A unified approach to handling uncertainty during cooperative consultations

Abstract

In this paper, we present a unified approach to handling uncertainty during plan inference in cooperative consultations. This approach assists with the following aspects of the plan inference process: inferring a user's intentions among a number of possibilities, deciding whether to admit an unlikely interpretation of the user's request or to actively acquire information from the user, determining whether a perceived ambiguity in a user's request is to be resolved by heuristics or by soliciting information from the user, and deciding whether a recognized intention is sufficiently detailed so that a plan may be proposed to satisfy it. We define an information-theoretic measure which allows us to determine the amount of information in an interpretation of a user's request, and show how this measure is combined with probabilities of interpretations to give preference to interpretations that are better defined. Our approach is implemented as part of a computerized consultant that operates as a travel agent.