Pennsylvania. It Was Supported by the Author While on a Teaching

The task of constructing knowledge bases is a difficult one due to their size and complexity. A useful aid for this task would be a system which has both knowledge about a particular knowledge representation scheme and tools with which to manipulate the representation’s components. Such a system would be a knowledge maniupulation sysThe remainder of the paper discusses taxonomic representation, TAXI’s design goals, its component tools, and possible applications. tern (KMS). This paper describes a KMS called TAXI which is used to manipulate knowledge in the form of a taxonomic knowledge representation scheme. The particular taxonomic representation used is discussed, along with support for the usefulness of a KMS for this particular representation scheme. Tools provided in the TAXI system are described, as are possible applications for the system. The taxonomic representation scheme was selected for severa1 reasons. Taxonornies are useful for organizing declarative knowledge, which is used in many knowledge based systems, particularly diagnostic systems. Constructing a taxonomy creates a classification process for a domain so that each domain object is uniquely defined by its attributes. A set of objects representing a domain are defined by associating a value with each of a set of attributes for each object. The domain is further subdivided into classes of objects which have the same Thii paper describes TAXI, a first version of a Knowlvalue for each of a given set of attributes. edge Manipulation System (KM!?) for taxonomic representation. A KMS is a system which contains knowledge about some representation scheme and its components, along with tools allowing users to manipulate and explore domain knowledge represented by that scheme. The final taxonomy will be the result of numerous decisions about attributes, values, application order of attributes, etc. Only by participating in these decisions will a user, whether human or machine, be fully aware of the implicit knowledge behind the final taxonomic structure. In order to most effectively utilize large amounts of Thus, the construction process will provide the construcdomain knowledge, it is essential that the knowledge be tor with a better understanding of the utilized domain. organized in some manner. In addition, the organization process enables a person to better understand the domain Except for already well defined and understood doand relationships among the knowledge being organized. mains, it is currently not possible to fully automate taxThe usefulness of this meta-knowledge and experimenonomy construction [Swartout, 19811. This is due to taxtation within a domain while trying to organize its inonomy construction being a meta-classification problem, formation is shown in the work of Swartout and Papert requiring that large amounts of domain knowledge be pos[Swartout, 19811 [Papert, 19791. sessed before construction begins. Automatic construction is also not practical since users will not obtain the better understanding of the domain from performing the construction. lThis work was done while the author was at the University of However, due to the needs of detecting generalities Pennsylvania. It was supported by the author while on a teaching among objects, keeping track of dependencies, determinor amistantship at all. aud by Penn via use of machines. No grants contracts 416 Knowledge Representation From: AAAI-87 Proceedings. Copyright ©1987, AAAI (www.aaai.org). All rights reserved.