Have you used DeepL, Linguee’s offspring?
I think we have all used Linguee, haven’t we? Well, the company changed its name last year to DeepL. Linguee is one of the most widely used computer translation engines: 1 billion users have made 10 billion consultations. In 2016, based on all the knowledge gathered through Linguee, the developers started working on a neural network system by training their neural networks with billions of translations gathered by the Linguee crawlers. In less than 10 years (Linguee was launched in 2009) the company has taken great strides into machine translation using deep learning and neural machine translation.
DeepL has claimed that it is the “most accurate and natural-sounding machine translation tool“ based mainly on the assessments they have made using the BLEU method (Bilingual Evaluation Understudy), which compares a human translation against machine translation measured on a 1 to 10 scale. It is indeed promising, but still a lot of work needs to be done (terminology is still a big challenge) and some experts agree that DeepL’s claim is a bit far-fetched. For example, Kirti Vashee, a technology consultant, commented: “I think they are using well understood and public test sets and claiming that they have better results. To my view they are slightly better, but far from revolutionary.” Read more about this in this post by Slator.
As a final note, I thought it is interesting to mention that DeepL’s supercomputer is based in Iceland, to take full advantage of that country’s renewable energy which makes operations cheaper. This supercomputer can translate 1 million words in less than a second. One thing is for sure, this company will probably give us a lot of good products, even beyond translation, as they say that their “neural networks have developed a level of text understanding that opens several exciting possibilities”.
So, if you haven’t given it a try, here is the link to DeepL: https://www.deepl.com/translator, where, by the way, you will find a tab for Linguee too. 2 in 1!