In the tail end of 2022, two major releases in the field of deep learning AI kicked off a snowball effect that, even now, is transforming the online world at a rate that no single person can perceive. Stable Diffusion – an AI Art generator, was released to the public with an open source; and a month or so later, ChatGPT – an AI text generator, was made publicly accessible as a ‘Free Research Preview‘ with scant restrictions.
I, like many others, hopped on and began experimenting right away, and in many cases, people began iterating and integrating it into software at a breakneck pace. I’m not going to go into all of those things here, as it’s very impractical to summarise when it’s still growing explosively, but anyone reading this within the next few years should have a very good idea of what’s going on, anyway. I figured out pretty quickly that this was about to become a new Wild West of the internet, and possibly some form of technological singularity.
Anywho, ChatGPT is the main focus of my article here, as I already did some early experiments with Stable Diffusion over on my Youtube Channel. After some rapid, mind-blowing experiments in December, I eventually embarked on a project that resulted in myself publishing what may quite possibly be the world’s first novel, completely translated by ChatGPT.
In fact, I published it back in January, and I just haven’t had the time to write about the experience. If you’d like, you can skip ahead and look at the Amazon Kindle store page by clicking here. It’s also available for less in my Patreon archive. (Well, I call it an archive but it’s a bit unorganised over there…) If that’s not to your taste – don’t worry, I’ll be sharing excerpts.
The entire process of translating a 70,000 word novel with this method took about three to four weeks, throughout January of 2023. And, I’m sure some of you are thinking, ‘Machine translating a whole book? It must be terrible!‘ – but of course, the fact that I somehow produced what I believe to be a 99% natural-sounding book with it is the entire point of my writing about this. I was very hesitant in the beginning, and initially pumped the first chapter into ChatGPT on a whim, and when it turned out to be too good, my curiosity made me continue with the second… and the third. And then, I thought to myself, ‘Well, now that the sample is done, all that’s left is a full extrapolation! Why not attempt the logical conclusion!‘
Illegible machine translated novels, however, are something I’m somewhat aware of. With the death of anime fansub culture in the early 2010s, the next generation of hobby translators turned to Japanese web and light novels – and these days, both Chinese and Japanese web novels, and their unofficial, clunky English translations, are very prolific. The bar is also much lower now, since it’s easy for someone with no language skills to pump text into a machine translator – unlike before, when you needed to be able to listen and transcribe subtitles. This isn’t a scene I follow too deeply though, outside of noticing that people complain about the poor English. That said, during my research, I found a Kindle version of the same novel I chose as my subject, filled with reviews about the ‘obvious’ machine translation.
Oh, I had ought to mention what I actually translated, too. There’s a bit of a story behind why I chose this book, but I decided to make my first subject ‘Around the World in Eighty Days‘, by Jules Verne. As a French author who contributed to the very concept of science fiction itself, it ended up feeling like a very appropriate decision to represent this new AI frontier. He’s also one of the most translated authors in the world – I wanted this project to also be very cognizant and respectful towards the field of translation, and out of the strong possibility that ChatGPT and others may heavily disrupt careers very soon, I wanted my first subject to be a story that had both a large sample to compare to, and not really affect the current industry at all. Was that the right decision to make? I have no idea.
I also have a particular set of high standards, in all my work: I didn’t want to simply copy-paste the whole thing. Even though I was experimenting with ChatGPT at the same time, I also wished to honour Jules Verne and present him fully, so I went through the entirety of a normal editing, review, and self-publishing process, with heavy amounts of historical, cultural, and language research. And as a cherry on top, I recreated the original first edition cover art from 1873.
I have a lot of random multimedia and technical skills.
Though incredibly fast by current standards, given the amount of effort I put in, I decided to initially publish this whole text as a Kindle book. I even chose a new pen name to separate this particular type of work – ‘Edward Sharpe‘ – though I’m not sure what I was going for, there. Is it strange to use a pen name behind another pen name, behind another pen name? It’s probably a little bit much.
All in all… well, what was I trying to say here again? I’ve had a lot of difficulties in the real world that prevented me from writing about this straight away, though I also admit that the sheer quality of ChatGPT’s translation is difficult to describe succinctly unless you just read it and compare for yourself. Or, maybe I’m overinflating the whole thing?
I’d like to publicly examine some of my results in a short series of articles, though I’m not really sure how long I’ll do this for. Let’s first look at the opening paragraphs:
1872 (Verne)
Dans lequel Phileas Fogg et Passepartout s’acceptent réciproquement, l’un comme maître, l’autre comme domestique.
En l’année 1872, la maison portant le numéro 7 de Saville-row, Burlington Gardens, — maison dans laquelle Shéridan mourut en 1814, — était habitée par Phileas Fogg, esq., l’un des membres les plus singuliers et les plus remarqués du Reform-Club de Londres, bien qu’il semblât prendre à tâche de ne rien faire qui pût attirer l’attention.
À l’un des plus grands orateurs qui honorent l’Angleterre, succédait donc ce Phileas Fogg, personnage énigmatique, dont on ne savait rien, sinon que c’était un fort galant homme et l’un des plus beaux gentlemen de la haute société anglaise.
On disait qu’il ressemblait à Byron, — par la tête, car il était irréprochable quant aux pieds, — mais un Byron à moustaches et à favoris, un Byron impassible, qui aurait vécu mille ans sans vieillir.
1873 (Towle)
In which Phileas Fogg and Passepartout accept each other, the one as master, the other as man
Mr. Phileas Fogg lived, in 1872, at No. 7, Saville Row, Burlington Gardens, the house in which Sheridan died in 1814. He was one of the most noticeable members of the Reform Club, though he seemed always to avoid attracting attention; an enigmatical personage, about whom little was known, except that he was a polished man of the world. People said that he resembled Byron–at least that his head was Byronic; but he was a bearded, tranquil Byron, who might live on a thousand years without growing old.
2023 (ChatGPT, raw)
In which Phileas Fogg and Passepartout accept each other reciprocally, one as master, the other as servant.
In the year 1872, the house at number 7 Saville Row, Burlington Gardens, which was the house in which Sheridan died in 1814, was inhabited by Phileas Fogg, esq., one of the most singular and most notable members of the Reform Club in London, although he seemed to take pains to do nothing that would attract attention.
To one of the greatest speakers who honor England, this Phileas Fogg, enigmatic character, succeeded, of whom nothing was known except that he was a very gallant man and one of the finest gentlemen of the upper English society.
It was said that he resembled Byron, in the head, because he was irreproachable in terms of feet, but a Byron with a mustache and a beard, an impassive Byron who would have lived a thousand years without aging.
After Editing
In which Phileas Fogg and Passepartout accept each other reciprocally, one as master, the other as assistant.
In the year 1872, the house at number 7 Saville Row, Burlington Gardens – which was the house in which Sheridan died in 1814 – was inhabited by Phileas Fogg, esquire, one of the most singular and most notable members of the Reform Club in London – although he seemed to take pains to do nothing that would attract attention.
To one of the greatest speakers who honour England, succeeded this Phileas Fogg, an enigmatic character, of whom nothing was known except that he was a very gallant man and one of the finest gentlemen of the upper English society.
It was said that he resembled Byron, in the head, because he was irreproachable in terms of feet, but a Byron with a moustache and a beard, an impassive Byron who would have lived a thousand years without ageing.
There are many things throughout my adventure here that took me aback, but to have a translation discrepancy in the very first paragraph of Verne’s novel was not something I expected. Towle, the original English translator, merges all of the first three chapters into a less flowery opening that’s half as long. The proper, more literal translation of the second paragraph, however, highlights just how flowery it was – it’s actually rather difficult to comprehend. In the published version, I actually rephrased this a little bit, but mostly kept it as-is.
Equally interesting, however, is the last word in the chapter title: ‘domestique‘, ‘man‘, and ‘servant‘. The proper translation of ‘domestique‘ is, in fact, ‘servant‘… of which I get the feeling that this liberty was taken due to the falling use of domestic servants in this time period, and the historical connotations of that word. Research told me that in English versions of this book, the word ‘valet‘ is predominantly used, but after studying the historical usage of various synonyms in the English language, I chose a third option.
From this, I hope you get the idea that I supervised and reviewed ChatGPT’s output with a fine-tooth comb… and barely changed a thing. So one could argue on a technicality that ChatGPT isn’t really capable of translating by itself… But does it really need to? Double-checking and fixing the work is simply a part of the process. It also happened to be the similar fates of other industries that were replaced by technology – for example, traditional animators became reference artists when 3D animation came into existence, and 3D animators became cleanup artists when motion capture became viable. Similarly, I have a passable background in translation, and here I cleaned up ChatGPT’s machine translation…
Is that ultimately a good thing, or a bad thing? I suppose you can make your own conclusion there. I’m just sharing a crazy experiment I turned up to eleven.
Eiri Sanada 