Why machine translation won't save you time or money

The rise of the machines

Unless you've been living under a rock, you've probably heard of machine translation. You might know it under a different name, like DeepL or Google Translate. But I'm sure you have a vague idea what it is and how it works.

Machine translation is 'a process that uses artificial intelligence (AI) to automatically translate text or speech from one language to another.' It has been around for a while and has been threatening to take my job for just as long.

Most savvy clients are already well aware that machine translation alone is unreliable and aren't willing to Google Translate and hope for the best. So, translators aren't out of a job just yet. Our job is simply changing, and we are learning to adapt with it.

Myth (and machine) busting

My experience tells me that many translation agencies are doing clients a disservice by offering this machine translation alternative as a cost-effective option.

One of the biggest translation myths is that having a translator edit a machine translation (known as machine translation post-editing, or MTPE) is more efficient than simply having a human translate in the first place.

The idea is that if you are only 'editing' a translation, rather than translating from scratch, you will spend less time researching and typing, so you can get the same outcome, only cheaper and faster.

Unless you are specifically looking for mediocre, robotic results, the myth that machine translation post-editing saves time and money is simply not true, for a number of reasons.

1. MTPE is completely hit and miss depending on the sector, subject matter, style of writing, quality and formatting of the original text.

   
   

All of the above dictates whether or not a text can be translated well by a machine. So what happens if your source text (the original document) is not as refined as you'd hope... say, your marketing team was in a rush that day?

If you were working with a human, they could take an educated guess at what these typos and line breaks were meant to do. A machine takes everything at face value.

Typos in machine translation are often not translated at all, leaving a random foreign word in a sentence. That is, unless it’s the kind of mistake that changes the meaning of a sentence, in which case you’re in even bigger trouble. Because the machine can’t quite figure out what you might have meant.

Think of the difference between ‘things’ and ‘thinks’. Or, in French ‘A très vite’ vs ‘À très bite’ (I have almost sent this to clients many times). The keys are right next to each other but there’s a biiiiiig difference ('vite' means 'quickly', 'bite' means 'd*ck'). They are both words in their own right, but with different meanings.

The same goes for line breaks in the middle of sentences. I have seen this in websites translated using the Wordpress plugin. A sentence gets split in two for visual purposes, but the machine thinks they are two different sentences and translates them accordingly (i.e. incorrectly).

So, if you’re opting for machine translation, you first need to make sure your ‘source text’ is pristine, meaning potentially more of your time and money spent (wasted) in the name of trying to ‘save time and money’.

2. Sometimes machine translation gets it so wrong that translators have to rewrite a whole sentence to try and save their own brains from melting

   
   

As a translator, if you spend all day reading text translated by a machine, you start to get Stockholm Syndrome. You second guess yourself. Maybe that sentence does sound okay, after all?

We have to make a call: if the sentence doesn’t make sense in the first five seconds of reading, just delete it and start from scratch. Trying to understand gobbledygook is inefficient. Don’t give it a second thought or the bad grammar and syntax will GET YOU.

Except, if you’ve been allocated less time to deliver an MTPE project than a straight translation in the name of ‘saving time’, you don’t have a lot of time to play with. You know?

The result? You end up with something that is ‘good enough’ rather than ‘good’ (don’t even think about ‘great’ or ‘engaging’). In my areas of specialisation (culture, marketing), ‘good enough’ is never good enough. So, this story just doesn’t fly.

3. Machines do not value consistency

   
   

I can usually tell a website has been machine translated if the same word (even a proper noun) has been translated in 2-3 different ways throughout the website. E.g. 'The church of Amy' vs 'Amy Church'.

If I spot this in a machine translation, I have to check these terms throughout and carry out additional checks to guarantee consistency. If I were translating, on the other hand, I would have used my professional translation software to create a glossary and be prompted to use the same word each time it appears in the original text. Or, the translator whose work I am revising would have done the same. Thanks to our mighty time-saving overlords, we get to spend more time sighing and wishing we were in a nature reserve instead.

4. Machine translation makes very different mistakes to a human

   
   

Best practice in the translation industry is for one linguist to translate a text and for another linguist to then edit it. Linguists can talk to one another or leave notes for a project manager to add context. It's fun! It's how we learn and develop as professionals, and it is particularly handy for providing context or options to a project manager or end client doesn't speak the target language.

Machine translation post-editing replicates this 'tandem' process in some ways, except the first human is replaced by a machine. And machines don’t really 'do' typos, which are usually easily fixed. Instead their crimes include inaccurate, literal or clunky translation, which takes far longer to rectify.

A machine can’t leave you a context note explaining a word choice, in the same way you can’t ask it why it translated a phrase in a particular way. You also can’t ask it where it found certain terminology and whether their sources are reliable or not. Or, as I mentioned with consistency, if it's translated the same word in two different ways, which one is best. So translators often end up checking terms the machine has translated to be double, triple sure. Again, this takes time that would be better spent outside touching grass.

So, Amy, how do you avoid these pitfalls?

My answer is fairly simple: I just don’t offer MTPE and I don't recommend it to my clients.

Whether my clients understand the limitations of a machine translation or not, I know they want and deserve better for their brand, their customers and their general aura points.

Firstly, I want them to get the most value out of my time as possible. Secondly, on an ethical level, the industries I work with are so human centric and experience focused, it makes no sense to remove the human from the process.

Instead, I offer human translation or transcreation.

This guarantees my and my clients' reputation isn’t sullied by work that I know will not be my best. It means I value my time and that of my clients. AND that I value the outcome of the translation and its context within their strategy, not just the translation job at hand.

Above all, it means I protect my creativity for future projects and make sure I’m not watering it down with nonsense that any old robot could write. That’s not me, and that’s not my clients with whom I’ve worked hard to build trust.

Maybe it's not that Deep(L)

As I said, ever since I entered the translation industry nine years ago, machine translation has been threatening to oust me. I'm still here and I firmly believe that I can do a better job for my creative, people-focused clients than a machine for all of the reasons above. Plus, you get to chat to a fun person and build a connection in the process. What's not to love?

If you’re looking for people-friendly translations and communications with personality and puns, I’m your woman.

If you’re looking for cheap and cheerful ‘good enough’, I can’t help you.

Get in touch and let's get creative together.