what is coordination?
TRANSCRIPT
What is coordination?$
Annabel Cormack*, Neil Smith
Department of Linguistics, University College London, Gower Street, London WC1E 6BT, UK
Received 25 August 2003; accepted 25 August 2003
Available online 17 December 2003
Abstract
We argue that there are no devices in the grammar specific to coordination. The grammar is only
capable of providing asymmetric structures, through particular lexical entries relating to semantic
conjunction. Such entries produce adjunction structures, rather than head-complement structures.
The interpretation of conjunction structures is a joint function of such lexical entries, processing
properties, and pragmatics. Coordination phenomena are the result of an unresolved ambivalence
between a ‘head initial’ and a ‘head final’ asymmetric conjunction structure, with the effect that there
are parallel representations.
# 2003 Elsevier B.V. All rights reserved.
Keywords: Coordination; Conjunction; Adjunction; Syntax; Semantics; Lexicon; Minimalism; CCG
1. Introduction
Our purpose in this paper is to argue for the position that there are no devices in the
grammar specific to coordination. Coordination appears to be symmetric, but the
grammar is only capable of providing asymmetric structures. In a standard Principles
and Parameters version of projection, two phrasal categories can be related in either of
two ways. They may be linked (asymmetrically) to a particular head as specifier or
complement of that head, or they may be linked (again asymmetrically) as adjunct and
host. We see the adjunct�host relation, like the complement�specifier relation, as also
essentially head-mediated and, in the case of conjunction, we argue that the particular
lexical entries encoding semantic conjunction relate the two conjuncts as adjunct and
Lingua 115 (2005) 395–418
$ This is a revised version of the paper given at the 4th NWCL International Conference ‘Coordination,
syntax, semantics and pragmatics’ at Salford in November 2001.* Corresponding author. Tel.: þ44-20-7679-7173; fax: þ44-20-7383-4108.
E-mail addresses: [email protected] (A. Cormack), [email protected] (N. Smith).
0024-3841/$ – see front matter # 2003 Elsevier B.V. All rights reserved.
doi:10.1016/j.lingua.2003.09.008
host, rather than as specifier and complement.1 The interpretation of conjunction
structures is then a joint function of such lexical entries, processing considerations,
and pragmatics. We argue that the grammar exploits both asymmetric (subordinating)
and symmetric (coordinating) conjunction structures, but that coordination requires no
further elaboration of the grammar: coordination phenomena result from the occurrence
of parallel ‘head initial’ and ‘head final’ asymmetric conjunction structures.
We first clarify the terminology and notation we use. Conjunction ‘^’, disjunction ‘_’,
and implication ‘�’ are logical operators—specifically, two-place operators. Negation is
a one-place logical operator. Lexical items (connectives) like and, but, and because, or,and if, are either Natural Language two-place operators corresponding to traditionaltruth-functional operators for the non-pragmatic part of their meaning, or they aremarkers, much like agreement markers, which are associated with such aNL operator(which may itself be phonologically null).
Coordination is a particular syntactic manifestation of conjunction (or disjunction),
which is symmetric with respect to the conjuncts/disjuncts (i.e. the conjuncts/disjuncts
have the same syntactic status). Without theoretical commitment, we may use &, instead of
^, to indicate that conjuncts are coordinated, and refer to this as coordinating conjunc-tion. Subordination, e.g. as induced by implication (if), is asymmetric: there is an
adjunct and a host. We may use $ to indicate asymmetric conjunction, instead of ^ (see
Smith, 1999). We mostly use a ‘bare’ category notation, where V, for instance, stands for
some projection of a verb. For the operators, we use the full categorial notation, which
includes the selection categories.
Our analysis of coordination exploits a simplified version of Minimalism, with the
addition of Combinators from Combinatorial Categorial Grammar (CCG). In such a theory,
most of the work of the grammar is done by the lexicon. We assume that the combinators
are included in the lexicon: they have syntax, semantics, and (null) phonology.
In contradistinction to standard CCG, but consistent with Minimalism, we assume that
Merge produces an LF representation within NL syntax, with LF being the representation
presented to the Conceptual-Intentional Interface for non-linguistic inferential processing.
All merge is driven by selection.2 We assume that the LF representation, like the PF
representation, is ordered: the unmarked option is that selector precedes selected.3 PF
ordering is derivative, and obtained by displacement. In particular, the PF-part of a head or
phrase may be displaced to some other position because of the morphological selection
properties of the head at that position. This process, which we refer to as PF-attraction, or
simply attraction, has no effect on LF ordering. For expository convenience, we talk of
‘movement’, though we have argued elsewhere (Cormack and Smith, 2001a) that there is
no real movement in the grammar of NL.
1 See Cinque (1994, 1999) for adjectival and adverbial phrases as Specifiers. Sportiche (1994) suggests for
adjuncts only a Spec-Complement relation. See also discussion in Cormack (1999).2 Cormack and Smith (1994); this assumption is fundamental in Categorial Grammar. Chomsky
(2000: 133–134) endorses this principle for ordinary heads, but excludes adjuncts.3 Of course, we do not know how a tree structure, or a tree with a precedence relation defined over it, is
represented in the mind/brain. However, given items of fixed arity, ‘functor first’ allows a linear bracket-free
representation of a tree with a suitable precedence relation (see Partee et al., 1990: 439; for Polish notation, see
McCawley, 1981).
396 A. Cormack, N. Smith / Lingua 115 (2005) 395–418
Arguments are discharged one at a time, resulting in binary branching. Because
argument noun phrases (DNPs) are always headed by a determiner which has the semantic
and syntactic selection properties of a generalised quantifier, noun phrase arguments never
have type hei, but rather hhe, ti, ti and so on.4 Hence, these arguments are functors over VPs
of type he, ti, and so typically precede the lexical selecting V head (though a movement
account from an underlying head-initial order would not be incompatible with our
analysis). VO order at PF is obtained by movement of V to some head Agr, equivalent
in effect to Larson’s (1988) higher V, Bowers’ (1993) Pr head, or Chomsky’s (1995: 60,
352) AgrO or ‘little v’. A comparable treatment of phrasal adjuncts requires either that they
be headed by a two-place operator, selecting first for the rest of the adjunct and finally for
the host, or that they be adjoined by means of some two-place operator which selects for the
adjunct and then for the host. This gives rise in the unmarked case to the order
‘Adjunct < ðprecedesÞ Host’.
We take it that syntax exploits the lexicon under the operation of a principle of default
reasoning which goes by a number of names, including most notably the ‘Penguin
Principle’.
(1) Penguin Principle: Default Reasoning (Elsewhere Condition; Panini Principle)5
If two rules/operations can apply to the same input and their application would
give rise to different outputs, then the one which applies more specifically to the
input has priority
That is, lexical entries compete, with more highly specified entries outranking more
general ones. For example, for the past tense form of the verb give, the competition is
between the form gived, arising from the general rule, and the form gave which is
idiosyncratically specified. It is the more specific form which wins.
By hypothesis, the grammar contains combinators. These occupy the head positions that
we have labelled Agr in our examples. We have suppressed most reference to combinators in
this paper but, if we exploit them fully, we can eliminate the ‘head-final’ versus ‘head initial’
microparameters from the grammar (Cormack and Smith, 2001b: Appendix).
We do not here offer an analysis of all aspects of coordination, omitting in particular
many issues relating to noun phrase coordination (including agreement and set-taking
predicates), and multiple conjunction structures. We concentrate on the syntactic asym-
metry versus symmetry issue, and the theoretical status of coordination itself.
2. Asymmetric versus symmetric conjunction
Ever since the introduction of X-bar theory, the symmetry of coordination has been a
problem. Standardly, X0 theory offers head-mediated phrasal relations between specifier
4 See also Cormack (1999). For an example with a proper name, see (48b). To take care of object and oblique
noun phrase arguments, we assume polymorphic quantifiers (see discussion in Heim and Kratzer, 1998: 180–182).5 See Briscoe et al. (1995) and Asher and Morreau (1991) for discussion, and also Gazdar (1987), Evans and
Gazdar (1989). For uses in phonology and morphology, see Kiparsky (1973), Anderson (1992: 132), Stump
(2001, e.g., 23 ff). Our informal formulation is based on the latter characterisations, for concreteness.
A. Cormack, N. Smith / Lingua 115 (2005) 395–418 397
and complement, or adjunct and host. Neither of these is a symmetric relation.6
Nevertheless, we argue that the symmetry of coordination can be accounted for using
the adjunction relation. In order to do this, we must first explore ASYMMETRIC con-
junction.
Consider the contrast between the two logical connectives and and if. What we get with
the logical connective if is ASYMMETRIC, with a host and an adjunct of differing status.
Extraction is possible from the host, but not from the adjunct, as we see in (2). In contrast,
what we see with and as the logical connective, in examples like (3), is SYMMETRIC, with
extraction allowed from NEITHER conjunct.
(2) a [[If [John takes the car]], [you will go by bus]]
b How, [[if [John takes the car]], [will [you taux go t]]]?
c �What, [if [John takes t]], [will [you taux go by bus]]?
(3) a [^ [You will go by bus] [and [John must take the car]]]
b �How [^ [will [you taux go t]] [and [John must take the car]]]
c �What [^ [ you will go by bus] [and [must [John taux take t]]]]
This syntactic symmetry or asymmetry reflects, and is motivated by, the inferential
symmetry in relation to the two operands of logical operators like and and or, in contrast
to the inferential asymmetry of the logical operator if. The differences between symmetric
and asymmetric operators can be seen clearly with closely related operators. Consider the
examples in (4).
(4) a John will leave soon, since Mary has left
What, since Mary has left, will John do? since: asymmetric
b John will leave soon, because Mary has left�What, because Mary has left, will John do? because: symmetric
The difference in grammatical status parallels differences in inferential status. In (4a),
that Mary has left is assumed to be already known, and so is the inferential/causal
connection between John’s leaving and Mary’s leaving. That John will leave soon is
asserted, and asserted to have been deducible from known information. In (4b), both
that John will leave soon and that Mary has left are asserted, as is the inferential/causal
connection.
We show next that not all conjunction is coordinating or SYMMETRIC. Indeed, not only is
subordinating, or ASYMMETRIC conjunction widespread, but coordinating (i.e. SYMMETRIC)
6 For a summary and bibliography of work on coordination, see Progovac (1998a,b). Attempts to base a
theory of coordination on the specifier–complement relation include Kayne (1994), Johannessen (1998), and
Zoerner (1999). Munn (1993, 2000), Thiersch (1996), Buring and Hartmann (1998), Velde (2000), and others
argue for a basis of adjunction. Borsley (1994) enumerates problems for symmetric coordination with most such
analyses. Many authors (e.g. Johannessen, 1998; Munn, 2000) claim that asymmetries in agreement point to an
inherent asymmetry in conjunction. We see this as an orthogonal issue.
398 A. Cormack, N. Smith / Lingua 115 (2005) 395–418
conjunction is derivative, and hence not anomalous. We base our arguments on symmetries
and asymmetries in EXTRACTION, SELECTION, and SEMANTIC TYPE.
We confine our comparison of asymmetric and symmetric operators mainly to the
logical connectives conjunction and implication, associated with and and if. Both the
logical connectives are semantically two-place operators, i.e. they select for two operands.
The default assumption is that their syntactic selection is the same. Given binary Merge, we
expect these operands to be selected one at a time, so that if if is the phonological realisation
of implication, the simplest structure is as in (5). We show this with the operator preceding
its operands, and indexed to show its arity.
(5) [[2if Q] P]
In general, a phrasal adjunct such as [if Q] is headed by a two-place operator, which
selects initially for the ADJUNCT Q and then for the HOST P. What differentiates minor heads
like if, which introduce adjuncts, from lexical or functional projections, is that the syntactic
category of the mother projection is not that of the head, but that of the host (i.e. the final
operand).7 The general two-place operator induces a tree like that in (6a i), exemplified as
shown in (6a ii) for the operator if. Using the Categorial Grammar notation, we code this in
the lexicon by assigning the operator the category shown in (6b), or (6c) for if. The features
with slashes represent the selections, with the selection discharged first being outermost.
Selection features are deleted as they are discharged.
Convention: variables over phrases of a given category, such as X, Y, will be used in such
a way that the alphabetic or subscript order reflects the LF linear order of the operands.
The assumption in (5) and (6c) is that the head if is the two-place connective. However,
there are reasons to suppose that a more elaborate structure might be needed for some
connectives. For example, we might propose the structures in (7), for if. Here, the
conditional operator itself is phonologically null, but there is an associated semantically
7 See Keenan and Faltz re modifiers (1985: 118 ff for a one-place adjunct, 197 for a two-place adjunct). The
categorial notation X/X for an adjunct instantiates Grimshaw’s (1991: 29) claim that functional categories may
be category neutral, transmitting the category of their complement.
A. Cormack, N. Smith / Lingua 115 (2005) 395–418 399
trivial one-place operator if on the antecedent of the conditional, and an optional one-place
operator then on the consequent, which are effectively MARKERS of agreement with the
phonologically empty two-place operator head.
(7) [[2� [1if Q]] [(1then) P]]
We will use the simpler version for if, although the other is probably right. Analogously,
one might expect conjunction structures of the form in (8):
(8) a [[2^ [(1both) Q]] [1and P]]
b [[2^ [(1either) Q]] [1or P]]
That and is a marker rather than a two-place operator is suggested by the fact that it occurs
in other conjunction-related and non-conjunction structures. Examples are alternately hot
and cold (where alternately is a two-place operator), and ‘modal subordination’ cases such
as Try and come and Come any closer and I’ll shoot. The last is syntactically as well as
semantically a conditional structure, as witness the licensing of any in the antecedent. But a
structure such as (7) still leaves the symmetry of coordination unexplained. Since we will
not give our explanation of symmetric coordination until Section 8, we may use structures
like (9) for conjunction.
(9) [^ [ Q] [and P]]
We will assume that in the unmarked case a head precedes its complements or operands.
3. Symmetric (coordinating) operators
3.1. Semantics and pragmatics
There is a sense in which conjunction is the most primitive binary operator. Assertion of
‘P’, followed by assertion of ‘Q’, is semantically (but not pragmatically) equivalent to
assertion of the conjunction ‘P ^ Q’. It is not surprising then that this operator may be
phonologically null (see Payne, 1985: 25–27).
We follow the now standard assumption that where logical conjunction is used, there
may be varied pragmatically derived implicatures or explicatures (Grice, 1967/1989;
Sperber and Wilson, 1995; Carston, 1988, 2002). The interpretations in (10) are plausible
and regularly available:
(10) a He [did some weeding and wrote a few pages
of the paper]
simple conjunction
b Hermione got pregnant and looked for a husband temporal ordering
c Jenny hit Billy and he began to cry cause-effect
d Lucinda nibbled biscuits and read a book simultaneity
400 A. Cormack, N. Smith / Lingua 115 (2005) 395–418
We see these same meanings arising in other languages, and for phonologically null
instances of the conjunction operator. In the asymmetric interpretations, the asymmetry is
based on ICONICITY, with temporal events paralleling the temporal appearance of the
relevant phrases, and the cause-effect ordering relying on the temporal non-commutativity
of cause and effect. We have argued in previous work (Cormack and Smith, 1999), that the
ordering underlying the iconicity should be identified not at PF, but at an ordered LF. We
return to this later.
3.2. Syntax
Informally, coordination behaves as if it were regulated by the following rule of thumb:
(11) Rule of thumb for coordination structures:
Each conjunct or disjunct behaves as if it were the host
In particular, each must have the same category and semantic type, or rather, each must
have a category and semantic type which is fit for the environment in which the whole
structure occurs (the ‘external homogeneity condition’ of Hohle, 1991). As shown in Sag
et al. (1985), conjuncts of mismatched category are not excluded by the grammar. The
structures in (12) are possible because predicational be is eclectic in its c-selection options,
accepting projections of P, V, A and N, and all the conjuncts are of the appropriate type, that
of a predicate, type he, ti:
(12) a John is [& [P in a temper] [and [V surrounded by fools]]
b John is [& [A hungry] [and [P in a temper]]]
The ‘rule of thumb’ permits distinct categories of conjuncts, each apparently acting as host,
as in the examples in (12), but says nothing about what the mother category is in the
conjunct phrase. In GPSG and HPSG, a cover-category corresponding to a lexical
projection (Lex) exploiting the features [�N, �V] is used. But this still leaves a puzzle
as to why the features should not project, in coordination, or if they do project, why they do
not clash, when in other circumstances the [�N,�V] features from the head must project in
order that there can be differential selection of a projection, say for A but not N. Our
coordination solution will sidestep this problem.8
We presume without argument that constituent conjunction is permitted. We assume
the combinatory account, using S (see Steedman, 1987, 1989, 1990, 1993, and
Szabolcsi, 1992).9 Under this account, although the lexical entry for the binary
operator of conjunction is of type ht, ht, tii (i.e. expecting two propositional arguments
to return a truth value), it may also combine with two operands of type a, where ais some unsaturated type such as he, ti. In effect, it behaves as if it had, type ha, ha, tii
8 See also Bayer (1996) for discussion.9 S is defined by the identity ðSfgÞx ¼ f ðxÞ � gðxÞ. That is, Sfg ¼ lx½f ðxÞ � gðxÞ�.
A. Cormack, N. Smith / Lingua 115 (2005) 395–418 401
(see Gazdar, 1980 or Keenan and Faltz, 1985 for theories assigning this type in the
lexicon). Concomitantly, there is a wide variety of syntactic categories that can be
coordinated.
The rule of thumb predicts correctly that there cannot be extraction from just one
conjunct. In (13b), [a father of twins] cannot be host, because [Who is John a father of
twins] is ungrammatical, so [a father of twins] and [fond of t] fail the rule of thumb test and
cannot be coordinated; and in (13c) [fond of his wife] cannot be host, because [How many
children is John fond of his wife] is ungrammatical, so that again, the two putative conjuncts
fail the rule of thumb test.
(13) a John is [[a father of twins] [and [fond of his wife]]]
b �Who is John taux [& [N a father of twins]] [and [A fond of t]]]?
c �How many children is John taux [& [the father of t] [and [fond of his wife]]]
There are of course other coordinating operators, such as or, but we will not discuss these
in detail here.
4. Asymmetric operators
Logical operators like^ (‘and’) and_ (‘or’) are symmetric with respect to the contribution
of the two operands, and may give rise to coordination structures. Other binary operators like
� (‘if’) are asymmetric. Asymmetric binary operators in natural language usually give rise to
asymmetric structures which are distinct from coordination structures. Since we are going to
argue that some conjunction structures in Natural Language are asymmetric rather than
symmetric, we discuss some of the relevant properties of structures headed by asymmetric
operators first, exemplifying with English if.
Subordinating operators are less promiscuous in their selection possibilities than
coordinating conjunction is, but do allow some variation. Consider again if. In (14), in
both the (a) and (b) versions, unless there is massive deletion we have two operands of type
he, ti.
(14) a Raw fish T is [[dangerous to eat] [if stale]]
b Raw fish T [[is dangerous to eat] [if stale]]
The semantics offered for constituent conjunction extends readily to the conditional, so we
may take this as prima facia evidence that if too may appear to have type ha, ha, tii for
appropriate a.
More interestingly, the asymmetric connective if may have semantically and syntacti-
cally unmatched operands. Stroik (1990) and Pesetsky (1995: 161) showed by using
binding possibilities that adjuncts can be attached lower in the clause than one might
expect. For example, in (15a), a bound variable reading is possible, so that the indirect
object must have scope over, and hence c-command, the adjunct. We assume that the
402 A. Cormack, N. Smith / Lingua 115 (2005) 395–418
underlying LF structure is something like that in (15 b), with the adjunct attached at the
lowest V-projection level. ‘Head movement’ of the verb to Agr gives the PF-order in (15a).
(15) a John gives [each girl]k money if shek asks for it (but not the boys)
b [John T [Agr [each girl]k [money [V [[if [C shek asks for it]][V gives]]]]]]
c if category: V/V/C; semantics: IF, where if ¼ lulv [u ! v] for u and v of
type hti
Here, if needs to select for operands headed by C and V. Of these, only the host, V, is
compatible with the higher selecting element, T. Not only do the categories of the two
operands of the connective if not match, but the types do not. The first operand of if is of
type hti, and the second is of type he, he, he, tiii. It is straightforward to give an appropriate
syntactico-semantic account of such a structure using the combinators of CCG. We assume
that if has type ht, ht, tii here, so that it combines with its first operand, the clause, by
function-argument application (A) as usual, giving type ht, ti. The next Merge however
must use generalised function composition, using the combinator B (Steedman, 1989,
1990, 1993).10 The representation of the VP in (16a) gives the correct semantics, equivalent
to (16b), and the correct syntax, in parallel.11
(16) a [B [ A IFht, ht, ti [SHE ASKS FOR IT]hti]ht, ti [GIVES]he, he, he, tiii]he, he, he, tiiib lz ly lx [[IF.[SHE ASKS FOR IT]][GIVES.z.y.x]
An example showing details is given in Appendix A.
We are crucially assuming that there are no traces which represent variables in the
grammar, so that extraction requires the use of the combinator B, which in effect passes the
selection information relating to the ‘gap’ up the tree. The use of B will permit adjunction
of some category to a host which is less saturated, as above, but not to a host which is more
saturated, so that it is predicted within CCG that there can be extraction from the host but
not the adjunct in an adjunction structure. It is essentially this fact that gives us an
explanation of the restrictions on extraction from constructions with conjunction that we
discuss below. Steedman (1987) shows that using the combinator S allows extraction from
both operands, giving parasitic gap structures.
5. Asymmetric conjunction
5.1. Head-initial and head-final operators
In this section, we illustrate instances of conjunction exhibiting asymmetries compar-
able to those shown by if, supporting our claim that the structures involved are adjunction
structures. We begin in Section 5.2 with examples of noun modification and of secondary
10 The combinator B is defined by ðBfgÞx ¼ f ðgðxÞÞ, so that in effect it allows the immediate combination of
f and g in anticipation of x ðBfgÞ ¼ lxf ðgðxÞÞ so that we may combine f having category X/Y and type hb, ai and
g having category Y/Z and type hg, bi to give a constituent of category X/Z and type hg, ai.11 Here, we have taken the pronouns to be referential, for simplicity.
A. Cormack, N. Smith / Lingua 115 (2005) 395–418 403
predication, where the conjunction operator is head-initial; and then in Section 5.3 we
argue that, somewhat surprisingly for English, there are other cases of adjunction to a
verbal head where the conjunction operator is head-final. In all these cases, we argue that
conjunction gives the correct semantics for the structures.
5.2. Head-initial asymmetric conjunction
Our first example concerns the modification of nouns. In the typical case of noun
modification, the relation between the meanings of the modifier and the noun is INTER-
SECTIVE, in the sense that if the modifier meaning and the noun meaning are taken to
correspond to sets, then the meaning of the whole corresponds to the intersection of the two
sets. In such cases we can represent the modification as headed by the conjunction operator,
as shown in (17).
(17) [D every [N ^ [Ared] [Nball]]]
Semantics: The conjunction entails that the phrase correctly represents ‘being red AND
being a ball’. Simply adjoining the adjective to the noun fails to give any syntax-semantics
correspondence.12
Syntax: Symmetric conjunction would require that D could select both N and A. But this
is not correct: D may only select N. If we have asymmetric conjunction, $, then only the
host category projects, and so the whole may properly be selected by D.
PF: There are two relevant observations concerning the phonological status of the
conjunction head. In the example cited, it is phonologically null, but in some languages it
may be overt, or there may be an overt marker (Rubin, 1996, Rebuschi this volume).
Indeed, there are cases in English where asymmetric conjunction is overtly marked, as we
will see in Section 5.3.
More importantly, the syntax we have proposed gives the wrong ordering for some noun
modification, as for example in (18b), where we now assume that the conjunction head is an
asymmetric head initial operator, shown as $.
(18) a [N [$ [P with red spots on]] N ball]]
b [N [$ [A proud of her successes]] N girl]
To handle this problem, we propose that the $ head PF-attracts a category Lex [�V]
(i.e. N or A). In (18a), this must be the N projection.13 In (17) or (18b), it might be the N
or A projection. We have argued in Cormack and Smith (2000) that word order
indeterminacies may be settled by Soft Constraints (understood as in OT), and suggest
12 The conjunction operator syntactically and semantically produces ‘predicate composition’ (Chomsky,
2002: 16), but this notion is not what is required for adjuncts with heads such as if.13 The alternative, of taking prenominal adjectives to be one-place operators of type hhe, ti, he, tii rather than
as predicates of type he, ti, is not viable in the light of the contrast between (i) �[the children happy] and (ii) [the
children at school and happy]. However, non-intersective adjectives like former and mere must be one-place
operators of category N/N.
404 A. Cormack, N. Smith / Lingua 115 (2005) 395–418
the same here. In these cases, there are conflicting desiderata. Moving the A projection
leaves the LF and PF orders in correspondence. Moving the N projection leaves a ‘heavy’
A projection last. We can postulate that English ranks the Soft Constraint HEAVY LAST
higher than PRESERVE CORRESPONDENCE, so that A moves in (17) and N in (18b) (see
Abeille and Godard, 2000 for French).
These considerations suggest that we need a lexical entry for asymmetric conjunction
whose syntax is as in (19) (where the information in parentheses is a reminder for the
reader, not part of the entry).
(19) N/N/X ^ head PF-attracts N or A (modification of N)
Such an entry licenses trees such as those shown in (20 a and b).14
The unspecified category X correctly allows for modification by any category of type he, ti,including PP, AP, passive and progressive VP, and relative clauses.
Our second example concerns adjunction to verbs in secondary predication structures. In
Cormack and Smith (1999), we argued that depictives and resultatives in English should be
given an account in terms of asymmetric conjunction. We take adjectives to be unac-
cusatives, assigning their semantically overt (internal) role to the object, but simulta-
neously selecting for an external argument with semantically vacuous role: that, is one
which contributes nothing to the associated meaning postulates. The analysis of a subject
depictive is illustrated in (21):
(21) a PF: Jo rode the horse weary
b LF: [Jo PAST Agr [VP [THE HORSE [V $ [AP WEARY t]] [V RODE ]]
adjunction to V0
14 On the standard definition of c-command, the noun in (20b) is apparently lowered to the position of the
conjunction head. It is however intuitively clear that a two-place operator should command both its operands,
and we take it that the relevant definition of command needs to be formulated accordingly.
A. Cormack, N. Smith / Lingua 115 (2005) 395–418 405
It is well-known that English does not have subject-orientated resultatives, but in languages
such as Korean, sentences comparable to (21a) do have such a reading (Kim and Maling,
1998). In Cormack & Smith (1999: 266f.), we argued that the difference lay in the head
initial versus head final setting for $ in these structures. Resultative readings of conjunction
structures depend on a pragmatic exploitation of iconicity, with causes preceding effects. In
(22a), but not (22b), the fall may be taken as the cause of the fracture.
(22) a Letitia fell over and broke her wrist
b Letitia broke her wrist and fell over
However, there are two possible representations over which this iconicity may be
exploited: PF and LF. Consider (21) again. There is a plausible causal connection between
riding a horse and becoming weary; the relevant iconicity is instantiated in the PF
(ride < weary), but not in the LF (weary < ride). We proposed then that pragmatic
interpretation depending on iconicity uses LF representations, not PF representations, so
that given a head-initial conjunction, (21) is correctly predicted not to have a subject-
orientated resultative reading.15 It may, of course, have a depictive reading and, for suitable
adjoined predicates, there may be a cause-effect interpretation, but it must be the adjunct
which gives the cause, as in (23), with IN A FIT OFtemper < break.
(23) a Mary broke the vase in a fit of temper
b [Mary PAST Agr [VP [THE VASE [V $ [AP IN A FIT OF TEMPER t]] [V BROKE ]]]]
The PF for these examples could be obtained by PF-attraction of V0 to Agr. However,
replacing the simple verb by a phrasal verb, shows that this is not sufficient.
(24) a Jo rubbed down the horse angry
b Jo rubbed the horse down angry
In (24a), we may suppose that the whole phrasal verb complex, which is arguably a V0
projection, moves to Agr. In (24b) however, the particle is stranded before the adjective. We
propose that there is PF-attraction of the V-projection to $, followed by the PF-attraction of
some V0 projection to Agr. The implication is that the lexical entry for conjunction required
for such examples is as in (25):
(25) V/V/X ^ head PF-attracts V (resultatives and depictives)
Crucially, our explanation for the absence of subject-orientated resultatives in English
relies on the $ being head-initial. In contrast, in a language such as Korean where subject-
orientated resultatives are grammatical, we postulate head-final $.
15 An alternative structure consisting of a single event gives rise to the object orientated resultative, without
appeal to iconicity (Cormack and Smith, 1999).
406 A. Cormack, N. Smith / Lingua 115 (2005) 395–418
Both the kinds of example discussed in this section have conjuncts mismatched for
selection by a higher head, which requires that the conjunct is asymmetric. In both, there is
displacement of some head or projection of a head to the syntactic position of the
conjunction operator: such displacement gives evidence that the operator is syntactically
present, even if phonologically null.16
5.3. Head-final asymmetric conjunction
So far, the $ position has been, as one would expect for English, head-initial. However,
that we need an ostensibly head-final $ can be shown not only for standard head-final
languages like Korean, but for English too.17
We can argue for head-final asymmetric conjunction on the basis of quasi-serials. At first
sight, examples like (26), discussed in Lakoff (1986), seem to allow extraction from either
conjunct:18
(26) a John ran to the shop and bought a paper
b Which shop did John run to t and buy a paper?
c What did John run to the shop and buy t?
However, in Cormack and Smith (1994, 1999), we argue that the correct analysis for such
sentences involves the conjunction at the canonic transitive level of two verbs or verb
projections, as in our analysis of a Serial Verb Construction. Further, the final result must be
a ‘QUASI-LEXICAL’ complex predicate in the sense that it has the syntactic and semantic
properties appropriate to a verbal lexical entry and must describe a ‘single event’. The
structure requires a ‘transitive’ form of run (available to unaccusative motion verbs), which
assigns a semantically null theta-role to an internal object argument (distinct from the goal
argument). This means that in (45), the argument a paper may be shared by run to the
shop and buy, which are conjoined as indicated in (27a). The whole complex V0 phrase is
PF-attracted to Agr to give the PF form. The occurrence of head-final conjunction is now
no obstacle to the extraction of a paper, as shown in (27b).
(27) a John T Agr [a paper [V [run to the shop] [[and buy] $] ]]
b Whatk did John T Agr [tk [V run to the shop] [[and buy] $] ]]
c Which shop did John T Agr [a paper [V [run [to t]] [[and buy] $] ]]?
In (27c), we have extraction from the first conjunct. If $ is head-final, then the first conjunct
will be the host, so that extraction is permitted, as discussed in Section 4.
16 Compare Steedman (1990: 215), where and is a marker, but the conjunction head is a combinator which is
not represented as a lexical item.17 Arguments for a host-initial adjunction version of asymmetric conjunction are made for German by
Thiersch (1996). For the asymmetries in German, see also discussion in Hohle (1989), Heycock and Kroch
(1994), Buring and Hartmann (1998), and Velde (2000).18 Lakoff was indeed arguing against the ‘Coordinate Structure Constraint’. The alternative we put forward
for the examples in this section can be seen as following the line suggested by Pauline Jacobson which is
recorded in the Appendix to Lakoff (1986).
A. Cormack, N. Smith / Lingua 115 (2005) 395–418 407
We conclude that there can be asymmetric conjunction with head-final $ in English. In
such examples, an overt marker and appears on the second conjunct. This suggests a lexical
entry for asymmetric conjunction of the form in (28):
(28) V/V/V{and} Head final; quasi-lexical
The feature on the adjunct selection in (28) is to force the one-place marker 1and to be
attached to this operand. The occurrence can be forced because 1and is a semantically and
syntactically trivial entity, which may be freely introduced into the array of lexical items to
be merged when required by syntax, rather than occurring essentially as part of the
meaning which is being constructed. When it occurs, it projects a feature {and} to the
mother category, as shown in (29).
Our second example is also from Lakoff (1986). Consider (30) and (31). For us, (30) is
acceptable or at worst only mildly deviant, while (31), with extraction from the second
conjunct, is unacceptable (Lakoff, 1986 gave both (30) and (31) as acceptable—see below).
If (31) is ungrammatical, then we suppose that we have head-final conjunction, so that not
get cancer is the adjunct of the adjunct�host pair. We may account for the extraction in
(30) as in (33a or b). In (33a), the parse has extraction from the first conjunct only. In (33b),
the conjuncts are mismatched for type, being of types he, he, tii and he, ti, respectively, as
are the operands of if in Stroik-style examples such as (15) in Section 4. Following the
discussion in Section 4, both these require that the first conjunct is the host in an adjunction
structure, and hence either of these bracketings entails that we have head-final asymmetric
conjunction, as shown.
(30) [What kind of herbs]k can you eat tk and not get cancer?
(31) �[What forms of cancer]k can you eat herbs and not get tk? (our�)
(32) �[What forms of cancer]k can you [eat herbs] and [[not get tk] $]
(33) a [What kind of herbs]k can you [V [V eat tk] [and [V not get cancer]] $]
b [What kind of herbs]k can you [V tk [V [V eat] [and [V not get cancer] ] $] ]
408 A. Cormack, N. Smith / Lingua 115 (2005) 395–418
We argue below that the structure for (30) too is quasi-lexical, so that the right form is that
in (33b).
Although the postulation of head-final conjunction for (30) and (31) gives the right
results for us, it is not right for Lakoff. Since Lakoff can extract from either conjunct of
(30), we need an explanation similar to that we gave for the dual extraction possibilities
from (26a). For this, we need get to assign a null thematic role to its patient argument, and
to take cancer as an oblique argument. Since cancer here is clearly not a canonic patient,
and get cancer can be paraphrased with ‘become ill with cancer’, where become is
unaccusative, this is not implausible. This will allow a shared argument what kind of herbs
to be extracted. If for Lakoff the conjunction is head-initial, then the second conjunct is the
host, and so alternatively, there can be extraction of what kind of cancer. The structures are
indicated in (34), where the null theta role assignment is correlated with selection for an
underdetermined type hnili.
(34) a [What kind of herbs]k can you [VP tk [V [$ [V eat]he, he, tii]
[and [V not get cancer]he, hnil, ti ]]
b [What forms of cancer]k can you [VP herbs [[$ [eat]he, he, tii]
and [[not get tk]he, he, hnil, tii]]
If this theta role assignment is available to all speakers for get, the conjunction of eat and
not get cancer may be at the canonic transitive level, allowing a quasi-lexical interpreta-
tion. In this instance, we see not get cancer as a state holding simultaneously with the eat
herbs process, so that we have something akin to a depictive, in (30), and to a result in
(34b), comparable to a resultative. Then for both Lakoff and ourselves, the required
selection is as in (28) above, with an obligatory quasi-lexical reading. The difference
resides only in the headedness of $ when the unaccusative is the second of the pair of
verbs.19
We have shown in this section that asymmetric head-initial conjunction is needed in the
analysis of noun modification, and of resultatives and depictives. We have also shown that
asymmetric head-final conjunction is available in some cases of sub-sentential conjunction
of verbal projections. On the assumption that symmetric (coordinating) conjunction also
exists, this finding leads immediately to the question of the distribution of the various kinds
of conjunction.
6. New question
Given both symmetric and asymmetric conjunction, how do we know when conjunction
must be symmetric, and when it must be asymmetric? It is clear that we must both answer
19 We believe that all Lakoff’s apparent violations of the coordinate structure constraint can be explained
away, and that one essential ingredient is an appeal to the option of null theta selection for the object argument
which may be associated with a subset of unaccusative verbs. We see his type A versus type B distinction as
having an essentially pragmatic explanation.
A. Cormack, N. Smith / Lingua 115 (2005) 395–418 409
this and provide some account of how symmetric conjunction emerges. We saw a contrast
between asymmetric if and necessarily symmetric and in examples (2) and (3), and the very
hypothesising of a coordinate structure constraint entails that not all examples of
coordination can be re-analysed as asymmetric structures.
Let us start with the question why coordination, i.e. symmetric conjunction, exists at
all, given its apparently anomalous status in the grammar. Unlike operators such as if,
conjunction and disjunction are inherently symmetric in their logical properties, and we
take it that this is the reason for their syntactic symmetry. Coordination is the unmarked
option.
What then is the motivation for introducing asymmetric conjunction? Our examples in
the last section were noun modification, and various adjuncts to verb projections, where the
product of the conjunction was a complex predicate of some sort. Let us suppose that it is
required on the grounds of expressive power and lexical parsimony that complex predicates
can be formed: that is, we do not want distinct lexical items to encode every complex of the
kind given by red ball or iron dry. Then in a language where adjectives, nouns, and verbs
belong to distinct syntactic categories, we NEED asymmetric conjunction. We do not
similarly expect asymmetric disjunction, because disjunctive concepts are of little if any
utility (cf. grue Goodman, 1955: Chapter 3). Accordingly, we need to set up lexical
selection restrictions licensing asymmetric conjunction, which we expect to be more
restricted that those for symmetric conjunction.
However, when we consider the distribution of symmetric and asymmetric conjunc-
tion, we have a problem. The distributions seem to overlap. We must have symmetric
conjunction of A and N projections in copular complements, as in (35a), but in noun
modification, we must have asymmetric conjunction of A and N projections, as in
(35b).
(35) a John is [& [A fond of his wife] [N and [N a happy man]]]
b the [$ [A happy] [N man]]
Similarly, we must have symmetric conjunction of V and A projections in (36a), but
asymmetric conjunction in resultatives and depictives as in (36b).
(36) a John is [& [A unhappy] [V and [V pursued by debtors]]]
b Mary [the vase [$ [P in a fit of temper t] [V broke]]] (¼(23))
Finally, we have symmetric conjunction of two V projections in examples like (37a), but
asymmetric conjunction in examples like those in (37b and c)
(37) a John [rare books] [& [V bought from the public] [V and [V sold to dealers]]]
(where neither PP can be extracted)
b John a paper [[V run] [V and [V fetch]] $]
c You can [those herbs [[V eat] [V and [V not get cancer]] $]
410 A. Cormack, N. Smith / Lingua 115 (2005) 395–418
For structures like (37), we claim that the condition restricting asymmetric conjunction to a
quasi-lexical interpretation of the whole is sufficient. In Cormack and Smith (1999) we
attempted to instantiate this in terms of event structure, and hence ultimately giving an
account in terms of a restriction on the types of the conjuncts. Although our account there is
flawed, we think this is the right approach, and will assume here that no further stipulation is
needed. For structures like those in (35) and (36), we suggest that the required distribution can
be obtained by exploiting checking theory to place restrictions on the lexical entries of the
host categories, where we restrict the host to being a projection of V or N.20
The solution we propose turns on special properties of V and N. In English, N must be
licensed by D, which is why characteristically determiners turn up in predicate NPs in
English.21 Comparably, V must be licensed by Inflection, which we have argued in
Cormack and Smith (1997) to be always a separate head in English, even when associated
with Tense. Consider the following LF structures:22
In the symmetric examples in (38), the required Determiner and Infl nodes are INSIDE the
scope of the conjunction; in the asymmetric examples in (39), the one relevant to the host
may be OUTSIDE.
We now tighten up our initial proposals for lexical entries for asymmetric conjunction,
given in (19), (25) and (29). The requirement for a checking feature is coded by requiring
the host to be of the form N{\D} or V{\Infl}, where the curly brackets indicate an
‘uninterpretable’ checking feature. We envisage that a feature {\X} is deleted under
sisterhood with X, perhaps after percolation.
(40) Revised lexical entries for the binary conjunction operator:
20 This means that there are no adjective or prepositional phrase modifiers which are adjoined by means of
covert conjunction. All such modifiers have to be headed by overt operators. As far as we can see, this is
harmless; the typical modifiers in both cases are measure phrases such as almost, as in almost hot, or almost out
of sight.21 We take it that in Bill is president, there is a null Determiner licensing president.22 It is of course also possible for there to be two instance of a Tense node, one in each conjunct.
A. Cormack, N. Smith / Lingua 115 (2005) 395–418 411
Note that because we have imposed the {\Infl} restriction on the host in (29), as in (40c),
any structure satisfying the syntactic description in (40c),23 which stipulates that the
conjunction operator must be head-final, will also satisfy (40b), where the conjunction
operator is head-initial. However, the various syntactic specifications are competing entries
in the lexicon, so that the Penguin Principle operates. Since the entry in (40c) is more
highly specified with respect to the items to be coordinated than that in (40b), it will
override the latter when applicable. In particular, quasi-serial constructions and examples
like Lakoff’s will always have $ final and an overt and.
It now seems that we predict incorrectly that structures with passives or progressives as
adjuncts, like that in (41a), instead of falling under (40b) as we assumed in Section 5.2, would
fall under (40c). This cannot be correct, since (40c) would require a form like that in (41b).
(41) a John ate his breakfast [standing up]/[surrounded by children]
b �John ate his breakfast and [standing up]/[surrounded by children]
Our reply is that passive and progressive phrases cannot occur in quasi-lexical projections
(see Cormack and Smith, 1997). If the fixed conditions of (40c) are not met, then this
specification cannot be used, and (40b) will be utilised.24
The question arises as to why it is the checking features {\Infl} and {\D} that are used to
restrictasymmetricconjunction.Weassumethat thefeaturesmustbecategorial, and thatusing
ordinary selection features would be incompatible with the ‘modification’ effects required of
the structures. If this is right, then a prerequisite for asymmetric conjunction is that the host
head is obligatorily associated with some higher head. Infl and D are local heads obligatorily
related to V and N, respectively, in English. Since we have not discussed all the variants of
asymmetric conjunction that we believe to occur in English, and do not have comparable
information from other languages, further generalisation and explanation must wait.
7. Coordination
Finally, we get to our proposal for symmetric conjunction. This has to meet the following
demands:
(42) i There are to be no principles of the grammar introduced especially for
coordination
ii In coordination, every conjunct behaves as if it were the head
iii We already have the three lexical entries given in (40) for the conjunction
operator
23 For Lakoff’s grammar, (40c) must presumably be replaced by two entries:
where the feature [location] is a cover term for the motion and location verbs possible with quasi-serials.24 It is possible that passive and progressive phrases do not even fall under the category V, but under Pass or
Mood, and Prog or Asp, respectively.
412 A. Cormack, N. Smith / Lingua 115 (2005) 395–418
Taking these three together leaves us with little room for manoeuvre. All we can do is to
construct further lexical entries. What is needed for an arbitrary coordination of X and Y, to
appear at LF in that linear order? In order for X to be the host, we need a head-final
conjunction operator, as specified in (43b). In order for Y to be the host, we need a
specification as in (43a):
(43) a Y/Y/X
b X/X/Y head final
For English, however, the marker and is mandatory in coordination, so we add this. The
entries shown amended in (45a and b) give rise to the trees in (44a and b).
We now argue that given these two lexical entries, the processing system will indeed parse
the appropriate strings as coordination structures. We show the whole collection of
specifications for the conjunction operator in (45).
(45) Lexical entries for the binary Conjunction operator:
Suppose we wish to conjoin P and Q, where none of the conditions for asymmetric
conjunction in (45c–e) is met. Then the less specified entries (a) and (b) are available.
Which is chosen? We claim not only that the speaker does not WANT the hearer to choose
between (a) and (b), but that the processing system is such that the hearer CANNOT choose
between them. There are two subparts to this claim, one about choosing, and the other
about the consequences of not choosing. We discuss each in turn.
A. Cormack, N. Smith / Lingua 115 (2005) 395–418 413
First, the grammar is normally set up in such a way as to minimise indeterminacy in
processing. Minimalism avoids optionality (Chomsky, 1991: 433). In most OT accounts,
this is enshrined in the requirement that soft constraints be totally rather than partially
ordered (Legendre, 2001: 3–5). In acquisition, Clark (1993) postulates a ‘Principle of
Contrast’ which states that ‘‘Speakers take every difference in form to mark a difference in
meaning’’. In the adult lexicon, there is a strong tendency for alternative items with the
same meaning to become differentiated (e.g. sheep, mutton). Where this appears not to
have happened, as with the free choice between mowed and mown, a decision is forced on
the speaker by the fact that although the items are identical semantically and equally
specified syntactically, they differ at PF (and almost certainly differ in related encyclo-
paedic content). In our case, although the two entries lead to distinct LF representations, the
two options are identical at the interfaces. That is, the PF output induced is the same for
both and, at the Conceptual�Intentional interface, we have first the contribution from the
symmetric logical operator conjunction, and secondly, any additional implicatures or
explicatures derived from the LF ordering of the conjuncts—which is the same for the
lexical options in (a) and (b). If the two entries lead to interface-identical output, there is no
basis for any decision, and no decision will be made.
This leads to the second question. How can a representation be indeterminate in this
way? There have been arguments for underspecified representations for both syntactic and
semantic ambiguity. For syntactic underspecification, see discussion in Gorrell (1995: §3.3
and §3.4), and for semantic underspecification, see Pinkal (1996), who says ‘‘Full
disambiguation may not even improve the usefulness of the utterance information, in
certain cases’’ (p. 186). It may be possible to produce an underspecified representation for
coordination from which the relevant syntactic and semantic inferences could be made.
Here, we assume that processing is done in parallel, with many options considered and
most rejected as the parse proceeds.25 If at the end there are still two viable representations,
so be it. That this is within human capacity is supported by the ubiquity of puns, where
because two distinct parses of some PF string are accessed almost simultaneously, both the
associated readings remain available, as in (46):
(46) a CHEMIST. WE DISPENSE WITH ACCURACY.
b The superconductivity revolution has come. Resistance is useless.
Our claim then is that there are two parallel parses as the proper output for a coordinate
structure.26 We are not claiming for our case, as others have done for other cases, that there
is real optionality in the grammar: we claim only that one cannot choose between equally
good syntactic possibilities.
25 We might assume some version of Ranked Parallel parsing: see Gorrell (1995) §3.6 and also §3.5 for an
introduction. It is clear that not all options can be pursued with equal commitment until shown non-viable, under
parallel processing, because of memory costs and the lack of explanation for Garden Path effects. That at least
limited parallel processing is necessary is argued for by Gorrell (1992), but see Gorrell (1995: 132) for an
alternative account.26 This makes our solution a bit like that of Goodall (1987): there are parallel structures in coordination. But
Goodall’s parallel structures contain material which is not identical at the Interfaces, whereas ours do not.
414 A. Cormack, N. Smith / Lingua 115 (2005) 395–418
Coordination, or symmetric conjunction, is essentially an ambivalent instance of
asymmetric conjunction. The symmetric properties of coordination arise from this
ambivalence. Either of the two conjuncts can be construed as the host, and the other
as the adjunct. Thus, as required, the two operands in certain conjunctions have equal
syntactic status for selection and extraction. Because there is no single representation, with
a single mother, for coordination, the problem of the category of the mother when
categorially mismatched conjuncts are coordinated disappears.
The symmetries of coordination are accounted for without the need to postulate any
symmetric structures in the grammar.
8. Conclusion
We have shown, we hope, several things. First, pretheoretically, grammars exhibit both
symmetric and asymmetric conjunction. Asymmetric conjunction is much like any other
adjunction structure, while symmetric conjunction is apparently anomalous. Secondly, we
have sketched a theory under which all the relevant distinctions, possibilities, and related
facts about selection, extraction, and so on, can be encoded in the lexicon. Coordination is a
by-product of an ambivalence in the lexicon.
Apart from the lexical entries themselves, what we have exploited in reaching this
solution is the background theory of CCG and Minimalism, pragmatics, and the Penguin
Principle interpretation of the lexicon. We needed crucially to claim that where the lexicon
provides no decision procedure for which choice of entries to make, no decision can be
made, and both offerings are to be used simultaneously.
Our take home message then is: There are no devices specific to coordination in the
grammar. Coordination is a syntactic pun.
Acknowledgements
We are grateful to Bob Borsley and to an anonymous referee for useful questions and
criticism, and to Deirdre Wilson for discussion of puns.
Appendix A. Adjunction with host and adjunct of unmatched types
We show in more detail here that the use of the combinator B derives the correct
meaning for an adjunction structure where adjunct and host have unmatched types
(Section 4). In the context in (47), the if clause needs to be part of the VP predicated of
Johnny, in order to give the natural sloppy reading where Tommy doesn’t cry if he,
Tommy, loses. The meaning of this VP is constructed on the lines of the simplified
representation in (48a). The subject is constructed as in (48b), where we assume that a
proper name is a noun, and that nouns must be associated with determiners—in this case,
a null determiner with the semantics of a type lifter. The whole is put together as in (48c),
which gives the correct meaning.
A. Cormack, N. Smith / Lingua 115 (2005) 395–418 415
References
Abeille, A., Godard, D., 2000. French word order and lexical weight. In: Borsley, R.D. (Ed.), Syntax
and Semantics: The Nature and Function of Syntactic Categories, vol. 32. Academic Press, San Diego,
pp. 325–360.
Anderson, S.R., 1992. A-Morphous Morphology. Cambridge University Press, Cambridge.
Asher, N., Morreau, M., 1991. Commonsense entailment: a modal theory of nonmonotonic reasoning.
In: Proceedings of the 12th International Conference on Computational Linguistics, Sydney, Australia,
pp. 387–392.
Bayer, S., 1996. The coordination of unlike categories. Language 72, 579–616.
Borsley, R.D., 1994. In defense of coordinate structures. Linguistic Analysis 24, 218–246.
Bowers, J., 1993. The syntax of predication. Linguistic Inquiry 24, 591–656.
Briscoe, T., Copestake, A., Lascarides, A., 1995. Blocking. In: Saint-Dizier, P., Viegas, E. (Eds.), Computational
Lexical Semantics. Cambridge University Press, Cambridge, pp. 273–302.
Buring, D., Hartmann, K., 1998. Asymmetrische koordination. Linguistische Berichte 174, 172–201.
Carston, R., 1988. Implicature, explicature, and truth-theoretic semantics. In: Kempson, R.M. (Ed.), Mental
Representations. Cambridge University Press, Cambridge, pp. 155–181.
Carston, R., 2002. Thoughts and Utterances: The Pragmatics of Explicit Communication. Blackwell, Oxford.
Chomsky, N., 1991. Some notes on economy of derivation and representation. In: Freidin, R. (Ed.), Principles
and Parameters of Comparative Grammar. MIT Press, Cambridge, MA, pp. 417–454.
Chomsky, N., 1995. The Minimalist Program. MIT Press, Cambridge, MA
Chomsky, N., 2000. Minimalist inquiries: the framework. In: Martin, R., Michaels, D., Uriagereka, J. (Eds.),
Step by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik. MIT Press, Cambridge, MA, pp. 89–155.
416 A. Cormack, N. Smith / Lingua 115 (2005) 395–418
Chomsky, N., 2002. Beyond Explanatory Adequacy. Ms, MIT.
Cinque, G., 1994. On the evidence for partial N-movement in the Romance DP. In: Cinque, G., Koster, J.,
Pollock, J.-Y., Rizzi, L., Zanuttini, R. (Eds.), Paths Towards Universal Grammar: Studies in Honor of
Richard S. Kayne. Georgetown University Press, Washington, DC, pp. 85�110.
Cinque, G., 1999. Adverbs and Functional Heads: A Cross-Linguistic Perspective. Oxford University Press,
Oxford.
Clark, E., 1993. The Lexicon in Acquisition. Cambridge University Press, Cambridge.
Cormack, A., 1999. Without specifiers. In: Adger, D., Pintzuk, S., Plunkett, B., Tsoulas, G. (Eds.), Specifiers:
Minimalist Approaches. Oxford University Press, Oxford, pp. 46–68.
Cormack, A., Smith, N., 1994. Serial verbs. UCL Working Papers in Linguistics 6, 63–88.
Cormack, A., Smith, N., 1997. Checking features and split signs. UCL Working Papers in Linguistics 9,
223–252.
Cormack, A., Smith, N., 1999. Why are depictives different from resultatives? UCL Working Papers in
Linguistics 11, 251–284.
Cormack, A., Smith, N., 2000. Head movement and negation in English. Transactions of the Philological Society
98, 49–85.
Cormack, A., Smith, N., 2001a. Don’t move! UCL Working Papers in Linguistics 13, 215–241.
Cormack, A., Smith, N., 2001b. What is coordination? Manuscript of talk given at the 4th NWCL International
Conference ‘Coordination, syntax, semantics and pragmatics’. Salford, November 2001.
Evans, R., Gazdar, G., 1989. Inference in DATR. In: Proceedings of the Fourth Conference of the European
Chapter of the Association for Computational Linguistics, pp. 66–71.
Gazdar, G., 1980. A cross-categorial semantics for conjunction. Linguistics and Philosophy 3, 407–410.
Gazdar, G., 1987. Linguistic applications of default inheritance mechanisms. In: Whitelock, P., Wood, M.M.,
Somers, H.L., Johnson, R., Bennett, P. (Eds.), Linguistic Theory and Computer Applications. Academic
Press, London, pp. 37–67.
Goodall, G., 1987. Parallel Structures in Syntax: Coordination, Causatives, and Restructuring. Cambridge
University Press, Cambridge.
Goodman, N., 1955. Fact, Fiction, and Forecast. Harvard University Press, Cambridge, MA.
Gorrell, P., 1992. Subcategorization and sentence processing. In: Berwick, R.C., Abney, S.P., Tenny, C. (Eds.),
Principle-Based Parsing: Computation and Psycholinguistics. Kluwer Academic Publishers, Dordrecht,
pp. 279–300.
Gorrell, P., 1995. Syntax and Parsing. Cambridge University Press, Cambridge.
Grice, P., 1967, 1989. Logic and conversation. In: Grice, P. (Ed.), Studies in the Way of Words. Harvard
University Press, Cambridge, MA, pp. 22–40.
Grimshaw, J., 1991. Extended Projection. Manuscript, Brandeis, Boston, MA.
Heim, I., Kratzer, A., 1998. Semantics in Generative Grammar. Blackwell, Oxford.
Heycock, C., Kroch, A., 1994. Verb movement and coordination in a dynamic theory of licensing. The
Linguistic Review 11, 257–283.
Hohle, T.N., 1989. Assumptions about asymmetric coordination in German. In: Mascaro, J., Nespor, M. (Eds.),
Grammar in Progress: GLOW Essays for Henk van Riemsdijk. Foris, Dordrecht, pp. 221–235.
Hohle, T.N., 1991. On reconstruction and coordination. In: Haider, H., Netter, K. (Eds.), Representation and
Derivation in the Theory of Grammar. Kluwer Academic Publishers, Dordrecht, pp. 139–197.
Johannessen, J.B., 1998. Coordination. Oxford University Press, Oxford.
Kayne, R.S., 1994. The Antisymmetry of Syntax. MIT Press, Cambridge, MA.
Keenan, E.L., Faltz, L.M., 1985. Boolean Semantics for Natural Language. Kluwer Academic Publishers,
Dordrecht.
Kim, S., Maling, J., 1998. Resultatives: English vs. Korean. In: Akatsuka, N., Hoji, H., Hiwasaki, S., Sohn, S.-
O., Strauss, S. (Eds.), Japanese Korean Linguistics 7. CSLI Publications, Stanford, CA, pp. 363–379.
Kiparsky, P., 1973. ‘‘Elsewhere’’ in phonology. In: Anderson, S.R., Kiparsky, P. (Eds.), A Festschrift for Morris
Halle. Holt, Reinhart and Winston, NY, pp. 93–106.
Lakoff, G., 1986. Frame semantic control and the coordinate structure constraint. CLS 22, 154–167.
Larson, R.K., 1988. Light Predicate Raising. Lexicon Project Working Papers 27. MIT Press, Cambridge, MA.
Legendre, G., 2001. Introduction to optimality theory in syntax. In: Grimshaw, J., Legendre, G., Vikner, S.
(Eds.), Optimality-Theoretic Syntax. MIT Press, Cambridge, MA, pp. 1–27.
A. Cormack, N. Smith / Lingua 115 (2005) 395–418 417
McCawley, J.D., 1981. Everything that Linguists have Always Wanted to Know about Logic. Blackwell, Oxford.
Munn, A.B., 1993. Topics in the Syntax and Semantics of Coordinate Structures. Ph.D. Dissertation, University
of Maryland.
Munn, A.B., 2000. Three types of coordination asymmetries. In: Schwabe, K., Zhang, N. (Eds.), Ellipsis in
Conjunction. Max Niemeyer, Tubingen, pp. 1–22.
Partee, B.H., ter Meulen, A., Wall, R.E., 1990. Mathematical Methods in Linguistics. Kluwer Academic
Publishers, Dordrecht.
Payne, J.R., 1985. Complex phrases and complex sentences. In: Shopen, T. (Ed.), Language Typology and
Syntactic Description, vol. II. Cambridge University Press, Cambridge, pp. 3–41.
Pesetsky, D., 1995. Zero Syntax. MIT Press, Cambridge, MA.
Pinkal, M., 1996. Vagueness, ambiguity, and underspecification. In: Galloway, T., Spence, J. (Eds.), Proceedings
from SALT VI. Cornell University, Ithaca, NY, pp. 185–201.
Progovac, L., 1998a. Structure for coordination part I. Glot International 3.7, 3–6.
Progovac, L., 1998b. Structure for coordination Part II. Glot International 3.8, 3–9.
Rubin, E.J., 1996. The transparent syntax and semantics of modifiers. WCCFL 15, 429–439.
Sag, I.A., Gazdar, G., Wasow, T., Weisler, S., 1985. Coordination and how to distinguish categories. Natural
Language and Linguistic Theory 3, 117–171.
Smith, N., 1999. $. Glot International 4.7, 7.
Sperber, D., Wilson, D.M.S., 1995. Relevance: Communication and Cognition, 2nd ed. Blackwell, Oxford.
Sportiche, D., 1994. Adjuncts and adjunction. GLOW Newsletter 32, 54–55.
Steedman, M., 1987. Combinatory grammars and parasitic gaps. Natural Language and Linguistic Theory 5,
403–439.
Steedman, M., 1989. Constituency and coordination in a combinatory grammar. In: Baltin, M.R., Kroch, A.S.
(Eds.), Alternative Conceptions of Phrase Structure. Chicago University Press, Chicago, IL, pp. 201–231.
Steedman, M., 1990. Gapping as constituent coordination. Linguistics and Philosophy 13, 207–263.
Steedman, M., 1993. Categorial grammar. Lingua 90, 221–258.
Stroik, T., 1990. Adverbs as V-sisters. Linguistic Inquiry 21, 654–661.
Stump, G.T., 2001. Inflectional Morphology: A Theory of Paradigm Structure. Cambridge University Press,
Cambridge.
Szabolcsi, A., 1992. Combinatory grammar and projection from the lexicon. In: Sag, I., Szabolcsi, A. (Eds.),
Lexical Matters. CSLI, Leland Stanford Junior University, Stanford, CA, pp. 241–268.
Thiersch, C., 1996. Asymmetrical coordination as adjunction. Manuscript, Tilburg. Available at http://
cwis.kub.nl/�fdl/general/people/thiersch/pubs.htm.
Velde, J. te., 2000. Assumptions about the structure of coordination. In: Schwabe, K., Zhang, N. (Eds.), Ellipsis
in Conjunction. Max Niemeyer Verlag, Tubingen, pp. 51–78.
Zoerner, E., 1999. One coordinator for all. Linguistic Analysis 29, 322–341.
418 A. Cormack, N. Smith / Lingua 115 (2005) 395–418