3.00 star(s) 6 Votes

Greily

Member
Mar 4, 2018
209
408
I don't really care about this game, but the translation discussion is interesting. Hope you keep turning positive results Dazedanon. Good luck!
 
  • Red Heart
Reactions: dazedanon

sanahtlig

Member
Jul 6, 2019
122
395
Yup. It would be interesting to see though. I actually have a few scenes in mind I could test it with but for now on the backburner.
How would you rate the performance of ChatGPT (using your methodology) compared to alternatives such as DeepL, Google Translate, and Sugoi's offline translation model? Is this assessment based purely on the quality of the English (writing fluency)? Have you considered comparing against a gold standard such as a professional translation to assess translation accuracy also? I could potentially assist with a manual translation if you'd be interested in conducting such a pilot study, though it would need to be from a game I've played (I play a lot of NTR games).
 

dazedanon

Engaged Member
Modder
Uploader
Donor
Jul 24, 2017
2,438
26,826
How would you rate the performance of ChatGPT (using your methodology) compared to alternatives such as DeepL, Google Translate, and Sugoi's offline translation model? Is this assessment based purely on the quality of the English (writing fluency)? Have you considered comparing against a gold standard such as a professional translation to assess translation accuracy also? I could potentially assist with a manual translation if you'd be interested in conducting such a pilot study, though it would need to be from a game I've played (I play a lot of NTR games).
In direct comparison I wouldn't call 3.5 a massive improvement over DeepL or Sugoi. What makes ChatGPT so much better for translation is that you can feed it context in the form of prompts to lead its output, something you can't do with the others. It also helps that it's really good at outputting proper english.

For example normally MTL's struggle with gender consistency. ChatGPT isn't different in this aspect. However if you feed it enough context, character names/genders, as well as who the speaker is, then it has no trouble distinguishing them (at least as far as I've tested).

If you would like to help me test the accuracy of the translation, this game is actually a good one to test on. I can send you the raw map files and you can compare a few scenes with the translation to see how close it actually gets.
 

NTR Angel

Member
Modder
Mar 17, 2021
466
5,488
I get where you are coming from but I'm not here to provide a professional translation for these games. I don't translate old games because I'm interested in new ones. I make these for me and I share them because I think the AI translation will improve with community support.

All I want from my MTL's are a clean interface and decent English. That's not something you are going to get using MTool or Translator++. It may be a fake good translation but ultimately, I as a non-japanese speaker can't generally tell the difference anyway.

Will these MTL's dissaude translators from translating these games? Idk, I really don't buy it. Majority of the games I see hit Dlsite remain untranslated for at least a year or two if at all, unless it's picked up by someone like Kagura Games. I doubt this game or any of the others I did would end up any different.

I don't really like the idea that we shouldn't MTL games because someone *might* not give it a real translation. But thats just my opinion I suppose.
"Will these MTL's dissaude translators from translating these games? Idk, I really don't buy it." This is a very common thing discussed in the translation community very often. Making MTLs is totally fine, I just hope you can reconsider doing it on older games.

It has never crossed my mind to make a proper translation of this game, but at this rate is likely you will publish an MTL of a game that is either getting a proper translation or has planned a proper translation from me or other translators. Which will only do damage to the perceived quality of these new games on the western side of the world. In my opinion, games translations should not be just understandable but also enjoyable. A decent game will turn to be just "meh" or "bad" with a translation with no character personality and context issues.

MTLing games that are older or actually have 0 chances of getting translated is guaranteed to benefit everyone, but just going for the new games will have high chances to ruin them forever, since once the MTL is up, it will set the perceived quality of the general public. Tons of reviews and comments will be made from it and even if the game gets a proper translation later down the line, those reviews will stay forever there.

I really hope that soon enough we will be able to upload entire game files and .jsons to these AIs to get proper translations but for now I hope you can give this conversation more thought and also ask around what other people think to become a much bigger positive force for the community.

With great power comes great responsibility
 
Last edited:

foolyGold

Member
Jun 18, 2019
124
275
I heard in terms of accuracy for 3.5 is slightly worse than Deepl and Sugoi, but the better context and wording of ChatGPT makes it more readable. Those context related issues (mis-gendered words) from Deepl and Sugoi really do stick out.

But then again 3.5 is the cheap version of ChatGPT compared to 4.0...
I do wonder how accurate 4.0 is.
 

sanahtlig

Member
Jul 6, 2019
122
395
In direct comparison I wouldn't call 3.5 a massive improvement over DeepL or Sugoi. What makes ChatGPT so much better for translation is that you can feed it context in the form of prompts to lead its output, something you can't do with the others. It also helps that it's really good at outputting proper english.

For example normally MTL's struggle with gender consistency. ChatGPT isn't different in this aspect. However if you feed it enough context, character names/genders, as well as who the speaker is, then it has no trouble distinguishing them (at least as far as I've tested).

If you would like to help me test the accuracy of the translation, this game is actually a good one to test on. I can send you the raw map files and you can compare a few scenes with the translation to see how close it actually gets.
One of the main issues I've encountered with MTL is subject/object confusion. I'm curious to see how GPT3.5 performs in this regard. Give me the JP and ENG text for the first scene and I'll see what I can do. I haven't played this game, nor do I intend to, but I might be able to make do by just playing that scene.
 

sanahtlig

Member
Jul 6, 2019
122
395
"Will these MTL's dissaude translators from translating these games? Idk, I really don't buy it." This is a very common thing discussed in the translation community very often. Making MTLs is totally fine, I just hope you can reconsider doing it on older games.

It has never crossed my mind to make a proper translation of this game, but at this rate is likely you will publish an MTL of a game that is either getting a proper translation or has planned a proper translation from me or other translators.
I propose a rough set of guidelines for machine translation to handle some of these concerns:
  1. Categorize releases into tiers: Commercial manual, fan manual, edited MTL, MTL. These tiers are chosen for simplicity and minimal ambiguity, unlike a theoretical translation quality assessment. Optionally, add a description of translation methodology (such as GPT3.5 with added context).
  2. Clearly distinguish release tiers: Tag the tier the release belongs to (this can currently be done by the original poster). But we also need filters that allow users to ignore edited MTL and/or MTL. This can only be implemented by the site admins, but it's necessary. (Tentative) Fan manual releases subsequent to MTL releases could get a new thread, though handling strings of incomplete/abandoned translation projects could get messy.
  3. Selecting titles to work on: Avoid titles with active translation projects at a higher tier (commercial manual > fan manual > edited MTL > MTL). Such projects should demonstrate evidence of progress or be announced by a group with a credible history of releases. Keep in mind that commercial projects will always have the option of simultaneous release, which MTL can never compete with.
Advances in AI will soon result in the site being flooded with MTL releases. Without an intentional and well-considered framework, we're all going to drown. Even English-original projects could suffer lack of visibility from being buried under the deluge.
 
Last edited:

NTR Angel

Member
Modder
Mar 17, 2021
466
5,488
I propose a rough set of guidelines for machine translation to handle some of these concerns:
  1. Categorize releases into tiers: Commercial manual, fan manual, edited MTL, MTL. These tiers are chosen for simplicity and minimal ambiguity, unlike a theoretical translation quality assessment. Optionally, add a description of translation methodology (such as GPT3.5 with added context).
  2. Clearly distinguish release tiers: Tag the tier the release belongs to (this can currently be done by the original poster). But we also need filters that allow users to ignore edited MTL and/or MTL. This can only be implemented by the site admins, but it's necessary.
  3. Selecting titles to work on: Avoid titles with active translation projects at a higher tier (commercial manual > fan manual > edited MTL > MTL). Such projects should demonstrate evidence of progress or be announced by a group with a credible history of releases. Keep in mind that commercial projects will always have the option of simultaneous release, which MTL can never compete with.
Advances in AI will soon result in the site being flooded with MTL releases. Without an intentional and well-considered framework, we're all going to drown. Even English-original projects could suffer lack of visibility from being buried under the deluge.
On paper it's a very good proposition. However it would only work if a complete new thread is made if a translation gets an upgrade.
I think it's okayish how MTL tags are handled right now, my issue is, once again, releasing MTLs of new games and permanently downgrading a "possible" much better experience is just sad.
 

sanahtlig

Member
Jul 6, 2019
122
395
I think it's okayish how MTL tags are handled right now, my issue is, once again, releasing MTLs of new games and permanently downgrading a "possible" much better experience is just sad.
Commercial projects aren't going to care whether an MTL is available on F95. The audiences are different.

Fan projects might care--visibility/attention is the social currency that tends to power such projects--but manpower is limited and the likelihood of a finished project in a reasonable timeframe is low for any given project. I could potentially see the fan TL scene atrophying, but that's already happening due to commercial projects cannibalizing the talented translators. Fan TL probably won't be able to compete with advanced MTL in a simple contest for interest; "free" users tend to care more about the appeal of the base game rather than the quality of the TL, and the scalability of MTL will always provide an advantage in number of titles to choose from. Therefore free fan TLs will likely become an endangered species. However, I could see Patreon-funded fan TLs living on: there will always be super fans willing to pay for a better experience, and where there's money there will be those willing to work for it. The crowdfunding model also provides an incentive for backers to promote the projects, which will help them stand out from the sea of MTL releases.
 
Last edited:
  • Like
Reactions: Toushiro

Spellslinger

New Member
Jun 14, 2019
12
20
This is a save for the Ugly Bastard NTR route. I still cannot figure out how to get the teacher route. I could use some help. Remove the last "." in the name of this save file.
Two days before her birthday, the girl should be at the school. Talk to her and when the option come up tell her "i agree". Ugly bastard seems to be default bad end, while i'm assuming the friend starts his route if you go to talk to him at the shrine at night like he asks.
 

John-721

Member
Feb 3, 2022
133
151
I have played the game and just saw the friend scenes in the gallery, it was a hard ntr painful for the mc and I can't even imagine how are the other routes, specially the ugly bastard because if you do the happy route which is the one that I did you discover that the story is very shady.

You don't have permission to view the spoiler content. Log in or register now.

After watching the story I agree with having the place where the antagonists are being burned, the deserve that and much more. I just don't understand why the friend antagonist is being burned by the kid of the insects, I think I miss the part where he treats him bad or something because for the ugly bastard he is being burned by Natsuka's dad I think and the teacher is being burned by some female teacher or girl that suffered harassment from him.
 
Last edited:

sanahtlig

Member
Jul 6, 2019
122
395
I always wonder what's the point of having no scenes for the avoidable path... it feels like your being punished for choosing to NOT let the LI get raped and corrupted. It also feels super lazy.
I like games that implement NTR in the following ways:
  1. Unavoidable scenes with differing levels of intensity depending on which route you're on / heading for. It might just be touching in the non-NTR route, but it's still sexual and stimulates feelings of jealousy. There may be suspense about just how far the action will go. Cli Pick does this pretty well.
  2. Games with multiple heroines where preventing NTR is difficult or has trade-offs. Preventing NTR all the time may result in bad things happening or bad ends, or be punished by game mechanics. It may be impossible to avoid all scenes with all heroines, as favoring/protecting one leaves the others vulnerable. School Days is a good example of this.
  3. Games that focus on suspense. It's boring if the outcome feels predetermined. Good NTR requires both a lingering feeling of anxiety and the carrot of hope--that everything could still turn out ok.
As for scenes that aren't NTR, it's helpful to have at least one scene between the protagonist and the love interest at the end to give the sense of conclusion/reward.
 
3.00 star(s) 6 Votes