[tei-council] Conformance inconsistency

[Conal Tuohy]

On Thu, 2007-10-25 at 19:53 -0400, Syd Bauman wrote:
> Thanks for the fast, interesting reply Conal. 
> 
> But you're implying that in order to be algorithmically conformable,
> my added stuff has to be "globally unique", 

Exactly, that's exactly the point which I think is implicit in the
definition of 23.3.

> which is clearly
> something I cannot completely control for, even if I do make up a
> namespace. (I cannot *prevent* other people from using my namespace!)

Of course neither you nor the guidelines can always compel people to use
namespaces correctly.  The best we can do is offer people a strategy for
using namespaces which will work for everyone who follows it. If we all
follow the simple rule of using their own namespaces for customisations,
then our customisations will always be distinguishable from each other,
and fully-automatic conversion will always be possible. 

> Besides, that is way too high a hurdle. 

Have we not already agreed that this is the recommended approach?

See the example of adding a "topic" attribute:
http://tei.oucs.ox.ac.uk/P5/Guidelines-web/en/html/USE.html#MDNS

> Why, when I send you a
> document, am I responsible to make sure you can transform both it and
> James's document with the same stylesheet? 

If not you, then who else should be responsible? :-)

> Surely it is enough on my
> end if you can transform my document with some stylesheet, whether or
> not the same one would work on James's documents?

That's not really "automatic" though is it? The existing definition says
the translation must be "automatic", but according to you I should first
have to look at the document, read the ODD, and then pick which
conversion process to apply. If the process of converting a
"Conformable" document to a "Conformant" document were truly automatic,
then  it would be possible to maintain an archive of "Conformable" TEI
files along with a single script which could convert any one of them to
Conformant TEI.

The process can't be fully automatic if it relies on the ODD, because
there's still no standard mechanism to link an instance to its ODD (is
there?)

> > e.g.
> >    A document is also said to be TEI Conformant if it is a
> >    well-formed XML document whose non-TEI features can be
> >    unambiguously recognised, and the document transformed
> >    algorithmically and automatically into a TEI Conformant document
> >    as defined above without loss of information.
> 
> But again, if the non-TEI feature is rendMe=, it can be unambiguously
> recognized as non-TEI (within my document instances), whether it is
> in my namespace or not.

It can be recognised as non-TEI, but it can't be distinguished from some
other rendMe.

> > Whereas if the new attribute were not in a namespace, then after
> > interchange it might not be clear what the attribute used to mean
> > (because of its name not being globally unique). 
> 
> Again, its name is unique within any documents *I* sent you. Its
> meaning should be clear because it's documented in my ODD (which I
> should have also sent you).
> Note that even if I add a subtype= to <title>, even though it is a
> name used by TEI, because it is not used by TEI on <title> it is
> still unambiguously recognizable as non-TEI and potentially
> algorithmically transformable to a TEI-Conformant document. (E.g., by
> appending the value of what used to be subtype= to the value of type=
> or some such.) This holds true whether or not I stick it into my own
> namespace.

I have already noted my objection to this: it's not sufficient for "automatic, algorithmic conversion".

> > An algorithm coud not, as a practical matter, distinguish the
> > attribute rendMe which you had added from a rendMe attribute which
> > I had added, intending it to mean something else altogether.
> 
> No, but the same algorithm is not going to be used on both your
> documents and mine. E.g., even if we use our own separate namespaces,

If same algorithm can be used (an algorithm that recognises the 2
distinct forms of markup, and handles them distinctly), then the
conversion can be automatic. If no algorithm can distinguish the 2
distinct forms of markup, then the conversion cannot be considered
automatic. According to the definition in 23.3, "algorithmic
conformance" requires automation, and automation (as I believe I've
demonstrated) requires global uniqueness of names.

> I may have created value lists for type= and subtype= of <title> that
> make it possible to transform by appending; you may have created
> values that would cause a loss if information if merely appended (but
> might be OK if appended with a "_" separator).
> 
> In case that possibility isn't clear:
> 
>             Syd             Conal
>             ---             -----
> type=       a, b, c, a_d    a, aa, aaa
> subtype=    d, e, f         a, ab, aab
> 
> The algorithm "append subtype to type" works for Syd (whether
> subtype= is in my namespace or not) but not Conal; the algorithm
> "append a '_' and then subtype" works for Conal (whether subtype= is
> in his namespace or not) but not Syd.

Yes here you've presented 2 markup schemes and 2 conversion algorithms,
but how do you decide which algorithm to apply? Because the 2 schemes
don't use globally unique names, you can't automatically pick the
appropriate algorithm. If they had used namespaced names, then the
choice of algorithm could be automated; that is to say, the 2 algorithms
could be combined into a single algorithm capable of handling both
cases.

-- 
Conal Tuohy
New Zealand Electronic Text Centre
www.nzetc.org