wangkongqiang at SemEval-2026 Task 1: MWAHAHA- Competition on Humor Generation

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

Summary

The system developed by Wang Kongqiang, Zhang Peng, and Tan Qingli for SemEval-2026 Task 1: MWAHAHA-Competition on Humor Generation focused on text-based and image-based humor generation. For Subtask A, Text-based Humor Generation, the system generated jokes from text constraints in English and Chinese. For Subtask B, Image-Based Caption Generation, it tackled multimodal humor from visual inputs in English, specifically B1 (Image-only) and B2 (Image and Prompt). The approach utilized BLIP and Qwen series LLMs. Evaluation used the Rating (95% CI) score. The system achieved notable rankings: 1st in Subtask A Chinese with a Rating of 1054 (95% CI [1024, 1104]), 16th in Subtask A English (Rating 950, 95% CI [922, 982]), 5th in Subtask B1 (Rating 976, 95% CI [941, 1007]), and 3rd in Subtask B2 (Rating 987, 95% CI [948, 1016]).

Key takeaway

For NLP Engineers or AI Scientists developing creative text generation systems, consider integrating multimodal LLMs like BLIP and Qwen series models. Your efforts in humor generation, especially across languages or with visual inputs, could benefit from these architectures, as demonstrated by their top rankings in SemEval-2026. This suggests a strong direction for improving the nuance and context-awareness of generated humor.

Key insights

A humor generation system using BLIP and Qwen LLMs achieved top rankings in SemEval-2026's multimodal and multilingual humor tasks.

Principles

• Multimodal LLMs enhance humor generation.
• Multilingual models expand humor task scope.
• Combining visual and text inputs improves results.

Method

The system employed BLIP and Qwen series LLMs for text-based humor generation in English and Chinese, and for image-based caption generation (B1, B2) in English, leveraging multimodal and large language models.

In practice

• Apply BLIP for image-to-text humor.
• Use Qwen LLMs for multilingual joke generation.
• Integrate visual and text prompts for richer humor.

Topics

• Humor Generation
• Multimodal LLMs
• Qwen LLM
• BLIP Model
• SemEval-2026
• Natural Language Processing
• Image Captioning

Best for: Research Scientist, AI Scientist, NLP Engineer, Computer Vision Engineer

Related on AIssential

• Linguistic Symptoms: Augmented Generation by Retrieval and Reasoning in LLMs under Portuguese-English Variation in Medical Contexts — Linguistic variation significantly impacts smaller LLMs in medical contexts, with larger models showing greater cross-language robustness.
• Benchmarking Multimodal LLMs on Code Generation for Complex Interactive Webpages — WebIGBench is the first benchmark to evaluate MLLMs on interactive webpage code generation, addressing gaps in existing static-focused evaluations.
• 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.
• LLMbench: A Comparative Close Reading Workbench for Large Language Models — LLMbench enables hermeneutic analysis of LLM outputs by visualizing token-level probabilities and textual differences.
• Combining Retrieval-Augmented Text Generation with LLMs for Reading Content Recommendations — RAG significantly enhances LLM-generated personalized reading content, improving relevance and groundedness.
• Most Influential ArXiv (Computation and Language) Papers (2026-04 Version) — Citation-based ranking reveals key trends in NLP research, emphasizing LLM reasoning, efficiency, and multimodal capabilities.

Open in AIssential →

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