✎ Edit this page

Pali

Pali is the language of the Theravada canon (Tipitaka) and its commentaries. It has rich, open text corpora, but no translation service supports it, and no public test measures translation into Pali.

Summary

   
Where Theravada countries and Pali students worldwide
Speakers not applicable (liturgical and scholarly language)
Script Roman for international use; Sinhala script in Sri Lanka; other local scripts
AI resources very low (for translation and speech); text corpora are good
Google Translate no [1] [2]
Microsoft Translator no [3]
Speech to text not in Whisper [4]; not in Meta Omnilingual ASR [5]
Best evidence on translation quality PaliBench, Pali to English only, 2026 [6]

Writing and text input

Pali has no script of its own. Sri Lankan readers use the Sinhala script, and international readers use Roman letters with diacritics. Tools in the Directory (for example paliscript and “Tipitaka Pali in 15 scripts”) convert between scripts. So an AI system can write Roman Pali, and a tool can convert it to the reader’s script.

OCR: a Sinhala and Pali OCR model and its training data (pathnirvana) are listed in the Directory below.

Data

Speech

Machine translation

Support. Google Translate does not support Pali [1] [2]. Microsoft Translator does not support Pali [3]. Pali is not in the FLORES-200 test set, so there are no FLORES scores for any system [8].

Benchmark (opposite direction): PaliBench, Pali to English, 10 models through OpenRouter, January–February 2026. Gemini 3 Pro scored best (similarity 0.946, chrF++ 68.5, BLEU 63.3, higher is better). Gemini 3 Flash scored 0.944 and Claude Opus 4.5 0.940. DeepSeek v3.2 scored 0.939, and GPT-5.2 was also tested. LLaMA 3.3 70B was last (0.888) [6].

These numbers are for translation from Pali. A model that reads Pali well does not always write good Pali. We found no test of translation into Pali by any system.

MITRA. Dharmamitra’s MITRA-MT model (Gemma 2, 9B) translates into English. Its paper reports no scores for translation into Pali [7].

AI projects

Recommendation

To translate into Pali today, use a current frontier model (Gemini 3 Pro and Claude Opus scored best on PaliBench, but only from Pali into English). Give it a glossary from the Digital Pāḷi Dictionary and similar SuttaCentral passages as examples. Ask it to write Roman Pali, then convert the script with a tool. A trained Pali reviewer must check every sentence. Confidence is low, because there is no test in this direction.

Test first: take 50 to 100 SuttaCentral segments with Sujato’s CC0 English translation. Give the model the English and ask for Pali. Score the output against the original Pali with chrF++, and ask a Pali scholar to review 20 items. Pali is not yet measured on the BuddhistAI leaderboard.

In the Directory

Projects and datasets in the Directory that list Pali. This list is made from the Directory data. To add an entry, suggest it.

Projects (21)

Datasets (12)

Sources

  1. Google Translate Help: more than 100 new languages, read 2026-11-17
  2. Google Cloud Translation: Language support, read 2026-11-17
  3. Microsoft Learn: Translator language support, read 2026-11-17
  4. OpenAI Whisper tokenizer.py (language list), read 2026-11-17
  5. Meta Omnilingual ASR language IDs, read 2026-11-17
  6. Metzger and Phophichit, PaliBench, arXiv 2605.16881, 2026-05-16, read 2026-11-17
  7. Nehrdich and Keutzer, MITRA, arXiv 2601.06400, 2026-01-10, read 2026-11-17
  8. Robinson et al. 2023, arXiv 2309.07423, read 2026-11-17

✎ Edit this page · History · Discuss