Clarifying the logical problem of language acquisition.

MacWhinney is to be commended for reopening questions about the logical problem of language acquisition in the light of new data and models. Unfortunately his discussion is marred by technical errors, false dichotomies, and inadequate attention to detail. The ‘logical problem of language acquisition’ (Baker & McCarthy, 1981; Pinker, 1979, 1989) is not the belief that ‘the input to the learner is too inconsistent to determine the acquisition of grammar. ’ The input CANNOT be ‘too inconsistent’, because children do acquire grammar. It is the question of how acquisition could work in principle – how a learner can correctly generalize from a finite sample of sentences in context to the infinite set of sentences that define the language from which the sample was drawn. As such, the logical problem is not a belief or theory or claim (either nativist or empiricist) but a research topic. The problem, as with all problems of induction, is that an infinite number of generalizations are consistent with any finite sample of data. Many curves can be drawn through a set of points, many laws are consistent with a set of observations, and many grammars are consistent with a set of sentences. Therefore any learner who correctly induces a function, theory, or grammar must respect prior (‘ innate’) constraints on its hypothesis space; the data alone are insufficient. This is a logical point which cannot be denied by any theory, nativist, empiricist, behaviourist, connectionist, constructivist, or emergentist (Quine, 1969). For the behaviourists, the innate constraints reside in the generalization gradients and response classes. For the connectionists, they reside in the features defining the units and the topology of the networks. For Chomskyans, they reside in categories, operations, and principles. For MacWhinney, they reside in the cues, items, alternatives pitted in competition, and categories whose absence constitutes ‘ indirect negative evidence.’ Thus ‘conservatism, item-based learning, indirect negative evidence, competition, cue construction, and