We are delighted to announce *SemEval 2026 Task 4: Narrative Story Similarity and Narrative Representation Learning*.
In the shared task SemEval-2026 Task 4: Narrative Story Similarity and Narrative Representation Learning, you (or rather your systems) are asked to identify narratively similar stories. We define narrative similarity by three core similarity components: the abstract theme, the course of action, and the outcomes of a story.
———————
Data
———————
In total, we annotated just over 1000 triples of story summaries. All summaries are sourced from the English Wikipedia. In each triple, the annotators decide which of the two alternatives is more narratively similar to the anchor.
———————
Tracks
———————
In Track A, participants operate on triples, akin to the annotation setup. Here, we encourage testing symbolic approaches to narrative understanding or representation. In Track B, on the other hand, participants are asked to create embedding representations for each story, with the embedding distances ideally corresponding to perceived narrative similarity.
———————
Participation
———————
The development phase will last until late December, followed by a testing phase in January. Task papers are due after that. Check out our website for more details: https://narrative-similarity-task.github.io/
You can submit your results on CodaBench: https://www.codabench.org/competitions/10273/
For future communication, please join our mailing list: narrative-similarity-task@googlegroups.com
If you want to contact the organiser team directly, use: narrative-similarity-task-organizers@googlegroups.com
———————
Organisers
———————
Hans Ole Hatzel, University of Hamburg
Ekaterina Artemova, Toloka AI
Haimo Stiemer, Technical University of Darmstadt
Evelyn Gius, Technical University of Darmstadt
Chris Biemann, University of Hamburg