Recovery and Common Sense Reasoning: Principles of Nonmonotonic Theory Recovery

Ninth Midwest Artificial intelligence and Cognitive Science Conference

Volume

Ninth Midwest Artificial intelligence and Cognitive Science Conference

Contents

We use common-sense reasoning to make predictions about what normally will be the case. Such reasoning is captured by using nonmonotonic semantics selecting a set of intended models from a larger collection of models. Such a selection process, however, might fail: although a theory might have a nonempty set of non-intended models, the set of intended models might be empty. We call such a theory nonmonotonically inconsistent. In order to restore consistency, we want to recover the theory, replacing it by a closely related one that has intended models. In this paper we present some rationality postulates for recovery of common-sense theories. As a result, we show that, contrary to intuition, in most cases contractions of a common-sense theory are inadequate to restore consistency, and that the only minimal recovery operations are expansions of it.