Could it autodetect the language using a few heuristics (or Bayesian analysis of existing submissions) instead of having the selection down the left hand side?