The Semantics of Ellipsis
77
Before we go on to look at how this conception of ellipsis facilitates the
analysis of ellipsis-containing antecedents and binderless sloppy readings, there
are a couple of potential problems to address. First, we should pause to consider
the case of simple bound variables in VP-ellipsis. Consider (42).
(42)
Every woman loves her mother. Even Mary does.
According to almost all current theories, her in the first sentence of this example
is to be analyzed as being or containing a bound individual variable.
6
I too will
assume this. The current theory, then, must maintain that we start off with the
LF structure in (43).
(43)
Every woman λ
2
T t
2
v love her
2
mother. Even
THE
Mary λ
2
does t
2
v
VP.
The process of LF ellipsis resolution produces the following:
(44)
Every woman λ
2
T t
2
v love her
2
mother. Even
THE
Mary λ
2
does t
2
v
love her
2
mother.
This produces the right reading, of course. And the use of the same index on the
expressions bound by the subjects of the first and second sentences violates no
prohibition that I know of. Difficulties with repeated use of the same index only
arise if the same index is used on referential expressions with different intended
referents, or on bound variables intended to be bound by different operators that
lie within the scope of both (unlike in this case), or on both bound variables
and an independent referential expression. Heim and Kratzer (1998:254) have
proposed a principle explicitly to deal with the latter case.
ellipsis with no linguistic antecedent in section 2.5. Meanwhile, if the idea of replacing a node
with something else causes unease, one could also think of ellipsis resolution as copying the
daughters of a node of the same category in the linguistic environment and pasting them sepa-
rately into position beneath the ellipsis node. But I personally find the version in the text less
awkward.
6
An exception is the variable-free semantics proposed by Szabolcsi (1989) and explored by
Jacobson (1999, 2000) in connection with Categorial Grammar. I will not attempt to assess this
work here. See Elbourne forthcoming for some critical discussion.
78
Paul Elbourne
The second matter that a simple copying theory of ellipsis, like the present
one, must address, is what Fiengo and May (1994: 218) call vehicle change.
The question is to deal with examples like the following:
(45)
I turned in my assignment, but most of the other students didn’t.
The problem, of course, is that a straightforward theory of copying (or, indeed,
of deletion under identity) seems to predict that the last sentence here will only
be able to mean “most of the other students didn’t turn in my assignment,”
when in fact it can mean that most of the other students did not turn in their
assignment. This need not be seen as ruling out the current approach, however.
One theory that has been proposed is to see pronouns like my here as simple
bound variables semantically devoid of φ-features, their φ-features being inher-
ited from their binders by an agreement process at PF (Kratzer 1998, Rullmann
2004, Heim 2005), and for present purposes I will assume that something like
this is the case.
2.2
Ellipsis-Containing Antecedents
Let us reconsider an example of an ellipsis-containing antecedent, (12), repeated
here as (46).
(46)
When John had to cook, he didn’t want to. When he had to clean, he
didn’t either.
Recall that the problem is that (46) can mean “. . . when he had to clean, he didn’t
want to clean.”
It can be seen that with the theory of ellipsis just sketched, the problem is
resolved quite easily. We start out with a simplified LF representation like that
in (47) for the utterance in question.
(47)
when John had to cook, he did not want to v VP
when he had to clean, he did not v VP
The Semantics of Ellipsis
79
There are bare VP nodes here, and we have a choice which one we fill in first.
Suppose we take the second and replace it with a copy of the matrix VP in the
first sentence. We obtain the following:
(48)
when John had to cook, he did not want to v VP
when he had to clean, he did not v want to v VP
We can then fill in the resultant bare VP nodes with simple VPs drawn from the
respective preceding sentences:
(49)
when John had to cook, he did not want to v cook
when he had to clean, he did not v want to v clean
The right meaning results. The other cases of ellipsis-containing antecedents,
(13)–(15), will work analogously.
We might very well wonder if this system overgenerates. The simple answer
is that it does indeed, if we suppose it not to be supplemented with other con-
siderations. For example, the theory as it stands predicts that the following will
also be a possible LF structure for (46):
(50)
when John had to cook, he did not want to v clean
when he had to clean, he did not v want to v cook
There is nothing in Theory the First to prevent us reaching the stage shown in
(48) and then looking forward to the second sentence and filling in the ellipsis
in the first with the VP clean, and looking back to the first sentence and under-
standing cook in the final ellipsis site. But I take it that these ellipsis resolutions
will be ruled out by independent factors. After all, the syntactic structure in (48)
is exactly that which we see overtly spelled out in (51):
(51)
When John had to cook, he didn’t want to. When he had to clean, he
didn’t want to either.
And it is a fact that this example cannot be understood as in (50) either. I take
it then that Theory the First is correct as far as it goes, but that it must be sup-
Dostları ilə paylaş: |