The nature of phoneme representation in spoken word recognition.

Four experiments used the psychological refractory period logic to examine whether integration of multiple sources of phonemic information has a decisional locus. All experiments made use of a dual-task paradigm in which participants made forced-choice color categorization (Task 1) and phoneme categorization (Task 2) decisions at varying stimulus onset asynchronies. In Experiment 1, Task 2 difficulty was manipulated using words containing matching or mismatching coarticulatory cues to the final consonant. The results showed that difficulty and onset asynchrony combined in an underadditive way, suggesting that the phonemic mismatch was resolved prior to a central decisional bottleneck. Similar results were found in Experiment 2 using nonwords. In Experiment 3, the manipulation of task difficulty involved lexical status, which once again revealed an underadditive pattern of response times. Finally, Experiment 4 compared this prebottleneck variable with a decisional variable: response key bias. The latter showed an additive pattern of responses. The experiments show that resolution of phonemic ambiguity can take advantage of cognitive slack time at short asynchronies, indicating that phonemic integration takes place at a relatively early stage of spoken word recognition. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Lexical characteristics of words used in emotional Stroop experiments.

Validity of the emotional Stroop task hinges on equivalence between the emotion and the control words in terms of lexical features related to word recognition. The authors evaluated the lexical features of 1,033 words used in 32 published emotional Stroop studies. Emotion words were significantly lower in frequency of use, longer in length, and had smaller orthographic neighborhoods than words used as controls. These lexical features contribute to slower word recognition and hence are likely to contribute to delayed latencies in color naming. The often-replicated slowdown in color naming of emotion words may be due, in part, to lexical differences between the emotion and control words used in the majority of such studies to date. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Comparison of opportunities to respond within a drill model when rehearsing sight words with a child with mental retardation.

Sight-word instruction can improve functioning with various daily, recreational, and work-related tasks among children with moderate to severe disabilities. Previous research has demonstrated the effectiveness of drill methods to teach sight words if the model contains at least 50% known items, which would also increase the number of opportunities to respond (OTR) to the target item. The current study taught five sight words each day to a child diagnosed as moderately mentally retarded using high OTR and moderate OTR conditions and a drill ratio of 10% unknown to 90% known. Results suggested that the high OTR condition led to better retention than the moderate OTR despite containing an equal ratio of unknown to known items between them. Implications for practicing school psychologists and future research are included. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Information technology at the National Computing Centre

An outline is given of the information services used in the National Computing Centre. An update is provided of a previous paper describing the local area network implemented in 1981. Three IT landmarks are highlighted, these include the introduction of word processing, the local area network and a digital PABX. The implications and subsequent developments are described. The facilities available on the local area network are discussed. The use of three electronic mail services are also considered, along with the use of facsimile.     

基于WWW的未登录词识别研究

一、引言当前,随着国民经济信息化的不断发展以及Internet的普及应用,全世界丰富的信息资源展现在我们每个人面前。如何从大量的信息中迅速有效地提取出所需信息极大地影响着我国计算机技术和信息技术的发展和应用推广。据统计,在信息领域中,80％以上的信息是以语言文字为载体的,因此,中文信息处理技术成为我国重要的计算机应用技术。未登录词的识别是中文信息处理技术中的难点之一。它在Internet数据挖掘、信息检索、图书馆图书文献管理、语音识别等应用中     

中文文本中外国人名与中国人名同步识别方法

根据中国人名和外国人名的构成特点产生潜在中国人名和外国人名，然后把它们作为节点词加入到句子的分词有向图中，利用上下文信息对有向图的边赋值．使有向图最短路径对应句子正确切分．在确定句子正确切分时识别出句子中的外国人名和中国人名，该方法可以避免由分词结果造成的人名不能被召回的现象，提高了人名识别的召回率．通过对真实语料的测试，在封闭测试中该方法对中国人名和外国人名识别的综合指标F值为97．30％．     

基于细粒度词表示的命名实体识别研究

命名实体识别(NER)是自然语言处理中的一项基础任务,其性能的优劣极大地影响着关系抽取、语义角色标注等后续任务。传统的统计模型特征设计难度大、领域适应性差,一些神经网络模型则忽略了词本身所具有的形态学信息。针对上述问题,该文构建了一种基于细粒度词表示的端到端模型(Finger-BiLSTM-CRF)来进行命名实体识别任务。该文首先提出一种基于注意力机制的字符级词表示模型Finger来融合形态学信息和单词的字符信息,然后将Finger与BiLSTM-CRF模型联合进行实体识别,最终该方法以端到端、无任何特征工程的方式在CoNLL 2003 数据集上取得了F₁为91.09%的结果。实验表明,该文设计的Finger模型显著提升NER系统的召回率,从而使得模型的识别能力显著提升。     

基于目标语词汇组合合理性评价的译文选择模型

机器翻译使用计算机实现自然语言的翻译功能,是计算机技术在语言学领域的新兴应用和研究,其中,源语言词汇的译文选择直接决定机器翻译的质量,是人们关注的难题,基于目标语词汇组合合理性评价的模糊分类模型试图以语言学和模糊数学为理论基础,尝试解决歧义词的译文选择问题,将该模型用于英汉机器翻译系统的译文选择实验,具有良好的译文选择性能,它的优势还在于可以同时为全句范围内的所有歧义词选择译文,因此适用于机器翻译的大规模开发任务。     

利用汉字二元语法关系解决汉语自动分词中的交集型歧义

本文提出了一种利用句内相邻之间的互信息及ｔ－测试差这两个统计量解决汉语自动分词中交集型歧义切分字段的方法。初步的实验结果显示，可以正确处理９０．３％的交集字段。     

参与中文配对词语联想学习记忆的脑区和神经机制——脑功能磁共振的研究

通过脑功能磁共振技术,研究健康人参与语言的词语配对联想学习记忆任务的脑区和神经机制。对16名右利手健康志愿者进行一项词语配对联想学习记忆任务作业的同时,进行脑功能磁共振扫描。实验采用组块设计,实验任务（包括记忆编码相和记忆提取相）与对照任务（共两个相）交替进行;数据采用SPM99软件进行数据分析和脑功能区定位。结果表明：左侧额叶,特别是左侧额叶的额中下回和枕叶的18,19区在词语联想学习记忆的编码阶段中起重要作用;而左侧顶上小叶、缘上回和角回则在进行记忆提取阶段起重要作用;左侧纹状体边缘区参与了人脑词语联想学习记忆作业的编码阶段。揭示了人大脑完成语言联想学习记忆任务时,除额、顶、枕和颞叶的皮层结构参与外,还新发现有皮层下结构如纹状体参与了词语联想学习记忆。在配对词语的编码和提取阶段,激活的脑区有所变化,显示了这两个语言阶段的神经活动变化机制。