At Dublin summit, two translators tell how AI re-shaped their craft

On Saturday morning, Fionnuala Ní Bhrógáin stood at a white podium that bore the logo of the Communication Workers’ Union (CWU).

She spoke in a spacious conference room at the offices of CWU on North Circular Road, filled with rows of red folding chairs holding bags of union swag. Two orange sofas sat on the stage behind a round white table. 

Ní Bhrógáin, head of organising and campaigning for CWU, told the crowd, as they trickled in, that they were invited because “a very large and very well-funded” conversation about the future of artificial intelligence (AI) was set to happen next week.

She was referring to the International AI Summit, slated for Wednesday at the Royal Dublin Society (RDS). It’s to be hosted by the Irish government as the current holder of the EU Council presidency. 

“We know what the conversation is going to be. It's going to be competitiveness, productivity, growth, and that the AI revolution is happening, and that it's inevitable,” Ní Bhrógáin said. 

That’s why it was salient to give voice to workers at an alternative AI summit, she said – bringing their experiences out of the shadows.

Then she called Trish Van Bolderen and Nick Bennett to the stage. They’re both seasoned translators.

Van Bolderen told Bennett of the gradual downgrade of translators' role since the AI surge. How some clients now want skilled translators to revise and proofread the machine's work, leaving those who can’t afford to say no disillusioned and exhausted.

It's a picture familiar across Europe. “‘Value-for-art-extraction’ paradigm, fostered by [automatic translation] tools, redefines translation from a creative, interpretive practice into a technology-driven support function oriented towards speed, efficiency, and market competitiveness,” said a recent paper published in the peer-reviewed journal Work and Occupations.

Van Bolderen and Bennett also talked of AI's slow crawl into the world of interpretation.

Their chat was just one of the conversations at the trade union's two-day summit, where workers compared notes, queried researchers, ate sandwiches from Rathmines' Middle-Eastern joint Shaku Maku, and even made protest art. 

Something changed

Van Bolderen, who specialises in French, English and Spanish, currently chairs the Association of Translators & Interpreters Ireland’s (ATII) executive committee. 

Bennett, who is skilled in carrying French and Japanese over to English, recently spent two years working as a generative AI data annotator – which involves labelling training data to improve the performance of large language models (LLMs) – for Covalen, part of outsourcing firm CPL. 

There, he helped boost the safety of “Meta's Llama LLM”, his website says.

Van Bolderen, who is tall and slim, wore a black blazer over a mint-green blouse. Bennett, with a shaved head, wore black-rimmed glasses, a faded-blue shirt, and jeans. 

He asked Van Bolderen to pinpoint the moment she realised AI was becoming more commonplace.

It was around 2022 or 2023, she said. Van Bolderen was translating a literary novel. “And I thought, ‘What would it look like to use?’”

She wasn't particularly excited, Van Bolderen said, but she was vaguely curious. At the time, the AI-powered tool in vogue for translation was DeepL, she said. 

She took out a paid subscription hoping that would keep her data safe, Van Bolderen said. “If you pay, there’s at least an illusion that no one’s scraping your data.”

So, she gave the tool a few pages to work on, just to see how it would fare. 

“And it didn't take very long, but it did take enough time to realise I was wasting my time to be revising the translation that had been given to me through DeepL,” said Van Bolderen.

And don’t get me wrong, she said. The work read alright.

As with many AI flops, Van Bolderen said, the models’ fluency can make shortcomings fly under the radar. “It can trick you into buying what’s being presented,” she later said.

The AI-translated pages she got? People didn’t talk like that in English, Van Bolderen said. 

“And then I got to spend time decontaminating that sentence, which takes so much time, and it's very unpleasurable,” she said. 

“In the context of literature, despite the swift progress of AI technologies, ongoing challenges persist in accurately conveying the subtle cultural and emotional nuances found in literary works,” said a recent book, AI and the Future of Translation: Trends and Challenges.

These days, a growing cohort of clients see the translator's job as just that, cleaning up after the machine, said Bennett and Van Bolderen.

"It's very common. When we talk to members in our association, we do hear that it's quite common for people to do that,” she said.

For Bennett, he said, noticing that trend was the first time it hit him that something had shifted. 

All the things we said

Van Bolderen, who teaches at Dublin City University (DCU), said she'd only revise machine-generated text in a classroom setting.

She’s lucky, she said, that teaching gives her the freedom to afford turning down work that doesn’t bring her joy – like cleaning up AI-made prose.

“I would say no, I'm so sorry, I'm not doing that work. There are lots of people who don't have that choice,” said Van Bolderen.

They’re agency workers, who have to say yes to everything or else risk losing their jobs, she said. 

And she knows ATII members who drifted away, or retired early, or were students who had dipped into the industry just for a couple of months and then left, she said.

“They say, ‘I can't do this anymore. I'm changing course because there's nothing. I love it, but there's nothing,’” Van Bolderen.

As for ATII members who work in interpretation, AI adoption hasn’t yet upended their jobs, Van Bolderen told Bennett. 

That said, she said, it's set to become commonplace, especially in court settings.

In March, Owen Harrison, the chief information officer for the Courts Service of Ireland, told the Dáil's Joint Committee on Artificial Intelligence of ongoing “proof of concept” efforts for a real-time translation service.

“It is hoped that we will be able to give members of the Judiciary and the courts options for moving cases along in respect of small procedural matters that need a translation capability in the room that is not immediately available and could otherwise lead to adjournments,” Harrison said.  

Even before the dawn of AI, ATII had raised concerns about the quality of human interpretation in important places like courts and asylum hearings.

Now, this new approach is what some judges want to see, Harrison said. It won't be rushing the roll-out, Harrison said. “We will go through it and test it thoroughly,” he said.

AI interpretation is not the only way crucial matters affecting immigrants can be outsourced to a machine.  

Paris Marx, a Canadian tech journalist and author, also told the workers' AI summit last week how the Canadian government had been using it to process immigration permit applications. 

He pointed to how at one point an AI agent had hallucinated some details and rejecting one woman's permanent residency application based on the made-up stuff.

Despite all that, Van Bolderen and Bennett ended their chat with a buoyant message.

She keeps stressing to her students the importance of staying the course right now, Van Bolderen said.

Some make AI out to be “future-oriented” and forward-looking, she said, when it was made by swallowing what has been said and recorded already then repeating that again and again. It is peering back into the past, said Van Bolderen.

“It's never going to be right there where things are happening, and it's not going to be looking ahead,” she said.