Dancing with the Devil? On the use of AI in literary translation and beyond

On entering the Aula Magna for the keynote speech by Dr Dorothy Kenny, I was struck by the contrast between the futuristic theme of AI in literary translation and the Modernista (aka Catalan Art Nouveau) architecture. The exquisite columns, tiles, bricks and dome were skilfully conceived and crafted by humans – but will our future be designed and built by robots? Must we accept AI’s dominance and a potential future in which translators are relegated to post-editing, just to line the pockets of a few billionaires?

I came to the METM conference with innate scepticism about AI and the suspicion that I would be told that it is already “king” and translators are obsolete. Since starting to work as a freelance translator in the 1990s, I have seen many technological changes. Some, such as Windows 95, Word, emails, the Internet, teleconferencing and mobiles, liberated us from the office and enabled us to expand internationally. Others are more suspect, such as requests to input our hard-gained glossaries and knowledge into remote programmes or training AI. What if, once we surrender our treasures, we are no longer needed?

Dr Kenny began by harking back to her previous speech at METM in 2011, fourteen years ago, when the subject under scrutiny was Statistical Machine Translation (SMT). She explained that, since then, machine translation has developed significantly, and the paradigm changed with the introduction of Neural Machine Translation (NMT) in about 2016. (Some translators embraced machine translation and others shunned it, and in both cases, some continued to work while others did not.) Now, the debate is about Large Language Models (LLMs) and whether they are replacing translators and interpreters, while literary translation is thought to be the last bastion of human translation.

First, Dr Kenny clarified the point that many people confuse AI with generative AI and LLMs, and that AI has in fact been around a long time and encompasses Machine Learning, Deep Learning and Deep Learning using Transformer Architectures (NMT and LLMs). The difference is that people are more aware of GenAI and more resistant to it because it has been hyped by corporations, has displayed inherent errors and hallucinations, and poses social and environmental risks. Protest has emerged, with many professions proclaiming “not us”. Although literary organisations are vocally opposed, AI has in fact already been used in literary production, such as Nordic novels.

Researchers contemplated three initial motivations for using AI in literary translation: the inherent challenges of the task (not just to communicate meaning, but style and textual effects); opportunism (with the emergence of the e-book and availability of legacy data) and altruism (to increase the range of foreign language books). Yet, in a summary of recent research (Kolb 2023; Kenny 2025), MT is perceived as offering lower quality, accuracy and creativity; MT + post-editing does not improve productivity; no uptick has been seen for translation into minoritised languages, and MT is associated with worsening economic conditions for translators.

In 2025, the future is uncertain. Media reports claim that generative AI pilots at companies are failing, while the stock markets suspect overvaluation of AI technology. AI pioneers have issued warnings and Byung-Chul Han, winner of the 2025 Princess of Asturias Award for Communication and Humanities, signals the risk “that human beings may end up becoming slaves to their own creation”. Dr Kenny’s research was enlightening and offered us some hope that humans cannot be completely supplanted by machines.

This METM25 keynote was chronicled by Sarah Downham.

Featured photo by METM25 photographer Julian Mayers.