Consider splitting a set in two based, one subset of which the elements satisfy the predicate, and one subset of which the elements don't satisfy the predicate. If the predicate is stronger, the Yes-set will be smaller ("the stronger the restriction, the narrower the class"). Therefore, the No-set will be bigger. And since it works both ways, this means that the negation of the predicate is now weaker.
This does not only work for predicates that become stronger or weaker. If we have a strong predicate, that means the Yes-set is small. Since the No-set is the complement of the Yes-set, it has to be big, and therefore the negation of the original predicate (that is, the predicate of the No-set) has to be weak.
The same, mathematically.
Suppose we have a set X and a predicate P. Then Y = {x ∈ X ∣ P x} and N = {x ∈ X | ¬P x} partition X, that is, Y ∪ N = X and Y ∩ N = ∅.
Since "the stronger the restriction, the narrower the class", Y will be smaller when P is stronger. And since N = X ∖ Y, this means that N will be greater, therefore, ¬P will be weaker. A similar argument can be made when P gets weaker: Y will be greater, so N smaller, therefore, ¬P will be stronger.