*** Second Call for Participation for HAHA at IberLEF 2026 https://sites.google.com/view/iberlef-2026 ***
Humor Analysis based on Human Annotation and Automatic Humor Generation
https://www.fing.edu.uy/inco/grupos/pln/haha/
Codabench page: https://www.codabench.org/competitions/14700/
NEWS:
The trial and development data have been released. You can now submit your systems for the development phase!
Can computers be funny? Can humans identify computer-generated humor?
While humor has been studied historically from psychological, cognitive, and linguistic perspectives, its computational study is an active area of research in Machine Learning and Computational Linguistics that has gained traction in recent years. There has been significant development mainly in the field of automatic humor detection and classification, but a characterization of humor that enables its automatic recognition and generation is far from being solved.
This task aims to gain better insight into what is humorous and what causes laughter, and to take some steps forward by assessing the capabilities of current LLMs to generate actual humorous content in Spanish and attempting to see whether it’s possible to automatically distinguish between computer-generated humor and humor written by humans. The target audience is NLP researchers interested in advancing the understanding of highly subjective and creative tasks, though anyone is welcome to participate.
Task description
This year, the HAHA evaluation campaign proposes three different subtasks related to automatic humor detection and generation, with the aim of deepening our understanding of computational humor.
Subtask 1 - Humor Detection: determining if a news headline is satirical or real. The main performance metric for this subtask will be the F1 score of the 'humorous' class. This subtask is similar to the first subtask proposed in previous editions of the HAHA shared task, but this time it's applied to a particular domain where humorous and non-humorous content might sometimes be difficult to tell apart.
Subtask 2 - LLM-generated humor detection: determining if a joke inspired by a news headline was generated by an LLM or written by a human. The main performance metric for this subtask will be the F1 score of the 'automatic' class.
Subtask 3 - Humor Generation: generating jokes from a news headline using computational methods. This subtask will be evaluated through human preference judgments, employing LLM arena-style battles between pairs of generated jokes, and ranking the systems using an Elo-based leaderboard.
How to Participate
The CodaBench page for the competition is available: https://www.codabench.org/competitions/14700/
Important Dates
March 18th, 2026: team registration page.
April 8th, 2026: development sets released and open for dev submissions.
May 27th, 2026: test sets released and open for test submissions.
June 3rd, 2026: end of test submissions, publication of results of subtasks 1 and 2.
June 10th, 2026: publication of results of subtask 3.
June 12th, 2026: paper submission.
June 23rd, 2026: notification of acceptance.
July 1st, 2026: camera-ready paper submission.
September 2026: IberLEF 2026 Workshop.