Listen

Description

Click above to hear what’s below.

Translation in transition

In translation, there’s a tsunami of change underway. I’ve been talking to some colleagues and it’s a grim picture. Meanwhile, the Justice Minister resigned because his department of Justice is broken. Like my statuette of her.

Thanks for reading The Endless Chain! Subscribe for free to receive new posts and support my work.

A few years ago, in 2021, I phoned an old contact to ask if he’d work with me on revising a translation of the annual report of a bank. He’s a banking specialist. He as good as laughed at me: as if he would work for me!

I pressed him. I asked what he’d charge, because it’s ultimately about money. Minimum 20 to 25 cents a word. I work for nine and less and count myself lucky. Clearly, he’s a far better translator than I’ll ever be. But I wonder if now, in 2023, he’s still got a ream of clients paying 25 cents a word. He posted recently on LinkedIn: We all need to specialise. I recounted a bitter experience, from long ago: I laughed at him, but this time in bitter irony. And one of his contacts commented: You’ve obviously got the wrong clients. It was a lady, but I think, if she’d been within arm’s-length distance, I’d have slapped her smug face.

The Italian Job

A big translation firm in Italy contacted me out of the blue and presented me ultimately with the final irony. They had a very large document that had been translated by a machine (DeepL, it’s called). Could I ‘post-edit’ it? I could have started with their e-mail, but I don’t like to be rude:

Recently many clients apply this type of service as post-editing of AT texts in order maintain the budget and reduce the costs and delivery dates.

Please see here attached a sample of pretranslated text.

The rate for post editing of pretranslated text is Euro 0,0350 (note:  taking into consideration new texts to arrive. The client has concluded a new contact so we expect many new texts to translate. That’s why we are looking for the translators to dedicate for this on going and future projects). [sic]

I was mildly amused that the price extends into the 10,000ths of a euro. Still, it looked good. A promise of a whole heap of future work. However, a caveat: if I had actually been given the whole heaps of future work that were ever promised to me, I’d be buried under Mount Everest by now. That said, hope springs eternal in the human breast.

You must understand that AI is how we now translate. We translate using what other people have already translated (because it’s so good) or what a machine thinks would be a good translation (so, we now have Europeanisms and Americanisms everywhere, because we are an ecumenical church). It can be quite a good method, but it holds dangers. A local food store here has to label British produce in Dutch. Their programming translated “still water” (no bubbles) as “nog steeds water”, or “hasn’t changed from being water” … it’s still water … (and mislabelling can result in an administrative fine for them).

* And some people go bananas—En sommige mensen gaan voor bananen—Et certaines personnes deviennent bananes

* and they split—en ze smeren ’em—et ils sont séparés

* text into bits—tekst in bits—texte en morceaux

(these are actual results from Google Translate). Put them all together:

* And some people go bananas and they split text into bits—En sommige mensen worden gek en splitsen tekst in stukjes—Et certaines personnes deviennent folles et divisent le texte en morceaux

However, what you get to edit is:

* En sommige mensen gaan voor bananen en ze smeren ’em tekst in bits, which roughly reverse-translates as “And some people go for bananas and clear off out of there text in units of an eighth of a byte.”

* Et certaines personnes deviennent bananes et ils sont séparés texte en morceaux, which is nonsense, because, as everyone knows, personne is feminine, so it should be ELLES sont séparéEs en moreaux, once they’ve become bananas, that is.

The text is in such disarray owing to poor segmentation (a segment should ordinarily be a complete sentence, or at least a complete clause up to a colon or semi-colon, but that is rare), and this split segmentation comes from poor preparation of the text by the agency or poor formatting by the writer. There are still people who use the spacebar instead of tabs and who fail to use full stops. They don’t read their own text before putting it into a machine to translate, and they then leave the translator to sort out, not only the translation but also the source text. All for three-and-a-half cents a word. Translation usually sells at around ten to 15 cents per word. (More for a bank, obviously.)

Poor segmentation was a feature of this job, so I asked if it could be sorted out. They didn’t do it. Remember: my job was post-editing. Their job was the machine translation part. But they machine-translated a text split into morceaux.

Where the text was actually coherently in full sentences, the AI translation was commendably acceptable. It wasn’t Shakespeare, but in any case the text was a telegraphic list of features of a cannon to be fitted on deck of a naval ship.

There was a price list for various aspects of this war material at the end, and the figures were mind-boggling. There was a time line for all aspects, including translation. Three months for a translation. I was given three days, with the great help of the machine, of course. I checked everything over. Vast swathes of it passed muster, even if the original text wasn’t in full sentences. I tidied a few bits of word order and made the terminology consistent where I understood it. I think I know a muzzle from a recoil.

The agency was indignant. I had changed 100% NOTHING, they said. No, I replied, I changed something, but it was on the whole a remarkably good AI production. Bravo AI, we are right to fear you and to look to our laurels, which will soon be yours! Summarily, the agency cancelled the job on the ground of quality deficiency and refused to pay for the three days I had devoted to the task. Not one cent.

I haven’t the energy to contest it. I will write and ask if they could send me the final version (as well as sending them this text), so that I can see what essential changes I missed. They will not refuse, I know. They will simply not answer. They’re in Italy and Italian law applies. Like I say, I haven’t the energy.

In short, I’m reduced to checking what a machine does, which is so good that it is taking my job from me, but if I don’t find enough fault with its perfection, I lose my job completely. Well, that’s what it seems like.

Reality check: what is checking?

Yes, what is “checking”? MTPE is the newish (since 2018) phrase meaning machine translation post-editing; it comes hot on the heels of revision and proofreading and checking, and all of these are different tasks, and they all depend on who gives you the task to do.

With proofreading, you read something just before it goes to press, or publication. Your job is to check spelling, grammar, context, flow, perhaps even style and the like. You read one document, and correct it.

Checking is whatever you want it to be. There is fact-checking, which carries great responsibility, and there is layout checking and there is “dirty checking”. When I worked at PricewaterhouseCoopers, they would come and ask for a dirty check. Not the full Monty, but just to take out the worst of the worst. That’s when the fun starts: Why did you change Member States to member states? I’d say, states are states, not States. States is not a proper noun, and only has a capital at the start of a sentence. The EU writes Member States, they’d say. The EU can do as it pleases. I proofread according to the rules of grammar, not the rules of the EU. You want me to learn the EU’s rules for my language? And who else’s rules? The Chicago Style Book? The Economist Style Book? The Washington Post? The Times? Encyclopaedia Britannica? Don’t tell me—DeepL? Do they even have a style book? You can probably ask DeepL to translate s…



Get full access to The Endless Chain at endlesschain.substack.com/subscribe