Recursive Noun Phrase

Recursive noun phrases (NPs), such as "my favorite new movie," present a challenge to natural language processing because their meaning depends on the order and interaction of modifiers. Current research focuses on evaluating the ability of large language models, particularly transformer-based architectures like GPT-2 and GPT-3, to correctly interpret the nuanced semantics of recursive NPs, often using specifically designed datasets and tasks. These studies reveal significant limitations in current models' understanding of such phrases, highlighting a crucial area for improvement in natural language understanding and impacting applications like sentiment analysis and harm detection.