The translation is really bad, no offense.

Could you post the PSD files? I want to make some fixes to the ESL-kun stuff (and obvious MTL mistakes) in the TL, since there is no doubt in my mind that this is a low-effort MTL (as someone who does higher-effort MTL, I can tell).

Example:

Page 13 "Isn't it in a position where it's easy to insert" should be "Wouldn't it be pretty easy to put it in in that position? Are you inviting me to put it in?" or a more idiomatic translation would just be like (for both bubbles) "That position… Are you inviting me to put it in?"

Page 14: "You're a complete sex toys" isn't english. In fact, this is just raw output from DeepL, as I suspected. He says "You're a sex/cum toilet (shou benki 性便器) hugging a toilet (benki 便器) and getting raped (hahaha)" (it's a joke bc he's holding the toilet and also being called a cum toilet).

You should be able to add the PSDs if you make a torrent with them.

Last edited on 18 September 2023, 04:15.

Base +19, saiox +6, Braltewrakcus +10, Sturer Emil +7, Nomake Wan -19, some fafg -6, crimsonraes -17, ROABE +6, turkdeturk -6, Ruckkus1411 +7, Th3 Sl0th -6, Jeriko78 +6, Nyxtani -6

>higher-effort MTL

Now I've seen everything

Base +7, slappydude +6, Nomake Wan +19, banana33 +10, fsadz +6, animadversion +6, Hazuma +10, UnknowDestroyer +19, JDhentai69 -6, turkdeturk +6, Babycakes +6, Th3 Sl0th +6, Nyxtani +6

Actually going through the MTL line by line and correcting it is higher effort than copy/pasting. Correlating results and messing around with multiple engines and looking up individual Kanji on Jisho is also more effort. This TL is dstractingly bad.

I also try to run any MTL I do by someone who actually reads japanese for checking. The number of legitimate japanese translators willing to work for free is much smaller than the amount of hentai manga people want to read, and hand-edited and proofread MTL can pick up some of the slack (and MTL is getting better every year that passes).

Do I make more mistakes than an actual japanese reader? For sure. No doubt. But my accuracy is ~90+% and doesn't contain shit that makes no sense in english. I also check my releases to see if anyone has any corrections to offer. I also try to put up my cleaning/typesetting as PSD files if anyone with more experience wants to fix stuff (except that one time that I nuked them by accident).

Last edited on 18 September 2023, 08:25.

Base +19, Nomake Wan -19, fsadz -6, some fafg -6, crimsonraes -17, bagiaaa -13, waltergeist -10, theleastdankperson -6, hajtos +8, Fr33sh00t -6, turkdeturk -6, Ruckkus1411 +7, etr +6, and 5 more...

@Negi-Sensei

slay, queen

Base +19, Nomake Wan -19, animadversion +6, sameguy -6, Hazuma +10, JDhentai69 -6, turkdeturk -6, 1d10t -6, Nyxtani -6, Shino22 -7

>But my accuracy is ~90+%
I love when they just give random percentages. It's always in the 90s too.

How did you figure out your TLs are "90%" accurate? Did you magically learn the language and compare MTL output with your own manual translation? Did you ask a native to do a comparison for everything you do? Could you tell me who if that's the case? I've had the displeasure of being given proofread MTL scripts before, and the only way I could salvage them is a full re-translation.

Not saying you can't put effort into "cleaning" MTLs, but this whole argument is moot. You can't suddenly clean up incorrect output (if you even realize it's wrong) unless you a) make a guess or b) already know the right answer (in which case you already know the language.) You could also go with route c) and painstakingly double-check everything. Look up every word in a dictionary and every grammar rule, and that way validate the MTL output. (Note: if you're doing this, you're actually just learning Japanese kek)

>and MTL is getting better every year that passes
Is it? Again, I'm not sure how you can make that statement without knowing the language you're translating from.

I've been unfortunately following its progress over the years, and honestly the output it provides hasn't changed much in the past 5 years. The biggest leap has been grammar, in the sense that modern MTLs will try their damnest to give you coherent output, but coherent =/= accurate. This has fooled many a MTL user (and reader.)

source: me after having tagged too many machine translated galleries

Base +24, fsadz +6, some fafg +6, crimsonraes +17, justineonetwothree +6, animadversion +6, Hazuma +10, Nomake Wan +19, Th3 Sl0th +6, Nyxtani +6

I can estimate accuracy because I also eMTL Web Novel chaps and compare vs human tled chapters that come out later, lol. You're acting like it's impossible to validate a methodology when it is far from it. Manga is a bit different, because it has less textual context, but where there is such ambiguity, I ask actual jpn speakers for help lol. Also, luckily, the stories in hmanga aren't usually that complex.

You can also do things like split a text into lexemes or morphemes, translate into romaji and sound things out (sometimes you can catch nonsense with this method), analyze sentence structure, etc. I guess if you want, you could call this stuff "learning japanese without characters" lol.

And then at the end of all that, you can put up your PSD files if a real translator wants to show up and fix things.

> I've been unfortunately following its progress over the years, and honestly the output it provides hasn't changed much in the past 5 years.

The problem is that most of the MTL on this website is really bad and low-effort while using outdated and inferior tools like DeepL.

GPT4 is immensely more accurate than earlier engines. Sadly it doesn't love hentai words so you have to replace words like butthole with birthday before asking it to translate, but this is fine if you know what you're doing. .

PS Meanwhile, no raw MTL engine will give you the benki/shou benki joke from this hmanga, but the joke is clear from the romaji. Seems like I am doing something right.

Last edited on 18 September 2023, 12:12.

Base +19, crimsonraes -17, waltergeist -10, Fr33sh00t -6, turkdeturk -6, Ruckkus1411 +7, etr +6, Nomake Wan -19, Th3 Sl0th -6, 1d10t -6, cumslut420 -6, Nyxtani -6, Shino22 -7

@Negi-Sensei

you're doing a lot of work to not learn japanese when you could just learn japanese

Base +19, crimsonraes +17, justineonetwothree +6, animadversion +6, Hazuma +10, waltergeist +10, lordoffools +6, Yosh_ +12, turkdeturk +6, Nomake Wan +19, Th3 Sl0th +6, Nyxtani +6, devildonger +6

I'll learn japanese when japanese people switch permanently to romaji. I'm not interested in learning/memorizing hieroglyphics.

Base +19, crimsonraes -17, clovis_486 -6, waltergeist -10, Rx2TF -6, Fr33sh00t -6, Kross25 -30, turkdeturk -6, Ruckkus1411 +7, etr +6, Nomake Wan -19, Th3 Sl0th -6, elvendrunk -11, and 3 more...

I wish the Japanese abandoned kana and embraced kanji.

Base +25, justineonetwothree +6, flash11 +25, ROABE +6, Rx2TF -6, Nomake Wan -19

Yeah, Kana is the worst of both worlds because they don't use spaces with it.

Base +19, justineonetwothree +6, Rx2TF -6, Fr33sh00t -6, Ruckkus1411 +7, Nomake Wan -19

There's human translators in the web novel scene? lmao

>You're acting like it's impossible to validate a methodology when it is far from it.
Didn't say it was impossible, just that with the effort required to go through with a proper validation, you might as well just learn the language and completely eliminate the guessing. I also have high reservations on what % constitutes an accurate TL (90% would be unacceptable to me, and enough to tag a gallery), but that's another conversation.

I'm gonna have to agree with Mr_Person. You're almost doing more effort than I am (assuming you are indeed as accurate as you portray yourself), and I had to learn this dumb language. That said, I haven't checked your own releases, so there's that.

At the end of the day, all I care about are good, accurate translations. I don't really care how someone gets there, but I've been less than unimpressed with MTL.

Base +24, clovis_486 +6, Nomake Wan +19, Th3 Sl0th +6, Nyxtani +6

04/05: Yuuto
- inconsistent spelling => Yuto
05: it's a cute girlfriend
- referencing a person => she's a cute girlfriend
07: turn our ass over here
- wrong reference? => turn your ass over here
08: A ano => S- Say
10: is it
- not interrogative => it is
11: you cried [...] you didn't like
- ongoing => you're crying [...] you don't like
11: first blowjobs come before first kisses?
- incoherent reaction? => first kisses come before first blowjobs?
- rhetorical? => don't first blowjobs come before first kisses?
12: it's a sense of achievement
- emphasize? => that's a sense of achievement
14: a complete sex toys
- inconsistent number => a complete sex toy
15: develop more and more
- object needed => develop you more and more
19: would be nice
- looking towards consequence => would make it nicer / would make it better
20: He's resting my mouth!
- nonsense
22: some fun toys [...] try it
- inconsistent number => some fun toys [...] try them
22: in front of you
- wrong reference? => in front of her

Base +8, Th3 Sl0th +6, Hazuma +10, Nyxtani +6

@cutegyaru Yeah, the Re:Zero web novel has a human translation for all of arcs 1,4,5,6 as well as parts of arc 2,3, and 7. At some point in arc 7, in order to improve chapter turnaround times, they moved to a hybrid emtl model where a rough tl was first performed by a group of mtlers and then a Japanese-English check and revision was done by the actual translators.

I've done full eMTL on my own without J-E check and compared line by line with the human TL. In my experience, my workflow is giving near-perfect fidelity on dialogue, but it sometimes has difficulty with very complicated or vague narration.

@orasmase I didn't check every page, but your "corrections" for page 14 are just wrong. It's something like "Look, the sex toilet is hugging the toilet while getting raped. Hahaha". The 'Hahaha' is in the raw, literally "ははは".

The problem w this tl is not just bad english grammar but that it's half bullshit (other half, lies). You should at least read the thread, since I mentioned page 14 in the first post...

Base +19, waltergeist -10, Fr33sh00t -6, turkdeturk -6, Ruckkus1411 +7, Nomake Wan -19, 1d10t -6, Nyxtani -6, Shino22 -7