Iconicity, Humour, and Pragmatic Loss in AI Translation: A Russian-Chinese Case Study

This paper investigates the challenges of translating linguistic creativity and pragmatic nuance, with particular attention to how humor and iconicity are lost in both human and artificial intelligence (AI) translation. Translation is not merely the transfer of lexical meaning but also the negotiation of pragmatic force, cultural resonance, and semiotic play. When these dimensions are overlooked, the communicative and emotional impact of the source text is diminished. Drawing on theories of diagrammatic iconicity (Fischer 1997), which describe how meaning is conveyed through the arrangement and relational structure of signs, and adaptive transcoding (Komissarov 1990), a pragmatic approach to translation that emphasizes functional equivalence, we analyze the communicative deficiencies arising in the translation of the Vladivostok café name Еда и точка. Азия (“Food and that’s it. Asia”) from Russian into Chinese. This café name skillfully employs diagrammatic iconicity and a calembour embedded in a phraseological collocation. The conjunction и (‘and’) is not simply a relator but part of the idiomatic unit и точка (‘and that’s it’), which carries a pragmatic force of finality and humor. The initial Chinese rendering, 食物和点 (“Food and point”), fails to convey this humor, dual meaning, and iconic effect, resulting in significant pragmatic loss. Such deficiencies are not confined to human translators: current AI translation tools also struggle acutely with these linguistic complexities. To explore this, we evaluated the Russian phrase Еда и точка across leading AI platforms. Baidu and ChatGPT produced largely literal translations (食物和点, 食物和句号), missing the calembour and pragmatic force. Google Translate offered a more adaptive but still imperfect version (食物，仅此而已 “food, that’s all”), which captures some pragmatic meaning but loses iconicity and playfulness. Copilot generated multiple variants, including 吃的，就这样 (“food, that’s it”), highlighting both potential and the need for human discernment even within AI-assisted translation. These results underscore that contemporary AI translation systems, despite their advancements, consistently fall short in handling nuanced linguistic phenomena such as humor and phraseological collocations. Their literal or partially adaptive renditions lead to unavoidable pragmatic loss, undermining the communicative and advertising effectiveness of the source text. The paper concludes by proposing several adaptive transcoding solutions grounded in human-centered descriptive translation approaches. These strategies preserve humor, pragmatic force, and iconicity, offering benchmarks for evaluating AI translation outputs. The study illustrates how insights from Chinese linguistics and translation studies can inform the enhancement of AI translation models, ensuring that systems more effectively capture linguistic creativity and cultural specificity. Such refinement represents an essential advancement in the age of AI.