ILab-NLP at SemEval-2026 Task 9: Comparing XLM-RoBERTa and LLaMA-2 for Multilingual Polarization Detection

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, quick

Summary

ILab-NLP's system for SemEval-2026 Task 9, Subtask 1, addresses binary detection of polarized versus non-polarized posts in English and Spanish. The team compared a fine-tuned multilingual encoder model, XLM-RoBERTa, against a prompted generative model, LLaMA-2 7B. Experiments revealed XLM-RoBERTa delivered stronger and more stable performance overall. LLaMA-2, conversely, showed a propensity for false positives in Spanish, attributed to a significant bias towards predicting the polarized class. Beyond headline results, the analysis included model behavior using confidence signals and SHAP, alongside efficiency measurements via CodeCarbon to detail performance-cost tradeoffs.

Key takeaway

For NLP engineers developing multilingual polarization detection systems, XLM-RoBERTa offers superior stability and performance compared to LLaMA-2 7B. You should prioritize fine-tuned encoder models for robust classification, especially when false positives are critical. If you consider generative models like LLaMA-2, rigorously test for language-specific biases, particularly a strong inclination towards polarized class predictions, and implement mitigation strategies to ensure reliable output.

Key insights

Fine-tuned encoder models like XLM-RoBERTa generally outperform prompted generative models such as LLaMA-2 for multilingual polarization detection.

Principles

• Encoder models provide stable performance for classification.
• Generative models can exhibit strong class bias, especially in specific languages.

Method

The study compared fine-tuned XLM-RoBERTa with prompted LLaMA-2 7B for binary classification, analyzing model behavior using confidence signals, SHAP, and computational efficiency with CodeCarbon.

In practice

• Prioritize XLM-RoBERTa for stable multilingual polarization detection.
• Carefully evaluate LLaMA-2 for bias, particularly in Spanish text.
• Use CodeCarbon to measure model efficiency tradeoffs.

Topics

• Multilingual NLP
• Polarization Detection
• XLM-RoBERTa
• LLaMA-2
• SemEval-2026
• Model Comparison
• Computational Efficiency

Best for: AI Engineer, Research Scientist, AI Scientist, Machine Learning Engineer, NLP Engineer

Related on AIssential

• mdok-style at SemEval-2026 Task 9: Finetuning LLMs for Multilingual Polarization Detection — Finetuning mid-sized LLMs with QLoRA and augmented data improves multilingual online polarization detection.
• Detecting HIV-Related Stigma in Clinical Narratives Using Large Language Models — LLMs can effectively detect HIV-related stigma in clinical notes, with encoder models outperforming generative models in zero-shot settings.
• Indi-RomCoM: Code-Mixed Benchmark for Evaluating LLMs on Romanized Indic-English Instructions — LLMs consistently underperform on Romanized Code-Mixed instructions, with performance degrading as code-mixing density increases.
• I spent 4 days training an AI model to understand my country. It barely works — Low-resource NLP projects face significant challenges in data collection, translation, and accurate labeling, often yielding modest baselines.
• Ancient Greek to Modern Greek Machine Translation: A Novel Benchmark and Fine-Tuning Experiments on LLMs and NMT Models — A new AG-MG parallel corpus and benchmark significantly advance Ancient Greek to Modern Greek machine translation.
• Cross-lingual Relation Extraction with Large Language Models: Zero-Shot, Few-Shot, and Fine-Tuned Evaluation on Romanian — QLoRA fine-tuning significantly boosts cross-lingual relation extraction performance for low-resource languages, outperforming prompt-only LLMs.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.