EAGER: Building Natural Language Processing Tools for a Low-Resource and Endangered Language

In recent decades, a concerted movement to document and revitalize endangered languages through educational efforts have saved such languages from extinction. However, progress has not kept up with endangered language community needs in modern society. Endangered languages often lack words for some modern science and technology concepts, and there is a growing need for educational materials in these languages. With recent advances in natural language processing, low-resource languages and endangered languages have been largely neglected. This project develops new user-facing language technologies for an endangered language, starting with machine translation and language models, which in collaboration with the language communities aids in preservation and revitalization. This project first collects, cleans, and consolidates texts and dictionaries from online sources, along with data from a digital repository of language data housed at the University of Hawai'i. Next, machine translation systems are trained on available data augmented with techniques such as backtranslation and methods from computational etymology and are applied to translate text materials both into and out of the endangered language. Due to differences in orthography, methods to develop a consistent orthography are also investigated. Finally, language models are developed to enable future computational study of low-resource and endangered languages. This project not only positively benefits the community of endangered language speakers under study, but also serves as a case study for applying and developing language technologies to preserve and revitalize other low-resource and endangered languages around the world. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.