4
Ellipsis and Syntactic Representation
(7)
a.
The Cubs start a more talented infield than [wh
£
the Sox start an
¥¦£
outfield]
b.
Jones produced as successful a film as [wh
£
Smith produced a
¥¦£
play]
(8)
a.
How talented
£
do the Sox start an
¥¦£
outfield?
b.
How successful
£
did Smith produce a
¥¦£
play?
The problem for a syntactic theory of ellipsis comes from examples like (9a) and
(9b), which appear to indicate that island effects disappear under ellipsis, contrary
to what we saw above in (1b) and (2b). If ellipsis involves deletion, then (9a) and
(9b) should be derived from (10a) and (10b), respectively, which should be just as
ill-formed as (7) and (8) above.
(9)
a.
The Cubs start a more talented infield than the Sox (do).
b.
Jones produced as successful a film as Smith (did).
(10)
a.
The Cubs start a more talented infield than [wh
£
the Sox (do)
[
VP
start [
DP
a
¥¦£
infield]]]
b.
Jones produced as successful a film as [wh
£
Smith (did)
[
VP
produced [
DP
a
¥¦£
film]]]
A second problem for syntactic analyses of ellipsis is that there are also
contexts in which binding effects seem to disappear. (11a), for example, can clearly
have a strict reading, despite the fact that it should be derived from (11b), which
does not support such a reading.
(11)
a.
Doug blamed himself for the band’s collapse because everyone else
did.
b.
Doug blamed himself for the band’s collapse because everyone else
did [
VP
blame himself]
Similarly, (12a) fails to show the disjoint reference effect that we would expect
to see if it were derived from the representation in (12b), which does not permit
coreference between the pronominal arguments in the second clause.
(12)
a.
Most Americans expected him
£
to be acquitted, and obviously he
£
did
too.
b.
Most Americans expected him
£
to be acquitted, and obviously he
£
did
[
VP
expect him
£
to be acquitted]
Christopher Kennedy
5
Finally, although it is the case that syntactic non-identity typically results in
judgments of unacceptability, it is a fact about English that many such examples
are attested. In particular, both of the examples presented above in (5) are naturally
occurring sentences:
(13)
a.
In yesterday’s elections, only 43 percent of registered voters did.
(heard on National Public Radio by CK in November 1996)
b.
A lot of this material can be presented in a fairly informal and ac-
cessible fashion, and often I do. (Chomsky 1982, p. 41; cited in
Dalrymple, Shieber, and Pereira 1991)
1.2 Semantic Analyses of Ellipsis
A second approach to ellipsis claims that elided constituents have no syntactic rep-
resentation at all, but rather can be fully explained in terms of a more general theory
of information retrieval (see e.g. Dalrymple et al. 1991; Hardt 1992, 1999; Jacob-
son 1992; Hendriks and de Hoop 2001). In the higher-order unification approach
advocated by Dalrymple et al. (1991), for example, a structure like (14a) is assigned
a semantic representation of the sort in (14b), where
§
is a free variable over prop-
erties that needs to be resolved. The problem of ellipsis is the problem of solving
the value of
§
, which is done by abstracting over parallel elements in some pre-
vious clause to generate a property-denoting expression, as shown in (14c), and
substituting this expression for
§
.
(14)
a.
Sterling quit the band because Lou did.
b.
¨©¥¥! #"$&%('()0¥21354768%@9BA
BECAUSE
§CDFEG©@A
c.
§
H
IQPSRT¨U©¥$PS)0¥213V4W68%X9YA
d.
¨©¥¥! #"$&%('()0¥21354768%@9BA
BECAUSE
IQPSRT¨U©¥$PS)0¥213V4W68%X9YADFEU©`A
One obvious positive aspect of this type of approach is that it doesn’t run into
the problems associated with purely syntactic accounts: because ellipsis does not
involve syntactic representation, we should not expect to find syntactic effects inside
the ellipsis site. However, this advantage is also its disadvantage: as noted above,
there are a number of contexts in which we do find clear evidence of syntactic
effects within the ellipsis site.
In order to account for facts like those discussed in section 1.1, we would
need to significantly weaken assumptions about the nature of the syntax-semantics
interface. For example, we could adopt Ha¨ik’s (1987) position that the ellipsis site
itself can serve as the “gap” for a syntactic operator, as illustrated in the following
examples.
6
Ellipsis and Syntactic Representation
(15)
a.
Sterling criticized every decision [
CP
wh
£
that Doug was upset [
AdvP
because Lou did
¥¦£
]]
b.
Dogs, I understand. Cats
£
I don’t know [
DP
a single person who does
¥¦£
]
Since this approach requires us to abandon the assumption that semantic type of
a bound variable corresponds to syntactic category of a corresponding gap (the
bound variable in the interpretation of the relative clause in (15a) has the type of
an individual, but the syntactic category of the gap in the syntactic representation is
VP), it should be adopted only if a less costly approach fails to materialize.
1.3 A “Mixed” Analysis
In recent work, Andy Kehler (1995; 2000) attempts to account for the apparently
paradoxical sensitivity of ellipsis to syntactic constraints by developing a “mixed”
syntactic/semantic analysis, in which whether an elided constituent has internal syn-
tactic structure or not depends on the discourse context in which it appears. In
particular, Kehler argues that the requirement for syntactic representation in ellipsis
depends on the type of “coherence relation” an elided VP participates in (see Kehler
1995; Hobbs 1979). Coherence relations determine coherence between sentences
in a discourse, which in turn affects acceptability. The two types of coherence rela-
tions that are relevant to ellipsis are
CAUSE
-
EFFECT
relations and
RESEMBLANCE
(parallelism and contrast) relations, the action of which is illustrated by the follow-
ing coherent (acceptable) and incoherent (unacceptable) discourses.
(16)
CAUSE
-
EFFECT
relations (coherent)
a.
Smith is a politician. He’s bound to be dishonest.
b.
Smith is dishonest because he’s a politician.
c.
Jones is a politician, but she’s honest.
(17)
CAUSE
-
EFFECT
relations (incoherent)
a.
?? Smith is a politician. He’s bound to wear long underwear.
b.
?? Smith is dishonest because he wears long underwear.
c.
?? Jones is a politician, but she has a nice daughter.
(18)
RESEMBLANCE
relations (coherent)
a.
Smith likes to play golf. Jones enjoys surfing the net.