For the First Time, AI Analyzes Language as Well as a Human Expert

OpenAI's o1 language model has achieved unprecedented performance on sophisticated linguistic tests, analyzing language with the skill of a graduate linguistics student. UC Berkeley researchers tested multiple AI models on complex tasks including parsing recursive sentences, resolving ambiguous meanings, and learning rules of invented languages—challenges designed to avoid memorized responses. While most models failed, o1 successfully diagrammed intricate sentence structures and identified phonological patterns in made-up languages, demonstrating metalinguistic reasoning abilities that challenge assumptions about uniquely human cognitive capabilities.