Computational Detection of Text Reuse in the Chinese Buddhist Canon

This paper presents the first corpus-wide computational analysis of text reuse in the Taishō Tripiṭaka (大正新脩大藏經 Dàzhèng xīnxiū dàzàngjīng), the standard modern edition of the Chinese Buddhist canon containing approximately 2,920 texts across 85 volumes. Using a five-stage pipeline (text extraction, n-gram fingerprinting, seed-and-extend alignment, scoring, and classification), I detected 3,610 significant textual relationships involving 1,558 unique texts. These include 135 excerpts (texts ≥80% derived from a single source), 539 digests (30–80% derived), 538 commentaries, 224 retranslations, and 2,174 shared-tradition pairs. The pipeline was validated against the well-established derivation of the Heart Sūtra from the Large Prajñāpāramitā, correctly classifying all six test cases. Notable discoveries include the systematic quantification of encyclopedic absorption (e.g., the 法苑珠林 Fǎyuàn zhūlín, "Forest of Gems in the Garden of the Dharma," incorporates 102 shorter texts at measurable coverage levels), the comprehensive mapping of dhāraṇī derivation networks, and the detection of a 28.6-percentage-point gap between same-translator and cross-translator matching: a computational signature of translator individuality previously described only qualitatively. A phonetic transliteration detection module identifies 270 text pairs with phonetically equivalent passages where different Chinese characters encode the same Sanskrit sounds. Independent validation using a unified Taishō–Tibetan concordance confirms 93.6% of computationally detected retranslation pairs when both texts have Tibetan parallel data.