In-Context Analogical Reasoning with Pre-Trained Language Models