they may occur as sisters of the head verb
(resultatives) or in a position adjoined to VP (depictives), as we have seen.
There was a total of 143 nominal Small Clause constructions in the corpus. Of
these 123 (86%) involved consider-type verbs, whereas only 20 constructions
contained appoint-type verbs (14%). The number of adjectival SC-constructions was
significantly higher: there were 211 such structures. 187 of these involved matrix
verbs from the consider-class (88.6%), and 24 (11.4%) involved verbs from the
appoint-class.*9 Comparing this last set of figures to those for the nominal SCs we
see that the ratio of consider-\erb& to appoint-vetbs remains constant across the
nominal and adjectival SC-constructions.
Let us now turn to a different matter. Apart from setting up two classes of verbs,
the consider-class and the appoint-class, there appears to be another way in which
we can subdivide [V NP XP] strings. Notice that some, but not all, verbs occurring
in this configuration allow an alternative construction with to be:
It has been proposed (Andersson 1985: 216ff) that the b-sentences in (201)—(203)
are derived from the a-sentences by inserting to be, while
Stockwell—Schachter—Partee (1973: 578) and Borkin (1973) have suggested the
opposite, namely that the a-sentences are derived from the b-sentences by a to be-
deletion transformation. Schematically, the proposed TO-BE-DEL rule operates as
follows:
Clauses with and without to be 69
PROP
NP AUX PROP
1
to be Λχ
PROP
NP PROP
χ
I
(Stockwell—Schachter—Partee 1973: 596)
In (204a) the verb find allows tofoe-deletion,but the same verb does not allow this
process to take place in (204b). Borkin argues instead for a rule which regulates the
application of to foe-deletion.
In the GB framework we cannot account for the data in (201)—(203) by positing
the existence of a to foe-deletion transformation. There are a number of reasons for
this. Firstly, the exact formulation of such a rule is bound to be problematic. For
example, in the formulation Stockwell—Schachter—Partee give (1973: 597), to in
AUX is deleted together with be under the PROP-node (see the trees above). This
means that a non-constituent is deleted. This is not allowed.51 Secondly, the rule
is construction-specific. In current generative theory there has been a move towards
deriving the properties of syntactic phenomena from general principles. To foe-
deletion has no place in such a framework. Thirdly, the to foe-deletion
transformation frequently changes the meaning of the sentence. Indeed, the whole
70 Small Clauses as constituents
point of Borkin's 1973 paper is to show what are the semantic effects of the
transformation. Finally, positing a to be-deletion transformation amounts to saying
that all sentences without to be containing consider or appoint-type verbs are
derived from sentences which originally did contain to be. It is not at all clear that
this should be the case (cf. (202 a/b)).
The problems noted here are a sufficient justification for rejecting a to be-deletion
rule. The correct treatment would seem to be to say that verbs which subcategorise
for clauses are marked in the lexicon for the type of clause(s) they subcategorise.
Thus a verb like consider in (201) can take ίο-infinitival clauses, ίΑαί-clauses and
Small Clauses as complements. Similarly, a verb like find in (204) subcategorises
either a Small Clause or a ίο-infinitival clause. This account is much more elegant.
We do, however, still have to explain why it is that in (204) I find Sam appealing
is fine, whereas ?I find Sam amusing to Charlie is not. In both cases the verb
subcategorises for an adjectival Small Clause. It seems reasonable to say that
syntactically the sentence I find Sam amusing to Charlie is well-formed, but that
semantic/pragmatic principles rule it out as being situationally anomalous. What sort
of semantic/pragmatic principles can we invoke here? Although some of the main
tenets of Borkin's paper should be rejected, she does make a useful suggestion with
regard to this matter. Her claim is that the conditions under which tofce-lessclauses
(i.e. SCs) and tofee-clausesare likely to occur are the following: Small Clauses are
more appropriate if the proposition expresses a situation which is related to personal
experience, whereas to fte-clauses express "an empirically oriented or discourse
given proposition." (1973:44). To illustrate, consider (205) and (206) (Borkin's (10
b/c)):
As Borkin suggests, (205) might be used by a speaker who has done a number of
tests on the chair in question, whereas (206) expresses a personal opinion. Now,
consider again the sentences in (204b), repeated here as (207 a/b):
3.5. Conclusion
4.1. Introduction
Apart from prototypical Small Clauses like (208) there are a number of further
structures that could be said to contain prepositional Small Clauses.
The first of the constructions I will be dealing with involves what have
traditionally been called phrasal verbs, such as those contained in sentences (209)
and (210):
In the next section I will show that in sentence (209) the postverbal NP and
(intransitive) PP taken together form a Small Clause constituent Sentence (210),
on the other hand, in which there is no subject-predicate relationship between the
postverbal NP and the (intransitive) PP, will be argued not to contain a Small
Clause.
In sections 4.3—4.6 I will discuss a number of constructions which at first sight
could also be said to contain prepositional Small Clauses. They are listed below:
In all these sentences, except (214a), the postverbal NP acts as the subject of what
74 Special constructions
comes after the elements from, as, for and 0/respectively. We might want to say
that (211)—(214) contain prepositional SCs, where in each case the postverbal NP
is the subject and what comes after the preposition the predicate. A question that
will be addressed in this chapter is whether there are arguments which support such
an analysis, or whether some other treatment is perhaps to be preferred. We will see
that in most of these cases the elements which I have so far called prepositions
should not be analysed as such. I will argue that the constructions in (211)—(213)
do not contain prepositional Small Clauses. Instead, they will be analysed as ECM
(Exceptional Case Marking) constructions (see chapter two) in the case of (211a),
(212b) and (213a/b), and as object control structures in the case of (211b) and
(212d). The elements from, as and for are taken to be inflectional elements. The
sentences in (212a) and (212c) will be analysed as involving nominal Small
Clauses, whereas the constructions in (214) will be argued to contain not Small
Clauses but NP and PP (or perhaps NP and CP) complements.
4.2.1. Introduction
Here I will be concerned only with constructions of the type in (215) and (216). In
the next section I will give a brief overview of previous theoretical treatments of
the verb-preposition construction, concentrating on three important Government-
Binding Theory treatments. In section 4.2.3 I will outline arguments which support
a different, and in my view, more adequate and elegant analysis of this type of
construction in GB terms. I will present this analysis in section 4.2.4. For the time
being the neutral term particle will be used for the final elements in (215) and
(216), though below they will be taken to be prepositions.
Verb-preposition constructions and Small Clauses 75
4.2.2. Background
In this section I will mainly be concerned with some of the theoretical proposals
that have been put forward in analysing the verb-preposition construction. The more
descriptive treatments (such as e.g. Bolinger 1971 and Dixon 1982), although they
provide large quantities of interesting data and potentially useful ideas, are not
aimed at explaining the phenomena in question. I have made some use of the very
wide range of material found in these studies.
Authors working in the earlier generative frameworks of the Standard Theory and
the Extended Standard Theory dealt with the verb-preposition construction by
making use of a rich transformational apparatus (see e.g. Chomsky 1957; Fräser
1974 and Emonds 1972, 1976). Some of these linguists have suggested that these
constructions involve particle movement. Thus, Chomsky assumed that at Deep
Structure the particle was adjacent to the verb, and that movement to the right
across the NP yielded the alternative configuration. This rule was obligatory if the
NP was a pronoun (see e.g. Chomsky 1957: 75—76 and 1964: 228). In Emonds'
analysis (1976: 82) the NP is adjacent to the verb at Deep Structure. Leftward
movement of the particle then derives the alternative order. Recent GB work has
not paid a great deal of attention to the verb-preposition construction. The most
important studies are Kayne (1984b), which provides a detailed analysis, and the
discussions in Radford (1988b) and Stowell (1981). Let us look at these studies in
a little more detail, starting with the latter.
Stowell assumes (1981: 296ff) that the particle in sentences such as (218),
The main motivation for this analysis is the Case Adjacency Principle (1981: 113)
which requires that for an NP to be assigned Case, it must be adjacent to the verb.
In (219) the NP the light is adjacent to the complex verb after application of the
rule of Particle Incorporation. Stowell accounts for structures such as (215), where
the particle appears to the right of the NP, as follows: first the word formation rule
of NP Incorporation applies, resulting in the creation of the complex verb switched-
the light; then the rule of Particle Incorporation applies to the output of this process.
The resulting S-Structure is (220):
The NP the light has the status of an incorporated object. Below I will show that
there are compelling arguments against this treatment
Radford (1988b: 90ff) argues for a structure like that in (221) for the sentences
in (215) and (216):
(221) S
V NP PP
I I I
I switched the light off
I looked the information up
(224) S
V Ρ
In this structure switch off and look up are complex verbs in which the prepositions
off and up can be regarded as word-level adjuncts (1988b: 257). With regard to the
relation between (221) and (224) Radford remarks (p.c.) "[t]o the extent that I
envisage any rule relating the two, it's one in which the Ρ originates as part of the
PP, but is incorporated into the V by REANALYSIS". Below I will argue against
positing the existence of such a rule.
Kayne (1984b) attempts to account for the constructions under investigation in
his binary branching model (see Kayne 1984a). He argues for the analysis of (215)
and (216) as in (225):
Verb-preposition constructions and Small Clauses 77
The verb subcategorises for a Small Clause which is headed by the particle and
whose subject is the NP. Kayne makes no syntactic distinction between (215) and
(216). He does remark, however, that in (215) the particle expresses a result
(1984b: 121), whereas this is not the case for the particle in sentence (216), which
is said to belong to the class of verb-preposition constructions that have "an idioma-
tic character" (1984b: 124).53 In Kayne's framework structures such as (222) and
(223), repeated here as (226) and (227),
in which the particles appear in a position adjacent to the veib, are derived by
moving the NPs to the right and by adjoining them to V'.54 For similar treatments
see Beukema—Verheijen (1987) and Hoekstra (1988).
In what follows I will adopt this rightward movement analysis. However, my
treatment will be different in two respects. Firstly, the displaced NPs adjoin to VP.
Secondly, and more importantly, I will be arguing that the Small Clause analysis
of the constructions under investigation is warranted only for verb-preposition
constructions such as (215), where there is a genuine subject-predicate relation
between the NP and the particle, but not for those of the type in (216), where no
such relationship holds. I will show that the semantic difference between spatial-
resultative constructions such as (215) and idiomatic constructions such as (216) is
paralleled by the different syntactic behaviour of these two constructions. Thus, the
claim here is that there are two distinct classes of verb-preposition construction.
I will call verbs such as switch in (215), which I will argue take SC
complements, Α-Verbs, and I will use the label B-Verbs for verbs like look up in
(216), which do not subcategorise for clausal complements.
The first argument that supports the distinction between Α-verbs and B-verbs
concerns the fact that only the [NP + particle] complements of A-verbs can occur
elsewhere as complements. Thus in (228)—(230) below such sequences occur as
the objects of prepositions in what van Riemsdijk has called absolute prepositional
phrases (see van Riemsdijk (1978), although note that for him they do not involve
clauses).
(228) He propped the bonnet of the car up; with the bonnet up he then drove o f f .
78 Special constructions
(229) Sally pushed the lever on the amplifier down; with the lever down her CD-
player was pre-programmed.
(230) Jim turned the radio o f f ; with the radio off he could finally relax.
(231) *He brought the kids up by himself; with the kids up he could go on holiday.
(232) *My teacher always puts his pupils down; with his pupils down he feels
superior.
(233) *Jim sold the car off to a friend; with the car off he could buy a boat.
Notice that the [NP + particle] sequence may also occur, though perhaps only
marginally for some speakers, after the comparative prepositions than and as:
(234) a. The oven off is less dangerous than the oven on.
b. The oven off is as dangerous as the oven on.
c. The ovens off is as at least as dangerous as the ovens on.
These facts clearly suggest that in (228)—(230) and in (234) the elements following
the prepositions form a constituent, whereas the elements in (231)—(233) do not.55
The sentences in (234) show up another interesting property of the [NP +
particle] complements to A-verbs, namely their ability to appear in subject position.
The inability of the [NP+particle] strings of B-verb constructions to do the same,
cf. (235) and (236), is a second argument in favour of analysing Α-verbs and B-
verbs differently.
— Case
However, this account is not without problems. Nigel Vincent has drawn my
attention to the danger that positing an abstract prepositional complementiser in
(237) (and in (239)/(240) below) leads to a circular argument: Why do the SC
subjects have Case? Because there is a prepositional complementiser. What is the
evidence for such an element? The fact that the SC subject must be assigned Case.
The analysis given here, however, is at least in part motivated in that sentences like
(241) below show that positing a complementiser position in this type of SC is
necessary on independent grounds. We could also argue that the main clause "I"
assigns Case to the SC subject. However, as one JL referee points out, this is
unlikely because objective Case is expected in this position (cf. Her out is a
problem / *She out is a problem). In any case, whatever the exact mechanism
assigning Case to the subject of the SCs in (237) and in (239)/(240), the main
concern here is demonstrating the fact that Α-verbs differ from B-verbs. I will have
more to say about Case assignment to SCs in subject position and to independent
SCs in chapter six.
As for the thematic properties of these constructions, the matrix clause predicate
does not assign a θ-role to the subject of the SC, but to the SC as a whole. As in
standard analyses, the predicate of the clause, in this case the particle, is in-
strumental in assigning a θ-role to the Small Clause subject.
[NP + particle] strings may also occur as complements in copular constructions:
(239) The room was extremely noisy: children shouting, the tv on, the record player
on, and little Jimmy kicking the cat.
(240) Hands up!58
(241) The thief, while out, swore never to end up in jail again.
80 Special constructions
(242) Mel and Kim were watching television in the dark when suddenly Rick burst
in; he switched the lights on and the tv off.
(243) It is difficult to arrange this furniture; let me see: I want the couch here and
the table there.
Given the standard assumptions regarding coordination, namely that units that can
be coordinated are constituents, both (242) and (243) offer strong empirical support
for the claim that the highlighted sequences are constituents. As there is a subject-
predicate relation between the NP and the particle, we must conclude that the units
in question are clauses.59
The behaviour of B-verbs is again different. For these verbs coordination of [NP
+ particle] sequences is impossible:
The ungrammatically of (244) and (245) suggests that the highlighted strings are
not constituents.
The semantic and syntactic evidence which has been accumulated in favour of
drawing a distinction between A-veibs and B-verbs suggests the following syntactic
structures for the relevant VPs (V' is irrelevant here and has been omitted):
Verb-preposition constructions and Small Clauses 81
VP
V IP (=SC)
NP PP
VP
V NP PP
The fact that (248) and (249) are well-formed constitutes empirical support for the
claim that the element off in these sentences is an independent maximal projection.
Right and completely are P'-specifiers in the structure [PP Spec [p. P]]. It might be
objected that although modification of the prepositional phrase is possible in [V NP
PP] configurations, it is not possible in [V PP NP] sequences, as is shown in
sentences (250) and (251):
These sentences seem to suggest that in these configurations the preposition cannot
be a maximal projection. Radford has taken data such as these to be evidence for
his claim that the preposition is incorporated in the verb, thus forming a complex
verb through Reanalysis. We will see below, however, that the ill-formedness of
these sentences can be explained without recourse to a Reanalysis rule.
Let us now turn to an account of (222) and (223), repeated here as (252) and
(253):
I propose that in both cases we have rightward movement of the NP deriving (254)
and (255) from (246) and (247) respectively:
(254) VP
VP NP;
V IP (=SC)
NP PP
(255) VP
VP NPj
V NP PP
Verb-preposition constructions and Small Clauses 83
Kayne, in dealing with such data, suggests that pronouns are "lighter" than other NP
types. He proposes the following condition:
(258) In ...[e], X NP;..., where NP, binds [e]i; NP, must be at least as heavily
weighted as X.
Because particles are "heavier" than pronouns, (258) prevents pronouns from
appearing to their right. Although this proposal is intuitively very appealing, as
pronouns do appear to behave idiosyncratically,64 it seems that because (258) is
rather complex it is implausible that it is part of a child's mental make-up.
In view of these considerations I propose to modify (258) in such a way that it
has the status of a condition on derivations resulting after rightward movement of
maximal projections:
84 Special constructions
The weightings are as follows: heavy XP: 2, regular XP: 1, light XP: 0. Of course,
the specified weightings are to be regarded as relative, not as absolute, values. We
may regard XPs that contain a PP or a clause as being heavy (cf. Whitney 1982:
299), and XPs that contain only a head, i.e. that have the structure in (260), as
being light.
(260) XP
I
X'
I
X
We are now in a position to explain why (256) and (257) are ill-formed. In both
sentences the pronominal NPs and the intransitive PPs have the structure in (260).
This means that they are light maximal projections. In these sentences the displaced
pronominal NPs appear in a position to the right of intransitive PPs which have the
same weight; (259) rules these structures out. In formulating (259) as above I am
proposing that movement of the NP to the right, as an instance of Move a, is
always a possible option. The condition, however, filters out some of the resulting
derivations. Because structures like (256) and (257), which result after movement
of the NP, have not violated any D-Structure or S-Structure principles, it is
reasonable to assume that (259) operates at the PF level.
It is to be expected that certain processes at this level affect the weight of the
pronoun. As has been observed in the literature, if the pronoun is appropriately
stressed, it can occur in final position, as in (261):
Because the pronoun carries heavy stress in this sentence its weight has increased
so that its appearance to the right of the PP does not violate (259).
Note that light pronominal NPs also increase in weight when they take
premodifying or postmodifying elements. Thus, (262) and (263) are well-formed
because the pronominal NPs, now heavier because of the added postmodifiers,
appear to the right of a light intransitive PP after movement:
(268) I cut ei [PP right o f f ] [NP all the branches that were keeping out the light],
(269) He switched e, [PP completely o f f ] [NP the radio that had been making funny
noises all the time],
However, although this is a possible derivation, more needs to be said here. I would
like to maintain that there is nothing that can block Reanalysis operating on
(266)/(267) in Radford's account along the lines suggested. To see this consider
again (215)/(252) and (216)/(253), repeated here as (270a/b) and (271a/b):
In Radford's analysis the relation between the a and b-sentences in (270) and (271)
is not mediated by movement. Instead, it is necessary that we "assume that the base
rules allow either the order V NP PP, or V PP NP, and that with the latter order
restructuring is required to avoid violating adjacency". (Radford, p.c.) Thus, in order
to avoid a violation of the Case Filter the PPs off and up in (270b) and (271b) are
incorporated into the verb. But if Radford is right, there is nothing to stop his
grammar from generating the structures in (266) and (267) directly. Reanalysis
could then apply, deriving the highly implausible complex verbs cut right off and
switch completely o f f . Thus my claim stands.
86 Special constructions
Radford (p.c.) also mentions (272) as evidence that incorporation is not possible
in (266)/(267):
(272) He cut — off the tree [all the branches that were withering]
It is not clear how this example is relevant. Here again, his analysis would allow
us to base-generate the structure [VP V PP NP], Reanalysis would again apply,
implausibly deriving the complex verb cut off the tree. If, instead, off the tree is
taken to be an adjunct, as in the structure [VP [Vp cut NP] off the tree]], then (272)
is simply an instance of Heavy-NP-Shift.
Radford (p.c.) also cites the sentences in (273) and (274) and claims that they are
a problem for the present analysis:
(273) *John will cut off the branches and Mary off the leaves.
(274) John will cut the branches off and Mary the leaves o f f .
(275) *John will cut off the branches and Mary ΨΠ1 Mi off the leaves.
This breaks the chain t — the leaves, leaving the NP the leaves uninterpretable as
a moved element. To the extent that sentence (274) is OK this is so presumably
because deleting the modal and the head veib (two separate constituents) is allowed.
This explanation, however, is also available for (274) in my framework where the
same elements get deleted.
The sentence in (276) is also cited by Radford:
(276) He will [cut off and burn] all the dead branches.
To the extent that coordination evidence is reliable, (276) seems to suggest that cut
off is a complex verb. However, it is not inconceivable that the VP in (276) has the
following structure [VP [VP cut f, off and burn tj] all the dead branchesy]]. Here we
would have to assume that one instance of the moved NPs is deleted under identity.
What does at first sight seem to be problematic for the present account is (277),
from Radford (1988b: 98):
(277) Drunks would put off the customers, and ΨόύΙά ptU i f f the waitress.
This sentence shows, Radford claims, that put off is to be regarded as a complex
verb which is deleted along with the modal (gapping). This sentence cannot be
explained by assuming deletion of the modal and the VP put t o f f , as that would
Verb-preposition constructions and Small Clauses 87
break the chain t — the waitress (cf. (275) above). However, one might question
whether the surface structure of (277) is indeed the result of gapping. (277) is
difficult, if not impossible, to distinguish from a structure in which the string the
customers and the waitress is a coordinated NP, even if there is a slight pause after
and.
Apart from the Reanalysis problem there is a further difficulty for an account
such as Radford's, and that is that there is no way, other than by stipulation, of
explaining the ungrammaticality of sentences like (256) and (257).
The present analysis has the added advantage that it can also account for other
types of Heavy-NP-Shift. Consider (278)—(281):
In each of these cases the Small Clause subject has moved to the right and has been
adjoined to VP, leaving behind a Case-marked, θ-marked trace. The resulting
structures do not violate (259) because the heavy NPs appear to the right of a
lighter regular XP.66 Notice that (259) is operative in preventing pronouns from
appearing to the right of Small Clause predicates:
In (286) the NP the student is syntactically heavier than the NP winner. (259)
wrongly predicts that this constituent may appear to the right of winner. In (287)
it should not be possible for the light NP idioms to appear to the right of the
equally light PP up. It would seem that the "bare NPs" winner and idioms in these
sentences are weighted like regular NPs, contrary to what is predicted by (260). If
this is indeed the case then (286) is ruled out, as it should be, and (287) is correctly
not ruled out. It is not entirely clear what should make these NPs heavier, but a
plausible possibility is that the NPs winner and idioms in (286) and (287) carry
extra "informational weight". Their natural position is then at the end of the
88 Special constructions
Although the superficial structures of these sentences are similar, (288) and (289)
differ crucially from (290) in that they involve a subject-predicate relation between
Jim and a fool in (288) and between those files and back in (289). These sentences
appear to have the following D-structures:
Note that the existence of she made out that Jim was a fool supports the analysis
of (291).
We derive (288)—(290) by moving the rightmost NPs in each case and by
adjoining them to VP:
Notice that in (294) the predicate of the SC has moved, whereas in (295) it is the
SC subject which is adjoined to VP. In (296) the displaced NP is a secondary
object. As before, the movements to the right are instances of Heavy-NP-Shift For
some reason, for which I have no explanation, movement of the NP a fool in (291)
is obligatory because the D-Structure in which it occurs cannot appear as a surface
form. (292) and (293), on the other hand, are acceptable as surface structures, and
Heavy-NP-Shift is optional, as expected.68
Consider next (297)—(299):
ruled out because in order to derive it from (291) the whole SC would have to be
moved to the right, adjoining to VP. In that position neither the subject nor the
predicate of the Small Clause would be Case-marked, thus violating the Case Filter.
The alternative, movement of the prepositional phrase out to the left, also results
in a violation of the Case Filter in that again neither Jim nor a fool would be
assigned Case in the resulting structure (if we assume that Case is assigned under
Adjacency).69 Furthermore, such a process would not be possible because it is
neither a case of Substitution, nor of Adjunction, the only two permissible types of
movement (cf. Chomsky 1986b: 4).
The two ways of deriving (294) from (298) are also illicit Movement of Jim
those files to the right involves movement of a non-constituent Furthermore, the
same problem as above with regard to the Case-marking of the displaced NPs
would obtain. Displacement of the prepositional phrase back to the left results in
a violation of the Case Filter because neither Jim nor those files would be adjacent
to the verb. Furthermore, as above, such movement is prohibited because it involves
neither Substitution nor Adjunction.
Finally, (299) cannot be derived from (293) for the same reasons: movement of
Jim a note is impossible because this string is not a constituent and because in the
derived structure these NPs would not be Case-mariced. Movement of out to the left
within VP is illicit, again for Case theoretical reasons, as above. Furthermore, as
in the discussion of (297) and (298), such movement does not involve Substitution
or Adjunction.
4.2.5. Conclusion
I have shown in this section that there is semantic as well as syntactic evidence for
making a distinction between two types of verb-preposition construction. I have
distinguished Α-verbs, which subcategorise for SCs, from B-verbs, which sub-
categorise for an NP and a PP complement. We can view B-verbs as transitive
prepositional verbs. Intransitive prepositional verbs comprise on the one hand verbs
of the type encountered in (217) above where the verb subcategorises for a
transitive PP only, and on the other hand verbs such as give (up) in a sentence like
I give up where we have an intransitive prepositional phrase complementing the
verb. So-called phrasal verbs do not exist. The analysis presented here, which posits
rightward movement of the NP and Adjunction to VP in accounting for the
alternations in (215)/(252) and (216)/(253), in conjunction with the condition on
derivations, (259), provides a principled account of verb-preposition constructions
in English.
90 Special constructions
4.3.1. Introduction
In this section I would like to deal with a relatively widely discussed but fairly
infrequently occurring group of verbs, the so-called verbs of negative causation
(Postal 1974). This group comprises verbs such as prevent (from), keep (from),
discourage (from) etc. The constructions in which these verbs occur have puzzled
linguists, as their syntactic properties are far from clear. In what follows I will
identify the problem areas and propose possible solutions.
As a starting point, consider the basic structural configurations that prevent, the
most typical of the verbs of negative causation, may occur in:
(305) a. The sailor was prevented from drowning the cat (by Jim).
b. *The sailor was prevented drowning the cat (by Jim).
(306) a. The cat was prevented from being drowned (by Jim).
b. *The cat was prevented being drowned (by Jim).
(307) Jim prevented the cat from being drowned (by the sailor).
(308) Jim prevented the cat being drowned (by the sailor).
With regard to these constructions we need to address the following issues: what
is the exact functional status of the postverbal NP in prevmi-constructions,
especially those in (303) and (304)? Is it a direct object or the subject of a
complement clause? And, if we do have a complement clause, is it a Small Clause?
What about from? Is it a preposition, a complementiser, or perhaps an inflectional
element? And is the construction with from (as in (303)) structurally the same as
that without from (as in (304))? Finally, how can we account for the passives of
sentences containing prevent-type verbs? Why, for example, are (305a), (306a),
(307) and (308) fine, but not (305b) and (306b)? I wül be discussing each of these
problems below and will then turn to to a discussion discourage-type verbs which
will be argued to be different from prevent-type verbs.
Verbs of negative causation 91
Let us start with (300) and (301) of the previous section. Their analysis presents no
particular problems. In each case prevent functions as a dyadic predicate, i.e. as a
predicate taking a subject argument and a direct object argument. This is
uncontroversial. Notice that the direct object must express a situation, or proposition
(in a wide sense of that term). This explains why we can have she prevented the
accident/ the arrest/ the war etc, but not she prevented John/ the mayor etc.71 The
significance of these facts will become apparent below. Sentence (302) would also
appear to involve a nominal direct object, namely Millie's (always) leaving the
house. This string functions like an NP in more than one way in that, firstly, it
contains the genitive construction, which is characteristic of NPs, and, secondly, it
may occur in other typical NP slots such as subject position and prepositional
complement position. It is therefore reasonable to analyse (302) on a par with (300)
and (301).72
While (300)—(302) are straightforward, (303) and (304) are not. There is little
agreement in the literature about the function of the NP following prevent-type
verbs in [V NP (from) -ing] constructions. Jespersen (1933: 326) regards it as a
direct object in the case of (303), but as the subject of the complement the sailor
drowning the cat in (304). Quirk et al. (1985: 1209—1211) take the NP in question
to be an indirect object in (303), but do not mention (304). A third possible analysis
is found in the transformational-generative literature of the late sixties and early
seventies. Rosenbaum (1967) and Postal (1974) have both argued that prevent-
constructions involve the rule of Subject-Raising-to-Object (henceforth "Raising"),
that is, a rule which moves the postverbal NP from the subject position of a
complement clause at Deep Structure to the direct object position of the matrix
clause at Surface Structure (cf. (7) above). Joan Bresnan, in an otherwise negative
review of Postal's book, remarks that "[a] Raising analysis is quite plausible for
these constructions." (Bresnan 1976: 497). She does not discuss the matter any
further though. In Postal—Pullum (1988) Postal's original claim that prevent-
constructions involve Raising is reasserted. McCawley (1988: 130) also opts for a
Raising analysis and makes the additional assumption that from is not present at D-
Structure but is inserted at the time Raising-to-Object applies. Fourthly, Chomsky
(1981: 147, note 108), who discusses prevent only fleetingly, seems to hold the
view that the postverbal NP is a direct object which is assigned a thematic role (Θ-
role) by the matrix verb. Finally, Kayne (1984a: 155) assigns prevenf-constructions
the structure V [s ΝΡ from S']. He calls prevent an object control verb, despite the
fact that the postverbal NP is regarded as a subject.
In what follows I will demonstrate that each of the analyses discussed above is
problematic in one way or another and I will argue instead that preveni-structures
of the type in (303) and (304) involve a complement clause headed by from which
is assigned a θ-role by prevent and whose subject is the NP immediately following
that verb.
92 Special constructions
The evidence for this position is twofold. First, let us consider the behaviour of
so-called pleonastic or dummy elements and idiomatic phrases in prevent-
constructions. Observe that these can occupy the NP position following the matrix
verb:
Existential there and nonreferential it in (309) and (310) are elements which can
only occur in θ'-positions (i.e. positions to which no θ-role is assigned). The same
is true for the idiom chunks in (311) and (312). This being so, (309)—(312) show
that the postverbal NP is not a direct object argument, but rather the subject of a
complement clause which is assigned a θ-role by prevent.
The sentences above are not wholly uncontroversial and have been subject to
some debate. (311) and (312) are sometimes regarded as less acceptable than (309)
and (310), but even the latter pair have been regarded as dubious by some.
Chomsky comments on the sentence they prevented it from raining as follows:
quite apart from its dubious status, it is difficult to see how any arguments can
be based on it, since the rules for generating it would appear to be
idiosyncratic, even if it is accepted as grammatical.
(Chomsky 1981: 147, note 108)
Two comments are relevant. First, nothing about [this sentence] is dubious. It
is perfectly grammatical for everyone we know of, and belongs to an example
category that has long been noted in the grammatical literature...Second, and
more important,...the allegedly "idiosyncratic" character of the rules defining
structures with verbs like prevent is, as a matter of logic, irrelevant to the key
issue, which is the existence of (even a single example of) a sentence type for
which the optimal account is a Raising-to-Object analysis.
(Postal—Pullum 1988:655)
Chomsky has suggested that his sentence is derivatively generated, but this is of
course entirely ad hoc. I would agree with Postal—Pullum that the sentence is
perfectly grammatical, at least for some speakers. The differing judgements may be
due to dialectal variation.
As a second piece of evidence in favour of the claim that prevent subcategorises
a complement clause consider the sentence in (313) below and notice that it is
Verbs of negative causation 93
(313) The chairperson prevented [the committee from taking the wrong decision]
(314) The chairperson prevented [the wrong decision from being taken by the
committee]
The fact that (313) and (314) are synonymous shows that the postverbal NP in both
cases is not an argument of the matrix verb and is therefore not assigned a θ-role
by it. If this is so the NP in question cannot be regarded as a direct object and must
be the subject of a complement clause.
As with the first set of data some further comments are in order here. Notice that
although the synonymy judgements in (313) and (314) are undisputed, it is possible
to construct sentences containing prevent, which, when passivised, appear to acquire
a different meaning. Thus, Andrew Radford (p.c.) claims that the pair of sentences
in (315) and (316) are not propositionally equivalent (see also Andersson 1985:
96—97):
To the extent that somebody is being acted upon it does indeed seem to be the case
that there is a direct thematic relation between prevent and the NPs that follow it,
i.e. the doctor in (315), and the patient in (316). These sentences thus appear to
contradict the claim that prevent-type verbs subcategorise for a clausal complement,
and not for an NP object If this is so, then we are faced with a conflict between
the predictions made by the synonymy judgements in (315) and (316) and the
evidence regarding dummy elements and idioms which denies the existence of a
predicate-argument relationship between the matrix verb and the postverbal NP in
these sentences. I believe this conflict is only an apparent one which can be
resolved by appealing to extra-grammatical circumstances. Susan Schmerling has
discussed the same phenomenon in constructions containing the verb allow
(Schmerling 1978). She reports that informants who are asked to judge whether the
pair of sentences I allowed the doctor to examine John and I allowed John to be
examined by the doctor are synonymous generally answer in the negative and that
these judgements conflict with the fact that allow, like prevent, can be followed by
elements like it and there (cf. I allowed there to be an investigation). Schmerling
explains this conflict by claiming that trutb-conditionally the sentences above are
equivalent and that any feeling of non-synonymity is brought about by pragmatic
factors. She observes that informants in performing synonymy judgements for
α/tovv-constructions tend to
opt for the most complex possible interpretation of a sentence. This means
that...informants will decide, in the absence of evidence to the contrary, on the
94 Special constructions
In other words, if the context allows it, the postverbal NP is pragmatically taken to
be an argument of the matrix verb. See also McCawley (1988: 150, note 12). I
propose to extend this account to the preve/tf-constructions in (315) and (316) in
order to explain how the contrast in interpretation of the argument status of the
postverbal NP in these sentences and those in (313) and (314) comes about. Thus
in (315) and (316) the fact that the postverbal NPs are felt to be arguments of the
matrix verb is a consequence of our contrasting the two sentences and creating a
context in which somebody can be prevented from doing something. We cannot do
the same in the case of (313) and (314) because in (314) the postverbal NP is not
an entity that can be prevented from doing something.
We could reformulate Schmerling's proposals in slightly different terms by
observing that (315) and (316), but not (313) and (314), have the flavour of what
are known as garden path sentences. These are sentences that are initially processed
in one way but on closer examination require a different analysis. This also appears
to be what is going on when processing (315) and (316): although it may initially
seem to be the case that the postverbal NPs are arguments of the verb prevent by
virtue of their being adjacent to this verb, it turns out that what is really prevented
is a situation, not an individual.73
From what has been said so far we must conclude that any analysis of prevent-
constructions in which the postverbal NP is taken to be an argument of the matrix
verb must be incorrect. The standard theoretical rule of Raising-to-Object is
therefore also excluded. There are incidentally also objections of a general
theoretical nature against a Raising account of preve/if-constructions in that this rule
violates both the Theta Criterion and the Projection Principle (see section 2.3.1). To
see why this is so, consider (317) below in which it is shown schematically how
Raising would apply in (303):
θ θ
ι 1 I =•
(317) Jim prevented [NP e][the sailor from drowning the cat]
1 I
Raising-to-Object
Verbs of negative causation 95
The bottom arrow shows how the subject of the complement clause is raised to the
object position in the matrix clause. This operation violates the Projection Principle
because the D-Structure subcategorisation properties of prevent do not remain
constant at all levels of representation. The Theta Criterion is violated because the
chain [NP e] — the sailor is assigned two θ-roles, one by the VP predicate of the
complement clause drowning the cat, and one by the matrix verb, as indicated by
the top two arrows.
The discussion above had led us to the conclusion that prevent-type verbs
occurring in constructions like (303) and (304) subcategorise for a clausal
complement of the form [NP (from) -ing]. The question now arises how the internal
structure of this complement is to be analysed. To answer this question we need
first to consider the status of from.
Before discussing the formal status of from I would like to make a few observations
about the history of preveni-constructions which will be seen to be relevant in the
ensuing discussion. Constructions containing the string [VNP from -ing] are very
old. Some of the earliest attested examples are cited in Visser (1973: 2370f): 74
(318) tat swote smirles...wit pat deade licome pat is ter-wid ismittet from rotunge
(cl200, in Hali Meidenhad, F.J. Furnival (ed), 1922, Early English Text
Society (henceforth EETS) 18, ρ 171)
'that sweet ointment prevents (wit) that dead body that is there-with rubbed
from rotting' 75
(319) ellis he my te not walke euenly in pe myddis of pe cawcee and keping hym
fro falling into pe watir... (c 1434, in Pecock, Reule of Crysten Religioun,
W.C. Greet (ed), 1927, EETS 217, p295)
'else he might not walk evenly in the middle of the causeway and keep him
from falling into the water...'
(320) pe seouede dole is of...hwat binimeö us his luuve & let us to luuien him
(C1225, Ancrene Riwle, Mabel Day (ed) EETS 1952, 6,25)
'the seventh part is about...what deprives us of His love and prevents (let) us
from loving him'
(321) Ah who shall hinder me to wail and weep (1594, Shakespeare, Richard III,
II, ii, 27)
This second construction, which, unlike the first, has not survived to the present,
behaves syntactically just like expect and believe in Modern English; that is, the
matrix verb subcategorises for a ίο-infinitival clause with an overt subject. In
Government Binding theory terms, the verbs in the second preve/tf-construction are
ECM (Exceptional Case Marking) verbs. These are verbs which subcategorise for
a clausal IP complement and assign Case directly to the subject of that complement.
For ease of exposition, let's refer to prevent-type verbs in [V NP from -ing]
constructions as prevent, and to prevent- verbs in [V NP to-infinitive] constructions
as prevent2. We might wonder whether we should perhaps posit the existence of a
third verb prevent, namely the one that occurs in (304), i.e. a structure without
from. There are reasons for not doing so. I will discuss this matter further in section
4.3.4. Let us now return to the question of the syntactic status of from.
It would be natural to consider first the possibility that from is a preposition.
After all, the English language does possess a lexical item from which is clearly a
preposition. Having argued that prevent takes a clausal complement in (303) this
sentence would then have to be analysed along the lines shown in (322):
(322) VP
In this phrase maiker the NP the sailor is regarded as the subject and the PP from
drowning the cat as the predicate of the clausal complement of prevent. However,
this analysis is semantically odd as it is clearly not the case that there is a subject-
predicate relationship between the NP and the PP, unlike in some other [V NP PP]
constructions, such as I want you in the car. We obviously do have such a
relationship between the strings the sailor and drowning the cat, but that leaves the
element from unaccounted for. Another objection to the analysis in (322) is that, if
Verbs of negative causation 97
accepted, we are bard put to explain why, if from is indeed a preposition, it cannot
normally take a regular NP as its complement, witness (323):
(324) VP
(c) From can optionally be omitted, as sentence (304) above shows. This is
also true for a number of complementisers (cf. I believe (that) she is
marvellous/ I want (for) her to leave), therefore from is likely to be a
complementiser also.
Under closer scrutiny these arguments turn out to be rather weak. As for the first
argument above, it is in fact unlikely that complementisers do not carry meaning.
If this were true the sentences in (325) below would be predicted to be
synonymous, as they differ only with regard to the choice of complementiser. They
are clearly not synonymous.
The second argument really only shows that from is a head, as Andrew Radford has
pointed out to me, and does therefore not exclude analysing this element as a
preposition or as an inflectional element. Finally, with regard to the third argument,
notice that it is not just complementisers that behave idiosyncratically in the way
illustrated above, as (326) shows:
Here the inflectional element to occurs in the passive, but not in the active. I will
return to sentences like (326) below.
Apart from these general observations, there are a number of further problems
with the analysis in (324) which make it unlikely that from is indeed a
complementiser positioned in "C". The most serious problem concerns the sentence
in (327):
The S-Structure of this sentence would be as in (328) on page 99. As the phrase
marker shows, the NP the sailor has moved from the lower Spec-of-IP (a subject
position) to Spec-of-CP of the complement clause. However, notice that once
this movement has taken place it is no longer possible for the Wh-element what
to move successive cyclically from its D-S tincture position following the verb
drown via Spec-of-CP of the lower clause to Spec-of-CP of the matrix clause.
The movement is blocked because the Spec-of-CP of the complement clause is
already filled by the NP the sailor. In the analysis proposed in Aarts (1990b)
there is thus no way for the grammar to generate sentences like (327) licitly.
Verbs of negative causation 99
NP I VP
-tns]
[+AGR] V NP
What, did Jim prevent sailor^ from ^ drowning t,
In this phrase marker the verb prevent is analysed as an ECM verb. The
complement clause is an IP rather than a CP in order for Case assignment to its
subject to be possible. If this analysis is correct prevem-constructions do not
involve prepositional Small Clauses, as was suggested as a possibility in section
4.3.1, because the IP complement clause in (329) does not have a PP predicate (cf.
the discussion of (322) above).
100 Special constructions
Before turning to the evidence for (329) let me briefly discuss some of the
arguments against this structure put forward in Postal—Pullum (1988). They argue
that "[u]nder the view we would take, word sequences like you from chewing the
tobacco [in 330] are never clauses, but rather are NP—PP sequences. They
therefore will only appear where NP—PP sequences are syntactically permitted."
(Postal—Pullum 1988: 646)
One piece of evidence that is cited against a clausal analysis of you from chewing
the tobacco is the possibility of intervening parenthetic material in prevent-
constructions, as in (331):
(331) We can prevent it, I assure you, from becoming known that we are here.
(Postal—Pullum's (30a))
(332) *We can prove that, I assure you, we are here. (Postal—Pullum's (30c))
Thus the claim is that the (im)possibility of insertion of lexical material in these
sentences shows that (331) does not involve a clausal complement of the form it
from becoming known that we are here, whereas (332) does involve a clausal
complement (namely that we are here). However, Postal—Pullum's argument
becomes less compelling when we take a wider range of data into account.
Consider the following sentences:
(335) *We can prevent, I assure you, it from becoming known that we are here.
Its non-occurrence, they claim, again shows that it from becoming known that we
are here is not a clause. However, there is a straightforward explanation for the
ungrammaticality of (335) in that it violates the Case Adjacency Principle which
requires Case assigner (prevent) and Case recipient (it) to be adjacent
Returning now to (329), I considered the possibility of from being inflectional in
Verbs of negative causation 101
Aarts (1990b: 160—161), but rejected it, partly on the grounds that there appeared
to be no evidence in the syntax of English for this position. However, it turns out
that there is some such evidence. Notice, for example, that there are cases where
the behaviour offrom parallels that of the infinitival marker to, standardly analysed
as an inflectional element Thus, sentences (336) and (337) show that from, like to,
can sometimes be left out:
Now, admittedly know is exceptional in that verbs that occur in the pattern [V NP
to-infinitive] do not normally allow an alternative structure without to. There are,
however, other parallels between to and from. Consider (338) and (339):
While the active sentences in (338a) and (339a) differ from each other in that in
(338a) to cannot be present, whereas from in (339a) may, but need not be present,
the passive counterparts of the a-sentences, (338b) and (339b), are alike to the
extent that the first of these must contain to, just as the second must contain from.
The similarities in the syntactic behaviour of to and from in (336)—(339) strongly
suggest that to and from are functionally identical. If it is true that to and from are
indeed both inflectional elements, then we would expect that the structures in which
prevent, and prevent2 occur(red), namely the [V NP from -ing] and [V NP to-
infinitive] constructions, are syntactically isomorphic, i.e. that their structure is as
in (329), with to and from heading "I". There is some historical evidence which
suggests that to and from are indeed functionally the same. As noted above, the
construction with prevent2 no longer exists and it is likely that it was pushed out
of the language by the prevenfj-construction. I found some support for this
hypothesis in the following data from Visser (1973: 2370):
102 Special constructions
(340) [To kepej pe naseperles from to moche haue lykynge in swote smelles, pe
tonge from to moche haue delite in goode metes and goode sauours (cl350,
in Book of Vices and Virtue, W. Ν. Francis (ed), 1942, EETS 217, p225, 25)
'to keep the nostrils from having much liking for sweet smells, the tongue
from having much delight in good foods and good tastes'
(341) keep tham fra gifies to gi/(cl391, in How Good Wijf Taught D, Skeat 111)
(342) Or who shall let me now, On this vile bodie from to wreake my wrong
(1590—1596, in Spenser, The Faerie Queene II, 8, 28)
'Or who shall prevent me now from avenging my wrong on this vile body'
Here, in each case from and to are both present. We can explain this by regarding
these examples as instances of syntactic contamination, that is, as structures in
which two different syntactic constructions are being confused. Contamination is
more common in the lexical domain where new lexemes are sometimes coined out
of affixes and lexemes which normally occur in different combinations. For
example, the word-form irregardless is probably a contamination of irrespective
and regardless. Similarly, to telephone up is likely to be a contaminative form
which combines to phone up and to telephone. In (340)—(342) it seems that we
have contaminative constructions involving prevent, and prevent2. The fact that
constructions like these have been attested not only suggests that prevent,-
constructions have been competing with prevenf2-constructions (and that prevent,
won out), but also that to and from were felt, at least by some, to be the same type
of element78 We thus have distributional as well as historical evidence to support
the contention that from, like to, is an inflectional element.
Sentences (303) and (304) above show that from in preveni-constructions is not
always overtly present. The /ram-less construction in (304) has often been frowned
upon by prescriptivists. Visser (1973: 2352) quotes Thomas Pyles who believes that
those who say prevent him getting back next term would probably not pass the
American College Board Examination (1964: 241). And Kahn (1985: 463) observes
that the structure in question is "still considered dubious, widespread though it has
now become". It is odd that some people should feel that the from-less construction
is bad in view of the fact that it has been in use for almost three centuries now
(Visser 1973: 2352).
Verbs of negative causation 103
I think it would be uncontroversial to say that (303) and (304) do not differ in
meaning, in that they express exactly the same proposition. One might wonder,
though, whether the sentences are syntactically the same as well. There are some
indications which suggest that they are not. Notice, for example, that (303) can be
passivised as in (343a), but (304) cannot be passivised as in (343b) (cf. also (306)):
(343) a. The sailor was prevented from drowning the cat (by Jim).
b. *The sailor was prevented drowning the cat (by Jim). (=(305))
(344) She had been prevented telling me her story. (1768, Laurence Sterne, A
Sentimental Journey, quoted in the OED, vol. VIII: 1337 and in Jespersen
1909—1949: V: 150)
(345) Sedgwick is prevented joining you by misfortune in his family. (1835, William
Whewell, in Todhunter Academic Writing, quoted in the OED, vol VIII: 1337)
(346) the prisoners were stopped writing any more letters. (1895, Owen Wister,
Red Men and White, quoted in Jespersen 1909—1949: V: 150)
These quotations show that until relatively recently, at least until the beginning of
this century, it was perfectly possible to passivise /?remtf-constructions which do
not contain from. Now, what does this mean? First of all, it means that, as regards
passivisation, less than a century ago there was no difference between prevent-
constructions with from and those without from. If we cannot find any other
syntactic differences between these structures this is an argument for saying that at
the time (303) and (304) were syntactically the same. Obviously, the contrast
between (343) on the one hand, and (344)—(346) on the other also tells us that
since about the beginning of this century there has been a change of some sort in
the language such that the latter are no longer possible. The question is: what sort
of change was it, a major one or a minor one? Given the fact that it has taken place
recently I would want to argue that it was a minor change which is perhaps part of
another process. However, let us first investigate the alternative, namely the
possibility that a major restructuring has taken place such that there is now a barrier
of some sort in (347b) preventing the operation of NP-Movement:
(347) a. [e] was prevented [the sailor from drowning the cat]
t HH
b. [e] was prevented [the sailor drowning the cat]
(348) a. Which patient did they prevent from seeing the doctor?
b. *Which patient did they prevent seeing the doctor.
The grammaticality judgements are Radford's. Not everyone, however, would agree
with them. I have checked the judgements in (348) with other native speakers, and
for some at least (348b) is fine. Consider also the sentences in (349):
Both these sentences seem grammatical. The data in (348) and (349) thus call into
doubt the existence of a barrier in (347b). Also, if the complement clause node was
a barrier we would have a problem with the active counterpart of (347b) as regards
Case assignment to its postverbal NP. We have some evidence, then, that the
structure of the complements in (347a) and (347b) and their associated S-Structures
in (343a) and (343b) is the same, namely as in (329). This is further supported by
the semantics of these constructions, in that, as we have seen, (303)/(304) and
(305a)/(305b) mean the same.
However, even if I am correct in thinking that (303) and (304) are diachronically
and synchronically one and the same construction, we still have to explain how the
language has changed such that (343b) and (344)—(346) are no longer possible. A
possible explanation would be to assume that preven/-constructions always have a
from in "I", which is either overt or covert in active structures, but always overt in
passive structures. When covert we are dealing with an empty allomorph of from.
We can then say that the change in the language which has made structures like
(343b) impossible is a minimal one and consists of the loss of ommissibility of
from in the passive. Another possible explanation would be to link the
unacceptability of (343b) to a more general pragmatic restriction. Before discussing
what this restriction might look like I would like to make some further diachronic
observations.
The historical development of /?reveni-constructions parallels that of [V NP bare
infinitive] constructions almost completely. We have seen some of these parallels
already, but table 1 on page 105 gives the complete picture. From this table we see
that the current [V NP bare infinitive] construction had an earlier variant which
allowed the presence of the inflectional element to both in the active as well as in
the passive. Table 2 gives some example sentences, most of which are taken from
Jespersen (1909—1949) (cf also Mustanoja 1960, e.g. 529, 532—33).
Verbs of negative causation 105
Diachronically Synchronically
Active
P: I observed many of them to bring out their mules.
(1886, R.L. Stevenson, Kidnapped, Jespersen
1909—1949: V: 280)
+to earlier pattern
(examples of +to only)
C: She had made the little girl to suffer. (1908, F.
Norris, The Pit, Jespersen 1909—1949: V: 291)
Passive
P: No attested example found.
±lo earlier pattern
(examples of -to only)
C: I've been made scream with pain in other ways.
(1914, G.B. Shaw, The Misalliance, The Dark
Lady, Fanny's First Play, Jespersen 1909—1949:
V: 317)
Active
P: They heard him cry.
-to current pattern
C: She made tum shudder.
Passive
P: He was heard to cry.
+to current pattern
C: He was made to shudder.
These data show that the English Language changed in such a way that the earlier
possibility of having to in active perception and causative constructions was lost:
the structures with to have been attested until at least 1910—1920, but now they
are possible only without to. Also, while perception and causative constructions
both with and without to were possible in the passive, nowadays the passive
structures must take to. The parallel with the verbs of negative causation becomes
clear when we consider table 3 with examples of constructions containing these
veibs:
Active
±from earlier pattern Jim prevented the sailor (from) drowning the cat.
Passive
±from earlier pattern The sailor was prevented (from) drowning the cat.
(see also (344)—(346))
Active
±from current pattern Jim prevented the sailor (from) drowning the cat.
Passive
+from current pattern The sailor was prevented from drowning the cat.
Table 3 shows that the verbs of negative causation have developed in a way that
is similar to the development of the [V NP bare infinitive] constructions, the only
difference being that in the latter construction-type to is no longer possible in the
active, while from in active preveni-constructions still is.
If the picture I have presented above of the parallel developments of [V NP bare
infinitive] constructions and prevenr-constructions is correct, we would expect that
in the course of time the active prevenf-construction with from will disappear. This
expectation is borne out in that there already seems to be a discernable trend to
leave out from, as becomes apparent from table 4 which shows frequencies of
occurrence for prevent in a sample of written and spoken current British English:
Table 4. Frequencies of occurrence for prevent in a sample of written and spoken current
British English. The frequecies for the passive occurrences are shown in brackets.
Table 4 shows that in written English prevent occurs one and a half times more
often with than without from. In spoken English the opposite situation seems to
hold, although the frequencies are perhaps too low to be significant. The data
suggest the following "frequency of use hierarchy" with regard to from in prevent-
constructions:
If the spoken language is the greater influence on determining future language use,
as is generally assumed, then perhaps these data suggest that prevent-type
constructions are on the way out If this is so, it would seem that, judging by their
low frequencies in spoken English, the constructions with from will disappear first.
Andersson (1985: 101) observes that all preve/U-constructions without from in his
corpus occurred in British English and none in American English.
Let us now return to (343b) and the reason for its unacceptability. Above I
suggested that this may be due to a general pragmatic restriction. This may have
sounded somewhat speculative but I believe my suggestion gains some credibility
when we consider that the unacceptability of (343b) differs sharply from the
ungrammaticality of, say, (350) below:
My point is this: considering both that sentences like (343b) have been attested
recently and that despite its being odd it is perfectly comprehensible (unlike (350)),
the hypothesis that what rules out (343b) is an extra-grammatical performance
factor of some sort seems quite reasonable. This factor could then also explain why
to is required in the passive versions of [V NP bare infinitive] constructions.
Specifying what that factor might be is by no means easy. The intuitive
generalisation seems to be that the passive versions of prevent and [V NP bare
infinitive] constructions are difficult to process if the inflectional elements from and
to are not overtly present. This difficulty in processing could be due in (343b) to
the fact that we have a clause involving a stranded nonfinite V-ing phrase which
follows a matrix verb with passive morphology. In (338b) (=*They were made work
hard) a bare infinitival clause is stranded after a passive matrix verb. That sentences
involving dangling nonfinite clauses preceded by passive matrix verbs may be hard
to process, and consequently unacceptable, seems to be confirmed by the passives
of [V NP -ing] constructions. Consider (351), (352) and (353):
108 Special constructions
A sentence like (303) (=Jim prevented the sailor from drowning the cat) takes three
passives:
(354) The sailor, was prevented [ti from drowning the cat]
(355) Jim prevented [the cat, (from) being drowned t j
(356) The cat, was prevented [i, from being drowned t j
direct object position of the complement clause to the subject position of the same
clause. Notice that from is optional. In (356) we have a combination of an external
and an internal passive: the NP the cat has moved successive cyclically from the
lowest trace position via the intermediate trace position to the matrix clause subject
position. In the previous section we have already discussed one restriction on
passivisation, namely that from must be present in external passive prevent-
constructions. Further restrictions on passivisation are exhibited by (357) and (358):
In these sentences the pleonastic elements there and it have been displaced under
NP-Movement. For most speakers this results in ungrammaticality. By contrast, the
sentences in (359)—(361) are acceptable, or, at worst, marginal:
We might try to account for the ungrammaticality of (357) and (358) by again
arguing that the passivisation restriction has an extragrammatical explanation such
that these sentences are syntactically and semantically well-formed, but
pragmatically anomalous because they pose processing difficulties (cf. the
discussion of (347b) above). Let's take a closer look at them. It would seem that
even a linguistically trained person requires some time to decide exactly what (357)
and (358) express. The linguistically naive hearer, when confronted with sentences
of this type, is even more at a loss trying to mentally compute the thematic
relations. It is very likely that there in (357) is initially interpreted as an element
introducing an existential main clause and it in (358) as referential it. On further
processing the hearer has to abandon these hypotheses, but by that time s/he will
be lost. To the extent that (359)—(361) are odd, this is probably due to processing
difficulties as well. Although I have not tested this, I suspect that (359) is more
difficult to process than (360) and (361) because a two-phase movement is
involved. (360) in turn would be slightly more difficult to process than (361)
because the verb in the subordinate clause is transitive, whereas in (361) it is
intransitive.
In this section I would like to discuss verbs like discourage, deter, dissuade and
restrain. These were included by Postal in his class of N-verbs (verbs of negative
causation). This is odd in view of the fact that he did seem aware of the differences
110 Special constructions
between these verbs and verbs like prevent. Thus, in his 1974 book he notes that
the pairs of sentences in (362) and (363) are not synonymous:
The reason why the a and b-sentences in each case are not truth-conditionally
equivalent is due to the thematic properties of the matrix verbs discourage and
deter in that they assign a θ-role to the postverbal NPs, which are therefore
arguments of these verbs. In addition, note that pleonastic elements like
nonreferential it and there cannot occupy the postverbal NP position:
This lends further support to the claim that these verbs behave syntactically
differently from prevent-verbs. Finally, unlike prevent-type verbs, discourage-type
verbs do not require the postverbal NP to express a propositional event (cf. section
4.3.2)).
The most plausible analysis of discourage-constructions is one in which the verb
takes an NP complement and a clausal argument This clause has a null PRO
subject which is coindexed with the direct object. Under this proposal (362a) is
analysed as follows:
(365) I lyP discouraged [NP the nurse, ] [CP PRO, from moving the patient]]
These sentences sound somewhat odd, but are nevertheless comprehensible. They
are markedly better when from is present. The oddness of (366) and (367) can be
explained by appealing to a performance factor like the one discussed in section
4.3.4: processing is made difficult by the fact that a nonfinite IPRO V-ing]
Predicative as-constructions 111
constituent has been left dangling, this time after the direct object of the matrix
clause.
4.3.7. Conclusion
4.4.1. Introduction
(a) The predicative as-string is related to a postverbal NP, the function of which
is yet to be determined. It is in any case not an indirect object.
e.g. They regarded, him as a leader. The Board of Directors criticised her
as a doctor.
(b) The predicative as-string is controlled by an indirect object.
e.g. ...which has given her a considerable "following" as a teacher.
(W.7.6.97)
(c) The predicative as-string is controlled by a prepositional object
e.g. This happened to me as a child.
(d) The predicative as-string is controlled by a subject NP.
e.g. Jonathan came to visit me as my lawyer.
(e) The predicative as-string is an NP postmodifier.
e.g. His achievement as a scholar is outstanding.
(0 The predicative as-string is not controlled by any overt element
e.g. As a mechanic it was marvellous to spend a year in Disneyland.
Notice that as can be followed by an Adjective Phrase in many, though not all, type
(a) constructions; thus we can have We described the case as hopeless in our report
and We regarded the outcome as obvious, but not *They elected her as incompetent.
Most of these types have been discussed in the literature (cf. Emonds 1984),
though never in great detail. Type (b), to my knowledge, has never been noted. In
my treatment of predicative as-constructions I will be assuming that the NP
following as must always have a subject. This subject may be overt, but need not
be.
(369) a. Last weekend...we instigated a meeting of the whole community with Ian
as chairman and laid bare our souls. (W.7.3-9)
b. The Vice-Chancellor made a good case for Universities as the
disseminators of wisdom.
In these sentences the [NP as NP] strings are complements of the prepositions with
and for respectively. Furthermore, as we have already seen in chapter three, [NP
as NP] strings can also occur in subject position or independently:
Thirdly, sentences like (371) involving coordinated [NP as NP] strings suggest
that Jim as a complete idiot and Maria as an equally stupid person function as
constituents:
(371) / regarded Jim as a complete idiot and Maria as an equally stupid person.
Here again we have clear evidence that the postverbal NPs in (372a) and (372b) are
not arguments of the matrix verb.
On the basis of these four pieces of evidence (three if we discount the
coordination data) we must conclude that regard-type verbs, at least in the sense
in which they are used here, do not take NP arguments.82 A sentence like (373)
is therefore plausibly analysed as in (374):
The question now arises how we should analyse the internal structure of the
constituent which has provisionally been labelled "clause" in (374). Let us first
concentrate on the cases where, apart from the element as, there is a verb, as in
(375) and (376):
In (375) as is followed by the main verb be, whereas in (376) it is followed by the
main verb have. One way of analysing the bracketed structures in (375) and (376)
is to regard the NP subjects (them and the policy) as being predicated of a PP
predicate headed by as:
However, here we face the same problem as we did when we discussed the
possibility of analysing from in prevent constructions as a preposition: in (377)
there is no subject-predicate relation between the NP and the PP (cf. *they are as
being clowns/ *the policy is as having no effect), but rather between the higher NP
and the lower NP. A second reason for not taking as to be a preposition is that it
can be followed by an Adjective Phrase, as we have seen, which as a rule is not
possible for prepositions. Finally, notice that PRO is governed in (377). This
violates the PRO-theorem.
A second possibility is to analyse a« as a complementiser (see Aarts 1990a). The
S-Structure of the sentences in (375) and (376) would then be as in (378) below
(only the relevant VP is shown):
Here the subjects of the IP-clause have moved from their D-Structure position in
Spec-of-IP to the Spec-of-CP position in the complement clause. Although this
Predicative ^-constructions 115
(381) [e] were regarded [CP [Spec ] [c as] [IP them being clowns]]
(382) [e] was regarded [CP [Spec ] [c as] [,P the policy having no effect]]
In order to derive (379) and (380) from (381) and (382) the subjects of the IP
clause would have to move via Spec-of-CP to the matrix clause subject position.
This movement is illicit as the NPs in question would be moving from one A-
position to another Α-position via an A'-position.
By far the most plausible analysis is to regard as as an inflectional element, like
from in preveni-constructions. Sentences (375) and (376) would then be analysed
as in (383) (again only the relevant VP is shown):
(383)
Regard-type verbs, like prevent-type verbs are analysed here as ECM verbs, which
can assign Case across IP. The constructions in which these verbs occur do not
therefore contain prepositional Small Clauses, as was suggested as a possibility in
section 4.4.1. Notice that as the verbs that occur in the IP complement clause are
always nonfinite -ing forms which do not move to "I", there is never the possibility
that as blocks V-Movement to "I".83
What about regard-constructions where there is no verb, as in (373), repeated
here as (384)?
We have already established that the string them as clowns is a clause. The question
116 Special constructions
is whether as is inflectional here too. I would like to claim that it is. The analysis
of (384) can then be represented as in (385) (where only the relevant VP is shown):
(385) VP
SC (=IP)
regard
The complement of regard is a Small Clause here whose syntactic status I take to
be IP. The lower VP in (385) is analysed as containing an implicit main verb "BE".
For further discussion of the syntactic status of SCs and their internal composition,
see chapter six.
Notice that in constructions containing the verb regard the element as is
obligatorily present. Thus, we cannot have (386):
This is not the case for all verbs which take a complement with as:
For verbs like imagine the element as shows the same type of idiosyncratic
behaviour as from in preve/if-constructions in that it can be left out As in the case
of from, I would say that in (387b) we have a phonetically null inflectional element
as.
Other verbs like regard...as are intend...as and take...as. Like imagine...(as) are
consider...(as), deem...(as), rate...(as) and reckon...(as), among others.
Secondly, notice that verbs like describe cannot be followed by dummy elements:
And, finally, notice that (390a) and (390b) do not mean the same, which again
suggests that the postverbal NPs in these sentences are arguments of the matrix
verb:
Thus, there is strong evidence suggesting that the postverbal NPs in describe-type
constructions are assigned a θ-role directly by the matrix verb. Sentences like those
in (391), which involve the verbs be and have in the as-string, can be analysed as
in (392) (only the relevant VP is shown):
(392) VP
NP I VP
V NP
The structure in (392), like constructions involving the verbs persuade and
discourage, is again an instance of what I have called controlled dependent
predication: the matrix clause object controls the PRO subject of the complement
clause. (392) does not therefore involve a prepositional Small Clause, as was
118 Special constructions
(394) VP
C IP (=SC)
NP I VP
As the NP him and the PP with envy in (395) are not in a subject-predicate relation
we cannot analyse these two phrases taken together as a Small Clause. The same
is true for him and highly in (396). In both sentences the pronoun him appears to
be an argument of the matrix verb. If this is so, our earlier claim that regard does
not take NP arguments is contradicted. However, it would seem that in (395) and
(396) regard has meanings which differ from the meaning of the verb discussed
previously. The meaning of regard in the sentences above is not 'consider', but
Predicative as-constructions 119
rather something like 'behold' or 'look at' in (395), and 'esteem' in (396). We are
therefore dealing with at least three different verbs regard. The verbs in (395) and
(396) take an NP argument, whereas the verb discussed in the previous section does
not
So far we have looked only at the predicative αί-strings of type (a) in section 4.4.1.
How do we analyse the other constructions listed there? Let us first re-establish our
earlier observation that the NP following as in ai-strings is a predicate and as such
it must be predicated of something. This means that there must be a subject, be it
overt or covert If we accept this, then, on the basis of the discussion in the
previous sections, we must analyse the example sentences in (a)—(f) of section
4.4.1, repeated here as (397a)—(401a), as in (397b)—(401b):
In each case the ai-strings form Small Clauses with a PRO subject
Let us first turn to the sentences in (397)—(399). The analysis of their internal
structure should not come as a surprise in the light of the discussion in the previous
sections. I have so far left unresolved the question of the functional status of the
as-strings. Are they to be regarded, perhaps, as extraposed postmodifying clauses?
That is, do the aj-strings in (397)—(399) postmodify her, me and Jonathan
respectively? Given the fact that in (399), and perhaps also in (397), the as-strings
may not occur in a position immediately following the head noun it would appear
that they are not postmodifying units:
(404) VP
V
PRO; as a teacher
PRO, as a child
PRO, as my lawyer
I should mention here the English construction involving the verb strike. This
verb is rather special in that it is one of a relatively small set of so-called subject
control verbs. Promise is another such verb. Consider the following sentence:
Here we have a subject-predicate relation between the subject of the matrix clause
and the AP intelligent, rather than between a postverbal NP and an XP predicate.
Notice that in (405) both the postverbal NP and the as-phrase appear to be
obligatory:
This suggests that strike takes two complements, namely an NP and an SC: [Jimj
[VP strikes me [sc PRO, as intelligent]]]. However, there is some evidence that
argues against such an analysis. Consider (408):
The question posed by speaker A in (408) shows that the αί-clause in B's response
has an adverbial feel to it. If this is correct perhaps (409) is a more appropriate
analysis of (405) (only the relevant VP is shown):
Predicative as -constructions 121
(409) VP
strikes me
(411) VP
strikes me
In both (409) and (411) as is positioned in "I", as in the regard and describe-
constructions discussed in previous sections.
We can contrast the sentences in (397)—(399) and in (402)—<403) with (400)
where we do have a postmodifying αί-clause occurring in a position immediately
following the head noun. Notice that (412) is unacceptable for most speakers, or
marginal at best:
(413) NP
SE
The Small Clause is analysed here as an adjunct to the head noun achievement. A
problem with this analysis is that the PRO subject of the SC is governed. We can
get around this by adjoining the Small Clause to the NP his achievement as in
(414):
(414) NP
NP
hisi achievement
Here the SC has the status of a phrasal adjunct (cf. Radford 1988b: 255), and PRO
is no longer governed.
Let us finally turn to sentence (401). Again we have an aj-clause with an empty
PRO subject, but here there is no NP with which PRO can be coindexed. Its
reference needs to be determined contextually. In this case it refers to the utterer
of the sentence.
4.5. (Mis)take...for
In these cases there is a subject-predicate relation between her and a lawyer in the
a-sentence, and between him and a lecturer in the b-sentence. Notice that the
postverbal personal pronouns in these sentences are not thematically related to the
hccxtse-type verbs 123
matrix verbs. After all, he did not take her in (415a), nor did she mistake him in
(415b), but rather he took [her to be α lawyer] and she mistook [him to be α
lecturer]. We cannot therefore regard her and him as direct objects of the matrix
verbs. The predicates, a lawyer and a lecturer, are complements of the element for
here. Interestingly, take...for and mistake...for mean more or less the same. Both
verbs roughly mean 'erroneously assume somebody/something to be something', so
that the sentences in (415) can be glossed respectively as 'He erroneously assumed
her to be a lawyer' and 'She erroneously assumed him to be a lecturer'. That the
assumption made is mistaken is only implied in (415a), whereas it is explicitly
conveyed by the prefix mis- in (415b).
SC-theory is forced to assign the following analyses to the sentences in (415):
Although this is perhaps somewhat counterintuitive, we could argue that for is again
an inflectional element here, like from in prevenf-constructions and as in regard-
constructions. However, it seems that for is more centrally a preposition than either
from or as. Notice, for example, that for, unlike as, cannot be followed by an
Adjective Phrase:
In this final section I would like to turn my attention to another group of verbs
which could be said to occur in a structure which involves a prepositional Small
Clause. These are the verbs which behave syntactically like accuse. They occur
with either
1 an NP
2 an NP and a PP
3 an NP + Ρ + -ing clause86
Notice that in (420) the NP James is the subject of murdering the cat. This leads
us to the following question: do the postverbal NP and the elements following it (of
murdering the cat) form a constituent? That is, is the string James of murdering the
cat in (420) a clause, perhaps even a (prepositional) Small Clause? This question
is straightforwardly answered in the negative. In all the sentences above the
postverbal NPs are separate arguments of the verb, not subjects of a clausal
complement. The proof of this is that both (419) and (420) entail (418).
Further proof is afforded by sentence (421), which results after passivisation of
the string following the verb in (420):
The unacceptability of this sentence can easily be explained if we assume that there
are selectional restrictions between the verb accuse and the following NP. This can
only be the case if the NP-position in question is assigned a θ-role by the matrix
verb, i.e. if the postverbal NP is an argument of the matrix verb.
We can draw the same conclusion on the basis of the sentences in (422), which
show that there is no possibility of pleonastic it or existential there occurring
directly after accuse:
These sentences again show that the NP which occurs in the position following
accuse is an argument of that verb.87
On the basis of all the observations made so far we must conclude that the NP-
position following verbs like accuse is a separate θ-position: the verb assigns a
patient θ-role to NPs occurring in this slot For this reason in (420) the postverbal
NP and the string introduced by of do not form a clausal constituent. We could of
course have a clause with a PRO subject following the postverbal NP. This
possibility will be discussed below.
Let us now address the question how to analyse the sentences in (418)—(420).
The analysis of (418) is straightforward: the NP is simply a direct object
Sentence (419) is plausibly analysed as in (423):
Accuse-iype verbs 125
The fact that there are selectional restrictions between the matrix verb and the PP
in (419) (cf. */ accused James on the murder) entails that the PP in (423) must be
an argument of accuse.
Let us now turn to the pattern instantiated by sentence (420), i.e. [V NP "P" -ing
clause]. We have already seen that the NP James in this sentence, apart from being
the direct object of the matrix clause, is semantically also the subject of the -ing
clause. One possible way of analysing (420) is as in (424) (only the relevant VP
is shown):
(425) VP
Under this analysis we would have to assume that the complementiser of, unlike the
preposition of, is not a governor because the IP subject PRO may not be governed.
On balance, the analysis in (424) is to be preferred over the analysis in (425)
because if of is regarded as a complementiser an unfortunate asymmetry obtains
between (425) and the analysis in (423), where this element is taken to be a
preposition.
It has been argued that the construction type discussed here is ditransitive (cf.
Quirk et al. 1985: 1210—1211; Andersson 1985: 93ff). In such analyses the
postverbal NP in sentences like (419) and (420) is regarded as an indirect object
which is followed by a prepositional object It is not clear what motivates this
analysis. The entailment relation between (419)/(420) on the one hand and (418) on
the other, noted above, would seem to be conclusive evidence that the postverbal
NP is not an indirect object argument, but a direct object argument Furthermore,
if the NP following the verb accuse were indeed an indirect object we would not
expect it to be possible for it to be moved to the right under Heavy-NP-Shift as
indirect objects cannot be displaced under such movement (cf. I told the woman
with the hat a story /*/ told a story the woman with the hat). Sentence (426) shows
that NPs following accuse can be shifted to the right under HNPS:
Notice that in sentences like (419) and (420) condition (259) of section 4.2.4
correctly predicts the impossibility of movement of the postverbal NP across the PP
Conclusion 127
to the end of the sentence. The reason why this movement is prohibited is that in
these cases the NP James is lighter than the PP. Thus we cannot have (427):
In this last case the NP and PP have equal weight Condition (259) disallows this
situation to obtain. Movement to the right of the NP is possible in those cases
where the NP in question is emphatic:
In (429) the NP this man has acquired extra phonological weight at PF allowing it
to appear in a position to the right of the PP of murder. See section 4.2 and also
chapter five for discussion of weight issues. Despite the differences, the analysis
presented here and those of Quirk et al. and Andersson agree that NPs following
accuse are arguments of that verb.
Other verbs like accuse are congratulate...(on), convict...(of), convince...(of),
sentence...(to), suspect...(of), thank...(for), wam...(of) etc.
4.7. Conclusion
and
only, and can be regarded as intransitive prepositional verbs (e.g depend [PP on
NP], look Ipp at NP] etc.).
In section 4.2 the following condition on derivations is proposed:
This condition accounts for the unacceptability of the sentences *I switched off it
and *I looked up it and the like, in which light pronoun NPs have moved across
equally light prepositional phrases. We will see in chapter five that the proposed
condition has a much wider application.
Sections 4.3—4.6 dealt with prevent-comtructiom, predicative aj-constructions,
(mis)take.. .for-constructions and accwse-constructions. I hope to have shown that
constructions containing prevent/discourage-type verbs and regard/describe-type
verbs are best analysed as involving complement clauses of the form [NP/PRO
from -ing] and [NP/PRO as XP] respectively. The elements resembling prepositions
in these constructions, namely from in prevent/discourage-consttuciions and as in
regard/describe-constructions, have here been analysed not as prepositions but as
inflectional elements. Perhaps for in (mis)take...for constructions is also an
inflectional element, but this is less obvious. Accuse-type verbs have been analysed
as subcategorising an NP and a PP argument
Chapter 5
Extractions from Small Clauses
5.1. Introduction
In chapter three I argued that there is substantial evidence which supports the
contention that Small Clauses exist as syntactic units. In chapter four I discussed
a number of special constructions. In this chapter I will discuss some further
characteristics of Small Clauses, more specifically the possibilities of extraction
from them. The emphasis will be on movement, both to the right and to the left,
from nominal and adjectival SCs. Nominal Small Clauses are [NP NP] strings the
two NPs of which are in a subject-predicate relationship with each other. Similarly,
adjectival Small Clauses are [NP AP] constructions in which the NP is the subject
of the AP. At the outset we must exclude structures like those in (430) and (431)
from the discussion:
In (430) there is no subject-predicate relation between the indirect object him and
the direct object an interesting story. Both these NPs are separate complements of
the verb. In (431) her is a direct object and last night functions as an adverbial.
Again, there is no sense in saying that her is the subject of last night.
In this section I will be discussing what types of extraction are possible from Small
Clauses. Let us first distinguish extractions to the right from extractions to the left,
and also movement of the subject of the SC from movement of the predicate. The
following picture emerges:
All of these movements are possible, except that predicate movement to the right
is extremely rare, hence the asterisk. It will not be discussed any further here. For
an example of such movement see sentence (294), section 4.2.4. It is also possible
130 Extractions from Small Clauses
Leftward Rightward
6188 1
(Passive) (Topical.)
1
1089 12 (io-inf.)
(Wta-Mov.) (Wh-Mov.)
Leftward Rightward
1690 1
(Passive)
(Topical.)
41 —
If we compare the extraction data for the nominal SCs with those for the adjectival
SCs we find that despite the great variety of rightward extractions (see below) they
hardly ever occur in the case of nominal SCs. The reverse situation holds for
leftward extractions: although there are only four types (Passivisation, Raising, Wh-
Movement and Topicalisation), they occur far more frequently. The picture for
Extraction possibilities from Small Clauses 131
adjectival SCs is the opposite from that of the nominal SCs: the number of
rightward extractions is relatively large. Just over 19% of the total number of
adjectival SCs involve such extraction (as opposed to 1% of the total number of
nominal SCs). The table for the adjectival Small Clauses also shows that there are
28 SCs which involve extractions to the left, just over 13% of the total (as opposed
to 59% of the nominal SCs).
All this means that some 68% of the adjectival SCs and 40% of the nominal SCs
involve no movements. A further observation we can make is that there are no
significant differences between the spoken and written varieties of the language as
regards the various extraction possibilities.
There are a number of different types of rightward and leftward extraction. They
are shown in (432) and (433):
Heavy-NP-Shift (HNPS)
a. Movement of NPs ·
•Movement of other NPs
/o-clause
b. »/- r/iar-clause -Extraposition
-mg-clause
L
adverbial clauses
c. 0 - • to-clause
-that-clause
a. Passivisation
b. Raising
c. Wh-Movement
d. Topicalisation
(432a)—(432c), while Detachment, (432d), will be dealt with in section 5.4. Section
5.5 discusses leftward extractions.
5.3.1.1. Heavy-NP-Shift
All derivations resulting after movement of phrases from Small Clauses to the right
are subject to condition (259) of section 4.2. Recall that this condition stated that
There are no instances of HNPS from nominal Small Clauses in the corpus, though
there are a handful of instances from adjectival SC-constructions. Here are two
examples:
(436) of course that leaves unknown the the the [sic] number we didn't find
(S.10.8a.62)
(437) At least some influential Labour lawyers argue that the decision makes
necessary early amendment of the Trade Disputes Act, 1906, to underpin the
workman's liberty to withhold his labour (W. 12.1.1-57)
Rightward extractions I: Extraposition 133
Let us now turn to the properties of nominal Small Clauses containing light NPs.
Recall that light XPs were defined in chapter four as being phrases whose head
does not take any dependents. Light NPs therefore have the following form:
(438) [ NP [N. [ Ν 7/
Such NPs can occur as SC-subjects and as SC-predicates. Examples are given in
(439) and (440):
In these sentences the SC-subject NPs him and her and the predicate NPs winner
and prisoner are light. Condition (259) (see above) correctly prevents him and her
from occurring to the right of winner and prisoner because the light pronouns may
not appear to the right of the equally light predicate NPs:
In the syntax we can now move the NP the student in (443) to the right across the
NP winner, and as the phrase the student is heavier than the NP winner, condition
(259) does not prevent it from appearing to the right of the latter. And yet the
resultant sentence is unacceptable:
(446) The committee declared [sc t, winner] [the student with the red jacket],
This case, however, is not problematic: a heavy NP93 (the student with the red
jacket) has moved across a light NP (winner). There is also no problem when both
NPs are heavy (i.e. have equal weight). Condition (259) then correctly predicts that
the heavy subject of the SC may not appear to the right of the equally heavy
predicate:
(447) ??The committee declared [sc i, winner of this week's contest] [the student
with the red jacket
In this section I will discuss movements to the right of regular NPs across light NPs
in adjectival SC constructions.94 Before doing so, consider again condition (259)
of section 4.2, repeated here
Observe that this condition predicts that sentences like (448) and (449) are
grammatical:
(448) Last weekend (...) we instigated a meeting of the whole community with Ian
as Chairman and laid bare our souls. (W.7.3-9)
(449) He tried to force open the horse's mouth to thrust in a "massive dose" of
liquid phenobarbitone in a plastic capsule. (W. 12.2.12)
Rightward extractions I: Extraposition 135
This prediction is borne out Both sentences are fine. What has happened in (448)
and (449) is that the regular Small Clause subject NPs our souls and the horse's
mouth have moved across the light predicate APs bare and open.
However, it is not difficult to find examples which (at first sight at least) refute
condition (259). Consider (450) and (451):
In these sentences too regular SC subject NPs have moved across light predicate
APs. Condition (259) does not block the result. And yet, (450) and (451) are not
possible structures of English. So what is going on here? Bolinger in his 1971 book
on phrasal verbs claims that the possibility of the subject NPs in question appearing
to the right of the predicate Adjective Phrases depends on any of three factors: the
type of matrix verb, the type of adjectival head that appears in SC-predicate
position and the semantic properties of the construction. He isolates three sets of
what he calls adjective phrasal verbs (1971: 7Iff). What I have been calling an AP
here he calls an adjectival particle. In each of these sets regular SC subject NPs
may appear to the right of light predicate APs.
Set (a)
This set is open as far as the adjectives are concerned, but closed as far as the
verbs are concerned. The matrix verb is the empty causative verb make (or
occasionally keep, leave, have, hold, render) followed by an unlimited (but
presumably not a large) group of adjectives.
Examples (taken from Bolinger 1971):
Set (b)
This set is closed as far as the adjectives are concerned, but open as far as the
verbs are concerned. The adjectives are open, loose, free and clear.
Examples (again from Bolinger 1971):
Set (c)
This set is open lexically, but closed semantically. Bolinger gives the example
Will it bleach white the undies? and comments "[b]leach and white are
synonymous, or represent some kind of cause-effect relationship in which the
effect is more or less intrinsic to the cause: to bleach something is to make it
white." (Bolinger 1971: 74)
Set (d)
Idiomatic expressions.
Examples:
In explaining the unacceptability of (452) and (453) we can again appeal to the
concept of informational weight discussed in chapter four and in the previous
section. In (452') and (453') movement of the postverbal NPs to the right across the
APs takes place in the syntax, deriving (452) and (453). In the latter pair of
sentences the APs happy and beautiful then acquire extra informational weight at
PF by virtue of the fact that they carry new information. As a result the APs and
NPs have equal weight. Condition (259) then rules these sentences out as
unacceptable.97
Now, this is all very well, it may be objected, but why is it that the same process
of assignment of extra informational weight at PF does not also result in the
unacceptability of the sentences in sets (a)—(d)? In other words: why don't the APs
occurring after the verbs in each of these sets also acquire extra informational
weight in such a way that condition (259) subsequently filters out the resulting
structures? We can explain the well-formedness of the sentences in sets (a)—(d) as
follows: notice that in most of these sentences the identity or type of adjective is,
to a greater or lesser degree, predictable. For example, in Bolinger's sentence Will
it bleach white the undies? no adjective other than white could have been used. I
would like to suggest that as a result of the fact that it is predictable in most of
these cases which adjectives combine with which verbs, the APs are not assigned
extra informational weight because they do not carry new (=unexpected)
information. They therefore remain light. As the NPs which move across them are
heavier they may appear in a position to their right Condition (259) therefore
correctly does not rule out the sentences in sets (a)—(d).
I should perhaps say a little more here about the idea of an element acquiring
extra informational weight. As before, in the model I am assuming anything may
move anywhere in the syntax. However, independently motivated principles and
constraints, such as e.g. (259), rule out certain resulting structures. This leads us to
say that (452) and (453) are grammatical, as they are generated by the grammar,
but they are ruled out at the PF-level where suprasyntactic processes such as, for
example, assignment of informational weight and assignment of contrastive stress
take place. I have been using notions such as informational weight and new
information somewhat loosely here. Fixing their exact meaning is the task of a
theory of pragmatics. See e.g. Sperber—Wilson (1986: 202ff). As I am more
concerned with the syntax of the constructions under investigation I will not pursue
these pragmatic issues any further here.
In this section we have seen how we can deal in a principled way with initially
puzzling data like (454)—(457):
The question we addressed was: why can we move an NP freely across the
adjective phrase in sentences like (454) and (455), but not in sentences like (456)
and (457)? These facts must be explained if an adequate treatment of these
constructions is to be given. In the present study the apparent anomaly of the data
has been explained by looking at the semantic, pragmatic and lexical properties of
the constructions under investigation and by appealing to (259), a condition which
was motivated independently in chapter four, and has proved to possess great
explanatory force.
Let us now consider some examples of the types of Extraposition listed under
(432b) and (432c) in section 5.2:
Although all these possibilities exist, there was only one instance of Extraposition
of a subject clause from a nominal Small Clause in the corpus, as we have seen.
This involved Extraposition of a /o-infinitival clause, as in (458). It is quite striking
that rightward Extraposition of subject clauses from nominal SCs is so infrequent.
In section 5.5 we will see that leftward displacements are much more common from
this type of SC. Quite the reverse situation holds for adjectival Small Clauses:
Extraposition to the right of subject clauses from such SCs is far more frequent, as
table 7 shows:
it — //wf-clause 8
0 — i/ial-clause100 8
it — fo-clause 18
0 — /o-clause 1
it — -ing clause 1
it — wAert-clause 0
it — »/-clause 1
In the next subsection I will discuss the question in which structural position
extraposed SC subject clauses end up. Section 5.3.2.3 deals with the Extraposition
of -ing clauses (cf. (460) above), while extraposed adverbial clauses (cf. (461a—b))
are dealt with in section 5.3.2.4. The (im)possibility of leaving out the dummy
subject it (cf. (462) and (463)) is discussed in section 5.2.3.5. Finally, the
obligatoriness of Extraposition of to and ίΛαί-clauses functioning as subjects from
Small Clauses is dealt with in section 5.2.3.6.
The accounts of clausal Extraposition in the literature usually deal only with
Extraposition of the clausal subjects of matrix clauses (e.g. That she is beautiful is
obvious It is obvious that she is beautifiil). As to the structural position where
these extraposed clauses end up, there seems to be a fairly broad consensus that
they are moved to a VP-final position, as we will see. A question I will address
here is whether this account also works for extraposed clausal SC-subjects.
The standard treatment of Extraposition of clausal subjects is to assume that at
the level of D-Structure they occur in their original (subject) position. They then
140 Extractions from Small Clauses
move to a position following the predicate. This movement leaves behind a trace
to which a propositional θ-role is assigned. This θ-role is subsequently transmitted
to the extraposed clause. Finally, the empty pronominal element it is inserted in the
position of the trace. Notice that it is not possible to base-generate it in the matrix
clause subject position as this dummy element would then be assigned a theta
role.101 The diagram in (464) shows how the process described above operates for
the sentence That she is beautiful is obvious:
(464)
Extraposition
//-Insertion
There has been some difference of opinion as to the exact syntactic position of
the extraposed clause (or Landing Site, to use a term introduced by the American
linguist Mark Baltin). Emonds (1976: 122) argues that the rule of Extraposition
should be formulated as in (465):
(465) Extraposition
The S-constituent moves from position 3 under NP into the base-generated empty
S-position 5, while it is inserted in the empty base-generated position 2 under NP.
This process is illustrated in (466):
Rightward extractions I: Extraposition 141
(466) S
NP Μ VP
Ν S V NP S
(467) *It proves (that) Mary is innocent that John has blood on his hands.
(Emonds' (21))
(468) VP
In (467) both the subject and the object Ss (S, and S2) have been extraposed as
Emonds assumes that Extraposition from object position is obligatory for clauses:
The claim is that (465) correctly predicts (467) to be ungrammatical as there is only
one possible landing site for extraposed clauses (namely ^ 0]). By contrast, Baltin
(1982: 11, 1978/1985: 143) deems sentences such as (467) not to be too bad. He
argues that configurations like (468) are not excluded,102 and claims that
constituency tests show that extraposed subject clauses end up at the end of VP in
the position of S2 in (468). Let us look at one of these tests in detail. Baltin
(1978/1985: 145) gives us the following pair of sentences (slightly amended):
(470) VP-Fronting
I warned you that it would upset Rosa that you smoke, and upset her that you
smoked it certainly did. (=Reinhart's (11))
(471) Though-Movemeat
a. Though it was unlikely [S2 that she would pass], Rosa still decided to take the
exam. (=Reinhart's (14a))
b. Unlikely that she would pass though it was, Rosa still decided to take the
exam. (=Reinhart's (14b))
In (471b) a subpart of the VP of the though-ctease has moved, namely the string
following the head was in (471a). Reinhart does not make clear whether extraposed
subjects end up as sisters of V, as in Baltin's framework, or in a Chomsky-adjoined
position. I will return to this matter presently.
Apart from the evidence surveyed so far there are further tests we can apply to
establish in which structural position extraposed subject clauses end up. Consider
(472) and (473):
(472) Do ^-Substitution
a. He said that he thought it is obvious that she likes him, and Clara does
so too.
b. *He said that he thought it is obvious that she likes him, and Clara does
so too that she likes him.
(473) Pseudoclefting
a. What they did was say it was a problem that she likes him.
b. *What they did that she likes him was say it was a problem.
(471b). This sentence does not establish unambiguously that the clause that she
would pass is part of the VP of the though-clause. Consider (474):
(474) Unlikely though it was that she would pass, Rosa still decided to take the
exam.
Here again a subpart of the VP of the concessive clause (the AP unlikely) has
moved, but notice that the ί/ιαί-clause has been stranded, and this could suggest that
it is not part of the VP of the though-clause. Furthermore, notice that we can insert
sentential adverbs in sentences with extraposed subject clauses (cf. also Emonds
1976: 123):
(475) It is true, however, that the Prime-Minister made an effort to unite her
cabinet.
(476) It is a problem, frankly, to send books abroad
Sentences (475) and (476) constitute evidence which favours an analysis in which
extraposed subject clauses are immediately dominated by IP, or adjoined to IP. On
balance, though, the accumulated evidence favours a VP-landing site for extraposed
subject clauses.
The question now arises whether extraposed subject clauses indeed end up in
position S2 in (468), as Baltin has claimed, i.e. as sisters of the head verb.
McCawley (1988: 98—99) adduces the sentences in (477) to show that this is not
the case, and that they are instead adjoined to VP, as in (478). The reason is that
"[b]oth the original VP and the combination of it with the extraposed subject
complement behave as units with regard to conjoining." (McCawley 1988: 98—99)
(477) a. It both surprised Alice and shocked Susan that John quit his job.
b. It both surprised Alice that John quit his job and shocked her that he
didn't seem concerned.
(478) a. S
NP VP
VP S
it both VP and VP
144 Extractions from Small Clauses
There are a number of problems with Baltin's treatment which also lead us to a
VP-adjunction analysis. First, note that his account of Extraposition is not structure-
preserving, as he himself observes (1982: 16). If Radford (1988b: 544—555) is
right in claiming that "[a] 11 transformations are structure-preserving, and comprise
either structure-preserving substitutions...or structure-preserving adjunctions" then
this is undesirable. Secondly, and more importantly, if extraposed clauses do end
up in the position of S2 in (468) then the Θ-Criterion is violated as this is a Θ-
position. Recall that the Θ-Criterion states that each argument must be assigned one
θ-role only, and that every θ-role must be assigned only once. Now, in Baltin's
theory the extraposed clause is assigned two θ-roles: one in its base-generated
subject position and one in its derived position.
We can, however, easily remedy the problems with Baltin's account if we
Chomsky-adjoin the extraposed subject clause to VP, as in (478). We then have an
account of Extraposition which is structure-preserving and which does not violate
the Θ-Criterion.103 Extraposition of clausal subjects then operates as follows104:
(479) IP IP
NP I VP NP I VP
it VP
insertion
The question we must now ask is whether the landing site for extraposed clausal
SC-subjects (like those in (458)—(463) of section 5.3.2.1) is also within VP. If we
apply the tests we find that again the results are mixed:
(480) VP-Deletion
a. Although not everybody thought it credible that she was a junky, there were
some who did.
Rightward extractions I: Extraposition 145
b. *Although not everybody thought it credible that she was a junky, there were
some who did that she was a junky.
(481) VP-Preposing
a. We said we would keep it a secret that she is coming, and keep it a secret that
she is coming we will.
b. *We said we would keep it a secret that she is coming, and keep it a secret we
will that she is coming.
(482) Pseudoclefting
(483) Do io-Substitution
a. Hans kept it a secret that he had inside information, and his girlfriend did so
too.
b. *Hans kept it a secret that he had inside information, and his girlfriend did
so too that he had inside information.
(484) TftoKgÄ-Movement
a. Keep it a secret that he sent the information though he did, the police
nevertheless found out about it.
b. Keep it a secret though he did that he sent the information, the police
nevertheless found out about it.
(486) VP VP
V SC VP Claus^
NP XP V SC
Clause XP
t
it
insertion
In (480)—(485) I have used ίΛαί-clauses in the landing site tests. The results are
the same for extraposed to and -ing subject clauses. I leave it to the reader to apply
the tests to these clause types. For extraposed adverbial subject clauses a different
situation obtains. See section 5.3.2.4.
The Small Clause construction which involves an -ing clause as subject is of some
interest as it is the only construction in which movement to the right of the clausal
subject takes place optionally. We will see why in section 5.3.2.5. The following
instance was found in the Survey of English Usage:
(487) but in this there are two points # one is # we ought to go on lecturing # and
the other is that we leave [it so late] # preparing it
(S. 1.4.44, brackets added)
There were no instances of -ing clauses moved out of nominal Small Clauses.
Constructions containing extraposed -ing clauses are exceptional in two further
respects. Firstly, notice that the extraposed clause in (487) is separated from the
preceding lexical material by a marked intonational break (#). This is not a
peculiarity of this particular sentence, but a characteristic of all sentences involving
extraposed -ing clauses. Secondly, notice that the dummy /f-pronoun must be
inserted in the Small Clause subject position after Extraposition of the -ing clause.
Thus, sentences like (488), where no it has been inserted are impossible in English:
As we will see in section 5.3.2.5, some of the other construction types involving
clausal Extrapostion do not always require it to be inserted.
Extraposition of -ing subject clauses is rare: the sentence in (487) was the only
instance found in the SEU. Furthermore, such movement is not always possible.
Quirk et al. (1985: 1393) quote the sentence in (489)
Rightward extractions I: Extraposition 147
(490) VP-Deletion
(491) VP-Preposing
(492) Do io-Substitution
a. The theology lecturer considers it a crime when people blaspheme, and his
students do so too.
b. ??The theology lecturer considers it a crime when people blaspheme, and his
students do so too when people blaspheme.
148 Extractions from Small Clauses
(493) 77ioMg/i-Movement
The pseudocleft test cannot be used here because if we apply it there occurs a clash
between the action verb do in the matrix clause, and the psychological verb
consider in the focus clause (*What we did was consider it a crime when people
blaspheme).
What do sentences (490)—(494) tell us? As in the case of extraposed that, to and
-ing subject clauses discussed in section 5.3.2.2, the results of applying the
constituency tests are mixed. This time, however, the weight of the evidence points
not to a VP landing site, but to a sentential landing site. I will explain why.
Observe that only the VP-Deletion test in (490) unambiguously demonstrates the
extraposed clause to be part of the Verb Phrase. Arguably, the do so substitution
test in (492b) does so too, although it would also be possible to say that this
sentence is bad for stylistic reasons only. The results of applying VP-Preposing and
Though-Movement in (491) and (493) are compatible both with a sentential and a
VP landing site. However, the results of Adverb Insertion in (494), and the sentence
in (495) below, in which the adverbial clause is situated in initial position, are
evidence which supports only the sentential landing site analysis:
This last sentence suggests that adverbial clauses of the type under investigation are
sentential adjuncts (cf. Quirk et al. 1985: 1392, note c). They should therefore be
in a position where they have scope over the whole sentence. An analysis of these
constructions should bear this out Jackendoff (1977: 97) in discussing sentences
like (496)
suggests that the w/zen-clause is base-generated and not extraposed at all. In view
of what has been said above this seems quite plausible, but does then beg the
question as to what is the status of the element it. The only possibility, it would
Rightward extractions I: Extraposition 149
Such an analysis is supported by the fact that we can paraphrase (496) as in (498):
If our conclusions are correct and adverbial clauses are not extraposed and not part
of the VP of the main clause, then the most likely place for them to be generated
is in a position adjoined to the CP node:
(499) CP
/criminal blaspheme
This analysis accounts straightforwardly for the fact that the wAen-clause has scope
over the matrix clause and can also occur sentence-initially, as in (495). Adjunction
to IP is excluded because pre-IP nodes can be filled, as in the following sentence:
In most of the cases of Extraposition of clausal Small Clause subjects that we have
looked at so far it was necessary to insert the pronoun it as a dummy subject in the
SC. Leaving out this element resulted in ungrammatically. However, certain types
of clauses, notably to and that clauses, allow non-insertion of it under certain
circumstances when they are extraposed, as (462) and (463), repeated here in (501)
and (502), show:
Table 7 in section 5.3.2.1 shows that tfctt-clauses allow this to happen more
frequently than ίο-clauses. Here are four SEU examples; the first two sentences
contain a dummy it, the second two sentences do not:
(503) I've found [sc iti necessary] [to join a public library in Theobalds's Road
which has a pretty good English department/, (S.3.3.42)
(504) Personally I do not find [sc itι surprising] [that lenders find this sort of
situation acceptable]t (W.7.11.35)
(505) The governor had felt earlier that he could handle the matter locally and had
not thought [sc 0 fit] [to communicate with the central government]
(W. 16.7.36-3)
(506) ...but the first thing that you must do is to make [sc 0 quite sure] [that the
purchasing power of his money is maintained...] (S.2.2&112)
Notice that the movement that has taken place in (505) and (506) resembles Heavy-
NP-Shift in that a heavy constituent is moved to the right leaving behind a trace.
However, the process instantiated by (505) and (506) differs from HNPS in that it
is obligatory, whereas HNPS is never obligatory. Observe also that omission of the
dummy pronoun is only ever possible in adjectival SC-constructions. We do not
find anything like the following:
(503') *I've found [sc 0 necessary] [to join a public library in Theobald's Road
which has a pretty good English department]
(504') *Personally I do not find [sc 0 surprising] [that lenders find this sort of
thing acceptable]
The obvious question to ask at this point is: what are the conditions that regulate
the presence or absence of the dummy subject pronoun it?
We can shed some light on this problem if we consider the relationship between
the matrix verb and the head of the Adjective Phrase predicate of the Small Clause.
It turns out that of the 8 constructions in the corpus which involve Extraposition of
a ίΛαί-clause without a dummy it having been inserted four involve the combination
make...sure, three involve the combination make...clear, and one construction
involves the combination think... likely. The one 0 — fo-clause construction (in
(505) above) involves the combination think...fit. As a first attempt at analysing
these constructions we could say that a process of Reanalysis takes place in such
a way that strings like make...sure, make...clear and the like are restructured as
complex verbs but only when the verb and the Adjective Phrase are juxtaposed.
After Extraposition a sentence like (51 la) would be reanalysed as (51 lb):
The ίΛαΐ-clause in (511b) then functions as the direct object of the newly created
complex verb make sure. There are, however, a number of strong arguments against
such a treatment Consider again sentence (506) above. In this particular case we
would have to say that the string make quite sure is reanalysed as a complex verb.
This is obviously highly undesirable as the lexicon will then contain scores of
make-adverb-sure combinations such as make very sure, make entirely sure etc. A
second problem is that we would have to explain why Reanalysis does not take
place in sentences like (503') and (504') where insertion of dummy it is obligatory.
I would like to offer an alternative explanation as to why it is not always inserted.
It runs as follows: a verb like make, in addition to its most frequent
subcategorisation frame
152 Extractions from Small Clauses
1 [ —,NP] make,
2 [ — , SC] make2
and
3 [ — » [ap—adjective...] CP ] make3
(512) They make damn good coffee in the town of Twin Peaks.
The second frame is selected for sentences like (513) and (514):
Finally, the third frame is chosen in those cases where no it is inserted, as in (506),
which is analysed as in (506'):
(506') ...but the first thing that you must do is to make [AP quite sure] [CP that the
purchasing power of his money is maintained]
(505') The governor had felt earlier that he could handle the matter locally and
had not thought [APfit] [CP to communicate with the central government]
The head of the AP needs to be specified in the lexical entry, otherwise the
grammar will overgenerate. In the third frame the dots represent possible modifying
elements such as very, quite etc. The type of clause selected (ίο-infinitival, that-
clause etc.) also needs to be specified in the lexicon. Sentences (505') and (506')
cannot be assumed to pattern with the second subcategorisation frame. If they did
it would be predicted that //-insertion is possible, which is not the case.105
We might refine our analysis by observing that there is a difference between
(505') and (506'): the Adjective Phrase head fit in (505') cannot be modified,
whereas the head sure in (506') can:
Rightward extractions I: Extraposition 153
(515) *...and [he] had not thought very fit to communicate with the central
government.
All this suggests that the degree of idiomaticisation is greater for think...fit and
make...ready constructions than for make...sure constructions. That is to say: the
bond between the matrix verb and the head of the following AP is greater for the
first type of construction than for the second. In fact, the degree of idiomaticisation
is so great for the think...fit/make...ready constructions that it is quite plausible that
Reanalysis applies to them, in such a way that we have the complex verbs think fit
and make ready. The objection against Reanalysis which I put forward for the
make...sure cases, namely that the lexicon would be cluttered with make-adverb-
sure combinations, does not hold for the think...fit and make...ready structures as
the adjectives fit and ready are never modified, as we have just seen. It would
appear that we now have to distinguish from the class of verbs taking frame 3
above a further, fourth class, namely one in which the degree of idiomaticisation
has increased to the point of restructuring. We end up with the following classes:
Class (a) The verb selects the frame [—, NP], e.g. make, in (512)
Class (b) Λ-insertion is obligatory, the verb selects the frame [—, SC], e.g.
make2 (514), find in (503), (504) and (507), keep in (508)
Class (c) //-insertion is impossible, the verb selects the frame [—,
[^...adjective...] CP], e.g. make3 in (506) which selects the frame [—,
[ ΑΡ...sure...] CP]]. The dots represent possible modificatory elements.
Class (d) Λ-inserüon is impossible, restructuring takes place such that a newly
formed complex verb selects the frame [—, clause], e.g. make4 in He
[VP [v made-ready] [to jump first]]. Also: think in (505)/(505').
Let us now turn to a more general problem: why is it that in the account of
Extraposition described earlier in this chapter the clauses which are base-generated
in SC-subject position must obligatorily be extraposed? In other words, why can we
not have sentences like (517) and (518) below:
Any treatment of a language which deals with the facts in a principled way and
which is aimed at explaining the data must give an account as to why Extraposition
of the subject clauses in sentences like (517) and (518) is obligatory.
We can try to tackle this problem by taking a closer look at the mechanism of
Case assignment. As we saw in chapter two, verbs assign objective Case to their
complements, and, exceptionally, to the subject of a Small Clause or to the subject
of a ίο-infinitival clause (e.g. believe assigns Case to Pat in we believe [Pat to be
an honest person J). It has been suggested in the literature that if a verb has a Case
to assign, it must assign it (cf. Lasnik—Uriagereka 1988: 29). Let us assume for the
moment that this is so, and also that Case must be assigned to a lexical NP. We can
then explain why (517) and (518) are ungrammatical. The verbs consider and think
must assign objective Case to an adjacent lexical NP. In these two sentences,
however, there is no adjacent lexical NP, only a nominal clause. As consider and
think cannot assign Case to clauses the sentences are ungrammatical (see also the
similar account of Stowell 1981: 148f, where ίΛαί-clauses and infinitival clauses are
held to be Case resistant). The only way to salvage (517) and (518) is to extrapose
the clausal subjects of the subordinate clauses and to insert the dummy pronoun it
so that this element can receive the objective Case which the transitive verbs
consider and think must assign.
The account sketched above works for fo-infinitive and ί/ωί-clauses which are
obligatorily extraposed. But what about sentences like (519), discussed earlier in
this chapter?
(519) but in this there are two points, one is that we ought to go on lecturing and
the other is that we leave [sc it so late] preparing it
In this sentence an -ing clause has been extraposed. But, as we have seen, and as
(520) shows, Extraposition is by no means obligatory here:
Thus, -ing clauses behave differently syntactically from to/that-clauses. The fact that
they do so is not all that surprising, however, in view of the fact that they behave
Rightward extractions I: Extraposition 155
differently in other areas of syntax too. Of the nominal clauses -ing clauses are the
most NP-like: for example (521)—(523) show that they are the only type of clause
that can appear as the complement of a preposition:
(521) By preparing it so late, we may not have time to mark the exam.
(522) *By to prepare it so late, we may not have time to mark the exam.
(523) *By that we prepare it so late, we may not have time to mark the exam.
We can now explain why a sentence like (520) above is grammatical and why
Extraposition is not obligatory by again claiming that the verb leave can, indeed
must, assign objective Case. Unlike ίο-infinitive clauses and rto-clauses, -ing
clauses are sufficiently similar to lexical NPs to be assigned Case. This means that
the string preparing it can be assigned objective Case. As the requirement that
leave must assign Case is met in (520) Extraposition is not obligatory. When
Extraposition does occur (as in (519)), Case is assigned to the empty element it
after the -ing clause has been moved.
Although the explanation for the phenomena discussed here in terms of Case
theory looks promising, it is not without problems. For one, we need to explain
passives like (524):
As (525) is not a possible surface structure the clause to eat it must move. In GB-
theory movement of NPs in passive D-Structures (e.g [e] was eaten the doughnutJ)
is triggered by the fact that the postverbal NP is required by the Case Filter to be
assigned Case. This Case cannot be assigned by the passive matrix verb and the NP
must move to the matrix clause subject position [e] to receive Case from "I" (see
section 2.3.1). However, in (525), as we are assuming that /o-infinitival clauses are
not assigned Case, and hence are not subject to the Case Filter, movement of this
clause to position [e] remains unmotivated. Another problem occurs when we
consider sentences like I considered that he was a fool. Here consider must assign
Case to an adjacent lexical NP, but cannot do so. A third problem, which Andrew
Radford has pointed out to me, is that in order for the proposed Case-theoretical
explanations to work we would have to set up some ad hoc mechanism to prevent
the it that gets inserted after the SC subject clause is moved (e.g. in (519)) from
transmitting its Case to the extraposed clause. All these complications suggest that
we should perhaps look for another explanation as to why SC subject clauses must
be extraposed.
Consider again (517) and (518). Rather than explaining that the clausal SC
156 Extractions from Small Clauses
Arguably a/or-clause (i.e. a CP) has been extraposed here. (527) would be the D-
Structure for (526b):
However, this is not obviously correct We might want to say that for them in
(526a) is a PP which is base-generated as a complement to the adjective within the
AP more difficult. The alternative D-Structure for (526a) is (528):
(528) ...it makes [sc to raise money more difficult for them]
It would appear that sentences like (526a) are structurally ambiguous in that both
the /or-clause Extraposition analysis and the /o-clause Extraposition analysis are
perfectly valid. The D-Structures in (527) and (528) can be paraphrased as in (529)
and (530):
Rightward extractions II: Detachment from nominal Small Clauses 157
One might ask whether these sentences mean the same. There seems to be a subtle
difference in meaning between them in that (529) focuses both on the money
raising activity as well as on the fact that it was undertaken by them, whereas (530)
focuses solely on the raising of money. Who undertook it is of secondary
importance. The difference in meaning between the two alternative structures in
(527) and (528) brought out by (529) and (530) is obscured in (526a) for which,
as noted, we have no way of knowing whether a /or-clause or a ίο-clause was
extraposed.106
...a constituent that detaches from subject position moves to a position outside
the verb phrase, whereas a constituent that detaches from a noun phrase within
the verb phrase ends up at the end of the verb phrase. (Baltin 1981: 266)
He cites the following sentences to show that the relative clause in (531) and the
PP in (532), both of which are detached from subjects, are not part of the VP which
has been deleted from the position indicated by the dash:
(531) Although nobody would ride with Fred who knew just him, people would —
who knew his brother.
(532) Although not many reviews appeared of Lucretia 's performance, one very
scathy review did — of Max's performance.
Consider now the sentences in (533) and (534), also from Baltin's work. In
(533b) and (534b) a relative clause and a PP respectively have been detached from
an NP within the VP:
If we apply some of the by now familiar VP-constituency tests, we see that the
158 Extractions from Small Clauses
relative clause in (533b) and the PP in (534b) remain inside the VP after
movement:
a. John said that he would call people up who are from Boston, and call people
up who are from Boston he will. (Baltin's (28a))
b. *John said that he would call people up who are from Boston, and call people
up he will who are from Boston. (Baltin's (28b))
a. Call people up from Boston though he may, he's generally pretty cheap about
long distance calls. (Baltin's (29a))
b. *Call people up though he may who are from Boston, he 's generally pretty
cheap about long-distance calls. (Baltin's (29b))
a. John said that he would call people up from Boston, and call people up from
Boston he will. (Baltin's (30a))
b. *John said that he would call people up from Boston, and call people up he
will from Boston. (Baltin's (30b))
a. Call people up from Boston though he may, he's generally pretty cheap about
long-distance calls. (Baltin's (31a))
b. *Call people up though he may from Boston, he's generally pretty cheap about
long-distance calls. (Baltin's (31b))
Baltin's claim seems to make the correct predictions in these cases. However, a
problem arises when we apply his "Detachment Rule" to constructions containing
Small Clauses. Consider (539):
Apart from it being possible to move the whole subject NP of the SC to the right
(as in e.g. I consider a fool any man who smokes) we can also move only the
relative clause:
Now, notice that Baltin's claim makes two contradictory predictions. The relative
clause who smokes is detached both from a subject NP and from an NP within the
Verb Phrase. It is predicted, then, that the moved relative clause ends up both
outside the VP and inside the VP. Obviously, only one of these predictions can be
correct. If we applied the VP constituency tests to the sentence in (540), we would
find that the detached relative clause remains inside VP. The question now arises
whether we should reject Baltin's "Detachment Rule" because it makes
contradictory predictions. The answer is "no". We can salvage his generalisation if
we can find a common characteristic for a postverbal argument NP like people in
(533a)/(534a) which occurs in VP and a Small Clause subject NP like any man who
smokes in (539) which is also part of VP. Notice that although the NP people in
(533a)/(534a) is a direct object and the NP any man who smokes in (539) a subject,
what they have in common is that both NPs are assigned objective Case by the
verbs call and consider respectively. The problem of the contradictory predictions
noted above can now be solved if we adjust Baltin's generalisation in the following
way:
It should perhaps be noted that, although there is no explicit claim about this in his
work, Baltin appears not to espouse Small Clause Theory. If instead he is an
adherent of Predication Theory then his framework makes the correct predictions
regarding (539) because in that theory the postverbal NP is not a subject. It is only
when Baltin's theory of Detachment interacts with Small Clause Theory that the
problem of the contradictory predictions rears its head.
It is possible to interpret the spirit of Baltin's claim to be that detached phrases
are moved to a position at the end of (or adjoined to) the maximal projection that
immediately dominates them, as Andrew Radford (p.c.) has suggested to me. This
may well be so, but in any case the revised version of Baltin's "Detachment Rule"
is an improvement of the original formulation as it stands, because the revised
version is more precise and does not allow multiple interpretations.
Detachment is best regarded as a stylistic rule. This view is also espoused in
Baltin (1981, 1982: 13) and in Quirk et al. (1985: 1397f). Gu6ron (1980: 645)
disagrees, but does not tell us why. There are a number of constraints on the
operation of Detachment. For discussion, see Gu6ron (1980).
160 Extractions from Small Clauses
We saw in section 5.2 that there are four types of leftward extractions:
Passivisation, Raising, Wh-Movement and Topicalisation. In what follows I will not
specifically be dealing with Wh-Movement and Topicalisation. These processes do,
however, play a role in the ensuing discussion.
Subsection 5.5.1 discusses Passivisation and Raising. The last two subsections of
this chapter deal with a variety of curious special constructions. One of these,
involving causative have, to be discussed in section 5.5.2, resembles a [V NP -ed]
construction of the type I had my car painted, but turns out to be structurally quite
different. The other constructions, to be discussed in section 5.5.3, involve predicate
movements to the left under Wh-Movement and Topicalisation.
Up to this point mostly active Small Clause constructions have been discussed. The
number of passive structures, however, is quite large, at least for the nominal SCs:
there were 78 passive structures, as opposed to 64 active structures. 61 of these
involved consider-type verbs, whereas 17 involved appoint-type verbs. We have
seen that leftward movements out of nominal SCs occur much more often than
leftward movements out of adjectival SCs. For some reason adjectival Small
Clauses resist leftward movement processes: there was only one case of
Topicalisation, there were only 11 cases of Wh-Movement and only 16 cases of
Passivisation.108
For consider-type verbs passive constructions always involve an empty subject
for the Small Clause at S-Structure. To illustrate, consider (541):
The NP that man in (542a) is moved to the empty base-generated matrix clause
subject position fe] at S-Structure for Case-theoretical reasons: the passive verb-
form considered in (542a) is unable to assign Case to the SC subject. In order not
to violate the Case Filter this NP must therefore move in order to receive its Case
from the INFL node in the matrix clause. Raising operates in exactly the same way.
In passive and Raising constructions the Small Clause never has a lexical subject,
except at D-Structure.109
The classic Raising verbs are verbs like seem and appear, but we should perhaps
Leftward extractions 161
add a verb to this class which at first sight does not seem to be a Raising predicate.
I am thinking here of make in one of its senses. Consider the sentence in (543)
below:
Notice that the verb make we have here is different from the verb make we have
in sentences like / made a nice cake or I made it clear that / don't want her to
come. Let us refer to make in (543) as make5 to distinguish it from the verbs
makediscussed in section 5.3.2.5.
My suggestion is that makes does not assign a θ-role to the NP a good professor
(i.e. this phrase is not a direct object) and that the D-Structure for (543) is as in
(544), where make5 subcategorises a Small Clause:
This seems a plausible analysis as the NPs you and a good professor are in a
subject-predicate relationship with each other. If (544) is indeed the correct D-
Structure representation for (543) then we must ask what motivates movement of
the NP you to the matrix clause subject position. If make, assigns Case to the SC
subject there is no motivation for such movement. Let us therefore assume that
make5, like verbs carrying passive morphology and like the Raising verbs, does not
assign Case to the NP that immediately follows it, in this case the SC-subject. So
as not to violate the Case Filter this phrase must then move to the matrix clause
subject slot in order to be assigned Case by the matrix clause INFL-node. Under
this analysis makes is regarded as a Raising verb. The Small Clause subject receives
its θ-role from the SC-predicate, the NP α good professor. The SC as a whole is
assigned a θ-role by the matrix verb. This account straightforwardly explains the
ungrammaticality of (545), pointed out to me by Andrew Radford:
This sentence (which contains an instance of make5) is ungrammatical for the same
reason that (546) is ungrammatical: Raising verbs do not occur' in the passive.
(551) He, would [VP make her [CP [sc PROt a good husband]]]
which complement verbs that have a preference for occurring in the passive. Here
are a few examples:
(555) With regard to personal details, I am a British citizen, [aged 22], married
with two children. (W.7.8.57)
(556) I write in reply to your letter [dated 18 March 1980]. (W.7.6.82)
(557) Silently I went to the shelf [marked sounds of pistol shots]. (S.11.3e.l2)
(558) My main area of teaching at St Mary's has been a two-year course [entitled
"Linguistics"] which is an important component of the BA degree.
(W.7.7.18)
In each of these sentences the matrix verbs subcategorise for a Small Clause. In
deriving the S-Structures the operators move to the empty base-generated subject
position of the matrix clause from where at LF they move to Spec-of-CP, an A1-
position. Passive be is deleted.
We would have to assume that the indirect objects are deleted at a later stage. We
can apply a syntactic test to see whether the matrix verbs here are simple transitive
verbs or ditransitive verbs. Consider (567):
Notice now that to the extent that (555)—(558) have active structures, these do
allow Heavy-NP-Shift of the postverbal NP:
(569) / dated 18 March 1980 the letter which I sent to my little sister.
(570) / marked sounds of pistol shots the shelves which had been painted red.
(571) I entitled "Linguistics" a course which I had always wanted to teach.
These facts suggest that the matrix verbs in (555)—(558) are not ditransitive verbs
but rather simple transitive verbs.
In the examples we have discussed the fossilised Small Clause constructions are
regarded as relative clauses with an empty operator functioning as a relative
pronoun. Consider now (572) and (573):
(572) [Aged 36], this man was the oldest first-year student.
Leftward extractions 165
(573) He began with flipping baseball cards, then moved, [aged about 10], to penny
poker, playing the game as though it were a shoot-out, with no reference to
other people's cards."3
In these sentences the highlighted strings are probably not relative clauses, but
rather adverbial clauses of time or reason, witness the fact that we can paraphrase
them as in (574) and (575) below:
(574) [When/because he was aged 36], this man was the oldest first-year student.
(575) He began with flipping baseball cards, then moved, [when he was aged about
10], to penny poker,...
(576) ...and we should have the whole lot made research assistants (S. 11.2.20)
Although [V NP -ed] constructions are outside the scope of this book, I would like
to discuss them briefly here in order to deal with a sentence like (576). (577) is
analysed as follows:
The NP my car is clearly not a direct object argument of have, and as this NP is
the subject of overhauled, the clausal analysis is not unreasonable. We might
wonder whether (578) is a base-generated construction or whether it is derived in
some other way. There are indications that it is not base-generated. Notice that
semantically my car is really the direct object of overhauled. (578) can therefore
be said to have the D-Structure in (579):
The verb overhaul assigns a θ-role to the NP my car. At S-Structure this NP must
appear in the empty position [e] in order to be assigned Case by the verb have
(recall that passive morphology on a verb absorbs its Case-assigning properties;
overhaul therefore does not assign Case to its postverbal NP). No external θ-role
166 Extractions from Small Clauses
(580) ...and we should have [chmse [e] made [sc the whole lot research assistants]]
In (580) there are two clauses: one embedded under have in the matrix clause, and
another (an SC) embedded under make. Notice that as the NP the whole lot is in
a subject-predicate relation with the NP research assistants it is reasonable to
assume that these two NPs together constitute a nominal Small Clause complement
to the verb make. However, (580) cannot be a surface form.114 We can explain
why this is so as follows: the SC-subject the whole lot cannot be assigned Case by
the passive verb-form made. It must therefore move into the empty position [e]
where it can be Case-marked by have. Observe that the NP research assistants is
a predicate, not an argument. By the Visibility Condition (see Chomsky 1986a: 95
and note 69 above) it does not require Case. The account I have given of (576)
differs in a number of respects from the analysis of (577). Firstly, the NP that
moves in (580) is a D-Structure subject, not a D-Structure direct object. Secondly,
this subject NP is θ-maiked not by the verb that precedes it, but by the SC-
predicate research assistants. The process which derives (576) from (580) is in fact
Passivisation. The putative active counterpart of (576) is then as in (581):
(581) ...and we should have [cbnist. NP make [sc the whole lot research assistants]]
Notice that the passive complement clause in (576) does not contain a form of to
be.
We must conclude that despite some similarities (576) is structurally quite
different from (577).
Leftward extractions 167
The tables in section 5.2 show that the vast majority of leftward extractions out of
Small Clauses involve Passivisation. In 42 out of 77 cases a Wh-phrase has been
passivised. Each time subjects have moved to the left. The tables show that
predicates too can move to the left Here are a few examples of moved Wh-
predicates:
In both (582) and (583) the SC-predicate what has moved to the CP-Specifier
position of the matrix clause.
An interesting situation obtains when Wh-movement of a predicate and
Passivisation take place at the same time, as in (584):
In the whatever-c]aase the subject and the predicate of the SC that complements
call have moved to a position further to the left. (585) is the resulting S-Structure:
(585) / dislike Margaret Thamchers or [CP whatever [IP shej is [VR called [sc tjj tJJJ]
i • • I
We have a free relative (i.e. nominal) clause here, which contains an SC the
predicate of which is the element whatever which moves to the Specifier-of-CP
position. The SC-node is not a barrier for this movement because it is L-marked.
VP, though, is inherently a barrier. However, if, in moving it up, we first adjoin the
predicate to this phrase (cf. section 2.3.4) there is no problem as it then crosses
only one segment of VP at a time. IP cannot be an inherent barrier. As for the SC-
subject, it also crosses no barriers in moving: SC is not a barrier because it is L-
marked and VP is not a barrier either if we adjoin the subject to it first, as above.
We end up with a Small Clause in (585) that contains no lexical material at all,
only traces.
There are also cases where there is Wh-movement of a non-overt Wh-predicate.
Consider (586):
(586) The new two-tier system, [as he sarcastically called it], would have
"fourpenny peers" and "fivepenny peers" — the voting and the non-voting
Lords. (W. 12.2.34)
case that the he concerned 'sarcastically called the two-tier system', but rather that
he 'sarcastically called it something'. In other words, the predicate is missing in the
αί-clause in (586). I propose that the D-Structure of this clause is (587):
In deriving (586) from (587) the empty operator moves under Wh-Movement from
its D-Structure position to the CP-Specifier position in the as-clause:
(588) ICP [Spec O, J l c [c as] [IP he sarcastically [VF called fsc it tJ]]]J
4 ι
Notice that in (588) the SC is not a barrier as it is L-marked. VP and IP are also
not barriers if we assume adjunction of the predicate to VP first, in the way
described above.115
The sentence in (585) is not the only construction that can involve movement of
both a subject and a predicate out of a Small Clause. Consider (589) which again
contains the complementiser as:
(589) The best theatre company in Britain is the Royal Shakespeare Company, or
the RSC, [as it is sometimes called].
The predicate here is again an empty operator. The movements operate as in (585)
and again result in a lexically empty Small Clause.
Two final examples of SCs not containing any lexical material are given in (591)
and (592):
In (591), an appomi-construction, the Small Clause predicate has moved under Wh-
movement resulting in a clause containing a PRO and a trace. In (592) the SC-
subject it as well as the predicate collate or something have been fronted; the first
under NP-Movement, the second under Topicalisation.
A further instance of a construction that we could handle under the heading
Topicalisation is given in (593):
(593) lAP brilliant], though we consider [sc him t j , we will not give him the job.
Conclusion 169
(594) [CP [Spec brillliantj [c [c though] [,P we consider [sc him tj]]j...
5.6. Conclusion
In this chapter further properties of Small Clauses have been discussed, more
specifically the possibilities of extracting elements from them, both to the right and
to the left.
I have given accounts of Extraposition to the right of heavy, light and regular
subject NPs. I have also accounted for Extraposition to the right of SC-subject
clauses and for the complex conditions under which the empty element it must be
inserted when such movement takes place. Furthermore, I have tried to explain why
such Extrapostion is obligatory in most cases. Detachment, i.e. rightward extraction
from NPs, was dealt with in section 5.4. The condition proposed in chapter four on
derivations resulting after rightward movement of constituents as well as other
pragmatic considerations have been seen to play an important role in this chapter.
In section 5.5 I discussed leftward movements out of Small Clauses. I dealt with
Passivisation and Raising, and proposed analyses of constructions which have
hitherto received little or no attention in the literature. One of these involves the
verb make in such sentences as he would make a good lecturer. I have also
discussed what I have called Fossilised Small Clause Constructions, that is,
constructions containing nominal SCs which complement matrix verbs which
always or predominantly occur in the passive. Empty operator movement was
posited in the analysis of these constructions. Finally, I discussed a structure
involving causative have and leftward movement of predicates.
Chapter 6
The Small Clause node
6.1. Introduction
Small Clauses are controversial entities. This is true in the field of linguistics as a
whole but we saw in chapter two that even some Government-Binding theorists
have questioned their existence. The situation turns out, however, to be more
complex still because there is also no agreement among the Small Clause
proponents. SC-theorists agree that Small Clauses exist as syntactic constructs; what
they disagree about is the exact categorial status of the Small Clause node. In this
book I have so far discussed old and new evidence in favour of recognising Small
Clauses as syntactic units. I have looked at a number of special SC-constructions
and I have dealt with extraction phenomena from Small Clauses. I have not,
however, discussed the issue of the syntactic status of Small Clauses and their
internal structure. Up to now I have mostly used the label "SC", thus not
committing myself to any particular view. I have, however, tentatively suggested
throughout that Small Clauses are simply IP (=S) constituents. In section 6.3 I will
defend this position. But first let us look at some of the other proposals that have
been put forward in the literature regarding the categorial status of the SC-node.
Small Clauses are very much a phenomenon of the eighties. The first proposals
regarding their analysis date back to 1981.116 In the following subsections I will
look at the frameworks of Stowell (1981, 1983); Manzini (1983); Chomsky (1981,
1986b); Kitagawa (1985); Kluender (1985); Chung—McCloskey (1987);
Hornstein—Lightfoot (1987) and Radford (1988a/1988b).
The theory of Stowell (1981) has already been discussed in chapter two. Let us
briefly recapitulate his ideas. In essence he claims that Small Clauses should be
analysed as in (595):
(595) U NP] X7
Thus, the Small Clause Sue intelligent is taken to be an Adjective Phrase the
subject of which is in the Specifier position. Why should this be? It is on the basis
of sentences like those in (598) that Stowell has claimed that the matrix verb is
sensitive to elements occurring inside the Small Clause:
(598) a. I expect [sc that sailor off my ship] (Stowell 1981: 257)
b. */ expect [sc that sailor very stupid] (Stowell 1981: 259)
In (598) the predicate of the Small Clause can be a PP, but not an AP. Therefore,
according to Stowell, expect subcategorises a Small Clause with the syntactic status
of a PP. The head of the SC is the predicate, of which the Small Clause is a
projection. See also Contreras (1987). Chomsky (1981: 169) largely follows
Stowell's conclusions regarding the structure of SCs except that he believes that the
Small Clause node cannot be a maximal projection, as Case must be assigned to the
SC-subjecL In the 1981 theory Case assignment (which takes place under
government) cannot occur across a maximal projection. However, Chomsky does
not commit himself to any particular view of the syntactic status of Small Clauses.
He simply labels them X*. In the 1986 Barriers framework it is no longer a
problem for the Small Clause node to be a maximal projection. In that theory the
SC is L-marked (lexically marked), and hence not a barrier (see section 2.3.4).
Chomsky's revised analysis of SCs is essentially that of Manzini (1983, 1989). It
involves an adjunction structure, in which the SC-subject is adjoined to its
predicate: [XP NP [XP ...]] (1986b: 20).
Stowell's analysis is without doubt currently still the dominant treatment of Small
Clauses. However, it is not without problems, as researchers have pointed out over
the last couple of years. In what follows I will summarise some of the arguments
put forward in the literature against Stowell's treatment. I will also present some
new counterevidence.
Kitagawa (1985: 21 If) has argued that the ungrammaticality of (598b) is due to
the selectional restrictions holding between the matrix verb and its SC-complement.
According to him a verb like consider selects a "state of affairs" Small Clause,
whereas verbs like expect select "change of state" complements. In (599) below
consider is followed by a "change of state" complement, hence the sentence is
ungrammatical. In (598b) expect is followed by a "state of affairs" complement; this
too leads to ungrammaticality.
(599) *The doctor considers [that patient dead tomorrow] (Kitagawa's (8a))
In order to explain why (601) and (602) are fine one would have to allow Wh-
Movement of heads out of SCs to be possible as a marked option. This is
undesirable.
Hornstein—Lightfoot (1987: 32f) put forward no fewer than five arguments
against Stowell's analysis. They argue that "Stowell commits himself to some
complications which range from unnecessary to unacceptable". (1987: 32) Here I
will concentrate on Hornstein—Lightfoot's two main arguments against Stowell's
analysis which concern what they consider the unacceptable complications of that
proposal. Firstly, like Kitagawa, they argue that the ungrammaticality of a sentence
like */ consider John off my ship is not due to a subcategorisation restriction
between consider and off my ship. They cite the sentence As soon as he sets foot
on the gangplank, I'll consider John off my ship to show that if the context is
changed, the sentence becomes acceptable. The subcategorisation relation holds,
they imply, between the verb and the SC as a whole. Secondly, in Stowell's
framework it is possible for consider-type verbs to subcategorise an NP whose head
is a maximal projection, as in / consider [NP John] [NP a friend]]
(Hornstein—Lightfoot's (27), cf. also (598a) above). This would require changing
the phrase structure rules. The same problem occurs in the framewoiks of Chomsky
(1981) and (1986b). See Williams (1983: 297—298) who first noticed these facts.
In assessing Stowell's proposals Radford (1988b: 517) makes the same point he
observes that SC-predicates are probably not single bar categories. On the basis of
sentences such as (603) he argues that they must be maximal projections because
they may contain Specifiers:117
(603) a. I've always considered [sc John THE best player in the team]]
b. I've never considered [sc John MY best friend]]
c. I'd never considered [sc John [AP THAT fond of his mother]]
(Radford's (187))
Radford further observes that subjects of SCs can be fronted, as in (604) and (605)
(see also chapter five of the present work):
174 The Small Clause node
Now, if SC-subjects are Specifiers how can we explain the grammaticality of these
sentences given that in general in English Specifiers cannot be fronted (though see
chapter two, note 24)?
The arguments against Stowell's treatment of Small Clauses found in the
literature by no means exhaust all the counterevidence. In chapter three I discussed
the possibility of sentential adverbs occurring in Small Clauses. Consider again
sentences (127) and (128) from section 3.2, repeated here as (606) and (607):
(606) I thought [it perhaps a pity] at the time, but his motivation was pessimism
(unnecessary in his case) about academic job prospects.
(607) I must admit that I have found [these summer international schools probably
the most rewarding part of my work].
(608) a. I thought that [it was perhaps a pity] at the time, but his motivation was
pessimism...
b. I thought that [perhaps it was a pity] at the time, but his motivation was
pessimism...
c. I thought that [it was a pity perhaps] at the time, but his motivation was
pessimism...
(609) a. Perhaps I thought [it a pity] at the time, but his motivation was
pessimism...
b. Perhaps I thought that [it was a pity] at the time, but his motivation was
pessimism...
c. I thought perhaps that [it was a pity] at the time, but his motivation was
pessimism...
(610) a. I must admit that I have found that [these summer international schools
are probably the most rewarding part of my work],
b. I must admit that I have found that [probably these summer international
schools are the most rewarding part of my work],
(611) a. I must admit that probably I have found [these summer international
schools the most rewarding part of my work],
b. I must admit that I have probably found [these summer international
schools the most rewarding part of my work].
The analysis of Small Clauses 175
(612) Wimbledon has developed into perhaps the most prestigious tennis
tournament in the world.
(613) We serve probably the best coffee in town.
Here we could say that perhaps in (612) is part of the NP the most prestigious
tennis tournament in the world which complements the preposition into, and that
probably in (613) is part of the NP the best coffee in town which complements the
verb serve. Andrew Radford has suggested to me that the structure of these NPs in
X-bar theoretical terms should be along the following lines:
(614) [NP perhaps [NP the most prestigious tennis tournament in the world]]
(615) [NP probably [NP the best coffee in town]]
In his own work he has proposed this kind of treatment for NPs like even the older
residents (cf. Radford 1988b: 255—256). The data in (606) and (607) Radford
believes to be irrelevant because, so he claims, the strings perhaps a pity and
probably the most rewarding part of my work ought to be analysed along the lines
in (614)/(615). However, firstly, as regards (606) and (607), their paraphrases in
(608) and (610) clearly refute an analysis of their bracketed strings as in
(614)/(615). Especially (608b/c) and (610b) are telling in that here the adveibials
occur either clause initially or clause finally, divorced from the NPs a pity and the
most rewarding part of my work. This type of moveability is typical of sentential
adverbials. Secondly, notice that the strings perhaps the most prestigious tennis
tournament in the world and probably the best coffee in town in (612) and (613)
have a prepositional flavour, witness the paraphrases in (616) and (617):
(616) Wimbledon has developed into what perhaps is the most prestigious tennis
tournament in the world.
(617) We serve what probably is the best coffee in town.
These sentences show that the complements of into and serve in (612) and (613)
are best regarded as reduced nominal relative clauses. If this is correct we have an
explanation as to why perhaps and probably can occur in strings that appear to be
NPs, but are in fact clauses.118 We can conclude on the basis of the distributional
behaviour of elements with sentential scope, such as perhaps and probably, which
do not as a rule occur in phrasal expansions of lexical categories, that Small
176 The Small Clause node
Kitagawa (1985) treats all complement Small Clauses as S-bars (=CPs) containing
an INFL-node (cf. also Stevenson—Coopmans 1991). He does not deal with non-
subcategorised SCs. It is argued that only if we assume that Small Clauses are S-
bars do we have no violation of Subjacency (see section 2.3.1) when extracting the
subject of an SC, as in (618a). Assuming that NP and S are the bounding nodes for
English, Subjacency is violated in (618b) as the Wh-element moves across two S-
nodes. In (618c), the structure Kitagawa proposes for (618a), Subjacency is not
violated because no two NP or S-nodes are crossed in one swoop.
If we stick to the 1981 version of GB-theory this argument makes excellent sense.
However, in the 1986 Barriers framework it no longer does. The SC in (618a) is
L-marked (see Chomsky 1986b: 15 and section 2.3.4), and therefore not a barrier.
The higher S is not a barrier either because Ss (=IPs) are not inherent barriers nor
does this node inherit barrierhood from another phrase. Therefore, in moving up,
the Wh-element in (618a) does not violate Subjacency. The corollary of this is that
in the Barriers model the structure in (618b) is a possible representation for (618a):
the lower S is L-marked and therefore not a barrier, nor is the higher S, as before.
(618c) is ruled out in the Barriers framework for Case-theoretical reasons: the
position t within S cannot be assigned Case by consider because S' (=CP), although
L-marked, is a barrier by inheritance from S, a blocking category.
There are other serious problems with Kitagawa's account One of these concerns
the question how subjects of Small Clauses receive Case. Schematically the
structure of SCs in Kitagawa's view is as in (619):
Case from a preceding matrix verb because the S' is a barrier to government in the
model Kitagawa is assuming. The only option available is for the INFL-node within
the SC to assign Case to the SC-subjecL However, this is problematic for two
reasons. Firstly, if Kitagawa is right in assuming there to be an empty copula BE
in INFL, which is intuitively appealing, then the question immediately arises how
this INFL-node assigns Case to the SC-subject. It is reasonable to assume that INFL
in (619) is untensed and an INFL-node which is marked [-tense] does not assign
Case. Even if we were to find a solution to this there is still a second problem: SC-
subjects require objective Case (cf. *I consider he a fool). INFL, however, assigns
nominative Case. So, for Kitagawa's account to go through we would have to
exceptionally allow an INFL-node which is marked [-tense] to assign accusative
Case. This is an unnecessary complication of the theory.
On the basis of the above observations I reject Kitagawa's claim that complement
Small Clauses are S-bars/CPs.
where X = N, A or Ρ
The node INFL0 takes NP, AP or PP complements. When INFL is marked [+tense]
or [-tense] its complement is a VP. The article makes a number of further claims
regarding the element PRO which need not concern us here.
The most important problem I can see in Hornstein-—Lightfoot's proposal is that
it necessitates a complication of the theoretical apparatus. We now have two types
of INFL-node: "regular" INFL which takes a VP as its complement and INFL0,
which takes NP, AP or PP complements.
A related problem concerns the fact that under Hornstein—Lightfoot's analysis
Small Clauses are regarded as being neither finite nor nonfinite. Such a view does
not square very well with linguistic intuitions.
A final problem is that if INFL in Small Clauses is indeed empty we have no
way of accounting for agreement phenomena holding between subjects and
predicates of Small Clauses. That there is agreement between the subjects of SCs
and their predicates can easily be demonstrated. Consider the sentence in (621). It
is ungrammatical because there is no number agreement between the two NPs
inside the SC:
178 The Small Clause node
Bob Borsley (p.c.) notes that in order to have agreement between two elements an
INFL node need not necessarily be present. He gives the example of agreement
between attributive adjectives and nouns in many languages. His observation is
correct. However, if we do not use INFL we will have to find some other
mechanism which takes care of the agreement requirement in sentences like (621).
On general theoretical grounds it is desirable not to have to add extra machinery
of any sort to the existing system. Below I will propose a mechanism which
handles the agreement phenomenon in sentences like (621) and which requires the
presence of an INFL-node.
On the basis of German and Irish data repectively, Kluender (1985) and
Chung—McCloskey (1987) conclude, like Hornstein—Lightfoot, that Small Clauses
are of the category S. However, their proposal, unlike that of Hornstein—Lightfoot
and unlike the proposal to be presented in section 6.3, does not allow an INFL-node
in SCs.
Chung—McCloskey's analysis is unattractive for a number of reasons. First of
all, there again seems to be no provision to deal with agreement phenomena of the
type in (621) above. In the absence of INFL some other as yet unknown mechanism
is then needed.
Secondly, as Kluender and Chung—McCloskey admit, their Small Clauses are
exocentric structures in that they are not properly headed. This means that they do
not conform to X'-theory. As a result, we lose the account of cross-categorial
generalisations that X'-theory furnishes. It may well be that Chung—McCloskey are
right in saying that it is "necessary for the theory of grammar to allow at least some
(language particular) phrase structure configurations that conform to no principled
cross-categorial pattern" (1987: 235), but it seems that for English at least the cross-
categorial symmetries are well-established, and exocentric structures should not be
allowed.
Radford (1988a/1988b) does not firmly commit himself to a position regarding the
status of Small Clauses. He does, however, suggest (1988b: 324 and 518) that they
have the structure in (622):
(622) [ s c NP XP]
A new analysis of Small Clauses 179
He presents arguments supporting the view that SCs lack a C-system and an I-
system. That is, he believes that Small Clauses do not have a complementiser node
or an I-node. That Small Clauses do not contain a COMP-node, and are therefore
not of the category CP, seems to be correct If they were it would be impossible for
the SC-subject to receive Case as the CP would be a barrier (by inheritance from
IP, which is a blocking category). See section 6.2.2 above. Radford's other claim,
namely that SCs do not contain an I-node, seems to me to be untenable for the
same reason that Hornstein—Lightfoot's positing of an INFL0-node is dubious: if
there is no INFL-node, how do we then account for the fact that there is agreement
between subjects and predicates of Small Clauses without positing the existence of
additional machinery?
Consider also the following statement:
With regard to the syntactic status of Small Clauses and their internal structure we
can summarise the preceding sections as in table 8 below. My own proposal, to be
discussed presently, has been included.
180 The Small Clause node
Table 8. The syntactic status of Small Clauses and their internal structure
Stowell (1981) XP No No
Chomsky (1981) X* No No
Kluender (1985)/
Chung—McCloskey (1987) s No No
Radford (1988a/1988b) — No No
I will now turn to my own proposal regarding the analysis of Small Clauses. So far
I have argued that Small Clauses must be sentential constituents not phrasal
expansions of lexical categories. Two possibilities now present themselves: either
SCs are CPs (as in Kitagawa (1985), see section 6.2.2), or they are IPs. I have
already argued against the CP-analysis by observing that if SCs were CPs there
would be no way for the SC-subject to receive Case. I therefore take Small Clauses
to be IPs.
What remains to be done now is to discuss the internal structure of Small
Clauses. Again, two possibilities present themselves: either Small Clause IPs have
an unmarked structure or they have a marked structure. Of these possibilities I
propose to adopt the former as it is the null hypothesis. I will thus take SCs to have
an I-node as well as a VP-node. The I-node is necessarily marked [-tense] as there
is no sense in assuming a clause without a lexically realised verb to be tensed.
Also, if "I" were marked [+tense] it would assign nominative Case to the subject
position. We saw above that this generates ungrammatical sentences as SC-subjects
A new analysis of Small Clauses 181
where X= N, A or Ρ
Let us now return to the I-node and to the question how we can account for the
agreement relation that holds between subjects and predicates of Small Clauses, as
in (624):
(625) [Of which frescos do you consider [IP(=SQINP restoration tj[vp BE [NP a
disaster]]]
In this sentence the subject NP the restoration of the lower clause, which is
positioned in Spec-of-IP, is not L-marked by consider and hence a blocking
category and a barrier. The IP which immediately dominates it is then also a barrier
(by inheritance, see section 2.3.4). It is thus predicted that (625), in which the PP
of which frescos has moved out of its containing NP, is ungrammatical. This
prediction is wrong because (625) is fine. Chomsky (1986b: 25) discusses this
problem with regard to [V NP to VP/-constructions (i.e. Exceptional Case Marking
constructions). He assumes that a verb like believe L-marks its complement IP, its
head (="1" containing to) and (under Spec-Head agreement) its subject. If the
subject of IP is L-marked in this way it cannot be a blocking category, and IP will
then not inherit barrierhood from it. Chomsky observes that
The advantage of the present analysis of SCs is that the mechanism that voids the
barrierhood of IP in fte/ieve-constructions with regard to extraction from the subject
position also voids the barrierhood of Small Clause IPs when such extraction takes
place because SCs are analysed here as containing an "I"-node. Thus, if we take
SCs to be IPs there is no need to make any stipulations about the existence of "an
agreement relation of the relevant sort" between subjects and heads of Small
Clauses.
A further motivation for taking Small Clauses to be IPs containing an INFL-node
is that such an analysis allows us to account in a straightforward way for SCs
containing the element as (e.g. in I regard them as wicked, see section 4.4). We
have seen that there are reasons for regarding as not as a preposition, but rather as
an inflectional word positioned in INFL. SCs containing as are a problem for the
other frameworks we discussed because they have no obvious way of
accommodating this element.
My final reason for analysing Small Clauses as containing an "Γ-node, alluded
to in previous sections, runs as follows: if, as I have argued, Small Clauses are
sentential constituents, not phrasal expansions of some lexical head, then SCs must
be either finite or nonfinite. In other words, they cannot not be marked for the
feature [±tense] and hence they cannot not have an I-node. To put it another way,
if SCs did not have an I-node we would end up with clauses which are neither
finite nor nonfinite. That would be rather like having human beings who are neither
male nor female.
A new analysis of Small Clauses 183
We need to tie up some loose ends regarding the assignment of Case to the
subjects of independent Small Clauses and to the subjects of SCs that themselves
function as subjects. Let us consider both these cases in turn. Consider first some
examples of independent SCs taken from previous chapters:
(626) The room was extremely noisy: children shouting [the tv on], [the record
player on}, and little Jimmy kicking the cat.
(627) [Doors open 20.30]
(628) [PRO closed]
A problem with the bracketed SC-structures is how their subjects receive Case. As
there is no verbal or prepositional Case assigner there is no straightforward way in
which Case can be assigned. I can see two possible ways of approaching this
problem. One of these was hinted at in chapter four. We could posit the existence
of a covert prepositional complementiser which is instrumental in assigning Case
to the subject of the independent SC as follows:
[ lCa] lsc NP XP ]]
Case
Such an account is problematic, however, in view of cases like (628) where the SC
has a PRO subject which would be governed by the abstract prepositional
complementiser. A more viable approach would be to say that subjects of
independent SCs receive what has been referred to as default Case (see e.g.
Chung—McCloskey 1987: 188, footnote 11) which, for English at least, is
objective.
Let us now turn to Small Clauses which function as subjects. Consider sentence
(629) below with the structure in (630):
(629) The Yorkshire Ripper free poses a greater threat than the Yorkshire Ripper
behind bars.
184 The Small Clause node
(630)
As noted in chapter four, subjects of Small Clauses which are themselves subjects
always carry objective rather than nominative Case (cf. Him free poses a greater
threat than him behind bars). The question is: how do these NPs acquire this
objective Case-marking? Notice that the closest governor and Case assigner for the
NP the Yorkshire Ripper in the structure in (630) is the higher I-node (the lower "I"
is marked [-tense] and cannot therefore assign Case). This node governs the subject
of the SC. That this is so is confirmed by the following facts (taken from Contreras
1987: 233):
( 6 3 5 ) a. VP b. PP C. IP
NP NP NP
I VP
Indirect government, by contrast, is what we get in all other cases, for example in
Exceptional Case Marking constructions where the subject of IP (in Spec-of-IP) is
governed in the following way:
(636)
indirect government
Here V governs IP, but also its Specifier and head (see chapter two). If we now
assume that under indirect government only objective Case can be assigned we have
an explanation for the fact that subjects of Small Clauses which themselves function
as subjects always take objective Case. The higher I-node in (630) then assigns
objective Case to the SC-subject the Yorkshire Ripper. The lower I-node, because
it is marked [-tense], does not assign Case, as we have seen.123 This proposal has
the additional benefit that it allows us to account straightforwardly for the fact that
subjects of gerund subject clauses also carry objective Case:
Andrew Radford suggests (p.c.) that in some cases subjects of Small Clauses in
subject position are nominative:
(638) [He in the front seat] and [she in the back seat] would be safer.
The fact that there is nominative rather than objective Case-marking on the SC-
subjects here may have something to do with the fact that we have coordinated SCs
in (638), as shown below:
(639) SC (=IP)
SC and SC
186 The Small Clause node
Some of these sentences are no doubt substandard. This is not, however, relevant
to the point made which is that Case in coordinate structures seems to be acquired
randomly.
To conclude this section I would like to claim that my proposal regarding the
syntactic status and internal structure of Small Clauses is maximally simple. It does
not require any additional complicated machinery in the theory, nor does it require
the existence of marked options. Furthermore, it avoids all the problems we
encountered with the earlier proposals.
In this section I would like to discuss the implications of two papers, Pollock
(1989) and Chomsky (1991), for the treatment of Small Clauses presented above.
In these studies, dealing mostly with English and French, it is suggested that the
structure of the I-node is more complex than was previously assumed in GB-
theory. Pollock (1989:379) proposes the following structure for CPs (irrelevant
details are omitted):
(646) CP
C TP
NP TP'
Τ NegP
Neg AgrP
Agr VP
(Adv) V
A new analysis of Small Clauses 187
The differences between (646) and (647) are the following: in Pollock's proposal
subjects of clauses are positioned in Spec-of-TP, for Chomsky they are positioned
in Spec-of-IP. Furthermore, in Chomsky's framework there are two Agreement
positions: AGR-S which regulates subject-verb concord (this node corresponds to
"I" in the Barriers framework), and AGR-0 which takes care of agreement between
verbs and their objects. The latter type of agreement is illustrated in the following
sentence:
Chomsky assumes that the participle repeintes moves to AGR-0 and that the Wh-
phrase combien de tables, before moving to Spec-of-CP, adjoins to the AGR-O
Phrase. Here it will be in a government relation with the participle, thus establishing
agreement between itself and this verb.
It is not immediately obvious how my proposal that SCs are IPs fits into either
of the two frameworks discussed here. The Structure of IP is very much an area of
ongoing research (cf. Iatridou 1990; Baker 1991; Chomsky 1992; Chomsky—Lasnik
forthcoming), so what I will have to say below will be speculative.
We might assume, as an anonymous reader suggests, that Small Clause IPs
188 The Small Clause node
correspond to Pollock's AgrP (Chomsky's AGR-O Phrase). This seems unlikely for
two reasons. Firstly, SCs would lack an S-Structure subject position. Spec-of-VP
is a possible subject slot in (646), but only at D-Structure. Furthermore, I have
argued that Small Clauses are untensed and should therefore have a position where
the [-tense] feature can be located. In Pollock's analysis that position would be "T".
We are thus led to the possibility that SCs are TPs. However, Pollock assumes TPs,
but not AgrPs, to be inherent barriers (Pollock 1989: 397). This poses no problems
for material that is extracted out of L-marked SCs, as in cons/der-constructions, but
it does pose problems for elements extracted out of adjunct SCs (which are not L-
marked), as in αρ/joini-constructions. Such elements would be crossing more than
one barrier when movement takes place to Spec-of-CP (namely the SC (=TP) itself
and the matrix TP). We can overcome this complication by assuming that the
inherent barrierhood of TP is voided in some way or other. One possible way of
achieving this is allowing for adjunction to TP in such a way that when material
is extracted it never crosses more than one segment of this category (cf. the
discussion of adjunction to VP in section 2.3.4). Pollock (1989: 406, footnote 38),
citing Frampton (1987), in fact mentions this possibility. Alternatively, we can
follow Chomsky's suggestion that IPs are structured as in (647). We can then
extract subjects from complement SCs freely. However, under Chomsky's proposal
subjects and predicates moved from adjunct SCs and predicates moved from
complement SCs (cf. chapter five) cross a barrier, namely TP. In the Barriers
model movement across one barrier is permitted, though this normally results in
decreased acceptability. As subject and predicate extraction from adjunct SCs and
predicate extraction from complement SCs doesn't result in decreased acceptability
(cf. Who, did you appoint [t, Senior Producer?]/How bright yellow, did you paint
[the wall tj? How generous, do you consider [him tj?) we are again led to the
possibility that adjunction to TP is a plausible option.
Summing up, under the split INFL hypothesis Small Clauses are either TPs, as
in (646), or IPs, as in (647).
Although verbal Small Clauses are outside the scope of this book, I cannot here
totally ignore them, as the proposals I have made in this chapter have consequences
for their analysis. What I will have to say about them is speculative.
There are three types of construction that are of interest, namely those in
(649)—(651):
In (649) the matrix verb is followed by an NP and a verb with an -ing ending, in
(650) by an NP and a bare infinitive, and in (651) by an NP and an -ed form.
Researchers have pointed out problems for a Small Clause analysis of these
sentences. It could be that none of them in fact involve SCs. Reuland (1983)
discusses [V NP -ing] constructions and concludes that their [NP -ing] substrings
are tensed S'-constituents. Williams has discussed sentences of the type in (650),
and rejects an SC-analysis of them (see section 2.3.3). One of the reasons why the
construction is problematic is that its passive counterpart contains the element to
(cf. She was made to laugh). Hornstein—Lightfoot (1987: 51) remark that "[o]f
course, this is an old problem to which there has been no satisfactory solution".
For the sake of argument let us assume that an analysis in which (649)—(651)
contain Small Clauses is correct. One good reason for taking such a position is that
the postverbal NPs in each of these sentences are not direct objects. That is, they
are not assigned θ-roles by the matrix verbs. This can easily be demonstrated by
citing sentences like (652):
It would seem, though, that verbal Small Clauses should receive a different
treatment from nonverbal SCs as far as their internal structure is concerned. The
reason is that the two types of SC differ in at least one important respect: in
nonverbal Small Clauses there is a copular relationship between the subject and the
predicate. For this reason I posited a phonetically null copular verb BE in SCs (cf.
(623) above). There is no such relationship in verbal Small Clauses. This being so,
the picture that emerges is one in which we have two classes of Small Clauses: a
class of copular SCs (i.e. nominal, adjectival and prepositional SCs) and a class of
verbal SCs (i.e. those in (653)—(655)).
6.4. Conclusion
In this chapter I have critically examined the proposals that have been made in the
literature regarding the categorial status and internal structure of Small Clauses. I
have shown that each of these proposals is problematic in one way or another.
These problems can be overcome if we analyse SCs as IPs.
Chapter 7
Conclusion
In this book I have examined the properties of nonverbal Small Clauses in English
making use of the Government-Binding framework of Chomsky (1981, 1986b) and
related work.
I have made and defended a number of claims. The most fundamental of these
is that Small Clauses exist as syntactic constructs. Various researchers have already
put forward some of the evidence for this contention, but here for the first time this
evidence is brought together. I have furthermore presented some new evidence
bearing on this issue.
Another claim I have made is that Small Clauses can appear in two syntactically
different environments: in complement positions and in non-complement positions.
Complement SCs are assigned a θ-role by a verb from a group of verbs which I
have called the consider-cla&s or by a preposition in so-called absolute
constructions. Non-complement Small Clauses are not assigned a θ-role by a
subcategorising element. They can appear in four positions. The first of these is a
position which is Chomsky-adjoined to VP, the head of which is a member of a
group of verbs which I have called the appoint-class. The second position is at
sentence boundaries; that is, preceding or following the matrix clause (e.g. Naked,
Jill ate the meat/Jill ate the meat, naked). Small Clauses of this type have a PRO
subject and function as sentence adjuncts. They can also appear in a position
adjoined to VP, but they then have a slightly different interpretation. A third
possible position for non-complement SCs is the subject position, as in [The
Yorkshire Ripper free] poses a threat to society. Here the subject-SC is assigned
a θ-role by the matrix clause predicate. This particular sentence refutes the claim
repeatedly made in the literature that Small Clauses may only appear in subject
position in copular and Raising environments. Finally, again contrary to claims
made by some researchers, Small Clauses may occur independently, as in the
following sentences: The room was extremely noisy: children shouting, [the tv on],
[the record player on], and little Jimmy kicking the cat! [Hands up!]/ [PRO
closed], I have referred to constructions containing verbs from the consider-class
as instances of dependent predication and to constructions containing verbs from
the appoint-class as instances of controlled predication.
In chapter four I discussed a number of special constructions which involve, or
at least at first sight appear to involve, Small Clauses. The first of these
constructions concerns a group of verbs which have traditionally been called phrasal
verbs. I argue that these do not exist. Instead we should recognise the existence of
two classes of verb-preposition construction: a class that involves a group of verbs
that subcategorises a prepositional Small Clause (e.g. switch in I switched [sc the
light on], where the element on is analysed as an intransitive PP), and a class that
involves a group of verbs that does not subcategorise an SC, but instead
192 Conclusion
This condition turns out to have great explanatory force, not only in chapter four
in the analysis of verb-preposition constructions, where it is first proposed, but also
in chapter five, where there is a more detailed discussion of extractions from Small
Clauses.
Conclusion 193
I have argued that the acceptability of these sentences decreases and that this is due
to increasing processing difficulties. Thus, (656) which involves NP-Movement
from a subordinate clause containing an intransitive verb poses no particular
problems. (657) and (658), by contrast, are slightly more difficult to interpret
because NP-Movement takes place from a subordinate clause containing a transitive
verb and from a clause containing a passive verb respectively. The unacceptability
of (659) and (660) can be explained by conjecturing that it and there, semantically
empty elements, are initially interpreted wrongly by hearers; the pronoun as
referential it, and there as an element which introduces an existential main clause.
All these sentences are syntactically well-formed but pragmatically deviant to some
extent. Sentence (660) especially has puzzled quite a few linguists who have tried
194 Conclusion
I hope to have shown that Small Clauses have an important role to play in the
syntactic description of English. Their existence should have consequences for other
areas of linguistics too (on this see Aarts, forthcoming). Small Clauses will
probably remain controversial for some time to come, but they are undoubtedly here
to stay.
Appendix
Unless otherwise indicated the phrase the corpus in this book refers to the following
set of texts (each containing 5,000 words), all from the Survey of English Usage
at University College London. The S-texts (20 in number) are spoken. The W-texts
(also 20 in number) are written. The total number of words is 200,000.
S-texts W-texts
Like the Bermuda Triangle, X-bar Theory has been claimed to exist in a
number of widely-read works, and credited with explanatory power,
198 Notes
without ever being delineated with much precision, and thus is generally
not referred to at all in woiics dealing with a more technical orientation.
(Pullum 1985: 323)
However, in recent years most linguists have started making use of the
standard version of X'-theory which is discussed here.
12. On pleonastic it see also Bennis (1986) who has claimed that it is referential.
13. The definition of governing category is in fact somewhat more complex, but
for present purposes the definition in the text will do.
14. In the work of R.S. Kayne double object constructions and particle
constructions are also analysed as SCs:
See Kayne (1984a, 1984b). The second of these two constructions will be
discussed in detail in section 4.2.
15. In fact, he has PrtP (Participle Phrase) instead of VP; the lower bracketed
string is not labelled. For further discussion of the verbal Small Clause
constructions in (52)—(54), see section 6.3.3.
16. For a definition of c-command see section 2.3.1.
17. Williams gives no clear examples of S' as predicate.
18. Emonds (1985: 36) has a similar analysis, without the coindexing.
19. Here again, Williams does not define the exact meaning of a new term he
introduces. All the reader is offered is the following:
The use of the notion theme is not critical here. In the worst possible
case, it will be necessary to specify which NP a VP-dominated predicate
modifies. In this worst case, theme is being used as a purely diacritic rule
feature. In a large number of cases, though, theme seems to give the
correct answer, at least to the extent that the notion theme is clear in the
first place. (Williams 1980: 207)
20. PXPs are predicative phrases of the type XP. For example, in Pete considers
Jim foolish the PXP is an AP.
21. Cf. Williams (1983: 290):
22. If there is movement across two barriers the resulting sentence is much less
acceptable than in those cases where only one barrier is crossed.
23. In following chapters the level of V' will only be shown if it is relevant to
the discussion.
24. Various researchers have proposed that the D-Structure position of subjects
is Spec-of-VP. From there they move to Spec-of-IP at S-Structure. See e.g.
Contreras (1987), Kitagawa (1986), Sportiche (1988) and especially
Koopman—Sportiche (1990). For our purposes the D-Structure position of
the subject is of no importance and I will assume that subjects are generated
in Spec-of-IP.
25. The Times, 24 November 1987.
26. We do of course find sentences like (i) where it appears to occur in direct
object position:
(i) If it wasn 7 for Tom's testimony, I wouldn 't have believed it that Jay hit
Harry.
The it we have here is also a dummy element However, we can account for
(i) as follows: first a θ-role is assigned to the direct object r/wr-clause before
it is extraposed. Then, after extraposition, the dummy pronoun it is inserted.
See also chapter 5 of the present work.
27. For example, the Russian for Ivan is a student is Ivan student, see e.g. Kayne
(1984a: 135).
28. In fact (97) is ambiguous: open could be taken to be a verb as well as an
adjective.
29. The Independent, 29 September 1988.
30. In this book all sentences with a classification number preceded by "W"
(written text) or "S" (spoken text) are taken from the Survey of English
Usage. A list and short description of the texts that were used is given in the
Appendix.
31. The Independent, 7 January 1989.
32. Incidentally, it would seem that the string daughter of a well-to-do and
rigidly conventional family in (108) is perhaps also best regarded as a Small
Clause. It would be another example of an SC which does not occur in
complement position and would be analysed as in (i):
The empty pronominal subject refers to the subject she of the matrix clause.
33. Both (112) and (113) are taken from The Independent of 23 February 1989.
34. Mee is the postposition form of met.
200 Notes
35. It should be noted that Kayne rejects an SC analysis for the bracketed string
(on the basis of sentences like *She was distressed at the thought of there
likely to be another war). He assigns it the structure [NP NP X].
36. See also Hornstein—Lightfoot (1987: 32, footnote 8) and Napoli (1989: 152).
In these studies Safir's claim regarding the distribution of HNPs is repeated.
Tim Stowell claims that SCs in subject position can only occur if the matrix
verb is identificational be (1988: 12).
37. The Independent 11 June 1988.
38. It has been suggested to me by Andrew Radford that the data in (120) are not
significant in view of sentences such as those quoted in (i) and (ii) below:
(i) Persistent abuses of civil rights was one of the main reasons why the
system was reformed.
(ii) Twelve roses is more romantic than seven.
However, notice that we could argue that the strings persistent abuses and
twelve roses are predicates too, just like angry about the pay in (120). This
observation is supported by the fact that we can paraphrase (i) and (ii) as in
(iii) and (iv), albeit somewhat clumsily:
(iii) There being persistent abuses of civil rights was one of the main
reasons why the system was reformed.
(iv) There being twelve roses is more romatic than there being seven.
39. See also Suzuki (1988: 61) who has independently made a similar
observation.
40. There are cases where elements such as perhaps and probably could be said
to be part of NPs. Consider the bracketed strings in (i) and (ii) which can be
argued to be NPs complementing a preposition in (i) and a verb in (ii):
(i) Wimbledon has developed into [perhaps the most prestigious tennis
tournament in the world].
(ii) We serve [probably the best coffee in town].
Witness the paraphrases in (129) and (131) in the text, however, perhaps and
probably in (127) and (128) do not occur in NPs. With regard to (i) and (ii),
notice that there arc reasons to believe that the complements of into and
serve are in fact reduced nominal relative clauses (cf. Wimbledon has
developed into [what perhaps is the most prestigious tennis tournament in the
world]/We serve [what perhaps is the best coffee in town]). If this is so, we
then have an explanation as to why perhaps and probably can appear in (i)
and (ii) inside what seem to be ordinary NPs, but are really clauses. For
further discussion see section 6.2.1.
Notes 201
53. The two most important reasons why the the strings in (215) and (216) are
regarded by Kayne as being syntactically identical in structure are the
following: the first is that their nominalised counteiparts in both cases are
ungrammatical, cf. (i) and (ii):
An example like this confirms that regard and verbs like it are indeed ECM
verbs.
84. A sentence like that in (i) is not a counterexample to the claim made in the
text:
Here the clause that he was angry is assigned a θ-role in its D-Structure
position following the verb describe. Then, after it has been extraposed, the
dummy element it is inserted. Thus, the claim that the position following
describe is a θ-position is upheld. There is more on Extraposition in chapter
five.
85. This sentence is of course grammatical if strike has the sense hit.
86. The Ρ is in italics as its syntactic status is at issue.
87. Frits Stuurman (p.c.) claims that sentence (i), which contains the veib
suspect, which is like accuse, is possible:
lowering rule. The same is true for Baltin's account. Radford suggests that
this problem is solved if we base-generate the subject clause in Spec-of-VP
and then raise it to the adjoined position in VP. The element it would then
have to be base-generated in Spec-of-IP.
105. Notice that when the third subcategorisation frame is selected Extraposition
does not take place. If this is correct adjustment of the figures in the table in
5.3.2.1 is necessary. There are then nine fewer cases of rightward movement.
106. There was one more instance of a construction like (526) in the corpus. In
the table in section 5.3.2.1 I have analysed both sentences as involving an
extraposed ίο-infinitive clause.
107. Where the PP from Boston is construed with people.
108. Strictly speaking, in passive appomi-constructions the Small Clause remains
intact. See note 88 and below.
109. Burzio (1986) extends the process shown in (542) to copular constructions.
In his analysis (i) is derived from the D-Structure in (ii):
112. There are a few apparent exceptions. Consider like and worth in He is much
like his brother, it is worth £10. See Huddleston (1984: 307). We can,
however, also analyse these elements as prepositions. Near, as in we came
nearer a solution, seems a real exception, as nearer takes the comparative -er
ending, and cannot therefore be analysed as a preposition.
113. The Independent Magazine, 18 February 1989.
114. Notice that the string We should have made the whole lot research assistants
can of course be a surface form if have is construed as an aspectual auxiliary.
In (580), however, have is a main verb.
115. We also have an empty operator in the lower clause of (i), the S-Structure/LF
representation of which is shown in (ii):
116. The term Small Clause, however, was first used in print much earlier by
Edwin Williams in the title of his 1975 paper "Small Clauses in English".
117. It should be noted that under the proposals of Abney (1987) these sentences
cease to be problematic. I will not go into this here.
118. I do not here want to go into the exact relation between (612)/(613) and
(616)/(617). Presumably (612) and (613) contain empty elements of some
sort.
119. Radford also regards [NP bare infinitive/-ing/-ed] strings in constructions
containing perception or causative verbs as Small Clauses.
120. Adjunction structure.
121. I am assuming here that the SC-subject is base-generated in Spec-of-IP, thus
remaining neutral with regard to the question whether perhaps it originates
in Spec-of-VP.
122. Contreras (1987) accounts for the facts in (631)—(634) in a way that cannot
be correct. For (632) he has the structure in (i) (following Stowell Small
Clauses are analysed as XPs headed by X'):
(i) CP
NP A' I VP
[+fin]
He assumes that the empty C is lexical and functions like the prepositional
complementiser for, but only if it is coindexed with a tensed "I", as is the
case here. "C" then assigns Case to the subject of the adjectival Small
Clause. However, notice that in (i) the Small Clause AP is a blocking
category as it is not L-marked. This means that IP is a barrier (by
inheritance). C cannot therefore govern the NP subject of the Small Clause
which must remain Caseless and thus violates the Case Filter.
123. A complication here is that the Minimality Condition is violated (Chomsky
1986b: 10). This condition states that an element a does not govern a phrase
c is there is a closer governor b. In (630) the lower "I" is a closer governor
for the subject of the SC than the higher "I". We can get around this problem
by stipulating that [-tense, +AGR] governors do not count for the Minimality
Condition.
124. Page references are to the manuscript version of this paper.
125. Perhaps also (i):
Aarts, Bas
1987 Complex transitive complementation revisited. [Unpublished
Master's dissertation, University of Utrecht]
1989 "Verb-preposition constructions and small clauses", Journal of
Linguistics 25 (2): 277—290.
1990a Small Clauses in English: The prepositional, nominal and adjectival
types. [Unpublished Ph.D. dissertation, University of London.]
1990b "Prevent-type verbs in a GB framework", UCL Working Papers in
Linguistics 2: 147—164.
1991 "Descriptive linguistics and theoretical linguistics: Some new
thoughts on a (still) uneasy relationship". [Paper presented at the
inaugural conference of the European Society for the Study of
English. To appear in the proceedings. D. Kastovsky (ed.)]
forthcoming "Applying Government-Binding theory: A case study", in: Ma
Adelaide Jurado Spuch (ed.).
Aarts, Flor — Jan Aarts
1982 English syntactic structures. Oxford: Pergamon Press.
Abney, Steve
1987 The English noun phrase in its sentential aspect. [Unpublished Phd
dissertation, Massachusetts Institute of Technology.]
Akmajian, Adrian
1984 "Sentence types and the form-function fit", Natural Language and
Linguistic Theory 2: 1—23.
Andersson, Evert
1985 On verb complementation in written English. (Lund Studies in
English 71.) Lund: Gleerup/Liber.
Anderson, Stephen R. — Paul Kiparsky (eds.)
1973 A Festschrift for Morris Halle. New York: Holt, Rinehart and
Winston.
Andrews, Avery
1982 "A note on the constituent structure of adverbials and auxiliaries",
Linguistic Inquiry 13 (3): 313—317.
Austin, Frances O.
1980 " Ά crescent-shaped jewel of an island': appositive nouns in
phrases separated by o f , English Studies 61: 357—366.
Bach, Emmon
1977 Review article of Postal (1974), Language 53 (3): 621—654.
Baker, Carl L. — John J. McCarthy (eds.)
1981 The logical problem of language acquisition. Cambridge MA:
Massachusetts Institute of Technology Press.
210 References
Bresnan, Joan W.
1970 "On complementizers: Toward a syntactic theory of complement
types", Foundations of Language 6: 297—321.
1972 Theory of complementation in English syntax. [Unpublished Ph.D.
dissertation, Massachusetts Institute of Technology. Published as
Bresnan 1979.]
1976 "Nonarguments for raising", Linguistic Inquiry 7 (3): 485—501.
1979 Theory of complementation in English syntax. New York —
London: Garland.
Brown, Keith — Jim Miller
1991 Syntax: A linguistic introduction to sentence structure (2nd edition).
London: HarperCollins.
Bunt, G. Η. V. — Eric S. Kooper — J. Lachlan Mackenzie — David R. M.
Wilkinson (eds.)
1987 One hundred years of English studies in Dutch universities.
Amsterdam: Rodopi.
Burton-Roberts, Noel
1986 Analysing sentences: An introduction to English syntax. London —
New York: Longman.
Burzio, Luigi
1986 Italian syntax. Dordrecht: Reidel.
Cardinaletti, Anna — Guglielmo Cinque — Giuliana Giusti (eds.)
1989 Constituent structure. (Papers from the 1987 GLOW Conference.)
Dordrecht Foris.
Chierchia, Gennaro — Sally McConnell-Ginet
1990 Meaning and grammar: An introduction to semantics. Cambridge
MA: Massachusetts Institute of Technology Press.
Chomsky, Noam
1957 Syntactic structures. The Hague: Mouton.
1964 "A transformational approach to syntax", in: Jerry A. Fodor —
Jerrold J. Katz (eds.), 211—245; also in Archibald A. Hill (ed.),
124—158.
1965 Aspects of the theory of syntax. Cambridge MA: Massachusetts
Institute of Technology Press.
1970 "Remarks on nominalisation", in: Roderick A. Jacobs — Peter S.
Rosenbaum (eds.), 184—221.
1973 "Conditions on transformations", in: Stephen R. Anderson — Paul
Kiparsky (eds.), 232—286. [Page references are to Chomsky 1977.]
1975a The logical structure of linguistic theory. New York — London:
Plenum Press. [MS, 1955—1956.]
1975b Reflections on language. New York: Pantheon.
1977 Essays on form and interpretation. New York: North Holland.
1980 Rules and representations. Oxford: Blackwell.
212 References
Emonds, Joseph
1972 "Evidence that indirect object movement is a structure-preserving
rule", Foundations of Language 8: 546—561.
1976 A tranformational approach to English syntax. New York:
Academic Press.
1984 "The prepositional copula as", Linguistic Analysis 13 (2):
127—144.
1985 A unified theory of syntactic categories. Dordrecht: Foris.
Fabb, Nigel
1984 Syntactic affixation. [Unpublished Ph.D. dissertation, Massachusetts
Institute of Technology.]
Fodor, Jerry A. — Jerrold J. Katz (eds.)
1964 The structure of language: Readings in the philosophy of language.
Englewood Cliffs NJ: Prentice Hall Inc.
Frampton, J.
1987 "Parasitic gaps and the theory of Wh-chains". [MS, Northeastern
University, Boston MA.]
Fraser, Bruce
1974 The verb-particle combination in English. Tokyo: Taishukan
Publishing Company.
Freidin, Robert (ed.)
1991 Principles and parameters in comparative grammar. Cambridge
MA: Massachusetts Institute of Technology Press.
Gazdar, Gerald — Ewan Klein — Geoffrey Pullum — Ivan Sag
1985 Generalised phrase structure grammar. Cambridge MA: Harvard
University Press.
Green, Georgia M.
1970 "How abstract is surface structure?", in: Papers from the sixth
regional meeting of the Chicago Linguistic Society, CLS 6:
270—281.
1972 "Some observations on the syntax and semantics of instrumental
veibs", in: Papers from the eighth regional meeting of the Chicago
Linguistic Society, CLS 8: 83—97.
1973 "A syntactic syncretism in English and French", in: Braj B. Kachru
et al. (eds.), 257—278.
Gu6ron, Jacqueline
1980 "On the syntax and semantics of PP extraposition", Linguistic
Inquiry 11 (4): 637—678.
Gu6ron, Jacqueline — Hans-Georg Obenauer — Jean-Yves Pollock (eds.)
1984 Grammatical representation. Dordrecht: Foris.
Haegeman, Liliane
1985 "The gei-passive and Burzio's generalisation", Lingua 66: 53—77.
1991 Introduction to government and binding theory. Oxford: Blackwell.
214 References
Halliday, Michael A. K.
1967 "Notes on transitivity and theme in English". (Part 1). Journal of
Linguistics 3 (1): 37—81.
Hayes, John R. (ed.)
1970 Cognition and the development of language. New York: J. Wiley
and Sons.
Hill, Archibald A. (ed.)
1962 Proceedings of the third Texas conference on problems of linguistic
analysis in English. Austin TX: University of Texas Press.
Hoekstra, Teun
1984 Transitivity: Grammatical relations in government-binding theory.
Dordrecht: Foris.
1988 "Small clause results", Lingua 74: 101—139.
Hornstein, Norbert — David W. Lightfoot
1987 "Predication and PRO", Language 63 (1): 23—52.
Huddleston, Rodney
1984 Introduction to the grammar of English. Cambridge: Cambridge
University Press.
Hudson, Richard A.
1988 "Coordination and grammatical relations", Journal ofLinguistics 24
(2): 303—342.
Huybregts, Rini — Henk van Riemsdijk (eds.)
1982 Noam Chomsky on the generative enterprise. Dordrecht: Foris.
Iatridou, Sabine
1990 "About Agr(P)", Linguistic Inquiry 21 (4): 551—577.
Jacobs, J. — A. von Stechow — W. Sternefeld — Τ. Vennemann (eds.)
forthcoming Syntax: An international handbook of comtemporary research.
Berlin: Walter de Gruyter.
Jacobs, Roderick A. — Peter S. Rosenbaum (eds.)
1970 Readings in English transformational grammar. Waltham MA:
Ginn and Co.
Jackendoff, Ray
1977 X'-Syntax: A study of phrase structure. Cambridge MA:
Massachusetts Institute of Technology Press.
Jespersen, Otto
1909—49 A modern English grammar on historical principles. London —
Copenhagen: George Allan & Unwin and Munksgaard.
1933a Essentials of English grammar. London: George Allen and Unwin
Ltd.
1933b "The system of grammar", in Jespersen 1933c.
1933c Linguistica: Selected papers in English, French and German.
Copenhagen: Levin and Munksgaard/ London: George Allen and
Unwin Ltd.
References 215
Phillipps, Κ. C.
1954 "Contamination in late ME", English Studies 35: 17—20. [Part II:
English Studies 37: 12—14.]
Pollock, Jean-Yves
1989 "Verb movement, Universal Grammar, and the structure of IP",
Linguistic Inquiry 20 (3): 365—424.
Postal, Paul
1974 On raising. Cambridge MA: Massachusetts Institute of Technology
Press.
Postal, Paul — Geoffrey K. Pullum
1988 "Expletive noun phrases in subcategorized positions", Linguistic
Inquiry 19 (4): 635—670.
Pullum, Geoffrey K.
1985 "Assuming some version of X-bar theory", in: Papers from the
general session at the twenty-first regional meeting of the Chicago
Linguistic Society, CLS 21: 323—353.
Pyles, Thomas
1982 The origins and development of the English language. (3rd edition.)
New York etc.: Harcourt Brace Jovanovich.
Quirk, Randolph — Sidney Greenbaum — Geofffrey Leech — Jan Svartvik
1972 A grammar of contemporary English. London: Longman.
1985 A comprehensive grammar of the English Language. London:
Longman.
Radford, Andrew
1978 "Agentive causatives in Romance: accessibility versus
passivisation", Journal of Linguistics 14 (1): 35—58.
1981 Transformational syntax: A student's guide to Chomsky's Extended
Standard Theory. Cambridge: Cambridge University Press.
1988a "Small children's small clauses", Transactions of the Philological
Society 86 (1): 1—43. Also in Research Papers in Linguistics 1:
1—38, University College of North Wales.
1988b Transformational grammar: A first course. Cambridge: Cambridge
University Press.
Reinhart, Tanya
1980 "On the position of extraposed clauses", Linguistic Inquiry 11 (3):
621—624.
Reuland, Eric J.
1983 "Governing -ing", Linguistic Inquiry 14 (1): 101—136.
Riemsdijk, Henk van
1978 A case study in syntactic markedness: The binding nature of
prepositions. Dordrecht: Foris.
218 References
Stowell, Tim
1981 Origins of phrase structure. [Unpublished Ph.D. dissertation,
Massachusetts Institute of Technology.]
1983 "Subjects across categories", The Linguistic Review 2: 285—312.
1988 "Small clause restructuring". [MS, University of California at Los
Angelis.]
Stuurman, Frits
1990 Two grammatical models of modern English. London — New
York: Routledge.
Suzuki, Torn
1988 "An analysis of small clauses and perception verb complements",
English Linguistics 5: 54—70.
Visser, Frans Th.
1973 An historical syntax of the English language. 4 vols. Leiden: Brill.
Wekker, Herman — Liliane Haegeman
1985 A modern course in English syntax. London: Croom Helm.
Whitney, Rosemarie
1982 "The syntactic unity of wA-movement and complex NP shift",
Linguistic Analysis 10 (4): 299—319.
1983 "The place of dative movement in a generative theory", Linguistic
Analysis 12 (3): 315—322.
Williams, Edwin S.
1975 "Small clauses in English", in: John P. Kimball (ed.), 249—273.
1980 "Predication", Linguistic Inquiry 11 (1): 203—238.
1983 "Against small clauses", Linguistic Inquiry 14 (2): 287—308.
1984 "77iere-insertion", Linguistic Inquiry 15 (1): 131—153.
Index
INFL (Inflection) 14—16, 176, 180f, 186f Lexicon 13, 14, 16—17, 22—23, 49, 58,
Inflection Phrase (IP) 31 64, 67, 70, 151—153
Informational weight 87, 134, 137 LF
see also Weight see Logical Form
Inherent barrier LGB
see Barrier see Lectures on Government and
Intervening NP 1 Binding
Intransitive preposition 81 Light XP 84, 133—134
Intrinsic barrier see also Weight
see Barrier Lightfoot, D. 11, 12, 30, 200 note 36, 49,
IP 52, 171, 173, 177—180, 189
see Inflection Phrase Logical Form (LF) 13,14, 17, 26—29,49,
//-Replacement 197 note 6 163, 206 note 115
— component 13
Jackendoff, R. 16
Jaworska, E. vii Mad Magazine Sentences 38—39
Mair, C. 12
Jespersen, O. 36, 39,43, 65,91, 103—105 make
— as a Raising verb 161
Kahn, J. 102 make...clear 151
Kayne, R. 199 note 27, 43, 47, 75—77, make...ready 153
83, 91 make...sure 151, 153
Kempson, R. vii, 204 note 81 make^ 153
Kiparsky, P. and C. 10 make5 161
Kitagawa, Y. 199 note 24, 171—173, make6 162
176—177, 180, 181 Mallen, Ε. 50, 53, 54, 57
Kluender, R. 171, 178—180 Manzini, Μ. R. vii, 171, 172, 175, 176,
Koopman, H. 199 note 24 180
Koot, H. van de vii Matthews, P. 35
Kruisinga, E. 40 McCawley, J. 61, 62, 91, 94, 143, 169
L-marking 33 McCloskey, J. 171, 178—180, 183
Landing site 139 McConnell-Ginet 197 note 3
— of extraposed SC subject clauses MDG
139f see Modern Descriptive Grammar
see also Extraposition Meat-eating construction 49
Larson, R. 202 note 59 Meillet, A. 21
Lasnik, H. 21, 35, 154, 187 Mental organ 13
Lectures on Government and Binding Meyer, C. 186
(LGB) 9 Miller, J. 35
Leftward Extractions 129, 131, 160f Mirror of the mind 13
— Passivisation 131, 160 (mis)take...for 122—123, 128, 192
— Raising 131, 160f Modern Descriptive Grammar (MDG) 2f
— Topicalisation 131, 160, 168 Move (a) 17, 84
— Wh-Movement 131, 160 Movement of NPs
Leftward movement of predicates 129, — Heavy-NP-Shift 131, 132
167f — Movement of other NPs 131, 133f
Levin, B. 201 note 45
Index 225
PRO 20, 24, 39—42, 49—50, 52—54, 67, "Remarks on Nominalisation" ("Remarks")
110, 128, 162, 168, 177, 184, 191, 192 5
theorem 20 Resultative predicate
Projection Principle 197 note 3, 17, 23, see Predicate
94—95 Reuland, E. 189
Pronoun 10, 19, 43, 83, 84, 128, 141 Riemsdijk, H. van 77
— Substitution 10 Rightward Extractions 129—131, 132f,
see also Binding Theory 157f
Proper government Rigter, B. 2
see Government Riley, K. 186
PS rules Roberts, I. 50—51, 53, 57, 66
see Phrase Structure rules Rosenbaum, P. 10, 12, 91, 97
Pseudoclefting 50, 66, 142, 145 Ross, J. 197 note 6, 34, 112
Pullum, G. 197 note 11, 38, 91, 92, 100 Rosta, A. vii
Pyles, T. 102 Rothstein, S. 21, 30, 39, 57—61, 63, 65,
68
Quantifier lowering 28 Rule of Predication
Quasi-argument 81, 121, 205 note 95 see Predication
Quirk et al. 2, 35, 91, 204 note 80, 112, Rules
126, 127, 146, 148, 159 — of anaphora 14
— of control 14
R-expression — of deletion 14
see Referential expression — of quantification 14
R-pronoun 43 — phonological 14
R-triggers 10
Radford, A. vii, 16, 29—32, 38—39, 200 S'-deletion (S-bar deletion) 20
note 38, 4 6 - ^ 7 , 75—76, 81—82, 202 S-selection
note 63, 85—87, 93, 98, 204 note 77, see Semantic selection
104, 122, 205 note 95, 144, 155, 159, S-Structure 14, 17
161, 169, 171, 173, 175, 178—180, Safir, K. 43—45, 78
185, 207 note 125 Schachter, P. 68f
Raising 10—12, 21, 44, 45, 201 note 49, Schein, Β. 21, 30, 57
91, 92, 94, 111, 130, 131, 160—162, Schmerling, S. 93—94
169, 191 Secondary predicate
— to-Object 10—12, 91, 92, 94, 111 see Predicate
— to-Subject 10 Semantic selection (S-selection) 23
Rappaport, M. 201 note 45 Sentence adverbs 45, 174
Reanalysis 76, 82, 85—87, 205 note 97, SEU
151, 153 see Survey of English Usage
see also Complex verb; Incorporation, Simpson, J. 58
Referential Expression (R-expression) 19 Single bar category 5, 16
see also Binding Theory see also X'-theory
Regard-type verbs 112f Small Clause (SC) 1
Regular XP 84, 87, 134f — announcement 39
see also Weight — as adjunction structures 172, 180
Reinhart, Τ. 142 — as adjuncts 39
Index 227
wwimj-contraction 4—5
want 10, 21, 24, 29, 73, 87, 98, 100
Weight 83—84
— condition on derivations see
Condition (259)
Weinberg, A. 30
Wekker, H. vii, 2, 35, 65
Wh-Movement 15,32, 63,130—131, 160,
167—168, 172—173
Whitney, R. 83—84
Wilder, C. vii
Wilkins, W. 21, 30, 53, 57
Williams, E. 14, 21, 25—30, 35, 57, 169,
207 note 116, 173, 176, 189
Wilson, D. 137
X-bar
see Single bar category
X'-Theory 14, 16
Zero INFL node 177