Towards a formal view of corrective feedback