Udated 26 April 2008
Something seems to be happening in the machine translation world. I've set up a crude (and methodologically somewhat flawed) comparison of two of the main methods (rules-based, and probabilistically-based). If you can spare 15 minutes, please contribute. [Link no longer active.] View results.
(For more on probabilistically-based machine translation, see 21 minutes into this May 2007 talk "Theorizing from data" by Google's Peter Norvig.)