Elamur Research⌕
☰
home / papers / 2015 / A hedging annotation scheme focused on epistemic phrases for informal language

A hedging annotation scheme focused on epistemic phrases for informal language

Paper · 2015 · Liliana Mamani Sanchez and Carl Vogel

Summary

Hedge annotation schemes have mostly been built for edited, formal prose such as biomedical and scholarly articles. They treat hedges as isolated speculative cue words and usually ignore who holds the uncertain stance. The authors studied posts from a commercial software vendor's user-support forum, where people hedge through casual phrasal forms: "not sure", IMO, AFAIK, "dunno", often with subject ellipsis and typos. They propose a scheme of entities (hedge, source, scope, non-hedge, other discourse markers), relations linking hedges to source and scope, and attributes. It has three main hedge classes: single hedges, syntactic hedges, and Not-Claiming-Knowledge (NCK) epistemic phrases. NCK phrases are first-person expressions that openly mark the writer's own involvement, and they form the central class. Source is split into an outer source, which is always the post's writer, and an inner source, whose view is actually hedged. Scope is marked by its syntactic dependency head. Annotating the forum data shows that NCK phrases are varied but can be normalized into a small set of groups. This matters for detecting hedges in noisy text, where deep parsing is unreliable.

Key points

  • Hedge classes: Single hedges, with a starting lexicon taken from Rubin (2010); NCK epistemic phrases, subdivided into primary, semantically extended (e.g. negated "know", "dunno") and lexically extended (e.g. "I am hopeful"); and Syntactic hedges. A miscellaneous group and a Non-hedge label cover the remaining cases.
  • Normalization handles abbreviations, contractions, misspellings, tense and number variation, colloquial forms and modifiers. It reduced single-hedge types from 272 to 189 and NCK types from 300 to 137. The "I do not know" group alone gathers 20 surface variants.
  • Syntactic hedges (e.g. if, or, when) cover 7 lexical types with 1,307 occurrences. The miscellaneous class has 314 occurrences over 210 types, and topic-specific self-deprecation such as "I am not a techie" was kept out of NCK so that NCK stays topic-neutral.
  • In 3.57% of cases (286 hedges), the source is someone other than the writer. In 18.26% of cases (1,496 hedges), the hedge has no attached scope.
  • No independent annotators or agreement study yet; the annotation is described as exploratory.

Interesting ideas

  • I1. Separating an outer source (who wrote the sentence) from an inner source (whose hedged view it is) means reported stances like "User1 suggested…" are not credited to the writer.
  • I2. Scope is anchored to the dependency head of the affected phrase instead of to clause boundaries, which avoids tagging unrelated words like a leading "however".
  • I3. An NCK phrase counts only when a first-person subject is present (explicit or elided), so "can not know" is classified differently depending on its subject.
  • I4. Surface variants (dunno, donno, i dont know, IMO) are collapsed into normalized groups, so a small set of canonical hedge types covers noisy input.
  • I5. An explicit Non-hedge label records cue words that look like hedges but are not used to hedge, which gives ready-made negative examples.

For Mova

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

  • I1 (frames): Speech-act and stance frames in prag could carry both a writer slot and an experiencer slot, so that "X suggested Y" records X's tentative belief rather than the narrator's.
  • I2 (tree): The UD tree already supplies heads, so a hedge's scope can be stored as a pointer to the head of the governed subtree rather than as a token span.
  • I3 (world): When building the world of a text, first-person NCK phrases can lower the writer's certainty about a proposition, yielding explainable states such as "writer does not know whether X".
  • I4 (tokens): The en crate could map informal contractions and acronyms to canonical lemma sequences before tagging, giving downstream stages a clean hedge signal.
  • I5 (learn): Hedge-cue detectors could be evaluated against these non-hedge occurrences as negative controls, checking that a rule does not fire on every "may" or "if".

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

🎓Trinity College DublinDublin · Ireland17 works · 1,222 citations

📚 Shelves 5 Computer Science · 9,086 works Artificial Intelligence · 8,889 works Arts and Humanities · 9,296 works Language and Linguistics · 9,206 works Literature and Literary Theory · 340 works

Cite

Cite as Liliana Mamani Sanchez and Carl Vogel. 2015. A hedging annotation scheme focused on epistemic phrases for informal language. Proceedings of the Workshop on Models for Modality Annotation.

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

@inproceedings{sanchez2015hedging,
  title = {{A hedging annotation scheme focused on epistemic phrases for informal language}},
  author = {Liliana Mamani Sanchez and Carl Vogel},
  booktitle = {Proceedings of the Workshop on Models for Modality Annotation},
  year = {2015},
  url = {https://aclanthology.org/W15-0302/}
}

References

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

  1. Farkas, R., V. Vincze, G. Mra, J. Csirik, and G. Szarvas (2010, July). The CoNLL-2010 Shared Task: Learning to Detect Hedges and their Scope in Natural Language Text. In Proceedings of the Fourteenth Conference on Computational Natural Language Learning, Uppsala, Sweden, pp. 1–12. ACL. our card
  2. Ganter, V. and M. Strube (2009). Finding hedges by chasing weasels: Hedge detection using Wikipedia tags and shallow linguistic features. In Proceedings of the ACL-IJCNLP 2009 Conference Short Papers, Suntec, Singapore, pp. 173–176. our card
  3. Hendrickx, I., A. Mendes, and S. Mencarelli (2012). Modality in text: a proposal for corpus annotation. In Proceedings of the Eight International Conference on Language Resources and Evaluation (LREC’12), Istanbul, Turkey. European Language Resources Association (ELRA). find
  4. Holmes, J. (1988). Doubt and certainty in ESL textbooks. Applied Linguistics 9(1), 21–44. ↗
  5. Hyland, K. (1998). Hedging in Scientific Research Articles. Pragmatics & beyond. John Benjamins Publishing Company. ↗ our card
  6. Iatridou, S. (1991). Topics in Conditionals. Ph. D. thesis, MIT, Cambridge, Massachusetts. Distributed by MIT Working Papers in Linguistics. find
  7. Kärkkäinen, E. (2010). Position and scope of epistemic phrases in planned and unplanned american english. In New approaches to hedging, pp. 207–241. Amsterdam: Elsevier. ↗
  8. Light, M., X. T. Qui, and P. Srinivasan (2004). The language of bioscience: Facts, speculations, and statements in between. Proceedings of BioLink 2004 Workshop on Linking Biological Literature, Ontologies and Databases: Tools for Users, 17 – 24. find
  9. Medlock, B. and T. Briscoe (2007). Weakly supervised learning for hedge classification in scientific literature. In Proceedings of the 45th Annual Meeting of the Association of Computational Linguistics, Prague, Czech Republic, pp. 992–999. our card
  10. Rubin, V. L. (2010). Epistemic modality: From uncertainty to certainty in the context of information seeking as interactions with texts. Information Processing & Management 46(5), 533–540. ↗
  11. Scheibman, J. (2001). Local patterns of subjectivity in person and verb type in. In J. L. Bybee and P. Hopper (Eds.), Frequency and the Emergence of Linguistic Structure, Volume 45 of Typological studies in language, pp. 61–89. Amsterdam; Philadelphia: John Benjamins Publishing Company. find
  12. Szarvas, G., V. Vincze, R. Farkas, and J. Csirik (2008). The BioScope corpus: annotation for negation, uncertainty and their scope in biomedical texts. In Proceedings of the Workshop on Current Trends in Biomedical Natural Language Processing, Columbus, Ohio, pp. 38–45. our card
  13. Thompson, S. A. and A. Mulac (1991). A quantitative perspective on the grammaticization of epistemic parentheticals in english. In Approaches to Grammaticalization, pp. 314–329. John Benjamins. find
  14. Vincze, V. (2013, October). Weasels, hedges and peacocks: Discourse-level uncertainty in wikipedia articles. In Proceedings of the Sixth International Joint Conference on Natural Language Processing, Nagoya, Japan, pp. 383–391. Asian Federation of Natural Language Processing. our card
  15. Vincze, V., I. K. Simkó, and V. Varga (2014). Proceedings of LAW VIII - The 8th Linguistic Annotation Workshop, Chapter Annotating Uncertainty in Hungarian Webtext, pp. 64–69. Association for Computational Linguistics and Dublin City University. find
  16. Vincze, V., G. Szarvas, R. Farkas, G. Mora, and J. Csirik (2008). The BioScope corpus: biomedical texts annotated for uncertainty, negation and their scopes. BMC Bioinformatics 9(Suppl 11), S9. ↗
  17. Wiebe, J., T. Wilson, and C. Cardie (2005). Annotating expressions of opinions and emotions in language ANN. Language Resources and Evaluation 39(2/3), 164–210. ↗
  18. Wierzbicka, A. (2006). English: meaning and culture. Oxford University Press, USA. find
  19. Wilson, T. and J. Wiebe (2005). Annotating attributions and private states. In Proceedings of the Workshop on Frontiers in Corpus Annotations II: Pie in the Sky, CorpusAnno’05, Stroudsburg, PA, USA, pp. 53–60. ACL. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/sanchez-2015-hedging-annotation-scheme-focused-epistemic/