Elamur Research⌕
☰
home / papers / 2026 / Santham: A Curated Sanskrit–Tamil Dataset with Anvaya and Segmentation for Building and Evaluating Machine Translation

Santham: A Curated Sanskrit–Tamil Dataset with Anvaya and Segmentation for Building and Evaluating Machine Translation

Summary

Sanskrit-to-Tamil machine translation has received little attention. Good parallel data is scarce, and the two languages differ deeply: Sanskrit is Indo-Aryan, with sandhi, compounds and free verse word order, while Tamil is Dravidian. The authors present Santham, a curated corpus of more than 90,000 Sanskrit–Tamil pairs. It covers classical epics and devotional works plus modern prose. The data was assembled through commissioned human translation, OCR of printed editions, web scraping and a cleaned, re-aligned portion of NLLB. On this corpus they fine-tune open LLMs (Gemma, Qwen) and compare them with IndicTrans2. They also test two input preprocessing steps: word segmentation with existing Sanskrit tools, and anvaya, which rearranges verses into prose order. The fine-tuned LLMs beat the specialized multilingual baseline. Segmentation gives modest gains, and anvaya raises BLEU on poetry by over 46% relative. A human evaluation adds nuance: anvaya does not always preserve meaning or Tamil fluency better than segmented input. The resource and benchmark fill a real gap for translation between classical and regional Indian languages.

Key points

  • Corpus size and sources: Santham contains over 90,000 parallel pairs. Its material draws on five classical and devotional texts (the Bhagavad Gita, the Rāmāyaṇa, the Narayaneeyam, the Mahābhārata and the Bhagavatam), along with eight modern prose collections, Saṃsādhanī prose translated by experts, and a curated NLLB subset. Google Vision API was chosen over Tesseract for OCR.
  • Pipeline: The five-stage preparation process includes OpenCV-based footer removal before OCR. Alignment was done at the sentence level for prose and the verse level for poetry, starting automatically and then corrected by hand.
  • Models and segmentation: The fine-tuned models include Qwen3-8B and Gemma3-4B, which outperform IndicTrans2. Segmentation was tested with Sanskrit Heritage (SH) and Dharmamitra (ByT5).
  • Human evaluation design: Two bilingual experts rated 115 samples on a 1–5 scale for adequacy, fluency, syntactic flow and technical consistency. Agreement on adequacy reached weighted κ = 0.683.
  • Human evaluation findings: Prose translations scored higher than poetry on every criterion. Among automatic metrics, COMET matched human judgments most closely.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

Fields: Machine translation · Natural language processing

Topics: Machine translation

Cite

The paper itself, not this page. BibTeX from the library's record of the work.

@inproceedings{s2026santham,
  title = {{Santham: A Curated Sanskrit–Tamil Dataset with Anvaya and Segmentation for Building and Evaluating Machine Translation}},
  author = {S, Prasanna Venkatesh T and Shetye, Ketaki Mangesh and Arjunasamy, Vishnuraj and Sahu, Ayush Kumar and Krishnan, Sriram and Krishnamurthy, Parameswari},
  booktitle = {Proceedings of the 8th International Sanskrit Computational Linguistics Symposium},
  year = {2026},
  url = {https://aclanthology.org/2026.iscls-1.5/}
}

References

23 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Rahul Aralikatte, Miryam de Lhoneux, Anoop Kunchukuttan, and Anders Søgaard. 2021. Itihasa: A large-scale corpus for Sanskrit to English translation. In Toshiaki Nakazawa, Hideki Nakayama, Isao Goto, Hideya Mino, Chenchen Ding, Raj Dabre, Anoop Kunchukuttan, Shohei Higashiyama, Hiroshi Manabe, Win Pa Pa, Shantipriya Parida, Ondřej Bojar, Chenhui Chu, Akiko Eriguchi, Kaori Abe, Yusuke Oda, Katsuhito Sudoh, Sadao Kurohashi, and Pushpak Bhattacharyya, editors, Proceedings of the 8th Workshop on Asian Translation (WAT2021), pages 191–197, Online, August. Association for Computational Linguistics. find
  2. Rhugved Pankaj Chaudhari, Bhakti Jadhav, Pushpak Bhattacharyya, and Malhar Kulkarni. 2024. Sans-GPT: Advancing generative pre-training in Sanskrit. In Proceedings of the 21st International Conference on Natural Language Processing (ICON), pages 432–441. NLP Association of India (NLPAI). find
  3. Rahul Chingamtotattil and Rajamma Gopikakumar. 2022. Neural machine translation for Sanskrit to Malayalam using morphology and evolutionary word sense disambiguation. Indonesian Journal of Electrical Engineering and Computer Science, 28(3):1709–1719, December. ↗
  4. Jay Gala, Pranjal A Chitale, A K Raghavan, Varun Gumma, Sumanth Doddapaneni, Aswanth Kumar M, Janki Atul Nawale, Anupama Sujatha, Ratish Puduppully, Vivek Raghavan, Pratyush Kumar, Mitesh M Khapra, Raj Dabre, and Anoop Kunchukuttan. 2023. Indictrans2: Towards high-quality and accessible machine translation models for all 22 scheduled indian languages. Transactions on Machine Learning Research. ↗ our card
  5. Gemma Team. 2025. Gemma 3 technical report. ↗
  6. Gérard Huet. 2009. Sanskrit Segmentation. In Proceedings of the South Asian Languages Analysis Roundtable XXVIII, October. find
  7. Prashanth Kammar, Parashuram Baraki, Sunil Kumar Ganganayaka, Manjunath Swamy Byranahalli Eraiah, and Kolakaluri Lakshman Arun Kumar. 2024. Translating Sanskrit to Hindi Language using Recurrent Neural Network (RNN)-L2 Regularization. International Journal of Intelligent Systems and Applications in Engineering, 12(3):199–208, March. ↗
  8. Amrith Krishna et al. 2018. Free as in free word order: An energy based model for word segmentation and morphological tagging in sanskrit. In Proceedings of EMNLP 2018, pages 2550–2561. ↗
  9. Sriram Krishnan, Amba Kulkarni, and Gérard Huet. 2024. Normalized dataset for sanskrit word segmentation and morphological parsing. Language Resources and Evaluation, 59(2):1279–1330, Aug. ↗
  10. Anil Kumar, Amba Kulkarni, and Nakka Shailaj. 2024. START: Sanskrit teaching; annotation; and research tool. In Proceedings of the 7th International Sanskrit Computational Linguistics Symposium, pages 113–124. Association for Computational Linguistics, February. find
  11. Lisha C.R. 2024. Techniques in Translating Sanskrit Poetry: From Literal to Creative Approaches. International Journal of Innovative Research in Technology, 11(4):241–248, August. find
  12. Llama Team. 2024. The llama 3 herd of models. find
  13. Ayush Maheshwari, Ashim Gupta, Amrith Krishna, Atul Kumar Singh, Ganesh Ramakrishnan, Anil Kumar Gourishetty, and Jitin Singla. 2024. Samayik: A benchmark and dataset for English-Sanskrit translation. In Nicoletta Calzolari, Min-Yen Kan, Veronique Hoste, Alessandro Lenci, Sakriani Sakti, and Nianwen Xue, editors, Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024), pages 14298–14304, Torino, Italia, May. ELRA and ICCL. ↗
  14. Sebastian Nehrdich and Kurt Keutzer. 2026. Mitra: A large-scale parallel corpus and multilingual pretrained language model for machine translation and semantic retrieval for pāli, sanskrit, buddhist chinese, and tibetan. ↗
  15. Sebastian Nehrdich, Oliver Hellwig, and Kurt Keutzer. 2024. One model is all you need: ByT5-Sanskrit, a unified model for Sanskrit NLP tasks. In Findings of the Association for Computational Linguistics: EMNLP 2024, pages 13742–13751. Association for Computational Linguistics, November. find
  16. Sebastian Nehrdich, David Allport, Sven Sellmer, Jivnesh Sandhan, Manoj Balaji Jagadeeshan, Pawan Goyal, Sujeet Kumar, and Kurt Keutzer. 2026. Mitrasamgraha: A comprehensive classical sanskrit machine translation dataset. ↗
  17. NLLB Team. 2022. No language left behind: Scaling human-centered machine translation. ↗
  18. Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. 2002. Bleu: a method for automatic evaluation of machine translation. In Pierre Isabelle, Eugene Charniak, and Dekang Lin, editors, Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics, pages 311– 318, Philadelphia, Pennsylvania, USA, July. Association for Computational Linguistics. find our card
  19. Maja Popović. 2017. chrF++: words helping character n-grams. In Ondřej Bojar, Christian Buck, Rajen Chatterjee, Christian Federmann, Yvette Graham, Barry Haddow, Matthias Huck, Antonio Jimeno Yepes, Philipp Koehn, and Julia Kreutzer, editors, Proceedings of the Second Conference on Machine Translation, pages 612–618, Copenhagen, Denmark, September. Association for Computational Linguistics. find our card
  20. Anagha Pradeep and Radhika Mamidi. 2025. Sandarśana: A Survey on Sanskrit Computational Linguistics and Digital Infrastructure for Sanskrit. ACM Computing Surveys, May. Publisher: ACM-PUB27New York, NY. ↗
  21. Qwen Team. 2025. Qwen3 technical report. find
  22. Ricardo Rei, José G. C. de Souza, Duarte Alves, Chrysoula Zerva, Ana C Farinha, Taisiya Glushkova, Alon Lavie, Luisa Coheur, and André F. T. Martins. 2022. COMET-22: Unbabel-IST 2022 submission for the metrics shared task. In Philipp Koehn, Loïc Barrault, Ondřej Bojar, Fethi Bougares, Rajen Chatterjee, Marta R. Costa-jussà, Christian Federmann, Mark Fishel, Alexander Fraser, Markus Freitag, Yvette Graham, Roman Grundkiewicz, Paco Guzman, Barry Haddow, Matthias Huck, Antonio Jimeno Yepes, Tom Kocmi, André Martins, Makoto Morishita, Christof Monz, Masaaki Nagata, Toshiaki Nakazawa, Matteo Negri, Aurélie Névéol, Mariana Neves, Martin Popel, Marco Turchi, and Marcos Zampieri, editors, Proceedings of the Seventh Conference on Machine Translation (WMT), pages 578–585, Abu Dhabi, United Arab Emirates (Hybrid), December. Association for Computational Linguistics. find our card
  23. Nandini Sethi, Amita Dev, and Poonam Bansal. 2023. A Novel Neural Machine Translation Approach for low-resource Sanskrit-Hindi Language pair. ACM Trans. Asian Low-Resour. Lang. Inf. Process., April. Just Accepted. ↗

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/papers/s-2026-santham-curated-sanskrit-tamil-dataset/