Multi-Million Dollar Grant Awarded to Create SignGPT for AI Sign Language Translation

The UK Engineering & Physical Sciences Research Council (EPSRC) has awarded a GBP 8.45m (USD 10.5m) grant to three universities to build a foundational large language model (LLM) for sign language.

The three universities — the University of Surrey, the University of Oxford, and University College London — will embark on a five-year project to create “SignGPT” to allow spoken languages to be automatically translated into photo-realistic sign language, and vice-versa.

The dataset is expected to become the largest sign language dataset in the world, giving the breadth of application to the Deaf community that current LLMs provide for written/spoken languages.

Richard Bowden, Professor of Computer Vision and Machine Learning at the University of Surrey, told Slator that this will be achieved through “both text-to-sign and sign-to-text. […] There are lots of ways to achieve this which we are exploring, so it’s difficult to pick any single candidate technology. But rest assured it will be photorealistic, not an avatar.”

On the question of measuring quality in sign language output with AI, Bowden told Slator, “That is one of the hardest research questions we currently face. Text is symbolic and standard metrics already exist. For sign-to-text we can use similar metrics, but trying to measure the performance of text-to-sign is more difficult. At the moment we rely heavily on human judgment, but we need better automatic metrics.”

Slator Translation as a Feature (TaaF) Report

The Slator Translation as a Feature (TaaF) Report is a vital and concise guide on how AI translation is becoming an integral feature in enterprise technology.

$260 BUY NOWIncluded in our Pro and Enterprise plan.
Subscribe now!

Impact on Commercial Solutions

Bowden, who is also Co-Founder of AI sign language provider Signapse, clarified that the EPSRC grant “is for academic use only.” Signapse — which is 5% owned by the University of Surrey — has already secured USD 2.4m in seed funding last year.

“We will endeavor to open source and publish as much of the research as we can for non-commercial use. […] Signapse represents one of the advisory partners as they have an interest in the technology developed. They would also be a potentially good exploration partner for the research but that would involve future license discussions,” he added.

While the research focuses on generating output for British Sign Language (BSL), Bowden is optimistic about future scalability to other sign languages. “Most of the challenges come from the grammar and use of space, so once we have solved translation at a fundamental level it should be relatively easy to change the lexicon and move to other languages.”

“Once we solve the problem for one sign language, we solve the problem for all sign languages,” he concluded.