Modeling language change and language acquisition

Consider the following two models of syntactic change, which we can call the drift model and the imperfect transmission model. Under the drift model, change is driven by a gradual shift in the frequencies with which linguistic forms are used. If, after some time, the frequency of some crucial form drops below the threshold of learnabil-ity, the grammar changes. The frequency changes are driven by factors like random variation, gradual bleaching of meaning, phonological erosion of weakly accented syl-lables or other properties of usage. This model is explicitly or implicitly assumed by many generative and non-generative linguists. Its advantage is that the cause of grammatical reanalysis is obvious: Once the frequency of a crucial grammatical form is low enough, language learners are no longer able to learn grammatical properties that depend on it. The problem with this approach is that the drift in frequencies that it requires has no obvious motivation. If we assume that frequencies are learned by children as part of their knowledge of the language, why should they be learned accurately in through a period of stability and then no longer so learned when a lan-guage starts to change? Similarly, if the frequencies are indirect reflections of other factors, as is more commonly believed, then why should these factors become unstable at certain times?