Shh…is silence really golden? A multivariate analysis of subjectivity in academic writings and translations

Abstract Although subjectivity has been extensively examined in original academic writings, it remains comparatively underexplored in academic translations. For many non-native English-speaking scholars, academic texts are often initially composed in their first language and subsequently translated into English, either manually or with the assistance of AI-based technologies. Addressing this underexamined area, this study offers a comparative analysis of subjectivity in original, human-translated and ChatGPT-translated academic texts, combining factor analysis and random forest modelling. The results identify six latent dimensions underpinning the construct of subjectivity: authorial assertion, dialogic engagement, modalised stance, evidentiality-related evaluation, hypothetical reasoning and indeterminacy. Importantly, the study provides empirical evidence of a general loss of subjectivity in academic translations, a tendency more pronounced in human-translated texts across several dimensions. The key difference between human translations and original academic writings lies in their orientation towards the source text or the authorial presence, while the primary divergence between ChatGPT translations and original texts is found in the degree of academic modesty. These findings suggest that students who use AI tools to translate their first-language writings into academic English should be encouraged to adopt an active authorial stance and to engage in critical evaluation of the AI-generated outputs.