Re: [xml-dev] [ANNOUNCE] Feature Grammars: a tool for featureextraction and representation in XML
Because the branch under one detection of a feature must be the same as the branch under any other feature, the tree structure is equivalent to a rooted acyclic graph: just instead of having graph links the branches get replicated.
The cost is potential for an explosion due to replication, but if you do need to make a list of all the causes of a particular feature F, the XPath is simple (//*[F]) and perhaps a niche requirement.
I would expect the number of feature of interest would be way smaller than the size of the document: (if there is no clustering and the number of alternative features gets closer to the number of possible element, you are left with a kind of trimmed instance not really feature extraction.)
Rick