Simple algorithms start to perform well at the task once they have been provided with sufficiently high training data. The researchers further proposed that the research community should direct efforts towards increasing the size of the annotated training collections.