A Proposed Study A.1 Project Title Cognitive Approaches to Learning in a Hybrid Semantic Network: An Examination of Localist and Dual Localist–Distributed Representations and their Role in Concept Formation

Semantic networks are structures that represent knowledge using patterns of interconnected nodes and links; typically, the nodes correspond to entities and concepts, while the links define how they are related. In the field of artificial intelligence, these networks have been used to build expert systems, and for natural language processing; due to their intuitive nature, and their scale-free structure (Steyvers and Tenenbaum, 2005), they have also been used to implement efficient associative memory systems, and context-dependent forms of reasoning. Within the cognitive sciences, semantic networks have been used to model the organisation of knowledge in memory (Collins and Quillian, 1969), to account for the structures and mechanisms of human working memory (Cowan, 1999), and in studies of phenomena including associative recall, priming and episode blending (Grinberg and Kokinov, 2003). They have even been used in the study of analogies and analogical reasoning (Holyoak and Thagard, 1989; Kokinov, 1994; Kokinov and Petrov, 2001). The networks ’ versatility is due, at least in part, to their diversity. In his treatise on this subject, John Sowa (1992) identified five basic types of semantic networks: definitional, assertional, implicational, executable and learning; included in the last of these categories were the artificial neural networks (ANNs). While semantic and neural networks are both graphical forms of knowledge representation, there ar