MINID - AN EXPERIMENT IN BUILDING KNOWLEDGE BASES WITH TEXTUAL AND NON-TEXTUAL INFORMATION.

Summary form only given. The authors discuss incorporation of nontextual information, such as visual images, speech, sound, etc. into knowledge bases. Some issues covered include: (1) development of an expert system that supports multiple-media; (2) how to use multiple-media in the user interface; (3) development of selection criteria for choosing information types, and (4) development of empirical tests of multimedia expert systems. As an initial test of some of these issues, the authors have built MINID, an expert system for mineral identification that uses both text and images in the user interface. MINID was built using AVIEN, a backwards-chaining inference engine implemented as a C function library, callable from C programs. AVIEN allows expert system developers to provide explanations of both reasoning and questions put to a user. These explanations can be text, a path to a text file, a path to an executable file, or a combination of these. MINID makes heavy use of both textual explanations and video images. Results of a classroom implementation of MINID are given, as are additional features to be implemented.