Probabilistic Head-Driven Parsing for Discourse Structure

We describe a data-driven approach to building interpretable discourse structures for appointment scheduling dialogues. We represent discourse structures as headed trees and model them with probabilistic head-driven parsing techniques. We show that dialogue-based features regarding turn-taking and domain specific goals have a large positive impact on performance. Our best model achieves an f score of 43.2% for labelled discourse relations and 67.9% for unlabelled ones, significantly beating a right-branching baseline that uses the most frequent relations.