I've seen the video and the results are certainly good, but, as I've already commented in that video, there are certain things I'm not sure how it would be handled. For instance:
Sometimes, in rpgs, there are really long sentences which need to be printed in chunks one after the other, in more than one text box. If that system only translates what is currently shown in the screen, an accurate translation of the entire sentence would be difficult or not possible at all.
As a side note, I've been thinking for a few weeks now about a system that would be a mix of Clyde Mandellin's Wanderbar and the one shown in this video. Wanderbar grabs the translated text from a html file and shows every line of text in a separate window, and it would be great if, instead of that, the translated text could be shown in an additional layer just on top of the original box with Japanese text. You could even have a key mapped to a function that makes the layer with the translation visible and invisible at your own will.
The text could work the same way as Wanderbar already does: you have a text file in a specific format, be it html, xml, json, etc, and we could add some metadata such as position X and Y of the text window, background color and transparency, width and height, line ID of the original Japanese text, and of course the translated text itself.
I know, it would be quite some work to implement such functionality in an emulator (believe me, I would implement it myself right now if I had the skills), but it would make translating games a lot easier, as romhackers would not have problems with having to expand the rom for translated text to fit, having overflows and unexpected bugs, etc, because the rom wouldn't be altered in any way. You're just showing text in an additional layer.
What do you think?