TITLE:
Challenges and Solutions for Large Language Models in Metaphor Translation of Political Texts
AUTHORS:
Shubin Chen, Zhen Wang
KEYWORDS:
Large Language Models (LLMs), Translation of Political Texts, Metaphor Comprehension
JOURNAL NAME:
Open Journal of Applied Sciences,
Vol.15 No.8,
August
19,
2025
ABSTRACT: This paper aims to analyze the dilemmas faced by large language models (LLMs) in the metaphor translation of current political texts and explore strategies to enhance the accuracy of such translations. Through the literature research method, it sorts out the working principles of LLMs, the characteristics and translation requirements of current political texts, and theories related to metaphors. The study finds that when LLMs translate metaphors in current political texts, they struggle to accurately capture the political connotations and semantic evolution behind metaphors at the semantic comprehension level; due to cultural differences, it is difficult to convey the cultural connotations when translating metaphors with local cultural colors; in complex contexts, their understanding of metaphors in long texts and those with implicit political intentions is insufficient; problems such as insufficient data quality and diversity, poor interpretability, and lack of robustness of the models themselves also restrict translation effectiveness. To break through these dilemmas, strategies are proposed, including adding high-quality current political text data with precise metaphor annotations, constructing a political and cultural knowledge graph, strengthening the model’s ability to analyze the context of long texts, improving the model architecture to enhance interpretability and robustness, developing metaphor translation auxiliary tools, and introducing human intervention. These strategies are expected to provide directions for the development of artificial intelligence translation.