No abstract

Grooming of children online is a legally punishable form of child sexual abuse. In the UK, for instance, the Serious Crime Act has since 2015 made it a criminal offence for an adult to send a sexual message to a child. 1 Nevertheless, this offence is on the increase. The UK-based charity NSPCC (National Society for the Prevention of Cruelty to Children), for example, reports a 16.8% growth in policerecorded cases of online child sexual grooming in England and Wales from 2015/16 to 2016/2017 (Bentley et al., 2018, p. 19). Moreover, these figures underestimate the real scale of the issue, which is known to be significantly under-reported by their victims because of a variety of reasons. These include the inability in some cases to realise for some time that they are being abused online, believing instead that they have a friendship-based and/or romantic relationship with their online groomers (e.g., Davidson and Gottschalk, 2011; NSPCC, 2018 2). The seriousness of the offence and its increasing prevalence over time may account for a growing body of academic scholarship investigating online child sexual grooming over the past decade or so. This scholarship has been primarily conducted within the disciplines of Criminology, Psychology and, to a lesser extent, Computational (primarily, Machine Learning) Text Analysis (See Section 2). In contrast, Linguistics scholarship into online child sexual grooming is still "in its infancy" (Chiang and Grant, 2018, p. 2). This is sadly ironic given that online child sexual grooming is an internet-enabled communicative 1

The presence of Daesh (also known as the Islamic State of Iraq and Syria, ISIS) on Twitter has greatly diminished over the past five years as Daesh's propaganda dissemination strategy has evolved. Yet some Daesh supporters have persevered in their use of Twitter, using throwaway accounts to share outlinks to pro-Daesh materials on other platforms. This article analyses 892 outlinks found in 11,520 tweets that contained the word Rumiyah (Daesh's online magazine). It evaluates Twitter's response to attempts to use its platform to signpost users to Rumiyah in the context of the wider social media ecosystem and highlights the role played by botnet activity in efforts to disseminate the magazine and the impact of traditional news media coverage.Five years have now passed since Daesh (also known as the Islamic State of Iraq and Syria, ISIS) enjoyed its 'Golden Age' on Twitter. 1 The ISIS Twitter Census conducted by Berger and Morgan in 2015 found that during October and November 2014 there were no fewer than 46,000 overt Daesh supporter accounts on Twitter -and possibly as many as 90,000. The average number of followers of these accounts was 1004, and each account posted an average of 7.3 tweets per day over its lifetime. 2 As well as proselytisation, recruitment and firming up the resolve of followers, Daesh utised the new capabilities offered by social media to employ the platform for psychological warfare purposes. 3 Since then Daesh's presence on Twitter has been reduced significantly. Towards the end of 2014 Twitter began an aggressive campaign of suspensions. Berger and Morgan found that, by February 2015, Daesh supporters on Twitter were having to devote far more time to rebuilding their networks. 4 A follow-up study conducted by Berger and Perez also found that suspension activity had a significant disruptive effect. 5 Individual users who repeatedly created new accounts after being suspended 'suffered devastating reductions in their follower counts' and declines in networks persisted even when suspension pressure eased, 'suggesting that suspensions diminish activity in ways that extend beyond the simple removal of accounts '. 6 In response, IS supporters resorted to the use of a variety of countermeasures. These included locking their accounts so that they were no longer publicly accessible, using an innocuous image or the default egg as the avatar image and selecting a random combination of letters and numbers as the user handle or screen name. 7 However, 'A conscious, supportive and influential virtual community is almost impossible to maintain in the face of the loss of access to such group or

No abstract

Online grooming has become a wide-spread and worryingly fast increasing issue in society. This thesis analyses a corpus of online grooming communication, made available by the Perverted Justice (PJ) archive, a non-profit organisation that from 2004 until 2019 employed volunteers, who pretended to be children and entered chat rooms to catch and convict groomers, collaborating with law enforcement. The archive consists of 622 grooming chat logs and approx. 3.7 million words of groomer language. A corpus of this database was built, and a Corpus-Assisted Discourse Studies (CADS) approach used to analyse the language therein. Specifically, the language was compared to a reference corpus of general chat language data (PAN2012) and duration of online grooming and manipulative requesting behaviour were also investigated. The following research questions were answered: 1)What are the features of a corpus of online groomer language compared to that of a general digital chat language reference corpus? Is online groomer language distinct? How are online grooming intentions realised linguistically by online groomers?2)Does duration of grooming influence the grooming process/intentions? Is usage of specific words/specific grooming intentions associated with different duration of grooming? Can different duration profiles be established and, if so, what are the cut-off points for these duration profiles?3)How are requests realised in online grooming and how does duration influence this? How do groomers make requests and what support move functions do they use? Does duration influence how requests are made, and the type of support move function that are used?The thesis newly identifies nuanced linguistic realisations of groomers’ intentions and strategies, proposing a new working terminology for discourse-based models of online grooming. This is based on a review of the literature followed by an empirical analysis refining this terminology, which has not been done before. It finds evidence for two distinct duration-based grooming approaches and yields a fine-grained qualitative analysis of groomer requests, also influenced by grooming duration. There have only been very few studies using a CADS analysis of such a large dataset of groomer language and this thesis will lead to new insights, implications and significance for the successful analysis, detection and prevention of online grooming.