Connectionist Models in Spatial Cognition

Abstract

This chapter reviews neural network approaches to the study of spatial cognition and spatial language with a focus on the representations and processes that are used by humans for encoding, storing, accessing, and referencing spatial knowledge. Such processes are used for recall of spatial information, for navigation through space, for spatial decision making, and for generating spatial descriptions. Connectionist models for representing the structure of cognitive maps and understanding the language of spatial relations are discussed in detail, using the representation adopted by the model to evaluate the usefulness of each approach. In addition, the functional differences within neural network models, such as distributed versus local representations, are discussed. Finally, an agenda for future development of connectionist models, including the exploration of alternative network structures and data types, is proposed.