GPT for Translation: A Systematic Literature Review

In the face of the increasing prevalence of GPT in translation, a significant research gap exists in examining the interplay between GPT and translation. This article aims to fill this gap by offering a systematic review of the literature, discerning the overarching trends and the primary focuses of pertinent papers, and encapsulating the key advantages and drawbacks of employing GPT for translation. This systematic review examines 26 papers related to GPT and translation published in five major databases. The findings reveal a substantial growth in the volume of related publications, especially in the realm of translation accuracy and quality for high-resource and European languages. Most of these studies utilized quantitative methods, with the results consistently indicating that translations produced by GPT are equivalent to those translated by humans and surpass the quality of neural machine translation outputs. GPT has demonstrated its proficiency in accurately translating cultural texts, complex structures, and sophisticated linguistic elements, such as poetry, humor, and puns. Moreover, the findings also suggest that GPT can be effectively leveraged not only for translation but also for post-editing and translation evaluation, thereby creating new possibilities. It is also found that the quality of GPT-generated translations largely depends on the design of prompts, in which adding specific details, such as purpose, target audience, task information, pre-edit scheme, and examples, would enhance the translation accuracy. This study underscores the evolving roles of the main stakeholders and concludes by suggesting potential areas for future research. It also discusses the potential impact on the translation education, illuminating how teaching and learning processes in the field of translation could be influenced by the application of GPT.