Inducing grammars from IGT

Abstract

We suggest a strategy for incremental construction of deep parsing grammars from Interlinear Glossed Text (IGT). Essential to the approach is that both the IGT and the grammar deploy a rich inventory of linguistic types and categories in their respective domains, with a view to be able to directly accommodate strings and constructions from ideally any type of language. This involves a heavy load of linguistically-based encoding in the initial phase, but on the other hand a simple technology residing basically in XML export from one application to the other, and the possibility for linguists working on specific languages to present data to the system in a form which does not go beyond common linguistic technique, and is useful for general linguistic purposes as well. The tool hosting the IGT is TypeCraft, and the system hosting the grammars is the LKB system, generally employed for implementing HPSG grammars. The strategy is illustrated with induction of a grammar fragment from a small set of IGTs from the Kwa language Ga. © 2014 Springer International Publishing.