A PRINCIPLED ANALOGICAL TOOL BASED ON EVALUATIONS OF PARTIAL CORRESPONDENCES OVER CONCEPTUAL GRAPHS
Date
1989-05-01
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Analogy is one reasoning method used extensively and effectively by
humans. Because analogy is so widely used, it is felt that giving computers
this ability will make them more useful in supporting and emulating human
reasoning. This thesis describes the development of a principled
Analogical Tool which deals with this issue.
The Analogical Tool results from instantiation of a general framework for
analogy. This framework is based on analysis of previous work and
characterizes analogies as common generalizations. Using conceptual
graphs [Sowa, 1984] as the knowledge representation scheme for the
general framework, the tool forms analogies and supports analogical
inferencing based on them.
Analogy formation in the tool derives minimal common generalizations as
preferred "stronger" analogies. This formation is under the control of four
formation evaluations and two ordering evaluations. The formation
evaluations operate to determine how and what parts of the analogues are
generalized and the ordering evaluations partially order the generalizations.
Analogical inferencing based on the tool proceeds by hypothesizing new
information in a target analogue based on information in a similar source
analogue. This inferencing results in extension of the target analogue and
formation of a new minimal common generalization.
Testing of the Analogical Tool on examples of analogy taken from
significant research, shows that the tool compares well with other systems.
Examples range from simple analogy formation, through schema abstraction to
metaphor understanding. Further critical evaluation of the tool shows
limitations arise due to relevancy issues. These relevancy limitations
are intrinsic to systems using inference based on similarity. An
extensibility principle is introduced to deal with some relevancy limitations
in inferencing based on the tool. Other relevancy issues are dealt with by
the introduction of determination rules from Russell [1987].
Description
Keywords
Computer Science