ViSyRelPer · Perceptual foundations of relational thinking

Perceptual foundations of relational thinking

The ability to represent relations, the properties holding between things, is regarded as a distinctive feature that can account for the complexity and uniqueness of the human mind. Yet, we do not know how relational representations come about in the human mind/brain. Are relations represented discretely and explicitly, separately from the objects participating in them? Or are they represented as basic properties of the objects in the scope of the relation? Relational representations have primarily been studied in the context of higher-level cognitive functions such as language and reasoning. Only recently, scholars have begun to investigate their perceptual underpinnings, yielding a new research field, which exploits the constructs and methods of vision science to investigate whether and how we see (i.e., encode) relations holding between objects, just like we see features such as colour, motion, shape and so on. This initial work on relation perception has shown that the early stages of the representation of relations between objects in visual scenes are encoded in visual areas. Visual areas would host specialized structures for representing objects as well as their relations, such as support (A is on B), containment (A is in B), and other relational aspect of the world such as causality (A hits B that starts moving) and social interaction (A does something to/with B). The working hypothesis of ViSyRelPer is that the visual system computes high-level relations, categorizing configurations of multiple objects into distinct relational types, independently from the objects that constitute the relation. On this hypothesis, in vision, depictions of a "tiger in a bath" and "a flower in a vase" will be recognized as instances of the same relation “in” (containment), despite their perceptual dissimilarities. In addressing this hypothesis, ViSyRelPer asks how far vision can go in representing the visual world, particularly whether it can afford high-level representations of relations that generalize across different situations that do not share the same visual featues. The overarching goal of ViSyRelPer is to study the role of visual perception in the representation of relations. This research will contribute to determine the computational goals of human vision, and to develop a model that accounts for how relational contents come about in the human mind/brain.

Data: CORDIS, © European Union