ABSTRACT
In contrast with the preceding explanations, others have proposed that category effects in AD are an emergent property of the way concepts are represented within the semantic memory store. Gonnerman et al. (1997) do not deny that the location of lesions might be part of the explanation but they provide the following additional explanation of the pattern of loss they anticipate longitudinally in AD subjects. They propose that concepts are represented by a pattern of activation across a distribution of features. Some features (“intercorrelated features”) are frequently activated together and so develop strong bonds among themselves, providing collateral support if one feature is weakened. Wings, for instance, are frequently associated with flight and beaks. Other features (distinguishing features) are more linked to specific concepts, such as black stripes in the case of “zebras” or serrated metal edges in the case of “saws”. These distinguishing features are less affected by the coactivation of other features. The critical difference between biological and non-biological items is that biological items possess many more intercorrelated features than are encountered in non-biological items. Semantic degradation, according to Gonnerman et al. (1997) involves the random loss of features, both correlated and distinguishing. The early random loss of distinguishing features that are critical to the recognition of an object should be more likely to impair artefact identification than living object recognition because they receive little collateral support from other coactivated or correlated features. As the weight of loss accumulates, however, correlated activity will fail to support concepts and whole categories of biological items that are mutually dependent on each other will fall. The prediction of the model was therefore that there would initially be a living objects advantage followed by a non-living objects advantage, and this was partly supported by two case studies.
