Research On Chinese And English Extra-propositional Scope Identification Via Neural Networks

Extra-propositional semantic in natural language mainly refers to negative semantic and speculative semantic.Among them,negative semantic reverses the proposition itself or the semantic of a certain aspect related to it by the negation operator;speculative semantic means that people's expressions of things are in a fuzzy state,unable to give a definitive boundary,which lies between negativity and certainty.The research on the scope identifica-tion of extra-propositional semantic aims to identify the keywords representing negative and speculative semantics and the scope of their semantics in sentences.The scope is usually a continuous segment within a sentence.This research is of great significance for downstream applications of natural language processing,such as information extraction,information re-trieval,and sentiment analysis.The existing extra-propositional semantic scope identifica-tion models have the disadvantages of ignoring sentence context information and cannot satisfy domain adaptability.In addition,the current Chinese-oriented extra-propositional se-mantic scope identification models have low performance and lack effective neural network models.In view of the above problems,this thesis proposes a neural network-based method for identifying Chinese and English extra-propositional semantic scope identification.The main research contents include the following three aspects:(1)Extra-propositional Scope Identification Based on Context RepresentationExisting extra-propositional semantic scope identification models have not fully mod-eled sentence context information.Based on this,this thesis integrates Bi-directional Long Short-Term Memory(BiLSTM)neural network and Conditional Random Field(CRF)for modeling.Among them,The LSTM network can learn the contextual features with the help of forward and backward sequences,and also learn the dependency relationship between the output tags with the help of the CRF layer.On the other hand,the existing methods often ignore the problem of domain adaptability.Therefore,this thesis adopts the Generative Ad-versarial Networks to learn the similarity and difference between different domains.In this thesis,the validity of the model is verified on the BioScope corpus.Among them,the ac-curacy of negative and speculative scope identification has increased by 4.73%and 1.68%,respectively.At the same time,the Generative Adversarial Networks has also achieved su-perior results in cross-domain identification,and the absolute performance of each sub-task has improved by an average of 2.65%(negative)and 2.44%(speculative).(2)Extra-propositional Scope Identification of Chinese Based on Neural NetworkAt present,the research on Chinese-oriented Extra-propositional scope identification is currently scarce.Most of the existing researches are based on traditional feature engineering methods,with low experimental performance and poor scalability.In this thesis,a two-way BiLSTM and a Convolutional Neural Network(CNN)are used to learn both the charac-teristics between contexts and local information in a sequence.Another important reason for the slow progress of Chinese-oriented research is the lack of Chinese corpus resources.Therefore,this thesis uses Generative Adversarial Networks for Cross-language learning,which can semantically map the source language and target language,thereby alleviating feature differences caused by different language distributions.We perform experiments on the CNeSp corpus,the accuracy rates of negative and speculative scope identification are 80.89%and 80.04%,respectively,which are 24.82%and 30.40%higher than the current best methods.(3)Implementation and display of Chinese and English Extra-propositional Scope Identification systemIn this thesis,the two research contents(English and Chinese Extra-propositional scope identification)are studied in detail,and the proposed model is experimentally verified on En-glish and Chinese data sets.The results show that the methods proposed in this thesisr have good effects on Chinese and English Extra-propositional scope identification.The research results have certain help and reference value for related natural language information ex-traction work.In addition,in order to visualize the research content,this thesis implements a prediction system with Chinese and English Extra-propositional scope identification,and puts the research into application.In summary,this thesis is devoted to the study of Extra-propositional scope identifica-tion.On the one hand,it proposes effective methods to improve the performance of related tasks,and on the other hand,it attempts to promote the research in Chinese.It is expected that the preliminary results obtained in this thesis will have certain reference value for relat-ed research in this field and promote the development of deep understanding technology of natural language.