Re: Comments on "SPARQL 1.1 Uniform HTTP Protocol for Managing RDF Graphs"

On Sun, 2011-03-20 at 22:31 +0000, Nathan wrote:
> David Booth wrote:
[ . . . ]
> > Just as the speed of an object is relative to the observer, the
> > ambiguity/uniqueness of a URI's identity is *relative* to the
> > application.
> 
> A URI/name can still reference a unique distinct unambiguous thing even 
> when the description of that thing leaves much to be desired and is 
> extremely ambiguous.
> 
> If I create a URI <http://webr3.org/nathans#tv> and publish one triple:
>    <http://webr3.org/nathans#tv> :color "black" .
> The name is extremely unambiguous, it only refers to a single unique 
> thing, the description on the other hand leaves much to be desired.

Hang on, you have used an intentionally evocative name by putting
"nathans" and "tv" in the URI.  As we all know, machines cannot rely on
guessing the identity of a URI's resource by applying natural language
processing to the text of the URI.  That would violate URI opacity:
http://www.w3.org/TR/webarch/#uri-opacity
So to be more fair, let me re-write your assertion as:

  <http://example/vxb24#a37f> :color "black" .

To *you* that URI may be unambiguous, but only because you have lots of
private, out-of-band knowledge about it, such as knowing that it is a
TV, that it belongs to you, etc.  But if the above statement is the only
definition that you published for that URI then an RDF consumer without
the luxury of your private, out-of-band knowledge would know only that
the URI identifies something with a "black" color.  This may be unique
*enough* for an application that is merely classifying objects by color,
but it would be very ambiguous to applications that are trying to
distinguish between TVs and shoes.

In effect then, *you* would be using one private definition of
<http://example/vxb24#a37f> and the rest of the world would have only
the very ambiguous definition that you published, which merely says that
it being something "black".

> 
> Does this mean that the name refers to anything which is black in color? 

To the rest of the world, yes.

> does it mean I can use the same URI to refer to everything I own that's 
> black in color? 

Yes, if you choose, but it probably would not be a good idea.  If you
*know* that you want to support applications that can distinguish
between the different items you own, then you should not, because for
*those* applications the URI's identity would be ambiguous.  This is
where good judgement comes into play.

> does it mean that the name is ambiguous?

Yes, to most applications, such as those that care who owns the item or
need to distinguish between shoes and TVs.  But not to applications that
only care about the thing's color.  Uniqueness of a URI's identity is
relative to the application.

> 
> Would your semantic web machinery be able to tell if I used the name to 
> refer to 3 different unique things which were all black? what about if 
> other people started using the name to describe those three things? what 
> if I fleshed out the description to describe each thing unambiguously 
> over time?

If you change the definition over time, then you risk breaking other
people's applications, if they have been using that URI.

> 
> Personally, I believe we'd have a big problem on our hands, and that one 
> name must only ever be used to refer to one distinct thing. 

That would be nice if it were possible, but it is not, as Pat Hayes (and
I'm sure several philosophers) have pointed out.  Ambiguity of reference
is inescapable.

> I also that 
> our descriptions can be as bare and ambiguous as we like, as they can 
> always be improved over time by publishing and linking more data.

Changing a URI's definition is feasible in a closed environment, where
you control *all* of the applications that rely on your URI's
definition.  But in an open environment like the web, if you change your
URI's definition -- even just monotonically adding assertions -- you
risk breaking existing applications by causing new contradictions. 

You *may* consider this to be an acceptable risk, as the authors of SKOS
have done.  But if so, you should publish your change policy so that RDF
consumers can make an informed decision about whether they wish to
accept this risk by using your URI.  Some may accept the risk; others
will opt for more stable URIs.



-- 
David Booth, Ph.D.
http://dbooth.org/

Opinions expressed herein are those of the author and do not necessarily
reflect those of his employer.

Received on Monday, 21 March 2011 13:41:49 UTC