A Comparison of Human and ChatGPT-Generated Translations of Chinese Architecture-Related Culture-Specific Items

Abstract

This study investigates the English translation of Chinese architecture-related culture-specific items (CSIs), particularly those related to vernacular dwellings, through a comparison between human and ChatGPT-generated translations. It examines human translations in terms of semantic accuracy, cultural retention, and reader comprehension, drawing on the theory of functional equivalence, with ChatGPT-generated versions used for contrast. The selected CSIs are treated as independent units and analyzed in isolation rather than within their original textual contexts, allowing culturally loaded meaning to be observed more directly in translation. The findings indicate that human translations are generally more adaptable, showing greater flexibility in balancing meaning transfer, cultural specificity, and reader accessibility. By contrast, ChatGPT-generated versions are less consistent. Some are concise and effective, whereas others follow the original too closely, with some loss of clarity and precision. The study therefore suggests that AI-assisted translation is useful for drafting and offering possible alternatives, but it functions more as a support tool than as a substitute for human judgment. The conclusion does not extend to broader discourse contexts or to other types of CSIs.