Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss
Proposal for a new l10n Tool
28 views
Skip to first unread message
Gautam Akiwate (gakiwate)
Mar 20, 2012, 4:40:55 PM3/20/12
to mozilla-t...@lists.mozilla.org, bugzi...@gmail.com
Greetings,
My name is Gautam Akiwate. And I would like to propose for a new l10n tool.
Details:
The proposed l10n tool will aim at helping standardize localizations.
In some languages, especially Indic languages, there are multiple possible translations for the same word. As a result everybody follows a different translation leading to a lot of cleaning up needing to be done by upstream contributors.
The idea is to create an exhaustive database for entries with their corresponding suggested translation. This should also work with small phrases.
I propose a tool which uses MT based on existing work. Simply put the idea is to take the existing l10n work. Put it through a statistical machine translation system like Moses. Based on this a list is created with preferred translations for the English words. Corrections to this are made if any. Organize it into a database of sorts and design a small web portal that will help people contributing to l10n find words and preferred translations real easy. This will be done for all languages.
Further to this the quality of the suggestions for existing strings compared to the localizations in our source will be done. As the MT system is not perfect errors can usually crop up due to a variety of reason. Thus, every MT system will give confidence levels for the translations and how these levels fare across the different languages. Moreover, the confidence levels can be indicator either that the MT system doesn't work well with the language or the existing localization sources quality is not consistent.
PS:
I would like to take up this project as part of GSoC 2012. Hence, I would appreciate if somebody would be willing to Mentor me for the same.
Thanks to Pike [Alex Hecht] for helping me.
Regards
Gautam Akiwate
gautam....@gmail.com
IRC Nick: gakiwate
My name is Gautam Akiwate. And I would like to propose for a new l10n tool.
Details:
The proposed l10n tool will aim at helping standardize localizations.
In some languages, especially Indic languages, there are multiple possible translations for the same word. As a result everybody follows a different translation leading to a lot of cleaning up needing to be done by upstream contributors.
The idea is to create an exhaustive database for entries with their corresponding suggested translation. This should also work with small phrases.
I propose a tool which uses MT based on existing work. Simply put the idea is to take the existing l10n work. Put it through a statistical machine translation system like Moses. Based on this a list is created with preferred translations for the English words. Corrections to this are made if any. Organize it into a database of sorts and design a small web portal that will help people contributing to l10n find words and preferred translations real easy. This will be done for all languages.
Further to this the quality of the suggestions for existing strings compared to the localizations in our source will be done. As the MT system is not perfect errors can usually crop up due to a variety of reason. Thus, every MT system will give confidence levels for the translations and how these levels fare across the different languages. Moreover, the confidence levels can be indicator either that the MT system doesn't work well with the language or the existing localization sources quality is not consistent.
PS:
I would like to take up this project as part of GSoC 2012. Hence, I would appreciate if somebody would be willing to Mentor me for the same.
Thanks to Pike [Alex Hecht] for helping me.
Regards
Gautam Akiwate
gautam....@gmail.com
IRC Nick: gakiwate
0 new messages