CophiWue at SemEval-2026 Task 4: Symbolic Narrative Profiling with Taxonomy-Guided Extraction and Contrastive Fine-Tuning

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

Summary

CophiWue's system for SemEval-2026 Task 4, particularly Track B (narrative embedding), employs a "Decompose & Align Cycle" approach. This method transforms each story into a structured NarrativeProfile, detailing abstract themes, a five-step course of action, and an outcome. An initial NarrativeTaxonomy is built from these extractions using agglomerative clustering, creating controlled vocabularies that guide a second extraction pass to standardize profiles across the dataset. Finally, the Qwen3-Embedding8B model is contrastively fine-tuned on profile text representations using TripletLoss to derive robust story embeddings. For Track A, the system adapts the provided baseline script, substituting Gemini 3 Pro as the judge with the organizers' default prompt on raw story texts.

Key takeaway

For NLP engineers or AI scientists developing systems for complex narrative analysis, CophiWue's structured approach offers a robust method for generating consistent narrative profiles and high-quality story embeddings. You should consider integrating taxonomy-guided extraction and contrastive fine-tuning into your own narrative understanding pipelines to improve data standardization and embedding performance, especially for large datasets.

Key insights

Symbolic narrative profiling uses taxonomy-guided extraction and contrastive fine-tuning for robust story embeddings.

Principles

• Structured narrative profiles enhance story understanding.
• Taxonomy-guided extraction standardizes narrative elements.
• Contrastive fine-tuning improves embedding quality.

Method

Convert stories to NarrativeProfiles (themes, 5-step action, outcome), build NarrativeTaxonomy via agglomerative clustering, guide second extraction for standardization, then contrastively fine-tune Qwen3-Embedding8B with TripletLoss for embeddings.

In practice

• Apply agglomerative clustering for vocabulary control.
• Use TripletLoss for narrative embedding fine-tuning.
• Integrate Gemini 3 Pro for judging raw story texts.

Topics

• SemEval-2026 Task 4
• Narrative Profiling
• Taxonomy-Guided Extraction
• Contrastive Fine-Tuning
• Qwen3-Embedding8B
• Story Embeddings

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, 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.
• 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.
• StoryCoder: Narrative Reformulation for Structured Reasoning in LLM Code Generation — Narrative reformulation of code problems significantly improves LLM code generation by structuring context.
• Specializing a Small Language Model for Closed-Domain Portuguese RAG using Knowledge Graph Supervision — Fine-tuning SLMs with knowledge graph supervision significantly enhances closed-domain RAG performance, surpassing larger LLMs.
• Semantic marking of named entities in The Lusiads — Hybrid NER workflows improve accuracy for historical literary texts, overcoming limitations of models trained on modern corpora.

Open in AIssential →

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