Statistics for Controlled Generation of Stylized Text Using Semantic and Phonetic Representations