Cumulative semantic interference as learning

Abstract

When aphasic individuals name pictures in a blocked-cyclic naming paradigm, they produce more semantic and omission errors when the repeatedly named items come from a single semantic category, relative to when the items are from different categories, an effect known as cumulative semantic interference. This effect is magnified in Broca’s aphasics and increases as patients repeatedly cycle through the pictures in each block (Schnur, Schwartz, Brecher, & Hodgson, 2006). A similar phenomenon manifests in non-aphasic populations as comparatively increased naming latencies for pictures from a single category (e.g. Howard, Nickels, Coltheart, & Cole-Virtue, 2006; Schnur et al., 2006). We propose a model in which incremental learning occurs during naming and is realized as weight change within the lexical system (e.g. Gordon & Dell, 2003), thereby producing cumulative semantic interference. To test our hypothesis that incremental learning alone could produce interference, we built a two-layer neural network with a distributed semantic input layer and a lexical output layer (Fig. 1a). We simulated blocked-cyclic naming as repeated access of lexical representations through semantic features with continued application of the error-based learning algorithm by which we initially trained the network.