The Conversational Lexical Affect Recognition Kit (CLARK) uses appraisal variables to better capture the dynamics of emotions in a conversation.
We are working towards an architecture for an interactive agent that responds to the users emotions. The agent integrates affect recognition models for language, facial expressions, and eye gaze. It uses a rules engine to reason about how the emotions affect how the agent responds to the user. The initial proof of concept demonstrates an agent helping train the user to be a better listener.