Signs of Localization Playing Across One’s Face

Hello all, and hope you are doing well. Today I thought I would comment further on a contentious topic that has been discussed a lot recently — localization. tl;dr: Don’t guesslate, translate faithfully, and if you do want to intentionally change something, have a clear reason to do so and do your due diligence first. Or in other words, like I’ve always said:

When in doubt, If It Ain’t Broke, Don’t Fix It.

Clearly 8-4 didn’t take this lesson to kor in the intervening decade since their previous debacle.

Background (press △ to skip)

Specifically, I’m of course talking about Unicorn Overlord. Someone did a comparison thread on TweetXElonZone demonstrating how the story was mistranslated/heavily localized. The responses were extremely curious. While there were some glaring mistranslations in the official localization, I didn’t think some of the examples were a big deal, and people would be well within their rights to hold the opinion that a lot of the examples shown were mostly normal and par for the course for localizations. Some people went quite a bit further than that, though:

1. Multiple people claimed that the thread starter doesn’t know Japanese and simply used machine translation. Any criticism he made would therefore be based on faulty information and thus invalid. This seemed very unlikely to me, but just in case, I double-checked, and Google and DeepL both translate it differently and get things wrong that he didn’t. It would be very impressive indeed if he somehow managed to hand-correct those errors while continuing to be a filthy monolingual “chud”. No one could offer the slightest evidence to support this accusation, even though I asked. Repeatedly. It seems like someone blindly made the accusation and then everyone believed them, likely because this explanation fit with their predetermined partisan sociopolitical worldview.
2. Some Australian linguistics professor claimed that extremely typical speech patterns for anime/games (omae, -te kure) are exceptionally rude (“abusive second person pronouns”, “aggro”) and thus it is proper translation/localization to totally change the lines to be much different, and ruder, than they literally are, to properly reflect this. In fact, he says, it’s mistranslation to not do so. This of course is absolute bullshit.
3. Related, the fetishization of Japanese, claiming it’s an ultra super deeply nuanced language and you can’t properly translate it without going deep down the rabbit hole with rich layers of meaning. This is roughly like saying you can’t translate the English “We took the bait and fell for it” into other languages without working in references to Louisianan fishing culture and Lucifer’s biblical fall from grace. No.
4. People claimed that the original script sucked, as evidenced by the basic translations provided by the thread starter, and the localization was thus an improvement. This misrepresents the purpose of providing a baseline translation. It should be as direct and unadorned as possible without missing the nuance and meaning, so you have something to compare to. A proper final translation should localize more than that, but that’s not the aim here.
5. Related, and completely opposite, some people claimed that you literally cannot do what that guy did, translating a line somewhat directly and comparing that against the localized line (while explaining any relevant surrounding context). You’d have to write an entire essay examining the surrounding context and evaluating if it works as a whole. Basically, there is no way to briefly objectively evaluate how accurate a translation is. What? Naturally, this would also mean you can’t actually judge the thread starter’s translation as sucking, whether for imagined MTL-induced inaccuracy or for dry writing.

Much like with Hearts R, I wonder if these people raising such bizarre objections are friends of the localization team (or team members themselves?), shills, trolls, or just genuinely massively ignorant. They seem unhealthily obsessed with starting with a certain conclusion in mind and then molding everything they see to fit that, rather the other way around, where you see things and then derive viewpoints and conclusions from your observations.

Luckily, another guy who clearly and unequivocally knows Japanese covered this issue with a stream/thread and lent another voice demonstrating that, no, the localization is just plain bad. I also wrote some tweet threads to address some of these points, especially #4: The original script was fine, and it was definitely hurt, not helped, by this poor localization. And even if we entirely ignore the original script and translation accuracy, the English writing is just bad on its own.

Analysis

Anyway, I bring all this up simply as background for what I want to talk about today. People can have different opinions, and they can defend the quality of the story localization as being based on personal preference, the expectations of the target audience, Japanese being super deep, or any other such nonsense to justify their position in the battle lines they have drawn. It’s a muddy and subjective topic. Guess what’s not, though? That’s right, system text localization!

One can wave one’s hands and say it all depends on point of view and there’s no possible objectivity all they want when it comes to dialogue, but when it comes to descriptions in the menus of how the game actually works, all that goes out the window. It should describe how the game works, full stop. If the Japanese did, and the English doesn’t, it’s mistranslation, and it’s impossible to argue otherwise. Let’s see how they did on that front, shall we? I’ll show a screenshot from the English localized game (from here), its transcription, the original Japanese (from the wiki), and the corrected translation of the Japanese. Or if you are feeling too lazy to read through the examples, you can Quick Jump to the end.

Defensive Curse Inflicts Phys. Attack -50%, Mag. Attack -50%, and Guard Seal on a row of enemies.	ディフェンスカース 敵一列の物理防御力-50%、魔法防御力-50% 対象にガード封じを付与する	Defense Curse Inflicts Phys. Defense -50%, Mag. Defense -50%, and Guard Seal on a row of enemies.
Official Localization	Japanese	Corrected Localization

Somehow, they have mistranslated Defense (防御) to Attack. I have no idea how one can even make this mistake in the first place, except perhaps flagrant carelessness with find and replace, or copy and paste, or maybe just straight up Japanese illiteracy. I would imagine such a mistake should have been caught in review, but apparently not.

Additionally, they also mistranslated the name of the ability to match. “Defensive” Curse indeed would make sense for a curse that would lower enemy Attack, like they mistranslated it as, but not for one that lowers enemy Defense. It’s like a game of Telephone; one mistranslation leads to another misguided localization change.

Aerial Shift Buff a row of allies. Grants +20% Attack and +40% Critical Rate to flying allies.	エリアルシフト 味方一列にバフを付与 飛行に対して攻撃力+20%、会心率+40%	Aerial Shift Buff a row of allies. Grants +20% Attack and +40% Critical Rate against flying [enemies].
Official Localization	Japanese	Corrected Localization

This one is so facepalm-worthy I can’t even. It doesn’t explicitly say enemies or allies, as you are expected to just infer the correct answer, as is common in Japanese. They failed, badly. If you don’t know Japanese, to give you a rough idea, imagine you are a translator from English to another language, and the text said, in English: “Vs. flying: Atk+20%, Crit+40%”. And you thought, “This means when MY flying allies are VERSUS enemies, their buff is active, right!?” It’s astoundingly, shockingly illiterate and incompetent.

Aerial Smite Attack a row of enemies. Cavalry targets cannot guard against this attack. Grants the user -1 AP if a target is at 100% HP.	エリアルスマイト 敵一列に攻撃する 対象が騎馬系の場合、この攻撃はガード不可 対象がHP100％の場合、AP	Aerial Smite Attack a row of enemies. Cavalry targets cannot guard against this attack. [Inflicts] -1 AP if a target is at 100% HP.
Official Localization	Japanese	Corrected Localization

Oh boy. Judging from the text on the Japanese wiki, in actuality, there likely was some kind of control code there in the text that would generate the special symbol and +/- value you see in the screenshot. Like the translators/editors would see “対象がHP100％の場合、AP<Whatever>” or something in their spreadsheet, and it wasn’t immediately clear what it meant, or whether it indicated gaining AP or losing AP, or how much AP. Ideally, the programmers in charge of the systems for handling text should give sufficient information to the translators/editors to know things like this, though depending on the circumstances, there could still be confusion. Also, the Japanese again doesn’t specify who is gaining or losing the AP, adding to the confusion.

But even though the translators were in Doubt, they didn’t bother to like, actually find out what the control code means, or how the skill works in-game, or familiarize themselves with the game’s mechanics so they would know what possibility might be more likely, and until then, make sure to Don’t Fix It. Instead, they decided they knew best, boldly made some random assumption about what it meant, and turned out to be the exact opposite of correct. It doesn’t grant the user AP, it reduces the target‘s AP.

You might be thinking, but Kaji, it was confusing, it was impossible to translate accurately without more knowledge, so you can’t blame them for this! Oh, yes I can. Imagine if they just wrote this instead:

If a target is at 100% HP, AP<Whatever>.

Boom. It doesn’t even matter what the control code meant, or how the skill works in-game, or anything. It is a direct translation that is 100% accurate in all cases.

Conclusion

And this is really the kor of the issue. This is what it means to accurately localize. Be careful, don’t make unnecessary assumptions, and stick as close as you can to the original meaning. Once you have fully grasped the meaning and how it fits into the larger context, you can possibly decide that certain things don’t work well or need a bit of extra flair in the localization, and you can have solid reasoning and justification for making those changes. Until then… Don’t! To change things willy-nilly at random with half-baked knowledge, like in Chesterton’s Fence, is irresponsible and leads to shitty results like this. Wow, yeah, the word “grants” sure sounds nice in that sentence… but it’s completely wrong. Great job.

As you can see from the above examples, they did mind-bogglingly atrocious work on the system text. You can’t handwave this away as personal preference; it’s just straight up mistranslated, and shockingly carelessly so. In cases where there was any ambiguity or Doubt, instead of just translating it straight and leaving it unchanged until they could confirm the true meaning, they went out of their way to (wrongly) overlocalize it anyway (i.e., failure to Don’t Fix It). Even our very bestest friends Guguru and FukaiEru could get these right — try it yourself if you don’t believe me! — and yet, the localizers missed the mark.

Gives you a lot of confidence that those complainers were wrong and the story localization was very definitely handled competently, doesn’t it?

Please, 8-4, you can surely do better than this. Take the time and effort and have the integrity to do it right. Remember that any other secondary localizations using the English localization as a base will suffer for your mistakes as well; from a brief glance at Reddit, it seems like FIGS got fumbled too. If you still need guidance, you may find my previous article on this very subject to be most useful.

Thanks again to a certain someone for bringing this facet of the localization to my attention, Duenan for his Reddit thread documenting the issue, Zakodgo and Poltfan for their comparison threads, and the Unicorn Overlord HyperWiki! Until next time!