AI for a corpus of Buddhist literature

Materials to consider

• Approx. 150 texts, by approx. 40 authors
• Original language of composition: Sanskrit
• Substantial number of texts survive only in Tibetan
• Some early ones also/only in Chinese translations

NOT systematically marked-up in any way

Problems for AI solutions?

Decomposition of Sanskrit euphonic combinations

• tathā + bhāva ==> tathābhāva
• “tathābhāva” can be analyzed as:
  • tathā + abhāva (possible)

  • OR tathā + bhāva (correct)

• Example solution: https://github.com/OliverHellwig/sanskrit/ (ca. 15% of lines contain an error);
• Possibly related class of problems: scriptio continua

Text alignment (editing assistance)

1. Discovery of (approximate) similar text passages
   • parallels/quotations/silent reuse
   • possibly across languages (mainly Sanskrit <–> Tibetan)
2. Alignment of main text ==> commentary ==> sub-commentary …
3. Alignment of translations

Realistic expectations?

1. Many texts available digitally (not systematically marked up)
2. Possible to create training corpus for Sanskrit <–> Tibetan alignment (mid-term)
3. Fairly good Sanskrit <–> Tibetan dictionaries

What we’d like:

1. Splitter for scriptio continua and compounds
2. Tools to align text passages of various types:
   • “Lemmas”: Main text ==> Commentary
   • “Quotations/Paraphrases”: Text A ==> Text B
   • Translations