page 2  (37 pages)
to previous section1
3to next section

Expanding the text here will generate a large amount of data for your browser to display

1 Introduction

A number of knowledge engineering methodologies are based on the notion of reusable components, which are used to build models of problem solving (e.g. the Generic Task approach [Chandrasekaran, 1986], the Method{to{Task approach [Musen et al., 1988], the Role{limiting Method approach [McDermott, 1988; Klinker et al., 1991], the Components of Expertise approach [Steels, 1990]). A central issue is how these reusable components are defined.

In this paper we discuss some of the problems with the current definition of reusable components, and we outline a possible solution in the context of the KADS methodology for constructing knowledge based systems. The basic reusable components in this methodology are the inferences or knowledge sources. The problems can roughly be divided into two categories: firstly the ambiguity of the names and descriptions of the inferences, and secondly the fact that the domain ontology on which the inferences are defined is too general for specific tasks. As a solution to the first problem, we propose to formalize inferences in first order logic, making all their assumptions and properties explicit, clarifying relations between inferences, and revealing new inferences. To deal with the second problem, we show how inferences based on other, task{specific ontologies, can be described and related to the inferences based on the KADS domain ontology.

This paper is structured as follows: the next section discusses the concept of reusable components in various approaches to knowledge acquisition, and outlines the main problems with these components. Section 3 gives a concise overview of those aspects of KADS which are relevant to this paper, and discusses how the problems outlined in section 2 reveal themselves within the KADS approach . Section 4 presents a solution to the problems presented in section 2, by means of formalization of the set of inferences. This section presents the formal language, outlines the choices that are to be made when formalizing inferences, describes how task{specific inferences can be described in the formalization, and how they relate to the ontology{specific inferences and gives example formalizations. Section 5 outlines how the formalization approach solves the problems sketched in the