Delineating the neural bases of semantic memory, even for single entities (objects, features of objects, categories, actions, etc.) has been fraught with numerous difficulties, including variable definitions of semantic terms and several different models of the functional organization of semantic memory (see previous chapters). The development of new investigative techniques has been a major asset in detecting regions associated with semantic memory, but has not led to a consensus on semantic organization, since the results from these studies have frequently been discordant. For example, the results of activation studies (positron emission tomography, functional magnetic resonance imaging event-related potential – PET, fMRI, ERP, etc.), which demonstrate regions likely “involved” in performing a task, have yet to be fully integrated with the results of lesion-based studies that show regions “essential” for performing a task.

Historically, attempts at delineating the anatomic substrates of semantic memory have been guided by one of two general models or classes of models: parallel distributed representation (McClelland & Rumelhart, 1985 and Hinton, 1981) and center processing (Geschwind, 1965). It is clear, however, that neither of these models in their pure form explains adequately the growing body of data from anatomic and functional studies. In other words, the brain comprises neither a homogeneous network of equivalent neuronal elements that encode every aspect of a memory, nor circumscribed processing centers that encode all memory elements.