ERIC: a generic rule-based framework for an affective embodied commentary agent

We present ERIC, an affective embodied agent for realtime commentary in many domains. The underlying architecture is rulebased, generic, and lightweight – based on Java/Jess modules. Apart from reasoning about dynamically changing events, the system can produce coherent natural language and non-verbal behaviour, based on a layered model of affect (personality, mood, emotion). We show how reasoning, template-based natural language generation and affective appraisal can be implemented within the same rulebased paradigm. To make the system domain independent we worked on two different domains, a virtual horse race and a multiplayer tank battle game. We empirically evaluated the genericness of the system by measuring the effort it takes to change the domain, and discuss the results.