I will present a general cognitive architecture which formalizes computational relations between the mind and the brain. Principles of neural computation yield an emergent property that constitutes a new principle of mental organization: mental processes compute representations that are optimal. Representations in a given cognitive component are optimal with respect to ‘soft’ constraints that characterize the world as cognized in that component. In addition to optimization, neural computation provides another key process: quantization. This process yields a fundamental property of higher cognition: mental representations are discrete, combinatorial structures. The architecture is illustrated in a domain particularly challenging for mental/neural integration: grammar.

Paul Smolensky is a professor in the Department of Cognitive Science at Johns Hopkins University.