Formalizing Belief Reports the Approach and a Case Study ? 1 Beliefs and Substitutivity

One of the most interesting puzzles in formalizing belief contexts is the fact that many belief reports can be given both an opaque and a transparent readings. A traditional explanation is that the two readings are related to the failure and success of the principle of sub-stitutivity respectively, and this in turn is explained with the de re/de dicto distinction. We propose an alternative analysis, based on the idea that another agent's beliefs can just be quoted (preserving opacity) or translated into the reporter's language (allowing for transparency). We show that MultiContext systems allow for the formalization of these two phenomena at the same time, thanks to their multi-language feature. A very important part of our intelligence is the ability to reason about our own and other people's beliefs. We have beliefs about our beliefs; we learn from other people's beliefs; we accept (don't accept) other people's beliefs among our beliefs; we share our beliefs with other people; we report other people's beliefs; we infer new information by combining diierent people's beliefs; we make decisions based on conjectures about other people's beliefs; many communication conventions are based on what we believe other people believe; and so on. Despite the pervasiveness of the notion of belief and the apparent easiness for humans to deal with it, representing beliefs and formalizing reasoning with and about beliefs raises very diicult problems. From a logical point of view, a well-known problem is that a very general and intuitive principle, the principle of substitutivity, fails in its unrestricted version 1. Intuitively, the unrestricted principle of substitutivity (UPS) states that whenever two terms (two sentences) ? We thank the Mechanized Reasoning Group at DISA (Trento), ITC-IRST (Trento) and DIST (Genoa). This work is part of the MRG project Distributed Representations and Systems