A good language model should learn correct language usage, free of biases and errors.