LIAAD INESCTEC at SemEval-2026 Task 4: Unsupervised Narrative Similarity via Discourse Representation Structures and Sentence Embeddings

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

Summary

LIAAD INESCTEC presented an unsupervised method for SemEval-2026 Task 4, which addressed the challenge of Narrative Similarity. This task was structured into two tracks, with the team focusing solely on Track A. In Track A, the system received a triplet consisting of an anchor story, story A, and story B, and its objective was to identify which of stories A or B was more similar to the anchor. Their approach involved parsing each narrative into a Discourse Representation Structure (DRS) format, subsequently extracting specific features from the DRS. After conducting ablation experiments on a development dataset, their strategy achieved an accuracy of 0.5975 on the official blind test set. This demonstrates a practical application of DRS for semantic comparison of narratives.

Key takeaway

For NLP Engineers developing narrative understanding systems, consider integrating Discourse Representation Structures (DRS) into your semantic parsing pipeline. This unsupervised approach, which achieved 0.5975 accuracy on SemEval-2026 Task 4, demonstrates that robust narrative similarity can be derived from structural features. You should explore DRS-based feature extraction to enhance your models' ability to compare complex story semantics without extensive labeled data. This could significantly reduce annotation costs for new narrative datasets.

Key insights

Unsupervised narrative similarity can be effectively determined by parsing stories into Discourse Representation Structures and extracting features.

Principles

• Discourse Representation Structures encode narrative meaning.
• Feature extraction from DRS enables semantic comparison.
• Unsupervised approaches can achieve competitive narrative similarity.

Method

Stories are parsed into Discourse Representation Structures (DRS). Features are then extracted from the DRS format. These features are used to determine narrative similarity between story triplets.

In practice

• Implement DRS parsing for complex text analysis.
• Extract structural features from DRS for semantic tasks.
• Design unsupervised systems for narrative comparison.

Topics

• Narrative Similarity
• Discourse Representation Structures
• SemEval-2026
• Unsupervised Learning
• Semantic Parsing
• Feature Extraction

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.
• 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.
• 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.
• Finding Video Games with Sense2Vec — Sense2Vec's contextualized phrase embeddings, trained on Reddit, effectively identify domain-specific entities like video games and their abbreviations.
• Discourse Relations and Connectives in Higher Text Structure — Local discourse analysis can reveal clues about higher text structure and global coherence.
• What They Saw, Not Just Where They Looked: Semantic Scanpath Similarity via VLMs and NLP metric — VLMs and NLP metrics can quantify semantic similarity in eye-tracking scanpaths, complementing traditional spatial analysis.

Open in AIssential →

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