The quality of translations is proportional to the quality of your original strings. You can ensure higher accuracy of translations by adding contextual information, but the original string itself is the source of the translation. If you put the same care into constructing your original strings as you do into your coding, you will have effective and intuitive translated software.
Just as you always read over a line of code before running it, always read over a string before using it. That single check will save you a lot of errors. Spellchecking is also useful in catching typos. You can spellcheck PO files using pofilter.
You can also receive reports from translators who have spotted typo or other errors in your original strings. Translators are a useful resource for this kind of error.
Once you make typo or grammar fixes in your original strings, unless you have actually changed the meaning of those strings, you want to unfuzzy them. Strings are automatically marked fuzzy on update when the original text has changed. This is useful when updating translations, but in the case of typo/grammar fixes, it means that every translator for every language will have to check every string you’ve fixed. This is a huge waste of resources, and thus a very unpopular situation with translators. Imagine having to check every line of code because someone had fixed the spelling in a comment.
So, good i18n practice is to unfuzzy strings after changes that don’t alter their meaning (e.g. typo/grammar fixes). How do we do that?
Oddly enough, since every project does this regularly, there doesn’t seem to be a quick and effective method. All we’re really doing is diffing the changed files with the previous files, then removing all fuzzy tags which have only appeared in the changed files. So it could be done with pogrep: if you work it out, please add the command here.
Meanwhile, the Debian project has its own method, and there are a couple of useful items in the gettext manual (Fuzzy Entries PO Mode Index).
Nicolas François adds:
In some specific cases, i.e. when all the PO files fully translated before the typos are fixed, msgattrib (gettext) is very fast and very safe.
The po4a package has a tool, msguntypot, which tries to deal with the generic case, but the process to use it (8 steps) is quite complicated and thus error-prone. msguntypot is not yet well-tested either, and all files should be backed-up before using it.
Unless you need to use msguntypot, I would recommend unfuzzying only complete translations and using msgattrib.
Friedel Wolff adds:
If you want to unfuzzy all fuzzies in a single file, this should work:
msgattrib --clear-fuzzy input.po -o output.po
If you want to work recursively on many files, or want to select the messages on which to do the unfuzzying, using pofilter/pogrep with pomerge (all Translate Toolkit programs) might be more powerful.