(Un)supervised morphological segmentation for low-resource languages

Abstract

There exists a growing asymmetry between the tools available to computational linguists on the one hand and language documentarians and linguists researching low-resource languages on the other. To this end, the current project details the application of modern natural language processing tools to the task of morphological segmentation, a necessary primary step in morphological description and analysis. Results from supervised and unsupervised versions of a Bayesian statistical learning model used for the task of morphological segmentation of a small dataset consisting of the inflectional paradigms of a handful of Bardi (Nyulnyulan; Australia) verbs are outlined. Moreover, implications for the use of such models for related documentary and analytic tasks are discussed.