I did a proof of concept for a better spell checker about 3 years ago. Query groups of two to three words in a search engine and look at the word count. Then replace the word in question with other words that are similarly spelled and run a query with each. The word with the highest result count is extremely likely to be the correct word. Really, it's surprising how accurate it is.
The glory of this is that it works with proper nouns that don't occur in dictionaries (IE: xkcd). In Google's initial demo for Wave, they showed "Icland is an icland" be corrected to "Iceland is an island." I'm fairly confident they took a similar approach. There's also a good chance it could work for other languages, because it doesn't use anything specific to English.
The disappointing part is that most of the accomplishment comes in "Suggestion Intelligence First," meaning that from a list of 5, the top result is the correct result. In most cases, the Suggestion Intelligence is just fine, you will just need to pick the right one yourself.
And this was the "research paper." Unfortunately it was a three week program, and myself and the other coder (IE: the ones who understood how the thing worked) didn't contribute much to the paper. Feel free to ask if you have any questions. The email in there isn't actually my email:
http://soe.rutgers.edu/sites/default/files/gset/Paper08-Hung...
The glory of this is that it works with proper nouns that don't occur in dictionaries (IE: xkcd). In Google's initial demo for Wave, they showed "Icland is an icland" be corrected to "Iceland is an island." I'm fairly confident they took a similar approach. There's also a good chance it could work for other languages, because it doesn't use anything specific to English.
The disappointing part is that most of the accomplishment comes in "Suggestion Intelligence First," meaning that from a list of 5, the top result is the correct result. In most cases, the Suggestion Intelligence is just fine, you will just need to pick the right one yourself.
If anyone's interested, this was my presentation: http://soe.rutgers.edu/sites/default/files/gset/Presentation...
And this was the "research paper." Unfortunately it was a three week program, and myself and the other coder (IE: the ones who understood how the thing worked) didn't contribute much to the paper. Feel free to ask if you have any questions. The email in there isn't actually my email: http://soe.rutgers.edu/sites/default/files/gset/Paper08-Hung...