idk, but I think they employ multiple tokenizing strategies, including case insensitive, case sensitive,by words, by n-grams of letters...