Forward logic evaluation : developing a compiler from a partially evaluated meta interpreter

Abstract

Pure horn logic does not prescribe any inference strategy. Clauses could be applied in forward and backward direction. This paper presents a translation of rules into forward clauses which simulate a forward chaining deduction if executed by Prolog's resolution procedure. Premises of forward rules are verified by Prolog's backward proof procedure using the original clauses. Thus, without any changes to the Prolog interpreter integrated bidirectional reasoning of horn rules is possible. The translation is obtained from a meta interpreter for forward reasoning written in horn logic. Data-driven partial evaluation of this meta interpreter wrt the original horn clauses results in a forward program. The approach is applied to the problem of recognizing production-specific features from a product model. A product model contains geometrical, topological, and technological information collected during the design phase. From these data features giving valuable hints about manufacturing are derived.