Research And Implementation On Judgment And Relationship Recognition Of Chinese Complex Sentence

Complex sentences as one of the important grammatical units in natural language,plays an important role in natural language.Complex sentences can be divided into explicit complex sentences with connectives and implicit complex sentences without connectives according to the connectives in complex sentences.Complex sentence not only enriches Chinese sentence structure,but also brings difficulties to automatic text comprehension.It is the basic task of natural language processing to accurately judge complex sentences and recognize their semantic relations,which provides support for Chinese abstract semantic representation(CAMR),machine translation,automatic q&a and other upper applications.Therefore,based on Chinese abstract semantic representation corpus and Chinese Discourse Treebank,this thesis uses neural network method to carry out experimental research on complex sentence judgment and semantic relationship recognition.The main works of this thesis are as follows.Construct the data set of complex sentence judgment and semantic relation recognition.The main work of this part is to determine the class of semantic relations contained in this data set by combining the relevant theoretical knowledge of complex sentences,and to construct the data set used in this thesis by manually calibrating sentences extracted from CAMR corpus and Chinese Discourse Treebank.The corpus includes 3644 explicit complex sentences,5144 implicit complex sentences and 5359 simple sentences.This thesis presents a method for automatic judgement of complex sentences based on neural network.In order to take advantage of the context semantic information,sentences are encoded by Bidirectional long short-term memory to obtain context semantic information at the sentence level,while sentence features are captured by attention mechanism to mine deeper semantic information.Through convolutional neural network,the partial information in the sentence is extracted to obtain the sentence feature representation containing rich semantic information.The experimental results show that this method can judge complex sentences effectively.An automatic recognition method for semantic relations of complex sentences based on neural network is proposed.This thesis focuses on the semantic relationship of complex sentences from the perspective of sentence structure features.Use the sentence-level word vector BERT to enhance the semantic representation of sentences.In order to make full use of the structural information of sentences,Tree-LSTM is used to model the syntactic tree tags of words and words in sentences.Because the absence of connectives makes semantic relationship recognition of implicit complex sentences difficult,this thesis integrates the classification information of connectives into semantic relationship recognition based on multi-task learning.Experiments show that the above method can achieve better recognition effect for semantic relation automatic recognition of complex sentences.A joint model is proposed for complex sentence judgment and automatic recognition of semantic relation.In order to reduce error transfer,this thesis constructs a joint model for complex sentence judgment and semantic relation recognition.Through joint model and parameter sharing,global features can be used to improve local prediction.The semantic relationship recognition task can effectively use the information in the complex sentence judgment task by using the attention mechanism,to enhance the performance of the model.Experiments show that the joint model can effectively reduce error transfer and achieve better recognition results in the two tasks.