💬 NLP

Cognitive Linguistic Identity Fusion Score (CLIFS): A Scalable Cognition-Informed Approach to Quantifying Identity Fusion from Text

The paper introduces the Cognitive Linguistic Identity Fusion Score (CLIFS), a novel, automated metric that leverages cognitive linguistics and large language models to quantify the psychological merging of self with external entities through text analysis, demonstrating superior accuracy and scalability compared to traditional methods and existing automated approaches.

Original authors: Devin R. Wright, Jisun An, Yong-Yeol Ahn

Published 2026-02-10

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Devin R. Wright, Jisun An, Yong-Yeol Ahn

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The "Soul-Mirror" Metric: How Computers Can Sense When You’ve Lost Yourself in a Cause

Imagine you are a die-hard fan of a sports team. For most people, it’s just a game. But for a "fused" fan, the team isn't just something they watch; the team is something they are. If the team loses, they feel a physical ache in their chest. If the team is insulted, they feel personally attacked. Their "self" and the "team" have melted together like two drops of water on a windowpane.

Psychologists call this Identity Fusion. It’s a powerful force that can make people incredibly brave and helpful (prosocial), but it can also drive people toward extreme, even violent, behavior if they feel their "target" (a religion, a political party, or an ideology) is under threat.

The problem? Until now, the only way to measure this was to sit people down and ask them, "On a scale of 1 to 7, how much are you one with your group?" This is slow, expensive, and people often lie or don't realize how they truly feel.

Enter CLIFS: The Digital Lie Detector for the Soul.

The Core Idea: The "Metaphor" Secret

The researchers behind this paper realized that when people are deeply fused with something, they don't just talk about it—they talk as it. They use language in a way that reveals their subconscious "oneness."

Think of it like this:

Normal Person: "I love my country. It has a great history." (The person and the country are two separate objects).
Fused Person: "My country is my heartbeat. Its struggles are my struggles." (The person and the country have become a single metaphor).

The researchers created a tool called CLIFS (Cognitive Linguistic Identity Fusion Score). Instead of asking questions, CLIFS reads text (like social media posts or manifestos) and looks for these "metaphorical leaks."

How It Works: The "Masked Actor" Trick

To find these hidden connections, the researchers used a clever AI trick inspired by how we learn language.

Imagine you are watching a play where the lead actor’s name is covered by a mask. If the script is written so that the actor and the character are totally different, you’d be surprised if someone called the actor by the character's name. But if the actor and the character are essentially the same person, the audience wouldn't even blink if the names were swapped.

CLIFS does exactly this with AI. It "masks" (hides) words like "I" or "me" and sees if the AI thinks words like "my religion" or "my party" fit perfectly in that empty space. If the AI finds that "I" and "My Group" are interchangeable in the text, it knows the person is likely experiencing Identity Fusion.

Why Does This Matter? (The "Safety Net" Application)

The researchers didn't just want to build a cool math formula; they wanted to see if it could actually predict danger. They tested CLIFS on "manifestos"—the writings of people who have committed acts of violence.

By adding the CLIFS "soul-mirror" score to existing security tools, they improved the ability to predict violence risk by a staggering 240%.

In short: While old tools were looking for "bad words" (like threats or insults), CLIFS was looking for the "emotional glue"—the deep, invisible bond that makes a person feel like they must sacrifice everything for an idea.

The Big Picture

CLIFS is like a high-tech thermometer for social tension. It doesn't just tell us what people are saying; it tells us how much they mean it. By automating this, we can move from studying a few people in a lab to understanding the massive, swirling currents of identity and emotion across the entire internet, helping us better understand both the beauty of human connection and the dangers of extreme devotion.

Technical Summary: Cognitive Linguistic Identity Fusion Score (CLIFS)

1. Problem Statement

Identity Fusion is a psychological phenomenon where an individual’s personal identity merges with a group or abstract target (e.g., religion, ideology, or nation), creating a "visceral feeling of oneness." This state is a powerful driver of both prosocial behavior (cohesion) and antisocial behavior (radicalization, violence).

Existing methods for quantifying identity fusion face significant limitations:

Traditional Measures: Scales like the Verbal Identity Fusion Scale (VIFS) rely on self-reporting, which is difficult to scale, requires direct contact, and is prone to social desirability bias.
Automated Measures: Prior NLP approaches, such as the Unquestioning Affiliation Index (UAI) or the Violence Risk Index (VRI), rely on explicit word counts or string-matching. These methods often fail to capture the subconscious, implicit metaphors that characterize deeply fused individuals and show weak correlation with established psychological gold standards.

2. Methodology

The authors propose CLIFS, a novel, automated metric that leverages Cognitive Linguistics and Large Language Models (LLMs) to quantify fusion through implicit metaphor detection.

A. Core Metric: Implicit Metaphor Detection
The central hypothesis is that fused individuals use "self" and "target" concepts interchangeably in their speech. CLIFS uses Masked Language Models (ModernBERT) to measure the conceptual proximity between these entities:

Directional Proximity ( $S_{x \to y}$ ): The model masks identity tokens (e.g., "I", "me") and calculates the probability that target-related terms (e.g., "America", "Religion") will fill the gap, and vice versa.
Reciprocity ( $f(I, T)$ ): Because fusion is a bidirectional relationship, the authors use a harmonic mean of the directional scores to ensure the metric captures the mutual reinforcement of identities.
Fictive Kinship ( $K_f$ ): Recognizing that fusion often manifests as "kin-like" bonds, the model measures how often kinship terms (e.g., "brother", "blood") are used as metaphors for the fusion target.

B. Feature Engineering & Model Architecture
CLIFS is not just a single score but an ensemble of features fed into a Random Forest classifier/regressor:

CLIFS Metrics: Fusion Proximity, Fictive Kinship, and directional scores.
Lexical Markers: Scores from UAI and VRI (affiliation and cognitive processing words).
Deep Learning Features: Semantic embeddings from SBERT and class probabilities from a fine-tuned ModernBERT.

C. Data Augmentation
To combat small dataset sizes, the authors employed Round-Trip Translation (RTT) (English $\to$ German/Chinese $\to$ English) and Generative AI (GPT-4o) to create diverse, synthetic training samples that maintain the linguistic nuances of different fusion levels.

3. Key Contributions

New NLP Task: Formally defines the task of predicting identity fusion levels from natural language.
Implicit Metaphor Framework: Moves beyond "keyword counting" to "conceptual proximity" using masked LLMs.
Scalable Automation: Provides a fully automated pipeline that outperforms human annotators in consistency and precision.
Open Science: Releases the CLIFS models and code to the research community.

4. Results

The authors evaluated CLIFS through two primary experiments:

Experiment 1 (Identity Fusion Prediction):
- Classification: CLIFS outperformed existing baselines by 6% to 154% in macro F1 scores.
- Regression: The model achieved a Spearman correlation ( $r_s$ ) of 0.633 with the VIFS gold standard, representing a massive improvement over prior methods (which showed near-zero or weak correlation).
- Human Comparison: CLIFS surpassed human annotation by 11–22% in F1 score and reduced error rates (MAE) by 25%.
Experiment 2 (Violence Risk Prediction - Proof of Concept):
- By integrating CLIFS features into the existing Violence Risk Index (VRI), the authors improved the accuracy of classifying violent manifestos by more than 240%.

5. Significance

CLIFS represents a paradigm shift in computational social science. By moving from explicit lexical analysis to implicit cognitive-linguistic analysis, it allows researchers to:

Monitor Social Dynamics: Track radicalization or social cohesion longitudinally across massive datasets (e.g., social media).
Enhance Security: Provide more accurate, automated tools for violence risk assessment in counter-terrorism.
Advance Theory: Validate psychological theories of identity fusion using large-scale, real-world linguistic data that was previously inaccessible to traditional survey methods.