A method of using a feedback loop for determining parts of speech on multi-parts-of-speech words as part of the machine translation process is described.