Conversational Analysis of Daily Dialog Data using Polite Emotional Dialogue Acts

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 2395–2400 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 2395 Conversational Analysis of Daily Dialog Data using Polite Emotional Dialogue Acts Chandrakant Bothe† and Stefan Wemter‡ †Foviatech GmbH chandrakant.bothe@foviatech.com ‡Knowledge Technology, Department of Informatics, University of Hamburg stefan.wermter@uni-hamburg.de Abstract Many socio-linguistic cues are used in conversational analysis, such as emotion, sentiment, and dialogue acts. One of the fundamental cues is politeness, which linguistically possesses properties such as social manners useful in conversational analysis. This article presents findings of polite emotional dialogue act associations, where we can correlate the relationships between the socio-linguistic cues. We confirm our hypothesis that the utterances with the emotion classes Anger and Disgust are more likely to be impolite. At the same time, Happiness and Sadness are more likely to be polite. A less expectable phenomenon occurs with dialogue acts Inform and Commissive which contain more polite utterances than Question and Directive. Finally, we conclude on the future work of these findings to extend the learning of social behaviours using politeness. Keywords: Politeness, Emotion, Dialogue Act, Conversational Analysis, Artificial Intelligence 1. Introduction Conversational analysis can potentially be enhanced with socio-linguistic politeness cues along with other linguistic cues (Kasper, 1990; Russell and Barrett, 1999; Danescu-Niculescu-Mizil et al., 2013). Liter- ature suggests that analysis and use of social cues is beneficial for human-robot interaction (Salem et al., 2014; Barros et al., 2015; Castro-Gonz´alez et al., 2016; Bothe, 2020). Our article focuses on an experiment by primarily adding linguistic politeness cues, expressed using polite phrases, to conversational analysis along with emotional states and dialogue acts where we aim to discover correlations between these cues. For our experiment, we will explore DailyDialog dataset (Li et al., 2017), a multi-turn dialogue dataset, which is pre-annotated with emotion and dialogue act labels. Furthermore, the dataset will be augmented with politeness labels ranging from 1 to 5 by leverag- ing a pre-trained model (Bao et al., 2021). By analyz- ing the correlation between politeness and emotional states, we discover that utterances with certain emotion classes are more polite or impolite than others. At least frequently, it appears natural to use polite utterances in a happy emotional state, whereas to use impolite utter- ances in an angry state. This phenomenon is precisely the motivation behind this experiment and we can find such utterance examples illustrated in Figure 1. The conversation in this example shows how an impolite utterance is used in the Anger emotion state and then shifts with a polite utterance to the Happiness emotion state. The results show this phenomenon statistically occurring significantly in the given dataset. The conversational behaviours learned with these cues might be beneficial to drive the dialogue flow in com- plex polite human-robot interaction setup, which was Hey. Why did you take that money? You are such a cheater! I should send you to jail! Emotion: Anger DialogAct: Question I am not cheating. When you pass go, you collect $200, …

Content truncated to 500 words