What I wish I’d known when I started working with machine translation
If you’re a regular reader of our blog, you might have noticed that I’m guilty of a blog cliché — I have a tendency to start articles with a personal confession. Well, here’s another one: I used to hate post-editing.
When I was studying languages at university, I was never that good at translation. I knew I had enough language to produce a passable translation, and a few times I’d make an attempt at post-editing purely because I couldn’t face translating properly, especially if it was into German. So I’d be there, sitting in the library thirty minutes before my assignment was due, trying hastily to post-edit something. Almost every time I’d get frustrated by the machine-translated text and end up giving up and producing something from scratch myself that was decidedly below average.
“I used to hate post-editing.”
You might find yourself wondering whether working at a translation AI company was really the best route for me, and perhaps you’d have a point. But some key things have changed in the last five years or so that made me change my mind.
Firstly, long gone are the days where machine translation couldn’t handle basic sentences. If you’re an editor at Unbabel, you’ll know that even though the machine translation isn’t always perfect (that’s why we have editors in the first place!), it’s come a long way from what it used to be — and the more we learn, the better it gets. Secondly, I spent some time working as a German-speaking customer service agent. It didn’t take me long to work out that translating everything from scratch in your head, especially into a language that isn’t your native one, is an absolute pain in a professional environment. In the real world, I’d be up against it for a deadline and the quality of the work I was producing had ramifications not just for my personal pride, but also for my ability to find and sustain work.
It’s then that I really started to embrace post-editing, long before I started to work at Unbabel. And perhaps when you’re reading this, you might be thinking: okay, that’s kind of an interesting story I guess, but it doesn’t really fit with the title of the article.
Well, now I’m going to take my personal experience and insights from my time working at Unbabel, and take you through a list of helpful post-editing tips. Read on to uncover the mysteries of post-editing…
The golden rule: work with the machine translation, not against it
There are always going to be times when you look at the machine-translated text and think “that’s completely different from the source” and you’ll have to start again. However, in most cases, even a mediocre machine translation can be a valuable starting point. I’m used to working with German, where there are often a huge number of different (but equally correct) ways to translate the English source text, and although, it can be tempting to restart and put in your preferred way of saying something, it can be unnecessary extra work.
“Our best editors who earn the most and have the best evaluation results are always those who make minimal edits and therefore stick to the source to produce fast, accurate translations.”
The issue is that once you make the decision to start again, it’s easy to stray from the source.I often speak to people who want to join our community who are obviously talented translators — and they make silly mistakes because they ignore the machine translation and put in their own version. These could be anything from poor lexical choices based on the source to adding in regional variations that go against the guidelines specified by the customer. Our best editors who earn the most and have the best evaluation results are always those who make minimal edits and therefore stick to the source to produce fast, accurate translations.
The machines are trained to help you
Did you know that the machine translation engines are often programmed for specific customer requests? Let’s say that you’re translating from English into French, and the customer has asked for you to use the informal tu. Maybe you’re not used to writing emails like that, and it’s not your personal preference — but we have to go with what the customer has asked for. The machine translated text should then have all of the verb conjugations in the informal second person, but if you take all of those out and start again, it can be really easy for the wrong register to slip through accidentally.
You don’t need to show off
I’d describe myself as someone who gets excited about language in general, and sometimes when I’m working with the written word, that flair tends to seep through into the translated text. Many times I’ve ended up looking at a word, thinking “I don’t like that”, and changing it for no reason other than weird personal preference. Even when I’m proofreading, I have to do a second run sometimes because I’ll find stuff I’ve corrected for absolutely no reason.
“Being a walking thesaurus can be fun, but it’s a recipe for disaster when you’re post-editing.”
But here’s the thing: as a translator, it’s not your job to show off. Changing that word to one you’d be more likely to use even though there’s no reason could change the meaning of the text and ruin the quality of the translation. Remember: as a post-editor, your job is to make sure that the machine-translated text is a completely faithful representation of the source text in the target language, and that it conforms to the brief and guidelines set. Being a walking thesaurus can be fun, but it’s a recipe for disaster when you’re post-editing.
Don’t assume the machine’s got it right
If you get a translation that’s overly technically written, it can be easy to think “oh, the machine can handle that” and skip over it. This is one of the most common errors we see at Unbabel — skipping over content that may not be familiar or focusing on the wrong part of the text.
And that brings us neatly to our final point and perhaps the most obvious of all tips — check your work. Time after time it pains me to see talented editors make silly mistakes, some of them even in the first few words, that would have been caught even with the most cursory of glances. Don’t be that person — take a second to get it right every time.
A note from the writer: thanks for reading this article! Over the coming weeks and months, we’re going to be changing our resources and guidelines to make them more user-friendly — watch this space!