Modelling and Optimizing on Syntactic N-Grams for Statistical Machine Translation