Elamur Research⌕
☰
home / papers / 2017 / A Hierarchical Neural Model for Learning Sequences of Dialogue Acts

A Hierarchical Neural Model for Learning Sequences of Dialogue Acts

Summary

The paper addresses dialogue act (DA) classification: assigning a DA label to each utterance in a conversation, where the label must be predicted immediately, without access to later turns. Earlier neural work, such as Shen and Lee (2016), classified utterances in isolation and lost conversational context. The authors propose HA-RNN, a hierarchical model with two nested LSTM layers. An inner RNN reads the tokens of each utterance, and an attention layer condenses them into one vector. An outer RNN then tracks the dialogue across utterances. The attention weights depend on each token's embedding, the conversation state so far, and the embedding of the previous DA. Decoding is greedy, proceeding left to right. The model was evaluated only on transcriptions of Switchboard and MapTask. It beat earlier transcription-only systems on both corpora, and ablations showed that each component contributes. The work matters because it shows that modelling dialogue structure explicitly helps DA tagging, and it identifies exposure bias as a central remaining obstacle.

Key points

  • Data:
    • Switchboard has 1155 conversations with 42 clustered tags, split 1115/21/19 following Stolcke et al. (2000).
    • MapTask has 128 dialogues with 13 tags, split randomly 80/10/10 into 103/12/13 conversations.
  • Main result: HA-RNN reaches 74.5% test accuracy on Switchboard and 63.3% on MapTask.
    • On Switchboard this beats Kalchbrenner and Blunsom (73.9%), Shen and Lee (72.6%) and the HMM of Stolcke et al. (71.0%).
  • Ablations: removing the utterance RNN reduces accuracy to 71.8% on Switchboard and 56.9% on MapTask. Removing the conversation RNN, or either context signal feeding the attention, also hurts performance.
  • Exposure bias: a variant called wDA2DA links the previous DA directly to the current prediction.
    • With gold (oracle) previous DAs it becomes the best model, at 75.0% on Switchboard and 65.0% on MapTask.
    • With greedy decoding it drops to 71.5% and 58.2%, because errors in earlier predictions propagate.
  • Attention analysis: the most-attended tokens fit each DA type intuitively, for example "move" for Instruct and "nope" for Reply n. However, Acknowledge and Reply y share very similar high-focus tokens.

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

Topics: Speech act theory

Cited by

14 papers of our library cite it (of the 140673 whose bibliographies we read); 3 of them have a page here, 11 not yet.

Cite

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

@inproceedings{tran2017hierarchical,
  title = {{A Hierarchical Neural Model for Learning Sequences of Dialogue Acts}},
  author = {Tran, Quan Hung and Zukerman, Ingrid and Haffari, Gholamreza},
  booktitle = {Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics: Volume 1, Long Papers},
  year = {2017},
  doi = {10.18653/v1/e17-1041},
  url = {https://doi.org/10.18653/v1/e17-1041}
}

References

25 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Di-amos, et al. 2015. Deep speech 2: End-to-end speech recognition in English and Mandarin. arXiv preprint arXiv:1512.02595. ↗
  2. Andrew McCallum, Dayne Freitag, and Fernando C.N. Pereira. 2000. Maximum entropy Markov models for information extraction and segmentation. In ICML’00 – Proceedings of the 17th International Conference on Machine Learning, pages 591–598, Stanford, California. find
  3. Anne H. Anderson, Miles Bader, Ellen Gurman Bard, Elizabeth Boyle, Gwyneth Doherty, Simon Garrod, Steven Isard, Jacqueline Kowtko, Jan McAllister, Jim Miller, et al. 1991. The HCRC MapTask corpus. Language and speech, 34(4):351–366. find
  4. Marc’Aurelio Ranzato, Sumit Chopra, Michael Auli, and Wojciech Zaremba. 2015. Sequence level training with recurrent neural networks. arXiv preprint arXiv:1511.06732. ↗ our card
  5. Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Bengio. 2014. Neural machine translation by jointly learning to align and translate. arXiv preprint arXiv:1409.0473. ↗ our card
  6. Sheng-syun Shen and Hung-yi Lee. 2016. Neural attention models for sequence classification: Analysis and application to key term extraction and dialogue act detection. arXiv preprint arXiv:1604.00077. ↗
  7. Peter F. Brown, Vincent J. Della Pietra, Stephen A. Della Pietra, and Robert L. Mercer. 1993. The mathematics of statistical machine translation: Parameter estimation. Computational Linguistics, 19(2):263–311. find
  8. Nitish Srivastava, Geoffrey Hinton, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov. 2014. Dropout: A simple way to prevent neural networks from overfitting. Journal of Machine Learning Research, 15(1):1929–1958. find
  9. John Duchi, Elad Hazan, and Yoram Singer. 2011. Adaptive subgradient methods for online learning and stochastic optimization. Journal of Machine Learning Research, 12(Jul):2121–2159. find
  10. Andreas Stolcke, Noah Coccaro, Rebecca Bates, Paul Taylor, Carol Van Ess-Dykema, Klaus Ries, Elizabeth Shriberg, Daniel Jurafsky, Rachel Martin, and Marie Meteer. 2000. Dialogue act modeling for automatic tagging and recognition of conversational speech. Computational linguistics, 26(3):339–373. ↗ our card
  11. Mark Gales and Steve Young. 2008. The application of hidden Markov models in speech recognition. Foundations and trends in signal processing, 1(3):195– 304. ↗
  12. Dinoj Surendran and Gina-Anne Levow. 2006. Dialog act tagging with Support Vector Machines and hidden Markov models. In Proceedings of Interspeech 2006, pages 1950–1953, Pittsburgh, Pennsylvania. find
  13. Björn Gambäck, Fredrik Olsson, and Oscar Täckström. 2011. Active learning for dialogue act classification. In Proceedings of Interspeech 2011, pages 1329– 1332, Florence, Italy. find
  14. Ilya Sutskever, Oriol Vinyals, and Quoc V. Le. 2014. Sequence to sequence learning with neural networks. In Advances in neural information processing systems, pages 3104–3112. ↗
  15. Alex Graves. 2013. Generating sequences with recurrent neural networks. arXiv preprint arXiv:1308.0850. ↗
  16. Paul A. Taylor, Simon King, Steve D. Isard, and Helen Wright Hastie. 1998. Intonation and dialogue context as constraints for speech recognition. Language and Speech, 41(3-4):493–512. find
  17. Sepp Hochreiter and Jürgen Schmidhuber. 1997. Long short-term memory. Neural computation, 9(8):1735–1780. ↗
  18. Kristina Toutanova and Christopher D Manning. 2000. Enriching the knowledge sources used in a maximum entropy part-of-speech tagger. In Proceedings of the 2000 Joint SIGDAT conference on Empirical Methods in Natural Language Processing and Very Large Corpora, pages 63–70, Hong Kong. find
  19. Fatema N. Julia, Khan M. Iftekharuddin, and Atiq U. Islam. 2010. Dialog act classification using acoustic and discourse information of MapTask data. International Journal of Computational Intelligence and Applications, 9(4):289–311. ↗
  20. Daniel Jurafsky, Elizabeth Shriberg, and Debra Bi-asca. 1997. Switchboard SWBD-DAMSL Shallow-Discourse-Function Annotation Coders Manual, Draft 13. Technical report, Stanford University. find
  21. Kristina Toutanova, Dan Klein, Christopher D. Manning, and Yoram Singer. 2003. Feature-rich part-of-speech tagging with a cyclic dependency network. find
  22. In NAACL’2003 – Proceedings of the 2003 Conference of the North American Chapter of the Association for Computational Linguistics on Human Language Technology, pages 173–180, Edmonton, Canada. find
  23. Nick Webb and Michael Ferguson. 2010. Automatic extraction of cue phrases for cross-corpus dialogue act classification. In Proceedings of the 23rd International Conference on Computational Linguistics, pages 1310–1317, Uppsala, Sweden. find
  24. Nick Webb, Mark Hepple, and Yorick Wilks. 2005. Dialogue act classification based on intra-utterance features. In Proceedings of the AAAI Workshop on Spoken Language Understanding, Pittsburgh, Pennsylvania. find
  25. Helen Wright Hastie, Massimo Poesio, and Steve D. Isard. 2002. Automatically predicting dialogue structure using prosodic features. Speech Communication, 36(1):63–79. ↗

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/tran-2017-hierarchical-neural-model-learning-sequences/