Brian Dupuis, Michael R. W. Dawson
Spatial learning and navigation have frequently been investigated using a reorientation task paradigm (Cheng, Cognition, 23(2), 149-78, 1986). However, implementing this task typically involves making tacit assumptions about the nature of spatial information. This has important theoretical consequences: Theories of reorientation typically focus on angles at corners as geometric cues and ignore information present at noncorner locations. We present a neural network model of reorientation that challenges these assumptions and use this model to generate predictions in a novel variant of the reorientation task. We test these predictions against human behavior in a virtual environment. Networks and humans alike exhibit reorientation behavior even when goal locations are not present at corners. Our simulated and our experimental results suggest that angles are processed in a manner more similar to features, acting as a focal point for reorientation, and that the mechanisms governing reorientation behavior may be inhibitory rather than excitatory.