Statistical learning and cognitive constraints on rule induction. An entropy model

What triggers the inductive leap from memorizing items and statistical regularities to abstract rules? We propose an innovative information-theoretic model for both learning statistical regularities and generalizing to new input. Our model predicts that rule induction is an encoding mechanism triggered by the discrepancy between input complexity (entropy) and human brain?s encoding limitations (channel capacity). Previous research proposed two different mechanisms underlying rule induction: statistical learning (Safran et al., 1996; Aslin & Newport, 2012) and abstract rule learning (Marcus et al, 1999). In our model, less input complexity facilitates memorization of items and item-based generalization, while a higher complexity overloading channel capacity drives category-based generalization (i.e. reduce the features that items can be coded for by grouping them in categories). In two artificial grammar experiments with adults we probed the effect of input complexity on rule induction. Results showed that when the input complexity increases, the tendency to infer abstract rules increases gradually. We are running two experiments (adults) to investigate the individual differences (channel capacity) that modulate rule induction. Results are expected to show that individuals with a lower memory capacity and a higher general pattern-recognition capacity are more likely to infer abstract rules, even from lower input complexity.