The last example in my previous message was incorrect and should have
looked like this:
version 1: <p>This is a dummy sentence.</p>
<p>Das Ziel des Lebens ist der Tod.</p>
version 2: <p>This is a dummy sentence. Das Ziel des Lebens ist der Tod.</p>
<p id="p0.0">This is a dummy sentence.</p>
<p id="p1.0" next="p1.1"/>
<p id="p1.1" prev="p1.0">Das Ziel des Lebens
ist der Tod.</p>
(This method is inspired by the fragmentation examples in the TEI
Guidelines, ch. 31.) The idea is a) in the lemma text, to disregard
the end of paragraph p1.0 (because it has a value for the
next-attribute) and b) for all versions, to disregard the beginning of
paragraph p1.1 (because it has a value for the prev-attribute). Only
in version 1 a new paragraph starts after the first sentence, because
the lem-element contains a p-element without a prev-attribute.
(One could even add a p0.1 in the beginning of the lemma and give p0.0
a next-attribute with value p0.1, but that seems overkill.)
In answer to prof. O'Donnell, of course it's undesirable to collate
all typographical details. But sometimes some typographical features
in a specific author's work are significant and should be collated.
The question is which encoding to use in those cases.
Bert Van Elsacker