{"id":39761,"date":"2015-01-09T22:01:55","date_gmt":"2015-01-09T21:01:55","guid":{"rendered":"https:\/\/www.voyages-d-affaires.com\/traduction-automatique-faire-tomber-les-barrieres-de-la-langue-20150109.html"},"modified":"2022-01-14T15:07:04","modified_gmt":"2022-01-14T14:07:04","slug":"machine-translation-breaking-down-language-barriers","status":"publish","type":"post","link":"https:\/\/www.voyages-d-affaires.com\/en\/machine-translation-breaking-down-language-barriers","title":{"rendered":"Machine translation: breaking down language barriers"},"content":{"rendered":"

With its websites - Babelfish, Linguee and Reverso among the best known, as well as Google Translate and Bing Translator - and its mobile applications galore, machine translation is now almost a technological standard. But at the end of the day, there's only one question that interests users: can we trust these solutions completely? The general consensus is that machine translation is accurate to between 75 and 80 %. This is often enough to grasp the general meaning of a text, but is hardly recommended for translating the terms of a contract worth billions of dollars. Just take one sentence at random to get an idea of the approximations involved: \"for a long time, I went to bed early<\/em>\"famous and seemingly simple. For Reverso, it's \"for a long time I am early lying\" and \"long time I went to bed early\" for Google Translate. But should we blame them? For more than a century, Proust's translators have still not agreed on what to write... Is it \"went to bed early\", \"used to go to bed early\" or \"would go to bed early\"?<\/p>\n

Singer Malinda Kathleen Reese has poked fun at these wanderings in her \"Google Translate Sings\" series of YouTube videos. Using Google's website, the artist translated the lyrics of well-known songs such as Queen's Bohemian Rhapsody and Let it Go, the catchphrase from Disney's latest Snow Queen. From English to Chinese, then from Chinese to French, then from French to Tamil and so on, before returning to English at the end, all set to music. One thing leads to another and the mistakes pile up. In the end, the laughter is guaranteed!<\/p>\n

\"Techno\"<\/p>\n

Researchers began working on machine translation in the 1950s, drawing in particular on experiments carried out during the Second World War to break enemy codes. Today, solutions abound, from websites to mobile applications for travellers.<\/em><\/span><\/span><\/strong><\/p>\n

This light-hearted humour highlights a fact: machine translation is still in its infancy. Google is not kidding itself about the absolute relevance of its solution. Firstly, because, to find segments of sentences, its systems will search through a corpus of texts which, while they may have the merit of existing in several languages, are often rather boring, like the official documents of the European Union. \"On the whole, our translation systems have been better at making sense of official or business documents than at helping people in their everyday communication.<\/em>\"At the beginning of the year, Aaron Babst, then director of the Google Translate programme, noted this before moving to Facebook in September.<\/p>\n

But, as we all know, you don't have to hope to undertake, nor do you have to succeed to persevere. Today's business travellers have a multitude of mobile applications at their disposal, which are like dictionaries to be carried in your pocket. Of course, the results are not always optimal, but it's still much more practical than having to carry around the entire Larousse, Harraps and Robert & Collins collections...<\/p>\n

Among the most widely used, iTranslate boasts the highest number of downloads (55 million since its inception), while Google Translate boasts that it translates nearly 100 billion words a day. A staggering figure, especially when compared with the capabilities of a professional translator, who struggles to translate 4,000 words a day.<\/p>\n

\n

Conversations unbounded<\/span><\/strong><\/h2>\n

\"Conversations<\/p>\n

After the written word, machine translation is now tackling the spoken word. Launched this year, Skype Translator has achieved a real tour de force, \"solving the gap between the way people speak and the way they write. This is one of the main areas my teams have been working on over the last year\", explains Arul Menezes, who is in charge of machine translation research at Microsoft, on the research.microsoft.com blog.<\/p>\n<\/div>\n

New technologies are breaking down language barriers. As the megaphone of a globalised world, major social networks such as Twitter and Facebook have also launched instant translations of messages exchanged. At the same time, innovative applications such as WordLens have emerged to rescue travellers \"lost in translation\". Without even needing an internet connection, WordLens lets you point a smartphone's camera at a menu or sign and immediately receive a translation in your own language. It's hardly surprising, then, that last year Google acquired the developer of this application, Quest Visual, to enhance its mobile translation offering.<\/p>\n

Better still, while solutions have long focused on the written word, oral translation has recently begun to make its appearance. iTranslate has added a Voice version to its range, enabling two people who have installed the application on their smartphone to converse in their own language, with the application acting as intermediary. For its part, Skype launched the Translator plug-in at the end of last year, enabling simultaneous translation. \"Speech translation seemed to us to be a natural evolution of the work we had undertaken on written translation.<\/em>\"explains Chris Wendt, director of this machine translation programme. Satya Nadella, CEO of Microsoft, Skype's parent company, is convinced that a tool like this could \"changing the lives of people and businesses<\/em>\".<\/p>\n

\n

Overview of applications<\/span> the most practical translation tools
\n<\/strong><\/h2>\n

\"Application\"<\/p>\n

Google Translate<\/strong><\/span><\/p>\n

The best-known with 90 languages, including 38 with voice recognition, and a camera function (free).<\/p>\n

\"Application\"<\/p>\n

Waygo<\/strong><\/span>
\nAn app for understanding inscriptions in Chinese, Japanese and Korean characters (free).<\/p>\n

\"Application\"<\/p>\n

Istone Travel<\/strong><\/span>
\n300 colloquial expressions translated into a dozen languages (free).<\/p>\n

\"Application\"Triplingo<\/span><\/strong>
\n2000 sentences translated into 13 languages, including colloquial and even slang expressions (free of charge).<\/p>\n

\"Application\"Ihandy Translator<\/span><\/strong>
\nA very user-friendly interface and 52 languages translated. Free, with a pro version at 2.99 $.<\/p>\n

\"Application\"Itranslate<\/strong><\/span>
\nThe other leader in machine translation, with 90 languages and a voice option (free).<\/p>\n

\"Application\"Itranslate Voice<\/span><\/strong>
\nSimultaneous translation via smartphone in 42 languages (6.99 $).<\/p>\n

\"Application\"SMS Translator<\/strong><\/span>
\nTranslates SMS messages into 16 languages (free).<\/p>\n

\"Application\"Sayhi Translate<\/span><\/strong>
\nWith 100 languages available, including translations of local dialects. (4.99 $).<\/p>\n

\"Application\"Reverso<\/span><\/strong>
\nInstant translation in 10 languages with examples of use in context (free).<\/p>\n<\/div>\n

Will new technologies save mankind from this divine punishment, the confusion of languages, as old as Genesis? The myth is well known. God's overweening desire to raise themselves up to heaven by building the Tower of Babel brought this proud, human project to its knees, and condemned them to drag out their labours in the midst of unfamiliar words. From the medieval mystic Hildegard of Bingen and her lingua ignota to Dr Zamenhof and his Esperanto at the end of the 19th century, philosophers and humanists have racked their brains to build artificial languages that can be understood by everyone. There have been almost 900 such initiatives, rarely crowned with success, if not esteem.<\/p>\n

The technological approach to the problem is the complete opposite. Since linguistic diversity is a fact of life, let's make the best of it and give people the means to communicate with each other over and above their differences. To put it simply, machine translation is based on statistical models of language, which the machines use to rummage through pages to find sequences of sentences for which they might find the equivalent in another language.<\/p>\n

\"Techno\"<\/p>\n

Cursed translation errors... Automated solutions do not have a monopoly on them. History is full of examples, some of them amusing, such as Jimmy Carter's trip to Poland, when his translator expressed his \"carnal desire for the Poles\" instead of \"his desire to better understand their desires\". More tragic was the Japanese Prime Minister's response, \"mokusatsu\", to the Allies' ultimatum. For some, this ambiguous little word meant that he refrained from commenting. But even the Nipponese press agencies interpreted it as a rejection. Ten days later, it was Hiroshima.<\/em><\/span><\/span><\/strong><\/p>\n

This long-term technological adventure, which began in the 1950s and was led by IBM and Systran in particular, is now really starting to bear fruit. To illustrate the difficulty of the task, Microsoft's research teams have been working on the subject for over a decade, with the first result being the launch of Bing Translator and then, more recently, Skype Translator. With voice translation, the researchers are tackling another difficulty. While written language systems can be based on texts, translating spoken language is much more difficult. In particular, it is necessary to take account of intonation and pauses, which make it difficult for a machine to understand where sentences begin and end. And that's not to mention the different ways in which spoken language is formulated! As a result, to enrich the vocabulary of the artificial neural networks, the researchers relied in particular on messages published on Facebook, which are closer to everyday language.<\/p>\n

To go even further, Google is now calling on... humans. The technology giant recently launched a collaborative programme to improve the algorithm of its Translate service, with users invited to suggest better translations for common phrases. Ultimately, users are part of the solution to the problem, because in the age of big data, the more solutions are used, the more the tools will refine their translations.<\/p>\n