As described by Lochbaum, Grosz and Sidner, approaches to computer generated conversation fall into one of two broad categories: the intentional model, in which dialogue structure is analysed from the perspective of user goals, and the more conventional informational model, in which the purpose of language is to convey information. This presentation starts with arguments for taking the intentional perspective and goes on to show how it is applied to developing descriptions of dialogue structure that explicitly address factors commonly grouped under the banner of social intelligence. The motivation for the work described has been the so called agent based approach to AI in which the focus is on the situated and autonomous nature of some software entities. Rather than focusing on what was said, the focus should be on what to say next in order to achieve goals. This line of investigation has highlighted the need for conversational computers to understand social conventions. What is the effect on a human if the machine follows or breaks these conversational norms? The talk is intended for both practitioners currently working with applied language technology such as chat bots or automated call handling, and for graduate students looking for potential research topics in what is traditionally called conversation analysis.
Peter Wallis has a bachelor of arts from Flinders University with majors in Computer Science and Philosophy, and a Ph.D. from RMIT in semantics for search engines. He has a long history of working on applied natural language: In his Ph.D. work, he used the Longman's Dictionary of Contemporary English (LDOCE) to produce canonical versions of text meaning. He then worked at Defence Science and Technology Organization on information extraction, and initiated the "fact extractor" architecture that has gone operational at several sites within Defence. Since 1998 he has been developing an interest in dialogue and is currently trying to commercialize some ideas on dialogue management for the VoiceXML community. Key publications have been on the evaluation of Language Technology, and when pressed he will argue for a functional view of semantics. [http://www.cs.mu.oz.au/~peter/]