STAR：面向文档到文档机器翻译的句子翻译对齐率
STAR : Sentence Translation Alignment Rate for Document-to-Document Machine Translation
浏览论文内容
中文总结 AI 辅助
该研究针对文档到文档机器翻译的结构对齐问题，提出STAR指标与StarPO框架，实验显示StarPO可提升翻译质量与结构完整性，还能让小型模型超越GPT-4o等大型系统并保持更优令牌效率。
中文摘要 AI 辅助
大型语言模型（LLMs）已推动机器翻译从句子级转向文档到文档（Doc2Doc）机器翻译，有望提升全局连贯性。但一次性完成的文档到文档生成常存在结构对齐问题，表现为句子遗漏或幻觉，违反源-目标对应这一核心要求。为解决该问题，我们引入句子翻译对齐率（STAR），这一辅助指标可明确量化句子级结构保真度。在此基础上，我们提出STAR掩码偏好优化（StarPO）框架，该框架按结构质量对文档级假设排序，并利用动态对齐掩码将优化聚焦于未对齐片段。在新闻和文学领域的实验表明，StarPO可显著提升翻译质量与结构完整性。值得注意的是，StarPO能让小型模型超越GPT-4o等大型专有系统的性能，同时保持更优的令牌效率。
英文摘要
Large Language Models (LLMs) have enabled a shift from sentence-level to document-to-document (Doc2Doc) machine translation, promising improved global coherence. However, document-to-document generation in a single pass frequently suffers from structural misalignment, manifesting as sentence omissions or hallucinations that violate the core requirement of source-target correspondence. To address this, we introduce Sentence Translation Alignment Rate (STAR), an auxiliary metric that explicitly quantifies sentence-level structural fidelity. Building on this, we propose STAR-masked Preference Optimization (StarPO), a framework that ranks document-level hypotheses by structural quality and utilizes a dynamic alignment mask to focus optimization on misaligned segments. Experimental results across news and literary domains demonstrate that StarPO significantly enhances translation quality and structural integrity. Notably, StarPO allows compact models to surpass the performance of massive proprietary systems like GPT-4o while maintaining superior token efficiency.
发表机构
- School of Computer Science and Technology, Soochow University（苏州大学计算机科学与技术学院）
- Alibaba Group（阿里巴巴集团）
机构由 AI 辅助整理，请以论文原文为准。
