Delving into Linguistic Nuances: Exploring Humor in Tulu Versus Analyzing Humor, Sarcasm and Irony in Kannada

The standard researching trend in Natural Language Processing scales down to sentiment analysis in various Indian languages. This analysis is frequently used to get more information about public opinion, analyze customer feedback, rate product reviews and also to monitor users’ interaction in social media. Since the linguistic perception and comprehension of Humor, Sarcasm and Irony are subjective, detecting these nuances can be a difficult task in the field of Natural Language Processing, pertaining to the lack of labeled datasets, lack of related resources, variation in dialects and limited research. The proposed work includes manually curating and building the dataset for Kannada and Tulu and exploring different individual, ensemble Machine Learning and Deep Learning models to perform Sarcasm, Humor and Irony detection in Kannada and Humor detection in Tulu. This work presents two architectures ensemble Machine Learning and ensemble Deep Learning. The proposed Deep Learning architecture outperforms the Machine Learning architecture by providing an accuracy of 75.23% for Humor, 79.79% for Sarcasm, 67.34% for Irony in Kannada and 82.71% for Tulu Humor.