In Mid-June 2025, the Chartered Institute of Linguists (CIOL) hosted a presentation by Ruth Ahmedzai Kemp entitled “AI and Literary Translation: Dos, Don’ts and What-Ifs”. It focused on the changing role of literary translators, advances in AI and how the industry has been affected by tools such as ChatGPT.
What is Literary Translation and What are Literary Translators
The start of Ruth’s presentation focused on the wide range in the field of literary translation, which in addition to novels includes poetry, children’s books, nonfiction, essays, graphic novels, journal articles and theatre (plays, musicals, etc). Literary translators work not only on published works but also on unpublished samples that are often used for scouting and pitching purposes.
One big difference from other types of translators is that literary translators often are copyright holders and are treated as co-authors under publishing laws and regulations. This gives them the right to be credited, control over edits and republications, royalties, and licencing and reuse rights as well. Because of these facts, it means that literary translations created by humans are protectable under the law. This then begs the question of the status of machine-generated or post-editing translations.
AI in Literary Translation
Ruth focused on two main concepts when talking about translations produced by machines: neural machine translation (NMT) and Large Language Models (LLMs). NMTs include translation services such as DeepL and Google Translation, which are trained on bilingual datasets for direct translations between languages. LLMs, which include ChatGPT and Google Gemini, are trained on vast and usually unlicenced monolingual bodies of texts (known as corpora) with the purpose of text generation and transformation.
LLMs in particular have raised a lot of red flags recently in the world of publishing due to the opacity of the process as well as possible copyright infringement, of which there have been many recent court cases. While LLMs are often more problematic due to these concerns, NMTs are also controversial in their own way, especially when used for copyrighted texts.
How Translators Use (and Don’t Use) AI
There are as many uses for AI as there are people using them. For translators, this can range from absolutely shunning the use of AI (often due to ethical or practical reasons) through stylistic consultations and terminology to post-editing of “zero drafts” that are produced first via NMT. Another use case is via collaboration of AI and monolingual or bilingual editors that review machine translation-produced drafts.
Ruth made it abundantly clear that caution needs to be used when using these kinds of translation processes for machine drafts that will eventually be published. This is due to the questions around ownership. In the case where text has not been completed edited by a human translator, who is the owner? Who takes accountability for this work?
Copyright, Contracts, and Post-Editing: What Counts as Translation?
A large part of Ruth’s presentation was focused on the legal and ethical concerns of Machine Translation Post-Editing (MTPE), which is where a machine (e.g. an LLM or NMT tool) creates the initial version and then a human edits that output.
One of the points she highlighted was a recent Danish legal position that came to the conclusion that MTPE is not considered a creative translation and therefore should not be granted copyright. While her own opinion was that if it is properly reviewed, meaning every word is revised and the translator takes responsibility, they should retain copyright, this highlighted the ongoing discussion about this legal issue.
Part of the issue can be resolved from the beginning within the structure of the contract. If a translator is considered “work for hire”, they do the translation and are paid for their work. As a result, the translator is not credited and receives no royalties. This is how most translators are employed outside of the literary industry. The alternative is a copyright-holding translator, who is entitled to royalites, receives credit and has control over reprints and reuse.
For as smooth a process as possible, this should be defined from the beginning. Ruth also recommended that translators should push for clauses that expressly prohibit AI use by publishers without the translators permission, disclose if the translator is using AI, and have a third party vet the contract. Examples of third parties that offer these kinds of services include the Society of Authors and the Author’s Guild.
Ethical, Legal and Environmental Concerns of AI Use
There are several concerns with AI use, many of which can cause issues for translators. For example most LLMs have been trained on copyrighted material without consent. As we have seen from the court cases, this leads to many questions about intellectual property. Furthermore using an LLM to translate copyrighted work without permission also usually breaches the original author’s rights as well.
Another legal concern that is not unique to literary translation is data retention. All information that is put into an LLM, whether during training or from a user, is usually kept and used for further training. The work of the translator, which may be proprietary for literary translations, will then be used by the AI for other potential translations, causing a (as per current laws) legally murky situation.
AI also has serious environmental concerns, the most pressing of which is the running and powering of data centres. These centres have massive carbon and water footprints, causing a serious impact on the environment at a time when humanity should be reducing the strain it is causing the planet.
Human-Only Superpowers
There are still many limitations to AI, LLMs and NMT. Only humans are really capable of understanding cultural nuances, idioms, irony and intertextual references. Only humans can truly collaborate with authors and editors. Only humans are concerned about musicality, rhythm and the flow of a text. Only humans make ethical choices as well as take and provide accountability. While AI certainly makes things sound “possible” or “plausible”, it certainly does not take those artistic components into consideration. AI can convert language but only humans can really interpret it.
Contract Clauses and Professional Practice
One way of solving some of these problems is through clauses in contracts and professional practices. For example Ruth shared examples of model contracts that restrict the use of AI systems to closed systems, meaning that the information input by the user is not allowed to be used for training and other purposes and therefore will not leave the confines of that project. Additional clauses should also limit the use of the translation for AI purposes, such as not allowing publishers to sublicence to these companies.
These clauses can also help protect the publishers as well, such as one that requires translators to review and approve every word when AI is used and even to clarify if AI-generated drafts are permissible.
Ruth recommended that as translators, you should be transparent with authors and editors if you use AI in your process. She also recommended that all translators should play around with AI but to make sure that they use out-of-copyright texts. Finally another recommendation was to reflect the use of AI within your pricing structure An example of this could manifest as per-hour pricing for post-editing tasks.
Advocacy and Collective Responsibility
The big take home message from this presentation was a call for advocacy. Ruth advocated for translators to follow some good practice guidelines. These includes sharing examples of good and back practices with trade bodies, pushing back against unethical or cheap AI use, helping to educate publishers and authors on the value of translations, and supporting ongoing initiatives such as the Manifesto Against Soulless Translation.
In particular translators should not assume whether AI use is permitted or banned but rather ask, declare if you use AI from the beginning, experiment with AI using non-sensitive and out-of-copyright or non-copyrighted texts, carefully vet contracts, and use AI as a tool rather than as a replacement.
My Thoughts
Literary translation is a very alien field to me as it operates very differently from the types of translations that I usually do. I had very little knowledge going into this presentation.
My initial predictions to the title of this talk were that basically AI was not going to be used in literary translation whatsoever and that the hour-long event was going to be focused on why that is. As you can imagine, I was pleasantly surprised to find out that it can be used as a tool and that it was not completely banned. I would have thought that in a more creative space that AI would have no place since that human component is critical, but this talk proved me wrong.
It also seemed to me that a lot of the problems that literary translators face with AI are very similar to those faced by the industry as a whole: unethical use, replacement of translators with “cheaper” AI options, and general lack of translators being valued by society.
While a lot of people (and here I am referring to people in general rather than authors, translators, project managers or others in the translation space) use AI without knowing or understanding the ethical dilemmas. I think due to the lack of education about AI, many people are ignorant of how LLMs/NMTs work and what kind of issues or concerns that there may be. On the other hand, there are some people that go in with full knowledge of those concerns and decide to use it regardless.
I think this talk highlights the need for education on the positives and negatives of AI, when to use it ethically and when it should be avoided for ethical, legal and environmental reasons. Only then can we safely integrate AI into facets of translation and other fields in an ethical and safe way.