The corpus RSTmulti contains English and German texts which have two separate Rhetorical Structure Theory annotations, for the purpose of studying disagreement in the context of discourse structure.
Repository with the dataset and scripts used in the The 19th Linguistic Annotation Workshop (LAW 2025) paper: Disagreements in analyses of rhetorical text structure: A new dataset and first analyses
Our doubly annotated data can be found in the data/ folder.
The files beginning with maz-* are from PCC*, with more information in the paper: Discourse Parsing for German with new RST Corpora (Shahmohammadi & Stede, KONVENS 2024).
The files beginning with pcc-* as well as impfenpro.rs3 and olympiacon.rs3 were doubly-annotated for the LAW paper. More information on the orignal PCC corpus can be found here.
The files beginning with UNSC-* are previously unpublished doubly-annotated files from the UNSC-RST corpus. More information can be found in the paper: Rhetorical Strategies in the UN Security Council: Rhetorical Structure Theory and Conflicts (Zaczynska & Stede, SIGDIAL 2024), as well as in the repository for the UP Multilayer UNSC Corpus.
The other files (which end with either -a2, -b1, or -or), are previously unpublished APA-RST files. More information can be found in the paper: APA-RST: A Text Simplification Corpus with RST Annotations (Hewett, CODI 2023).
The RST-DT is available from the Linguistic Data Consortium. We used the following files in our analysis: wsj_0615, wsj_0624, wsj_0630, wsj_0639, wsj_0651, wsj_0669, wsj_0684, wsj_1100, wsj_1102, wsj_1114, wsj_1117, wsj_1123, wsj_1129, wsj_1132, wsj_1141, wsj_1153, wsj_1168, wsj_1304, wsj_1314, wsj_1358, wsj_1924, wsj_1998, wsj_2303, wsj_2328, wsj_2349, wsj_2367.
The folder tace_output/ contains the output from RSTTace.
If you would like to use your own data, download Tace and parse your files accordingly.
The folder scripts/ contains the scripts which are used to create the CSV files, which contain the categories which we use in our paper (interchangeable relations, etc.).
To run the scripts, first download the requirements as outlined in requirements.txt.
The CSV files can then be created as follows:
python create_categories.py
If using your own data, or if you want to change the name of the output files, adapt the following arguments (run python create_categories.py -h for more details):
-tace_path
-corpus_path
-output_file
More information on the files can be found in our paper. If you use any of the data please cite our paper:
Freya Hewett and Manfred Stede. Disagreements in analyses of rhetorical text structure: A new dataset and first analyses. In Proceedings of the 19th Linguistic Annotation Workshop (LAW) at ACL. Vienna, 2025. (to appear).
Shield:
This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.
