让大语言模型自主复述，打破与人类对话的壁垒 Rephrase and Respond: Let Large Language Models Ask Better Questions for Thems