Autotically Synthesized Term Denotation Predicates: A Proof Aid

In goal-directed proofs, such as those used in HOL, tactics often must operate on one or more speciic assumptions. But goals often have many assumptions. Currently there is no good way to select or denote assumptions in HOL88. Most mechanisms are sensitive to inconsequential changes or are diicult to use. Denoting assumptions by lters (matching) makes it easier to maintain large proofs and reuse pieces. But writing the lter predicate can be time-consuming and distracting. We describe an aid to proof building which synthesizes lter functions from terms. Given examples of terms which should and should not be matched, the function creates an ML predicate which can be used, for example, with filter or FILTER_ASM_REWRITE_TAC. This paper reviews past discussions on denotation methods, the design and implementation of the lter synthesizer, applicable AI classiication techniques, and possible application to more general term handling and recognition.