@inproceedings{bevers-etal-2026-duluth,
    title = "{D}uluth at {S}em{E}val-2026 Task 4: A Hybrid Approach to Narrative Similarity using Bi-Encoder Embeddings with Cross-Encoder Tie breaking using Learned Weights",
    author = "Bevers, Maxwell  and
      Carlson, Aidan  and
      Pedersen, Ted",
    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-1.127/",
    pages = "927--931",
    ISBN = "979-8-89176-414-9",
    abstract = "We present a hybrid system for SemEval-2026 Task 4 on Narrative Similarity. Our approach decomposes the stories into four narrative components: theme, plot, emotion, and outcome. Each component is then encoded using a biencoder (all-mpnet-base-v2), and cosine similarities are combined through a learned pairwise ranking model. When similarity scores between candidate stories fall within a small margin of error, a cross-encoder (ms-marcoMiniLM-L-6-v2) is used as a tie-breaker. Our final system achieves 58.5{\%} accuracy on the official test set, placing us at 39th overall. Error analysis reveals that the system struggles with complex themes, multiple protagonists, and contrasting outcomes."
}