Typology as Language, or Why Loops are a Symptom of a Larger Problem