✎ Edit this page

Vietnamese

Vietnamese is the language of Vietnamese Buddhism, in Vietnam and in the diaspora (for example Plum Village). Older Buddhist texts are in Hán (Classical Chinese) and Nôm, together called Hán-Nôm. Short answer: AI handles modern Vietnamese well on general tests, but there is no translation benchmark like WMT for it. For Hán-Nôm, a Vietnamese system does OCR and transliteration.

Summary

   
Where Vietnam and the diaspora
Speakers about 97 million in total (Ethnologue 2026) [1]
Script Latin alphabet (Quốc ngữ); Hán-Nôm in older texts
AI resources high
Google Translate yes, Translation LLM and NMT [2]
Microsoft Translator yes, with LLM translation [3]
Speech to text yes; Whisper large-v3 WER 8.9 on FLEURS [6]
Best evidence on translation quality FLORES-200 (2023) [5]; no WMT test

Writing and text input

Modern Vietnamese uses the Latin alphabet with many diacritics. Unicode and input are mature. Spaces separate syllables, not words.

Hán-Nôm OCR: the “Kim Hán Nôm” system (Centre for Computational Linguistics, University of Science, VNU-HCM) reports over 95% accuracy on clean printed scans. Weathered inscriptions and handwriting still need experts. Its partners include the Tran Nhan Tong Institute, a Trúc Lâm Buddhist heritage institute [9]. An open OCR dataset (nom-ocr-data, 57.1k rows, CC BY 4.0) exists; it is not Buddhist [10]. The Vietnamese Nôm Preservation Foundation (1999–2018) worked on Nôm in Unicode and digitized texts, including at Thắng Nghiêm Temple [11]. We did not verify how many Nôm characters are in Unicode.

Data

Speech

Machine translation

Benchmark, FLORES-200, English to Vietnamese, chrF++ (higher is better): NLLB about 59.3 [4]; in 2023 NLLB-MoE 59.5 and ChatGPT 0-shot 58.5 [5]. This is high, but chrF++ cannot be compared across languages.

There is no WMT human evaluation for Vietnamese. Indirect benchmark: in a 2026 study, normalizing dialect text raised acceptance of Gemini 2.0 Flash translations from 61.83% to 67.00% [15]. Benchmark (knowledge, not translation): on VMLU, a Vietnamese multiple-choice test, Llama-3-70B scored 66.44 and GPT-4 65.53 (2024) [14].

We found no Buddhist test into Vietnamese. Older results may not show what current models can do.

Classical form: Hán-Nôm

AI projects

Recommendation

In the Directory

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

Projects (2)

Datasets (1)

Sources

  1. Wikipedia: List of languages by total number of speakers (Ethnologue 2026), 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. NLLB metrics, public copy (Mara4444/MSc_Thesis), read 2026-11-17
  5. Robinson et al. 2023, ChatGPT MT, arXiv 2309.07423, read 2026-11-17
  6. Dolphin paper, arXiv 2503.20212, read 2026-11-17
  7. Omnilingual ASR per-language results (7B LLM-ASR), read 2026-11-17
  8. openai/whisper tokenizer.py, read 2026-11-17
  9. HCMUS: AI unlocks Hán Nôm heritage, 2026-07-19, read 2026-11-17
  10. nom-ocr-data on Hugging Face, read 2026-11-17
  11. Wikipedia: Vietnamese Nôm Preservation Foundation, read 2026-11-17
  12. Vietnam Government portal: Vĩnh Nghiêm woodblocks, 2012-05-17, read 2026-11-17
  13. VanMach-7B on Hugging Face, read 2026-11-17
  14. VMLU on GitHub, read 2026-11-17
  15. ViDia2Std, arXiv 2603.10211, read 2026-11-17
  16. vanloc1808/buddism-qa-dataset, read 2026-11-17
  17. nphuc/thich-nhat-hanh-dataset, read 2026-11-17
  18. CBETA Copyright Notice, read 2026-11-17
  19. tnh-scholar preliminary feasibility study, read 2026-11-17

✎ Edit this page · History · Discuss