Extracting tree transducer rules for syntactic MT systems can be hindered by word alignment errors that violate syntactic correspondences. We propose a novel model for unsupervised word alignment which explicitly takes into account target language constituent structure, while retaining the robustness and efficiency of the HMM alignment model. Our model’s predictions improve the yield of a tree transducer extraction system, without sacrificing alignment quality. We also discuss the impact of various posteriorbased methods of reconciling bidirectional alignments.