The Role of Humor Theories for Computational Humor in the World of LLMs
Studies in American Humor · 08.10.2026 · 12 (2) · 202-226
DOI: 10.5325/studamerhumor.12.2.0202 ↗Kurz gefasst
Fachartikel in der Zeitschrift Studies in American Humor, erschienen am 08.10.2026. Verfasst von Julia Rayz.
Abstract (Originalsprache)
Recent advances in LLMs have significantly improved their ability to generate and explain humorous text, raising questions about whether these systems truly capture the cognitive mechanisms underlying humor or merely reproduce linguistic patterns. This paper examines LLM performance through the lens of established humor theories, particularly incongruity-resolution models and the general theory of verbal humor, which emphasize script opposition, logical mechanisms, and contextual interpretation. Using systematically modified variants of canonical jokes, this article evaluates the responses of GPT-5.2, focusing on its ability to identify scripts, distinguish foreground and background incongruities, and provide theoretically consistent explanations. The findings show that LLMs exhibit strong formal linguistic competence but limited functional competence, particularly in situational reasoning. Performance improves substantially when prompts use precise, theory-grounded definitions rather than loosely invoking humor terminology. This demonstrates that LLMs can be used to probe the hypotheses of humor theory: imprecise definitions lead to inconsistent interpretations, while precise formulations produce more theoretically aligned results, revealing both the strengths and limitations of current LLMs.
Zitieren
APA
Rayz, J. (2026). The Role of Humor Theories for Computational Humor in the World of LLMs. Studies in American Humor, 12(2), 202-226. https://doi.org/10.5325/studamerhumor.12.2.0202
Vancouver
Rayz J. The Role of Humor Theories for Computational Humor in the World of LLMs. Studies in American Humor. 2026;12(2):202-226. doi:10.5325/studamerhumor.12.2.0202
BibTeX
@article{Rayz2026,
author = {Rayz, Julia},
title = {The Role of Humor Theories for Computational Humor in the World of LLMs},
journal = {Studies in American Humor},
year = {2026},
volume = {12},
number = {2},
pages = {202--226},
doi = {10.5325/studamerhumor.12.2.0202}
}
EUVIMED