Abstract
One driver of perceptual learning (PL) in complex domains is the development of selective extraction of information that distinguishes different categories (Gibson, 1969). Between-category comparisons have been shown to improve this component of PL (Kang & Pashler, 2011; Higgins & Ross, 2011), particularly when the compared items are similar (Dwyer & Vladeanu, 2009). Here, we ask whether adaptive methods that use learner performance can further enhance the benefits of comparisons. We used a face identification paradigm with 22 categories of faces, each containing five unique exemplars of one individual (Min, Kose, & Dugelay, 2014). In Experiment 1, training was structured such that on standard trials a single face image was presented and learners attempted to select the correct name. Adaptive schedules in the ARTS system (Mettler, Massey & Kellman, 2016) guided category spacing based on learner performance. In one condition, two sequential errors involving the same pair of categories led to an adaptively triggered comparison trial (ATC condition). On ATC trials, participants were presented with a name and instructed to choose between two exemplars from the confused categories before resuming standard trials. In Experiment 2, we compared an ATC condition to a control condition containing the same number of random comparison trials (exemplars from randomly selected categories). All participants learned to mastery criteria of accuracy and fluency and completed immediate and delayed (one-week) posttests. Efficiency scores -- defined as posttest accuracy divided by the number of learning trials invested -- were compared across conditions. In both experiments, the ATC condition required the fewest average trials to reach mastery and resulted in more efficient learning than the control condition. These results suggest that adaptively triggered comparisons enhance the efficiency of PL. Using learner performance to determine the contents and timing of comparison trials can be beneficial in optimizing perceptual category learning.