Distinguishing ChatGPT‐Generated Translation From Neural Machine Translation and Human Translation: A Linguistic and Stylistic Approach

ABSTRACT The widespread adoption of large language models (LLMs) for translation tasks necessitates a deeper understanding of the stylistic characteristics of their generated translations, an area that remains largely underexplored. To address this gap, this study examines whether a distinct LLM‐translationese emerges in diplomatic translation and identifies the linguistic features that differentiate LLM‐based translation from neural machine translation (NMT) and human translation (HT). Using a purpose‐built corpus of Chinese‐English translation of Spokesperson's Remarks from ChatGPT, Google Translate, and institutional human translators, we employed both confirmatory (machine learning classification) and exploratory (multidimensional analysis, distance computation) methods on 121 linguistic features. Classification results confirmed the stylistic distinguishability of ChatGPT‐generated translation from the other two variants. Multidimensional analysis revealed that ChatGPT's output exhibits greater interactivity, informality, and simplicity compared to NMT and HT, while also being more attitudinal and evaluative than HT. Furthermore, distance computation established that ChatGPT's translation aligns more closely with NMT, whereas HT remains more distinct. Collectively, these findings provide critical insights into the unique stylistic profiles of LLM‐based translation, offering an empirical basis for practitioners to leverage this technology effectively in diplomatic settings and beyond.