- Nov 17, 2017
- 333
- 806
Damn... so far SLR v4 with the dictionary seems to be the best so far. Looking forward to it!
- Dec 29, 2017
- 3,108
- 5,382
SLR with the dictionary is what is being used right now.
The point of the new system would be to give Sugoi V4 and Levi the chance to translate it first so that it's not always the same generic preset inserts from my dictionary, but more varied.
I probably didn't pick the best examples for that, considering only in the first example Levi didn't crash.
- Dec 29, 2017
- 3,108
- 5,382
Basically during my test translation with roughly 220000 characters, Sugoi V4 crashed 13 times, Sugoi Levi was able to recover it 6 times, and 7 times my dictionary system needed to be used to prevent an error.
Edit: If that still didn't make sense, don't worry about it, it will happen in the background without you even noticing.
Because among thousands of cells it will only really do that a few times, it might even actually be faster because dictionary checks will only happen if it was actually used. Meaning during that entire translation it ran those checks only 7 times.
Edit: If that still didn't make sense, don't worry about it, it will happen in the background without you even noticing.
Because among thousands of cells it will only really do that a few times, it might even actually be faster because dictionary checks will only happen if it was actually used. Meaning during that entire translation it ran those checks only 7 times.
Last edited:
- Nov 17, 2017
- 333
- 806
What do you mean by crash? Like did the program crash or did Sugoi Crash? Was it because of intensive GPU/CPU usage?
- Dec 29, 2017
- 3,108
- 5,382
Sugoi like basically all "AI" translators can hang up on certain inputs. They just loose the plot for whatever reason.
Sugoi for example will suddenly start spamming the same character or word 50 times and loose the rest of the input.
That's what you're seeing in those examples.
I'm preventing those hangups/crashes by not feeding it certain patterns I know will "most of the time" cause it.
That's the main purpose of my dictionary system.
But different models hang up on different stuff.
So while the Sugoi V4 model crashed on Example 1.
The Sugoi Levi model actually didn't.
And so even though Sugoi V4 crashed, using the dictionary was not necessary in Example 1, because Sugoi Levi translated it correctly instead.
- Dec 29, 2017
- 3,108
- 5,382
Will it give a gigantic night and day translation quality improvement?
Probably not, but as weird as that sounds it might still actually make the translation faster, because it's not checking the thousands of entries of my dictionary for every single cell.
I will also make the 2 Levi steps optional, so that if you don't have a Levi model running it just skips them.
Probably not, but as weird as that sounds it might still actually make the translation faster, because it's not checking the thousands of entries of my dictionary for every single cell.
I will also make the 2 Levi steps optional, so that if you don't have a Levi model running it just skips them.
- Nov 17, 2017
- 333
- 806
Will have to test it out myself more when you release it. Not gonna lie I think it'll have a somewhat considerable improvement regarding phrasing and maybe even fix some of the pronoun issues.
- Apr 19, 2024
- 94
- 102
- Mar 12, 2022
- 155
- 34
- Apr 19, 2024
- 94
- 102
u have old data(folder) right? if yes just copy and paste the map091 file. if some dialogue is still in jp. translate again ( not all text in those map091 just the dialogue only)
- Dec 29, 2017
- 3,108
- 5,382
If you mean load loose files into SLR Translator to translate with SLR, yeah you can.
It just needs to be "normally readable", as in if you can open and edit it with Notepad++ you can also load it into SLR Translator.
You can do that using the Custom Parser option.
There's an in-game tool to create parser, but it's pretty bad/broken.
The best way currently is to create parsers manually.
There's some examples in the "Custom Parsers" folder.
For example this is a parser that will just grab everything from all ".txt" files in a given folder:
Code:
{"files":[{"pattern":"*.txt", "rules":[{"type":"regex","pattern":"/\\r?\\n?([\\s\\S]+?)\\r?\\n/g","captureGroups":"1"}]}]}If you actually just meant manually edit the files, For MV and MZ you can just open the Json and Js with Notepad++ and edit them directly.
You can use a code beautifier to make them better readable.
For VX and Ace your best bet is using the official editors. You may or may not find versions in the Additional Tools package.
- Dec 29, 2017
- 3,108
- 5,382
In v1.96 I've added (among other stuff) a new warning message that can appear during batch translation, which might make stuff seem broken, spamming about recovered placeholders.
It will show up if a placeholder was recovered using less ideal means and "could technically" mean that the sentence is garbled up which could be important for someone who wants to keep all parts of the translation legible.
It will show the way the placeholder was matched, the original placeholder, the original Japanese text, and the translation without the placeholder being replaced.
For the most part you can ignore those warnings, it was already doing that recovery process before 1.96, it just didn't tell you about it.
It will show up if a placeholder was recovered using less ideal means and "could technically" mean that the sentence is garbled up which could be important for someone who wants to keep all parts of the translation legible.
It will show the way the placeholder was matched, the original placeholder, the original Japanese text, and the translation without the placeholder being replaced.
For the most part you can ignore those warnings, it was already doing that recovery process before 1.96, it just didn't tell you about it.
Last edited:
- Dec 29, 2017
- 3,108
- 5,382
To clarify since I should have mentioned this:
It will not report shortening/trimming of the placeholder as long as the main identifying features are present and connected, it will only report if there is real doubt, that it actually found the correct remains. Otherwise this would spam on every single batch.
If all attempts to recover a placeholder fail, it defaults to Reddo's original system for "guessing" based on the original position before translation.
It will also only use the recovery system that allows the placeholder to change, if you use the SLR placeholder, it will not even try on any of the other options. If another placeholder is used it defaults to Reddo's recovery system.
It will also check the original Japanese text if something that looks like a placeholder is already present, and if it finds something it will again default to Reddo's system.
It will not report shortening/trimming of the placeholder as long as the main identifying features are present and connected, it will only report if there is real doubt, that it actually found the correct remains. Otherwise this would spam on every single batch.
If all attempts to recover a placeholder fail, it defaults to Reddo's original system for "guessing" based on the original position before translation.
It will also only use the recovery system that allows the placeholder to change, if you use the SLR placeholder, it will not even try on any of the other options. If another placeholder is used it defaults to Reddo's recovery system.
It will also check the original Japanese text if something that looks like a placeholder is already present, and if it finds something it will again default to Reddo's system.
Last edited:
- May 7, 2023
- 601
- 365
Hello, I'm having a problem with a Japanese Wolf game, I've already managed to translate it using SLR, but now my problem has turned to its font, which is cutting off the accents and everything else there and I can't change its font, how can I do that?
when i open a .json file to translate doesn't recognize the japanese characters, only the translated text that i put into the file manually
- Dec 29, 2017
- 3,108
- 5,382
What exactly are you trying to do and how are you trying to do it?
I mean RPGM files don't work like that, so I assume this is a json from some kind of translation tool?
I also assume you want it to grab the Japanese, but insert the translation in the other entry?
Edit:
If translating the Japanese, but placing it in the second entry is what you want, you could either make a custom parser, which is pretty annoying, or just place the Japanese in the second entry first.
To do that just open the json in Notepad++ open the search window, replace, turn on regex, and do
Code:
"(.*?)"(:.*?)"(.*?)"
Code:
"$1"$2"$1"
Then you can just use the json parser in SLR Translator and batch translate, but a reminder that SLR is only for translating to English.
Also there's some entries at the end of the json that probably shouldn't be MTLed, at least they look like they would break.
Last edited:
- Dec 29, 2017
- 3,108
- 5,382
If you don't really care about optimization you can use any of the official Sugoi Toolkit versions (except 9-9.4) and run Sugoi in offline mode. (The port will already be the correct one, you do not need to change anything.)
Or you can read this post:
https://f95zone.to/threads/slr-tran...aker-vx-vx-ace-mv-and-mz.208393/post-13751939
I do not want to give any specific tutorial or recommendation, because it's highly dependent on your hardware and personal preference.