Celtic languages like Cornish, Irish, Scottish Gaelic and Welsh have deep roots in the United Kingdom. To help keep these languages alive and useful in the digital age, the UK-LLM project is creating artificial intelligence tools that understand and work in both English and Welsh. These tools are powered by advanced technology from NVIDIA, such as the Nemotron family of AI models.
Making AI work fluently in Welsh will help people in Wales access important services like healthcare, education and legal help in their own language. The UK Prime Minister says this is all about making sure everyone can benefit from artificial intelligence, no matter which language they use every day.
The UK-LLM project, led by University College London and started in 2023, has already developed two AI models for UK languages. Their latest effort focuses on Welsh and is made through close teamwork with Bangor University in Wales and NVIDIA. This ties in with the Welsh government’s goal to reach a million Welsh speakers by 2050 through the Cymraeg 2050 initiative.
The new Welsh language AI model will be available for developers using an online service from Nscale, a UK-based cloud company. Gruffudd Prys from Bangor University’s language technology center explained that AI can really help both new learners and fluent speakers make progress with Welsh.
For businesses and schools, this model can help translate content and add Welsh chatbot features to apps and websites. This makes it easier for organizations like hospitals, teachers, broadcasters and shop owners to share information in both Welsh and English.
The team plans to use what they have learned to build AI models for other UK languages, including Cornish, Irish, Scots and Scottish Gaelic. They also hope to help create models for languages in Africa and Southeast Asia by working with partners worldwide.
Building this AI model was not easy because there is much less training data in Welsh than in English or Spanish. To solve this, the team used NVIDIA tools to translate huge collections of English data into Welsh, giving the AI more examples to learn from. They used powerful computers, including the Isambard AI supercomputer at the University of Bristol, to speed up the process.
Bangor University brought in experts to make sure the AI really understands the details of Welsh, such as how words change depending on the context. The team checked both the translated training data and the AI’s responses for accuracy and natural language use.
All these efforts will soon make the Welsh language AI model and its training data available to companies, government organizations and researchers. This means more apps and services in Welsh and more chances for the language to thrive.
The methods and technology used for this Welsh project can help build AI for other minority languages anywhere in the world. NVIDIA’s Nemotron models and tools are openly available, so developers can make AI that works for many languages and purposes. From learning support to business and public services, AI is becoming a powerful way to support language communities—making technology more inclusive for everyone.
Original article and image: https://blogs.nvidia.com/blog/uk-llm-nemotron/
