Results of the classification experiment.

Left: pre-training; right: post-training. We counted how often every stimulus was identified as object B. These proportions were averaged across participants. Then a cumulative Gaussian was fit to the data. From this sigmoidal function the PSE and JND can be retrieved. Participants categorize A and B correctly before training, but the morphed objects are perceived as a smooth transition. After training the curve is significantly steeper, indicating that training lead to a sharpened categorical boundary. The blue dotted lines indicate the locations on the shape continuum of the 25%, 50% and 75% categorization values. Error bars  =  SEM.