User-Oriented Explanation for Expert Systems

McKeown, Kathleen

In this project we are concerned with extending the range and effectiveness of explanation facilities currently available in knowledge-based expert systems by incorporating the full power of natural language techniques. As expert systems become larger and more sophisticated, it is important that they be able to communicate their reasoning effectively to their users, whether they be naive users, domain experts, or system maintainers, thus allowing them to evaluate advice the system provides, one main problem with current explanation facilities is the inability to adequately tailor an explanation to a particular user of the system. In this project, we are examining the effects of the user and discourse situation on the type of explanation provided. In particular, we are interested In developing capabilities for providing re-explanations when the user is dissatisfied with an explanation, for providing explanations in an appropriate level of detail for a given user, for adapting the explanation to the point of view the user has taken, and for using a persuasive strategy that the user is likely to accept. These efforts require the development of a semantically rich representation of the domain, capable of supporting abstractions and different points of view; the investigation of strategies that are effectively used for justification and causal explanation; and a development of the influences on level of detail, strategy selected, and point of view. We expect the results of this research to be both practical and theoretical they will improve- the range of man-machine communication currently available, making expert systems more accessible to their users, and will improve our understanding of how appropriate explanations can be produced.



More About This Work

Academic Units
Computer Science
Department of Computer Science, Columbia University
Columbia University Computer Science Technical Reports, CUCS-085-83
Published Here
October 26, 2011