Measuring socio-psychological drivers of extreme violence in online terrorist manifestos: an alternative linguistic risk assessment model

ABSTRACT

This paper develops a novel method of assessing the risk that online users will engage in acts of violent extremism based on linguistic markers detectable in terrorist manifestos. A comparative NLP analysis was carried out across fifteen manifestos on a scale from violent terrorist to non-violent politically moderate. We used a dictionary approach to measure the statistical significance of narratives previously identified in terrorism literature in predicting violence. The NLP analysis confirmed our research hypothesis, finding that the linguistic markers of identity fusion (an extreme form of group alignment whereby personal and group identities become functionally equivalent), dehumanising language towards the out-group and violence condoning norms were statistically significantly higher in manifestos of authors who engaged in acts of violent extremism. Building on our prior qualitative text analysis of terrorist manifestos, this study is among the first to offer a statistical analysis of the narrative patterns and associated linguistic markers distilled from terrorist manifestos. Beyond its academic contribution, the assessment framework presented here might assist security and counter-terrorism professionals in using psycholinguistic indicators to estimate the risk that online users will engage in offline violence and to make decisions on internal resource allocation in ongoing investigations.

KEYWORDS:

• Terrorist manifestos
• violent extremism
• violence risk assessment
• linguistic predictors
• identity fusion

Acknowledgements

The authors would like to express their gratitude to Philip Kreißel, data scientist at University of Bamberg, Hateaid and Volksverpetzer for his tech expertise, mentoring and advice in the NLP coding process.

Data availability statement

Due to the sensitivity of the analyzed content, the researchers refrain from publishing any raw datasets. However, all full manifestos and coding sheets can be made available upon request to academics and experts who can provide proof of their affiliation with an independent research institution.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 We chose the non-parametric Mann-Whitney U test instead of an unpaired t-test due to the relatively small sample sizes.

Additional information

Funding

This work was supported by the UK’s Economic and Social Research Council and St John’s College at Oxford University under the ESRC Grand Union Doctoral Training Partnership (DTP) Studentship and an Advanced Grant from the European Research Council (ERC) under the European Union’s HORIZON EUROPE 2020 Research and Innovation Programme (#694986).