I don‘t know how to get the alignment reference and training data reported in the paper, could you give me some clue ?