---
title: "Unsupervised Approach for Dialogue Act Classification (2008)"
description: "The paper asks whether dialogue act (DA) labels can be found without first fixing a tag set and… 37.5% accuracy. 12 references."
authors: ["Kiyonori Ohtake"]
year: 2008
venue: "Proceedings of the 22nd Pacific Asia Conference on Language, Information and Computation"
source: "https://aclanthology.org/Y08-1047/"
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/ohtake-2008-unsupervised-approach-dialogue-act-classification/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/ohtake-2008-unsupervised-approach-dialogue-act-classification/)

[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/page/2/) / Unsupervised Approach for Dialogue Act Classification

# Unsupervised Approach for Dialogue Act Classification

**Authors:** Kiyonori Ohtake · **Year:** 2008 · **Venue:** [Proceedings of the 22nd Pacific Asia Conference on Language, Information and Computation](https://md.research.elamur.ai/free/en/venues/paclic/) ([PACLIC 2008](https://md.research.elamur.ai/free/en/venues/paclic/2008/)) · **Publisher:** [De La Salle University, Manila, Philippines](https://md.research.elamur.ai/free/en/publishers/de-la-salle-university-manila-philippines/) · **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/Y08-1047/) · [PDF](https://aclanthology.org/Y08-1047.pdf) · [OpenAlex](https://openalex.org/W2178125101)

**Citations:** 3 citations (OpenAlex, 2026-10-14)

## Summary
The paper asks whether dialogue act (DA) labels can be found without first fixing a tag set and hand-annotating a corpus. Both steps are costly and depend on the domain. The author works with Japanese travel dialogues from the ATR Dialogue Database. Each utterance, together with the utterance before it, becomes a sparse vector of morpheme unigram and bigram counts. PLSI compresses this vector into a short topic vector, and K-means groups the topic vectors.

A rule-based paraphraser runs before modelling. It maps the many Japanese politeness and honorific endings onto one standard form, to address the pragmatic gap between surface form and intent.

Clusters were compared with manual labels on a small annotated subset. Some labels, such as THANK and WH-Q, lined up well with single clusters. Paraphrasing clearly helped with action requests. The catch-all label INFORM spread across many clusters. In a tiny nearest-centroid experiment, paraphrasing unexpectedly lowered accuracy. The work matters as an early look at label-free DA induction and at how normalizing surface forms affects it.

## Key points
- Data: the ATR Dialogue Database has 1,983 dialogues (83,052 utterances). PLSI was trained on 1,970 dialogues, and 13 dialogues (489 utterances) were annotated with the Tanaka and Yokoo tag set for evaluation. That set has 26 tags, of which 16 occurred. Utterances with several labels kept only their last tag.
- Features are pairs of morpheme basic form and POS. Numbers and proper names are reduced to their POS only.
- Separability was measured as the mean cosine between per-label average topic vectors (lower is better). For 10, 50, 100, 200 and 300 latent variables it was 0.607, 0.334, 0.288, 0.290 and 0.275, so 100 variables were judged sufficient. Tempered EM used the parameter 0.9.
- The paraphraser rewrote 56,027 utterances. It made ACT-REQ utterances much more coherent in clusters, while the generic INFORM label stayed scattered.
- Semi-supervised nearest-centroid test (12 dialogues for training, 32 test utterances):
  - 37.5% accuracy without paraphrasing;
  - 21.9% with paraphrasing;
  - 31.3% after the paraphrased test utterances were relabelled.

## Interesting ideas
- A conservative rule-based normalizer can collapse many polite or indirect ways of asking someone to do something into one canonical form before intent classification, so that surface variety does not split one act into many.
- Representation quality can be checked without a downstream task: average the vectors per gold label and measure how similar these label centroids are to each other, which also gives a cheap way to choose the model size.
- The unit for classifying an utterance's act is the utterance together with its predecessor, because what an act means depends on what came just before it.
- A rewrite step can quietly make gold labels wrong: relabelling the paraphrased test utterances recovered much of the lost accuracy, showing that the transformation had changed the meaning.
- Numbers and proper names are abstracted to their POS class, which shrinks sparsity without discarding the fact that such a slot exists.

## For Mova

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

- **I1** (frames): prag could map indirect requests ("could you possibly…", "would you mind…") to a canonical request frame, but keep the original wording alongside so the cues the normalizer removes are not lost.
- **I2** (learn): the same centroid-similarity score could compare frame or embedding variants and flag when added capacity stops improving separation.
- **I3** (frames): speech-act labelling in prag could always look at the previous turn or sentence, for example to tell an answer from a fresh statement.
- **I4** (learn): any pass that rewrites silver or gold data should re-check that the labels still hold, with a negative control comparing the original and rewritten versions.
- **I5** (tokens): the tagger could expose a slot-abstracted lemma view (NUM, PROPN) for statistical components, while the real values stay attached for the world model.

*Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.*

📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 18,614 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 17,965 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 17,450 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 17,296 works

🧭 Topics (2): [🗨️ Dialogue systems](https://md.research.elamur.ai/free/en/topics/dialogue-systems/) · 654 works · [💬 Speech act theory](https://md.research.elamur.ai/free/en/topics/speech-act-theory/) · 229 works

## Cite

Cite as: Kiyonori Ohtake. 2008. Unsupervised Approach for Dialogue Act Classification. Proceedings of the 22nd Pacific Asia Conference on Language, Information and Computation.

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

```bibtex
@inproceedings{ohtake2008unsupervised,
  title = {{Unsupervised Approach for Dialogue Act Classification}},
  author = {Kiyonori Ohtake},
  booktitle = {Proceedings of the 22nd Pacific Asia Conference on Language, Information and Computation},
  year = {2008},
  url = {https://aclanthology.org/Y08-1047/}
}
```

- [← previous The Vav-Prefixed Verb Forms in Elementary Hebrew Grammar](https://md.research.elamur.ai/free/en/papers/cook-2008-vav-prefixed-verb-forms-elementary/)
- [next → Exploitation of an Arabic Language Resource for Machine Translation Evaluation: using Buckwalter-based Lookup Tool to Augment CMU Alignment Algorithm](https://md.research.elamur.ai/free/en/papers/voss-2008-exploitation-arabic-language-resource-machine/)
- [↑ back to papers of 2008](https://md.research.elamur.ai/free/en/papers/2008/page/2/)

## References

12 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Allen, James and Mark Core. 1997. Draft of DAMSL: Dialog act markup in several layers. Technical Report, Discourse Research Initiative. [find](https://scholar.google.com/scholar?q=Allen%2C+James+and+Mark+Core%2E+1997%2E+Draft+of+DAMSL%3A+Dialog+act+markup+in+several+layers%2E+Technical+Report%2C+Discourse+Research+Initiative%2E)
2. Blei, David M., Andrew Y. Ng, and Michael I. Jordan. 2003. [Latent Dirichlet allocation](https://md.research.elamur.ai/free/en/papers/blei-2003-latent-dirichlet-allocation/). Journal of Machine Learning Research, 3:993-1022. [↗](https://doi.org/10.5555/944919.944937)
3. Duda, Richard O., Peter E. Hart, and David G. Stork. 2000. Pattern Classification. A Wiley-Interscience Publication. [find](https://scholar.google.com/scholar?q=Duda%2C+Richard+O%2E%2C+Peter+E%2E+Hart%2C+and+David+G%2E+Stork%2E+2000%2E+Pattern+Classification%2E+A+Wiley%2DInterscience+Publication%2E)
4. Hinarejos, Carlos D. Martínez, Ramón Granell, and José Miguel Benedí. 2006. [Segmented and unsegmented dialogue-act annotation with statistical dialogue models](https://md.research.elamur.ai/free/en/papers/hinarejos-2006-segmented-unsegmented-dialogue-act-annotation/). In Proceedings of the COLING/ACL 2006, pp. 563-570.
5. Hofmann, Thomas. 1999. Probabilistic Latent Semantic Indexing. In Proceedings of the 22 nd Annual ACM Conference on Research and Development in Information Retrieval, pp. 50-57. [find](https://scholar.google.com/scholar?q=Hofmann%2C+Thomas%2E+1999%2E+Probabilistic+Latent+Semantic+Indexing%2E+In+Proceedings+of+the+22+nd+Annual+ACM+Conference+on+Research+and+Development+in+Information+Retrieval%2C+pp%2E+50%2D57%2E)
6. Morimoto, Tsuyoshi, N. Uratani, T. Takezawa, O. Furuse, Y. Sobashima, H. Iida, A. Nakamura, Y. Sagisaka, N. Higuchi and Y. Yamazaki. 1994. A speech and language database for speech translation research. In Proceedings of ICSLP ’94, pp. 1791-1794. [find](https://scholar.google.com/scholar?q=Morimoto%2C+Tsuyoshi%2C+N%2E+Uratani%2C+T%2E+Takezawa%2C+O%2E+Furuse%2C+Y%2E+Sobashima%2C+H%2E+Iida%2C+A%2E+Nakamura%2C+Y%2E+Sagisaka%2C+N%2E+Higuchi+and+Y%2E+Yamazaki%2E+1994%2E+A+speech+and+language+database+for+speech+translation+researc)
7. Ohtake, Kiyonori and Kazuhide Yamamoto. 2001. Paraphrasing honorifics. In Workshop Proceedings of Automatic Paraphrasing: Theories and Applications (NLPRS2001 Post-conference Workshop), pp. 13-20. [find](https://scholar.google.com/scholar?q=Ohtake%2C+Kiyonori+and+Kazuhide+Yamamoto%2E+2001%2E+Paraphrasing+honorifics%2E+In+Workshop+Proceedings+of+Automatic+Paraphrasing%3A+Theories+and+Applications+%28NLPRS2001+Post%2Dconference+Workshop%29%2C+pp%2E+13%2D20%2E)
8. Ohtake, Kiyonori. 2005. Evaluating contextual dependency of paraphrases using a latent variable model. In Proceedings of the Third International Workshop on Paraphrasing (IWP2005) conjunct with IJCNLP 2005, pp. 65-72. [find](https://scholar.google.com/scholar?q=Ohtake%2C+Kiyonori%2E+2005%2E+Evaluating+contextual+dependency+of+paraphrases+using+a+latent+variable+model%2E+In+Proceedings+of+the+Third+International+Workshop+on+Paraphrasing+%28IWP2005%29+conjunct+with+IJCNLP)
9. Serafin, Riccardo and Brbara Di Eugenio. 2004. [FLSA: Extending latent semantic analysis with features for dialogue act classification](https://md.research.elamur.ai/free/en/papers/serafin-2004-flsa-extending-latent-semantic-analysis/). In Proceedings of the 42 nd Meeting of the Association for Computational Linguistics (ACL’04), pp. 692-699.
10. Stolcke, Andreas, 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.
11. Tanaka, Hideki and Akio Yokoo. 1999. An efficient statistical speech act type tagging system for speech translation systems. In Proceedings of the Thirty Seventh Annual Meeting of the Association for Computational Linguistics (ACL’99), pp. 381-388. [find](https://scholar.google.com/scholar?q=Tanaka%2C+Hideki+and+Akio+Yokoo%2E+1999%2E+An+efficient+statistical+speech+act+type+tagging+system+for+speech+translation+systems%2E+In+Proceedings+of+the+Thirty+Seventh+Annual+Meeting+of+the+Association+for+)
12. Zhao, Ying and George Karypis. 2005. Hierarchical clustering algorithms for document datasets . Data Mining and Knowledge Discovery, 10: 141-168. [find](https://scholar.google.com/scholar?q=Zhao%2C+Ying+and+George+Karypis%2E+2005%2E+Hierarchical+clustering+algorithms+for+document+datasets+%2E+Data+Mining+and+Knowledge+Discovery%2C+10%3A+141%2D168%2E)

*Updated 1 November 2026*