The responsible guide to using AI translation
When you need to translate and localise your marketing content ahead of a big expansion it can often feel like you have a dizzying number of choices before you.
Should you speak to a translation agency or a freelance translator? Is there someone in-house who can handle the job or does it need an external specialist?
And now, ever since ChatGPT exploded onto the professional scene last year, there’s been another question to add to the list: “What about AI?”
It’s a question we hear a lot, and not just since the rise of ChatGPT – whether it’s AI large language models or machine translation tools like DeepL and Google Translate, every new advance in technology has people wondering whether it’s finally good enough to be a reliable translation provider.
So what can AI do for translation now? What limitations does it still have? And if it is good enough to translate your marketing, how can you make sure you’re using it responsibly?
But really, is AI good enough now?
The honest answer is that it depends on what you’re using AI for. If you’re using it for a piece of content that won’t be seen by many people or for very long – like an email or a social media post – it might be a useful translation aid. But if you’re looking to use AI to translate core marketing content like blogs or a product page, it just isn’t where it needs to be in order to beat out a human translator.
To understand why, it helps to understand a bit more about how AI and large language models like ChatGPT really work. Despite the name, artificial intelligence doesn’t really do much thinking – it’s a tool for answering queries, and it does so by using its training data to predict what answer will be most likely to help you.
That’s why it’s so prone to hallucinations – the fancy word for when it spits out obviously false information. The hardest thing for an algorithm to do is to say it can’t give an answer, so if it lacks the data to answer your question accurately, it will try to answer it inaccurately instead.
What does that mean for translation specifically? It means that an AI or machine translation tool can’t draw on outside context or ask questions about ambiguous language (and language is often ambiguous). It has to choose a translation, whether it’s the correct one in that context or not. It also can’t advise you that translating a phrase like “eliminates pores” into German won’t fly because German consumers tend to call out claims that aren’t physically possible. It can only choose what its training data says is the correct answer and return that.
Nor can it translate marketing copy into your brand’s specific tone of voice. Because AI and machine translation draws on the data it’s been trained on and doesn’t make creative decisions, its translations will naturally sound like everything else on the market because that’s what it learns from.
When is AI translation worth the risk?
If you really can’t afford to engage a human translator, it’s easy to think that AI or machine translation is better than nothing. At least it gets your content out in front of your target market, right? And if there are any mistakes, surely your customers will understand and fill in the blanks themselves?
But when AI translation goes sideways, that doesn’t just mean your customers might have to wrap their heads around some sticky phrasing or the wrong tense. It could also mean that crucial ingredients aren’t included on health and allergy labels, for example, or that your products promise something they can’t possibly do. In that case, a bad AI translation can actually be worse than no translation at all.
Even if a piece of content doesn’t have the potential to harm customers if translated poorly, you still have to weigh up if the risk of error is worth what you can save initially by using machine translation. Even content that’s lower down the impact list – your website FAQs, for example – still have to be translated right if they’re going to have the right impression on your customers.
Ultimately, there are two questions you need to ask yourself when you’re judging whether AI translation is worth the risk for your brand and marketing:
- Will enough people see this content that any glaring errors will damage our brand reputation?
- If it’s low-value enough that mistakes don’t matter, why is it representing our brand in the first place?
So how can you use AI translation responsibly?
With everything we’ve said about the risks of using AI for your marketing translations, you might be wondering if we’re just taking the scenic route to say that AI translation is never, ever, under any circumstances a useful or responsible solution.
But while it’s true that AI just isn’t a match for a flesh-and-blood human translator, we’d be doing you a disservice if we said it couldn’t be used responsibly to assist with certain translation work. With “low-value” content for example, AI can help to speed up translating content that’s more about relating information and doesn’t require any creativity or wordplay – but only if there’s a human translator at the other end to thoroughly review the results.
The main trick to using AI responsibly is to understand the limits of machine translation and AI translation. Translation is about communication at the end of the day, and so long as you know where and why AI isn’t able to help that communication, you can put the right quality control in place to make sure any AI or machine-assisted translations aren’t confusing – or worse, misleading – your customers. We offer AI translation consulting services to make sure you benefit from the cost-savings of AI, without compromising on quality and putting your brand at risk.
However, when we’re talking about using AI responsibly, there are more than a few elephants in the room that have nothing to do with hallucinations or bland translations.
Ethical considerations of AI translation
For one, any AI use brings up ethical questions around when its training data crosses over from learning into copyright infringement. Several news outlets such as The New York Times, Intercept, Raw Story and AlterNet, for example, have filed law suits against OpenAI and Microsoft for violating copyright laws and illegally training large language models by using journalists’ work.
Environmental considerations of AI translation
But also there’s the environmental cost of AI, especially around large language models like ChatGPT. While it’s hard to pin down an exact carbon footprint because the tech and its usage is evolving so rapidly, one study from November 2023 estimates that ChatGPT uses anywhere from 1 million to 23 million kilowatt hours (kWh) of electricity every month. If carbon accounting is part of your environmental activities, you’ll have to be prepared to factor this in if you’re planning to use AI translations.
Ask: can your brand be candid about its use of AI?
Responsible AI translation might soon mean putting up clear disclaimers wherever it’s used. Regulations around AI are evolving every day, and we’re already seeing similar rules when it comes to AI-generated content – for example, YouTube now requires all creators to add a disclosure label to any video containing sound or images created by AI.
The question to ask yourself is where you’d be comfortable disclosing when you’ve used AI to translate something. If you have a language barrier with a client or supplier, for example, you might be perfectly happy letting them know that Google Translate is helping you to understand and respond to their emails. But what about telling your customers that the marketing blog they’re reading might contain errors because it was translated with AI?
If the answer is no, that’s probably a good sign that the responsible thing is to not use AI after all, or at least to temper it with a heavy dose of human editing and quality control.
To find out more about how we can help with translation, transcreation and localisation for your European marketing content, get in touch.