Learning of context-sensitive language acceptors through regular inference and constrained induction

Recently, Augmented Regular Expressions (AREs) have been proposed as a formalism to describe and recognize a non-trivial class of context-sensitive languages (CSLs), that covers planar shapes with symmetries 1, 2]. AREs augment the expressive power of Regular Expressions (REs) by including a set of constraints, that involve the number of instances in a string of the operands of the star operations of an RE. A general method to infer AREs from string examples has been reported 2] that is based on a regular grammatical inference (RGI) step followed by a constraint induction process. This approach avoids the diiculty of learning context-sensitive grammars. In this paper, a speciic method for learning AREs from positive examples is described, in which the RGI step is carried out by training a recurrent neural network for a prediction task 3] and extracting a DFA from the network dynamics 4]. The ARE learning method has been applied to the inference of a set of eight test CSLs, and good experimental results have been obtained.