A Post-Editing Dataset in the Legal Domain: Do we Underestimate Neural Machine Translation Quality? (2020)