Figure 5

Distributed functional cortical networks associated with each individual semantic component. Each cluster’s profile contains diverse information including the cortical distribution of the electrodes contributing to that cluster (1), distribution over single subjects to show that all reported clusters included multiple subjects (2), distribution over cortical areas (3), distribution over the five semantic components (3), activation time course of the cluster (4) and frames that trigger and deactivate the cluster (5). Cortical maps are the surface-based projections of the similarity of each electrode in the cluster to the cluster exemplar. The similarity is measured as Pearson correlation of the \(\beta\)-weights across all electrodes. The \(\beta\)-weights are vectors of regression coefficients over 50 semantic components produced by the ridge linear regression fit. Distributions over subjects (inner pie charts), per cluster, show a ratio of all electrodes that came from each single subject to the total number of electrodes. None of the reported clusters were subject-specific, meaning that no more that 30% of electrodes per cluster came from a single subject. Distributions over cortical labels (middle pie charts) are color-coded and only five groups of labels are highlighted: frontal, sensorimotor, parietal, temporal and occipital regions. The proportion of labels within each color-coded region is also informative, such that in Cluster 1, for example, the largest contributing label is the fusiform gyrus, which is a temporal region. It is not additionally color-coded, but one can see that among the temporal regions there was only one with the largest contribution (which corresponds to the fusiform gyrus). Other times, the contribution can be more equally distributed over multiple regions with the same color-coded labels. Distribution over the semantic components (outer pie charts) was calculated by performing a signed difference test on the top 10% of frames associated with peaks in the cluster’s activation time course and the bottom 10% of frames associated with dips in the cluster’s activation time course. Per cluster, the signed difference test (two-sided Wilcoxon ranked test) was performed by comparing values along each semantic component between ‘peak frames’ and ‘dip frames’. The pie charts represent the test statistic, significant at \(p<0.001\) (Bonferroni corrected for the number of clusters and semantic components), adjusted for the decreasing percentage of explained variance from the first to the fifth semantic component (see “Methods” for details). The cluster activation time courses show the dot product of the semantic components and the \(\beta\)-weights of the exemplar of each cluster. Shading represents the standard error of the mean calculated on the dot product using \(\beta\)-weights of all electrodes of the cluster. Examples of frames associated with peaks in the cluster’s activation are displayed above the activation time course. Examples of frames associated with dips in the cluster’s activation are shown below the activation time course.