Hierarchical Knowledge Bases and Efficient Disjunctive Reasoning

We combine ideas from relation-based data management with class hierarchies to obtain Hierarchical Knowledge Bases, which have greater expressive power while maintaining the beneets of predictable and eecient information processing. We then consider the problem of reasoning with certain limited forms of disjunctive information. We show that hierarchical knowledge bases can be used for eecient approximate reasoning with such information. The signiicant features of our approach include a well-conditioned trade between ee-ciency and accuracy, with a sound and complete limit case, and approximations guided by the structure of the domain theory. Because of the structure imposed on the knowledge base, it is possible to characterize the potential error in any approximation.