Second language speech recognition using multiple-pass decoding with lexicon represented by multiple reduced phoneme sets.