Newari (Nepal Bhasa)
Newari (Nepal Bhasa) is the language of the Newar people of the Kathmandu Valley, the home of Newar Buddhism. The language is endangered [1]. Google Translate added it in 2024, but there is no public test of translation into Newari.
Summary
| Where | Nepal (Kathmandu Valley) |
| Speakers | about 880,000 native speakers (2021 census); UNESCO “Definitely endangered” [1] |
| Script | Ranjana, Devanagari, Newar script (Nepal Lipi), Bhujimol [1] |
| AI resources | low |
| Google Translate | yes, added in June 2024 [2] [3]; NMT only in the Cloud API [4] |
| Microsoft Translator | no [5] |
| Speech to text | not in Whisper [6]; Meta Omnilingual ASR covers it [7]; Nwāchā Munā corpus (2026) [8] |
| Best evidence on translation quality | none found |
Writing and text input
Newari is written in several scripts: Ranjana, Devanagari, the Newar script (Nepal Lipi) and Bhujimol [1]. Most digital Newari text and AI work uses Devanagari. For example, the Nwāchā Munā speech corpus uses Devanagari transcripts [8]. A Unicode block for the Newa script exists; we did not verify its details in this research.
OCR: we found no verified public OCR model or ground-truth data for Ranjana or the Newar script. Two items on Ranjana recognition (from Khwopa College and Kathmandu University) appeared in search results, but we did not open them.
Data
- We found no Newari–English parallel corpus of useful size.
- Manuscripts: the Nepalese-German Manuscript Cataloguing Project (NGMCP) catalogue includes many Buddhist Sanskrit and Newar manuscripts. We saw it only in search results and did not verify the details.
- Web corpus sizes for Newari: not verified.
Speech
- Nwāchā Munā (Kathmandu University ILPRL, 2026) is the first public Newari speech benchmark we found: 5.39 hours of transcribed speech, 5,727 utterances, 18 native speakers, Devanagari transcripts [8].
- Best character error rate (CER): 17.59% (NepConformer with data augmentation). Best word error rate (WER): 48.46% (NepConformer with a 5-gram language model). Lower is better.
- Fine-tuned Whisper-Small: 18.76% CER, 51.44% WER.
- A Nepali model used zero-shot: 52.54% CER, 98.68% WER. So a Nepali model does not work for Newari without training.
- The paper is CC BY-NC-SA 4.0. The data license is not stated separately.
- OpenAI Whisper does not list Newari [6].
- Meta Omnilingual ASR (November 2025) reports a CER of 2.2 for new_Deva, with 37.9 hours of training audio [7]. This test set is small and narrow, so do not compare it with the Nwāchā Munā numbers.
Machine translation
Support. Google Translate added “Nepalbhasa (Newari)” in its June 2024 expansion [2]. Local news covered this on 2024-06-28 [3]. In the Google Cloud API, Newari is NMT only, not in the Translation LLM [4]. Microsoft Translator does not support Newari [5]. Newari is not in the FLORES-200 test set, so there are no FLORES scores [9].
Evidence. None. We found no benchmark and no published human evaluation for translation into Newari with any system. Any quality claim would be anecdotal.
AI projects
- Nwāchā Munā: Newari speech corpus and models by Kathmandu University ILPRL. Code: github.com/ilprl/nwacha-muna. Data and models are on Hugging Face (ilprl-docse collection) [8].
Recommendation
To translate into Newari today, use Google Translate (the only system with official support) and test a current frontier model next to it. A native Newari reviewer must check all output, and should check that Buddhist terms follow Newar usage. Confidence is low: there is no test of any system.
Test first: ask native speakers to translate 50 to 100 short English sentences into Newari in Devanagari, half general and half Buddhist (for example, from a prayer or a festival text). Use them as a small test set with chrF++ and native review. This set would also be the first public test for translation into Newari. Newari is not yet measured on the BuddhistAI leaderboard.
In the Directory
Projects and datasets in the Directory that list Newari. This list is made from the Directory data. To add an entry, suggest it.
Projects (1)
- Nepalese-German Manuscript Cataloguing Project (NGMCP) (University of Hamburg with the National Archives of Nepal): Catalogue of the Nepalese manuscripts that the Nepal-German Manuscript Preservation Project microfilmed (more than 114,500 items in Sanskrit, Newari, Nepali and other languages).
Datasets (1)
- Nwāchā Munā (Newari speech) (Kathmandu University (ILPRL)): 5.39 hours of transcribed Newari (Nepal Bhasa) speech in Devanagari, with speech recognition baseline models.
Sources
- Wikipedia: Newar language, read 2026-11-17
- Google Translate Help: more than 100 new languages, read 2026-11-17
- ekantipur / Kathmandu Post: Newari language included in Google Translate, 2024-06-28, read 2026-11-17
- Google Cloud Translation: Language support, read 2026-11-17
- Microsoft Learn: Translator language support, read 2026-11-17
- OpenAI Whisper tokenizer.py (language list), read 2026-11-17
- Meta Omnilingual ASR, per-language results (7B LLM-ASR), read 2026-11-17
- Nwāchā Munā, arXiv 2603.07554v2, revised 2026-03-30, read 2026-11-17
- Robinson et al. 2023, arXiv 2309.07423, read 2026-11-17

BuddhistAI Knowledge base