ICT-NLP at SemEval-2026 Task 1: Humor Generation via RAG-based Augmentation and Multi-LLM Internal-External Voting

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

Summary

ICT-NLP developed a system for SemEval-2026 Task 1: Humor Generation, a competition focused on creating genuinely humorous content under various constraints. Their approach integrates Retrieval-Augmented Generation (RAG) to preprocess news headlines and generate content summaries. The system employs a unified humor generation mode designed to adapt to two distinct types of generation constraints. A crucial final step involves an internal-external voting process, which refines and selects the optimal joke output. This methodology proved highly effective, achieving a competitive performance by ranking 1st (tied) among all participating teams in the Chinese track of Subtask A. This demonstrates a robust framework for automated humor generation.

Key takeaway

For NLP Engineers developing humor generation systems, this approach offers a proven strategy for competitive performance. You should consider integrating Retrieval-Augmented Generation for context preprocessing and a multi-LLM voting mechanism to refine output quality. This combination, particularly effective for constrained generation tasks, could significantly enhance your system's ability to produce genuinely humorous content, as demonstrated by its 1st place (tied) ranking in SemEval-2026 Task 1's Chinese track.

Key insights

A multi-stage RAG and voting system achieved top performance in humor generation for SemEval-2026 Task 1.

Method

The system uses RAG for news headline preprocessing and summarization, applies a unified humor generation mode for constraints, and employs multi-LLM internal-external voting for final joke selection.

Topics

• Humor Generation
• Retrieval-Augmented Generation
• Large Language Models
• Multi-LLM Voting
• SemEval-2026
• Natural Language Processing

Best for: Research Scientist, AI Scientist, NLP Engineer, Machine Learning Engineer

Related on AIssential

• Retrieval-Augmented Generation as a Bridge Between Classical NLP Pipelines and Modern LLM Systems — Retrieval-augmented generation (RAG) addresses the limitation of static knowledge in large language models (LLMs) by connecting them to external, up-to-date knowledge sources.
• I built a public voting benchmark where models have to make memes out of daily news — Memebench evaluates LLM humor and creativity by generating memes from daily news for public A/B voting.
• Re-defining Humor Data Objects for AI Humor Research — AI humor research improves by defining humor as a social interaction with context and explanations, not just a binary state.
• 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.
• X+Slides: Benchmarking Audience-Conditioned Slide Generation — Benchmarking LLM-generated slides requires audience-specific utility and source grounding, not just completeness or visual quality.
• TA-RAG: Tone-Aware Retrieval-Augmented Generation for Peer-Support Health Communication — TA-RAG enhances RAG with prompt-based tone control for sensitive health communication, improving empathy and readability without fine-tuning.

Open in AIssential →

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