aiqbench analysis
This study identifies "meaning abstraction" in the intermediate layers of large models as the primary driver of their ability to predict human brain activity via fMRI and ECoG. By linking intrinsic dimensionality directly to semantic richness during pre-training, the research clarifies why specific hidden states align with neural responses rather than output predictions alone.
Abstraction Induces the Brain Alignment of Language and Speech Models
arXiv:2602.04081v2 Announce Type: replace Abstract: Research has repeatedly demonstrated that intermediate hidden states extracted from large language models and speech audio models predict measured brain response to natural language stimuli. Yet, very little is known about the representation properties that enable this high prediction performance. Why is it the intermediate layers, and not the output layers, that are most effective for this unique and highly general transfer task? We give evidence that the correspondence between speech and language models and the brain derives from shared meaning abstraction and not their next-word prediction properties. In particular, models construct higher-order linguistic features in their middle layers, cued by a peak in the layerwise intrinsic dimension, a measure of feature complexity. We show that a layer's intrinsic dimension strongly predicts how well it explains fMRI and ECoG signals; that the relation between intrinsic dimension and brain predictivity arises over model pre-training; and finetuning models to better predict the brain causally increases both representations' intrinsic dimension and their semantic content. Results suggest that semantic richness, high intrinsic dimension, and brain predictivity mirror each other, and that the key driver of model-brain similarity is rich meaning abstraction of the inputs, where language modeling is a task complex enough (but perhaps not the only) to require it.
Abstract:Research has repeatedly demonstrated that intermediate hidden states extracted from large language models and speech audio models predict measured brain response to natural language stimuli. Yet, very little is known about the representation properties that enable this high prediction performance. Why is it the intermediate layers, and not the output layers, that are most effective for this unique and highly general transfer task? We give evidence that the correspondence between speech and language models and the brain derives from shared meaning abstraction and not their next-word prediction properties. In particular, models construct higher-order linguistic features in their middle layers, cued by a peak in the layerwise intrinsic dimension, a measure of feature complexity. We show that a layer's intrinsic dimension strongly predicts how well it explains fMRI and ECoG signals; that the relation between intrinsic dimension and brain predictivity arises over model pre-training; and finetuning models to better predict the brain causally increases both representations' intrinsic dimension and their semantic content. Results suggest that semantic richness, high intrinsic dimension, and brain predictivity mirror each other, and that the key driver of model-brain similarity is rich meaning abstraction of the inputs, where language modeling is a task complex enough (but perhaps not the only) to require it.
Computation and Language (cs.CL)
arXiv:2602.04081 [cs.CL]
(or arXiv:2602.04081v2 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2602.04081 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 299 words
Related coverage
BAT-CLIP: Trimodal Alignment of Brain, Audio and Text
arXiv · Sep 28, 2026
MoDAl: Self-Supervised Neural Modality Discovery via Decorrelation for Speech Neuroprosthesis
arXiv · Aug 7, 2026
Models Can't Remember Their Training. Neither Can You.
LessWrong · Jul 19, 2026
Beyond representational alignment with brain-guided language models for robust reasoning
arXiv · Jun 12, 2026