On Apr 18, 2007, at 3:57 AM, Alan Ruttenberg wrote:
Well, this is a conundrum. As you know, if anything is inconsistent
in DL, then everything logically follows.
In the base consequence relation, yes. But we can build things on top
of that.
What do you expect the behavior of the reasoner and the query
engine to be in such a case? Some sort of
bounds around inconsistencies? If so, how to set the bounds.
One resilient behavior is that if under every possible minimal repair
of the ontology something is entailed, return that.
Another is to report every possible non-inconsistency based way of
entailing something, with the entailments.
There are others.
Obviously, the desirable behavior is heavily dependent on the
application.
[snip]
A reasonable strategy might be to have 2 layers of representation.
Assert the biology as best you can. Perhaps even overstate (we will
do some of that for the demo). Have a reasoner identify the
contradictions and then, based on a curator's best judgement,
remove those findings that seem contradictory. Represent those
separately, as statements, in a database of controversial
findings. As a scientist, understand that you will need to query
both the consistent facts and the inconsistent facts if you want to
know the whole story.
Ah, yes, this is similar to the above points. But you can automate it
to extract and present the findings which *are* contradictory.
This seems a better option than what I usually see - silently
leaving in a portion of actual encoding errors with a smidgen of
controversy, and not marking any of it.
[snip]
Sometimes controversy is contradictory, sometimes not. Sometimes the
effort to make a controversy show up as a contradiction is too much.
Given enough background knowledge it might not be.
I do think intelligent contradiction management, whether the
contradictions are mere bugs or are useful information stored in the
system is very important. Recent work (by me and aditya and loads of
others) has brought us to a place where we can do very well in this
area.
So, yay!
Cheers,
Bijan.