---
title: "Hedge Detection Using the RelHunter Approach (2010)"
description: "The paper addresses the CoNLL-2010 Shared Task (Closed Task 2), which asks a system to find… 16 references, cited by 2 papers here."
authors: ["Eraldo Fernandes", "Carlos Crestana", "Ruy Milidiú"]
year: 2010
venue: "Proceedings of the Fourteenth Conference on Computational Natural Language Learning – Shared Task"
source: "https://aclanthology.org/W10-3009/"
licence: "CC BY-NC-SA 3.0 (ACL Anthology, before 2016)"
licence_url: "https://creativecommons.org/licenses/by-nc-sa/3.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/fernandes-2010-hedge-detection-relhunter-approach/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/fernandes-2010-hedge-detection-relhunter-approach/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2010](https://md.research.elamur.ai/free/en/papers/2010/page/2/) / Hedge Detection Using the RelHunter Approach

# Hedge Detection Using the RelHunter Approach

**Authors:** [Eraldo Fernandes](https://md.research.elamur.ai/free/en/authors/eraldo-fernandes/), Carlos Crestana and [Ruy Milidiú](https://md.research.elamur.ai/free/en/authors/ruy-milidiu/) · **Year:** 2010 · **Venue:** [Proceedings of the Fourteenth Conference on Computational Natural Language Learning – Shared Task](https://md.research.elamur.ai/free/en/venues/conll/) ([CoNLL 2010](https://md.research.elamur.ai/free/en/venues/conll/2010/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY-NC-SA 3.0 (ACL Anthology, before 2016)](https://creativecommons.org/licenses/by-nc-sa/3.0/)

**Links:** [Publisher's page](https://aclanthology.org/W10-3009/) · [PDF](https://aclanthology.org/W10-3009.pdf) · [OpenAlex](https://openalex.org/W2252250653)

**Citations:** 7 citations (OpenAlex, 2026-10-14) · 2 in our library, of the 141,840 papers whose bibliographies we read

## Summary
The paper addresses the CoNLL-2010 Shared Task (Closed Task 2), which asks a system to find hedge cues, the words that signal uncertainty, together with the span of text each cue makes uncertain. The authors apply RelHunter, their general method for extracting structured information. RelHunter treats each hedge as a relation linking three entities: a cue chunk, a scope-start token and a scope-end token.

The pipeline works in three stages:
1. Separate token classifiers detect each entity type.
2. A constrained procedure combines the detected entities into candidate triples.
3. A binary classifier keeps the true triples and rejects the rest.

All four classifiers are trained with Entropy Guided Transformation Learning (ETL). ETL can model dependencies among examples in the same segment, which suits a setting where accepting one candidate rules out others. The data is the BioScope-based CoNLL corpus, enriched with stems, POS tags, chunks and clauses.

The official run ranked 7th. A refined end-scope baseline developed after the competition raised F-score to 54.05, which would have placed fourth. The work shows that splitting a structured task into simple local decisions plus a joint filtering step can be competitive.

## Key points
- **Data split:** The corpus (derived from BioScope) was divided into ABST (abstracts) and FULL (full papers).
  - For development, models were trained on ABST and tested on FULL. This is a cautious setup, because the official test set contains only full articles.
  - For the final evaluation, models were trained on ABST and FULL together.
- **Candidate constraints:** A triple is kept only if the scope start does not come after the scope end and the cue lies between them.
- **Features for relation recognition:**
  - Local features come from each entity. For a multi-token cue, the rightmost token's features are used.
  - Global features are computed over 12 sentence fragments: the regions preceding, within and following each of the three entities, and likewise the regions preceding, within and following the full candidate.
  - Every fragment receives an occurrence flag and a count for each element of interest (entity types, verbal chunks).
  - The resulting dataset has km local and 6r(k+1) global features.
- **Entity classifiers on the evaluation corpus:** ETL reached cue F 77.88, compared with 51.52 for the Cue Dictionary baseline. Scope-start F was 83.15 and scope-end F was 76.55, both measured with gold cues.
- **Overall hedge detection:**
  - With the END-X baseline: P 57.84, R 50.73, F 54.05.
  - With the plain END baseline (the official entry): F 41.63, rank 7.

## Interesting ideas
- Model a span structure as a relation over a few simple anchor entities, detect each anchor with its own local classifier, then judge whole candidates jointly; this keeps each step simple and inspectable.
- Use cheap structural constraints, such as ordering rules and containment of the cue, to prune the candidate set before classification, instead of scoring every possible combination.
- Global features of the form "does element X occur before, inside or after this candidate, and how often" give a relation classifier sentence-level context without needing a bag of words.
- A tiny targeted heuristic for one frequent error type (moving scope ends out of sentence-final parenthesized asides) lifted overall F by more than ten points.
- When developing, train on the easier domain (abstracts) and validate on the harder target domain (full papers), so that progress is not overestimated.

## For Mova

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

- **I1** (frames): Hedge or negation scope in prag could be found by tagging cues and scope boundaries, then accepting or rejecting explainable (cue, start, end) triples.
- **I2** (tree): UD subtree boundaries could serve as an even stronger filter, so that only spans forming valid tree fragments become scope candidates.
- **I3** (frames): Frame or role candidates could carry explicit flags and counts for verbs, clauses and cues in each region, keeping every decision traceable.
- **I4** (learn): Self-induced repair rules could focus on the largest systematic error classes, such as trailing parentheticals, and be measured before and after on held-out data.
- **I5** (learn): Evaluations could reserve a register-shifted dev split, for example tales versus school texts, to catch overfitting to one genre.

*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

- [Pontifícia Universidade Católica do Rio de Janeiro](https://md.research.elamur.ai/free/en/universities/pontificia-universidade-catolica-do-rio-de-janeiro/) — University, Rio de Janeiro, Brazil · 4 works · 239 citations

📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 9,085 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 8,888 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

## Cite

Cite as: Eraldo Fernandes, Carlos Crestana and Ruy Milidiú. 2010. Hedge Detection Using the RelHunter Approach. Proceedings of the Fourteenth Conference on Computational Natural Language Learning – Shared Task.

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

```bibtex
@inproceedings{fernandes2010hedge,
  title = {{Hedge Detection Using the RelHunter Approach}},
  author = {Eraldo Fernandes, Carlos Crestana and Ruy Milidiú},
  booktitle = {Proceedings of the Fourteenth Conference on Computational Natural Language Learning – Shared Task},
  year = {2010},
  url = {https://aclanthology.org/W10-3009/}
}
```

- [← previous Exploiting CCG Structures with Tree Kernels for Speculation Detection](https://md.research.elamur.ai/free/en/papers/sanchez-2010-exploiting-ccg-structures-tree-kernels/)
- [next → A Syntactic Lexicon for Arabic Verbs](https://md.research.elamur.ai/free/en/papers/loukil-2010-syntactic-lexicon-arabic-verbs/)
- [↑ back to papers of 2010](https://md.research.elamur.ai/free/en/papers/2010/page/2/)

## References

16 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Eric Brill. 1995. [Transformation-based error-driven learning and natural language processing: a case study in part-of-speech tagging](https://md.research.elamur.ai/free/en/papers/brill-1995-transformation-based-error-driven-learning/). Computational Linguistics, 21(4):543–565.
2. Xavier Carreras, Lluı́s Màrquez, Vasin Punyakanok, and Dan Roth. 2002. Learning and inference for clause identification. In Proceedings of the Thirteenth European Conference on Machine Learning, pages 35–47. [find](https://scholar.google.com/scholar?q=Xavier+Carreras%2C+Llu%C4%B1%CC%81s+Ma%CC%80rquez%2C+Vasin+Punyakanok%2C+and+Dan+Roth%2E+2002%2E+Learning+and+inference+for+clause+identification%2E+In+Proceedings+of+the+Thirteenth+European+Conference+on+Machine+Learning%2C+page)
3. Xavier Carreras, Lluı́s Màrquez, and Jorge Castro. 2005. Filtering-ranking perceptron learning for partial parsing. Machine Learning, 60(1–3):41–71. [↗](https://doi.org/10.1007/s10994-005-0917-x)
4. Cı́cero N. dos Santos and Ruy L. Milidiú, 2009. Foundations of Computational Intelligence, Volume 1: Learning and Approximation, volume 201 of Studies in Computational Intelligence, chapter Entropy Guided Transformation Learning, pages 159–184. Springer. [find](https://scholar.google.com/scholar?q=C%C4%B1%CC%81cero+N%2E+dos+Santos+and+Ruy+L%2E+Milidiu%CC%81%2C+2009%2E+Foundations+of+Computational+Intelligence%2C+Volume+1%3A+Learning+and+Approximation%2C+volume+201+of+Studies+in+Computational+Intelligence%2C+chapter+Entropy+G)
5. Richárd Farkas, Veronika Vincze, György Móra, János Csirik, and György Szarvas. 2010. [The CoNLL-2010 Shared Task: Learning to Detect Hedges and their Scope in Natural Language Text](https://md.research.elamur.ai/free/en/papers/farkas-2010-conll-2010-shared-task-learning/). In Proceedings of the Fourteenth Conference on Computational Natural Language Learning (CoNLL-2010): Shared Task, pages 1–12, Uppsala, Sweden, July. Association for Computational Linguistics.
6. Eraldo R. Fernandes, Cı́cero N. dos Santos, and Ruy L. Milidiú. 2009a. Portuguese language processing service. In Proceedings of the Web in Ibero-America Alternate Track of the 18th World Wide Web Conference (WWW’2009), Madrid. [find](https://scholar.google.com/scholar?q=Eraldo+R%2E+Fernandes%2C+C%C4%B1%CC%81cero+N%2E+dos+Santos%2C+and+Ruy+L%2E+Milidiu%CC%81%2E+2009a%2E+Portuguese+language+processing+service%2E+In+Proceedings+of+the+Web+in+Ibero%2DAmerica+Alternate+Track+of+the+18th+World+Wide+Web+Co)
7. Eraldo R. Fernandes, Bernardo A. Pires, Cı́cero N. dos Santos, and Ruy L. Milidiú. 2009b. Clause identification using entropy guided transformation learning. In Proceedings of the 7th Brazilian Symposium in Information and Human Language Technology (STIL), São Carlos, Brazil. [find](https://scholar.google.com/scholar?q=Eraldo+R%2E+Fernandes%2C+Bernardo+A%2E+Pires%2C+C%C4%B1%CC%81cero+N%2E+dos+Santos%2C+and+Ruy+L%2E+Milidiu%CC%81%2E+2009b%2E+Clause+identification+using+entropy+guided+transformation+learning%2E+In+Proceedings+of+the+7th+Brazilian+Sympo)
8. Eraldo R. Fernandes, Bernardo A. Pires, Cı́cero N. dos Santos, and Ruy L. Milidiú. 2010. A machine learning approach to Portuguese clause identification. In Proceedings of the Nineth International Conference on Computational Processing of the Portuguese Language (PROPOR), volume 6001 of Lecture Notes in Artificial Intelligence, pages 55– 64, Porto Alegre, Brazil. Springer. [find](https://scholar.google.com/scholar?q=Eraldo+R%2E+Fernandes%2C+Bernardo+A%2E+Pires%2C+C%C4%B1%CC%81cero+N%2E+dos+Santos%2C+and+Ruy+L%2E+Milidiu%CC%81%2E+2010%2E+A+machine+learning+approach+to+Portuguese+clause+identification%2E+In+Proceedings+of+the+Nineth+International+Co)
9. Ruy L. Milidiú, Cı́cero N. dos Santos, and Julio C. Duarte. 2008. Phrase chunking using entropy guided transformation learning. In Proceedings of ACL-08: HLT, pages 647–655, Columbus, USA. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Ruy+L%2E+Milidiu%CC%81%2C+C%C4%B1%CC%81cero+N%2E+dos+Santos%2C+and+Julio+C%2E+Duarte%2E+2008%2E+Phrase+chunking+using+entropy+guided+transformation+learning%2E+In+Proceedings+of+ACL%2D08%3A+HLT%2C+pages+647%E2%80%93655%2C+Columbus%2C+USA%2E+Associatio)
10. Ruy L. Milidiú, Cı́cero N. dos Santos, and Carlos E. M. Crestana. 2009. A token classification approach to dependency parsing. In Proceedings of the 7th Brazilian Symposium in Information and Human Language Technology (STIL’2009), São Carlos, Brazil. [find](https://scholar.google.com/scholar?q=Ruy+L%2E+Milidiu%CC%81%2C+C%C4%B1%CC%81cero+N%2E+dos+Santos%2C+and+Carlos+E%2E+M%2E+Crestana%2E+2009%2E+A+token+classification+approach+to+dependency+parsing%2E+In+Proceedings+of+the+7th+Brazilian+Symposium+in+Information+and+Human+L)
11. Roser Morante and Walter Daelemans. 2009. [Learning the scope of hedge cues in biomedical texts](https://md.research.elamur.ai/free/en/papers/morante-2009-learning-scope-hedge-cues-biomedical/). In Proceedings of the BioNLP 2009 Workshop, pages 28–36, Boulder, USA, June. Association for Computational Linguistics.
12. Martin F. Porter. 1980. An algorithm for suffix stripping. Program, 14(3):130–137. [↗](https://doi.org/10.1108/eb046814)
13. Vasin Punyakanok and Dan Roth. 2001. The use of classifiers in sequential inference. In Proceedings of the Conference on Advances in Neural Information Processing Systems (NIPS), pages 995–1001. MIT Press. [↗](https://doi.org/10.48550/arxiv.cs/0111003)
14. Erik F. Tjong Kim Sang and Sabine Buchholz. 2000. [Introduction to the CoNLL-2000 shared task: Chunking](https://md.research.elamur.ai/free/en/papers/sang-2000-introduction-conll-2000-shared-task/). In Proceedings of CoNLL-2000 and LLL-2000, Lisbon, Portugal.
15. Erik F. T. K. Sang and Hervé Déjean. 2001. Introduction to the CoNLL-2001 shared task: Clause identification. In Proceedings of Fifth Conference on Computational Natural Language Learning, Toulouse, France. [↗](https://doi.org/10.3115/1117822.1455626)
16. Veronika Vincze, György Szarvas, Richárd Farkas, György Móra, and János Csirik. 2008. The Bio-Scope corpus: biomedical texts annotated for uncertainty, negation and their scopes. BMC Bioinformatics, 9 (Suppl 11):S9. [find](https://scholar.google.com/scholar?q=Veronika+Vincze%2C+Gyo%CC%88rgy+Szarvas%2C+Richa%CC%81rd+Farkas%2C+Gyo%CC%88rgy+Mo%CC%81ra%2C+and+Ja%CC%81nos+Csirik%2E+2008%2E+The+Bio%2DScope+corpus%3A+biomedical+texts+annotated+for+uncertainty%2C+negation+and+their+scopes%2E+BMC+Bioinformati)

*Updated 24 October 2026*