The Sum Is Less Than Its Parts: Decomposing the Attention Write by Source Detects Contextual Hallucinations
Abstract
When a language model answers from a retrieved document, its attention heads mix evidence from the context with evidence from the model’s own self-generated tokens and write the sum into the residual stream. We split each head’s write into these two source-specific contributions before they are summed, and find that the two sides are complementary. The context write, read out along the model’s unembedding, gives a training-free groundedness score, GroundHOG-U. The response write (its magnitude and directional agreement with the other residual writes) gives a supervised hallucination detector, GroundHOG. Across grounded-generation benchmarks and model families, GroundHOG beats prior internal-state detectors, and GroundHOG-U is the strongest training-free groundedness signal on the multi-model benchmarks.
Steffen Castle
PhD Candidate
Leonhard Hennig
Senior Researcher
