43 Keelung Road, Section 4, Taipei, TAIWAN 106 +Telecommunication Laboratories, Ministry of Transportation and Communications 9 Lane 74, Hsin-Yi Road, Section 4, Taipei, TAIWAN AbslraclThe knowledge acquisition process is technically challenging and time consumlng. Although many interactive knowledge acquisition tools are developed for overcoming the obstacle, these tools are domain dependent and difficult to apply to other domains. G A S is a primitives-based generic knowledge acquisltlon shell which allows users to easily construct and then execute a domainspecific knowledge acquisition tool for specific problem domains.It contains problem solving primitives, acquisition primitives, interaction primitives, representation schemas, learning primitives, and knowledge processing primitives in order to construct specific knowledge acquisition tools efficiently.This approach relieves knowledge engineers from selecting and applying different knowledge acquisition tools to different domains. We have also developed G A S as an open architecture so that further enhancement can be done easily.
INIRODUCI'IONThe process of knowledge acquisition (KA) is technically challenging and time consuming owing to many difficulties [17]. These difficulties restate the fact that knowledge acquisition is one of the major bottlenecks in the development of expert systems. Basically three approaches, traditional psychological methodologies, interactive KA tools, and machine learning methodr are taken in the research of knowledge acquisition. This interactive KA tools approach is interesting because it could make the psychological approach more productive by realizing proper psychological theories into specific KA tools, such as ETS [4]. It also facilitates the development of a framework that involves techniques from both learning and KA tools approaches so that sources for knowledge acquisition can be broadened, such as KRITON [ 141. Many KA tools have been developed under a rather unified idea that a KA tool could be developed based upon the analyses of domain knowledge, domain problem solving methods, and their inter-play since 1983. Examples include KNACK [21], MOLE [15], MORE [18], ROGET [31, S A L T [24], SIZZLE [29], etc.. Contrast to these KA tools which focus on specific domains for knowledge acquisition, some of the other KA tools try to do knowledge acquisition by integrating a variety of methodologies (or utilities) into a single workbench. Examples include AQUINAS [5], KADS 1301, KREME [l], TDE [19], and TEIRESIA [13].Although the aforementioned KA tools have made some progress in solving the knowledge acquisition bottleneck, several problems remain. First, many of the tools are too domain-specific and difficult to be applied (even after revision) to other domains. Developers need to redesign or refine these tools before applying to other domains. Secondly, integration of knowledge generated by the KA tools with other sources is difficult (in fact, impossible most of times), because of a single knowledge representation being u...
