AI as translator: Help or hindrance?

Artificial intelligence is rapidly entering different spheres of life, including book translation. This fact evokes mixed feelings within the professional community. Will AI become a useful tool? Or, on the contrary, pose a threat to the translation profession? Can AI adequately convey the style, rhythm, and cultural nuances of a literary text, and most importantly, is it capable of replacing a human?

To explore these questions, we spoke with translators about their attitudes toward AI, its potential benefits, possible threats to the profession, aspects of literary translation that remain beyond AI’s reach for now, and the need for ethical and legal regulation. Here are insights from translators, an editor, and a publisher on the future of book translation in the age of AI.

Bogdana Romantsova

Editor of publishing projects at Tempora publishing house, lecturer, and literary critic.

Photo credit: Nastia Telikova

In my opinion, AI can be helpful in translating nonfiction, at least to a certain extent, but more as an additional tool — like the software we use for checking spelling and punctuation. When it comes to fiction, style plays a much more significant role. The author’s perspective is the most interesting thing that the translator must convey when interpreting to another language. I really don’t believe AI, at least at this stage of development, is capable of fully handling that.

Editing is also significantly more creative and intricate than people usually realize. Most edits aren’t just basic corrections (a missing letter, for example) but rather subjective decisions made by the editor. I don’t think AI, for example, can detect a complex allusion. I’m not afraid of losing my job.

Sandra Ljubas

Ljubas holds a PhD in linguistics, works as a literary translator, and is a translation studies scholar. She has been active in the profession for eight years and has since translated over thirty literary works from Swedish, Norwegian, Danish and German into Croatian, most notably Astrid Lindgren, Solvej Balle, and Carsten Jensen. Her main research interests are machine translation and literary translation.

Photo credit: translator’s personal archive

Based on my almost 10 years of researching automated translation, I firmly believe that a more technocentric approach to translation – even literary translation – poses no threat to our profession. Translators have always been resilient; we adapt quickly, continuously learn, and refine our craft. That is the beauty of our profession, and new technology can serve to amplify our capabilities rather than replace them. In some cases, AI might assist by generating drafts for post-editing, but in literary translation its potential very often extends beyond that.

Generative AI can be a valuable tool – whether as a dictionary, a thesaurus, a source of inspiration, or even a simulated proofreader or colleague to discuss challenges with. And yes, AI is prone to hallucinations and should never be relied upon unquestioningly – that is precisely why the human remains the sculptor.

In general, the potential of AI is often overstated; it is a tool, nothing more, nothing less.

The decision to use it should remain in the hands of individual translators. Acknowledging AI’s role should not be taboo, nor should its use be imposed – especially given the ethical concerns surrounding AI, from environmental impact to legal issues related to training data. Publishers should also recognize this. While some, driven by profit, may be tempted to cut corners by lowering translation fees or tightening deadlines, it is crucial that they do not lose sight of the value of human expertise.

RELATED: Ukraine’s book industry debates the rise of AI in publishing, balancing ‘innovation’ with ethical concerns

Ela Varošanec

Translator from Spanish and English to Croatian. Member of the board of the Croatian Association of Literary Translators. Member of Conseil européen des associations de traducteurs littéraires (European Council of Literary Translators’ Associations or CEATL) working groups on copyright and artificial intelligence.

Photo credit: Edi Matić

CEATL’s stance is that there’s no place for AI in literary translation. Machines are not translators – they are “translatoids” (they only mimic translation). They don’t translate; they generate textual content. A machine doesn’t feel or empathize, which defies the core purpose and value of literature itself. Literature is a profoundly human creation, and AI’s inability to grasp emotional and cultural depth means it cannot fulfill the role of literary translator.

While AI might have some uses as a tool for literary translators, it should be used very cautiously and with the full awareness of its limits. In that sense, it’s interesting to note that in the CEATL’s 2023 AI survey, just about 10% of 1,500 respondents found AI translation tools useful or very useful.

AI models operate statistically, which is fundamentally opposed to the nature of language in literature – playing with words, subverting the norm, creating new images and feelings –  almost a new language for every book. AI-generated texts tend to simplify language and lack a true sense of literary flow.

AI models can sometimes identify the surface layer of meaning but fail to recognize cultural nuances and allusions as well as deeper emotional context – elements that are essential to literature.

We believe, and see it every day, that AI poses a significant threat to the profession in multiple ways, which can be grouped in three key areas: economic, professional, and creative.

Across Europe, translators are increasingly asked to post-edit AI-generated texts for severely reduced fees while being denied authorship and therefore royalties and professional recognition, whereas reworking a text translated by a machine is a creative full authorship process and should be fairly remunerated: the task is indeed more complex, time-consuming, and exhausting than standard translation work.

For all of the above mentioned reasons, lots of colleagues are deciding for a career change and abandoning translation all together. That is a terrible trend that will have long-lasting effects on European culture and society.

The big irony here is that AI models are trained, among other things, on translated works. This means that machines generate content based on the work of literary translators, and then that content is in direct competition with their work on the market.

We want to be very clear: literary translators are irreplaceable because literature is, at its core, a deeply human activity.

That, unfortunately, won’t stop various enterprises from trying to convince readers and the general public that the machines are “good enough.” But “good enough” should never be enough. Every book and every reader, and every literary translator, for that matter, deserves excellence – and machines simply can’t provide that. Also, the idea that some books are better suited than others to undergoing generative AI processing fosters a dangerous distinction between high literature and every other text, impeding societal development through reading.

We call for an ethical and legally effective approach: literary translation exists through ART*: Authorisation, Remuneration, Transparency. Any transfer of copyrighted material for commercial use, such as AI training, should always be negotiated by the author as an opt-in clause.

There should be clear labeling for books where AI was used in the translation process. Readers have a right to know when AI has been involved in creating a literary work.

At the same time, we strongly oppose any form of imposition of the AI tools on the literary translators and their creative process. Literary translation is an art, and forcing translators to work with AI-generated texts undermines the artistic integrity of their work.

Ivan Riabchyy

Translator, publisher, and public figure. Special consultant to UNESCO since 2024. Recipient of numerous awards. Co-founder and director of Pinzel Ukrainian publishing house. Executive Director of the Ukrainian Institute in France, 2023 to 2024. Author of over 100 book translations, including works by Michel Houellebecq, Patrick Modiano, Salvador Dalí, Éric-Emmanuel Schmitt, Florian Zeller, Henri Bauchau, and other francophone authors.

Photo credit: Oleksandr Zakletskyy

I’ll begin with an anecdote. Every time I come to France or any French-speaking country, I start writing something in French. It’s more of an exercise than actual creativity. However, this time it coincided with the beginning of a passionate relationship between me and ChatGPT. Back then, my job required me to familiarize myself with these technologies. I asked Lorelei — that’s what I called my AI-almost-friend — to analyze a text I had written and suggest how it might be developed. In a few minutes, I received a complete analysis, and I was surprised. It was the response I might have written myself. Lorelei gave a thorough and accurate critique, and her suggestions for how the story could progress was entirely reasonable. After that, I hesitated to share anything too literary with her. It felt like gazing into the abyss.

As I’ve already mentioned, any office position in France requires knowledge of AI technologies. These can include programs specifically tailored for translation (for instance, at UNESCO, DeepL has been widely used for years already), for organizing one’s life (calendars), project management, or team development (Asana), and so on. Naturally, when you’re preparing many corporate documents that won’t have your name on them, time, accuracy, and quality matter the most. AI can provide all of this: it saves time, processes massive amounts of information, and gives clear, well-done summaries. Of course, you still need to read everything carefully and make changes if needed.

AI can take all monotonous and uninteresting tasks, such as writing business emails, résumés, follow-ups, and so on. There are very convenient programs that join video meetings as a separate participant and instantly generate a transcript of the conversation, sending it to your email right after the meeting ends (like Otter) — and another program can immediately turn that transcript into a follow-up message. This helps you use less time and energy.

If we set aside purely literary translation, I personally find AI tools to be very useful and convenient. And yes, I regularly use Lorelei’s help in my work.

At the same time, I’m completely against using AI for translating books. For me, translating literature has always been something I enjoy — a creative process I share with the author — and only later did it become a job. But I’m lucky because I never have to translate anything that doesn’t mean something to me. When translation turns into a job and the translator takes as many assignments as possible to earn a living (which is an understandable motivation, of course), that’s when they are tempted to use AI. The text isn’t yours and you don’t really care about it — what matters is getting paid, meeting the deadline, and landing the next assignment. I call this pressured detachment. In such cases, translation is no longer a creative act — it’s just a job, nothing more.

The main threat posed by AI is that it is constantly improving. Constantly — meaning right now, as I’m writing these lines. I don’t think there’s anything that can stop it. So what should we do about this threat?

Since 2019, AI has been one of UNESCO’s main topics — especially the ethical issues related to it. During Italy’s presidency of the G7, UNESCO and the and the Organization for Economic Cooperation and Development (OECD) developed a toolkit for the use of AI in the public sector. UNESCO’s Global AI Ethics and Governance Observatory has issued specific recommendations. No other field needs a strong sense of ethics as much as AI. While it can bring advantages in many spheres, without ethical limits, it can repeat real-world biases and discrimination, make inequalities worse, and put basic human rights and freedoms at risk.

That’s why AI is among UNESCO’s global priorities and is central to the science-policy interface the organization is striving to shape and manage. Right now UNESCO is a global leader in the field of AI ethics, as well as in the ethics of neurotechnology and quantum science. If all goes according to plan, an AI Ethics Code will soon be developed and adopted by all UNESCO member states, and then implemented by AI-developing companies. This would significantly reduce the threat of job displacement caused by the introduction of AI.

Julia Didokha

Translator and program curator of the Ukrainian Literary and Translation Festival Translatorium. She has 12 years of translation experience and primarily works between English and Ukrainian. Didokha’s portfolio includes translations for Pascal Gielen and Danielle Krysa, as well as poetry by leading voices of the American Beat Generation, including Lawrence Ferlinghetti, Gary Snyder, and Jack Kerouac.

Photo credit: Daryna Deineko-Kazmiruk

AI is becoming a topic of discussion among literary translators. Some see it as a tool that saves time and helps in a big way, others as a soulless competitor, and some refuse to acknowledge its impact at all. One thing is clear: the future of book translation will not be possible without AI. In any case, we will all have to adapt to this new reality and learn to use it to our advantage.

When AI translation tools first appeared, I was quite skeptical about using them in my work. Just think of Google Translate: everyone had their opinions about this tool, and translators often joked about its “skills” and the “masterpieces” it produced. Because of that, my skepticism carried over to subsequent AI tools that emerged later — tools that, nevertheless, quickly convinced me that they are capable of doing their job very well. Now I use AI quite often in my work, but, as people often say, there are details that make a difference.

Usually, when we talk about book translation, we envision a creative process that requires not only language proficiency, but a deep sensitivity to the text, intuition, and emotional awareness. A translator must grasp the rhythm and style of the work, convey humor, allusions, and cultural nuances. It is precisely this emotional component that makes literary translation an art. Can AI grasp these subtleties? Not yet.

From my own experience, I can say that its algorithms handle syntax and vocabulary quite well, but definitely not subtext and metaphors. It doesn’t know how to read between the lines — and most likely won’t learn to do so for a long time.

At the same time, we shouldn’t underestimate the usefulness of AI as a tool. As I’ve said, I use it often now because it really helps me work faster: it quickly checks terminology, gives useful lists of synonyms, and offers good ways to translate common phrases — and there’s even more it can do. Of course, sometimes its suggestions sound awkward or unnatural — that happens too.

I do agree that for translators, AI is more like a helper than a competitor, and I believe the problem begins when we start relying on it too heavily. The more the market leans toward machine translation, the more we risk losing the uniqueness of literary translations. If we get used to working only with machine-generated drafts, the very principle of creative engagement with the text will gradually fade. And that’s where we’ll be facing an ethical issue in the field of translation.

Olha Liubarska

Translator with 15 years of experience, working in the English-Ukrainian language pair. Her translations include “The Bell Jar” by Sylvia Plath, “Mort” by Terry Pratchett, and “Cities for People” by Jan Gehl.

Photo credit: translator’s personal archive

You shouldn’t expect AI to do everything quickly and well while you sit and drink your coffee. AI-generated output always needs editing, sometimes a lot of it. AI can’t make translation decisions for a human translator, nor can it develop a strategy or approach to the linguistic nuances of characters, the author’s syntax, or idiolect. It’s unaware of cultural codes and the challenges of cross-cultural translation. Imagine a person translating something for the first time — someone who knows the language only with the help  of a dictionary — being handed a poetic work without knowing who the author is or having time to study the author’s style and do all the things professional translators routinely do. What kind of result would such an unprepared person produce? Unpredictable. That’s the kind of output you get from freely available AI-powered  translation tools. If you want to train AI capable of translating a specific author well, it will take a lot of time and financial resources — in other words, it won’t be cheap.

Personally, I don’t use AI for book translation. I mostly use it for more routine or formulaic texts — the kind I come across in daily life or when I need to save time and don’t want to spend 16 hours translating something that clearly only took 4 hours to write. A tool like DeepL gives me a basic version, and I then edit it to make it more readable. I usually use this method when translating from Ukrainian to English.

I’d also like to share my experience from another professional context. I work as a medical interpreter (English-Ukrainian) at a hospital in the Netherlands, where the demand for interpreting conversations between medical staff and patients is enormous, but the supply is very limited. As a result, there are many situations where communication happens via Google Translate. If you think about it, translating dialogue for film or comics is very similar to interpreting real-time conversations — except in live conversation, speakers don’t have the time to think and formulate like writers or screenwriters do. AI handles this kind of translation terribly, so I have good news for my colleagues: we’re not going to be replaced by machines any time soon.

To protect our profession from AI, we need to keep in mind that AI doesn’t learn on its own — it learns from texts created by people. That’s why whenever we create content that can be copyrighted, we should make it clear to publishers that we do not grant them the right to use the fruits of our labor for training machines.

We have every right to do so. If the goal of big businesses were to use AI to free up funds for social responsibility — say, for a universal basic income for people in creative professions — then we could at least begin to talk about some form of ethical use.

The threat to the profession of book translation is not AI itself, but its irresponsible use and the lack of understanding about translation quality. These are the same threats we faced when the Ukrainian market was flooded with unedited translations done by programs from Russian intermediaries. There are publishers who have no shame in releasing such work. There are readers who don’t mind reading it. The same will happen with AI translations — unfortunately, there is a buyer for everything. I comfort myself with the thought that we work for readers who understand and seek quality and who rejoice when they find it. Maybe out of a print run of two thousand, there will be ten such readers over five years, but that’s already something.

Translators In Action (TIA) is a group initiative that fosters development within the translation community and creates projects and tools to strengthen and increase the professional security of Ukrainian book translators.

The TIA aims to: keep translators informed on legal aspects of their work; help translators become more visible and secure concerning their legal status; inform, advise, organize or otherwise support translators’ initiatives; organize educational and cultural projects related to translation; represent the interests of the community; conduct research and surveys to identify specific issues within the community and explore ways to address them.

RELATED: A Ukrainian magazine released, created entirely with the help of AI

Translation: Iryna Saviuk

Copy editing: Sheri Liguori