🚨 What Is This Task About?

This new SemEval task tackles a socially urgent and computationally complex challenge: detecting polarization in online texts across different languages, cultures, and events.

This task asks:
How can we build NLP systems that detect, interpret, and explain polarized language globally?

💡 Why Join This Task?

• Tackle one of the most pressing social problems of our time
• Work with rich, diverse datasets from 5 continents
• Develop models for low- and mid-resource languages
• Participate in a global NLP benchmark on polarization
• Contribute to socially responsible, explainable AI
• Gain visibility at SemEval 2026 and the ACL workshop

🌍 What's Unique About This Task?

• Multilingual coverage across high, mid, and low-resource languages
• Datasets from real-world events: elections, protests, conflicts
• Three subtasks: from detection to fine-grained explanation
• Focus on interpretability, sociocultural context, and ethical NLP
• Annotations by native speakers using expert guidelines

🧪 Task Snapshot

• Subtask 1: Detect whether a text is polarized
• Subtask 2: Classify the target of polarization (e.g., religion, politics, race)
• Subtask 3: Identify how polarization is expressed (e.g., dehumanization, stereotyping)

🗃️ Data Release Schedule

• Pilot data available: 8 August 2025
• Full training/dev/test splits: 1 September 2025
• Evaluation phase: 10 January – 31 January 2026
• Paper submission to SemEval: February 2026

📈 Who Can Participate?

This task is open to individuals or teams interested in:

• NLP and low-resource language processing
• Computational social science and ethics
• Misinformation, hate speech, and polarization
• Cross-cultural communication and peacebuilding

📥 Get Involved

• Follow us on GitHub 🔗
• Email us: polarization-semeval-2026-organisers@googlegroups.comm

🧑‍💻 Starter Kits and Baselines

• Preprocessing scripts and tokenizers
• Baseline models (transformers, multilingual)
• Annotation guidelines and tutorials
• Evaluation metrics and scripts