Reference: Fikes, R. & Rice, J. The Stanford KSL Knowledge Base Merging Critical Component Experiment. Knowledge Systems Laboratory, October, 1999.
Abstract: Large-scale knowledge bases (KBs) are an essential enabling component of the next generation of intelligent systems. The high cost of producing KBs has motivated the development of technology and methods for generating reusable KB modules by multiple authors, maintaining those modules in knowledge libraries, and producing KBs for specific applications by assembling and extending modules from those libraries. This methodology for building KBs requires that KB modules produced by independent authors containing overlapping content using differing representations and vocabularies be reconciled (i.e., "merged") so that those modules can be used as compatible KB building blocks. Although KB merging can be arbitrarily difficult, software tools can provide substantial help with major steps in the process. In this paper, we present experimental results showing the benefits of using Chimæra, a new software tool designed to aid with the merging of taxonomies, which is a substantial portion of the overall KB merging process. The experimental evidence we cite shows that Chimaera is a significant improvement over general-purpose KB editing tools and text editing tools.