Given a certain base of "prior beliefs"(as given by innate universal grammar), what is the second step of Bayesian inference learning?