Describir: Detecting and Exploiting User Familiarity in Natural Language Human-computer Dialogue