There are predictable alternations, and are also unpredictable alternations. The rule that formulates unpredictable alternation isn't predictable, also isn't proper, yet we cannot say that the rule isn't rule if according to definition of rule. Because a rule is defined, whatever it is, as anything that formulates alternation. That rule is of course just non-proper on the view of veridity of rule. It is merely a problem of interpretation that a relation of two different surface forms [x] and [y] are alternation or the former and latter form of change. Regardless of if The phenomenon applied in [x] and [y] is interpreted in alternation or change, it is obvious fact that some phonological phenomenon had been applied between [x] and [y]. Some people severely have restricted synchronic alternation only to conjugation and declension. But if it does so, we cannot suppose synchronic alternation into word. A word can have one more than allomorph. And there isn't any evidence that a phonological phenomena appeared into word is diachronic one. Underlying form is just a existence interpreted. So we have to decide whether any alternation is synchronic one or diachronic one according to property of phenomenon itself.