The dependency-data from an automatical translation would be very useful for detecting, stopping and reversing vandalism.
Let me illustrate this point.
In Wikipedia when I make a change to an article it takes me some time, lots of concentration, Googleing, backtracking my subscribed RSS-feeds and consulting books which makes it very likely that I will put it on my Watchlist to see if someone axes my edits or what further info people input on the subject. I believe most people go about this Watchlist matter in the same way, which results in numerous eyeballs ready to catch vandalism, minor puns, POVs and so on.
In Wiktionary the contribution of adding a translation usually takes 10-30 seconds and when you get into the flow, you'll do these for half an hours straight and I have no interest to watch these articles (most likely thing to catch would be someone adding a translation in a language I don't have a clue about). Therefore it is much simpler to vandalise Wiktionary e.g. just change some translation to an obcenity and mark something else in the summary.
When utilising Wikipedia to get information, one can use common sense to filter out possibly unreliable information.
When utilising Wiktionary to get a translation, I'm really vulnerable to practical jokes, puns and obcenities whether human or bot created
This vulnerability (and the redundant manual work I've mentioned before) increases the chances that some people who feel almost religiously about the future of XML will fork a separate project from the Wiktionary to illustrate the power of meta-data and alleviate some headaches and frustrations
Comments, arguments and further thinking are very welcome.