Large Language Models as Neurolinguistic Subjects: Identifying Internal Representations for Form and Meaning