This research will develop a flexible knowledge representation medium and sufficient natural language processing tools to allow the automated extension of an expert system's knowledge base through access to machine-readable natural language texts such as dictionaries, thesauri, and technical references. The work on an early prototype has used the Merck manual as a source ?Merck 1977!. Initial work will include the building of a parsing system consisting of a core grammar and the means to expand the core grammar semi-automatically. Using this parsing system, the effort will develop the incremental augmentation of the knowledge system directly from the analysis of the text. The ability of an expert system to extend its own knowledge base by extracting information from machine readable text is valuable for a wide variety of systems, including expert systems for large technical applications in a domain in which the information is continually being updated. Other projects rely on human intervention during the process that extracts new knowledge ?Lenat et al. 1986; Humphrey 1989; Antonacci et al. 1989!. This makes dealing with problems such as contradictory information much easier. In contrast, here the feasibility of automating the actual extraction of information, in a disciplined way, from a structured technical reference will be examined.
