AI helps you reading Science

AI generates interpretation videos

AI extracts and analyses the key points of the paper to generate videos automatically


pub
Go Generating

AI Traceability

AI parses the academic lineage of this thesis


Master Reading Tree
Generate MRT

AI Insight

AI extracts a summary of this paper


Weibo:
We instantiate these face act exchanges in the context of persuasion and propose a dataset of 296 conversations annotated with face acts

Keeping Up Appearances: Computational Modeling of Face Acts in Persuasion Oriented Discussions

EMNLP 2020, pp.7473-7485, (2020)

Cited by: 0|Views197
Full Text
Bibtex
Weibo

Abstract

The notion of face refers to the public self-image of an individual that emerges both from the individual’s own actions as well as from the interaction with others. Modeling face and understanding its state changes throughout a conversation is critical to the study of maintenance of basic human needs in and through interaction. Grounded i...More

Code:

Data:

0
Introduction
  • Politeness principles, displayed in practice in dayto-day language usage, play a central role in shaping human interaction.
  • One of the most influential politeness theories from linguistics is proposed in (Brown and Levinson, 1978), in which a detailed exposition is offered of the individual actions whose cumulative effect results in saving face and losing face, along with a consideration of cost
  • Using this framework, it is possible to analyze how interlocutors make decisions about where and how these devices should be used based on an intricate cost-benefit analysis (Brown et al, 1987).
  • The authors refer to these component actions here as face acts
Highlights
  • Politeness principles, displayed in practice in dayto-day language usage, play a central role in shaping human interaction
  • The difference in F1 is less noticeable between EE and All, despite an additional face act (8 for All), possibly due to the increase in labelled data
  • We present a generalized computational framework based on the notion of face to operationalize face dynamics in conversations
  • We instantiate these face act exchanges in the context of persuasion and propose a dataset of 296 conversations annotated with face acts
  • We develop computational models for predicting face acts as well as observe the impact of these predicted face acts on the donation outcome
  • We use BERT-HIGRU-f as the model since it achieves the highest performance on face act prediction
  • We believe that our work may be extended to language generation in chatbots for producing more polite language to mediate face threats
Methods
  • 3.1 Face act prediction

    The authors model the task of computationally operationalizing face acts as a dialogue act classification task.
  • The authors adopt a modified hierarchical neural network architecture of Jiao et al (2019) that leverages both the contextualized utterance embedding and the previous conversational context for classification.
  • The authors hereby adopt this as the foundation architecture for the work and refer to the instantiation of the architecture as BERT-HIGRU
Results
  • The authors put forward the following research questions and attempt to answer the same.

    Q1.
  • The authors put forward the following research questions and attempt to answer the same.
  • How well does BERT-HIGRU predict face acts?
  • (Section 5.3).
  • Model Performance: The authors present the results of the models for face act prediction in Table 3 and glean several insights.
  • The authors observe that all models consistently perform better for ER than EE due to the more skewed distribution of EE and the presence of an extra face act (6 for ER vs 7 for EE).
  • The difference in F1 is less noticeable between EE and All, despite an additional face act (8 for All), possibly due to the increase in labelled data
Conclusion
  • The authors present a generalized computational framework based on the notion of face to operationalize face dynamics in conversations.
  • The authors instantiate these face act exchanges in the context of persuasion and propose a dataset of 296 conversations annotated with face acts.
  • One important limitation of the current work is the assumption that all face acts have the same intensity/ranking.
  • The authors intend to instantiate the proposed framework to other domains such as teacher/student conversations and other types of discourse such as social media narratives
Summary
  • Introduction:

    Politeness principles, displayed in practice in dayto-day language usage, play a central role in shaping human interaction.
  • One of the most influential politeness theories from linguistics is proposed in (Brown and Levinson, 1978), in which a detailed exposition is offered of the individual actions whose cumulative effect results in saving face and losing face, along with a consideration of cost
  • Using this framework, it is possible to analyze how interlocutors make decisions about where and how these devices should be used based on an intricate cost-benefit analysis (Brown et al, 1987).
  • The authors refer to these component actions here as face acts
  • Methods:

    3.1 Face act prediction

    The authors model the task of computationally operationalizing face acts as a dialogue act classification task.
  • The authors adopt a modified hierarchical neural network architecture of Jiao et al (2019) that leverages both the contextualized utterance embedding and the previous conversational context for classification.
  • The authors hereby adopt this as the foundation architecture for the work and refer to the instantiation of the architecture as BERT-HIGRU
  • Results:

    The authors put forward the following research questions and attempt to answer the same.

    Q1.
  • The authors put forward the following research questions and attempt to answer the same.
  • How well does BERT-HIGRU predict face acts?
  • (Section 5.3).
  • Model Performance: The authors present the results of the models for face act prediction in Table 3 and glean several insights.
  • The authors observe that all models consistently perform better for ER than EE due to the more skewed distribution of EE and the presence of an extra face act (6 for ER vs 7 for EE).
  • The difference in F1 is less noticeable between EE and All, despite an additional face act (8 for All), possibly due to the increase in labelled data
  • Conclusion:

    The authors present a generalized computational framework based on the notion of face to operationalize face dynamics in conversations.
  • The authors instantiate these face act exchanges in the context of persuasion and propose a dataset of 296 conversations annotated with face acts.
  • One important limitation of the current work is the assumption that all face acts have the same intensity/ranking.
  • The authors intend to instantiate the proposed framework to other domains such as teacher/student conversations and other types of discourse such as social media narratives
Tables
  • Table1: Generalized framework for situating and operationalizing face acts in conversations. The predicates for each of the face act are highlighted in bold
  • Table2: Distribution of different face acts for the donor (D) and non-donor (N) for ER and EE. *, **, and *** signify that the specific act is statistically significant for D and N according to the independent t-test with p-values ≤ 0.05, 0.01, and 0.001 respectively
  • Table3: Performance of the various models on face act prediction. The best results are shown in bold
  • Table4: An example conversation consisting of true and predicted face acts, along with donation probabilities. The persuader was unsuccessful in convincing the EE to donate. For brevity, the utterances of the EE are in cyan
  • Table5: Coefficients of the face acts, for ER and EE obtained from linear regression. A positive coefficient implies positive correlation. *, *** indicate statistical significance with p values ≤ 0.05 and 0.001
  • Table6: Here we describe the search-space of all the hyper-parameters used in our experiments and describe the search space we used to find the hyper-parameters. All the experiments were run on a single 1080Ti GPU. dh1, dh2 and dfc represents the hidden dimensions of Utterance GRU, Conversation GRU, and the Face act classifier. α is the hyper-parameter used to combine the face-act loss and donation loss denoted by Lf and LD respectively
  • Table7: Number of parameters for each model in our experiments
  • Table8: Instantiating predicates corresponding to the different face acts in the context of persuasion
Download tables as Excel
Related work
  • Although politeness derailment and politeness evolution in dialogue have been previously investigated in the NLP literature (Chang and DanescuNiculescu-Mizil, 2019; Danescu-Niculescu-Mizil et al, 2013), the prior work is distinguished from our own in that they do not explicitly model face changes of both parties over time. Rather, DanescuNiculescu-Mizil et al (2013) utilizes requests annotated for politeness to create a framework specifically to relate politeness and social power. Other previous work attempt to computationally model politeness, using politeness as a feature to identify conversations that appear to go awry in online discussions (Zhang et al, 2018a). Previous work has also explored indirect speech acts as potential sources of face-threatening acts through blame (Briggs and Scheutz, 2014) and as face-saving acts in parliamentary debates (Naderi and Hirst, 2018).

    The closest semblance of our work is with Kluwer (2011, 2015), which builds upon the notion of face provided by Goffman (1967) and invents its own set of face acts specifically in the context of “small-talk” conversations. In contrast, our work specifically operationalizes the notion of the positive and negative face of Brown et al (1987); Brown and Levinson (1978), which is well established in the Pragmatics literature and heavily acknowledged in the NLP community (DanescuNiculescu-Mizil et al, 2013; Zhang et al, 2018a; Wang et al, 2012; Musi et al, 2018). Moreover, we focus on analysing the effects of face acts in a “goal-oriented” task like persuasion, where there is an explicit threat or attack on face as opposed to small-talk scenarios, where the goal is building rapport or passing the time. Thus our work can be considered to be complementary to the prior work of Kluwer (2011) and Kluwer (2015). It also enables us to draw insights from recent work in persuasion strategy to analyze face act exchanges in persuasion (Wang et al, 2019; Yang et al, 2019).
Funding
  • This research was funded in part by NSF Grants (IIS 1917668 and IIS 1822831) and from Dow Chemicals
Reference
  • Muhammad Abdul-Mageed and Mona T Diab. 2012. Awatif: A multi-genre corpus for modern standard arabic subjectivity and sentiment analysis. In LREC, volume 515, pages 3907–3914. Citeseer.
    Google ScholarLocate open access versionFindings
  • Gordon Briggs and Matthias Scheutz. 2014. Modeling blame to avoid positive face threats in natural language generation. In Proceedings of the INLG and SIGDIAL 2014 Joint Session, pages 157–161, Philadelphia, Pennsylvania, U.S.A. Association for Computational Linguistics.
    Google ScholarLocate open access versionFindings
  • Penelope Brown and Stephen C Levinson. 1978. Universals in language usage: Politeness phenomena. In Questions and politeness: Strategies in social interaction, pages 56–311. Cambridge University Press.
    Google ScholarLocate open access versionFindings
  • Penelope Brown, Stephen C Levinson, and Stephen C Levinson. 1987. Politeness: Some universals in language usage, volume Cambridge university press.
    Google ScholarFindings
  • Paweł Budzianowski, Tsung-Hsien Wen, Bo-Hsiang Tseng, Inigo Casanueva, Stefan Ultes, Osman Ramadan, and Milica Gasic. 2018. Multiwoz-a largescale multi-domain wizard-of-oz dataset for taskoriented dialogue modelling. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 5016–5026.
    Google ScholarLocate open access versionFindings
  • Jonathan P. Chang and Cristian Danescu-NiculescuMizil. 2019. Trouble on the horizon: Forecasting the derailment of online conversations as they develop. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 4742–4753, Hong Kong, China. Association for Computational Linguistics.
    Google ScholarLocate open access versionFindings
  • Zheqian Chen, Rongqin Yang, Zhou Zhao, Deng Cai, and Xiaofei He. 2018. Dialogue act recognition via crf-attentive structured network. In The 41st International ACM SIGIR Conference on Research & Development in Information Retrieval, pages 225–234.
    Google ScholarLocate open access versionFindings
  • Jacob Cohen. 1960. A coefficient of agreement for nominal scales. Educational and psychological measurement, 20(1):37–46.
    Google ScholarLocate open access versionFindings
  • Cristian Danescu-Niculescu-Mizil, Moritz Sudhof, Dan Jurafsky, Jure Leskovec, and Christopher Potts. 2013. A computational approach to politeness with application to social factors. In Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 250–259, Sofia, Bulgaria. Association for Computational Linguistics.
    Google ScholarLocate open access versionFindings
  • Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. Bert: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association for
    Google ScholarLocate open access versionFindings
  • Erving Goffman. 1967. Interaction ritual: essays on face-to-face interaction.
    Google ScholarFindings
  • H Paul Grice, Peter Cole, Jerry Morgan, et al. 1975. Logic and conversation. 1975, pages 41–58.
    Google ScholarFindings
  • Swati Gupta, Marilyn A Walker, and Daniela M Romano. 2007. How rude are you?: Evaluating politeness and affect in interaction. In International Conference on Affective Computing and Intelligent Interaction, pages 203–217. Springer.
    Google ScholarLocate open access versionFindings
  • Wenxiang Jiao, Haiqin Yang, Irwin King, and Michael R. Lyu. 2019. HiGRU: Hierarchical gated recurrent units for utterance-level emotion recognition. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), pages 397– 406, Minneapolis, Minnesota. Association for Computational Linguistics.
    Google ScholarLocate open access versionFindings
  • Tina Kluwer. 2011. i like your shirt-dialogue acts for enabling social talk in conversational agents. In International Workshop on Intelligent Virtual Agents, pages 14–27. Springer.
    Google ScholarLocate open access versionFindings
  • Tina Kluwer. 2015. Social talk capabilities for dialogue systems.
    Google ScholarFindings
  • Gaurav Kumar, Rishabh Joshi, Jaspreet Singh, and Promod Yenigalla. 2020. AMUSED: A multi-stream vector representation method for use in natural dialogue. In Proceedings of The 12th Language Resources and Evaluation Conference, pages 750–758, Marseille, France. European Language Resources Association.
    Google ScholarLocate open access versionFindings
  • Tuan Manh Lai, Quan Hung Tran, Trung Bui, and Daisuke Kihara. 2019. A simple but effective bert model for dialog state tracking on resource-limited systems.
    Google ScholarFindings
  • Geoffrey N Leech. 2016. Principles of pragmatics. Routledge.
    Google ScholarFindings
  • Navonil Majumder, Soujanya Poria, Devamanyu Hazarika, Rada Mihalcea, Alexander Gelbukh, and Erik Cambria. 2019. Dialoguernn: An attentive rnn for emotion detection in conversations. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 33, pages 6818–6825.
    Google ScholarLocate open access versionFindings
  • Quinn McNemar. 1947. Note on the sampling error of the difference between correlated proportions or percentages. Psychometrika, 12(2):153–157.
    Google ScholarLocate open access versionFindings
  • Elena Musi, Debanjan Ghosh, and Smaranda Muresan. 2018. Changemyview through concessions: Do concessions increase persuasion? Dialogue & Discourse, 9(1):107–127.
    Google ScholarLocate open access versionFindings
  • Nona Naderi and Graeme Hirst. 2018. Using context to identify the language of face-saving. In Proceedings of the 5th Workshop on Argument Mining, pages 111–120, Brussels, Belgium. Association for Computational Linguistics.
    Google ScholarLocate open access versionFindings
  • Jeffrey Pennington, Richard Socher, and Christopher D Manning. 2014. Glove: Global vectors for word representation. In Proceedings of the 2014 conference on empirical methods in natural language processing (EMNLP), pages 1532–1543.
    Google ScholarLocate open access versionFindings
  • Vipul Raheja and Joel Tetreault. 2019. Dialogue Act Classification with Context-Aware Self-Attention. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), pages 3727–3733, Minneapolis, Minnesota. Association for Computational Linguistics.
    Google ScholarLocate open access versionFindings
  • Justine Zhang, Jonathan Chang, Cristian DanescuNiculescu-Mizil, Lucas Dixon, Yiqing Hua, Dario Taraborelli, and Nithum Thain. 2018a. Conversations gone awry: Detecting early signs of conversational failure. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1350– 1361.
    Google ScholarLocate open access versionFindings
  • Saizheng Zhang, Emily Dinan, Jack Urbanek, Arthur Szlam, Douwe Kiela, and Jason Weston. 2018b. Personalizing dialogue agents: I have a dog, do you have pets too? In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 2204– 2213, Melbourne, Australia. Association for Computational Linguistics.
    Google ScholarLocate open access versionFindings
  • Marilyn A Walker, Janet E Cahn, and Stephen J Whittaker. 1997. Improvising linguistic style: Social and affective bases for agent personality. arXiv preprint cmp-lg/9702015.
    Google ScholarFindings
  • William Yang Wang, Samantha Finkelstein, Amy Ogan, Alan W Black, and Justine Cassell. 2012. Love ya, jerkface: using sparse log-linear models to build positive (and impolite) relationships with teens. In Proceedings of the 13th annual meeting of the special interest group on discourse and dialogue, pages 20–Association for Computational Linguistics.
    Google ScholarLocate open access versionFindings
  • Xuewei Wang, Weiyan Shi, Richard Kim, Yoojung Oh, Sijia Yang, Jingwen Zhang, and Zhou Yu. 2019. Persuasion for good: Towards a personalized persuasive dialogue system for social good. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 5635–5649, Florence, Italy. Association for Computational Linguistics.
    Google ScholarLocate open access versionFindings
  • Steven R Wilson, Carlos G Aleman, and Geoff B Leatham. 1998. Identity implications of influence goals a revised analysis of face-threatening acts and application to seeking compliance with same-sex friends. Human Communication Research, 25(1):64–96.
    Google ScholarLocate open access versionFindings
  • Diyi Yang, Jiaao Chen, Zichao Yang, Dan Jurafsky, and Eduard Hovy. 2019. Let’s make your request more persuasive: Modeling persuasive strategies via semi-supervised neural nets on crowdfunding platforms. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), pages 3620–3630, Minneapolis, Minnesota. Association for Computational Linguistics.
    Google ScholarLocate open access versionFindings
  • Dian Yu and Zhou Yu. 2019. Midas: A dialog act annotation scheme for open domain human machine spoken conversations. arXiv preprint arXiv:1908.10023.
    Findings
Author
Ritam Dutt
Ritam Dutt
Rishabh Joshi
Rishabh Joshi
Your rating :
0

 

Tags
Comments
数据免责声明
页面数据均来自互联网公开来源、合作出版商和通过AI技术自动分析结果,我们不对页面数据的有效性、准确性、正确性、可靠性、完整性和及时性做出任何承诺和保证。若有疑问,可以通过电子邮件方式联系我们:report@aminer.cn
小科