Computational Humor with Multimodal LLMs: Methods, Datasets, Evaluation, and Challenges

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Tuo Liang, Zhe Hu, Disheng Liu, Jing Li, Yu Yin
• Venue: arXiv: cs.CL
• Published: 2026-07-21
• Collected: 2026-07-22
• URL: http://arxiv.org/abs/2607.19011v1
• PDF: https://arxiv.org/pdf/2607.19011v1
• DOI: UNKNOWN

Plain-Language Brief

The paper provides a survey on computational humor, focusing on visual humor in memes, cartoons, and comics. The study outlines the challenges faced by AI systems in understanding and generating such humor due to non-literal mechanisms, cultural knowledge, and communicative intent. It discusses the evolution of models from task-specific fusion models to large-model approaches based on multimodal alignment, evidence-grounded reasoning, and controlled generation. The paper also highlights the main obstacles to progress in the field.

What the Document Says

• The paper discusses the difficulty for AI systems in understanding and generating multimodal humor (in memes, cartoons, comics) due to non-literal mechanisms, shared cultural knowledge, and communicative intent.
• The study focuses on visual humor understanding in single-image and multi-panel artifacts and treats humor generation as an emerging downstream frontier.
• The paper positions the literature using a capability-centric hierarchy: recognition, interpretation and reasoning, and generation.
• It synthesizes benchmark design, evaluation protocols, and modeling paradigms, tracing the field's shift from task-specific fusion models to large-model approaches based on multimodal alignment, evidence-grounded reasoning, and controlled generation.
• The paper concludes by highlighting the main barriers to progress: shortcut-prone evaluation, limited cultural and narrative coverage, weak evidence grounding, and unresolved safety and ownership concerns.

Key Concepts Mentioned

• AI systems
• multimodal humor
• memes
• cartoons
• comics
• shared cultural knowledge
• communicative intent
• recognition
• interpretation and reasoning
• generation
• benchmark design
• evaluation protocols
• modeling paradigms

Claims Made by the Document

• AI systems face challenges in understanding and generating multimodal humor.
• The field has shifted from task-specific fusion models to large-model approaches based on multimodal alignment, evidence-grounded reasoning, and controlled generation.

Evidence Used

• The text discusses the difficulties faced by AI systems in understanding and generating multimodal humor.
• The paper mentions the evolution of models from task-specific fusion models to large-model approaches.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

Multimodal humor in memes, cartoons, and comics remains difficult for AI systems because intended meaning depends on non-literal mechanisms, shared cultural knowledge, and communicative intent rather than literal scene description. This survey focuses on visual humor understanding in single-image and multi-panel artifacts, while treating humor generation as an emerging downstream frontier. We position the literature against prior humor, sarcasm, and general MLLM surveys and organize it using a capability-centric hierarchy spanning recognition, interpretation and reasoning, and generation. Under this lens, we synthesize benchmark design, evaluation protocols, and modeling paradigms, tracing the field's shift from task-specific fusion models to large-model approaches based on multimodal alignment, evidence-grounded reasoning, and controlled generation. We conclude by highlighting the main barriers to progress: shortcut-prone evaluation, limited cultural and narrative coverage, weak evidence grounding, and unresolved safety and ownership concerns.