GermEval 2026: Harmful Content Detection

We are pleased to announce the second Shared Task on Harmful Content Detection in the GermEval 2026 competition series in natural language processing (NLP). Building on the success of our first edition, this year’s task once again challenges participants to develop innovative NLP methods for the automatic detection and classification of harmful and problematic content on social media. Such systems are of great practical importance both for supporting content moderation on online platforms and for assisting law enforcement agencies in analyzing large-scale communication datasets.

This year’s shared task encompasses four subtasks, each targeting a type of harmful or relevant language not often addressed in research so far:

1. Call to Action Detection: The aim is to identify commands or calls to action, such as calls for demonstrations or even criminally relevant actions.
2. Classification of Attacks on the Liberal Basic Democratic Order: A distinction is made between neutral statements, legitimate criticism, agitational content, and calls for the overthrow of the government.
3. Violence Detection: Going beyond a binary judgement, this year’s subtask requires — unlike in the previous edition — not only a decision on whether a post expresses a concerning attitude towards violence, but also the differentiation between various forms, such as propensity, endorsement, glorification, and incitement.
4. Detection of Defamatory Offences: The aim is to decide whether the statement in a post constitutes a defamatory offence under the German Criminal Code or not.

All participants will have access to rich datasets for each subtask drawn from an extremist movement’s environment, comprising up to 16,000 posts and comments annotated by students and staff at the University of Applied Sciences Mittweida.

Join us: We warmly invite researchers and practitioners alike to take part and contribute their ideas to this important challenge. Please register in advance of the official competition launch on 26 March, if possible.

Detailed information on the task, further deadlines, and the registration form can be found here. For questions about registration or participation, please feel free to contact us over the Competition forum or by email.