汉字识别的计算机模拟  

Computer simulation of Chinese character recognition

作　　者：彭聃龄[1] 刘颖[1] 陈鹰[1] 

机构地区：[1]北京师范大学心理学系

出　　处：《应用心理学》1996年第1期9-16,共8页Chinese Journal of Applied Psychology

基　　金：国家教委! ( 1993～ 1995 );国家自然科学基金!会 ( 1995～ 1997)的资助

摘　　要：本文总结了作者近年来在连结主义的理论框架下 ,用计算机模拟汉字识别的工作。文章分三部分。第一部分说明了计算机模拟与人工智能的关系。第二部分介绍了作者提出的两个模型 :汉字识别与命名的连结主义模型和基于语义的词汇判断的计算模型。两个模型分别成功地模拟了汉字识别中的频率效应、形声字读音中的规则效应、声旁效应、语义启动效应、语境与频率的交互作用等。第三部分讨论了模拟工作的意义、分布表征、学习算法等问题。研究表明 :认知的计算机模拟能验证人类认知实验的结果 ,对结果提出合理的解释 ,并能指导进一步的实验研究。This paper summarizes the recent work conducted by the authors on the computer simulation of Chinese character recognition. In Part 1, the distinction between computer simulation (CS) and artificial intelligence(AI) is made. The main goal of CS is to substantiate and test the adequacy of the psychological theory of human performance; it is a “processing” simulation. In contrast, the aim of AI is to get a machine to do a particular job as efficiently as possible, such as translating texts, diagnosing a disease, or keeping accounts; it is a so called “function” simulation. In Part 2, two models developed by the authors are introduced. One is the connectionist model of Chinese character recognition and pronunciation in which three sets of units and their connections are constructed. The orthographic and phonological information of Chinese characters are represented distributively. Using the back propagation learning algorithm, the network learned to read and pronounce 1108 Chinese characters. The performance of the model captured some of the characteristic aspects of the naming task, such asfrequency effects, regularity effects, phonetic effects, and the interactions between both frequency and regularity as well as between frequency and phonetic. The other one is the computational lexical decision model based on semantics. The model consisted of a forward network of five sets of orthographic, hidden, semantic, lexicon, and decision units. Weights on connections between orthographic, hidden and semantic units were modified during the training phase using the back propagation learning algorithm. The model simulated some aspects of human performance in the lexical decision task, including the frequency effect, semantic priming, the interaction between frequency and context, the word repetition effect, and the degradation effect. In Part 3, some questions are discussed, such as the role of computer simulation in cognitive research, distributed and local representation, and the learning algorith

关 键 词：汉字识别 命名 词汇判断 计算机模拟 

分 类 号：B842[哲学宗教—基础心理学]