I agree! It’s a lot of work to get a good localization.
It’s not like you get all the text in a simple text file, translate or improve/correct all the lines and then you’ll have a perfect localization.
How translations should be done in an optimal world:
(from my experience as lead tester - optimal world means with no time limit, with unlimited budget and good native speaking translator and localization testers available)
For the part of the game you want to translate (could be everything from a chapter up to the whole game) you’ve achieved text-freeze. That means for this part all the text relevant content is in and is basically working (logically not necessarily technically). This also means the puzzles because without them you don’t know if additional hints are needed.
A) Translation
The translator gets the localization kit. This lockit should be structured in a way, so that for every line some info is provided. The type (comment, conversation line, interface), speaking character, mood, conversation partner, current relation to the conversation partner. These info are very important for the translator and later on for the voice recordings. The lockit is upated during the whole localization process so that it’s up to date when the voice recording starts.
b) Proofreading
After the translation the translated content is checked for spelling, grammar, wording and continuity aspects either by a different translator or a speciliazed localization tester. This is done with the lockit.
c) Ingame testing
Finally the translated part of the game is checked ingame for textual and technical localization aspects by localization testers. Do the translations really fit in the context? Are there missing translations or switched lines (technical problems during the implementation of the updated lockit)? The lockit is also used to make sure that all lines are checked ingame. After a complete round the lockit is updated and implemented again and a new round of ingame testing starts. This happens until all is fine.
In the real world you have milestone plans, other budget restrictions and sometimes good native speaking translator/localization testers aren’t available. Or you suddenly encounter limits in the way the lockit has been build.
Typical examples:
1) It’s not clear from the lockit to which conversation a certain line belongs (added at the end of a list) or what the relationship between the conversation partners is. The translator is in trouble then. He can only guess. This way you easily get some continuity problems (e.g. Du/Sie, mood). I’m sure you’ve seen a lot of games with voice over that seems to be missing moods. It’s not always the actor doing a bad job. Problems with the lockit like described can also be a reason for it. Since if the actor don’t know the correct mood and/or relationship of the conversation partners it’s the safest way to be neutral.
2) You don’t have the budget or time to do a lot ingame testing. This way you get conversation lines that don’t really fit in the context. Or object/topics descriptions that don't make sense since there are often nouns that can stand for different objects/topics in the original language. And the translator can sometimes only guess what it means in the game. And sometimes he guesses wrong.
From what I’ve seen so far I guess that at least not much ingame testing have been done. And I think the easiest way to improve the translation is to do just that. Do ingame testing, note done the original translation and the improved translation for each faulty line. And then update the dictionary asset file and the respective texture files.
But before we fans start doing this I think we should wait and see what POS is up to.