Early templated-based systems use handcrafted ontologies tailored to specific system. For example, Light Bulb Joke Generator (Raskin and Attardo, 1994) produces jokes of the pattern “How many <group name> does it take to screw in a light bulb? <NumberX>. One to <activity 1> and <numberY> to <activity2>”. And can produce jokes such as “‘How many Californians does it take to change a light bulb? Twelve. One to screw it in and eleven to share in the experience.”. Some recent systems uses freely-available ontologies and databases, such as WordNet and ConceptNet.

Ontologies For Variable Selection

Joker generators using quantitative measures would assume some relations between variables quantitatively, which is not expressed in explicit variable relations. For example, Labutov and Lipson (2012a) choose variables and their relations that form jokes from ConceptNet in a probabilistic way. Based on the circuits in ConceptNet, they assume that two paths that have the same start and end nodes are joke candidates.

Quantitative Measures For Variable Selection

Some joke generation system combined ontologies and quantitative measures. For example, Sjöbergh and Araki (2008) implemented a system generating performing Japanese stand-up comedy. The system starts with a proverb joke, continues with a riddle joke, and ends with a joke retrieved from a human generated joke database.

Hybrid Systems

In linguistic, a template is a piece of text with slots that can be filled with different variables. To compose a joke, a template need to be associated with a schema, which stipulate the relationships between variables. Schema is typically chosen to provoke incongruity and its resolution. There are severals ways of selecting variables:

1.	Ontologies for variable selection

2.	Quantitative measures for variable selection

3.	Hybrid systems


University of Notre Dame

Humor generation can be viewed as a special case of automatic text generation, and current humor generation systems can be categorized to two major approaches:

1.	Neural network based systems

2.	Template-based systems


Types Of Humor Generation Systems

Amin, M., & Burghardt, M. (2020, December). A survey on approaches to computational humor generation. In Proceedings of the The 4th Joint SIGHUM Workshop on Computational Linguistics for Cultural Heritage, Social Sciences, Humanities and Literature (pp. 29-41).

A Survey on Approaches to Computational Humor Generation

Neural network based system used deep neural network model to generate short humorous short text. This category of model started with Yang and Sheng (2017). They used LSTM RNN as there neural network model, and trained the model on a dataset of 7,699 jokes written by Conan O’Brien and news data. 

Learn Before

Related

Learn After