Currently the sentences for fluency evaluations are obtained in a manual way by querying Claude and are only available on a few language. We should have a fully automated way to do it in all our 35 languages.