Summary
Scientists studied how large language models (LLMs), which are computer programs that understand and generate language, connect to human brain activity when listening to language. They found that only a tiny part of the model's internal data—just 1 to 3% of it—matches brain activity well. This small section works across different people and types of brain scans, showing a shared, simple connection. The key parts that predict brain activity are different from the parts that vary the most inside the model, and by focusing on these parts, the researchers found features that track brain responses better. This helps explain why and how parts of these LLMs reflect human brain language processing.
What this means in practice
- •For neural engineers: Identify compact brain-predictive model components to improve brain-machine language interfaces during speech listening.
- •For cognitive computational modelers: Use the identified low-dimensional subspaces to enhance models that simulate human brain language processing efficiently.
Abstract
Large language models (LLMs) are increasingly used as computational models of human language processing, yet it remains unclear which components of their representations drive alignment with neural activity. Here, we use ridge reduced-rank regression to identify the brain-predictive subspace of LLM representations: the representational directions that support linear prediction of neural responses. Across three naturalistic language-listening datasets spanning intracranial electrophysiology and fMRI, and six model conditions spanning autoregressive, bidirectional masked, and randomly initialized language models, we find that full linear encoding performance can be recovered using only 1-3% of the original embedding dimensionality. These low-dimensional subspaces generalize across held-out participants and, to a lesser extent, across datasets, suggesting that neural predictivity is concentrated in a compact and partially shared representational space. We next compare these brain-predictive directions with the principal axes of variation in the embeddings. Principal component analysis requires substantially more dimensions to recover neural prediction, demonstrating that the directions most useful for predicting neural activity are distinct from those capturing the most overall variance within LLM representations themselves. Finally, characterizing the learned subspaces allows us to surface interpretable features that are overrepresented in neural activity above and beyond the LLM representations. Together, these results show that LLM-brain alignment is supported by a small, structured subset of model representations and provides a framework for better understanding exactly why these representations align. Code can be found at: https://github.com/thefinnlab/low-rank-alignment