Last month's oddest NLP research: 65 papers on jokes, memes, Minesweeper rules and origami-by-video

Asked:

“Can you find all niche fun NLP/AI papers from the last month?”

A broad live-web scan surfaced 65 distinct papers dated August 20 – September 20, 2026, whose subject itself is playful, quirky, or unusually narrow. The scan covered 1,445 pre-deduped pages across two diversified search passes; 334 distinct dated titles fell in the window, and a stricter “niche fun” relevance pass plus targeted gap searches produced this catalog. Because no formal index of “fun” papers exists, treat this as a broad discovered catalog rather than a mathematically exhaustive universe.

The paper constellation: every paper by date and theme

What the constellation can't say alone

Humor is the field's favorite playground: 25 of the 65 papers study jokes, memes, puns or sarcasm — from measuring “I like my X like my Y” jokes with embeddings opentrain.ai to multi-agent meme “transcreation” across cultures arxiv.org.

Culturally specific phenomena get whole benchmarks: Chinese “yinyang guaiqi” passive-aggression alphaxiv.org, Nigerian Pidgin emotions like “Hustle Fatigue” arxiv.org, and one entire study on a single begging-hand emoji pbijournals.ipbcirebon.ac.id.

AI's own quirks became a subject: 26 firsthand anecdotes of algorithms misbehaving creatively arxiv.org, and a study finding models dodge tedious chores and doodle with Mandelbrot sets when given free time arxiv.org.

The most corroborated paper in the catalog is a graph-of-thought method for understanding memes and satire, echoed by 109 pages across 12 hosts d6108366.hf-mirror.com.

All 65 papers

#	Paper & hook	Date	Theme	Source

Method: catalog of 65 papers dated 2026-08-20 to 2026-09-20, gathered September 2026 by a broad live-web scan (1,445 pre-deduped result pages, two diversified search passes, 334 dated in-window titles) with a subjective “niche fun” inclusion rule — the paper itself must study a playful, quirky, or unusually narrow subject. Deduplication by normalized title; all 65 rows carry a source URL across 65 distinct URLs on 21 independent hosts, and no URL supports more than one paper (1.5% max share). The fun rank is a discovery aid, not a quality ranking; five theme groups were assigned for this report from each paper's stated subject and hook. 24 papers have unresolved authors and their author cells are left blank. “All” means all qualifying papers found by this methodology, not every paper published worldwide. Full summaries were shortened in the table for space; hover or tap a dot for details.

This report was generated automatically by Keenable SELECT at a user's request, from publicly available web sources linked herein. Keenable does not review, verify, or endorse its contents and makes no representation as to accuracy, completeness, or timeliness; AI-based extraction may contain errors. Nothing in this report is investment, legal, financial, or other professional advice. All trademarks and referenced content remain the property of their respective owners; no affiliation or endorsement is implied. To report an error, rights concern, or request removal: legal@keenable.ai.