Ambiguity minimization makes sense when you are following a principle of computational kindness. That is if Alice would like to reduce the computational burden on Bob then she can reduce or minimize the ambiguity of her sentence. This is because in order for Bob to interpret an ambiguous sentence Bob must in essence sort all possible interpretations of that sentence from most likely interpretation to least likely interpretation, and before he can even sort he must first search in order to find all possible or at least plausible interpretations.
This is very computationally expensive for Bob but very cheap for Alice. Alice knows exactly what she means but Bob has no clue what Alice REALLY means.