Lol this is a crazy question... Has anyone ever tried Optical Character Recognition software to translate manga? Basically the OCR program would (ideally) "read" the scanned pages of a raw manga and then produce a translation or semi-translation. Maybe even a bad translation, but at least something...?
The thing is, I have loads of mangas that I love that are only partially scanlated or only up to a cliffhanger, and I desperately want to know what happens next! I don't need a "perfect" translation, just something even vague so that I know what's going on. This is my pathetic otakuness lol. Of course I'm planning to learn Japanese during the holidays but I figured it definitely couldn't hurt to have OCR help me out if it can...
Has anyone tried this and does it work? I'd try it myself but I don't have OCR software right now and I don't want to buy it and spend money on it if it isn't gonna work. Thought I'd ask first in case someone has tried it already... Thx!
Manga Poll
Manga is the Japanese equivalent of comics
with a unique style and following. Join the revolution! Read some manga today!
Join #baka-updates @irc.irchighway.net
RSS Feed
Forums
OCR to translate manga?
You must be registered to post!
From User
Message Body
Member
5:02 pm, Nov 25 2009
Posts: 1354
Member
6:48 am, Nov 27 2009
Posts: 142
I've tried it, just to see if it is indeed possible. And, well...
It is, and it isn't.
The first problem is that OCR programs get very confused in the presence of artwork/bubbles/whatnot inbetween the precious text they're trying to extract.
The second problem is that most OCR programs suck ass at Japanese. And those that are at least decent require you to know a fair bit of Japanese in the first place, if you want to use them with any sort of confidence.
The third problem is that you can forget them doing the hand written text. (That includes sfx, of course.)
And, though this isn't really a problem as such, you also need high resolution scans. Don't expect tolerable results from Share/Winny scans.
But let's say you can extract maybe 80% of the text from a manga, give or take a bit depending on the type. (For some rare species of seinen, I wouldn't be surprised if you can get above 95%. And for some not so rare species of shounen, I wouldn't be surprised if you can't even get half.)
Which bring us to the last problem, that machine translations work horribly on manga. They might work reasonably well on technical documents, but there's very little in most manga that resembles that kind of text.
So, it's possible to try, at least. But you'll get a terrible "translation" that doesn't make any sort of sense, of maybe 3/4ths of the source text. And it won't be as automatic as you might be thinking, either, all OCR programs I've tried so far require quite a bit of manual coaching to get it right.
What OCR programs actually can help with, though, is if you have something that's really heavy on the obscure kanji, it can cut down on the time required to look them up.
It is, and it isn't.
The first problem is that OCR programs get very confused in the presence of artwork/bubbles/whatnot inbetween the precious text they're trying to extract.
The second problem is that most OCR programs suck ass at Japanese. And those that are at least decent require you to know a fair bit of Japanese in the first place, if you want to use them with any sort of confidence.
The third problem is that you can forget them doing the hand written text. (That includes sfx, of course.)
And, though this isn't really a problem as such, you also need high resolution scans. Don't expect tolerable results from Share/Winny scans.
But let's say you can extract maybe 80% of the text from a manga, give or take a bit depending on the type. (For some rare species of seinen, I wouldn't be surprised if you can get above 95%. And for some not so rare species of shounen, I wouldn't be surprised if you can't even get half.)
Which bring us to the last problem, that machine translations work horribly on manga. They might work reasonably well on technical documents, but there's very little in most manga that resembles that kind of text.
So, it's possible to try, at least. But you'll get a terrible "translation" that doesn't make any sort of sense, of maybe 3/4ths of the source text. And it won't be as automatic as you might be thinking, either, all OCR programs I've tried so far require quite a bit of manual coaching to get it right.
What OCR programs actually can help with, though, is if you have something that's really heavy on the obscure kanji, it can cut down on the time required to look them up.
Member
7:13 am, Nov 27 2009
Posts: 1354
THANK YOU.
Seriously, this is the most helpful and amazing reply ever... *bows repeatedly* Lol I wasn't sure anyone was going to reply after all this time, but you did and so thoroughly too! You addressed all my concerns... I was worried about the quality of the translations and whether it was even possible - now thanks to you I know what to do. I think I'll wait until I learn Japanese better before trying to decipher OCR-rendered translations lol.
Much love for you for replying to this thread! Thank you SO SO much! <3 <3 <3 <3
PS - I also think you're the one who did that really helpful scanlation tutorial?
Thanks so much for that too!
Seriously, this is the most helpful and amazing reply ever... *bows repeatedly* Lol I wasn't sure anyone was going to reply after all this time, but you did and so thoroughly too! You addressed all my concerns... I was worried about the quality of the translations and whether it was even possible - now thanks to you I know what to do. I think I'll wait until I learn Japanese better before trying to decipher OCR-rendered translations lol.
Much love for you for replying to this thread! Thank you SO SO much! <3 <3 <3 <3
PS - I also think you're the one who did that really helpful scanlation tutorial?
Thanks so much for that too!Member
11:06 am, Nov 27 2009
Posts: 142
Well, since I already went through the trouble of trying 5-6 different programs for it, the least I can do is share my impressions and possibly get some use out of the experience. Seeing as how the programs themselves turned out to be mostly useless, I mean.
If you want to try it out yourself, I recommend this program as being the least crap of those I've tried: http://www.kyodo-networks.net/RealReaderLiteTop.html
Sure, the Lite version has some limitations that would be annoying if you actually wanted to use it for real. But for a test, it works fine. (Which is kind of the point, I assume.)
Edit: I was slightly unfair. That program is actually quite good, the task is not an easy one. I was just being negative because I don't know the terms for things like "vertical right to left reading", but have to chance it.
Last edited by pnyxtr at 10:19 am, Nov 28 2009
If you want to try it out yourself, I recommend this program as being the least crap of those I've tried: http://www.kyodo-networks.net/RealReaderLiteTop.html
Sure, the Lite version has some limitations that would be annoying if you actually wanted to use it for real. But for a test, it works fine. (Which is kind of the point, I assume.)
Edit: I was slightly unfair. That program is actually quite good, the task is not an easy one. I was just being negative because I don't know the terms for things like "vertical right to left reading", but have to chance it.
Last edited by pnyxtr at 10:19 am, Nov 28 2009
You must be registered to post!
