A bigram model learns eight station announcements, from Kumasi to Mombasa, by counting which word follows which. It scores a new announcement and writes its own. Then use raw counts, add one to every count, or turn the temperature up. This lesson needs JavaScript.