English to Sinhala Neural Machine Translation

Neural Machine Translation (NMT) is the current state-of-the-art machine translation technique available among all other techniques. It has indeed shown promising results for resourceful languages. However, NMT heavily underperforms in low resource settings. English proficiency of Sri Lankans is very low and only a handful of people can understand and speak English. Since Sinhala language is the most widely used language in Sri Lanka, there is a huge demand for quality English to Sinhala translations in order to share knowledge among locals. Sinhala is a language with different morphology and syntax compared to English. Hence, translating English text to Sinhala is immensely challenging. In this paper, we introduce an effective NMT system along with Byte Pair Encoding (BPE) for the English-Sinhala language pair focusing on the Sri Lankan official government documents.