Welsh technology and artificial intelligence (AI): summary 2024 to 2025

Overview

We've been working hard over the last few years to create a lot of things to make it easier to use technology in Welsh.

Our priorities are:

• improving technology to increase the daily use of Cymraeg
• making sure everyone can access Welsh language technology
• improving Welsh language artificial intelligence (AI) and speech and language technologies (by sharing data and other means)

Here's a summary of our work for 2024 to 2025.

The latest is available at Welsh language technology and artificial intelligence (AI).

A report of what we've achieved so far is available at Welsh language technology action plan: final report.

New priorities

In announcing the Welsh language technology action plan in 2018, our aim was to create technological developments that would enable us to use more Welsh in a wide range of situations. In February 2024, the then Minister for Education and Welsh Language published the final report on this Plan, looking back at our progress, including in the area of artificial intelligence (AI).

We then launched a call for information asking for views and ideas on the next steps for Welsh language technology. This call for information closed on 1 October 2024, and a summary of responses is available. A huge thank you to everyone who sent in their ideas!

On 7 January 2025, the then Cabinet Secretary for Finance and the Welsh Language, issued a written statementoutlining 3 new priorities for our work in the field of Welsh language technology:

• technology as a way of increasing the daily use of the Welsh language
• ensuring that everyone is able to access Welsh language technology
• improving artificial intelligence (AI) and speech technologies (through data sharing and other ways)

By building the infrastructure for Welsh language technology in this way, we're encouraging people to use more Welsh, which helps us reach our goal for Cymraeg 2050: A million Welsh speakers.

We're also supporting the Digital Strategy for Wales by developing digital tools and technologies that enable people to access better public services in Welsh.

Partnerships

The Welsh Government announced a new data partnership with OpenAI in June 2024 to improve how AI technologies work in Welsh. During the partnership, we shared a number of resources and components that we as Welsh Government have funded, and we continue to release the technology we fund for free. And this is available for anyone to use, including OpenAI.

As part of the partnership, and as a joint project with the National Library of Wales, OpenAI used AI technologies to transcribe Welsh handwriting. The results are available.

Microsoft has announced that they are now supporting Welsh and Catalan in Copilot . This means that you can now use prompts and questions in Welsh and that Copilot learns how to respond yn Gymraeg. As part of our partnership with Microsoft, we've already collaborated to create a simultaneousinterpretation facility in Teams, and our guidance is available. Work with Microsoft continues to develop this further, and we will now also work with them to improve how Copilot processes the Welsh language.

Technology for specific needs

An important part of our next steps in Welsh language technology will be technology to help people with specific needs. This could include disabled people, older people, people with additional learning needs, or people with needs resulting from a medical treatment, a cognitive or a physical impairment.

For example, Welsh speakers at risk of losing their voice due to medical reasons such as Motor Neurone Disease or throat cancer can now continue to communicate in Welsh, thanks to the Lleisiwr service. The service is financed by the Welsh Government through the Welsh language technology action Plan and developed by Bangor University.

Another example is the Welsh Government partnership with Tobii Dynavox, a leading AAC (augmentative and alternative communication) technologies company, which has led to TD Snap Express Cymraeg. TD Snap Express is designed to be simple and easy to use, focused on communication intent rather than on grammar, to support both children and adults in developing their communication and language skills.

This work isn't only so that people can use the Welsh language but also so that they can use it easily, without additional challenges.

Technology grants

The 2024 to 2025 technology grants helped build the infrastructure, data and resources needed for the Welsh language in the digital age. From mapping place names and expanding Welsh content online, to developing AI models, speech technology and new open data resources, the projects have strengthened the Welsh language technology ecosystem in practical ways. While technology and AI priorities were not announced when these grants were awarded, much of the work done supports the same strategic direction.

Here are some highlights:

Iriaith (Bangor University)

By creating open data, developing new models and building tools that people can use on a daily basis, Iriaith has strengthened many of the key foundations needed for Welsh language technology and AI.

• It collected and published over 440,000 parallel Welsh-English segments under an open licence to support research and development of Welsh language technology (OGL Wales data).
• It created a Welsh version of the Llama 3 AI model, helping to build an open Welsh AI capability (Llama 3 Welsh).
• It made significant progress in speech recognition and Welsh transcription (Trawsgrifiwr).
• It developed practical tools such as Macsen and new text-to-speech services that help more people use Welsh in everyday life and work (Macsen, TTS Cymru, Trosleisiwr).

Want to know more about the work of the Language Technologies Unit at Bangor University? Here is a link to the Welsh National Language Technologies Portal.

Mapio Cymru (Data Orchard)

Mapio Cymru helps to make the Welsh language more visible in the digital world by ensuring that Welsh names are used on maps and services that are used by people every day. The work also shows how open data can strengthen the presence of the Welsh language online.

• The project helped increase the number of Welsh names on OpenStreetMap to over 30,000.
• It continued to support the Welsh language tier on an official Welsh Government map (Data Map Wales).
• It developed and promoted new Welsh language mapping services for the public (OpenStreetMap Cymru, Mapio Cymru).
• It worked with communities across Wales to encourage more people to contribute open Welsh language data.

Wici, AI and Place Names (National Library of Wales)

This project has shown how open data, Wiki and artificial intelligence can work together to increase the amount of Welsh language information available online. It has also strengthened the links between important Welsh language databases and the organisations that use them.

• Welsh labels were added to 11,264 items and published via Wikidata and SNARC (SNARC Fund).
• New methods of using AI and entity identification to detect Welsh place names in digital texts have been developed and tested (Welsh NER Report).
• The work supported the development of the Welsh Language Commissioner's Standard Place Names website and collaborated with Mapio Cymru and other stakeholders.
• A History of Wales Hackathon and editing events were held to expand Welsh language content and data online (History Hackathon 2025).

A Small Welsh Language Model Pilot for Sentiment Analysis Testing (Cardiff University)

This project has addressed an important gap in Welsh-language artificial intelligence infrastructure by creating the first resources to support sentiment analysis in Welsh.

• SENTimental was created to collect and label data for training AI models.
• Over 109,000 Welsh words were reviewed and over 4,500 responses were collected to create new training data.
• The model showed about 80% agreement with human marks within a single scoring level.

The resource was released as open software for other researchers to build on. (SENTimental GitHub).