State-of-the-art systems on CoNLL-2014 data set. The systems are divided into three categories: classifiers, MT, and combined. For each system, we also show the size of the annotated learner data used to train the system. Depending on the type of training data available, a specific approach should be preferred. In particular, classification systems can be trained from significantly less annotated data and can focus on specific errors, whereas MT-based systems do not have this capability. Best result in each category is in bold.