PLlama at SemEval-2026 Task 4: Zero-shot Prompting with Llama-3.2 for Narrative Similarity

AbstractThis paper describes our submission to the SemEval-2026 Task 4 on Narrative Story Similarity and Narrative Representation Learning. The shared task focuses on modeling the similarity across narratives on the basis of perceived relatedness between events’ causality. The task frames narrative similarity as a binary classification problem in which the models determine which of the two stories is more narratively similar to a given anchor story. Our approach leverages the pre-trained language model Llama-3.2-3B-Instruct with prompt engineering, allowing the system to assess narrative similarity without explicit fine-tuning. On the test data, our system achieved an accuracy of approximately 55% in Track A. While modest, our results establish a baseline for narrative similarity detection in large language models (LLMs) highlighting both their potential and challenges of applying computationally efficient instruction-tuned models to this task. Our analysis highlights the struggle of LLMs in capturing event causality and long range narrative dependencies.