Compute word stems in various languages (http://hackage.haskell.org/package/snowball)
It appears that using Data.Text.Encoding to go back and forth to utf8 should suffice and reduce the footprint, since text-icu is sort of a pain to get working reliably cross-platform?
