Elamur Research⌕
☰
home / papers / 2020 / Will I Sound Like Me? Improving Persona Consistency in Dialogues through Pragmatic Self-Consciousness

Will I Sound Like Me? Improving Persona Consistency in Dialogues through Pragmatic Self-Consciousness

Summary

Persona-based chatbots often say things that conflict with the persona they were given. Earlier fixes usually rely on NLI annotations or separately trained consistency modules, which are costly to build. The authors first show that three strong generators (ControlSeq2Seq, TransferTransfo, Blender) on the Dialogue NLI evaluation set pick contradictory candidates more often than ground-truth replies. These models also assign lower perplexity to contradicting words than to words the persona supports.

Inspired by the psychology of public self-consciousness, the authors add an imaginary listener at decoding time using the Rational Speech Acts (RSA) framework. The speaker favours tokens that would help this listener identify its own persona rather than distractor personas. They extend RSA in two ways:

  • Distractor Memory, a learned method for choosing distractors instead of picking them manually or at random.
  • A new world-prior update based on the literal listener, which better accumulates information across decoding steps.

Without extra labels or retraining, the method reduces contradictions and improves accuracy on Dialogue NLI and PersonaChat. It also improves context consistency beyond persona, for example on EmpatheticDialogue. This matters because the approach can be added to existing generators at low cost.

Key points

  • The Dialogue NLI evaluation set gives each dialogue 31 candidates: 10 entailing, 10 neutral, 10 contradictory and 1 ground-truth. Models are scored with Hits@1, Entail@1, Neutral@1 and Contradict@1.
  • In base models, contradictory words had an average perplexity of about 106.7, versus 280.1 for their ground-truth counterparts. This suggests the models act more like plain language models than persona-conditioned ones.
  • The self-conscious agent (S1) is built on top of a base speaker (S0) and works with any persona-based generator. It needs no consistency labels, NLI models or additional training.
  • Updating the world prior with the literal listener L0 helped significantly, whereas using L1 performed about the same as a uniform prior. A listener rationality β ≤ 1 worked best.
  • Speaker rationality α controls how much persona text is copied into replies. Very large values (e.g. α = 10) produce ungrammatical, persona-fragment outputs.

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

Cited by

35 papers of our library cite it (of the 140673 whose bibliographies we read); 4 of them have a page here, 31 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2020.emnlp-main.65 · arXiv 2004.05816

@inproceedings{kim2020will,
  title = {{Will I Sound Like Me? Improving Persona Consistency in Dialogues through Pragmatic Self-Consciousness}},
  author = {Kim, Hyunwoo and Kim, Byeongchang and Kim, Gunhee},
  booktitle = {Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP)},
  year = {2020},
  doi = {10.18653/v1/2020.emnlp-main.65},
  eprint = {2004.05816},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.18653/v1/2020.emnlp-main.65}
}

References

40 references, 16 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Peter Anderson, Basura Fernando, Mark Johnson, and Stephen Gould. 2016. SPICE: Semantic Propositional Image Caption Evaluation. In ECCV, pages 382–398. Springer. find our card
  2. Demis Hassabis, R Nathan Spreng, Andrei A Rusu, Clifford A Robbins, Raymond A Mar, and Daniel L Schacter. 2013. Imagine All the People: How the Brain Creates and Uses Personality Models to Predict Behavior. Cerebral Cortex, 24(8):1979–1987. find
  3. Łukasz Kaiser, Ofir Nachum, Aurko Roy, and Samy Bengio. 2017. Learning to Remember Rare Events. In ICLR. find
  4. Jacob Andreas and Dan Klein. 2016. Reasoning about Pragmatics with Neural Listeners and Speakers. In EMNLP. find our card
  5. Reuben Cohn-Gordon and Noah Goodman. 2019. Lost in Machine Translation: A Method to Reduce Meaning Loss. In NAACL-HLT. find
  6. Reuben Cohn-Gordon, Noah Goodman, and Christopher Potts. 2018. Pragmatically Informative Image Captioning With Character-level Inference. In NAACL-HLT. find our card
  7. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. In NAACL-HLT. find
  8. Emily Dinan, Varvara Logacheva, Valentin Malykh, Alexander Miller, Kurt Shuster, Jack Urbanek, Douwe Kiela, Arthur Szlam, Iulian Serban, Ryan Lowe, et al. 2019. The Second Conversational Intelligence Challenge (ConvAI2). arXiv:1902.00098. find our card
  9. Kevin Doherty and Barry R Schlenker. 1991. Self-Consciousness and Strategic Self-Presentation. Journal of Personality, 59(1):1–18. find
  10. Nouha Dziri, Ehsan Kamalloo, Kory W Mathewson, and Osmar Zaiane. 2019. Evaluating Coherence in Dialogue Systems Using Entailment. In NAACL-HLT. find
  11. Allan Fenigstein, Michael F Scheier, and Arnold H Buss. 1975. Public and Private Self-Consciousness: Assessment and Theory. Journal of Consulting and Clinical Psychology, 43(4):522. find
  12. Leon Festinger. 1962. A Theory of Cognitive Dissonance, volume 2. Stanford University Press. find
  13. Michael C Frank and Noah D Goodman. 2012. Predicting Pragmatic Reasoning in Language Games. Science, 336(6084):998–998. find our card
  14. Daniel Fried, Jacob Andreas, and Dan Klein. 2017. Unified Pragmatic Models for Generating and Following Instructions. In NAACL-HLT. find our card
  15. Daniel Fried, Ronghang Hu, Volkan Cirik, Anna Rohrbach, Jacob Andreas, Louis-Philippe Morency, Taylor Berg-Kirkpatrick, Kate Saenko, Dan Klein, and Trevor Darrell. 2018. Speaker-follower models for vision-and-language navigation. In NeurIPS. find
  16. Alison Gopnik and Henry M Wellman. 1992. Why the Child’s Theory of Mind Really is a Theory. Mind & Language, 7(1-2):145–171. find
  17. Diederik Kingma and Jimmy Ba. 2015. Adam: A Method for Stochastic Optimization. In ICLR. find
  18. Ilia Kulikov, Alexander Miller, Kyunghyun Cho, and Jason Weston. 2019. Importance of Search and Evaluation Strategies in Neural Dialogue Modeling. In INLG. find
  19. Jiwei Li, Michel Galley, Chris Brockett, Georgios P Spithourakis, Jianfeng Gao, and Bill Dolan. 2016. A Persona-Based Neural Conversation Model. In ACL. find our card
  20. Margaret Li, Stephen Roller, Ilia Kulikov, Sean Welleck, Y-Lan Boureau, Kyunghyun Cho, and Jason Weston. 2020. Don’t Say That! Making Inconsistent Dialogue Unlikely with Unlikelihood Training. In ACL. find
  21. Chin-Yew Lin. 2004. Rouge: A Package for Automatic Evaluation of Summaries. In Text Summarization Branches Out, pages 74–81. find
  22. Qian Liu, Yihong Chen, Bei Chen, Jian-Guang Lou, Zixuan Chen, Bin Zhou, and Dongmei Zhang. 2020. You Impress Me: Dialogue Generation via Mutual Persona Perception. In ACL. find
  23. Andrea Madotto, Zhaojiang Lin, Chien-Sheng Wu, and Pascale Fung. 2019. Personalizing Dialogue Agents via Meta-Learning. In ACL. find
  24. Junhua Mao, Jonathan Huang, Alexander Toshev, Oana Camburu, Alan L Yuille, and Kevin Murphy. 2016. Generation and Comprehension of Unambiguous Object Descriptions. In CVPR. find our card
  25. A. H. Miller, W. Feng, A. Fisch, J. Lu, D. Batra, A. Bordes, D. Parikh, and J. Weston. 2017. ParlAI: A Dialog Research Software Platform. arXiv:1705.06476. find our card
  26. Alec Radford, Karthik Narasimhan, Tim Salimans, and Ilya Sutskever. 2018. Improving Language Understanding with Unsupervised Learning. Technical report, Technical report, OpenAI. find
  27. Hannah Rashkin, Eric Michael Smith, Margaret Li, and Y-Lan Boureau. 2019. Towards Empathetic Open-domain Conversation Models: A New Benchmark and Dataset. In ACL. find
  28. Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M Smith, et al. 2020. Recipes for Building an Open-Domain Chatbot. arXiv:2004.13637. find our card
  29. A Abigail See, Stephen Roller, Douwe Kiela, and Jason Weston. 2019. What Makes a Good Conversation? How Controllable Attributes Affect Human Judgments. In NAACL-HLT. find our card
  30. Results on Variants of Distractor Selection (Section 4.2) Model Sheng Shen, Daniel Fried, Jacob Andreas, and Dan Klein. 2019. Pragmatically Informative Text Generation. In NAACL-HLT. find our card
  31. Hits@1 ↑ Entail@1 ↑ Contradict@1 ↓ ControlSeq2Seq (See et al., 2019) Random 8.5 32.8 Nearest 7.6 32.8 Farthest 9.4 33.6 BERT-Classifier 9.2 33.6 BERT-Ranker 9.6 33.3 DM 11.1 36.0 find
  32. Michael Shum, Stephan Zheng, Wojciech Kryściński, Caiming Xiong, and Richard Socher. 2019. Sketch-Fill-AR: A Persona-Grounded Chit-Chat Generation Framework. arXiv:1910.13008. 37.6 36.5 35.4 35.6 35.1 28.2 *Table 8: Quantitative results of the proposed Distractor Memory (DM) and other distractor s… find
  33. Haoyu Song, Wei-Nan Zhang, Jingwen Hu, and Tiu Liu. 2019. Generating Persona Consistent Dialogues by Exploiting Natural Language Inference. arXiv:1911.05889. — [arXiv:1911.05889](https://arxiv.org/abs/1911.05889) find
  34. Ramakrishna Vedantam, Samy Bengio, Kevin Murphy, Devi Parikh, and Gal Chechik. 2017. Context-Aware Captions from Context-Agnostic Supervision. In CVPR. find
  35. Sean Welleck, Jason Weston, Arthur Szlam, and Kyunghyun Cho. 2019. Dialogue Natural Language Inference. In ACL. find our card
  36. Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pier-ric Cistac, Tim Rault, Rémi Louf, Morgan Funtow-icz, et al. 2019a. Transformers: State-of-the-art Natural Language Processing. arXiv:1910.03771. find our card
  37. Thomas Wolf, Victor Sanh, Julien Chaumond, and Clement Delangue. 2019b. TransferTransfo: A Transfer Learning Approach for Neural Network based Conversational Agents. arXiv:1901.08149. — [arXiv:1901.08149](https://arxiv.org/abs/1901.08149) find
  38. Sina Zarrieß and David Schlangen. 2019. Know What You Don’t Know: Modeling a Pragmatic Speaker that Refers to Objects of Unknown Categories. In ACL. find our card
  39. Saizheng Zhang, Emily Dinan, Jack Urbanek, Arthur Szlam, Douwe Kiela, and Jason Weston. 2018. Personalizing Dialogue Agents: I Have a Dog, Do You Have Pets Too? In ACL. find our card
  40. Yizhe Zhang, Xiang Gao, Sungjin Lee, Chris Brockett, Michel Galley, Jianfeng Gao, and Bill Dolan. 2019. Consistent Dialogue Generation with Self-supervised Feature Learning. arXiv:1903.05759. — [arXiv:1903.05759](https://arxiv.org/abs/1903.05759) find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/kim-2020-will-i-sound-like-me/