---
title: "A Simple Global Neural Discourse Parser — Zhou 2020"
description: "The authors build a CKY-style chart parser that scores spans, relations and nuclearity with small neural networks over learned… 19 references."
authors: ["Yichu Zhou", "Omri Koshorek", "Vivek Srikumar", "Jonathan Berant"]
year: 2020
venue: "arXiv:2009.01312"
source: "https://arxiv.org/abs/2009.01312"
doi: "10.48550/arxiv.2009.01312"
arxiv: "2009.01312"
licence: "arXiv non-exclusive distribution licence (read only, no redistribution)"
licence_url: "http://arxiv.org/licenses/nonexclusive-distrib/1.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/zhou-2020-simple-global-neural-discourse-parser/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/zhou-2020-simple-global-neural-discourse-parser/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2020](https://md.research.elamur.ai/free/en/papers/2020/page/9/) / A Simple Global Neural Discourse Parser

# A Simple Global Neural Discourse Parser

**Authors:** Yichu Zhou, [Omri Koshorek](https://md.research.elamur.ai/free/en/authors/omri-koshorek/), [Vivek Srikumar](https://md.research.elamur.ai/free/en/authors/vivek-srikumar/) and [Jonathan Berant](https://md.research.elamur.ai/free/en/authors/jonathan-berant/) · **Year:** 2020 · **Venue:** arXiv:2009.01312 · **Licence:** [arXiv non-exclusive distribution licence (read only, no redistribution)](http://arxiv.org/licenses/nonexclusive-distrib/1.0/)

**Links:** [Publisher's page](https://arxiv.org/abs/2009.01312) · [PDF](https://arxiv.org/pdf/2009.01312) · [DOI 10.48550/arxiv.2009.01312](https://doi.org/10.48550/arxiv.2009.01312) · [arXiv 2009.01312](https://arxiv.org/abs/2009.01312) · [OpenAlex](https://openalex.org/W3081670140)

**Citations:** 0 citations (OpenAlex, 2026-10-14)

## Summary
The paper addresses document-level discourse parsing in the RST framework. Most existing parsers are greedy and depend on hand-built features or extra resources, because exact global search is expensive: a node's relation and nuclearity labels depend on where its children are split. The authors build a CKY-style chart parser that scores spans, relations and nuclearity with small neural networks over learned EDU representations, using no engineered features. The key move is a "partial independence" assumption. The split point is chosen from span and subtree scores without looking at labels. The labels are then predicted given that split. This lowers decoding cost while keeping a one-way dependency, so the model stays more coupled than the fully independent approach of Li et al. (2016). Training uses a structured max-margin loss with decoding inside the loop. On the RST Discourse Treebank, the parser is the strongest global model and is competitive with leading greedy systems. This suggests that global inference can stand in for heavy feature engineering.

## Key points
- Data: RST Discourse Treebank, 385 documents (347 train, 38 test), with 47 training documents held out as a dev set. The 78 relation types are collapsed into 19 classes, and gold (manual) EDU segmentation is used.
- Encoder: each word is the concatenation of frozen GloVe, ELMo (with tuned mixing weights and scale) and trainable spaCy POS embeddings, fed into a bi-LSTM. An EDU is represented by the forward and backward states at its first and last words. Two-layer feedforward networks produce S_span, S_rel and S_nuc.
- Decoding complexity drops from O(Gn³) for joint label-and-split search to O(n²(n+G)). Training uses a structured-SVM loss with Hamming distance and loss-augmented decoding, optimized with Adam.
- Macro F1 (Span/Nuclearity/Relation) is 87.2/74.9/61.9 for partial independence, against 85.7/72.2/56.7 for complete independence and 86.6/73.8/61.6 for a shift-reduce baseline with the same encoder. Human agreement is 89.6/78.3/66.7.
- On micro F1 the model reaches 84.5/71.1/57.5. Only Yu et al. (2018) scores higher (85.5/73.1/59.9), and the authors attribute that to external resources such as a Bi-Affine parser.

## Interesting ideas
- Pick the structure (split) from structural scores first, then predict labels conditioned on it, keeping a one-way dependency instead of full joint search. This keeps global decoding affordable when the label set is large.
- Measure "missing predictions": during training, count how often the approximate decoder returns a tree that scores below the gold tree. This separates search failures from modelling failures.
- Compare a greedy and a global parser that share exactly the same encoder, so any gain can be credited to inference alone.
- Represent a span cheaply by concatenating the bidirectional encoder states at its two boundaries, with no span-internal pooling.
- Train with a margin scaled by Hamming distance to the gold tree, using loss-augmented decoding. Wrong trees that differ in many spans are then penalized more strongly.

## For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

- **I1** (tree): gparse could choose arcs globally and then assign UD relations given each chosen head, so the large deprel inventory stops multiplying the decoding cost.
- **I2** (learn): evaluation runs could log how often each parser's decoder misses a higher-scoring gold tree, which shows whether to fix the search or the model.
- **I3** (tree): comparing the perceptron parser with gparse on identical input features would show whether improvements come from search or from representations.
- **I4** (frames): clause or argument spans in prag could get vectors this way for scoring discourse links between clauses, while the span boundaries stay explicit and inspectable.
- **I5** (learn): the perceptron parser's updates could be made cost-sensitive in the same way, so silver trees with many disagreeing arcs push the weights harder than near-misses.

*Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.*

## Institutions

- [Tel Aviv University](https://md.research.elamur.ai/free/en/universities/tel-aviv-university/) — University, Tel Aviv, Israel · 67 works · 10,006 citations
- [University of Utah](https://md.research.elamur.ai/free/en/universities/university-of-utah/) — University, Salt Lake City, United States · 29 works · 5,781 citations

📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 9,086 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 8,889 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 9,296 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 9,206 works

🧭 Topics (1): [🧵 Discourse analysis](https://md.research.elamur.ai/free/en/topics/discourse-analysis/) · 394 works

## Cite

Cite as: Yichu Zhou, Omri Koshorek, Vivek Srikumar and Jonathan Berant. 2020. A Simple Global Neural Discourse Parser. arXiv:2009.01312.

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@misc{zhou2020simple,
  title = {{A Simple Global Neural Discourse Parser}},
  author = {Yichu Zhou, Omri Koshorek, Vivek Srikumar and Jonathan Berant},
  year = {2020},
  doi = {10.48550/arxiv.2009.01312},
  eprint = {2009.01312},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2009.01312}
}
```

- [← previous A Rule-Based Parsing System for Encoding Kennicott's Collation of the Hebrew Bible](https://md.research.elamur.ai/free/en/papers/bambaci-2020-rule-based-parsing-system-encoding/)
- [next → A Survey of Qualitative Error Analysis for Neural Machine Translation Systems](https://md.research.elamur.ai/free/en/papers/diaz-2020-survey-qualitative-error-analysis-neural/)
- [↑ back to papers of 2020](https://md.research.elamur.ai/free/en/papers/2020/page/9/)

## References

19 references, 14 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Bengio. 2015. [Neural machine translation by jointly learning to align and translate](https://md.research.elamur.ai/free/en/papers/bahdanau-2014-neural-machine-translation-jointly-learning/). In 3rd International Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings. [↗](https://doi.org/10.48550/arxiv.1409.0473)
2. Chloé Braud, Maximin Coavoux, and Anders Søgaard. 2017. [Cross-lingual RST discourse parsing](https://md.research.elamur.ai/free/en/papers/braud-2017-cross-lingual-rst-discourse-parsing/). In Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics, EACL 2017, Valencia, Spain, April 3-7, 2017, Volume 1: Long Papers, pages 292–304. [↗](https://doi.org/10.48550/arxiv.1701.02946)
3. Chloé Braud, Barbara Plank, and Anders Søgaard. 2016. [Multi-view and multi-task training of rst discourse parsers](https://md.research.elamur.ai/free/en/papers/braud-2016-multi-view-multi-task-training/). In Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 1903–1913.
4. Lynn Carlson, Daniel Marcu, and Mary Ellen Okurovsky. 2001. [Building a discourse-tagged corpus in the framework of rhetorical structure theory](https://md.research.elamur.ai/free/en/papers/carlson-2001-building-discourse-tagged-corpus-framework/). In Proceedings of the Second SIGdial Workshop on Discourse and Dialogue. [↗](https://doi.org/10.3115/1118078.1118083)
5. Vanessa Wei Feng and Graeme Hirst. 2014. [A linear-time bottom-up discourse parser with constraints and post-editing](https://md.research.elamur.ai/free/en/papers/feng-2014-linear-time-bottom-up-discourse/). In Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), volume 1, pages 511–521.
6. Katsuhiko Hayashi, Tsutomu Hirao, and Masaaki Nagata. 2016. Empirical comparison of dependency conversions for rst discourse trees. In Proceedings of the 17th Annual Meeting of the Special Interest Group on Discourse and Dialogue, pages 128–136. [find](https://scholar.google.com/scholar?q=Katsuhiko+Hayashi%2C+Tsutomu+Hirao%2C+and+Masaaki+Nagata%2E+2016%2E+Empirical+comparison+of+dependency+conversions+for+rst+discourse+trees%2E+In+Proceedings+of+the+17th+Annual+Meeting+of+the+Special+Interest+Gr)
7. Yangfeng Ji and Jacob Eisenstein. 2014. [Representation learning for text-level discourse parsing](https://md.research.elamur.ai/free/en/papers/ji-2014-representation-learning-text-level-discourse/). In Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), volume 1, pages 13–24.
8. Shafiq Joty, Giuseppe Carenini, and Raymond T Ng. 2015. [Codra: A novel discriminative framework for rhetorical analysis](https://md.research.elamur.ai/free/en/papers/joty-2015-codra-novel-discriminative-framework-rhetorical/). Computational Linguistics, 41(3):385–435. [↗](https://doi.org/10.1162/coli_a_00226)
9. Diederik P. Kingma and Jimmy Ba. 2015. Adam: A method for stochastic optimization. In 3rd International Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings. [↗](https://doi.org/10.48550/arxiv.1412.6980)
10. Qi Li, Tianshi Li, and Baobao Chang. 2016. [Discourse parsing with attention-based hierarchical neural networks](https://md.research.elamur.ai/free/en/papers/li-2016-discourse-parsing-attention-based-hierarchical/). In Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 362–371.
11. William C Mann and Sandra A Thompson. 1988. [Rhetorical structure theory: Toward a functional theory of text organization](https://md.research.elamur.ai/free/en/papers/mann-1988-rhetorical-structure-theory-functional/). Text-Interdisciplinary Journal for the Study of Discourse, 8(3):243–281. [↗](https://doi.org/10.1515/text.1.1988.8.3.243)
12. Daniel Marcu. 2000. The theory and practice of discourse parsing and summarization. MIT press. [↗](https://doi.org/10.7551/mitpress/6754.001.0001)
13. Mathieu Morey, Philippe Muller, and Nicholas Asher. 2017. [How much progress have we made on rst discourse parsing? a replication study of recent results on the rst-dt](https://md.research.elamur.ai/free/en/papers/morey-2017-how-much-progress-have-we/). In Conference on Empirical Methods on Natural Language Processing (EMNLP 2017), pages pp–1330.
14. Jeffrey Pennington, Richard Socher, and Christopher Manning. 2014. [Glove: Global vectors for word representation](https://md.research.elamur.ai/free/en/papers/pennington-2014-glove-global-vectors-word-representation/). In Proceedings of the 2014 conference on empirical methods in natural language processing (EMNLP), pages 1532–1543. [↗](https://doi.org/10.3115/v1/d14-1162)
15. Matthew Peters, Mark Neumann, Mohit Iyyer, Matt Gardner, Christopher Clark, Kenton Lee, and Luke Zettlemoyer. 2018. [Deep contextualized word representations](https://md.research.elamur.ai/free/en/papers/peters-2018-deep-contextualized-word-representations/). In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers), pages 2227–2237.
16. Mitchell Stern, Jacob Andreas, and Dan Klein. 2017. [A minimal span-based neural constituency parser](https://md.research.elamur.ai/free/en/papers/stern-2017-minimal-span-based-neural-constituency/). In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, ACL 2017, Vancouver, Canada, July 30 - August 4, Volume 1: Long Papers, pages 818–827.
17. Mihai Surdeanu, Tom Hicks, and Marco Antonio Valenzuela-Escarcega. 2015. [Two practical rhetorical structure theory parsers](https://md.research.elamur.ai/free/en/papers/surdeanu-2015-two-practical-rhetorical-structure-theory/). In Proceedings of the 2015 conference of the North American chapter of the association for computational linguistics: Demonstrations, pages 1–5.
18. Ben Taskar, Vassil Chatalbashev, Daphne Koller, and Carlos Guestrin. 2005. Learning structured prediction models: A large margin approach. In Proceedings of the 22nd international conference on Machine learning, pages 896–903. ACM. [find](https://scholar.google.com/scholar?q=Ben+Taskar%2C+Vassil+Chatalbashev%2C+Daphne+Koller%2C+and+Carlos+Guestrin%2E+2005%2E+Learning+structured+prediction+models%3A+A+large+margin+approach%2E+In+Proceedings+of+the+22nd+international+conference+on+Machin)
19. Nan Yu, Meishan Zhang, and Guohong Fu. 2018. Transition-based neural rst parsing with implicit syntax features. In Proceedings of the 27th International Conference on Computational Linguistics, pages 559–570. [find](https://scholar.google.com/scholar?q=Nan+Yu%2C+Meishan+Zhang%2C+and+Guohong+Fu%2E+2018%2E+Transition%2Dbased+neural+rst+parsing+with+implicit+syntax+features%2E+In+Proceedings+of+the+27th+International+Conference+on+Computational+Linguistics%2C+pages+)

*Updated 24 October 2026*