How Linguistics Learned to Stop Worrying and Love the Language Models

The article discusses the evolving relationship between linguistics and language models (LMs), which can generate fluent and grammatical text. While some argue that LMs do not truly learn language and that their success diminishes the need for linguistic theory, the authors contend that both views are incorrect. They assert that LMs can significantly contribute to understanding linguistic structure, language processing, and learning. Furthermore, LMs challenge traditional arguments in linguistics, prompting a reevaluation of foundational concepts. The authors present an optimistic perspective on how LMs can serve as model systems and proofs of concept for gradient and usage-based approaches to language.