Despite the money and resources of major internet companies, it is DeepL that can boast the world’s most accurate and natural-sounding machine translation tool. When users enter a text, DeepL is able to capture even the slightest nuances and reproduce them in translation unlike any other service. From today, it is available to everyone free of charge.
In blind tests pitting DeepL Translator against the competition, translators preferred DeepL’s results by a factor of 3:1. Academic tests bear this out, as well. Within the field of machine translation, the gold standard for measuring system performance is the bilingual evaluation understudy (BLEU) score, which compares machine-translated texts with those produced by a translator. DeepL Translator achieves record BLEU scores.*
We have achieved several significant improvements in neural network architecture,” says Gereon Frahling, the company’s founder and CEO. “By arranging the neurons and their connections differently, we have enabled our networks to map natural language more comprehensively than any other neural network to date. DeepL’s ambitions are not limited to translation. Our neural networks have developed a level of text understanding that opens several exciting possibilities.”
DeepL’s revolutionary neural architecture runs on a supercomputer in Iceland, capable of 5.1 petaFLOPS, enough power to translate a million words in under a second. “Due to the abundance of renewable energy, we can train our neural networks very cost-efficiently in Iceland. We will continue to invest in high-performance hardware,” explains CTO Jaroslaw Kutylowski.
The supercomputer is used to train neural translation networks on a huge collection of multilingual texts. Here, DeepL can draw on the success of its first product, Linguee, the world’s largest translation search engine. Over the past decade, DeepL has gathered over a billion high-quality translated texts, the finest possible training material for a neural translation network.
DeepL Translator currently supports 42 language combinations between English, German, French, Spanish, Italian, Polish, and Dutch. The neural networks are already training to master more languages like Mandarin, Japanese, and Russian. DeepL also intends to release an API in the coming months, allowing its superior translations to enhance other products such as digital assistants, dictionaries, language learning apps, and professional translation programs.
*The previous record on the widely-used WMT 2014 newstest set for English-German was 28.4, achieved by the Google research department. On the same set, DeepL’s public translation service obtains a BLEU score of 31.1. For the 2014 newstest set for English-French, DeepL Translator achieves a BLEU score of 44.7, besting the previous record by more than 3 points.
Sign up for the free insideAI News newsletter.
