Myanmar-Wa Machine Translation using LSTM-based Encoder-Decoder Model

Abstract

This work contributes to the quality evaluation of Machine Translation between Myanmar and Wa and provides the research on Long Short-Term Memory (LSTM)-based Deep Learning encoder-decoder mode. The Parallel Myanmar-Wa Corpus also includes over 20000 sentences based on Myanmar. According to previous research, Neural Machine Translation (NMT) is still needed for the development of Natural Language Processing (NLP) research field in Myanmar. Machine translation systems, especially statistical machine translation systems, require large amount of parallel corpora. The lack of a large parallel corpus for proposed system development is a major problem in development of machine translation. Myanmar and WA are very different languages not only in terms of basic sentence structure, but also in terms of syntax, grammar and morphology. This reason can cause great complexity in any NLP task. Furthermore, the analysis presented in this study provides valuable information for future studies using interethnic MT in Myanmar.