An Application of Hugging-Face Model for Exegesis of the Shlokas from Sanskrit to Bengali

Sanskrit is one of the most ancient languages in the world. It was commonly spoken in India around 600 B.C. This paper is focused on producing Bengali exegesis of the Shlokas of the Bhagavad Gita, which were originally written in Sanskrit. By utilizing a retrained Hugging-Face(HF) model and combining parallel data from both Sanskrit and Bengali sources, our goal is to overcome the linguistic and cultural barrier between these two ancient languages. This paper presents an analysis of the translation models utilized for this conversion. The models under consideration are IndicTrans2, iNLTK, and an HF model. The evaluation of these models is based on metrics such as BLEU, NIST, and LEPOR scores. This will reveal that the HF model outperforms the existing models, producing translations that closely align with three renowned Bengali versions of the Bhagavad Gita by Swami Prabhupada, Dhanapati Halder, and Swami Jagadiswarananda.