The neuroscience of perception has recently been revolutionized with an integrative modeling approach in which computation, brain function, and behavior are linked across many datasets and many computational models. By revealing trends across models, this approach yields novel insights into cognitive and neural mechanisms in the target domain. We here present a systematic study taking this approach to higher-level cognition: human language processing, our species’ signature cognitive skill. We find that the most powerful “transformer” models predict nearly 100% of explainable variance in neural responses to sentences and generalize across different datasets and imaging modalities (functional MRI and electrocorticography). Models’ neural fits (“brain score”) and fits to behavioral responses are both strongly correlated with model accuracy on the next-word prediction task (but not other language tasks). Model architecture appears to substantially contribute to neural fit. These results provide computationally explicit evidence that predictive processing fundamentally shapes the language comprehension mechanisms in the human brain.

What is the relationship between language and other high-level cognitive functions? Neuroimaging studies have begun to illuminate this question, revealing that some brain regions are quite selectively engaged during language processing, whereas other "multiple-demand" (MD) regions are broadly engaged by diverse cognitive tasks. Nonetheless, the functional dissociation between the language and MD systems remains controversial. Here, we tackle this question with a synergistic combination of functional MRI methods: we first define candidate language-specific and MD regions in each subject individually (using functional localizers) and then measure blood oxygen level-dependent signal fluctuations in these regions during two naturalistic conditions ("rest" and story-comprehension). In both conditions, signal fluctuations strongly correlate among language regions as well as among MD regions, but correlations across systems are weak or negative. Moreover, data-driven clustering analyses based on these inter-region correlations consistently recover two clusters corresponding to the language and MD systems. Thus although each system forms an internally integrated whole, the two systems dissociate sharply from each other. This independent recruitment of the language and MD systems during cognitive processing is consistent with the hypothesis that these two systems support distinct cognitive functions.

No abstract