Semantic Specificity in Spoken Dialogue Requests