一种用于模拟汉字认知过程的多层自组织神经网络

为了模拟汉语初学者的汉字认知过程,在Kohonen神经网络的基础上,改进了其网络结构和算法,并且将改进后的网络输出层根据Hebbian学习规则连接,构建了一个多Kohonen网络协同工作的汉字认知自组织神经网络模型.模拟研究结果表明,模型能够成功地学习到汉字的结构类型,且能有效识别出汉字的部件,在一定程度上模拟了汉字认知的部分过程,说明该模型用于汉字认知乃至汉语言习得的可行性.

Simulation of Chinese characters learning with improved multi-SOM network

In order to simulate the Chinese character acquisition process, this paper set up a multilayer self-organizing maps (SOM) network model based on improved Kohonen network. The model's output maps, which adapt modified algorithm and expand neuron's neighborhood, were connected via associative links updated by Hebbian learning. After training the model could learn Chinese character architecture successfully and also do well in Chinese character component recognition. The simulation results demonstrated that the feasibility of further research in Chinese character acquisition and even Chinese language learning with this model was possible.