🚀 go-pugleaf

			    ITK-COLLOQUIUM

			   ROBBERT-JAN BEUN
				(IPO)


    A FRAMEWORK FOR SPEECH ACT GENERATION IN COOPERATIVE DIALOGUE


  A central theme in Artificial Intelligence approaches to speech acts
is the modeling of the mental states changes of the participants in
terms of beliefs and intentions. Although this so called
`context-change' approach seems to offer an attractive formal
treatment of the mental state changes of dialogue participants, the
approach lacks the power to model the generation of speech acts in
conversation. Generation of a speech act is of crucial importance to
determine, for instance, adequate behavior of an information exchange
system in response to a user's primary communicative goals.

  To determine adequate communicative behavior of participants (human or
machine) in a cooperative dialogue, a theoretical framework will be
presented that is based on the concepts of `preference organization',
`dialogue game' and `mental state(-change)' of the participants. Main
goal of the dialogue is to exchange information in a collaborative
manner so that, if the information is available, the participants
mutually believe in the final state the answer to an initial question.
The resulting structure of the information exchange can be complex,
since knowledge to find the answer to the initial question may be
distributed among the participants and may therefore result in
subdialogues an the generation of counter-questions.

  It will be shown that the application of different communicative
strategies and different initial belief states gives important insights
in the conversational structure of the dialogue. Moreover it is
claimed that, if the type of the dialogue is well-restricted, only
three types of belief are significant to determine adequate behavior
of an information exchange system in a cooperative situation: (i) the
system's belief about the domain of discourse, (ii) the system's
belief about mutual belief of the system and the user and (iii) the
system's belief about what the user doesn't believe.

  Robbert-Jan Beun has studied Electrical Engineering at the University
of Eindhoven. In 1989 he defended his thesis called `the recognition
of communicative functions in information dialogues'. At present he is
co-coordinator of the `Dialogue management and knowledge acquisition'
(DenK), a joint research project of the universities of Eindhoven and
Tilburg. 

	       Thursday, Nov. 19, 1992, 16.00h - 17.00h

    Institute for Language Technology and Artificial Intelligence
		   Warandelaan 2, 5037 GC, Tilburg
			     Room: X 6