Scaling Laws for Neural Machine Translation