Despite the central role of attention heads in Transformers, we lack tools to understand why a model attends to a particular token. To address this, we study the query-key (QK) space -- the bilinear joint embedding space between queries and keys. We present a contrastive covariance method to decompose the QK space into low-rank, human-interpretable components. It is when features in keys and queries align in these low-rank subspaces that high attention scores are produced. We first study our method both analytically and empirically in a simplified setting. We then apply our method to large language models to identify human-interpretable QK subspaces for categorical semantic features and binding features. Finally, we demonstrate how attention scores can be attributed to our identified features.
@article{arxiv.2602.04752,
title = {Decomposing Query-Key Feature Interactions Using Contrastive Covariances},
author = {Andrew Lee and Yonatan Belinkov and Fernanda Viégas and Martin Wattenberg},
journal= {arXiv preprint arXiv:2602.04752},
year = {2026}
}