Anna K. Lawrance¹ (), Johannes Schultz-Coulon², Brett D. Roads³, James W. Tanaka¹; ¹University of Victoria, ²Maastricht University, ³University College London

Experts identify objects in their domain of expertise faster, more accurately, and at a more specific level of abstraction than novices (Tanaka & Taylor, 1993). Whereas a novice sees the yellow bird flitting in the bush, the expert instantly recognizes this object as a Cape May Warbler. Although substantial research has explored the behavioral and neural correlates of the expert’s downward shift in recognition, less is known about how their mental structure mediates such speeded identification. In our experiment, 75 participants were trained to identify ten images of Cape May, Magnolia, Prairie, and Townsend warblers to a criterion of 90% accuracy. Before and after training, category structure was assessed with PsiZ. PsiZ (https://psiz.readthedocs.io) is a machine learning package that generates a multi-dimensional category representation (i.e., psychological embedding) based on the participant’s judgments of image similarity. The key finding was that training produced profound changes in category structure. Specifically, warbler images belonging to different species became significantly more differentiated, while warbler images of the same species became more compact; hence, training produced between-category expansion and within-category compression. What is the relationship between category structure and category performance? Once participants completed their post-training PsiZ judgments, participants were given a recognition test where they were asked to identify the species of novel Warbler images and images used in training. Based on their recognition accuracy, the group of top 25% and bottom 25% performers were identified. The psychological embeddings were then inferred for each group and compared. The PsiZ results revealed significant differentiation between species, particularly among the lower quartile participants following training. Moreover, after training, top-performers showed denser within-species clusters than lower performers. Collectively, subordinate-level training produced significant category restructuring. Further, the quality of this reorganization appears to play a functional role in one’s expert recognition performance.