Tatyana G. Polyakova

The present study aims to compare approaches to translating puns in the TV show Young Sheldon when translation is performed by: 1) a professional translator; 2) students of the Translation in Professional Communication programme; 3) artificial intelligence systems. Since sitcom humour relies on cognitive ambiguity, character-specific speech patterns, and culturally embedded references, the accurate translation of puns becomes a pertinent test for assessing the potential of current AI systems. The main methods used in the research include contextual, translational, linguostylistic, and pragmatic analysis. 

The results show that all groups of translators employ a similar set of techniques: pun retention, pun omission, and humour recreation. This research provides evidence that ChatGPT and DeepSeek are more efficient at conveying the comic effect when translating puns than the voice assistant Alisa. In addition, each neural network exhibits distinctive features of an "individual translation style." Artificial intelligence systems such as ChatGPT and DeepSeek demonstrate a certain capacity to identify and translate puns based on polysemy, generally choosing techniques similar to those of professional translators. DeepSeek more frequently relies on a broad context and explains its translation choices, whereas ChatGPT and Alisa do not take the pun's environment into account and, consequently, are unable to preserve a consistently constructed character image with specific communicative traits during translation; Alisa refuses to translate puns on religious topics and tends toward a literal rendering of meaning. 

As for student translations, they tend to recreate puns, and the statistical pattern of translation techniques they use is different. Moreover, in the proposed variants, there is a shift in the pun's theme, which can negatively affect the text's integrity but at the same time indicates the linguistic creativity of the learners. 

The problem of asymmetry in the presuppositional knowledge of the recipients of the original and translated media texts remains unresolved in the translations offered by artificial intelligence, while anthropogenic translation aims to compensate for it. 

The findings further reveal that, despite significant progress, AI systems still fail to reproduce the pragmatic depth and cultural presuppositions inherent in polysemy-based puns. The findings highlight the need for further development of humour-sensitive algorithms and point to the pedagogical value of AI translations in translator training.