A tale of two ‘contexts’: ideological differences in the translations of UN political speeches by human interpreters and by ChatGPT4o

This study interrogates the ideological disparities between ChatGPT4o and human interpreters in rendering Chinese political speeches delivered at the United Nations General Assembly (UNGA) from 2008 to 2023. Drawing on van Dijk’s socio-cognitive Context Model, it examines how ‘context’ is operationalised regarding (1) role-relation engagement through deictic pronouns, (2) the modulation of agency via modal verbs, and (3) the treatment of culturally embedded metaphors, which carry significant ideological implications. With corpus-assisted critical discourse analysis and qualitative interpretation, the study illuminates that human interpreters, informed by sociopolitical cognisance and cognitive adaptability, actively mediate discourse to align with the speakers’ ideological positioning through discursive amplification. Conversely, ChatGPT4o, limited by its token-based algorithmic architecture, tends toward literal translations in parallel with its diluting key ideological markers. These divergences underscore the epistemologically grounded differences between human cognition and algorithmic processing, pointing to the sensitivity of human interpreters in ideologically laden contexts alongside the opacity of large language models.