Search

A Study on the Accuracy and Comparison of Metaphor Translation of the Character "Liu" (Willow) in Ancient Chinese Poetry by Three Large Language Models—An Empirical Analysis based on ChatGPT, DeepSeek and Doubao

Author(s)

Zheng Mengli

Affiliation(s)

Hubei University of Technology, Wuhan, Hubei, China

Abstract

Objective: To evaluate the accuracy and differences in metaphor translation of the character "Liu" (willow) in ancient Chinese poetry by three large language models: ChatGPT, DeepSeek, and Doubao.Methods: A comparative method was adopted, in which the three models were used to translate 61 ancient poems containing "Liu" in batches. The accuracy of metaphor translation (translated/ untranslated) was judged manually, the accuracy rate was quantified, and visual analysis was conducted via Python.Results: The accuracy rates of GPT and DeepSeek were both 86.89%, while that of Doubao was slightly lower at 85.25%. GPT performed well in metaphors unrelated to lovesickness and loneliness between men and women. Doubao exhibited stable translation of high-frequency metaphors with the smallest temporal fluctuation. However, all three models performed poorly in low-frequency metaphors such as frustration in official career and gestures of beauties, especially Doubao.Conclusion: Mainstream large language models possess basic capabilities in metaphor translation of ancient Chinese poetry. Model selection should be based on specific scenarios, and their ability to process low-frequency and uncommon metaphors needs targeted improvement.

Keywords

Large Language Models; Translation of Ancient Chinese Poetry; Metaphor of "Liu" (Willow); Translation Accuracy; Multidimensional Evaluation

References

[1]Ed-Dali R. Assessing DeepSeek R1 and ChatGPT 4.5 in Arabic-English literary translation[J].Cogent Arts & Humanities,2025,12(1). [2]Li, RC; Nawi, AM and Kang, MS. Human-machine Translation Model Evaluation Based on Artificial Intelligence Translation[J].EMITTER-INTERNATIONAL JOURNAL OF ENGINEERING TECHNOLOGY,2023,11(2):145-159. [3]Wael Etaiwi, Bushra Alhijawi.Comparative Evaluation of ChatGPT and DeepSeek Across Key NLP Tasks: Strengths, Weaknesses, and Domain-Specific Performance[J].ARRAY，2025，27. [4]Hui Jiao,Bei Peng,Lu Zong,Xiaojun Zhang,Xinwei Li.Gradable ChatGPT Translation Evaluation [J].PROCESAMIENTO DEL LENGUAJE NATURAL,2024,72:73-85. [5]L. David Ritchie & Xuede Zhao.To “Face the Powder” or “Powder the Face”? Contemporary Metaphor Theory and the Art of Chinese to English Translation [J].METAPHOR AND SYMBOL,2020,35(2):122-135. [6]Li Hongyu, Che Mingming. Comparative Analysis of Linguistic Features of Scientific Texts Based on ChatGPT and DeepSeek Translation Models [J]. Journal of North China University of Science and Technology (Social Science Edition), 2025, 25(4): 88-96 [7]Du Lijuan. My View on the Translation of "Willow" in Farewell Poems [J]. Journal of Tongren University, 2014, 16(3): 67-69 [8]. Sun Xiuhua. An Overview of Plant Imagery and Cultural Implication Interpretation in "Nineteen Ancient Poems"[J]. Ningxia Social Sciences, 2010, 162(5): 168-172 [9]Feng Quangong, Li Lin. A Study on Emotional Metaphors in Chinese Classical Poetry and Their English Translation [J]. Language and Translation, 2017, 3: 57-64 [10]An Yong, Xie Xiaoli. Fragrant Grass and Elegant Trees Enhance Implication, Carefully Deliberate and Translate Elegantly and Truthfully - A Probe into the Translation of Plant Names and Related Expressions in Tang and Song Dynasty Poetry [J]. Journal of Shanghai Customs College, 2010, 2: 99-102 [11]Xu Yuanchong. Three Hundred Tang Poems in Chinese-English Parallel Text [M]. Beijing: Higher Education Press, 2000 [12]Xu Yuanchong, Lu Peixian, Wu Juntao. A New Translation of Three Hundred Tang Poems [M]. Beijing: China Translation Publishing House, 1997 [13]Yang Xianyi, Gladys Yang, et al. Ancient Poetry Garden: A Collection of Chinese-English Translations [M]. Beijing: Foreign Languages Press, 2001