Linguistic markers in the writings of violent offenders, high-risk individuals, and online users

This study investigates whether warning signs of targeted violence can be identified in written communication by analyzing linguistic markers across three groups: violent offenders, high-risk persons of concern (POCs), and a diverse online population. We focus on linguistic markers for five warning signs: anger, grievance, leakage (intent to do harm), identification (taking the role of a previous attacker, violent ideology, or warrior persona), and operational details (planning and preparation of an attack). Using a unique dataset of writings from 115 individuals (80 violent offenders and 35 POCs) and communication from 14 online communities, we have assessed the prevalence of these markers. Our results show that all five warning signs were significantly more common among violent offenders and POCs than in the online population. In a comparison between the violent offenders and the POCs, we found that operational details were significantly more frequent in texts written by violent offenders than by POCs, suggesting that the presence of operational details is important when separating those who actually commit violent attacks from persons of concern. Our findings also show that POC express more anger than lone offenders, which may suggest emotional instability without a clear intention to act violently. Importantly, these linguistic markers demonstrate discriminative capacity between groups and are intended to support the identification and prioritization of elevated risk, rather than to predict individual acts of violence. Our findings demonstrate the value of linguistic markers in digital threat assessment and highlight how specific online environments may amplify risk indicators. This approach offers a scalable method to support early detection and intervention in cases of potential targeted violence.