@inproceedings{orsanigo-etal-2026-digis,
    title = "{D}igi{S}-{FBK} at {S}em{E}val-2026 Task 9: Multi-task Learning for Multilingual and Cross-cultural Polarization Classification",
    author = "Orsanigo, Veronica  and
      Ramponi, Alan  and
      Leonardelli, Elisa",
    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-1.357/",
    pages = "2838--2851",
    ISBN = "979-8-89176-414-9",
    abstract = "Online polarization promotes social fragmentation, misinformation, hate, and toxic language. Polarization has been studied from social and communication perspectives, but it can also be addressed computationally as a text classification task. Due to the variety of polarization targets and manifestations, polarization is a complex phenomenon to study, and both detecting and characterizing it are challenging tasks.In this paper, we present the systems submitted by the DigiS-FBK team to SemEval-2026 Task 9 POLAR aimed at detecting polarization in textual content (subtask 1) and identifying its type (subtask 2) and manifestation (subtask 3) in a multilingual, multicultural, and multievent context. Considering the strong link between subtasks, we propose an approach that leverages a multi-task learning paradigm. Our results reveal that, despite the variability in scores across languages, the overall performance when using multi-task learning is higher than when adopting a single task approach in all subtasks"
}