Distributional Semantics Tracing: A Framework for Explaining Hallucinations in Large Language Models

Researchers introduce Distributional Semantics Tracing (DST), a new framework for explaining hallucinations in large language models by tracking how semantic representations drift across neural network layers. The method reveals that hallucinations occur when models are pulled toward contextually inconsistent concepts based on training correlations rather than actual prompt context.

• →DST provides a model-native method to trace and explain hallucination formation in LLMs by building layer-wise semantic maps.
• →Hallucinations arise from correlation-driven representational drift where models favor familiar concept neighborhoods over contextual accuracy.
• →The framework outperforms existing attribution and probing methods in explaining model failures under LLM-judge evaluation.
• →The Contextual Alignment Score (CAS) effectively predicts when models will produce hallucinated outputs.
• →The research provides new insights into the mechanistic causes of AI model unreliability and potential mitigation strategies.