eRST: A Signaled Graph Theory of Discourse Relations and Organization

TL;DR

This work introduces eRST, a signaled graph theory that extends RST by allowing tree-breaking, non-projective, and concurrent discourse relations anchored to token-level signals. It defines a formal three-part framework (primary tree, secondary edges, and signals), builds a large, publicly available eRST-annotated GUM corpus across genres, and delivers a parsing baseline combining existing state-of-the-art components for primary trees, connectives, morphosyntax, and coreference with a novel transformer-based predictor for secondary edges. Empirical results show strong performance for primary-tree parsing but highlight the ongoing challenge of reliably predicting secondary edges and signal anchors, underscoring the importance of correct primary structures for downstream tasks. The paper argues that eRST enhances explainability, enables richer relation extraction, and opens avenues for detailed attribution analysis and downstream applications, while providing data and tools to support further development and cross-lingual extension.

Abstract

In this article we present Enhanced Rhetorical Structure Theory (eRST), a new theoretical framework for computational discourse analysis, based on an expansion of Rhetorical Structure Theory (RST). The framework encompasses discourse relation graphs with tree-breaking, non-projective and concurrent relations, as well as implicit and explicit signals which give explainable rationales to our analyses. We survey shortcomings of RST and other existing frameworks, such as Segmented Discourse Representation Theory (SDRT), the Penn Discourse Treebank (PDTB) and Discourse Dependencies, and address these using constructs in the proposed theory. We provide annotation, search and visualization tools for data, and present and evaluate a freely available corpus of English annotated according to our framework, encompassing 12 spoken and written genres with over 200K tokens. Finally, we discuss automatic parsing, evaluation metrics and applications for data in our framework.

Figures (15)

• Figure 1: RST Fragment from GUM Zeldes2017. The most central point is the nucleus in [24], to which other units are direct or indirect satellites (manner and concession). Symmetrical relations such as list are multinuclear nodes ([26]-[27]).
• Figure 2: RST fragment from RST-DT: Satellites point to nuclei (e.g. [28] is a consequence of [27]) while the symmetrical sequence relation connects equally prominent nodes. [33] and [35] form a discontinuous same-unit.
• Figure 3: Fragment of an SDRT graph in the Glozz tool. The large blue discourse unit on the bottom has two incoming relations, contrast from the large blue unit at the top, and elaboration from the gray EDU with the text 'Inhibition de contact' at the top.
• Figure 4: PDTB annotation interface for the same fragment from Figure \ref{['fig:rstdt-ex']}. Two concurrent relations are recognized, corresponding to but and then respectively.
• Figure 5: Head-ordered DDS converted from the RST fragment in \ref{['fig:gum-rst-example']}.