POLAR-SemEval 2026 Trial Dataset

This is a trial dataset sample for POLAR-SemEval 2026.

Shared Task Overview

This shared task includes 3 subtasks:

• Subtask 1: Detect whether a text is polarized
• Subtask 2: Classify the target of polarization - includes Gender/Sexual, Political, Religious, Racial/Ethnic, and Other categories
• Subtask 3: Identify how polarization is expressed - includes Vilification, Extreme Language, Stereotype, Invalidation, Lack of Empathy, and Dehumanization

Languages

This dataset includes the following languages:

• amh (Amharic)
• ara (Arabic)
• ben (Bengali)
• deu (German)
• eng (English)
• spa (Spanish)
• fas (Persian)
• hau (Hausa)
• hin (Hindi)
• ita (Italian)
• mya (Myanmar)
• nep (Nepali)
• rus (Russian)
• tel (Telugu)
• tur (Turkish)
• urd (Urdu)
• zho (Chinese)

Participation

Participants may participate in all three subtasks or any one or two.

Dataset Schema

The CSV file contains the following columns:

Column	Description	Subtask
Text	The original text sample	-
Lang	Language code	-
ID	Unique identifier for each text sample	-
Polarization	Boolean indicating if content is polarizing (TRUE/FALSE)	Subtask 1
Political	Binary flag (0/1) for political polarization target	Subtask 2
Racial/Ethnic	Binary flag (0/1) for racial/ethnic polarization target	Subtask 2
Religious	Binary flag (0/1) for religious polarization target	Subtask 2
Gender/Sexual	Binary flag (0/1) for gender/sexual polarization target	Subtask 2
Other	Binary flag (0/1) for other polarization targets	Subtask 2
Vilification	Binary flag (0/1) for vilification expression	Subtask 3
Dehumanization	Binary flag (0/1) for dehumanization expression	Subtask 3
Extreme Language	Binary flag (0/1) for extreme language expression	Subtask 3
Lack of Empathy	Binary flag (0/1) for lack of empathy expression	Subtask 3
Stereotype	Binary flag (0/1) for stereotyping expression	Subtask 3
Invalidation	Binary flag (0/1) for invalidation expression	Subtask 3