Context-Adaptive Humor Rewriting: A First-Order Logic Framework Using Large Language Models

Abstract

Humor is an important conversational skill that conveys approachability. When addressing different audiences, we often need to adjust and adapt jokes to lower the comprehension barrier while also preventing the original punchline from becoming offensive in a new context. To this end, we propose a humor-rewriting-agent framework. The system converts jokes from an existing humor corpus into a First-Order Logic representation, extracting predicates and constants. The agent then performs context-aware semantic mapping, replacing relevant elements with more suitable equivalents in the target scenario. Finally, the mapped logical structure is realized as fluent, natural text. Through evaluation by participants, this framework provides a controllable, traceable, and interpretable implementation pathway for contextually grounded humor adaptation.