Research On Key Technologies Of Chinese Pronoun Resolution

The phenomena of anaphora and ellipsis exist extensively which results in ambiguity,make it more difficult to understand the natural language,especially in the application scene of chatting robot.The study of anaphora resolution has a long history,which was from the early manual rules and theoretical knowledge to the technology of computer automatic processing derived from a large-scale corpus,to the introduction of various machine-learning methods,the performance of which promotes constantly.However,due to the understanding and representation of semantics in natural language is still not mature enough,the use of deep language knowledge and semantic features are relatively simple,there is no deep enough digging for the different features of word,sentence,chapter or other multi-levels.There is no effective use of contextual information.This thesis focuses on the improvement of contextual understanding as well as the key techniques of Chinese pronoun resolution and ellipsis restoration.The main contents are as follows:(1)In this paper,we propose a multi-feature fusion Chinese pronoun digestion algorithm,which introduces various types of features such as empirical vectorization,semantic role annotation and word vector to describe the se mantics,structure,and so on.This paper expatiates on the construction and implementation of the whole algorithm framework of Chinese pronouns based on the formulation of the model.On this basis,the different performance of the multi-class features in the task is discussed,and the validity of several feature fusion me thods is proposed and compared based on the vector splicing method,the influence of different classifier parameters,word vector dimension and classifier threshold on the experimental results is verified,and the best experimental results are obtained.(2)In this paper,we propose the depth of learning technology into the pronoun digestion task.In particular,the long and short memory network model of the appropriate serialization input used to describe the deep feature representation of the context.In this paper,we propose a Chinese zero-substitution algorithm based on bidirectional circular network,and try to summarize and summarize the problems existing in zero-generation pronoun recognition task,and put forward the corresponding rule optimization scheme.This paper also stud ies the performance of the deep learning model of different network structures in the omission restoration task of Chinese pronouns,and obtains better model and parameter configuration by contrast test.(3)This paper realizes the intelligent chatting robot system based on wechat platform,introduces the overall structure,module design and system display of the system in detail,and explains the generation a nd elimination of the pronoun.In practice,we discuss the effectiveness of Chinese pronoun digest t echnique and pronoun recovery technology in intelligent robot syst em.The semantic complement task is analyzed and optimized.