Location information in mature word representations.

Mean generalization accuracy of the linear classifier on input and hidden layer word representations, after training of the model (Standard error of the mean in parenthesis). Patterns corresponded to horizontally and vertically presented words, and were classified into 2 or 6 locations. For the 6 target locations classifier, percent classification to adjacent locations (Adjac.) are provided along with an index of adjacency effect (Adj. Index).