Annotating the Semantic Consistency of Speech Recognition Hypotheses