Jessica Golding¹, Tianye Ma¹, Weiwei Zhang¹; ¹University of California Riverside

Facial expressions are represented with both continuous sensory fidelity and categorical semantic labeling. While classical models debate the primacy of these representations, recent evidence suggests that the balance between continuous and categorical encoding is malleable, influenced by top-down biases and individual differences. The present study investigated individual differences in working memory (WM) for facial expressions and their potential relationship with psychopathology. We collected mouse-tracking data from 32 participants during a continuous estimation task for facial expression. Stimuli consisted of morphed faces from the NimStim set, forming a continuous valence half-circle response scale from 150% Angry to 150% Happy. Participants recalled expressions on the continuous scale. Analysis of mouse trajectories revealed significant early mu-shifts toward negative valence, with final responses showing a robust negative bias. Next, we integrated the Category Adjustment Model (CAM, prototype-biased) into the Target Confusability Competition (TCC, linear similarity-based) and developed 10 candidate models to capture categorical and continuous WM representations. While half of the participants were best fit by the continuous TCC-Linear model, the other half exhibited performance favoring categorical encoding models. Crucially, these strategy differences tracked with variations in depression, anxiety, and schizotypy. We propose that while normative facial processing preserves graded fidelity (TCC), psychopathological traits may potentiate top-down categorical warping (CAM), specifically strengthening negative attractors. These findings unify psychophysical models with emotional bias, demonstrating that the "geometry" of memory is reshaped by the observer’s emotional state.