James K. Tauber asks:
 
> 1. In <profileDesc><creation>...</></> we'd like not only to include
> <date>...</> but also the region the manuscript is believed to have been
> written (eg Egypt). Does the content model for <creation> allow for a
> suitable tag to put this in?
 
Well, the Guidelines (this group's Gospel) say:
 
  The <creation> element may be used to record details of a text's creation,
 e.g. the date and place it was composed, if these are of interest...
 
and the content model is
 
 <!ELEMENT creation - O (%phrase.seq;) >
 
so you can put almost anything you like in there. Speaking as a
prescriptivist, I've always had a problem with <creation> (as, I suppose, have
some biblical scholars). I would rather that the element, instead of, in
effect, being allowed to contain a single paragraph of rambling, were more
structured. If you want to impose more structure on your particular
<creation>, the DTD customization allowed by the Recommendations allows you to
do so: make your own element, say, <creationStruct>, which has a structured
content model which is also allowable under the unstructured model of
%phrase.seq, and which has a TEIform attribute with the default value of
"creation". (The Guidelines also allow you to mangle the content model to a
greater extent, but that's impolite and unnecessary in this case. The change I
suggest would allow others to parse your documents against the standard TEI DTD
with the single addition (for example in the DTD subset)
 
<!entity % n.creation 'creationStruct' >
 
> 2. We are using <milestone> to mark traditional chapter/verse boundaries,
> reserving <div> for marking pages, columns and lines. But what element do
> we make the text at the lowest level? <div> can't take character data and
> <p> doesn't seem appropriate.
 
I suspect you need <s> here. Its description, "a sentence-like division of a
text", is intentionally vague: it's up to you to decide what is
"sentence-like". For example, in the British National Corpus, an <s> is
anything that automatic parsing software said was an <s>. In the
conventionally punctuated prose, <s>s correspond closely to what most would
regard as sentences; in verse, no <s> is longer than a verse line, even if the
line is not a complete semantic unit (or whatever); and in the absence of cues
(for example, in tabular material), <s>s have rather arbitrary boundaries.
Buts that's all fine: the documentation in the BNC's header <segmentation>
element says that this is what's happening, and that's all that needs to be
done. So, define what constitutes a lowest-level element, document it in
<segmentation>, and mark it with <s>.
 
Maybe you already have <s> (which can't be nested) ear-marked for some other
function. If so, you could use <p> for lowest level elements. If you think
that its name carries too much semantic baggage, give it another name. Again,
the Guidelines allow you to do this. (It's all in chapter 29.) Alternatively,
consider using <phr> instead of <s>, freeing up <s> for those lowest-level
elements.
 
> 3. How do I avoid getting "reference to nonexistent ID" errors in nsgmls
> for all the resp attributes?
 
This should not be: the type of the RESP attribute is CDATA, and so should not
excite "reference to nonexistent ID" errors in sgmls (or any other parser).
Are you certain that it is the value of the RESP attributes that is causing the
problem? It could be something else on the same source line, for example. If
you have another validator to hand, its error messages might throw more light.
 
> I noticed I could avoid this error for hand
> attributes by including an id attribute in the <hand>s in the header. In
> this latter case I made the id attribute the same as the hand attribute.
 
This is a requirement of SGML: for every specified attribute of type IDREF,
there must be somewhere in the same document instance an attribute having the
same value, and a type of ID. But, as I say, RESP does not have a type of
IDREF, so the issue should not be arising.