A Local Models Semantics for Propositional Attitudes

We are interested in the representation and mechanization of propositional attitudes, and belief in particular, inside complex reasoning programs, e g knowledge representation systems, natural language understanding systems, or multiagent systems. Modal logics are the obvious formalism for this goal, the one which has been most widely proposed and studied in the logic and philosophical literature. However, as already discussed in detail in [Giunchiglia and Serafini, 1994; Giunchiglia, 1995], modal logics are hardly used in the existing implemented systems. From the point of view of implementors, the problem is that it is very hard to codify in modal logics all the needed information (e.g. the agents’ knowledge, their — usually very different — reasoning capabilities, the — usually very complicated — interactions among them), in a way to have efficient, easy to develop and to maintain implementations. Modal logics allow for compact representations, where one has basically to provide the “appropriate” axioms. This is very elegant, and beautiful from the point of view of the logician. The drawback is that implementors do not find in the formalism enough structure for directly representing all they would like to represent.