tation schemes in different projects are usually different, since the underlying linguistic theories vary and have different ways to explain the same language phenomena. Though statistical NLP systems usually are not bound to specific annotation standards, almost all of them assume homogeneous annotation in the training corpus.