From Choice to Probability: Domestication and Default Literalism in Machine-Translated Uzbek Proverbs and Idioms

Machine translation (MT) systems and chatbots are now widely used to translate culturally loaded language, including proverbs and idioms. Lawrence Venuti's distinction between domesticating and foreignizing translation is the standard framework for describing how a translation treats the "foreign" [7]. However, this framework was developed for human translators who make deliberate choices, not for MT systems. The question is especially relevant for Uzbek, which is classified as a low-resource language in natural language processing, so MT systems tend to behave unevenly and unpredictably with it [6, 4]. In addition, idioms are already a known weak point for neural MT, since such systems often translate them too literally [2]. Venuti's concepts assume a translator who takes a deliberate, ethical stance toward the foreign text. An MT system, however, takes no stance: it simply outputs whatever its training data makes most probable. Therefore, it remains unclear whether "domestication" and "foreignization" can meaningfully describe MT output at all, particularly for a low-resource Turkic language such as Uzbek. The aim of this study is to determine whether Venuti's domestication–foreignization distinction can describe how Google Translate and ChatGPT render Uzbek proverbs and idioms into English, and to propose adjusted concepts where the original distinction does not apply.