98
Paul Elbourne
is, that the final clause is I swam the English Channel and climbed
Kilimanjaro. That the elided occurrences must be conjoined in the
representation of ([104]) just follows from the way they can “fit”
into its structure.
They further suggest the following (1994:200), referring to examples like (78a)
(“Whenever Max uses the fax or Oscar uses the Xerox. . . ”):
As a general rule, the discourse sentence is also the domain from
which the elided coordinating element is drawn. In a case of dis-
junction, for instance, or is reconstructed.
It appears, then, that Fiengo and May are assuming that a conjunction must be
reconstructed somehow on the basis of the linguistic environment. A possible
problem with this theory is that there are split antecedent cases when there is no
and or or in the linguistic environment (Elbourne 2001):
(105)
Mary swam the English Channel. Mary climbed Kilimanjaro. I did too.
It is unclear to me how (105) could be dealt with in Fiengo and May’s theory
of split antecedent cases. It poses no problem for the theory advocated in this
article, of course, since this theory does not rely on a conjunction being present
in the linguistic environment.
Fiengo and May do not deal with cases of ellipsis-containing antecedents.
3.3
Hardt 1999
As far as I know, Hardt’s (1999) theory is the only one previously published
that attempts to account for binderless sloppy readings, ellipsis-containing an-
tecedents and split antecedents. It is also not difficult to see how it might be
extended to deal with cases of ellipsis with no linguistic antecedent. For the
purpose of illustration, I will give an informal summary of how Hardt analyzes
binderless sloppy readings and ellipsis-containing antecedents.
The Semantics of Ellipsis
99
Hardt uses a dynamic semantics incorporating a notion of discourse cen-
ter, based on the Centering framework of Grosz, Joshi and Weinstein 1995. In
particular, discourse representation structures (the “boxes” of traditional DRT
(Kamp 1981)) contain special discourse markers (variables) assigned to the cen-
ter (roughly, topic) of the discourse. This enables Hardt to explain binderless
sloppy readings along the following lines. Take (106).
(106)
If Tom was having trouble in school, I would help him. If Harry was
having trouble, I wouldn’t.
In the first sentence, the discourse center is Tom. The overt pronoun him is
translated by a special discourse marker assigned to the center. So help him
means something like “help the current center.” This meaning is then understood
at the ellipsis site; more precisely, it is the assigned as the value of the INFL of
the second sentence, which is a deictic element in this theory. In the meantime,
however, the center has changed. Harry is the discourse center of the second
sentence. So the second sentence ends up meaning “I wouldn’t help Harry,” as
desired.
The same principle can be used in cases of ellipsis-containing antecedents.
Take our standard example:
(107)
When John had to cook, he didn’t want to. When he had to clean, he
didn’t either.
Hardt supposes that in the first sentence the discourse center is the property of
cooking. The first VP-ellipsis is resolved by having to be a deictic expression
that picks up the property that is the current center. We arrive at the meaning
“want to cook,” then, for the end of the first sentence. The value of the deictic
INFL of the second sentence is taken to be the DRT representation of want
to, something interpretable as “want to perform the kind of action that is the
current center.” And again by the time we get to this point the center has shifted,
100
Paul Elbourne
according to Hardt. It is now the property of cleaning, and the so the correct
interpretation is obtained.
Even from this informal summary, it can be seen that it is important to
Hardt’s system that there is no syntactic structure, or at least no syntactic com-
plexity, at ellipsis sites. The correct interpretation is arrived at by assigning de-
notations involving variables picking out the current center to the INFL of the
ellipsis sentence. (One could also imagine a variant in which there was a deictic
element in the VP position, as opposed to having INFL do this job.) But this
feature of the theory, which lends it a useful flexibility and power, means that
it is ill-equipped to deal with cases where there seems to be movement from
ellipsis sites, as in the following examples.
(108)
Which book did John read? And which book did Bill?
(109)
John read every book that Bill did.
See Johnson 2001 for a summary of the controversy on whether theories without
normal syntactic structures in the ellipsis sites can deal with examples like these.
The upshot is not encouraging for those theories, and things seem especially
difficult for the particular version that Hardt puts forward, according to which
there is nothing whatsoever in ellipsis sites. By contrast, the theory advocated
in the current article has normal syntactic structure in all ellipsis sites.
14
14
Hardt’s theory also faces a knotty technical difficulty in analyzing certain seemingly simple
cases of sloppy identity. Take Tom loves his cat and John does too (Hardt 1999:194). Using the
device described above, Hardt analyzes loves his cat as roughly “loves the cat of the current
center,” and wants this meaning understood at the ellipsis site, by which time the center has
changed to John. But the actual DRT representation used to express this meaning by Hardt also,
necessarily given the system, contains a discourse marker for Tom’s cat. Roughly speaking,
and in slightly incongruous terms, we can think of his cat here as meaning something like
“the unique x such that x is the cat of the current center and x is identical to c,” where c is
a constant referring to Tom’s cat. If we understand this at the ellipsis site, then, we obtain a
contradiction: the claim would be that John loves the cat of the current center (himself) that is
identical to c (not his own, but Tom’s cat). To avoid this difficulty Hardt proposes to reconstruct
an “alphabetic variant” of the original property, one that replaces the troublesome discourse
marker referring to Tom’s cat by another one (Hardt 1999:195). But this seems like the merest