Presentation
Search Abstracts | Symposia | Slide Sessions | Poster Sessions | Poster Slams
Heterogeneous neural responses distributed across the language network revealed by electrocorticography
Poster A61 in Poster Session A, Thursday, October 6, 10:15 am - 12:00 pm EDT, Millennium Hall
Tamar I Regev1, Colton Casto1, Eghbal A Hosseini1, Markus Adamek2,3, Peter Brunner2,3,4, Evelina Fedorenko1; 1Massachusetts Institute of Technology, 2National Center for Adaptive Neurotechnologies, Albany, NY, 3Washington University School of Medicine, 4Albany Medical College
A left-lateralized network of frontal and temporal brain regions is specialized for language processing, spoken, written, or signed. Different regions of this ‘language network’ have all been shown to be sensitive to various kinds of linguistic information, from combinatorial sentence structure, to word meanings, to sub-lexical/phonotactic regularities (Fedorenko et al., 2020; Regev et al., 2021). However, whether neural computations are the same across these different brain regions remains debated. A key limitation of many past studies is the poor temporal resolution of fMRI—a dominant imaging modality in language research. Smearing the fine details of neural responses across time potentially obscures important aspects of the neural computations that support language comprehension. To shed light on this debate, we present data from 6 patients with intractable epilepsy, who were implanted with electrodes placed directly on the brain surface and agreed to perform a language task that has been extensively used in past fMRI work. The task included reading sentences, lists of unconnected words, ‘Jabberwocky’ sentences (which contain a syntactic frame made up of function words and functional morphological endings, but in which content words are replaced with nonwords), and lists of unconnected nonwords, followed by a word/nonword memory probe. These data (from n=106 language-selective electrodes, defined by a significant Sentences>Nonwords effect, as in Fedorenko et al., 2016) reveal functionally heterogeneous responses (high gamma power, 70-150Hz range; Crone et al., 1998) to these four conditions. Furthermore, electrodes appear to differ in their temporal dynamics over the course of the 8-word/nonword-long stimulus, and in the degree of their time-locking to the stimulus. To formally evaluate these apparently distinct functional-temporal profiles, we performed k-means clustering on all language-selective electrode responses and identified 4-6 main response types, which together explained over 80% of the variance in the data. One response type showed a strong preference for sentences over all other conditions (presumably supporting sentence structure building) and showed weak stimulus time-locking. Another response type showed stronger responses to sentences and word lists over the other conditions (presumably supporting the processing of word meanings) and more time-locking to the stimulus. Yet another response type showed a relatively strong response to nonwords (presumably sensitive to sub-lexical regularities) and a strong degree of stimulus time-locking. Interestingly, all of these response types were distributed across the various parts of the language network, in both temporal and frontal regions. Notably, in line with previous fMRI reports, no electrode showed stronger responses to the Jabberwocky condition, which aims to isolate syntactic processing, relative to the word-list condition, which targets lexico-semantic processing. These results demonstrate the functional heterogeneity of neural responses in the language network, and highlight the diverse temporal dynamics that may give rise to neural computations needed in order to extract meaning from linguistic input. The mosaic of neural responses across the language network suggests that all regions of the language network have direct access to distinct response types—a property that may be crucial for the function of the language network.
Topic Areas: Computational Approaches, Meaning: Lexical Semantics