Cryptix at SemEval-2026 Task 4: Zero-Shot Bi-Encoder Modeling for Narrative Story Similarity - A Sentence Transformer Approach
Year: 2026
Type: conference-paper
Abstract: This paper describes our submission to SemEval-2026 Task 4 on Narrative Story Similarity and Narrative Representation Learning.The task evaluates systems on their ability to model narrative similarity between story summaries using a contrastive triple-based setup.We participate in both Track A (comparative narrative similarity) and Track B (narrative representation learning).Our approach employs a zero-shot bi-encoder architecture based on the pretrained sentence-transformers/all-mpnetbase-v2 model.Narrative similarity is modeled using cosine similarity between independently encoded story embeddings.For Track A, the system selects the candidate story with higher cosine similarity to the anchor; for Track B, we directly output dense embedding representations evaluated via cosine distance.This system achieves 59.50% accuracy on Track A (rank 37) and 57.50% accuracy on Track B (rank 25).Through quantitative margin analysis and qualitative error categorization, we find that generalpurpose semantic encoders capture topical similarity effectively but struggle to model higherlevel narrative abstractions such as causal progression and outcome alignment.
Authors: Sushmitha M, Sarath Kumar P, Thanalaxmi S, Beulah A
Language: English
Topic: Topic Modeling
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7166806008