Research On Chinese Functional Constituent Parsing

Syntactic function shows the relationship between all the components of a language.The so-called functional component,usually refers to the subject,predicate,object,adverbial.Besides the constituent parsing and dependency parsing,function constituent parsing is another research content of sentence analysis.The constituent parsing and dependency parsing exist the respective shortcomings,namely,the constituent parsing clearly reflects the hierarchical tree structure of all phrases between sentences,but lacks analysis of the relationship parsing between the phrase;although the dependency parsing analyzes and marks the relationship between phrases in sentence,but lacks hierarchy within the sentence.However,the syntactic function constituent parsing can combines well with the characteristics of the both analysis.Hierarchical relationships within all components of the sentence are given,and the relationship between the various components are analyzed.Moreover,function constituent information of sentence will have great positive effect on many tasks in Natural Language Processing in the field.For example,in the Machine Translation task,we can add the function constituent information to the process of word alignment;in semantic analysis tasks,function co nstituent information can be used as a limitation in the analysis process.However,the function constituent parsing of sentence,so far at home and abroad is not specifically studied.Accordingly,we put forward the construction of Chinese sentence function constituent Treebank annotation system and method for analysis of function constituent parsing in Chinese sentences.The main research contents and achievements of this paper are as follows:(1)according to Linguistics for various function constituent in the relevant definitions,we detailed annotation guidelines about function constituent in the function constituent Treebank,and ultimately forms a annotation system of a complete Chinese sentence function constituent Treebank.(2)taking the Treebank annotation system as the standard,we carried out the error correction and the appropriate extension of the corpus of existing function constituent.At present,training corpus has 23758 Chinese sentences,the manualcorrection test set consists 1000 sentences.(3)on the basis of function constituent Treebank,we studied the method of analysis of Chinese sentences.By comparing the method of the function constituent parsing based on conditional random fields,based on deep learning,and transition-based function function constituent parsing,finally found the model trained by transition-based function constituent parsing performed better.It has not only high analysis precision,but also the output is an analysis tree contained hierarchical structure and syntax information,functio nal information.(4)finally,we made a data statistics and analysis about collocation regularities between relevant function constituent on Chinese function constituent Treebank.The results confirm the results in the previous part of the work on the one hand,on the other hand,will help to further study on function constituent parsing.