They are about twenty, half computer scientists, half linguists, and they already challenge Google. They work for DeepL, an online machine translation service, boosted with artificial intelligence. Focus on this start-up created in Cologne, Germany.

Launched in August 2017, DeepL offers 42 combinations from seven languages: English, Spanish, German, French, Italian, Polish and Dutch. “We plan to launch more languages in a few months,” says J. Kutylowski, the company’s technical director. Chinese, Japanese, Russian and Portuguese are on the way.

But already, the translations are much more natural than those of Google Translate, according to the results of the reference tests “BLUE” (Bilingual Evaluation Understudy), put forward by DeepL and confirmed in particular by TechCrunch.

A billion text’s collection

For this result, DeepL relies on its first product, Linguee, a database of texts already translated and collected via Webcrawlers. Since its launch in 2009, Linguee has collected a billion translations and “responded to more than 10 billion requests from more than a billion users,” DeepL assures. DeepL, like deep learning, is therefore a neural network that is now convolutional and runs on a supercomputer based in Iceland.

Huge machinery for a company that is still a start-up. “Linguee was initially financed like everyone else via investment funds,” explains Mr Kutylowski. Its first capital raising in 2011 was initiated by Swiss BTov partners, Ukrainian TA Ventures and British N. Waesche, says Crunchbase. According to the German business register, DeepL posted a net income of 1.3 million euros in 2017.

Advertising and development

“Linguee works through advertising. DeepL is freely accessible on the internet but we are starting to develop the commercial offer for companies by adapting our software according to needs,” says Mr Kutylowski. DeepL also concludes contracts with translation companies, charging by the number of signs.

Translation agencies, international companies that want to translate their internal documents quickly… “the range of our clients is quite wide and spread all over the world,” he says.

Growing up without selling?

Still start up or already SME? “Between the two”, notes the technical director, “for the long term, we are an SME. For the agile structure and with a horizontal hierarchy, we look more like a start-up. Even if we prepare to grow, we want to keep this atmosphere.

Led by former DropBox alumnus L. Fink and Google researcher G. Frahling, the company employs 23 DropBox alumnus Leonard Fink people in Cologne and 500 external editors. For the time being, DeepL focuses on improving language quality and remains discreet about its ambitions. “We are not building a company to sell it,” says J. Kutylowski.