Yes, there even was such a tool posted on HN some time ago. I can't remember its name, but basically it got its corpus from twitter data. Maybe someone else can post a link?