What is SignGPT and who is involved?
SignGPT is a five year research programme, funded by the Engineering and Physical Sciences Research Council (EPSRC). It started in May 2025. It is a collaboration between Surrey University, Oxford University, and DCAL at University College London. The Surrey and Oxford teams are specialists in AI and machine vision, while the DCAL team bring their skills in sign linguistics and sign language corpora. The project is overseen by an Advisory Board, whose members include organisations such as the British Deaf Association (BDA) and Royal Association of Deaf People (RAD), as well as deaf and hearing academics from around the UK and abroad.
What is SignGPT working on?
In the long term, we want to create AI chatbots for BSL and other sign languages. This would be something like ChatGPT or Gemini, except that you would ask questions directly in BSL, and get a response from a generated avatar that looks and signs like a real person. For example, you could use it to produce a short BSL video summarising a longer BSL video or a text web page. However, to get to that point, there are two major problems to solve:
- Teaching computers to recognise BSL from video input (e.g. a webcam)
- Producing realistic and believable signing on the screen (so a computer can reply in BSL without using English at any stage)
To solve these problems, we need models based on high quality sign language data. But to get that high quality data, we first need better visual language “tools”.
What kind of tools do linguists need?
Historically, turning sign language videos into data has been a very slow and intensive process. For example, even after many thousands of hours of work by deaf and hearing linguists, only about 15% of the BSL Corpus at DCAL has been analysed. To make that work go faster, the project is working on tools that will speed up tasks like these:
- Recognising where individual signs begin and end
- Recognising which sign is being produced, based on a specialised dictionary
- Recognising meaningful facial expressions and body movements
At the end of the project, tools like this will be published, so anyone will be able to use them.
What about regional variation in BSL, and different signing styles?
These are important issues. If we want high quality sign language AI, we need high quality data that is diverse but also well organised. Simply scraping random videos off the internet is not good enough. SignGPT prioritises BSL video data which represents different UK regions, different age groups, and different ethnic backgrounds. We also make a distinction between deaf and hearing signers, and between spontaneous and interpreted signing.
Are you trying to replace human interpreters?
No! This kind of technology is not a replacement for human interpreters. We believe that human interpreters are essential, especially for complex, sensitive, or high-stakes communication. This project focuses on everyday low risk situations where an interpreter isn’t practical. We think that people should be free to choose whether or not they want to use technology to help with everyday communication.
How are deaf people involved in this project?
Deaf BSL users are involved at every level, not just as participants. They are:
- Members of the research teams
- Members of the Advisory Board
- Central to planning, designing, testing and evaluation
The project is built on the idea that technology for BSL must be co-created with the Deaf community, otherwise it will not be trusted or useful.