Forum - View topicINTEREST: Manga Localization Company Amimaru Allegedly Trials Machine Translation For Commercial Man
|
Goto page Previous 1, 2, 3, 4, 5 Next Note: this is the discussion thread for this article |
| Author | Message | |||
|---|---|---|---|---|
garfield15
 Posts: 1541 |
|
|||
|
Until it can get names and proper nouns right without shitting itself, MTL for Japanese will never be viable
|
||||
|
||||
Suxinn
 Posts: 251 |
|
|||
|
Hoo boy, remember when Tokyopop ended up machine translating several volumes of manga before the manga bubble burst? And how, when other publishers later license rescued some of their series, they had to go on record to say, "Yes, we are going to retranslate this completely because the previous translations are terrible"? This reminds me of that.
I feel like those advocating for machine translators are 1) people who don't actually know a second language (and especially not a second language that is so vastly dissimilar from their first) or 2) greedy companies trying to earn a quick buck by underpaying their translators as much as possible. And, from what I can tell, this issue isn't unique to a manga industry. I've seen plenty of translators being signed on as "editors" for translations that turned out to be churned out by a machine -- and according to them, they've had to basically retranslate from scratch in order to put out a good product, for far less pay because they're purportedly only "editors". (Most translator forums will encourage people never to accept jobs like these.) As someone who's fluent in two languages (Chinese and English), I mentioned in a separate thread that all machine translations, including the fan-beloved DeepL, can't match up to a single human translator with a competent grasp of the language, and that somehow got turned into a debate over "idiomatic" translation. But, no, that wasn't what I was referring to; I was referring solely to plain accuracy. Most machine translations still get very, very basic words wrong that even a beginning learner in the language wouldn't. This is because they are designed for the most common meanings rather than the most accurate. An example I always use is the Chinese word 穿 which is most commonly used in the context of "wearing clothes". So when a machine translation can't recognize a phrase with that word in it, it defaults the meaning of 穿 to "wear". But that's not actually what the word means in Chinese. At the very basic level, 穿 means to "pass through", so literally 穿衣服 actually means you're passing your body through your clothes. But of course no one would translate it like that, since English has its own dedicated word for "wear". (This is also why, while you can say "wear a hat", it's more idiomatic in Chinese to say 戴 (put on) a hat, since you're not passing through anything. Ditto for glasses, which, again, doesn't pass through anything.) And, of course, all other instances of 穿, such as 穿過, 穿到, etc. uses the "passing through" definition, but these are much less common phrases than the idea of "wearing your clothes", so this incorrect definition sticks in machine translations, because machines don't know any of this. Can an AI maybe, eventually, learn all these language rules that seem second nature to humans? Sure, maybe, but at that point, they'd need to be a pretty full-featured AI whose way of thinking has to be functionally similar to humans. This is not a "in a matter of a few years" thing. And, yes, I'd expect that kind of AI to be prohibitively expensive even once invented, under our current capitalist system, so even if it was possible, I'd expect these kinds of companies to still default to terrible, free machine translators in order to maximize profits. The issue here isn't about machines vs. humans; it's about companies trying to screw over its employees and customers in order to turn a quick buck. |
||||
|
||||
|
FilthyCasual
Posts: 2438 |
|
|||
Your comparison is farcical beyond belief. |
||||
|
||||
|
Tenebrae
Posts: 494 |
|
|||
|
Well, those translations will at least be very easy to detect. I mean if Google, after years and years and who knows how much money, still can't create a decent translator, there ain't gonna be any miracle software for this job.
Anyone who speaks two languages should be able to appreciate the complexities of the process, and how literal translation isn't gonna cut it. Which was what some fan translators insisted on in the early days and which made their stuff read, to put it mildly, a bit poorly. And some stuff can't be translated too well without understanding the context, which may in manga be purely visual. Here's for the next Star War The Third Gathers: The Backstroke Of The West... |
||||
|
||||
|
Arxana
Posts: 43 |
|
|||
|
Kodansha had to rerelease Sailor Moon with a brand new translation because people found their first release’s translation way too literal and stiff, and that was with a human translator. If they even think of releasing something with a machine-based translation, they will be found out pretty quickly and people will call them out on it.
Everything I wanted to say about this topic has already been said in this thread, so I’ll just add: give me a localization over a literal translation any day of the week.people who can make something like Japanese sound natural in English deserves so much more money and praise than they get. |
||||
|
||||
|
Utsuro no Hako
Posts: 1054 |
|
|||
MTLs from Japanese aren't just stilted. They're missing words because Japanese relies on contextual cues for readers to understand things that have to be explicitly stated in English. Until AI gets to the point that it can tell who the unstated subject and object of a Japanese sentence is, you need to have a human go through and rewrite every line of text. As for Internet memes appearing in translations, 99% of the time it's from something like Steins;Gate, WataMote or Isekai Smartphone, where the Japanese version is also full of memes and slang. You remind me of the review I read once of Squid Girl complaining that the subtitles were full of squid puns. Like, duh. |
||||
|
||||
|
meiam
Posts: 3471 |
|
|||
|
I don't think you have to go that far back in time to find peoples saying that computers would never be able to identify face in picture and now they're near perfect at it. If the company gets serious, they could make MT software specializing in manga or even genre and they could probably take care of 95+% of the text and translator can deal with the few edge cases and then use those cases to refine the process.
Although I guess the process will happen behind close door, rather than big announcement, we'll find out that some publisher has been using it for X number of years already. |
||||
|
||||
|
DaResidentDouche
Posts: 30 |
|
|||
|
The translators won't be replaced by the machine. At least not in all the stages. Rather the human translators will work on after the level 1 or 2 transformation of the base by the machine. The human translator will be then there be to resolve any ambiguity related issues or provide the necessary input needed for the machine to resolve the issue or process it further.
|
||||
|
||||
|
lossthief
ANN Reviewer
 Posts: 1440 |
|
|||
Or get rid of a redundancy and just pay the translator to, y'know, translate the book. Instead of creating massively less efficient or effective method that needlessly complicates the process. |
||||
|
||||
icomeanon6
 SubscriberPosts: 116 |
|
|||
|
Putting aside the technological fact that machine translation is so far from suited for this task that it would be of virtually no help to a competent translator, this is disgusting. This is literally de-humanizing the creative process. (And yes, translation is part of the creative process. Making a good translated text without creative writing aptitude is impossible.)
|
||||
|
||||
|
Jeff Bauersfeld
Posts: 110 |
|
|||
Maybe I'm a year or two behind the research, but last I checked facial recognition is very much not perfect, particularly for non-white people. And language is much more contextual than a face, so translations are even farther from perfect. |
||||
|
||||
|
DaResidentDouche
Posts: 30 |
|
|||
The machine's purpose will be there (or be built) to do the boilerplate and/or 'by the book' translation quickly compared to the human. The point/goal (mostly for the business) is to cut down cost of the human translation or more precisely set the pricing to "flat" for the human translator instead of "per x". |
||||
|
||||
|
lossthief
ANN Reviewer
Posts: 1440 |
|
|||
Yeah no, logistically all that does is put more work on the translator (who would really be an editor in this proposed scenario) who now has to double check the work of a program rather than just translating tie text themselves, before also having to turn a raw text into a workable script for a comic. Meaning editing or rearranging lines to better fit word bubbles, accounting for character voice and any other eccentricities that comes from writing fiction like in-universe terminology. Notti mention the monumentally thankless task of preserving humor across languages. But now they have to do it with extra steps for less pay. What you're proposing is downsizing and consolidating positions onto fewer workers, which universally leads ti more stress, worse performance, and higher rates of burnout and turnover. So like I said, the MO of shortsighted, poorly managed businesses |
||||
|
||||
#Neothegreenland
 Posts: 42 |
|
|||
|
i don't know whether to this this seriously or not. can't tell if it's a joke or not
-it could potentially undercut translators' wage because their paycheck is not fixed but rather depends on the amount they translate a lof of times -machine translation is not ready yet to translate adequately for it to streamline translation. so this is just an experimental run of the mill that everyone will forget about in a couple of months like with tokyopop |
||||
|
||||
Covnam
 Posts: 3876 |
|
|||
|
If any MT makes it to print in a commercial release I hope ANN puts the headline at the top of the page so I can make sure to never buy that release.
|
||||
|
||||
 |
All times are GMT - 5 Hours |
|
|
|
Powered by phpBB © 2001, 2005 phpBB Group