If you speak the English language rather as sub-middle as I am, one is grateful for tools, which grab one under the arms when one reads English texts. For whole pages, a Google Translate or even the Microsoft Translator. I personally swear for years on the offer von Linguee, when I sometimes need a translation for a word or a phrase.
The results are provided based on texts already published by other people. So you can not just translate a particular word, but you can also pick out the right context from different examples. This often works better than the Google Translate algorithm, which translates rather stubbornly word by word into a sentence and sometimes delivers strange sounding translations.
Founded in 2009, company.com has now been renamed DeepL to provide a new translation software of the same name, which is intended to put Google Translate and Microsoft Translator into their own limits. The name is also used to indicate the future direction because one wants to focus on deep learning and sophisticated neural networks.
DeepL relies on an artificial intelligence running on a supercomputer in Iceland, which can run 5.1 petaFLOPS (5 100 000 000 000 000 operations per second). This is enough power to translate one million words in less than a second. Why is the crate in Iceland? Because of the oversupply of renewable energies, the neural networks in Iceland can be trained very cost-effectively, according to CTO Jaroslaw Kutylowski. It is also planned to invest in new hardware in the future.
The Cologne-based company, therefore, relies on neural networks and wants to set a new standard for machine translation. Fear of Google or other competitors is not. You are so convinced of your own system because you have cut off so brilliantly in a blind test.
Professional translators were presented with 100 sentences in August 2017, each of which was translated by DeepL, Google, Microsoft, and Facebook – the translations of DeepL were rated three times as often as better. Graphically prepared looks like this:
Also in the case of the BLEU score – the reference for machine-translated texts, which are compared with those of a translator – Cologne’s competitors clearly depend on their competitors:
DeepL is now also online, so you can try the fun even times on this site. I have times translated a few sentences and here and there it hangs still, is sometimes a word not recognized or a sentence a little wooden built. On the whole, however, impressive results come out. Google and Microsoft do this already on a very high level, but some translations of DeepL do not actually realize that they are done by machine:
DeepL also benefits from the fact that one has a decade of experience at its fingertips. The focus of this learning process, with the help of many already translated texts, is not only to offer the correct translation, but also to take into account the context and the correct formulations.
DeepL currently has 42 language combinations between German, English, French, Spanish, Italian, Polish and Dutch. However, as the German company knows in the press release, the neural networks are already training in order to be able to master additional languages such as Mandarin, Japanese and Russian in the future.
This is a nice new translation world, but still not the end of the flagpole: In the next few months you want to make an API available so that other manufacturers and programmers can also access DeepL. So, imagine that DeepL will not only be used as a standalone product, but will also be used in dictionaries, translation programs and not least in digital assistants.
Currently, you need to call the browser version of the translator, if you want to use DeepL now. The colleagues of golem.de could, however, learn that already on Apps for iOS and Android is tinkered, which should be finished in three months. Your Christmas wishes from abroad, you can then probably have already translated through the DeepL app.
Personally, I am very impressed with what DeepL has on the box and yes, I find it also really great that this new evolutionary level of machine translation was not achieved with a software from Facebook, Microsoft, Apple or even Google, but by a company from Germany. We like to be a bit small and do as if there is no one here, who can offer the Great Paroli. DeepL is a good example of the fact that it is still possible.
Product image: Pixabay