From the ashes of Linguee, an online dictionary born in 2009 in Cologne, DeepL Translate debuts in August 2017 immediately attracting the attention of the trade press (and users) by promising far better translations than the closest rival Google Translate. Initially available only for a handful of languages (among these, also Europen), it immediately seems capable of nibbling away at small slices of the market at the service of the Mountain View giant.
A success based on the naturalness and ability of DeepL Translator to grasp the nuances and the “subtext” and to be able to read “between the lines”
This is thanks to the mix between algoritmi di machine learning highly specialized, convolutional neural networks which replicate the functioning of the visual cortex of animals and a huge database of online searches and old translations which allow you to “train” the translation engine of the German service and continuously improve its performance. Furthermore, the artificial neurons have been arranged so that their analysis and understanding of human language is as complete and precise as possible.
Mechanical translations thus appear more natural and more similar to those that a human translator could create. DeepL engineers quickly trained the AI to create aweb application that excels at replicating meaning original sentences in the target language, offering precise and professional formulations. Ability that is valid for both literary texts and technical documents.
How DeepL Translator works
Although the German startup does not reveal much about the model chosen to create what is defined as the best online translator in the worldit is possible to obtain some information that helps us understand how DeepL Translate works.
First of all, in creating their online translator the German engineers did not start from scratch: DeepL rests on the solid foundations deriving from Linguee’s experience and artificial intelligence algorithms and machine learning already used (and on its research database obviously). Added to this are new translation algorithms and a artificial neural network system capable of offering high performance in any sector.
Thanks to the computing power of video cards, DeepL Translate is able to translate texts of all types perfectly, or almost
This combination is supported by a extremely powerful supercomputer (according to the DeepL Translator technicians, it is the 23rd supercomputer in the world in terms of computing power) based on GPU computing. The supercomputer, hosted in Iceland so it can be powered by renewable energy sources present in the Nordic country, it is capable of translating over a million words per second and, being able to count on an infinite or almost infinite database of linguistic-themed searches and translations, “feeds” one of the most efficient online translators around.
How to use DeepL Translate
To start using DeepL Translator, simply visit the official home page. So, pasting the text to be translated (CTRL+V) in the left box, DeepL will automatically recognize the language in which it is written, providing the translation in the language specified in the box on the right.
In the free version, which is accessible without the need to register, DeepL Translate allows the translation of texts up to 1,500 characters long3 non-editable documents per month and access to a maximum of 10 entries in the integrated dictionary.
For advanced features, you can subscribe to one of the subscription plans offered by the service. Although, to translate longer texts, it is possible to do it in multiple passes. For some languages, there is also the possibility of activating a glossary that allows you to customize the translation with specific terms or alternative expressionswhich is particularly useful when dealing with texts of a technical nature.
Furthermore, you can adapt the translation depending on the recipient by adjusting the tone of the translation and then, opting for the formal or informal voice option, which updates the pronouns and surrounding text.
DeepL to translate from pdf files
In spite of other competing solutions, DeepL Translate offers the possibility of translate documents while preserving their formatting original. Therefore it is possible to exploit artificial intelligence for translate from PDF, DOCX and PPTX filesincluding Word documents and PowerPoint presentations.
To translate a document or presentation from one language to another, that’s enough drag the file from your workstation working directly in the DeepL window opened in the browser. Alternatively, you can click “Translate tabs“then up”Select from your computer” and then on “to be translated“.
Once the file to be translated has been uploaded, you can choose the target language from the right pane, to start automatic translation. The quality of translations for documents is equal to that offered for the translation of simple texts, but with the added advantage of preserving the structure or layout of the original document.
However, the free version of DeepL Translator does not offer the ability to translate from PDF, DOCX, and PPTX files larger than 5 MB. The service may refuse to translate from PDF files, even if you have compressed the images in it. In these cases, you can split the document into multiple pages or temporarily remove the largest images.
How to translate from pdf files with DeepL Translate
We just saw that DeepL is able to translate from PDF files, but to process this format, documents come temporarily sent to Adobe servers. However, if you want to avoid additional data transfer or prefer to use DeepL’s built-in Word file management engine, you can convert the original PDF file.
To do this, you can take advantage of online services like SmallPDF, where you simply press CTRL+SHIFT+N in the browser (CTRL+SHIFT+P in Firefox) and visit the specific page. After uploading the PDF file of interest, you can get a Word version by selecting the “Converti in Word“.
At this point, you can proceed by sending the Word file to DeepL Translator and wait for the corresponding file to be produced with the text in the desired language.