- Dec 29, 2017
- 3,181
- 5,505
You need to parse the raw game again, because you don't have the staging data.
Please start a new project and go through the translation process again, I've made a lot of fixes and most of them wont work with old data.
Please start a new project and go through the translation process again, I've made a lot of fixes and most of them wont work with old data.
- Mar 12, 2022
- 183
- 47
- Dec 29, 2017
- 3,181
- 5,505
I've released v1.92.
My definitely not overly complicated regex patterns to fix wrong punctuation of the AI failed, and was actually deleting dots at the end of sentences and putting random spaces in text.
I've fixed it now and it totally will only do the right stuff now.
My definitely not overly complicated regex patterns to fix wrong punctuation of the AI failed, and was actually deleting dots at the end of sentences and putting random spaces in text.
I've fixed it now and it totally will only do the right stuff now.
- Oct 16, 2016
- 195
- 300
Hey Shisaye - maybe not a bug, but, I've been using the v1.92 for a couple of games now - coming up with a bunch of errors (like more than 350+ in some cases) - SHISAYE.LOSTBACKSLASH1, which did not occur before (prev version v1.89).
Sharing a couple of translation files with differences of RJ01098164 but I've noticed the same for at least 3+ games. Could you take a look?
Sharing a couple of translation files with differences of RJ01098164 but I've noticed the same for at least 3+ games. Could you take a look?
- Dec 29, 2017
- 3,181
- 5,505
At least the error system is working correctly.
Please try v1.93.
Edit: Btw feel free to always report error cells. Not just if it's hundreds of them.
The goal would be to always have 0.
Last edited:
hey i cant seem to get this working with Sugoi Toolkit V9 (public version) it looks like this
Last edited:
- Dec 29, 2017
- 3,181
- 5,505
Edit:
V9-V9.4 do not work with SLR.
But all other versions including V9.5 work.
Also for example all of Entai's repacks work with SLR:
You must be registered to see the links
Last edited:
- Dec 29, 2017
- 3,181
- 5,505
A lot of those UNUSUALLYSHORT ones were false alarms, I will fix that next version. (It is currently counting whitespace characters.)
As for your error it seems like you need to leave the 2 names of the main characters in the "Actors.json" untranslated because they are used by a custom plugin. (Just leave those 2 cells empty.)
As far as I can tell that wont actually impact their names in the dialogue, just in the menu, so it's probably not worth it to modify the plugin to be able to translate them.
In general that dev used a lot of weird plugins it seems, it's not even using the normal font system.
Last edited:
- Dec 29, 2017
- 3,181
- 5,505
Just to make people aware, get some opinions and stuff, I'm not sure how long this is still going to take, I'm working on a substantial change to SLR translation.
You may have noticed annoying SUGOIERROR or UNUSUALLYSHORT errors. Those are generally caused by Sugoi crashing and loosing parts of the input.
To prevent that I've built this large dictionary of pre-defined translations meant to prevent things being fed to Sugoi that crash it.
But the downside of that is that parts of the text sometimes feel very repetitive, generic or even out of place, because obviously my dictionary doesn't consider context and the entries are applied somewhat broader than they have to be, because I rather have slightly worse translation than no translation.
So currently it works like this:
Text found in the dictionary gets removed from the cell before the rest gets send to SugoiV4, if the output does not look like a viable translation even after applying all the patterns to fix issues with Sugoi, it places an error code.
And the plan instead now is:
Send all text without considering dictionary entries to the SugoiV4 model.
If the output fails the checks for a viable translation, send it to the Sugoi Levi model instead.
If that output fails the checks as well, apply all dictionary translations and send the rest to the SugoiV4 model again.
If that output still fails the checks, apply all dictionary translations and send the rest to the Sugoi Levi model again.
If that output still fails the check insert an error code.
That way context will be lost much less often, translations will feel much more varied and accurate, and you will see much less error codes.
Obvious downside is that it makes translation slower and you would need to have both models.
Although the average translation would only have like 20 cases, in which it would actually need to re-translate something.
Edit: The main obstacle so far is to actually accurately determine if a translation is viable or not.
I will have to do A LOT of testing.
You may have noticed annoying SUGOIERROR or UNUSUALLYSHORT errors. Those are generally caused by Sugoi crashing and loosing parts of the input.
To prevent that I've built this large dictionary of pre-defined translations meant to prevent things being fed to Sugoi that crash it.
But the downside of that is that parts of the text sometimes feel very repetitive, generic or even out of place, because obviously my dictionary doesn't consider context and the entries are applied somewhat broader than they have to be, because I rather have slightly worse translation than no translation.
So currently it works like this:
Text found in the dictionary gets removed from the cell before the rest gets send to SugoiV4, if the output does not look like a viable translation even after applying all the patterns to fix issues with Sugoi, it places an error code.
And the plan instead now is:
Send all text without considering dictionary entries to the SugoiV4 model.
If the output fails the checks for a viable translation, send it to the Sugoi Levi model instead.
If that output fails the checks as well, apply all dictionary translations and send the rest to the SugoiV4 model again.
If that output still fails the checks, apply all dictionary translations and send the rest to the Sugoi Levi model again.
If that output still fails the check insert an error code.
That way context will be lost much less often, translations will feel much more varied and accurate, and you will see much less error codes.
Obvious downside is that it makes translation slower and you would need to have both models.
Although the average translation would only have like 20 cases, in which it would actually need to re-translate something.
Edit: The main obstacle so far is to actually accurately determine if a translation is viable or not.
I will have to do A LOT of testing.
Last edited:
- Nov 3, 2021
- 391
- 934
That sounds like a lot of work. Good luck!
Aside from errors, I usually don't notice Sugoi-weirdness until I'm playtesting and at that point I'm happy to consult other machines or my brain to improve the translation manually. I hope you still leave the option to use the old method (maybe as a toggle in the options?). Keeping Levi on had seems like a poor use of storage space in my use-case since I do as much hand-editing as needed anyway.
Aside from errors, I usually don't notice Sugoi-weirdness until I'm playtesting and at that point I'm happy to consult other machines or my brain to improve the translation manually. I hope you still leave the option to use the old method (maybe as a toggle in the options?). Keeping Levi on had seems like a poor use of storage space in my use-case since I do as much hand-editing as needed anyway.
- Dec 29, 2017
- 3,181
- 5,505
I'm sorry, but Sugoi can only be used to translate to English. You would need something else for Chinese.
对不起，Sugoi 只能用来翻译成英文。
(我不会说中文，我只是用 DeepL 翻译的。)
- Dec 29, 2017
- 3,181
- 5,505
Well, if it doesn't find a Levi model it would not see a viable translation and just go to the next option.
So technically you wouldn't even notice a difference.
Also in case you forgot, Levi is only 1gb. The entire thing.
- Dec 29, 2017
- 3,181
- 5,505
I've released v1.94.
Obviously this is not the new system, yet, that will still take a while, but this already introduces necessary infrastructure.
Among some not yet active stuff I've significantly improved the UNUSUALLYSHORT (Sugoi has crashed) error code.
It's still not perfect and you might have some false alarms, but it should be quite a bit better now.
I've also changed the way excessively repeated characters and words are handled.
Not only will the system be much better at catching them all, it will also be more varied in the output so that it will not always be the same amount.
Another new thing is that the fix cells button will now no longer process cells if there is already a translation in Initial Corrections, Wrapped Translations, or Override. That means for example after you wrapped the project, if you need to press the button again to fill in some pre-defined translations in a plugin or whatever, the process will be much faster because all those wrapped cells are skipped.
Edit: I noticed a potential issue with my new patterns. I forgot to protect repeated dots and dashes. So it will now also reduce those to 10 characters or less.
Usually stuff like that isn't translated in the first place, and it's not like that would cause crashes, but I will still fix that next version.
Obviously this is not the new system, yet, that will still take a while, but this already introduces necessary infrastructure.
Among some not yet active stuff I've significantly improved the UNUSUALLYSHORT (Sugoi has crashed) error code.
It's still not perfect and you might have some false alarms, but it should be quite a bit better now.
I've also changed the way excessively repeated characters and words are handled.
Not only will the system be much better at catching them all, it will also be more varied in the output so that it will not always be the same amount.
Another new thing is that the fix cells button will now no longer process cells if there is already a translation in Initial Corrections, Wrapped Translations, or Override. That means for example after you wrapped the project, if you need to press the button again to fill in some pre-defined translations in a plugin or whatever, the process will be much faster because all those wrapped cells are skipped.
Edit: I noticed a potential issue with my new patterns. I forgot to protect repeated dots and dashes. So it will now also reduce those to 10 characters or less.
Usually stuff like that isn't translated in the first place, and it's not like that would cause crashes, but I will still fix that next version.
Last edited:
- Dec 29, 2017
- 3,181
- 5,505
I would assume that a standing picture file name was translated in plugins.js that shouldn't have, but I can't know that from just looking at the error.
I would need the actual game.
- Mar 12, 2022
- 183
- 47
the game is this
You must be registered to see the links
i download from h-sharingand this is the translation file
- Dec 29, 2017
- 3,181
- 5,505
You technically pressed a "Yes" button when loading the plugins, in which you agreed to check the plugins.js for weird cells, because the system cannot automatically detect custom plugins.
But Maybe this is some of those cases, in which something just seems obvious to me and nobody else...
Anyway you need to get rid of the translations for the cells 37 - 80 in plugins.js, or the "Macro" plugin.
Basically all that stuff that if translated says something like "The main character appears from the right", "Slide in from the right as the second fades".
Those are commands for the text boxes and standing pictures, but it will only accept Japanese commands because it's a Japanese plugin.
- Dec 29, 2017
- 3,181
- 5,505
I've made some rough tests, and to explain it better here's some translations in which re-translation happened in my prototype.
You don't have permission to view the spoiler content.
Log in or register now.
You don't have permission to view the spoiler content.
Log in or register now.
You don't have permission to view the spoiler content.
Log in or register now.