Mean sentence decoding accuracy across participants (top left). Mean ± SEM drop in decoding accuracy induced by withholding semantic vectors associated with each grammatical element from the sentence decoding model for the highest scoring ROIs (right). ROIs are arranged in order of decreasing decoding accuracy (LSTS at the top left gives the strongest result). Significant results from each ROI from the grammatically reduced model analysis (right) are identified on a brain map (bottom left). ROI colors for this plot (bottom left) are arbitrary and used to delineate ROIs. Supplementary Figure 4 illustrates a companion post hoc analysis undertaken on anterior, mid and posterior subregions of LSTS, LSTG, and LMTG. Because different grammatical elements occurred with different frequencies across the sentences (and withholding elements consequently affected different subsets of sentences) different bars correspond to tests on different subsets of the experimental sentences. See main text for further details. The number of test pairs contributing to each comparison were: Sent-S: 28667 tests, Sent-V: 27733 tests, Sent-O: 22464 tests, Sent-IO: 6102 tests, Sent-Cop: 9568 tests, Sent-Adju: 14985 tests.