The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge

Research Paper #473

Title:Understandable Explanations: the Fudge Discourse Generator
Date:May 1990
Presented:SUbmitted to the Proceedings of the 5th UK Explanation Workshop (Ed. N. Filer) 1990
Abstract:This paper develops two principles for generating understandable explanations, and presents a simple computational system based on those principles. First, explanations should be planned so that, in principle, the concepts in them may be understood by the particular hearer. This involves making use of both general learning/teaching methods (such as by analogy or by example) and information about how different types of knowledge (such as procedures, objects or processes) may be described using these. Then, giving assumptions about the user's knowledge it should be possible to generate understandable explanations in a fairly principled manner. Second, our model of the hearer's knowledge is very unlikely to be correct (though it may be a useful approximation). It is therefore vital to allow the user to signal whether or not they are following, and to ask clarification questions as necessary. In a complex explanation these may take place within the explanation, and influence the progress of the explanation. These principles motivated a revised version of the EDGE explanatory discourse generator [Cawsey 89, Cawsey 90], which we will call FUDGE (Fairly Understandable Discourse GEnerator). This paper introduces the FUDGE system.

[Search These Pages] [DAI Home Page] [Comment]