Distributional Learning of Some Context-Free Languages with a Minimally Adequate Teacher

Angluin showed that the class of regular languages could be learned from a Minimally Adequate Teacher (mat) providing membership and equivalence queries. Clark and Eyraud (2007) showed that some context free grammars can be identified in the limit from positive data alone by identifying the congruence classes of the language. In this paper we consider learnability of context free languages using a mat. We show that there is a natural class of context free languages, that includes the class of regular languages, that can be polynomially learned from a mat, using an algorithm that is an extension of Angluin's lstar algorithm.

• Type: Conference
• Year: 2010
• Where: ICGI
• Authors: Alexander Clark
• Keywords: Context Free Grammars, Context Free Languages, ICGI 2010, Natural Language Processing, Regular Languages