海事文本的生成式AI翻译质量对比研究

A Comparative Study on the Translation Quality of Generative AI on Maritime Texts

摘要: 本研究针对生成式AI在海事领域的翻译质量展开系统评估。选取DeepSeek、豆包、ChatGPT和Gemini四款主流模型，以海事专业文本为基础，采用BLEU指标对英中/中英双向翻译质量进行量化分析。研究发现：四个模型在海事文本汉译英任务中表现均优于英译汉，Gemini综合表现最优且稳定，DeepSeek在技术文本中可靠，ChatGPT输出波动较大，豆包更适用于通用文本。研究成果为海事相关文本的翻译选择AI工具提供了实证依据。

Abstract: The study systematically evaluates the translation quality of generative AI models in the maritime domain. We selected four mainstream models—DeepSeek, Doubao, ChatGPT, and Gemini—and used specialized maritime texts as the basis for quantitative analysis of English-to-Chinese (E-C) and Chinese-to-English (C-E) bidirectional translation quality using the BLEU metric. The findings indicate that all four models perform better in the maritime C-E translation task than in the E-C task. Gemini demonstrates the best overall and most stable performance, DeepSeek proves reliable for technical documents, ChatGPT’s output shows significant variability, and Doubao is more suitable for general texts. The research results provide empirical evidence for selecting AI tools for translating maritime-related texts.