A new language to ensure peace in the world. This is the iconoclastic project carried by Louis-Lazare Zamenhof, father of Esperanto. This Pole, born in 1859, was convinced that language barriers were tearing humanity apart. And that a universal language would prevent many conflicts. As candid as it was inventive, the initiative did not succeed in becoming massively democratized. No more than other similar enterprises, from “interlingua” to “solrésol” via “volapük”. But in 2024, artificial intelligence is about to succeed where these crazy projects have failed. For twenty years, tech giants have been sweating to create a universal translator. We have to admit that they have made immense progress.
Gone are the crude translations, where “word for word” emptying sentences of their meaning. Over the years, groups such as Google have refined their techniques in order to take into account the context in which an element appears. They now even manage to translate one language into another, without any example of correspondences – which was the main pebble in their shoe.
Today, DeepL’s finely-tuned text translations attract cohorts of students and workers every day. And the German unicorn is working on a real-time voice translation tool. “Voice is a bigger challenge. There are more variables to manage: accent, ambient noise, microphone quality… It’s a more complex basis than text, but we have very good results,” says David Parry-Jones, the company’s revenue manager.
OpenAI, Google, Meta and others are also transforming their tools into all-terrain vehicles capable of juggling between written and spoken language, with a multitude of languages. Bringing us at full speed closer to these fascinating “universal translators” that populate science fiction works.
The boom in AI translation
The American online training giant Coursera is a good example of the scale of the change induced. “We used AI to translate 4,000 courses into 17 different languages. Previously, it cost $13,000 to have one translated into one language. Today, a subscription of around twenty dollars is enough,” its CEO, Jeff Maggioncalda, revealed to us last November.
The AI translation market is booming. According to Statistait should be around 7 billion dollars in 2024 and could exceed 27 billion by 2030. This will obviously transform the translation professions. But it would be foolish to think that this sector will be the only one to be disrupted.
With these tools, a host of companies will be able to tackle new foreign markets. The downside – that’s the game – is that they will soon see new competitors from elsewhere bursting into their backyard.
Tomorrow, all merchant sites will be able to launch versions in English, Mandarin, Spanish or Hindi. The new generation of AI allows them to translate vast catalogs of product sheets much faster, at almost no cost. But above all, to translate in real time all the modules that really bring these platforms to life: consumer reviews, or even “chats” allowing sellers to be questioned. Not to mention the strategic customer service, which must imperatively understand users’ complaints in order to provide them with the appropriate solutions.
“All multinationals need translation to function,” points out David Parry-Jones of DeepL. For tourism professionals, too, it’s an old dream come true: communicating with these enthusiastic but somewhat lost tourists. In Japanese transport, ticket counters are starting to be equipped with translation screens. liveExperts are already dreaming of the help that these types of tools could bring to Japanese taxi drivers, less than 1% of whom speak good English.
The music industry is not left out. Universal Music Group recently partnered with SoundLab to provide artists with several AI-backed tools, including one that realistically translates their voice and lyrics into multiple languages. The same trend is true for influencers. With HeyGen or Synthesia, they can translate their videos into multiple languages, automatically adapting lip movements for greater realism.
On social media, AI translation may unfortunately do some damage by helping disinformers speak to many more people. Fortunately, it will “redeem” itself by allowing Internet users to access more information – half of the Web is indecipherable to those who do not speak English – and by helping scientists around the world share questions or findings.
“AI is also very interesting in the manufacturing sector,” observes David Parry-Jones of DeepL. This is not surprising. Automotive, household appliances… All manufacturers who sell worldwide have a lot of technical documents to decline.
English “dethroned” by artificial intelligence?
“With AI-powered translation, the geographical boundaries of recruitment will be shattered,” points out Jeff Maggioncalda of Coursera. “Increasingly, tomorrow’s workers are being born in Africa and South Asia. Added to this is an expanding range of online training and the rise of teleworking. Artificial intelligence has removed the last obstacle to international recruitment: language barriers.” English may not be the lingua franca of the business world.
Despite their impressive progress, translation AIs must be handled with caution. First, because not all of them offer the same level of security and confidentiality. Free tools often keep the texts submitted by Internet users in order to improve. At DeepL, the “pro” version offers an adequate level of protection for corporate documents. “Our paid product does not keep any data, it deletes all translations, and all exchanges are encrypted,” says David Parry-Jones, noting that even the Swiss government has been tempted.
“Compared to experienced human translators, the limits of AI are obvious,” he also recalls. a ticket from Xiao Weiqinghead of the Department of Translation and Interpretation at Shanghai University. Ideological nuances, cultural expressions, colloquial language confuse them […] They lack the linguistic sensitivity to make subtle distinctions, nor the creativity required for artistic texts. They produce generic results, which lack personality and sparkle.”
And when it comes to sensitive texts, even DeepL is unequivocal: “As soon as a document engages your legal responsibility, human verification is required. But the fact that the AI has prepared the basic version saves a lot of time,” assures David Parry-Jones.
In this strange new world, the key will ultimately be to learn how to use these new digital companions properly. If an exchange takes an unexpectedly unpleasant turn, systematically check whether it was initiated by the man or the machine. For world peace, there is indeed a worse threat than the inability to understand each other: wrongly believing that someone is picking on you.
.