Learning Language from a Large (Unannotated) Corpus