Structured definitions of discourse relations in the Supracorpora Database of Connectives

The paper presents initial outcomes resulting from development of structured definitions of discourse relations based on novel classification principles and describes how these definitions are captured in the Supracorpora Database of Connectives (SCDB). The authors provide an overview of existing approaches to definition of discourse relations and propose novel principles for capturing structured definitions of discourse relations in the SCDB based on several aspects which include ($i$) the basic semantic operation that the logical-semantic relation (LSR) is based on: implication, relative timeline positioning, comparison, and correlation between general and specific, between an element and a set; ($ii$) the linguistic level that the LSR is established on: propositional level, utterance (illocutionary) level, and metalinguistic level; ($iii$) the polarity, i. e., whether the LSR is established directly between the provisions $p$ and $q$ featuring in the text or whether their negative correlates should also be considered in the relation interpretation; and ($i\nu$) the semantic and pragmatic features of the context. The paper provides some examples of such structured definitions. The structured definitions are captured within the SCDB by a set of interrelated tables. In addition, the “Family” table is introduced to offer information about conceptual closeness of some sets of classification features. The proposed structure allows researchers to access similarities and distinctions between various LSRs — as of today, this functionality is not implemented in any of the existing corpora that include annotation of discourse relations.