Bridging Auxiliary Constraints to Resolve Instruction Following in Large Reasoning Models
2026-06-02 • Artificial Intelligence
Artificial IntelligenceComputation and Language
AI summaryⓘ
The authors study how large reasoning models (LRMs) sometimes fail to follow complex sets of instructions properly. They define this problem as the Constraint Adherence Problem (CAP) and propose a new method called Constraint Relationship Graph Completion (CRGC). This method represents instructions as graphs that highlight relationships between constraints and finds special 'bridge constraints' to help the model better understand and balance these instructions. Their tests show that CRGC cuts down errors by 39% compared to usual methods, without hurting the model's reasoning skills.
Large Reasoning ModelsConstraint Adherence ProblemInstruction FollowingKnowledge GraphConstraint Relationship Graph CompletionBridge ConstraintsPromptingConstraint SatisfactionReasoning Abilities
Authors
Zhengyi Zhao, Shubo Zhang, Huimin Wang, Zezhong Wang, Yutian Zhao, Yefeng Zheng, Binyang Li, Yulan He, Kam-Fai Wong, Xian Wu
Abstract
Large Reasoning Models (LRMs) have demonstrated impressive capabilities in many tasks, yet they struggle with reliably following multiple instructions, either by failing to satisfy individual constraints or by struggling to balance competing constraints simultaneously. We formalize this challenge as the Constraint Adherence Problem (CAP). This paper introduces a novel framework that addresses CAP by representing instructions as a structured knowledge graph of constraints. Our approach, Constraint Relationship Graph Completion (CRGC), explicitly models relationships between constraints, identifies adherence challenges, and discovers ``bridge constraints'' that help the model better focus on and reconcile requirements. Bridge constraints act as auxiliary instructions that make primary constraints more salient and compatible. Unlike existing approaches that enhance instruction following through general training methods, CRGC specifically improves constraint satisfaction by leveraging the model's own knowledge to create better pathways for generation. Experiments across three popular instruction following datasets demonstrate that our approach reduces constraint violations by 39% compared to standard prompting while maintaining reasoning abilities of large reasoning models.