If we’re to believe what we read on the web these days, machine translation (MT) and artificial intelligence (AI) tools can whip up a high-quality yet low-cost translation at breakneck speed. Yet, as the age-old saying goes, if it sounds too good to be true, it probably is. Our project manager Laura Germaney offers some insight.
Many of us have used such tools in our personal lives. When holidaying abroad with others who don’t speak the lingo, I admittedly prefer to let my phone do the work rather than having to translate the entire restaurant menu myself. After all, I’m officially on vacation. How useful, however, is AI in a professional context?
Technological advances are indeed part and parcel of the working world and translation is no exception. Long gone are the days when translators would sit at their desks, pen in hand, scribbling away on paper. Nowadays, translators often work in dedicated computer-based tools that help to speed up workflows and check for terminological consistency. Yet we must not throw caution to the wind.
First off, you’ve likely heard of ‘Buy cheap, buy twice’. I’d like add another to the mix: ‘If it’s free, you’re the product’. The use of free online translation tools still comes at a cost. Whatever content you enter is no longer yours, but rather the property of the online translation tool provider and it will be used for their own purposes. Even for certain paid platforms, data scraping is still automatically enabled as standard. Entering confidential information is therefore an absolute no-go. Once it’s in the tool, you cannot go back. Professional language service providers like First Edition, however, adhere to strict data privacy regulations to ensure your content is kept safe and secure.
Whilst free online tools will scrape your data, we have strict confidentiality agreements in place across our supply chain and are certified in line with the UK government Cyber Essentials scheme.
Another key aspect to highlight is that AI cannot think. Computer systems ‘think’ exclusively in 0s and 1s. At the heart of it, 0 represents nothingness and 1 represents somethingness. AI isn’t actually clever. What it is doing is retrieving data and reaching conclusions based on input. It’s a bit like when your phone predicts the next word in your text – it is guessing your next move based on previous texts you’ve written.
It’s therefore fair to say AI systems don’t have a grasp on how to trigger certain emotions, they’re making logical, rational decisions and equations based on data. This clashes particularly with editorial content such as museum brochures. Here, you are not just portraying a message in words. You are seeking to trigger feelings about the artwork, artist or story, and that’s an abstract concept – open to interpretation – that belongs in the outside world, beyond the realm of AI. So we certainly need a human to evaluate how something is or can be perceived by the target audience. Reading between the lines and thinking outside the box are innately human processes.
Even for medical projects, say for pharmaceutical drug marketing brochures aimed at healthcare professionals (HCPs), we need to consider how the text will be received by the reader. Does for instance the wording instil trust in the drug’s capabilities? Ultimately, texts are made for human reading. And we humans are notoriously whimsical creatures whose emotions cannot be boiled down into 0s and 1s comprehensible to a computer.
This leads us to our next point: Machine-driven translation tools are not aware of differences between varying cultural groups or target readerships. Editorial texts, for example, often contain cultural references, idioms and context-specific references that may not have direct equivalents in the target language. Unlike machines, linguists will pick up on this and know to culturally adapt so that the new target reader, who is often embedded in an entirely different culture, still understands and enjoys the text.
“Translation is not a matter of words only: it is a matter of making intelligible a whole culture.”
British writer, composer and translator Anthony Burgess
Lastly, it is not uncommon for there to be mistakes in the source text. Whilst AI may be able to override an error that’s categorically incorrect (misspelling of a word), it’s not fact-checking. Quite the contrary, it carries the risk of ‘hallucinations’, i.e. presenting false or misleading information as true. In contrast, our expert translators have been known to tell us that something in a source text just doesn’t seem quite right to them – a gut instinct, if you will. For example, when recently translating product packaging, we noticed that the best before dates were in the past, which didn’t seem logical. By engaging in such thought processes and going beyond the words on the screen, we as humans can efficiently correct errors originating from the source text, saving the client time, bother and, in such cases, reprinting costs.
At the AI Translation Slam event, which I attended in December 2024 at the Institut Francais d’Écosse, translators were pitted against AI models. In one text, three source text errors were not recognised by the AI – but they were corrected by the linguists.
Here at First Edition, we are certainly not averse to technology and its advantages in localisation processes. We embrace technological advancements to optimise our workflows to the benefit of our customers and linguists alike. Where appropriate, we are happy to discuss a hybrid human-machine translation service known as post-editing. We do, however, believe it is necessary to heed caution when hearing lofty claims about AI. Put simply, we believe greatly in the immense value that our professional translators bring to the table.
Whatever the size of your project and whatever languages you need, if you’d like to chat with us about your translation requirements, we are happy to hear from you under enquiries@firstedit.co.uk