A cognitively motivated architecture for an anthropomorphic artificial communicator
Nadine Leßmann, Ipke Wachsmuth · PUB – Publications at Bielefeld University (Bielefeld University) · 2003
This poster presents demonstrable work towards developing an architecture for an artificial conversational agent which aims to integrate communicative abilities with autonomous intelligent behavior. The context for our research is the Collaborative Research Center SFB 360 which is concerned with situated artificial communicators. Since situatedness involves embodiment, emphasis is laid on the agent's physical representation. The application scenario consists of a task-oriented, mixed-initiative discourse between an instructor and a constructor who collaboratively build aggregates, like a model aeroplane, from a wooden toykit (“Baufix”). Multimodal interaction including interpretation abilities as well as production competency plays a central role in the scenario. One realization exhibiting such abilities is an artificial anthropomorphic agent embodied in large-scale virtual reality, “Max” (cf. Figure 1). Max is capable of interpreting multimodal input comprising speech and gesture and of producing multimodal output including gestures, facial expressions, and speech. The agent’s movements are calculated on the basis of a kinematic skeleton as real-time computer animations (Kopp & Wachsmuth, 2000). Max has access to a knowledge base about possible connections of Baufix-parts and has additional knowledge on how to construct specific aggregates. The production of multimodal utterances and intelligent behavior on the one hand and the perception of the environment and interpretation of multimodal communicative acts on the other hand demand a complex interplay of sensory, cognitive and actoric abilities: • the agent should be able to perceive his environment and especially the user, interpreting perceived input in a cognitively motivated way • situated communication goes beyond a pure processing of instructions insofar as it requires dialog competences in understanding and generating context-dependent utterances • to approach this aim a discourse memory and a partner model have to be integrated • turn-taking is important in mixed-initiative dialog characterized by asynchrony, change of initiative, openness and unpredictability • the agent’s embodiment plays an important role in providing the possibility to use several channels to convey information