Figure 3.

Can the model read novel words? Generalization performance of the model on the nonwords of Olson et al. [19] and Rastle & Coltheart [20]. The nonword sets were repeatedly tested during the course of learning to read (i.e. after every 50 000 word presentations).