Answers are grounded in this paper and cite their sources.
This paper presents our multi-stage pipeline for information extraction from a vast corpus of hand-written Tibetan manuscripts documenting indigenous pagan ritual and mythic traditions, surviving alongside later Bön and Buddhist works. Diplomatic transcriptions, produced through our collaborative Handwritten Text Recognition (HTR) workflow (Griffiths & Meelen 2026), are normalised using a combination of rule-based replacements, a trained Seq2Seq model, and a KenLM model for ranking (Meelen & Griffiths forth.), building on the work of Faggionato et al. (2022) and extended with a custom abbreviation dictionary. Normalised eTexts are then uploaded to a custom-made annotation tool, where we apply a multilayered workflow combining animacy, semantic role labelling (SRL), and content features (through Named Entity Recognition, NER) to support downstream NLP tasks. This multilayered annotation system will allow us to categorise textual variants, trace the evolution of actors and ritual motifs across chronological and geographical layers, and separate indigenous pagan texts from those from Bön and Buddhist traditions. Beyond our own work, this scalable computational pipeline can aid others looking to extract information from large, highly ambiguous, and low-resource language materials.
No takes yet. Share an insight, caveat, or question.
Faggionato et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: