GPT and Translation: A Systematic Review
Venus Chan, William Ko-Wai Tang
Abstract
Despite the growing use of GPT for translation, little research has analyzed studies on GPT and translation. This paper aims to provide a systematic literature review to explore the overall trend and summarize the major benefits and limitations of using GPT for translation based on 16 reviewed papers. The results demonstrate that there has been a significant increase in the number of publications, particularly in the field of translation quality of the European languages. Most studies used a quantitative method, and the findings mostly suggest that GPT-generated translations are on par with human translations and perform better than neutral machine translation outputs. GPT can accurately translate cultural texts, complex structures, and advanced linguistic features, such as poetry, humor, and pun. It is also found that GPT can be effectively used not only for translation but also for post-editing and translation evaluation, leading to new challenges and ethical concerns. This research highlights the changing roles of the key stakeholders and concludes with some direction for future studies.