Discourse Analysis for Insights into Group Identity and Intent

ABSTRACT:In our Phase 1 SBIR effort (Discourse Analysis for Insights into Group Identity and Intent), SAE conducted a pilot study to assess whether its suite of natural language processing techniques could be adapted to automatically extract meaningful discourse measures from extremist group literature (cognitive complexity, idea density, vocabulary diversity, and sentiments expressed toward in and out-groups). We also demonstrated that these discourse indicators significantly increased our ability to predict future extremist group attacks, at least in the case of the Peoples War Group (PWG) in India.In Phase 2 of the project, we propose to extend this line of work by (1) refining our measures and assessing their generalizability to other English-speaking groups (2) assessing the extent to which we can automatically process discourse in multiple western (French and Spanish) and non-Western (Urdu, Pashto, Mandarin, and Arabic) foreign languages and (3) developing a prototype system that will allow analysts to apply our techniques themselves in useful ways to enhance their missions.BENEFIT:Military and intelligence analysts (i.e., the National Air and Space Intelligence Center (NASIC), Defense Intelligence Agency (DIA), Central Intelligence Agency (CIA), routinely scrutinize statements, blog postings, videos and other literature generated by extremist organizations for clues about their future potential and intent to engage in violent attacks on US interests.This is a painstaking, labor-intensive process to identify what are often subtle shifts in rhetorical expositions.Moreover, even well-trained analysts will have difficulty systematically assessing how changes in one discourse indicator interact with changes in other discourse indicators, and how these interactions are likely to affect the probability of increases in violent attacks.If current trends continue, volume of extremist literature should continue to increase, while budgetary pressures reduce manpower allocations required to keep up with it. Our automated Discourse analysis system could be a significant asset to the extent that it could be applied in near real time to monitor trends and shifts in the use of discourse mechanisms across multiple organizations using multiple languages, and cue the analysts to specific document clusters that may require a more thorough examination and a more focused, nuanced interpretation.Thus, rather than spending 90% of their time collecting data and 10% of their time assessing it, with the aid of our tools, we can alter this ratio in a more cost-effective and useful direction.