which, when added to a phrase’s feature matrix, triggers the special pronun-
ciation rule in (7c). (For present purposes, we could equally well suppose that
the mapping algorithm itself were sensitive to some aspect of the structure, or
that there were a ‘construction’ where this special phonology is stated.) On
this approach, the speaker who utters (4a) has produced a sentence
syntactic
, a
sentence
semantic
, and a sentence
pragmatic
.
(5)
a.
TP
¨
¨
¨
¨
r
r
r
r
Bill
T
¨
¨
¨
¨
r
r
r
r
should
VP
¨
¨
¨
r
r
r
collect
NP
butterflies
b.
TP
¨
¨
¨
¨
r
r
r
r
Jill
T
¨
¨
¨
¨
r
r
r
r
should
T
E
>
¨
¨
¨
r
r
r
collect
NP
butterflies
(6)
TP:
should
(collect( jill, butterflies))
¨
¨
¨
¨
¨
¨
r
r
r
r
r
r
Jill:
jill
T :
λ z.should(collect(z, butterflies))
¨
¨
¨
¨
¨
¨
r
r
r
r
r
r
should
T
:
λ Pλ z.should(P(z))
E
>:
λ x.collect(x, butterflies)
¨
¨
¨
¨
r
r
r
r
collect:
λ yλ x.collect(x, y)
NP:
butterflies
butterflies
(7)
Rules
a. Syntactic combinatoric rules: should [ _ VP ] (equivalently, T →
should
VP), etc.
b. Semantic combinatoric rules: (β -reduction / λ -conversion) If f
is a expression of type τ containing one or more instances of a
free variable h of type σ and g is an expression of type σ and h
is free for g in f , then λ h
σ
[ f
τ
](g
σ
)
f
τ
g
/h
.
c. Phonological interpretation rules:
should
p
/S2d/, X
E
p
/0, etc.
A second possibility, which Stainton calls ellipsis
semantic
, posits no unpro-
nounced syntactic structure at all. This view is compatible with complicating
the mapping S ⇔
sem
M
in the appropriate way. One specific proposal along
these lines is given in Culicover and Jackendoff 2005: they posit syntactic
representations such as (8a) for examples like (4a), as part of their program
for ‘Simpler Syntax’. This is simpler in the sense that there are no syntactic
nodes that lack pronunciation. It is more complex, however, in that the subcat-
egorization requirements of auxiliaries like should must be modified by some
rule, presumably operating on the lexical entry of should to produce a new
lexical item should
/V P
, indicated in (8a). (Equivalently, the phrase-structure
rules for expanding S or VP, which normally require that a clause contain a
VP, could be suspended or altered. Their hypothesis is compatible with either
route.) The semantic representations for the nonelliptical (4a) and elliptical
(4b) would be equivalent, given in a standard notation in (8b).
(8)
a.
S
¨
¨
¨
r
r
r
Jill
should
/V P
b. should(collect( jill, butterflies))
Culicover and Jackendoff 2005 use a slightly different semantic represen-
tation, called conceptual structure (CS) (see their work for details). Culicover
2008 uses a representation of CS which is similar to predicate logic formulae
supplemented by thematic role annotations on the arguments of certain pred-
icates. The usual mapping between a nonelliptical syntactic structure and its
corresponding CS is given in the lower half of Figure 1. Each arrow repre-
sents a mapping rule, and it is clear that there is no necessary connection
between the hierarchical structure in the semantics and that in the syntax;
for this clause, four mapping rules are needed. The resulting rule system is
given in (9); they give a rule for Bare Argument Ellipsis (BAE), which I re-
turn to in much more detail below, not for VP-ellipsis, but the mechanism
(so-called ‘Indirect Licensing’ plus pragmatic establishment of the value for
f
) is presumably the same in both cases. (Their system is merely the most
recent and well-worked-out of a range of similar proposals; cf. Hardt 1993,
Dalrymple et al. 1991, Ginzburg and Sag 2000, and Schlangen 2003.) On this
approach, a sentence
semantic
is produced without a correspondingly complete
sentence
syntactic
.
Natural Language Syntax, © 2005, 2006 by Peter W. Culicover
13-14
characteristics of a full sentence – it has a subject, tense and an auxiliary, but no VP.
Hence it is unlike BAE.
As indicated earlier in this chapter, there are two basic ways to analyze VP
ellipsis syntactically. Either the VP is present, but invisible, or it is simply not present.
These two alternatives are illustrated in (33), for Robin can speak German. For
concreteness we show the CS representation of the modal as an operator that takes as its
argument the entire proposition.
(33)
a.
Empty VP
b.
No VP
Note that we are assuming here that the syntactic rules of English permit an S that
contains a subject and I
0
, but no VP.
Figure 1.
A ‘missing’ VP and its antecedent for Culicover and Jackendoff
(9)
Rules
a. Syntactic combinatoric rules: S → NP I
0
(VP), etc.
should
[ _ (VP) ], etc.
b. Semantic combinatoric rules:
i. Argument/Modifier Rule
CS: [F ... X
i
... ] ⇔
de f ault
Syntax: {..., YP
i
, ...}
ii. R1 : If X is the meaning of the NP-daughter-of-S whose pred-
icate meaning is PRED, then let PRED(AGENT:_, ...) = PRED(AGENT:X ,
...)
iii. Bare Argument Ellipsis (C&J 2005:265)
Syntax: [
U
XP
i
ORPH
]
IL
Semantics: [ f (X
i
)]