Wow, I like it! That was very well done. I have this great big Japanese transcription database that some other engineers and I have been trying to use to cook up a speech recognizer for the language, but the transcription database has a bunch of crap in it. One way to fix this, for the recognizer is just to give it all available options, wherever there is ambiguous 'crap'. We were having a lot of trouble with this silly problem today though. So thanks a lot!