@inproceedings{marhoefer-etal-2026-ucsc,
    title = "{UCSC} {NLP} at {S}em{E}val-2026 Task 10: Boundary-Aware Span Extraction and {R}o{BERT}a Classification for Conspiracy Detection",
    author = "Marhoefer, Dom  and
      Suvakovic, Milos  and
      Grant-Richards, Glenn  and
      Pinero, Aidan  and
      King, Ryan",
    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-1.194/",
    pages = "1495--1500",
    ISBN = "979-8-89176-414-9",
    abstract = "We present our systems for SemEval-2026 Task10 (PsyCoMark), addressing conspiracy markerextraction (Subtask 1) and document-level con-spiracy detection (Subtask 2). For marker ex-traction, we formulate the task as multi-labelspan classification over enumerated candidatespans, using IoU{\ensuremath{\geq}}0.95 positive labeling, hard-negative sampling, and containment-based non-maximum suppression (NMS) with boundary-aware span representations. Document classi-fication is modeled independently using a se-quence classifier with label smoothing and astratified train{--}validation split. Analysis showsthat entity-like roles (Actor, Victim) are de-tected robustly, while abstract roles (Action,Effect, Evidence) remain sensitive to boundarycriteria. On the official test set, our systemsrank 7th in Subtask 1 (0.2251 macro F1) and12th in Subtask 2 (0.7694 weighted F1)."
}