Search diversification techniques for grammatical inference

Abstract

Grammatical Inference (GI) addresses the problem of learning a grammar G, from a finite set of strings generated by G. By using GI techniques we want to be able to learn relations between syntactically structured sequences. This process of inferring the target grammar G can easily be posed as a search problem through a lattice of possible solutions. The vast majority of research being carried out in this area focuses on non-monotonic searches, i.e. use the same heuristic function to perform a depth first search into the lattice until a hypothesis is chosen. EDSM and S-EDSM are prime examples of this technique. In this paper we discuss the introduction of diversification into our search space [5]. By introducing diversification through pairwise incompatible merges, we traverse multiple disjoint paths in the search lattice and obtain better results for the inference process.