Towards chinese metaphor comprehension based on attribute statistic analysis

Abstract

Chinese metaphor computation is an interdisciplinary frontier research topic, and it focuses on natural language understanding at the semantic level. In current work, metaphor comprehension is mainly processed by using rule-based methods, which are subjective and difficult to be expanded. In addition, the similarities of metaphor in each instance are not distinguished according to specific surrounding contexts. In contrast to the current work, this paper firstly reviews the metaphor processing mechanism in human brains, so as to establish an attribute essence based approach, called “attribute matching method”, to present metaphor knowledge. Based on the “context dependence hypothesis”, this paper describes three different levels of contexts, which include a language knowledge system, objective contexts, and subjective cognitive states. The similarities of metaphor in the dynamic construction process via the procedure of metaphor comprehension are regarded as an attribute classification problem, which concentrates metaphors as the kernel and is restricted by the context information of the instance between tenors and vehicles. This paper also presents a matching algorithm of metaphorical similarities based on ensemble classifier method. This method constructs the algorithm’s framework from a perspective on the multi-value classification. The experimental results demonstrate that an effective similarity extraction method of Chinese metaphor based on statistical analysis of attributes has bee successfully established. This work enriches the Chinese metaphor computation method, and promotes the development of the computational metaphor.