A modular attractor model of semantic access
This paper presents results from lesion experiments on a modular attractor neural network model of semantic access. Real picture data forms the basis of perceptual input to the model. An ultrametric attractor space is used to represent semantic memory and is implemented using a biologically plausible variant of the Hopfield model. Lesioned performance is observed to be in agreement with neuropsychological data. A local field analysis of the attractor states of semantic space forms a basis for interpreting these results.