We present how to use English translation for unsupervised word segmentation of low resource languages. The inference uses a dynamic programming algorithm for efficient blocked Gibbs sampling. We apply the model to Inupiaq morphology analysis and get better results than monolingual model as well as Morfessor output.