Earlier this year, a game translation project which I had been trying to get off the ground for a very long time was surprise-announced by another team member, with a release implied to be imminent. I'm not going to mention the name of the project in the body of this post for Woolie Madden reasons—I don't think we should have tempted the Fates then, and I don't think we should any further—but if you've happened across this page, you know what it is. I had myself mentioned that I had already translated 90% of this game's Super Famicom script, a number the team member making the announcement quoted, and one which I stand by. Yet here we are several months later, and I'm still deep in work.
In short: the translation's the holdup. I'm informed that technical reasons make using the original SFAM game for a patch prohibitively difficult. However: the version of the game I understand we have to use has expanded greatly on that original SFAM game, adding a number of events and much more text—plus a second protagonist choice with mostly unique events and dialogue—while revising some of the existing script. Translating & editing all this new material, in addition to editing the entire script, old and new, for length and inserting it line-by-line, is unavoidably taking a good amount of time. While I've been presented with a method that attempts to create a script for the game using a combination of Google Translate, what I translated for the SFAM version, and guesses from non-translating members of the team, I strongly felt that the results did not deliver in terms of quality, accuracy, fluency, or characterization, so I asked, and was promised, complete responsibility for the script side of the project.
That's the reason for the delay. I, as the translator, am the rate-determining step. That's due to a decision I made for the quality of the project—I want to deliver a good patch, one that reflects the game & characters accurately and of which everyone can be proud—and the consequences of that decision are my responsibility. Given my current work rate, I'm going to shoot to complete the translation side of this project six months from now; emergencies or unexpected developments may change this, but I think it's doable.
That's the nutshell version, and I apologize to anyone whose hopes were gotten up by the announcement earlier in the year. There really isn't a way to short-circuit the time investment required for the translation, for reasons explained at length below. If you'd a more in-depth explanation, please proceed past the cut.
A bit of background
A bit of background: I've been a longtime fan of a romance-oriented Japanese game & media franchise aimed at women that founded an entire gaming genre but, except for one TV series, has never had any sort of official release outside Japan. There'd been some talk in the fandom of how nice it would be if the franchise's foundational, genre-defining game, for the Super Famicom, finally got an English patch—for historical purposes; to give the franchise a bit more recognition outside Japan; just so that the English-speaking fans could play. However: the game's script was notoriously huge, and no one wanted to climb that mountain. (The game's lovey-dovey, frilly subject material also doesn't typically appeal to the folks who fill up romhacking.)
I was exactly the type of unreasonable person equipped for this task, however. I've been involved in hobby game-translation projects for...well, decades now. I also translate professionally and have worked in game translation on projects for Square, Sega, Bandai Namco, Koei, and others, as chronicled in part on this blog. I also had one huge advantage for this game: the publisher had released a book containing almost its entire script. It took a great deal of time, but a few years ago, I finished and posted a translation of the script book.
Attracting someone to insert this script into a patch, though, proved a problem. One obstacle, I learned, was the immediate lack of a file pairing the English translations with the Japanese lines; I'd thought I'd attack this once a patch project started and I got proper script dump files, but not having a paired version of what had been translated immediately prepared was a turnoff, I guess. I therefore set to work producing one. In the meantime, I was indeed contacted by a few folks with some interest in producing a patch, but I didn't subsequently hear of any progress.
About a year ago, though, shortly after I started on that "paired" version of the script, I was contacted by a very interested and motivated individual who claimed they could use the script to produce a patch for the game. One thing, however: there were issues with the coding of the SFAM version that made producing a patch for it prohibitively difficult. (I know very little of the tech side of producing patches and am not equipped to provide an explanation, but this was not the first time I'd heard this about the SFAM game. I had persisted in translating the SFAM script in the hopes that the issue was resolvable, as the SFAM version is the historic, genre-founding title and the folks with reservations self-admittedly hadn't looked very deeply, but evidently, the problem is indeed a serious obstacle.) As a solution, this person asked if I'd be willing to work on a patch for a later, more-technically-manageable rerelease of the game on the DS. I readily agreed.
A while after, two more folks joined the project on the tech side, evidently with a solution to an issue that was frustrating progress on the DS version. They offered to produce a version of the patch for the PS1 version alongside the DS version; I agreed. (I eventually learned, however, that this would involve extra work, that of producing a second version of the script to accommodate the PS1 version's different per-line character limits.)
What exactly are you doing right now?
In brief: when agreeing to produce a version of the script for the rereleases, I didn't anticipate how much they expanded on the original game.
- While the original version has a single, set protagonist, the expanded versions allow you to choose to play as her rival. The choice isn't limited to a simple swapping out of names in dialogue; each protagonist has her own unique versions of events, plus unique dialogue from the nine love interests. (Some dialogue is shared, but about two-thirds of it, I'd say, is tailored to the choice of protagonist.)
- Dialogue from the love interests has been greatly expanded to flesh out characterization. For example, in the original game, if you asked one of the love interests his opinion on one of the others, he'd respond with a generic message of approval, disapproval, or indifference based on his current regard for the other man. In the remakes, however, each love interest has a unique set of approval/disapproval/indifference messages for each of the other eight suitors. In other words, where there were once three different messages, there are now twenty-four, each on average significantly longer than the originals.
- There are numerous new events and activities: cafe and skygazing dates, letter-writing, cooking, finding lost objects, etc., all with unique dialogue for each protagonist and love interest, making for a base of eighteen variations. Many have additional variations based on other parameters.
- Some of the dialogue that's still present in the remake has been revised from the original.
Again: what I had originally produced was a text translation of the script book to the Super Famicom version of the game. The script book was not complete—it focused on exhaustive enumerations of love interest dialogue and elided NPC dialogue—but I believe it has about 90% of the original game's text. With the remake, though, there's a lot more work to do.
My tasks are numerous. The script files I've been given generally sort dialogue by event. For each file, I have to identify the specific event they concern, then go line by line to ascertain the speaker of each line and the circumstances under which each is said from context (as this information isn't marked) as well as determine if the text is old or new. If it's old, I have to find the translation in the script-book file and match it to the source text; if it's new, I have to translate it—and also check it against a new script book for the new version, as certain changeable or user-definable proper nouns, such as protagonist names and various locations an event can take place, are for tech reasons not reliably represented in the script dump files. For all dialogue, I have to edit translations to match character and line limits, putting in line breaks and sometimes rewording text or, as a last resort, cutting it—and since the character limit per line is different for the PS1 and DS versions, I have to produce two versions. (In fairness, I don't _have_ to, but the extra space in the DS version is useful on occasion, and dialogue boxes might look odd if I don't use real estate properly.)
It's a lot. However: it's not that different than what I do as a professional game translator (with extra steps), and I'm uniquely equipped to tackle this particular job thanks to having translated the SFAM script book and being very familiar with the characters from following the franchise for years.
I can do all this, but it takes time. That's what holding up the patch.
No, AI can't help
Not long after the announcement, one of the team members informed me that they intended to produce the patch using a method where Google Translate versions of the script files were combined with parts of my original translation and guesses from non-translating members of the team as to what lines meant. They shared an example—an attempt at an English-language version of the introductory segment of the DS game produced using this method. (I had translated the SFAM introduction, but, as with all matters, the DS version greatly expanded on the SFAM version.) I had not been informed previously about this plan.
I took a look, and I felt very strongly that the results did not live up to the standard I wanted for this franchise's English-language debut. There were numerous errors in translation, spelling, grammar, and punctuation. The English used wasn't fluent. Characters no longer had unique voices—they all sounded the same, and characterization was completely absent.
(I have to say: despite the importance of the other issues, the gutting of characterization was the biggest knife for me. The game was so successful in pioneering its genre due in great part to its attention to characterization, which I worked hard to reflect in my translation. Seeing all that work torn up in favor of an AI translation that was not only not up to the task linguistically but torched everything that made the game unique frankly hurt.)
This isn't Google Translate's fault. The script files are typically formatted to present, line by line, just one side of a conversation in all its numerous variations—either for one love interest in all possible permutations (variants depending on your affinity with him etc.), or for all love interests who can have this conversation. There's no overt signal in the file that the speaker or scenario is shifting; you have to realize who's speaking and in what circumstances from context and speaking style. Google Translate can't glean the context it needs to deliver an effective translation when dialogue is presented in this manner, and you can't infuse it with the franchise or script knowledge it needs to make the right calls about speaker/scenario/etc.
Other obstacles: The dialogue is presented line by line, so Google Translate is going to translate line by line, which leads to choppy dialogue. Characterization is frequently communicated in elements like verb endings and sentence-ending particles that Google Translate overlooks or doesn't convert well or that can be translated in multiple ways depending on context. Conversations also have a number of short lines that are also heavily context-dependent, where trying to translate them context-free via the line-by-line method leads to inane translations. There are more; any experienced translator could probably expound further on why using Google Translate in this scenario is a bad idea. You need an experienced translator with franchise knowledge to tackle this project.
I also don't expect the folks on the technical side of the team to attempt to translate—it's not their job, any more than romhacking is my job.
I would have been derelict in my duty as a translator if I let the patch go forward in this state. I didn't, and don't, feel that a Google Translate patch would deliver any value—there are programs out there that can Google Translate dialogue boxes in roms automatically, if that's what you want. This is a historic game, and I feel it deserves a patch that showcases the qualities that made it special for its English-language debut. If we delivered a patch that was incoherent and characterless, the franchise would be tainted in the English market, and the team would be nailed to the wall. From my job, being the one equipped to gauge quality in this department, I knew what public reception of a patch in this state would be, and I didn't want the rest of the team or me to suffer that. (To be selfish, a poor-quality patch would also have professional repercussions for me, as pros do watch the fan scene, and trying to pass off AI, much less AI that's not up to the task, as your own work is, rightfully, a huge black mark on your reputation.) Also, to be blunt: I didn't start this project years ago to spit AI all over this game.
I sent them an email explaining the issues above to the relevant team member, and I asked to have full responsibility for the script side of the project. They agreed.
There was also an attempt by a team member to use AI to help with editing that didn't bear fruit. One of the non-translating team members tried to help with my workload by taking the lines I'd edited for the 38-character DS line length and trying to re-edit them for the 33-character PS1 line length with AI assistance. I was not initially informed of this difference in line lengths or the need for two scripts. I discovered it only upon accessing the group versions of the script files on the cloud drive and discovering that a number of formatting and translation errors had been introduced into files I'd previously edited, some rather basic: the Japanese verb for "to eat" had been mistaken for the one meaning "to rest," for instance. Upon inquiring, I discovered what had happened, and as a result, it was clarified that I would be handling editing for both versions and reworking the files affected to restore the translation and remove the errors produced. I appreciate the attempt to alleviate my workload, but there's by nature a hard linguistic wall involved with editing, and the attempt just created more work.
So the other team members have experimented with using AI, but it hasn't produced anything usable—honestly, it just kind of screws up everything it touches. (If I'd been aware of the attempts to use AI on the project beforehand, I would have reached out and explained why they wouldn't have worked.) I'll state right now that my vision for this patch absolutely does not, and never did, include Google Translate or other AI or LLMs: as a professional translator, I have no need to use AI to translate; it would not expedite the translation due to the particulars detailed above; and I care too much about the quality of this project to farm any part of it out to AI. I mention it because a lot of folks, particularly tech folks, have more confidence in AI than I do, having worked with it extensively for my job and knowing well its capabilities and limitations. It can be useful in certain situations, but it's not omnipotent, and it's not suited for this project.
"Hey, would it help to get an additional translator?"
Honestly, not at this point, no, for several reasons.
- One, the work to be done requires a great deal of previous familiarity with the script. Remember, the script for the roms exists in a bunch of files with the lines are generally (but not always) sorted by event, where the individual speakers and the nature of the event itself have to be inferred from context or just plain recognized from experience. The lines also aren't typically organized to present a complete conversation, usually going through all the iterations of one side of the conversation character by character. A good deal of the work is recognizing context and speakers for the lines and whether they're already-translated material or entirely new or old lines that have been changed, so that existing translations can be matched or new, context-appropriate translations can be written. You really need to know your way around the script to do that. I'm equipped to do that, since I've translated the script book for the SFAM version; that's a tall ask for someone coming in cold. We tried it previously, and it didn't work.
(It also helps if you have previous familiarity with the characters and how they speak, not only to identify when they're speaking, but to translate what they're saying in a character-appropriate manner. The latter isn't impossible without that familiarity—it's how games are typically translated—but it's way easier if you come in with that knowledge.) - Two, this is a character-intensive game with a huge amount of nuanced character writing to reflect in the English translation, and that calls for a certain degree of translation and just plain writing ability. Many of the devices used to communicate that nuance are easily overlooked unless you have a lot of translation experience. (I'm speaking from my own development as a translator here: you can reach a point where declarative sentences like the narration in a world guide or Ultimania are within your grasp but the nuances of conversation are just going to elude you.) Plus: even if you identify and understand the Japanese text, you still have to be able to phrase it in English fluently, in an accurate and character-appropriate manner. This whole set of skills is another ask.
- Three, this is a niche title for the West whose massive script was notorious for repelling translators. "Hey, guess what! I'm working on a version that I've discovered is even more massive and interminable!" does not make the project more enticing.
- Four, it's way harder to vet translators nowadays, and it would consume resources on my end I'd rather put toward getting the dang thing done. The rise of LLMs has given rise to a number of folks eager to take over translation duties on a project who have just one or two semesters of Japanese but think Google Translate can make up for the rest. (You might say, "hold on, you just mentioned that this project repels translators; why would someone like that sign up for this?", but that's kind of the thing: it'd attract those who don't know enough to know what they're getting into.) In a conversation- and context-heavy project like this, where AI is going to fall down hard, those people are going to hit a wall very quickly, which leads to difficult conversations that I'd really rather not have—but would have to have, since I'm the translator who started this whole thing, and it's my responsibility to deliver a quality product.
Furthermore, a lot of this can be difficult to communicate to the type of people who would be most enthusiastic about AI at the present moment. I love you, tech people. I love you. But I have gotten the impression over the years that some tech folks in the rom-patching scene regard translators and translations as kind of valueless and interchangeable, because, you know, words are all the same, who really cares. It's already an uphill battle to convince some folks that words, the type of words you use, matter, and I'm afraid my forces are already fully engaged on one front (the script itself).
Even if the volunteer were heaven-sent, due diligence in ascertaining their divine status would take a lot of energy, and after the stumbles we've had, I just want to get this done, and just want to rely on a method I know from experience will work.
Real-life issues! We've got real-life issues here!
I will also note, by way of explanation, that I have had a number of personal things happen that have resulted in delays on my end, which I will recite here for entertainment's sake: I got bronchitis twice in succession (I had it for a week, got better for a week, and then got it again, worse, for another week) and became as sick as I've been in my adult life. There's been an issue with a property in my family to which I've had to tend, involving lawyers extensively. I got an opportunity this year to work professionally on one of the most storied franchises in gaming, to which I had to dedicate a suitable amount of time, as well as other professional translation projects that demanded my attention. I've had to travel extensively for personal reasons, and not fun, vacation-type travel—travel that involved a good deal of work, physical and professional. I had fallout from a client who wanted me to cheat on my taxes to cover for an accounting issue on their end and submitted falsified documents to the IRS. Also, after a lifetime of avoiding, by luck or otherwise, such situations, I have had a problem with being harassed by not one but two completely unrelated individuals, one of whom became physically violent with me. (Not very successfully, but the attempt itself was disturbing.)
To paraphrase from Jurassic Park, I'm sure, dear reader, you are sorry for my problems, but they are my problems. But I recite this to communicate it's not been all Lazy Translator Fun Time over here, and at times, though my spirit has been willing to work on this translation, my schedule has been weak. Or too strong, rather. This doesn't make the delay go down any more easily, but it has been a practical concern.
So that's the status report. In brief: I'm asking for six months to get this done, AI-free, and I thank you for your patience (both with the delay and in reading this).
