Welsh Government AI summary: bilingual transcription and more Welsh data

The Welsh technology and artificial intelligence summary 2025–26 outlines new partnerships and continued investment in Welsh infrastructure.

Four people sharing ideas in a colourful newsroom

Published the full summary today as part of the technology programme for Cymraeg 2050. The main points:

Partnership with Microsoft

The summary describes collaboration with Microsoft. Microsoft Teams can now transcribe Welsh and English simultaneously — an ability the Welsh Government describes as the first of its kind for any language pair — and users can give Welsh instructions to Copilot. The Geraint and Gwyneth synthetic voices, developed with RNIB, are also freely available through Windows assistive technology.

The recent Coleg Cymraeg Cenedlaethol report also noted that the Government has a data-sharing partnership with OpenAI. This suggests that the priority established in January 2025 — improving Welsh AI through data sharing — is now being implemented with more than one major lab.

Grants building infrastructure

Iriaith (Bangor University, renewed for 2026–27): around 500,000 Welsh–English sentence pairs have been collected, and the work includes machine translation, large language models, Y Trawsgrifiwr, text-to-speech technology and the Macsen digital assistant.
Mapio Cymru (Data Orchard): over 30,000 Welsh place names in OpenStreetMap.
Wici-Fflics (National Library of Wales): over 1,000 new Wikipedia articles.
CEFR Analyser (Cardiff): a tool to assess the language level of learning materials — already producing the dataset welsh-cefr and two open classification models.

The Commissioner: ‘technology is no excuse’

This work takes place within the regulatory framework established in the Welsh Language Commissioner’s AI policy statement in July 2025. The statement sets out four principles: linguistic fairness; accuracy and naturalness; data security and privacy; and continued compliance with Welsh language standards. It also emphasises that using these technologies is no excuse for failing to comply with Welsh language duties.

What does this mean?

Compared with the lack of Welsh data from international labs, Wales’s strategy is clear: Welsh’s presence in AI depends on good-quality data, concrete partnerships and long-term research. These projects build the foundations rather than waiting for the wider market to do so.

Sources: Welsh Government 2025–26 summary (Welsh and English); Welsh Language Commissioner’s policy statement (22 July 2025); UWTSD/Coleg Cymraeg Cenedlaethol report 2026.