First steps towards an ISO standard for annotating discourse relations

This paper describes initial studies in the context of a new effort within ISO to design an international standard for the annotation of discourse with semantic relations that are important for its coherence, “discourse relations”. This effort takes the Penn Discourse Treebank (PDTB) as its starting point, and applies a methodology for defining semantic annotation languages which distinguishes an abstract syntax, defining annotation structures as set-theoretical constructs, a concrete syntax, that defines a reference XML-based format for representing annotation structures, and a formal semantics. A first attempt is described to formulate an abstract syntax and a concrete syntax for the annotation scheme underlying the PDTB. The abstract syntax clearly shows an overall structure for a general-purpose standard for annotating discourse relations, while the resulting concrete syntax is much more readable and semantically transparent than the original format. Moreover, some additional elements are introduced which have an optional status, making the proposed representation format compatible not only with the PDTB but also with other approaches.