A Stylistic Comparative Study for Evaluating Neural Machine Translation Technologies

Neural network-based online machine translation technologies have brought great convenience to foreign language teaching and research. However, performance evaluation of these translation technology models from the perspective of stylistic comparisons are currently scarce. In this study, Wenxin 4.0, DeepSeek-V3, and ChatGPT are used to translate a selection from an ancient Chinese story, and a stylistic comparison is conducted among the three versions of translation. By analyzing the transitivity processes of the original text, along with the three translated versions, it is found that the translations processed by the three technological models are closely similar to the original text in terms of the number and type distribution of transitivity processes. Nevertheless, there were certain differences in the use of mental and behavioral processes compared to the original text. Comparing the semantic configuration, as well as the stylistic feature distributions of the translated versions and the original text, is conducive to examining the differences in existing online translation technology models, both at home and abroad, when handling document-level translation units. This provides a basis for performance evaluation to support technological updates.