This article introduces a mechanism to compare two conceptual graphs and produce a similarity measure, reflecting incomplete, imprecise, or inconsistent information. It is then shown that conceptual graphs, in conjunction with the similarity measure, can be effectively used to define concepts by means of a series of examples. The descriptions of these examples provide the basis for a recognition system. Support logic programming is also introduced and it is demonstrated that conceptual graphs, together with the comparison metric, provide a natural knowledge representation framework. 0 1993 John Wiley & Sons, Inc.
I. OUTLINE OF REQUIREMENTS
Within the field of knowledge representation, the facility to make comparisons between pieces of information is a common requirement. In many situations that arise in real-world problems, the available information may be inaccurate, incomplete, or inconsistent. Thus, any measure of similarity, intended to reflect the result of comparing pieces of information, must be designed with the flexibility to handle such knowledge.'
Examples of applications in which a similarity measure would be essential are as follows: 0 vision analysis-the nature of a certain object in view must be determined by comparing it to descriptions of existing objects. Although the descriptions of the existing objects are complete and accurate, the object being observed may be only partially visible, or it may be indistinct. classification-scientists may discover a "new" creature and wish to classify it with respect to all previously known creatures. 0 selection-attempting to determine whether a possible future employee would be suited to the job for which he or she has applied.
Conceptual graphs provide a very flexible, and powerful means of representing knowledge.*.' To demonstrate the ways in which the uncertainty requirements, presented above, may be modeled by conceptual graphs, the following