On Thu, Dec 29, 2022 at 10:21:55AM +0100, Heinz Tuechler wrote:

[...]

> Agreed. I think that "allow certain nodes to only have one text element
> at the very beginning" is an arbitrary choice.

There are many arbitrary choices when designing a data structure to
represent text.

> Inconvenient, but logically cleaner would be to allow for text either
> only in leaves,

You'd regret that quickly: what do you do for inline markup (e.g.
bold, italics)? You'd have to introduce a special markup for
"normal" text. Good-bye "lightweight markup" :)

> or in any place of the nodes.

This is XML's "mixed content". Note that specific XMLs can and
do limit that, via DTDs or other validation restrictions (aka
schema). If you're doing text using some suitable XML schema,
you most probably won't be able to put a whole paragraph whithin
a section title. Or a table. If you want a whole italics stretch
straddling a paragraph boundary, you'll have to close it before
the paragraph's end and re-open it next paragraph or something
(although they might, logically or philosophically, belong
together).

Human language is too exciting to fit into a tree structure.

> The latter appears more natural to me, as it would allow for connecting
> words between sub sections and closing remarks at the end of a section.
> Usually, one would circumvent the problem by inserting a connecting text
> at the end of subsections, although this would offend the hierarchy.
> best regards, Heinz

You'll always end up offending some hierarchy some of the time.
In some places you can be sent to jail for that ;-)

No, seriously: Document models sometimes get stretched to data
description languages (I always cringe when people talk about
a JSON or YAML "document") and vice-versa. This is a strength,
but also a weakness.

And don't forget, OASIS's OpenDocument format also doesn't seem
to support "going back". Perhaps for a reason.

Cheers
-- 
t