Evaluating Large Language Models and Neural Machine Translation Systems in Translating Liaozhaizhiyi: A Cross-Cultural Literary Study

Abstract

Neural machine translation has demonstrated strong performance in high-resource languages and commercial translation contexts. However, its effectiveness in translating classical Chinese literature remains insufficiently examined. This study conducts a comparative evaluation of three translation systems—ChatGPT, Google Translate, and Youdao Translate—using selected texts from Liaozhaizhiyi as the research corpus. The analysis focuses on four dimensions: semantic alignment, measured by BLEU scores; translation fluency; stylistic fidelity; and the detectability of machine-generated translation patterns. The results indicate that ChatGPT achieves superior performance in stylistic fidelity, particularly in preserving poetic tone, as well as in semantic alignment when translating idiomatic expressions and culturally embedded references. In addition, translations produced by ChatGPT exhibit fewer mechanical artefacts commonly associated with neural machine translation outputs. Further experiments demonstrate that structured prompt engineering strategies contribute to improved literary naturalness and greater cultural coherence in the translated texts. These findings suggest that large language models offer notable advantages in the translation of classical literary works and provide empirical insights into the role of artificial intelligence in facilitating cross-cultural interpretation and the international transmission of Chinese literary traditions.

Ähnliche Arbeiten