For my own convenience I programmed a small PHP utility that scans my large corpus for two words, and compare the words around those, giving the most word-bound token strings as output.
This way it is easy to make antipatterns for rules that present too much false positives.
If you would like to try, just say so.