On 7 and 8 April 2025, around 30 participants gathered in Brussels to discuss the impact of artificial intelligence on sign language and accessibility for deaf people.
Organised by the EUD (European Union of the Deaf), the event ‘Sign language in the age of artificial intelligence: what digital rights and ethical dilemmas?’ brought together researchers, deaf teachers, interpreters, and professionals from the sector to assess the risks, opportunities, and ethical issues associated with the development of AI in this field.

From technology to technology with

The debates were moderated by two renowned deaf researchers: Maartje De Meulder (University of Utrecht, Netherlands) and Robin Angelini (University of Vienna, Austria). Their conclusion is clear: AI technologies are still largely designed without deaf people in mind, sometimes even against their interests.

They denounce an audist approach, centred on hearing people, which produces tools unsuited to the real needs and linguistic practices of deaf signers. They warn of the risk of linguistic subordination to technology: when tools are developed without deaf communities, technology tends to freeze a single norm, to the detriment of the diversity and richness of sign language.

The two researchers also addressed the concept of techno-solutionism—the belief that technology can solve all social problems. This approach is dangerous if it ignores the structural causes of inequality and the realities experienced by minorities. The EUD has highlighted the danger of replacing human translation with avatars in official videos: behind the immediate accessibility, we lose human, linguistic, and cultural quality.

However, they emphasise that AI can also become a real lever for accessibility if it is designed with and for deaf people, for example, for website translation or subtitling.

When accessibility becomes a market

Given the high costs of human interpreters, AI technologies are attracting increasing public and private funding. Here are a few examples:

•  The French SignToKids project, funded by the National Research Agency (ANR) (€543,860), aims to develop digital resources in LSF for children, with signing avatars. It raises questions about linguistic quality and the role of deaf teachers.
•  The Austrian SiMAX project, funded by the EU (€1,094,360), is developing a 3D avatar to automatically translate texts into sign language. Promoted in Europe, it is presented as an economical alternative to human interpretation in public information videos.
• *SignON, a project funded by the European Horizon 2020 programme (support for research and innovation), is coordinated from Ireland. It aims to develop an automatic translation application between several sign languages (LSF, BSL, ISL, etc.) and the corresponding spoken languages. The project is ambitious, but raises doubts about its feasibility, as sign languages are visual, spatial and strongly culturally rooted.

But at what cost to the linguistic rights of deaf people? Without real participation from the communities concerned, automation risks producing standardised solutions that are far removed from actual usage and the diversity of sign languages. This would result in superficial accessibility, which could ultimately weaken rather than strengthen these rights.

The EUD’s position: the same caution for all languages

The EUD has highlighted an inequality in the use of AI: for spoken languages, it is designed as a tool to support interpreters (note-taking, transcription, terminology assistance); for sign language, it is often seen as a replacement.

This difference reflects unequal treatment: on the one hand, the human role is valued, while on the other, the aim is to reduce costs. Thus, signing avatars are proposed to replace LSF interpreters, to the detriment of the linguistic quality and relational dimension of interpretation.

The EUD pointed out that:

• AI still lacks the linguistic precision and cultural sensitivity to guarantee accurate interpretation in sign language.
• Current solutions, which are poorly calibrated, can hinder access to vital information.
• The same level of caution applied to the spoken word must be applied to the visual.

The EUD intends to continue its work with European institutions to regulate the development of AI and defend accessibility that respects sign languages and their users.

efsli: between vigilance and building solutions

The European Forum of Sign Language Interpreters (efsli), which brings together sign language interpreter associations across Europe, has presented its recommendations on the development of artificial intelligence in the field of accessibility.

The organisation emphasises the importance of guaranteeing the right to human interpretation, particularly in sensitive areas such as health, education and information. It also encourages supporting technological developments by creating new opportunities for deaf interpreters, for example in the fields of ethics, quality and technological mediation. However, efsli warns of a risk: that certain tasks, such as museum translations or public announcements, will be entrusted to avatars at the expense of deaf interpreters.

‘Interpreting is not just about translating words. It’s about connecting worlds, emotions and cultural references. No algorithm can do that,’ emphasised the efsli representative.

What next?

This seminar did not end the debate: it opened it. Other meetings are planned to explore these crucial issues in greater depth. Because one thing is certain: innovation is only worthwhile if it protects the dignity and rights of deaf people.
At a time when algorithms are penetrating all areas of life, it is more necessary than ever to sign, think and create differently.
To learn more about sign language avatars, check out our article published on the Deaf Journalism Europe website: Sign language avatars: Germany’s gold rush to reduce the number of ‘skin’ interpreters. It looks at the history, methods used and concrete issues, based on the German example.

 

Written by Charlotte Berthier and presented by Laura Guernalec from Médiapi