It is argued that while the hippocampus is not vital for all classes of configural learning, it is vital for a specific subclass of configural tasks called "structural learning." The defining feature of structural learning is that in addition to binding stimulus elements to make unique arrays (as in all configural learning), the relationship of these elements to each other, be it spatial or temporal, is specified. Direct evidence supporting the proposal that the hippocampus is required for structural learning comes from recent lesion studies with rats. While rats with hippocampal lesions were impaired at relearning a set of spatial structural problems, they showed no impairment when relearning two configural tasks (transverse patterning and the biconditional discrimination), neither of which required structural learning. Other support comes from surveys of spatial and temporal learning by amnesic patients with hippocampal damage, and from imaging studies of both humans and rats. While these studies offer consistent support for the structural hypothesis, very few provide a rigorous test as the tasks can often be solved by other strategies. For this reason, the present review details the design features of future stringent tests of the structural hypothesis.
(c) 2007 Wiley-Liss, Inc.