Just to clarify, are you picturing some sort of feedback loop?  I'm
just trying to get a better picture of the process (sounds like an
interesting project).

In other words, do you have something like:

1) take in a full-text document (like, say, a novel?)
2) Run it through NER, pull out locations, places, things.
3) Have a user who's read the novel (or perhaps display those words in
context?) go through each the locations and pick a lat & long using
Google Maps as an interface.  (Ie says this "Dublin" is Dublin, OH not
Dublin, Ireland).
4) Do something similar with names, only using some sort of resource
like dbpedia to display possible individuals?
5) markup the original file in an XML doc w/ identifiers around those
occurrences?

Is that what you're picturing?

Jon G.

Who doesn't really know enough about linked data to contribute, but is
interested nonetheless.