Multimodal Machine Translation for Sanskrit-Hindi: An Empirical Analysis

Due to its extensive use in ancient Indian religious scriptures, Sanskrit is among the oldest indigenous languages and is rightfully referred to as the language of the gods. However, it is losing favour in contemporary India. Sanskrit is not widely used in current times due in large part to the lack of resources for translation into and out of it. In recent years, machine translation (MT) has improved above and beyond the norm and is now typically performed utilising supervised learning approaches. Due to the paucity of comparable corpora for Sanskrit, new research in the unsupervised MT domain appears to have promise for Sanskrit. With the aid of manually created parallel corpora for the Sanskrit-Hindi language pair, an analysis is conducted between various modelling techniques of building a machine translation system, namely Statistical and Neural, in order to bridge the gap between Sanskrit and its contemporary successor Hindi. In order to provide a fresh viewpoint on the area as a whole, the primary benefits and drawbacks of statistical and neural machine translation has been examined in this work. Our results suggest that Neural machine translation modelling technique performs better than Statistical machine translation.