Learning to Recognize Plans Involving Affect

Abstract

This chapter reviews a system called AMAL that uses knowledge of plans and emotions to construct explanations of actions, and allows the learning of new plan recognition rules based on these explanations. Motivation analysis and plan recognition, the task of understanding the mental states underlying observed actions, requires knowledge about the causal relationships between emotions and actions. Emotions serve to focus the recognition process on specific actions or events in cases where people select particular plans of action based upon their emotional state such as running away from a feared object or striking someone who angers them. AMAL draws upon three sources of input: a library of cases, an emotion law library, and a general law library. At present, there is a library of over 150 cases that can be explained by AMAL. The cases are almost exclusively drawn from a diary study of emotional events by Turner. The system also draws upon a general law library, where facts concerning plans and causal relations are stored. Other rules, such as what actions are considered blameworthy and what objects or people are typically regarded as appealing are also kept in the general law library.