Elamur Research⌕
☰
home / papers / 2026 / On the Structure of Address in Multi-Party Dialogue: From Discrete Labels to Continuous Levels

On the Structure of Address in Multi-Party Dialogue: From Discrete Labels to Continuous Levels

Summary

Addressee detection in multi-party dialogue is usually treated as choosing one discrete target, either a single participant or the whole group. Under that framing, annotator disagreement counts as noise. Mori et al. question this assumption. They use the Japanese Teidan corpus from Inoue et al. (2025), which contains 36 casual three-person conversations among lab members. Three annotators independently labelled each turn with one of eight addressee categories, and listener gaze was also annotated.

From these judgments the authors derive two listener-level representations:

  • a binary "addressed" label based on majority vote;
  • a continuous address level inferred with a latent-variable model.

Bayesian mixed-effects regressions then relate each representation to next-speaker selection, listener gaze toward the speaker, and backchannel production. Both representations are associated with all three behaviors, but the continuous levels fit the data better. This suggests that address is graded rather than strictly categorical. For systems talking with several users, addressee modeling could become a continuous estimation task tied to listener behavior, not only to turn-taking.

Key points

  • Data: 12 triads each discussed 3 topics, giving 36 dialogues. Total length is about 3 h 38 min, with 3,121 turns. After one pilot dialogue, 3,078 turns in 35 dialogues were triple-annotated.

  • Label set: The scheme follows Kadota et al. (2024): A/B/C (one person), E (everyone), W (whoever), N (none), O (other) and U (unknown). E was the most frequent raw label (3,529).

  • Annotator disagreement: Individual labels (A, B, C) co-occurred with the same label in over 50% of annotator pairs, and with E in about 30%. The boundary between individual and group address is therefore fairly stable, but not perfectly so.

  • Model coefficients: Positive effects appeared in every model.

    OutcomeBinary labelContinuous level
    Next speaker0.721.06
    Gaze proportion (5,744 observations)0.771.15
    Backchannels (4,920 observations)0.180.31

    The 95% credible intervals exclude zero for all six coefficients.

  • Random effects: Listener identity was the largest source of variation for turn-taking and backchannels. For gaze, turn-level variation dominated.

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

Fields: Linguistics & languages

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.sigdial-1.47

@inproceedings{mori2026structure,
  title = {{On the Structure of Address in Multi-Party Dialogue: From Discrete Labels to Continuous Levels}},
  author = {Mori, Taiga and Inoue, Koji and Lala, Divesh and Kawahara, Tatsuya},
  booktitle = {Proceedings of the 27th Annual Meeting of the Special Interest Group on Discourse and Dialogue},
  year = {2026},
  doi = {10.18653/v1/2026.sigdial-1.47},
  url = {https://doi.org/10.18653/v1/2026.sigdial-1.47}
}

References

15 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Suman V Ravuri and Andreas Stolcke. 2014. Neural network models for lexical addressee detection. In Interspeech, pages 298–302. find
  2. Koji Inoue, Divesh Lala, Mikey Elmers, Keiko Ochi, and Tatsuya Kawahara. 2025. An LLM benchmark for addressee recognition in multi-modal multi-party dialogue. In Proceedings of the 15th International Workshop on Spoken Dialogue Systems Technology, pages 330–334. find
  3. Harvey Sacks, Emanuel A Schegloff, and Gail Jefferson. 1974. A simplest systematics for the organization of turn-taking for conversation. language, 50(4):696– 735. find
  4. Natasa Jovanovic, Rieks op den Akker, and Anton Nijholt. 2006. Addressee identification in face-to-face meetings. In 11th Conference of the European Chapter of the Association for Computational Linguistics, pages 169–176. find
  5. Keisuke Kadota, Seima Oyama, and Yasuharu Den. 2024. Annotation of addressing behavior in multi-party conversation. In 2024 27th Conference of the Oriental COCOSDA International Committee for the Co-ordination and Standardisation of Speech Databases and Assessment Techniques (O-COCOSDA), pages 1–6.… find
  6. Motoki Sato, Hiroki Ouchi, and Yuta Tsuboi. 2018. Addressee and response selection for multilingual conversation. In Proceedings of the 27th International Conference on Computational Linguistics, pages 3631–3644. find
  7. Elizabeth Shriberg, Andreas Stolcke, Dilek Hakkani-Tür, and Larry Heck. 2012. Learning when to listen: Detecting system-addressed speech in human-human-computer dialog. In Interspeech, pages 334–337. find
  8. Michael Smithson and Jay Verkuilen. 2006. A better lemon squeezer? maximum-likelihood regression with beta-distributed dependent variables. Psychological methods, 11(1):54. find
  9. Thao Minh Le, Nobuyuki Shimizu, Takashi Miyazaki, and Koichi Shinoda. 2018. Deep learning based multi-modal addressee recognition in visual scenes with utterances. arXiv preprint arXiv:1809.04288. — [arXiv:1809.04288](https://arxiv.org/abs/1809.04288) find
  10. TJ Tsai, Andreas Stolcke, and Malcolm Slaney. 2015. A study of multimodal addressee detection in human-human-computer interaction. IEEE Transactions on Multimedia, 17(9):1550–1561. find
  11. Taiga Mori, Koji Inoue, Divesh Lala, Keiko Ochi, and Tatsuya Kawahara. 2026. Analysing next speaker prediction in multi-party conversation using multimodal large language models. In Proceedings of the 16th International Workshop on Spoken Dialogue System Technology, pages 83–94. find
  12. Rui Zhang, Honglak Lee, Lazaros Polymenakos, and Dragomir Radev. 2018. Addressee and response selection in multi-party conversations with speaker interaction RNNs. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 32. find
  13. Yukiko I Nakano, Naoya Baba, Hung-Hsuan Huang, and Yuki Hayashi. 2013. Implementation and evaluation of a multimodal addressee identification mechanism for multiparty conversation systems. In Proceedings of the 15th ACM International Conference on Multimodal Interaction, pages 35–42. find
  14. Harm op den Akker and Rieks op den Akker. 2009. Are you being addressed? - real-time addressee detection to support remote participants in hybrid meetings. In Proceedings of the SIGDIAL 2009 Conference, pages 21–28, London, UK. Association for Computational Linguistics. find
  15. Pengcheng Zhu, Wei Zhou, Kuncai Zhang, Yuankai Ma, and Haiqing Chen. 2023. Robust learning for multi-party addressee recognition with discrete addressee codebook. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 571–578. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/papers/mori-2026-structure-address-multi-party-dialogue/