Human vs AI in the Translation of Humour: A Case Study of Roald Dahl’s The Twits in English, Russian, Italian and Dutch

Description

The paper examines whether AI translation can preserve humour at the lexical and tone level in children’s literature, comparing its output to that of human translators.
Translating humorous elements is a complex task that requires an understanding of children’s developmental stages, as well as cultural intuition and creative flexibility (Vandaele, 2002; Tisgam, 2009). Despite ongoing conversations about AI likely replacing human translators (Tomlinson et al., 2025), Mirowski et al. (2024) found that AI systems struggled to translate humour due to their reliance on training data and architectural constraints. It is relevant to re-evaluate these conclusions in 2026, given the improvements in AI systems.
This study employs close reading to compare three sentences from the chapter “Mr Twit” in Roald Dahl’s “The Twits” (1980) with translations submitted to ChatGPT and Claude Sonnet from English into Russian, Italian and Dutch. The outputs are analysed alongside human-translated versions. The study focuses on the translation of the main characters’ surnames, bodily grotesque imagery, exaggerated language, and humorous tone.
The findings show that while both AI models reproduce the semantic content, they fail to maintain the intended humorous effect. In Russian, for example, ChatGPT tended to overexplain the grotesque, whereas Claude defaulted to adult-register vocabulary incompatible with Dahl’s child-oriented tone. On the other hand, the human translator employed creative strategies, most notably constructing a new morphological wordplay system around the name “Svin" (pig), which replicated the intended humorous effect for Russian-speaking children. The results currently confirm that human translation remains necessary in children's literature.

References
Dahl, R. (1980). The Twits. Knopf. https://primarysite-prod-sorted.s3.amazonaws.com/white-rock/UploadedDocument/febe287074f74a6db03a08bc021e9847/the-twits-by-roald-dahl.pdf
Dahl, R. (2015). Svintusy [The Twits] (T. Karpova, Trans.). Rosman-Press. https://litres.com/book/roald-dal/svintusy-50216699/
Mirowski, P., Love, J., Mathewson, K., & Mohamed, S. (2024). A robot walks into a bar: Can language models serve as creativity support tools for comedy? An evaluation of LLMs' humour alignment with comedians. In Proceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency (pp. 1622-1636). Association for Computing Machinery. https://doi.org/10.1145/3630106.3658993
Tisgam, K. H. (2009). Translating cultural humour: Theory and practice. Wasit Journal for Humanities, 5(9), 79-121.
Tomlinson, K., Jaffe, S., Wang, W., Counts, S., & Suri, S. (2025). Working with AI: Measuring the applicability of generative AI to occupations (Microsoft Research Technical Report). https://doi.org/10.48550/arXiv.2507.07935
Vandaele, J. (2002). (Re-)constructing humor: Meanings and means. The Translator, 8(2), 149-172.

Period	29 May 2026
Event title	35th Anéla-/VIOT Juniorendag
Event type	Conference
Conference number	35
Location	Amsterdam, NetherlandsShow on map
Degree of Recognition	National