Ellipsis and Syntactic Representation
Christopher Kennedy
Northwestern University
This paper addresses a question that has been of interest to researchers
on ellipsis since the very early days of work in generative grammar: do
constituents targeted by various types of ellipsis operations have syntac-
tic structure at some level (or levels) of representation, or can the various
properties of ellipsis constructions be accounted for purely in terms of
recovery of meanings, without positing syntactic representation at the
ellipsis site? Focusing on the interaction of ellipsis and several differ-
ent grammatical phenomena and constraints, including parasitic gaps,
binding theory, and extraction islands, I will present evidence that ellip-
sis constructions are sensitive to configurational constraints on syntactic
representations, but not to constraints that are based on morphophono-
logical properties of lexical items, thus supporting a view of ellipsis as
deletion of syntactic material.
1 The Representation of Nothing
Since at least Hankamer and Sag 1976, a central question in research on ellipsis
has been what sorts of representations are involved in the resolution and licensing
of unpronounced linguistic information? Two lines of thought have predominated,
which differ in their assumptions about the role of syntax in ellipsis. The first
approach, which has a long tradition in generative grammar, postulates that elided
material has syntactic structure at some level of representation, but the grammar
contains a means of blocking its pronunciation in the surface form. The second
approach rejects the claim that unpronounced material has syntactic representation,
hypothesizing instead that general mechanisms governing the recovery of meanings
from context can be put to work to resolve ellipsis. The purpose of this paper is to
provide arguments in favor of a version of the first approach, and to show that an
analysis in which ellipsis involves only the recovery of meanings, without reference
to syntax, fails to provide an empirically adequate account of the facts. Before
making an argument in favor of this position, however, I will present an overview
¡
I am grateful to Susanne Winkler and Caterina Donati for their thoughtful comments on an
earlier draft of this paper, and to the participants in the Workshop on Adding and Omitting at the
21st Meeting of the DGfS for very helpful discussion of the material presented here.
1
2
Ellipsis and Syntactic Representation
of the positive and negative aspects of several approaches that are representative of
these different answers to the question of representation.
1.1 Syntactic Analyses of Ellipsis
The hypothesis that ellipsis involves syntactic representation can be implemented
in two ways: deletion of syntactic material from the representation that is the input
to the phonological component or recovery of syntactic structure at some level of
logical representation. The former approach goes back to the very early days of
generative grammar, and has been revitalized in recent work in the Principles and
Parameters framework (see e.g. Hankamer 1979; Sag 1976; Tancredi 1992; Wilder
1995; Merchant 2001; Kennedy and Merchant 2000); the latter approach has also
appeared in different forms over the past twenty-five years (see e.g. Wasow 1972;
Williams 1977; Ha¨ik 1987; Kitagawa 1991; Fiengo and May 1994). The crucial
assumption that both sorts of syntactic analyses share is that elided material has
syntactic structure at some level of representation. A central result of such ap-
proaches, therefore, is that they can account for syntactic effects within the ellipsis
site.
For example, the fact that island effects appear under ellipsis, as shown by
the contrast between the (a) and (b) examples in (1) and (2), receives a straightfor-
ward explanation: if ellipsis involves deletion, then the (b) sentences are derived
from the representations in (1c) and (2c) (where struck-through text indicates ma-
terial deleted from the pronounced form), which involve extraction out of an island
(indicated by a subscript
¢
).
(1)
a.
Sterling criticized every decision that Lou did.
b.
Sterling criticized every decision that Doug was upset because Lou
did.
c.
Sterling criticized every decision [ wh
£
that Doug was upset [
¤
because
Lou did [
VP
criticized
¥¦£
]]]
(2)
a.
Dogs, I understand, but cats, I don’t.
b.
Dogs, I understand, but cats, I don’t know a single person who does.
c.
Cats
£
I don’t know [
¤
a single person who does
[
VP
understand
¥¦£
]]
Note that in the absence of an A-dependency into the ellipsis site, the relation be-
tween an elided VP and its antecedent is not sensitive to island constraints, as orig-
inally observed by Ross (1967).
Christopher Kennedy
3
The appearance of Binding Theory effects in the ellipsis site is another prop-
erty that is expected in a syntactic analysis. For example, the fact that (3a) strongly
disfavors a ‘strict’ interpretation, in which Sterling also blames Doug for the band’s
collapse, follows from the fact that it is derived from (3b): the strict reading would
violate Condition A, which requires a reflexive pronoun to find its antecedent lo-
cally. Likewise, the disjoint reference effect in (4a) is a direct consequence of Con-
dition B of the Binding Theory, which rules out coreference between a pronoun and
a co-argument.
(3)
a.
Doug blamed himself for the band’s collapse, and Sterling did too.
b.
Doug blamed himself for the band’s collapse, and Sterling did
[
VP
blame himself] too.
(4)
a.
Kim takes care of him
£
because he
£
won’t.
b.
Kim takes care of him
£
because he
£
won’t [
VP
take care of him
£
]
Finally, a syntactic approach to ellipsis, augmented with sufficiently strict
requirements on the type of identity relation that licenses deletion, provides the
basis of an account of the unacceptability of examples involving syntactic non-
identity.
(5)
a.
?? Only 43 percent of registered voters did.
b.
?? A lot of this material can be presented in a fairly informal and acces-
sible fashion, and often I do.
There appear to be a number of serious problems for a syntactic account of
ellipsis, however, the most important of which is the fact that there are contexts in
which syntactic effects within the ellipsis site seem to disappear. One such context
involves comparatives constructed out of attributive adjective phrases, such as those
in (6).
(6)
a.
The Cubs start a more talented infield than the Sox start an outfield.
b.
Jones produced as successful a film as Smith produced a play.
Kennedy and Merchant (2000) demonstrate that the unacceptability of the exam-
ples in (6) is due to the Left Branch Constraint (LBC), which blocks movement of
left branch attributive modifiers (see also Pinkham 1982). Assuming that compar-
atives are derived through an operation of A-movement that targets the compared
constituent in the than-clause (Ross 1967; Chomsky 1977), the syntactic structures
assigned to the examples in (6) are those in (7), which, just like the questions in (8),
violate the LBC.
Dostları ilə paylaş: |