Far from the Shallow: Brain-Predictive Reasoning Embedding through Residual Disentanglement

Understanding how the human brain progresses from processing simple linguistic inputs to performing high-level reasoning is a fundamental challenge in neuroscience. While modern large language models (LLMs) are increasingly used to model neural responses to language, their internal representations are highly "entangled," mixing information about lexicon, syntax, meaning, and reasoning. This entanglement biases conventional brain encoding analyses toward linguistically shallow features (e.g., lexicon and syntax), making it difficult to isolate the neural substrates of cognitively deeper processes. Here, we introduce a residual disentanglement method that computationally isolates these components. By first probing an LM to identify feature-specific layers, our method iteratively regresses out lower-level representations to produce four nearly orthogonal embeddings for lexicon, syntax, meaning, and, critically, reasoning. We used these disentangled embeddings to model intracranial (ECoG) brain recordings from neurosurgical patients listening to natural speech. We show that: 1) This isolated reasoning embedding exhibits unique predictive power, accounting for variance in neural activity not explained by other linguistic features and even extending to the recruitment of visual regions beyond classical language areas. 2) The neural signature for reasoning is temporally distinct, peaking later ( 350-400ms) than signals related to lexicon, syntax, and meaning, consistent with its position atop a processing hierarchy. 3) Standard, non-disentangled LLM embeddings can be misleading, as their predictive success is primarily attributable to linguistically shallow features, masking the more subtle contributions of deeper cognitive processing. Our work provides compelling neural evidence for an abstract reasoning computation during language comprehension and offers a robust framework for mapping distinct cognitive functions from artificial models to the human brain 2 .

A growing body of work has shown that LLMs exhibit strong representational alignment with human brain activity during language comprehension. These studies employ language encoding models, powerful predictors of measured brain activity in response to a language stimulus [41,19,10,20,38,5,13,34,17,7,27,2,3,8,9]. Typically, these models are created by inputting a language stimulus into an LLM, to create a contextual embedding of the stimulus using the models' internal hidden state. A linear mapping is then learned between these hidden states and the measured brain response to that stimulus. Modern language encoding models represent some of the most advanced computational approaches for studying neural processing across sensory modalities, and they provide a unique window into the semantic processing landscape of human cortex. * Equal Contribution. 2 Code available here. 39th Conference on Neural Information Processing Systems (NeurIPS 2025).