The development and application of machine translation provides a machine-translation perspective to reflect and shed light on the three key questions of translation in nature, i.e. translator, process, and product. In this weak AI era, as the translator overlaps between machine and human, machine translation facilitates the division of labour in translation but generalizes the translator’s identity, proving the great value of human translation; as translation process dichotomizes between human translation’s multiple transformation and machine translation’s probabilistic matching, their differences may be rooted in their different translation views: the former holds a dynamic outside-inside view of translation, while the latter adopts a static inside-outside view of translation; as translation products serve functionally and connotationally, machine translation can achieve the intratextual functions, but not necessarily convey extratextual connotations. Too much attention paid to the function of machine translation at the cost of human translation’s connotations can result in linguistic “pollution” and language sense “lossing”.