This paper discusses the empirical validation of annotation schemes proposed for discourse relations, when signaled explicitly by discourse connectives, through their application to texts in several languages. Considering a monolingual annotation scheme as a starting point, the paper explains the reasons for either specifying or generalizing some labels, illustrating them with a review of experiments in translation spotting of connectives. Then, an experiment with the PDTB scheme applied to five languages (EN, FR, DE, NL, and IT) shows how specification and generalization are put to work in order to build a scheme which has an improved empirical validity for several languages.
Zufferey, S., Degand, E., Popescu-Belis, A., & Sanders, T. (2012). Empirical validations of multilingual annotation schemes for discourse relations. In Bunt, Harry (ed.), Proceedings of the 8th Joint ISO-ACL SIGSEM Workshop on Interoperable Semantic Annotation (pp. 77-84).