Recovering the Calcutta Edition of the Mahābhārata for Computational Analysis

Paper · 2026 · Sujoy Sarkar, Manoj Balaji Jagadeeshan and Pawan Goyal

Summary

The Calcutta Edition (1834–1839) was the first complete printed Mahābhārata and is the main witness of the Northern recension. Whereas the critical text from BORI and the Kumbhakonam version of the South are available in machine-readable form, it has existed only as page scans and could not be used in computational work. The authors recover it from the digitized M. N. Dutta text in the Itihāsa dataset. That text follows Calcutta content but uses a different arrangement of books, chapters and verse numbers. They first restructured it by hand to match the original Parva–Adhyāya layout, fixing split, merged and missing chapters. They then aligned verses semi-automatically, using manually checked anchor verses and interpolating between them. About 88% of the Calcutta Edition now has verse-level alignment, and uncertain segments are flagged for later review.

With this corpus, they compare the Calcutta, BORI and Kumbhakonam texts on Parva-level expansion, speaker distribution and character embeddings. Textual volume and structure differ substantially between recensions, but higher-level narrative and semantic patterns stay fairly stable. The corpus is openly released under a Creative Commons license.

Key points

Source and restructuring: The source is the Mahābhārata portion of Itihāsa (~73K verses, based on M. N. Dutta). Dutta's nine books were remapped onto the Calcutta Edition's 18 Parvas, with corrections such as the missing Adhyāya 257 in the Śānti Parva.
Verse and chapter counts: BORI has ~73K verses, Kumbhakonam ~96K, Sastri–Vavilla ~95K and Calcutta ~90K. Adhyāya totals are 1995, 2311, 1981 and 2112 respectively.
Alignment result: About 88% of the Calcutta Edition is aligned at verse level, combining anchor verses with automatic interpolation.
Speaker analysis: Speakers were attributed with rule-based matching of markers such as uvāca. Calcutta's narrative proportions closely follow BORI, while Kumbhakonam departs more.
Embedding analysis and release: FastText models were trained per recension and projected with PCA. In all three, the Pāṇḍavas and Kauravas separate clearly, and Kṛṣṇa sits apart from both groups. The corpus is available at github.com/sujoysarkarai/mahabharatace.

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, topics and concepts

📚 Shelves 2Computer Science · 16,667 works Artificial Intelligence · 16,194 works

Cite

Cite as Sujoy Sarkar, Manoj Balaji Jagadeeshan and Pawan Goyal. 2026. Recovering the Calcutta Edition of the Mahābhārata for Computational Analysis. Proceedings of the 8th International Sanskrit Computational Linguistics Symposium.

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

@inproceedings{sarkar2026recovering,
  title = {{Recovering the Calcutta Edition of the Mahābhārata for Computational Analysis}},
  author = {Sujoy Sarkar, Manoj Balaji Jagadeeshan and Pawan Goyal},
  booktitle = {Proceedings of the 8th International Sanskrit Computational Linguistics Symposium},
  year = {2026},
  url = {https://aclanthology.org/2026.iscls-1.11/}
}

References

12 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

Srinivasa Kumar N. Acharya and S. R. Arjuna. 2016. Grammatical analysis and subject indexing of the mahābhārata and tātparyanirṇaya. In Proceedings of the Sanskrit Computational Linguistics Workshop (SCLWS). find
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 P… find
Piotr Bojanowski, Edouard Grave, Armand Joulin, and Tomas Mikolov. 2017. Enriching word vectors with subword information. Transactions of the Association for Computational Linguistics, 5:135–146.
Johannes Adrianus Bernardus Buitenen and James L Fitzgerald. 1973. The Mahābhārata, volume 1. University of Chicago Press. find
Ian Jolliffe. 2011. Principal component analysis. In International Encyclopedia of Statistical Science, pages 1094–1096. Springer, Berlin, Heidelberg. find
Tilak Bahadur Khatri. 2023. Historical development of the epic mahābhārata. Patan Prospective Journal, 3(01):173–180. find
Krishna Dwaipāyana and Manmatha Nāth Duttā. 1895. Mahābhārata. Elysium Press, Calcutta. find
Tonape Ramacharya Krishnacharya and Ṭī Ār Vyāsācārya. 1907. Sriman Mahabharatam: a new edition mainly based on the South Indian texts, with footnotes and readings, volume 5. ”Nirnaya-sagar” Press. find
Sujoy Sarkar, Gourav Sarkar, Manoj Balaji Jagadeeshan, Jivnesh Sandhan, Amrith Krishna, and Pawan Goyal. 2025. Mahānāma: A unique testbed for literary entity discovery and linking. In Christos Christodoulopoulos, Tanmoy Chakraborty, Carolyn Rose, and Violet Peng, editors, Proceedings of the 2025 Con… find
Søren Sørensen. 1904. An Index to the Names in the Mahabharata: With Short Explanations and a Concordance to the Bombay and Calcutta Editions and P.C. Roy’s Translation, volume 1. Williams & Norgate, London. find
Vishnu Sitaram Sukthankar. 1933. Prolegomena [to the critical edition of the Ādiparvan, Book 1 of the Mahābhārata]. Bhandarkar Oriental Research Institute, Poona. find
Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. In Proceedings of the 31st International Conference on Neural Information Processing Systems, NIPS’17, page 6000–6010, Red Hook, NY, USA. Cur…

Updated 1 November 2026

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/sarkar-2026-recovering-calcutta-edition-mahabharata-computational/