I want one of these running free (as speech) software; imagine the possibilities, the size of the market, the captive audience, your splashscreen in so many public places.
Jim Gettys has an interesting post about the challenge of localising software on large scale (in the context of OLPC).
I personally like the AbiWord model: internally strings are identified by numerical id’s; only one set of strings is built into the software (en_US), and localised strings for a given locale are read from an external file, falling back on the internal stringset. Interestingly, Maemo uses similar system and there is a lengthy discussion about the advantages / disadvantages of such system vs. gettext on the Maemo developers list. I find none of the arguments in there that were put forward for gettext particularly convincing; true, gettext has some neat features, like plural handling, but the application pays for that with significant processing overhead, which, depending on the HW and how often strings need to be looked up, might or might not matter. On the other hand, what gettext has undoubtedly in its favour, in the GNOME context is its estiblished use.