@inproceedings{rosenfeld-etal-2026-jct,
    title = "{JCT} at {S}em{E}val-2026 Task 4: A Multi-Method Approach to Narrative Story Similarity",
    author = "Rosenfeld, Dvori  and
      Walles, Rinat  and
      Liebeskind, Chaya",
    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-1.387/",
    pages = "3089--3094",
    ISBN = "979-8-89176-414-9",
    abstract = "Narrative similarity detection involves under-standing the underlying structure of a storyrather than just matching similar words orphrases. This paper details our multi-strategyapproach to the SemEval-2026 Task on Nar-rative Similarity, which requires identifyingwhich of two candidate stories most closelyresembles an anchor story based on three di-mensions: abstract themes, the sequence ofevents, and the final outcomes.We developed three distinct but complemen-tary methods to address this challenge. First,we fine-tuned a specialized story-embeddingmodel using parameter-efficient techniques onsynthetic data. Second, we utilized a ``Distill-then-Embed'' workflow, where a large languagemodel extracts the essential narrative core ofeach story before computing similarity. Third,we employed direct zero-shot prompting to al-low a high-reasoning model to compare thestories organically.Our analysis reveals that each approach excelsat different types of narrative comparisons, andtheir combination leads to robust performance.We demonstrate the importance of narrative dis-tillation in removing surface-level distractorsand the effectiveness of carefully engineeredprompts in guiding language models to focuson narrative structure"
}