Baselines available and clarification on resubmissions

22 views

Skip to first unread message

Luis Chiruzzo

unread,

Jan 27, 2026, 4:57:38 PMJan 27

to semeval-2026-task-1-humor-gen

Dear participants,

We just published the baseline used during the Evaluation Trial phase on the competition web page (https://pln-fing-udelar.github.io/semeval-2026-humor-gen/). Apologies for uploading it late.

For task A, the baseline consists of prompting the Gemini 2.5 Flash model with simple, task-specific prompts for both the two-word constraint and the news title. The prompts are written in English. For Spanish and Chinese, a simple indication of the expected output language is appended to the prompt. For tasks B1 and B2, the first frame of the GIF is extracted and provided to the same model, together with task-specific prompts.

As you can see in the leaderboard for the dev phase (https://thefunnier.com/leaderboard), the baseline system achieved really competitive results. We hope these results and the publication of the new baseline will help you prepare stronger systems for the ongoing Evaluation phase.

There are two things you have to keep in mind about this phase, ending January 31st:

In your submission, you can also include the source code of your systems attached to your outputs, as well as a text explanation of how it works. We encourage participants to do so.
Remember, only one submission is allowed per participant, and your submission must pass the basic check (score 1 on the leaderboard) to be considered for evaluation. However, given the issues that have happened to some participants, where they couldn’t delete the previous submission in order to make a new one, we have increased the maximum number of submissions to 5, but only one submission will be considered for evaluation per participant: the last submission that achieved score 1 on the basic check.