From implicit learning to explicit representations