Integrating bottom-up and top-down reasoning in COLAB

Abstract

The knowledge compilation laboratory COLAB integrates declarative knowledge representation formalisms, providing source-to-source and source-to-code compilers of various knowledge types. Its architecture separates taxonomical and assertional knowledge. The assertional component consists of a constraint system and a rule system, which supports bottom-up and top-down reasoning of Horn clauses. Two approaches for forward reasoning have been implemented. The first set-oriented approach uses a fixpoint computation. It allows top-down verification of selected premises. Goal-directed bottom-up reasoning is achieved by a magic-set transformation of the rules with respect to a goal. The second tuple-oriented approach reasons forward to derive the consequences of an explicitly given set of facts. This is achieved by a transformation of the rules to top-down executable Horn clauses. The paper gives an overview of the various forward reasoning approaches, their compilation into an abstract machine and their integration into the COLAB shell.