@inproceedings{camacho-gonzalez-etal-2026-l52,
    title = "L52+-{IIMAS}-{UNAM} at {S}em{E}val-2026 Task 1 ({MWAHAHA}): Joke Selection Through a Multi-Stage Prompt-Engineering and Heuristic Pipeline",
    author = "Camacho Gonzalez, Adolfo Tonatihu  and
      Cruz, Ximena  and
      God{\'i}nez-Aldana, Natalia  and
      Palacios-Pati{\~n}o, Lizeth  and
      Rangel, Ram{\'o}n  and
      Meza Ruiz, Ivan Vladimir",
    editor = "Kochmar, Ekaterina  and
      Ghosh, Debanjan  and
      North, Kai  and
      Komachi, Mamoru",
    booktitle = "Proceedings of the 20th {I}nternational {W}orkshop on {S}emantic {E}valuation (2026)",
    month = jul,
    year = "2026",
    address = "San Diego, California, USA",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingest-acl-workshops/2026.semeval-1.291/",
    pages = "2301--2306",
    ISBN = "979-8-89176-414-9",
    abstract = "Humor generation remains one of the most challenging tasks in natural language processing, requiring creativity, incongruity resolution, cultural sensitivity, and strict structural control. We present a fully prompt-based system for headline-conditioned joke generation in SemEval-2026 Task 1 (MWAHAHA) for both English and Spanish. Deliberately avoiding fine-tuning, our approach relies on structured prompt engineering combined with a multi-stage heuristic pipeline. For Spanish, we extract 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 performed by the team (n=180) shows substantial gains over single-pass generation, particularly in funniness and punchline clarity. Official shared-task results were modest (12th/16 Spanish, 24th/31 English), underscoring that limited originality remains a key bottleneck. In an era dominated by large language models (LLMs) such as GPT-4o and Grok, our work demonstrates the value of linguistically grounded heuristics as an efficient, interpretable, and low-cost complement to black-box generation systems."
}