PSK at SemEval-2026 Task 9: Multilingual Polarization Detection Using Ensemble Gemma Models with Synthetic Data Augmentation

TL;DR

This paper describes a multilingual polarization detection system using fine-tuned Gemma models with synthetic data augmentation, achieving high accuracy across 22 languages in SemEval-2026 Task 9.

Contribution

The authors introduce a novel ensemble approach combining large Gemma models and synthetic data strategies, improving multilingual polarization classification performance.

Findings

Achieved a mean macro-F1 of 0.811 across 22 languages.

Per-language threshold tuning improved F1 by 2-4%.

Alternative architectures underperformed on test data, emphasizing generalization importance.

Abstract

We present our system for SemEval-2026 Task 9: Multilingual Polarization Detection, a binary classification task spanning 22 languages. Our approach fine-tunes separate Gemma~3 models (12B and 27B parameters) per language using Low-Rank Adaptation (LoRA), augmented with synthetic data generated by a large language model (LLM). We employ three synthetic data strategies (direct generation, paraphrasing, and contrastive pair creation) using GPT-4o-mini, with a multi-stage quality filtering pipeline including embedding-based deduplication. We find that per-language threshold tuning on the development set yields 2 to 4\% F1 improvements without retraining. We also use weighted ensembles of 12B and 27B model predictions with per-language strategy selection. Our final system achieves a mean macro-F1 of 0.811 across all 22 languages, ranking 2nd overall of the participating teams, with 1st place finishes in 3 languages and top-3 in 8 languages. We also find that alternative architectures (XLM-RoBERTa, Qwen3) that showed strong development set performance suffered 30 to 50\% F1 drops on the test set, highlighting the importance of generalization.