16 years after Google Translate, has AI replaced translators?
While the latest developments in AI are worrying writers and artists, the translation industry has been affected by this problem for much longer.
In recent months, artificial intelligences have become the stars of tech news, as they have made impressive progress. Developments that also create concern and anger, because they put to the fore the question of replacing people by robots and AI. The translation sector has been plagued by these questions for decades.
Technologies are changing, so is the labor market
For a long time, machine translation software like Google Translate drew derision for being word-for-word and not understanding context. This changed in the 2010s with the advent of neural networks, which manage to process entire sentences simultaneously to create smoother and more nuanced translations based on context.
The best known neural machine translation software – because it’s free – is DeepL. self-expressive “best translator in the world” on its site, DeepL was created in 2017 by the German company Linguee, known for the site of the same name, a multilingual dictionary that allows you to find the translation of a word or expression by comparing texts in same language DeepL therefore relies on this first site’s database to produce more accurate translations.
This has slightly changed the work of freelance translators, as clients sometimes approach them with automatically translated texts and simply offer them to proofread and correct. A faster and cheaper solution for clients, but frustrating for translators, because sometimes entire paragraphs have to be reworked while paying less. Because translation software is infallible, its ubiquity gives translators the bitter impression that their clients prefer quantity over quality.
Work around the limitations
Since I have been interested in machine translation tools for a long time, I regularly try to push them to their limits in a standard French language expression: “I drank the cup”. Even a child knows that, depending on the context, this phrase can mean I accidentally swallowed water while swimming. However, advanced machine translation software like DeepL is unable to translate it for me non-literally, even when I add the context of a swim. This is actually an idiom that doesn’t necessarily have an equivalent in another language, so Linguee’s comparative text database is of no help here. When the swimming context is defined, it is possible by clicking on drink (bu) to obtain multiple alternative translations. Theirs, strangled (choke). Therefore, it requires human intervention to guide DeepL in the right direction. The site offered me to save it to my glossary drink was translated by choke but not to get the full expression.
“The examples given so far seem harmless, but in critical contexts where the translation must be absolutely accurate, such as diplomacy, the smallest mistake can have serious consequences. For literature and audiovisual, people will always be better at transcribing a certain style, rhythm or feeling. »
For some languages, it also happens that the software makes a detour through English, because it is the language with the most complete database. This sometimes leads to strange translation errors, as teacher-researcher Pascale Elbaz explained last May during the conference “Will neural machine translation replace humans?” » : he gives an example of Chinese text to a calligrapher who also likes to carve seals. sealin English, is seal. sealalways in English, can also mean seal. Thus the French translation describes a calligrapher who likes to carve seals.
The examples given so far seem harmless, but in critical contexts where the translation must be absolutely accurate, such as diplomacy, the smallest mistake can have serious consequences. For literature and audiovisual, people will always be better at transcribing a certain style, rhythm or emotion. Therefore, human translators remain indispensable for many reasons.
On the positive side, artificial intelligence can also be an ally of these translators. This is called CAT, computer-assisted translation. This software allows them to maintain stylistic and terminological consistency, for example by storing translations of specific expressions in memory. While translators may lament that prices are being driven down by artificial intelligence, they are far from being replaced and can become even more efficient thanks to these technologies.
