Translation and the Machine
How the European Parliament’s linguists are adapting to the brave new world of translation technologies.
The European Parliament (EP) is the most multilingual government institution in the world, by a long shot. It operates in 24 official languages on a daily basis—four times more than the United Nations, with its 6 official languages. This multilingualism is considered a fundamental part of the democratic process in the EP. The idea is that every member state should be able to speak and be spoken to in its own language to ensure the equal representation of all 28 members.
“I think language is very much tied up with identity and democracy in the EU,” says Róisín Abbott, a public relations officer for the EP’s translation service. “It means that citizens have access to documents in their own language. They can petition the Parliament to look into any area of concern in their own language. Multilingualism is democracy, ultimately.”
By law, this means that Members of Parliament (MEPs) have the right to speak in any of the EU’s 24 official languages. During sessions of the Parliament, every word spoken by an MEP is interpreted into all 23 other languages by the EU’s Interpretation Department.
For the EP’s Translation Service, which deals with all the written texts and legislations, this means every document coming out of the Parliament must be published in every official language. This can be a fairly straightforward process for a standard French to German translation, for example. But when you consider that there are 552 possible language combinations between the EU’s 24 official languages, things can get complicated rather quickly. How many people do you know who can translate from Spanish to Swedish, for example? From Czech to Portuguese?
Want to keep reading?
This story is free! But if you want to support us, you could spoil yourself with a printed version of this story.
As it turns out, the EU does not hire 552 specific translators for each language combination. Instead, it uses what it calls a “relay system” to manage its more obscure translation requests.
Under this system, the EP’s three most commonly spoken languages—French, German, and English—act as the middle men between more difficult language combinations. The idea is that the EP is equipped to translate all 24 official languages into at least one of these three relay languages. From there, the text can then be re-translated into any one of the other 23 languages.
For instance, if material must be translated from say, Maltese into Croatian—“though that’s probably not as rare as it might sound,” the EP’s translators are quick to point out—the Maltese text will first be translated in German, and then from German into Croatian.
“Multilingualism can be a complicated thing,” a senior translator for the EP explained. “We have to translate back and forth between all these language combinations. It’s hard! It’s very difficult to find, let’s say, Greeks who can speak Estonian, and it’s difficult to find Portuguese people who can speak Lithuanian. Concentrating on three main languages is just a pragmatic approach to a huge problem.”
The relay system is an invaluable battle strategy for EU translators. But like other European institutions, the EP is also relying increasingly on new technologies to assist in its daily operations. Since 2009, the EP has consistently used an integrated Computer Assisted Translation (CAT) system to process all of its translation requests. The CAT tool combines a patchwork of digital resources to help translators in their daily tasks. One of the most useful of these is the system’s “translation memory depository,” a massive virtual library of all of the translation department's previous work.
“The CAT tool goes back to our repositories, recovers all previous translations that might be useful, and packages them for translators and translation assistants to have available when they work,” said Paula Mărginean, the Head of the EP’s Applications & IT Systems Development Unit. “We re-use as much as possible from our previous documents.”
This allows the translators to reduce typing, save time, and focus their energies on the more cognitively challenging parts of the job.
The Parliament also has its own machine translation software incorporated into its CAT system. This software, developed in-house by the European Commission and shared between all EU institutions, uses data fed into it by programmers to produce computer-generated results. Think of it as Google Translate, but made specifically by and for the EU.
Though machine translation can be helpful when there are no previous human examples to pull from, it doesn’t always provide the most accurate results—and all language combinations are not created equal. For example, while French to Spanish translations get a green light, translations between English and Hungarian are particularly difficult for the machine.
“It took the Commission team a lot of time to train the engine to get good results there,” Mărginean said. “It’s a matter of differences in the structure and syntax between the two languages. The more they are different from each other, the poorer the quality of the machine’s results.”
There are also certain types of texts that the Parliament prefers to leave to its human translators—like political speeches by Members of Parliament. These tend to be full of metaphors, jokes, and references that the machine can’t handle quite as well as a human can.
“For instance, a cultural reference is not the same in all languages,” Mărginean explained. “You need to have the background and the understanding of that perspective and to be a very fine connoisseur of that culture to be able to translate those into your own language. That is where the creativity of a human is irreplaceable.”
Still, there is no denying that these technologies have made the position of translators more precarious than it was only a decade ago. The EP admits that the CAT tools have allowed it to cut its translation staff in the last few years, despite having an ever-increasing amount of content to translate. But the EP doesn’t foresee a world without its human translators—at least not yet.
“In 2006, when I did the examination to become an official at the European Parliament, one of the questions I was asked was: ‘Are you afraid that you will be replaced by machine translation?’” Mărginean said. “At the time, I said, ‘Not at all, because a good translator needs to use his or her mind and soul to do a good translation. Knowing the language is not the only thing.’ And I think that answer is still valid today.”
This article appears in Are We Europe #5: Code of Conscience