文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > A COMPARISON OF NATURAL AND MACHINE LEARNING OF STRESS
侵权投诉

A COMPARISON OF NATURAL AND MACHINE LEARNING OF STRESS

In the language acquisition literature children are reported to make typical errors in assigning stress to words. These errors are taken to be unequivocal evidence that children learn rules for stress assignment as opposed to learning on a word-by-word bas

rules is so strong as to take effect when rules are not only necessary for the production ofcorrect forms [...], but are also obscured by large numbers of both real and apparentexceptional forms." (Hochberg, 1988b, p. 685) In a Universal Grammar (UG) approachchildren's propensity to learn rules is taken for granted and can be elegantly explained:children learn the parameter values of the language. The (metrical) parameters areinnately given and the process of determining their appropriate values is a consequenceof children's innate endowment to acquire language given suitable input. Although not allstudies of stress acquisition explicitly adhere to the UG theory, they share the idea thatchildren learn rules for stress assignment.

Why are children hypothesized to learn rules for stress assignment, and why is learningstress on a word-by-word basis dismissed? To start with the latter, learning stressassignment on a word-by-word basis is empirically inadequate. There are two convincingarguments for the empirical inadequacy of an approach which advocates memorization ofindividual words and their stress patterns without explicitly abstracting the regularities ofthe domain in the form of rules. First of all, such an approach lacks the necessarygeneralization capacity. And secondly, it is inadequate in accounting for developmentalphenomena such as overgeneralizations. We will briefly review both arguments.

The first argument - lack of generalization capacity: if children (and adults) would simplymemorize the stress pattern of individual words, they would never succeed in assigningcorrect stress to novel words. Dresher and Kaye (1990) and Dresher (1992) explicitlydiscussed such an approach. A table is constructed in which the learner stores words (orstrings of weighted syllables in the framework of metrical theory) together with theirappropriate stress patterns. However, such a learner would be unable to project itsgrammar to assign stress to words not encountered before: if the table look-up fails (noexact match is found because the word is not in the lexicon), the system is at a lossbecause it has no way to assign stress in that case. Hence, they conclude that "[...] suchlearning theories are empirically inadequate." (Dresher & Kaye, 1990, p. 301)

In the developmental literature, this issue is discussed without an extensive investigationof specific models. For instance, Hochberg 's (1988a, b) prime issue is that “[...] childrenlearning Spanish as a first language learn rules for assigning stress, as opposed to simplymemorizing stress on a word-by-word basis.” (Hochberg, 1988b, p. 683) It can only beinferred that learning on a word-by-word basis is understood as storing all words in atable (the lexicon) and looking up a word in the table for the purpose of stress

上一页第4页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email