---
title: "Recognizing Speculative Language in Biomedical Research"
description: "Kilicoglu and Bergler build a rule-based detector grounded in linguistics instead. 15 references, cited by 39 papers here."
authors: ["Halil Kilicoglu", "Sabine Bergler"]
year: 2008
venue: "Proceedings of the Workshop on Current Trends in Biomedical Natural Language Processing"
source: "https://aclanthology.org/W08-0607/"
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/kilicoglu-2008-recognizing-speculative-language-biomedical-research/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/kilicoglu-2008-recognizing-speculative-language-biomedical-research/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2008](https://md.research.elamur.ai/free/en/papers/2008/) / Recognizing Speculative Language in Biomedical Research Articles: A Linguistically Motivated Perspective

# Recognizing Speculative Language in Biomedical Research Articles: A Linguistically Motivated Perspective

**Authors:** [Halil Kilicoglu](https://md.research.elamur.ai/free/en/authors/halil-kilicoglu/) and [Sabine Bergler](https://md.research.elamur.ai/free/en/authors/sabine-bergler/) · **Year:** 2008 · **Venue:** Proceedings of the Workshop on Current Trends in Biomedical Natural Language Processing · **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/W08-0607/) · [PDF](https://aclanthology.org/W08-0607.pdf) · [OpenAlex](https://openalex.org/W2159230276)

**Citations:** 123 citations (OpenAlex, 2026-10-14) · [39](https://md.research.elamur.ai/free/en/papers/kilicoglu-2008-recognizing-speculative-language-biomedical-research/cited-by/) in our library, of the 149,439 papers whose bibliographies we read

## Summary
Biomedical articles are full of tentative claims. Extraction systems that treat every sentence as a plain assertion can misread them, and in some cases a hedge even flips the meaning of the fact. Earlier work classified sentences as speculative using single-word features or simple substring matching. Kilicoglu and Bergler build a rule-based detector grounded in linguistics instead. They start from Hyland's (1998) inventory of hedging devices: modals, epistemic verbs, adjectives, adverbs and nouns. They enlarge this inventory by studying 521 hand-annotated sentences and add "unhedgers", which are cues signalling certainty. The cue lists are then expanded semi-automatically with WordNet and the UMLS SPECIALIST Lexicon and cleaned by hand. Syntactic patterns are matched over dependency output from the Stanford Lexicalized Parser. Each cue carries a weight that nearby structures can strengthen or weaken, and the weights add up to a sentence hedging score. On the manually annotated Drosophila test set of Medlock and Briscoe (2007), the system reaches a recall/precision BEP of about 0.85, against baselines of 0.60 and 0.76. A compact, transparent rulebase can therefore beat bag-of-words learning on this task.

## Key points
- **Data:**
  - The test set comes from Medlock and Briscoe (2007): 1537 sentences, 380 of them speculative, drawn from six full-text fruit-fly articles.
  - A distinct training set of 521 sentences (213 speculative) was labelled by the first author.
  - The authors set aside the automatically induced 13,964-sentence training set because it over-represents the seed cues *suggest* and *likely*.
- **Representativeness check:** In the small training sample, hedge realizations break down as epistemic verbs 32%, adverbs 26%, modals 21% and adjectives 19%. This roughly matches Hyland's (1998) distribution.
- **Tools and resources:**
  - Lexicon expansion used WordNet and the UMLS SPECIALIST Lexicon.
  - Syntactic patterns were matched with the Stanford Lexicalized Parser and its typed-dependency representation.
  - Significance against the baselines was tested with the binomial sign test (p \< 0.01).
- **Results:**
  - The system reaches a recall/precision BEP of about 0.85.
  - Baseline1 (Light et al. substring matching) scores 0.60, and Baseline2 scores 0.76.
  - At the best threshold, precision and recall are both about 0.85.
- **Error types:**
  - Missed negative quantifiers, as in "little was known".
  - Missed derived forms such as *suggestive*.
  - *could* misread as a hedge when it really expresses ability or constraint, for example after "we".
  - Weak cues such as *conclude*, *usually* and *assumption*.

## Interesting ideas
- The method lists cues of certainty ("unhedgers") alongside hedges and treats a negated unhedger, such as "little is known", as a hedge. This captures admissions of missing knowledge, which word-feature classifiers tend to miss.
- Cue weights are adjusted by syntactic context; for instance, *appear* counts as a strong hedge only with a *to* or *that* complement, and the authors suggest per-verb handling rather than uniform discounting.
- Instead of trusting a large auto-induced training set, the authors hand-labelled a small sample and checked that its category distribution matched a published linguistic survey. This is a cheap sanity check for skew introduced by seed words.
- The error analysis points to WordNet's derivationally related forms as a way to carry cue status from *suggest* to *suggestive*, extending a lexicon systematically rather than word by word.
- False positives with *could* mostly involve first-person "we" and objective enabling conditions, where the modal marks ability or a past constraint rather than doubt. This shows modal sense depends on subject and context.

## For Mova

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

- **I1** (frames): `prag` could attach an epistemic-status field to each frame, flipping it when a certainty verb like *know* is negated or quantified by *little/few*.
- **I2** (world): states entered into the world of the text could carry a graded certainty value computed from the cue and its dependency context, so the answer stage can distinguish "X does Y" from "X appears to do Y".
- **I3** (learn): silver data from witness parsers or LLM tiers could be compared against reference distributions of constructions to catch seed or teacher bias before training.
- **I4** (world): the compiled WordNet knowledge could propagate epistemic and other verb-class features along derivation links, giving adjectives and nominalizations the semantics of their base verbs.
- **I5** (frames): modal frames could be disambiguated into epistemic versus ability or root readings using the UD subject and negation, so that "we could not find" becomes a fact about a failed search, not uncertainty.

*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

- [Concordia University](https://md.research.elamur.ai/free/en/universities/concordia-university/) — University, Montreal, Canada · 9 works · 673 citations

📚 Shelves (5): [🖥️ 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 · [📚 Literature and Literary Theory](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/literature-and-literary-theory/) · 519 works

## Cite

Cite as: Halil Kilicoglu and Sabine Bergler. 2008. Recognizing Speculative Language in Biomedical Research Articles: A Linguistically Motivated Perspective. Proceedings of the Workshop on Current Trends in Biomedical Natural Language Processing.

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

```bibtex
@inproceedings{kilicoglu2008recognizing,
  title = {{Recognizing Speculative Language in Biomedical Research Articles: A Linguistically Motivated Perspective}},
  author = {Halil Kilicoglu and Sabine Bergler},
  booktitle = {Proceedings of the Workshop on Current Trends in Biomedical Natural Language Processing},
  year = {2008},
  doi = {10.1186/1471-2105-9-s11-s10},
  url = {https://doi.org/10.1186/1471-2105-9-s11-s10}
}
```

- [← previous Specialized Models and Ranking for Coreference Resolution](https://md.research.elamur.ai/free/en/papers/denis-2008-specialized-models-ranking-coreference-resolution/)
- [next → Towards Robust Semantic Role Labeling](https://md.research.elamur.ai/free/en/papers/pradhan-2008-robust-semantic-role-labeling/)
- [↑ back to papers of 2008](https://md.research.elamur.ai/free/en/papers/2008/)

## References

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

1. deMarneffe, M. C., MacCartney B., Manning C.D. 2006. [Generating Typed Dependency Parses from Phrase Structure Parses](https://md.research.elamur.ai/free/en/papers/marneffe-2006-generating-typed-dependency-parses-phrase/). In Proc of 5th International Conference on Language Resources and Evaluation, pp. 449-54.
2. DiMarco C. and Mercer R.E. 2004. Hedging in Scientific Articles as a Means of Classifying Citations. In Exploring Attitude and Affect in Text: Theories and Applications AAAI-EAAT 2004. pp.50-4. [find](https://scholar.google.com/scholar?q=DiMarco+C%2E+and+Mercer+R%2EE%2E+2004%2E+Hedging+in+Scientific+Articles+as+a+Means+of+Classifying+Citations%2E+In+Exploring+Attitude+and+Affect+in+Text%3A+Theories+and+Applications+AAAI%2DEAAT+2004%2E+pp%2E50%2D4%2E)
3. Fellbaum, C. 1998. WordNet: An Electronic Lexical Database. MIT Press, Cambridge, MA. [find](https://scholar.google.com/scholar?q=Fellbaum%2C+C%2E+1998%2E+WordNet%3A+An+Electronic+Lexical+Database%2E+MIT+Press%2C+Cambridge%2C+MA%2E)
4. Friedman C., Alderson P., Austin J., Cimino J.J., Johnson S.B. 1994. A general natural-language text processor for clinical radiology. Journal of the American Medical Informatics Association, 1(2): 161-74. [↗](https://doi.org/10.1136/jamia.1994.95236146)
5. Hyland K. 1998. [Hedging in Scientific Research Articles](https://md.research.elamur.ai/free/en/papers/hyland-1998-hedging-scientific-research-articles/). John Benjamins B.V., Amsterdam, Netherlands. [↗](https://doi.org/10.1075/pbns.54)
6. Kipper Schuler, K. 2005. VerbNet: A broad-coverage, comprehensive verb lexicon. PhD thesis, University of Pennsylvania. [↗](https://doi.org/10.5281/zenodo.8100881)
7. Klein D. and Manning C. D. 2003. [Accurate unlexicalized parsing](https://md.research.elamur.ai/free/en/papers/klein-2003-accurate-unlexicalized-parsing/). In Proc of 41st Meeting of the Association for Computational Linguistics. pp. 423-30.
8. Lakoff G. 1972. [Hedges: A Study in Meaning Criteria and the Logic of Fuzzy Concepts](https://md.research.elamur.ai/free/en/papers/lakoff-1973-hedges-study-meaning-criteria/). Chicago Linguistics Society Papers, 8, pp.183-228. [↗](https://doi.org/10.1007/bf00262952)
9. Light M., Qiu X.Y., Srinivasan P. 2004. The Language of Bioscience: Facts, Speculations, and Statements in between. In BioLINK 2004: Linking Biological Literature, Ontologies and Databases, pp. 17-24. [find](https://scholar.google.com/scholar?q=Light+M%2E%2C+Qiu+X%2EY%2E%2C+Srinivasan+P%2E+2004%2E+The+Language+of+Bioscience%3A+Facts%2C+Speculations%2C+and+Statements+in+between%2E+In+BioLINK+2004%3A+Linking+Biological+Literature%2C+Ontologies+and+Databases%2C+pp%2E+17%2D24%2E)
10. McCray A. T., Srinivasan S., Browne A. C. 1994. Lexical methods for managing variation in biomedical terminologies. In Proc of 18th Annual Symposium on Computer Applications in Medical Care, pp. 235-9. [find](https://scholar.google.com/scholar?q=McCray+A%2E+T%2E%2C+Srinivasan+S%2E%2C+Browne+A%2E+C%2E+1994%2E+Lexical+methods+for+managing+variation+in+biomedical+terminologies%2E+In+Proc+of+18th+Annual+Symposium+on+Computer+Applications+in+Medical+Care%2C+pp%2E+235%2D9)
11. Medlock B. and Briscoe T. 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 45 th Meeting of the Association for Computational Linguistics. pp.992-9.
12. Palmer F.R. 1986. Mood and Modality. Cambridge University Press, Cambridge, UK. [find](https://scholar.google.com/scholar?q=Palmer+F%2ER%2E+1986%2E+Mood+and+Modality%2E+Cambridge+University+Press%2C+Cambridge%2C+UK%2E)
13. Pustejovsky J., Hanks P., Saurí R., See A., Gaizauskas R., Setzer A., Radev D., Sundheim B., Day D. Ferro L., Lazo M. 2003. The TimeBank Corpus. In Proc of Corpus Linguistics. pp. 647-56. [find](https://scholar.google.com/scholar?q=Pustejovsky+J%2E%2C+Hanks+P%2E%2C+Saur%C3%AD+R%2E%2C+See+A%2E%2C+Gaizauskas+R%2E%2C+Setzer+A%2E%2C+Radev+D%2E%2C+Sundheim+B%2E%2C+Day+D%2E+Ferro+L%2E%2C+Lazo+M%2E+2003%2E+The+TimeBank+Corpus%2E+In+Proc+of+Corpus+Linguistics%2E+pp%2E+647%2D56%2E)
14. Saurí R., Verhagen M., Pustejovsky J. 2006. SlinkET: a partial modal parser for events. In Proc of 5 th International Conference on Language Resources and Evaluation. [find](https://scholar.google.com/scholar?q=Saur%C3%AD+R%2E%2C+Verhagen+M%2E%2C+Pustejovsky+J%2E+2006%2E+SlinkET%3A+a+partial+modal+parser+for+events%2E+In+Proc+of+5+th+International+Conference+on+Language+Resources+and+Evaluation%2E)
15. Wilbur W.J., Rzhetsky A., Shatkay H. 2006. New Directions in Biomedical Text Annotations: Definitions, Guidelines and Corpus Construction. BMC Bioinformatics, 7:356. [find](https://scholar.google.com/scholar?q=Wilbur+W%2EJ%2E%2C+Rzhetsky+A%2E%2C+Shatkay+H%2E+2006%2E+New+Directions+in+Biomedical+Text+Annotations%3A+Definitions%2C+Guidelines+and+Corpus+Construction%2E+BMC+Bioinformatics%2C+7%3A356%2E)

*Updated 1 November 2026*