Abstract:
The concept being proposed here is to use prototype semantics to represent an agent's belief state. Prototype semantics is a linguistic theory that emerged in the 1980s -- the key idea is to describe the meaning of an utterance, or a notion, by defining the prototype (the most typical example to which the notion refers) and the extension rules describing 'family resemblances' between various entities and, in consequence, allowing to derive less typical instances from more typical ones.

The intuition behind this paper is that in situations of incomplete information such a representation of the agent's knowledge may be easier to deal with than a straightforward probabilistic representation -- especially when belief changes are not necessarily monotonic, and informations being acquired by the agent may be vague themselves. Moreover, when an exhaustive search through all the possibilities is impossible the agent may benefit from analyzing the typical situations instead of random ones. This property was tested for a family of games with incomplete information on finite binary trees.

Keywords: belief change, prototype semantics, games with incomplete information, language communication, knowledge representation, nonmonotonic reasoning, multiagent systems.

Computational Intelligence Group @ Technical University of Clausthal
Human Media Interaction Group @ University of Twente
Computer Science Group @ University of Gdansk	Last modified 2001-05-11