Re: "meaningless terms" verbage for Primer

Patrick Stickler wrote:

> 
>>Patrick Stickler wrote:
>>
>>>Here is my proposed verbage regarding terms with undefined
>>>semantics, for inclusion into the Primer. Frank, feel free
>>>to wordsmith freely.
>>>
>>>Note that the list was based on there being no defined semantics
>>>for the term in the MT, even if there might be range, domain,
>>>type, or other assertions made about the terms. The list (or the MT)
>>>may need adjusting if it is incorrect...
>>>
>>>rdf:li was not mentioned in the MT, but that may be an omission
>>>since it's in the syntax doc. It is included in the list
>>>below.
>>>
>>>--
>>>The RDF/S vocabulary includes several terms the meaning of which
>>>was undefined or ambiguous in earlier specifications of RDF and
>>>which remain undefined in the present RDF MT.
>>>
>>>Specifically:
>>>
>>>   rdf:value
>>>   rdf:Bag
>>>   rdf:Seq
>>>   rdf:Alt
>>>   rdf:li
>>>   rdf:_n
>>>   rdf:List
>>>   rdf:first
>>>   rdf:rest
>>>   rdf:nil
>>>   rdfs:comment
>>>   rdfs:seeAlso
>>>   rdfs:isDefinedBy
>>>   rdfs:label
>>>
>>>These terms remain in the RDF/S vocabulary for various historical
>>>reasons. Their lack of an explicit or clear interpretation has
>>>resulted in their being used in incompatable ways by different
>>>applications. Nevertheless, as they provide utility to certain
>>>RDF applications, and in the interest of backwards compatability,
>>>they have not been deprecated or removed.
>>>
>>I suspect a reader familiar with M&S would think many of these terms
>>*did* have a clear and explicit interpretation.  In some cases we've
>>backed off from what was said in M&S.  
>>
> 
> Then we need to say where we've backed off from M&S.


I agree.  However, I don't think the Primer is the place to do this 
(example:  isDefinedBy)

 
> 
>>In other cases, the problem was
>>the lack of a *machine-interpretable* interpretation.
>>
> 
> Hmmmm.... isn't that the only point to RDF? Enabling *machines*
> to do something useful with our content?
> 
> If there is no machine interpretable interpretation, then IMO
> there is no interpretation whatsoever. Eh?


Of course we want machines to do something useful with our content. 
However, there is quite a lot of machine interoperability going on 
without machine-interpretable *definitions* of the terms and data 
structures being used.  Instead, people read definitions, and write the 
software to behave in accordance with those definitions.  Practically 
every database in the world operates this way, for example (and most of 
the RDF currently works this way too).  I understand that, without 
machine-interpretable definitions, there is no guarantee that people 
will interpret the terms, and hence write the software, consistently. 
That's what we're trying to *appropach* in the Semantic Web.  However, 
there is no way we're going to be able to totally throw away the idea of 
people having to build a certain amount of additional understanding into 
software, barring the creating of the perfect description languages (I'm 
not holding my breath).  Some variant of your "here be dragons" warning 
is fine, along with carefully distinguishing what can be *guranteed* by 
virtue of having precise machine-interpretable semantics and what can't, 
based on the need for human understanding of terms (and reflection of 
that understanding in software), and the possible mistakes (or 
deliberate evasions) that may occur.  However, talking as if the use of 
such "software defined understanding" is totally out the door now seems 
similar to refusing to take a lifeboat and radio when embarking on a 
leaky ship.


> 
> 
>>>Note that some of these terms do have certain constraints defined
>>>in the MT for their use, such as their domain, range or type, but
>>>their actual meaning is not specified.
>>>
>>>Users should take care when employing these terms, as there is no
>>>guaruntee that any RDF applications will interpret them as intended.
>>>It should also be noted that no valid inferences may be drawn from
>>>statements using these terms, insofar as the model theory is concerned.
>>>Any interpretation or inference based on these terms is entirely
>>>application specific.
>>>
>>Can we explain this without references to "inferences"?  As valuable as
>>the concept is, at this point I'd just as soon not get into explaining
>>in the Primer what "inferences" have to do with anything if I can help
>>it.
>>
> 
> Fine. I stole that from comments made by Pat. The key point is that
> users should not expect RDF applications to do anything consistently,
> or anything at all, based on those terms as their interpretation is
> always application specific, if any interpretation occurs at all.
> 
> Human intuition about what the natural language mnemonics of the
> terms suggests has no normative weight here, and folks are free
> to use the terms in whatever way they choose without violating
> the RDF specs in any way.


I don't know about that.  Do you mean that even if something is 
specified in English in an RDF normative spec, that doesn't matter, and 
only what is describable in the model theory must be adhered to?


> 
> In short: here be dragons...
> 


--Frank

 


-- 
Frank Manola                   The MITRE Corporation
202 Burlington Road, MS A345   Bedford, MA 01730-1420
mailto:fmanola@mitre.org       voice: 781-271-8147   FAX: 781-271-875

Received on Tuesday, 10 December 2002 11:06:43 UTC