Compositionality in Perception

Perception is a fundamental source of information about the empirical world. But in what form is “the testimony of the senses” delivered ? In this talk, I show how we can answer this question in a systematic, empirically constrained, and philosophically illuminating way. As a case study, I draw on vision science research on how the bounding contours of objects are “coded” in mid-level vision in order to develop a compositional semantics for such contour representations. Second, I show how this semantic theory can illuminate recent debates over whether perception is fundamentally language-like (Quilty-Dunn, Porot, Mandelbaum, 2022 ; Cavanagh, 2021) or image-like (Block, 2023). The account that I develop implies that visual representations of object contours have “domain-specific form” of the sort that is paradigmatic of diagrams and maps but uncharacteristic of sentences and logical formulas. Finally, I discuss the epistemic significance of having such domain-specific form, which I argue functions to regiment and insulate against certain possibilities for error.