汉语句法分析的交互激活竞争模型

A Interactive Activation and Competition Parsing Network for Chinese Sentence

本文根据联结主义提出的交互激活竞爭理论,把句法分析看做为一个通过竞爭机制从句法规则集中选取合适句法规则的过程,并建立了一个汉语句法分析的网络模型。这一模型的特点是:(1)能用最少单元把任何短语文法转换成网络表达;(2)能根据一般原则对网络赋权;(3)网络具有较好的收敛性。这一模型己成功地应用于若干汉语基本句型的分析和歧义词的处理。本文还对通常的网络函数计算公式作了改进,使网络能克服“层次障碍”的影响,并保持原有的特性。

We show that parsing can be formulated as a competition process by which proper rule can be selected from a collection of context-free productions.We build a model in terms of Interactive Activation and Competition(IAC) mechanism of connectionism. The advantanges of this treatment are. (1) any context-free grammar can be transformed into an IAC network with fewest number of processing units; (2) the weights of the network can be assigned in accordance with general rules; (3) the network demonstrates a relatively high rate of convergence.The network has been successfully tested for a number of Chinese sentences, including some cases of lexical ambiguity. We also modified the commonly used net function such that the 'level barrier' of IAC network can be lessened without losing its merits