Research On The Recognition Method Of Chinese Syntax Elements Based On Deep Neural Network

Syntactic elements refer to the syntactic elements such as subject,predicate,object,and time in the sentence.Syntactic element recognition can obtain grammatical and semantic information of sentences,which supports natural language processing applications such as Chinese word segmentation,knowledge graph construction,machine translation,and automatic question answering.Due to the hieroglyphs composed of ideographic symbols in Chinese characters,the structure between words and words is loose and a lack of morphological changes of words.And predicate word formation is diverse and cannot be distinguished morphologically.Therefore,Chinese syntax analysis and Chinese word segmentation are extremely error-prone.In addition,the expression of Chinese sentence patterns is flexible,and there are special sentence patterns such as object preposition and inversion,which leads to serious structural ambiguity in Chinese sentences.The recognition of Chinese syntactic elements is always a difficulty in Chinese information processing.This paper focuses on two aspects of Chinese syntactic element recognition.?1?Make the standard of syntactic element annotation.Based on the characteristics of Chinese,a flat standard of Chinese syntactic elements annotation is proposed.The standard defines six syntactic elements?subject mention?action mention?attribute mention?raising mention?temporal mention?local mention?with predicate as the core.It focuses on the top-level syntactic structure of the sentence,avoids the structural ambiguity caused by the phenomenon of language nesting,and reduces the workload of annotation.According to the proposed annotation standard,750 documents are annotated and the annotation data set is published free of charge.?2?Propose recognition model for Chinese syntactic elements.In view of the strong correlation between syntactic elements and predicate verbs,attention mechanism is introduced into syntactic elements recognition to learn the internal dependency of input sequence and obtain the global information of sentences.First,the text content of the fact description part of the case is learned by word vector.Then,the att-Bi-LSTM-CRF model based on the Bi-directional Long Short-Term Memory model and attention mechanism is used to learn the context grammar information of syntactic elements.Finally,the maximum annotation path is output through CRF layer.This model can effectively improve the recognition efficiency of syntactic elements.The ₁F value is 83.38%,which is 2.85%higher than the traditional CRF model.