Data Complexity Issues in Grammatical Inference

Abstract

Grammatical inference (also known as grammar induction) is a field transversal to a number of research areas including machine learning, formal language theory, syntactic and structural pattern recognition, computational linguistics, computational biology, and speech recognition. Specificities of the problems that are studied include those related to data complexity. We argue that there are three levels at which data complexity for grammatical inference can be studied: at the first (inner) level the data can be strings, trees, or graphs; these are nontrivial objects on which topologies may not always be easy to manage. A second level is concerned with the classes and the representations of the classes used for classification; formal language theory provides us with an elegant setting based on rewriting systems and recursivity, but which is not easy to work with for classification or learning tasks. The combinatoric problems usually attached to these tasks prove to be indeed difficult. The third level relates the objects to the classes. Membership may be problematic, and this is even more the case when approximations (of the strings or the languages) are used, for instance in a noisy setting. We argue that the main difficulties arise from the fact that the structural definitions of the languages and the topological measures do not match.