W3C home > Mailing lists > Public > www-archive@w3.org > September 2013

Re: defn of Named Graph

From: Pat Hayes <phayes@ihmc.us>
Date: Fri, 20 Sep 2013 01:16:55 -0700
Cc: Jeremy J Carroll <jjc@syapse.com>, Sandro Hawke <sandro@w3.org>, www-archive <www-archive@w3.org>
Message-Id: <6E78DD9C-949E-4536-8D7F-D892FEE80C27@ihmc.us>
To: Gregg Reynolds <dev@mobileink.com>

On Sep 19, 2013, at 2:20 PM, Gregg Reynolds wrote:

> On Thu, Sep 19, 2013 at 2:35 PM, Jeremy J Carroll <jjc@syapse.com> wrote:
> ...

> In particular you said:
> [[
> metonymy [] already accounts for the way RDF properties and classes are construed by RDF Semantics; we can just do the same thing for graphs.
> ]]
> which, if we could flesh that out, would achieve the goals I think.
> Then an RDF property is not a set of pairs, and an RDF class is not a set of resources, and a named graph is not a set of triples  but there is one step of indirection which you are referring to as being a metonym.
> I would shy away from "is" and use "denotes" instead.

No, Jeremy is exactly right, and "denotes" here would be incorrect. An IRI in a property position in an RDF triple *denotes* an (individual element of the RDF universe which is an) RDF property (by definition). And this RDF property has an associated extension, ie a set of pairs (again, by definition, in this case the definition of RDF interpretation). What Jeremy said is that the property and the extension are distinct, which indeed is (correct, and) an unusual feature of the RDF model theory. But this is not metonymy. There is no reason to suppose that such an IRI in a property position has its 'normal' meaning shifted to refer to something other than what it correctly or normally denotes. 

>  In my experience, is-talk, especially in discussions about language, is often frustrating, since "x is y" can be taken to refer either to the symbol 'x' or whatever it denotes.  Restricting usage to either "denotes" or "refers by metonymy" (or whatever phrase works best) avoids the ambiguity.
> How to say that in a simple ways seems to be the issue
> Indeed.  I think consideration of metonymy can help clarify the issue by relating the technical, structural aspects of RDF semantics to ordinary natural language usage, but the term itself risks alienating at least some readers.

I think that the idea that metonymy has anything whatever to do with RDF is wildly misleading. There is nothing in the RDF specs to suggest this idea. 

> Another idiom that might conceivably be useful is the distinction between de dicto and de re interpretation. I haven't worked this out in detail, but I think it's worth thinking about.

We thought about it, quite hard, when defining RDF in 2004. We debated whether to have a de dicto reading of reification, in particular, but rejected the idea, I believe quite correctly. I would vehemently oppose trying to bring back any kind of de dicto machinery into RDF. For a start, that would immediately break all OWL reasoners. The referential transparency achieved by a rigorously de re semantics is a very valuable property, particularly in a language intended for distributed use on the Web.

> Here's an unintentionally amusing example (invented to illustrate the de re/de dicto disinction) from a book published in 1994:
>    The president of the United States will be black by the year 2000.
> Since Bill Clinton was president in 1994, read de re this would mean "Bill Clinton will be black by the year 2000."  Read de dicto, it means that "The president of the United States is black" will turn out to be true by the year 2000.  In other words, whatever it is that "The president of the US" refers to, between 1994 and 2000 it will have been black at least once.
> Now suppose we have a two names :g and :h that refer to the same graph:
>    GRAPH :g { :a :b :c.}
>    GRAPH :h {:a :b :c.}
> And we have :g :createdBy :Jones.  On a de re reading, this would (I think) mean Jones created {:a :b :c.}, which would entail :h :createdBy :Jones (also on a de re reading).

This depends on exactly what we take GRAPH to mean, which is what this thread is all about :-)

>  But on a de dicto reading, no such entailment would follow; it would mean something like "whatever it is that :g refers to, was created by Jones."  Since that "whatever" might not be the graph referred to by :h, no entailment involving :h follows.

Indeed, but no entailment of any use follows at all, since the de dicto interpretation removes all referential contact between the name and its referent. De dicto semantics effectively puts unknown modalities into the logic at random places, breaking virtually all useful entailments. 

> To put it another way:  the de re reading commits the reader to the statement that Jones created {:a :b :c}.  The de dicto reading does not.  Example:  "The Fab Four had moptop haircuts" does not commit the de dicto reader to the statement "The Beatles had moptop haircuts", since that reader may not think that the Fab Four and The Beatles have the same extension (John, Paul, George, and Ringo).  She might think the Fab Four are the Monkees.
> This account is different from the metonymy account

When you say "account", this suggests that there are two detailed proposals here. AFAIKS, there are no proposals at all. 

> in that it assigns no determinate denotation/reference to :g on the de dicto reading.  We would still have to say it denotes something in IR which maps to the graph, but you can say something about that something without knowing its identity.  Just like we can say "The next president of the United States will be female" without knowing who that will be.  Or "The 12th president of the United States was male".  I haven't the slightest idea who the 12th president was, but I know he was male.

And when you say that sentence, you are in fact referring to someone, even if you don't know who that actually is. Which is WHY you know he is male, because you know that referent is in an all-male set. But this is all de re reasoning. 

> I'm not sure at this point if something like this works, or could be worked into a clear and simple account that satisfies the requirements, but it seems promising.  At the very least, you'd have to provide a way of explicitly indicating which reading is the intended reading.

This is all completely beside the point. None of this (metonymy, de dicto reasoning, modality) has got anything to do with either the topic we are discussing or RDF semantics. 


> Cheers,
> Gregg

IHMC                                     (850)434 8903 home
40 South Alcaniz St.            (850)202 4416   office
Pensacola                            (850)202 4440   fax
FL 32502                              (850)291 0667   mobile (preferred)
phayes@ihmc.us       http://www.ihmc.us/users/phayes
Received on Friday, 20 September 2013 08:17:27 UTC

This archive was generated by hypermail 2.4.0 : Friday, 17 January 2020 22:34:48 UTC