AI and literary translation: how to preserve translation quality with human postediting

As early as 2022, the BookTok phenomenon had already generated considerable views: nearly 1.6 billion views across around 376,000 videos using the hashtag #BookTok, according to Actualitté, a specialist trade publication in the publishing industry. This reader-driven phenomenon, which emerged on TikTok, now has a major influence on the book industry. When a book goes viral, readers quickly demand its translation into multiple languages, thereby pushing publishers to speed up their release schedules.

A book’s success on BookTok leads to rapid growth in sales, tighter editorial deadlines and the ability to respond quickly to the expectations of an international audience.

Faced with this pressure, AI-based translation may seem like a quick solution. However, literary translation is not about replacing one word with another. It requires reading, writing, coherence and nuance. It is precisely when deadlines become tighter that text quality can be weakened.

Rather than setting AI and humans against each other, it is more relevant to focus on the conditions: what deadlines, what level of proofreading and who is responsible for the final version?

Tight deadlines: how translation quality can deteriorate

When the launch window narrows, literary translation comes under significant pressure. In this context, the limitations of AI-generated translations quickly become apparent: variations in register, inconsistencies and wording that is too closely tied to the source language.

Sometimes, small details are enough to reveal an inaccurate translation: a character who suddenly changes register, an expression that feels unnatural or an element of the narrative phrased differently from one chapter to the next. Conversely, when a translation is successful, the reader no longer thinks about the original text: they read naturally.

Why literary translation is a profession centred around writing and research

Translating a literary text is not just about conveying meaning. It involves producing a target text that remains consistent throughout: dialogue, rhythm, tone, humour, subtext and overall coherence. In other words, translation is also an act of writing. In an interview published by the British Council, the literary translator Daniel Hahn describes translation as an interpretative and creative act, in which every translation differs from the original and involves constant decision-making about how to render meaning in another language.

In practical terms, high-quality literary translation often hinges on four elements:

• Narrative style
• Rhythm
• Implicit meanings (subtext)
• Consistency throughout the work

A translation may be accurate in terms of meaning and yet still be inadequate if it fails to convey the energy of the text: irony, emotion, tension or deliberate ambiguities.

Research is often invisible, yet decisive. Checking a cultural reference, understanding dialect expressions, identifying the source of a quotation or standardising recurring terminology helps to prevent inconsistencies across chapters.

Machine translation and postediting: what impact do they have on quality?

Before making a judgement, it is worth clarifying the terms. Some of the misunderstandings stem from the fact that people confuse “machine translation” and “postediting”, even though their objectives and outcomes are not the same.

Machine translation produces a text generated by a computer system. It can help to understand the general meaning or generate a first draft, but it does not guarantee the style, coherence or literary intent of the text.

As stated in ISO 17100, translation quality is based on a structured process including translation, revision by a second professional and a final check, in order to ensure the consistency and reliability of the text.

Postediting consists of correcting and improving a machine-translated text. Depending on the objective, corrections may be minor or more substantial, sometimes going as far as completely rewriting the text. At this stage, the focus is on adapting specialised content while ensuring terminological accuracy and consistency of meaning, a requirement that also applies to technical translation. For further insight into the methodology and levels of revision, the article “What is postediting?” describes this process in a professional context.

In literature, postediting is not limited to simple correction: it involves genuine editorial work, including style harmonisation and careful revision.

Publishing and artificial intelligence: why the debate is becoming increasingly tangible

The recent decision by Dutch publisher VBK to trial AI-generated English translations for a limited number of books illustrates the increasingly concrete choices facing publishers: how to balance speed, cost, and responsibility.

This trend is also affecting other cultural industries. In manga localisation, the race to achieve simultaneous releases is also becoming a key issue. Slator’s report on GlobalComix shows how some platforms are trying to accelerate publication, particularly in response to the circulation of unauthorised content.

In digital publishing, website translation also plays a key role in ensuring the international visibility of works and their availability across different languages.

This acceleration, however, comes at a human cost. A recent report on the video game industry highlights how the pursuit of productivity can push expertise into the background, with potential consequences for the quality and nuance of the final text.

What AI can actually contribute

When used as a support tool, AI can be useful for specific tasks:

• Documentation
• Spotting repetitions
• Helping to maintain terminological consistency
• Initial wording suggestions when a sentence is difficult to translate

By contrast, it does not guarantee what gives a literary translation its value: textual coherence, irony, humour, subtext or subtle cultural adaptation. The real question is not whether AI is good or bad, but the framework in which it is used.

A realistic compromise is to reserve AI for preparatory stages: identification of terms, glossary building and initial phrasing suggestions. Human proofreading should then be focused on style: accuracy of imagery and consistent register throughout. This process helps prevent overly standardised translations while still meeting tight deadlines.

A practical method for preserving quality under tight deadlines

When time is limited, the solution is not to accelerate the process at any cost, but to safeguard the coherence of the text. A simple method can reduce errors.

1) Define the scope before translating

Clear upfront scoping helps avoid a range of errors: choice of register, whether to use informal or formal address, lists of proper names, a short glossary of recurring elements and narrative voice (narrator, characters). Even a brief scoping phase can avoid many unnecessary variations.

2) Treat postediting as rewriting

If postediting is used, the objective must be clear: to produce a final text that is readable and natural, without a machine-generated feel. This sometimes involves rewriting entire paragraphs.

A good habit to adopt is to reread key scenes as a reader: the opening, narrative turning points and the ending.

3) Implement targeted quality control

Targeted quality control can help prevent major inconsistencies:

• Checking recurring elements
• Checking style consistency across distant passages
• Careful proofreading of the opening, a key moment and the ending

The idea is not to review everything at the same level, but to focus effort on the passages that most affect the perceived quality.

4) Do not neglect the final editorial stage

When the translation is coherent, the format must follow: punctuation, typography, consistency of choices and layout. A layout or desktop publishing (DTP) stage helps ensure perceived quality and reading comfort, especially for dense texts.

How to improve speed without sacrificing style

To balance quality demands and deadlines, the most reliable approach remains a process in which a human is responsible for the final text. That is what a structured translation service must ensure: scoping, proofreading, editorial consistency and quality control.

When the project involves multiple translators, a publication series or content released across different platforms, organisation becomes even more important. It facilitates feedback, harmonises choices and preserves overall consistency.

Conclusion

AI can become a useful support, but only if what gives a literary text its value is protected: voice, nuance, rhythm and editorial responsibility. Without a clear framework, postediting may save time on the surface, but result in a loss of quality. With a rigorous methodology, however, it can help meet deadlines while respecting the writing.

Ultimately, the essential question remains: what level of quality do you want to offer the reader? In literary translation, AI can serve as a support, but only rigorous human postediting can preserve the quality of the text in the long term.