handling rdf:value

After reading Franks section in the primer more carefully, I would 
like to make the following suggestion for how to handle rdf:value, 
which I think codifies the intent rather better than any other idea 
we've had so far.  I've rewritten Frank's section 4.2 along these 
lines in the version at 
http://www.coginst.uwf.edu/~phayes/RDF-Primer-modified.html, but of 
course this rewrite is only OK if people agree to the treatment.

-------

Frank characterizes the typical use of rdf:value as a way to indicate 
a 'primary' value of a multi-arity relation. (Note, this is not at 
all the same notion as a primary field in a DB.) That is, when R is a 
more-than-binary relation, but can be abbreviated usefully as a 
binary one by ignoring some of its arguments, then the rdf:value is 
the argument that should not be ignored. In cases like this, we can 
typically think of the binary form as an abbreviation or summary of 
the longer formulation, where some detail has been omitted or 
suppressed.

I think that this very nicely captures the intended range of uses for 
rdf:value, and doesn't get it confused with issues like 
distinguishing dimensions from values or textual forms from real 
values, which I had often gotten it confused with. But it suggests 
the following slightly modified treatment.

The *strictly correct* use of rdf:value is to do *exactly* the above, 
and no more; ie to say, when a relation with more than two arguments 
is described by having a structured value which itself has the other 
arguments as values, which one of those arguments can be 
appropriately used as the single argument when the n-ary relation is 
abbreviated or summarized as a binary relation, ie a simple property. 
For example, using the address example that Frank gives:

exstaff:85740  exterms:address  _:johnaddress .
_:johnaddress  exterms:street   "1501 Grant Avenue" .
_:johnaddress  exterms:city     "Bedford" .
_:johnaddress  exterms:state    "Massachusetts" .
_:johnaddress  exterms:Zip      "01730" .

an appropriate use of rdf:value here might be to add:

exstaff:85740  exterms:address  _:johnaddress .
_:johnaddress  exterms:street   "1501 Grant Avenue" .
_:johnaddress  exterms:city     "Bedford" .
_:johnaddress  exterms:state    "Massachusetts" .
_:johnaddress  exterms:Zip      "01730" .
_:johnaddress rdf:value "01730" .

which would say that the way to succinctly abbreviate this in binary 
form would be to just use the Zip code as the address, ie that it is 
correct, even if less informative, to also write:

exstaff:85740 ex:terms:address "01730" .

Now, of course, this kind of strictly correct usage means that one 
has to say the value twice; once with its correct attaching property 
and once again with rdf:value; and so users may wish to abbreviate 
this by omitting the 'correct' property, and leaving it implicit; but 
that strategy is inherently risky, as the intended meaning of 
rdf:value is now contextual and liable to be misunderstood if taken 
in isolation. So, caution.

On this view, the 'proper' way to write the kilogram example would be

aaa weightIs _:x .
_:x ex:quantity "24" .
_:x rdf:value "24"
_:x ex:units ex:kilograms .

And omitting the second triple is an obvious economizing strategy, 
but users are cautioned that it has its risks.

-----

If people like this idea than it could be captured formally as a RDF 
semantic condition corresponding to the inference rule:

aaa ppp bbb .
bbb rdf:value ccc .
-->
aaa ppp ccc .

for any property ppp. This would fit very naturally into 
rdf-entailment. But as this goes beyond 
http://www.w3.org/2000/03/rdf-tracking/#rdfms-replace-value.
I hereby REQUEST feedback from the WG before inserting it into the 
MT. If people think it should be there then I can put it in one 
evening this week. All the proofs and so on are transparent to this 
addition.

Pat

PS. BTW, this account allow you to use rdf:value for more than one of 
the properties, and the semantics then would be that *either* of them 
could be correctly used as the abbreviating property, eg if you also 
said that the city was an rdf:value, then it would be correct to use 
either the zip code or the city as the value of the simple address 
property instead of the structured value encoding all the aspects of 
the address.


-- 
---------------------------------------------------------------------
IHMC					(850)434 8903   home
40 South Alcaniz St.			(850)202 4416   office
Pensacola               			(850)202 4440   fax
FL 32501            				(850)291 0667    cell
phayes@ai.uwf.edu	          http://www.coginst.uwf.edu/~phayes
s.pam@ai.uwf.edu   for spam

Received on Sunday, 8 December 2002 23:59:50 UTC