There's no special formatting requirements. It will just exclude all special characters when extracting each word (separated by space) and store those words individually. If it already has the word, then it will increase its priority value by 1, which it uses to determine using that word over other similar possibilities when assembling a response.