Information-based methods for evaluating the semantics of automatically generated test items.