Identification Of Chinese Verb Metaphor Based On Topic Model

Research on metaphor computation has become a focal topic in natural language processing.The effectiveness of metaphor processing will directly influence the result of natural language understanding,machine translation,search engines and other related application fields.Metaphor identification is the first step in metaphor processing.Most of existing methods for Chinese metaphors recognition aim at the metaphors in noun phrase level,while researchers pay less attention to the verbal metaphors which are more frequent in real text.The main purpose of this paper is to develop models for verb metaphor identification at sentence level.We proposed two metaphor identification models Chinese verb metaphors based on topic model.Firstly,details of topic model are introduced,a method of Chinese verb metaphor recognition based on the topic model is proposed.In this method,the LDA(Latent Dirichlet Allocation)is first used in the identification process of Chinese Metaphor.We apply topic model to sentences in the corpus,and obtain the topic distribution of every sentence.Then,we manually annotate each topic with a topical word and take it as a feature of SVM,which is used as a classifier in metaphor identification.Experimental results show the validity of the suggested method based on topic model.Secondly,we further give an improved method for Chinese verb metaphor recognition.The method avoids the manually annotation step,while introduces semantic knowledge into topic model.The semantic knowledge base includes Cilin in Chinese and Word Net in English.We choose topic distribution of a sentence,the word abstraction degree and semantic category information as the features of SVM classifier to Chinese verb metaphor identification.This method eliminates the influence of manual annotation,therefore,the experiment can be executed automatically.The experimental result shows the effectiveness of the method.