The prediction-irrelevance problem in grammar learning
Rainer Spiegel, F. W. Jones, I.P.L. McLaren · 2002
The Elman recurrent network (SRN) has been considered a good model of language acquisition including grammar learning. Until recently, however, it was reported that it cannot master the prediction-irrelevance criterion, which, if true, would clearly limit its success of being an adequate neural network in this context. The paper shows that the SRN can deal with prediction-irrelevant information.