Approaches to annotation of discourse relations in linguistic corpora

This paper examines the Supracorpora Database of Connectives (SCDB-Connectives) that is based on data from parallel corpora. The SCDB-Connectives provides structural and semantic annotation of Russian connectives and their translation correspondences in French (and, eventually, in other languages). The SCDB-Connectives annotation approach is compared to the latest developments in the area of annotation of discourse relations — the annotated corpus of discourse relations Penn Discourse Treebank (PDTB) and the proposed standard for annotation of semantic relations ISO 24617-8, some of the important differences are discussed. Penn Discourse Treebank and ISO 24617-8 support annotation of both explicit and implicit discourse relations while SCDB-Connectives only annotates explicit relations, i. e., those expressed by connectives. Furthermore, PDTB and ISO 24617-8 provide a superior framework for annotating text spans as relation arguments, which allows annotating attribution for these arguments, such as source and type of the linked propositions. In addition, ISO 24617-8 specifies argument roles for asymmetrical discourse relations. On the other hand, the principle advantage of the SCDB-Connectives is that it supports annotation of both connectives and their translation correspondences in parallel corpora, opening up new possibilities for contrastive studies. The SCDB-Connectives is based on a relational database rather than on the XML format, which helps to manage complex cross-linguistic data efficiently. Benefits of semantic annotation of connectives for both theoretical and practical purposes are also discussed.