superman [Was: [All] Proposal: RDF Graph Identification]

* David Wood <david@3roundstones.com> [2012-08-16 16:56-0400]
> Hi Pat,
> 
> Thanks for the useful feedback.  I can fold in the changes where we agree, hopefully by tomorrow midday.  I've also asked some questions back to you below.
> 
> On Aug 16, 2012, at 13:45, Pat Hayes <phayes@ihmc.us> wrote:
> > ...
> > 6. I do not understand what this refers to: "This quoting behavior is considered to be important; it avoids the “superman” effects that plagued RDF reification."
> 
> Ivan?
> 
> /me refuses to be drawn on http-range-14 or reification ever again :)

Here's an old doc where I tried to capture Superman: <http://www.w3.org/2001/12/attributions/#superman>. It's essentially a fallacy resulting from applying one person's axioms to another person's data and acting surprised when that other person appeared to draw conclusions they couldn't know about. Reification encourages us to take n different, um, let's call them "graphs", and put them in one reified graph. This is a problem if we then apply e.g. owl:sameAs to the terms in the rdf:Statements. We end up inventing assertions on behalf of people, e.g. recording that Lois Lane said that Clark Kent can fly (which apparenlty she doesn't know).

In the same way that inferring over reified statements from different sources appears to put words in people's mouths, so to does inferring over different sources sequestered into named graphs (albeit with fancier punctuation). I read "avoids the superman effect" in section 3.1 as saying "don't do that." The problem is boxed up in a way that gives people some guidance for how to package their data in order to impede inference. I guess "interpretation of an RDF dataset is the interpretation of its default graph" says that you aren't held directly responsible for mistruths that you put in a graph called <damnedLies>, or even <foo>. Some assertions in the default graph could follow conventions for asserting the veracity or endorsement of a particular named graph; so people can again sue you. Yay!

I'm not sure that we can or even want to solve the Superman problem by gating inference. It would be challenging to leave no corruptable assertions in the default graph; plus I'm kind of hoping that the SemWeb will have sufficient lattitude to tell Lois Lane that Clark Kent *is* superman (and that she effectively just said that Clark Kent can fly.) I personally think we should skip the superman issue and let provenance tell us how to report who conluded what.

-- 
-ericP

Received on Thursday, 16 August 2012 22:44:27 UTC