Isabella DeStefano¹ (), Timothy Brady¹; ¹University of California San Diego

When multiple stimuli are perceived or remembered, separated in space or time, reproductions of one item tend to be biased towards or away from the other item(s). Previous work has proposed that these biases could arise from assuming that items belong to the same or different objects (Chetverikov, 2023). Here we use a bayesian model of concept generalization (Tenenbaum & Griffiths, 2001) in conjunction with a signal detection framework (Schurgin et al., 2020) to demonstrate that attractive and repulsive biases arise as a natural consequence of assuming items belong to either the same or a different ‘concept’. Under this model, concepts provide groupings over stimuli (i.e., for numbers, concepts might be “20-30”); generalization is governed by how likely stimuli are to arise from the same underlying concept; and similarity between stimuli arises from generalization. When concepts are assumed to reflect ranges of different sizes across stimuli space (as in Shepard, 1989), this model accurately predicts full error distributions. When a same/different assumption is incorporated into this model, unique patterns of responses arise: for repulsion, responses are skewed away from the other stimulus with the true stimulus as the median; for attraction, responses are symmetric with the median shifted away from the true stimulus. The magnitude of bias can be modulated via manipulating the maximum ranges of concepts considered. Additionally, by defining heterogeneous concepts (e.g., for color, greater ranges within yellows than across the yellow-orange boundary), stimulus-specific biases can emerge. This model suggests that attractive and repulsive biases are not unique to vision, but instead apply broadly; consistent with this, we provide preliminary empirical evidence of conceptual rule-based repulsion using symbolic number. Overall, understanding this putatively visual phenomena under a framework of concept generalization broadens our understanding of the cause and nature of such biases.