[Exim] Regex Help - new spam catching trick.

OK - I have yet another idea on catching spam and I need a little regex
help. Here's what I need to do:

I want to grab the subject header and create a second header containing
only words that match a predetermined list.

If my list contains (sale|get|free|win|prize) and the subject is:

Get $10 or $50 free and win $20,000!

the expression will return "getfreewin"

What I'm thinking is taking subject lines - getting rid of all but
certian words that I compress into combo words that bayesian filters
will recognize and learn and start scoring.

So - can anyone do my regex for me? I'll share the results.