GraPHIA: a computational model for identifying phonological jokes

Currently in humor research, there exists a dearth of computational models for humor perception. The existing theories are not quantifiable and efforts need to be made to quantify the models and incorporate neuropsychological findings in humor research. We propose a new computational model (GraPHIA) for perceiving phonological jokes or puns. GraPHIA consists of a semantic network and a phonological network where words are represented by nodes in both the networks. Novel features based on graph theoretical concepts are proposed and computed for the identification of homophonic jokes. The data set for evaluating the model consisted of homophonic puns, normal sentences, and ambiguous nonsense sentences. The classification results show that the feature values result in successful identification of phonological jokes and ambiguous nonsense sentences suggesting that the proposed model is a plausible model for humor perception. Further work is needed to extend the model for identification of other types of phonological jokes.