What's Not Yours Isn't Theirs, Either
Never has a first-party game been more honest and self-aware about its circumstances than Segagaga. A coda to the Dreamcast's troubled life released a few short months after Sega announced its withdrawal as a platform maker, Segagaga morbidly revels in the dire situation of the eponymous(-ish) company and its workers with the sort of bleak gallows humor you'd expect to find in, say, World War II movies about "the front." Only, instead of soldiers inside grimy trenches quipping about bad rations and dwindling ammo supplies, it's unkempt, heavily overworked developers all but shackled to their cubicles inside of dingy vaults, trapped from seeing the light of day as protagonist Taro wards off roaming bands of influenza.
This is no allegory. Segagaga has nary a romanticized bone in its gaunt body to be found. "Everything you've heard about Japanese game development and more is true, and nowhere is it more true than here at Sega at the turn of the millennium," the game morosely reiterates with every mission doled out, every NPC conversation, and every combat encounter. A certain undercurrent of cynicism permeates throughout that can only come from those who have truly lived the life they're depicting and witnessed how the digital sausage is made, exaggerated or not. Even so, cynicism by itself is rarely constructive and Segagaga isn't a game that takes the easy way out by glibly giving into it. What it actually is, in no uncertain terms, is a game about loving somebody at their worst, somebody who hits the bottom again and again, only to break through the floor and fall even deeper with each passing day.
A quarter of a century later, it's a game that I myself deeply relate to in some ways, having entered my 12th year in Japanese-to-English game localization this past April. The only game industry I know is the Japanese one and never is it for the faint of heart. I have taken on weighty workloads with inadvisable deadlines made viable only with the energy and single-minded dedication of youth. I have had the long nights on the job, staying up until the break of dawn, workshopping and refining the most minute details in pursuit of a translation that sings. Things where the raw dollar-and-cents return I was poised to earn for such minuscule amounts of material couldn't possibly justify the investment I put into them. I have had—and sometimes lost—the battles with my superiors, my agencies, and my clients to do right by their games and their audience. I have had projects cancelled on me after months of working onto them, their divulgence forever verboten by NDAs. I have lost contracts and promises of work and money even after winning it all fair and square because I just happened to rub the wrong person in power the wrong way a little too much for a little too long. And I have stuck it out in this industry doing what I do despite it All, despite the politics and the pay and the lack of job security, because I genuinely believed in what I was doing like nothing else I have done in my life. For the love of the game that is Japanese video games.
Segagaga languished for decades as one of those stubborn holdouts whose English translation just never seemed like it could come together, even as white whales that eluded fan translators for decades were tamed in the end. I would know better than most: I was once asked early in my career if I would join the ranks of one such project that was always on-again, off-again. From what I gather, even Sega itself explored ways to bring out an official localization, belated as it would've been. Yet it remained the exceptionally rare kind of fan translation target where no amount of desire, reverse engineering insights, and translator prowess ever proved to be enough to coalesce into a final, playable patch. Even Tokimeki Memorial, a similarly troublesome game that many romhackers had written off despite its significance, fell a few years ago at last and you can now play it in a language you probably understand... kind of. (Look, there's still work to be done and it's a matter of public record that I have some strong opinions on the version those of you reading this can actually play as of this writing.)
This state of affairs ostensibly changed a few months ago when an English patch at last emerged almost 25 years to the day of its original launch. That is, with a lede that was buried by both the team behind it and especially by many enthusiast sites promoting its release: the translation wasn't produced from scratch by humans from start to finish. Rather, the script was, in the project's own words on its GitHub page, "developed using a combination of DeepL and ChatGPT 4o/4.5. That translation then went through a substantial, months-long human translator review." In professional circles, we call such a process MTPE, or "machine translation post-editing." During it, the first pass of a translation is done by a machine translator like, indeed, DeepL, ChatGPT, or Google. It's then a human translator's job to review that output by comparing the generated translation against the source text, make any corrections where necessary, and deliver a polished final product, or at minimum something more presentable than the machine could've managed by itself.
At least, that's the elevator pitch that translation agencies routinely sell to spendthrift clients. I've already litigated the practical realities of MTPE on the ground on this very blog at length, so I won't repeat myself here. (Though, at the risk of adding another 2500 words of homework, I encourage you to go back and read it first if you aren't familiar with MTPE and its myriad shortcomings.) As anyone familiar with that post and my work in general can likely surmise, suffice it to say, the abrupt conclusion of Segagaga's protracted localization journey in such a manner was deeply demoralizing. While preliminary translation patches made with more rudimentary machine translators have, for better or worse, existed for a long time, the precipitous increase in LLM usage combined with MTPE as a supposed guardrail tells me that these teams are either insufficiently considering the practical and ethical implications of their methodology, or disregarding them outright. For my money, the fact this is happening more frequently with high profile games such as Segagaga suggests to me that it may well often be a case of the latter.
As much as I genuinely urge such teams to rescind their work and take down any patches, even after the fact, I also recognize that the cat is unfortunately out of the bag and many more cats will probably be let out of just as many bags so long as access to LLM models remains affordable, something which itself may not remain true for long. Be that as it may, speaking as a former fan translator myself, LLMs and MTPE present further problems specific to fan translation that merit discussion in addition to the broader issues I've already explored at length in my previous post linked above. At the root of it all is an important question that I want to ask. Namely, what does translation do to the translated work and for it?
I don't mean what a translation, when done properly, accomplishes on a mechanical level. You know what my basic job description is. Nor am I here to debate the supposed merits of translation versus localization writ large. (It's a false dichotomy, even outside creative mediums.) What I'm really asking is, what happens to a work once it's been translated, especially for the first time? In entertainment media and video games in particular, where translations are often the most time and labor-intensive to realize, most works make the jump to another given language only once. Barring exceptional situations, in the vast majority of cases, the first translation into that language is the only one that work ever receives, officially sanctioned or otherwise. In turn, it becomes the de facto way that audience can and will ever engage with that piece, the very grammar and semantics available to that language informing how people observe even objective realities of that content.
In other words, when you're translating somebody else's creation and especially when that translation is taking place in your chosen language for the very first time, you're not only speaking on another creator's behalf. As I've said many times, translation is an act that inherently requires that you put words into people's mouths. Not only the fictional mouths of the characters you're borrowing, but also those of the real people behind them. To a large extent, you're defining nothing less than that work's very legacy and that of its creators in abstentia. That new audience's first and quite likely only impression of it will be profoundly dictated by the choices you make throughout that process, the ensuing conversations being ones in which those creators likely cannot participate.
That is a heavy, heavy responsibility. I say this not to discourage other translators from making bold, creative choices in their approach. On the contrary, the job often demands it and no amount of attempted deferment to the supposed objectivity of translation dictionaries (or, indeed, LLMs and machine translators) will lighten the weight of that burden. Rather, I say this to stress what is always at stake when you decide to sit down and attempt to represent the work and mindset of somebody else with your own words. It may well be the only shot that media ever gets at reaching people within your sphere, making it imperative to aim both well and true.
More than other fields of translation, games and other entertainment exhibit what I've come to term as "self-evident context." That is, as bespoke creations, the words and ideas contained within them aren't strictly confined to our own reality. They can and often do have their own underlying histories and etymology informing their usage within that world. Any respectable translation concerned with accuracy must take those differences into consideration, as well as how audiences in the source and target languages alike each perceive those word choices. Something in mecha fiction might in its native language, for instance, call its take on giant robots "dolls" because of their lack of autonomy. It's then up to the translator to determine if a direct translation of the word "doll" adequately maintains the intended metaphor, or if they need to adopt a superficially different term that can nevertheless evoke similar ideas.
This is what makes the use of machine translators of any sort insidiously dangerous to deploy on creative projects. The algorithms underpinning them by their very nature force them to bias potential translations based on likelihood, which they in turn ultimately determine by their average frequency in A:B comparisons, as well as potential proximity to other, related words. While they may sometimes happen to be incidentally fed fictional material as a matter of course, they have no means of discerning it as uniquely fiction in contrast with the rest of its archived material derived from the real world. All of it is simply lumped together as "language" in a murky stew comprised of infinite ingredients in varying proportions. This makes the systems mathematically incapable of weighing potential translation candidates rooted in fiction more heavily than anything else they can unearth when tasked with translating fiction. Doing so would essentially require them to be fed more fictional sources of material than real ones such that the probabilities and averages arrived at favor the former, which is inconceivable.
When considering that diligent translations also must bear in mind the holistic, big picture needs of the source's themes and narrative, not simply moment-to-moment, or even line-by-line semantic parity, machine translators and LLMs quickly turn from a time-saving shortcut to an outright liability. After all, if ChatGPT and other AI models can't be expected to genuinely remember past interactions with them to any reliable, meaningful degree, why should they be trusted to have the reading comprehension skills necessary to consume a script in its entirety and then make judgment calls about terminology, character speaking styles, and many other pressing variables? They simply can't, and if you can't trust them to maintain a working memory of what they're translating as they "translate" it, then it turns out that the better, quicker, and safer option is always, in fact, to turn to a knowledgeable, capable human who can from the outset.
But the concerns about LLMs and MTPEs in fan translation extend well past the immediate practicalities and responsibilities of the work itself. Any discussion of them must also grapple with the reality of how those initial translations are generated. When you as a fan translator take that source material and paste it into something like ChatGPT, you're not merely receiving a one-time output before the technology purges itself of your query. You're feeding the work of creators who toiled months, if not years, into a system that will ingest it, permanently retain it, and disperse it to other users globally without those creators' permission for as long as that system exists. That material, which was never yours and will never belong to you, is, by extension, not yours to effectively donate to the (aspiring) profit engines of Sam Altman and his ilk. While permission in general might be relative when it comes to fan translation, what is absolutely indisputable is who owns the work being built upon and it certainly isn't the fan translation teams engaging in this behavior.
Having almost certainly seen my work fed into generative AI by at least one renowned client by my count, I can say that I don't spend those long nights playing Japanese builds of games (sometimes at breakneck pace for prolonged periods), writing dialogue, and troubleshooting problems with my colleagues just to have those same clients turn around and ask a computer to produce a limp facsimile of my writing in a brazen investment grab. Is that client legally allowed to do so because they own the copyright to my translations? Yes, but it doesn't change the fact that I put in the time, effort, and quality that I do for the games themselves, for their developers, and for their players, not for a technology whose stated end goal is to make me redundant in the eyes of employers. And while I can't speak for the teams behind the games now receiving MTPE-based patches, I can't imagine they crunched and made sacrifices in their lives during development just to have their work served up on a platter to an objectifying, averaging, muddying algorithm, either, least of all those whose games have waited decades to make their international break. None of those individuals slept under their desks or were hospitalized so that a small group of people on the Internet many thousands of miles away could enjoy a hit of clout at their expense for but a fleeting moment.
For many game developers and other creatives in Japan, the extent of their impact on the world such as they see and feel it doesn't extend beyond the archipelago. Even acclaimed, globally celebrated individuals can go their entire careers and enter retirement feeling wholly anonymous, never even imagining an audience spanning oceans and continents. It may not be until a foreign fan reaches them, and especially in their own language, that they understand their work did travel abroad and make a difference well beyond their homeland. In my own experience, "white foreigner with a penchant for retro dating sims that time forgot" can make for a surprisingly quick and potent ice-breaker when introducing yourself in the right circles, it turns out.
As a translator who will never be contractually allowed to reveal the bulk of my work with the world, it's this kind of life spent in obscurity, one that has a way of minimizing your self-worth from a lack of recognition, that causes me to feel a deep, knowing sadness when I see yet another fan effort needlessly succumb to the siren calls of MTPE-tinted machine translations. Beyond the practicality of the whole exercise, beyond the ethics and morals of churning through material in that particular way, is a group of people whose efforts made and prices paid we'll never know.
Were they taken care of?
Were they taken advantage of?
Were they bled dry?
Did they watch life pass them by?
Were they happy to do it all anyway?
Or was it all they could do to get their pay?
We'll never know. All I do know is, when what remains of that is taken out of human hands, no matter for how long or how little, they and their legacies are condemned to their own vault, their chance of escape vanishingly little. Only a person can make a translation that respects and reveres the humanity imparted upon this earth. Either you do this for the love of the games themselves and the people behind them, or for the love of yourself. There is no line to be drawn down the middle.