Can ChatGPT Understand Malapropism Correctly? Challenges to Davidson's Passing Theory in Generative AI

This study examines ChatGPT's performance in understanding Japanese malapropisms, aiming to explore its capacity for linguistic inference compared to humans. Despite its remarkable fluency in conversation, ChatGPT shows significant limitations in comprehending malapropisms, particularly in handling phonetic, lexical, and contextual errors. Using a specialized dataset, the research highlights these gaps, suggesting that while ChatGPT excels in fluency, its understanding of nuanced language phenomena remains distinct from human comprehension. The findings contribute to the discourse on the potential and limitations of generative AI, advocating for a reevaluation of linguistic-philosophical theories in light of AI advancements.