The Translate task lets you pick a start token and an end token to delimit 
keys, but they must be single characters. This single character limitation 
makes it hard to pick a good delimiter character when you're running Translate 
against, say, an XSLT -- there just aren't many single characters left.

Also, the Task actually has more internal conditions that are required before 
a string is considered a match. If I'm using # as my tokens, it corresponds to 
something like the regular expression:

#[^\s=:#]#

It might be nice to allow the user to specify a regular expression to 
determine what a string keys may look like, to reduce spurious matches.