Constraints in a Hybrid Knowledge Representation System

In our research group, the hybrid knowledge representation system Babylon has been developed providing formalisms for rules, prolog and frames. Beyond it, we implemented Consat, a system for constraint satisfaction. Since applications of Babylon for process diagnosis, planning etc. required constraints. we integrated Consat into the Babylon environment. The paper describes the integration of Consat into Babylon, regarding two aspects. First, constraints should be available as another Babylon formalism by using the functional interface of Consat. On the other hand, it is important to have constraints implicitly controlling other Babylon formalisms, for instance, in order to keep the system's database consistent. While with respect to the first point, the paper describes work already finished, the second form of integration is work in progress. 1. Introduction Constraint networks are evolving as another style of knowledge representation beside rules, logic, frames, etc. in many domains such as qualitative reasoning and planning. A hybrid knowledge representation system called Babylon has been developed in our group at GMD Sankt Augustin [Forschungsgruppe Expertensysteme 1985], [diPrimio, Brewka 1985], that-so far-provides rules, prolog, and frames. However, Babylon is open to integrate any other formalisms which can be achieved by simple extensions of the system. Currently, in our group a planning system, a knowledge acquisition system [Diederich et al. 1986], and a system for process diagnosis are being developed, all of which depend on the availability of some constraint mechanism (cf. Molgen [Stefik 1981] as a planning system using constraints, and [Davis 1984] and [deKleer, Williams 1986] as diagnosis systems using constraints). Thus, there is an urgent need to extend Babylon to a constraint formalism. Independently from Babylon but with the purpose of later integration, a constraint system called Consat [Guesgen 1986] has been developed. Consat allows to define hierarchical networks of domain independent constraints. The variables in a network can be assigned sets of values that are propagated to compute the maximal locally consistent assignment, the globally consistent assignments, or to detect an inconsistency in the original assignment.