Should machine-translated chapters be appropriately tagged?

Power Uploader
Joined
Oct 1, 2020
Messages
312
MTL has gotten a lot better since 'back in the day' in the last few years it's made huge leaps, it's not perfect but it's still enough to get good translations out if you work with it. And people who don't know enough Japanese to translate on their own have used it. You need to look at what is currently happening not your past.
Wrong.

By and large MTL has the same fundamental problems today that it's always had. Google translate has been using AI since 2016 is maintained, receives updates/improvements, and is still shit. Even the relatively newer stuff like chatgpt will still be confidently wrong and make up random shit just like every other MTL. Google has so much data from websites, books, user input with the translator, and more, but it still can't even translate a website with sufficient accuracy, which is one of the things it was designed to do. Comics have images, and these MTL are blind. Even non-fluent humans checking words and grammar in a dictionary can't hope to match the experience of having read and spoken a language in context for the hundreds or thousands of hours it takes to understand those langs to the level that their TLs can even begin reach their full potential.

It's true that GPT and to a lesser extent some contemporars like deepl have better output with respects to producing something that sounds like it's been proofread, but the issue is that fundamentally it's still not accurate enough in keeping the meaning and character of the text across langs. GPT you can talk to, and ask for alternative translations, or ask for clarification, which to be fair is a big improvement over other competing products, but its not fundamentally changing the game. When it comes to not making huge mistakes/making shit up, or in preserving the character of the text, just like every other MTL so far, it's simply not sufficient.
Don't let chat GPT fool you with extra spit polish in the form of outputting text that sounds like it's been proofread, a turd is still a turd no matter how much you polish it. It was trained to write text that mimics human conversation/ Its translation capability is influenced by this in ways that can result in the model not properly conveying the character of a text, even more than what you'd expect of a normal MTL. GPT all too often blatantly ignores instructions, and gives unreliable, robotic output.
In my opinion, this is in some respects reminiciant of how output from AI which generates digital paintings is noticeable because such ai generates a painting all at once, and not with brushes/tools like humans do in photoshop or csp/ect. GPT is a much better tool for people who already speak Japanese than it is for people who expect to be able to use it as a crutch, just like all MTL to this point, particularly for comic translation.

MTL has made modest improvements over the years, and future products may one day have potential, but as I said before, with the state of the tech being what it is, MTL is still woefully deficient when it comes to the fundamental parts of what makes for an acceptable translation, especially for comics/manga. "it's not perfect" is an understatement. Obviously human translators aren't going to be perfect either, but not in the way I think you mean, because a lot of the times there is no "perfect" translation. Translation is an artform past a certain threshold of competency. MTL does not pass the threshold of competency, or understand human language in a way that's conducive to making a good TL. This has been the case for years, and will probably continue to be the case for years.

Second the main point of the thing I suggested is not about reporting, that's just an extra use, the main point of it is to convey to people what the rest of the community feels about the chapter without them having to read through a bunch of forum post. If they don't want to add on the investigating they don't have to. But adding something like this might actually lower the number of reports for things that aren't offenses. Because people like that are reporting to vent, having a 'dislike' button for that same thing means they can 'vent' there without actually putting in a report. Not everything is a 1 dimensional system with just one feature and use.

The false positive reports are due to misunderstanding the rules, not just a desire to complain, altho obviously that can play a role. People go to the comments section for that. Rating individual chapters in some capacity does sound interesting, but I really dont know what that could look like, esp if the site is to be kept relatively free of useless clutter and bloaty tech debt. Not gonna say any more on that since your responses are so nonspecific.
 
Power Uploader
Joined
Oct 1, 2020
Messages
312
At this point the guy should just change his screen name to ostrich.
What kind of lame ass ESL insult is that? At this point you should just change your screen name to illiterate. drrrr.
You have the chutzpah to be indignant about getting a detailed response on an internet forum? Still too busy reading classics to spare a moment to read the forum post you'll be complaining about? If you want somewhere to piss and moan about shit you make up or heard in a youtube video essay or whatever go back to plebbit, facebook or twitter.
 
Dex-chan lover
Joined
Jan 18, 2018
Messages
716
What kind of lame ass ESL insult is that? At this point you should just change your screen name to illiterate. drrrr.
You have the chutzpah to be indignant about getting a detailed response on an internet forum? Still too busy reading classics to spare a moment to read the forum post you'll be complaining about? If you want somewhere to piss and moan about shit you make up or heard in a youtube video essay or whatever go back to plebbit, facebook or twitter.
Firstly it's not ESL, if you can't make out that the insult is you have your head in the ground like an ostrich maybe you're the one that's ESL. Secondly don't try to make yourself look smart by saying chutzpah. Using words that are unnecessary or overly extravagant to try and convey a point only makes you look like you're trying too hard.

Now for the meat and potatoes. Your 'detailed response' is well honestly meaningless. Almost all of what you said falls apart upon the realization that a person does not have have to take what the MTL gives it and just shove it into the scanlation. There are a number of things, from research, cross-referencing, and contextualizing that an individual without the ability to translate things far enough to do it on their own can do to improve the result of the scanlation. Your problem is that you are still stuck in the past. The simple FACT is that good MTLs of manga have been produced. It's that simple. So in reality anything you say about it not being a viable tool is just not true. Because it is viable, it's just a matter of how much effort and dedication you're willing to put in.

Also, the one who has been pissing and moaning is you. All you've done is whine about "MTL is bad, I sucked when I first started" You've actually said nothing of substance related to this actual discussion firstly. And you went off the rails the moment you said that there is no way of auditing MTL. Like just how binary and dumb is your mental track for you to think it's impossible to audit MTL?

And to add on to all that? You got caught being stupid with my suggestion, by making huge sweeping assumptions as to the goal, and then not only that but you tried to criticize the idea, based on those assumptions being the flaw in the method. You literally made up a problem just so you could argue it to try to look cool.

But most of all, you don't even know what you're talking about. Like at this point you are just saying stuff, you have no direction with the information you present. Your literal entire basis was in your own words:
The issue with MTL isn't that the output doesn't sound human, or that it needs proofreading, the issue is that you're delegating your ignorance of a language to an error-prone machine you have no way of auditing.

But you have yet to realize not only does that mean that the problem isn't MTL, but how people use it and at what skill level, but you also have literally said lines along the lines of "Most people don't use it right". Meaning that some people do, and it works. You have literally all but admitted to the fact that MTL can produce a viable and acceptable result. But yet here you still are trying to say MTL sucks, but the problem isn't the MTL, it's how it's used. Which is literally what I said in my first post, and why I disagree with an MTL tag. In other words, this means literally you have said nothing to actually argue the core of my initial statement this entire time.

Come back when you're going to actually discuss the point at hand, and not just rant.
 
Group Leader
Joined
Jul 15, 2019
Messages
6,600
Unmmm...... akchually
MTL this MTL that.

why dont you just appreciate those who clean the bubble instead and spread some love. 🥰🥰🥰🥰🥰

It's literally already in the name "Machine" why'd be so far discussing that this Machine would be actually bad or good.
It is there to assist human, nothing more.

Telling MTL is bad or telling MTL is good, this and that are not different than a shitpost in general lounge. :)
 
Abolish Harems
Staff
Developer
Joined
Jun 4, 2019
Messages
238
I just realized this thread is in the god damn lounge instead of the suggestions forum
@AlexEliot WHAT ARE YOU DOING ASKING IT HERE

This post is not meant to provide a suggestion, but rather invite discussion.
On the opening message of this post. And the title of the post is not even a suggestion—but rather a question (notice the question mark at the end, and the sentence beginning with "should"). So I don't know why you think it should be in the suggestions forum. Should I write down a grammatical analysis on why it's not a suggestion?
 
Dex-chan lover
Joined
Jan 20, 2023
Messages
264
Answering the tittle of the thread: IMO yupperino-dino-doo! Some google tl chapters are the complete opposite of what's written on the raws, so ofc they should be marked, althought that would mean human tl cchapters would need to be marked as such too, still, it would be pretty nice to know what we're looking at before even opening the chapter. The mark could be a green robot to the right of the group/ uploader's name? Something like that
 
Dex-chan lover
Joined
Jan 20, 2023
Messages
264
How about only the uploader can tag it? I can only see that coming to Pumpkin Night. wheeezeee
Wouldn't this defeat the point of the rating system? Like, the uploader can just overinflate the rating based on being proud of their work or outright lie and say that this piece of crap that looks like it was made by an incapable toddler is 5/5, imo it could work if it was user rated, like the manga rating system or youtube-5-star-review-like. Maybe if it was a tab for the uploader and group showing a rating chosen by the comunity? like "Manatuga-scanlation 2/10 vmman 1/10"
 
Group Leader
Joined
Jul 15, 2019
Messages
6,600
Wouldn't this defeat the point of the rating system? Like, the uploader can just overinflate the rating based on being proud of their work or outright lie and say that this piece of crap that looks like it was made by an incapable toddler is 5/5, imo it could work if it was user rated, like the manga rating system or youtube-5-star-review-like. Maybe if it was a tab for the uploader and group showing a rating chosen by the comunity? like "Manatuga-scanlation 2/10 vmman 1/10"
bro i said how about, not might be
 
Dex-chan lover
Joined
Jan 18, 2018
Messages
716
The problem with tagging something as "MTL" is that is carries no meaning other than to 'punish' people who are just trying to help out the community. Bad MTL will generally be easy to spot anyway, and if you can't spot it right away, then that just means there could (and likely are) human translations of the same quality floating around. I know I've seen my fair share of bad human translation. But the big problem is that GOOD MTL exist, and with a tag we'll have people deciding not to read perfectly good translations that had a lot of work go into them just because they used MTL software.

There is way too much focus on 'MTL'. Literally what people complain about 9 times out of 10 isn't even the translators job, it's the proof reader or editor. And no/poor proof reading or editing will result in a bad scanlation no matter how perfect the translation is.
 
Dex-chan lover
Joined
Feb 17, 2018
Messages
1,857
@Plykiya Gotta agree with @AlexEliot, on that one.

After, a well thought out discussion and agreements by all parties cough cough, I'm sure there will be a crystallization of an acceptable suggestion that can be then swiftly posted to the suggestions board and immediately accepted due to the merits and mitigated risks of the final proposal.

Someone help me rewrite that with Chat GPT. I'm curious to see how different it looks. I still haven't made an account with them.
 
Power Uploader
Joined
Oct 1, 2020
Messages
312
Firstly it's not ESL, if you can't make out that the insult is you have your head in the ground like an ostrich maybe you're the one that's ESL.
ofc I've heard of the myth that ostriches are so stupid they stick their heads in sand if startled, but that's a playground tier insult. Your attitude, and the fact you thought your insult was a worthy provocation was much more obnoxious to me than being compared to a bird. You'll have to forgive me for thinking english might be your second language when you write bangers like "Yes you're the ignornat one." and "Secondly, I said have knowledge of a language doesn't mean their language skills are prefect(...)"
Now that I think about it, your limited english skills may explain why you overlook my concerns that MTL betrays the characterization of the raw text. As I've reiterated many times till this point, translations are not good when they're free of errors which totally ruin the text. A translation without critical flaws with respects to semantic meaning can still fail to respect the integrity and character of the original text, the cornerstone of a good translation, by altering the character of the text which is something MTL is profoundly deficient and inconsistent with.

Secondly don't try to make yourself look smart by saying chutzpah. Using words that are unnecessary or overly extravagant to try and convey a point only makes you look like you're trying too hard.
It's a word I hear all the time where I'm from. My goal is not looking smart, it is to communicate effectively. If I wanted to put on airs and write like a pompous asshole it would require a lot more effort than I am willing to waste on you.

Now for the meat and potatoes. Your 'detailed response' is well honestly meaningless. Almost all of what you said falls apart upon the realization that a person does not have have to take what the MTL gives it and just shove it into the scanlation.
As I've said before here and in other posts, MTL is a tool. Most of the people who use MTL demonstrate a lack of understanding or respect for it as a tool, and JP as a lang, by acting like tools themselves by deferring to the tool when TLs should be made from knowledge not ignorance. If you don't speak Japanese, MTL or not, you can't reliably make translations that reflect the integrity and character of the original text, or guarantee MTL output its free from critical errors.

Your problem is that you are still stuck in the past. The simple FACT is that good MTLs of manga have been produced. It's that simple. So in reality anything you say about it not being a viable tool is just not true. Because it is viable, it's just a matter of how much effort and dedication you're willing to put in.
MTL is not always going to be completely worthless. This has been the case for many years. Even a stopped clock is right twice a day. In my posts I have been detailed and freely acknowledged where MTL has improved over the years, but that I am still critical of its use because I understand how it works and believe that it is widely misused. I'm not arguing it's always going to be totally incorrect all the time. It is a fact that there are plenty of legible MTL out there, even a considerable amount that don't make critical errors (altho they are still fail to maintain the character of the text), but the devil's in the details.
  1. Difficulty/complexity of source text
  2. Accuracy of the OCR used
  3. Specific MTL product used
  4. language knowledge of the people involved (if any)
  5. How revisions made to MTL output affect translation accuracy is not readily understood by revisor
  6. MTL is not gonna get the context looking at the images brings
All of these factors and more add extra complexity to the translation process, usually SOMETHING in this list makes the MTL go off course in my experience. With really basic short texts its totally possible MTL output will be free of critical errors. I've done it myself in the past and seen it done recently. What you seem to have overlooked is that even in cases where the MTL is able to make an accurate translation the odds are usually against it, and most of the time it is not reliable, especially not reliable enough to be used responsibly for manga TL with the expectation of consistency.

Also, the one who has been pissing and moaning is you. All you've done is whine about "MTL is bad, I sucked when I first started" You've actually said nothing of substance related to this actual discussion firstly. And you went off the rails the moment you said that there is no way of auditing MTL. Like just how binary and dumb is your mental track for you to think it's impossible to audit MTL?
As I've said before and will continue to reiterate: People who disproportionately rely on MTL are simply not going to have the capacity to audit its output effectively! In order to understand a target language on an intuitive level, language learners need to be exposed to a huge depth and variety of concepts. It's one thing to read about a grammatical structure in a textbook because you're trying to audit mtl and don't have the confidence or skill to know better than the AI, it's another to see the grammar or vocab hundreds or thousands of times in different contexts to the point that you don't need to rely on the MTL to confidently produce correct output. The MOST you can reasonably expect of MTL is that there are no critical story altering errors, and most of the time in my experiences with MTL in the past and today it doesn't even meet that abysmal standard all too often.

And to add on to all that? You got caught being stupid with my suggestion, by making huge sweeping assumptions as to the goal, and then not only that but you tried to criticize the idea, based on those assumptions being the flaw in the method. You literally made up a problem just so you could argue it to try to look cool. But most of all, you don't even know what you're talking about. Like at this point you are just saying stuff, you have no direction with the information you present. Your literal entire basis was in your own words: "The issue with MTL isn't that the output doesn't sound human, or that it needs proofreading, the issue is that you're delegating your ignorance of a language to an error-prone machine you have no way of auditing."
And I stand by those words, altho you do not seem to understand them. Right out of the gate in your first post you said "Because MTL isn't the problem, it's low effort and little to no proofreading that is the problem." People who aren't fluent with the langs can't reliably audit the MTL because they dont have the skill to! You are fundamentally misunderstanding the issue. If you can't operate independently of the MTL you can't audit it or "proofread" it to make it more accurate! Proofreading is not the same as TLQC! Most of the people who proofread the MTL are just guessing at what the true character of the text is, rather than making an informed decision based on fluency of the language, even if, as in some cases, the result is accurate, the result was not reached from a reliable method like human translation, so it should be discouraged! Instances where the MTL is flawless are by far the exception and not the norm.

But you have yet to realize not only does that mean that the problem isn't MTL, but how people use it and at what skill level, but you also have literally said lines along the lines of "Most people don't use it right". Meaning that some people do, and it works. You have literally all but admitted to the fact that MTL can produce a viable and acceptable result. But yet here you still are trying to say MTL sucks, but the problem isn't the MTL, it's how it's used. Which is literally what I said in my first post, and why I disagree with an MTL tag. In other words, this means literally you have said nothing to actually argue the core of my initial statement this entire time.
Why do you keep trying to move the goalposts on my own argument? People who use MTL expect it to be accurate, and since it is not accurate, the MTL is deficient!
The right way to use MTL is with:
1. projects that are not going to be distributed
2. for getting a rough/inaccurate translation
3. by someone who knows the source and target language
Obviously its not impossible for MTL to produce viable and acceptable results depending on the circumstances or difficulty of the text, however, for the purpose of making clear and consistent, accurate, comic book translations that are free of major errors but keep the tone and style of the text, MTL is sorely deficient. As I've said before, even a stopped clock can be right twice a day. These MTL products make it too easy for people to get output that seems good enough to the untrained eye because its not totally incomprehensible, while being prone to errors that only people who don't need to rely on the MTL are going to be able to mitigate. The fault does not only lie in the uneducated people who believe the MTL is good enough for comics, but in with the MTL products themselves for offering inaccurate services that are not marketed as not being accurate enough for use cases it wasn't designed for. Some MTL producers will even say flat out that their products are not meant to be used in situations where accuracy is a necessity (i.e. artistic works like comics), but its clearly not sufficient from the part of the MTL creator, because there are still people who use the MTL and rely on it rather than their own language skills. (google "when should i use machine translation")

In my responses I have also argued against your inaccurate portrayal of MTL's recent developments, to which you have provided no substantial counterclaims. Not only are your responses to my arguments reductive and unfounded, but I believe you fundamentally misunderstand what MTL is, how it works, and when its use is acceptable in translating Japanese comics. It's kinda starting to look to me like you yourself are guilty of the accusations you level towards me, particularly along the lines that I'm not arguing reasonably, because rather than offer a detailed 1:1 response to my claims as i have for you, you generalize my positions while making broad statements and complain that I give detailed responses. All the explicit claims you've made so far have either not been relevant, are not in contention, or are unsubstantiated.

To tie things up closer to on-topic:
The problem with tagging something as "MTL" is that is carries no meaning other than to 'punish' people who are just trying to help out the community. Bad MTL will generally be easy to spot anyway, and if you can't spot it right away, then that just means there could (and likely are) human translations of the same quality floating around. I know I've seen my fair share of bad human translation. But the big problem is that GOOD MTL exist, and with a tag we'll have people deciding not to read perfectly good translations that had a lot of work go into them just because they used MTL software.
The tag would have to be self-enabled, as I advocated for in my first post. If the uploader wants to tell people they aren't confident in the quality of the translation because they have used machine translation, they should have the ability to mark their upload, this would not be compulsory. There are plenty of people who will want to be open with what tools they use. Even if the quality of the MTL will vary with the complexity of the manga and depend on circumstance as MTLers so often say, "Its better than nothing"
There is way too much focus on 'MTL'. Literally what people complain about 9 times out of 10 isn't even the translators job, it's the proof reader or editor. And no/poor proof reading or editing will result in a bad scanlation no matter how perfect the translation is.
Complete misconception. "If you can't operate independently of the MTL you can't audit it or 'proofread' it to make it more accurate! Proofreading is not the same as TLQC! Most of the people who proofread the MTL are just guessing at what the true character of the text is, rather than making an informed decision."
Is a non searchable, opt-in only vanity tag for chapter uploads really so bad? You can't say that having a chapter marked as MTL is a punishment if it's the uploaders own personal decision.
 
Dex-chan lover
Joined
Apr 30, 2018
Messages
944
@Plykiya Gotta agree with @AlexEliot, on that one.

After, a well thought out discussion and agreements by all parties cough cough, I'm sure there will be a crystallization of an acceptable suggestion that can be then swiftly posted to the suggestions board and immediately accepted due to the merits and mitigated risks of the final proposal.

Someone help me rewrite that with Chat GPT. I'm curious to see how different it looks. I still haven't made an account with them.
Was Chat GPT the one that tried to tie a user to a past murder case. That was funny.
 
Dex-chan lover
Joined
Apr 26, 2020
Messages
817
Can we have a discussion about inappropriately tagging MTL chapters next? I want to see tags like 'dodecaphonic' and '3.73 limited-slip' get thrown around.
 
Joined
Jun 27, 2023
Messages
5
While a suggestion for this has already been made here, the matter was quickly dropped due to the short-lived characteristic of suggestion threads. This post is not meant to provide a suggestion, but rather invite discussion.

So let me start it off with a reason in its favor: considering there are countless chapters that are preceded by a disclaimer page specifying the chapter is machine-translated, an MTL tag will relieve scanlators of the need for such a page, and readers will be properly informed in a more streamlined way. The tag would only be decorative; meaning that it will not be filterable, ensuring no kind of penalty is given, so there would be no effect of chapters tagged as machine-translated getting less publicity than those without it.

Would that solve the MTL problem once and for all? If not, what is preventing it?
A question maam/sir... what if the chapters were scanlated through the use of MTL but with the involvement of the scanlator himself, editing it, proofreading it, making it decent and human-readable.. Does that still under the MTL Tag? or Does it not?
 
is a Reindeer
VIP
Joined
Jan 24, 2018
Messages
3,231
On the opening message of this post. And the title of the post is not even a suggestion—but rather a question (notice the question mark at the end, and the sentence beginning with "should"). So I don't know why you think it should be in the suggestions forum. Should I write down a grammatical analysis on why it's not a suggestion?
people post probing questions in the suggestions forum all the time to figure out whether people are in support of it or not
 

Users who are viewing this thread

Top