Consistent identification in the limit of any of the classes k-valued is NP-hard

Abstract

In [Bus87], [BP90] 'discovery procedures' for CCGs were defined that accept a sequence of structures as input and yield a set of grammars. In [Kan98] it was shown that some of the classes based on these procedures are learnable (in the technical sense of [Gol67]). In [CF00] it was shown that learning some of these classes by means of a consistent learning function is NP-hard. The complexity of learning classes from one particular family, Gk-valued, was still left open. In this paper it is shown that learning any (except one) class from this family by means of a consistent learning function is NP-hard as well.