On the left there are three windows with translated, current, an next
sentences from the source text.  Central windows are for translated and
current sentences, and the bottom central window is for current word.
The right window is for statistics, and (not shown here) Wordnet
(/usr/bin/wn) lookup.

The idea is to have some words (in bold in the sreenshot) that are
controlled, so that while translating them you can keep trace of all
other occurencies and prior translations.  So every word in the source
material need to be indexed and referenced to a (possibly empty) word in
the ongoing translation.

Work happens in the very central frame, where words are presented
untranslated at first, and you can move them around or substitute them
with prior or new (including empty) translations.  After a while, it
gets fast.

I am attaching the code.  Most of it is a painful and messy tratment of
the publisher markup, and all of it is intended for personal use and for
the particular book I was translating.  But maybe you can adapt some of
it to your needs.  Regards,

  Giovanni