Abstract
A growing body of research suggests that an individual's willingness to fight and die for groups is rooted in the fusion of personal and group identities, especially when the group is threatened, violence is condoned, and the group's enemies are dehumanised or demonised. Here we consider whether the language used by extremists can help with early detection of these risk factors associated with violent extremism. We applied a new fusion-based linguistic violence risk assessment framework to a range of far-right extremist online groups from across the violence spectrum. We conducted an R-based NLP analysis to produce a Violence Risk Index, integrating statistically significant linguistic markers of terrorist manifestos as opposed to non-violent communiqués into one weighted risk assessment score for each group. The language-based violence risk scores for the far-right extremist groups were then compared to those of non-extremist control groups. We complemented our quantitative NLP analysis with qualitative insights that contextualise the violence markers detected in each group. Our results show that the fusion markers combined with several other variables identified across the different online datasets are indeed indicative of the real-world violence level associated with the relevant groups, pointing to new ways of detecting and preventing violent terrorism.