Humor understanding has attracted increasing attention in recent NLP research; however, the automatic generation of humorous content remains a particularly challenging problem, especially for underrepresented languages such as Arabic. Humor generation requires not only linguistic fluency, but also cultural awareness, creativity, and the ability to surprise while remaining socially appropriate.
ARHAHA 2026 is a shared task dedicated to Arabic humor generation, aiming to provide a structured evaluation framework for studying creative language generation in Arabic. The task focuses on encouraging systems to produce original, safe, and culturally grounded humorous texts, rather than reproducing memorized jokes.
To register, please fill the form
To participate (and download the input data), please visit our CodaBench page.
Despite the richness and diversity of Arabic linguistic and cultural traditions, there is currently no standardized benchmark for Arabic humor generation. ARHAHA addresses this gap by introducing carefully designed constraints and evaluation protocols that promote novelty and fairness, while allowing meaningful comparison between systems.
Participants are invited to develop models that generate short humorous texts in Arabic under explicit constraints. Evaluation combines automatic validation of format and constraints with human assessment of humor quality, originality, fluency, and cultural appropriateness.
The task aims to advance research on creative language generation for Arabic and has potential applications in conversational agents, entertainment-oriented AI, educational tools, and culturally aware human–computer interaction.
The goal of the shared task is to develop systems capable of generating genuinely humorous content under various constraints. By imposing constraints, we foster uniqueness and ensure that systems cannot simply reproduce pre-existing jokes found online.
Humor Task Constraints
1. No Offensive or Derogatory Content
Generated humor must not contain insults, degradation, or negative stereotyping of any individual or group, including those defined by nationality, ethnicity, religion, race, gender, disability, or political affiliation.
2. No Sexual or Explicit Humor
Sexual jokes including indirect references, innuendo, double meanings, or suggestive phrasing are strictly prohibited.
3. No Hate Speech or Discriminatory Remarks
Humor must not include racist, xenophobic, sexist, ableist, or otherwise discriminatory language. Any content that mocks, belittles, or dehumanizes a protected group is unacceptable.
4. Dialect-Based Misunderstandings Are Allowed When Benign
Benign linguistic or dialectal misunderstandings are permitted only when:
· the humor arises from differences in vocabulary or regional expressions, and
· no group is demeaned or negatively portrayed.
For example, misunderstandings caused by differing meanings of a word across dialects are acceptable as long as the joke remains respectful and non-harmful.
مصرية ضاعت بنتها قالت لها قصيمية هي دقاقه ]يعنى نحيفه بالقصيمي[
..قالت: لا مش دجاجه دي انسانه
5. No Humor Targeting Nationalities or Cultures in a Harmful Way
While jokes involving language differences or benign cultural misunderstandings are allowed, any humor that mocks, stereotypes, or implies inferiority of a nationality or cultural group is prohibited.
6. No Personal Attacks
The humor must not target identifiable individuals with harmful, insulting, or defamatory statements.
7. Permissible Lighthearted Elements
The system may include harmless, non-sensitive comedic elements such as:
animals, fictional characters, exaggerated situations, common joke archetypes (e.g., نملة, واحد مضيع، محشش), provided they do not cross into offensive territory.
8. Humor Must Preserve a Benign Tone
All jokes should remain within the bounds of benign violation i.e., surprising or playful without causing harm, discomfort, or disrespect.
9. Original and Non-Recycled Jokes
Generated humor must be novel and not a direct copy or minor rewording of existing jokes (whether from known joke databases, social media, or popular culture). Systems should create new jokes rather than reproducing memorized or widely-circulating ones. Submissions that simply replicate well-known jokes, memes, or existing punchlines will be considered invalid.
December 15, 2025: Release of the official dataset and evaluation scripts.
February 10, 2026: Participant registration deadline and release of the final test set.
February 17, 2026: Submission portal closes and evaluation phase ends.
February 24, 2026: Announcement of the final leaderboard results.
March 10, 2026: Deadline for system description paper submissions.
March 20, 2026: Notification of acceptance for system description papers.
March 30, 2026: Camera-ready versions due.
Ameera Almasoud*, Hend Alkhalifa*, Manal Albahlal*, Nora Alangari*, Reem Alqifari*
*King Saud University
For all inquiries, please post a message in our Slack Group (preferred method) or email us (for private communication): arhaha2026@gmail.com