The reading task can stay largely automatic until both streams try to use the same speech-production machinery at once