---
title: "Semi-supervised Speech Act Recognition in Emails and Forums"
description: "Recognizing speech acts sentence by sentence in emails and forum threads is hard because these… 24 references, cited by 27 papers here."
authors: ["Minwoo Jeong", "Chin-Yew Lin", "Gary Geunbae Lee"]
year: 2009
venue: "Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing"
source: "https://aclanthology.org/D09-1130/"
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/jeong-2009-semi-supervised-speech-act-recognition/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/jeong-2009-semi-supervised-speech-act-recognition/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2009](https://md.research.elamur.ai/free/en/papers/2009/) / Semi-supervised Speech Act Recognition in Emails and Forums

# Semi-supervised Speech Act Recognition in Emails and Forums

**Authors:** Minwoo Jeong, [Chin-Yew Lin](https://md.research.elamur.ai/free/en/authors/chin-yew-lin/) and Gary Geunbae Lee · **Year:** 2009 · **Venue:** [Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing](https://md.research.elamur.ai/free/en/venues/emnlp/) ([EMNLP 2009](https://md.research.elamur.ai/free/en/venues/emnlp/2009/)) · **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/D09-1130/) · [PDF](https://aclanthology.org/D09-1130.pdf) · [OpenAlex](https://openalex.org/W2089285937)

**Citations:** 79 citations (OpenAlex, 2026-10-14) · [27](https://md.research.elamur.ai/free/en/papers/jeong-2009-semi-supervised-speech-act-recognition/cited-by/) in our library, of the 149,439 papers whose bibliographies we read

## Summary
Recognizing speech acts sentence by sentence in emails and forum threads is hard because these genres have almost no annotated data. The authors work around this by training on spoken-dialogue corpora that are already labeled (Switchboard-DAMSL and the Meeting Recorder Dialog Act corpus). They collapse both annotation schemes into 12 domain-independent tags derived from MRDA. They then adapt the resulting model with large unlabeled collections: Enron emails and TripAdvisor travel-forum posts.

The classifier relies on mined "subtree features": variable-length n-grams, part-of-speech patterns and fragments of dependency parses, chosen by boosting so that only the most discriminative ones survive. The paper compares two semi-supervised schemes, BOOTSTRAP and SEMIBOOST, against a source-only BASELINE and an in-domain SUPERVISED upper bound. Both semi-supervised methods beat the baseline, which reaches about 79 micro-F1. On forums, SEMIBOOST nearly matches the supervised model, and rare tags benefit the most. The work shows that discourse-level labels can be transferred across genres without annotating new target data.

## Key points
- **Data and annotation**
  - Labeled training data: SWBD with 96,553 training sentences and MRDA with 50,865.
  - Unlabeled data: 22,391 Enron emails and 55,743 TripAdvisor posts in 11,602 threads.
  - Evaluation sets: 40 BC3 email threads and 100 forum threads, with inter-annotator κ = 0.79 (EMAIL) and 0.73 (FORUM).
  - Tag set: 12 tags, keeping MRDA categories and dropping speech-only ones such as backchannel and floorgrabber. Examples include S, QY, QW, QR, QH, QO, AC, P, AA, A, R and U.
- **Micro-F1 results**

  | Model | EMAIL | FORUM |
  |---|---|---|
  | BASELINE | 78.87 | 78.93 |
  | BOOTSTRAP | 83.11 | 79.09 |
  | SEMIBOOST | 82.80 | 81.76 |
  | SUPERVISED (5-fold CV) | 90.95 | 83.67 |

- **Gains on rare tags (FORUM):** the polite mechanism tag (P) saw its error rate drop by 40%, the action motivator tag (AC) by 25%, and the or-clause question tag (QR) by 15%.
- **Why EMAIL lags:** the authors attribute the gap mainly to a mismatch between the adaptation data (Enron) and the evaluation data (BC3, W3C mail).
- **Thread-structure features:** forum sentences carry extra features absent from spoken data, namely initial vs. reply post, authorship across posts, and relative position of the sentence in its post.

## Interesting ideas
- Two telltale cues can decide a speech act: a sentence-initial Wh-adverb tag paired with a final "?" is enough to flag a Wh-question, so a few structural anchors at sentence edges can replace a large lexical feature space.
- Features are mined as subtrees mixing words, POS tags and partial dependency structures, and boosting keeps only the discriminative ones, which yields compact and inspectable models.
- Several source annotation schemes are mapped into one genre-neutral tag set, and categories that cannot occur in the target genre are dropped.
- The confusion matrix is sorted by tag semantics, which shows that most errors stay within the question family or within the statement family; this motivates two-stage or hierarchy-aware classification.
- An in-domain supervised cross-validation score serves as a ceiling for judging how much transfer from out-of-domain data actually achieves.

## For Mova

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

- **I1** (frames): crate prag could detect question types from tree-edge patterns (sentence-initial WRB, aux inversion, final punctuation) as explicit, explainable rules instead of word lists.
- **I2** (frames): speech-act and implicature classifiers could be built from small sets of mined UD subtree patterns, each of which can be shown as the reason for a decision.
- **I3** (learn): when merging treebanks or speech-act corpora for silver data, publish explicit mapping tables and prune labels impossible in written text before training.
- **I4** (frames): classify coarse illocution first (question vs. statement vs. directive) and refine afterwards, so that mistakes stay semantically close and easier to explain.
- **I5** (learn): evaluation of silver or transferred models should always report a small gold in-domain upper bound next to the source-only baseline.

*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

- [Microsoft Research Asia](https://md.research.elamur.ai/free/en/companies/microsoft-research-asia/) — Company, Beijing, China · 73 works · 60,539 citations
- [Pohang University of Science and Technology](https://md.research.elamur.ai/free/en/universities/pohang-university-of-science-and-technology/) — University, Pohang, South Korea · 7 works · 464 citations

📚 Shelves (4): [🖥️ 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

🧭 Topics (1): [💬 Speech act theory](https://md.research.elamur.ai/free/en/topics/speech-act-theory/) · 229 works

## Cite

Cite as: Minwoo Jeong, Chin-Yew Lin and Gary Geunbae Lee. 2009. Semi-supervised Speech Act Recognition in Emails and Forums. Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing.

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

```bibtex
@inproceedings{jeong2009semi,
  title = {{Semi-supervised Speech Act Recognition in Emails and Forums}},
  author = {Minwoo Jeong, Chin-Yew Lin and Gary Geunbae Lee},
  booktitle = {Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing},
  year = {2009},
  doi = {10.3115/1699648.1699671},
  url = {https://doi.org/10.3115/1699648.1699671}
}
```

- [← previous Sentence Boundary Detection and the Problem with the U.S.](https://md.research.elamur.ai/free/en/papers/gillick-2009-sentence-boundary-detection-problem-u/)
- [next → The Lie Detector: Explorations in the Automatic Recognition of Deceptive Language](https://md.research.elamur.ai/free/en/papers/mihalcea-2009-lie-detector-explorations-automatic-recognition/)
- [↑ back to papers of 2009](https://md.research.elamur.ai/free/en/papers/2009/)

## References

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

1. R. Ando and T. Zhang. 2005. A framework for learning predictive structures from multiple tasks and unlabeled data. Journal of Machine Learning Research, 6:1817–1853. [↗](https://doi.org/10.5555/1046920.1194905)
2. J. Ang, Y. Liu, and E. Shriberg. 2005. Automatic dialog act segmentation and classification in multiparty meetings. In Proceedings of ICASSP, pages 1061– 106. [find](https://scholar.google.com/scholar?q=J%2E+Ang%2C+Y%2E+Liu%2C+and+E%2E+Shriberg%2E+2005%2E+Automatic+dialog+act+segmentation+and+classification+in+multiparty+meetings%2E+In+Proceedings+of+ICASSP%2C+pages+1061%E2%80%93+106%2E)
3. J. Austin. 1962. How to Do Things With Words. Harvard Univ. Press, Cambridge, MA. [find](https://scholar.google.com/scholar?q=J%2E+Austin%2E+1962%2E+How+to+Do+Things+With+Words%2E+Harvard+Univ%2E+Press%2C+Cambridge%2C+MA%2E)
4. K.P. Bennett, A. Demiriz, and R. Maclin. 2002. Exploiting unlabeled data in ensemble methods. In Proceedings of ACM SIGKDD, pages 289–296. [find](https://scholar.google.com/scholar?q=K%2EP%2E+Bennett%2C+A%2E+Demiriz%2C+and+R%2E+Maclin%2E+2002%2E+Exploiting+unlabeled+data+in+ensemble+methods%2E+In+Proceedings+of+ACM+SIGKDD%2C+pages+289%E2%80%93296%2E)
5. W.W. Cohen, V.R. Carvalho, and T. Mitchell. 2004. [Learning to classify email into “speech acts”](https://md.research.elamur.ai/free/en/papers/cohen-2004-learning-classify-email-into-speech/). In Proceedings of EMNLP, pages 309–316.
6. R. Dhillon, S. Bhagat, H. Carvey, and E. Shriberg. 2004. Meeting recorder project: Dialog act labeling guide. Technical report, International Computer Science Institute. [↗](https://doi.org/10.21236/ada607947)
7. G. Druck, G. Mann, and A. McCallum. 2008. Learning from labeled features using generalized expectation criteria. In Proceedings of ACM SIGIR, pages 595–602. [find](https://scholar.google.com/scholar?q=G%2E+Druck%2C+G%2E+Mann%2C+and+A%2E+McCallum%2E+2008%2E+Learning+from+labeled+features+using+generalized+expectation+criteria%2E+In+Proceedings+of+ACM+SIGIR%2C+pages+595%E2%80%93602%2E)
8. D. Feng, E. Shaw, J. Kim, and E. H. Hovy. 2006. Learning to detect conversation focus of threaded discussions. In Proceedings of HLT-NAACL, pages 208–215. BASELINE BOOTSTRAP SEMIBOOST (c) Confusion matrix M. Galley, K. McKeown, J. Hirschberg, and E. Shriberg. 2004. [Identifying agreement and disagreement in conversational speech: use of bayesian networks to model pragmatic dependencies](https://md.research.elamur.ai/free/en/papers/galley-2004-identifying-agreement-disagreement-conversational-speech/). In Proceedings of ACL.
9. L. Gillick and S. Cox. 1989. Some statistical issues in the comparison of speech recognition algorithms. In Proceedings of ICASSP, pages 532–535. [find](https://scholar.google.com/scholar?q=L%2E+Gillick+and+S%2E+Cox%2E+1989%2E+Some+statistical+issues+in+the+comparison+of+speech+recognition+algorithms%2E+In+Proceedings+of+ICASSP%2C+pages+532%E2%80%93535%2E)
10. D. Jurafsky, E. Shriberg, and D. Biasca. 1997. Switchboard SWBD-DAMSL labeling project coder’s manual, draft 13. Technical report, Univ. of Colorado Institute of Cognitive Science. [find](https://scholar.google.com/scholar?q=D%2E+Jurafsky%2C+E%2E+Shriberg%2C+and+D%2E+Biasca%2E+1997%2E+Switchboard+SWBD%2DDAMSL+labeling+project+coder%E2%80%99s+manual%2C+draft+13%2E+Technical+report%2C+Univ%2E+of+Colorado+Institute+of+Cognitive+Science%2E)
11. T. Kudo and Y. Matsumoto. 2004. A boosting algorithm for classification of semi-structured text. In Proceedings of EMNLP, pages 301–308. [find](https://scholar.google.com/scholar?q=T%2E+Kudo+and+Y%2E+Matsumoto%2E+2004%2E+A+boosting+algorithm+for+classification+of+semi%2Dstructured+text%2E+In+Proceedings+of+EMNLP%2C+pages+301%E2%80%93308%2E)
12. S. Levinson. 1983. Pragmatics. Cambridge Univ. Press, Cambridge. [find](https://scholar.google.com/scholar?q=S%2E+Levinson%2E+1983%2E+Pragmatics%2E+Cambridge+Univ%2E+Press%2C+Cambridge%2E)
13. L. Mason, P. Bartlett, J. Baxter, and M. Frean. 2000. Functional gradient techniques for combining hypotheses. In A.J. Smola, P.L. Bartlett, B. Schölkopf, and D. Schuurmans, editors, Advances in Large Margin Classifiers, pages 221–246. MIT Press, Cambridge, MA. [↗](https://doi.org/10.7551/mitpress/1113.003.0017)
14. K. McKeown, L. Shrestha, and O. Rambow. 2007. Using question-answer pairs in extractive summarization of email conversations. In Proceedings of CI-CLing, volume 4394 of Lecture Notes in Computer Science, pages 542–550. [find](https://scholar.google.com/scholar?q=K%2E+McKeown%2C+L%2E+Shrestha%2C+and+O%2E+Rambow%2E+2007%2E+Using+question%2Danswer+pairs+in+extractive+summarization+of+email+conversations%2E+In+Proceedings+of+CI%2DCLing%2C+volume+4394+of+Lecture+Notes+in+Computer+Scien)
15. J. Nastri, J. Pe na, and J. T. Hancock. 2006. The construction of away messages: A speech act analysis. Journal of Computer-Mediated Communication, 11(4):article 7. [find](https://scholar.google.com/scholar?q=J%2E+Nastri%2C+J%2E+Pe+na%2C+and+J%2E+T%2E+Hancock%2E+2006%2E+The+construction+of+away+messages%3A+A+speech+act+analysis%2E+Journal+of+Computer%2DMediated+Communication%2C+11%284%29%3Aarticle+7%2E)
16. S. Ravi and J. Kim. 2007. Profiling student interactions in threaded discussions with speech act classifiers. In Proceedings of the AI in Education Conference. [find](https://scholar.google.com/scholar?q=S%2E+Ravi+and+J%2E+Kim%2E+2007%2E+Profiling+student+interactions+in+threaded+discussions+with+speech+act+classifiers%2E+In+Proceedings+of+the+AI+in+Education+Conference%2E)
17. C. Rosé, Y. Wang, Y. Cui, J. Arguello, K. Stegmann, A. Weinberger, and F. Fischer. 2008. Analyzing collaborative learning processes automatically: Exploiting the advances of computational linguistics in computer-supported collaborative learning. International Journal of Computer-Supported Collaborative Learning, 3(3):237–271. [↗](https://doi.org/10.1007/s11412-007-9034-0)
18. R.E. Schapire and Y. Singer. 1999. Improved boosting algorithms using confidence-rated predictions. Machine Learning, 37(3):297–336. [↗](https://doi.org/10.1023/a:1007614523901)
19. J. Searle. 1969. [Speech Acts](https://md.research.elamur.ai/free/en/papers/searle-1969-speech-acts/). Cambridge Univ. Press, Cambridge.
20. D. Shasha and K. Zhang. 1990. Fast algorithms for the unit cost editing distance between trees. Journal of Algorithms, 11(4):581–621. [↗](https://doi.org/10.1016/0196-6774(90)90011-3)
21. R. Soricut and D. Marcu. 2003. [Sentence level discourse parsing using syntactic and lexical information](https://md.research.elamur.ai/free/en/papers/soricut-2003-sentence-level-discourse-parsing-syntactic/). In Proceedings of NAACL-HLT, pages 149– 156.
22. A. Stolcke, K. Ries, N. Coccaro, E. Shriberg, R. Bates, D. Jurafsky, P. Taylor, R. Martin, C. Van Ess-Dykema, and M. Meteer. 2000. [Dialogue act modeling for automatic tagging and recognition of conversational speech](https://md.research.elamur.ai/free/en/papers/stolcke-2000-dialogue-act-modeling-automatic-tagging/). Computational Linguistics, 26(3):339–373. [↗](https://doi.org/10.1162/089120100561737)
23. G. Tur, D. Hakkani-Tür, and R. E. Schapire. 2005. Combining active and semi-supervised learning for spoken language understanding. Speech Communication, 45(2):171–186. [↗](https://doi.org/10.1016/j.specom.2004.08.002)
24. D. P. Twitchell, J. F. Nunamaker, and J. K. Burgoon. 2004. Using speech act profiling for deception detection. In Second Symposium on Intelligence and Security Informatics, volume 3073 of Lecture Notes in Computer Science, pages 403–410. [find](https://scholar.google.com/scholar?q=D%2E+P%2E+Twitchell%2C+J%2E+F%2E+Nunamaker%2C+and+J%2E+K%2E+Burgoon%2E+2004%2E+Using+speech+act+profiling+for+deception+detection%2E+In+Second+Symposium+on+Intelligence+and+Security+Informatics%2C+volume+3073+of+Lecture+Note)

*Updated 1 November 2026*