ELSEVIER Expert Systems with Applications 14 (1998) 83-90
Expert
Systems
with
Applications
Abduction, that ubiquitous form of reasoning
Ram6n Brena*
Centro de lnteligencia Artificial, Instituto Tecnol6gico y de Estudios Superiores de Monterrey, Monterrey, N.L., M~xico
Abstract
In recent years, attention has been devoted to abduction, a hypothetical form of non-monotonic reasoning that tries to fit the best
'explanation' to a given observation. In this paper we present a collection of applications of automated abductive reasoning
developed in the Center for Artificial Intelligence of the ITESM (Monterrey, Mexico) in the last five years, covering a range from
natural language understanding to software re-use. © 1998 Elsevier Science Ltd. All rights reserved
1. INTRODUCTION
A usual common-sense reasoning problem consists in
finding a 'good' explanation for an observed fact, given
a base of knowledge and assumptions. For example, a
physician tries to diagnose a disease that could produce
the observed patients symptoms, based on his/her
knowledge of diseases and their symptoms, the patients
clinical history, and whatever she/he knows about the
case.
In abductive reasoning we try to find an explanation of
a set of observed facts from a given knowledge base
(Charniak, 1985). Abductive reasoning can be illustrated
in the following way (Charniak, 1985):
X causes Y
Y is known to be true
then X can be an explanation of Y.
This kind of reasoning can be applied to a variety of
tasks, including diagnosis and interpretation, which can
be applied to many fields like medicine, machine ~:
troubleshooting, natural language understanding and
others; examples of all of the preceding will be presented
in the following. A
The structure of this paper is as follows: in Section 1.1
a precise description of the abductive reasoning frame-
work is presented; in Section 2 we present an application
of abductive reasoning to expert systems prototyping; in
Section 3 we describe an application of automated
abduction to software re-use, and finally in Section 4 we
present an application of automated abduction to natural
language understanding, followed by a general conclu-
sion.
1.1. The Theorist Framework
Theorist is a precise framework for abductive reasoning,
as well as an implementation for automated abduction,
proposed by David Poole (1994). It is based on the idea
of 'theory formation'--that is, incremental formation of
the 'theory' that explains the observed fact--from a fixed
set of possible hypotheses. In other words, theory
formation is seen as accepting as true instances of some
of the hypotheses stored in the hypotheses set. Clearly
the hypotheses set is composed of a set of declarations
that could explain some of the observed facts. The user is
responsible for providing an adequate set of plausible
hypotheses.
In Theorist, the user provides three sets of first order
logic formulas: ~
is a set of closed formulas called facts. They are
supposed to be true in the world that is being
modeled.
is a set of (possibly open) formulas that act as
possible hypotheses, some of which could be used,
once instantiated, as a part of an explanation, if it
is consistent with the facts and the constraints.
is a set of closed formulas taken as restrictions.
* Email: rbrena@campus.mty.itesm.mx.
~This definition can be consulted in Poole (1988); they are included
here just for the sake of completeness.
0957-4174/98/$19.00 Copyright © 1998 Elsevier Science Ltd. All rights reserved.
PII S0957-4174(97)00066-3