Behavioral correlates of cortical semantic representations modeled by word vectors
Fig 4
Cortical mapping of prediction accuracy.
Participant-averaged prediction accuracy of voxelwise models (vector dimension = 1000) was mapped onto the cortical surface of a reference brain for each of fastText (A and B), GloVe (C and D), and word2vec (E and F) vectors and for each of movie sets 1 (A, C, and E) and 2 (B, D, and F). The prediction accuracy was averaged within each of the cortical regions that were anatomically segmented. Brighter colors in the surface maps indicate cortical regions that have higher prediction accuracy. We showed only regions with mean prediction accuracy above 0.11, which reaches a significance level (i.e., p = 0.05) of prediction accuracy after Bonferroni correction for multiple comparisons among 148 cortical regions (i.e., p = 0.0001 ~ 0.05/148). The five cortical regions with the highest mean prediction accuracy are numbered in a descending order separately for each type of word vectors and for each movie set. The names of these regions are shown in Table 1. LH, left hemisphere; RH, right hemisphere; A, anterior; P, posterior.