The semantics of KBSSF, a language for KBS design

Abstract

This article exemplifies the semantics of KBSSF, a formal specification language for the design of knowledge based systems. KBSSF has been developed within the context of the ESPRIT II project VITAL—a methodology based workbench for KBS development. A formal specification language shows its value especially in the validation and verification of knowledge based systems. In VITAL this validation and verification also has its impact on the knowledge acquisition, because KBSSF is closely related to graphical primitives representing the conceptual model of the domain. To perform validation and verification, tool support is inevitable. The definition of the language semantics is on its turn a prerequisite for the development of such tools. The KBSSF language combines dynamic knowledge and static knowledge. The dynamic specification, i.e. problem solving behaviour, is specified by means of an imperative language with so called inference operators. The static knowledge specification is composed of logical theories which are a combination of Abstract Data Types and order sorted predicate logic. The inference operators form the link between the dynamic knowledge and the static knowledge, they infer new knowledge from the logic theories.The semantics of KBSSF reflects the distinction in dynamic knowledge and static knowledge. The semantics of the behaviour is described in terms of the Plotkin formalism (Plotkin, 1981), a formalism which describes the semantics in terms of transitions between configurations (machine states). The semantics of the logical theories is described in terms of Herbrand models, which are tailored for order sorted logic with equations, restricted to order sorted Horn clauses. The inference operators are given meaning in terms of these Herbrand models.The coupling between the two semantics is achieved by describing the evaluation of the inference operators as one step transitions in the Plotkin formalism, with the meaning in terms of Herbrand models attached to them.Thus, we show that two different semantic description techniques can be combined in an elegant way, which implies that KBSSF can be given a meaningful and pragmatic semantics, for which support tools can be developed.