---
title: "Argument Mining with Structured SVMs and RNNs — Niculae 2017"
description: "The authors release CDCP (Cornell eRulemaking Corpus), a set of annotated user comments on… 25 references, cited by 62 papers here."
authors: ["Vlad Niculae", "Joonsuk Park", "Claire Cardie"]
year: 2017
venue: "Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)"
source: "https://aclanthology.org/P17-1091/"
doi: "10.18653/v1/p17-1091"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/niculae-2017-argument-mining-structured-svms-rnns/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/niculae-2017-argument-mining-structured-svms-rnns/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2017](https://md.research.elamur.ai/free/en/papers/2017/page/3/) / Argument Mining with Structured SVMs and RNNs

# Argument Mining with Structured SVMs and RNNs

**Authors:** [Vlad Niculae](https://md.research.elamur.ai/free/en/authors/vlad-niculae/), [Joonsuk Park](https://md.research.elamur.ai/free/en/authors/joonsuk-park/) and [Claire Cardie](https://md.research.elamur.ai/free/en/authors/claire-cardie/) · **Year:** 2017 · **Venue:** [Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)](https://md.research.elamur.ai/free/en/venues/acl/) ([ACL 2017](https://md.research.elamur.ai/free/en/venues/acl/2017/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/P17-1091/) · [PDF](https://aclanthology.org/P17-1091.pdf) · [DOI 10.18653/v1/p17-1091](https://doi.org/10.18653/v1/p17-1091) · [OpenAlex](https://openalex.org/W2609722168)

**Citations:** 98 citations (OpenAlex, 2026-10-14) · [62](https://md.research.elamur.ai/free/en/papers/niculae-2017-argument-mining-structured-svms-rnns/cited-by/) in our library, of the 141,840 papers whose bibliographies we read

## Summary
The paper addresses argument mining when support relations between argumentative units do not have to form trees. Earlier systems assumed tree or forest structures, but web discussions are often messier. The authors release CDCP (Cornell eRulemaking Corpus), a set of annotated user comments on Consumer Debt Collection Practices rule proposals from regulationroom.org. In CDCP, links form general directed graphs, and over 20% of comments contain non-tree configurations.

They cast the task as joint inference in a factor graph that does two things at once: it classifies each proposition's type and predicts whether each directed link exists. The graph can include higher-order link factors, such as chains, siblings and co-parents, along with hard structural constraints like transitivity or trees. Inference uses AD3. The scoring functions are parametrized either by linear structured SVMs or by RNNs trained with a structured loss.

On both CDCP and the UKP argumentative essays v2 corpus, the structured models beat unstructured baselines. On UKP they set a new best link-prediction result. The code is released as Marseille.

## Key points
- **CDCP corpus size and labels:** 731 comments, about 4,700 propositions and 88k words.
  - Propositions have five types: POLICY, VALUE, FACT, TESTIMONY and REFERENCE.
  - Links are labeled reason or evidence, but they are treated as binary in the experiments.
- **Link sparsity in CDCP:** only about 3% of the 43k candidate pairs are linked, and 28% of comments have no links at all.
  - By comparison, 17% of candidate pairs are linked in UKP, which has 402 essays and ~6,100 propositions.
- **Annotation agreement (Krippendorff's α):** 64.8% for proposition types and 44.1% for links.
  - The released data has the transitive closure of links applied and nested propositions removed, with code provided to reproduce these steps.
- **RNN training loss matters:** RNNs did poorly when trained with plain classification losses. With the structured loss, they became competitive with feature-engineered structured SVMs.
- **Higher-order factors on UKP:** across the 856 co-parent configurations found in UKP's test data, they lifted F1 from 45% (baseline) and 47% (basic structured SVM) to 53% (full structured SVM).
  - On CDCP, F1 on siblings and co-parents stayed below 10%.

## Interesting ideas
- **Joint modeling instead of post-hoc combination.** Unit labels and relations are modeled together, and the model learns how compatible they are. This replaces combining separate classifiers with a hand-tuned ILP score, and the same idea could carry over to any task that labels nodes and edges together, such as relation extraction or discourse parsing.
- **Structure as swappable constraint factors.** Tree, transitivity or no constraint can be plugged in as parameter-free factors in one framework (via AD3). This lets a single model serve corpora with different annotation conventions.
- **Neural scorers inside a structured objective.** A neural scorer can fall short under local losses yet become competitive once trained against the global structured objective. This is worth trying whenever local classifiers underperform on structured outputs.
- **Error analysis beyond F1.** Besides F1, the authors compare how many links each model predicts against how precise those links are. In one case, a strict baseline predicted 796 links at 16% precision, while the structured RNN predicted 52 links at 33% precision. This kind of check reveals whether a model is over- or under-predicting.
- **Releasing raw and preprocessed data together.** The corpus is shipped without the preprocessing applied, plus scripts that reproduce it. This keeps harder phenomena such as nested units available for future work.

*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

- [Cornell University](https://md.research.elamur.ai/free/en/universities/cornell-university/) — University, Ithaca, United States · 89 works · 58,422 citations
- [Williams College](https://md.research.elamur.ai/free/en/universities/williams-college/) — University, Williamstown, United States · 2 works · 395 citations

📚 Shelves (5): [🖥️ 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 · [🤔 Philosophy](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/philosophy/) · 211 works

## Cite

Cite as: Vlad Niculae, Joonsuk Park and Claire Cardie. 2017. Argument Mining with Structured SVMs and RNNs. Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).

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

```bibtex
@inproceedings{niculae2017argument,
  title = {{Argument Mining with Structured SVMs and RNNs}},
  author = {Vlad Niculae, Joonsuk Park and Claire Cardie},
  booktitle = {Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2017},
  doi = {10.18653/v1/p17-1091},
  url = {https://doi.org/10.18653/v1/p17-1091}
}
```

- [← previous Stronger Baselines for Trustable Results in Neural Machine Translation](https://md.research.elamur.ai/free/en/papers/denkowski-2017-stronger-baselines-trustable-results-neural/)
- [next → Frame-Semantic Parsing with Softmax-Margin Segmental RNNs and a Syntactic Scaffold](https://md.research.elamur.ai/free/en/papers/swayamdipta-2017-frame-semantic-parsing-softmax-margin/)
- [↑ back to papers of 2017](https://md.research.elamur.ai/free/en/papers/2017/page/3/)

## References

25 references, 8 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. André FT Martins, Miguel B Almeida, and Noah A Smith. 2013. [Turning on the Turbo: Fast third-order non-projective Turbo Parsers](https://md.research.elamur.ai/free/en/papers/martins-2013-turning-turbo-fast-third-order/). In Proceedings of ACL.
2. Stergos Afantenos and Nicholas Asher. 2014. Counter-argumentation and discourse: A case study. In Proceedings of ArgNLP. [find](https://scholar.google.com/scholar?q=Stergos+Afantenos+and+Nicholas+Asher%2E+2014%2E+Counter%2Dargumentation+and+discourse%3A+A+case+study%2E+In+Proceedings+of+ArgNLP%2E)
3. Mathieu Blondel, Masakazu Ishihata, Akinori Fujino, and Naonori Ueda. 2016. Polynomial networks and factorization machines: New insights and efficient training algorithms. In Proceedings of ICML. [find](https://scholar.google.com/scholar?q=Mathieu+Blondel%2C+Masakazu+Ishihata%2C+Akinori+Fujino%2C+and+Naonori+Ueda%2E+2016%2E+Polynomial+networks+and+factorization+machines%3A+New+insights+and+efficient+training+algorithms%2E+In+Proceedings+of+ICML%2E)
4. André FT Martins, Mário AT Figueiredo, Pedro MQ Aguiar, Noah A Smith, and Eric P Xing. 2015. AD3: Alternating directions dual decomposition for MAP inference in graphical models. Journal of Machine Learning Research 16:495–545. [find](https://scholar.google.com/scholar?q=Andre%CC%81+FT+Martins%2C+Ma%CC%81rio+AT+Figueiredo%2C+Pedro+MQ+Aguiar%2C+Noah+A+Smith%2C+and+Eric+P+Xing%2E+2015%2E+AD3%3A+Alternating+directions+dual+decomposition+for+MAP+inference+in+graphical+models%2E+Journal+of+Machine+)
5. Mathieu Blondel and Fabian Pedregosa. 2016. Lightning: large-scale linear classification, regression and ranking in Python. https://doi.org/10.5281/zenodo.200504. — [doi:10.5281/zenodo.200504](https://doi.org/10.5281/zenodo.200504) [↗](https://doi.org/10.5281/zenodo.200504)
6. Ryan McDonald, Fernando Pereira, Kiril Ribarov, and Jan Hajič. 2005. [Non-projective dependency parsing using spanning tree algorithms](https://md.research.elamur.ai/free/en/papers/mcdonald-2005-non-projective-dependency-parsing-spanning/). In Proceedings of EMNLP.
7. Hamish Cunningham, Diana Maynard, Kalina Bontcheva, Valentin Tablan, Niraj Aswani, Ian Roberts, Genevieve Gorrell, Adam Funk, Angus Roberts, Danica Damljanovic, Thomas Heitz, Mark A. Greenwood, Horacio Saggion, Johann Petrak, Yaoyong Li, and Wim Peters. 2011. Text Processing with GATE (Version 6). [find](https://scholar.google.com/scholar?q=Hamish+Cunningham%2C+Diana+Maynard%2C+Kalina+Bontcheva%2C+Valentin+Tablan%2C+Niraj+Aswani%2C+Ian+Roberts%2C+Genevieve+Gorrell%2C+Adam+Funk%2C+Angus+Roberts%2C+Danica+Damljanovic%2C+Thomas+Heitz%2C+Mark+A%2E+Greenwood%2C+Horaci)
8. Ofer Meshi, Mehrdad Mahdavi, Adrian Weller, and David Sontag. 2016. Train and test tightness of LP relaxations in structured prediction. In Proceedings of ICML. [find](https://scholar.google.com/scholar?q=Ofer+Meshi%2C+Mehrdad+Mahdavi%2C+Adrian+Weller%2C+and+David+Sontag%2E+2016%2E+Train+and+test+tightness+of+LP+relaxations+in+structured+prediction%2E+In+Proceedings+of+ICML%2E)
9. Andreas C Müller and Sven Behnke. 2014. PyStruct: learning structured prediction in Python. Journal of Machine Learning Research 15(1):2055–2060. [find](https://scholar.google.com/scholar?q=Andreas+C+Mu%CC%88ller+and+Sven+Behnke%2E+2014%2E+PyStruct%3A+learning+structured+prediction+in+Python%2E+Journal+of+Machine+Learning+Research+15%281%29%3A2055%E2%80%932060%2E)
10. Aaron Defazio, Francis Bach, and Simon Lacoste-Julien. 2014. SAGA: A fast incremental gradient method with support for non-strongly convex composite objectives. In Proceedings of NIPS. [find](https://scholar.google.com/scholar?q=Aaron+Defazio%2C+Francis+Bach%2C+and+Simon+Lacoste%2DJulien%2E+2014%2E+SAGA%3A+A+fast+incremental+gradient+method+with+support+for+non%2Dstrongly+convex+composite+objectives%2E+In+Proceedings+of+NIPS%2E)
11. [Graham Neubig, Chris Dyer, Yoav Goldberg, Austin Matthews, Waleed Ammar, Antonios Anastasopoulos, Miguel Ballesteros, David Chiang, Daniel Clothiaux, Trevor Cohn, Kevin Duh, Manaal Faruqui, Cynthia Gan, Dan Garrette, Yangfeng Ji, Lingpeng Kong, Adhiguna Kuncoro, Gaurav Kumar, Chaitanya Malaviya, Pau…](https://md.research.elamur.ai/free/en/papers/neubig-2017-dynet-dynamic-neural-network-toolkit/)
12. Ivan Habernal and Iryna Gurevych. 2016. [Argumentation mining in user-generated web discourse](https://md.research.elamur.ai/free/en/papers/habernal-2017-argumentation-mining-user-generated-web/). Computational Linguistics .
13. Sepp Hochreiter and Jürgen Schmidhuber. 1997. Long short-term memory. Neural Computation 9(8):1735–1780. [find](https://scholar.google.com/scholar?q=Sepp+Hochreiter+and+Ju%CC%88rgen+Schmidhuber%2E+1997%2E+Long+short%2Dterm+memory%2E+Neural+Computation+9%288%29%3A1735%E2%80%931780%2E)
14. Arzoo Katiyar and Claire Cardie. 2016. Investigating LSTMs for joint extraction of opinion entities and relations. In Proceedings of ACL. [find](https://scholar.google.com/scholar?q=Arzoo+Katiyar+and+Claire+Cardie%2E+2016%2E+Investigating+LSTMs+for+joint+extraction+of+opinion+entities+and+relations%2E+In+Proceedings+of+ACL%2E)
15. Joonsuk Park, Cheryl Blake, and Claire Cardie. 2015a. Toward machine-assisted participation in eRulemaking: An argumentation model of evaluability. In Proceedings of ICAIL. [find](https://scholar.google.com/scholar?q=Joonsuk+Park%2C+Cheryl+Blake%2C+and+Claire+Cardie%2E+2015a%2E+Toward+machine%2Dassisted+participation+in+eRulemaking%3A+An+argumentation+model+of+evaluability%2E+In+Proceedings+of+ICAIL%2E)
16. Eliyahu Kiperwasser and Yoav Goldberg. 2016. [Simple and accurate dependency parsing using bidirectional LSTM feature representations](https://md.research.elamur.ai/free/en/papers/kiperwasser-2016-simple-accurate-dependency-parsing-bidirectional/). arXiv:1603.04351 preprint.
17. Joonsuk Park, Arzoo Katiyar, and Bishan Yang. 2015b. Conditional random fields for identifying appropriate types of support for propositions in online user comments. In Proceedings of the 2nd Workshop on Argumentation Mining. Association for Computational Linguistics, Denver, CO, pages 39–44. [find](https://scholar.google.com/scholar?q=Joonsuk+Park%2C+Arzoo+Katiyar%2C+and+Bishan+Yang%2E+2015b%2E+Conditional+random+fields+for+identifying+appropriate+types+of+support+for+propositions+in+online+user+comments%2E+In+Proceedings+of+the+2nd+Workshop)
18. Klaus Krippendorff. 1980. Content Analysis: An Introduction to Its Methodology. Commtext. Sage. [find](https://scholar.google.com/scholar?q=Klaus+Krippendorff%2E+1980%2E+Content+Analysis%3A+An+Introduction+to+Its+Methodology%2E+Commtext%2E+Sage%2E)
19. Frank R Kschischang, Brendan J Frey, and H-A Loeliger. 2001. Factor graphs and the sum-product algorithm. IEEE Transactions on Information Theory 47(2):498–519. [find](https://scholar.google.com/scholar?q=Frank+R+Kschischang%2C+Brendan+J+Frey%2C+and+H%2DA+Loeliger%2E+2001%2E+Factor+graphs+and+the+sum%2Dproduct+algorithm%2E+IEEE+Transactions+on+Information+Theory+47%282%29%3A498%E2%80%93519%2E)
20. Andreas Peldszus and Manfred Stede. 2015. [Joint prediction in MST-style discourse parsing for argumentation mining](https://md.research.elamur.ai/free/en/papers/peldszus-2015-joint-prediction-mst-style-discourse/). In Proceedings of EMNLP.
21. Jeffrey Pennington, Richard Socher, and Christopher D 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 EMNLP.
22. Peter Potash, Alexey Romanov, and Anna Rumshisky. 2016. Here’s my point: Argumentation mining with pointer networks. arXiv:1612.08994 preprint. — [arXiv:1612.08994](https://arxiv.org/abs/1612.08994) [find](https://scholar.google.com/scholar?q=Peter+Potash%2C+Alexey+Romanov%2C+and+Anna+Rumshisky%2E+2016%2E+Here%E2%80%99s+my+point%3A+Argumentation+mining+with+pointer+networks%2E+arXiv%3A1612%2E08994+preprint%2E+%E2%80%94+%5BarXiv%3A1612%2E08994%5D%28https%3A%2F%2Farxiv%2Eorg%2Fabs%2F1612%2E08994%29)
23. Christian Stab and Iryna Gurevych. 2016. [Parsing argumentation structures in persuasive essays](https://md.research.elamur.ai/free/en/papers/stab-2017-parsing-argumentation-structures-persuasive-essays/). arXiv:1604.07370 preprint.
24. Ben Taskar, Carlos Guestrin, and Daphne Koller. 2004. Max-margin Markov networks. In Proceedings of NIPS. [find](https://scholar.google.com/scholar?q=Ben+Taskar%2C+Carlos+Guestrin%2C+and+Daphne+Koller%2E+2004%2E+Max%2Dmargin+Markov+networks%2E+In+Proceedings+of+NIPS%2E)
25. Ioannis Tsochantaridis, Thorsten Joachims, Thomas Hofmann, and Yasemin Altun. 2005. Large margin methods for structured and interdependent output variables. Journal of Machine Learning Research 6(Sep):1453–1484. [find](https://scholar.google.com/scholar?q=Ioannis+Tsochantaridis%2C+Thorsten+Joachims%2C+Thomas+Hofmann%2C+and+Yasemin+Altun%2E+2005%2E+Large+margin+methods+for+structured+and+interdependent+output+variables%2E+Journal+of+Machine+Learning+Research+6%28Sep%29)

*Updated 24 October 2026*