PROLOG META-INTERPRETERS FOR RULE-BASED INFERENCE UNDER UNCERTAINTY

Abstract

Uncertain facts and inexact rules can be represented and processed in standard Prolog through meta-interpretation. This requires the specification of appropriate parsers and belief calculi. We present a meta-interpreter that takes a rule-based belief calculus as an external variable. The certainty-factors calculus and a heuristic Bayesian belief-update model are then implemented as stand-alone Prolog predicates. These, in turn, are bound to the meta-interpreter environment through second-order programming. The resulting system is a powerful experimental tool which enables inquiry into the impact of various designs of belief calculi on the external validity of expert systems. The paper also demonstrates the (well-known) role of Prolog meta-interpreters in building expert system shells.