Abstract

Psychology and philosophy have long studied the nature and role of explanation. More recently, artificial intelligence research has developed promising theories of how explanation facilitates learning and generalization. By using explanations to guide learning, explanation-based methods allow reliable learning of new concepts in complex situations. This volume addresses fundamental issues in generating and judging explanations: When to explain, what constitutes an explanation, how to build explanations, and how to evaluate candidate explanations. It argues that standard models, which are neutral to context and experience, are inadequate to deal with the problems that arise in everyday explanation. It presents an alternative theory in which context---involving explainer beliefs, goals and experience---is crucial in generating and judging explanations. It examines the role of anomalies in motivating explanation, describes a model of pattern-based anomaly detection, and shows how an analysis of the content of anomalies, explanations, and explanation purposes can be used to guide generation and evaluation of explanations by a case-based explanation system. The theory is implemented in ACCEPTER, a computer system that understands stories, detects anomalous events, guides retrieval and adaptation of stored explanations, and evaluates candidate explanations.