JCT at SemEval-2026 Task 1: Let the Best Joke Win - A Generate - and-Rank Approach to Constrained Humor

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning · Depth: Advanced, medium

Summary

The JCT system, developed for SemEval-2026 Task 1, Subtask A, generates short jokes under lexical or headline-based constraints. It utilizes a large language model to produce multiple candidate jokes, employing diverse humor styles and prompting strategies, including zero-shot, few-shot, and structured prompting. Constraint satisfaction is explicitly enforced, either by requiring exact lexical inclusion or by approximating semantic relevance to a headline using sentence-embedding similarity. All valid candidates are then ranked using a weighted humor score that combines semantic incongruity, emotion-based humor potential, irony likelihood, linguistic fluency, and novelty with respect to a large external jokes corpus. The single highest-scoring joke is selected for each constraint, following a best-candidate selection paradigm without task-specific fine-tuning.

Key takeaway

For NLP Engineers developing creative text generation systems, this work suggests that a generate-and-rank architecture can achieve high-quality, constrained outputs without extensive task-specific fine-tuning. You should consider integrating automated humor proxies like semantic incongruity and novelty into your ranking functions to refine generated content. This approach offers a robust method for producing nuanced, context-aware text.

Key insights

A generate-and-rank approach with automated humor proxies effectively creates constrained jokes without task-specific fine-tuning.

Principles

• Automated humor proxies improve joke quality.
• Constraint satisfaction can be enforced lexically or semantically.

Method

Generate diverse joke candidates using an LLM, enforce constraints, then rank valid jokes with a multi-factor humor score combining incongruity, emotion, irony, fluency, and novelty to select the best.

In practice

• Employ diverse LLM prompting strategies.
• Combine multiple humor metrics for ranking.

Topics

• Humor Generation
• Large Language Models
• Generate-and-Rank
• Semantic Evaluation
• Prompt Engineering
• Text Constraints

Best for: Research Scientist, AI Scientist, NLP Engineer, AI Student

Related on AIssential

• 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.
• Off for adventures — AI image generation still struggles with realistic depiction, often producing humorous artifacts.
• Humor Style Drives Laughter, Topic Shapes Acceptability: Evaluating Bilingual Personal and Political Robot-Delivered AI Jokes — Aggressive and Affiliative humor styles enhance robot-delivered joke funniness, while person-related topics improve appropriateness.
• 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.
• How Far Can Machine Translation Quality Take You? Extrinsic Discourse Evaluation in Goal-Oriented Setups — Extrinsic discourse evaluation reveals intrinsic MT quality doesn't guarantee downstream success, especially in goal-oriented interactions.

Open in AIssential →

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