MKJ at SemEval-2026 Task 9: A Comparative Study of Generalist, Specialist, and Ensemble Strategies for Multilingual Polarization
Apr 13, 2026·
·
0 min read
Maziar Kianimoghadam Jouneghani
Abstract
We present a systematic study of multilingual polarization detection across 22 languages for SemEval-2026 Task 9 (Subtask 1), contrasting multilingual generalists with language-specific specialists and hybrid ensembles. While a standard generalist like XLM-RoBERTa suffices when its tokenizer aligns with the target text, it may struggle with distinct scripts (e.g., Khmer, Odia) where monolingual specialists yield significant gains. Rather than enforcing a single universal architecture, we adopt a language-adaptive selection strategy that chooses among multilingual generalists, language-specific specialists, and hybrid ensembles based on development performance. Additionally, cross-lingual augmentation via NLLB-200 yielded mixed results, often underperforming native architecture selection and degrading morphologically rich tracks. Our final system achieves an overall macro-averaged F1 score of 0.796 and an average accuracy of 0.826 across all 22 tracks. Code and final test predictions are publicly available at: https://github.com/Maziarkiani/SemEval2026-Task9-Subtask1-Polarization.
Type
Publication
Proceedings of the 20th International Workshop on Semantic Evaluation (SemEval-2026)
Authors
PhD Candidate | NLP Researcher
Hi! My name is Maziar, though you can call me Maaz. I’m currently a PhD candidate in Computational Sciences and Technological Innovation, specializing in Artificial Intelligence and Data Science, at the DISIT department, University of Eastern Piedmont (UPO). I’m part of the ARC2S group, working under the supervision of Prof. Giancarlo Ruffo and Prof. Anna Sapienza. I finished my Master’s in Language Technologies and Digital Humanities at the University of Turin in July 2026.
My research over the past two years has centered on multilingual and cross-cultural NLP, and on harmful online content, specifically how it spreads and how it is perceived differently across cultures and communities.
This interest comes from my background. I grew up bilingual in Persian and Qashqai Turkish (a Turkic language of the Oghuz branch), and then did my bachelor’s in English Language Translation, with a focus on linguistics and translation studies. This is the starting point of my interest in meaning, cultures, and languages. We didn’t have interdisciplinary NLP master’s program in my country that I could join with that background (and we still don’t!), so I applied abroad and came to Italy to study NLP.
My master’s thesis, supervised by Prof. Viviana Patti and Dr. Marco Antonio Stranisci, builds on a community-developed dataset on information disorder, InDor. The idea was that an LLM’s fluent reasoning about manipulative content is not necessarily culturally aligned. So I thought maybe retrieving items and explanations from native speakers based on semantic similarity (instead of static few-shot) could fill this gap a bit. I built this retrieval framework and tested it on Persian and Italian, with automatic metrics, a qualitative analysis, and a human evaluation study.
In general, my research interests are: Multilingual & Cross-Cultural NLP, Computational Social Science (CSS), Explainable AI, Fairness, In-Context Learning, Human-in-the-Loop AI, and Information Disorder.
Beyond academia, I spent over 3 years in SEO, web content, and digital marketing.
My research over the past two years has centered on multilingual and cross-cultural NLP, and on harmful online content, specifically how it spreads and how it is perceived differently across cultures and communities.
This interest comes from my background. I grew up bilingual in Persian and Qashqai Turkish (a Turkic language of the Oghuz branch), and then did my bachelor’s in English Language Translation, with a focus on linguistics and translation studies. This is the starting point of my interest in meaning, cultures, and languages. We didn’t have interdisciplinary NLP master’s program in my country that I could join with that background (and we still don’t!), so I applied abroad and came to Italy to study NLP.
My master’s thesis, supervised by Prof. Viviana Patti and Dr. Marco Antonio Stranisci, builds on a community-developed dataset on information disorder, InDor. The idea was that an LLM’s fluent reasoning about manipulative content is not necessarily culturally aligned. So I thought maybe retrieving items and explanations from native speakers based on semantic similarity (instead of static few-shot) could fill this gap a bit. I built this retrieval framework and tested it on Persian and Italian, with automatic metrics, a qualitative analysis, and a human evaluation study.
In general, my research interests are: Multilingual & Cross-Cultural NLP, Computational Social Science (CSS), Explainable AI, Fairness, In-Context Learning, Human-in-the-Loop AI, and Information Disorder.
Beyond academia, I spent over 3 years in SEO, web content, and digital marketing.