That may be the way to reduce duplication and bloat in the Lojban [1] root-word list. Currently, about 1300 root words exist. A lot of them have overlapping or

That may be the way to reduce duplication and bloat in the Lojban [1] root-word list. Currently, about 1300 root words exist. A lot of them have overlapping or redundant meanings, and it has already been proposed to (manually) unify them and reduce their number. Combined with the already existing rules for word composition, and the grammar, it could make a very nice language. The idea is still very far from possible completion though. One (of several) things to look at is: how do we characterize word composition? word vector adding? averaging? I guess the operation could be learned too. [1] Lojban is an artificial language based in predicate logic http://lojbo.org/ http://lojbo.org/ . Source for the "could be used in AI" claim in that website: http://www.goertzel.org/new_research/lojban_AI.pdf http://www.goertzel.org/new_research/lojban_AI.pdf .