I’ve always been a strong opponent of machine translation, the real reason of course being that I was afraid I’d be out of work in no-time if machines ever replaced us, but this reasoning was strongly supported by the fact that until a few years ago, machine translation looked absolutely ridiculous. Those days are gone. Enter neural networks.
From the frying pan into the fire
What made machine translation even more hated, was the fact that some translation agencies presented the incredibly bad machine-translated drafts as something that would actually help and benefit translators, saving them time and effort. But editing the existing machine translation was so excrutiatingly slow, that is often was better to just delete the whole sentence and restart from scratch. PEMT (Post-Edited Machine Translation) they called it, and quite some agencies still demand huge discounts for the “help” their pre-translated texts offer.
So yes, when I started testing machine translation a few years ago, it was mainly for laughs, to reconfirm how good I was and to once more show the world that translators had nothing to fear. Boy oh boy, was I wrong. About the first part that is.
Because by that time, a new technology had entered: neural networks. Now, without going into the precise details, trust me when I say that this is indeed a new revolutionary technology that makes machine “think” differently. So differently in fact, that some translations these networks generate are eerily close to what a human translator would come up with. And I have to admit it: sometimes the suggestions are even better than what I would come up with myself.
Foe turned friend
It then occurred to me that machines are not our enemies. They can also be our friends. Tools are only as bad or good as you make them to be. And here was a tool that could actually benefit me: saving me time when it came up with a good suggestion, and making me laugh when the suggestion was way off.
One thing was still bothering me though: these neural networks did not learn. You had to train them for days and after that they’d be static until you retrained them with newly added translations. Enter ModernMT. Here’s a new neural network technology that is adaptive, meaning it constantly trains and renews itself based on translations you add to it.
The dual sword master (STR +13)
So now we can harness two weapons: our old and trusted translation memories, for which Cattitude offers the best matches in the world (no CAT tool goes as deep as Cattitude!), and this new revolutionary technology called ModernMT.
And it feels great. It feels great to know that while I’m translating, not only do my translation memories get better, also the machine translation engine behind Cattitude improves. And I benefit from both. Because first Cattitude looks for a good match from my translation memory, and when it doesn’t find one, only then will it start generating a machine translation. And Cattitude doesn’t do this for each segment; it does this for every sentence in a segment, contrary to any other CAT tool. This means that a match generated by Cattitude often is a mixture of exact matches and machine translation.
Still not convinced? Cattitude also offers a cheaper subscription plan for dinosaurs healthily sceptical translators. This version has no link with ModernMT, but only with DeepL, another machine translation engine that does not learn and which you can turn off if you don’t like it.
Does discount as a sentence?
That said: one piece of advice. One client excepted, who only gets a discount of one cent per word on PEMT (even though Cattitude immediately scraps their suggestions and replaces them with a match of machine translation generated by Cattitude/ModernMT), I offer no discounts on machine translation. First, because Cattitude generates far better machine translations than my clients can come up with. I really don’t see why I should offer a discount on something I can do better myself. And second because this technology costs money (in my case thousands of hours of development — in your case, the investment in a CAT tool with plug-ins). As long as my clients are not willing to pay for that, I’m not planning to return the non-favour.
So be very wary with discounts on PEMT. It is absolutely true that if handled well, machine translation saves us time and can even improve quality. But even though Boeings are put on auto-pilot during your flight, no one in their right mind would suggest lowering the pilot’s wages because of that. Because when auto-pilot fails, you need that pilot to save your life. And it still takes years’ of experience and a tremendous amount of skill to handle that airplane when it fails. Translators are no different. Besides, I don’t know about your rates, but mine haven’t been indexed for 25 years. I think it’s our turn to benefit from technology now.
If I do a job in 30 minutes, it’s because I spent 25 years learning how to do it in 30 minutes.
