Displaying screenshots to translators
44 views
Skip to first unread message
Geza Kovacs
May 25, 2012, 11:03:55 AM5/25/12
to transif...@googlegroups.com
(moving conversation over from
https://github.com/transifex/transifex/issues/65 at mpessas' request)
I've developed a system for providing a screenshot to translators to
illustrate the context in which a message is used in the program, and
wanted to help integrate it into Transifex if there was sufficient
interest in such a feature. Basically, the way it would work is, the
developer can upload some screenshots in addition to the message
template file, and they would automatically be matched to messages, so
that when the translator is translating a message, a screenshot will
be shown.
The system works by OCR-ing screenshots and then matching each message
to the best-matching screenshot. If interested in more details, I've
published a paper on it in CHI 2012 (
http://groups.csail.mit.edu/uid/other-pubs/chi2012-screenshots-for-translation-context.pdf
; the "Message-Screenshot Matcher" subsection is the part which
describes the algorithm)
In terms of code, the prototype implementation I used for the paper is
MIT-licensed, and is written in Java, is at
https://github.com/gkovacs/textmatch . I realize Java isn't very
popular in the FOSS world, so I can reimplement the algorithm in
another language (python, C, whatever's most convenient) if needed. I
was primarily interested if there was sufficient interest in
integrating such a feature into Transifex (wouldn't want to work on
this for several months only to have the work rejected due to
integration issues), and if yes, I'd appreciate specific steps that I
should be taking to do so.
Thanks,
Geza
https://github.com/transifex/transifex/issues/65 at mpessas' request)
I've developed a system for providing a screenshot to translators to
illustrate the context in which a message is used in the program, and
wanted to help integrate it into Transifex if there was sufficient
interest in such a feature. Basically, the way it would work is, the
developer can upload some screenshots in addition to the message
template file, and they would automatically be matched to messages, so
that when the translator is translating a message, a screenshot will
be shown.
The system works by OCR-ing screenshots and then matching each message
to the best-matching screenshot. If interested in more details, I've
published a paper on it in CHI 2012 (
http://groups.csail.mit.edu/uid/other-pubs/chi2012-screenshots-for-translation-context.pdf
; the "Message-Screenshot Matcher" subsection is the part which
describes the algorithm)
In terms of code, the prototype implementation I used for the paper is
MIT-licensed, and is written in Java, is at
https://github.com/gkovacs/textmatch . I realize Java isn't very
popular in the FOSS world, so I can reimplement the algorithm in
another language (python, C, whatever's most convenient) if needed. I
was primarily interested if there was sufficient interest in
integrating such a feature into Transifex (wouldn't want to work on
this for several months only to have the work rejected due to
integration issues), and if yes, I'd appreciate specific steps that I
should be taking to do so.
Thanks,
Geza
Tim Babych
May 25, 2012, 2:41:39 PM5/25/12
to transif...@googlegroups.com
Apostolis Bessas
May 30, 2012, 6:52:32 AM5/30/12
to transif...@googlegroups.com
Hi Geza,
Could you send me some details on how to test this? I mean, how do you
use the test data?
Additionally, what are the (software) requirements of this project?
There are two ways we could go forward with this:
a) Use the java version, but make it a service. We could then send the
po files and screenshots to that service for further processing (i.e.,
do it in the background).
b) Rewrite it in Python.
How fast is it? If it is really fast (within a couple of seconds), we
could have that written in Python and call it directly from within the
upload process.
If it needs more time, we would have to implement this as a service.
That means, that hte langauge is not the most important issue here.
Apostolis
--
Apostolis Bessas
Systems Engineer
https://www.transifex.com/
Could you send me some details on how to test this? I mean, how do you
use the test data?
Additionally, what are the (software) requirements of this project?
There are two ways we could go forward with this:
a) Use the java version, but make it a service. We could then send the
po files and screenshots to that service for further processing (i.e.,
do it in the background).
b) Rewrite it in Python.
How fast is it? If it is really fast (within a couple of seconds), we
could have that written in Python and call it directly from within the
upload process.
If it needs more time, we would have to implement this as a service.
That means, that hte langauge is not the most important issue here.
Apostolis
Apostolis Bessas
Systems Engineer
https://www.transifex.com/
Geza Kovacs
May 30, 2012, 8:10:45 PM5/30/12
to transif...@googlegroups.com
Hi Apostolis,
I've updated https://github.com/gkovacs/textmatch/blob/master/README.md
with additional details about prereqs, setup instructions, and usage.
In terms of speed, it's roughly O(#messages * #screenshots), so it's
slower on larger projects. For pcmanfm, it's 17 sec; with
xfce-terminal, it's 1:50, and with gnome-control-center it's 3 minutes
(this was on a laptop with a Core 2 processor).
Thanks,
Geza
2012/5/30 Apostolis Bessas <mpe...@transifex.com>:
I've updated https://github.com/gkovacs/textmatch/blob/master/README.md
with additional details about prereqs, setup instructions, and usage.
In terms of speed, it's roughly O(#messages * #screenshots), so it's
slower on larger projects. For pcmanfm, it's 17 sec; with
xfce-terminal, it's 1:50, and with gnome-control-center it's 3 minutes
(this was on a laptop with a Core 2 processor).
Thanks,
Geza
2012/5/30 Apostolis Bessas <mpe...@transifex.com>:
Reply all
Reply to author
Forward
0 new messages