Learning Causality from Textual Data