Recognition And Application Of Chinese Simple Noun Phrase

The noun phrase often acts as important syntactic ingredients in a sentence, such as subject, object and attributive. Recognition of noun phrase is the foundation of many natural language processing fields. As a special type of noun phrase, Chinese simple noun phrase, which has a combination of structure simplicity and semantic accuracy, can not only keep structure completely, but also meet the needs of syntax analysis. This paper takes the recognition of Chinese simple noun phrase as main task, and applies it to a specific syntactic analysis task. In this paper, the main work are as follows:Analyze differences between simple noun phrase and other types noun phrase, while clarify its definition and identification task. Penn Chinese Treebank (CTB) is selected as original data, and experimental data set is finally obtained with it.To recognize the simple noun phrase, we select three mainstream statistical machine learning methods, namely the maximum entropy, conditional random field and support vector machine to build models, and word, part of speech and sentimantic informations are employed as features. Furthermore, the superposition method is used to build composite classifier on basis of single model, which improves the result effectively, and finally gets a F value of 90.91%.Based on the statistical recognition method, a kind of algorithm is proposed, by using structure parallelism and semantic similarity characteristics of parallel structure.This algorithm not only improves the recognition effect of simple noun phrase, but also analysis its inner parallel constituent. Experiments show that we have solved the recognition problem of multi noun coordinate structure to a certain extent.Simple noun phrase is applied in the maximal noun phrase identification in this paper after analysis of low precision of Chinese maximal-length noun phrase identification. The complexity of maximal-length noun phrase is reduced by breviary and replacement strategy with using simple noun phrase. With using a single classifier which has the best effect in this task, F value increases by 1% when the automatic identification simple noun phrases are used in underlayer, while it'll increase by 4% under ideal circumstance when totally correct simple noun phrases are used.