Robust Numeric Recognition in Spoken Language Dialogue