There is a newer version of the record available.

Published February 2, 2023 | Version 1.1
Dataset Open

PAN23 Trigger Detection

  • 1. Bauhaus-Universität Weimar
  • 2. Leipzig University


This is the dataset for the shared task on Trigger Detection at PAN@CLEF2023. Please consult the task's page for further details on the format, the dataset's creation, and links to baselines and utility code.

Task: In trigger detection, we want to assign trigger warning labels to documents that contain potentially discomforting or distressing (triggering) content. We model trigger detection as a multi-label document classification challenge: assign each document all appropriate trigger warnings, but not more. All warnings are chosen from the author's perspective, i.e. the work's author decided which kind of trigger the document contains.

Dataset: This dataset contains annotated works of fanfiction, extracted from (AO3). Each work is between 50 and 6,000 words long and has between 1 and many trigger warnings assigned. Our training dataset contains 307,102 examples, with 17,104 in validation and 17,040 in the test split. The label set contains 32 different trigger warnings. All labels are based on the freeform content warnings added to a fanwork by its author.



  • 1.0: initial upload
  • 1.1 fixed a minor bug where some works in the labels.jsonl contained labels that are not used in the competition (heteronormativity and religious-discrimination). Those labels have been removed.  


Files (1.8 GB)

Name Size Download all
1.8 GB Preview Download