Test-set performance of our best results (according to UAS or LAS, whichever has the larger difference), compared with state-of-the-art greedy transition-based parsers (“Best Greedy Result”) and best results reported (“Best Published Result”). Note that B+′13 and B+′14 are a combination of parsers and use unlabeled data; our models do not. We only use unlabeled data for the pretrained word embeddings of DWPP. W refers to Words; C refers to Chars; WP refers to Words + Pos; WPM refers to Words + Pos + Morph; CP refers to Chars + Pos; CPM refers to Chars + Pos + Morph; B′13 is (Ballesteros 2013); N+′06a is (Nivre et al. 2006); DWPP is our model Word + Pos with pretrained embeddings and dynamic oracles as in Section 6.5; WPM is our model Word + Pos + Morph; B+′13 is (Björkelund et al. 2013); B+′14 is (Björkelund et al. 2014); W+′15 is (Weiss et al. 2015).