Translation strategies in literary collocation translation: A comparative analysis of human and machine-assisted translation
TL;DR
It is demonstrated that collocational competence remains a key factor distinguishing novice human translators from large language models in human–AI translation contexts and demonstrating that collocational competence remains a key factor distinguishing novice human translators from large language models.
Abstract
Literary collocations present persistent challenges in translation because their meanings are often culturally embedded and cannot always be inferred from individual lexical items. Limited studies explain how human translators and AI systems differ in their strategic handling of literary collocations. Addressing this gap, this study compares the translation strategies employed by students, ChatGPT, and DeepSeek. Using a qualitative comparative design, the study analyzed 35 collocations extracted from the novel Gadis Kretek, translated by 21 English language education students and two AI systems, ChatGPT and DeepSeek. A framework of translation strategies was used to classify and compare the translation outputs. The findings reveal that all three translator groups predominantly relied on paraphrasing with related words; however, significant differences emerged in their strategic behaviour. Students frequently employed omission and occasionally produced semantically unrelated paraphrases, whereas ChatGPT and DeepSeek consistently preserved collocational meaning through semantically related reformulations. DeepSeek further demonstrated greater lexical creativity through selective borrowing and cultural adaptation. The study contributes to Translation Studies by extending Baker’s framework to human–AI translation contexts and demonstrating that collocational competence remains a key factor distinguishing novice human translators from large language models. AI helps, but human judgment remains essential.