Grounding LLM Reasoning under Incomplete Graph Evidence

2026-06-29

AI Summary

This work addresses the unreliability of large language model (LLM) reasoning under incomplete knowledge graph evidence by proposing a soft grounding framework. The approach formalizes incomplete graph evidence in terms of entity anchors, relation residuals, path energies, and support regions, then integrates these with the LLM’s prior trajectory distribution through KL-regularized soft constraints. It is the first to characterize, in an open-world setting, how graph evidence constrains LLM inference without discarding “true but unobserved” reasoning paths—a limitation inherent to hard-constraint methods. Leveraging information-theoretic analysis and graph representation learning, the paper establishes stability bounds under evidence perturbations, clarifies suitable constraint mechanisms for scenarios such as GraphRAG and KGQA, and redefines knowledge graph compatibility not in terms of factual truth but as declarative support.

Abstract

Knowledge graphs can guide large language models (LLMs) reasoning, but the graph seen by a system is usually a retrieved, linked, temporally scoped, and incomplete evidence state rather than a complete account of truth. We develop a theoretical perspective on grounding observable LLM trajectories under such incomplete graph evidence.The evidence state induces entity anchors, typed relation residuals, path energies, and support regions, while the language model supplies a prior over candidate trajectories. We show that, under open-world incompleteness, no hard rule based only on the observed state can both reject every false unsupported trajectory and retain every true-but-unobserved one.We then characterize soft grounding as a KL-regularized deformation of the LLM prior: finite slack preserves support for unsupported but non-contradicted trajectories, whereas hard conditioning appears as an infinite-penalty limit.The framework also yields stability bounds under evidence perturbations and clarifies the constraint regimes appropriate for GraphRAG, KGQA, graph agents, constrained decoding, and faithful generation. The claims are evidence-relative: KG compatibility is treated as declared support, not factual truth.

Problem

• incomplete knowledge graph
• LLM reasoning
• evidence grounding
• open-world assumption
• trajectory support

Innovation

• soft grounding
• incomplete knowledge graph
• LLM reasoning
• KL regularization
• evidence-relative support