In this ongoing research, Gilad Gutman explores the tragedies of Shakespeare and his contemporaries by applying NLP methods (Natural Language Processing) to analyze various figurative tropes, such as metaphors, similes, metonymies, personifications and others. In comparison to the computational analysis of literal language with NLP, there are certain difficulties that arise due to the very structure of figurative language. For example, metaphors bring together words that are unlikely to appear together in literal language, which complicates the application of probabilistic models.
Gutman's solution involves a deep learning algorithm that combines BERT, a large language model, in order to train a multi-label topic classifier. In each figurative instance, the topic classifier assigns a particular topic to the different words or disregards a word when used literally.
