I think that having a file for each language with the strings is OK.
But I think we don't need any other file for RTL. We need just an array (sorry Derick :)) with the languages that use RTL. Which languages use RTL? and which charset?
A lot of strings can be extracted from the phpdoc build system (entities used for translation). Also the phpdoc build system is aware of what language is RTL or LTR, so there is no need to duplicate this.
Goba
