Formal Models for Learning in the Principles and Parameters Framework
Partha Nyogi, Robert C. Berwick · 2000
Abstract One of the goals of linguistic theory is to develop a formal scheme to represent ‘linguistic knowledge’: in other words, what it is that we know when we know a language. One approach to this is the ‘principles and parameters’ framework (Chomsky 1981) whose central thesis is that knowledge of language can be reduced to knowledge of a (finite) number of principles. These universal principles take specific forms in the different natural languages of the world, and correspondingly associated with them are a number of parameters which capture the parametric variation across the world’s languages. Roughly speaking then, there are a number of parameters: setting them to one set of values might represent knowledge of German (and correspondingly define the grammar of German), setting them to another might define knowledge of Japanese, etc. This is a scheme to represent knowledge and a natural question to ask is how can this knowledge be acquired? The problem of language acquisition is reduced to learning to set the parameter values for the target language on the basis of sentences from that target. Thus a child is born in some linguistic environment; it receives evidence about the parental (target) parameter settings in the form of parental (societal) linguistic input, and on the basis of this evidence, is able eventually to set the parameters to the appropriate values and thus acquire the language of its parents. From a computational perspective, we are interested in algorithms whereby the child might set parameter values on the basis of data; whether these algorithms converge to target values in the limit and further the rate at which they do so Figure 11.1 provides a schematic view of the processes involved in language acquisition.