The idea would be to read in the entire dictionary and mark the Bloom Filter for each word. Then whenever you get a word you want to test, see whether all of the places it would mark were previously marked. If it misses any, you know it's not a word. Unfortunately, if it doesn't miss, you're only sure it's probably a word (with the worst case probability controlled by the number of hash functions you use).