🔬 physics

Relationship between ideology and language in the Catalan independence context

This study analyzes Twitter data to reveal that while political polarization regarding Catalan independence creates a bimodal opinion distribution where active users hold extreme views, language use patterns show that pro-independence supporters speak both Catalan and Spanish, whereas anti-independence and a distinct intermediate third pole predominantly use Spanish.

Original authors: Julia Atienza-Barthelemy, Samuel Martin-Gutierrez, Juan C. Losada, Rosa M. Benito

Published 2026-02-25

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Julia Atienza-Barthelemy, Samuel Martin-Gutierrez, Juan C. Losada, Rosa M. Benito

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

Imagine a massive, chaotic town square where everyone is shouting about whether their neighborhood should become its own independent city or stay part of the larger country. This is the Catalan independence debate, and the authors of this paper decided to listen in on the digital version of this town square: Twitter.

Here is the story of what they found, explained simply.

1. The Setup: A Digital Echo Chamber

The researchers looked at millions of tweets posted between September and November 2017. This was a very intense time in Spain, marked by a controversial independence referendum, a declaration of independence, and the Spanish government stepping in to take control.

They didn't just count tweets; they looked at who was retweeting whom. Think of a retweet as a nod of agreement. If Person A tweets something and Person B retweets it, Person B is essentially saying, "I agree with this, and I want my friends to hear it too." By mapping these nods, the researchers built a giant web of influence.

2. The "Elite" and the "Listeners"

To figure out what people actually thought, the researchers needed a starting point. They identified two groups:

The Elites (The Seeds): These are the super-influential users (like politicians and famous activists) who have very strong, unchanging opinions. Some are 100% for independence; others are 100% against it. They are the "seeds" that plant ideas.
The Listeners (The Crowd): Everyone else. The researchers used a mathematical model (like a game of "telephone" but in reverse) to guess the opinions of the listeners. If a listener keeps retweeting the "Pro-Independence" seeds, their opinion score moves toward "Pro." If they retweet the "Anti-Independence" seeds, their score moves toward "Anti."

3. The Big Surprise: It's Not Just Two Sides

Most people assume political fights are like a tug-of-war: Team A vs. Team B.

The Finding: The researchers found that while there were indeed two big teams, there was a third, middle team.
The Analogy: Imagine a seesaw. You expect people to be either on the far left or the far right. But they found a group of people sitting right in the middle, slightly leaning toward the "Pro-Independence" side, but not fully committed. This "Third Pole" represents people who are undecided or have moderate views.

4. The "Super-Users" Are More Extreme

The study looked at who talks the most.

The Finding: The people who tweet the most, get the most retweets, and show up every day are the most extreme.
The Analogy: Think of a loud party. The people shouting the loudest and dancing on the tables are usually the ones with the strongest feelings. The quiet people in the corner (who tweet less) are more likely to have mixed or moderate opinions. As you filter for the "loudest" users, the middle group shrinks, and the two extreme sides get sharper and more distinct.

5. The Language Connection: The "Accent" of Opinion

This is the most fascinating part. In Catalonia, people speak two languages: Spanish and Catalan. The researchers wanted to see if your political opinion matched the language you used.

The "Anti-Independence" Group: These users spoke almost exclusively Spanish. It was like a uniform; if you were against independence, you spoke Spanish.
The "Pro-Independence" Group: These users were bilingual. They spoke both Catalan and Spanish, mixing them freely.
The "Middle" Group (The Third Pole): Here is the twist. Even though this group was politically closer to the "Pro-Independence" side, they spoke like the "Anti-Independence" side. They mostly spoke Spanish.

The Metaphor: Imagine a sports team. The "Pro-Independence" team wears red and blue jerseys (Catalan and Spanish). The "Anti-Independence" team wears only red jerseys (Spanish). The "Middle" team, even though they are fans of the red-and-blue team, decided to wear only red jerseys. They are politically close to one side, but culturally (linguistically) they behave like the other.

6. Why Does This Matter?

The paper shows that political polarization isn't just about "Yes" or "No." It's a complex landscape where:

Activity matters: The loudest voices are the most extreme.
Language is a signal: Speaking a certain language is a strong clue about your politics, but it's not a perfect map (as seen with the "Middle" group).
The Middle exists: Even in a highly divided society, there is a significant group of people who don't fit neatly into the two extreme boxes.

In a nutshell: The researchers used Twitter to prove that in a heated political fight, the people shouting the loudest are the most extreme, and while your language often reveals your politics, sometimes the "middle ground" speaks the language of the opposition even if their heart is with the other side.

1. Problem Statement

The paper addresses the challenge of understanding political polarization and its relationship with language use in the context of the Catalan independence movement. While sociological studies often link territorial disputes to language, most existing research relies on discrete categorizations of opinion (e.g., pro vs. anti) and aggregated data. The authors aim to:

Infer a continuous opinion distribution among Twitter users regarding Catalan independence, moving beyond binary classifications.
Analyze the interplay between political ideology and language choice (Catalan vs. Spanish) at the individual user level.
Determine if polarization observed in political elites translates to the general population and how user activity levels influence these distributions.

2. Methodology

The study employs a computational social science approach using a large-scale Twitter dataset collected between September 15, 2017, and November 3, 2017, covering key events like the independence referendum and the subsequent unilateral declaration of independence.

A. Data Collection and Preprocessing

Dataset: 36,090,661 tweets from 2,511,319 users, retrieved via the Twitter Streaming API using a broad, neutral set of keywords in both Catalan and Spanish (e.g., "procés," "referendum," "155").
Network Construction: A directed, weighted retweet network was built. Retweets were chosen as the proxy for influence and agreement, assuming that retweeting implies backing the original message.

B. Opinion Inference Model

The authors adapted a model based on the DeGroot process to infer continuous opinions ( $X \in [-1, 1]$ ) from network interactions:

Elite Selection: A subset of "Elite" users was identified as opinion seeds. Selection criteria included:
- High Engagement: Participation ratio > 80% of the time interval.
- High Influence: > 5,500 retweets received.
- Community Structure: Using a Nested Stochastic Block Model, the authors identified two distinct communities within this elite group: one containing politicians strictly in favor of independence (assigned $X = +1$ ) and one strictly against (assigned $X = -1$ ).
Listeners: The remaining users ("Listeners") were initialized with a neutral opinion ( $X=0$ ).
Iterative Update: Listeners iteratively updated their opinions based on the weighted average of their incoming neighbors (those they retweeted) until convergence:
$X_i(t) = \frac{\sum_j A_{ij} X_j(t-1)}{k_{out}^i}$
Where $A_{ij}$ is the adjacency matrix and $k_{out}^i$ is the out-degree.

C. Language Analysis

Language Detection: The langdetect library was used to classify tweets as Catalan or Spanish (precision/recall > 97%).
Language Index ( $L_i$ ): Defined for each user as the average value of their tweets, where Catalan = $+1$ and Spanish = $-1$. This yields a continuous index $L_i \in [-1, 1]$ .

D. Quantitative Metrics

Three parameters were used to characterize the opinion distribution:

Pole Distance ( $d$ ): Normalized distance between the gravity centers of the pro- and anti-independence poles.
Population Difference ( $\Delta A$ ): Normalized difference in the size of the two poles.
Polarization Index ( $\mu$ ): Defined as $\mu = (1 - \Delta A)d$ , measuring the degree of polarization (maximized when two equal-sized groups hold opposite views).

3. Key Results

A. Opinion Distribution Characteristics

Bimodality with a Third Pole: The overall distribution is not strictly binary. It exhibits a bimodal behavior with a distinct third intermediate pole.
- Global Distribution: Roughly 55% (anti) vs. 45% (pro), with low polarization ( $\mu = 0.35$ ) due to the proximity of the poles ( $d = 0.39$ ).
- Active Users: When filtering for users with >10 tweets, the pro-independence pole grows to 62%, and the third pole becomes more distinct.
Activity Correlation: There is a strong correlation between user activity and ideological extremity.
- Active/Influential/Engaged Users: As thresholds for activity (tweets), influence (retweets received), and engagement (days active) increase, the distribution narrows around the three poles.
- Entropy: The entropy of the opinion distribution decreases significantly as user activity increases (from $h_{10}=0.54$ to $h_{210}=0.08$ ), indicating that the most active users hold more extreme and less variable opinions.
Temporal Dynamics: During major offline events (e.g., the referendum day), the opinion distribution becomes more diverse (broader), while periods of low activity show a return to strict bimodality at the extremes.

B. Interplay Between Ideology and Language

A clear, non-linear relationship exists between political opinion ( $X$ ) and language use ( $L$ ):

Anti-Independence Users ( $X \in [-1, 0.3]$ ): Speak almost exclusively Spanish ( $L \approx -1$ ).
Pro-Independence Users ( $X \in [0.3, 1]$ ): Exhibit a wider range of language use, shifting monotonically toward Catalan as the opinion becomes more extreme.
The "Third Pole" Anomaly: Users in the intermediate/central pole (closer to pro-independence in opinion, $X \approx 0.2$ ) behave linguistically like the anti-independence group, speaking almost exclusively Spanish.
Correlation: The Spearman correlation coefficient between the opinion index and language index is 0.46, confirming a significant but not deterministic relationship.

4. Key Contributions

Continuous Opinion Inference: The study moves beyond binary classification, successfully inferring a continuous opinion spectrum that reveals a spontaneous third intermediate pole, likely representing moderate or undecided political stances.
Granular User-Level Analysis: Unlike previous aggregated studies, this work analyzes data at the individual user level, revealing that active users are more polarized and that the "third pole" behaves differently linguistically than its political proximity suggests.
Language-Ideology Decoupling: The paper demonstrates that political alignment does not always predict language choice linearly. Specifically, the intermediate political group aligns linguistically with the anti-independence camp (Spanish), challenging the assumption that pro-independence sentiment is the sole driver of Catalan language use in this context.
Methodological Framework: The authors provide a robust framework combining retweet network analysis, DeGroot opinion dynamics, and community detection to study complex socio-political phenomena.

5. Significance

This research offers critical insights into the mechanisms of polarization in territorial disputes. It highlights that:

Polarization is not uniform: The most vocal and influential members of a society are often the most extreme, while the "silent" majority may hold more moderate views.
Language is a complex marker: In the Catalan context, speaking Catalan is not a universal marker of pro-independence sentiment; rather, it is a marker of extreme pro-independence sentiment. Moderate or intermediate political stances often retain the use of Spanish.
Policy and Social Implications: Understanding these nuances is vital for policymakers and sociologists, as it suggests that bridging the divide requires addressing not just political positions but also the linguistic and cultural identities associated with different levels of political engagement.

The study concludes that the proposed methodology effectively uncovers complex patterns in the ideological landscape, providing a template for analyzing other social dimensions (e.g., gender, income) in polarized environments.