In the last few years, several enhancements for the hierarchical phrase-based translation model have been proposed. They aim to include additional syntactic information in the translation process in order to achieve better fluency in the generated output. In this work we review and compare three such methods: parsematch, soft syntactic labels and string-to-dependency. Our goal is to find out if these models complement each other of if they rather address the same deficiencies in the translation process. Furthermore, we present a novel method for extending the translation model in the same direction without the need for parse trees, since they may not be available for some languages. Our approach is based only on automatic clustering of phrases, without the need for additional information. Our findings show that we are able to achieve similar results as when applying syntax models.