Cologne-based translator and copywriter Ian Winick talks about the importance of checking your parachute before you jump off the plane.
Ian, you’ve been a professional translator for 24 years. How do you use AI in your work?
I don’t – or at least not unless a customer asks me to. If it were possible to produce a high-quality translation more quickly based on AI output, I’m sure many translators would include it in their process. But I don’t use it myself and don’t know any other professional translators who do.
Why not?
Leaving aside the environmental issues that everyone knows but seems to gloss over, there are serious confidentiality concerns. If customers entrust their internal files to me, I’m not going to let them anywhere near a machine that is specifically designed to hoover up all the data it can find. Especially if – wait for it – my customers are paying me to translate it. Not everyone has jumped on the AI bandwagon.
But wouldn’t it speed up the translation process?
Maybe if you haven’t been a translator for very long. But any colleague who has ever outsourced a job to the wrong person will tell you that it’s almost always easier to retranslate from scratch than to polish a turd. And yes, I can think of no better image than that.
So AI doesn’t help to lay the groundwork?
Maybe if you’re translating operating instructions for toasters, but it doesn’t help with the more complex and nuanced translation work that generally crosses my desk. If you’ve been doing this for as long as I have, it will take you less time to produce an elegant, natural-sounding translation than it would to turn machine output into anything usable. And you’ll still have the will to live afterwards!
Is this because the quality of machine translation is hit-and-miss?
Yes, very much so. The misses are frequent and the hits are adequate at best. Just imagine trying to cobble together a coherent translation by assigning a sentence each to a few hundred translators of varying skill, experience and hallucinogenic tendencies who have no contact with one another – because that’s essentially what machine translation is. And as far as translation is concerned, AI is just machine translation in a fancy hat.
Have you looked into AI yourself?
Yes, I’ve just completed an excellent ten-week course on AI in Translation with the Institute of Translation and Interpreting (ITI), which was designed to explore the uses – and limitations – of AI in a translation context.
What did you gain from this course?
Apart from allowing translators to look under the bonnet of AI, it gave us a clear understanding of the risks involved and of the role that translators have in helping to manage these risks. And “looking under the bonnet” only goes so far, of course – after all, even AI creators will admit to not fully understanding the beast they are unleashing on the world.
What risks are you talking about?
Well, despite what tech bros claim, AI doesn’t actually understand anything – what it does is predict what, statistically speaking, the next “token” (word or part of a word) is likely to be. And you can’t walk that tightrope without a safety net. Particularly if there are crocodiles lurking below in the form of legal and reputational risks.
So the risk is transferred to the translator?
Yes, which is how it’s always been – that’s why translators have liability insurance. But if you let a hallucinating AI loose on a translation without having it carefully checked, good luck getting OpenAI to take responsibility. So it’s always a good idea to transfer the risk to a professional, to have not just a human in the loop but an experienced translator who knows how AI “thinks”. And preferably one with a big red pen who is not afraid to use it.
So you’re recommending using AI with input from a real translator?
No, a microwave dinner is still a microwave dinner, regardless of how much parsley you sprinkle on it. In other words, no matter how much they are tweaked, AI translations are a poor substitute for an experienced translator with excellent writing skills. But if you are going to go the AI route – whether for cost reasons or fear of missing out (FOMO) – at least make sure your parachute is on properly before you jump out of the plane.
Meaning what exactly?
Get a human translator to make sure that the translation is at least technically correct.
Is this what is meant by “transferring the risk”?
Yes, experienced professionals will check if the translation is accurate and can ask questions if necessary. I’ve always said that you shouldn’t trust a translator who never queries anything – if something is unclear, professionals ask questions, amateurs guess and, in our brave new world, AI simply makes stuff up. Translators don’t “hallucinate”, no matter how much coffee they have drunk!
But isn’t that a tough sell to companies who are under pressure to keep costs to a minimum?
If your website is a calling card that’s going to be seen by thousands of potential customers, it’s worth pulling out all the stops. AI slop is a real turn-off for many customers, who are going to wonder where else you are cutting corners. But it’s also true that some translations just need to be “fit for purpose”.
What does that mean exactly?
If you’re posting a video on LinkedIn and want a line or two to accompany it, you’d be forgiven for running it through a machine and hoping for the best. And if your budget is tight, you might decide to translate a set of internal company minutes using AI. Fair enough, but at least get an experienced translator to invest an hour or two so that you don’t end up with egg on your face. Or your CEO’s face.
But won’t AI inevitably get better over time?
Think about it – ChatGPT has been around since 2022 but AI-generated text and images are still pretty dreadful. And how exactly do people think AI translations will get better? Are there massive teams of professional translators providing expert input for a pittance so that the next generation of AI can replace them? Of course there aren’t.
Hasn’t machine translation improved as well?
I don’t use MT either – for the same reason I don’t eat processed cheese – but I have heard that quality has taken a nose-dive recently. I imagine this is because MT is being fed with its own dodgy output. And this, as any livestock farmer will tell you, is not going to end well. Now that the internet is being inundated with AI slop that is fed back into the system again, I can’t see things getting any better. Garbage in, garbage out.
Do you expect a backlash against AI?
It’s already happening. Consumers are sick of having AI foisted upon them – it’s an answer to a question that none of them ever asked. For instance, AI technology can be used to create podcasts and novels. Impressive, yes, but who on earth wants to listen to a podcast with zero human warmth or a read a story scraped together from thousands of other novels? I imagine that most people, like me, have a pile of books at home patiently waiting to be read. So why would I read a book that no one has bothered to write?
And how do translation customers feel about AI?
The feeling I’m getting from talking to colleagues is that customers are realising that AI translations are more trouble than they are worth, especially now that they won’t be a cheap option for much longer. But more importantly, they are also seeing that quality content – written or translated with style, energy and a discernible voice – is a breath of fresh air for the AI-jaded masses.
(Questions asked by Nicki Wain)
