A Note on the Relationship between Different Types of Correction Queries

Abstract

The adult-child interaction which takes place during the child’s language acquisition process has been the inspiration for Angluin’s teacher-learner model [1], the forerunner of today’s active learning field. But the initial types of queries have some drawbacks: equivalence queries are both unrealistic and computationally costly; membership queries, on the other hand, are not informative enough, not being able to capture the feedback received by the child when he or she makes mistakes. This is why a new type of query (called correction query), weaker than the first one and more informative than the second, appeared. While in the case of natural languages it is well understood what correcting means, in formal language theory different objects may require different types of corrections. Therefore, several types of correction queries have been introduced so far. In this paper we investigate the relations existing between different models of correction queries, as well as their connection to other well-known Gold-style and query learning models. The study comprises results obtained in the general case when time complexity issues are ignored, and in the restricted case when efficiency constraints are imposed.Keywordsquery learningGold-style learningcorrection query