Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Precision fMRI reveals how different cortical networks contribute to lexical semantics
Poster Session B, Wednesday, September 30, 4:30 - 6:30 pm, Wangari Maathai
Leonardo Fernandino1, Jia-Qing Tong1, Stephen Mazurchuk1, Joseph Heffernan1; 1Medical College of Wisconsin
Introduction: Cortical parcellations based on resting-state connectivity have identified several large-scale networks, including a perisylvian network often assumed to support high-level language processing. This network largely overlaps with the areas identified in the contrast Sentence > Pseudoword List, proposed by Fedorenko et al. (2011) as a localizer for language-specific regions. We investigated the extent to which each of these networks contributes to lexical semantic processing by examining activation patterns elicited by hundreds of individual words across three fMRI studies. Methods: We analyzed three fMRI datasets in which neurotypical adults rated the familiarity of visually presented English nouns. Participants (N = 39, 40, and 26) and word sets (320, 300, and 600 nouns) were different in each dataset. Each word was repeated 3–6 times across three sessions. Dataset 3 also included 40 pseudowords and three resting-state scans, enabling a Word > Pseudoword contrast and individual-level resting-state parcellation. General linear models included word frequency, concreteness, and nine orthophonological variables, yielding whole-brain maps for each factor. Additionally, activation maps for individual words were generated and used for representational similarity analyses (RSA), again controlling for non-semantic psycholinguistic variables. Results: Across all datasets, semantic RSA searchlight revealed several regions engaged in lexical semantic processing, including angular gyrus, inferior frontal gyrus, medial parietal cortex, medial superior frontal gyrus, parahippocampal sulcus, orbitofrontal cortex, posterior middle temporal gyrus (pMTG), and anterior superior temporal sulcus (aSTS). These peaks closely matched regions activated by the Word > Pseudoword contrast in dataset 3. Only two of those regions, pMTG and aSTS, overlapped the Language network at the group level. Individual resting-state parcellation (Yeo et al., 2011) in dataset 3 showed that lexical semantic regions overlapped mainly with Default Mode, Frontoparietal, and Language networks, in decreasing order. Network-level RSA showed significant lexical semantic coding in eight networks, strongest in DMN-A and DMN-C, followed by FPN-A, Language, FPN-B, DMN-B, Salience-A, and Salience-B. Activity in the Language network was negatively correlated with concreteness and positively correlated with sublexical factors, particularly trigram frequency. Word frequency correlated with activation in DMN-A and DMN-C. These patterns were consistently observed in individual participants in dataset 3. Discussion: These results indicate that other large-scale cortical networks, especially the DMN, contribute more to lexical semantic processing than the “Language network”. Because abstract words presented in isolation carry less determinate semantic information than concrete words, increased Language-network activity for abstract words may indicate that this network is tuned to non-semantic—possibly phonological—information. Supporting this idea, activity in the Language network was independently correlated with sub-lexical variables such as trigram frequency, number of letters, number of phonemes, and biphone frequency. In contrast, activity in DMN regions was strongly correlated with concreteness and word frequency but not with sub-lexical variables, supporting a role in lexical-semantic processing. DMN semantic activity may be missed by the localizer proposed by Fedorenko et al. because it includes an implicit resting-state baseline—which is known to activate the DMN—and because the pseudoword condition requires little attention to the stimuli, which probably leads to task-unrelated thoughts.
Topic Areas: Meaning: Lexical Semantics, Computational Approaches