Understanding and detecting behavioural and ideological escalation in online communities: a linguistic approach

Harmful online communities can support individuals in becoming more ideologically extreme and more skilled at o↵ending. A key task for law-enforcement is to detect these various forms of escalation as a means of assessing potential threats posed by particular online groups. This task is, however, complicated by two issues: first, a lack of definitional consensus around related and overlapping terms including escalation, polarisation, radicalisation, grooming, extremism and terrorism (Schmidt, 2013), and second, a lack of formalised, empirically tested method for analysing and detecting various forms of escalation. In this talk, we present preliminary findings from ongoing project ADOBE (Analysing and Describing Online Behaviours around Escalation); a research project directly tackling these twin issues. We first report findings from a corpus-assisted analysis of academic and policy literature around escalation and related terms, exploring their uses and meanings across varied fields of study and addressing the question of whether and how we can provide some terminological clarity and consistency. Having reviewed strengths and weaknesses of previous approaches, we present our own form- and function-based linguistic analysis of data taken from a far-right extremist online forum to identify the lexico-grammatical and pragmatic features of the discourse that indicate ideological and behavioural escalation of its members.
We hope that this work will assist scholars interested in the processes of online escalation/radicalisation/extremism, and be of practicable assistance to law-enforcement bodies tasked with analysing and detecting escalation in online communities, assessing risk, and developing appropriate responses.