Of the three sentences that are misclassified in the simulation, two were used in the experiment. It appears that these are the very two sentences that were most misclassified by subjects as well, yielding 38 and 43% correct classification, respectively, meaning that subjects classified them ...