Flexible Pattern Recognizers are also Concept Formers

Abstract

The typical pattern recognizer (PR) applies a set of characterizers to an input. Each characterizer implies a set of possible names, and the single most highly implied name is chosen. The typical concept former (CF) applies a binary test to the input. This test implies either another test to apply, or a name to output. Pattern recognizers have almost always applied probabilistic (usually two-valued but occasionally multi- valued) characterizers in parallel; whereas concept formers have always applied deterministic two-valued tests, in series. This paper presents and examines simple computer programs (coded in EASEy-2, a language that is relatively easy to understand) for (1)parallel "pattern recognition" (NAMER), and (2)serial "concept formation" (CONCEIVER). Finally, these programs are generalized to give a single " flexible" pattern recognizer-concept former (FLEXIBLE PR-CF) that combines the desirable features of parallel-probabilistic and serial-deterministic systems.`