NMT (neural machine translation) is a topic this blog first covered back in 2015. But at the time, it was just “machine translation”, without the neural. And machine translation was still pretty much synonymous with Google Translate.
A lot has changed since then. The relatively rudimentary Google Translate (or Giggle Translate as some linguists were fond of calling it) has fallen by the wayside. Probably the best-known machine translation engine today is DeepL.
DeepL developed from the Linguee online dictionary and is a “neural” machine translation engine. As the name suggests, neural machine translation uses neural networks, a form of machine learning, to boost relatively clumsy statistical based translation.
The results are impressive. So much so, that many translators of European languages use NMT translation as a base for their translations. But as with any tool, NMT’s effectiveness depends on how it is used, not to mention whether it is the right tool for the job in the first place.
Here are a few quick observations and tips based on my own personal experience using DeepL.
Consider DeepL for creative texts
You might think machine translation would be well suited to logical and repetitive texts, and sometimes it is. But the DeepL engine in particular appears to have been “trained” on a corpus of excellent idiomatic translations. At a world, phrase and sentence level, it often outputs very natural sounding English. So, if you are looking for inspiration with a creative translation, seeing what it has to offer can help.
NMT is not a dictionary
This cannot be stressed enough. The difference between a dictionary and even the best machine translation is that a dictionary at least claims to give an authoritative translation. But NMT only uses its own mysterious algorithms to give the most “likely” one. That’s why it can sometimes get it dramatically and comically wrong. See following point.
Don’t trust anything translated by machine
Because unexpected and mission-critical errors can hide anywhere in even the most “plausible” sounding NMT text, the only way to use NMT is to check every word of the output against the original text. That times time, and naturally it’s a job for a professional translator.
Avoid the “terror of the blank page”
NMT is great if you actually prefer working from a draft to translating from scratch. Obviously, you need to watch out the correcting NMT doesn’t longer than translating without it. But when it works well, it can save time and stress.
Another weapon in the translator’s armoury?
I’m not one of those translators who catastrophize about NMT making our profession obsolete. Short of machine consciousness and translation engines that actually “understand” the texts they translate, I expect machine translation to remain nothing more than a powerful tool in the hands of translators. All the more reason to learn how to use it well!
Finally… does it improved the translation?
Ask yourself this question: is the final translation better for using DeepL or any other translation engine? It is one thing for someone to use machine translation to translate an email in a language they don’t understand or read a menu on holiday. But there is only one reason for a translator to use DeepL – and that is to improve quality of their translation.
Contact WeDoJapan for a chat about how our translation and transcreation services can help your business. +44 (0)1373 301853 or moc.n1680267016apajo1680267016dew@t1680267016catno1680267016c1680267016
We’d love to hear from you.