Year: 2026
Type: conference-paper
Abstract: This paper describes our system for Subtask 1 of SemEval-2026 Task 9: POLAR, which focuses on multilingual polarization detection.The task is formulated as a binary classification problem across 22 languages drawn from diverse online platforms and real-world events.We investigate three complementary approaches: supervised fine-tuning of multilingual encoder-only transformer models, zeroand few-shot classification using large language models (LLMs), and transfer learning from related harmful language tasks such as hate speech, toxicity, abusive language, and genderbased violence.Among the supervised models, mDeBERTa achieved the strongest baseline performance.Prompt-based methods with open-weight LLMs showed limited effectiveness, particularly in zero-shot settings.The best results were obtained using transfer learning, where the model was first fine-tuned on related task datasets and then adapted to the polarization task, achieving a Macro-F1 score of 0.781.Our findings indicate that supervised multilingual encoders remain highly effective for polarization detection and that incorporating related harmful language tasks can substantially improve performance, especially for nuanced and context-dependent expressions of polarization.
Author: Aatman Vaidya
Institution: University of Tübingen
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7166799024