LODES of grammar: Syntactic error analysis for intelligent computer-assisted language instruction.

摘要

The objective of this thesis is to investigate a principled theory of error analysis for Intelligent Computer-Assisted Language Instruction (ICALI). The research question addresses: how can a natural language processing program, namely LODES, can detect and classify errors in syntax according to a model of Universal Grammar based on the Principles and Parameters of Government and Binding theory? This is a feasibility study demonstrating how an error-detection ICALI program can make judgements of grammaticality and report violations of well-formed sentences in a manner similar to a human tutor. In the past error detection for conventional Computer Assisted Language Instruction (CALI) has been limited by reliance on many rules that describe the surface structure idiosyncracies of student input. CALI programs have also been restricted by the lack of a linguistic model. The present research attempts to overcome these shortcomings by relying on the modular linguistic theory of Noam Chomsky and techniques from Artificial Intelligence research that perform natural language processing. Drawing on both theoretical linguistics and artificial intelligence we show how to detect errors and infer an explanation for the cause of an error based principles and parameters rather than rules. Our data comes from native Spanish-speaking undergraduates at the University of Puerto Rico enrolled in low level English as a foreign language classes. Our analysis of student errors is a result of the modularity of the natural language processor which allows us to pinpoint a particular error type as a violation of a particular linguistic principle because the module in which the error occurs shows a particular kind of violation that we can classify. We find that a third of the sample sentences could be described and explained as violations of one or more principles elaborated in Universal Grammar theory. The remaining sentences demonstrate that the errors made by beginning second language learners are too numerous to be handled by a fixed order application of principles. It is recommended that additional computational techniques be combined with a modular linguistic theory to increase the number and type of sentences that can be handled.