Automated Gloss Mapping for Inferring Grammatical Properties

Abstract

This thesis describes a software system that maps glosses from interlinear glossed text (IGT) to an internally consistent set. This study hypothesizes that mapping glosses supports better inference of grammatical properties. Inference refers to analyzing information from IGT to automatically determine grammatical properties of a language for which a computational grammar can be constructed. The IGT will likely contain unknown or non-standard glosses. By mapping all glosses to an internally consistent set the non-standard rate should decrease which would provide more precise information for inferring grammatical properties. The inference procedure matches standard grams from a language to tense, aspect, and mood categories. These inferred gram and category pairs called choices are used to create computational grammars. The final results demonstrate that the methodology successfully reduces the non-standard gloss rate.