If you compare to Norvig's first corrector you should also compare to the much faster http://norvig.com/ngrams http://norvig.com/ngrams (which follows an algori

If you compare to Norvig's first corrector you should also compare to the much faster http://norvig.com/ngrams http://norvig.com/ngrams (which follows an algorithm I sent him). Added: I'd guess the newer Norvig corrector is still slower, though it is more accurate (taking into account probabilities of different typos). There's a curiosity not mentioned in his article, that the new algorithm computes a very slightly different set of candidates. I only noticed the difference when I got around to checking these sets against each other for all the words in a sample document -- I'd meant them to be the same. Might be fun to check the OP's code this way too.