Automatic translations
No thanks
Last week, Substack sent out their regular monthly update, highlighting what’s been going on behind the scenes of this platform. One thing caught my eye immediately (it’s in the title), which is that they now offer ‘post translations’.
I have nothing against translations per se. Indeed, since I’ve worked in the translation and interpreting professions for most of my career, it would be a bit contradictory of me to write a post against them.
No, what upsets me about this development is how the translation is going to be done. You see, translation done by humans is a process that takes time. Translation done by a machine takes no time at all. Sometimes the results can work, but a reader who is unable to compare the languages in the source and target texts isn’t the best placed to judge the quality of the work in question.
When I write my posts, I put a lot of thought, sometimes, into what ends up on the page. Obviously on other occasions I just sort of word-vomit the first thing that comes into my head into the document growing in front of me and move on to the next paragraph quickly. But when I do think about the words, sometimes I think a lot about them.
There are half-remembered rules we were told at school, things about not repeating a word too often in a short space of time. I vaguely remember stuff about not following the same structure in every sentence, because that way monotony lies, and other rules that I don’t think I remember, but which I apply anyway (Mr Valentine, my A-level English teacher would be proud, I think).
If I’ve struggled - yes, I admit, I’m being melodramatic, but ‘struggled’ brings to mind an image of me crafting my posts in a chilly garret somewhere, shivering because I haven’t got another coin to feed into the electricity meter – sorry, that was a long aside, and so I’ll start again, not least so I can break the rule I quoted in the previous paragraph and get ‘struggled’ into the same paragraph 4 times. If I’ve struggled to find just the right word, then perhaps I, or one of my peers working as a translator, would be the best placed to determine what the translation for that word should be in any translation of the original text.
A couple of weeks ago, I interpreted a speech by a man who is quite well known. In fact you’ve definitely heard of him, and the organisation which he leads. His speech referenced Dickens’ A Tale of Two Cities, as fine an example of carefully-chosen words as you’ll find anywhere, right from the very first paragraph. The book has been translated into many languages, many times, but I bet that every translator who has ever worked on it has spent a lot of time wondering how on earth to start and, in many cases, I bet the more canny translators among them have left the first few paragraphs alone until much later.
A machine won’t do that, because it doesn’t have the same capacity to weigh up alternatives. As Dickens might have said, if he’d had to consider AI translation, it might have been the best of translations, it might have been the worst of translations.
Another example for you. I translated a legal document a few weeks ago. Again, a machine could have done the translation in the blink of a computer eye, but it took me over a week (which was reflected in the size of my invoice when I delivered the text). At one point, I put a sample, duly anonymised, into an AI tool, just to see what would happen. The results looked very good, but also, once you looked more closely, completely wrong – the machine had managed to get the Donor and the Grantee mixed up, which meant that the person who should have been receiving the money, in the automatic text would have ended up having to hand it over instead.
That would have been fun for the courts to sort out later.
Apologies, most of my posts are much lighter in tone than this one, but the short version of this post is that I like to imagine, modestly, that I, and everybody else who writes here, has thought about their words, and a machine is never going to understand that.
Or am I, perhaps, just a luddite, tilting at windmills or trying to stop the tide from coming in.
Yes, that’s probably it. And I do accept that automatic translations have their place - if they can help understanding, then that can only be a good thing.
But I like translating, and I enjoy interpreting. So consider this post my somewhat pathetic attempt to fight back on behalf of Team Human Translator.
Normal service will be resumed in my next post. Promise.
I don’t have many photos of translating or interpreting (well, I do, but I’m in them which makes them rather boring) and so I’ve illustrated this post with some photos I took on a single day interpreting around London – I wonder what Dickens would say.
If, quite understandably, the thought of subscribing is too much to contemplate, but you’ve enjoyed reading something I’ve written, then you also have the option to simply buy me a pen so I can continue to translate the old fashioned way (contributions may in fact be used to buy beer, but I won’t tell if you don’t ask).






As I finished the third leg of a five-leg translation ultra-marathon today, what struck me was just how often we (the humans) rearrange the connective tissue in a text so that it is readable, cohesive, and, therefore, engaging, yet manages to retain the original intended meaning and purpose. The gangly bits sticking out of a so-called automatic translation (a literal translation from many Latin languages, as it happens, and as you are well aware, of what began life in English as "machine translation") are nowhere to be seen. This is because the human is not automatic: he or she has years of thought behind every deft victory of functional equivalence and a sure-footed knowingness that the spat-out word strings of automatic piglets wearing lipstick cannot hope to match, no matter how much "compute" is thrown at them.
Aren't you glad I did not bring up the polished turd trope? Oh, but I will now! I have a Finnish friend whose Facebook posts in Finnish would be translated into Portuguese because of my FB settings at the time. What a mess! (I could tell that English had been the bridging language in that FI-PT process.) Sometimes, she would come back later and post her own translation in competent English (her L2). Suffice to say that the FB instant trashlation was way off the mark more often than not. Sure, AI and LLMs have made everything look a little more, er, polished, but they are not texts I care to read. They irk me something terrible. (I wrote that last sentence like that on purpose, btw.)
I, too, am a word-girl, words mean things and I pick them very carefully.
I also spent many years as an American Sign Language Interpreter and know the struggles of picking the right words in English to match the ASL and the right ASL to match the English, interpreting words into pictures and pictures into words. So I totally feel you on that (that’s another thing altogether, idioms!)
However, Substack is a global thing and without some type of translation tool, I would be lost 🐝