SPOKEN LANGUAGE PROCESSING USING WEIGHTED FINITE STATE TRANSDUCERS