Disambiguation by Prioritized Circumscription

This paper presents a method of resolving ambiguity by using a variant of circumscription, prioritized circumscription. In a disambiguation task, human seems to use various preferences which have various strength. In prioritized circumscription, we can express these preferences as defeasible constraints with various strength and we infer the most preferable logical models which satisfy stronger constraints as much as possible. This representation is very natural for disambiguation since we can regard a logical interprentation as a possible reading and the most preferable logical models as the most preferable readings. We argue that prioritized circumscription is another promising method for the task. We also discuss an implementation of prioritized circumscription by a hirarchical logic programming (HCLP) language.