From Sentence Embeddings to Large Language Models to Detect and Understand Wordplay

A pun is a form of wordplay in which a word or phrase evokes the meaning of another word or phrase with a similar or identical pronunciation. In this study, we present our work for JOKER 2023, particularly the pun detection, location, and interpretation tasks. The methods used demonstrate the evolution of the field from sentence embeddings with various classifiers, sequence, and token classification using BERT-based models, to inference with prompt engineering using LLMs. Experimental results demonstrate varying effectiveness across methodologies, highlighting the strengths and limitations of each approach. Additionally, challenges such as handling nuances, diverse languages, interpreting contextually diverse word meanings, and integrating external sense dictionaries are discussed. This study provides insights into the evolution of natural language processing techniques for detecting and understanding wordplay, paving the way for future advancements in computational humor analysis.