Abstract
High variability in category features creates challenges in learning and search for those categories. Compared to low variability categories, highly variable categories require more learning time to form a generalizable search template (Gaultier et al., 1998) and produce slower and less efficient search (Hout et al., 2015). This study investigated how search is impacted by category variability and frequency of exemplar exposure (i.e., prevalence) during category learning. We hypothesized that variability would serve as a moderator to the effect of learning prevalence on search processes. That is, variability will affect search for novel exemplars more so than for familiar exemplars viewed during learning. In one session, 52 participants learned to classify 12 rock categories that were diverse in feature variability. For some categories, participants were exposed to the same exemplars with varying prevalence. Following training, participants were tested on their ability to classify new exemplars. Finally, participants completed a search task where they were cued with the name of a rock category and searched for the rock among four distractor rocks. Target rocks included those from training with varying prevalence and never-before-seen exemplars. Eye tracking was used to measure attentional guidance and target verification. We found interactions between prevalence and variability in response time (RT) and target verification, but not attentional guidance. Variability impacted search for novel exemplars but not for exemplars previously viewed during training. Based on participants’ poorer search accuracy for high variability rocks (70% compared to 87% for low variability), we also analyzed testing accuracy as a measure for overall learning that may have influenced search. This measure of category learning was predictive of many aspects of search, primarily RT and target verification. These results not only demonstrate the effect of familiarity and variability on search, but also suggest that category learning is predictive of search performance.