Detecting Contextual Hallucinations in LLMs with Frequency-Aware Attention

TL;DR

This paper introduces a frequency-aware attention analysis method to detect hallucinations in large language models, revealing that hallucinated tokens are linked to high-frequency attention signals, and demonstrates improved detection performance.

Contribution

It proposes a novel frequency-based analysis of attention signals to identify hallucinations, offering a lightweight and effective detection approach.

Findings

High-frequency attention energy correlates with hallucinated tokens.

The method outperforms existing hallucination detection techniques.

Effective across different models and tasks.

Abstract

Hallucination detection is critical for ensuring the reliability of large language models (LLMs) in context-based generation. Prior work has explored intrinsic signals available during generation, among which attention offers a direct view of grounding behavior. However, existing approaches typically rely on coarse summaries that fail to capture fine-grained instabilities in attention. Inspired by signal processing, we introduce a frequency-aware perspective on attention by analyzing its variation during generation. We model attention distributions as discrete signals and extract high-frequency components that reflect rapid local changes in attention. Our analysis reveals that hallucinated tokens are associated with high-frequency attention energy, reflecting fragmented and unstable grounding behavior. Based on this insight, we develop a lightweight hallucination detector using high-frequency attention features. Experiments on the RAGTruth and HalluRAG benchmarks show that our approach achieves performance gains over verification-based, internal-representation-based, and attention-based methods across models and tasks.