Re: Denotation of datatype values

On 2002-04-08 14:58, "ext Brian McBride" <bwm@hplb.hpl.hp.com> wrote:

> At 12:38 08/04/2002 +0300, Patrick Stickler wrote:
> 
>> Jeremy's recent posts regarding entailments between the idioms
>> suggests that there is an important question that seems to need
>> answering, namely, do we wish/expect/need that a given datatype
>> value (member of a datatype value space) always and only be
>> denoted by a single node in the graph? (as opposed to it being
>> denoted by a combination of nodes or statements such as the
>> inline idiom with an rdfd:range assertion).
> 
> That question doesn't seem to me to accurately capture the concern.  The
> answer is obviously no, since several bnodes in a graph may denote the same
> value, but that does not get at the essence of Jeremy's issue.

I am not talking about unique nodes. I am talking about a single
node denoting the value, rather than a "constellation" of nodes or
triples denoting the value.

In the case of the inline idiom, no single node denotes the value.
The denotation of the value is implicit in the combination of the
literal node (lexical form) and rdfd:range assertion (datatype
context).

> Doesn't Jeremy's entailment capture the issue well enough?

It does touch upon the question, but I don't think it explicitly
asks the question directly, no.

To rephrase Jeremy's entailment in terms of the above question:

Is it unacceptable for a literal node without an rdfd:range assertion
to not denote a value, but only a literal whereas the same literal node
along with an rdfd:range assertion (but not by itself) can denote a value.

I.e. given only

   Jane ex:age "25" .

"25" alone does not denote the value twenty-five. But given

   ex:age rdfd:range xsd:integer .
   Jane ex:age "25" .

then "25" and the rdfd:range assertion *together* denote the
value twenty-five. Yet still, "25" alone does not denote the
value twenty-five. There is no single node in the graph which
denotes the value twenty-five. The value remains implicit in
the datatype interpretation.

Contrast this with the lexical form idiom. Given only

   Jane ex:age _:x .
   _:x rdfd:lex "25" .

the node _:x does not denote the value twenty-five, it only
denotes some presently unknown value that has a lexical
representation of "25". As above, if we expand that to

   ex:age rdfd:range xsd:integer .
   Jane ex:age _:x .
   _:x rdfd:lex "25" .

then, and only then, does _:x denote the value twenty-five,
but we then have a single explicit node denoting the value,
it does not remain implicit in the interpretation, even
though the value is fixed by the datatype interpretation.

The first case, with the inline idiom and literal node, may
at first seem to be non-monotonic, as if in one case the literal
node denotes a string but in the second case it denotes a value.
But isn't non-monotonic because in both cases the literal node
itself does not denote the value, but always the literal string.
Knowledge does not change, it is simply expanded by additional
information provided by the rdfd:range assertions.

Insofar as Jeremy's actual entailment example is concerned:

Does

  Jenny age "10" .
  Tommy age _:b .
  _:b integer "10" .

entail

  Jenny age _:x .
  Tommy age _:x .

I.e. do Jenny and Tommy have the same age?

In the absence of an rdfd:range assertion, i.e.

  age rdfd:range integer .

then Jeremy's entailment cannot hold because the inline idiom does
not denote a value. Jenny's actual age is underspecified.

In the presence of the rdfd:range assertion above, however,
then I consider that Jeremy's entailment does hold insofar as the
datatyping interpretation is concerned, i.e. both Jenny and Tommy
have the same age, which is 'ten'. Whether this entailment is
provided by some closure rule that can derive the datatype triple
idiom from the inline idiom with rdfd:range assertion, e.g.

   { 
      ?property rdfd:range ?datatype .
      ?resource ?property ?literal .
   }
   log:implies
   { 
      ?resource ?property ?value .
      ?value ?datatype ?literal .
   }

so that Jenny's age is denoted by a single explicit node in the graph,
seems secondary to the real question of whether Jenny and Tommy have
the same age, which in the presence of the rdfd:range assertion, they
do.

Now, getting back to the actual question at hand...

If we need/desire that the value always have an explicit denotation
by a single (not necessarily unique) node in the graph, then the
present definition of the inline idiom is unacceptable and either

1) we eliminate the inline idiom

This is basically to return to the original "convergence" proposal,
using only the lexical form and datatype property idioms. I think
this is the least acceptable of the two options, because users have
clearly stated that they want the inline idiom, and it's our job
to figure out how to make the MT work with it.

2) we adopt untidy literal nodes which denote values according to
   rdfd:range assertions.

I.e. literal nodes become just like blank nodes in the lexical form
idiom, denoting "some" value, only with literal (lexical form) labels
rather than no label. They become in essence a contraction of the lexical
form idiom with the same semantics as the lexical form idiom (and we can
then eliminate the lexical form idiom as redundant). This would still
not be non-monotonic, as the lexical form idiom is monotonic.

--

However if we are OK with values sometimes not having any explicit single
node denotation in the graph, then the present treatment is fine, and
we do nothing (other than get the WD finished ;-)

Thus the issue boils down to the particular question that I'm asking.
Is it OK if values remain implicit in the interpretation or must
they always be denoted by some specific node in the graph? Whatever
we decide, the options seem fairly clear, but we need to decide.

If the WG decides yes, then we proceed as we have been. If the WG
decides no, then I strongly support option 2 above.

Patrick

--
               
Patrick Stickler              Phone: +358 50 483 9453
Senior Research Scientist     Fax:   +358 7180 35409
Nokia Research Center         Email: patrick.stickler@nokia.com

Received on Tuesday, 9 April 2002 03:17:03 UTC