The State of AI in Literary Translation

The State of AI in Literary Translation

So-called translation fails can get plenty of attention on social media worldwide — particularly when they involve big-name buyers and/or shocking mistranslations in marketing materials. Interest in literary translation, on the other hand, tends to stay more niche.

But a recent article published by The Atlantic brought some mainstream attention to the subject, describing literary translation as “the last frontier” of machine translation (MT). 

In 2024, with the ever-improving quality of MT (for certain language pairs, at least) complemented by the advent of AI, some literary translators have reported a shift in the work their publishing house clients offer.

As French translator Thomas Crespellière related on LinkedIn, some of the clients are publishers that contract with different, larger publishing companies to provide the translation.

Immediately Suspicious

“What made me immediately suspicious was the fact that they wanted me to POST-EDIT the book they would previously translate with machine translation,” Crespellière wrote, explaining that the client offered between EUR 0.03-0.04 per word for this work.

“With this type of publisher, there’s little chance of being credited as an author, and even less of receiving the royalties corresponding to the book’s sales,” he added. “And it’s a safe bet that they will sell the MTPE as if it were a 100% human translation.”

If literature currently remains an outlier amid widespread acceptance and adoption of MT, it begs the question: Who is willing to give this content the “MT treatment” — and acknowledge that MT is part of the process? 

While Crespellière did not name names, The Atlantic piece cited several companies already using AI to translate literature, including audiobook specialist Storytel and European online bookseller Bol.

Robert Casten Carlberg, the CEO and co-founder of Nuanxed, wrote that he was “proud to share our point of view” in the Atlantic article: “To successfully take on this field with all of its challenges and complexity, we believe in a holistic approach where our close collaboration with linguists to “really nail human-machine translations” is key.”

Mincor, a Mexico-based subcontractor for larger language services providers (LSPs), offers a combination of MT-only, MT-and-human, and human-only book translation plans, according to its rates page. The full breadth of combinations, and use of MT, however, are not mentioned elsewhere on the website.

Taking a different approach, Tencent Holdings-backed China Literature credited AI translation with boosting sales by 40% in 2022, as the South China Morning Post reported in December 2023. 

China Literature’s fiction platform apparently “plans to use AI to translate its works into more languages, including English, Spanish, Bahasa Indonesia, German, French, Japanese, and Portuguese,” though reviews of the translation quality are mixed.

MT Quality for Literature: A Cliffhanger

Amazon’s Kindle Direct Publishing (KDP) acknowledges that some authors self-publishing their works through the platform might be tempted to use AI for translation to reach a wider audience, but recommends against it. 

“A poor translation, such as a machine translation, will damage both your reputation as an author and the reputation of your book,” KDP’s Resources page states. “Technical content is much easier to translate than literary content, such as a novel.”

According to KDP, authors who choose to forge ahead with AI-generated translations are required to disclose the content as such, “even if [the author] applied substantial edits afterwards.” This is in contradistinction to the use of AI-based tools to edit, refine, error-check, or otherwise improve that content, which would be considered “AI-assisted.” 

Will academia and translators eventually come around to new, AI-centered workflows for literary translation? Institut für Translationswissenschaft UIBK in Austria just held a conference about AI and creativity translation, which included a session on AI in the translation of children’s literature. 

By contrast, En chair et en os (“Flesh and Blood”), a movement of writers and translators opposed to AI literary translation, has already gained the support of a significant number of signatories of their “manifesto”.

One translator promoting the manifesto noted that “my last 2 literary translation contracts included clauses forbidding AI (not that I would…).” For now, it seems, she and those clients are on the same page.