In the Latvian l10n community we often deal with the problem of mistranslation of common words such as "file", "preferences" and "advanced". There are several words that fit each of those and over the years preferences in the community have shifted back and forth.
I just came up with an idea of making a script that would download all the translations, search for these common words in the source strings and see if any of the common translations and mistranslations are in the translation for that string. Then it could form up a web page with statistics of the mistranslations and possibly even links to specific instances to allow for corrections. It would also find new mistranslations :)
How other translation teams handle this problem?