New framework improves legal AI by decomposing complex questions

Researchers have developed a new framework called Decompose-and-Refine (DaR) to improve legal question answering using large language models. DaR addresses the challenge of accurately retrieving relevant legal statutes for complex, multi-hop questions by breaking them down into smaller, manageable sub-questions. It then uses parametric knowledge to refine queries for each sub-question, ensuring better alignment with statutory text and reducing the risk of hallucination. Evaluations on the Korean KoBLEX benchmark showed DaR enhances both retrieval accuracy and final answer quality. AI

IMPACT Enhances LLM accuracy in legal applications by improving statutory grounding and reducing hallucinations.

RANK_REASON The cluster contains a research paper detailing a new framework for legal question answering using LLMs. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• Decompose-and-Refine (DaR)
• Gemma3-27B
• KoBLEX
• Qwen3-32B

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →