Candide: An Interactive System for the Acquisition of Domain Specific Knowledge

摘要

How is the knowledge that is embodied in a computer system acquired. In most present-day systems, it is painstakingly encoded in the actual algorithms that implement the system. Even in most artificial intelligence systems, which may have a level of explicit representation of the knowledge, this knowledge must be entered into the system as expressions in some formal computer language. Usually this must be done by a specialized programmer or knowledge engineer. The result is a knowledge bottleneck. Significant efficiency could be gained by making it possible for the needed knowledge to be installed by those who are knowledgeable about the domain of some computer system, but who do not necessarily possess expertise about the internal workings or language of the system itself. The Candide project, whose results are described in this report, has been concerned with developing tools that will help to alleviate the knowledge bottleneck problem. Under its auspices, the Candide system was designed and implemented. Candide is a multimodal interpretation system designed specifically for knowledge acquisition tasks. Candide makes it possible for someone who is knowledgeable about the domain of some computer system to create, add to, or update the system's knowledge base using a combination of ordinary English discourse and a simple graphical interface tool. Keywords: Passing; Syntax; Phonology; Computers; Semantics; User manuals; PRS(Procedural Reasoning System). (kr)