The EU Commission is building, among other things, on AI-controlled translation programs. They are getting better and better – and the number of EU translators is falling. But to what extent can machines replace humans?
They ensure that Europe understands each other: interpreters translate speeches such as those by Commission President Ursula von der Leyen into the 24 official languages of the EU. Translators translate bills so that governments and MEPs know what they are passing.
Alongside the United Nations, the EU is the largest language project since the Tower of Babel. The Commission’s translation service alone has over 2,000 employees, two-thirds of whom are women, most of whom are based in Luxembourg and Brussels. But there are fewer: 18 percent over the past ten years – also because artificial intelligence, self-learning computers, do the work.
Machines cannot completely replace people, emphasizes the responsible general director Christos Ellinides. “If the translation into one of the 24 official languages is incorrect, the implementation of legislation in one member state may be different than in another.” So the human element is needed to control the quality of what the machines are outputting. “I am firmly convinced that we cannot simply create legal texts with the help of AI or machine translation, for example.”
EU Commission relies on artificial intelligence
Ellinides’ general management did not fire anyone, but not all who retired were replaced. The EU Commission has been building on machine translation for a long time, and for four years on neural networks – i.e. structures that are modeled on nerve cells to evaluate data and find patterns. The latest achievement is an AI-controlled program that summarizes extensive texts on a maximum of three pages. The more data there is, the better the machines become.
So when translators feed the databases, they practically get rid of themselves, complains Marion Thur, who worked as a translator for the Council – the representative of the member states. “As we translate more and more, we’re expanding the corpus of what exists, and then we’re becoming more and more redundant afterwards. I find it a great pity that the profession is basically abolishing itself and that we’re contributing to it ourselves.”
“We are becoming less and producing more”
Technology doesn’t take away our work, it changes it, explains Mads Nyegaard Outzen from Denmark. He has been a translator in the EU Parliament for 20 years. “And that also enables us to work faster, to work better and of course we can also save resources, that’s very clear.” One big change is that translators are now more likely to revise texts than they used to translate. “Now there is a suggestion from the computer that we can revise.”
We’re getting fewer and producing more, says translator Outzen, and machines help with that. Last year, 2.9 million pages were translated in the European Parliament – a stack as high as the Eiffel Tower.
People always hit the tone better
Parliamentary interpreter Andrea Seidenstücker also uses computers, which are constantly evolving. Meanwhile, they also display proper names and numbers on the screen during a speech, which is a great help. Also: “If you get a speech ten minutes before someone comes to the lectern, it is now possible to send it through a machine translation system. Not to read afterwards what the machine has produced, but to get a basis to have.”
The staff of the parliamentary interpreters was also reduced, says Seidenstücker, but not because of AI – money is scarce everywhere. She doesn’t care about her job. In EU politics, the tone makes the music and people would always hit it better than machines. “Whether a politician is from the government or from the opposition party, i.e. which political direction he or she belongs to, plays a very important role in how I have to express myself,” explains Seidenstücker.
And also which topic is addressed when and how: “It makes a difference whether it’s a small working group or whether it’s the commemoration event for the Soviet deportations from the Baltic States – because you simply have to use a completely different tone. So far, AI can even do that not.”