xml-dev - Re: XML Schemas: Best Practices

["Roger L. Costello" <costello@mitre.org>]

I am forwarding a message from Mike Ripley:

>The idea of "universal semantics" is as outdated and inapplicable here 
>as it is in natural language 

I should have qualified the "universal semantics" statement more 
precisely.  I believe we should strive for common semantics within 
particular domains, and at points where domains intersect.  This is, 
I believe, the way real languages work - there are many common 
semantics so anyone who knows English can read and understand this 
e-mail thread, and there are American English specifics (since Roger 
and I are Americans), and there are occasional work specifics only 
someone from our company will understand.  I believe XML semantics 
can and should work the same way.

It's not all one way or the other, and I'm sorry to have implied that 
with the term "universal semantics".

I do believe very strongly, on the other hand, that we should 
advocate going beyond pure application specific semantics as a Best 
Practice.  This may not be the state-of-the-art right now, but we are 
talking Best Practices here, not documenting as-is practices.

On Mon, 27 Nov 2000 Ed Hodder wrote:

>I don't think there is an option but to 'allow' semantic-morph. Because 
>XML uses natural language  to structure content and, more importantly,
>communicate that content it will follow natural language laws. Dialects 
>will naturally evolve, tags may change meaning based on syntatic 
>position or context since of course the same word can describe 
>different things. If 'title' were universally tied to the 'name of 
>something' then how can it also describe a document establishing 
>ownership? Or an honorific? Or a sports championship?

This particular example would be handled with namespaces to eliminate 
the confusion the simple tag 'title' would invoke.  And I agree 
completely.  But I also believe that within the domain "sports 
championship" the tag 'title' should have a common semantic meaning.

>So to my mind there is no absolute semantics, or more precisely 
>meaning, to jdkdsfjkds that is application specific. Meaning is always 
>derived from context.

Concur.  I like very much the suggestion from Mary Pulvermacher and 
Ray Spinosa of a hierarchy of definitions.  This does the things 
we've discussed - provide some framework for semantics, allows for 
semantic morphing/refining/enhancing, and can be as broad-based or 
specific as the need dictates.  This also gets it out of the 
application, which is my primary objection.  Application specific 
semantics do not facilitate data interoperability.

>Could a hyper-context be defined so that jdkdsfjkds always means
>the same thing regardless of the application? Yes, but then XML is no 
>longer a plain language description of content and it will lose power. 
>Instead of creating a universal meaning you'd be better off moving to a 
>universal description like 
>noun_transportation_four-wheeled_generalized_en to deliniate 'car' or 
>you'd be forced to looked up which meaning you want before you wrote 
>the tag <car id Definition="4" dictionary="Miriam Webster Unabridged 
>8.5">. But I don't see that.

This starts to get into the big problem of how to implement common 
semantics.  It would be great if namespaces equated to semantic 
definitions, so a namespace qualifier would give an application the 
references needed to define the semantics each element in that 
namespace has.

> > What do you think?  When you create a schema component should that
>> component be expected to have the same semantics regardless of the
>> application that uses it, i.e., universal semantics?  Or, should the
>> component be able to "semantic-morph" to each application, i.e.,
>> localized semantics?

This summary description is more all-or-nothing than what I intended. 
As a Best Practice, I believe schema semantics should be common 
within a domain.  A domain is as big or small as the need dictates, 
and will consist of semantic morphing/refining/enhancing from other 
domains (both "higher" more abstract domains and "horizontal" cross 
domains).  It would be wicked awesome (a Boston Massachusetts area 
colloquialism for 'very nice') if namespaces = domains, but that's 
another discussion.

rip