At least I think, because of many references I found in /usr/share/maliit/, it uses Nuances XT9 prediction. XT9 is available in many languages, so that's good, but it's closed source and I can't find the place where the dictionaries are stored... Maybe it's a good start to find out in which format they're stored, where they are located and if it is possible to add other languages, like Dutch. It could be doable since we've got an open source dictionary team, found here.