What are the differences? A comparative study of generative artificial intelligence translation and human translation of scientific texts

Generative artificial intelligence translation (GenAIT) brings convenience yet also imposes severe challenges on the dissemination of knowledge. The respective (dis)advantages of GenAIT and human translation (HT), and the ways to promote their effective interaction have not been sufficiently explored yet. This study investigates the linguistic features of GenAIT and HT of scientific texts rendered from English to Chinese from lexical and syntactic levels. The GenAIT is generated by ChatGPT 3.5, a representative GenAI platform, while HTs are done by 19 Master-of-Translation-and-Interpreting students in China. Data shows that GenAIT and HTs present distinguished linguistic features in both levels. At the lexical level, HT exhibits lengthier texts with a lower average word diversity; GenAIT presents higher accuracy in translating terminology. At the syntactic level, the average sentence count in HT is greater, whereas its average sentence length measured in tokens is shorter. Moreover, human translators tend to transform sentences from passive voice into active voice more frequently than ChatGPT 3.5 does. Furthermore, human translators exhibit superior skills in deconstructing lengthy and complex sentences into shorter, more comprehensible clauses. This study reveals that ChatGPT 3.5 and human translators exhibit complementary capabilities in the translation of scientific texts, thereby suggesting an optimisation of the strengths inherent to both parties. Implications are provided for future translator training, language service providers and the ongoing development of GenAIT and HT.

Linling Fu, Lei Liu