L52+-IIMAS-UNAM at SemEval-2026 Task 1 (MWAHAHA): Joke Selection Through a Multi-Stage Prompt-Engineering and Heuristic Pipeline

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning · Depth: Expert, quick

Summary

L52+-IIMAS-UNAM developed a fully prompt-based system for headline-conditioned joke generation in English and Spanish for SemEval-2026 Task 1 (MWAHAHA). This approach deliberately avoids fine-tuning, relying instead on structured prompt engineering combined with a multi-stage heuristic pipeline. For Spanish, the system extracts a "stylistic-humor DNA" from a public joke corpus to guide generation. The pipeline integrates multi-candidate generation, diversity enhancement, iterative refinement, LLM-based rewriting, and constraint-aware selection. Human evaluation (n=180) demonstrated substantial gains in funniness and punchline clarity compared to single-pass generation. Despite modest official shared-task results (12th/16 Spanish, 24th/31 English) due to limited originality, the work highlights the value of linguistically grounded heuristics as an efficient, interpretable, and low-cost complement to large language models like GPT-4o and Grok.

Key takeaway

For NLP Engineers developing creative text generation systems, you should integrate structured prompt engineering and linguistically grounded heuristics alongside large language models. This approach, demonstrated in joke generation, offers greater interpretability, control over output, and cost efficiency compared to purely black-box LLM fine-tuning. Consider implementing multi-stage pipelines with iterative refinement and "stylistic-humor DNA" extraction to enhance quality and address originality bottlenecks in your applications.

Key insights

Structured prompt engineering and heuristics enhance LLM-based humor generation, offering interpretability and cost efficiency.

Principles

• Heuristics complement LLMs for interpretability.
• Multi-stage pipelines improve generation quality.
• Prompt engineering can replace fine-tuning.

Method

The method involves multi-candidate generation, diversity enhancement, iterative refinement, LLM-based rewriting, and constraint-aware selection within a prompt-engineered pipeline.

In practice

• Extract stylistic DNA from corpora.
• Integrate LLM rewriting into pipelines.
• Employ human evaluation for nuanced feedback.

Topics

• Humor Generation
• Prompt Engineering
• Heuristic Pipelines
• Large Language Models
• Natural Language Processing
• SemEval-2026

Best for: Research Scientist, AI Scientist, NLP Engineer, Prompt Engineer

Related on AIssential

• Re-defining Humor Data Objects for AI Humor Research — AI humor research improves by defining humor as a social interaction with context and explanations, not just a binary state.
• Automatic Generation of Highlights for Academic Paper Via Prompt-based Learning — Prompt-based learning with LLMs can generate academic paper highlights effectively without extensive labeled data.
• Surpassing Scale by Efficiency: A Compact 135M Parameter Foundational LLM Natively Adapted for the Bangla Language — A compact 135M parameter LLM for Bangla achieves 1B-tier performance by efficiently merging tokens, overcoming low-resource script challenges.
• As Easy as Rocket Science: Assessing the Ability of Large Language Models to Interpret Negation in Figurative Language — LLMs struggle with combined negation and figurative language, with prompt style significantly impacting performance.
• Token-Level Pun Location Using Multi-Layer BERT with Mixture of Experts — A two-stage BERT-MoE model effectively locates puns in Portuguese by specializing expert components.
• From Raw Text to Smart Predictions: A Beginner-Friendly Guide to the Complete NLP Pipeline — An NLP pipeline transforms raw text into machine-understandable numerical representations through systematic cleaning and vectorization.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.