A Semantic Approach to Combining Preference Formalisms.

We propose a versatile framework for combining knowledge bases in modular systems with preferences. In our formalism, each module (knowledge base) can be specified in a different language. We define the notion of a preference-based modular system that includes a formalization of meta-preferences. We prove that our formalism is robust in the sense that the operations for combining modules preserve the notion of a preference-based modular system. Finally, we formally demonstrate correspondences between our framework and the related preference formalisms of cp-nets, preference-based planning, and answer set optimization. Our framework allows one to use these preference formalisms (and others) in combination, in the same modular system.