Humor Generation Project

We proposed building a system that can produce humorous proverbs. This proposed system uses the punch-line framework called Sukashi. By changing the end of a proverb, the proposed system produced an amusing sentence that diverged from an individual’s expectations. The proposed system has three distinct features. The novel Sukashi framework was employed. A varied and large number of Sukashi candidates were generated because the Web Japanese N-gram corpus constructed from 255 billion words was utilized. Finally, the factors of laughing — imageability (That quality in a physical object that gives it a high probability of becoming a strong mental image in the minds of viewers) and the concreteness of the targeted words — were added to sound length, sound difference, and concepts, which are frequently used in pun generation. Because of these factors, more humorous Sukashis could be selected from candidates using fuzzy rules. We confirmed that the proposed system was able to generate Sukashis comparable to human-made Sukashis in terms of amusement and unpredictability.
The intersection of proposal and the project is that we developed numerical membership functions for NLP to make the system to understand human-like word senses. As such, the system was able to successfully generate humorous sentences that are comparable to those produced by humans.

For more details, please click following link below.

Hiroaki Yamane and Masafumi Hagiwara, “A funny proverb generation system based on Sukashi,” ICANN’10 Proceedings of the 20th international conference on Artificial neural networks: Part III, Lecture Notes in Computer Science Volume 6354, pp.368-377, 2010.