Catalogue of Artificial Intelligence Techniques
Keywords: learning, semantic evaluation
Author(s): Maarten van Someren
A method to discover and correct faulty hypotheses in a theory. If a proposition that is derived from a set of hypotheses turns out to be false in terms of a model, the derivation can be used to identify the faulty hypotheses. Contradiction backtracing uses the trace of a Resolution proof, in which the false proposition was the goal. The proof is traced backwards from the empty clause. Each clause is semantically evaluated by the user or in a standard model. If it is false, the negated parent clause is considered next, and otherwise the positive parent. The substitutions that allowed the resolution steps are accumulated and applied to each following clause. The procedure will finally lead to a hypothesis which is false.
- Shapiro, E.Y., An algorithm that infers theories from facts, Proceedings of IJCAI-81, vol.1, 1981, pp.446--451.