Elamur Research⌕
☰
home / papers / 2006 / Dimensions in Dialogue Act Annotation

Dimensions in Dialogue Act Annotation

Paper · 2006 · Harry Bunt

Summary

The paper addresses a conceptual weakness in multidimensional dialogue act annotation: what should count as a "dimension". The literature mixes two readings. One treats a dimension as a kind of information. The other treats it as a group of tags that exclude each other. DAMSL uses both. Bunt shows with short dialogue fragments that mutual exclusivity is a poor test. DAMSL lets one utterance be tagged as both an assertion and a question, which is incoherent. Yet a single utterance can genuinely question one thing while answering another. He instead defines a dimension as an aspect of taking part in a dialogue that dialogue acts can address independently, and he uses empirically checkable criteria to decide which aspects qualify. The result is 11 dimensions, among them Task/Activity, Auto- and Allo-Feedback, Turn Management and Social Obligations Management. Questions, informs, answers, requests and offers are recast as general-purpose functions usable in any dimension. This gives a principled way to constrain multiple tagging and to extend the taxonomy, and it fed into the LIRICS and ISO semantic annotation standardisation work.

Key points

  • Clark and Popescu-Belis (2004) found that annotators of meeting data used only 220 DAMSL tag combinations out of roughly 4 million possible ones.
  • DAMSL is organised into four layers, which cover functions that look ahead in the dialogue, functions that look back to earlier utterances, the status of the information, and the level it belongs to. Its 8 forward-looking and 4 backward-looking "dimensions" include Statement, Info-request, Answer, Agreement and Understanding.
  • The paper's 11 proposed dimensions are:
    • Task/Activity
    • Auto-Feedback and Allo-Feedback
    • Turn Management and Time Management
    • Contact Management
    • Own and Partner Communication Management
    • Topic Management
    • Dialogue Structuring
    • Social Obligations Management
  • General-purpose functions fall into four families: information-seeking, information-providing (including answers and confirmations), commissive and directive.
  • The work is tied to the EU project LIRICS and ISO TC 37/SC 4/TDG 3. The scheme is being used in a multi-agent dialogue manager and in an annotation usability study.

Interesting ideas

  • I1. Separate what an act does (its communicative function, e.g. question) from which aspect of the interaction it targets (its dimension, e.g. feedback); this avoids duplicating "question about the task" vs. "question about hearing".
  • I2. Allow at most one value per dimension while allowing many dimensions per utterance; this gives built-in consistency checks on multi-labelling.
  • I3. Define a category by whether it can be addressed independently, not by whether its labels exclude each other. This is a reusable test for whether any label set deserves to be an axis.
  • I4. Treat indirect and multifunctional utterances, such as a question that also answers a previous question, as normal and annotatable rather than as noise.
  • I5. Even social formulas like "Thank you" carry hidden feedback, here a signal that the previous answer was perceived, understood and accepted.

For Mova

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

  • I1 (frames): speech acts in prag could carry a function label plus a separate target-dimension slot, so "Did you say Thursday?" is a check aimed at auto-feedback rather than at the task.
  • I2 (learn): when witness parsers or LLM teachers propose speech-act labels, two values in one dimension can be rejected automatically as a contradiction, while labels in different dimensions are kept together.
  • I3 (world): the same test can audit Mova's state and property axes in the ontology, merging or splitting attributes that fail independence.
  • I4 (answer): fable and tale QA readers could record that a character's question also answers or declines an earlier request, which supports explainable answers about dialogue intent.
  • I5 (frames): implicature rules in prag could attach a positive auto-feedback inference to thanking acts, which the world model can use as evidence that information was accepted.

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

🎓Tilburg UniversityTilburg · Netherlands50 works · 6,429 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 works

🧭 Topics 2Research subjects and languages the work is about.Dialogue systems · 624 worksSpeech act theory · 229 works

Cite

Cite as Harry Bunt. 2006. Dimensions in Dialogue Act Annotation. Proceedings of the Fifth International Conference on Language Resources and Evaluation (LREC'06).

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

@inproceedings{bunt2006dimensions,
  title = {{Dimensions in Dialogue Act Annotation}},
  author = {Harry Bunt},
  booktitle = {Proceedings of the Fifth International Conference on Language Resources and Evaluation (LREC'06)},
  year = {2006},
  doi = {10.63317/2jchd238vsa7},
  url = {https://doi.org/10.63317/2jchd238vsa7}
}

References

18 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Ahrenberg, L., N. Dahlbäck & A. Jönsson (1995) Codings Schemes for Studies of Natural Language Dialogue. In: Working Notes from the AAAI Spring Symposium, Stanford. find
  2. Akker, R. Op den, H. Bunt, S. Keizen & B. van Schooten (2005) From Question Answering to Spoken: Dialogue: Towards an Information Search Assistant for Interactive Multimodal Information Extraction. In Proceedings Ninth European Conference on Speech Communication and Technology Interspeech 2006, Lisbon, 2793-2797. find
  3. Allen, J. & M. Core (1997) DAMSL: Dialogue Act Markup in Several Layers (Draft 2.1). Technical Report, Multiparty Discourse Group, Discourse Resource Initiative, September/October 1997. http://www.cs.rochester.edu/research/cisd/resources/damsl/RevisedManual/ ↗
  4. Allwood, J., J. Nivre & E. Ahlsén (1994) Semantics and Spoken Language Manual for Coding Interaction Management. Report from the HSFR project Semantik och talsprak Gothenburg University. find
  5. Beun, R.J.. (1989) The Recognition of Declarative Questions in Information Dialogues. Ph.D. thesis, Tilburg University. find
  6. Bunt, H. (1995) Dynamic Interpretation and Dialogue Theory. in M.Taylor, D. Bouwhuis & F. Néel (eds.) The Structure of Multimodal Dialogue, Vol. 2. Amsterdam: Benjamins, 139–166. find
  7. Bunt, H. (2000) Dialogue pragmatics and context specification. In H. Bunt & W. Black (eds.) Abduction, Belief and Context in Dialogue. Studies in Computational Pragmatics. Amsterdam: Benjamins, 81–150. ↗
  8. Bunt, H. (2005) A Framework for Dialogue Act Specification. 4th Joint ISO-SIGSEM Workshop on the Representation of Multimodal Semantic Information, Tilburg, January 2005. http://let.uvt.nl/research/ti/sigsem/wg ↗
  9. Bunt, H. & Y. Girard (2005) Designing an open, multidimensional dialogue act taxonomy. In C. Gardent & B. Gaiffe (eds.) DIALOR’05, Proceedings of the Ninth Workshop on the Semantics and Pragmatics of Dialogue, Nancy, June 2005, 37–44. find
  10. Bunt, H. & S. Keizer (2005) Dialogue semantics links annotation for context representation. In Joint TALK/AMI Workshop on Standards or Multimodal Dialogue Context, Edinburgh, December 2005. http://homepages.inf.ed.ac.uk/olemon/standcon-SOI.html ↗
  11. Bunt, H. & L. Romary (2004) Standardization in Multimodal Content Representation: Some Methodological Issues. Proceedings 4 th International Conference on Language Resources and Evaluation LREC 2004, June, Lisbon, 2219–2222. find
  12. Bunt, H. & A. Schiffrin (2006) Methodological aspects of semantic annotation. In Proceedings 5 th International Conference on Language Resources and Evaluation LREC 2006, Genova, May 2006. find
  13. Carletta, J., A. Isard, S. Isard, J.Kowtko & G. Doherty-Sneddon (1996) HCRC dialogue structure coding manual. Technical Report HCRC/TR-82. find
  14. Clark, D. & A. Popescu-Belis, A. (2004) Multi-level Dialogue Act Tags. In Proccedings of SIGDial 2004, 5th SIGDial Workshop on Discourse and Dialogue, Cambridge, MA, 163–170.
  15. Core, M., M. Ishizaki, J. Moore, C. Nakatani, N. Reithinger, D. Traum & S. Tutiya (1998) Report of The Third Workshop of the Discourse Resource Initiative, May 18-22, 1998, Chiba University, Chiba, Japan. http://cogsci.l.chiba-u.ac.jp/$\sim$tutiya/Publications/DRI98Report.pdf ↗
  16. Geertzen, J. & H. Bunt (2006, submitted) Evaluating DIT: Annotator agreement in a multidimensional dialogue act annotation scheme. find
  17. Petukhova, V. V. (2005) Multidimensional interaction of multimodal dialogue acts in meetings. MA thesis, Tilburg University. find
  18. Soria, C. & V. Pirrelli (2003) A multi-level annotation meta-scheme for dialogue acts. In A. Zampolli, N. Calzolari & L. Cignoni (eds.) Computational Linguistics in Pisa. Linguistica Computazionale, Special Issue, vol. XVIII-XIX, Pisa-Roma: IEPI, 865-900. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/bunt-2006-dimensions-dialogue-act-annotation/