Recent research demonstrates that advanced AI models are capable of analyzing complex sentences, including linguistic features like ambiguity and recursion.

A UC Berkeley study led by Professor Gašper Beguš reveals that AI chatbots such as ChatGPT now possess metalinguistic abilities, allowing them to reflect on and manipulate language structures similarly to trained linguists.

The study, published in the IEEE Transactions on Artificial Intelligence, suggests that AI models like OpenAI's o1 are bridging the gap in metalinguistic skills, which were once thought to be uniquely human.

OpenAI's o1 model outperformed other versions in detecting linguistic features like ambiguity and recursion, successfully analyzing complex sentences and producing accurate syntactic diagrams.

Specifically, the o1 model accurately identified the ambiguity in sentences such as 'Eliza wanted her cast out,' and extended recursive structures, demonstrating significant improvements in linguistic understanding.

The research involved testing 120 complex sentences across multiple AI models, including versions of ChatGPT and Meta's Llama, to evaluate their linguistic capabilities.

These findings contribute to the ongoing debate about whether AI truly understands language or merely mimics it, providing a scientific benchmark for evaluating AI linguistic progress.

The study challenges the notion of human uniqueness in language processing and emphasizes the importance of distinguishing between technological hype and genuine AI understanding.