Combining Semi-Unsupervised Acquisition of Corpora and Supervised Learning of Textual Entailment Rules