Distributed word representations capture relational similarities by means of vector arithmetics, giving high accuracies on analogy detection. We empirically investigate the use of syntactic dependencies on improving Chinese analogy detection based on