Towards data-driven machine translation for Lumasaaba

Abstract

This paper reports on results from initial efforts towards the application of data-driven machine translation for low-resourced East African languages. In particular, the paper evaluates the application of phrase-based statistical machine translation (PBSMT) and neural machine translation (NMT) for automating translation between Lumasaaba (an East African Bantu language) and English. As expected, the PBSMT approach outperforms the NMT approach on a small Bible-based corpus of parallel sentences. The parallel corpus and the respective machine translation evaluations presented in this paper can be used as baselines for future machine translation quality improvement efforts involving Lumasaaba and other related East African Bantu languages.