The Words Matter in Early Reading Acquisition: A Simulation Study

The development of word reading skills fundamentally depends on children's interactions with printed words. A significant amount of research in the psychological and educational sciences has been conducted on the learning processes that contribute to skilled word recognition in children and adults. However, little is known about how ensembles of words encountered during early learning affect learning outcomes over timescales larger than what can be accomplished in a controlled experimental setting. In order to move past this limitation, using an established connectionist learning architecture (Plaut, McClelland, Seidenberg, & Patterson, 1996; Seidenberg & McClelland, 1989) we examined how variation in word ensembles influence learning and generalization during early reading acquisition. We simulated learners (N = 30,000) exposed to different sets of monosyllabic printed words drawn randomly from children’s literature at a large scale. In order to examine how different ensembles might impact learning depending on the learner's ability to represent the relationship between letters and sounds, we also manipulated the representational capacity of models by systematically changing the number of hidden units, holding the ensemble of words constant. Across levels of representational capacity each model had "twins" which differed only in terms of the number of hidden units they possessed. Results demonstrate significant variation in learning outcomes in a set of generalization words as a function of the word, ensemble and representational capacity. These findings inform educational practices by highlighting the importance of deliberate word selection to support early development.