An Easy-to-use and Robust Approach for the Differentially Private De-Identification of Clinical Textual Documents.