The role of inhibition in a spreading-activation model of language production. II. The simulational perspective

This is the second part of a project aimed at establishing the role of lateral inhibition in spreading-activation models of language production. While the first part (Berg &; Schade, 1992) reviewed the psycholinguistic evidence for inhibition and established that indeed inhibition takes place, this paper addresses the issue from the implementational perspective. A series of computer simulations are reported in which spreading-activation models with or without inhibition are directly contrasted. The performance of the competing models is evaluated against two sets of criteria. On the more technical side, the “heat death” and the selection problems have to be solved. That is, the target must have a high and the nontarget units a low level of activation. Additionally, there are certain empirical effects from the production literature which may serve as test cases for deciding between the rival models. The results of the simulations indicate that only the inhibition-based model proves capable of meeting all challenges. All inhibitionless variants fail in one or more respects. It is further shown that the need for the inhibitory component increases with the size of the processing network and that one well-known production model can do without lateral inhibition only at the expense of introducing a concept which resembles inhibition in its effect but cannot replace it.