There are a number of other logical systems competently specified in the six books in question. Some of these (e.g., the sorted logic featured in Alan Frisch's ``The Substitutional Framework for Sorted Deduction'', in Brachman et al. which marries classical logic to semantic networks) serve narrow but important purposes. Others are systems that anyone seriously interested in LAI (and related mathematical, linguistic, and philosophical matters) ought to be familiar with. One family of such ``required reading'' systems are those formed in the attempt to genuinely come to grips with the expressive power of natural language. Invariably, members of this family build upon the logical systems already described in this paper; the perfect example is Montague semantics, and perhaps the perfect introduction to Richard Montague's work (and related matters) comes by way of Thayse 1989 and Thayse 1991.

Montague, in a series of three seminal papers (all of which
are conveniently
reprinted in Thomason 1974), initiated a program intended to
formulate a syntax and semantics for natural language having the same rigor and
precision as the syntax and semantics for formal languages.
The introduction to this program found in Thayse 1989 and Thayse 1991 is
first rate.
The Thayse 1989 part of this treatment
starts with an extensional, type-theoretic logic; proceeds to
consider Montague's intensional logic; pauses to
engagingly discuss the historical,
philosophical, and linguistic context of Montague's program; then presents
for converting natural language to formal language;
and, finally, briefly discusses
some alternative approaches to reaching Montague's dream. Thayse 1991
is an
articulate discussion of some of the limitations of Montague's approach
(e.g., anaphora isn't accommodated) and some of the proposed solutions - solutions
that show LAI making genuine progress.Anaphoric constructions
are handled by
infusing Montague's approach
with Hans Kamp's **discourse representation theory** (1984). Another problem
afflicting Montague's original approach is that it can't deal with contexts
that are *not* referentially opaque. Thayse 1991 finds a solution in
the **situation semantics** of Barwise & Perry 1983.
We recommend that interested readers
begin with the initial portion of the advanced
material in Thayse 1991, which constitutes a quick
introduction to Montague's logic; from there the reader can move back to the
coverage in Thayse 1989, and then back to the advanced topics treated in
Thayse 1991. We attempt now to give a sense of Montague's work.

Montague's approach borrows numerous elements from the logical systems visited above. For example, his logic is ``multimodal'', in that the logic combines operators for possibility and necessity with temporal operators:

- - necessarily
- - possibly
**F**- - always in the future
- F - sometimes in the future
**P**- - always in the past
- P - sometimes in the past

Belief and knowledge, which we formalized above using the two basic operators from
modal logic, are formalized by Montague with a two-place predicate for each
concept - e.g., *believe*(*a*,*e*) - the first
place occupied by a symbol referring to the agent, the second
to an event. There is
insufficient space here for even a compressed presentation of how the
step-by-step process of Montague's
translation scheme,
when applied to natural language, yields a formal counterpart. But
central aspects of the process are as follows: At the heart
of the translation is an isomorphism between the analysis of an expression
in natural language and its associated analysis in the intensional logic. This
isomorphism comes by way of a **type-theoretic syntax** for the logic and
a **translation function** that maps each syntactic category of
natural language to a
type of the intensional logic. Each natural-language expression *A* of
syntactic category is assigned a logical translation *A*' in the
logic; and *f* is the logical type of category . The approach
assumes that most sentences of natural language can be analyzed according
to two different moods: *de re* and *de dicto*.The distinction
here will be familiar to many philosophers, but others may be seeing it
for the first time. Consider the sentence `John believes that a woman talks'.
The *de dicto* reading is that this sentence means that `John believes that
there exists at least one woman who talks', where this belief doesn't refer
to a particular woman. On the other hand, a *de re* reading would be
that the sentence means that `John believes that there exists a well-determined
woman who is talking'.
Two additional operators handle the *de re* and *de dicto*
moods; these are the intension and extension
operators - and , respectively
the first indicating *de re*, the second *de dicto*.
The final tool used in Montague's translation is the **lambda operator**. Our
readers are doubtless familiar with such ``set builder'' statements as

that is, the set of all people able to beat Deep Blue, a famous chess-playing
computer program. The -operator allows us to say similar things
in the logic itself. For example, if
the two-place relation *Wxy*
is true if and only if *x* has written *y*, and if *u*
is a constant denoting Umberto Eco, then intuitively speaking
the following represents the set of works that Eco has written.

In order to perhaps whet your appetite for Montague's program, here is what all of this machinery produces in the logic for the English sentences ``John examines and solves a problem'' and ``John will not talk'', respectively:

Those interested not only in Montague's work, but in the full range of proposals for how to genuinely deal with the formidable power of natural language (especially natural language's ability to allow reference to nonexistent objects), are encouraged to read Graeme Hirst's ``Existence Assumptions in Knowledge Representation'', in Brachman et al. (We strongly recommend that readers searching for answers to the problems lurking in this area also investigate SNePS (Shapiro & Rapaport 1987).)

Mon Nov 17 14:57:06 EST 1997