Is Perceptual Similarity the underlying mechanism of categorical perception?
Summary
Based on previous research by Hartendorp, Van der Stigchel, Burnett, Jellema, Eilers & Postma (2010), morphed objects are perceived as their dominant category, which is known as categorical perception (CP). Subsequently, they found higher perceptual similarity ratings for CP morph series compared to non-CP series. In the current study, we try to predict CP from similarity ratings. To do so, in addition to the existing Hartendorp dataset, a new stimulus-dataset is developed and tested to be able to generalize the CP/similarity effect. In experiment 1, we conducted a verification task, showing that almost all stimuli were representative for their category, except for three stimuli. In Experiment 2, a free-naming task was conducted using the two datasets. We also recorded eye-tracking data during this task, to gain insight in focus patterns when detecting and categorizing an object.
Our similarity experiment revealed that similarity ratings did not differ with those in the Hartendorp study. Also, we found overlap between CP series in the Hartendorp study and current experiment with respect to the Hartendorp dataset. However, we did not find a correlation between perceptual similarity and CP series. We conclude that, based on this study, similarity does not seem to be the basic assumption of CP.
Concerning the eye tracking data, we found a strong correlation between focus and free-naming responses. This indicates that within-category, viewing patterns are consistent. When the category boundary is passed, focus also shifts to another area of interest (AOI). Based on these findings, we can conclude that human object recognition is preceded by efficient viewing patterns.