On the Relationship between Sentence Analogy Identification and Sentence Structure Encoding in Large Language Models