Mathias Sablé-Meyer

Title:

Human Cognition of Geometric Shapes: A Window Into the Neural Representation of Abstract Concepts

Abstract:

Natural language is not the only hallmark of humans’ striking cognitive abilities. I propose that cognition involving geometric shapes requires a set of discrete, symbolic mental representations that act as a mental language; that perceiving a shape entails performing mental program induction: finding the shape’s shortest representation in this internal language. First, I show that all humans share a sense of geometric complexity, but that baboons lack this even after training. Artificial neural networks of object recognition fit baboons’ data, but explaining humans’ behavior requires using additional symbolic properties such as the presence of right angles. Then, I identify two distinct neural dynamics for the visual and symbolic strategies of shape perception using brain imaging methods, and I provide preliminary evidence for the existence of the symbolic strategy in infants. I then propose and test a mental generative language of geometric shapes, and derive more general rules that alternative proposal for such mental languages will have to satisfy. Finally, I will point toward a possible mechanistic implementation of models of this nature, and present early evidence in favor of this implementation.