Abstract
Multi-voxel pattern analysis was used to determine whether a support vector machine (SVM) could distinguish between grammatically correct sentences and sentences with a morphological violation related to tense-marking in 7-year-old children (N = 92) undergoing functional magnetic resonance imaging. SVM classifier accuracy was examined in regions of interest (ROIs) related to phonological processing (inferior frontal gyrus, pars opercularis and posterior superior temporal gyrus) and semantic processing (inferior frontal gyrus, pars triangularis and posterior middle temporal gyrus) determined with independent functional localizers using sound and meaning judgments. In these ROIs, the classifier performed around chance indicating that grammatical sentences and sentences with a morphological violation have a similar underlying representation for the whole sample. We also investigated correlations of standardized measures of language skill with classifier accuracy in the ROIs and in the whole brain. Whole brain analyses showed lower skill was associated with higher classification accuracy in bilateral brain areas implicated in semantic processing, in articulation and phonological processing and in verbal working memory. Overall, children with lower language skill seem to use widespread mechanisms in both hemispheres for sentences comprehension, possibly indicating increasing use of rehearsal mechanisms and semantic compensation for detecting morphological errors.