Structural network modelling in cognitive science

Abstract

Many objects of study of cognitive sciences can be naturally represented in a form of networks. In network model nodes (cells, people, groups, words, categories, etc.) are defined by connections that they (don’t) have, form and lose. Thus, network model shifts emphasis from attributes of elements to their relations, evolution of those relations and – consequently – wholesome structure of the system. Traditional approach to network modelling in cognitive science has been conceptual approach (models of A.Collins and E.Loftus, J.Anderson, neural networks of D.Rumelhart, D.Hinton, etc). Main flaw of this approach is that its models represent not the structure of cognitive system per se, but authors’ ideas about that structure, they also often use hypothetical constructs, such as chunks and artificial neurons. At the start of XXI century another – structural – approach to network modelling has emerged. In contrast to conceptual model, structural model is a direct visualization of a data array that describes the system (e.g., MRI results, orthographic dictionary, social connections data, log of incoming and outgoing information packages, associative thesaurus, etc.). Topology of resulting network is analyzed by mathematical apparatus of computational network science. Based on results of this analysis the author can then produce hypotheses about evolutionary mechanisms that formed that structure and also dynamical consequences – how system’s structure influences cognitive processes. Our article describes main principles, notions and goals of structural network modelling. Brief history of network models is presented: from regular graphs to complex scale-free networks. Special emphasis is made on specificity of structural modelling in cognitive science.