The Semantics of Ellipsis
73
TP,
∃t(t <
NOW
& at t : ∃e(stabbing(e)
& Theme(e, Caesar) & Agent(e, Brutus)))
Brutus
λx.∃t(t <
NOW
& at t : ∃e(stabbing(e)
& Theme(e, Caesar) & Agent(e, x)))
λ
2
T ,
∃t(t <
NOW
& at t : ∃e(stabbing(e)
& Theme(e, Caesar) & Agent(e, g(2))))
T
past
vP,
λe.stabbing(e) & Theme
(e, Caesar) & Agent(e, g(2))
t
2
λy.λe.stabbing(e) & Theme
(e, Caesar) & Agent(e, y)
v
VP,
λe.stabbing(e) &
Theme
(e, Caesar)
stab Caesar
Figure 1: Brutus stabbed Caesar
as the (give or take φ-features), the interpretation of pronoun plus index in this
case will be “the unique x such that x is identical to John,” or, in other words,
“John.” This position has the advantage of unifying the referential and bound
occurrences of pronouns with their use as donkey anaphors (Elbourne forth-
coming), assuming a theory whereby donkey pronouns are analyzed as definite
descriptions (Cooper 1979, Neale 1990, Heim 1990, Elbourne 2001, forthcom-
ing).
I will follow Burge (1973), Recanati (1993), Larson and Segal (1995) and
74
Paul Elbourne
Elbourne (forthcoming) in assuming that names are basically nouns. We often
see them occurring with overt determiners, as in (37).
(37)
a.
An embattled Tony Blair addressed the Commons this afternoon.
b.
Which Alfred did you mean? This Alfred?
When they appear to stand alone, they will be preceded by a special phono-
logically null definite determiner
THE
. This is paralleled by those languages
like Classical Greek and some dialects of German in which names are regularly
preceded by an overt definite article.
As for the semantics of names on this view, Burge’s (1973) basic idea is that,
for example, Alfred means something like “entity called Alfred,” and variants of
this have been proposed by the other authors just cited. In Elbourne forthcoming
I propose that on most occasions of use Alfred will mean “entity called Alfred
and identical to a,” where a is an individual constant picking out a particular
entity called Alfred. In this article I will just assume things like [λx.x is an
Alfred] for the meaning of names, since their exact semantics is orthogonal to
the issues of primary concern.
I will just assume that nouns are of type e,t , and that definite articles,
including pronouns, are functions from predicates of type e,t to individuals
(Heim 1991, Heim and Kratzer 1998, von Fintel 2004, Elbourne 2005, forth-
coming), as proposed originally by Frege (1893). The semantics for some rele-
vant lexical items is shown in (38).
4
(38)
[[the]]
g,h
= λf
e,t
.ιx f(x) = 1
[[him]]
g,h
= λf
e,t
.ιx f(x) = 1
4
The semantics given in (38) is a simplification in that we probably need to embed the
whole lexical entry in a situation semantics or possible worlds semantics for full adequacy, and
have definite articles be functions from properties to individual concepts, i.e. functions from
circumstances of evaluation to individuals, as in Elbourne 2005, forthcoming. I overlook this
complication here and continue to operate with an extensional semantics. I also overlook the
φ-features on the pronouns, for the sake of simplicity.
The Semantics of Ellipsis
75
[[cat]]
g,h
= λx.x is a cat
[[Alfred]]
g,h
= λx.x is an Alfred
The semantics of the metalanguage operator ι is as follows: for any function
f, the denotation of ιx f(x) = 1 will be of type e, if it is defined; if there
is exactly one entity x such f
(x) = 1, the denotation of ιx f(x) = 1 will
be that very individual; if there is no such individual, the whole expression
will have no value. (So the expression in effect introduces a presupposition that
there is exactly one such individual, since an utterance containing it will not
be felicitous otherwise.) The individual that is the value of the expression will
naturally vary from model to model. For example, if our universe is
{2, 3, 4},
then the denotation of ιx x >
3 is 4; if the universe is {2, 3, 5}, the value
of the same expression will be 5. This, simply put, is how definite descriptions
differ from constants.
2.1.3
Ellipsis
In this section I will sketch a theory of ellipsis that will enable us to give a
straightforward account of the sentences involving ellipsis-containing anteced-
ents and binderless sloppy readings, and I will apply it to some relatively simple
data. In the next sections I will use to analyze the problematic data that we saw
in section 1.
The theory is as follows:
(39)
Theory the First
VP-ellipsis and NP-deletion consist in the generation of bare VP and
NP nodes, respectively. These structures are sent to PF. There is an LF
process of resolving the ellipsis, whereby the bare nodes are replaced
with a copy of a phrase of the same syntactic category drawn from the
linguistic environment.
76
Paul Elbourne
TP
DP
THE
Bill
T
λ
2
T
does
vP
t
2
v
v VP
Figure 2: Bill does too
According to this account, then, a sentence involving VP-ellipsis or NP-deletion
begins life as a structure that is syntactically incomplete. For example, the last
sentence of (40) will have the (possibly simplified) structure in Figure 2.
(40)
John loves Mary and Bill does too.
Note that in Figure 2 we have a VP node that is simply not spelled out any
further. This will be possible if we adhere to a traditional conception of phrase
structure rules that allows things like (41).
(41)
v
→ v VP
It is obviously incompatible with Chomsky’s (1995) Bare Phrase Structure, ac-
cording to which the idea of a phrasal node with no daughters does not make
sense. I hope to show that significant empirical advantages can be gained from
the traditional conception of syntactic rules.
We generate, then, a structure like that in Figure 2, and this is what is pro-
nounced. Ellipsis resolution will then be an LF process that replaces the bare
phrasal node with a copy of a phrase of the same syntactic category drawn from
the linguistic environment.
5
In the case of the current example, then, we copy
the antecedent VP [
VP
love Mary] and replace the empty VP node with it.
5
We will need to make an addition to our theory when we return to the consideration of
Dostları ilə paylaş: |