The access methods for a knowledge base provide ways to locate information. Although they are critically important for virtually all applications of a knowledge base, current methods are inadequate, especially for large, structured knowledge bases. To locate information about a concept using current access methods, the user must provide the address (usually a frame name) of the concept within the knowledge base, which requires an unrealistic level of omniscience. Moreover, only those concepts reified in the knowledge base can be located, which excludes information about the many concepts that are implicit in the knowledge base. Our solution to these problems is to provide, through our access methods, an abstraction of the knowledge base, one in which concepts can be located by a partial description of their contents and in which implicit concepts are automatically reified when they are requested. After locating a concept, our access methods provide an additional service: selecting coherent subsets of facts about the concept. Conventional methods either return all the facts about the concept or select a single fact (usually the filler of a specified frame-slot). Our access methods extract viewpoints --- coherent collections of facts that describe a concept from a particular perspective. We have identified many types of viewpoints and developed methods for extracting them from knowledge bases, either singly or in combinations. Our evaluation indicates that viewpoints extracted by our methods are comparable in coherence to those people construct

PS