Machine translation overcomes language barriers

For a while, the results of machine translation tools were often laughable. For all those who have turned their backs on them for a number of years, it's time to give them a try again to revise their judgement. Today's solutions rival professional translations.

By the turn of the 2000s, specialist publishers had employed an army of linguists to try and model the semantic subtleties and syntax. This involved deconstructing the source sentence and then, taking into account the context and certain words, generating the target sentence according to the rules specific to each language.

The statistical approach pioneered by Google, and made possible by the rise of artificial intelligence (AI), has shook up the codes. There are no grammatical rules or glossaries: this time, the idea is to use a very large number of validated translations to establish occurrences. The approach is similar to that of Deep Blue, IBM's supercomputer that beat world chess champion Garry Kasparov in 1997 by referring to memorised games of the grand masters. Another advantage of AI: translation engines are self-learning. In the case of Google TranslateThe service is enriched by feedback from its users, who can suggest better translations. Thanks to this community of contributors, Google is able to improve its machine translation tool free of charge.

The digital giant has announced that it will soon be taking on on-the-fly translation of an audio sourcewhich will act as an interpreter at conferences. The Google Translate application for Android will use the smartphone's microphone to do this. According to Google, it will then be able to adjust the translated text by modifying certain words according to the context or by adding punctuation.

As a direct competitor to Google, DeepL announced in early February that it was rolling out a new translation system based on the latest advances in neural networks. According to the German publisher, it will enable large, complicated texts to be translated without a hitch. Visit convince the undecidedFor the first time, it has posted a comparative evaluation of DeepL's results against Google Translate and Microsoft's Bing Translator. We can add Translate, a standalone application from Apple introduced this summer with the iOS 14 version.

Can we let Google translate professional documents?

Gaelle Bou, Sales and Marketing Director of Systran, a company with a 50-year history in machine translation, warns against using these free online translators, whose business model is based on use of personal data. "Business travellers using Google Translator to translate confidential documents raises questions"she stresses. Enacted in the United States in 2018, the Cloud Act forces US providers to disclose personal information about their users as part of legal investigations, even if the data is not stored on US territory. In contrast, Systran complies with the RGPD and defends itself against any exploitation of data, including for its free online translator, Systran Translate.

Systran also points out that "the majority of online translation services use data that is not their own, collected on the web by aligning data from multilingual sites. This explains why translation results on these platforms may be incorrect, or, in the worst cases, highly controversial..."Like Reverso and Prompt, this French publisher has positioned itself in the market for automatic translation dedicated to professionals. In SaaS or on-premise mode - i.e. accessible remotely or installed in-house - its solutions can be integrated into a workflow or translate PDFs on the fly. The main aim is to meet business needs by adding specialised dictionaries and glossaries. "The AI model is powered by the words used by the company, not by the mundane, explains Gaelle Bou. For a given company, we will look for translations that have already been validated in-house by a professional."

To multiply its efforts, at the end of 2019 Systran launched a marketplace from which translation agencies, freelance translators, linguists and universities can drive translation engines specialising in the legal, health, IT or tourism sectors. Provided they have an internet connection, business travellers can access on their mobile devices to these customised translation engines.

Interpreter in the pocket

Rather than using a smartphone, business travellers can also use a dedicated device such as Pocketalk. This box, which looks like a vintage MP3 player, can translate 74 languages in both sens. It's so easy to use, all you have to do is press and hold the single central button and speak the sentence to be translated into the microphone. The sentence is displayed, then translated into the target language before being spoken by the synthesised voice. Pocketalk also has a photo sensor for translating text, such as a restaurant menu. As the computing power is transferred to the cloud, the device needs to be connected to Wi-Fi or a 3G/4G network. In the latter case, it incorporates an on-board international SIM card (eSIM) providing two years' unlimited data free of charge.

For Tomoaki Kojima, Managing Director of Sourcenext, the Japanese distributor of Pocketalk, this product meets the needs of business travellers by guaranteeing the confidentiality of exchanges. It also offers greater security than a smartphone using a dedicated machine translation application. "Pocketalk's eSIM is for translation purposes only and cannot be used for web browsing. This limits exposure to risk"he says. What's more, "business travellers sometimes visit several countries in a single tripcontinues Tomoaki Kojima. There's no need to change any settings. The built-in SIM card automatically connects to the local line."However, in the tests we carried out, Pocketalk's performance depended on the quality of the mobile network.