Should a Chatbot be Sarcastic? Understanding User Preferences Towards Sarcasm Generation

Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 7686 - 7700 May 22-27, 2022 c⃝2022 Association for Computational Linguistics Should a Chatbot be Sarcastic? Understanding User Preferences Towards Sarcasm Generation Silviu Vlad Oprea1 and Steven R. Wilson3 and Walid Magdy1, 2 1 School of Informatics, University of Edinburgh, Edinburgh, United Kingdom 2 The Alan Turing Institute, London, United Kingdom 3 School of Engineering and Computer Science, Oakland University, Rochester, MI, USA silviu.oprea@ed.ac.uk, stevenwilson@oakland.edu, wmagdy@inf.ed.ac.uk Abstract Previous sarcasm generation research has fo- cused on how to generate text that people per- ceive as sarcastic to create more human-like interactions. In this paper, we argue that we should first turn our attention to the question of when sarcasm should be generated, finding that humans consider sarcastic responses inap- propriate to many input utterances. Next, we use a theory-driven framework for generating sarcastic responses, which allows us to con- trol the linguistic devices included during gen- eration. For each device, we investigate how much humans associate it with sarcasm, find- ing that pragmatic insincerity and emotional markers are devices crucial for making sar- casm recognisable. 1 Introduction The prevalence of sarcasm on the social web (Kho- dak et al., 2018; Sykora et al., 2020) has motivated computational investigations across the NLP com- munity. Most focus on textual sarcasm detection, the task of classifying whether or not a given text is sarcastic (Riloff et al., 2013; Wallace et al., 2015; Rajadesingan et al., 2015; Bamman and Smith, 2015; Joshi et al., 2016; Amir et al., 2016; Hazarika et al., 2018; Oprea and Magdy, 2019; Abu Farha et al., 2022). A recent research direction considers sarcasm generation. Approaches to sarcasm generation in- troduced so far (Joshi et al., 2015; Mishra et al., 2019; Chakrabarty et al., 2020) are mainly moti- vated by the potential to create more approachable, human-like conversational agents, considering that sarcasm is a natural part of human discourse. We suggest reconsidering this motivation, as a commu- nity, for two reasons. First, in human discourse, sarcasm is not a communicative goal in itself. Rather, it can be used to achieve a wide variety of goals. Some of these goals, such as to diminish the impact of criticism (Dews and Winner, 1995), to create hu- mour (Kreuz et al., 1991; Colston and O’Brien, 2000b,a), to praise (Bruntsch and Ruch, 2017), or to strengthen relationships (Jorgensen, 1996; Pex- man and Zvaigzne, 2004), might be desirable in human-machine interactions as well. However, other goals, such as criticising, mocking, or ex- pressing dissociation, often with surface contempt or derogation (Wilson, 2006), might not be desir- able in human-machine interactions. Second, the communicative goals mentioned above were observed in human interactions. Even when a machine seeks potentially desirable goals, it is unclear whether sarcastic utterances have the same effect on humans when coming from ma- chines. Therefore, we suggest it is imperative, not least from an ethical perspective, to consider the follow- ing research questions (RQs): RQ1. When should …

Content truncated to 500 words