Word classes of White Hmong: a computational approach

Abstract

This thesis employs a computational approach to study the word classes in White Hmong, a minority language of Mainland Southeast Asia. It proposes an automatic discovery procedure for word classes based on a careful review and comparison of existing algorithms. Motivated by the distributional hypothesis, which posits that similar words occur in similar environments, the procedure represents words as vectors defined by pairwise co-occurrence. It then measures their grammatical similarity in terms of spatial proximity and clusters them into a hierarchical taxonomy. The procedure is applied to an unannotated corpus of White Hmong, yielding a classification of its lexicon. The classification is evaluated against known grammatical properties of the language, demonstrating the linguistic meaningfulness of the results.