Developing Phrase-Based SMT for low-Resource Language Translation: Myanmar and Tedim Chin Language

This paper presents an initial evaluation of a translation system developed to address the linguistic gap between the Myanmar and Tedim Chin languages. As part of this project, we have also compiled a parallel corpus for Myanmar-Tedim Chin, containing around 50,000 sentences sourced from the UCSY corpus. Translation involves converting the symbols or characters of one writing system into those of another, with the aim of preserving the original pronunciation as closely as possible. For the Myanmar and Tedim Chin languages, accurate translation is particularly challenging due to the complexity of their writing systems and the limited availability of relevant data. In this study, we perform statistical experiments using the Phrase-Based Statistical Machine Translation (PBSMT) model, employing 5-gram language models during the decoding phase. We conduct experiments with a test set of 1,000 sentences and a development set of 7,000 sentences from the newly created dataset. To assess the performance of the translation system, we use the Bilingual Evaluation Understudy (BLEU) score, evaluating both Myanmar-to-Tedim Chin and Tedim Chin-to-Myanmar translation directions.