Assessment of deep learning based Hindi Odia bidirectional machine translation system

Abstract

India is a vast nation with a diverse range of cultures and languages. Most Indians choose to use their native languages when communicating with machines. To integrate smart technologies into every facet of Indian society, efficient systems that can positively identify Indian languages must be established. Machine translation (MT) studies comprise most of the natural language processing (NLP) in the era of multilingual computer-human interaction. Till now, less emphasis has been placed to develop MT systems among Indian languages. Yet again, building a qualitative and quantitative corpus in these languages is challenging. This work focuses on two Indic languages for the development of a Hindi to Odia bidirectional machine translation system (HOBMT). Bilingual evaluation understudy (BLEU), word error rate (WER), character error rate (CER), and metric for evaluation of translation with explicit ordering (METEOR) evaluation metrics are used to assess the accuracy of the translation. The most advanced sequential deep learning (DL) models, such as recurrent neural network (RNN), long short term memory (LSTM), and gated recurrent unit (GRU), are used in this study. In this research, RNN is observed with improved translation results due to its sequential data handling with context preservation.