Deep Fusion of Bi-LSTM Attention Mechanism for the Enchantment of Machine Translation Performance

Abstract

Language translation is in high demand for a variety of reasons, including business travel and news comprehension, in this modern-day when people are increasingly reliant on technology. Bengali is the world's seventh most widely spoken language; yet, when compared to a language with more resources, such as English, the work done on Bengali machine translation falls short. The accuracy could be increased utilizing a state-of-the-art technique. We were inspired to investigate English to Bengali machine translation after finding research gaps. We utilized Neural Machine Translation namely BiLSTM. The encoder and decoder both employed BiLSTM; the encoder and decoder map the English sentences to Bengali sentences. In the decoder, the attention mechanism was implemented for better mapping. We found that this method works well for accurate machine translation. Because the sequences of the input language of the BiLSTM, are taken from both the left and the right, the mapping becomes more accurate. We compared our approaches to other standard results of English-Bengali translation. And finally, we showed that our result outperformed other claimed results.