Learning Greedy Policies for the Easy-First Framework