This paper describes Fudan's submission to CoNLL 2018's shared task Universal Dependency Parsing. We jointly train models when two languages are similar according to linguistic typology and then do an ensemble of the models using a simple re-parse algorithm. Our system outperforms the baseline method by 4.4% and 2.1% on the development and test set of CoNLL 2018 UD Shared Task, separately. 1. Our code is available on https://github.com/ taineleau/FudanParser. * Authors contributed equally. 1 Unfortunately, we did not finish the run before the deadline. As a result, the official accuracy gain for test set is only 0.54% and we ranks 17th out of 27 teams.