Neural machine translation (NMT) is becoming increasingly popular as an effective method of automated language translation. However, due to a scarcity of training datasets, its effectiveness is limited when used with low-resource languages, such as Indian Languages (ILs). The lack of parallel datasets in Natural Language Processing (NLP) makes it difficult to investigate many ILs for Machine Translation (MT). A data augmentation approach such as Backtranslation (BT) can be used to enhance the size of the training dataset. This paper presents the development of a NMT model for ILs within the context of a MT system. To address the issue of data scarcity, the paper examines the effectiveness of a BT approach for ILs that uses both monolingual and parallel datasets. Experimental results reveal that while the BT has improved the model’s performance, however, it is not as significant as expected. It has also been observed that, even though the English-ILs and ILs-English models are trained on the same dataset, the ILs-English models perform better in all evaluation metrics. The reason for this is that ILs frequently differ in sentence structure, word order, and morphological richness from English. The paper also includes error analysis for translations between languages that were utilized in experiments utilizing the Multidimensional Quality Metrics (MQM) framework.

Multilingual Neural Machine Translation (MNMT) exhibits incredible performance with the development of a single translation model for many languages. Previous studies on multilingual translation reveal that multilingual training is effective for languages with limited corpus. This paper presents our submission (Team Id: NITR) in the WAT 2022 for “MultiIndicMT shared task” where the objective of the task is the translation between 5 Indic languages from OPUS Corpus (which are newly added in WAT 2022 corpus) into English and vice versa using the corpus provided by the organizer of WAT. Our system is based on a transformer-based NMT using fairseq modelling toolkit with ensemble techniques. Heuristic pre-processing approaches are carried out before keeping the model under training. Our multilingual NMT systems are trained with shared encoder and decoder parameters followed by assigning language embeddings to each token in both encoder and decoder. Our final multilingual system was examined by using BLEU and RIBES metrics scores. In future, we look forward to extend our research that will help in fine-tuning of both encoder and decoder during the monolingual unsupervised training in order to improve the quality of the synthetic data generated during the process.