“The first thing we do is, let’s kill all the translators”

Politics, AI and the "last-mile problem" of getting Japanese content to audiences abroad

Localization is the art of adapting content for foreign audiences. Translators play a key role in this process. And lately, they can’t seem to catch a break. A small but vocal contingent of angry fans want them fired for their supposed liberal biases. (You probably never imagined Fox News taking a stand on Japanese translation, yet here we are.) And the entertainment industry that depends upon the skills of translators to reach paying customers looks to replace them at every turn. Thus the title of this essay, paraphrased from Shakespeare’s quip in Henry VI about lawyers. The homicidal urge is metaphor, I hope, but the tensions are very real. How did we get here?

There is a huge and growing demand for Japanese content around the globe. But the vast majority of this content is primarily produced by and for Japanese audiences, with export a secondary concern. This means the task of preparing it for foreign consumption must fall to specialists who translate and adapt content for local audiences: the localizers. Their position, standing astride cultures and between content makers and consumers, makes them inviting targets for corporate cost-cutters and culture warriors both.

I have spent most of my professional career in the field of localization. My partner Hiroko Yoda and I began working together in 1998, then co-founded a company that specializes in producing foreign-language versions of Japanese entertainment, mainly video games and manga. Over twenty five years, we’ve worked on some of the biggest franchises in the Japanese canon, either as an independent unit, part of a larger team, or managing a team we hired ourselves. The arc of our career represents almost the entirety of the modern J-E entertainment localization industry, modern being defined as “when a majority of content producers finally started taking translation seriously.” I’ve seen it all, from having to painstakingly explain the word “localization” to clients early on, to the word becoming industry commonsense, to it emerging as an unlikely lightning rod.

Japanese creators tend to shy away from revealing much about themselves personally, much less their political views. This studied vagueness makes it easy for consumers of all political stripes to identify with the content. It’s good business, and it’s also why you can see everyone from fighters for social justice to Neo-nazis rallying around the same Japanese cartoon characters. When a creator does tip their hand, either deliberately or inadvertently, fans who’ve projected their own politics upon the work can short-circuit. They don’t want to blame the creators, so they turn their ire on a more convenient target: the localizers, and more specifically the translators. Thus a formerly arcane art, practiced quietly by nerdy desk-bound bilinguals toiling in dark cubicles, has emerged as a live-wire issue in the outrage economy.

Among businesses that publish and distribute Japanese content in translation, translators are a live-wire for different reasons: not political, but economic. Translating all of those words takes a lot of time and money. This can lead managers to view translators less as collaborators or partners than burdens and expenses. This is particularly true in the manga publishing industry, where margins can be be razor-thin, and the costs of translating a title need to be carefully weighed against its sales potential. The switch from paper to digital distribution has helped bring more content to market that wouldn’t have justified printing costs in the all-paper era. But there’s no getting around those pesky translators, with their insatiable demands for living wages, sane working conditions, and bathroom breaks.

Now a pair of Japan-based startups are aiming to disrupt the world of localization with AI technology. One is Mantra, funded in part by manga-industry titan Shueisha, publisher of Shonen Jump. Their goal, according to a glowing Government of Japan report, is to “support the overall translation process” with AI tools. The other is Orange, which aims to eliminate translators from the localization process entirely. Instead, an AI will translate, and proofreaders and editors will clean up the output to make it presentable.

Anyone who has worked seriously with AI translations knows this process is akin to changing bedsheets after they’ve been shat upon. But the general argument of these startups seems to be that manga translators actually stand in the way of getting manga translated, because there are too few translators capable of doing it, and they charge too much for their work.

The concept seems to be resonating with investors. Mantra has gathered a reported 250 million yen in funding. And last week Orange announced that it had attracted a whopping three billion yen ($19 million USD) in seed money from a consortium that includes mega-publisher Shogakukan — not coincidentally the arch-rival of Shueisha. (If you needed any further evidence that we are living in a Gibsonian cyberpunk world, look no further than two megacorps waging a proxy war with rival AI models trained on competing manga.) All the while, that subset of fans who are furious at translators cheer this development on, buoyed in turns by schadenfreude and the fact that current AI systems often deliver decidedly illiberal results.

Despite the eye-popping sums, details remain scant, which makes it difficult to predict the actual utility of these systems. What AI models are these companies using? How are they being trained? What sorts of manga that weren’t being translated before are now going to be made profitable through this process? I am no Luddite when it comes to new technologies, even AI — I have been using computer-assisted translation tools since the beginning, and currently subscribe to ChatGPT. I understand the dream of unlocking vast reservoirs of untapped content. But is it a dream, or a siren song? Even if the technology works (a very, very big if, as you’ll soon see) nobody is asking the bigger question: how much of manga’s success in English is due to curation? Westerners currently enjoy the cream of the crop; will flooding the market with unsold leftovers satisfy, or nauseate?

Manga share superficial characteristics with fiction, games, and anime but offer a unique set of localization challenges that set them apart from these forms of content. They tend to have fewer words than novels or video games, but narration and dialogue are embedded in tiny word balloons. Many of those balloons are oriented vertically, which is great for Japanese, but virtually unusable for English text. Sound effects, few of which correspond to English-language onomatopoeia, are splashed across the page in hand-drawn scripts, more a part of the art than the exposition. And translations inevitably tend to run longer than their originals, so text has to be cut back down to fit in the word balloons, which are then meticulously lettered by hand. When even expert editing and lettering can’t “save” a line, the balloon has to be carefully re-drawn, which also necessitates re-drawing the art.

In times of old, when translators from Japanese into English were even rarer beasts than today, they enjoyed remuneration that would make the current crop of translators’ eyes water like those of an envious manga character. I’ve heard from insiders that back in the Eighties, publishers sometimes even offered to share royalties with translators. This would be utterly unthinkable today. Now it is far more common to pay page rates in the mid single-digits. In my own experience, flat fees for books are even more common, which can really hurt when a manga’s page count expands dramatically over the arc of a series, as it often does.

All of this makes localizing manga a painstaking process requiring the efforts of numerous specialists: not only the dreaded translators, but also proofreaders, editors, letterers, and sometimes artists. It is very much a team effort, and translators are not the ones with final say over the content. My company has translated a great many pages of manga. I’ve personally translated, proofread, and edited, and I’ve worked closely with letterers and artists. I cannot imagine AI replacing any human in this process, and should an AI emerge that can, we are going to have a lot bigger problems on our hands than localizing manga.

I became a localizer because I believed it was important and necessary work. There weren’t many manga available in translation when I was growing up in the Eighties, so my first inklings about the field came through games. I was thirteen in 1986 when I got my hands on a Nintendo Entertainment System. The quality and craftsmanship of the games struck me early on. It was obvious someone had put a lot of time and effort into making these things, yet the English often seemed silly, stilted, or outright garbled. Even as a teen, this bothered me. Not because I felt any particular love for the English language at that time, nor even because I felt inconvenienced or confused. I just thought that the games deserved better. This is the era that gave us classics like “A winner is you!” and “this guy are sick!” This latter example is from 1997’s Final Fantasy VII, a game Square spent the equivalent of $40 million USD making, and Sony spent $20 million marketing in the US — without anyone ever thinking to hire a proofreader.

The problem isn’t that there aren’t enough qualified translators. Look at this survey conducted by the Japan Foundation. In 2021, there were more than half a million people studying Japanese in North America, the British Isles, and Oceania. Of course, not every student is going to go on to become fluent in the language, let alone qualified to work as an entertainment translator, which is as much or more about English writing talent as it is about Japanese ability. But there are more English speakers studying Japanese now than ever before. The problem is that manga translation rates have plunged to the point that most experienced translators are shopping their skills to higher-paying fields (such as that of video game development, which interestingly doesn’t seem to be anywhere nearly as enthusiastic about AI translation.)

Despite it all, I cheer on entrepreneurs and I love new technologies. Perhaps a better way of looking at things is that localization is the linguistic equivalent of the last-mile problem. When it comes to logistics, the most expensive part of the shipping process isn’t getting widgets from China to America, or from port to warehouse. It’s that last little bit from distribution center to the customer’s home. A similar effect is at work with localization of manga and other forms of content.

The expense of translating all those words represents a juicy target for tech solutions. But manga are made by people, for people, and people are always going to play the key role in bringing them to life for customers. Which leads to the perhaps biggest question of all: how far would $19 million go in training and paying more translators? With all this money seemingly there for the taking, surely there must be a way to budget for a human touch.