DUTIR at SemEval-2026 Task 4: Narrative Story Similarity and Narrative Representation Learning

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Software Development & Engineering · Depth: Advanced, quick

Summary

DUTIR's approach for SemEval 2026 Task 4 focuses on narrative story similarity and representation learning. Their primary method for Track A involved fine-tuning a large language model using Low-Rank Adaptation (LoRA), incorporating data cleaning, and employing a multi-prompt strategy. This system was trained on the official synthetic dataset and achieved an official score of 0.70 on Track A. Additionally, the authors explored an alternative contrastive learning framework, initially designed for Track B, to learn narrative-structure embeddings. This framework was then applied to Track A through similarity comparisons. The analysis from this work suggests that a direct supervised adaptation approach is likely more effective for narrative reasoning tasks compared to the contrastive learning method explored.

Key takeaway

For NLP Engineers developing narrative understanding systems, you should prioritize direct supervised adaptation over contrastive learning for tasks like story similarity. Fine-tuning large language models with Low-Rank Adaptation (LoRA) and employing multi-prompt strategies, alongside robust data cleaning, can yield reasonable performance. Consider these techniques to efficiently adapt LLMs for specific narrative reasoning challenges, aiming for higher accuracy on complex semantic tasks.

Key insights

Direct supervised adaptation of LLMs is more effective for narrative reasoning than contrastive learning.

Principles

• LoRA fine-tuning improves LLM adaptation.
• Data cleaning enhances model performance.
• Multi-prompt strategies aid complex tasks.

Method

Fine-tune a large language model with Low-Rank Adaptation (LoRA), clean data, and use a multi-prompt strategy on synthetic datasets for narrative similarity.

In practice

• Apply LoRA for efficient LLM fine-tuning.
• Implement multi-prompting for complex NLP.
• Prioritize supervised methods for narrative tasks.

Topics

• SemEval 2026
• Narrative Story Similarity
• Large Language Models
• Low-Rank Adaptation
• Contrastive Learning
• Narrative Representation Learning

Best for: Research Scientist, AI Scientist, NLP Engineer, AI Student

Related on AIssential

• ttda704 at SemEval-2026 Task 4: Modeling Narrative Structures via Pseudonymization and Multi-View Sentence Alignment — Using contrastive learning with fine-tuned sentence transformers and multi-view alignment effectively models narrative structures and similarity.
• What is LoRA actually? — LoRA efficiently adapts large models by learning low-rank weight updates, drastically reducing parameters without sacrificing quality or adding latency.
• Narrative Theory-Driven LLM Methods for Automatic Story Generation and Understanding: A Survey — LLMs bridge abstract narrative theories with NLP, enabling scalable story generation and understanding despite definitional challenges.
• System Report for CCL25-Eval Task 5: New Dataset and LoRA-Fine-Tuned Qwen2.5 — Domain-specific datasets and LoRA fine-tuning significantly enhance LLM performance in specialized tasks like classical poetry analysis.
• SemEval-2026 Task 4: Narrative Story Similarity and Narrative Representation Learning — The NSNRL task operationalizes narrative similarity for evaluation and identifies areas for automated system improvement.
• Multilinguality of Large Language Models From a Structural Perspective — LLMs' multilinguality involves structural differences, with low-resource languages diverging more from English.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.