evidoria

← Back to browse

Good practice Imported

Dzongkha NLP — Bhutan's AI Translation, Speech and Text-to-Speech Tool for the National Language, Built on a Nu 21 Million Budget

Bhutan · Thimphu · See the Bhutan profile · See the Thimphu profile

Evidence: Descriptive / self-reported Top 47% 60/100 · Ask Evidence Copilot about this practice

Bhutan's GovTech Agency and a Royal University college built Dzongkha NLP — machine translation, speech recognition and text-to-speech for the national language — but officials openly report accuracy of only 30–50% given limited training data.

21 Nu million (~USD250,000)
Project budget
30-50%
Reported translation accuracy
~50
Volunteer engineers in prototype challenge (Aug-Oct 2022)
Dzongkha NLP — Bhutan's AI Translation, Speech and Text-to-Speech Tool for the National Language, Built on a Nu 21 Million Budget

Details

Maturity
Pilot
Promoter
College of Science and Technology (Royal University of Bhutan), GovTech Agency, Department of Culture and Dzongkha Development, Druk Holding & Investments, Thimphu TechPark, Omdena
Period
2019–2023 (Digital Drukyul Flagship Programme)
Keywords
language technology, digital government, cultural preservation, citizen services

Context

Dzongkha, Bhutan's official language, is unsupported by major commercial translation engines. Under the Digital Drukyul Flagship Programme, a 2022 Omdena AI Innovation Challenge (~50 volunteer engineers) built the first Dzongkha-to-English machine translator prototype.

Activities

The prototype fed into Dzongkha NLP, an Android app launched 9 August 2023 by the College of Science and Technology and the Department of Culture and Dzongkha Development, combining neural machine translation, speech recognition and text-to-speech, on a Nu 21 million (~USD250,000) budget.

Results

Officials reported translation accuracy of only 30-50%, attributed to the small training dataset, with dataset expansion, grammar/spell-checkers and OCR planned; no iOS version had shipped at launch.

Implementation
Indicative cost
Medium (€50k–€500k) — Nu 21 million (~USD250,000) stated project budget
Time to results
Medium (1–3 years)
Staffing & skills
College of Science and Technology (Royal University of Bhutan), GovTech Agency, Department of Culture and Dzongkha Development, Omdena volunteer engineers

Conditions for success

  • expanding the training dataset to raise translation accuracy
  • adding grammar/spell-checker and OCR features

Common failure modes

  • small training dataset limiting accuracy to 30-50%

Commonly funded by

National / regional programmes

Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Attachments

Similar practices you may find useful