?
К вопросу о формировании набора отношений для корпуса с дискурсивной разметкой текста
The work discusses the problem of discourse annotation and the consequences of the relations set simplification for the sake of higher interannotator agreement. One of the theoretical approaches to discourse structure representation is the Rhetoric Structure Theory by William Mann and Sandra Thompson [1]. There is a set of rhetoric relations between discourse units that were suggested within this theory. Modern discourse treebanks, including the discourse annotated treebank for Russian (Ru-RSTreebank ), exploit this set of relations. Considering the conceptual rather than operational basis of relations definitions and the lack of strict grammatical tests for specifying relation types, the developers of these treebanks face the problem of low interannotator agreement. One of the ways to overcome this difficulty is to simplify the annotation scheme, e.g. to unite confusing relations in one relation. This option was chosen by the developers of the Ru-RSTreebank . We discuss the pitfalls of this approach...