GRAFT: A Graph-based Flow-aware Agentic Framework for Document-level Machine Translation
Himanshu Dutta, Sunny Manchanda, Prakhar Bapat, Meva Ram Gurjar, Pushpak Bhattacharyya · 2025
Enterprises, public organizations, and localization providers increasingly rely on Documentlevel Machine Translation (DocMT) to process contracts, reports, manuals, and multimedia transcripts across languages.However, existing MT systems often struggle to handle discourse-level phenomena such as pronoun resolution, lexical cohesion, and ellipsis, resulting in inconsistent or incoherent translations.We propose GRAFT, a modular graph-based DocMT framework that leverages Large Language Model (LLM) agents to segment documents into discourse units, infer inter-discourse dependencies, extract structured memory, and generate context-aware translations.GRAFT transforms documents into directed acyclic graphs (DAGs) to explicitly model translation flow and discourse structure.Experiments across eight language directions and six domains show GRAFT outperforms commercial systems (e.g., Google Translate) and closed LLMs (e.g., GPT-4) by an average of 2.8 d-BLEU, and improves terminology consistency and discourse handling.GRAFT supports deployment with opensource LLMs (e.g., LLaMA, Qwen), making it cost-effective and privacy-preserving.These results position GRAFT as a robust solution for scalable, document-level translation in real-world applications.The codebase and data can be found at https://github.com/ himanshu-dutta/graft. Maria glanced at the clock.It was already past 6 PM, and she hadn't heard from John all day.The sun dipped below the horizon as Maria stepped out onto the balcony.The air smelled of rain, but the storm hadn't arrived yet.John was stuck in traffic.His phone buzzed repeatedly, but he couldn't answer.Maria sighed."If only he would pick up," she thought, staring at the darkening sky.The storm finally broke while Maria was preparing dinner.Rain lashed against the windows, and thunder rumbled in the distance.In another part of town, John finally pulled into his driveway.He was drenched and exhausted, but the sight of the glowing porch light made him smile.Maria's phone buzzed.She hesitated before picking it up.It was John.