Testing The Usefulness Of Rst And More General Representations For Discourse Analysis Across Domains And Applications