Efficient Language Learning with Correction Queries

We investigate two of the language classes intensively stud ied by the algorithmic learning theory community in the context of learning wit h correction queries. More precisely, we show that any pattern language can be infe rred in polynomial time in length of the pattern by asking just a linear numb er of correction queries, and that k-reversible languages are efficiently learnable within thi s setting. Note that although the class of all pattern languages i s learnable with membership queries, this cannot be done in polynomial time. Mor eover, the class of k-reversible languages is not learnable at all using member ship queries only. Furthermore, we present results on a newly introduced class of l anguages, namely the class of injective languages.