Abstract
The Language of Thought (LoT) hypothesis posits that mental representations are best understood as programme-like objects; indeed, "thoughts" share properties such as productivity and systematicity with programming languages. I tackle questions that arise from taking this hypothesis at face value and unfolding its predictions, from computational accounts to mechanistic implementation. First, zooming on humans' cognition of geometric shapes, I show that in all human groups tested (adults, children, congenitally blind), the perception of shapes is heavily influenced by geometric features. Then, I show using MEG and fMRI that the neural signature of these exact geometric properties is separate both in timing and localisation from typical visual processes. To generalize beyond quadrilaterals, I commit to a proposition for a generative language of shapes to account for the complexity of geometric shapes in humans, while implementing an algorithm for perception-as-program-inference. Finally, building on recent results in rodent neuroscience, I sketch a research programme and give preliminary results on a mechanistic understanding of how program-like representations might be implemented by populations of neurons.
