Research On Anaphora Resolution In Noun Phrase Based On Semantic Information

With the rapid development of artificial intelligence,natural language processing plays an important role in the field of artificial intelligence.It is necessary to solve the problem of anaphora disambiguation in order to realize the barrier-free communication between human and computer using natural language.In order to make computer understand natural language correctly,the premise is to understand the anaphora information in natural language.Natural language often uses anaphora for various purposes,which makes it difficult for computer to understand natural language.The difficulty is that the anaphoric relationship between noun phrases can not be accurately identified.This paper studies the role of semantic information in denotation disambiguation of Uygur NOUN phrases,from a large amount of semantic information carried in typical word vectors,to semantic categories,and the extraction of semantic relations.This paper adopts the method of in-depth learning,constructs in-depth learning network,extracts the semantic information of Uygur language,and completes the research of denotation disambiguation of Uygur noun phrases with the semantic information of word vectors.The main work of this paper is as follows:1)Based on the spatial and temporal features of the language,a CNN_BiLSTM dual-channel Uighur noun phrase anaphora disambiguation model is constructed.Convolutional Neural Networks(CNN)is used to extract spatial features of text,and Bi-directional long short term memory(Bi-LSTM)is used to extract temporal features of text.The combination of the two features extracts the semantic information of the word vector from the two channels,and excavates the deeper semantic information in the language.Combination feature is regarded as a binary classification problem,which completes the task of denotation disambiguation of Uygur NOUN phrases.2)A method of in-depth learning is proposed to extract the semantic role and category information of the text as new features to join the Uygur anaphora disambiguation task.Fully Convolutional Networks(FCN)is adopted.After convolution and deconvolution,the input sentences are classified and their semantic information is determined.They are added to the feature vectors of words,and then used as input of the model to complete the task of denotation disambiguation of Uyghur NOUN phrases.At the same time,in order to speed up the training and reduce the amount of data,the feature set is used to do the preliminary screening of data.(3)In view of the above two studies,which only focus on the semantic information of a single word,but do not pay attention to the relationship between the word and the word,we propose adding the semantic dependency feature.First,we add the manual tagging semantic dependency feature to the original feature to verify its validity.Then we use CNN_Attention model to extract the raw corpus(where the raw corpus has been tagged).The semantic dependency feature in reference information is not labeled,and the disambiguation task of Uygur noun phrase reference is completed by adding feature set.After repeated experiments,the two-channel model of Uygur noun phrase anaphora disambiguation effectively improves the F value of Uygur noun phrase anaphora disambiguation.After adding semantic features and semantic dependence,the F value of the model increases more obviously.The most obvious is that it effectively improves the recall rate of Uygur noun phrase anaphora disambiguation.