Search papers, labs, and topics across Lattice.
ClueWeaver introduces a dual-agent framework designed for long-narrative question answering using compact language models, addressing the limitations of existing costly long-context models. The Finder agent retrieves critical evidence through retrieval-guided segmentation, while the Interpreter processes this evidence to provide accurate answers and rationales, all optimized via reward-guided reinforcement learning. Experiments demonstrate that ClueWeaver significantly enhances the performance of local models in terms of evidence coverage and reasoning transparency, making it a valuable tool for humanities and social science research.
ClueWeaver transforms how compact language models tackle long narratives, achieving superior evidence retrieval and reasoning transparency.
Humanities and social science research requires close reading of long narrative materials such as novels, scripts, archives, and case reports, yet many users have limited access to costly proprietary long-context models. Compact, locally deployable language models are a practical alternative, but directly feeding them an entire long context remains costly, hard to inspect, and prone to missing sparse evidence. We present ClueWeaver, an evidence-aware dual-agent framework for long-narrative question answering with compact local models. A Finder identifies passages containing answer-critical clues through retrieval-guided segmentation, while an Interpreter derives the answer from the selected evidence, produces rationales with paragraph-ID citations, and applies an internal self-calibration pass for high-risk questions. Both agents are optimized with reward-guided reinforcement learning: Finder rewards emphasize evidence retention and faithful paragraph-ID references, and Interpreter rewards emphasize correctness, grounding, and concise explanations. This decomposition makes evidence selection and reasoning more inspectable than end-to-end prompting. Experiments across multiple long-context narrative question answering and claim verification settings show that ClueWeaver substantially improves local end-to-end language models while providing evidence coverage and paragraph-referenced reasoning traces. Code is available at https://github.com/Ameame1/ClueWeaver.
