Research on Quality Assessment of English Translation of Chinese Classics from the Perspective of Large Language Models—A Case Study of The Library of Chinese Classics
School of Foreign Languages, Xinjiang University, Urumqi, China
Keywords:
Large Language Models; English Translation of Chinese Classics; Text Type Theory; Translation Quality Assessment; The Library of Chinese Classics
Abstract:
This study takes nine works from The Library of Chinese Classics as its corpus, categorizes them into informative, expressive, and operative types based on Reiss’ Text Type Theory, with a total corpus of around 1,800 Chinese characters. With ChatGPT-5.4 and DeepSeek-V3.2 as test models, translation quality is assessed across accuracy, literary quality, and persuasiveness by employing the MQM error hierarchy scale, BLEU score, and BERT-based sentiment similarity analysis. The results show that for informative texts, DeepSeek-V3.2 registers a lower total error count (38.33) than ChatGPT-5.4 (57.67), and it outperforms the latter in minimizing undertranslation; for expressive texts, DeepSeek-V3.2 achieves a higher average BLEU score (18.89) than ChatGPT-5.4 (15.03), demonstrating better reproduction of poetic prosody; for operative texts, both models yield a cosine similarity score above 0.98, demonstrating comparable ability to deliver emotional nuances. Through comparative analysis, this paper aims to provide empirical evidence for the selection of intelligent translation models for classics of different text types, and to offer a reference for the quality improvement of Chinese culture outbound translation and the construction of a multi-dimensional assessment system.
DOI: 10.35534/lin.0802014 (registering DOI)
Cite: Wei, Z., & Bai, L. (2026). Research on Quality Assessment of English Translation of Chinese Classics from the Perspective of Large Language Models—A Case Study of The Library of Chinese Classics. Advances in Linguistics Research, 8 (2), 171-182.