SubDocTrans: Enhancing Document-level Machine Translation with Plug-and-play Multi-granularity Knowledge Augmentation

Large language models (LLMs) have recently achieved remarkable progress in sentence-level machine translation, but scaling to documentlevel machine translation (DocMT) remains challenging, particularly in modeling contexts and discourse phenomena across sentences and paragraphs. Document translations generated by LLMs often suffer from poor consistency, weak coherence, and omission errors. To address the issues, we propose a novel DocMT framework named SubDocTrans, that enables LLMs to produce high-quality translations via plug-and-play multi-granularity knowledge augmentation. SubDocTrans first performs topic segmentation to divide a document into coherent sub-documents, enabling efficient document-level translation while preserving contextual coherence. For each sub-document, both global and local knowledge are extracted including bilingual summary, theme, proper nouns, topics, and transition hint. We then incorporate the multi-granularity knowledge into the prompting strategy, to guide LLMs in producing consistent, coherent, and accurate translations. We also conduct extensive experiments across various DocMT tasks, and the results demonstrate the effectiveness of our framework, particularly in improving consistency and coherence, reducing omission errors, and mitigating hallucinations.