So I did the first translation (I had to use Word 2000 since 2003 is apparently incompatible with Trados 5.5).
True (you need T6.5.x to support Office System 2003), but probably not relevant here.
A couple weeks later my client asked me to translate a revised version of the original document. Instead of using "track changes" (which I believe is not compatible with trados?), new text was underlined and deleted text was done in "strikethrough."
I wonder why they should do that? The very purpose of using a CAT tool is to be able to analyse exactly where something has been changed.
My client ran an analysis of the document before sending it to me which showed that the material was about 79% new. When I did my translation, however, that number seemed way off - more like 25% new, 75% repeats.
Which could well be due to a misinterpretation of segments that were 'shown as deleted' (as opposed to really deleted). Text shown using 'strikethrough' is still there and comes up as a match, albeit with a penalty for the formatting difference (depending on your settings).
I ran my own analysis after translation, but perhaps I should have run the analysis at an earlier stage (before the translation memory was updated??), because my analysis now shows 100% match.
You must analyse before translating: afterwards - as you found out - the entire text will be 'known'.
Or perhaps all the formatting changes (underline, strikethrough) have affected the count?
Not the changes per se, but the formatting (as opposed to deletion) is bound to have an effect.
Is there a way to use this to create a translation memory - which I could then use to try one more time to analyze the new file?
Yes - simply use a copy of the pre-cleanup file, and clean it up, making sure that the Update TM option is activated.
In the case at hand, try to get a clean file of the updated version, which you can then analyse. You can also directly compare the two versions (provided they do not contain revision marks) - I explained how to do this in an earlier post.