elided (as in VP-ellipsis), one of more *ts will remain in the structure, trig-
gering deviance. The structure in (28) illustrates these two possibilities.
(25)
* Ben wants to hire someone who speaks a Balkan language, but I
don’t remember which he wants to hire someone who speaks.
(26)
Ben wants to hire someone who speaks a Balkan language, but I don’t
remember which.
(27)
* Abby wants to hire someone who speaks a Balkan language, but I
don’t remember what kind of language Ben does. (=
someone who speaks>)
(28)
CP
¨
¨
¨
¨
¨
r
r
r
r
r
DP
2
which / what
kind of language
¨
¨
¨
r
r
r
C
TP
¨
¨
¨
¨
r
r
r
r
*t
2
TP
¨
¨
¨
¨
r
r
r
r
Ben
¨
¨
¨
r
r
r
(does)
v
P
¨
¨
r
r
*t
2
v
P
want to hire someone
[
island
who speaks t
2
]
This account can be extended to observed island effects in fragment an-
swers, as in (29) by positing an unelided *t in the final structure, as in (30).
(29)
a. A: Did each candidate
1
try to feed questions to the journalist who
will ask him
1
about abortion
(at the debate)?
b. B: *No, [about foreign policy].
c. cf. B: No, each candidate
1
tried to feed questions to
the journalist who will ask him
1
about foreign policy
.
(30)
FP
¨
¨
¨
¨
¨
r
r
r
r
r
PP
2
about foreign policy
¨
¨
¨
r
r
r
F
CP
¨
¨
¨
r
r
r
*t
2
¨
¨
r
r
C
each candidate tried to feed
questions to the journalist
[
island
who will ask him t
2
]
But the empirical picture appears to be less uniform than the data in Mor-
gan 1973 and Merchant 2004a would indicate. Both Culicover and Jackend-
off 2005:244ff. and Stainton 2006:138 produce examples which they judge
acceptable. Culicover and Jackendoff’s examples include (31a) and (32a),
whose putative unelided counterparts (under the movement+deletion analy-
sis) are given in (31b) and (32b) and are unacceptable.
(31)
Is Sviatoslav pro-communist or anti-communist these days?
a. —Pro.
b. *Pro, Sviatoslav is [t-communist these days.]
(32)
A: John met a woman who speaks French.
a. B: And Bengali?
b. *And Bengali, did John meet a woman who speaks French t?
For Culicover and Jackendoff, such ‘fragments’ (or ‘Bare Argument El-
lipsis’) have no clausal syntactic source: instead, such items are generated
directly by the syntax as ‘orphans’ whose properties (such as case, gender,
etc.) are determined by an algorithm of ‘indirect licensing’ and whose mean-
ing is given by the rule in (33b) subject to the algorithm in (33c), repeated
from above:
(33)
Bare Argument Ellipsis
a. Syntax: [
U
XP
i
ORPH
]
IL
b. Semantics: [ f (X
i
)]
c. If f is an expression in CS
a
and f cannot be determined from
SYNTAX
a
by application of Rules R
1
...R
n
, then “ f amounts to
the presupposition of the antecedent, constructed by substitut-
ing variables for the [necessary elements] in the CS of the an-
tecedent” (Culicover and Jackendoff 2005:276)
They show that ‘indirect licensing’ is useful for accounting for why an En-
glish speaker, pointing at a pair of scissors, will say, ‘Please hand me those’
with a semantically otiose plural demonstrative whose morphological plural
form is determined by the fact that the English word ‘scissors’ is pluralia tan-
tum. But they give no account of the form connectivity facts in (20) above
(let alone connect such an account to the ill-formedness of (21)) or in (23).
While I’m happy to admit that language-specific linguistic aspects of the ob-
jects in a context can influence choice of demonstrative (the alternative being
that the demonstrative itself contains an instance of NP-ellipsis of scissors,
as on Elbourne’s 2005 account), doing so does not make sense of the voice or
P-stranding facts.
5
In any event, a closer look at some of Culicover and Jackendoff’s data is
in order. First, the fact that (31a) is acceptable seems irrelevant, given that
(34) is also acceptable: under some circumstances, even bound prefixes can
be used without their usual hosts. Whatever accounts for this fact will allow
a maximal projection to be moved in (31a). Second, consider carefully the
range of interpretations available to (32a); these are sketched in (35).
(34)
Sviatslav is anti-communist and Derzhinsky is pro-.
(35)
a. = Did John meet a woman who speaks French and Bengali?
b. = Does she speak French and Bengali?
c. = And does she speak Bengali (too)?
d. = And did John also meet a different woman who speaks Bengali
(in addition to meeting the woman who speaks French)?
The crucial thing to notice is that the reading given in (35d) is absent from
B’s ‘fragment’ utterance in (32a). On my account, this is because to generate
5
And note that it’s unclear that Stainton would be too happy about their conclusion either,
as it seems to require, even for discourse-initial uses, that speakers can make use of peculiari-
ties of linguistic coding for ‘deep’ anaphora as well—Culicover and Jackendoff give examples
with nonsemantic gender features in several languages as well. See section 5 below for more
discussion.