Knowledge Level Engineering Ontological Analysis

Knowledge engineering suffers from a lack of formal tools for understanding domains of interest. Current practice relies on an intuitive, informal approach for collecting expert knowledge and formulating it into a representation scheme adequate for symbolic processing. Implicit in this process, the knowledge engineer formulates a model of the domain, and creates formal data structures (knowledge base) and procedures (inference engine) to solve the task at hand. Newell (1982) has proposed that there should be a knowledge level analysis to aid the development of AI systems in general and knowledge-based expert systems in particular. This paper describes a methodology, called ontological analysis, which provides this level of analysis. The methodology consists of an analysis tool and its principles of use that result in a formal specification of the knowledge elements in a task domain, 1. Knowledge Engineering needs a methodology. Traditionally, knowledge engineering has been a difficult process. Neophyte knowledge engineers often “don’t know where to start.” The difficulty in getting started is related to confusions over how to encode or classify relevant knowledge items from the task domain. Clancey (1985) provides a typical example from MYCIN: Perhaps one of the most perplexing difficulties we encounter is distinguishing between subtype and cause, and between state and process . . . For example, a physician might speak of a brain-tumor as a kind of brainmass lesion. It is certainly a kind of brain-mass, but it causes a lesion (cut); it is not a kind of lesion. Thus, the concept bundles cause with effect and location: a lesion in the brain caused by a mass of some kind is a brainmass-lesion. (pg. 3 11) This experience is familiar to any knowledge engineer. Misunderstandings about the knowledge elements in a system often pervade mature systems and cause endless problems, In response to this problem Newell (1982) has suggested that there should be a knowledge level analysis of domains which would guide knowledge-based systems development. In this paper we discuss a methodology for analyzing problem domains we call ontological analysis. Most problems encountered in knowledge-based systems derive from ad hoc design of the knowledge structures. Often, knowledge is collected by writing rules or frames in a language-specific syntax, without a systematic consideration of the underlying structure of knowledge elements. Ontological analysis focuses attention on the elements of knowledge in their own right, independent of implementation techniques. An ontological analysis is distinctly different from knowledge representation languages in that it presents only a high level description of a problem’s knowledge structure. Ontological analysis is used to identify and construct an adequate knowledge representation for a problem. 2. Ontological Analysis. To philosophers, ontology is the branch of metaphysics concerned with the nature of existence, and the cataloguing of existent entities (Quine, 1980). The role of ontology in AI has been noted previously (Hayes, 1985; Hobbs, 1985; McCarthy, J., 1980). We use the term to emphasize that a knowledge-based system is best designed by careful attention to the step-by-step composition of knowledge structures. An ontology is a collection of abstract objects, relationships and transformations that represent the physical and cognitive entities necessary for accomplishing some task, Our experience indicates that complex ontologies are most easily constructed in a three step process that concentrates first on the (static) physical objects and relationships, then on the (dynamic) operations that can change the task world, and finally on the (epistemic) knowledge structures that guide the selection and use of these operations. Any useful methodology must contain both formal tools for constructing an analysis, and informal principles of practice to guide application of the formal tools. Our research has indicated that several different formal tools are useful for extracting and defining ontologies. We are developing a family of languages collectively called SPOONS (Specification Of ONtological Structure) to encompass tools based respectively on domain equations, equational logic, and semantic grammars. 2.1 Domain Equations in Ontological Analysis The most useful and concise of these languages is SUPESPOONS (SUPErstructure SPOONS), which is based on the domain equations of denotational semantics (Gordon 1979; Stoy 1977) and algebraic specification (Guttag and Horning, 1980). Because of the rich ontologies found in most knowledge engineering problems, domain equations provide a concise and reasonably abstract characterization of the necessary knowledge structures. Furthermore, they do not encourage the knowledge engineer to get prematurely involved in details. 2.1.1 SUPE-SPOONS Syntax and Semantics SUPE-SPOONS consists of two basic statement types: l Domain equations: Site = Building x Campus. These statements define domains, or types of knowledge structures.* l Domain element declarations: add-meeting: Meeting + [Meetings + Meetings] These statements declare the type of specific domain elements. The right hand side of statements can be composed of one or more domains or constant elements with operators relating these elements, Four primitive domains, STRING, BOOLEAN, INTEGER and NUMBER, are always assumed to be defined. Other primitive domains can be defined by explicit enumeration of their elements, or by open assignment to some collection of atomic elements. *For most purposes, it suffices to think of domains as sets. A more complex semantics is needed if domains are defined recursively (Stoy 1977) or with multiple equations. KNOWLEDGE ACQUISITION I 963 From: AAAI-86 Proceedings. Copyright ©1986, AAAI (www.aaai.org). All rights reserved.