headed by the existential quantiﬁer (in order to introduce another woman).
The mutually equivalent readings in (35a-c) are straightforwardly derived
not by extraction out of an island, but by extraction out of a simple clause
whose subject (e.g. she as in (35c) is assigned the same value as the woman
who speaks French (see Merchant 2001:ch. 5 for extensive discussion of such
E-type readings under sluicing). Culicover and Jackendoff have no obvious
explanation for this pattern of judgments, since their account builds on the
intuition that the value of f in indirect licensing contexts follows the same
route that gives us interpretations for idioms like the following:
a. And what about Bengali?
In (36), the interpretation of what/how about X? does seem to allow a wider
range of possibilities, including, crucially but damningly, a reading like that
in (35d). So their proposal, while perhaps appropriate for examples like (36),
overgenerates if applied to (32a).
Stainton 2006:138 provides more challenging examples, namely the fol-
lowing (though he doesn’t try to address the difference between these exam-
ples and the island-sensitive ones like (29) and (32a), I will below):
Q: The Pope likes beer and what?
A: HEIMer. (Cf. # HEIM.)
Q: Do you pronounce it Can[kun] or Can[k2n]? A: kun.
The ﬁrst involves focus on a subpart of a word (see Artstein 2004 for a semantics for these),
and the second of an aspect of the linguistic form itself. These examples might be the best
candidates for a ‘replacive’ or ‘completive’ construction with the properties Stainton seeks.
Its use in these contexts would be licensed by the fact that the fragment answer could not have
been moved in the regular construction. This seems like the only way to block this construction
from overapplying and voiding all sorts of connectivity effects.
A: They want to hire someone who speaks a Balkan language.
B: Really? Which one?
In this latter example, what is voiding the island-violating movement of
the fragment is apparently the fact that the question itself involves island-
violating wh-movement. One way to implement this would be to encode this
difference in the ellipsis in the fragment answer in some way, allowing for
deletion of the higher, CP, node just in case the antecedent involved sluic-
ing (making it sensitive to the fact that an E feature was present in the an-
tecedent, in other words). A less mechanical solution would be to follow Fox
and Lasnik’s 2003 analysis of variable island effects: they claim that island-
violating sluicing involves long wh-movement without intermediate traces,
and that this is licensed just in case the correlate is an indeﬁnite bound by
a choice-function existential closure operator (meaning there are no inter-
mediate traces of QR in the antecedent), satisfying a structural parallelism
requirement. For (39), this would mean that Albanian could move in one-fell-
swoop, violating islands, but only if the antecedent to the ellipsis still satisﬁes
parallelism—and this will only be possible if the antecedent contains island-
violating wh-movement itself, as in sluicing. (But see Agüero-Bautista 2007
for complications in this picture.)
If the example in (39) is indicative of what is going on in the Pope exam-
ples, then we have to posit that such ‘quizmaster’ questions (like the ‘reprise’
questions of Ginzburg and Sag 2000, sometimes called ‘echo’ questions) can
involve covert long movement. Such a conclusion makes most sense on a
theory of islandhood which takes it to be primarily a consequence of PF con-
siderations. (Other locality effects of covert movement, such as those found
for scoping in QR and for multiple questions as in Dayal 2002, must have
another source: they are not merely sensitive to islands, but to stricter locality
conditions, in fact.)
The truth is that none of this matters too much for Stainton. He’s interested
in antecedentless cases, not short answers. The main point of this discussion
is to show how a particular theory of ellipsis
can be implemented. It
is the possible extension of this implementation to Stainton’s cases that is of
most interest next.
The limits of the ‘limited ellipsis
X and labels
A syntactic ellipsis analysis could be given of some antecedentless subsen-
tences if one is allowed to elide syntactic structure in certain circumstances
with no linguistic antecedent. Merchant 2004a proposes such a ‘limited el-
lipsis’ strategy for some examples, claiming that an expletive, deictic, or
demonstrative subject (there/it, he/she/it, this/these/that/those) and an appro-
priate form of the verb be (appropriate in person, number, aspect, and tense)
can be elided if a referent for the deictic or demonstrative is salient enough
to resolve it (in other words, in the same circumstances that such elements
can be used without linguistic antecedents, period). Representative proposed
structures are the following, before movement of the remnant and with the
unpronounced material in angled brackets.
Properties applied to a manifest object
b. Moving pretty fast!
c. From Spain!
d. Driven exactly 10,000km.
e. Recommended for ages 6 and older.
f. one of Anthony Carroll’s best men.
Individuals as arguments of a manifest property
b. Rob’s mom.
c. Nova Scotia.
d. The ngo-gyin, the song of mourning.
Compare these to uses of the copula with a demonstrative subject like the
That’s Max (all right, all over again, for sure, for you).
present context has raised Max to a high degree of salience. (Such an exam-
ple could be used on seeing walls painted with Max’s typical style, a messy