Grammatical Inference by specialization as a state splitting strategy

Abstract

We exhibit connexions between two already known learning algorithms developed in different backgrounds. This allows to show that learning classical (or AB) categorial grammars by specialization can be identified with a “state splitting” strategy, in a search space made of extended automata. It also leads to a new interpretation of why it is possible to learn categorial grammars from semantically typed (in Montague's sense) examples.