Elamur Research⌕
☰
home / papers / 2013 / A Generative Joint, Additive, Sequential Model of Topics and Speech Acts in Patient-Doctor Communication

A Generative Joint, Additive, Sequential Model of Topics and Speech Acts in Patient-Doctor Communication

Paper · 2013 · Byron C. Wallace, Thomas A. Trikalinos, M. Barton Laws et al.

Summary

Clinical conversation research needs each utterance coded for two things at once: what it is about (topic) and what the speaker does with it (speech act). Earlier statistical work treated speech acts alone, as HMM-style sequence labeling that ignores subject matter. Wallace et al. propose JAS, a generative model that is Joint, Additive and Sequential. It builds on SAGE-style additive log-linear components. Word emissions and transitions to the next topic and to the next speech act are each a background log-frequency plus separate offsets for topic, for speech act, and for their interaction. This avoids the sparse Cartesian product of the two label sets. Current topic and current act are conditionally independent given the previous pair, and the coupling flows through that previous pair. Trained on annotated patient-physician visits, JAS beats a univariate Markov-Multinomial baseline that predicts topics and speech acts separately. It also reproduces the counts behind a randomized trial on antiretroviral (ARV) adherence counseling more faithfully than the baseline. Automated coding could therefore substitute for some expensive manual annotation in health-communication studies.

Key points

  • The baseline is the Markov-Multinomial (MM) model from Stolcke et al. (2000):
    • first-order label transitions;
    • a naive-Bayes multinomial over words;
    • maximum-likelihood counts and Viterbi decoding;
    • one independent model each for topics and for speech acts.
  • A χ² test shows topics and speech acts are not independent (p < .001), which motivates joint modeling.
  • JAS extends SAGE (Eisenstein et al., 2011) to supervised sequential tasks. The authors call this the first such extension.
    • Background terms are estimated from observed proportions.
    • The other components are fitted by gradient-based optimization.
  • The label scheme uses 10 high-level speech acts. Stolcke et al. used 42, of which only 8 exceeded 1% frequency. Example topics include Biomedical, Logistics and Socializing.
  • In the reanalysis of 58 control and intervention visits, JAS-predicted medians and quartiles of ARV/information-giving utterances matched the human counts better than MM in every case but one, where the two models tied.

Interesting ideas

  • I1. Instead of modeling a crossed label space directly, express each probability as a log-linear sum of a background term, per-factor offsets and an interaction offset; this keeps sparse combinations estimable.
  • I2. Let the previous (topic, act) pair drive both the next topic and the next act, rather than conditioning simultaneous labels on each other; this gives correlation without a circular dependency.
  • I3. Judge a model by whether automated labels reproduce a downstream study's aggregate conclusion (trial medians/quartiles), not only by per-item accuracy.
  • I4. Use a small set of coarse, high-frequency speech-act classes rather than a long tail of fine ones, because most fine classes are too rare to learn.
  • I5. Positing a latent intent that generates both what is said and how it is said is a clean conceptual decomposition.

For Mova

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

  • I1 (frames): Speech-act and topic or frame-type scores in prag could be scored additively with an explicit interaction term, so rare act×frame combinations still get sensible, inspectable weights.
  • I2 (frames): Dialogue-level speech-act tracking could condition the next utterance's act on the prior act plus its frame/topic, an explainable sequential rule.
  • I3 (learn): Evaluation could include checking that silver-labeled corpora yield the same aggregate statistics as gold, catching systematic bias that accuracy alone hides.
  • I4 (frames): The prag speech-act inventory could keep a coarse top layer for decisions and treat fine subtypes as optional refinements.
  • I5 (world): Speaker goals could be represented as world-level entities that explain both an utterance's frame content and its speech act, linking pragmatics to the text's world.

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

🎓Brown UniversityProvidence · United States58 works · 25,310 citations

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 9,085 worksArtificial Intelligence · 8,888 worksArts and Humanities · 9,296 worksLanguage and Linguistics · 9,206 works

🧭 Topics 1Research subjects and languages the work is about.Speech act theory · 160 works

Cite

Cite as Byron C. Wallace, Thomas A. Trikalinos, M. Barton Laws, Ira B. Wilson and Eugene Charniak. 2013. A Generative Joint, Additive, Sequential Model of Topics and Speech Acts in Patient-Doctor Communication. Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing.

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

@inproceedings{wallace2013generative,
  title = {{A Generative Joint, Additive, Sequential Model of Topics and Speech Acts in Patient-Doctor Communication}},
  author = {Byron C. Wallace, Thomas A. Trikalinos, M. Barton Laws, Ira B. Wilson and Eugene Charniak},
  booktitle = {Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing},
  year = {2013},
  doi = {10.18653/v1/d13-1182},
  url = {https://doi.org/10.18653/v1/d13-1182}
}

References

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

  1. Jeremy Ang, Yang Liu, and Elizabeth Shriberg. 2005. Automatic dialog act segmentation and classification in multiparty meetings. In Proc. ICASSP, volume 1, pages 1061–1064. find
  2. Daniel Angus, Bernadette Watson, Andrew Smith, Cindy Gallois, and Janet Wiles. 2012. Visualising conversation structure across time: Insights into effective doctor-patient consultations. PloS one, 7(6). ↗
  3. John Langshaw Austin. 1955. How to do things with words, volume 88. Harvard University Press. find
  4. Matthew Brand, Nuria Oliver, and Alex Pentland. 1997. Coupled hidden Markov models for complex action recognition. In Computer Vision and Pattern Recognition, 1997. Proceedings., 1997 IEEE Computer Society Conference on, pages 994–999. IEEE. find
  5. Alexander Clark and Andrei Popescu-Belis. 2004. Multi-level dialogue act tags. In Proc. SIGdial, pages 163–170.
  6. William W Cohen, Vitor R Carvalho, and Tom M Mitchell. 2004. Learning to classify email into speech acts. In Proceedings of EMNLP, volume 4. sn.
  7. Julia Cretchley, Cindy Gallois, Helen Chenery, and Andrew Smith. 2010. Conversations between carers and people with schizophrenia: a qualitative analysis using leximancer. Qualitative Health Research, 20(12):1611–1628. ↗
  8. Jacob Eisenstein and Regina Barzilay. 2008. Bayesian unsupervised topic segmentation. In Proceedings of the Conference on Empirical Methods in Natural Language Processing, pages 334–343. Association for Computational Linguistics.
  9. J. Eisenstein, A. Ahmed, and E.P. Xing. 2011. Sparse additive generative models of text. In Proceedings of ICML, pages 1041–1048. find
  10. Oliver Ferschke, Iryna Gurevych, and Yevgen Chebotar. 2012. Behind the article: Recognizing dialog acts in wikipedia talk pages. In Proceedings of the 13th Conference of the European Chapter of the Association for Computational Linguistics (EACL 2012), pages 777– 786. Citeseer.
  11. Jeffrey D Ford and Laurie W Ford. 1995. The role of conversations in producing intentional change in organizations. Academy of Management Review, pages 541–570. ↗
  12. George Forman and Martin Scholz. 2010. Apples-to-apples in cross-validation studies: pitfalls in classifier performance measurement. volume 12, pages 49–57. ACM. find
  13. Michel Galley, Kathleen McKeown, Eric Fosler-Lussier, and Hongyan Jing. 2003. Discourse segmentation of multi-party conversation. In Proceedings of the 41st Annual Meeting on Association for Computational Linguistics, pages 562–569. Association for Computational Linguistics.
  14. Zoubin Ghahramani and Michael I Jordan. 1997. Factorial hidden Markov models. Machine learning, 29(2-3):245–273. ↗
  15. Jade Goldstein, Andrew Kwasinski, Paul Kingsbury, R Sabin, and Albert McDowell. 2006. Annotating subsets of the enron email corpus. In Proceedings of the Third Conference on Email and Anti-Spam. Citeseer. find
  16. P-Y Hsueh and Johanna D Moore. 2006. Automatic topic segmentation and labeling in multiparty dialogue. In Spoken Language Technology Workshop, 2006. IEEE, pages 98–101. IEEE. find
  17. Richard S Irwin and Naomi D Richardson. 2006. Patient-focused careusing the right tools. CHEST Journal, 130(1 suppl):73S–82S. find
  18. Su Nam Kim, Lawrence Cavedon, and Timothy Baldwin. 2010. Classifying dialogue acts in one-on-one live chats. In Proceedings of the 2010 Conference on Empirical Methods in Natural Language Processing, pages 862–871. Association for Computational Linguistics.
  19. Su Nam Kim, Lawrence Cavedon, and Timothy Baldwin. 2012. Classifying dialogue acts in multi-party live chats.
  20. Michael Barton Laws, Ylisabyth S Bradshaw, Steven A Safren, Mary Catherine Beach, Yoojin Lee, William Rogers, and Ira B Wilson. 2011a. Discussion of sexual risk behavior in HIV care is infrequent and appears ineffectual: a mixed methods study. AIDS and Behavior, 15(4):812–822. ↗
  21. Michael Barton Laws, Lauren Epstein, Yoojin Lee, William Rogers, Mary Catherine Beach, and Ira B Wilson. 2011b. The association of visit length and measures of patient-centered communication in HIV care: A mixed methods study. Patient Education and Counseling, 85(3):e183–e188. ↗
  22. Michael Barton Laws, Mary Catherine Beach, Yoojin Lee, William H Rogers, Somnath Saha, P Todd Korthuis, Victoria Sharp, and Ira B Wilson. 2012. Provider-patient adherence dialogue in HIV care: results of a multisite study. AIDS and Behavior, pages 1–12. ↗
  23. Gregory Makoul. 2001. Essential elements of communication in medical encounters: the kalamazoo consensus statement. Academic Medicine, 76(4):390–393. ↗
  24. Marie W Meteer, Ann A Taylor, Robert MacIntyre, and Rukmini Iyer. 1995. Dysfluency annotation stylebook for the switchboard corpus. University of Pennsylvania. find
  25. Lucille ML Ong, Johanna CJM De Haes, Alaysia M Hoos, and Frits B Lammes. 1995. Doctor-patient communication: a review of the literature. Social science & medicine, 40(7):903–918. ↗
  26. Michael Paul and Mark Dredze. 2012. Factorial lda: Sparse multi-dimensional text models. In Advances in Neural Information Processing Systems 25, pages 2591–2599. find
  27. Michael J. Paul, Byron C. Wallace, and Mark Dredze. 2013. What affects patient (dis)satisfaction? analyzing online doctor ratings with a joint topic-sentiment model. In AAAI Workshop on Expanding the Boundaries of Health Informatics Using AI (HIAI). find
  28. Michael J Paul. 2012. Mixed membership Markov models for unsupervised conversation modeling. In Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pages 94–104. Association for Computational Linguistics. find
  29. C Raymond Perrault and James F Allen. 1980. A plan-based analysis of indirect speech acts. Computational Linguistics, 6(3-4):167–182. ↗
  30. Ashequl Qadir and Ellen Riloff. 2011. Classifying sentences as speech acts in message board posts. In Proceedings of the Conference on Empirical Methods in Natural Language Processing, pages 748–758. Association for Computational Linguistics.
  31. Lawrence Rabiner and B Juang. 1986. An introduction to hidden Markov models. ASSP Magazine, IEEE, 3(1):4–16. ↗
  32. Debra Roter and Susan Larson. 2002. The roter interaction analysis system (rias): utility and flexibility for analysis of medical interactions. Patient education and counseling, 46(4):243–251. ↗
  33. John R Searle. 1969. Speech acts: An essay in the philosophy of language. Cambridge university press. ↗
  34. John R Searle. 1985. Expression and meaning: Studies in the theory of speech acts. Cambridge University Press. find
  35. Andreas Stolcke, Elizabeth Shriberg, Rebecca Bates, Noah Coccaro, Daniel Jurafsky, Rachel Martin, Marie Meteer, Klaus Ries, Paul Taylor, and Carol Van Ess-Dykema. 1998. Dialog act modeling for conversational speech. In AAAI Spring Symposium on Applying Machine Learning to Discourse Processing, pages 98–105. ↗
  36. Andreas Stolcke, Klaus Ries, Noah Coccaro, Elizabeth Shriberg, Rebecca Bates, Daniel Jurafsky, Paul Taylor, Rachel Martin, Carol Van Ess-Dykema, and Marie Meteer. 2000. Dialogue act modeling for automatic tagging and recognition of conversational speech. Computational linguistics, 26(3):339–373. ↗
  37. Charles Sutton, Andrew McCallum, and Khashayar Rohanimanesh. 2007. Dynamic conditional random fields: Factorized probabilistic models for labeling and segmenting sequence data. The Journal of Machine Learning Research, 8:693–723. ↗
  38. Carol Teutsch. 2003. Patient-doctor communication. The medical clinics of North America, 87(5):1115. ↗
  39. David R Traum and Staffan Larsson. 2003. The information state approach to dialogue management. In Current and new directions in discourse and dialogue, pages 325–353. Springer. ↗
  40. Jurgen Van Gael, Yee Whye Teh, and Zoubin Ghahramani. 2008. The infinite factorial hidden Markov model. In Neural Information Processing Systems, volume 21. find
  41. Ira B Wilson, M Barton Laws, Steven A Safren, Yoojin Lee, Minyi Lu, William Coady, Paul R Skolnik, and William H Rogers. 2010. Provider-focused intervention increases adherence-related dialogue, but does not improve antiretroviral therapy adherence in persons with HIV. Journal of acquired immune deficiency syndromes, 53(3):338. ↗
  42. Renxian Zhang, Dehong Gao, and Wenjie Li. 2012. Towards scalable speech act recognition in twitter: Tackling insufficient training data. EACL 2012, page 18.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/wallace-2013-generative-joint-additive-sequential-model/