Search strategy and selection function for an inferential relational system

Abstract

An inferential relational system is one in which data in the system consists of both explicit facts and general axioms (or “views”). The general axioms are used together with the explicit facts to derive the facts that are implicit (virtual relations) within the system. A top-down algorithm, as used in artificial intelligence work, is described to develop inferences within the system. The top-down approach starts with the query, a conjunction of relations, to be answered. Either a relational fact solves a given relation in a conjunct, or the relation is replaced by a conjunct of relations which must be solved to solve the given relation. The approach requires that one and only one relation in a conjunction be replaced (or expanded) by the given facts and general axioms. The decision to expand only a single relation is termed a selection function. It is shown for relational systems that such a restriction still guarantees that a solution to the problem will be found if one exists. The algorithm provides for heuristic direction in the search process. Experimental results are presented which illustrate the techniques. A bookkeeping mechanism is described which permits one to know when subproblems are solved. It further facilitates the outputting of reasons for the deductively found answer in a coherent fashion.