Authors: (1) Yi-Ling Chung, The Alan Turing Institute (ychung@turing.ac.uk); (2) Gavin Abercrombie, The Interaction Lab, Heriot-Watt University (g.abercrombie@hw.ac.uk); (3) Florence Enock, The Alan Turing Institute (fenock@turing.ac.uk); (4) Jonathan Bright, The Alan Turing Institute (jbright@turing.ac.uk); (5) Verena Rieser, The Interaction Lab, Heriot-Watt University and now at Google DeepMind (v.t.rieser@hw.ac.uk). Authors: Authors: (1) Yi-Ling Chung, The Alan Turing Institute (ychung@turing.ac.uk); (2) Gavin Abercrombie, The Interaction Lab, Heriot-Watt University (g.abercrombie@hw.ac.uk); (3) Florence Enock, The Alan Turing Institute (fenock@turing.ac.uk); (4) Jonathan Bright, The Alan Turing Institute (jbright@turing.ac.uk); (5) Verena Rieser, The Interaction Lab, Heriot-Watt University and now at Google DeepMind (v.t.rieser@hw.ac.uk). Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 Background 2 Background 3 Review Methodology 3 Review Methodology 4 Defining counterspeech 4 Defining counterspeech 4.1 Classifying counterspeech 4.1 Classifying counterspeech 5 The Impact of Counterspeech 5 The Impact of Counterspeech 6 Computational Approaches to Counterspeech and 6.1 Counterspeech Datasets 6 Computational Approaches to Counterspeech and 6.1 Counterspeech Datasets 6.2 Approaches to Counterspeech Detection and 6.3 Approaches to Counterspeech Generation 6.2 Approaches to Counterspeech Detection and 6.3 Approaches to Counterspeech Generation 7 Future Perspectives 7 Future Perspectives 8 Conclusion, Acknowledgements, and References 8 Conclusion, Acknowledgements, and References 4 Defining counterspeech Counterspeech is multifaceted and can be characterised in several different ways. In Table 1 we outline a framework for describing and designing counterspeech, covering who (speaker) sends what kinds of messages (strategies) to whom (recipients), and for what purpose (purpose). Using this structure, we summarise how counterspeech has typically been categorised in past studies. Most studies in the field use one of three main terms: counterspeech, counter-narratives (Reynolds and Tuck, 2016; Carthy and Sarma, 2021; Tuck and Silverman, 2016; Iqbal et al., 2019) and hope speech (Snyder et al., 2018). These three terms broadly refer to a similar concept: content that challenges and rebuts hateful discourse and propaganda (Saltman and Russell, 2014; Bartlett and Krasodomski-Jones, 2015; Benesch et al., 2016; Saltman et al., 2021; Garland et al., 2022) using non-aggressive speech (Benesch et al., 2016; Reynolds and Tuck, 2016; Schieb and Preuss, 2016). There are some differences between the terms. Ferguson (2016) considers counter-narratives as intentional strategic communication within a political, policy, or military context. Additionally, the term counter-narrative also refers to narratives that challenge a much broader view or category such as forms of education, propaganda, and public information (Benesch et al., 2016). Such counternarratives are often discussed in the context of the prevention of violent extremism. Hope speech, meanwhile, could be seen as a particular type of counterspeech: it promotes positive engagement in online discourse to lessen the consequences of abuse, and places a particular emphasis on delivering optimism, resilience, and the values of equality, diversity and inclusion (Chakravarthi, 2022). In this paper, we review work that relates to all of these three concepts, and largely make use of the catch-all term counterspeech, while acknowledging the slight differences between the concepts. 4.1 Classifying counterspeech Researchers have identified a variety of different types of counterspeech. Here, we outline four main ways in which counterspeech can vary, in terms of the identity of the counterspeaker, the strategies employed, the recipient of the counterspeech and the purpose of counterspeech. Counterspeakers (who) Psychological studies show that the identity of a speaker plays a key role in how large an audience their message reaches and how persuasive the message is. Common crucial factors include group identity (such as race, religion, and nationality), level of influence, and socioeconomic status. For instance, counterspeech provided by users with large numbers of followers and from an in-group member is more likely to lead to changes in the behaviour of perpetrators of hate (Munger, 2017). Counterspeakers (who) Some studies characterise individuals who use counterspeech and suggest that these users exhibit different characteristics and interests than users who spread hate (Mathew et al., 2018, 2019; Buerger, 2021b). Through lexical, linguistic and psycholinguistic analysis of users who generate hate speech or counterspeech on Twitter, Mathew et al. (2018) find that counterspeakers are higher in agreeableness, displaying traits such as altruism, modesty, and sympathy, and display higher levels of self-discipline and conscientiousness. Possibly driven by a motive to help combat hate speech, counterspeakers tend to use words related to government, law, leadership, pride, and religion. Regarding the impact of being a counterspeaker, in an ethnographic study, members of a counterspeech campaign reported feeling more courageous and keen to engage in challenging discussions after expressing opinions publicly (Buerger, 2021b). Strategies (how) Counterspeech can take many forms. Benesch et al. (2016) first identify eight types of counterspeech used on Twitter: (1) presentation of facts, (2) pointing out hypocrisy or contradiction, (3) warning of consequences, (4) affiliation [i.e. establishing an emotional bond with the perpetrators or targets of hate], (5) denouncing, (6) humour/sarcasm, (7) tone [a tendency or style adopted for communication, e.g., empathetic and hostile], and (8) use of media. Based on this taxonomy, follow-up studies on counterspeech make minor modifications to cover strategies in a broader scope. Mathew et al. (2018) analyzed and classified counterspeech on Twitter, taking Benesch et al. (2016)’s taxonomy but dropping the use of media and adding hostile language and positive tone, which replaces general strategy tone. Similarly, Mathew et al. (2019) collected and annotated counterspeech comments from Youtube, adopting Benesch et al. (2016)’s taxonomy but excluding tone and adding positive tone, hostile language and miscellaneous. Chung et al. (2019) collaborated with NGOs to collect manually written counterspeech. For data annotation, they followed the taxonomies provided by Benesch et al. (2016) and Mathew et al. (2019), while adding counter question and discarding the use of media. Counterspeech examples for each strategy are provided in Table 2. Strategies (how) Counterspeech recipients (whom) Depending on the purpose of the counterspeech, the target audience may be perpetrators, victims or bystanders (see Figure 1). Identifying the appropriate target audience or ‘Movable Middle’ is crucial to maximise the efficacy of counterspeech. Movable middle refers to individuals who do not yet hold firm opinions on a topic and can hence be potentially open to persuasion. They are also receptive to arguments and more willing to listen. These individuals often serve as ideal recipients of messages addressing social issues such as vaccination hesitancy (Litaker et al., 2022). In the context of counterspeech, previous studies show that a small group of counterspeakers can shape online discussion when the audience holds moderate views (Schieb and Preuss, 2016; Buerger, 2021b). Counterspeech recipients (whom) Wright et al. (2017) group counterspeech acts into four categories based on the number of people involved in the discussion: one-to-one, one-to-many, many-to-one, or many-to-many. Some successful cases where counterspeech induces favourable changes in the discourse happen in a one-to-one discussion. This allows for dedicated opinion exchange over an ideology, which in some cases even yields long-lasting changes in beliefs. The use of hashtags is a good example of one-to-many and many-to-many interaction where conversations surge quickly (Benesch et al., 2016; Wright et al., 2017). For instance, Twitter users often include hashtags to express support (e.g., #BlackLivesMatter) or disagreement with haters (e.g., #StopHate) to demonstrate their perspective. The purpose of counterspeech Hateful language online can serve to reinforce prejudice (Citron and Norton, 2011), encourage further division, promote power of the ingroup, sway political votes, provoke or justify offline violence, and psychologically damage targets of hate (Jay, 2009). Just as the effects of hate are wide-ranging, counterspeech may be used to fulfil a variety of purposes. The purpose of counterspeech • Changing the attitudes and behaviours of perpetrators In directly challenging hateful language, one key aim of counterspeech can be to change the attitudes of the perpetrators of hate themselves. The strategy here is often to persuade the perpetrator that their attitudes are mistaken or unacceptable, and to deconstruct, discredit or delegitimise extremist narratives and propaganda (Reynolds and Tuck, 2016). Counterspeech aimed at changing the attitudes of spreaders of hate may address the hate speaker directly, countering claims with facts or by employing empathy and affiliation. Challenging attitudes is often seen as a stepping stone to altering behaviours (Stroebe, 2008). In attempting to change the minds of perpetrators, counterspeakers ultimately hope to discourage associated behaviours such as sharing such content again in the future or showing support for other hateful content (i.e., stopping the spread of hate). In changing the minds of perpetrators, counterspeakers may also hope to prevent them from engaging in more extreme behaviours such as offline violence. • Changing the attitudes and behaviours of perpetrators • Changing the attitudes and behaviours of bystanders More commonly, counterspeech is initiated with the intention of reaching the wider audience of bystanders rather than perpetrators of hate themselves (Buerger, 2022). These bystanders are not (at least yet) generating hateful language themselves, but rather are people exposed to hateful content either incidentally or by active engagement. Here, counterspeakers hope to persuade bystanders that the hateful content is wrong or unacceptable, again by deconstructing and delegitimising the hateful narrative. The strategy here may be to offer facts, point out hypocrisy, denounce the content, or use humour to discredit the speaker. Additionally, counterspeakers will often invoke empathy for targets of hate. In preventing bystanders from forming attitudes and opinions in line with the hateful narrative, counterspeakers hope to mitigate further intergroup division and related behaviours such as support for or engagement with additional abuse or physical violence. Counterspeakers may also hope to encourage others to generate rebutals and rally support for victims (Benesch, 2014a), bringing positive changes in online discourse. • Changing the attitudes and behaviours of bystanders • Showing support for targets of hate A third key way in which counterspeech functions is to show support directly to targets of hate. Online abuse can psychologically damage the wellbeing of targets and leave them feeling fearful, threatened, and even in doubt of their physical safety (Benesch, 2014b; Leader Maynard and Benesch, 2016; Saha et al., 2019; Siegel, 2020). By challenging such abuse, counterspeakers can offer support to targets and encourage bystanders to do the same (Buerger, 2021b). This support aims to alleviate negative emotion brought on by hate by demonstrating to targets that they are not alone and that many people do not hold the attitudes of the perpetrator. Here the particular strategies may be to denounce the hate and express positive sentiment towards the target group. Intergroup solidarity may in turn reduce retaliated antagonism. • Showing support for targets of hate This paper is available on arxiv under CC BY-SA 4.0 DEED license. This paper is available on arxiv under CC BY-SA 4.0 DEED license. available on arxiv
