Comparing Chinese–English Translation Performance of Large Language Models: A Multidimensional Evaluation of Financial, Technological, and Political Texts
Journal of Computer Science and Frontier Technologies · 26.06.2026 · 3 (3) · 61
DOI: 10.63313/jcsft.9087 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Computer Science and Frontier Technologies, online erschienen am 26.06.2026. Verfasst von Yaowei Xu und 1 weiteren Autor:innen. Die Arbeit nennt 17 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-SA 4.0).
Abstract (Originalsprache)
Large language models (LLMs) are increasingly used for machine translation; however, their reliability in professional Chinese–English translation, particularly for domain-specific texts, remains insufficiently examined. This challenge is especially evident in texts containing specialized terminology, numerical information, complex logical relations, and formal institutional phrasing. To address this issue, this study conducts a controlled comparison of six LLMs: ChatGPT-5.0, Claude Sonnet 4.5, DeepSeek-R1, GLM-4.6, Qwen3-MAX, and Grok 4.0. Three representative Chinese texts from the financial, technological, and political domains were selected and translated under three prompt conditions: zero-shot prompting, domain-enhanced prompting, and glossary-based prompting. Translation quality was assessed through a multidimensional evaluation framework. Standard reference-based metrics, including BLEU, ChrF++, TER, ROUGE, BERTScore, and COMET, were used to measure automatic translation quality. In addition, lexical richness and syntactic-discourse indicators were incorporated to examine stylistic and structural differences across model outputs. The results reveal a clear domain effect, indicating that financial, technological, and political texts present different levels and types of translation difficulty for LLMs. By contrast, prompt strategy did not produce significant main effects under the three tested conditions. At the model level, GLM-4.6 achieved the highest COMET-based ranking across all prompt settings, while DeepSeek-R1 showed comparatively stable second-place performance. Overall, the findings suggest that, within this controlled exploratory dataset, model type and text domain exert a stronger influence on translation performance than prompt modification alone. Rather than offering a definitive ranking of LLM translation capability, this study provides empirical evidence on how selected LLMs handle specialized Chinese–English translation under comparable experimental conditions. Future research should expand the dataset, incorporate human evaluation, and further investigate the role of terminology constraints in improving translation accuracy and stylistic appropriateness.
Zitieren
APA
Xu, Y., & Fu, Z. (2026). Comparing Chinese–English Translation Performance of Large Language Models: A Multidimensional Evaluation of Financial, Technological, and Political Texts. Journal of Computer Science and Frontier Technologies, 3(3), 61. https://doi.org/10.63313/jcsft.9087
Vancouver
Xu Y, Fu Z. Comparing Chinese–English Translation Performance of Large Language Models: A Multidimensional Evaluation of Financial, Technological, and Political Texts. Journal of Computer Science and Frontier Technologies. 2026;3(3):61. doi:10.63313/jcsft.9087
BibTeX
@article{Xu2026,
author = {Xu, Yaowei and Fu, Zhihan},
title = {Comparing Chinese–English Translation Performance of Large Language Models: A Multidimensional Evaluation of Financial, Technological, and Political Texts},
journal = {Journal of Computer Science and Frontier Technologies},
year = {2026},
volume = {3},
number = {3},
pages = {61},
doi = {10.63313/jcsft.9087}
}
EUVIMED