A method for developing a voice user interface for a statistical semantic system is described. A set of semantic meanings is defined that reflect semantic classification of a user input dialog. Then, a set of speech dialog prompts is automatically developed from an annotated transcription corpus for directing user inputs to corresponding final semantic meanings. The statistical semantic system may be a call routing application where the semantic meanings are call routing destinations.