Mangalam Corpus of Buddhist Sanskrit Literature

This is a Sanskrit corpus developed at the Mangalam Research Center (Berkeley, California) for the study of Buddhist Sanskrit lexicology.

It comprises:

•  436 lemmatized and metadata-enriched Buddhist Sanskrit texts for a total of ~ 7.5 million words. 
• a lemmatized reference corpus of general Sanskrit including 397 texts for a total of ~ 15.5 million words
• a metadata table with information about each text in the Buddhist and Reference corpora
• stemmed and normalised version of the Buddhist corpus & sketch grammar for use in Sketch Engine

Lemmatization notes

The corpora are in romanised Sanskrit (UTF-8 encoding). Where multiple spelling variants involving a nasals are attested,  we have normalised the spelling to ṃ. Verbs are lemmatized to the stem of the present indicative of third person singular; the verb root can be found in the Root column. We have replaced avagraha with a.

Data Quality & Limitations
We are grateful to have received an Ashoka grant from the Khyentse Foundation to proofread samples of the Buddhist Sanskrit Corpus. Still, only 1.6% of the corpus has been proofread and many segmentation and lemmatization errors are likely to remain. Quantitative evaluation based on ~9000 proofread sentences puts pre-processing accuracy at ~90% (0.914 accuracy and 0.901 F1 score) for version .0

Acknowledgments
The corpus had been first realised as part of the project 'Lexis and Tradition: variation in the vocabulary of Sanskrit Mahāyāna literature'. This project was funded by the British Academy through a Newton International Fellowship (NF161436) and hosted at the Department of Theology and Religious Studies at King's College London under the supervision of Prof. Henrietta Kate Crosby. It has subsequently been expanded and its accuracy improved with funding from the Khyentse Foundation (Ashoka Grant 2021-2022).

Dr. Bruno Galasek-Hul has contributed to versions 1.4 - 1.7 thanks to funding from the Mangalam Research Center for Buddhist Languages.

Dr Anuja Ajotikar, Madhusudan Rimal & Jai Paranjape have proofread sentences sampled from versions 1.7 to 2.0, thanks to funding from the Khyentse Foundation.

The reference corpus of general Sanskrit has been tokenised by Matej Martinc within the project 'Computing the Dharma' funded by the National Endowment for the Humanities (HAA-277246-21).

Thanks to GRETIL, CTS e-texts, Vinita Tseng, Jowita Kramer and Prof. Steinkellner for kindly giving their permission to include automatically processed versions of some of their editions in this corpus.

Changelog

version 2.1 reprocessed reference corpus; both corpora expanded

version 2.0 changes the title of the corpus, adds more Buddhist texts and improves pre-processing accuracy.

version 1.9 adds more Buddhist texts, has better segmentation and lemmatization and is partially proofread

version 1.8 adds more Buddhist texts and is partially proofread

version 1.7 adds more Buddhist texts and a new pre-processed corpus of general Sanskrit

version 1.6 adds more Buddhist texts, improves segmentation and adds an initial iteration of the lemmatised corpus

version 1.5 adds more Buddhist texts, removes the reference corpus and improves segmentation

version 1.4 adds 59 Buddhist texts and fixes some recurrent segmentation errors

version 1.4.1 corrects some spacing and sentence parsing errors