Language Predicts Identity Fusion Across Cultures and Reveals Divergent Pathways to Violence

Abstract

In light of increasing polarization and political violence, understanding the psychological roots of extremism is increasingly important. Prior research shows that identity fusion predicts willingness to engage in extreme acts. We evaluate the Cognitive Linguistic Identity Fusion Score, a method that uses cognitive linguistic patterns, LLMs, and implicit metaphor to measure fusion from language. Across datasets from the United Kingdom and Singapore, this approach outperforms existing methods in predicting validated fusion scores. Applied to extremist manifestos, two distinct high-fusion pathways to violence emerge: ideologues tend to frame themselves in terms of group, forming kinship bonds; whereas grievance-driven individuals frame the group in terms of their personal identity. These results refine theories of identity fusion and provide a scalable tool aiding fusion research and extremism detection.

Figures (3)

• Figure 1: Performance of fusion-prediction methods across text-length and sampling conditions. Heat maps show Spearman rank correlations ($r_s$; top), associated $p$ values (middle), and mean absolute error (MAE; bottom) between model scores and ground-truth identity fusion ratings for four methods (CLIFS, UAI, VRI-Fusion, and nUAI) across minimum-length and corpus conditions. Rows correspond to filtering thresholds based on the number of sentences or words per document in the MTurk corpus, as well as the Singapore corpus in original and chunked form. Higher $r_s$ and lower MAE indicate better agreement with true scores. Statistical significance is indicated by asterisks ($^*p<0.05$, $^{**}p<0.01$).
• Figure 2: Distributional comparison of CLIFS scores by motivational framing. CLIFS score distributions for victim- and ideologue-motivated texts are shown using density-normalized histograms and kernel density estimates (KDE), with interquartile ranges (IQR; shaded) and medians (dashed). Overlayed KDEs (top right) and empirical cumulative distribution functions (ECDF; bottom right) enable direct shape and distribution-wide comparison. Differences in central tendency and effect sizes (Victim $-$ Ideologue) are small, and the Wasserstein distance indicates only a modest global shift between distributions.
• Figure 3: Distributions of CLIFS masked-LM implicit-metaphor scores for victim- and ideologue-motivated texts. Histograms and kernel density estimates show the distributions of the four CLIFS components—fusion proximity $f_{(I,T)}$, fictive kinship $K_f$, fusion target-to-identity $S_{T \rightarrow I}$, and identity-to-target $S_{I \rightarrow T}$ directional scores—for text chunks labeled High Fusion by the coarse-grained CLIFS classifier. Texts classified as Victim are shown in blue (top panels) and Ideologue in orange (bottom panels). Dashed vertical lines indicate group medians, and shaded regions indicate interquartile ranges (IQRs). Numeric annotations report median values. Insets summarize median differences (Victim $-$ Ideologue) with 95% bootstrap confidence intervals and Cohen's $d$.