@inproceedings{elschenbroich-britz-2026-aivengers,
    title = "{AI}vengers at {S}em{E}val-2026 Task 9: Utilizing Language Specific Encoders for Multilingual Text Classification",
    author = "Elschenbroich, Boon  and
      Britz, Lars",
    editor = "Kochmar, Ekaterina  and
      Ghosh, Debanjan  and
      North, Kai  and
      Komachi, Mamoru",
    booktitle = "Proceedings of the 20th {I}nternational {W}orkshop on {S}emantic {E}valuation (2026)",
    month = jul,
    year = "2026",
    address = "San Diego, California, USA",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingest-acl-workshops/2026.semeval-1.108/",
    pages = "761--776",
    ISBN = "979-8-89176-414-9",
    abstract = "Polarizing language has evolved from a social media phenomenon into a pervasive feature of public and everyday discourse across cultures and geographies. And, this is not limited to certain countries, but a world wide trend. As we will show, detecting polarization, it{'}s type and manifestation is not a simple task for one ML model, but, it requires multiple different approaches depending on the language and culture. In this paper, we provide the best methods that we found for each language in all three SemEval 2026 - Task 9 multilingual text classification challenge subtasks. We achieved the best results with language specific pre-trained BERT and RoBERTa models, rather than using a general approach and using a generic multi-language model. Our approach secured a high to medium rank in all subtasks and languages."
}