DocAtlas: Multilingual Document Understanding Across 80+ Languages

TL;DR

DocAtlas introduces a comprehensive multilingual document understanding framework for over 80 languages, utilizing high-fidelity OCR datasets, novel annotation methods, and effective training strategies to address low-resource language challenges.

Contribution

It presents a new dataset construction pipeline, a unified annotation format, and a training approach that enhances multilingual performance without degrading base-language accuracy.

Findings

Evaluating 16 models reveals significant gaps in low-resource scripts.

DPO with rendering-derived ground truth improves multilingual adaptation.

DocAtlas-DeepSeek outperforms baseline by 1.7% in accuracy.

Abstract

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 .