(62)
a. λ P
et
[P(partee)]
b. λ P
et
[P(partee)](Q
et
)
c. Q(partee)
(62c) is an expression of type
, whose truth will be evaluated relative
to what the assignment function gives for the value of the variable Q.
One might object that this is asking too much of the assignment function.
But such complex semantic objects determined by the assignment function
are not needed merely in the cases at hand. They are also required to account
for the meaning of anaphoric elements like that and it and their null counter-
part in the following kinds of examples:
(63)
a. Every patient
1
was told that he
1
was sick. But then most of them
forgot it.
b. Most reservists
2
found out by mail that they
2
were being sent to
Iraq and that pissed them
2
off.
c. Everyone
3
remembered to bring their
3
swimtrunks. No-one for-
got.
d. Everyone
4
remembered that they
4
wanted to marry their
4
cousin.
No-one forgot.
These have readings that are equivalent to the following.
(64)
a. Every patient
1
was told that he
1
was sick. But then [most of
them]
5
forgot that they
5
were sick.
b. Most reservists
2
found out by mail that they
2
were being sent to
Iraq and the fact that they
2
found out by mail pissed them
2
off.
c. Everyone
3
remembered to bring their
3
swimtrunks. No-one
6
for-
got to bring their
6
swimtrunks.
d. Everyone
4
remembered that they
4
wanted to marry their
4
cousin.
No-one
7
forgot that they
7
wanted to marry their
7
cousin.
For these and similar cases, we seem to need the assignment function to
be able to assign pronouns like it values like [x
5
_was_sick], allowing the vari-
able x
5
to be bound by a higher quantifier to capture the attested covariance
with the quantificational elements. So it seems plausible that such objects in
the semantic representation are available to the assignment function, and can
therefore serve as possible values for higher-type variables in ‘slot-filling’.
Stainton anticipates something like this account in a paragraph on p. 185
(and also on p. 55), where he discusses the idea that Alice holding up a pen
and saying ‘Red’ to Bruce can be translated as Red(x
3
)
, where, assuming
an assignment function g where g(x
3
) =the pen Alice held up, Red(x
3
)
does
express what Alice asserted. To this idea he writes that ‘it is absurd to suggest
that the thought Alice got across is grasped via [Red(x
3
)
], since Bruce, qua
ordinary English speaker, could not have used the latter to understand the
proposition—this being a made-up language.’ But this is precisely what’s at
issue: on the claim pursued here, the English word red can have the semantic
value red(x), where the variable x can be bound or not. If free, the assignment
function (whose values are determined by pragmatics) must yield a value.
On this approach, then, there really are more ‘slots’ to be filled: these
slots, by design, cover exactly the same ground that Stainton’s three subcases
cover. This is no accident: this account is quite close to previous versions,
which differ however from this in introducing the variable as part of the el-
lipsis resolution algorithm (Dalrymple et al. 1991, Culicover and Jackendoff
2005). Here the variables are already there, as parts of the meaning of the
items used. What their values are is determined by context, just as the ac-
tual content of the assignment function or accessibility relation is. So this has
precisely the same effect as Stainton’s account in this way, since it is the prag-
matics that does this. But it ‘semanticizes’ the variables in a familiar way. The
difference between this account and Stainton’s is pretty tiny indeed: the only
real difference is that by having the semantic ‘slots’ in the meaning (semantic
value) of the phrase uttered, they can all be type , propositional. The prag-
matics does its work in the same way it does in determining the denotations
in a context of other kinds of variables, nothing more.
This proposal comes very close in spirit to that of Culicover and Jackend-
off 2005; it differs in its implementation. For them, the variable over contex-
tually specified meanings (which they posit is part of the semantic represen-
tation ‘Conceptual Structure’ of the utterance, as here and pace Stainton) is
introduced by a special rule that is the grammatically specified interpretation
rule for the Bare Argument Ellipsis construction. In the present view, on the
other hand, no special or construction-specific rules are employed: only the
regular semantic mechanisms independently needed.
12
12
And unlike Culicover and Jackendoff, I emphatically do not intend for the present account
to also apply to syntactic ellipsis structures. Doing so I believe overgenerates. If there is no
syntax internal to an ellipsis site (and its meaning is the product of a special interpretation
rule), there is no explanation for the ill-formedness of pseudogapping in (iii) on a reading