marina@ai.toronto.edu (Marina Haloulos) (12/21/89)
Department of Computer Science, University of Toronto
(GB = Gailbraith Building, 35 St. George Street)
-------------------------------------------------------------
ARTIFICIAL INTELLIGENCE SEMINAR
GB305, at 11:00 a.m., Thursday 11 January 1990
Dr. David McAllester
M.I.T.
"A Functional Motivation for Natural Language Syntax"
One can define a class of semi-automated reasoning systems based on a
distinction between ``obvious'' statements, that do not require explicit
justification, and non-obvious statements that must be explicitly
justified. The notion of obviousness can be based on decision procedures
for certain inference problems. We show that, within this semi-automated
framework, and for a given class of inferential decision procedures, the
power of the automated reasoning underlying the notion of obviousness is
extremely sensitive to the syntax of the formal langauge. The syntax that
currently yields the strongest known decision procedure (i.e., that
maximizes that class of obvious statements) is based on a fragment of
English under mantague semantics. This talk will present the general
framework, discuss various choices of syntax, and argue for the functional
superiority of natural language syntax under Montague semantics.