HAM-RPM: Natural Dialogues with an Artificial Partner

This paper introduces the understanding system HAM-RPM, which simulates a dialogue partner conversing about a real-life domain. After outlining the system's overall structure, we discuss three of its distinguishing features: The first is its organisation of spatial data in a redundant multiple data base, inspired by certain aspects characteristic of visual search in humans. A new algorithm for noun-phrase generation is then sketched which is sensitive to the conversational state and uses a 'worst-case-first' strategy. Finally, we describe in some detail a specific operationalisation of the notion of the communicative relevance of objects. The paper concludes with a summary of the objectives of this research. DESCRIPTIVE TERMS: natural language processing, dialogue simulation, focus of attention, reference semantics, visual search, FUZZY