Ah, what Bryan Cholfin has identified is a good point, that what you
really want in a word processor is that it be able to convert its
internal format into an SGML ASCII output for interchange to
another word processor. That is, it should be able to display
multiple fonts, underlining, superscripts, etc. but if you
ask it to write the data out in plain ASCII it should also
be able to put into the ASCII a translation of its own
codes into SGML tags.
This is I believe the point behind several emerging software systems.
I would contend that rather than be concerned about what today's
word processors are doing, we should be looking forward to better
word processors that can speak, decode and encode in SGML.
It goes back to pointing out why standard for SGML tags are essential.
I.e. if you use <para>, I use <graf> and someone else uses <pgf>
there will be endless conversions even between SGML+ASCII. Thus
enters the TEI by making some arbitrary decisions to call the tag
one name, rather than 3 or 30.