Category-Specific Semantic Deficits in Focal and Widespread Brain Damage: A Computational Account

Abstract

Category-specific semantic impairments have been explained in terms of preferential damage to different types of features (e.g., perceptual vs. functional). This account is compatible with cases in which the impairments were the result of relatively focal lesions, as in herpes encephalitis. Recently, however, there have been reports of category-specific impairments associated with Alzheimer's disease, in which there is more widespread, patchy damage. We present experiments with a connectionist model that show how ficategory-specificfl impairments can arise in cases of both localized and wide-spread damage; in this model, types of features are topographically organized, but specific categories are not. These effects mainly depend on differences between categories in the distribution of correlated features. The model's predictions about degree of impairment on natural kinds and artifacts over the course of semantic deterioration are shown to be consistent with existing patient data. The model shows how the probabilistic nature of damage in Alzheimer's disease interacts with the structure of semantic memory to yield different patterns of impairment between patients and categories over time.