No abstract

The study we report in this article addresses the results of comparing the rhetorical trees from two different languages carried out by two annotators starting from the Rhetorical Structure Theory (RST). Furthermore, we investigate the methodology for a suitable evaluation, both quantitative and qualitative, of these trees. Our corpus contains abstracts of medical research articles written both in Spanish and Basque, and extracted from Gaceta Médica de Bilbao ('Medical Journal of Bilbao'). The results demonstrate that almost half of the annotator disagreement is due to the use of translation strategies that notably affect rhetorical structures.

This article presents a discourse annotation methodology based on Rhetorical Structure Theory and an empirical study of annotating a corpus of specialized medical texts in Basque. The annotation process includes two phases: segmentation and annotation of rhetorical relations. Phase one entails an initial study which leads to establishing linguistic criteria for sentence-based segmentation; a second phase focuses on annotation of rhetorical relations. After establishing discourse segments and rhetorical relations, the annotation process is analyzed and evaluated by means of the method commonly used in RST (Marcu 2000). Inconsistencies detected in the evaluation method lead the authors to redefine some criteria of the evaluation method. As a result of this work, a small annotated Basque-language corpus is provided to scientific community.

In 2021, we organized the second iteration of a shared task dedicated to the underlying units used in discourse parsing across formalisms: the DISRPT Shared Task (Discourse Relation Parsing and Treebanking). Adding to the 2019 tasks on Elementary Discourse Unit Segmentation and Connective Detection, this iteration of the Shared Task included for the first time a track on discourse relation classification across three formalisms: RST, SDRT, and PDTB. In this paper we review the data included in the Shared Task, which covers nearly 3 million manually annotated tokens from 16 datasets in 11 languages, survey and compare submitted systems and report on system performance on each task for both annotated and plain-tokenized versions of the data.

No abstract