Paraphrasing is often performed with less concern for controlled style\nconversion. Especially for questions and commands, style-variant paraphrasing\ncan be crucial in tone and manner, which also matters with industrial\napplications such as dialog systems. In this paper, we attack this issue with a\ncorpus construction scheme that simultaneously considers the core content and\nstyle of directives, namely intent and formality, for the Korean language.\nUtilizing manually generated natural language queries on six daily topics, we\nexpand the corpus to formal and informal sentences by human rewriting and\ntransferring. We verify the validity and industrial applicability of our\napproach by checking the adequate classification and inference performance that\nfit with conventional fine-tuning approaches, at the same time proposing a\nsupervised formality transfer task.\n