METU-Treebank is a Turkish dataset that is built from manually collected newspapers, journal issues, and books. The corpus involves 5659 sentences. You will take %70 of the data as training (for Task 1) and rest of it as test data (for Task 2 and Task 3). In other words, the rst 3960 lines will be used for training and the last 1699 lines will be used for testing.