---
title: "Hedge Classification with Syntactic Dependency Features Based"
description: "The paper describes a system for Task 1 of the CoNLL-2010 Shared Task: deciding whether a sentence… 7 references, cited by 1 paper here."
authors: ["Yi Zheng", "Qifeng Dai", "Qiming Luo", "Enhong Chen"]
year: 2010
venue: "Proceedings of the Fourteenth Conference on Computational Natural Language Learning – Shared Task"
source: "https://aclanthology.org/W10-3023/"
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/zheng-2010-hedge-classification-syntactic-dependency-features/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/zheng-2010-hedge-classification-syntactic-dependency-features/)

[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/3/) / Hedge Classification with Syntactic Dependency Features Based on an Ensemble Classifier

# Hedge Classification with Syntactic Dependency Features Based on an Ensemble Classifier

**Authors:** [Yi Zheng](https://md.research.elamur.ai/free/en/authors/yi-zheng/), Qifeng Dai, Qiming Luo and [Enhong Chen](https://md.research.elamur.ai/free/en/authors/enhong-chen/) · **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/W10-3023/) · [PDF](https://aclanthology.org/W10-3023.pdf) · [OpenAlex](https://openalex.org/W213220957)

**Citations:** 2 citations (OpenAlex, 2026-10-14) · 1 in our library, of the 149,439 papers whose bibliographies we read

## Summary
The paper describes a system for Task 1 of the CoNLL-2010 Shared Task: deciding whether a sentence carries uncertain or hedged information. The authors do not classify whole sentences directly. They tag each token with BIO-cue labels that mark the start of a hedge cue, the inside of one, or neither. Any sentence with at least one cue token counts as uncertain. Preprocessing uses PTBTokenizer, the OpenNLP chunker and Stanford Parser, which supplies POS, lemmas and Stanford typed dependencies. Features cover the token itself, its syntactic head, a window of neighbours, document metadata and n-grams. Two taggers, CRF++ and OpenNLP MaxEnt, are trained separately, and their outputs are merged so that a token is flagged as a cue if either model flags it. The system was tested on BioScope (biomedical) and on Wikipedia weasel-word data. The union ensemble raised recall and clearly helped on Wikipedia, where single models missed many hedges. Dependency features helped only on in-domain biomedical text. The work shows that the main difficulty with hedges in open-domain text is recall on multiword phrases.

## Key points
- Official F-scores for the uncertainty class were 81.11 (Bio in-domain), 67.99 (Bio cross-domain), 55.48 (Wiki in-domain) and 55.41 (Wiki cross-domain).
- The pipeline is PTBTokenizer → OpenNLP chunker → Stanford Parser for POS, lemma and typed dependencies. Dependency features are parent index, token, lemma, POS and relation. Neighbour features use a window of 3 tokens on each side. Metadata features are documentPart type, domain, and abstract vs. article.
- The ensemble merge rule works per token. If both models predict O, the label is O. Otherwise the token gets B if either model predicts B, and I in the remaining cases. CRF++ ran with -f 3 and -c 1.5; MaxEnt used 100 iterations, cutoff 0 and no smoothing.
- On the Wikipedia test set the ensemble reached 55.48 F, against 48.67 for MaxEnt and 47.02 for CRF alone. On the biomedical in-domain test set, CRF alone did better (82.99 vs. 81.11).
- Error analysis showed false negatives far outnumbering false positives. Wikipedia hedges are mostly phrases of several words, while biomedical cues are mostly single words. Cues of three or more words were especially hard to learn.

## Interesting ideas
- The sentence-level label is derived from token-level cue tags, so every "uncertain" verdict points at the exact words that caused it, which makes the decision auditable.
- When error analysis shows misses dominate, the authors merge models by OR rather than by vote, deliberately trading precision for recall. This only pays off where single models have low recall.
- The syntactic head of a token (its lemma, POS and relation) is used as a feature for cue detection, and it helped on regular, well-formed text but not on noisy text.
- Errors are broken down by cue length, revealing that multiword hedge phrases follow a few recurring patterns. The authors suggest rules for these phrases combined with statistics for single words.
- Degenerate inputs are handled with an explicit default: empty sentences are labeled certain. The authors note this has negligible impact because such sentences are rare.

## For Mova

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

- **I1** (frames): prag could mark epistemic status on a frame and store the cue token IDs as the justification, so a reader can see why a claim was treated as hedged.
- **I2** (frames): uncertainty and speech-act detectors that miss many hedges could be combined by union, after first checking the FP/FN balance on a held-out set.
- **I3** (frames): hedge detection in prag could read the UD head of candidate words like "may" or "suggest" rather than linear neighbours, while expecting smaller gains on informal input.
- **I4** (learn): self-induced repair rules could target recurring multiword hedge patterns that the statistical tagger misses, with the residual FN count tracked to confirm the gain.
- **I5** (answer): readers should have documented fallbacks for empty or unparsable sentences so that failures stay explainable instead of silent.

*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 Science and Technology of China](https://md.research.elamur.ai/free/en/universities/university-of-science-and-technology-of-china/) — University, Hefei, China · 108 works · 112,698 citations

📚 Shelves (6): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 16,039 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 15,891 works · [🧬 Biochemistry, Genetics and Molecular Biology](https://md.research.elamur.ai/free/en/shelves/biochemistry-genetics-and-molecular-biology/) · 234 works · [🧬 Molecular Biology](https://md.research.elamur.ai/free/en/shelves/biochemistry-genetics-and-molecular-biology/molecular-biology/) · 231 works

## Cite

Cite as: Yi Zheng, Qifeng Dai, Qiming Luo and Enhong Chen. 2010. Hedge Classification with Syntactic Dependency Features Based on an Ensemble Classifier. 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{zheng2010hedge,
  title = {{Hedge Classification with Syntactic Dependency Features Based on an Ensemble Classifier}},
  author = {Yi Zheng, Qifeng Dai, Qiming Luo and Enhong Chen},
  booktitle = {Proceedings of the Fourteenth Conference on Computational Natural Language Learning – Shared Task},
  year = {2010},
  url = {https://aclanthology.org/W10-3023/}
}
```

- [← previous Exploiting Multi-Features to Detect Hedges and their Scope in Biomedical Texts](https://md.research.elamur.ai/free/en/papers/zhou-2010-exploiting-multi-features-detect-hedges/)
- [next → NameDat: A Database of English Proper Names Spoken by Native Norwegians](https://md.research.elamur.ai/free/en/papers/adde-2010-namedat-database-english-proper-names/)
- [↑ back to papers of 2010](https://md.research.elamur.ai/free/en/papers/2010/page/3/)

## References

7 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. 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 CoNLL-2010: Shared Task , pages 1–12.
2. 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 the ACL-IJCNLP 2009 Conference Short Papers , pages 173–176.
3. Marie-Catherine de Marneffe, and Christopher D. Manning. 2008. Stanford typed dependencies manual , September 2008. [find](https://scholar.google.com/scholar?q=Marie%2DCatherine+de+Marneffe%2C+and+Christopher+D%2E+Manning%2E+2008%2E+Stanford+typed+dependencies+manual+%2C+September+2008%2E)
4. 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 Proc. of ACL 2007 , pages 992–999.
5. 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 Proc. of the BioNLP 2009 Workshop , pages 28–36.
6. 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 Proc. of ACL 2008 , pages 281–289.
7. 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+Gy%C3%B6rgy+Szarvas%2C+Rich%C3%A1rd+Farkas%2C+Gy%C3%B6rgy+M%C3%B3ra%2C+and+J%C3%A1nos+Csirik%2E+2008%2E+The+Bio%2DScope+corpus%3A+biomedical+texts+annotated+for+uncertainty%2C+negation+and+their+scopes%2E+BMC+Bioinformatics+%2C+)

*Updated 30 October 2026*