语言可学性与语言习得的计算模式  

Language Learnability and Computational Models of Language Acquisition

作　　者：戴曼纯[1] 

机构地区：[1]湖南大学外国语学院

出　　处：《现代外语》2001年第2期206-215,共10页Modern Foreign Languages

摘　　要：语言可学性研究不是关于语言是否可以被学得,而是试图解释语言习得机制完成习得任务的方式。除了原则与参数框架下的大量实证研究外,还有些学者提出了不同的语言习得计算模式。本文主要介绍计算模式的核心思想,如触发、触发性学习运算系统、参数空间、参数设置运算系统、马尔可夫链、结构触发学习者和以线索为基础的学习者等。虽然这些计算模式存在一些问题,但是它们是非常有益的理论尝试,是在不断发展的理论思路。The study of language learnability is not oriented to whether language is learnable but to explaining how the language acquisition device does its work. Those who follow Chomsky in theoretical orientation attempt to construct language acquisition theories in line with UG theory. For example, the study of Manzini & Wexler (1987) investigates the parametric setting issue of binding theory and has been extensively cited in related studies. Apart from extensive empirical studies, some computational models of language acquisition have been proposed within the framework of principles and parameters theory. The alternative approach to language learnability can be found in the computational models of Clark & Roberts (1993), Gibson & Wexler (1994), Frank & Kapur (1996), Berwick & Niyogi (1996), Fodor (1998) and Dresher (1999). This paper is an investigation into the core concepts and notions of those computational models. Triggering data, language input from the target language, is believed to have the function of setting parameters. The learner can parse the triggering data successfully, if and only if one target language parameter has been set properly. In contrast with the hypothesis-testing model, triggering is not a creative process, but a simple and mechanical process (Fodor 1998:1). Triggers can be further classified into two classes, global and local. 'A global trigger for value v of parameter Pi, Pi (v), is a sentence S from the target grammar L such that S is grammatical if and only if the value for Pi is v, no matter what the values for parameters other than Pi are卆 local trigger for value v of parameter Pi, Pi (v), is a sentence S from the target grammar L such that S is grammatical if and only if the value for Pi is v.' (Gibson & Wexler 1994:409). The Triggering Learning Algorithm (TLA) proposed by them assumes that the learner analyzes incoming sentences from the target language. If S is analyzed successfully, then the parameter remains unchanged. If the learner cannot analyze S, then he uniformly selects

关 键 词：语言可学性 语言习得机制 参数空间 马尔可夫链 计算模式 

分 类 号：H09[语言文字—语言学]