GRU-based Encoder-Decoder Attention Model for English to Bangla Translation on Novel Dataset
Al Mahmud, Md. Mohaimin Al Barat, Soomanib Kamruzzaman · 2021
Neural machine translation (NMT) is being used widely for its better translation quality for many languages. Though Bangla is a widely used language, many experiments on different NMT approaches have not been carried out for the English-Bangla language pair. In this paper, we experimented with a specific NMT model to analyze the results on the language pair. We have chosen a model on which research for translating from English to Bangla has not been conducted yet. We used a GRU-based unidirectional RNN model with an encoder-decoder-attention for this research. The research is conducted on a small-size balanced dataset. We have created a dataset of 1,014 sentences by translating from the classic novel ‘Gulliver’s Travels’ and appended them with another balanced dataset. This model gives a huge BLEU score of 50.07. Qualitative and quantitative analysis on the resulting outcome and the BLEU score has been made in the research.