Analysis of a probabilistic model of redundancy in unsupervised information extraction