Should machine-translated chapters be appropriately tagged?
- Thread starter AlexEliot
- Start date
Group Leader
- Joined
- Jul 15, 2019
- Messages
- 7,470
I guess sooner or later machines will be able to talk to you with violence.
Dex-chan lover
- Joined
- Mar 15, 2019
- Messages
- 3,074
Group Leader
- Joined
- Jul 15, 2019
- Messages
- 7,470
the loli we dont deserve, but we need. i hope her 3D A.I will come soon and says "you low life dont deserve my feet."We are not there yet,
But what we currently have
Is quite glorious.
Dex-chan lover
- Joined
- Mar 15, 2019
- Messages
- 3,074
Dex-chan lover
- Joined
- May 18, 2019
- Messages
- 3,747
As much ad I loathe MTL dumbasses, a dedicated tag probably won't work in practice and they'll just find a way around it. The "bad translation" (for incorrect translation that fails to follow the idea of the original text) and "bad grammar" (shit to read) tags on sadpanda need a lot of manual moderation and don't even work properly. Bad grammar is easy to prove, the same can't be said for unfaithful translation: the good old strict translation vs. liberal localization debate.
If anything, I just want a bigger limit for my block list. Something like 500 groups and 4000 uploaders would do.
A decorative unfilterable tag is probably the worst solution you can implement.
If anything, I just want a bigger limit for my block list. Something like 500 groups and 4000 uploaders would do.
A decorative unfilterable tag is probably the worst solution you can implement.
Last edited:
- Joined
- Jun 4, 2019
- Messages
- 246
I really like your idea. Instead of "shaming" bad quality, praising good quality seems much healthier and will probably even incentivise groups to do better to get that highlight. Plus, it has been proven to have a positive community effect on nearly every platform.
To ensure that groups who upload first don't get an unfair advantage, I imagine it would be based on a percentage of the views rather than just comparing two integers. How to do it safely would need a little data analysis with a sample of a period where we'd run this without any visible effect—there certainely is a perfect point we can have decent confidence rate.
Basically having a "what did you like about this chapter?" with options such as "good translation" / "good lettering" etc.. While also displaying a message that advises the user to vote based on their honest opinion and ignore the scanlator asking for commendations on their credit page (because they will do it).
In fact, this is such a sound idea that I believe deserves its own thread.
Last edited:
Contributor
- Joined
- Oct 1, 2020
- Messages
- 382
would really love a self-reporting MTL tag, since people will often admit to MTLing. It just cant be required otherwise some definitely-not-mtlers will piss themselves trying to prove their innocence.
- Joined
- Mar 27, 2023
- Messages
- 292
yes, it should have had that long ago
I hate the idea of doing anything of pointing out MTL. Because MTL isn't the problem, it's low effort and little to no proofreading that is the problem. I've read plenty of MTL chapters that were perfectly legible. And at the same time I've literally had to argue with someone on reddit that the MTLer's chapter was of equal quality to a certain group's. And the guys evidence of bad translation was a list of effects in a spell name being in a different order, and a replacement of a idiom, which the MTL'er literally had a note that said "The normal is this [insert idiom] but I don't think that makes sense so I replaced it." So the only difference ended up being a human decision. And that was all the person could find but they were insisting on "It's MTL so it's bad!" People have an emotional and irrational hate for MTL because in the early days it was bad, letting them know it's MTL will have them looking for problems.
Rather maybe we should do something like Nyaa does. Like they have colors for their post that pretty much mean if a group is in good respects with the community, normal, or if they're in bad respects (or a reuploader) at least I think that's how the colors work. So like you could have it so that chapters that get a lot of positive feedback on a 'like/dislike' counter get green text. And if they get a lot of dislikes they get red text. And maybe if it goes to a certain ratio, the mods investigate it to see if there is actually something legitimately bad that needs to be addressed.
Rather maybe we should do something like Nyaa does. Like they have colors for their post that pretty much mean if a group is in good respects with the community, normal, or if they're in bad respects (or a reuploader) at least I think that's how the colors work. So like you could have it so that chapters that get a lot of positive feedback on a 'like/dislike' counter get green text. And if they get a lot of dislikes they get red text. And maybe if it goes to a certain ratio, the mods investigate it to see if there is actually something legitimately bad that needs to be addressed.
Contributor
- Joined
- Oct 1, 2020
- Messages
- 382
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. Anyone who doesn't know how to speak japanese is gonna be at a loss on if the MTL is making the right choices because they themselves have no way of knowing. MTL can't even translate the text free of inaccuracies, let alone faithfully maintain the actual character or prose of writing, which is the entire fucking purpose of a good translation.
MTL may be perfectly legible, but since most MTL's are an outdated large language models glued to a dictionary, ofc its gonna make grammatically correct output with no spelling errors, thats not the issue anyway, because the underlying translation is often both inaccurate and usually not faithful to the form of the text. MTL is often too literal, in the next sentence too vague, even the best large language models still have the same limitations and faults that every other MTL has where it will make shit up totally out of the blue and almost never correctly preserve the character of the writing.
MTL was designed to be quick and dirty, the most it hopes for is that there are no huge errors, and even then it fails far too often. Even if there are no dire inaccuracies, the tone and character of the text is not correctly translated by the MTL under ideal conditions, MTL manufacturers know this, officials know this, everyone know this, but for some reason there are still redditors like you who don't actually know how any of this works but still feel free to go around preaching to other people about how MTL is good/bad because XYZ.
Even in cases where the differences between two languages aren't as large as jp > en, like en > es, while the MTL will make far less fatal errors, people will still prefer to read something translated by a human because human translators aim to maintain the actual character and form of the writing, rather than aim to have as few critical errors as possible and fail.
Saying MTL is perfectly legible is like telling people your keyboard works because the backlight turns on even though all the switches are dead. You want your keyboard to accept input, not glow when you plug it in. NO SHIT the mtl is gonna be legible, its literally a spellcheck machine with a dictionary, MTL is shit at what people expect from a good translation, and a lot of the times it cant even make a bad translation because its so prone to massive errors.
red posts on nyaa are repacks/ect, so right off the bat its not 1:1 applicable for MD. If MD were to adopt the nyaa system there would be a lot of recency bias with what groups get selected for green. And even then, if ALL of a group's uploads get greened, for a lot of groups with long histories and who's early chapters were bad, they would still get green, making tagging good groups meaningless in the first place. It would make more sense to show green or red ratings on individual chapters, because instead of evaluating the quality and consistency of every chapter a group's ever made its just a single chapter, but even then, readers have no idea what goes into a good scan, and it's just gonna turn into a scanlation group popularity contest at best, and a totally worthless bit of extra clutter and wasted dev time at worse.
Thats not how Mangadex works. READ THE RULES. The site does not police scanlation quality. If you report a chapter for having missing text, bad redraws, low resolution, or what have you, your report is gonna get ignored. No way no how is MD gonna start policing scans or scan quality. The only relevant exception is when a chapter is entirely AI generated (HMM, I wonder why?!)
I'm not even going to read your entire post, you're just really thick headed and stuck in your ways. I'm gonna point a few things out for you.
Firstly MTL does not mean that the translation is gonna be poor quality, becuase MTL doesn't mean that the person translating it has no knowledge of the language they're translating from. Also with every year MTL solutions get better and better. I litearlly put in my post that I had an entire argument about the quality of an MTL, and it was upt to the same quality of a good group. So no it's not impossible for MTL to be good if done responsibly.
Second thing I'm going to point out, I did not say police quality. Take off your stupid filter and stop trying to preach your dumb personal bias and listen to people. I said "And maybe if it goes to a certain ratio, the mods investigate it to see if there is actually something legitimately bad that needs to be addressed." No where does that say "if the quality is so bad it needs to be removed" I said "SOMETHING LEGITIMATELY BAD".
Now get your self indulgent text walls out of here.
Contributor
- Joined
- Oct 1, 2020
- Messages
- 382
you can't even bring yourself to read my post, but I'm ignorant? In my post right out of the gate i'm making it clear the issue is people delegating ignorance of a language to an error-prone machine they have no way of auditing. Most people who use MTL are doing exactly that. If you already know how to speak japanese the MTL is at best redundant! Obviously!
what way do you mean this other than the mods policing it? Its not the mods job to do that, they've got their hands full with taking care of site data. What would the mods be doing in "investigate it to see if there is actually something legitimately bad that needs to be addressed." if not policing the scanlation quality?
Supporter
- Joined
- Apr 26, 2020
- Messages
- 2,331
After a day and a half of rolling this around in my head on and off, I would agree with this sentiment. There doesn't seem to be a way to implement something that's not vulnerable to being weaponized and otherwise abused.Perhaps we should just
Maintain the status quo and
And just kick this can....
Yes you're the ignornat one. Why? Because like I said MTL can be good. It doesn't matter if "Most the people" that you see doing MTL use it badly. If it can be good than that means MTL isn't the problem so you don't try to make rules based on MTL. That's why you're ignorant, because you don't know how to address an actual problem. Secondly, I said have knowledge of a language doesn't mean their language skills are prefect, they could be comparing their own knowledge to the MTl, and cross checking. They could be using MTL to get the gist of it but then using things like novel translations or what they know from a group that dropped the series to try and recreate the right speaking pattern. It might not be perfect, but plenty of human translations aren't perfect and even take liberties with character speech so what, the point is not every MTL is trash, and that's literally all that matters.
If you would get off your high horse for even like I don't know 20 seconds you could've actually got enough oxygen to your head to think. And you'd notice that my idea does not limit itself to 'quality'. In fact I only say the word 'quality' once in the entire post and that is literally talking specifically about TML translation quality, not when I'm talking about my idea. I said the for "feed back" and "like/dislike" This could be for ANYTHING. Yeah it would usually be quality related but not always. It's another tool that can be used to detect problems, because people would be more likely to make a vote on that than to actually report. Because for reporting they'd have to know the rules, they don't have to know the rules to know they dislike something. So if there is a huge dislike ratio, investigating to see if there is a legitimate problem of any kind, be it a crazy water mark (because that has shown up on site before) a group page that violates rules, advertising, a scam, or bot farming dislikes...all of those could be reasons people hit dislike on a chapter.
Contributor
- Joined
- Oct 1, 2020
- Messages
- 382
As I said, MTL is at best a redundant tool when its in the hands of a good translator who doesnt need it, but is most times a crutch that makes bad decisions for people who cant speak the language. You're saying MTL can be good, but what you mean is that there are people who can make MTL good because they already saeak japanese well enough to rein MTL in if its being inaccurate... That's not the MTL being good, thats the people being good. I would like to reiterate: the issue with MTLers is that they CANT tell if mtl output is accurate because they DONT speak the language.
I'm not advocating for calling people out for MTLing, either. It would accomplish nothing worthwhile, be hard to do at scale, and practically impossible to do in a fair way. If you'd bothered to read my one sentence first post you'd know what I am advocating for is giving people who use MTL as a crutch without knowing the language the option to add a tag to their uploads if it was MTL'd, if they want to. There are already tons of people who put warnings right at the start of an upload about how its an MTL, and that people should read at their own discretion, ect ect...
Regardless of anything else, I think it would be better for everyone if people who make scans for themselves with MTL but also post to MD who want to put it out there that they're using MTL can get support in being transparent. Years ago, when I was just starting to get into scanlation, before I learned Japanese and came to understand through learning JP just how bad MTL is, I used MTL in some scans. I'd love it if I had the ability to go back and tag a few chapters I made without help from a real human TL as MTL for the sake of transparency. Even if some people don't want to use the feature, which is fine, I think it should still be there for people who do.
People on MD are already using the reports feature to its full potential, stuff gets reported all the time, regardless of if it actually breaks the rules, and one of the many things I see the site staff laugh about frequently over discord is just how often people make reports for no good reason, and how they're always swamped with reports. If they were to add a system where scans with a certain % of downvotes would essentially be automatic reported for mods to investigate if there's an issue would waste even more of the moderation team's already very limited time on chasing issues that may or may not exist rather than looking into something specific that's worthy of catching attention.
The report system is already way more efficient than what you're proposing because with reports there's no need for mods to look through the situation to seek a possible issue rather than get a report with a specific issue to adjudicate. And with reports, even when users make false-positives (i.e. report something that does not break rules) at least the staff can be sure its a false positive, and not just them missing the reason for people to leave downvotes.
- Joined
- Jun 15, 2023
- Messages
- 3
It should, but Mangadex staff probably won't do it. Wait, aren't you one of the devs?
Maybe you could convince other staff to fast-track and prioritize so the forum comments can show up on the manga title page like in the old days?
Let the readers warn their fellow readers about it. They can post saying that the chapters done by this group are using MTL, so read them at your own risk.
With a featured post on the manga page, any future readers will now be aware of it and can make their own decision whether to read it or not.
Maybe you could convince other staff to fast-track and prioritize so the forum comments can show up on the manga title page like in the old days?
Let the readers warn their fellow readers about it. They can post saying that the chapters done by this group are using MTL, so read them at your own risk.
With a featured post on the manga page, any future readers will now be aware of it and can make their own decision whether to read it or not.
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.
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.
Similar threads
- Suggestion
