Which format you get is chosen per word: recognition while a word is still shaky, typed recall once it is solid.