CAN POPPERIANS LEARN TO TALK?

In several recent publications (Sampson [1978], [1980a]) Geoffrey Sampson has argued that an essentially Popperian language acquisition device could learn language much as a human child does. The device Sampson envisions would freely (or perhaps randomly) generate hypotheses about the grammar the child seeks to learn, and test these hypotheses against the data available to the child. If the data are incompatible with an hypothesis, the hypothesis is rejected and another one tried. If any hypothesis does not conflict with the data, it is retained for further testing. Sampson's language acquisition device exhibits a Popperian proclivity for strong, simple hypotheses, but when it sets out on its acquisition task it has no a priori information about the nature of the grammar it seeks to acquire. If Sampson is right, then there is no need to postulate rationalist learning mechanisms of the sort advocated by Noam Chomsky (in Chomsky [1975] and many other publications) to account for the child's ability to learn language. However, it is my view that Sampson's argument is fatally flawed. He has not shown that a Popperian learning mechanism could duplicate the child's feat. Indeed, I think it can be shown that Sampson's Popperian learner could not possibly match the language learning achievement of the normal child, save by miraculous accident. In an earlier paper (Stich [1979]) I set out the argument that a Popperian device could not learn language, and I explained where Sampsons defence of the Popperian learning mechanism had gone astray. My conclusion was not that Chomsky's rationalist account of language learning must be correct, for I see serious problems with that view as well. Rather, I urged that the correct theory of language acquisition ought to be sought in the very substantial domain that lies between Chomskian rationalism and Popperian empiricism.