Neural Machine Translation from English to Marathi Using Various Techniques

Machine translation (MT) is a term used to describe computerized systems that generate translations from one linguistic communication to another, either with or without the need of humans. Text can be used to evaluate knowledge and converting that information to visuals can help in communication and information acquisition. There have been limited attempts to analyze the performance of state-of-the-art NMT algorithms on Indian languages, with a significant number of attempts in translating English to Hindi, Tamil and Bangla. The paper explores alternative strategies for dealing with low-resource hassle in neural machine translation (NMT), with a particular focus on the English-Marathi NMT pair. To provide high-quality translations, NMT algorithms involve a large number of parallel corpora. In order to tackle the low-resource dilemma, NMT models have been trained, along with transformers and attention models, as well as try hands-on sequence-to-sequence models. The data has been trained for sentence limit of 50 words and then fine-tune the default parameters of these NMT models to obtain the most optimum results and translations.