Using Spoken Word Posterior Features in Neural Machine Translation