An Exploratory Analysis of the Relation between Offensive Language and Mental Health

Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021, pages 3600–3606 August 1–6, 2021. ©2021 Association for Computational Linguistics 3600 An Exploratory Analysis of the Relation Between Offensive Language and Mental Health Ana-Maria Bucur1, Marcos Zampieri2, and Liviu P. Dinu1 1University of Bucharest, Romania 2Rochester Institute of Technology, USA ana-maria.bucur@drd.unibuc.ro, marcos.zampieri@rit.edu ldinu@fmi.unibuc.ro Abstract In this paper, we analyze the interplay be- tween the use of offensive language and men- tal health. We acquired publicly available datasets created for offensive language identi- fication and depression detection and we train computational models to compare the use of offensive language in social media posts writ- ten by groups of individuals with and without self-reported depression diagnosis. We also look at samples written by groups of individ- uals whose posts show signs of depression ac- cording to recent related studies. Our analysis indicates that offensive language is more fre- quently used in the samples written by individ- uals with self-reported depression as well as individuals showing signs of depression. The results discussed here open new avenues in re- search in politeness/offensiveness and mental health. 1 Introduction The use of offensive language is pervasive in social media and it has been studied from different per- spectives. A popular line of research is the study of computational models to identify offensive con- tent online relying on traditional machine learning classifiers (e.g. naive bayes and SVMs) (Xu et al., 2012; Dadvar et al., 2013), neural networks (e.g. LSTMs, GRUs) with word embeddings (Aroyehun and Gelbukh, 2018; Majumder et al., 2018), and more recently, transformer models like ELMO (Pe- ters et al., 2018) and BERT (Devlin et al., 2019) which have shown to obtain competitive scores topping the leaderboards in recent shared tasks on offensive language and hate speech detection (Liu et al., 2019). Offensive language is related to the notion of im- politeness (Culpeper, 2011) and it can take various forms from general and often harmless profanity WARNING: This paper contains offensive words. to abusive language intended to cause harm, such as cyberbullying and hate speech (Waseem et al., 2017). Computational models have been applied not only to identify the various types of offensive content (Basile et al., 2019) but also to, for ex- ample, study the relation between profanity and hate speech (Malmasi and Zampieri, 2018) and the different functions and intentions of vulgarity in social media (Holgate et al., 2018). Most of the datasets used in the aforementioned studies contain data sampled from the general pop- ulation and therefore very little light has been shed on the use of offensive language in online commu- nication by specific groups such as individuals with mental health conditions. A notable exception is the recent study by Birnbaum et al. (2020) which shows that users with mood disorders (bipolar dis- order, major depressive disorder) and schizophre- nia spectrum disorders use more swear words in their Facebook messages than healthy users. To address this shortcoming, in this paper, we build on recent work on offensive language iden- tification and apply …

Content truncated to 500 words