Elamur Research⌕
☰
home / papers / 2021 / Speaker and Time-aware Joint Contextual Learning for Dialogue-act Classification in Counselling Conversations

Speaker and Time-aware Joint Contextual Learning for Dialogue-act Classification in Counselling Conversations

Paper · 2021 · Ganeshan Malhotra, Abdul Waheed, Aseem Srivastava et al.

Summary

Counselling dialogues differ from task-oriented and chit-chat conversations. Therapist and patient exchange long turns, much of what the patient needs stays implicit, and the whole session history matters. The authors argue that recognising the dialogue act behind each turn is a basic building block for any therapy assistant, but no suitable data existed. They therefore built HOPE: about 12.9K utterances from 212 publicly available counselling videos on YouTube, covering CBT, child therapy, family therapy and similar settings. The transcripts were cleaned and labelled with twelve counselling-specific dialogue-act labels organised hierarchically. They also propose SPARTA, a transformer-based classifier that labels every utterance of a dialogue jointly. SPARTA combines three signals: who is speaking, a recency-weighted view of nearby turns, and a recurrent summary of the whole conversation so far. In their experiments SPARTA beats several baselines and reaches state-of-the-art on HOPE, and ablations support each module. The work matters as an early dataset and model for the language-understanding part of mental-health dialogue systems.

Key points

  • HOPE contains about 12.9K utterances from 212 counselling sessions, transcribed from public YouTube videos and manually annotated.
  • The annotation scheme is hierarchical, with 12 dialogue-act labels tailored to therapy, such as Greeting, Information Request/Delivery and Clarification Request/Delivery.
  • SPARTA treats the task as sequence classification at the dialogue level, predicting one act per utterance.
  • Global context comes from a GRU run over utterance representations. Local context comes from a novel time-aware attention: the current utterance cross-attends to a sliding-window memory of recent turns. Speaker dynamics are modelled explicitly.
  • Code and part of the dataset are released at github.com/LCS2-IIITD/SPARTA_WSDM2022.

Interesting ideas

  • I1. Design the dialogue-act inventory around the domain instead of reusing a generic tag set, and pair requests with deliveries (information vs. clarification) so the label itself records the move and its expected reply.
  • I2. Read a conversation through two context channels at once: a short sliding window of recent turns that the current turn attends to, weighted by recency, plus a running summary of the entire history. Different decisions depend on different distances.
  • I3. Treat the speaker identity and speaker changes as a first-class input rather than leaving them implicit in the text.
  • I4. Use the predictable shape of a session (greetings, then inquiry, then deeper probing, then advice) as a structural prior over which act comes next.
  • I5. Back each architectural component with an ablation that removes it, showing that every module earns its place.

For Mova

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

  • I1 (frames): crate prag could type speech acts as request/delivery pairs, so an answer frame explicitly links back to the question frame it satisfies.
  • I2 (world): event ordering and state tracking could keep a local window for immediate antecedents and a separate global store for long-lived facts, and log which one justified each update.
  • I3 (frames): coreference and speech-act assignment in dialogues or fables with quoted speech could condition on who holds the turn, which helps resolve "I"/"you" and turn-taking acts.
  • I4 (answer): QA over dialogues could check whether a predicted act sequence is plausible, flagging an answer that has no preceding request as a likely parsing error.
  • I5 (learn): every new repair rule or context feature could be evaluated with a switched-off variant as a negative control before it is 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

🎓Birla Institute of Technology and Science, PilaniPilāni · India8 works · 40,950 citations🎓Indraprastha Institute of Information Technology DelhiDelhi · India7 works · 1,233 citations🎓Birla Institute of Technology and Science, Pilani - Goa CampusSancoale · India2 works · 254 citations

📚 Shelves 6 Computer Science · 9,086 works Artificial Intelligence · 8,889 works Arts and Humanities · 9,296 works Language and Linguistics · 9,206 works Psychology · 798 works Social Psychology · 129 works

🧭 Topics 2 Dialogue systems · 401 works Speech act theory · 160 works

Cite

Cite as Ganeshan Malhotra, Abdul Waheed, Aseem Srivastava, Md Shad Akhtar and Tanmoy Chakraborty. 2021. Speaker and Time-aware Joint Contextual Learning for Dialogue-act Classification in Counselling Conversations. arXiv:2111.06647.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2111.06647 · arXiv 2111.06647

@misc{malhotra2021speaker,
  title = {{Speaker and Time-aware Joint Contextual Learning for Dialogue-act Classification in Counselling Conversations}},
  author = {Ganeshan Malhotra, Abdul Waheed, Aseem Srivastava, Md Shad Akhtar and Tanmoy Chakraborty},
  year = {2021},
  doi = {10.48550/arxiv.2111.06647},
  eprint = {2111.06647},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2111.06647}
}

References

22 references, 8 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Ali Ahmadvand, Jason Ingyu Choi, and Eugene Agichtein. 2019. Contextual dialogue act classification for open-domain conversational agents. In SIGIR. 1273– 1276. our card
  2. Inci M. Baytas, Cao Xiao, Xi Zhang, Fei Wang, Anil K. Jain, and Jiayu Zhou. 2017. Patient Subtyping via Time-Aware LSTM Networks. In SIGKDD. 65–74. find
  3. Christophe Cerisara, Somayeh Jafaritazehjani, Adedayo Oluokun, and Hoa T Le. 2018. Multi-task dialog act and sentiment recognition on Mastodon. In COLING. 745–754. ↗ our card
  4. Zheqian Chen, Rongqin Yang, Zhou Zhao, Deng Cai, and Xiaofei He. 2018. Dialogue act recognition via crf-attentive structured network. In SIGIR. 225–234. our card
  5. Kyunghyun Cho, Bart van Merriënboer, Caglar Gulcehre, Dzmitry Bahdanau, Fethi Bougares, Holger Schwenk, and Yoshua Bengio. 2014. Learning Phrase Representations using RNN Encoder–Decoder for Statistical Machine Translation. In EMNLP. 1724–1734. ↗
  6. Jacob Cohen. 1960. A Coefficient of Agreement for Nominal Scales. Educational and Psychological Measurement 20, 1 (1960), 37–46. ↗
  7. Pierre Colombo, Emile Chapuis, Matteo Manica, Emmanuel Vignon, Giovanna Varni, and C. Clavel. 2020. Guiding attention in Sequence-to-sequence models for Dialogue Act prediction. In AAAI. ↗ our card
  8. Alexis Conneau, Holger Schwenk, Loïc Barrault, and Yann Lecun. 2017. Very Deep Convolutional Networks for Text Classification. In EACL. 1107–1116. our card
  9. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. In NAACL. 4171–4186. our card
  10. Alex Fine, Patrick Crutchley, Jenny Blase, Joshua Carroll, and Glen Coppersmith. 2020. Assessing population-level symptoms of anxiety, depression, and suicide risk in real time using NLP applied to social media data. In Proceedings of the Fourth Workshop on Natural Language Processing and Computational Social Science. 50–54. find
  11. J. J. Godfrey, E. C. Holliman, and J. McDaniel. 1992. SWITCHBOARD: telephone speech corpus for research and development. In [Proceedings] ICASSP-92: 1992 IEEE International Conference on Acoustics, Speech, and Signal Processing, Vol. 1. 517–520 vol.1. https://doi.org/10.1109/ICASSP.1992.225858 ↗
  12. Jonathan Gratch, Ron Artstein, Gale M Lucas, Giota Stratou, Stefan Scherer, Angela Nazarian, Rachel Wood, Jill Boberg, David DeVault, Stacy Marsella, et al. 2014. The distress analysis interview corpus of human and computer interviews.. In LREC. 3123–3128. find
  13. Sergio Grau, Emilio Sanchis, Maria Jose Castro, and David Vilar. 2004. Dialogue act classification using a Bayesian approach. In 9th Conference Speech and Computer. find
  14. Kai Hua, Zhiyuan Feng, Chongyang Tao, Rui Yan, and Lu Zhang. 2020. Learning to Detect Relevant Contexts and Knowledge for Response Selection in Retrieval-Based Dialogue Systems. In CIKM. 525–534. find
  15. Zornitsa Kozareva and Sujith Ravi. 2019. ProSeqo: Projection Sequence Networks for On-Device Text Classification. In EMNLP-IJCNLP. 3894–3903. find
  16. K. Kretzschmar, H. Tyroll, G. Pavarini, A. Manzini, and I. Singh. 2019. Can Your Phone Be Your Therapist? Young People’s Ethical Perspectives on the Use of Fully Automated Conversational Agents (Chatbots) in Mental Health Support. Biomed Inform Insights 11 (2019), 1178222619829083. find
  17. John D. Lafferty, Andrew McCallum, and Fernando C. N. Pereira. 2001. Conditional Random Fields: Probabilistic Models for Segmenting and Labeling Sequence Data. In ICML. Morgan Kaufmann Publishers Inc., San Francisco, CA, USA, 282–289. find
  18. Y. LeCun, B. Boser, J. S. Denker, D. Henderson, R. E. Howard, W. Hubbard, and L. D. Jackel. 1989. Backpropagation Applied to Handwritten Zip Code Recognition. Neural Comput. (Dec. 1989), 541–551. ↗
  19. Ji Young Lee and Franck Dernoncourt. 2016. Sequential Short-Text Classification with Recurrent and Convolutional Neural Networks. In NAACL. Association for Computational Linguistics, San Diego, California, 515–520. our card
  20. Maxwell Levis, Christine Leonard Westgate, Jiang Gui, Bradley V. Watts, and Brian Shiner. 2020. Natural language processing of clinical mental health notes may add predictive value to existing suicide risk models. Psychological Medicine (2020), 1–10. ↗
  21. Yanran Li, Hui Su, Xiaoyu Shen, Wenjie Li, Ziqiang Cao, and Shuzi Niu. 2017. DailyDialog: A Manually Labelled Multi-turn Dialogue Dataset. In IJCNLP. ↗ our card
  22. Pengfei Liu, Xipeng Qiu, and Xuanjing Huang. 2016. Recurrent Neural Network for Text Classification with Multi-Task Learning. In IJCAI. AAAI Press, 2873–2879. *Table 6: Training hyperparameters* LSTM for conversational sentiment analysis. Neural Networks 133 (2021), 40 – 56. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/malhotra-2021-speaker-time-aware-joint-contextual/