Machine translation has always been notorious for just being bad. It’s great if you’re playing around or lost in a country and need help, but when applied to professional translations the results are, more often than not, shockingly terrible. When applied to creative translation, the results are even worse.
Let’s take a look at what machine translation is and why it’s so bad for creative translation and the Japanese media localization industry as a whole.
What is Machine Translation?
If you don’t know, machine translation is when a machine translates text from one language into another without human intervention.
You put X sentence into the machine translation program, and it comes out in another language, but there’s no guarantee the translation you get is accurate. (You can’t tell unless you’re fluent in both languages!)
How a machine approaches a translation depends on how it’s been programed and the data it’s pulling from.
For example, Google Translate bases its results on publicly accessible documents which are available in multiple languages (such as UN documents and transcripts) as well as previous machine translations that have been corrected by humans. (This is why Google Translate tends to do an OK job for European languages but has poor results for non-European languages such as Japanese.)
Like any AI, how “smart” the machine translation is, is based on what data you feed it.
Machine translation is not to be confused with Computer Aided Translation (CAT) tools, which are software that aid translators to translate a lot of text in a more efficient manner.
CAT tools keep track of past translations (done by humans) and glossaries, so the translator can easily keep terminology and language styles consistent. These are particularly great for technical and legal documents which have set styles of language. Many game localization companies also use CAT tools too due to the sheer amount of text and terms many games have.
Machine Translation Helps Translators, Right?
CAT tools help. Machine translation hinders.
CAT tools are designed to improve quality by helping human translators keep terms and style consistent and improving translation efficiency.
Machine translations create more work, more hassle, and more headaches for translators. They don’t keep terms, styles, or tone consistent, and it’s hard to use them as references when the English often doesn’t make any sense. Ask any professional translator and they’ll tell you that trying to edit a machine translation is a nightmare, and creates poorer quality work and takes more time to fix.
What happens, more often than not, with machine translation is; the client throws the text through a machine translation, they pass it onto a translator or editor to “edit” (normally at a reduced pay rate), that person is given a context-less mess of broken English and now has two choices; 1) make it sounds somewhat like English and pray it works in context, or 2) compare it to the Japanese and original context, effectively re-translating (not editing) it to sound good and make sense.
Machine translation screws over either the translator/editor who has to spend extra time (that they’re not paid for) to fix the text so it makes sense, or the end-user who gets a poor quality product that they probably paid full price for.
The Machine Translation of Media
Machine translation is a tool that can be used in some fields such as technical translation. (Although note that even these industries mostly use CAT tools which are based on carefully curated translation memories of human translations.) This is because something like technical texts tend to have almost exactly the same wording and sentence structures. The point of these standardized documents is to convey important information to the end-user.
Creative translation is, by definition, creative. Media is there to entertain and inspire. It’s heavily rooted in culture and often contains many levels of nuances that the AI in machine translation software can’t pick up and then translate into an appropriate equivalence.
For example…
Original Japanese (人間失格 “No Longer Human” by Osamu Dazai)
Note: https://www.aozora.gr.jp/ is a collection of classic Japanese literature that are no longer under copyright.
「可愛い坊ちゃんですね」
といい加減なお世辞を言っても、まんざら空お世辞に聞えないくらいの、謂わば通俗の「可愛らしさ」みたいな影もその子供の笑顔に無いわけではない
Direct Translation (J-EN Translations)
“What a cute boy.”
Even if you say this lukewarm compliment you do not hear a completely empty compliment, the shadow of something like so-called common “cuteness” isn’t not present on the child’s smiling face.
Machine Translation (Google)
“It’s a cute boy.”
Even if you say a sloppy flattery, it’s not that the child’s smile doesn’t have a shadow like the so-called “cuteness” that you can’t even hear from the sky.
Creative Translation (J-EN Translations)
Even a half-hearted compliment like, “What a cute little boy!” can ring somewhat true as a shadow of what is widely considered ‘cute’ graces a child’s smiling face.
As you can see, machine translation doesn’t convey the meaning or feel of the original work well and doesn’t make sense in some parts. (Even a direct translation comes off stilted and strange in English.)
Machine translation doesn’t work well for a language like Japanese, not just because sentence structures are different, but because the cultural expectations of language and culture are so different from English. In the above example the machine translation struggles with the use of archaic kanji. But in popular media like manga and games, they struggle with slang, accents, cultural references, jokes, etc.
You need a human to judge the most appropriate translation for the type of story and characters, as well as what the target audience would enjoy the most.
Machine Translating Japanese Media
Where do I start with the can of worms that is machine translation in Japanese media?
Let’s start with the machine translations themselves.
Remember I mentioned how an AI is as good as the data you feed it?
Google is OK but it’s not great, especially when it comes to Japanese. Particularly so when you put in casual Japanese because it’s trying to pull from publicly available translations which are normally technical, not everyday Japanese.
Just take a look at Legend of Localization’s experiment machine translating Final Fantasy IV. Clyde Mandelin decided to see what would happen if he ran this game through Google Translate, and recorded the results on his website, as well as in a book. A hilarious read on a Saturday afternoon, but can you imagine paying $30+ for the official release of the game and getting that!?
Is there an alternative?
So what if you don’t use Google and instead create a machine translation that’s only been fed creative translations? That should work…right?
In response to that I would like to ask, what publicly available creative translations are there? Practically none due to who owns the rights of the original text as well as the translated text.
You can’t use any translation to feed a corpus (database of texts used for machine translation.) Translations are owned by the translator or by the person who licensed the translation (which is most common in creative translation). Using texts and translations you don’t own is illegal.
What creative texts are “available” are fan translations. (With large air quotes because the use of fan translations for commercial purposes is also legally and morally questionable.)
Fan translations are unofficial translations, often done by aspiring translators or bilingual members of certain fandom communities. These translations are often done by fans for a small group of fans. (Sadly these are often taken by aggregator sites and distributed as pirate translations, but that’s a whole other issue.)
The main point I want to make is fan translations are different from official translations. They tend to have less quality control and stick closer to Japanese sentence structures, resulting in unnatural and awkward English. (This isn’t always the case but happens a lot of the time.)
Another option is to hire translators to translate text that will be fed into the machine translation. However, no self-respecting translator would do this. The rates are insultingly low to create a product that will undermine the translator’s livelihood in the future. Which means that inexperienced translators or hobbyists might take these low-paying jobs. Which, as mentioned, doesn’t guarantee quality.
In both of these cases, feeding awkward translations into a machine translation database means the machine will spit out awkward translations.
Machine Translation in the Video Games and Manga Industries
There are two sections within the Japanese media localization industry where machine translation is notorious, games and manga. (And machine translation is, unfortunately, on the rise in these industries.)
Video Games
In video games you have some localization agencies who are trying to increase their profit margins by implementing the use of machine translation and hiring “bilingual editors” to “edit” the translation. (See above why this doesn’t work!)
I’ve even heard of professional translators misusing machine translation to pump out ridiculous amounts of text then leaving them to the poor editors to fix. (This is morally wrong on so many levels, including being paid for a quality product and not delivering on that promise.)
Manga
In the manga industry there are two infamous companies who are pushing to use machine translation. One is Mantra, which is run by a group of young graduates who developed a machine translation software during college. That’s fine, but now they’re trying to market it as a legitimate translation and lettering option for Japanese publishers, even though it’s clear the results are low quality.
In the above example you can see how the sentence is too long to fit the bubble, making it overflow. The English is understandable but awkward with unnecessary repetition. A better translation would be something like “You’re still a doctor… The patient doesn’t care if you’re inexperienced.”
The other issue that this company are also advertising machine lettering. Lettering is literally art. The original manga artist put work into the SFX, fonts, and typesetting. It feels like a giant middle finger to the original artist when the translated version looks like this:
The other is INKR, a Vietnamese company previously known as the infamous MangaRock.
MangaRock was a manga pirating application and aggregator who stole fan scanlations and official translations then charged users monthly subscriptions to read them.
INKR draws in $3.1M to make more comics accessible to worldwide audiences
As mentioned above, machine translations don’t help translators deliver accurate translations more quickly…
Machine Translating Japanese Media is a BAD Idea
The push to use machine translation in Japanese media isn’t for the benefit of the translators, the original creators, the target audience, or the industry as a whole. It’s for the benefit of lining the pockets of a few people in the drive to undercut competition and increase their profit margins.
This kind of competition is bad for the entertainment industry as a whole because it drives down the quality of entertainment localization as people cut more and more corners to save money. Which means you, the person buying the products, end up paying full price for poorer quality products.
It’s also incredibly difficult to increase the standard after this as the original IP and license holders often want the cheapest localization possible.
And spoiler alert, you can’t have cheap localization that’s also good quality!