Macquarie University, Sydney, Australia.
Cogn Neuropsychol. 2006;23(1):96-109. doi: 10.1080/02643290500202649.
Two current approaches to the computational modelling of reading are the connectionist triangle model approach (parallel processing; distributed representations; models developed via a connectionist training algorithm such as backpropagation) and the DRC ("dual route cascaded") model (serial processing components permitted; local representations; model architecture specified by the modeller rather than acquired by a learning algorithm). One way of testing such computational models is to lesion the computer programs that instantiate the models and study how well the impaired reading of such damaged models corresponds to the patterns of impaired reading seen in people with acquired dyslexia. This is computational cognitive neuropsychology. It has been used with both types of model in attempts to simulate acquired surface dyslexia and acquired phonological dyslexia. The results of this body of work currently favour the DRC model over connectionist models developed within the triangle model framework.
目前有两种方法可以对阅读进行计算建模,分别是连接主义三角模型方法(并行处理;分布式表示;通过连接主义训练算法(如反向传播)开发的模型)和 DRC(“双重路径级联”)模型(允许串行处理组件;局部表示;由建模者指定的模型架构,而不是通过学习算法获得)。测试此类计算模型的一种方法是损伤实例化模型的计算机程序,并研究受损模型的阅读能力与后天性失读症患者所表现出的阅读障碍模式的吻合程度。这就是计算认知神经心理学。它已经被用于这两种类型的模型,试图模拟后天性表层失读症和后天性语音失读症。目前,这方面的研究结果倾向于 DRC 模型而不是连接主义三角模型框架内开发的模型。