Grateful thanks to all who responded to my enquiry on this subject. I am
encouraged to persevere.
Several people advised to do the conversion as a series of small steps, and
I will keep that in mind. However, I am developing my conversion on a
non-final version of a part of the dictionary (the letter D), so it must be
re-runnable on the other letters, as well as on the final version of D!
I will see how far I can get with a simple series of "template match"
operations, before I think about anything more advanced. Whatever can't be
done that way may be feasible manually; and in any case I expect to do
manual tidying-up of odd cases and data errors which would not be covered
by a programmed conversion.
Although there are plenty of webpages about XSLT, it is difficult to find
the information you want. For example, I spent most of a day putting this
together:
<xsl:template match="my_element[@font-weight='bold' and
contains(preceding-sibling::node()[1], '=')]">
But I suppose there are a few things in it that I will not need to learn a
second time.
There were some interesting comments on the initial doc/rtf to xml
conversion. I tried several, and here are some figures, for the letter D.
.doc file 772 KB
Word save as rtf 711 KB
Word save as txt 286 KB
Word 2003 save as xml 4592 KB
Novosoft rtf-to-fo 2624 KB (http://www.rtf-to-xml.com/)
Walter rtf-to-xml 400 KB (http://thewalter.net/stef/software/rtfx/)
Walter rtf-to-xml with -p parameter 536 KB
Yawc doc-to-xml 433 KB (http://www.yawcpro.com/)
Unfortunately (for me) the Walter and Yawc conversions both discarded
small-caps info. I haven't noticed anything important discarded in the
Novosoft conversion, but it is still a lot smaller than the MS one.
Incidentally, the original Word files make no use of styles, but are all in
"normal".
Many thanks again,
Ciaran S Duibhmn