Interhemispheric Effects of Simulated Lesions in a Neural Model of Single-Word Reading

Neurocomputational models of the remote effects of focal brain damage

“…To make the results concrete and to illustrate how these models work, as a specific example, we consider the phoneme sequence generation (or ''word naming'') model, our first attempt to investigate computationally the factors influencing lateralization [45,46]. Its recurrently connected network is trained using supervised learning to take representations of 50 written three- [48] letter words (CAD, MOP, etc.)…”

“…Lateralization was measured as the difference between the output error when the left hemispheric region alone controlled the output versus when the right hemispheric region alone did. Using methods developed and described in detail elsewhere [46], a lateralization coefficient q was measured after training. Negative values of q indicate left lateralization of functionality while positive values indicate right lateralization.…”

“…Accordingly, several versions of each of these three models were subjected to systematic lesioning studies (a lesion is a region of damage) [46,[52][53][54]. Each lesion was introduced into an intact model by clamping a subset of elements in one hemispheric region to be nonfunctional.…”

“…Accordingly, we studied two neural models with excitatory callosal influences that are similar to those described in the previous section (Table 1) except that now subcortical regions with cross-midline inhibitory connections are present. As an example here, we again consider the phoneme sequence generation model [46,53].…”

Effects of callosal lesions in a model of letter perception

“…In the context of this "callosal dilemma" about the excitatory/inhibitory nature of callosal connections, a number of computational models, consisting of paired left and right cortical regions, have recently been developed to study hemispheric interactions and cerebral functional asymmetries (Cook, 1999;Cook & Beech, 1990;Jacobs & Kosslyn, 1994;Levitan & Reggia, 1999Reggia, Gittens, & Chhabra, 2000;Reggia, Goodall, & Shkuro, 1998;Ringo, Doty, Demeter, & Simard, 1994;Shevtsova & Reggia, 1999Shkuro, Glezer, & Reggia, 2000). Among other things, these neural network models have been very successful in demonstrating that a variety of underlying hemispheric asymmetries (such as asymmetries in region size, excitability, receptive field size, feedback intensity, or synaptic plasticity) can, in theory, lead to hemispheric specialization.…”