Associations and propositions: The case for a dual-process account of learning in humans