| The widespread absence and reference in natural language make the text contain the problem of semantic loss in the global level,which pose a severe challenge to machine understanding.In this paper,the Uyghur has the typical characteristics of the generation refers to the personal pronouns to refer the disambiguation research,from two aspects of person pronouns and zero pronouns to construct Uyghur anaphora disambiguation framework,using context information in the deep semantic text default semantics of Uyghur discourse of the recovery,thus for the machine to understand,supporting for machine translation research and information extraction.In studying the process of anaphora resolution,we find that the recognition of the missing items can avoid unnecessary noise in the process of the anaphora resolution.Therefore,we do some research on the recognition of the dominant personal pronouns.At present,most of the researches on the anaphora resolution are based on the semantic features of artificial extraction,while ignoring the deep semantic mining of the text.In view of Uyghur personal pronoun anaphora,the following research are done in this paper.(1)Aiming at personal pronouns in Uyghur,the framework of personal pronoun based on deep learning mechanism and word embedding is constructed.The use of Bi-directional long short term memory network(Bi-LSTM)to capture the target vocabulary context hidden semantic features.Design of double Bi-LSTM to excavate the antecedent and anaphor candidate in the hidden semantic association context level,starting from the deep semantic level,exploring word embedding and deep learning algorithm in Uyghur pronoun on behalf of the elimination of the rationality and validity of the anaphora resolution task.(2)For Uyghur zero pronouns,construct the Stacked Denoising Autoencoder(SDAE)of the zero pronoun anaphora resolution framework.The word embedding is used as the semantic feature of candidate antecedent and default zero pronoun.In addition,according to the characteristics of the Uyghur default null digits,we construct the hand-crafted feature set containing 14 features.The fusion of word embedding semantic features and hand-crafted feature set are used as input of SDAE,so that model can explore the deep semantic features of features and effectively accomplish Uyghur zero pronoun resolution task.(3)In the anaphora resolution task,there is definite pronoun refers to an entity to above problems,if the anaphora resolution process in the introduction of non essential non coreference invalid,affect refers to the disambiguation performance,therefore,it is necessary to refer to the disambiguation of coreference identification research.In view of Uyghur dominant personal pronouns,this paper constructs a Deep Belief Networks(DBN)based Uyghur personal pronouns to resolved anaphoricity determination.To mention itself and its context information based on the selected 10 feature set as the input of DBN,by training the RBM layer(Restricted Boltzmann Machine,RBM),in an unsupervised manner to learn semantic features in the text implied,and in the last layer of BP network,the global supervised training network and fine-tuning the feature vector of the output of the RBM classification,and finally complete the Uyghur pronoun anaphoricity determination task. |
