We'll be using the BIP39 implementation present in Bitcoinj as slush says. 

>Proper Unicode handling is a serious issue however. You don't want
>someone to move from one input method / machine to another and
>suddenly find that their coins are inaccessible, because of an issue
>of decomposed vs. compatibility forms or whatever.

We generate the word list internally (12,18,24) and confirm it is entered correctly through a retyping operation. This will allow us to detect character encoding transpositions (e.g. u0032 vs u00a0) and alert the user before it becomes an issue.

While English is the language of the first word list to be implemented, we would definitely integrate alternative non-English word lists to make life easier for the global community. In general the approach would be for the user to select their language (implying a locale) and then the word list to be selected from that locale if available with a fallback to English. This follows the same approach as resource bundles in Java.