(A) frame-and horn claude-based hybrid knowledge representation

Abstract

Intelligent systems have various kinds of representational needs to be used in real world. To support multiple representational formalisms and specialized reasoning for each formalism in a well-defined and precise way, a new strategy called hybrid systems for knowledge representation has been studied. In this thesis, a new method for hybrid knowledge representation is developed by integrating frames and Horn clauses. A hybrid system called Sphinx is also developed on the basis of this formalism. Sphinx is a theorem prover-based hybrid system of which proof mechanism is a SLD-resolution strategy augmented with a special-purpose inference system for frame-based descriptions. From systematic point of view, Sphinx also consists of two major components for terminological and assertional reasoning. It is also shown that Horn clause can be combined efficiency with the conventional frame-based description system if some useful assumptions are given. By integrating theses two kinds of reasoners, more expressive power and deductive capability than simple database-style hybrid systems can be offered, while retaining the reasoning efficiency. To reflect the terminological knowledge completely in proving a query, deduction rules are included automatically in the ABox and an extended logical query language involving explicit logical connectives and quantifiers is designed. An efficient and sound query evaluation procedure with domain closure assumption and the notion of complete knowledge base is also presented. To permit incremental additions and retractions of factual knowledge, a new truth maintenance strategy is presented in this thesis. The truth maintenance strategy uses semantic dependencies of predicates by referring the TBox. With this approach, it is shown that even a theorem prover-based hybrid systems can support the dynamic revision of knowledge. It is also ascertained that the use of the negation-as-failure rule can simplify greatly the query processing and tru...