Uncovering redundancy and rule-inconsistency in knowledge bases via deduction

Abstract

Two examples of dangerous rule interactions-redundancy and rule inconsistency-are reviewed. Their implementation is described in the context of the Defense Advanced Research Projects Agency (DARPA) Expert-System Validation Associate (DEVA). Emphasis has been placed on devising strategies that can be used on atypical knowledge bases, i.e., those containing an especially hostile search space. A hostile search space is one which is very bushy and/or contains possibly many cycles in the rules. Techniques for detecting redundancy and rule-inconsistency anomalies in the absence of facts are discussed. Two approaches are considered: the restricted generate-and-test approach and residue analysis.<>