@inproceedings{rabehi-etal-2026-team,
    title = "Team Macaroni at {S}em{E}val-2026 Task 10: {P}sy{C}o{M}ark: Psycholinguistic Conspiracy Marker Extraction and Detection",
    author = "Rabehi, Rofaida  and
      Plenk, Nicolai  and
      Han, Miriam",
    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-1.50/",
    pages = "338--342",
    ISBN = "979-8-89176-414-9",
    abstract = "This paper describes our submission to SemEval-2026 Task 10: PsyCoMark, which addresses span-level identification of psycholinguistic conspiracy markers and document-level conspiracy classification. For Subtask 1, we fine-tune several pretrained transformer encoders and analyse their behaviour under different training configurations. For Subtask 2, we develop a hybrid system that combines ModernBERT-large with surface-level linguistic features.Our results show that straightforward fine-tuning of strong pretrained models is more effective than more complex pipelines and that additional handcrafted features do not yield consistent improvements. On the official test set, we rank 18th in Subtask 1 (overlap-based macro F1 = 0.16) and 20th in Subtask 2 (macro F1 = 0.76)."
}