This is work in progress. I'm working on grammar rules now. Last night though I posted the spell checker poem in just to see how the system would respond to it and I was really surprised (sometimes we're our own harshest critics). I have a plugin for Wordpress too at <a href="http://www.afterthedeadline.com" rel="nofollow">http://www.afterthedeadline.com</a>
This is pretty neat to see. But this is probably the wrong metric to use to measure performance of a spell checker. In the real world you don't want to spell check writing that is 0.5 spelling mistakes per word, and many of these misspellings are not representative of how people actually misspell words. (Who would actually type "mist ache" for "mistake"?) Also, this may just reflect the fact that you have your thresholds for tagging a misspelling set lower. For good metrics you would want to have a large corpus of real documents and measure both the false positive and false negative rates.<p>(To be fair, the author probably knows this stuff already, and is just having fun.)
I tried the demo at <a href="http://www.polishmywriting.com/" rel="nofollow">http://www.polishmywriting.com/</a> and it found some mistakes I missed in my writing. I am impressed with it's suggestions although it suggested many technical words as spelling errors. Check it out on your writing.
It seems in this case you can just break them down into sets of phonemes and do a frequency analysis to get the most common usages.<p>That would take care of the "Eye have"
Just fyi, there are other startups in Australia that also work on this spelling problem <a href="http://spellr.us/" rel="nofollow">http://spellr.us/</a>