构建82种语言的高精度文档数据集，解决低资源语言理解难题
DocAtlas: Multilingual Document Understanding Across 80+ Languages
- 通过无模型渲染与合成生成双管道构建统一格式标注数据
- 多语言测试显示低资源语言仍有显著性能差距
- 用渲染真值优化模型，提升跨语言泛化能力且不损害母语表现
多语言文档理解在低资源语言上受限于训练数据稀缺和基于模型的标注流程带来的偏见。我们提出DocAtlas框架，构建覆盖82种语言和9项评估任务的高保真OCR数据集与基准。双通道流水线分别对原生DOCX文档进行差异渲染，以及针对从右到左书写的文本采用合成LaTeX生成，以统一的DocTag格式精确编码布局、文本和组件类型，无需依赖学习模型完成核心标注。评估16个先进模型发现，低资源脚本仍存在明显性能缺口。我们证明，使用渲染生成的真值作为正样本进行直接偏好优化（DPO），可实现稳定的多语言适应，使域内准确率提升1.9%，域外准确率提升1.8%，且无明显母语性能下降；而监督微调反而导致域外性能下降高达21%。最佳变体DocAtlas-DeepSeek相较最强基线提升1.7%。代码已开源：https://github.com/ahmedheakl/DocAtlas。
原文摘要 · Abstract (English)
Multilingual document understanding remains limited for low-resource languages due to scarce training data and model-based annotation pipelines that perpetuate existing biases. We introduce DocAtlas, a framework that constructs high-fidelity OCR datasets and benchmarks covering 82 languages and 9 evaluation tasks. Our dual pipelines, differential rendering of native DOCX documents and synthetic LaTeX-based generation for right-to-left scripts produce precise structural annotations in a unified DocTag format encoding layout, text, and component types, without learned models for core annotation. Evaluating 16 state-of-the-art models reveals persistent gaps in low-resource scripts. We show that Direct Preference Optimization (DPO) using rendering-derived ground truth as positive signal achieves stable multilingual adaptation, improving both in-domain (+1.9%) and out-of-domain (+1.8%) accuracy without measurable base-language degradation, where supervised fine-tuning degrades out-of-domain performance by up to 21%. Our best variant, DocAtlas-DeepSeek, improves +1.7% over the strongest baseline. Code is available at https://github.com/ahmedheakl/DocAtlas .
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。