Matt Kettler wrote:
Of course you could train your spell checker to your companies local
mail words.. however, at that point you've implemented a low-quality
version of a bayes checker.
and he can just use a bayesian classifier to implement his "feature".
training is easy:
- ham = all words from (some|many|all) dictionaries
- spam = random words that aren't in a dictionary
<story>
This reminds me of a funny story that happened in a former company.
One of our sales guys sent an email that was hardly readable because of
too many digits, and was replied to "can't understand...". He then
called over the phone complaining that his keyboard "doesn't work".
Because 'i' generates "1", ... etc. So he can't compose all those
important messages to important people...
It turned out he had accidentally activated the "Fn key" on his notebook...
</story>
