Simulating Alzheimer’s Discourse for Caregiver Training in Artificial Intelligence-based Dialogue Systems

Abstract

A relatively new focus of computer science research in artificial intelligence (AI) is the development of embodied conversational agents(ECAs) (Cassell et al. 2000a; Johnson et al. 2000). Created by a computer program, an ECA is an animated character whose modes of communication may include speech, gesture, and facial expressions. The words uttered by an ECA may be scripted by a human author, or partially or completely synthesized from more abstract representations (e.g. of speaker goals and affect and propositional content) via natural language generation techniques (Reiter & Dale 2000). Similarly, intonation, gesture and facial expression may be scripted, or generated by the computer program based upon functional models of paralinguistic communication (Cassell et al. 2000b). ECAs can be programmed to react to sensed or simulated events in their environment, or to respond to spoken or typewritten language provided by a computer user. AI models of emotion and personality may be used to enhance an ECA’s believability (Elliott & Brzezinski 1998). The motivation for developing ECAs is two-fold. First, an ECA can be used to submit a linguistic or psychological model to testing, i.e., via simulation instead of testing with human subjects. Second, ECAs provide a new and compelling mode of human-computer interaction for education and entertainment.