I like how they have a way to test between different dialogue models. Very interesting! Will read the paper.