CSRL: A Language for Expert Systems for Diagnosis

Abstract

We present CSRL (Conceptual Structures Representation Language) as a language to facilitate the development of expert diagnosis systems based on a paradigm of “cooperating diagnostic specialists.” In our approach diagnostic reasoning is one of several generic tasks, each of which calls for a particular organizational and problem-solving structure. A diagnostic structure is composed of a collection of specialists, each of which corresponds to a potential hypothesis about the current case. They are organized as a classification or diagnostic hierarchy, e.g. a classification of diseases. A top-down strategy called establish—refine is used, in which either a specialist establishes and then refines itself, or the specialist rejects itself, pruning the hierarchy that it heads. CSRL is a language for representing the specialists of a diagnostic hierarchy and the diagnostic knowledge within them. The diagnostic knowledge is encoded at various levels of abstractions: message procedures, which describe the specialist's behavior in response to messages from other specialists; knowledge groups, which determine how data relate to features of the hypothesis; and rule-like knowledge, which is contained within knowledge groups.