Hopscotch Summer 2023 Translators Forum: A.I. & I
“I readily admit that a few of his pages are worthwhile, but these pages are not my salvation, perhaps because good writing belongs to no one in particular, not even to my other, but rather to language and tradition. (…) I don’t know which of us wrote this.”
– Jorge Luis Borges, “Borges and I,”
translated by Kenneth Krabbenhoft,
Selected Poems
THE PROMPT
It’s hard to imagine any field of human endeavor that will have been spared a bout of self-reflection, not to say soul-searching, since ChatGPT was launched on the world in November of last year. The chatbot’s ability to fabricate works of art has led many commentators – and artists – to reassess their understanding of the creative process. Yet in the field of literary translation, where the notion of artistic originality has always been radically unstable, such questions are hardly new. We at Hopscotch thought it was high time to hear from the translators themselves…
We asked a number of our peers whether the advent of OpenAI and the increasing sophistication of other automatic translation tools had changed the way they thought about their own literary translation practice, whether in linguistic, aesthetic, ethical, or commercial terms. Below is what they had to say!
THE RESPONSES
Ilze Duarte
The increasing encroachment of artificial intelligence tools on the publishing industry has changed the way I think about my literary translation practice not in the sense that it has altered it—it has not—but in the sense that it has led to an intensification of my beliefs about what I am and what I do as a literary translator: I am a creative writer. Even as artificial intelligence becomes more and more sophisticated, I do not believe machine translation will ever produce texts of literary merit. Writers and translators create texts of literary merit.
Unlike a translating machine, I do not use an algorithm that inputs A (the text in Portuguese) and outputs B (the text in English). My translation work starts with a deep reading of the text so that I can understand its meaning, that is, the story that it tells as well as other layers of meaning such as metaphors and cultural and historical references. This deep reading also allows me to comprehend and appreciate the elements that make the work unique, including its tone, register, and musicality and the ways in which these qualities are evidenced in each passage and interwoven in the work as a whole. As I start composing the translation, I use my knowledge of Portuguese and English as well as Brazilian and U.S. culture, my literary sensibilities, and the linguistic devices at my disposal to recreate the work in English. My translation will bear my own signature, my unique writing style.
A machine lacks all these attributes. It has no lived experience, no judgment, no aesthetic sense, no voice. The software will look for patterns and regularities in its language model and choose the most likely synonym phrase by phrase. As every literary translator knows, the unit of translation is not the word or the phrase but the entire work. Even those of us who start translating a book without having read it in its entirety beforehand will go through multiple revisions to arrive at a translation that befits the text as a whole rather than engage in high-probability guesses of equivalent words at the sentence or even paragraph level. Machines will continue to improve in their ability to generate text, but I doubt they will ever capture sarcasm, humor, ambiguity, cultural context. I doubt machines will ever do the interpretive work necessary to render a literary text well—artistically—from one language to another.
Perhaps I am being terribly naïve and there is indeed a good chance machines will produce translations of literary merit in the future. If that is the case, we will need to keep on fighting, and fight hard, for the right to make a living out of translation and for the right to enjoy through this work the quintessentially human endeavor that is literature.
For the moment, I would like to think that publishers committed to promoting literature in translation will help protect the integrity of the art and craft that allow it to exist by continuing to pay translators for their work and rejecting any form of machine translation. However, I understand publishers may be tempted to use machine translations because they are faster and cheaper. I suspect many readers will be unable to distinguish between a text translated by a qualified translator and a text cobbled together by a machine. Further, I suspect many readers will not care how the text they are reading has come to be.
It is our job as translators to inform the public of the artistry that goes into the recreation of a literary text from one language to another and to help the public appreciate the uniqueness of each translation as well as the possibility of multiple interpretations of any given work of literature. We must write to the general press to demand more coverage of the art of translation and offer our expertise in the form of essays and interviews. The recent New York Times series on translation, including the outstanding essay by Sophie Hughes unveiling her line-by-line translation process, exemplifies the sort of promotion the work of literary translation needs and deserves. As the #namethetranslator movement has been successful in highlighting the importance of acknowledging translators for the essential work they do, so can a widespread movement succeed in raising awareness of the type of work involved in literary translation and the reasons why only we—humans, writers, artists—can do it.
The increased threat of artificial intelligence to translation work has also strengthened my belief that I need to be fully engaged with my colleagues and participate actively in our professional organizations. To stay informed of the issues facing literary translators and engage in action to address them, in addition to renewing my membership with American Literary Translators Association I have recently joined Portuguese to English Literary Translators Association and The Authors Guild. I have signed The Authors Guild’s letter calling on the leaders of generative AI companies to compensate writers fairly for the use of their work, which these companies have been feeding massively to their so-called large language models at will and for free.
Of course, such initiatives are just the beginning of a long and probably protracted fight. We need to campaign for legislation to protect our intellectual property and for the adequate enforcement of any new laws.
Georgina Fooks
AI is seemingly everywhere, and yet I have been trying my best not to think about it. Artificial intelligence, two terms that are both slippery if you try to pin them down. But lately I feel the need to try, to untangle myself from the web (if such a thing is possible) and to see things clearly.
The way AI works mimics our journey with language. Models are exposed to vast amounts of data, just as babies swim through worlds of words, slowly pinning meaning onto referents. This is green, this is blue, this is sky, this is tree. It learns how to connect concepts and to construct sentences. But is learning the right verb?
One reason I feel such a personal affinity with the act of translation is the way it exposes me intimately to the languages I am learning (and no matter how ‘fluent’ I get, I will always feel like I’m learning). Translation is an interface between two idioms, the language of the writer and the language of the translator. My English – or my Englishes, shaped by all the people I have met, the places I have been – touching their Spanish, their Spanishes. And in the gaps, where we don’t quite align, where our languages don’t touch, is where I learn the most. It’s the spark of electricity, the lightning in the sky seeking earth.
AI can write, but it can’t read. It can only process. And it’s the reading of the translator I value most. As translators’ labour has become increasingly visibilised (although there’s a long way to go), there’s a reason why readers gravitate towards the works of certain translators. The role of curation, firstly, but also the translator’s style. When Yves Bonnefoy translated Hamlet into French, he described the experience as lending the intensity of an accent to a voice. Whether the translation is the sharing of accents or the blending of voices – I’m not sure – there is a humanity that flows through, the possibility of connection.
I don’t feel threatened by the impact of AI on translation as a creative pursuit, because humans are always wanting to create. It’s heartening to see my favourite translators brush off the threat to their livelihoods – from Sophie Hughes in the New York Times saying that the rudimentary skills of Google Translate don’t keep her up at night, to Anton Hur affirming the importance of a translator’s reading of the text. And I’m also intrigued to learn more about the practices of translators who use AI as a starting point alongside other tools, working with the original and the AI translation as an intermediate version. Importantly, the agency of the translator remains intact.
But I’m always wary of the looming presence of a hypercapitalist future, and the obsessive way corporations invest time and money in the name of ‘optimising’, which usually ends up being a euphemism for the elimination of every human element from the process. I also hear more and more about the environmental impact of running the servers needed for AI. In July, Uruguay experienced a record drought while plans were underway to guzzle millions of litres of water a day to cool the servers at a proposed Google data centre. And in a moment of increasing inequality and global precarity, and when there’s already so little money in translation, it’s easy to imagine AI as one of the tools used to drive down the price of our creative work. I keep returning to the same question. Why would we delegate to a machine that which humans have always loved to do? To create.
As I look ahead to teaching literature and translation as part of my PhD, I have been thinking of what intimacy with a language has taught me. Me at my desk, the dictionary, the pencil, the text. Me in the street, on the bus, at a café, learning from every misstep, every embarrassing mistake. I don’t want to put AI between me and my languages. I want to be there, translating like lightning. Tracing patterns through the air, writing my way through, touching the ground.
May Huang
A thought crossed my mind as I spent the better half of a recent long weekend poring over my translation manuscript, fact-checking two dozen poems head-to-toe and proofreading each last punctuation mark: for me, translation has never been about “efficiency.”
When people (who are selling AI products) tout the benefits of AI, the same phrases crop up time and time again: increased efficiency. More streamlined workflows. Improved speeds. AI helps accelerate your processes, freeing up time to focus on “work that truly matters.”
When I translate, I like to maximize the amount of time I spend with a text—re-reading, revising, revisiting. The work of a poet and translator is iterative by nature. Especially when I translate poetry, spending “quality time” with the work feels crucial to the experience of translating it; you’re slowly peeling back the layers of the “onion” to get to the core of what you’re trying to translate. I’ve always liked to say that literary translation is personal—the translated text is shaped by my relationship to the text, to the author, to a shared literary imagination. When I approach a translation project, the “dataset” I pull from comprises my upbringing, the books I’ve read, the poets I’ve studied, the emotions I bring to a text. As intelligent as today’s automatic tools may be, the dataset they haven’t been trained on is my own lived experiences that shape who I am as a translator.
Granted, there are people and professions that benefit greatly from AI tools like ChatGPT, which can make life easier by doing the reading and analysis for you. Feed a text into an AI-powered tool and ask it to pull out key themes, or have it tell you the plot summary of a book. ChatGPT can do this and more, because it has mined reams and reams of available data on the internet to give you an answer. All of this saves time.
In a society where “productivity” always seems to be the goal (in labor, in technology), I believe there’s infinite value to a creative process that’s not anchored in so-called “efficiency.” My literary translation practice is rooted in the slow, steady work of spending a lot of time with and thinking about a text. Ultimately, to me, that’s the “work that truly matters.”
Julia Pelosi-Thorpe & Lourdes Contreras
A dialogic reflection between Julia Pelosi-Thorpe and Lourdes Contreras, co-translators of Marzia Grillo’s 2022 short-story collection The Sun’s Point of View.
JP: I’m not sure to what extent my experience is shared by you, Lourdes, or others, but OpenAI and digital translation tools exist for me as reference works in the same way as a dictionary. And although the particular works I consult constantly shift—and the way I read in Italian, Dialects, Latin (and English) changes through the years too—my overall process pivots on how much I enjoy thinking about the senses words can have in different contexts. I like to make a messy, fast first draft alone with the page, but this is followed by waves of editing that hinge on consultation and deliberation, and I know we’ve turned to a suite of references in the translations we’ve done together…
LC: These AI translation tools take on a similar role in my practice, although I find that each has a particular function in making my translations feel complete. Before my final edits, I utilize OpenAI tools to confirm certain expressions that are not part of my quotidian speech. I grew up bilingual, and the version of English I spoke at home was melded with adages that were always seemingly one or two words off of what is commonly said in the United States. For both Spanish and English, ChatGPT assists in confirming what is idiomatic and what is an invention of a bilingual upbringing. In turn, translation tools such as DeepL help me correct translation errors in all languages but particularly in Italian rough drafts. These tools, I’d like to think, exist outside the act of translation and help smooth out the edges that come with speaking, reading and writing in different languages.
JP: We have such different experiences of English and Italian, and I think this informs not just our approaches to reference works like AI but our deep enjoyment of language as we tinker with syllables, words, phrases. We were both raised in multilingual families but on opposite sides of the globe… Despite popular culture, I didn’t expect as many divergences between Australian and American Englishes (catchall terms that themselves hold multiplicities) as I found when I moved. I realise writing this that I don’t even have a clear understanding of which Englishes OpenAI are trained on. And then there are the regional differences within Italian…
LC: Often we come across Italian phrases that are region-specific, unfamiliar, or don’t have an English equivalent. If they do not have an exact English counterpart an understanding of the variety of English idioms is necessary before we can even begin an accurate search. Our discussions about these phrases often enrich my understanding of the text and also of idiomatic speech in English in both Australia and the US.
JP: Consulting different texts feels so enriching, and reminds me of the Borges quote prefacing our reflection, and that people who create something invariably have references. And so they organically do the same process that ChatGPT’s model is doing artificially… I imagine things will change in the future, but right now these tools are not really on my radar for literary translations. Take this sentence we recently translated from Marzia Grillo’s short story Matryoshka:
E i padri, allora? recitava un manifesto sul muro, accanto al civico 15 di una casa in affitto a un prezzo ancora ragionevole.
What about fathers? read a poster on the wall near number 15, a house that had been leased at a still-reasonable price.
I can’t remember whether we consulted ChatGPT, but when I ask for an English translation today, August 2023, it does not toy with the literal as much as we did:
And the fathers, then? It recited a manifesto on the wall, next to civic number 15 of a house for rent at a still reasonable price.
DeepL comes closer to our own choices for individual words, giving poster as manifesto and house number as civico:
What about fathers, then? said a poster on the wall, next to house number 15 of a house rented at a still reasonable price.
All three AI translators keep the syntax closer to the source text’s own architecture and therefore Italy’s palazzo architecture, while our reading is more ambiguous… ChatGPT could modify its translation in “collaboration” with our wishes. But whenever we problem solve together and experiment with language, there is always an outcome that depends on the context of the piece, and this is something I appreciate about the way we can play with sentences together and that AI perhaps cannot currently grasp.
LC: Even the translation of the word allora is interesting. Our English seems to evoke the same interrogative mood as the Italian text, the word then feels unnecessary. However, AI translators can only work off of the exact text given. Co-translation brings these kinds of conversations to the surface because literary translations require an understanding of changing colloquial and formal language. AI would need to keep up with the velocity at which words change meaning, or become less or more intense.
JP: Yes! It’s wild to me that ChatGPT would be so good at so many things but less at literary translation; it reiterates this feeling that I have that the nature of spoken language is so intensely human.
Trask Roberts
Over the past nine months speculation has abounded (some of which has already been borne out) about which jobs are most likely to be taken by ChatGPT. The media provide new headlines at a dizzying speed: like this New York Times’ “Tinkering With ChatGPT, Workers Wonder: Will This Take My Job”; The Wall Street Journal’s “The Robots Have Finally Come for My Job”; or the Economist’s “ChatGPT could replace telemarketers, teachers and traders,” with its unfortunate and almost gleeful subtitle, “Here’s why that is no bad thing.” As translators, teachers, and researchers, we likely find ourselves at odds with this conclusion. But it is undeniable that natural language processing models will have profound effects on the work we do and the texts we read. That said, I would like to focus on a particular interest of mine, self-translation. Self-translation is a practice less likely to be “disrupted,” as start-ups are so wont to do, by ChatGPT, for the two simple reasons that one, few self-translators engage in the practice for pecuniary benefits, and two, on a more foundational level, you cannot self-translate with AI since ChatGPT is not yourself (this does not rule out the possibility, of course, for collaborative translation). For the moment, ChatGPT has no self at all. Let’s hope it stays that way.
The program will admit just as much when asked. Ask ChatGPT if it is capable of self-translation and it will tell you: “As an AI language model, I can’t translate myself directly, as I am just a program running on servers and do not have a distinct identity or language of my own.” I would wholeheartedly agree with the premise that at the heart of self-translation is a complicated, humanistic play of identity and language. But when pressed, and only slightly, with a prompt such as, “write a haiku and translate it into French,” ChatGPT responds:
Sure, here’s a haiku:
Amidst falling leaves,
Whispers of nature’s secrets,
Harmony prevails.
Now, let’s translate it into French:
Parmi les feuilles tombantes,
Chuchotements des secrets de la nature,
L’harmonie prévaut.
It would seem that ChatGPT’s first answer about being incapable of self-translation was a demonstration of false modesty, if not deliberate misdirection. This is a haiku as defined generically. And this is a translation, though one that favors a strict adherence to propositional meaning over poetic form—which is to say, the second poem is not a haiku due to the number of syllables per line (though by refining the prompt it is capable of adhering to any meter and rhyme we demand). All this is done by a machine without an identity—a selfless self-translator, no doubt, both in the sense of not having a self, but also in the common sense of the word, as it was carried out for my benefit. We won’t find ChatGPT scribbling out poems and translations in its journal behind closed doors out of pure love for language and beauty. And though it could be done, I will abstain from any close reading of ChatGPT’s poems or prose as, for now, I believe our close reading energy expenditures to be better spent on those texts written by authors whose humanity is less questionable.
But as we are dragged (or perhaps march, if so inclined) into this age of ChatGPT, it may be worth looking to self-translators to remember why we write at all. Endless streams of productivity hacks peddled on LinkedIn and similar sites promise to increase your content creation tenfold. Offering solutions (at last!) to the travails of putting the ideas in your head (or someone else’s) onto paper. In a world where efficiency reigns supreme, thought takes a backseat. The self-translators we find in our literary landscapes are less selfless (again, in both senses) than our NLP models. Samuel Beckett, for example, explains that he self-translates out of a “foolish feeling of protectiveness towards the work.” Nancy Huston, looking less to protect her work than her own psyche explains that when she successfully brings her text into a new language “je me sens bien […] comme si ça prouvait qu’en fait je ne suis pas schizophrène, pas folle, puisque finalement la même personne dans les deux langues.” [I feel good […] as though it proves that I am not in fact schizophrenic, not crazy, since in the end the same person exists in both languages.] Whereas Jhumpa Lahiri, also interested in the reparative potential of self-translation, but for the text, writes, “The act of self-translation enables the author to restore a previously published work to its most vital and dynamic state—that of a work-in-progress—and to repair and recalibrate as needed.” These avowed goals of this small sample of self-translators are incongruous with the rapidity and selflessness of ChatGPT. As I think is obvious, self-translation is not in the pursuit of efficiency, but rather something deeper within both the text and its author.
Though self-translation, in the strictest sense, is an activity performed by relatively few, it can remind us of a truism too often overlooked or conveniently forgotten: writing is not a simple conveyor of thought, but also a generative act. Delegating the task of writing to AI also consequently means delegating the task of thinking, and, as I’ve been reminding my students over the past year, we must be wary of letting anyone (and, even more so, any thing) think for you. The journey of writing often leads us to unexpected places, and as our study of self-translators demonstrates, we find ourselves changed in the process. The self-translators who rethink their texts also rethink their selves. This act reminds us that the self is a work-in-progress, one that needs protecting, and one that needs healing.
Damion Searls
The following excerpted text has been published with the generous permission of the author and publisher of the forthcoming volume: The Philosophy of Translation by Damion Searls, Yale University Press, 2024. Our sincere thanks!
ChatGPT is raising once again the topic of artificial intelligence and whether or not it will replace the human writer. But while Google Translate is relevant to translation, ChatGPT is not, for a reason that reveals something deep about its limitations.
You can ask ChatGPT to “Write a story about X in Proust’s style,” but that is different from “Translate this particular Proust story into English”: the latter is precisely what ChatGPT can’t do, because to translate a story requires reading it. Whatever your definition of translation, it is a certain kind of writing linked to a certain kind of reading of another text, the original. While ChatGPT can comb and cull and copy and crib and collage—and to that extent, it can “write”—what it can’t do is read. This is why it generates references to nonexistent legal precedents and fake articles, quotes passages that aren’t real, and so on: it operates on texts rather than referring to what’s actually in the world.
Translation reveals, in other words, that when we talk about “the deeply human enterprise” of creativity or thought that ChatGPT cannot replace, what we’re really talking about is the deeply human act of reading: the act, subjective and objective at once, of engaging as an individual person with something that exists out there in shared reality.
As for Google Translate and other translation AI, I am as fascinated as anyone else by the impressive technological magic of such programs; I am grateful for being able to read websites in other languages, point my phone at wall text in a museum in another country and get a good amount of the information it contains, and so forth. As a citizen, I am concerned by the labor issue of artificial intelligence combing the world for work by other writers and then appropriating it without credit or payment. But as a translator I find talk of the existential threat posed to my career completely pointless.
In effect, Google Translate is the same kind of tool as a bilingual dictionary. I could stand at that wall text in the museum with a Spanish-English dictionary and look up all the words, maybe refer to a Spanish grammar book written for English speakers, and in half an hour or so I would get a good amount of the information the wall text contains. (I wouldn’t be able to do this with Chinese or Arabic text, of course.) If I point my phone and tap the screen, Google Translate will in effect look up all the words at once, providing English counterparts in something like the right grammatical relationships. (It doesn’t literally work by using dictionaries in this way, but that is the effect.) This is undoubtedly more efficient, but no more threatening to my job or my humanity than a bilingual dictionary.
Google Translate as it were accesses every possible meaning out there, but it cannot reliably choose among the different possible meanings. Although it seems to make different characteristic mistakes in each of the language pairs I have experimented with, I have noticed particular problems with languages where the same word means many different things in the language being translated into. “Baguettes” in French means baguettes, drumsticks, chopsticks… The word will be translated correctly in close proximity to words about a bakery, a drum set, or Chinese food, but the program can’t otherwise use common sense about the context, because it has none.
More fundamental than glitches of word choice is the fact that AI is incapable of producing utterances—it can produce only words and sentences. Semantic and grammatical units mean nothing by themselves: they are not intended, inflected, given force until spoken or written in an utterance by an actual mind. Some kinds of statement are minimally inflected, produced with minimal human subjectivity behind them—digital camera instruction booklets, a hotel receptionist telling you the checkout time, informational wall texts, online listicles—and these can be effectively produced or indeed translated by artificial intelligence. But any real utterance, which certainly includes any piece of a work of literature, argument, or rhetoric, requires inflection and intentionality in its use of language. In short, a text run through AI translation still has to be translated.
Here are two examples taken more or less at random from an art history text I recently translated, by Anne Bertrand. In each example, the first passage in English is the original put through Google Translate in 2023; the second is my translation:
Des essais littéraires du jeune homme subsistent plusieurs textes courts, certains plutôt réussis, au point que l’on peut se demander comment il se fait qu’il n’ait pu parvenir à les faire publier.
Of the young man’s literary essays there are several short texts, some quite successful, to the point that one wonders how it is that he could not manage to have them published.
Among the young man’s literary efforts were several short prose meditations good enough that it’s surprising he couldn’t get them published.
“Essais” does mean essays, but it also means attempts, ventures, which of course is why Montaigne named his new genre of exploratory prose “essays.” Here the context of literature makes the translation of “essais” as “essays” superficially more plausible; in fact, a sensitive reader would realize that “literary essays” is redundant and that saying “there are several short texts among the essays” is pointless, so the noun must be different. The French “textes,” meanwhile, does mean texts; it was the translator’s mind and knowledge of English conventions that made him decide “texts” was too vague and so sounded wrong here, and prompted him to query the author about which more specific noun—essays, stories, reviews, poems?—was correct in this case.
In the following example, the all-at-once dictionary gets each word pretty much correct but is unable to put the pieces of the sentence together in a way that makes sense in English:
Peu d’éléments subsistent quant aux tentatives de publication de l’apprenti écrivain et traducteur, alors que la littérature l’occupe encore largement, même s’il se met peu à peu à la photographie.
Few elements remain as to the attempts at publication of the apprentice writer and translator, while literature still largely occupied him, even if he gradually took up photography.
He continued to work primarily as an apprentice writer and translator, even while gradually turning to photography, but little survives of his attempts to get published.
Elements (“éléments”) is not what we’d call cover letters and pieces of writing submitted for publication, and occupied him is a bit too literal for “l’occupe”—English wants people as active subjects of its verbs, so “he was busy with work” or “he worked” is better than “work occupied him”—but the main task of the translator of this sentence is to rearrange the order the information comes in to fit the expectations of the English-language reader. The French opens with the state of the archive, moves to the nature of the documents in question and the man producing them, then reminds us that this writer is the photographer we’re reading a book about (Walker Evans); English, in contrast, puts the facts in chronological order: he wrote and translated, then he started photographing, then papers got lost. The French text isn’t trying to do anything fancy or counterintuitive to surprise the French reader, so the translation should conform just as smoothly to expected usage in English.
None of the issues in these two examples is difficult to handle, and several would never have come up if the text weren’t put through Google Translate in the first place, but I hope these examples of translation in action show the kind of engagement with language that AI doesn’t have. What we might call overall “naturalness”—a product of real readers’ assumptions and expectations—is what an actual translation has to be sensitive to.
Russell Valentino
I’ve run a half-dozen experiments using ChatGPT for translation, entering text passages of both prose and poetry from languages I know and trying different prompts. I’ve also introduced the software into my teaching, allowing student translators to use it for their translation projects provided that they include how they’ve used it in their process reflection pieces. The biggest potential virtue to my mind is similar to other computer assisted translation tools—speed, especially when dealing with long texts that have a lot of repetition. This means probably not literary texts, especially not those with pretensions to “high” art. But formulaic works such as romance novels or low-brow mysteries might work well. In one of his YouTube videos on ChatGPT and translation, Tom Gally at the University of Tokyo makes a similar claim about popular Japanese literature. (See approx. minute 8 here: https://www.youtube.com/watch?v=5KKDCp3OaMo.) The software seems to do much better with fiction and literary nonfiction than with poetry, especially poetry with any sort of sound painting in it. I tried several times, unsuccessfully, to get it to create slant rhymes. While the software cannot technically “hear,” the recognition and creation of phonetic representations of sounds is probably something it will do better in subsequent versions.
For use with literary texts that require, let’s say, greater writerly skills in the receiving culture, I could see the software being used to generate a first draft or even an alternate draft, something like the “fast pass” that many translators will do before setting about the critical tasks of researching references, inserting hidden explanatory phrasing, differentiating the idioms of characters, controlling the pace, and attending to other nuances. Would I use it to do this? Probably not, as my practice does not generally include a quick first draft. I usually try to polish from the very start, re-reading and revising multiple times as I proceed to the next portion of a text. This is just how I work. I would also be suspicious of how a quick first draft generated by AI might influence all my subsequent drafts, up to and including the finished text.
The biggest weakness I have found so far is what software developers call AI’s “hallucination” challenge. In short, this means that it makes shit up, and it does so with apparent confidence. In practice, given the trust problem that this creates, one would need to read whatever it generated almost as if one already had a translation finished in one’s mind. Otherwise, where the AI might slip in something it hallucinated, something that sounded like it could go there, might be hard to notice. This would not require proofing so much as checking every word and phrase against the source. I mean, imagine editing the text of a translator you knew had a tendency to make shit up and pass it off as authoritative. Maybe you could get good at recognizing the patterns, predicting where the text was likely to veer into pure invention. This does not sound like a time saver.
It’s moving fast, however, so I’m keeping my eye on it. I’ll likely continue to allow my translation students to use it occasionally (with appropriate commentary and reflection) and will keep experimenting to see if there are discrete tasks I can entrust to it. These look to me to be less about translation per se at this point—given its trust problem—than about invention. I could imagine, for instance, noticing something stylistically familiar in a text, e.g., this character sounds to me like a Croatian Winston Churchill, and then asking the AI to give me some Churchillisms for my English version. I believe it could do this fairly well, as long as the character didn’t have to speak in slant rhymes.
CONTRIBUTORS
Lourdes Contreras researches Italian literature, visual arts, and oral histories at the intersection of Ecocriticism, Mediterranean Studies and Gender Studies. Lourdes is a co-editor for the Bibliotheca Dantesca journal and teaches language and content courses as a PhD candidate in the Francophone, Italian and Germanic Studies department at the University of Pennsylvania.
Ilze Duarte writes short prose and translates literary works by contemporary Brazilian writers. Her original work appears in New Plains Review, Please See Me, Dear Damsels, FlashFlood and Thanatos Review, and her translations in Your Impossible Voice, The Massachusetts Review, Columbia Journal Online, Ambit, Northwest Review, Exchanges Journal, and Asymptote Journal. Her essay on becoming a literary translator is featured in Hopscotch Translation. She lives in Milpitas, California.
Georgina Fooks is a writer and translator based in England. She is the Director of Outreach at Asymptote, and her writing and translations have been published in Asymptote, The Oxonian Review, and Viceversa Magazine. She is working towards a doctorate in Latin American literature at Oxford, looking at the multimodal poetry of Alejandra Pizarnik and Susana Thénon.
May Huang (黃鴻霙) is a writer and translator from Hong Kong and Taiwan. Her work has appeared in Circumference, Electric Literature, Words Without Borders, Asymptote, and elsewhere. She graduated from the University of Chicago with honors in English and Comparative Literature in 2019. You can follow her on Twitter as @mayhuangwrites.
Julia Pelosi-Thorpe’s translations from Latin, Italian, and Italian Dialects can be found in the Journal of Italian Translation, Asymptote, Modern Poetry in Translation, The Poetry Review, and other literary magazines, and her website is jpelosithorpe.com. With Lourdes Contreras, she is currently cotranslating Marzia Grillo’s debut collection of creative autofiction The Sun’s Point of View (2022).
Trask Roberts is an assistant professor of French and translation at Kent State University. His research focuses on questions of translation, broadly construed, within the context of modern French and English language literatures. He is currently preparing a monograph focused on the interplay of self-translation and self-writing.
Damion Searls has translated more than fifty books from German, Norwegian, French, and Dutch, most recently Thomas Mann’s New Selected Stories, Jon Fosse’s Septology, Victoria Kielland’s My Men, and Bambi. His own writing includes fiction, poetry, criticism, The Inkblots (a history of the Rorschach Test and biography of its creator, Hermann Rorschach), and The Philosophy of Translation, forthcoming. He is currently a Distinguished Writer in Residence at Wesleyan University. Photo: Beowulf Sheehan.
Russell Scott Valentino is a professor of Slavic and East European Studies at Indiana University. His work has been published by the NY Times, Reaktion Books, The Harvard Review, Yale University Press, and a dozen other literary magazines and book publishers. A former editor at The Iowa Review and former President of ALTA, Valentino served on the 2022 jury for the National Book Awards. He is the founder and publisher of Autumn Hill Books and blogs at russellv.com.
Originally published on Hopscotch Translation
Tuesday, September 19, 2023
hopscotchtranslation.com 