in each Hebrew document, when typing some Hebrew words after the 1st time, OOo
suggests completey incurrect auto-complete for those words, adding letters that
where never in the original word.
Thus, if the user wants to enter them and still use the auto complete for other
words, he must finish typing the word while OOo displays a completly non-sese
word, and then go back and delete the characters that don't belong.

I went in the auto-correct settings, and the word list there looks ok, so I am
not sure where OOo is getting the wrong words from.

I am attaching two sample documents that show the problem, so you can repro it
yourself:

* rel.sxw: Try typing the 4th word in the document (going RTL) and/or the 15th
word in the first paragraph (going RTL, not counting the title).
* pkg_linux.sxw: Try typing the 2nd word in the document (going RTL) and/or the
3rd word in the document (going RTL) and/or the 5th word in the document (there
are more examples in the same doc, but I think the idea is pretty clear).