Context: https://en.wikipedia.org/wiki/Statistical_machine_translation
Since we can use the statistical translation models to calculate the probability that the F is the translation of the E, why we can't calculate the probability that the E is the translation of the F using the statistical translation models.
$$\mathop{\arg\,\max}\limits_{e} p(e \mid f) = \mathop{\arg\,\max}\limits_{e} p(e)\times p(f \mid e)$$