词类标注的目的是排除自然语言文本中多类词的词类歧义．在描述词类标注问题的基础上，讨论了单层高阶动态网络（ＳＨＤＮ）和自联想递归网络（ＡＡＲＮ）用于汉语文本词类标注的实现方法．介绍了两种神经网络模型的结构和训练算法．对小语料的标注实验表明，两种网络的标注正确率分别达到了９２％和９５％以上．对这两种网络的性能进行了比较和分析 The purpose of part-of-speech labeling is to exclude word-type ambiguities of multi-type words in natural language texts. Based on the description of part of speech tagging, this paper discusses how to use single-layer high-order dynamic network (SHDN) and self-associative recursive network (AARN) for Chinese word class annotation. The structures and training algorithms of two neural network models are introduced. Experiments on annotation of small corpora indicate that the annotation accuracy of the two networks is up to 92% and 95% respectively. The performance of these two networks are compared and analyzed