The Rise of the Translation Machines
There’s no doubt that online translation engines are incredibly useful. To be able to paste in some text and get an instant, reasonably accurate, translation is something that would have been unimaginable 20 years ago. There is also no doubt that translation engines have improved significantly over the years. With the introduction of Neural Translation AIs (Artificial Intelligence) — which use more contextual data and AI “learning” — we have seen a rapid rise in the reliability of Machine Translation.
But which works better? Google Translate is of course the most well-known, but DeepL has impressed many since it launched in 2017. In this article we look at the differences between the two: Google Translate vs. DeepL.
Google Translate
Free & Pro options | Translates into 100+ languages
Google translate launched in 2006 and at the beginning it used statistical methods to translate. This means translations were word-for-word, and based on usage — ie. the most popular or common translation was offered. However they have since upgraded to an AI Neural Network (for some, not all, languages offered) which analyses the context and takes it into consideration. As a result, the quality of the translations is much improved, although Google Translate can still be quite literal.
Google translates into over 100 languages, although the most common language combinations give better results. However, if you need to translate into a less spoken language, Google might be your only free option.
It is worth noting that Google collects a vast amount of data, including all of your source texts. So if you are translating company documents or any sensitive information, be aware that Google keeps and stores them.
DeepL
Free & Pro options | Translates (currently) into 28 languages
DeepL launched in 2017 and since then it has consistently scored better in translation tests (for the languages it offers). For translation agencies, a key feature of DeepL is that it links conveniently to APIs and software plugins for CAT tools.
DeepL also uses an AI Neural Network, but the networks that each platform uses are different. Google Translate (like most other machine translation engines) uses what are known as Recurrent Neural Networks. DeepL, however, uses Convolutional Neural Networks (CNNs), based on “deep learning”, a branch of AI that attempts to create learning models which simulate the human brain. CNNs produce better all-round results for continuous sequences of words. Generally speaking, DeepL translations are considered to be more natural sounding, as well as more accurate. As with all AI-powered translation engines, the most common language combinations (such as English-Spanish and other European language pairs) produce the best results.
The CNNs that DeepL uses have ‘learnt’ from data produced by their own online dictionary, Linguee, which has access to over one billion translated texts. Linguee searches the net for translations, adds them to its database, and uses algorithms and user feedback to evaluate them.
DeepL is based in Germany, so naturally they meet EU data protection regulations. However, only the Pro (paid) version provides end-to-end encryption, and the option to delete your source text after translation.
API Interfaces and options
Google Translate offers more features that DeepL: it can translate a whole website directly, and offers voice and image translation.
Both services offer API services that can connect to your own interface to conveniently translate your own website. If you are a developer you can easily set this up yourself. Otherwise, you can use an additional tool, such as WordPress, with an a plugin, such as TranslatePress or WPML.
Find out more: Go multilingual with WordPress WPML
Human translators
In texts where the style matters, such as creative marketing copy, a literal translation almost certainly won’t be sufficient. Style, register and nuance are key to really effective communication. Consider, if it takes six marketing executives half-a-day to decide on the phrasing in the original document, that subjectivity will be multiplied when it comes to the translation. Human translators can adjust the style to the context, and you can request a particular tone or register appropriate to the target audience you are translating for.
Furthermore, texts that feature important safety or legal information (technical documentation, installation guides, medical documents, legal contracts, etc.) should always be translated by humans! If you want to avoid personal injury and/or claims for damages, your translated documents cannot contain errors. A cost-effective option is to use Machine Translation together with a human translator or editor, this process is known as post-editing (MTPE: Machine Translation Post Editing ).
