On triggered learning

In current linguistic theory, natural languages are thought to depend on extensive interaction between systems of principles or violable constraints. The principles or constraints are considered to be innate, and subject to a small degree of variation. This variation is expressed in parameters or, alternately, in the ranking of constraints. Under such accounts, the child acquires the language of his or her community, in part, by establishing the settings of parameters or the ordering of constraints. The forms which the learner uses in this task are known as triggers. The study of triggered learning emphasises a number of dimensions of language acquisition. Here I consider three such dimensions: the space of possible languages, the nature of the input the child receives, and the nature of the learning algorithm. Chapter 1 provides an overview of the phenomena to be explained, the linguistic theory and the specific theory of learning which I assume. Chapter 2 investigates the extent to which small changes in the internal representation of grammars correspond to small changes in the resulting language. Chapter 3 explores the role that forms not accounted for by the target grammar may have on parameter setting. Chapter 4 shows that aspects of the learning scenario which lead to a failure of one kind of learning algorithm may not be as severe for another. In conclusion, I try to clarify the role that computational studies of language acquisition can play in the construction of linguistic theory, and the generation of testable hypotheses.Arts, Faculty ofLinguistics, Department ofGraduat