Mitigating Context-Memory Conflicts in LLMs through Dynamic Cognitive Reconciliation Decoding

2026-05-12

AI Summary

This work addresses the challenge that existing decoding strategies struggle to balance knowledge conflict resolution and generation quality when parametric knowledge in large language models (LLMs) contradicts external information provided in the context. We propose Dynamic Cognitive Reconciliation Decoding (DCRD), which introduces a novel two-stage dynamic decoding mechanism: it first analyzes attention maps to predict potential knowledge conflicts and then adaptively selects between greedy decoding and a context-faithful contrastive decoding path. DCRD maintains strong performance in non-conflicting scenarios while significantly improving knowledge integration accuracy in complex, real-world settings. We evaluate our approach across four prominent LLMs and six question-answering benchmarks, and further introduce ConflictKG, a new evaluation suite for knowledge conflict scenarios. Results demonstrate that DCRD consistently outperforms current state-of-the-art baselines.

Abstract

Large language models accumulate extensive parametric knowledge through pre-training. However, knowledge conflicts occur when outdated or incorrect parametric knowledge conflicts with external knowledge in the context. Existing methods address knowledge conflicts through contrastive decoding, but in conflict-free scenarios, static approaches disrupt output distribution. Other dynamic decoding methods attempt to measure the degree of conflict but still struggle with complex real-world situations. In this paper, we propose a two-stage decoding method called Dynamic Cognitive Reconciliation Decoding (DCRD), to predict and mitigate context-memory conflicts. DCRD first analyzes the attention map to assess context fidelity and predict potential conflicts. Based on this prediction, the input is directed to one of two decoding paths: (1) greedy decoding, or (2) context fidelity-based dynamic decoding. This design enables DCRD to handle conflicts efficiently while maintaining high accuracy and decoding efficiency in conflict-free cases. Additionally, to simulate scenarios with frequent knowledge updates, we constructed ConflictKG, a knowledge conflict QA benchmark. Experiments on four LLMs across six QA datasets show that DCRD outperforms all baselines, achieving state-of-the-art performance.

Problem

• context-memory conflicts
• knowledge conflicts
• large language models
• dynamic decoding
• parametric knowledge

Innovation

• Dynamic Cognitive Reconciliation Decoding
• context-memory conflict
• attention map analysis
• dynamic decoding
• ConflictKG