Computer Humor and Human Humor: Construction of Japanese “Nazokake” Riddle Generation Systems

Abstract

Humor is important in smooth human communications, however, computer-generated humor is still distinguishable from humor that arises naturally in human communication. The purpose of this study is to construct a computer system that can generate humor in a human-like manner. The method involves using “nazokake” riddles, which comprise a type of Japanese word game. The game creates humorous links between two incongruous premises by linking them semantically to homophones: “Why is A like B? Because X/X^′,” where A and B are independent premises and X and X^′ are homophones linked to A and B, respectively. In a previous study, a system was constructed to generate such riddles based on a simple word similarity between two nouns that are homophones. This study builds on the previous study by generating more complex riddles based on the dependency relationships between homophonic verb-noun combinations. Subsequently, the two systems are compared with each other by evaluating them against riddles created by humans. The results show that the system based on dependency relationships generated more humorous, unexpected, and natural riddles than that based on word similarities. However, these riddles were not equal to those created by humans.