DReUD: Discourse Relations in Universal Dependencies

Abstract

We present a proposal for an annotation scheme and data representation of shallow discourse relations annotation in the Universal Dependencies (UD) framework, as a theoretically appropriate and also practically oriented extension of the established morphosyntactic analysis. We outline the design requirements for the annotation scheme, encompassing simplicity, comprehensibility, theoretical grounding, practical applicability and technical robustness, while accommodating the specific constraints of shallow discourse analysis. At the same time, we present a work-in-progress baseline version of DReUD (Discourse Relations in Universal Dependencies), a modular shallow discourse parser for Universal Dependencies as a command-line program, a web client and a REST API service for Czech and English, designed for a seamless and rapid integration of discourse relations analysis both in the theoretical research and in NLP applications.

Resources

Details