Consistency of rule-based expert systems

Abstract

Consistency of a knowledge-based system has become a topic of growing concern. Every notion of consistency presupposes a notion of semantics. We present a theoretical framework in which both the semantics and the consistency of a knowledge base can be studied. This framework is based on first order flat many-sorted predicate logic and is sufficiently rich to capture an interesting class of rule-based expert systems and deductive databases. We analyse the feasibility of the consistency test and prove that this test is feasible for knowledge bases in Horn format without quantification.