I partially agree and partially disagree.

>> http://www.tei-c.org.uk/wiki/index.php/Conformance is perfectly clear:  
>> while it may be seen as liberal, it defines conformance only with  
>> respect with element /grammar/, not element /usage/. And you explicitly  
>> stated latter that the element usage is not interesting, providing  
>> element grammar is correct. This is a major reduction of what  
>> "conformance" mean.

> I agree, this is a serious problem. But what can we do? We have no  
> mechanistic way of checking for tag abuse or appropriate encoding 
> depth. 

Yes we have: humain brain. This is not because we have no mechanistic way of validating practices that practices should not be taken into account in a definition of TEI-conformance (what is a (good) TEI text for the TEI). In fact I don't see how practices, scientific choices, etc., could ever be checked by a mechanistic way. And even I we (imagine we) could, I don't know if we should.

It is the question of the normativity: which is the instance responsible for? which instance grant garanties? Should it be a formalism, or something else? In other words, are the essentiel aspect of a TEI annotating practice (or a TEI document content) defined by the XML grammar? This is a very strong and insightful choice of the TEI, as I understood it, to say: "while formalism can express a part of the important content (the content important for the document being exchangable), it is *not* the only ground for conformance, because formalism is related to encoding practice, choice, and assumption; which have to be made explicit"; formalised content cannot without drastic oversimplification be defined as the TEI content, as in a Docbook document, so it cannot be the source for normativity. I think the TEI made a courageous and clear choice, absolutly required for objects of the humanity computing fields. Practices have to be related to hypothesis, choices, goal of the annotation, social and material conditions, etc. The formalism cannot be the /norm/ for such an object (the text), only a /tool/ for expressing parts of the claimed norm of a document. With this position, the TEI has a philologic dimension, which will be loose if conformance is defined only in formalism terms.

I think that if we think of long-term exchangability, it is certainly not an ODD that can make a document reusable in 50 years, when vocabulary (and perhaps even syntax) will have changed. It is very rather the tagUsage, the project description, the conditions, etc. Similarly, if one interchange corpora with a very different field of research, the ODD has no real use. It is rather the definition of "w" (Compound words are words? "fire engine" is one or two word? "pomme de terre" is one or two words?) Which algorithm and heuristic have been used? what where you assumption?) etc.

> I think "conformance" is *one* of the  
> preconditions for being a Good TEI Text.

So you have to make explicit what is the other. And in the definition of Good TEI Text proposed on the wiki, it is defined only in terms of formalism. (I assume you mean "conformance" as "conformance to a grammar" (formalism-conformance), while I use conformance as "how the TEI define a Good TEI Text" (TEI-conformance)).

> If I get some money to encode a text, my funder may say that I must  
> produce a GTT, and they may appoint
> a reviewer. I would expect that reviewer to a) check that I conform to  
> the TEI, and b) check (by human inspection)
> that my encoding meets current academic norms. It would surely be  
> surprising if they *only* did b)?

Of course, a project should be able to define a formal conformance, and validate it. But the point is not the risk of reducing a) to b), which nobody claim, but reducing b) to a) in the definition of a TEI document (reducing conformance to formal-conformance).

>>  If one choose a new set of elements, which cannot be "mapped" to  
>> existing one with <equiv> element, but are correctly and finely  
>> documented in the Header, why should he been told less conformant than  
>> someone who has not this need?
> I am regarding  "conformance types" as being an unordered set, so the  
> word "less" is not appropriate.
> Syd has used the word "degrees" in that wiki page, which I think is a  
> bit unfortunate as it may
> reintroduce the ordering; the point is that that there are different  
> _ways_ to be conformant
> (or "interchangeable" if you prefer not to use the C-word), it is not
> a binary divide. In your example, the project is TEI Conformant  
> according to Type 3 rules - its just
> hard to find names for these types.

I think it is useless to deny that formalism-conformance *does* have degrees. This is an ordered set of formalisation. If you say that they are different ways, not hierarchised degrees, this is because you have *another* point of view. You have the assumption of the TEI as allowing different practices. And this is a requirement to make explicit your assumption. If there is different practices, it entails for the TEI users responsibility and justification of his choices. That's why the choice of the TEI, not to make TEI-conformance relying *only* on formalism conformance, sounds very insightful and precious to me. Formalism can help to give garanties that some part of the claimed content of the document is effectively conformant to the claim. But the point is that *it remains a claim* what a TEI document is (or: there is always a claim of what a document is, even in a docbook one, but given the complexity of the object of the TEI and its philologic dimension (I claim that) this claim should be in the definition). For instance, if you said that you use "w" in the grammatical meaning of the TEI, you can check that is grammatically true. But your word does have another definition important for exchangeability (what exactly is a word for you?), while in docbook a para is not subject to interpretation of that importance. The TEI is precious because it states and encourage to recognise that there is another normativity "*behind*" the grammar:

    The Guidelines provide a means of documenting 
    the encoding in such a way that a user of the 
    text can know the reasoning *behind* that encoding, 
    and the general interpretive decisions on which 
    it is based. It is *strongly recommended* that 
    the TEI header be used to give an account of 
    these aspects of the encoding. 

    (http://www.tei-c.org/P4X/AB.html, my bolds)

This "recommendation" should perhaps not disear in the definition of conformance.

>> And in fact I don't see why some people can not benefit from the TEI  
>> only because it is a clear list of features and it provides immediate  
>> good suggestions, because they are sure not to choose element names  
>> that will conflict with new elements if they extend vocabulary, because  
>> they can create a header, etc., and who produce document validating  
>> only against somewhere hacked schemas, providing they are documenting  
>> their choice.
> I don't disagree at all. I think its a perfectly fine use of the work  
> that has gone into the TEI. However, wearing its other hat as a 
> pseudo-standard, the TEI must address conformance.

Yes, but how should the TEI adress it? The only point I'm in trouble with in your definition is that conformance is defined *only* in terms of formalism. I think it should take into account the documentation, and said more generaly: a TEI conformant document is a documented document, with the following possible degrees of formalisation, *and* a human-documentation for which there is this whole vocabulary. Selecting a degree of formalism is related to "local processing" needs; while the TEI can (usefully, I admit! :-) define different degrees of formalism, the TEI should perhaps not *be defined* by this degree. Of course this degrees of formalisation are usefull, and could, for instance, help to make tools reusable.

>> ... I think that documentation of practice (element usage) is at least  
>> as important for perenity than documenting element grammar. Why  
>> conformance should be defined regarding only the latter?
>>
> because the latter cannot be formally coded at present.

Because it cannot be formally coded, it should not be take into account, while it does obviously exist?

-- 
Sylvain Loiseau
[log in to unmask]
http://panini.u-paris10.fr/~sloiseau