UCSC NLP uses boundary-aware span extraction for SemEval-2026 conspiracy detection

UCSC NLP presents systems for SemEval-2026 Task 10 (PsyCoMark), addressing conspiracy marker extraction and document-level conspiracy detection. For marker extraction, the team formulates the task as multi-label span classification using boundary-aware representations and non-maximum suppression.

• Marker extraction uses IoU >= 0.95 positive labeling and hard-negative sampling.
• Document classification employs a sequence classifier with label smoothing.
• Entity-like roles are detected robustly, while abstract roles remain sensitive to boundary criteria.
• The system ranks 7th in Subtask 1 (0.2251 macro F1) and 11th in Subtask 2 (0.7694 weighted F1).