PARAMETERS, LEARN ABILITY, AND THE NULL SUBJECT PARAMETER

Abstract

Parameters were originally proposed within the Principles and Parameters model of linguistic analysis (Chomsky, 1981) as a means of explaining the relative ease and rapidity with which children construct grammars for their target languages. These parameters were seen as linking together seemingly unrelated surface syntactic features so that once "set" on the basis of simple input data, readily available to the child, a number of syntactic features would be "learned" simultaneously. In this paper, we consider the evolution of the concept of the parameter and show how leamability considerations and cross-linguistic variation have led to a very different notion of parameters based on the goal of preventing overgeneralization, in contrast to the original goal of preventing undergeneralization. Turning to the specific case of the null subject parameter, we evaluate various proposals (Hyams, 1986, 1987a; Bloom, 1990, 1993; Valian, 1990b) for what the initial setting of that parameter is, what the triggering data is for setting it, and what markedness relation exists between different values of the parameter. It becomes clear that the original conceptualization of a parameter faces serious empirical challenges when one attempts to specify precisely how it might work in particular cases.