Machine Translation Experiment

Machine Translation Experiment

Aug 02

We all know Google Translate is bad at Japanese, but just how bad is it? I decided to run the previous chapter (62) through Google Translate, and this is the result. Here are the rules:

  1. Put all text (SFX too) verbatim into Google Translate. (Japanese⇒English)
  2. If a sentence is split across bubbles, do not combine them.
  3. All translations must be copied verbatim into the speech bubbles.



For full effect, I got Comic Sans MS to help me out.


  1. Dude


  2. axemtitanium

    While this is rightly recognized as atrocious, I just wanted to tip you off to this thing called Project Naptha ( which has the potential to revolutionize scanlation if they can add Japanese language support. tl;dr version is that it allows real-time, live text recognition and editing within any image file which is completely bonkers. Imagine if you could just open a scanned raw file, double click on the text bubbles, delete it, and paste in your translation in 30 seconds. Hope this project goes to its natural conclusion!

    • Calculuswhiz

      Seems like a pretty cool project. I’ll sign up for it ‘cuz I want to see it on Firefox. Actually, it’s probably not out of the question to extend Japanese support, since they’re all just symbols.
      You know, I’ve actually thought about something like this myself. Been kind of curious about computer vision, so this might be a fun (several weekend) coding project to start on. Really though, all I’d need is text removal because typesetting still requires precise alignment. So… I think probably the best way would be via GIMP plug-in? Worth a shot, right?

Leave a Reply

Your email address will not be published. Required fields are marked *