yasaminal at Semeval2026: Constraint-Aware Humor Generation with Knowledge Graph Guidance

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

Summary

The yasaminal system, presented at Semeval2026, introduces a knowledge-guided approach for humor generation, capable of creating humorous text from either word pairs or news headlines. This system integrates structured semantic reasoning, derived from a knowledge graph (KG), with controlled generation capabilities of large language models (LLMs). It operates by constructing an intermediate KG hint, comprising related concepts in the target language, which is then appended to the prompt to guide the LLM. The system generates a single candidate joke per input using controlled top-p decoding. Experimental results indicate that KG reasoning significantly improves relevance and constraint satisfaction, while LLM-based generation ensures fluency and creativity. This method offers a structured and interpretable framework for humor generation across multiple languages.

Key takeaway

For NLP engineers developing advanced text generation systems, this approach offers a robust method for incorporating external knowledge and control. You should consider integrating knowledge graph hints into your LLM prompting strategies to enhance output relevance and ensure specific constraints are met, particularly for creative or domain-specific content generation tasks. This can lead to more interpretable and higher-quality generated text.

Key insights

The yasaminal system combines knowledge graphs and LLMs for structured, constraint-aware, and multi-language humor generation.

Principles

• KG reasoning improves relevance and constraint satisfaction.
• LLM generation ensures fluency and creativity.

Method

Construct an intermediate knowledge graph hint with related concepts, append it to the prompt, then guide the LLM using controlled top-p decoding to generate a single joke.

In practice

• Generate humor from word pairs.
• Generate humor from news headlines.

Topics

• Humor Generation
• Knowledge Graphs
• Large Language Models
• Semantic Reasoning
• Text Generation
• SemEval2026

Best for: Research Scientist, AI Scientist, NLP Engineer

Related on AIssential

• Emotion-Aware Image Generation from Korean Diary Text via LLM-based Prompt Translation and LoRA Fine-Tuning — A pipeline generates emotion-aware, children's drawing style images from Korean diary text using LLM sentiment analysis and LoRA fine-tuned Stable Diffusion.
• BLINKG: A Benchmark for LLM-Integrated Knowledge Graph Generation — BLINKG benchmarks LLM capabilities in mapping diverse data to ontologies for Knowledge Graph generation.
• [R] Dynamic Large Concept Models: Latent Reasoning in an Adaptive Semantic Space — Dynamic Large Concept Models explore latent generative modeling for text, a technique common in image/video diffusion.
• 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.
• Constrained Semantic Decompression in LLMs through Persian Proverb-Conditioned Story Generation — LLMs struggle with "constrained semantic decompression," failing to translate abstract cultural knowledge like proverbs into faithful narratives.
• Improving Parametric Knowledge Access in Reasoning Language Models — Reasoning language models can be trained to better access their internal world knowledge.

Open in AIssential →

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