Research on the Efficacy and Paradigm of AI-Assisted English Translation of Forestry Ecological Security Legal Texts— A Comparative Experiment Based on Zero-Shot, Few-Shot, and Chain-of-Thought Prompting Strategies

Abstract

The international dissemination of China's ecological rule-of-law discourse relies heavily on high-quality English translation of legal texts. However, English translation of forestry ecological security legal texts has long faced quality bottlenecks due to inconsistent species nomenclature, complex interdisciplinary semantics of ecological terms, and difficulties in restructuring complex legal sentences. The emergence of Generative Artificial Intelligence (LLMs) offers a novel path for legal translation in domain-specific contexts. Using the Forest Law and the Wetland Protection Law of the People's Republic of China as corpora, this study constructs a Chinese-English parallel dataset and designs a comparative experiment across three prompting strategies—Zero-shot, Few-shot, and Chain-of-Thought (CoT)—based on a 5-point Fluency and Adequacy scale. The results indicate that Chain-of-Thought prompting achieves the best performance in restructuring complex sentences and converting legal jurisdiction concepts; few-shot examples significantly reduce mistranslation rates; and general-purpose large language models can efficiently adapt to forestry legal translation scenarios through prompt engineering optimization. This study constructs the 'Prompt Guide for AI-Assisted Forestry Legal Translation,' providing a reusable solution for cross-border forestry law enforcement cooperation and the international dissemination of ecological rule of law.