This info is a few months old so may now be inaccurate.....
I believe that they were working on some integration With .msi packages, including allowing file definitions to be read directly from the package definition rather than having to be identified manually from scans based on a hard disk.
This type of integration would make teaching any application based around an msi intaller much more efficient, accuracy could in theory also be improved as the package definition would identify all files and directories installed for the application as well as other elements such as registry keys (used in some of the more advanced recognition code post V7).
I think for now that this is about as far as you can reasonably go with automation of regognition training. The problem with 100% automation is that there are no good standards for reporting application descriptions to a library, just look at some SMS 2.0 output to see how bad things can get if you move too far down that route
hope this helps ;-)