(to appear) Hyewon Jang & Diego Frassinelli, Generalizable Sarcasm Detection is Just Around the Corner, of Course!, Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL 2024).
Summary
In this paper, we test the generalizability of sarcasm detection models by comparing three different language models finetuned on several datasets of sarcasm, including a new dataset we release (CSC), collected from multiple psycholinguistic experiments. We show that all language models finetuned on one dataset perform a lot worse on the other datasets, but that (CSC) can handle generalizable sarcasm detection relatively well. We discuss the reasons for the results in terms of the varied domains, styles, and label sources of sarcasm.