Non-instructional linguistic communication with virtual actors

We explore a new paradigm for natural language communication with autonomous agents. While the dominant paradigm is to use natural language instructions from which the agent behaviour is generated, we investigate how natural language input can influence a pre-existing plan-based behaviour, by interfering with the various types of sub-goals in such a plan. Using as a testbed a fully implemented interactive storytelling application based on virtual actors, we identify various forms of communicative actions and how these can influence actors behaviours. We report early results from the use of a speech processing system used within the storytelling application to influence the behaviour of the artificial actors.