Thestate-of-the-artmethodsusedforrelationclassificationareprimarilybasedonstatisticalmachinelearning,andtheirperformancestronglydependsonthequalityoftheextractedfeatures.Theextractedfeaturesareoftenderivedfromtheoutputofpre-existingnaturallanguageprocessing(NLP)systems,whichleadstothepropagationoftheerrorsintheexistingtoolsandhinderstheperformanceofthesesystems.