ChulaNLP at SemEval-2026 Task 4: Neural Aspect Composition for Narrative Story Embeddings

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Natural Language Processing, Data Science & Analytics · Depth: Expert, medium

Summary

ChulaNLP's Neural Aspect Composition addresses SemEval-2026 Task 4 by generating vector embeddings for narrative story comparison, a task where traditional representations often fail to capture multi-faceted story elements like theme and plot. The proposed method utilizes a Large Language Model (LLM) to decompose stories into 13 distinct semantic narrative aspects, including theme, course of action, and outcomes. Each of these aspects is then independently encoded using an encoder model, and a trained weighting layer assigns a global importance weight to each aspect. This approach achieved official test scores of 0.64 on Track A and 0.61 on Track B. Validation demonstrated its superior performance compared to direct raw story text encoding and a sentence-averaging baseline. Analysis of learned weights indicated that thematic elements and narrative resolutions were the primary factors driving perceived similarity, receiving higher importance than intermediate plot events or minor details like character introductions.

Key takeaway

For NLP Engineers developing narrative understanding systems, you should consider adopting an aspect-based embedding strategy. This method, which decomposes stories into weighted semantic components, demonstrably improves similarity comparisons over traditional raw text encoding. Focus your model's attention on thematic elements and narrative resolutions, as these are critical drivers of perceived story similarity. Implementing an LLM for initial aspect decomposition can enhance the accuracy of your narrative embeddings.

Key insights

Decomposing narratives into weighted semantic aspects significantly improves story similarity embeddings.

Principles

• Thematic elements and resolutions drive narrative similarity.
• Aspect-based encoding outperforms raw text or sentence averaging.

Method

An LLM decomposes stories into 13 aspects, which are then separately encoded and weighted by a trained layer to form embeddings.

In practice

• Prioritize thematic and resolution aspects in narrative analysis.
• Use LLMs for granular story decomposition before encoding.

Topics

• Narrative Embeddings
• SemEval-2026 Task 4
• Neural Aspect Composition
• Large Language Models
• Semantic Narrative Aspects
• Story Similarity

Best for: Research Scientist, AI Scientist, NLP Engineer

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.
• AspectRAG: An Architecture of Retrieval and Generation for Aspect-Based Sentiment Analysis — AspectRAG leverages aspect-guided retrieval and generation for unsupervised, high-performance sentiment analysis in Portuguese.
• Characterizing Narrative Content in Web-scale LLM Pretraining Data — Narrative structure in web-scale LLM pretraining data is measurable and unevenly distributed, impacting model reasoning.
• Which sentence is doing the most work in your favourite novel ? I tried to find out. — Machine learning can identify semantically distinctive sentences in texts, revealing structural or narrative pivots.
• Principles of Concept Representation in Sentence Encoders — Sentence encoders' concept representation quality hinges on matching supervision to semantic composition types.
• Solving the Embedding Step in NLP Projects — Effective embeddings are crucial for NLP model generalization and reliable downstream task performance.

Open in AIssential →

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