A Characterization of the Language Classes Learnable with Correction Queries

Abstract

Formal language learning models have been widely investigated in the last four decades. But it was not until recently that the model of learning from corrections was introduced. The aim of this paper is to make a further step towards the understanding of the classes of languages learnable with correction queries. We characterize these classes in terms of triples of definite finite tell-tales. This result allowed us to show that learning with correction queries is strictly more powerful than learning with membership queries, but weaker than the model of learning in the limit from positive data.