Not so many years ago machine translations had a deserved reputation for being unreliable to the point of being laughable. You may well have filled a few minutos muertos playing the game of translating a normal English phrase into clunky Spanish and then translating it back into amusingly surreal English.
Britain’s The Guardian newspaper published an article in 2013 that tells of how one Turkish newspaper ill-advisedly printed a machine translation of a Noam Chomsky interview. Leading to gems like “Contrary to what happens when everything that milk port, enters the work order, then begins to bustle in the West.”
As the article pointed out a flesh and blood translator would have offered up something like: On the contrary, when everything has calmed down, then this will be when the West starts panicking. Read the full article in The Guardian
If you have been tracking the progress of Google/Bing Translate you have probably noticed a lot of improvement in the quality of the translations served up. Although they are still often far from perfect, used intelligently they can be a useful tool.
It’s clear that the AI grows smarter when fed Big, BIG Data. Google has Ginormous Data and its ubiquity makes it the default choice for machine translation.
However it’s not only Google who have been working hard in this hugely lucrative market.
Last year a Spanish lady in one of my English classes asked me if I was using DeepL.
I had a look at the website but at first glance it didn’t seem too different from Google Translate. Combined with the the fact that I could get a translation from GT just by selecting text and hitting copy meant that I didn’t go any further with DeepL.
While watching one of Nachotime’s excellent videos on YouTube I was interested in the author’s experience. He was talking about a great exercise for written practice, that I’d being doing for a while using GT. As Nacho said he was getting better results with Deepl I decided to give it another try.
Deepl takes its name from ‘deep learning’ technology which uses artificial neural networks to analyse data.
Smart technology is one thing but as they say ‘rubbish in, rubbish out’. So how does DeepL produce better translations?
Being linked to the brilliant Linguee means that the data that feeds the algorithms isn’t just big, it’s incredibly rich in nuanced turns of phrase created by humans translators.
The layout of the webpage is very straightforward, little more than one dialogue box to write/paste the text to be translated and a second box showing the translation.
I really like the fact that you can just start typing or paste some text and DeepL will recognise the language without having to remember to switch manually.
I would love an app for my phone and tablet. The only extra feature that I really need is a search history and would be happy to make a one off payment for the option.
I’m sure that the available Pro version is very worthwhile for professional translators but unfortunately the recurring 6€-40€ monthly subscriptions are beyond my very modest means.
The quality of the machine translations produced by DeepL is impressive and takes us a step nearer to instant communication without the language barrier.