I think we could write a program to do this. Search any word and then make a list of all (or at least a few) words not in the result pages. Shouldn't take long for a decent crawler to find at least a few.

Problem with any simple algorithms is that Google quite often returns results within which "Search on page" can't find the word requested. It appears that they often include "sounds like" results and in some cases may include words that they imagine might be synonyms for your search term, or that might be variant spellings.

John