Điều này tiến triển đặc trưng xuất hiện dễ dàng interpretable nhất về ngữ nghĩa của một hệ thống phân cấp có cấu trúc, trong đó thông tin cụ thể được đại diện tại các chi của một nhánh "cây kiến thức" phân biệt cơ bản | John R. Hodges 78 Figure Distributed representation microfeature model illustrating penguin thin line owl dashed line and robin thick line . the disease they produce prototypical . horse for hippopotamus and for any other large animal or superordinate responses animal but only in very advanced cases are cross-category errors produced. This characteristic progression appears most readily interpretable in terms of a hierarchically structured semantic system in which specific information is represented at the extremities of a branching tree of knowledge. More fundamental distinctions such as the division of animate beings into land animals water creatures and birds are thought to be represented closer to the origin of the putative hierarchy with living versus nonliving things at the very top. The defining characteristics of higher levels are inherited by all lower points Collins Quillian 1969 . Such a model has intuitive appeal and the deficits of semantic dementia can be seen as a progressive pruning back of the semantic tree Warrington 1975 . An alternative account which we favor is based on the concept of microfeatures in a distributed connectionist network McClelland et al. 1995 McClelland Rumelhart 1985 . The basic idea is illustrated in figure . An advantage of such a model is that the low-level features of individual concepts need only be represented once while a hierarchical model requires distinctive features to be represented separately for every concept for which they are true . has a mane for both lion and horse . Category membership is then understood as an emergent property of the sharing of elements of these patterns between concepts and thus becomes a matter of degree another intuitively appealing property. A distributed feature network could predict preservation of superordinate at the expense of finer-grained knowledge as seen in semantic dementia because even in a network that had lost the representations of many individual attributes .
