@inproceedings{tangkulung-tsuruoka-2026-utokyo,
    title = "{UT}okyo Tsuruoka Lab at {S}em{E}val-2026 Task 9: Efficient Single Forward Pass Inference for Multi-Label Polarization Classification",
    author = "Tangkulung, Howard  and
      Tsuruoka, Yoshimasa",
    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-1.211/",
    pages = "1641--1651",
    ISBN = "979-8-89176-414-9",
    abstract = "Detecting and interpreting polarized online content is increasingly crucial as online platforms become central to public information exchange. We present an efficient adaptation of large language models for multi-label polarization classification in SemEval-2026 Task 9: Detecting Multilingual, Multicultural and Multievent Online Polarization. Our single-forward-pass inference method outperforms baseline multi-step decoding approaches for multi-label classification by reducing error propagation while improving inference efficiency. Beyond performance and efficiency analysis, we investigate the cross-lingual transferability of the system, observing statistically significant generalization within language families, a result that offers a practical path for low-resource language adaptation. Our system ranked 1st in 8 languages for Subtask 1 and 6 languages for Subtask 2, and placed in the top 5 for 16 out of 22 languages across both subtasks.Overall, we provide a simple, effective, and efficient solution for multilingual polarization classification."
}