Visual illusions as a tool to hijack numerical perception: Disentangling nonsymbolic number from its continuous visual properties.

The past few years have witnessed a fervent theoretical debate about the exact visual mechanisms supporting nonsymbolic number processing. The idea that quantity information is extracted through a primitive visual segmentation algorithm has been challenged by recent models, which rather tap on low-level features confounded with numerosity (i.e., density, convex hull, or total area). Here we used an original manipulation based on visual illusions to disentangle whether visual numerosity processing operates over discrete units or rather over continuous variables. In particular, we generated a set of stimuli composed by open inducers (e.g., like a pac-man shape) that simulate physical connections with Kanizsa-like illusory contours (ICs). Test sets contained pairs of collinear open inducers items that prompted 0 IC, 2 IC, or 4 IC lines connecting 2 objects. Critically, low-level visual features were fully controlled across connectedness levels. We found a systematic underestimation as we increased the IC connections when participants had to select the larger between 2 sets of objects (Experiment 1) but not in the case of aligned closed inducers preventing illusory lines (Experiments 2A and 2B). We also found a systematic numerosity underestimation when both IC connections and continuous features (e.g., convex hull) were independently manipulated in test stimuli (Experiment 3). Finally, these results were shown to be task independent because the same effects of IC connections were replicated in an estimation task (Experiment 4). Taken together, our findings indicate that numerosity perception relies on basic visual-segmentation mechanisms, pointing out the need of new theoretical frameworks integrating both continuous and discrete perceptual number signals. (PsycInfo Database Record (c) 2021 APA, all rights reserved).