A Bot Topic: Debate continues over the role AI interpreting has in BSL access industry

A conversation around artificial intelligence’s place in the British Sign Language (BSL) interpreting profession is continuing to develop, sparking difficult and contentious issues around ethics, interpreter numbers and sustainability.

In his latest feature for The Limping Chicken, Liam O’Dell investigates what role – if at all – bots and AI could play in the future of interpreting.

“Hello, I’m Cassie, [a] BSL interpreter helping the Deaf community,” signs the avatar. It appears to me in a video on my Instagram feed earlier this month, and naturally piques my interest. It’s from the company Robotica, and is an artificial intelligence solution to British Sign Language (BSL) interpretation.

Of course, AI has the potential sparks a wide range of scepticism – from the existential and apocalyptic prophecies a Terminator-like machine wiping out humanity, to the more current concern around its intrusion into modern industries such as retail and e-commerce. It was only a matter of time before such a conversation would arrive at its place in the interpreting industry.

A significant intervention came in 2018, when the World Federation of the Deaf and the World Association of Sign Language Interpreters expressed concerns around signing avatars in a joint statement.

“Computer-generated machine translations cannot render culturally appropriate translations as would be provided by live interpretations from a human sign language interpreter,” it reads. “Work in this field has seen great improvements with the image quality and appearance of signing avatars.

“Whilst the technology has progressed and offers real potential for wider use of signing avatars, these computerised products do not surpass the natural quality and skill provided by appropriately trained and qualified interpreters and translators,” it continues. “Individuals who are fluent in a signed language and qualified to present information on particular subjects not only use the hands, arms, shoulders and torso, movements of the head, facial expression and mouth patterns, but also include cultural information where necessary to convey the intended meaning contained within a message.”

Over email, Robotica’s founder Adrian Pickering is honest in his assessment of AI’s future in interpreting. “Humans will always be preferable,” he writes. “Artificial intelligence will never offer empathy and trust.  Nobody wants a computer reading their medical diagnosis or explaining a legal ruling.

“Only when either no human signer is available, or when human signing is not feasible, should our AI signers be used,” adds Pickering. “We do not ever foresee a time when AI will replace human translators and interpreters.”

Meet Cassie, our team-leading virtual #interpreter pic.twitter.com/R5V4iXY2Qz

— Robotica Machine Learning (@roboticaml) September 6, 2022

Though at the same time, the low number of interpreters is no secret, and to many, it could certainly be higher. The current registration figures listed on the website of the National Registers of Communication Professionals working with Deaf and Deafblind people (NRCPD) states there’s only 1,391 sign language interpreters on its register in the UK.

“Few are media trained and fewer still are willing to be on television,” Pickering continues. “Those translators are working flat out, and demand is increasing.

“We talk frequently with just about all the UK’s major TV companies, and they really do desperately want to put BSL everywhere – just as they are with subtitles – but they just can’t find enough people,” he says. “This means that only a small amount of television is being translated. It also limits which shows are offered.”

According to Pickering, AI signing using an avatar offers up new possibilities. “For example, AI can sometimes make translations easier to follow. We can produce digital signers that look like the characters in movies or dramas,” he explains, “or have avatar clothing that changes colour according to which character is speaking – just as often happens with subtitles.

“We are not asking people to choose between human and AI translations. Why would we? Humans will win every time,” stresses Pickering. “We are offering people a choice between AI translations and subtitles. A choice between AI translations and no translations.”

A choice which may not be easy for some Deaf people to make. For Belgian academic Dr Maartje De Meulder – who just a few weeks ago gave a presentation on machine-based interpreting to the European Forum of Sign Language Interpreters (EFSLI) – there is what she describes as a “critical need” to answer some “urgent questions”.

“Who invents the technologies, and what is their motivation for developing them? How are data being collected to make machines learn? Who evaluates the outcomes, and how,” she writes in a 2021 paper on the ethics of sign language technologies published by the Association for Machine Translation in the Americas. “Is there an actual demand from the communities? Who are the end users and who decides that? Who benefits from these technologies, and who is at risk of being left behind?

“What are the current and potential future applications of those technologies,” Dr De Meulder continues. “How will language rights keep pace with the development of language technologies? What are the ideologies behind these technologies?”

It’s this last question which the researcher underscores in the document’s conclusion. “What has been done technologically so far is very promising,” she writes, “but if continued on the same path, there is a risk that technologies developed in the end will not be voluntarily adopted by end users. This uptake in use is important, because the more ‘we’ use AI, the better it will become. There must be a consideration though of who this ‘we’ is – who is the language technology for, and why?”

Sign language interpreters certainly aren’t the ‘end users’ here, argues Dr De Meulder.  “Nor should they be seen as the benchmark for language use. Placing interpreters on the centre of deaf peoples’ lives (a constructed dependence) comes from a biased and hearing- centred view on communication.”

They are, she explains, deaf people, and calls on developers of new sign language interpretation technology to include a “widely varied group” when co-designing such software.

The National Union of British Sign Language Interpreters (NUBSLI) also mentions the end user in their response to a request for comment. “Ultimately it is for those end users to decide whether or not they want to use it,” they write. “We would welcome the use of AI in appropriate settings where it might enable improved or more widely available access for deaf communities. The emphasis here would be on appropriate use, for example, at train stations or in other settings where an AI could be used to provide announcements to deaf travellers.

“Any concerns we might have about the use of AI in our field would primarily be around the potential effect on the ongoing viability of our profession, our members terms and conditions, and the knock-on effect that any detriment to these areas would have on deaf communities,” NUBSLI goes on to add. “If AI is eventually deployed as a means to replace human interpreting and translation, or to provide a cheaper alternative that drives down fees in the industry, we would expect to see numbers in the profession plummet as the career becomes less and less sustainable.

“Ultimately this would harm deaf communities and users of our services, and we would always work to avoid this as much as possible, with the means at our disposal,” their statement concludes.

In the age of subtitles glasses and sign language gloves, new AI tech could soon be added into the mix. As for the part it plays in the community and the debates it will spark, there are – as of yet – no clear signs.

Photo: @RoboticaML/Twitter.

By Liam O’Dell. Liam is an award-winning Deaf freelance journalist and campaigner from Bedfordshire. He can be found talking about disability, theatre, politics and more on Twitter and on his website.