ling@ria.ccs.uwo.ca (Charles X. Ling) (02/22/90)
NOTICE: THIS TALK IS ON WEDNESDAY !!!
Wednesday, February 28, 1990. 2:00 pm. Middlesex College, Room 316
Dr. James P. Delgrande
School of Computing Science
Simon Fraser University
Burnaby, B.C.
Canada V5A 1S6
will speak on:
Default Logic Revisited
A B S T R A C T
Perhaps the best known and most thoroughly examined approach to defeasible
inference in Artificial Intelligence is Reiter's default logic.
In this approach, classical first-order logic is augmented with a set of
domain-specific rules of inference of the form $A:B/C$. The intended
reading of a rule is ``if $A$ can be proven from what's known, and $B$
is consistent, then conclude $C$''. Unfortunately, the approach does not
always provide the results that one might intuitively expect:
in some cases ``reasonable'' default conclusions are not forthcoming;
in others, ``unreasonable'' conclusions are. In this paper we address
these problems. We argue that the problems arise because, first, the
requirement of proving the antecedent $A$ is too strong and second,
the notion of consistency given for the justification $B$ is too weak.
Three variants of default logic are presented:
the first addresses the difficulties with prerequisites; the second
addresses justifications; while the third is the union of the first two.
In the first case we show that the full generality of default logic isn't
required, and instead argue that one just needs defaults of the form
$:B/C$, where $B \supset C$. In the second case we modify the manner
in which consistency is used in the definition of a default extension.
The combined approaches lead to a more general formulation of default
theories which arguably better conforms to intuitions regarding
consistency-based default reasoning. We show too that one can now reason
{\em about} a set of defaults and can determine, for example,
if a particular default in a set is redundant. The overall result,
we claim, is a variant of default logic that retains the
desirable properties of the original formulation and fully addresses
objections that have arisen.
*********
Coffee and cookies will be served after the colloquium in room 300