function predict_next(context_window): # e.g. the previous n tokens

   return model.predict(context_window)

 context = ["the", "cat", "sat", "on", "the"]
 predict_next(context) # "mat" -- meaning depends on surrounding words, not just
 one
