Knowledge-Base Revision Using Implications as Hypotheses

The integration of an assertion stemming from some sender agent into the knowledge base of a receiver agent may be hindered by logical incompatibilities. In the belief-revision scenario the receiver agent trusts the incoming assertion and starts a revision of his knowledge base— resulting in the elimination of some formulas of the knowledge base. In this paper a new family of belief-revision operators is analyzed, which exploits the idea of revising hypotheses on the semantic relatedness of the symbols belonging to the name spaces of the sender and the receiver. Ambiguous use of the same symbol p across different agents can thus be handled by setting up an initial set of hypotheses of the form p→ p′, p′ → p where p′ belongs to the receiver and p belongs to the sender. This approach generalizes the revision operators of J. Delgrande and T. Schaub, which are based on biimplications as hypotheses. It can be shown that finite representations for the implication based revision operators exist. Additionally, the non-sceptical versions of these new revision operators can be axiomatically characterized by postulates, which rely on the equivalent representation of a knowledge base as the set of prime implicates it entails.