Re: literals, again.

On 2002-07-09 2:42, "ext pat hayes" <phayes@ai.uwf.edu> wrote:


>> , there is the additional
>> constraint that the value denoted by the literal node is the
>> specific value to which the lexical form embodied in the label
>> maps, according to the lexical to value mapping of that datatype.
> 
> But there is in general no way to connect the literal to the
> particular datatype 'through' the value, that is the (your) point. So
> just knowing that something has the rdf:type of a certain datatype
> class is no longer sufficient to tell us how to interpret a literal
> connected to that thing.

Yes there is. This is defined by the very nature of datatypes.

A datatype has a value space, a lexical space, and an N:1 mapping
from the lexical space to the value space.

If we have a member of a value space, i.e. a value of rdf:type D
where D is rdf:type rdfs:Datatype, then if there is a lexical form
(literal) associated with that value, the semantics of the *datatype*,
not rdf:type, require there to be a mapping defined from that lexical
form to that value, or else the graph is ill-formed according to
the datatyping.

The point of connection between the value and the literal is the
fact that the rdf:type D of the value is rdf:type rdfs:Datatype.
I.e., because we're dealing with a datatype, not some arbitrary
RDF Class, and datatypes define precisely such connections between
literals and values.

>> The rdfs:range semantics tells us the literal node denotes the value.
>> The rdfs:Datatype semantics tells us the label is the lexical
>> representation of the value.
>> 
>> The fact that the class extension of the datatype is only its value
>> space is only a problem for query engines or similar applications
>> which might separate a lexical form from its original datatype
>> context,
> 
> Any engine might do that, since there is no such thing in as this
> 'context' in an RDF graph, right?

Sigh. Again you're misunderstanding "context". I've explained
it dozens of times to you Pat. By context here, I mean that
the datatype itself provides a scope of interpretation for
the literal. You use the word 'context' yourself in your latest
edited version of the stake-in-the-ground proposal. Go see how
the word is used there. I'm using it here in the same way.

>> based on the original triple in which it occurs (by e.g.
>> applying subPropertyOf relations, etc.)
>> 
>> As long as implementers are aware that subPropertyOf and subClassOf
>> relations are valid for datatype values but not mappings, and
>> that the original context must be preserved, then we're fine.
> 
> But what does it even *mean* to preserve this 'context' ? And if
> datatype mappings are properties, then why is subPropertyOf not valid
> on them?

You yourself just answered this question to Brian. Because inheritance
in RDF/S is downwards, not upwards. If a given property X has
an rdfs:range of datatype D, and X rdfs:subPropertyOf Y, then Y
does not have an rdfs:range of datatype D, insofar as the
subproperty relation is concerned, and thus, if one separates the
literal from datatype D per the original statement using property X,
then it may not be possible to get the same interpretation, or
any interpretation at all if in fact property Y has no datatype
associated with it.

Thus,

   John age "10" .
   age rdfs:range xsd:integer .
   age rdfs:subPropertyOf personProperty .

implies

   John personProperty "10" .

but does *not* imply

   personProperty rdfs:range xsd:integer .

and thus, when trying to interpret the latter implied statement

   John personProperty "10" .

since personProperty has no defined range (in this example),
the pairing ("10", ?) is incomplete and has no datatyping
interpretation.

Alternate, the property personProperty could have as its range
a datatype other than D where the value space subsumes that of
D but has a different lexical space and/or lexical to value mapping.

So, to separate a literal from the "datatyping context" of the
original statement may preclude or corrupt the correct datatyping
interpretation expressed by the original statement.

Eh?

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 Friday, 12 July 2002 05:12:44 UTC