Just Like a Human Would, Direct Access to Sarcasm Augmented with Potential Result and Reaction

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 10172–10183 July 9-14, 2023 ©2023 Association for Computational Linguistics Just Like a Human Would, Direct Access to Sarcasm Augmented with Potential Result and Reaction Changrong Mina, Ximing Li∗b,c, Liang Yanga, Zhilin Wangb,c, Bo Xua, Hongfei Lina aSchool of Computer Science and Technology, Dalian University of Technology, China bCollege of Computer Science and Technology, Jilin University, China cKey Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education, Jilin University, China 11909060@mail.dlut.edu.cn,liximing86@gmail.com wangzl5521@mails.jlu.edu.cn,(hflin,liang,xubo)@dlut.edu.cn Abstract Sarcasm, as a form of irony conveying mockery and contempt, has been widespread in social media such as Twitter and Weibo, where the sarcastic text is commonly characterized as an incongruity between the surface positive and negative situation. Naturally, it has an urgent demand to automatically identify sarcasm from social media, so as to illustrate people’s real views toward specific targets. In this paper, we develop a novel sarcasm detection method, namely Sarcasm Detector with Augmentation of Potential Result and Reaction (SD-APRR). Inspired by the direct access view, we treat each sarcastic text as an incomplete version without latent content associated with implied negative situations, including the result and human re- action caused by its observable content. To fill the latent content, we estimate the potential result and human reaction for each given train- ing sample by [xEffect] and [xReact] relations inferred by the pre-trained common- sense reasoning tool COMET, and integrate the sample with them as an augmented one. We can then employ those augmented samples to train the sarcasm detector, whose encoder is a graph neural network with a denoising module. We conduct extensive empirical experiments to evaluate the effectiveness of SD-APRR. The results demonstrate that SD-APRR can outper- form strong baselines on benchmark datasets. 1 Introduction Sarcasm, as subtle figures of speech, serves many communicative purposes in human daily life (Ivanko and Pexman, 2003), commonly used to crit- icize an individual. Refer to the formal description of sarcasm from the Oxford English Dictionary:1 “A way of using words that are the opposite of what you mean in order to be unpleasant to somebody or to make fun of them.” ∗* Corresponding Author 1https://www.oed.com The sarcastic text is typically characterized as an incongruity between the positive surface and nega- tive situation (Riloff et al., 2013; Liu et al., 2022). For example, as an obvious sarcasm “I love work- ing for six hours every day for free”, its surface meaning tends to be positive, conveyed by the sen- timent word “love”, but it corresponds to a negative situation “work for free”, conveying people’s com- plaint. Detecting sarcasm from social media is a signifi- cant task due to the universal existence of sarcasm, but its complicated nature makes the task challeng- ing. To resolve this task, the community has re- cently proposed a number of Sarcasm Detection (SD) methods, whose major idea is to capture the incongruity characteristic of sarcasm (Joshi et al., 2017; Xiong et al., 2019; Pan …

Content truncated to 500 words