In (57a), we posit that a predicative expression like sick denotes the char-
acteristic function of a set of individuals. When applied to an individual such
as john in (57b), we have a proposition of type . This holds equally if the
property is applied to a variable like x
3
as in (57c): in this case, the truth of the
proposition is evaluated relative to the value of x
3
in the context-determined
assignment function g. But crucially the expression itself is of the same type,
namely , that (57b) is. Last, we can bind the variable with a λ -operator as
in (57d), yielding again a characteristic function of a set. These expressions
are interpreted in a model theory using a model M = and a denotation
function.
The above represents a typical way of modeling meanings in a typed sys-
tem, using standard definitions such as the following (from Bernardi 2002:16):
(58)
D
EFINITION
[Typed λ -terms]. Let VAR
a
be a countably infinite set
of variables of type a and CON
a
a collection of constants of type a.
The set TERM
a
of λ -terms of type a is defined by mutual recursion
as the smallest set such that the following holds:
i. VAR
a
⊆ TERM
a
ii. CON
a
⊆ TERM
a
iii. (α(β )) ∈ TERM
a
if α ∈ TERM
and β ∈ TERM
b
,
iv. λ x.α ∈ TERM
, if x ∈ VAR
a
and α ∈ TERM
b
.
A common practice in work in natural language semantics is to assign
λ -terms as the translation of lexical items, such as the following.
(59)
a.
every
= λ P
et
λ Q
et
[∀x
e
(P(x) → Q(x))]
b.
boy
= λ x
e
[boy(x)]
c.
see
= λ x
e
λ y
e
[see(x)(y)]
But this use of the λ -operator is not a necessary one. Imagine instead
that λ -abstraction occurs in the course of or as part of the semantic com-
position, not as stipulated in lexical entries. This is in fact a common view:
Carpenter 1997 for example uses a system that can apply variables and λ
binders separately to terms, and systems like Heim and Kratzer’s 1998 intro-
duce λ -binders as the result of certain movement operations. On this view,
then, λ -abstraction occurs as necessary to enable semantic composition, but
not otherwise. It is a possible precursor to function application (other sys-
tems are conceivable, of course: see Chung and Ladusaw 2004 for an explicit
proposal for other semantic composition operations in addition to function
application, and recall that Heim and Kratzer 1998 also use an operation of
function ‘identification’ as well as application). The result of this view of the
semantics is that predicates have a variable in them, but no λ -binder. When
used in isolation, they will therefore have a free variable.
This is all that needs to be said to account for two of Stainton’s three
main subcases. Stainton assumes that the semantic value of a phrase like on
the stoop
or quite a lot of children, used in isolation, will be either what the
interpretation function I returns or an appropriate λ -translation: in either case,
on the stoop
will be (as in (57a,d) above) and quite a lot of children will
be , as follows, for example (assuming for simplicity that the PP denotes
a predicate and that quite a lot
∗
C
is predicate true of plural individuals x iff
the cardinality of x exceeds some contextually given amount C):
(60)
a. λ x
2
[on.the.stoop(x
2
)]
b. λ Q
et
[∃z[quite.a.lot
∗
C
(z) ∧ children(z) ∧ Q(z)]]
But if introduction of variables—here x
3
and P, with β -reduction—is an
available option (as in (58iii) above), then there is a further possibility:
(61)
a. on.the.stoop(x
3
)
b. ∃z[quite.a.lot
∗
C
(z) ∧ children(z) ∧ P(z)]
These expressions have free variables—x
3
and P: ‘slots’, in other words.
What the values of these variables will be is determined by the assignment
function. So Stainton is right that the pragmatics is crucial, and that our intu-
itions require that it be the context that determines what individual or prop-
erty is used, but once we admit that the assignment function is responsible
for ‘slot-filling’ of unbound variables, we already have in place the semantic
mechanism needed.
One additional assumption is needed to account for the third major sub-
case, that of individual-denoting phrases like Barbara Partee. For such ex-
pressions, we have to assume, with Partee and Rooth 1983, Jacobson 1999,
Barker to appear, and many others, that an individual-denoting expression
can lift into a generalized quantifier type (whether freely so or due to require-
ments of the context is immaterial: this seems necessary for the interpretation
of conjunctions like John and every woman, etc.). Given this option, Barbara
Partee
can lift into the expression in (62a), to which variable introduction and
β -reduction apply, yielding (62c).