Statistical Learning of Linguistic Structure

语言结构的统计学习

• 批准号: 0446929
• 负责人: Paul Smolensky
• 金额: --
• 依托单位: Johns Hopkins University
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2005
• 资助国家: 美国
• 起止时间: 2005-06-01 至 2009-06-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0446929&HistoricalAwards=false
• 关键词: Statistical; Learning; Linguistic; Structure

It is a commonplace observation that children, in contrast to non-human primates or sophisticated computers, are able to learn a language in what seems to be an effortless fashion. Yet, the basis of this ability remains a point of contentious debate. Some theorists emphasize the importance of nature, focusing on the exquisitely abstract and subtle generalizations that form part of a child's linguistic knowledge. Others stress the importance of nurture, highlighting the remarkable ability of child language learners to detect and exploit subtle statistical properties of the language input they receive. While there is no incompatibility between these two perspectives on language learning, there has to date been little systematic investigation of the relative contributions of nature and nurture in an empirically rich domain. With support from the National Science Foundation, Drs. Robert Frank, William Badecker, and Donald Mathis will explore this middle ground, focusing on a diverse and crosslinguistically variable set of phenomena in the domain of sentence structure (syntax). This research project will use computational simulations of artificial neural networks to probe the precise character of the generalizations that emerge when statistical learning is applied to such complex syntactic data. These generalizations will be compared to patterns of human linguistic behavior via a parallel set of psycholinguistic studies, allowing the identification of contexts in which human linguistic knowledge is a close reflection of statistically-induced patterns and contexts in which it bears the stamp of a learner's innate nature.The broader impacts of this project include an understanding of factors that might distinguish between the processes normal and abnormal language development and point to possible lines of intervention, as well as a potential characterization of certain types of language breakdown. The interdisciplinary character of this work, encompassing ideas from linguistics, psycholinguistics, and computer science, will enhance the integration of research and education both through the unique research opportunities the project will provide for undergraduate and graduate students as well as through the development of innovative courses on the mental structure of language that will incorporate a laboratory component to be made publicly available on the web.

与非人类灵长类动物或复杂的计算机相比，儿童能够毫不费力地学习一门语言，这是一个司空见惯的观察。然而，这种能力的基础仍然是一个有争议的争论点。一些理论家强调自然的重要性，关注构成儿童语言知识一部分的精致抽象和微妙概括。另一些人则强调后天培养的重要性，强调儿童语言学习者发现和利用他们接收到的语言输入的细微统计特性的非凡能力。虽然这两种语言学习观点之间没有不相容之处，但迄今为止，在经验丰富的领域中，对先天和后天的相对贡献的系统调查很少。在美国国家科学基金会的支持下。罗伯特·弗兰克、威廉·贝德克尔和唐纳德·马西斯将探索这一中间地带，重点关注句子结构（句法）领域中多种多样、跨语言可变的现象。该研究项目将使用人工神经网络的计算模拟来探索当统计学习应用于如此复杂的句法数据时出现的泛化的精确特征。这些概括将通过一组平行的心理语言学研究与人类语言行为模式进行比较，从而确定人类语言知识是统计诱导模式和语境的密切反映，而在这些语境中，语言知识带有学习者天生的印记。这个项目更广泛的影响包括了解可能区分正常和异常语言发展过程的因素，并指出可能的干预路线，以及某些类型的语言障碍的潜在特征。这项工作的跨学科特点，涵盖了语言学、心理语言学和计算机科学的思想，将通过该项目为本科生和研究生提供独特的研究机会，以及通过开发关于语言心理结构的创新课程（将包含实验室组成部分，并在网络上公开提供），加强研究和教育的整合。