Is human translation more conservative than machine translation?

Abstract The present study investigates whether conservatism exists in human- and machine-translated texts from Chinese into English, and whether this tendency is consistently observable across different registers and multiple lexico-grammatical features by applying profile-based correspondence analysis and mixed-effects logistic regression modelling. The results reveal that human translation is characterised by a higher level of conservatism than both machine translation and original writing, irrespective of registers and lexico-grammatical features. In contrast, machine translation tends to be more conservative compared to non-translations only in journalistic and fictional texts, and the degree of conservatism varies across machine translation platforms. These findings suggest that human translators are more risk avoidant than original writers are, providing strong support for the risk aversion hypothesis. Moreover, the lack of understanding of translation norms or standards in machine translation, as well as the linguistic distinctions from human translation, implies the immense potential of future human-machine collaborative translation models.