From Shloka to Shabda: Low-Resource Sanskrit to Bengali Translation

Sanskrit, one of the world’s oldest classical languages, holds a vast repository of historical, religious, and literary knowledge. Yet, most Bengali speakers face significant barriers accessing these resources due to language differences and the lack of effective Sanskrit-to-Bengali machine translation systems. To overcome this challenge, we developed a neural machine translation (NMT) system that translates Sanskrit texts into Bengali, assisted by a novel Sanskrit-Bengali translation dataset. Our main contribution involved manually compiling a parallel corpus of 31,639 Sanskrit-Bengali sentence pairs from religious scriptures, textbooks, tutorials, literary works, and existing translations. We trained and evaluated the transformer-based multilingual mBART model, pre-trained on several Indian languages, which shows enhanced translation performance in low-resource scenarios owing to its linguistic proximity to Sanskrit through both transfer learning and linguistic conceptual cross-lingual transfer. During fine-tuning, the model’s training BLEU score for Sanskrit-to-Bengali translation improved from 18 to 25.73. We evaluated the model using standard translation metrics and qualitative analysis. The results achieved were a BLEU score of 21.11, METEOR score of 40.10, ROUGE-1 score of 19.41, ROUGE-2 score of 8.72, ROUGE-L score of 19.29, chrF score of 41.41, and a TER of 77.83 on the test set. These results demonstrate that our method produces meaningful translations even with limited data.

Authors

Institutions

Publication Details

Journal
ACM Transactions on Asian and Low-Resource Language Information Processing
Published
2026-11-19
DOI
https://doi.org/10.1145/3864375
Primary Topic
Natural Language Processing Techniques
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

From Shloka to Shabda: Low-Resource Sanskrit to Bengali Translation

Mayna Talukdar, Mahir Mahbub, Saurav Das, Toukir Ahammed et al.
ACM Transactions on Asian and Low-Resource Language Information Processing
Natural Language Processing Techniques
article

From Shloka to Shabda: Low-Resource Sanskrit to Bengali Translation

Mayna Talukdar, Mahir Mahbub, Saurav Das, Toukir Ahammed, Md. Sanaullah, KOUSHIK BISWAS, Rupu Ghosh
article en

Abstract

Sanskrit, one of the world’s oldest classical languages, holds a vast repository of historical, religious, and literary knowledge. Yet, most Bengali speakers face significant barriers accessing these resources due to language differences and the lack of effective Sanskrit-to-Bengali machine translation systems. To overcome this challenge, we developed a neural machine translation (NMT) system that translates Sanskrit texts into Bengali, assisted by a novel Sanskrit-Bengali translation dataset. Our main contribution involved manually compiling a parallel corpus of 31,639 Sanskrit-Bengali sentence pairs from religious scriptures, textbooks, tutorials, literary works, and existing translations. We trained and evaluated the transformer-based multilingual mBART model, pre-trained on several Indian languages, which shows enhanced translation performance in low-resource scenarios owing to its linguistic proximity to Sanskrit through both transfer learning and linguistic conceptual cross-lingual transfer. During fine-tuning, the model’s training BLEU score for Sanskrit-to-Bengali translation improved from 18 to 25.73. We evaluated the model using standard translation metrics and qualitative analysis. The results achieved were a BLEU score of 21.11, METEOR score of 40.10, ROUGE-1 score of 19.41, ROUGE-2 score of 8.72, ROUGE-L score of 19.29, chrF score of 41.41, and a TER of 77.83 on the test set. These results demonstrate that our method produces meaningful translations even with limited data.

ACM Transactions on Asian and Low-Resource Language Information Processing
University of Dhaka (BD), Jagannath University (BD), University of Frontier Technology, Bangladesh (BD)
Natural Language Processing Techniques
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon