A crucial aspect of a knowledge base population system that extracts new
facts from text corpora, is the generation of training data for its relation
extractors. In this paper, we present a method that maximizes the effectiveness
of newly trained relation extractors at a minimal annota