In the field of knowledge based system development, the content and structure of knowledge bases is more and more considered as a central topic of research. In this paper, we will argue that for several reasons, a dynamic evaluation is preferable over a static consideration of background knowledge in an implemented system. We introduce a statistical method designed for the evaluation of knowledge bases and present first results of a study made about the LEU/2 knowledge base of the LILOG text understanding system. We give examples of three kinds of improvement that a statistical tool can facilitate: optimizing knowledge base content, revising knowledge base structure, and hence achieving a better interaction of knowledge processing with other modules in the implemented system.
Evaluation of Knowledge Based Systems, LILOG, LEU/2
Ingo Glöckner and Gudrun Klose, Static and dynamic evaluation of knowledge bases. In Kai v. Luck, Hans-Joachim Novak, and Claus-Rainer Rollinger (eds.), Fully Implemented Natural Language Understanding Systems (Proceedings of the Workshop held at Trento, Italy, March 30, 1992), IWBS Report 236 (IBM TR-80.92-033), Institut für Wissensbasierte Systeme, Heidelberg, pp. 50-60.
If you want to receive a copy of the manuscript, just contact me by email.