Problem
Research questions and friction points this paper is trying to address.
Buddhist texts
Sinhala
Pali
corpus
language models
SiPaKosa: A Comprehensive Corpus of Canonical and Classical Buddhist Texts in Sinhala and Pali
🤖 AI Summary
This study addresses the critical scarcity of high-quality Sinhala–Pali Buddhist textual corpora, which has significantly hindered domain-specific language model development and digital Buddhist studies. We present SiPaKosa, the first systematically constructed bilingual corpus comprising 786,000 sentences and 9.25 million words, derived from 16 copyright-cleared historical texts and a comprehensive web crawl of the Tipiṭaka. The corpus was built using high-accuracy OCR via Google Document AI, custom web crawlers, and rigorous cleaning pipelines, and is accompanied by structured metadata. Evaluation of ten pretrained language models on SiPaKosa reveals that closed-source models achieve perplexity as low as 1.09, outperforming open-source alternatives by a factor of three to six. This resource establishes a foundational benchmark for Buddhist text processing, historical linguistic analysis, and information retrieval in low-resource sacred languages.
📝 Abstract
SiPaKosa is a comprehensive corpus of Sinhala and Pali doctrinal texts comprising approximately 786K sentences and 9.25M words, incorporating 16 copyright-cleared historical Buddhist documents alongside the complete web-scraped Tripitaka canonical texts. The corpus was created through high-quality OCR using Google Document AI on historical manuscripts, combined with systematic web scraping of canonical repositories, followed by rigorous quality control and metadata annotation. The corpus is organised into language-specific subcorpora: Sinhala and Mixed Sinhala-Pali. We evaluate the performance of language models using ten pretrained models, with perplexity scores ranging from 1.09 to 189.67 on our corpus. This analysis shows that proprietary models significantly outperform open-source alternatives by factors of three to six times. This corpus supports the pretraining of domain-adapted language models, facilitates historical language analysis, and aids in the development of information retrieval systems for Buddhist scholarship while preserving Sinhala cultural heritage.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Buddhist corpus
OCR and web scraping
domain-adapted language models
perplexity evaluation
Sinhala-Pali texts
🔎 Similar Papers
No similar papers found.
