Abstract
Scientific Reasoning Graph Extraction (SRGE) aims to recover explicit links among observations, evidence, intermediate claims, and paper-level conclusions. While LLMs can produce graph-like scientific explanations, their outputs often suffer from malformed syntax, drifting edge labels, incorrectly oriented roots, and weak source anchors. To address this, we propose PEARL (Peircean Extraction via Abstraction and Repair Layer), a training-free framework that transforms noisy LLM graph responses into auditable reasoning graphs and repairs them for strict semantic validity.
PEARL first materializes explicit graph content under a closed Peircean schema, then employs matched evidence-grounded judge feedback to repair rejected edge types, local inference steps, and terminal roots while preserving an audit trail. On five 70-paper model archives from ARCHE, PEARL raises strict gate passes from 0/350 for the LLM baseline to 300/350, with average REA improving from 0.339 to 0.906. The graphs provide a reliability layer for research-agent and AI scientist workflows that require inspectable reasoning traces rather than unconstrained graph regeneration. Code and audit artifacts are available at GitHub.
Blogger's Review: The introduction of the PEARL framework offers a fresh perspective on the accuracy and auditability of scientific reasoning graphs. By integrating feedback mechanisms with a strict semantic structure, it significantly enhances the reliability of LLM-generated content, which is crucial for the transparency and credibility of scientific research.