A Software Architecture for Simulated Human-Human Conversation with User Intervention

Abstract

We present the high-level design of a prototype AIbased interactive system that simulates conversation between two animated virtual characters. The conversational choices made by each character are influenced by dynamic models of each virtual participant's mental state. Also, the choices made by one character, the user's Conversational Avatar, are at certain times under user control. This style of simulated conversation with user intervention is applicable to a range of potential applications in education and entertainment. Our current application is a training system for coaching caregivers on having social conversations with persons with Alzheimer's Disease.