Representing Emotions with Linguistic Acuity

For a robot to make effective and friendly interaction with human users, it is important to keep track of emotional changes in utterance properly. Emotions have traditionally been characterized by intuitive but atomic categories or as points in evaluation-activity dimensions. However, this characterization falls short of capturing subtle emotional changes either in narration or in text, where the vast majority of information is presented with a host of linguistic constructions that convey emotional information. We propose a novel representation scheme for emotions, so that such important features as duration, target and intensity can also be treated as first-class citizens and systematically accounted for. We argue that it is with this new mode of representation that the subtlety of the emotional flow in utterance can be properly addressed. We use this representation to encode the emotional states and intentions of characters in the drama scripts for soap opera and describe how it is utilized in conjunction with parsing for lexicalized grammars.