I remember last year when Greplin was still working on improving search in Gmail - unfortunately I was never that impressed with the results, but I love the premise.<p>I would especially expect Google to do a better job with email search than they do. They know pretty literally everything about me, and have some amazing natural language processing tech. Of course, using all of the available data and, worse, the NLP algorithms, immediately becomes very expensive computationally. As covered in this Quora thread (<a href="http://www.quora.com/Why-does-Gmail-take-longer-to-search-my-inbox-than-Google-takes-to-search-the-entire-web" rel="nofollow">http://www.quora.com/Why-does-Gmail-take-longer-to-search-my...</a>), there's an awful lot of email out there to search.<p>Maybe local clients like Sparrow are the solution to the high costs of good email search - I don't really mind if my computer bogs down for the half second it takes to get really great results.
Sounds like a job for latent semantic indexing. This is a very accessible piece on how LSI works and how it can be applied. <a href="http://www.knowledgesearch.org/lsi/" rel="nofollow">http://www.knowledgesearch.org/lsi/</a>