Chat GPT for Literary Translation: Test No. 1

I’ve been following Tom Gally’s YouTube videos on various aspects of Chat GPT and foreign languages, including his several videos on the software’s potential use in translation. It’s surprisingly good, the software, I mean. I thought why not try it on a text I had already translated to see what it could do. So I gave it the text below, from the opening pages of Miljenko Jergović’s Rod, and the simple prompt, “Please translate the following text from Croatian to English.”

The source:

Čudni su, priča se, bili nedjeljni ručkovi kada bi se okupila cijela obitelj. Postojala je stroga jezička procedura, kakve danas, vjerojatno, ima samo u sjedištu Europske Unije, a da se nitko nije upitao zašto tako treba biti. Otati Karlu silno je bilo stalo do njegovoga nijemstva, i njegove odabranosti među drugim ljudima, čemu su se svi drugi morali prilagoditi. Zauzvrat, nitko im, a ponajmanje on, nije branio da budu ono što jesu i da između sebe govore na jezicima koji su im po volji. Otata je volio svoje zetove i nije mu smetalo što nisu Nijemci, ali je, recimo, bio ponosan na njihova građanska zanimanja. Pripadanje željezničarskome cehu za njega je bilo poput pripadanja nekoj tajnoj organizaciji, masonskoj loži, čiji se članovi od drugih ljudi razlikuju po pogledu na svijet, ali i po vlastitoj ulozi u svijetu. Nijemac željezničar sa željezničarom Hrvatom čini bratstvo koje se međusobno bolje razumije nego bilo koji narod unutar sebe. Otata Karlo bio je ljevičar, pa je početkom dvadesetih godina dvadesetog stoljeća bio zaglavio u zatvoru, a onda i izgubio posao, jer je podržao štrajk željezničara. Ne bi to bio skandal da nije bio šef stanice i da nije bio Nijemac među divljim Slavenima, pa je zbog iznevjeravanja svoga nacionalnog i kastinskog položaja oštro kažnjen od kraljevskih vlasti.

Chat GPT’s response:

It is said that the Sunday lunches when the whole family gathered were strange. There was a strict linguistic procedure, which probably only exists today at the headquarters of the European Union, and no one wondered why it had to be that way. It was very important to Otata Karlo that his silence and his chosen status among other people were maintained, to which everyone else had to adapt. In return, no one, least of all him, prohibited them from being who they were and speaking in languages that suited them among themselves. Otata loved his sons-in-law, and it did not bother him that they were not Germans, but he was proud of their civil professions, for example. Belonging to the railway guild was for him like belonging to a secret organization, a Masonic lodge, whose members differ from others in their worldview and their own role in the world. A German railwayman with a Croatian railwayman forms a brotherhood that understands each other better than any people within themselves. Otata Karlo was a leftist, so in the early 1920s, he was imprisoned and then lost his job because he supported the railway workers’ strike. It wouldn’t have been a scandal if he hadn’t been a station chief and a German among wild Slavs, so he was severely punished by the royal authorities for betraying his national and caste position.

And for comparison’s sake, here is the published version in my translation, from 2021:

They say the weekly meals at which the whole family would gather were quite something. There was a strict language protocol of the sort that today probably only exists at the headquarters of the European Union, though no one seems to have wondered why it had to be that way. Karlo’s Germanness was especially important to him; everyone else around him would have to adapt. In return, no one, least of all he, prevented them from being something other than who they were or from speaking whatever languages they pleased. My great grandfather loved his sons-in-law, and it didn’t bother him that they weren’t German; rather, he was proud, I should note, of their civic calling. Belonging to the railroad workers’ trade was for him something like being in a secret society, a Masonic lodge of sorts, whose members differed from other people by their understanding of the world and their own role within it. A German rail man and a Croatian rail man enjoy a brotherhood that allowed them greater mutual understanding than any members of a single nation among themselves. Opapa was a leftist, and in the early 1920s he ended up in prison and later lost his job for backing a rail men’s strike. It wouldn’t have been a scandal if he had not been a stationmaster and a German among the barbarous Slavs. He was harshly punished by the royal government for the betrayal of his caste and his nation..

Chat GPT’s version is remarkably good. There’s really only one “mistake” per se, the interpretation of “nijemstvo” as “silence” rather than “Germanness,” which renders the passage a little nonsensical. But if you were reading quickly, you might not notice. Now this could be a bigger problem that it might at first seem. The software is really good at making things sound right, giving the impression of smoothness, that commonplace of translation discussions, such that one might not question the result.

Otherwise, the differences between the two versions can be understood as differences in interpretation and stylistic choices. If I were to use the software as one tool in my box, generating a first draft, for instance, I wonder how it would affect the overall translation. Would it seem less spontaneous, less alive somehow, or could I, watching out for this, prevent it and the various other erroneous inventions it might come up with?

I’ll be checking out other passages, from other languages, with dialogue, slang, and other challenges translators often face. As well as how to use the software in teaching, which I’ve already tried with a group of students this term.



  1. I’m wondering whether ChatGPT simply scanned the web until it found your translation and then used bits and pieces of it. Isn’t that why the Hollywood writers are about to go on strike: they do not want to have to fix ChatGPT content, since the program freely plagiarizes.

    1. A good question. Comparing the two versions suggests to me that Chat GPT didn’t plagiarize my version, but the only way to be sure would be to give it a text that I was sure had not been translated. Tom Gally did that with Japanese (something in a Japanese language magazine published a few days before he did his test) and found the results were the same. I’m convinced by his test, but I can certainly do one of my own next. That can be Test No. 2.

Leave a comment

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: