Elamur Research
☰
home / papers / 2023 / Mapping Explicit and Implicit Discourse Relations between the RST-DT and the PDTB 3.0

Mapping Explicit and Implicit Discourse Relations between the RST-DT and the PDTB 3.0

What it is about

On the 365 Wall Street Journal articles annotated in both RST-DT (tree relations over elementary discourse units) and PDTB 3.0 (connective-anchored relations between argument spans), the authors align and map relations empirically. Each PDTB argument is aligned with the RST segment that maximises character overlap with fewest extra characters (85% of continuous arguments align perfectly, 50% one-to-one with a single EDU; discontinuous ones only 47%). A PDTB relation is then mapped to the lowest RST relation covering both aligned segments, checked by the strong nuclearity principle (following nucleus paths to see that each argument lies on its segment's nucleus); relations embedded inside one EDU or reaching a multinuclear node are left unmapped. 80% of relations align (implicit 86.7%, explicit 73.7%, the latter hurt by discontinuous arguments). Explicit relations map more unambiguously to RST labels than implicit ones.

Ideas

For Mova

languages/en/36-discourse-relations.md §3–4 (RST and PDTB), §11.6 (eRST in GUM not yet used for measurement). Code: a scorer next to prag/src/fbench.rs reading GUM RST on dev (evaluation only). Measures: link precision by link type and by explicit vs implicit.

Library summary

The Research library's own summary of the paper, in our words — not the paper's text.

Written for the Research library (claude-opus-5-5) from the paper text. Original — aclanthology.org/2023.ranlp-1.39.pdf

Summary

The RST-DT and the PDTB are the two main discourse-annotated corpora, but they rest on different frameworks. They segment text differently and use different relation labels, so they are hard to combine when training discourse models. Earlier empirical mappings only covered PDTB 2.0. This paper builds the first empirical mapping between the RST-DT and PDTB 3.0.

The authors use the 365 Wall Street Journal articles that both corpora annotate and take PDTB relations as the starting point. Their algorithm works in two steps:

  • Alignment: each PDTB argument, including each part of a discontinuous argument, is matched to the RST-DT segment it overlaps most. Matches count as perfect when they differ only by punctuation or explicit connectives.
  • Mapping: each aligned PDTB relation, at the second sense level, is assigned the lowest RST-DT relation covering both aligned segments.

The method maps 80% of explicit and implicit relations, about 4 points more than prior PDTB 2.0 work. Implicit relations are easier to align, while aligned explicit relations map less ambiguously, likely because connectives make their meaning clearer. The mapping helps reuse both corpora together, for example to fine-tune large language models.

Key points

  • Corpus sizes:
    • RST-DT: 385 WSJ articles, 20,017 relations.
    • PDTB 3.0: 2,162 articles, 53,631 relations.
    • Overlap: 365 articles, containing 8,134 explicit and implicit PDTB relations. AltLex, AltLexC, EntRel, NoRel and hypophora were excluded.
  • Coverage: 6,510 relations (80.0%) were mapped, compared with 76% reported by Demberg et al. (2019) for PDTB 2.0.
  • Alignment quality:
    • Continuous PDTB segments: 85% aligned perfectly, and 50% were perfect one-argument-to-one-EDU matches.
    • Discontinuous segment constituents: only 47% aligned perfectly.
  • Example correspondences:
    • Explicit relations: TEMPORAL-SAME-TIME maps to Temporal.Synchronous 98% of the time, and LIST to Expansion.Conjunction 95%.
    • Implicit relations: PURPOSE maps to Contingency.Purpose 96%. ELABORATION-ADDITIONAL spreads across several senses.
  • Implementation and future work: the code uses Java and the Gate Embedded API and is available at github.com/CLaC-Lab/Mapping-Discourse-Relations. Planned extensions are covering AltLex and AltLexC relations and annotating the non-overlapping WSJ articles automatically.

Topics: Discourse analysis

Cite

The paper itself, not this card. BibTeX from the library's record of the work. DOI 10.26615/978-954-452-092-2_039

@inproceedings{costa2023mapping,
  title = {{Mapping Explicit and Implicit Discourse Relations between the RST-DT and the PDTB 3.0}},
  author = {Costa, Nelson Filipe and Sheikh, Nadia and Kosseim, Leila},
  booktitle = {Proceedings of the 14th International Conference on Recent Advances in Natural Language Processing},
  year = {2023},
  doi = {10.26615/978-954-452-092-2\_039},
  url = {https://doi.org/10.26615/978-954-452-092-2_039}
}

References

19 references, 7 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Harry Bunt and Rashmi Prasad. 2016. ISO DR-Core (ISO 24617-8): Core Concepts for the Annotation of Discourse Relations. In Proceedings of the 12th Joint ACL-ISO Workshop on Interoperable Semantic Annotation (ISA’16), pages 45–54, Portorož, Slovenia. European Language Resources Association (ELRA). find
  2. Lynn Carlson, Daniel Marcu, and Mary Ellen Okurovsky. 2001. Building a Discourse-Tagged Corpus in the Framework of Rhetorical Structure Theory. In Proceedings of the Second Annual Meeting of the Special Interest Group on Discourse and Dialogue (SIGdial’01). ↗ our card
  3. Lynn Carlson, Daniel Marcu, and Mary Ellen Okurowski. 2002. RST Discourse Treebank. LDC2002T07. Web Download. Philadelphia: Linguistic Data Consortium. find
  4. Ta-Chung Chi and Alexander Rudnicky. 2022. Structured Dialogue Discourse Parsing. In Proceedings of the 23rd Annual Meeting of the Special Interest Group on Discourse and Dialogue (SIGdial’22), pages 325–335, Edinburgh, UK. Association for Computational Linguistics. ↗
  5. Christian Chiarcos. 2014. Towards interoperable discourse annotation. Discourse features in the Ontologies of Linguistic Annotation. In Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC’14), pages 4569– 4577, Reykjavik, Iceland. European Language Resources Association (ELRA). find
  6. Vera Demberg, Merel CJ Scholman, and Fatemeh Torabi Asr. 2019. How compatible are our discourse annotation frameworks? Insights from mapping RST-DT and PDTB annotations. Dialogue & Discourse, 10(1):87–135. ↗
  7. Wei Liu and Michael Strube. 2023. Annotation-Inspired Implicit Discourse Relation Classification with Auxiliary Discourse Connective Generation. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics, pages 15696–15712, Toronto, Ontario, Canada. Association for Computational Linguistics. ↗
  8. William C Mann and Sandra A Thompson. 1988. Rhetorical Structure Theory: Toward a functional theory of text organization. Text-interdisciplinary Journal for the Study of Discourse, 8(3):243–281. ↗ our card
  9. Daniel Marcu. 2000. The Theory and Practice of Discourse Parsing and Summarization. MIT Press, Cambridge, MA, USA. ↗
  10. Eleni Miltsakaki, Rashmi Prasad, Aravind Joshi, and Bonnie Webber. 2004. The Penn Discourse Treebank. In Proceedings of the Fourth International Conference on Language Resources and Evaluation (LREC’04), pages 2237–2240, Lisbon, Portugal. European Language Resources Association (ELRA). ↗ our card
  11. Lucie Polakova, Jiřı́ Mı́rovský, and Pavlı́na Synková. 2017. Signalling implicit relations: A PDTB - RST comparison. Dialogue & Discourse, 8(2):225–248. ↗
  12. Rashmi Prasad, Nikhil Dinesh, Alan Lee, Eleni Milt-sakaki, Livio Robaldo, Aravind Joshi, and Bonnie Webber. 2008. The Penn Discourse TreeBank 2.0. In Proceedings of the Sixth International Conference on Language Resources and Evaluation (LREC’08), pages 2961–2968, Marrakech, Morocco. European Language Resources Association (ELRA). ↗ our card
  13. Rashmi Prasad, Eleni Miltsakaki, Nikhil Dinesh, Alan Lee, Aravind Joshi, Livio Robaldo, and Bonnie Webber. 2007. The Penn Discourse Treebank 2.0 Annotation Manual. Technical report, University of Pennsylvania. ↗ our card
  14. Rashmi Prasad, Eleni Miltsakaki, Nikhil Dinesh, Alan Lee, Aravind Joshi, and Bonnie Webber. 2006. The Penn Discourse Treebank 1.0 Annotation Manual. Technical report, University of Pennsylvania. find our card
  15. Rashmi Prasad, Bonnie Webber, Alan Lee, and Aravind Joshi. 2019. Penn Discourse Treebank Version 3.0. LDC2019T05. Web Download. Philadelphia: Linguistic Data Consortium. find
  16. Ted JM Sanders, Vera Demberg, Jet Hoek, Merel CJ Scholman, Fatemeh Torabi Asr, Sandrine Zufferey, and Jacqueline Evers-Vermeul. 2021. Unifying dimensions in coherence relations: How various annotation frameworks are related. Corpus Linguistics and Linguistic Theory, 17(1):1–71. find
  17. Symon Stevens-Guille, Aleksandre Maskharashvili, Xintong Li, and Michael White. 2022. Generating Discourse Connectives with Pre-trained Language Models: Conditioning on Discourse Relations Helps Reconstruct the PDTB. In Proceedings of the 23rd Annual Meeting of the Special Interest Group on Discourse and Dialogue (SIGdial’22), pages 500–515, Edinburgh, UK. Association for Computational Linguistics. ↗
  18. Maite Taboada and William C Mann. 2006. Rhetorical Structure Theory: Looking back and moving ahead. Discourse Studies, 8(3):423–459. ↗
  19. Bonnie Webber, Rashmi Prasad, Alan Lee, and Aravind Joshi. 2019. The Penn Discourse Treebank 3.0 Annotation Manual. Technical report, University of Pennsylvania. find our card