---
title: "A Hedgehop over a Max-Margin Framework Using Hedge Cues"
description: "A naive \"any cue present\" rule came close at 59%. 17 references, cited by 7 papers here."
authors: ["Maria Georgescul"]
year: 2010
venue: "Proceedings of the Fourteenth Conference on Computational Natural Language Learning – Shared Task"
source: "https://aclanthology.org/W10-3004/"
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/georgescul-2010-hedgehop-over-max-margin-framework/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/georgescul-2010-hedgehop-over-max-margin-framework/)

[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/) / A Hedgehop over a Max-Margin Framework Using Hedge Cues

# A Hedgehop over a Max-Margin Framework Using Hedge Cues

**Authors:** Maria Georgescul · **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/W10-3004/) · [PDF](https://aclanthology.org/W10-3004.pdf) · [OpenAlex](https://openalex.org/W208016162)

**Citations:** 19 citations (OpenAlex, 2026-10-14) · [7](https://md.research.elamur.ai/free/en/papers/georgescul-2010-hedgehop-over-max-margin-framework/cited-by/) in our library, of the 141,840 papers whose bibliographies we read

## Summary
This CoNLL 2010 shared task entry tackles sentence-level uncertainty detection: deciding whether a sentence contains hedged, speculative or "weasel" content. Data came from two domains, BioScope (biomedical abstracts and full articles) and Wikipedia. The author asks whether the linguist-annotated hedge cues shipped with the training corpora are enough on their own. Each sentence is encoded as counts of the known cues, with 2-grams and 3-grams of cue expressions used during tuning. An SVM with an RBF kernel learns from these features. For each domain, class weights offset the label imbalance, and gamma and C are chosen by grid search with 10-fold cross-validation.

On Wikipedia the system placed first, with an F-score of 60.17%. A naive "any cue present" rule came close at 59%. On BioScope it scored 0.78, while the top system reached 0.86. Further analysis shows full articles are inherently harder than abstracts, even when training size is equal. The work matters because it shows how much a compact, interpretable cue inventory achieves without bag-of-words features.

## Key points
- **Baseline.** Tagging every sentence that holds any training cue as uncertain gives near-perfect recall on BioScope (0.98–0.99) but precision below 0.5 on all four datasets. Many cues are ambiguous and also occur in certain sentences.
- **Bag-of-words is worse.** A bag-of-words SVM got about 42% F on Wikipedia, against 49% for the cue baseline. Its input would exceed 150,000 columns, needing over 6GB of RAM in dense form.
- **Best settings.** Wikipedia used gamma = 0.0625 and C = 10 (CV error 17.94%). BioScope used gamma = 0.0625 and C = 110 (CV error 3.73%).
- **Tuning cost.** The grid search took roughly 13 hours on Wikipedia and 4 hours on BioScope, on a Pentium 4.
- **Abstracts vs. full articles.** On held-out BioScope training data, F was 0.90 for abstracts and 0.79 for full articles. A learning curve showed that about 2,000 abstract sentences already give around 87% F, so an 8-point gap remains at matched data size.

## Interesting ideas
- A trivial lexicon-match rule can be nearly as strong as the trained model (59% vs. 60.17%), so every reported gain should be measured against such a rule.
- Restricting features to a curated inventory of annotated cue expressions beat using all words, and it stays small and inspectable.
- Counting "ambiguous cues" (cues that also appear in certain sentences) quantifies how often surface markers fail to signal hedging.
- To tell whether a weak domain suffers from scarce data or is intrinsically harder, the author compared domains at matched training sizes via a learning curve.

## For Mova

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

- **I1** (learn): Add "cue/keyword present" baselines as standing negative controls in evaluation, so a learned component must clearly beat simple lexical matching to count as progress.
- **I2** (frames): Speech-act and certainty detection in prag could start from an explicit hedge-cue lexicon, with each decision traceable to the cue that triggered it.
- **I3** (world): When asserting facts into the world of a text, flag statements with ambiguous hedge cues as epistemically uncertain rather than treating them as plain facts or plain hedges.
- **I4** (tree): When gparse or the perceptron parser does worse on one treebank genre, retrain on a size-matched subset of the strong genre to tell data scarcity apart from genuine genre difficulty.

*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

- [University of Geneva](https://md.research.elamur.ai/free/en/universities/university-of-geneva/) — University, Geneva, Switzerland · 23 works · 2,820 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

## Cite

Cite as: Maria Georgescul. 2010. A Hedgehop over a Max-Margin Framework Using Hedge Cues. 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{georgescul2010hedgehop,
  title = {{A Hedgehop over a Max-Margin Framework Using Hedge Cues}},
  author = {Maria Georgescul},
  booktitle = {Proceedings of the Fourteenth Conference on Computational Natural Language Learning – Shared Task},
  year = {2010},
  url = {https://aclanthology.org/W10-3004/}
}
```

- [← previous Speculation and negation annotation in natural language texts: what the case of BioScope might (not) reveal](https://md.research.elamur.ai/free/en/papers/vincze-2010-speculation-negation-annotation-natural-language/)
- [next → Automatic Extraction of Cue Phrases for Cross-Corpus Dialogue Act Classification](https://md.research.elamur.ai/free/en/papers/webb-2010-automatic-extraction-cue-phrases-cross/)
- [↑ back to papers of 2010](https://md.research.elamur.ai/free/en/papers/2010/page/2/)

## References

17 references, 6 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Rong-En Fan, Pai-Hsuen Chen, and Chih-Jen Lin. 2005. Working set selection using the second order information for training SVM. Journal of Machine Learning Research , 6: 1889-918. [find](https://scholar.google.com/scholar?q=Rong%2DEn+Fan%2C+Pai%2DHsuen+Chen%2C+and+Chih%2DJen+Lin%2E+2005%2E+Working+set+selection+using+the+second+order+information+for+training+SVM%2E+Journal+of+Machine+Learning+Research+%2C+6%3A+1889%2D918%2E)
2. 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.
3. An Fellbaum. 1998. WordNet: Electronic Lexical Database . MIT Press, Christiane Cambridge, MA. [find](https://scholar.google.com/scholar?q=An+Fellbaum%2E+1998%2E+WordNet%3A+Electronic+Lexical+Database+%2E+MIT+Press%2C+Christiane+Cambridge%2C+MA%2E)
4. Viola Ganter, and Michael Strube. 2009. [Finding hedges by chasing weasels: Hedge detection using Wikipedia tags and shallow linguistic features](https://md.research.elamur.ai/free/en/papers/ganter-2009-finding-hedges-chasing-weasels-hedge/). In Proceedings of joint conference of the 47th Annual Meeting of the ACL-IJCNLP.
5. Halil Kilicoglu, and Sabine Bergler. 2008. Recognizing Speculative Language in Biomedical Research Articles: A Linguistically Motivated Perspective. In Proceedings of Current Trends in Biomedical Natural Language Processing (BioNLP) , Columbus, Ohio, USA. [find](https://scholar.google.com/scholar?q=Halil+Kilicoglu%2C+and+Sabine+Bergler%2E+2008%2E+Recognizing+Speculative+Language+in+Biomedical+Research+Articles%3A+A+Linguistically+Motivated+Perspective%2E+In+Proceedings+of+Current+Trends+in+Biomedical+Natu)
6. Marc Light, Xin Ying Qiu, and Padmini Srinivasan. 2004. The Language of Bioscience: Facts, Speculations, and Statements in between. In Proceedings of the HLT BioLINK . [find](https://scholar.google.com/scholar?q=Marc+Light%2C+Xin+Ying+Qiu%2C+and+Padmini+Srinivasan%2E+2004%2E+The+Language+of+Bioscience%3A+Facts%2C+Speculations%2C+and+Statements+in+between%2E+In+Proceedings+of+the+HLT+BioLINK+%2E)
7. Chrysanne Di Marco, and Robert E. Mercer. 2004. Hedging in Scientific Articles as a Means of Classifying Citations. In Proceedings of Working Notes of AAAI Spring Symposium on Exploring Attitude and Affect in Text: Theories and Applications , Stanford University. [find](https://scholar.google.com/scholar?q=Chrysanne+Di+Marco%2C+and+Robert+E%2E+Mercer%2E+2004%2E+Hedging+in+Scientific+Articles+as+a+Means+of+Classifying+Citations%2E+In+Proceedings+of+Working+Notes+of+AAAI+Spring+Symposium+on+Exploring+Attitude+and+A)
8. Alexa T. McCray, Suresh Srinivasan, and Allen C. Browne. 1994. Lexical methods for managing variation in biomedical terminologies. In Proceedings of the 18th Annual Symposium on Computer Applications in Medical Care . [find](https://scholar.google.com/scholar?q=Alexa+T%2E+McCray%2C+Suresh+Srinivasan%2C+and+Allen+C%2E+Browne%2E+1994%2E+Lexical+methods+for+managing+variation+in+biomedical+terminologies%2E+In+Proceedings+of+the+18th+Annual+Symposium+on+Computer+Applications+)
9. Ben Medlock. 2008. Exploring hedge identification in biomedical literature. Journal of Biomedical Informatics , 41:636-54. [↗](https://doi.org/10.1016/j.jbi.2008.01.001)
10. Ben Medlock, and Ted Briscoe. 2007. [Weakly Supervised Learning for Hedge Classification in Scientific Literature](https://md.research.elamur.ai/free/en/papers/medlock-2007-weakly-supervised-learning-hedge-classification/). In P roceedings of the 45th Annual Meeting of the Computational Linguistics . Association of Robert E.Mercer, Chrysanne Di Marco, and Frederick Kroon. 2004. The frequency of hedging cues in citation contexts in scientific writing. In Proceedings of the Canadian Society for the Computational Studies of Intelligence (CSCSI) , London, Ontario.
11. György Móra, Richárd Farkas, György Szarvas, and Zsolt Molnár. 2009. Exploring ways beyond the simple supervised learning approach for biological event extraction. In Proceedings of the BioNLP 2009 Workshop Companion Volume for Shared Task . [find](https://scholar.google.com/scholar?q=Gy%C3%B6rgy+M%C3%B3ra%2C+Rich%C3%A1rd+Farkas%2C+Gy%C3%B6rgy+Szarvas%2C+and+Zsolt+Moln%C3%A1r%2E+2009%2E+Exploring+ways+beyond+the+simple+supervised+learning+approach+for+biological+event+extraction%2E+In+Proceedings+of+the+BioNLP+2009+Wo)
12. Roser Morante, and Walter Daelemans. 2009a. [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 , Boulder, Colorado. Association for Computational Linguistics.
13. Roser Morante, and Walter Daelemans. 2009b. [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 Workshop on BioNLP.
14. György Szarvas. 2008. [Hedge classification in biomedical texts with a weakly supervised selection of keywords](https://md.research.elamur.ai/free/en/papers/szarvas-2008-hedge-classification-biomedical-texts-weakly/). In Proceedings of the ACL-08: HLT .
15. Vladimir N. Vapnik. 1998. Statistical learning theory . Wiley, New York. [find](https://scholar.google.com/scholar?q=Vladimir+N%2E+Vapnik%2E+1998%2E+Statistical+learning+theory+%2E+Wiley%2C+New+York%2E)
16. Veronika Vincze, György Szarvas, Richárd Farkas, György Móra, and János Csirik. 2008. The BioScope corpus: biomedical texts annotated for uncertainty, negation and their scopes. BMC Bioinformatics , 9. [↗](https://doi.org/10.1186/1471-2105-9-s11-s9)
17. Dennis Wegener, Thierry Sengstag, Stelios R. Sfakianakis, and Anthony Assi. 2007. GridR: An R-based grid-enabled tool for data analysis in ACGT clinico-genomic trials. In Proceedings of the 3rd International Conference on e-Science and Grid Computing (eScience 2007). [find](https://scholar.google.com/scholar?q=Dennis+Wegener%2C+Thierry+Sengstag%2C+Stelios+R%2E+Sfakianakis%2C+and+Anthony+Assi%2E+2007%2E+GridR%3A+An+R%2Dbased+grid%2Denabled+tool+for+data+analysis+in+ACGT+clinico%2Dgenomic+trials%2E+In+Proceedings+of+the+3rd+Intern)

*Updated 24 October 2026*