This study aims to conduct an exhaustive comparison between the performance of human translators and artificial intelligence-powered machine translation systems, specifically examining the top three systems: Spider-AI, Metacate, and DeepL. A variety of texts from distinct categories were evaluated to gain a profound understanding of the qualitative differences, as well as the strengths and weaknesses, between human and machine translations. The results demonstrated that human translation significantly outperforms machine translation, with larger gaps in literary texts and texts characterized by high linguistic complexity. However, the performance of machine translation systems, particularly DeepL, has improved and in some contexts approached that of human performance. The distinct performance differences across various text categories suggest the potential for developing systems tailored to specific fields. These findings indicate that machine translation has the capacity to bridge the gap in translation productivity inefficiencies inherent in human translation, yet it still falls short of fully replicating human capabilities. In the future, a combination of human translation and machine translation systems is likely to be the most effective approach for leveraging the strengths of each and ensuring optimal performance. This study contributes empirical support and findings that can aid in the development and future research in the field of machine translation and translation studies. Despite some limitations associated with the corpus used and the systems analysed, where the focus was on English and texts within the field of machine translation, future studies could explore more extensive linguistic sampling and evaluation of human effort. The collaborative efforts of specialists in artificial intelligence, translation studies, linguistics, and related fields can help achieve a world where linguistic diversity no longer poses a barrier.
