Understanding multimodal deixis with gaze and gesture in conversational interfaces