Syntax, More or Less

Author(s): John Collins

Source: Mind , Oct., 2007, New Series, Vol. 116, No. 464 (Oct., 2007), pp. 805-850
Published by: Oxford University Press on behalf of the Mind Association

Stable URL: https://www.jstor.org/stable/30166512

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

and Oxford University Press are collaborating with JSTOR to digitize, preserve and extend
access to Mind

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
Syntax, More or Less
JOHN COLLINS

Much of the best contemporary work in the philosophy of language and content
makes appeal to the theories developed in generative syntax. In particular, there is a
presumption that-at some level and in some way-the structures provided by syn-
tactic theory mesh with or support our conception of content/linguistic meaning as
grounded in our first-person understanding of our communicative speech acts. This
paper will suggest that there is no such tight fit. Its claim will be that, if recent gener-
ative theories are on the right lines, syntactic structure provides both too much and too
little to serve as the structural partner for content, at least as that notion is generally
understood in philosophy. The paper will substantiate these claims by an assessment
of the recent work of King, Stanley, and others.

1. Introduction

d-structure, s-structure, and LF just don't exist.

(Chomsky 2004, p. 152)

Conceptually, we don't want to seat interpretive motivations in the driver's

seat of our syntactic car.
(Lasnik and Uriagereka 2005, p. 154)

Much of the best contemporary work in the philosophy of language

and content makes appeal to the principles and results of generative
syntactic theory. In particular, there is a presumption that-at some
level and in some way-the structures specified by syntactic theory
mesh with or support our conception of content/linguistic meaning as
grounded in our first-person understanding of our communicative
speech acts. This paper will suggest that there is no such tight fit. Its
claim will be that, if recent generative theories are on the right lines,
syntactic structure provides both too much and too little to serve as the
structural partner for content, at least as that notion is generally under-
stood in philosophy. This is not to suggest, of course, that syntax offers
no structural contribution to linguistic meaning; the suggestion, rather,
is that the contribution from syntax is by no means straightforward and
is certainly not clearly reflected in the kind of structure that is familiarly
taken to constitute propositional content. On this view, the philoso-

Mind, Vol. 116 . 464 . October 2007 CC Collins 2007

doi:10.1093/mind/fzm805

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
806 John Collins

pher's content, as it were, is the result of a massive cognitive interaction

effect as opposed to an isomorphic map onto syntactic structure. Piet-
roski (2005b), after Chomsky, has advanced some intriguing ideas in
this area. My purpose here is not to say much at all about the positive
side of this picture, but to argue for the negative side of the thesis-that
syntax fails to match up with content in a principled way. This situation
is not the least problematic for syntactic theory itself, for it is under no
obligation whatsoever to offer any structure that may partner our intu-
itive notion of a proposition. Properly to establish this claim would
involve, just as a preface, the Herculean task of surveying the relevant
work in syntax, much of it unfamiliar to philosophers, and the many
varied philosophical appeals to syntax made over the past forty years.
Space demands a different method. The paper will focus on what I take
to be the most sophisticated and philosophically fecund appeals to
syntax-the work of Jeff King, Jason Stanley, and Zoltan SzabO-and
show thereof that the mismatch between propositional and syntactic
structure is, at the very least, cause for serious concern; that is, I shall
invite you to generalize from the best cases.
Before we begin in earnest, some general points of orientation will be
useful. First off, my targeting of King et al. should not be read as a sug-
gestion that their work is peculiarly confused; quite the contrary. It is a
virtue of their work that it may be confronted with research from a
related field. Such vulnerability stands in happy contrast to nigh-on all
of the 'metaphysics of content' literature, which, much to its detriment,
eschews the very question of how content is linguistically encoded. Sec-
ondly, I take all relevant parties to be minimally naturalistic in the sense
of holding to the principle that philosophical elaborations of content
impose no a priori constraint on how syntactic theory will or should
develop. Indeed, as we shall see, much of the attraction of generative
syntax is precisely that it is perceived to have converged on the same
point as the philosophers; the syntactic results, therefore, might be
properly employed as evidence for the philosophical claims. Thirdly and
much more prosaically, I shall take the philosophers at their word in
their explicit commitments to current or the 'best' syntactic theory in
the generative camp. Many of the questions I shall raise, however, are
independent of particular syntactic frameworks, and if the picture
looks bad for the philosophers under the framework I shall employ, it is
far from obvious that it would look better under the numerous other
extant frameworks. Besides, it is the responsibility of philosophers who
make claims about syntax to confirm that it is as they assert. Fourthly,
for present purposes, the syntactic principles to which I shall appeal

Mind, Vol. 116 . 464 . October 2007 @ Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
 Syntax, More or Less 807

need not be correct. My claim is conditional: if current syntactic theory

is on the right lines, then it provides too much and too little to support
or structure content. I am simply presuming the truth of the anteced-
ent. If my reasoning is valid, then the philosophers have but two
options: radically alter their conception of content to fit the syntax,
or-being minimally naturalistic-come up with a different syntax.
Syntactic structure is not a mere excrescence for King, Stanley, et al., it
is integral to their claims; absent the syntax and their claims become
indistinguishable from much of the 'competition'.
Finally, the issue of the relation between natural language syntax and
propositional structure has dominated much of philosophy for the past
century or so. It could be said that modern philosophy of language was
born in the realization that the structure of the proposition is essen-
tially logical in some sense rather than linguistic, for natural language
syntax appears to be 'systematically misleading' as to the meanings sen-
tences express. Cutting a long and complex story short, the leading con-
temporary diagnosis of this traditional thought is that it laboured
under a conception of syntax that was too much in the thrall of how
sentences appear to be structured. By positing various 'hidden' levels of
structure, generative linguistics can be understood to have at least
established the possibility that meaning is indeed linguistically struc-
tured. In other words, the traditional error-the original, albeit very
fruitful, sin-was to think that there is no more to syntax than the 'sur-
face' organization of words. Besides, whatever one takes propositions to
be, they are things that are expressed by sentences, so we do not want
the pairing of the two to be ultimately mysterious. In a sense, then, my
position is conservative just to the extent that I am arguing for a mis-
match between linguistic and propositional structure, much as Frege
and Russell argued. This, concord, however, might be misleading. I
happily accept-indeed, insist upon-the broad generativist claim
that syntactic structure is not exhausted by 'surface structure'; further-
more, syntactic structure is open to empirical discovery and should
constrain our semantic inquires. In this light, while I accept the Frege-
Russell conclusion that there is a mismatch, the consequence I draw is
diametric to theirs. Far from natural language being relegated in our
semantic inquiries, we need to think again about the theoretical sali-
ence of notions such as proposition and truth conditions. Put in radical
terms, it is the very Frege-Russell conception of meaning as proposi-
tional (as truth conditions) that is at fault. Given the empirical mis-
match and the need to explain how meanings are paired with
structures, we should not take our inherited ideas to be a constraint on

Mind, Vol. 116 . 464 . October 2007 CC Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
808 John Collins

our linguistic theorising. In short, as revealed by generative inquiry, it

looks as if language narrowly construed is just not in the business of
expressing propositions.
In the first part, I shall consider the claim that syntactic structure is
the actual structure of content. In a series of articles and a recent book,
King has argued that it is. My riposte will be that there is just far too
much structure in the syntax for this to be plausible. If we keep syntax
constant, the only avenue open to King and others will be to reshape
content to fit the syntax. It will be argued that following this course has
some unwelcome consequences, at least if one still wishes to trade in an
idea of propositions.
The second part will assess the claims of Stanley and others that the
syntax provides covert variables to serve as the bearers for all constitu-
ents of linguistic content. My riposte will be that the syntax provides
too little structure for this to be true. In simple terms, the relevant vari-
ables appear to be absent from the syntax. So, if Stanley wishes to retain
his semantics and have syntactic support for it, he must hold out for a
different syntactic approach than the one that currently prevails in the
generative field.

2. Too much syntax: the structure of the proposition

It is commonly assumed by philosophers of language that their ulti-
mate business is an account of propositional content: the p of the
schema

(PC) Sentence S means that p.

This conception is inherited from Frege and Russell and continues to

animate contemporary work in the philosophy of language under the
guise of capturing the explicature (the what is said of an utterance).

(EPC) Sentence S as uttered by U in context C expresses the proposi-

tion that p.

The orthodox view here is that the composition of S (inclusive of any

explicit context-sensitive items) determines explicature, with other
information effects being provided by C (e.g. Grice 1989; Soames 2002;
Cappelen and Lepore 2005). My present concern is not with the many
`contextualist' rivals to this orthodoxy, but with an internal problem for
the view, namely, that it faces the question of what is the S such that it
might determine p. Some philosophers have turned to generative lin-
guistics in the hope that it might shed light on the nature of proposi-

Mind, Vol. 116 . 464 . October 2007 CC Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
 Syntax, More or Less 809

tions via an independent account of the nature of sentences. The

attraction of generative linguistics might be elaborated as follows:

(i) The proposition a sentence expresses (relative to context etc.)

constitutes the sentence's meaning, which in turn determines
the assignment of truth conditions.'

(ii) Propositions, however, pose a problem. The structure a sen-

tence possesses such that it expresses the proposition it does is
not encoded on the sentence's surface

(iii) If, then, we take a proposition to be made up of elements such

that the sentence that expresses it is true just when those ele-
ments are structured so as to express the truth conditions of the
sentence, we are left in search of what structure that is, for the
'surface' sentence does not provide it.

(iv) Generative linguistics posits a distinct level of syntactic struc-

ture that encodes the structure of a sentence such that it ex-
presses the proposition it does.

(v) Therefore, syntactic theory promises to provide the structure of

the proposition, for that structure is none other than the un-
derlying form of a sentence such that it means what it means,
that is, such that it possess the truth conditions it possesses.
Thus, a proposition is the range of values of the lexical items of
a syntactic structure that are organized just as the lexical items
are. 2

This position is most closely associated with Jeff King; we shall also see
that Stanley shares much the same conception. Many others, less
directly concerned with the ontology of propositions, also make the
presumption that, at some level, syntactic structure encodes the organ-

The phrasing here is intended to be ecumenical between various conceptions of the relations
between sentences, propositions, and truth conditions. On one view, there is a two stage process,
where a proposition is determined from a sentence (relative to context) and truth conditions are
then assigned to the proposition (e.g. King 1995). This dual process allows for distinct processes of
composition between proposition and truth conditions. For my purposes, this complication may
be sidelined (for discussion, see King and Stanley 2005). As we shall see, the problem I present is
one of an overdetermination of syntactic structure vis-is propositional structure, not underde-
termination between a sententially determined proposition and its truth conditions.

2 Many 'traditionalists' would reject this reasoning in toto, claiming that there is no structural
map between language and thought (cf. Fodor 2001; Schiffer 1987). For the purposes of this paper,
I shall ignore any such position. For the record, however, I share the view of the theorists with
whom I shall be concerned that natural language semantics is a serious pursuit and is not merely
an indirect way of thinking about thought, whatever that might be.

Mind, vol. 116 . 464 . October 2007 CC Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
810 John Collins

ization of sentence meaning.3 The basic thought can be crudely

depicted as follows:

(1)a. Bill sleeps

b. <Bill, property of sleeping>

c. [s [NP Bill] [vp sleeps]]

In (i)a, we have a simple sentence of monadic predication that

expresses the proposition that Bill sleeps, which, as depicted in (i)b,
involves just the two constituents, Bill and the property of sleeping. The
question is: how are these two constituents structured so that (i)a gets
to express (i)b? How, in other words, are we to understand the angled
brackets and comma? (1)c provides the answer: the structure of the
proposition just is the structure of the sentence. So, the proposition
expressed is that the object Bill refers to-Bill himself, as it were-
stands in a relation of instantiation to the property sleeps expresses-
the property of sleeping. The syntactic structure, then, simply provides
the structure to determine the required relation of instantiation to save
the putative proposition from merely being a list or realising a weird
structure wholly divorced from how we understand the sentence in
(i)a.
My substantive complaint against this proposal is that syntax pro-
vides far too much structure, much more than can be accommodated
as the values that determine a sentence's truth conditions, the proposi-
tion it expresses. This, further, is not a mere quirk of some construc-
tions: the surplus structure is generally exhibited and reflects the fact
that current syntactic theory is not in the business of providing struc-
tures that answer to philosophical conceptions of propositions. That, at
least, is my charge. Before substantiating it, a little excursus on syntax
will be useful.
The proposal under discussion presumes that syntactic theory arrives
at the same point, from a different direction, as the philosopher is aim-
ing at. Yet the history of generative linguistics presents a somewhat dif-
ferent picture. It is, for example, simply false that generative linguistics
has always posited a level of meaning-relevant syntactic structure-
'See King 1994, 1995, 1996, 2002, 2007; King and Stanley 2005. This work explicitly appeals to
propositions. What of those, such as Harman (1974), Higginbotham (1986), Larson and Ludlow
(1993), and Larson and Segal (1995), among others, who favour (interpreted) LF structures as en-
tering into the semantics of' ... believes ... ' and ' ... means ... '?I think the considerations below
also pose complications for such accounts. Certainly such accounts, as so far presented, make no
sense on the phase derivation model of syntax (see n. 4), but perhaps they can be somehow recon-
structed. The issues are highly complex and deserve a separate treatment that current space pre-
cludes. (I thank Gabe Segal (p.c.) for some clarification here.)

Mind, Vol. 116 . 464 . October 2007 CC Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
 Syntax, More or Less 811

'logical form' or 'propositional structure'. Chomsky's (1955-56/75,1957)

earliest theories had no level of structure that uniquely encoded all that
was relevant to 'meaning'. Instead, the grammars provided a T(ransfor-
mational)-marker that detailed the derivational 'history' of a structure;
the properties relevant to meaning were distributed across the marker.
In the early 19705, two levels of structure were posited that determined
meaning-relevant properties (see Chomsky 1972 and Jackendoff 1972).
In the present day, Chomsky and many other syntacticians eschew any
level of logical form; instead they propose theories in which, in a man-
ner similar to that of the earliest theories, the structure relevant to
meaning is distributed across a derivational history ofphases.4 My point
here is not to suggest that there is anything essentially wrong with a
level of syntax that encodes 'meaning', but only to point out that while
some generative theories have posited a unique level of logical form,
many others have not. Hence, there is nothing about the generative
enterprise as such that constitutively informs the philosopher's search
for propositional structure. Perhaps the 'best' syntactic theory should
answer the philosopher's demands, but there is little empirical reason
to think so and some empirical reasons to think not.
To continue the excursus with a technical point: philosophers often
elide many aspects of syntactic structure in their presentations. For
example, no one serious in the field believes that the structure of even a
simple sentence like (1)a is accurately represented by (1)c; (1)c is inade-
quate as regards both its architecture and its labelling. Two simple
points will suffice for our purposes. Firstly, the notion of a sentence
('S') has been effectively redundant since the early 1970s for good
empirical and theoretical reasons. For instance, as will be noted, the
structure in (1)c is exocentric, that is, it is marked as an S without its
S-ness, as it were, being determined by any of its constituents. This
For the development of level-free approaches, see Chomsky wooa, 2001, 2004, 2005a, 2005b;
Epstein et al. 1998; Epstein 1999; Epstein and Seely 2002, 2006; Uriagereka 1999, 2002. On such ap-
proaches, the syntax inputs to 'semantics' in cascades or phases, where at phase n+i, the structure
transferred to semantics at phase n is no longer accessible to operations or probes within the syntax
(a Phase Impenetrability Condition holds). Chomsky's working assumption is that CP and vP (with
full thematic assignment, thus not unaccusatives and passives) are phases, where the complement
domains (i.e. minus edges: SPEC-head configurations) of each are transferred, i.e. TP and VP re-
spectively. See the references above for disagreements over the size of phases. One may still speak
of `LF' on such approaches, although the notion simply designates an interface between syntax
and semantics, not a level of representation defined by its satisfaction of unique conditions; e.g.
one can no longer speak of LF as where binding theory holds or where QR applies, and so on. It
bears noting that the level-free approach, and minimalism generally, is a not a recent quirk. Con-
siderations of economy have been central to syntactic theory from the beginning, even if they have
not always shaped the theoretical developments. Minimalism is, in essence, simply an approach
which puts such considerations to the forefront of research. See Collins 2001; Chomsky 1998; and
Martin and Uriagereka 2000, for general discussions of economy.

Mind, Vol. 116 . 464 . October 2007 @ Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
812 John Collins

makes a sentence categorically distinct from a verb phrase, or a noun

phrase, or a prepositional phrase, and so on-these are all endocentric,
that is, they are x phrases because they contain a constituent that is an
x, which is said to 'head' the phrase. Empirically, furthermore, syntactic
structure looks to be nigh-on uniformly endocentric. We should like,
therefore, for the integrity a sentence has to flow from its constituents;
we do not want it to be a 'top down' a priori stipulation that just certain
structures will be sentences. Descriptively, there is no harm in the stip-
ulation, but we fall into the explanatory red if we take the shorthand
seriously. Secondly and correlatively, just as S is assumed, so tense and
agreement are ignored. Again, just for shorthand, no harm is done, but
when we include tense and inflectional properties we see the trouble for
S double. Not only is S floating above the structure as a stipulation, but
the tense/inflectional features do not project, that is, it is as if they were
merely members of a list rather than an integrated part of the structure.
The obvious solution, and the one adopted, is to ditch the idea of a sen-
tence and let tense/inflection project and be the head of what we think
of as a sentence, just as, for example, it is the verb hit that determines
that hit the ball is a verb phrase rather than a sentence or a noun phrase.
As we shall see, it is precisely this well motivated move in linguistics
that causes trouble for the philosophers.
From what has just been said, we might imagine the structure of (i)a
to be (2) (where IT' is a tense phrase and '4)' is the tense head being
'spelt out' as -s on the verb):

(2) [Tp Bill [T, 4) [vp sleep] ] ]

(2) appears to give us everything we want for the proposition, including

tense, yet it remains inadequate. According to theories of the past
fifteen years or so, every full clausal structure involves some movement
or copying, which produces a surplus of elements relative to property
instantiation of the kind apparently required for propositional struc-
ture. Rather than merely stipulate this principle or go through the
detailed syntactic argument, let us see if we can motivate the basic idea.
Consider the following three theses that are widely held in contem-
porary syntactic theory.5

5 See, for example, Lasnik and Uriagereka 2005 for a textbook account of these claims. The basic
idea that any sentence-like structure (featuring tense) will involve movement in fact follows from
the verb-internal subject hypothesis that has been discussed since the early 198os. For the classic
defence of the hypothesis, see Koopman and Sportiche 1991; for an overview of the hypothesis, see
McClosky 1997. See den Dikken 2006 for the hypothesis in a minimalist setting.

Mind, Vol. 116 . 464 . October 2007 0 Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
 Syntax, More or Less 813

(TA) Verbal theta assignment takes place in verbal domains.'

(TEN) Tense heads are merged post the formation of verbal domains
and carry uninterpretable features.

(MC) All uninterpretable features must be valued either by a match-

ing of features or by movement/copying.

(It is important to note that my argument does not require each of

these precise theses to be true; minimally, I only require syntactic 'sur-
plus' from the perspective of propositions. For example, in their differ-
ent ways, Kratzer (1996), Hornstein (2001), Borer (2005), and others
reject (TA) in favour of at least some arguments receiving their roles
outside of verbal domains, but, as we shall see, the arguments still move
and so create the problematic copies.) We can briefly explain these ideas
by way of our simple example.
First assume that each lexical item (roughly, word) carries inherent
features and unvalued features. The former are unaffected by syntactic
operations (Merge) that build structures. The latter are affected; they
acquire a definite value or content only when combined with other
items. It turns out that this distinction closely matches the distinction
between interpretable features (semantically significant), which are
inherent, and uninterpretable features (semantically empty, but poten-
tially morphologically significant), which are unvalued. Assume this
holds. In our example, then, the verb sleep assigns a single theta role,
that is, it requires an item to be interpreted as the agent of the sleeping.
According to (TA), this role can only be assigned to items that are dom-
inated by the VP that is headed by sleep. (2) contradicts (TA), for Bill is
not in the domain of the VP. Let the derivation of the structure of (2),
then, start with Bill being merged with sleep from which it acquires its
theta role. Now we need tense, T, which merges with the VP. According
to (TEN), however, T carries uninterpretable features that need to be
valued relative to other items (e.g. agreement features of number, per
son). Indeed, Bill carries such a feature (Case). According to (CH), such
features must be valued if the structure is to be interpretable at the
semantic interface and marked for morphological purposes (in other
words, valuing a feature relative to other features is a way of marking
the item as only interpretable (potentially) morphologically, not
semantically). If we have a match between (unvalued) uninterpretable

6 Theta roles are features constituents of a structure have in relation to verbs and other theta-
assigning items, such as prepositions. For example, in Mary hit Bill with a bat, Mary is the agent,
Bill is the patient (the thing affected), and the bat is the instrument. For our purposes, the details of
theta roles and their assignment may be happily sidelined.

Mind, Vol. 116 . 464 . October 2007 @ Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
814 John Collins

features on T and interpretable features on Bill, we can think of the later

as cancelling out the former (valuation is a way of marking an item as
only of morphological significance); thus, the uninterpretable agree-
ment features of person and number on T are cancelled by such inter-
pretable features (3rd person, singular) that are part of Bill (we can
assume here that Case on Bill is reflexively valued as well, marked as
nominative for morphology, if required, as in he rather than him). But
we have not finished yet, for T in English is strong, that is, it attracts
items into its domain in order to value a remaining uninterpretable fea-
ture (so-called 'EPP', a mere name for we know not what'). To make all
this somewhat clearer, we have the following derivation (where +1-
mark interpretable and uninterpretable features respectively and
underlining marks the valuing of uninterpretable features; the usual
abbreviations mark how morphology treats the items):

(3)a. Bill sleeps

b. [vp Bill{+lst, +sing, _Case} sleep]

c. [T' {+Pres, -3rd Per, -Sing Num, -EPP} [vp Bill{ + 3-rd, +sing, -Nom} sleep]

d. [Tp Bill{ +3rd, +sing, -Nom} [T' {+Pres, -3rd Per, -Sing Num, -EPP}
[VP <Bill{+3rd, +sing, _Nom}> sleep]

There are two crucial things to note here. Firstly, the derivation is driven
to value uninterpretable features, not to provide a propositional struc-
ture. This motive is semantic in nature, for an uninterpretable feature is
precisely a feature that has no semantic significance, but the structure
produced departs from what we have been imagining is required for
propositional structure. This is the second point. The copying of Bill
higher up the structure creates an item that looks to be surplus relative
to property instantiation; that is, the lower copy of Bill exhausts what
propositional structure appears to be interested in, namely, Bill is fixed
as the agent of sleep, the instantiater of the property of sleeping.' Fur-
ther, after the introduction of tense, we have a temporal dimension.
Still, Bill gets copied for a reason that is not recorded in the proposition,

7A common idea is that EPP is in fact reducible to Case valuation. This would be a nice result,
but would create complications for the simple picture I am presenting here. For our purposes, the
reason for the copying is much less important than the fact that the copying appears to lack the
'right' semantic significance.

Propositions might feature the same object twice, if we are dealing with a reflexive relation
(e.g. suicide). My present point is that the syntax appears to demand at least a pair of copies inde-
pendently of the conceptual adicity or reflexivness of the verb.

Mind, Vol. 116 . 464 . October 2007 CC Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
 Syntax, More or Less 815

as it were (perhaps Case valuation or EPP elimination). Of course, per

the first point, this is no problem or mystery at all from the perspective
of the syntax. The syntax operates to value uninterpretable features, and
copying is part of that mechanism, a displacement of items to meet
interface demands, not to create a propositional structure.
I have chosen this most simple of examples to show that the relevant
properties hold generally. A more complex example, however, will per-
haps make the philosophical points here less opaque. Consider a classic
raising construction (features and other properties of the derivation are
elided for expository convenience):

(4)a. Bill seems to love Mary

b. [vp love Mary]

c. [v v [vp love Mary]]

d. v+love [vp <love> Mary]]

e. [vp Bill [v v+love [vp <love> Mary]

f. [T, to [vp Bill [v v+love [vp <love> Mary]

g. [Tp Bill [T, to [vp <Bill> v+love [vp <love> Mary]]]

h. [vp seem [Tp Bill [T, to [vp <Bill> [v v+love [vp <love> Mary]]]

i. [T, (13. [vp seem [Tip Bill [T to [vp <Bill> [vv+love [vp <love> Mary]

j. [Tp Bill [T, 43. [vp seem [Tp <Bill> [T, to [vp <Bill> [vv+love
[vp <love> Mary] ]
Again, we see an item Bill copied (this time twice) so as to value unin-
terpretable features on functional heads-in order: infinite T to and
finite T -s. Semantically, such structure appears not to be required, for
the predicate seem appears to express the character of an experience
that Bill loves Mary (technically, seem assigns just the single theta role to
its TP complement; Bill does not do any seeming). The confirmation of
this is given by the paraphrase It seems that Bill loves Mary, where the
subject of seem is pleonastic and Bill is the agent of love.
The details of these constructions and movement/copying in general
are quite complex, but a generalization can be readily extracted: copy-
ing occurs after everything relevant to a propositional (truth condi-
tional) structure has been determined. But copying creates new
structure, and so there is more structure than appears to be demanded

Mind, Vol. 116 . 464 . October 2007 @ Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
816 John Collins

by the requirements of the encoding of structure relevant to proposi-

tions.9 Again, the crucial point is that, according to recent theory, copy-
ing is motivated by semantic requirements, that is, the valuing of
uninterpretable features on functional heads, such as Tense. So, all of
the structure is playing a semantic role in one sense, just not the sense
apparently required for the encoding of propositions.' It also bears
emphasis that the problematic copying is not a mere notational arte-
fact. If the reasoning I have sketched is correct, then all valuing occurs
in certain 'local' environments; thus, copying is simply a mechanism to
ensure that items stand in the right local relations to have their features
valued and so be acceptable at the interpretive interfaces. Put crudely,
an item cannot be in two localities at the same time unless it is copied.
I do not take the above reasoning to signal a refutation of any partic-
ular thesis. The apparent mismatch of structure, however, poses a seri-
ous challenge to philosophers minded to take syntax to encode
propositional structure. It is far from obvious that any match can be
properly elaborated. Indeed, as mentioned, Chomsky's recent work and
those of many others renders the dream null and void, as the syntax
proceeds in phases, with no level of structure designed just to encode
meaning relevant relations, much as the earliest theories proceeded. Let
us now look at some options how one might accommodate the syntac-
tic surplus.

2.1 King on syntax

King does not explicitly address the challenge posed above, but he has
made some recent remarks that might be taken to ameliorate the
problem.

9 This situation also threatens Stanley and Williamson's (2001) analysis of 'know how'. They ar-
gue that 'know how' is a species of 'know that' in part on the basis that the infinite complements of
complementiser how are propositional due to the presence of covert PRO in subject position (see
below for more on PRO). As far as it goes, this thought is accurate, but infinite complements, as
we are seeing, contain structure that looks redundant if the complement is simply taken to encode
or reflect propositions. For example:

(i) Bill knows [cp how [Tp PRO to [vp <PRO> swim]
(ii) Bill knows [cp how [Tp PRO to [ vp <PRO> v+ride [VP <ride> a bike]

1 Chomsky's (2001, 2004, 2005a, 2005b, 2007) speculation that semantic interpretation has a
dual interface would establish a 'functional' semantic role (independent of the syntax that operates
blindly) for every element of a structure, most typically an operator-variable relation (especially
see Chomsky 2007, p. 12). Serious complications abound. The semantic role of the later structure
is speaker/discourse-information orientated; further, the proposal is made in a level-free setting,
under which there would be no unitary proposition structured by the syntax, for there would be
no unitary structure. This, note, makes perfect sense if the later structure serves to situate the ear-
lier structure relative to speaker/discourse-information.

Mind, Vol. 116 . 464 . October 2007 CC Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
 Syntax, More or Less 817

King (2007, p. 47, n. 44) writes: '[syntactic] structure is likely consid-

erably more complex than I have represented it as being ... I don't
think this affects issues that concern me'. In the absence of an account
of the nature of the additional complexity, however, it is impossible to
judge whether the affect will be negligible or disastrous. I am suggesting
the affect is somewhere towards the latter pole, for King does present
syntax as realizing an instantiation function in the simple way pre-
sented above. If the syntactic structure of even the simplest sentences
involves structure via copying that goes beyond the instantiation
required of a proposition, then syntactic structure is always in surplus
as regards propositional structure. That is the problem.
King's sanguinity is explainable, I think, by remarks he makes on the
nature of syntactic relations:

If we assume ... that any two objects that are combined by Merge stand in
the same syntactic relation R, this means that in a given tree all we will ever
have are pairs of expressions standing in R ... This means that to give an ac-
count of what sentential relations are, all we need to do is say what R is.
(King 2007, p. 48)

King (2007, p. 48, n. 45) notes that in now standard treatments, syntac-
tic movement/copying is construed as a species of Merge (Internal
Merge'), so, again, copying or movement in itself does not raise any
additional worries for the very idea of syntactic structure doubling as
propositional structure. I think King's reasoning is correct as far as it
goes, and he is certainly correct to think that the ultimate question of
what syntactic relations really are need not be settled before we can
constructively employ the notion (King 2007, pp. 49-50). None of this,
however, speaks to my concerns. King (2007, p. 47) is concerned to
answer a Benacerraf-style problem of the overdetermination of propo-
sitional structure by the potential myriad of ways of understanding syn-
tactic structure. That is not my concern. However we ultimately
understand the notion of a syntactic relation, if current theory is on the
right lines, then syntactic structure realizes relations between items that
go beyond what propositions require. King's reasoning, as presented,
can only be an answer to this problem if his hypothesis is merely that
propositional structure realizes the same kind of relations as syntax
realizes. Yet that is not King's claim, for he explicitly holds that LF rep-
resentations map onto structured propositions (King 2007, p. 29);
indeed, they are 'identical' save for a surplus on the propositional side to
accommodate the same syntactic organization realising distinct instan-
tiation functions (King 2007, p. 38). Thus, King's hypothesis is not
merely that the same kind of relations are realized in language and

Mind, vol. 116 . 464 . October 2007 0 Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
818 John Collins

propositions, but that the very syntactic structure language realizes is

recruited to structure propositions, that is, the proposition a sentence
expresses (relative to context) is structured by the very sentence (as rep-
resented at LF). Thus, our problem of too much syntax remains, for
there is a fundamental mismatch between the two structures, regardless
of precisely what we take syntactic relations to be and regardless of the
correct observation that movement/copying is just another species of
Merge.

2.2 Biting the bullet

We have so far been assuming a fixed intuitive conception of what
propositions must be like if they are to play the role philosophers envis-
age for them. For some, at least, this might be an unwarranted assump-
tion. In a naturalistic spirit, the bullet might be bit; that is, granting
that syntactic theory provides too much structure to fit into the com-
mon notion of a proposition, we might be enjoined to let the notion of
a proposition float until anchored to our 'best' conception of the syn-
tax. The advice here would be to elaborate propositions on the basis of
our syntactic findings as opposed to seek to constrain syntax by way of
our inchoate intuitions as to what propositions must be. Thus, our
argument would be accepted in full, but its impact stymied, for we have
assumed a fixed notion of a proposition independent of syntax (cf.
King 2007, pp. loo--1). There are two problems with this gambit, one
motivational, the other empirical.
The very idea of a proposition brings with it a conception of truth
conditions-a proposition determines truth conditions, either directly
or by way of further composition (see n. 1) -and abstractness, that is,
propositions can be shared and communicated (Frege's 'common store
of thoughts')." If these minimal criteria are surrendered, then it
becomes opaque why one should talk about propositions. The position
becomes indistinguishable from a version of sententialism or, more
generally, propositional eliminativism. After all, I take it that the issue is
not merely semantic, a dispute about how to use the word 'proposition'.
The crucial aspect of the above considerations is precisely that the first
criterion mentioned is seriously threatened: if propositions are struc-
tured by syntax, then propositions appear to contain elements that play
no role in the determination of truth conditions. Any bullet can be
bitten, but if this one is, I cease to understand the position.

"One might take propositions to determine less than truth conditions, but such a complica-
tion does not by itself ameliorate the present problem, where we have too much structure, not too
little (cf. Soames's, 2005, change of mind from his 2002).

Mind, Vol. 116 . 464 . October 2007 CC Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
 Syntax, More or Less 819

One can imagine various options to make the bullet more palatable.
One thought might be that only one aspect of the syntactic structure
goes to structure the proposition, the rest is somehow periphera1.12 This
is an old demarcation. Certain aspects of a structure, such as theta
assignment, are determined initially, that is, no item is copied so as to
get a theta role, while other aspects, more discourse related, are deter-
mined later or by copying, such as scope, topic, focus, and elements of
mood (see n. io). This is one reason for the positing of levels within
syntax. But a syntactic derivation is not complete until a structure is
built that does not correspond to a structure apt to encode a proposi-
tion. Again, from the point of view of the syntax, the task is to value the
features that have mere morphological significance; the task is not to
deliver a structure for a proposition. I do not wish to be dogmatic here.
Syntax, as currently understood, does contain sub-structures that look
like propositional structures. The problem is to isolate them in a princi-
pled way-that is, to define a structure that is complete relative to syn-
tactic demands and the putative demands of propositional structure.
Such a demarcation is what looks to be unavailable. The problem is
exacerbated in the recent phase derivation model. Here, uninterpretable
features are valued in, following Chomsky's proposal, something like a
'propositional' structure (Chomsky 200oa, 2001), but it is not the prop-
ositional structure itself that becomes interpretable, it remains uninter-
pretable; it is only the complement of the phase head that becomes
interpretable. For example, if we look at a derivation of Bill loves Mary
at the stage [vp Bill love [vp <love> Mary] ], we have a 'propositional'
structure (a phase) with all theta roles assigned, but it remains 'incom-
plete' relative to interface demands: inter alia, the uninterpretable Case
of Bill remains unvalued. So, in effect, the only thing the derivation has
finished with is [vp love Mary], which, needless to say, is not the struc-
ture of a proposition. We have perhaps already ventured too far into
technicalities. The basic idea that syntactic structure has an inherent
duality, with one aspect dealing with propositional structure, is not a

12 A natural way of substantiating this idea would be to take propositions to be the values of a
function from S*+X, where `Si' is a partial aspect of the syntactic structure and 'X' is the other in-
gredient. The problem here is to say what goes into `S*' without merely imposing a top-down con-
dition from our inchoate notion of a proposition, for such a move would apparently exclude
interpretable copies that satisfy all syntactic and interface demands.
There is a principled argument for an operation of 'deletion' of all but the highest copy based
on the need for linearization at PF (see Nunes 2004). Such reasoning, however, does not translate
into a reason for deletion at LF, or its equivalent, at least not straightforwardly. Firstly, lower copies
are interpretable at the 'semantic' interface, even if theta roles are featural rather than configura-
tional. Secondly and correlatively, because linearization does not apply at the 'semantic' interface,
there is an absence of an independent reason to view any process of deletion as applying at LF.

Mind, Vol. 116 . 464 . October 2007 @ Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
820 John Collins

daft idea at all; it is probably true. The present point is simply that how
this duality appears to play out precludes the recovery of a level of syn-
tax that codes just for propositional structure, at least if that notion is
intimate with that of truth conditions.
Biting the bullet also has empirical consequences. On the proposal
envisaged, we should let the notion of a proposition float until
anchored to some settled syntactic account. Yet, without getting into
any technicalities at all, there is a problem here. The proposal cuts
propositions as finely as syntax, but there are surface 'transformations'
that preserve something we might wish to call synonymy, although dis-
course factors change, for example:

Passives: Bill kicked the ball - The ball was kicked by Bill

Expletives: A fly is in my soup - There is a fly in my soup

Clefts: Bill wants a car - What Bill wants is a car - It is a car that
Bill wants

Raising: Bill appears to be tired - It appears that Bill is tired

Prima facie, discourse effects apart, such 'transformations' do not affect

meaning. On reflection, it is easy to see that there are indefinite ways of
'saying the same thing' with distinct syntactic structures. Intuitively, an
account of propositions should furnish what is shared across the struc-
tures. If, however, propositions are cut as finely as the syntax, we lose
anything that could be shared. It bears emphasis that this is an empiri-
cal problem. The data are that competent English speakers understand
the pairs above as 'saying the same thing', and it is this intuition (inter
alia) that the notion of a proposition is meant to capture.
The fineness of cut problem is not new. My present worries, however,
do not turn on translation between languages or propositional attitude
embeddings, both of which typically animate propositional cut worries
(cf. King 2007, pp. 95-101). My concern is that the data appear to tell us
that syntactic constituency has parameters of freedom that are inde -
pendent of the determination of propositions (the same truth condi-
tions). Thus, tying propositions so intimately to syntax is misleading
regarding both the nature of syntax and the higher level determination
of truth conditions. The appropriate inquiry, it seems, is to determine
the parameters of freedom within the syntax that preserve a notion of
'same meaning' (modulo discourse factors). For this inquiry, we pre-
cisely do not want propositions individuated by syntax; the idea of such
a proposition would be a supererogatory abstraction.

Mind, Vol. 116 . 464 . October 2007 @ Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
 Syntax, More or Less 821

2.3 Scope, variables, and reconstruction

Part of the attraction of syntax for the philosopher of language is that it
promises to account for scope phenomena, which surface structure
alone notoriously fails to capture, at least in the absence of much gerry-
mandering; indeed, the syntactic level of LF was principally posited to
account for such data (Chomsky 1975, 1977; May 1977, 1985). Here at last
appears to be a case where syntactic difference marches in step with
propositional difference. Consider this simple example:

(5)a. Everyone loves someone

b. (every person x)(some person y)(x loves y)

c. [Tp [DP Everyone], [Tp {DP someone]) [Tp ti loves ti] 1

d. (some person y)(every person x)(x loves y)

e. [Tp [DP someone]) [Tp [DP everyone] i [Tp ti loves tj] J

The syntactic structures c and e reflect the respective interpretations of

ambiguous a given in b and d. As things stand here, we have movement,
but not a surplus of syntax: the moved DPs are read as variable binding
operators with the variables being the traces-t-left behind in the
launch sites of the movements.
The operation (Quantifier Raising-QR) that moves the DPs in this
case was precisely designed to capture the semantic operator-variable
readings given in b. and d. It is for this reason that QR is very suspicious
from a purely syntactic point of view' (a more faithful rendering, on
the lines witnessed earlier, would see the subject DP copied from an
internal SPEC-vP position to a SPEC-TP position, which here is occu-
pied by ti). Let us put such qualms aside. For present purposes, I am
happy to let the structures structure the corresponding propositions
(see King 2007 for a detailed account of how this might go). After all,
QR was devised to capture scope construal rather than narrow syntactic
demands of feature valuation. The question is: Is it plausible to con-
strue all cases of copying as creating structure relevant to the encoding
of propositions? The idea here might be to take a proposition to be
determined from what we might call a Tull-reconstruction', where every
copy plays a propositional role. There are many technical issues here,

13 For suspicions over QR, see Hornstein 1995; Kitahara 1996; Uriagereka 1998; but see also Fox
2000, 2003.

Mind, Vol. 116 . 464 . October 2007 CC Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
822 John Collins

which I shall sideline." Let us, though, consider the proposal in rough
outline to see how far it might take us.
It is certainly true that intermediate copies have scopal and binding
effects. Consider, for example, the following case:

(6) Bill seems to her [<Bill> to appear to X [<Bill> to <Bill> like

Mary] ] (X = himself, her, *herself, *him)

Here we see that, where X = himself, the intermediate copy in the sub-
ject position of to appear acts as a clause-mate antecedent, thus ruling
out X = herself. Thus, we may construe the highest copy as an operator
and all lower copies as bound items (cf. Chomsky 2007). This produces
a propositional form along the following lines:

(7) Bill is such that it seems to her that Bill appears to himself to be
such a Bill that is a Bill that likes Mary

Here we have four instances of Bill to match the four copies of Bill in
(6). Similarly, our simple example of Bill sleeps should be rendered as,
say, Bill is a Bill that sleeps. Questions might be raised about the fidelity
of such forms to the original 'meanings', but, in such cases at least, let us
accept them as accurate full-reconstructions. Two problems face this
general strategy, one conceptual, one empirical.
The conceptual problem is one suggested above. Were we to follow
this line, we would radically depart from any intuitive idea of a proposi-
tion that encodes truth conditions, or something as simple as, say, Bill
instantiating the property of sleeping. Now this might be a price many
would happily pay, but then they land in the problems above that face
the 'empirical' idea of a proposition. Perhaps all the reconstructions
show us is that all copies in this case must be interpretable at the seman-
tic interface; why should we conclude from that that a proposition is
thus structured? Indeed, if the phase derivation model is accepted, then
the syntax will not deliver the structure of a proposition even in the
case of (6), for the VP like Mary, will be spelt out once the phase [ vp Bill
like Mary] is constructed. This problem will hold for all structures that
feature copying through phases. At best, then, the proposal will give us
the structure of a proposition for non-phase including TP comple-
ments of phase CP. It bears emphasis here that this does not affect the

14The status of reconstruction is very much a live one. Suffice it to say that issues turn on which
copies should be interpretable and which should be, in some sense, deleted; thus, there is broad
consensus that low A-bar copies are uninterpretable. The status of A-copies is much less certain
(e.g. Chomsky 1995b; Boecloc 2001). In general, therefore, the idea that a proposition is going to be
reconstructed from the full set of copies looks to be a non-starter.

Mind, Vol. 116 . 464 . October 2007 CC Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
 Syntax, More or Less 823

general point intimated just above that the reconstruction of copying

simply shows the interpretability of copies.'
The empirical problem is that it is difficult to see how the proposal
would work for all cases of copying. The idea, in effect, is to construe
the proposition expressed by a given sentence as the proposition
expressed by the syntactic structure of the reconstruction of the given
sentence. Yet not all instances of copying have reconstructions.
Consider verb raising, which is widely believed to hold for all transi-
tive verbs.

(8)a. Bill loves Mary

b. [Tp Bill [T, (I) [ vp <Bill> [v v+<love> [vp love Mary]

The problem here is to say what the semantic relation is between the
copies of love. It is not difficult, of course, to find paraphrases of (8)a
that feature two instances of Bill and love. The problem is to relate the
copies of love so that they might realize a semantic relation analogous
to an operator-variable relation that might be underwritten by the syn-
tax. For example, Bill is a Bill that loves the loving of Mary gets the
meaning wrong. Worse, Bill is a Bill that loves and the loved is Mary (or,
say, ... the loving is of Mary) just signals a surrender of the idea that
syntax is to be a guide to the proposition-that is, while the meaning is
in some sense captured, the reconstruction has, in effect, the verb love
agreeing with the DPs, as if there were two distinct relations in the syn-
tax, one for Bill and one for Mary. It might be that the syntax works in
this manner, distinguishing internal and external arguments as mon-
adic predicates of the same event, but the proposition appears to
involve just the one dyadic relation. The problem becomes more vivid if
we look at ergative/causative verbs, such as break, sink, boil, etc. With
such verbs in transitive form, it is plausible to take the copied verb as
expressing causation:

(9)a. Bill boiled the soup

b. Bill caused-to-boil the soup

c. [Tp Bill [T,(13. [vp <Bill> [v v+boil [vp <boil> the soup] ]

15 The general strategy of taking lower copies to be bound items of a higher operator within a
phase depends upon the integrity of chains as syntactic objects legible at the SEM interface. Al-
though chains are widely assumed, it is far from obvious that chains in general should be sanc-
tioned; see Epstein and Seely 2006. A rejection of chains would, in effect, be a rejection of any
reading of copies as related as operator to a variable.

Mind, Vol. 116 . 464 . October 2007 CC Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
824 John Collins

In such cases, we precisely do not want the originally merged verb

showing up in the proposition, as it were, for that, again, would give us
the wrong meaning: Bill is a Bill that boiled the boiling of the soup. Of
course, the interesting fact about such verbs is that the original copy
remains interpretable:

(io)a. The soup boiled

b. [Tp The soup [T, .4) [-vp boil <the soup>]]]

(Thus, we take the verb boil to be the same item in both constructions
that assigns a theme role to the soup; boil acquires the causative reading
via copying to the functional v head, this option being necessary, if
there is another nominal available to serve as the agent.16)
There are many other instances of movement/copying, including
topicalization, passives, unaccusatives, the full range of interrogatives,
focus constructions, relative clause formation, 'tough movement',
quantifier floating. Indeed, the problem looks insuperable if we take
projection itself to be copying, as in bare phrase structure. Correla-
tively, if labels are required, perhaps for purposes of extension and line-
arization, then they too must be rendered semantically interpretable.
To account for all of these varied structures in terms of some form of
reconstruction, which does indeed appear to work in some cases for
argument copying to an operator position, is an onerous task. I should
not want to say that the task is impossible, but I suspect that once the
complexity of the problem is clearly in view, the likely return on the
effort will be seen to be not worth the outlay.
As an addendum, it is worth noting the difference between traces and
copies. Traces look amenable to a variable construal, for they are not
proper lexical items; they merely mark a position in a relation (a so-
called `chain'). Copies are different; they are genuine lexical items with
a full set of features-a copy is not simply a variable. From the point of
view of the syntax, copies are much to be preferred. Traces must be
added to a derivation, but it is opaque from whence they might come. It
appears that they are simply posited to account for movement. Copies,
on the other hand, come for free. The lexical items are already there and
the copying of them is too in so far as they are copied from the lexicon

16 See Pietroski 2003 for an interesting account of the semantics of causatives in line with the
presented syntax. More generally, Pietroski (2005a) has presented a framework where lexical items
are interpreted as monadic predicates of event variables The structures, however, clearly do not
support the linguistically structured proposition idea; the structures are, rather, an interpretation
the syntax provides that treats Merge as conjunction.

Mind, Vol. 116 . 464 . October 2007 CC Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
 Syntax, More or Less 825

as a token of a type. Again, we reach a familiar conclusion: syntax

appears not to care about propositions.
Throughout this section, I have argued that syntactic structure is out
of step with the structure that appears to be required for propositional
structure: there is too much syntax. This is not the least surprising or
disturbing if we do not begin with the thought that syntax must answer
to what philosophy or intuition otherwise mandates. On the feature-
valuation model I sketched, syntax is driven by semantic constraints,
but from the point of view of the syntax, this might be seen as a sorting
out of what belongs purely to morphology and what is of semantic sig-
nificance. This syntactic sorting appears not to respect the notion of a
proposition. As things stand, therefore, we have no good reason to
think that syntax, as currently understood, provides the structure for
propositions.

3. Too little syntax: the case of covert variables

In a series of recent articles, Jason Stanley has proposed and defended
the thesis that all elements of a proposition (explicature) are encoded as
values of the lexical items (or encoded via syntactic operations on such
items) of the sentence that expresses the proposition (relative to con-
text).17 In Stanley's words, 'all truth-conditional effects of extra-linguis-
tic context can be traced to logical form' (moo, p. 391). Stanley's thesis
has attracted not a little criticism:8 Almost all of the extant responses,
however, leave unmolested Stanley's syntactic assumptions; their focus
is firmly on his semantic claims:9 Further, the responses variously
defend some species of pragmatism, or otherwise question whether all
'truth conditional effects' can be accommodated as values of lexical
items. In short, Stanley's semantics is questioned. In contrast, my claim
will be that syntax, at least as currently conceived within the generative

'Stanley 2000, 2002a, 2002b, 2005; Stanley and SzabO 2000a, 2000b. Stanley's thesis amounts
to the denial of unarticulated constituents: constituents of a proposition that are not expressed by
any lexical item of the sentence that express the proposition (see Perry 1986; Crimmins and Perry
1989). Taylor (2001) also appeals to covert variables in the syntax to account for putative contextual
effects on content. Unlike Stanley, however, he neglects to offer any syntactic argument for his the-
sis.

18 From Carston (2002a, 2002b), Breheny (am), Bach (2000), Neale (moo, 2004, 2005), Reca-
nati (2002, 2004), Cappelen and Lepore (2002, 2005a, 2005b), Elgardo and Stainton (2004), Lepore
(2004), and Pagin (2005).

19Breheny (2002, p. 183), Recanati (2004, p. io8, n. 19), and Cappelen and Lepore (2005b, Ch. 6)
express some doubt about the justification for (not the truth of) Stanley's syntactic assumptions,
although neither press the issue. To my knowledge, only Blair (2005) questions Stanley on syntax,
in relation to cross-over effects.

Mind, Vol. 116 . 464 . October 2007 CC Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
826 John Collins

tradition, provides too little for this thesis to be true. I should empha-
size that my arguments are not intended to lend weight to the positions
of Stanley's critics; indeed, Stanley is to be commended for taking seri-
ously the syntactic constraints on semantic interpretation, which are
highly problematic for `contextualise positions. My complaint is only
that such constraints do not play out in favour of Stanley's positive
view.
Stanley does not present a full-dress account of logical form; rather,
he simply appeals to 'correct syntactic theory' (2000, p. 397), and
assumes that such a validated theory will (i) vindicate a level of logical
form and (ii) support the presence of items at logical form that may
serve as vehicles for 'all truth-conditional effects of extra-linguistic con-
text'. Stanley (2002b, 2005) does acknowledge, of course, that this thesis
about syntax can not be beamed directly from semantics, as it were..
some syntactic motivation must be provided for the claim that the rele-
vant items occur at 'logical form'. It is this syntactic argumentation that
will be my focus.
This relatively narrow approach might seem to leave me open to a
simple polemical rejoinder: 'So much the worse for generative theory!'
My intent, however, as expressed in my introductory remarks, is not to
categorically refute Stanley's thesis and his attendant semantic account.
My contention is conditional: if recent generative theory is correct,
Stanley's account is incorrect, or at least lacks the support he claims for
it, for there is too little syntax. The imagined rejoinder, then, is at least
coherent, but it has little impact. It is crucial for Stanley's account and
his general anti-(contextualise stance that some syntactic theory does
actively support his account, and he assumes that the generative
approach does so; as we shall see, he is perfectly explicit in this regard.
So, one committed to a Stanley-style approach to the relevant semantic
phenomena is free to pick any syntactic account that coheres with it,
but, if I am right, recent generative theory is not such an account.
Further, for present purposes, it might well be that Stanley has the
semantics right; my concern is only to cast doubt on his syntactic
assumptions. To repeat, this is not a marginal issue, but one that goes to
the heart of Stanley's approach and those of many others. Stanley
(2000, p. 413, 2005, p. 235, n. 8) readily admits that variables of the kind.
he posits are often appealed to in the semantics literature; what distin-
guishes his position is precisely the thought that such variables are 'syn-
tactically represented'. If Stanley's position has no support from
syntactic theory, then the values of his posited variables would be de
facto unarticulated constituents. But Stanley's very claim is that there

Mind, Vol. 116 . 464 . October 2007 CC Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
 Syntax, More or Less 827

are no unarticulated constituents. It follows that Stanley's thesis is, in

part, a syntactic one; its syntactic assumptions are precisely what distin-
guishes it from the competition (cf. Cappelen and Lepore's (2005b,
pp. 8-9) geography of 'moderate contextualism').

3.1 Stanley on context and logical form

Stanley's thesis is that 'all truth-conditional effects of extra-linguistic
context can be traced to logical form' (moo, p. 391), where 'logical
form' is a level of syntactic structure underlying surface form postu-
lated by 'correct syntactic theory' (2000, p. 397). Stanley discusses a
range of data to support his thesis; the focus of the sequel will be on
what I take to be the best case for Stanley, namely, quantifier domain
restriction and binding more generally, although many of the points to
be raised are intended to have a general import.2
The phenomenon of domain restriction is that quantifier DPs are
typically read as being restricted to a domain narrower than is lexically
determined by the complement nominal. Since the restriction is not
lexically encoded, it seems that it must be provided non-lexically, such
as by a pragmatic process of 'enrichment' or a determination of 'com-
municative relevance', and that just means that not every item of a
proposition will be encoded at LF. Consider:

(ii)a. Every student got i00%

b. Every student answered every question

c. Every student answered every question on her exam

A typical statement of (ii)a does not communicate the thought that

every student in the universe got ioe/o. The subject DP is typically
restricted to cover just some contextually salient domain (e.g. every stu-
dent in one's class). Likewise, the object DP of (ii)b clearly does not
typically cover every question; again, the domain is restricted. We may
assume that (ii)c expresses the typical explicature of an utterance of
(n)b. The prepositional adjunct in (11)c, however, is not lexically

20 The following considerations also critically bear on other theorists; for example, King (2001)
suggests that where a bare demonstrative forms the subject of a clause, then it has an empty nomi-
nal complement; thus, e.g. [Tp [DP That till [is tall]]. This posit is required to support King's gen-
eral account of demonstratives as 2-place quantifiers (in fact, his theory posits two more places,
but these are not lexically realized). There is, however, no clear syntactic reason for such a phono-
logically null item outside of ellipsis. Further, it would appear that demonstrative determiners
(that, these, those) along with many other kinds, although not all, carry the right (13s-features (3rd
person, singular/plural-number, perhaps null Case) to match and value uninterpretable features of
T (their external Merge or raising from vP eliminates the EPP feature of T). Thus, syntactically, it is
not obvious that there is a nominal apart from ellipsis. See Longobardi 2001 and Chomsky 2001.

Mind, Vol. 116 . 464 . October 2007 @ Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
828 John Collins

encoded in (n)b. Without further ado, therefore, it appears that the

explicature does not compositionally devolve onto the constituent lexi-
cal items of the sentence that expresses it.
Although disarmingly quick, I think that such reasoning is inescapa-
ble if we are working with 'surface form' alone (i.e. if we recognize no
items in a linguistic structure other than those that are phonologically
realized), and if we want (11)b to express (11)c as its explicature (relative
to context C). As remarked above, however, the received understanding
in generative linguistics is that phonological realization does not
exhaust the items of a linguistic structure. Indeed, the very idea of 'log-
ical form' in the sense Stanley intends enshrines the claim that 'surface
form' is not the level at which (all) features determinant of semantic
interpretation are encoded. The above quick argument is thus incom-
plete at best. Indeed, we might take it to militate for the hypothesis that
'surface form' is not the level at which semantic interpretation applies.
Such is Stanley's approach. He contends that, when we consider 'logical
form', we may readily postulate phonologically null items that serve to
realize the propositional constituents unexpressed at the surface.
Stanley claims that, at 'logical form', the 'object' DP in (11)b contains
a covert function-variable complex that is 'associated' with the nominal
question, or 'co-habits a node with it', where the argument variable is
bound by the subject DP every student, and the function on the argu-
ment is determined by context (perhaps a speaker's intention), in this
case a function from individuals (students) to exam questions.' In gen-
eral, we have the following schematic proposals, where (12)a gives the
semantics for the schematic syntax in (12)b, which enables the binding
relation displayed in (12)c:

(12)a. D [<N,f(j)>]c (B) = D [(N n Ix: X E C(f)(c(0)}], {B}

b. [xp [DP [D N+ f(j)]
c. [ [Every <student, f(j)>]i [answered [every question f(i)]]]

The restriction on a Det D is the intersection of the set determined by

the lexical nominal complement and the set determined by the values
the functional complex takes relative to context c. Where not bound, as
in (ii)a, the variable complex takes contextually determined values,
which explains typical domain restriction. That there is such a variable

'In certain respects, the proposal is a development of Partee's (1989) model of implicit varia-
bles. Partee, however, is very cautious indeed about her proposal (especially see pp. 267-8). Fur-
ther, her semantic model is DRS, and, in general, she seeks no support from a notion of 'logical
form'.

Mind, Vol. 116 . 464 . October 2007 CC Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
 Syntax, More or Less 829

complex is supported by the fact that it can be bound, as apparently is

the case in (11)a. If there were no item in the syntax, then there would
be nothing to be bound, and the bound readings would thus be impos-
sible, or at least anomalous. So goes the thought.
Thus, relative to the context we are imagining, (11)b gets to express
the proposition expressed by (11)c, whose structure is schematized in
(12)c, because the object DP has its otherwise contextually fixed varia-
ble i bound by the higher DP every student. So, Stanley's point here is
not that context does not determine truth conditions, but rather that it
does so only in (roughly) the same way as it determines indexical val-
ues. That is, context determines values of elements of the structure that
take variable values, but context does not introduce new elements that
are not encoded syntactically, either lexically or as a consequence of a
syntactic operation such as movement. Let us assume that the existence
of such covert variable complexes would serve to provide the right val-
ues to deflect the contextualists' radical pragmatics construal of the
`underdetermination' data. Indeed, let us further assume that no extant
pragmatic process could secure the bound reading of (11)b. The ques-
tion still remains whether there are such items at 'logical form'.
Stanley offers two general considerations in support of the existential
claim. I shall present them and raise doubts about each in turn. Then,
in the following section, we shall see that his underlying assumptions
about syntax are at odds with recent generative theory.

3.2 Stanley's syntactic arguments

Stanley (2000, p. 412) appeals to an 'innocent ... presupposition':
for explicit quantifier expressions, within a clause, semantic and syntactic
binding coincide. That is, bound readings within a clause are due to the ex-
istence of a variable binding operator standing in a certain structural rela-
tionship to a co-indexed variable in that clause.

Granting the 'presupposition', where we find semantic binding, there

we should also find syntactic binding. Consider, then, (11)b,c. The
explicature of (11)1) (relative to context) is expressed by (n)c, which
contains a semantic binding relation between the matrix DP every stu-
dent, and the pronoun her in the prepositional phrase on her exam
adjoined to the complement DP every exam. Yet, trivially, (ii)b appears
not to express any such relation syntactically, for it lacks the bound pro-
noun. But, by the 'presupposition', there must be a syntactic binding
relation. The hypothesis that the nominal of the complement DP con-
tains a functional complex whose variable argument may be bound by
the antecedent matrix DP satisfies the presupposition. Thus, on the

Mind, vol. 116 . 464 . October 2007 @ Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
830 John Collins

assumption that the coincidence of syntactic and semantic relations

holds generally, we have an argument for the hypothesis. Further, we
have a refutation of a contextualist account of the phenomenon, for
pragmatics does not determine syntax, but if Stanley's presupposition is
correct, then the semantic reading rides on the back of syntax.
In general, the methodology here exemplified is to show that clauses
that putatively express 'unarticulated constituents' have a (semantic)
reading that involves the constituent being bound by a higher operator;
it follows from the coincidence of syntactic and semantic binding that
there is a variable in the syntax that is bound by the explicit quantifier.
Thus, for the quantificational cases at hand, we are interested in those
constructions where one or more quantificational DPs occur whose
explicatures appear to demand a reading under which the domain of
one varies with the domain of the other. (11)b provides an example
(repeated here as (13)a). Here are two more examples, with the relevant
readings supplied:

(13)a. Every student answered every question.

'For every x such that x is a student, every question asked x was
answered by x.'

b. In most of Bill's classes, he fails exactly three Frenchmen.

Tor most x, such that x is a class of Bill's, Bill fails exactly three
Frenchmen in x:

c. In every room in Bill's house, every bottle is in the corner.

'For every x such that x is a room in Bill's house, every bottle in
x is in the corner.'

So, the evidence is that the domains of the DPs-every question, three
Frenchmen, every bottle-are fixed by a preceding DP. In (13)a, not
every question, just every question asked every student, was answered;
in (13)b, not any three Frenchmen were failed, but three each in most of
Bill's classes; and in (13)c, not every bottle is in the corner, but only every
bottle in every room. Thus, the hypothesis is that the three latter DPs
contain a variable whose value may be fixed by the nominal of the pre-
ceding DP. For example, the object variable of the complex 'associated'
with question is satisfied by students, which then serves as the argument
of a function contextually fixed that maps to the domain of the
quantifier-perhaps, the set of questions students were asked, or
maybe the set of questions on the students' exam. Let us accept these
readings.

Mind, vol. 116 . 464 . October 2007 @ Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
 Syntax, More or Less 831

Stanley does not say what kind of syntactic relation is supposed to

realize the putative semantic binding. The idea seems to be that co-
indexation under c-command suffices. One might raise concerns here
over the status of indexation, but let this worry pass.22 If Stanley means
that where we have a binding reading, there we have a c-commanding
DP which supports that reading, then the claim appears trivial. Of
course we shall have a c-commanding DP for the phonologically null
item, for we have a c-commanding DP for the DP to which the item is
adjoined on the fully realized readings as offered in the above exam-
ples.' That is, c-command itself does not determine that the putative
covert item is bound; that the item is c-commanded by the higher DP
would only tell us that it could be bound. But even this triviality
appears to fail for Stanley's proposal. The bound item is supposed to be
in some sense essentially associated with the relevant nominal,
although not merged with it, that is, the association is not syntactic.
Therefore, the complex is not projected in the tree as either an adjunct
or argument. But if this is so, then how is syntax supposed to make any
deliverance on the matter given that the item is not even in a position to
be bound, for it does not project to be in a position of the higher DP's
c-command domain? Here, we are effectively being asked to imagine
that c-command extends to something outside of the syntactic configu-
ration, while Stanley wants us to think that syntactic theory supports
the proposal. In other words, there appears to be no syntactic principle
or generalization that admits the imagined binding relation; a fortiori,
there is none that militates for the existence within the syntax of the
phonologically null item.
Stanley (2005, p. 245, n. 15) does suggest that the covert variables
might be in adjoined positions, which marks a departure from the
notion of 'co-habitation' or 'association' from the earlier papers,
although it does mirror the supposed lexical elaboration of the proposi-
tions expressed by sentences with covert variables (cf. Marti 2006).
Adjunction has been a theoretically slippery notion for many years (see,
for example, Chomsky 1981, 1986b, 1995a, 2004; Kayne 1994; Uriagereka

22 Minimalist scruples, in the form of inclusiveness (see below), might lead us to view indexes as
merely notational, for their only mandate appears to be the construals they mark; that is, indexes
are extra-syntactic posits. For perhaps the most fully developed non-index account of binding, see
Safir (2004a, b). For a defence of indexes as inherent features of nominals, see Baker (2003), espe-
cially pp. 96-7.

23 Here I assume that distinct scope readings are supported by distinct c-command relations. It
might be noted, familiarly, that c-command is not necessary for semantic binding:
(i) [Every lover [of Mozart]] thinks hei was the greatest of composers
(ii) [The men [who prepared the fishi]] hated theiri smell

Mind, Vol. 116 . 464 . October 2007 @ Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
832 John Collins

1998, 1999). If we think of adjunction as essentially modification, then

the proposal does fit with what Stanley wants.' Further, since adjuncts
do not enter into either feature elimination or theta-role assignment
relations, then the proposal would lift explanatory syntactic burdens
from Stanley's shoulders. However, the proposal is problematic.
Although adjunction is not well understood, it seems that adjunction
extends phrasal projections without changing the category of the pro-
jection, that is, adjuncts are adjoined as phrasal sisters of the projec-
tions of heads and inherit the phrasal category of their sisters-they
exist in a different dimension. Thus, adjunction falls outside of SPEC--
head-complement configurations. (In the adjunction proposal pre-
sented by Marti (2006, pp. 142-43), the putative variable adjuncts are
mistakenly presented as arguments in first Merge position.) Two prob-
lems arise. Firstly, it is difficult to see how the covert variables could be
understood to be phrasal. An adjunct appears to inherit its phrasal sta-
tus from its sister. The idea of a variable, however, as understood in
mathematics/logic, just makes no sense in terms of syntax. Every lexical
item is a bundle of interpretable/uninterpretable features to which
operations within the syntax are sensitive. But a variable, as understood
by Stanley, has no features at all; it is essentially a semantic element, as
it were, an item which can take a contextual value (and so be free in the
syntax) or be bound, but this is not something to which the syntax can
be sensitive.' Even if we put this worry to one side, a greater second
problem remains.
Adjuncts are essentially optional. For example, the English
wh-adjuncts (how, when, where, why) mark, respectively, manner, time,
place, and purpose. None of this information enters into thematic
assignment or agreement relations. This can be seen by the fact that un/

24 Being an adjunct is not a property of a kind of lexical item, but a configurational position
items can realize. The operation of adjunction may be viewed as a deviant form of Merge. Merge
extends a structure to create a new kind of projection. Adjunction, on the other hand, adds a sec-
ond tier, as it were, to an extant projection. Being hierarchically outside of the SPEC-head-com-
plement configuration, adjunction creates familiar ambiguities, for the site of adjunction is
thematically open and not marked by agreement (e.g. Bill shot the elephant in his pajamas).

25 This is not to say that items cannot be construed as variables. Such an understanding is what
animated the early work on 'logical form' (Chomsky 1975, 1977; Chomsky and Lasnik 1977; May
1977). It bears emphasis, however, that the 'variable' construals proposed covered PRO and trace t,
i.e. items that are bound. Free variables, of the kind required by Stanley to take contextual values,
were strictly ruled out within the syntax (Chomsky 1977, pp. 10-11; Chomsky and Lasnik 1977,
pp. 429, 432). Similar remarks apply to covert operators posited to account for 'tough construc-
tions' (inter alia), i.e. the operator has to move in order to bind a lower copy/trace, with the whole
structure forming a complex predicate. By more recent theory, we can say that free variables are
uninterpretable.

Mind, VOL 116 . 464 . October 2007 @ Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
 Syntax, More or Less 833

acceptability is invariant over phrases and phrases+adjunct. In effect,

adjunction is neutral as to the interpretability of a structure.26
Marti (2006) takes optionality of adjuncts to be a boon for the thesis
that variables are adjuncts; but how precisely are we to understand this
claim? There are three choices, each problematic. First, the variables
might be strictly optional like genuine adjuncts. The problem with this
option is that Stanley's very argument is that the semantic binding phe-
nomena necessitate the covert variables, given 'innocent' assumptions.
Thus, optional adjuncts fail in the work Stanley would ask of them; that
is, if we take the relevant structures to be invariant over grammatical
criteria of acceptability, whether they feature covert variables or not,
then there simply is no syntactic argument for thinking that they are
extant in the structures. Second, the variables might be understood to
be necessary. This would, however, just render the 'adjuncts' non-
adjuncts. As remarked, adjuncts do not enter into feature valuation or
thematic relations, and so do not determine acceptability. Third, the
variables might serve as adjuncts as and when required. Following
Marti (2006), Stanley (2005) appears to favour this option: he suggests
that arithmetical nominals do not contain covert variables because
their semantics do not require domain restriction.27 But we are after a
syntactic reason to think that covert variables exist. Even if one were to
sideline the other problems, which look quite recalcitrant, to posit an
empty adjunct to capture a semantic reading is not to furnish a syntac-
tic reason.

Overall, Stanley's argument here, as far as I can see, is premised upon

the silent assumption that the nature of syntax should be read off our
stable semantic intuitions. The specific issue of binding is a red herring

26 Adjuncts do enter into binding relations and negative polarity/parasitic gap licensing, which
appears to establish that adjuncts enter into c-command relations. The fact of their inability to af-
fect acceptability, however, remains unaffected. Chomsky's (2004) pair-set account of adjuncts
views them as entering the derivation late at the interface. The account is designed to cater for ad-
juncts' extra-dimensionality while establishing a c-command relation. None of this complexity
appears to offer succour to Stanley.

27 Cappelen and Lepore (2005b) appeal to arithmetical predicates as a counterexample to

Stanley. Their argument runs on the thought that the covert variables must be bound or have val-
ues fixed referentially, but, of course, mathematical truths are not context sensitive. Stanley (2005,
pp. 243-4) responds by claiming that arithmetical predicates do not have variables associated with
them because they do not support bound readings (cf. Marti 2006). He gives no syntactic reason
for this claim (cf. Pagin 2005, p. 325, n. 22, where Pagin appears to endorse Stanley's response but
rejects Stanley's broader thesis for independent reasons). It has, alternatively, been suggested to me
that variables could take a universal value in the cases under consideration. There are at least two
problems with this proposal: (i) apart from universal quantification, it is difficult to see how a var-
iable could take a universal value and (ii), in the absence of an implicit universal quantifier, the
values would be contextually determined, but what kind of context could determine a universal
value?

Mind, Vol. 116 . 464 . October 2007 @ Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
834 John Collins

in so far as the examples offered do not in fact exhibit syntactic binding

as there is no merged item to be bound; rather, it is particular readings
that purport to exhibit binding; and even if the readings are correct,
nothing would ipso facto follow as regards syntax save for the c-com-
mand relation, which is already in place. None of this is yet to say that
Stanley's proposal is wrong; it is only to say that it is opaque what prin-
ciple within syntax should lead us to think it is correct. The syntax
appears to contain too little.
Stanley's second general argument from syntax to support his bind-
ing proposal is more rhetorical. Stanley (2002a, p. 152-3) asks us to con-
sider the following pair (inter alia):

(14)a. *The ship sank [PRO to <PRO> collect the insurance] 28

b. The ship was sunk [PRO to <PRO> collect the insurance]

Stanley takes (14)a to be 'ungrammatical' because 'the available local

[controller is not an expression] that denote [s] things capable of col-
lecting insurance' (2002a, p. 153). In other words, PRO-the phonolog-
ically null SPEC of TP to collect the insurance-requires an antecedent
or controller that may realize an agent theta role of the verb collect, but
the only available expression is the ship, which does not play the
required agent role. (Note that this assumption depends on certain
ideas about what ships can and can not do. For example, in the context
of a child's story about ships with mental states, (14)a would be accept-
able. Let this complication pass.) Why, in contrast, is (14)1) grammati-
cal? Stanley's explanation is that there is a phonologically null
prepositional phrase, by a, whose complement is the thematic agent of
passive sunk. Such an expression, a, may serve as the controller of PRO
in the subordinate clause. For example:

(15) The ship was sunk {by the crew i } [PROi to <PRO> collect the
insurance]

Stanley claims that his argument from binding rehearsed above is 'of the
very same structure', that is, the postulation of phonologically null items

28 If we grant 'PRO', we ought not to conceive of it as a peculiar lexical item that essentially
serves the special role of realising semantic arguments for control infinitives etc. We take the lexi-
con to contain a generic item PRONOUN, which is merged at SPEC vP to satisfy theta require-
ments. The merging of infinitive T to (with EPP and Case valuing features) attracts PRONOUN to
its SPEC, where its Case is valued as [null], and the uninterpretable features of T are eliminated.
This entails that the item has no spell-out, it remains covert. 'PRO' simply marks the fact that
there is a non-spelt-out item in the structure. Thus, the structures support control construals
without the derivation being driven to realize any such readings. For a contentious overview of
PRO and control, see Hornstein 2003.

Mind, vol. 116 . 464 . October 2007 @ Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
 Syntax, More or Less 835

in the syntax explains otherwise inexplicable phenomena. Thus, if one

denies the binding argument, one should, by parity of logic, deny the
above reasoning. Stanley's argumentative strategy here is questionable.
First off, Stanley is correct that passive verbs imply agency in a way
that ergatives (sink, open, close, break, etc.) do not. Compare the follow-
ing cases parallel with the sank/ sunk case:

(16)a. Bill broke the glass

b. The glass broke

c. The glass was broken deliberately

d. *The glass broke deliberately

e. The glass was broken in order to create a disturbance

f. *The glass broke in order to create a disturbance

The adjuncts in (16)c and (16)e imply agency and are perfectly accepta-
ble; the same adjuncts in d. and f. create unacceptable structures
because the ergative verbs do not imply agency. It seems, therefore, that
semantic agency is certainly obligatory in these cases, but it just does
not follow that this is recorded syntactically.
The phonologically null phrase to which Stanley appeals is usually
referred to as an implicit argument. Such items are perhaps the worst
possible for Stanley's analogising purpose, for such arguments are typi-
cally taken not to project in the syntax and are so precluded from syn-
tactic relations: plausibly, relations that involve implicit arguments are
ones of construal rather than being clearly syntactic (see, for example,
Williams 1985; Chomsky 1986a; Rizzi 1986; Jackendoff 1987; Brody and
Manzini 1988; Bhatt and Pancheva 2006). For example, theta conditions
do not demand the occurrence of thematic agents for passive predi-
cates. Likewise, no agreement or EPP factors demand their occurrence.
To see the point here note that, in (14)b, whoever does the sinking need
not be the person or persons who collect the insurance; it is left open
that, say, the crew sank the ship under the orders of the owner, who col-
lects the insurance; that is, control is optional, and so a phonologically
null phrase is not necessary. Likewise, a clever burglar who engineers it
so that a security guard breaks a glass (in order to cause a distraction)
might rightly be said to be the agent of the distraction. The case of
adverbial adjuncts is clearer still: deliberately (and similar intentional
adverbs) imply agency, but they do not require a syntactically realized
agent. In sum, the effect appears to be semantic, not syntactic.

Mind, vol. 116 . 464 . October 2007 CC Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
836 John Collins

Of course, an explicit provision of a prepositional phrase with a suit-

able agent complement does make for obligatory control. In effect, an
implicit argument is posited as an elaboration of the lexical content to
support a control construal; implicit arguments are not independently
identifiable syntactic or lexical items.
To buttress the point here, consider that explicit thematic arguments
can be provided that satisfy all syntactic requirements yet do not serve
as suitable controllers.

(17)a. The ship was sunk by an iceberg [in order [PRO to <PRO> col-
lect the insurance without arousing suspicion] ]

b. [We engineered it so that] the glass was broken by the guard

[PRO to <PRO> keep everyone busy while we robbed the safe].

These are fine to my ears, even though an iceberg is not the kind of thing
that can collect insurance and the guard is clearly not the agent of the
subordinate clause. Both DPs (an iceberg' the guard) play a kind of
instrument role, normally introduced by with, which in turn 'implies' an
agent or perhaps serves as an analogue of one. We can look at the struc-
tures as specifying an event in the matrix passive, with the subordinate
clause explaining the reason for the event.' In both cases PRO is not
controlled. If the clauses are preposed, then PRO is clearly arbitrary as
opposed to controlled and the proposed readings are readily available:

(18)a. [In order [PRO to <PRO> collect the insurance without arous-
ing suspicion] ], the ship was sunk by an iceberg

b. [PRO to <PRO> keep everyone busy while we robbed the safe],

we engineered it so that the glass was broken by the guard

c. [PRO to <PRO> collect the insurance], the ship was sunk

All of this suggests that the implicit control readings do not lead to any
covert item in the syntax.

'Some informants have reported a slight infelicity, but this is ameliorated once the 'rationale'
clause is fronted (see below). This suggests that the degree of unacceptability is pragmatic, not
syntactic.
Control is also affected by semantic features. Consider:
(i) Mary did not want to be parked outside the pub all evening
(ii) The Anglo-French decision to ban the exports angered the

In (i), it is understood that it is not Mary herself that is to be parked but her vehicle, which is not
mentioned. Similarly, in (ii) there is no controller at all, but we understand the agent of the ban to
be the English and the French, rather than arbitrary. These uses might be peripheral in the sense of
Chomsky 1981. My present point is just to suggest that determination of control is not well-suited
to Stanley's analogical purpose.

Mind, vol. 116 . 464 . October 2007 CC Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
 Syntax, More or Less 837

Even if we were to take PRO to be obligatorily controlled, we would

still be left with the need to posit another by-phrase, for an iceberg can
not be the controller, but evidently, any such phrase would not be an
argument of passive sunk. Also note that a passive relative-for exam-
ple, towed by the crew-merged with iceberg also fails to provide an
appropriate controller. In short, it appears to be a control construal that
leads us to posit the phonologically null argument. On such a con-
strual, we take the implicit argument in the cases at hand to be an
abstract agent that answers to the semantic selection requirements of
collect on a control construal. There is nothing antecedent in the syntax
to support this posit, and the construal does not demand that we put
anything into the syntax to support it. Indeed, the alternative con-
struals just discussed show that the containing prepositional phrase
does not project.'
So, let us return to Stanley's challenge: if one rejects the binding
argument for phonologically null functional complexes being associ-
ated with quantificational DP nominal complements, then one must,
by parity of logic, reject the phonologically null controllers in examples
such as the above. To be swayed by such reasoning, we must view the
cases as analogous; that is, Stanley's putative covert items must corre-
spond to implicit arguments. For example, no analogy holds between,
say, the positing of phonologically null PRO and Stanley's items, for
PRO is supported by straightforward syntactic considerations from
theta theory and the uniformity of SPEC-head relations, as well as, of
course, its role in explaining binding and construal data. In the cases at
hand, PRO is required to satisfy the theta requirements of collect and to
value the uninterpretable features of non-finite to. Stanley's items have
no such general syntactic support. It would thus be perfectly rational
for one to accept the positions of PRO in the above structures and
reject Stanley's proposal. But if Stanley's items are a variation on
implicit arguments, then we lose any argument from syntax. Again,
implicit arguments are just posits to account for the control data, they
do not syntactically project. But Stanley's whole idea is that some prin-
ciples of syntax mandate his covert items. This is not so, at least not if
we accept the analogy offered.

'The implicit argument data are somewhat variable. I have discussed the best and most famil-
iar case. Stanley (2002a) also appeals to the following pair, with a covert prepositional phrase ap-
parently required in (ii):

(i) *The record broke [in PRO winning the race]

(ii) The record was broken [in PRO winning the race]
(ii) sounds marginal to my ears.

Mind, vol. 116 . 464 . October 2007 CC Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
838 John Collins

3.3 Stanley on logical form and syntax

So far we have seen that there is no convincing syntactic argument for
the existence of variables of the kind Stanley posits within LF structure.
This section will further support the case against their existence by way
of considering Stanley's conception of logical form.
Nowhere does Stanley say what he means by 'logical form' save that it
is a syntactic level that inputs to semantics. In one sense, this lack of
specificity is perfectly innocent. Stanley's proposal, on the assumption
that syntax does interface with semantics, is that at the interface the
syntax should record the kind of readings that the contextual variables
make available. So, the proposal is neutral about which syntax would
best provide for this service. However, the lack of specificity raises seri-
ous worries that the proposal is quite unhinged from any account of
syntax. The problem is simple: What reason do we have, semantics
apart, for thinking that the 'best' syntax admits the covert contextual
variables? The question is not rhetorical. No level of syntax is a mere
receptacle for whatever we may identify at the semantic level. Anything'
occurring in the syntax must enter into syntactic relations, not just seman-
tic ones. There is an all too easy equivocation between construing 'logi-
cal form' as a level of meaning and construing it as a level of syntax that
is sometimes called 'logical form'. As previously mentioned, recent the-
ory within the minimalist program has ditched LF, and other frame-
works have never sanctioned such a level. So, far from the best syntactic
theory mandating contextual variables at logical form, the 'best theory'
may not even admit logical form.
If we sideline the recent level-free' developments, we might ask:
What is a syntactic relation as opposed to a semantic one? The question
cannot be answered a priori, but only on the basis of theory. Let us
make two assumptions:

(i) The Merge Assumption: Every syntactic relation is a function of

a binary combinatorial operation.

(ii) The Inclusiveness of Interpretation: All output of the syntactic

computation must be exclusively interpretable at one or anoth-
er interface (morphology/phonology or semantics) on the basis
of just the lexical features that enter into the computation.

Assumption (i) provides us with a binary operation, Merge, that simply

builds complex structure from atomic items in a pair-wise fashion. All
syntactic relations must be derivable from Merge. Thus, for example,
movement/copying will be simply (Internal) Merge of an already

Mind, vol. 116 . 464 . October 2007 CC Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
 Syntax, More or Less 839

merged item to extend the structure; c-command, as we saw earlier, will

reflect the fact that the c-commander is merged with the head projec-
tion that contains all that it c-commands (i.e. sister-hood becomes a
simple reflex of Merge). Assumption (ii) tells us, in effect, that the com-
putation is essentially and only driven to 'sort' lexical features between
those that are, respectively, interpretable at 'sound' and 'meaning:31
Following these two assumptions leads us to a conception of principled
explanation, where the structure of the faculty is explained either by
non-language particular principles of computation (Merge) or external
demands from the interfaces (see Chomsky 200113, 2004, 2005a). Thus,
in one sense, as earlier suggested, the syntactic derivation is driven to
answer semantic demands, especially as regards the dual nature of
Merge. It would be a mistake, however, to think of minimalism as a
functionalist semantic agenda. For any item at LF, we need to ask: How
did it arrive at LF -what is its merge history? What features of the item
are un/interpretable? Without answers to these questions, the putative
item is simply beamed from semantic intuition.32
The speculation behind the goal of principled explanation is that the
faculty computation answers to the most minimal/economical design
specifications (per assumption (i)) to satisfy the demands of the exter-
nal components (per assumption (ii)); that is, the computation pro-
ceeds as if it follows the instructions of the external components. Pace
Stanley, the faculty does not in fact follow any such instructions. As
explained above, the computation operates simply to value or elimi-
nate features, with the effect that its outputs blindly and optimally
meet the demands of the external components. What features drive the
computation?-minimally, those to which the computation is sensi-
tive; any other features will not be recorded in the syntax. For example,
a noun carries person, number and Case features (what Chomsky
(1981, 1995a) calls 43s-features). It carries the first two because it must
enter into agreement relations; and these features are semantically
interpretable, of course. Case is also necessary (although mainly silent
in English) because it has a morphological reflex depending on the
configurational position of the noun. But Case is not semantically

31 The assumption here fuses two earlier assumptions for convenience: full interpretation and
inclusiveness. See Chomsky 1986a, 1995a,b. Both assumptions are methodological (guides to re-
search); they are not a priori truths.

32 Neale (2004, p. 113, n. 59, and 2005, p. 230, n. 107) rightly chides his earlier self (Neale 1994)
and various others for treating LF as a receptacle for semantics as opposed to a level of syntax. It is
curious, therefore, that Neale should focus on what an element is 'doing' at LF vis-is semantics
instead of focusing on how the syntactic derivation treats the element such that it has its particular
LF position.

Mind, Vol. 116 . 464 . October 2007 Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
840 John Collins

interpretable. Thus, for any noun entering a derivation, the computa-

tion must value Case so that the semantic interface 'recognizes' it as
uninterpretable. Similar reasoning applies to other categories. Now,
obviously, a noun such as [cat] carries many features beyond its 4)-fea-
tures. Many features appear to be structural (e.g. [cat] can be used as
count, mass, and abstract). But it would seem that the syntax is not
sensitive to any of [cat] 's substantial semantic features, beyond mini-
mal requirements of theta role as determined by verbal configuration.
For example, the internal structure proposed by Pustejovsky (1995)
appears not to enter into the syntax. In short, much of the lexical infor-
mation appears to be carried, as it were, through the syntactic deriva-
tion. It only becomes visible after the structure is handed over to the
external semantic component. In general, then, the derivation works as
if it were solving an equation between lexical items and divergent exter-
nal requirements from sound and meaning. The minimalist assump-
tion is that the derivation is optimal (as encoded in our two
assumptions above). The derivation works as if it 'knows' just what is
interpretable to the respective interfaces, with uninterpretable features
and copying serving as a joint mechanism to satisfy the interface condi-
tions, but the optimality is achieved independent of any access to what
sense, if any, the external components may make of the structures that
input to them. Here is the general moral: an item of syntax must be
such that it is (i) merged and (ii) carries un/interpretable features to
which the derivation is sensitive. In general, as Chomsky puts it, 'Deri-
vations are driven by the narrow mechanical requirement of feature
checking [that is, valuation of uninterpretable features] only, not by a
((search for intelligibility" or the like' (Chomsky, 1995a, p. 201). Let us
now consider how Stanley understands 'logical form', again ignoring
the recent level-free approaches.
First off, Stanley appears to understand LF as a syntactic receptacle
for whatever semantics demands, where semantics here is the inchoate
notion of a proposition or truth conditions. Thus, Stanley writes:
Syntax associates with each occurrence of a natural language expression a
lexically and perhaps also structurally disambiguated structure which differs
from its apparent structure, and is the primary object of semantic interpre-
tation. In accord with standard usage in [generative] syntax, I shall call such
structures logical forms. (2000, p. 393)

Stanley further explains that context 'help [s] us to decide which logical
form is the one that has been uttered', where this solves the equation:
'utterance + X = logical form' (2000, p. 399). If this is intended as an
account of the notion of 'logical form' (= LF) as it occurs in contempo-

Mind, Vol. 116 . 464 . October 2007 CC Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
 Syntax, More or Less 841

rary generative linguistics, then it is off target.33 (i) LF is a mental struc-

ture; we cannot utter logical forms; nor can we utter something with a
logical form. PF is an orthogonal structure to LF. (ii) Logical forms are
not the 'objects' of interpretation. We can understand an utterance, and
in so doing, we will presumably map information from LF onto an
internal structure formed from our perception of the acoustical proper-
ties of the utterance, but we cannot conceivably interpret an LF struc-
ture of another speaker; LF is 'inside the head' it is not something
expressed externally. (iii) Logical forms (LF structures) are not disam-
biguated structures corresponding to natural language sentences. The
idea that there is such a map was proposed by May in his 1977, but
rejected by him in his 1985, and is, anyhow, by no means the consensus.
As previously noted, following minimalist constraints, QR has fallen
under a certain suspicion. May's Quantifier Raising (QR) operation, for
example, raises object DPs whose accusative Case has already been val-
ued. On a phase derivation model, the DP should thus be 'frozen' or
'transferred' before it can be moved higher up the structure to capture
distinct scope readings. Perhaps the quantifier carries some further
uninterpretable feature. These are matters of live debate. (iv) Perhaps
most fundamentally, it is a mistake to think that there are logical forms,
as if they were a kind of object. LF is a level at which operations on lexi-
cal features accessible to external systems input to those systems. It is
wholly an abstraction to turn the level into a set of objects independent
of the operations of the faculty, as if they were determined by an associa-
tion with natural language sentences under favoured semantic readings.
In short, Stanley conceives as LF as part of an interpretive process,
with speakers assigning (LFs' to one another, which, with context,
determine truth conditions. Given this view, it is little wonder that
Stanley feels so free to posit elements at LF that have no syntactic sanc-
tion; they are posited wholly on the basis of the needs of contextual
interpretation, not the internal demands of interface legibility.
Consider the structure Stanley and SzabO (mom, p. 251) propose:

(19) [s [NP [pet Every] [N <man, f(i)>]][vp [NT runs]

" No level of syntactic structure has ever essentially implicated a truth conditional conception
of semantics: Stanley's equation is devised from a certain conception of semantics and how that
conception might integrate into syntactic theory (e.g. Higginbotham 1985; Larson and Segal 1995).
Indeed, Chomsky (2000b, p.132, 2003, p. 295) has remarked that his implicit notion of 'meaning'
has always been more akin to Austin's or Wittgenstein's. Notice that where truth conditional se-
mantics has been most clearly fruitful is in exploring certain structural properties of constructions
that have little to do with mind-world relations or inchoate notions of communication and pub-
licity. For discussion, see Chomsky 1977, 1996, 2000b; Hornstein 1984; McGilvray 1998; and Piet-
roski 2003a.

Mind, Vol. 116 . 464 . October 2007 @ Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
842 John Collins

Stanley and SzabO explain that the variable complex `f(i)' is 'associated'
with the nominal and 'co-habits a node with it'. This proposal might
attract some plausibility if one is treating structures like (19) as repre-
sentations mapped onto utterances by an interpreting speaker, but it is
not a serious proposal about LF unless a story is told about from where
the variable complex arises. That is, it is one thing to say where the
complex should occur, it is quite another thing to say how it arrives in
that position. Without an explanation of the origins of the complex, the
structure is effectively unconstrained by syntax. The problem for the
proposal is that, derivationally, the complex has no place in LF struc-
ture: it is parachuted in from semantics. This essential syntactic con-
straint will be missed if LF is construed as a representation of properties
of a sentence, rather than as the output of a derivational process.
By minimalist assumptions, every element at LF must be drawn from
a selection from the lexicon; no element can be present that is not a lex-
ical feature. Well, are variable complexes in the lexicon? All we are told
is that they are 'associated' with nominals and 'co-habit' nodes.
First, as Stanley (2002b) tentatively suggests, we might try to read the
variables as incorporated into the nominal projection (see Baker 1988,
for the classic account of incorporation). This would make the relation
between variable and nominal much like the relation between the noun
deer and the verb hunt in the complex intransitive verb deer-hunt. In
general, cases of incorporation reflect thematic assignment. But
Stanley's proposed covert variables are not constrained to satisfy the-
matic roles; quite the opposite: they appear to be intended to be inter-
preted as adjuncts, which, as we saw above, are essentially optional
elements of strucutre. It thus seems that there is no basis to view the
relation between overt noun and covert variable as one of incorpora-
tion.
A related option is take the relation to be one of conflation in the
sense of Hale and Keyser (2002). Conflation may be understood as 'lex-
ical internal incorporation'. For example, we may view the unaccusative
redden as a deadjectivalised verb, of the form [v red-en [ADJ <red>]].
Yet the internal structure hypothesized by Hale and Keyser does not
project in the 'sentential' syntax; in particular, the structure does not
fall under c-command relations from outside the particular lexical
items. Thus, no binding relation will be supported. Again, it seems that
we lack an extant model for what Stanley requires.
One might suggest the idea that the complexes are 'triggered' by the
merging of N with Det. This is another non-starter. Nothing can occur
in the merged object that is not present in the constituent objects; this

Mind, Vol. 116 . 464 . October 2007 CC Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
 Syntax, More or Less 843

follows from our assumptions. The nominal can be restricted, of

course, but only from further merging of lexical items, not from the
mere fact that a DP is a merged object.

4. Concluding remarks
We have seen that syntactic structure, as understood in recent gener-
ative theory, is out of step with the kinds of demands philosophers
make of it. It contains both too much and too little. As hopefully made
clear, this dissonance is by no means problematic for the linguistic the
ories, for they do not presume to capture the same intuitions or notions
that animate philosophical research. They might be off course in other
ways, but it would be obtuse to condemn them on grounds quite out-
side their empirical and theoretical range. Equally, my plea is not for a
veil of silence to fall between linguists and philosophers-quite the
opposite. The work of King, Stanley, and others exemplifies the fecun-
dity of a philosophy of language oriented to our best science of
language. If my considerations show anything, it is perhaps only that a
closer integration is required if linguistic results are to play a genuine
role in shaping philosophical accounts of language, as they must, if
such philosophy is to be a serious area of inquiry.'

School of Philosophy JOHN COLLINS

University of East Anglia

Norwich NR4 7171
UK

References

Abraham, W., S. D. Epstein, H. Thrainsson, and C. J-W. Zwart (eds.)

1996: Minimal Ideas: Syntactic Studies in the Minimalist Framework.
Amsterdam: John Benjamins.
Antony, L. M. and N. Hornstein (eds.) 2003: Chomsky and his Critics.
Oxford: Blackwell.
Bach, K. 2000: 'Quantification, Qualification, and Context: a Reply to
Stanley and SzabO's Mind and Language, 15, pp. 262-83.

m My thanks go to Dan Blair, Noam Chomsky, Gareth Fitzgerald, Keith Hossack, Jeff King, Pe-
ter Ludlow, Gabe Segal, and Barry Smith. I am especially grateful to Guy Longworth, for getting
me to see what I wanted to say, and to Paul Pietroski, for more than this note will accommodate.
Versions of the first half of this paper were delivered at the Dubrovnik Philosophy of Linguistics con-
ference in September 2006 and at the Joint Session meeting in Bristol, July 2007.

Mind, Vol. 116 . 464 . October 2007 CC Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
844 John Collins

Baker, M. 1988: Incorporation: A Theory of Grammatical Function

Changing. Chicago: Chicago University Press.
Baker, M. 2003: Lexical Categories: Verbs, Nouns, and Adjectives. Cam-
bridge: Cambridge University Press.
Baltin, M. and C. Collins (eds.) 2001: The Handbook of Contemporary
Syntactic Theory. Oxford: Blackwell.
Barber, A. (ed.) 2003: Epistemology of Language. Oxford: Oxford Uni-
versity Press.
Barbosa, P., D. Fox, P. Hagstrom, M. McGuiness, and D. Pesetsky (eds.)
1998: Is the Best Good Enough? Optimality and Competition in Syn-
tax. Cambridge, MA: MIT Press.
Belletti, A. (ed.) 2004: Structures and Beyond: The Cartography of Syn-
tactic Structures, Vol. III. Oxford: Oxford University Press.
Bezuidenhout, A. and M. Reimer (eds) 2004: Descriptions and Beyond.
Oxford: University Press.
Bhatt, R. and R. Pancheva 2006: 'Implicit Arguments'. In Everaert and
Riemsdijk 2006, pp. 558-588.
Blair, D. 2005: 'Contexts Crossed Over'. Protosociology, 21, pp. 88-104.
Boeckx, C. 2001: 'Scope Reconstruction and A-Movement'. Natural
Language and Linguistic Theory, 19, pp. 503-548.
Borer. H. 2005: The Normal Course of Events. Structuring Sense, Volume
II. Oxford: Oxford University Press.
Breheny, R. 2002: 'The Current State of (Radical) Pragmatics in the
Cognitive Sciences'. Mind and Language, 17, pp. 169-87.
Brody, M. and R. Manzini 1988: 'On Implicit Arguments'. In Kempson
1988, pp. 105-130.
Cappelen, H. and E. Lepore 2002: Inde)dcality, Binding, Anaphora, and
A Priori Truth'. Analysis, 62.4, pp. 271-81.
Cappelen, H. and Lepore, E. 2005a: 'Radical and Moderate Pragmatics:
Does Meaning Determine Truth Conditions?'. In SzabO 2005,
pp. 45-71.
Cappelen, H. and Lepore, E. 2005b: Insensitive Semantics: A Defense of
Semantic Minimalism and Speech Act Pluralism. Oxford: Blackwell.
Carston, R. 2002a: 'Linguistic Meaning, Communicated Meaning and
Cognitive Pragmatics'. Mind and Language, 17, pp. 127-48.
Carston, R. 2002b: Thoughts and Utterances: The Pragmatics of Explicit
Communication. Oxford: Blackwell.
Chomsky, N. 1955-56/75: The Logical Structure of Linguistic Theory. New
York: Plenum.
Chomsky, N. 1957: Syntactic Structures. The Hague: Mouton.

Mind, Vol. 116 . 464 . October 2007 CC Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
 Syntax, More or Less 845

Chomsky, N. 1965: Aspects of the Theory of Syntax. Cambridge, MA:

MIT Press.

Chomsky, N. 1972: Studies on Semantics in Generative Grammar. The

Hague: Mouton.
Chomsky, N. 1975: Reflections on Language. London: Fontana.
Chomsky, N. 1977: Essays on Form and Interpretation. Amsterdam:
North-Holland.
Chomsky, N. 1980: Rules and Representations. New York: Columbia
University Press
Chomsky, N. 1981: Lectures on Government and Binding. Dordrecht:
Foris.

Chomsky, N. 1986a: Knowledge of Language: Its Nature, Origin and Use.

Westport: Prager.
Chomsky, N. 1986b: Barriers. Cambridge, MA: MIT Press.
Chomsky, N. 1995a: The Minimalist Program. Cambridge, MA: MIT
Press.

Chomsky, N. 1995b: 'Bare Phrase Structure'. In Webelhuth 1995,

pp. 481-539.
Chomsky, N. 1996: Powers and Prospects: Reflections on Human Nature
and the Social Order. London: Pluto Press.
Chomsky, N. 1998: 'Some Observations on Economy in Generative
Grammar'. In Barbosa, Fox, Hagstrom, McGuiness, and Pesetsky
1998, pp. 115-27.
Chomsky, N. 2000a: 'Minimalist Inquiries: the Framework'. In Martin,
Michaels, and Uriagereka 2000, pp. 89-155.
Chomsky, N. 2000b: New Horizons in the Study of Language and Mind.
Cambridge: Cambridge University Press.
Chomsky, N. 2001: 'Derivation by Phase'. In Kenstowicz 2001, pp. 1-52.
Chomsky, N. 2003: 'Reply to Horwich'. In Antony and Hornstein 2003,
pp. 295-304.
Chomsky, N. 2o04a: 'Beyond Explanatory Adequacy'. In A. Belletti
2004, pp. 104-31.
Chomsky, N. 20041): The Generative Enterprise Revisited. Berlin: Mou-
ton de Gruyter.
Chomsky, N. 2o05a: 'Three Factors in Language Design. Linguistic
Inquiry, 36, pp. 1-22.
Chomsky, N. 2005b: 'On Phases'. Unpublished ms.
Chomsky, N. 2007: 'Approaching UG from Below'. In Sauerland and
Gartner 2007, pp. 1-29.
Chomsky, N. and H. Lasnik 1977: 'Filters and Control'. Linguistic
Inquiry, 8, pp. 425-504.

Mind, Vol. 116 . 464 . October 2007 CC Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
846 John Collins

Collins, C. 2001: 'Economy Conditions in Syntax'. In M. Baltin and C.

Collins 2001, pp. 45-61.
Crimmins, M. and Perry, J. 1989: 'The Prince and the Phone Booth:
Reporting Puzzling Beliefs'. Journal of Philosophy, 86, pp. 685-711.
den Dildcen, M. 2006: Relators and Linkers: The Syntax of Predication,
Predicate Inversion, and Copulas. Cambridge, MA: MIT Press.
Elgardo, R. and Stainton, R. 2004: 'Shorthand, Syntactic Ellipsis, and
the Pragmatic Determinants of What is Said'. Mind and Language,
19, pp. 442-71.
Epstein, S. 1999: 'Un-Principled Syntax and the Derivation of Syntactic
Relations'. In Epstein and Hornstein 1999, pp. 317-45.
Epstein, S., E. Groat, R. Kawashima, and H. Kitahara 1998: A Deriva-
tional Approach to Syntactic Relations. Oxford: Oxford University
Press.

Epstein, S. and N. Hornstein (eds.) 1999: Working Minimalism. Cam-

bridge, MA: MIT Press.
Epstein, S. and T. D. Seely 2002a: 'Rule Applications as Cycles in a
Level-Free Syntax'. In Epstein and Seely 2002b, pp. 63-89.
Epstein, S. and T. D. Seely (eds) 2002b: Derivation and Explanation in
the Minimalist Program. Oxford: Blackwell.
Epstein, S. and T. D. Seely 2006: Derivations in Minimalism. Cam-
bridge: Cambridge University Press.
Everaert, M. and H. van Riemsdijk (eds.) 2006: The Blackwell Compan-
ion to Syntax, Volume II. Oxford: Blackwell.
Fodor, J. 2001: 'Language, Thought and Compositionality's Mind and
Language, 16, pp. 1-15.
Fox, D. 2000: Economy and Semantic Interpretation. Cambridge, MA:
MIT Press.

Fox, D. 2003: 'On Logical Form'. In Hendrick 2003, pp. 82-123.

Grice, P. 1989: Studies in the Ways of Words. Cambridge, MA: Harvard
university Press.
Haegeman, L. (ed.) 1997: Elements of Grammar: Handbook of Generative
Syntax. Dordrecht: Kluwer.
Hale, K. and S. Keyser 2001: Prolegomenon to a Theory of Argument
Structure. Cambridge, MA: MIT Press.
Harman, G. 1974: 'Logical Form'. Foundations of Language, 9, pp. 38-65.
Hendrick, R. (ed.) 2003: Minimalist Syntax. Oxford: Blackwell.
Higginbotham, J. 1985: 'On Semantics'. Linguistic Inquiry, 16, pp. 547-
93

Higginbotham, J. 1986: 'Linguistic Theory and Davidson's Program in

Semantics'. In Lepore 1986, pp. 29-48.

Mind, Vol. 116 . 464 . October 2007 CC Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
 Syntax, More or Less 847

Hornstein, N. 1984: Logic as Grammar: An Approach to Meaning in Nat-

ural Language. Cambridge, MA: MIT Press.
Hornstein, N. 1995: Logical Form: From GB to Minimalism. Oxford:
Blackwell.
Hornstein, N. 2001: Move! A Minimalist Theory of Construal. Oxford:
Blackwell.
Hornstein, N. 2003: 'On Control'. In Hendrick 2003, pp. 6-81.
Jackendoff, R. 1972: Semantic Interpretation in Generative Grammar.
Cambridge, MA: MIT Press.
Jackendoff, R. 1987: 'The Status of Thematic Relations in Linguistic
Theory'. Linguistic Inquiry, 18, pp. 369-411.
Kayne, R. 1994: The Antisymmetry of Syntax. Cambridge, MA: MIT
Press.

Kempson, R. (ed.) 1988: Mental Representations: The Interface between

Language and Reality. Cambridge: Cambridge University Press.
Kenstowicz, M. (ed.) 2001: Ken Hale: A Life in Language. Cambridge,
MA: MIT Press.

King, J. 1994: 'Can Propositions be Naturalistically Acceptable?' Mid-

west Studies in Philosophy, 19, pp. 53-75.
King, J. 1995: 'Structured Propositions and Complex Predicates'. Notts,
29, pp. 516-35.
King, J. 1996: 'Structured Propositions and Sentence Structure'. Journal
of Philosophical Logic, 25) pp. 495-521.
King, J. 2ool: Complex Demonstratives: A Quantificational Approach.
Cambridge, MA: MIT Press.
King, J. 2002: 'Two Sorts of Claim about "Logical Form". In Preyer and
Peter 2002, pp. 118-31.
King, J. 2007: The Nature and Structure of Content. Oxford: Clarendon
Press.

King, J. and J. Stanley 2005: 'Semantics, Pragmatics, and the Role of

Semantic Content'. In SzabO 2005, pp. 111-64.
Kitahara, H. 1996: 'Raising Quantifiers without Quantifier Raising'. In
Abraham, Epstein, Thrainsson, and Zwart 1996, pp. 189-98.
Koopman, H. and D. Sportiche 1991: 'The Position of Subjects'. Lingua,
85, pp. 211-58.
Kratzer, A. 1996: 'Severing the External Argument from the Verb'. In
Rooryck and Zaring 1996, pp. 109-37.
Larson, R. and P. Ludlow 1993: 'Interpreted Logical Forms'. Synthese, 95,
pp. 305-55.
Larson, R. and G. Segal. 1995: Knowledge of Meaning: An Introduction to
Semantic Theory. Cambridge, MA: MIT Press.

Mind, vol. 116 . 464 . October 2007 CC Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
848 John Collins

Lasnik, H. and J. Uriagereka 2005: A Course in Minimalist Syntax: Foun-

dations and Prospects. Oxford: Blackwell.
Lepore, E. (ed.) 1986: Truth and Interpretation: Perspectives on the Phi-
losophy of Donald Davidson. Oxford: Blackwell.
Lepore, E. 2004: 'An Abuse Context in Semantics: the Case of Incom-
plete Definite Descriptions'. In Bezuidenhout and Reimer 2004,
pp. 41-67.
Longobardi, G. 2001: `The Structure of DPs: Some Principles, Parame-
ters and Problems: In Baltin and Collins 2001, pp. 562-603.
Marti, L. 2006: 'Unarticulated Constituents Revisited'. Linguistics and
Philosophy, 29, pp. 135-66.
Martin, R., D. Michaels, and J. Uriagereka (eds) 2000: Step by Step:
Essays on Minimalist Syntax in Honor of Howard Lasnik. Cam-
bridge, MA: MIT Press.
Martin, R. and J. Uriagereka 2000: `Some Possible Foundations of the
Minimalist Program'. In Martin, Michaels, and Uriagereka 2000,
pp. 1-30.
May, R. 1977: The Grammar of Quantification. Unpublished disserta-
tion. MIT.

May, R. 1985: Logical Form: Its Structure and Derivation. Cambridge,

MA: MIT Press.

McClosky, J. 1997: (Subjecthood and Subject Positions'. In L. Haegeman

1997, pp. 197-253.
McGilvray, J. 1998: 'Meanings are Syntactically Individuated and Found
in the Head'. Mind and Language, 13, pp. 225-80.
Neale, S. 1994: 'Logical Form and LF'. In C. Otero 1994, pp. 788-838.
Neale, S. 2000: 'On Being Explicit: Comments on Stanley and SzabO,
and on Bach'. Mind and Language, 15, pp. 284-94.
Neale, S. 2004, `This, That, and the Other'. In Reimer and Bezuidenhout
2004, pp. 68-181.
Neale, S. 2005: `Pragmatism and Binding'. In Szab6 2005, pp. 163-285.
Nunes, J. 2004: Linearization of Chains and Sideward Movement. Cam-
bridge, MA: MIT Press.
Otero, C. (ed.) 2004: Noam Chomsky: Critical Assessments. London:
Routledge.
Pagin, P. 2005: `Compositionality and Context'. In G. Preyer and G.
Peter 2005, pp. 303-48.
Partee, B. 1989: `Binding Implicit Variables in Quantified Contexts'. In
Papers from CLS 25, pp. 342-56. Reprinted in B. Partee 2004, Com-
positionality in Formal Semantics. Oxford: Blackwell, pp. 259-81.

Mind, Vol. 116 . 464 . October 2007 CC Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
 Syntax, More or Less 849

Perry, J. 1986: 'Thoughts without Representation'. Supplementary Pro-

ceedings of the Aristotelian Society, 6o, pp. 137-52.
Pietroski, P. 2oo3a: 'The Character of Natural Language Semantics'. In
Barber 2003, pp. 217-56.
Pietroski, P. 2003b: 'Small Verbs, Complex Events: Analyticity without
Synonymy'. In Antony and Hornstein 2003, pp. 179-214.
Pietroski, P. 2oo5a: Events and Semantic Architecture. Oxford: Oxford
University Press.
Pietroski, P. 2005b: 'Meaning before Truth'. In Preyer and Peter 2005,
pp. 255-302.
Preyer, G. and G. Peter (eds) 2002: Logical Form and Language. Oxford:
Oxford University Press.
Preyer, G. and G. Peter (eds) 2005: Contextualism in Philosophy: Knowl-
edge, Meaning, and Truth. Oxford: Oxford University Press.
Pustejovsky, J. 1995: The Generative Lexicon. Cambridge, MA: MIT
Press.

Recanati, F. 2002: 'Unarticulated Constituents'. Linguistics and Philoso-

phy, 25, pp. 299-345.
Recanati, F. 2004: Literal Meaning. Cambridge: Cambridge University
Press.

Rizzi, L. 1986: 'Null Objects in Italian and the Theory of pro'. Linguistic
Inquiry, 17, pp. 501-57.
Rooryck, J. and L. Zaring (eds.) 1996: Phrase Structure and the Lexicon.
Dordrecht: Kluwer.
Safir, K. 2004a: The Syntax of (In)Dependence. Cambridge, MA: MIT
Press.

Safir, K. 2004b: The Syntax of Anaphora. Oxford: Oxford University

Press.

Sauerland, U. and H-M. Gartner (eds.) 2007: Interfaces + Recursion =

Language? Chomsky's Minimalism and the View from Syntax-
Semantics. Berlin: Mouton de Gruyter.
Schiffer, 5.1987: Remnants of Meaning. Cambridge, MA: MIT Press.
Soames, S. 2002: Beyond Rigidity: The Unfinished Semantic Agenda of
Naming and Necessity. Oxford: Oxford University Press.
Soames, S. 2005: 'Naming and Asserting'. In Z. SzabO 2005, pp. 356-82.
Stanley, J. 2000: 'Context and Logical Form'. Linguistics and Philosophy,
23, pp. 391-434.
Stanley, J. 20 02a: 'Making it Articulated'. Mind and Language, 17,
pp. 149-68.
Stanley, J. 2002b: 'Nominal Restriction'. In Preyer and Peter 2002,
pp. 365-88.

Mind, vol. 116 . 464 . October 2007 @ Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms
850 John Collins

Stanley, J. 2005: 'Semantics in Context'. In Preyer and Peter 2005,

pp. 221-54.
Stanley, J. and Z. G. SzabO 2000a: 'On Quantifier Domain Restriction'.
Mind and Language, 15, pp. 219-61.
Stanley, J. and Z. G. SzabO 2000b: 'Reply to Bach and Neale'. Mind and
Language, 15, pp. 295-98.
Stanley, J. and T. Williamson 2001: 'Knowing How'. Journal of Philoso-
phy, 98, pp. 411-44.
SzabO, Z. G. (ed.) 2005: Semantics versus Pragmatics. Oxford: Oxford
University Press.
Taylor, K. 2001: 'Sex, Breakfast, and Descriptions'. Synthese,128, pp. 45-
6i.

Uriagereka, J. 1998: Rhyme and Reason: An Introduction to Minimalist

Syntax. Cambridge, MA: MIT Press.
Uriagereka, J. 1999: 'Multiple Spell-Out'. In Epstein and Hornstein 1999,
pp. 251-82.
Uriagereka, J. 2002: Derivations: Exploring the Dynamics of Syntax. Lon-
don: Routledge.
Webelhuth, G. (ed.) 1995: Government and Binding Theory and the Min-
imalist Program. Oxford: Blackwell.
Williams, E. 1985: 'PRO and the Subject of NP'. Natural Language and
Linguistic Theory, 3, pp. 397-415.

Mind, Vol. 116 . 464 . October 2007 CC Collins 2007

This content downloaded from

49.36.171.116 on Fri, 16 Jun 2023 07:52:52 +00:00
All use subject to https://about.jstor.org/terms