The simplest answer is: try it and you will find out. What aspect will you look at first? When will the job be finished? What will you take into account? Since the answers to none of these questions is obvious, it is obvious that there is a problem here.
The heart of the problem is defining parameters or criteria for comparison.
The first determined attempt in this direction was that by Kitty van Leuven-Zwart (in ‘Translation and Original’, Target 1, 1989, 2, 151-82 and Target 2, 1990, 1, 69-96). She began by dividing both texts into comparable phrases and then aligned the phrases with a view to isolating the semantic overlap between them. She subsequently tried to determine whether the overlap was complete or only partial. In most cases, of course, the overlap is only partial, and Van Leuven-Zwart sought to label all the different kinds of variation. She then counted how many instances of this or that kind of variation occurred and tried to figure out what these quantitative data meant in qualitative terms. In other words, what does it mean for our perception of a text (a) and a text (b) that (a) contains many terms that are more specific that those in (b)?
When you reflect on the amount of interpretation that needs to be done to label and categorize a particular choice, and how many pigeon-holes you might need to put all sorts of nuances and differences in, you begin to appreciate the problem.
The attempts to develop comparative models along these lines have not been abandoned, though. Today they are mostly carried out on written texts, with the help of computers and software that facilitates the alignment of texts in different languages. The methodologies tend to combine linguistic and narratological categories.
As an exercise, think of exactly how you would align and compare the three versions (machine; Human A; Human B) in Translation 2.