权益分类	功能权益	普通用户	{{item.name}}会员
{{category.name}}	{{benefitItem.name}}

Construction of an Optimality-Theoretic Theory of Case and Linking and Its Realization on a Neural Network

案例和链接的最优理论的构建及其在神经网络上的实现

基本信息

批准号：
14510616
负责人：
NAKAMURA Wataru
金额：
$ 2.05万
依托单位：
Tohoku University (2003-2004)The University of Electro-Communications (2002)
依托单位国家：
日本
项目类别：
Grant-in-Aid for Scientific Research (C)
财政年份：
2002
资助国家：
日本
起止时间：
2002 至 2004
项目状态：
已结题

项目摘要

The aim of this year's study was to provide a foundation for transforming a mufti-layered perceptron with a back propagation learning algorithm which realizes an accusative/ergative case system into a neural neywork which adopts a Hebbian learning algorithm as used in our brain. Specifically, er presented and wrote a paper which examined the computational capabilities of a quantum neural network (see Matsumoto, Okubo, and Nishino 2004). This work investigated whether it is possible to apply a very powerful computational capability of a quantum neural network to neuroidal network which is is designed to simulate the process of language acquisition.We were not able to simulate a post-'Verb Island' phase on a neuroidal network, but we are in the process of investigating how to simulate the process of acquiring an accusative linking system as found in English and Japanese under the assumption that syntactic categories such as NP, VP, or AP are innate. Specifically, we are planning to construct a neuroidal network which acquires an accusative system of grammatical relations on the basis of a bundle of semantic features as proposed by Dowry's (1995) and Ackerman and Moore's (2003) protorole theory.Finally, we examined how the human brain comprehends an active/passive sentence, using the functional brain imaging with a particular focus on how Japanese understands (see Yokoyama., Uchida, Nakamura, Kawashima, and others 2004). a Japanese or English sentence.

今年的研究的目的是提供一个基础，将一个多层感知器与反向传播学习算法，实现了一个推理/作格的情况下系统到一个神经网络，采用了Hebbian学习算法，在我们的大脑中使用。具体来说，er提出并撰写了一篇论文，研究了量子神经网络的计算能力（参见松本，大久保和西野2004）。本文研究了量子神经网络的强大计算能力是否可以应用于模拟语言习得过程的类神经网络。我们无法在类神经网络上模拟后“动词岛”阶段，但我们正在研究如何模拟在英语和日语中发现的获得替代连接系统的过程，假设句法范畴如NP、VP或AP是天生的。具体地说，我们计划构建一个神经网络，该网络基于Dowry（1995）和阿克曼和摩尔（2003）的原型理论提出的一组语义特征来获得语法关系的解释系统。最后，我们考察了人脑如何解释主动/被动句，使用功能性脑成像，特别关注日本人如何理解（参见Yokoyama.，Uchida，中村，Kawashima等2004）。日语或英语句子。