Mixed Initiative Dialog using Goal Tagged Activities

Peter Wallis

Institution

Date and time: 11.30am-12.30pm, Friday 2nd July, 2004

Venue: 10.11.04

Chair: Michael Winikoff

Abstract:

Agency has highlighted the situated and autonomous nature of many kinds of software. The idea of an agent reasoning about its effect on its environment is relatively old in the field of natural language processing (NLP) with the notion of "speech acts" being introduced by Searle in the early seventies(Searle 69). The study of language however has a very long history and new ideas are often embraced, extended and extinguished well before they can take root. An expert in NLP is thus not necessarily going to ``get the point'' of a paper that talks about goals in a social context for example, nor might they think of Embodied Conversational Agents as anything but toys. This paper starts by explicitly contrasting the agency model of dialog with other approaches to natural language processing, and arguing for BDI as the framework. It then presents a variant on classic BDI which uses Goal Tagged Activities in place of plans to do things in the agent's environment. The contribution of this paper is to point out that the BDI algorithm can be "unwrapped" to provide a state-space description of the agents plan library. Using BDI style planning over activities might be the ultimate aim of those engineering dialog systems, but in the mean time, a variation on the classic chat-bot mechanism is a serviceable research tool for those interested in the intentional structure of dialog.

About the speaker:

Seminars are free and open to the general public. No booking is necessary.

If you are interested in giving a presentation in this seminar series, or to make suggestions for speakers, please contact James Harland, the seminar co-ordinator.