When Do Categories Change How the World Looks? It Depends on the Tools Used to Measure the Change
Benjamin Chin, Vassar College ’15 and Prof. Jan Andrews
The purpose of this study was to investigate inconsistencies in the reporting of observed learned categorical perception (CP) effects in the research literature. The two effects in question are compression and expansion. Compression refers to the phenomenon in which learning that two stimuli belong to the same category makes them appear perceptually more similar to each other, while expansion refers to the occurrence of the opposite effect where learning that two stimuli belong to different categories renders them more distinguishable. Numerous studies report finding either compression or expansion, or more rarely, both, but there have been no attempts to directly address what accounts for the different patterns of results.
We hypothesize that the aforementioned inconsistencies arise due to large variations in the methods and stimuli used to assess CP effects. To investigate this, we ran a basic learned CP experiment online using Mechanical Turk with three different assessment tasks and two sets of stimuli that differed in discriminability. These stimuli, which resembled micro-organisms, varied on one category-relevant dimension and one category-irrelevant dimension. We found both compression and expansion effects, but these varied according to both the task used (e.g., similarity rating vs. same-different judgment) and to a lesser extent, the degree of stimulus discriminability. These results confirm our initial hypotheses, particularly with regard to assessment task. The next step is to generate and test specific hypotheses about why these different patterns of results occur.