Knowledge Base Merging by Majority

A common practice for resolving connicts among a group of agents is to let the majority decide. We formalize this principle and apply it to the problem of merging the knowledge of multiple agents. We then postulate logical properties that all knowledge merging operators should satisfy and give a model-theoretic characterization of all merging operators that satisfy the postulates. It turns out that the operators that satisfy the postulates are precisely those that induce a certain kind of partial pre-order over the set of possible worlds such that the models of the merged knowledge base are the possible worlds that are minimal with respect to the pre-order. Then, we review several previous approaches and study them in light of the proposed postulates. We then present a particular method for knowledge merging, CMerge, that satisses all the postulates. We show by example that CMerge appears to resolve connicts among knowledge bases in a plausible way. Finally, we show that CMerge can be implemented by an eecient syntactical transformation from the set of knowledge bases to be merged.