Null Instantiation Identification And Resolution Based On The Chinese Frame Semantic Relations

Traditional semantic role labeling only focus on semantic roles which have explicit expression in the sentence while ignore the implicit arguments,resulting in some loss of semantic argument.In the Chinese Framenet,these implicit semantic roles are called Null Instantiations.Some of the default semantic roles can be understood without the need to retrieve or find the corresponding antecedent in the discourse context,which is called Indefinite Null Instantiations.Some of the default semantic roles can establish a specific semantic association between each argument,and the corresponding antecedent needs to be found in the context,which is referred to as Definite Null Instantiations.Discovering and finding these missing semantic roles contributes to a deeper understanding of textual semantics.With the support of the National Natural Science Foundation Project "Chinese discourse frame semantic network automatically extract and semantic reasoning",this paper based on Chinese frame semantic relations,studies the three sub tasks: the Null Instantiations identification,the Null Instantiations classification and the Definite Null Instantiations resolution.The main research contents and achievements are as follows:(1)For the Null Instantiations identification task,this paper focuses on the relationship between the Null Instantiations core-frame element sets,and then improve the Null Instantiations identification rules,proposes the Null Instantiations identification method based on frame element relations.(2)For the Null Instantiations classification tasks,we found that the Null Instantiations have different contexts,belongs to a different framework,different target words,different core frame elements missing,the classification result is likely to be different.Therefore,this paper selected four kinds of features of word level and frame-level,and then established the SVM classification model and maximum entropy classification model to achieve the Null Instantiations classification.(3)For the Definite Null Instantiations resolution tasks,this paper studied the relationship between frame-to-frame of Chinese FrameNet.Then,considered the problem as a two-classification problem of anaphora resolution,and establish the maximum entropy classification model and the SVM classification model,selected the best feature combination and candidate word sets through experiments;Finally,combined with the framework semantic features and frame-to-frame relations,realized the filling of definite null instantiations.Based on Chinese frame semantic relations,this paper studies the methods of Null Instantiations identification and resolution,and validates the effectiveness on Chinese authentic corpus,which provides a new way for deeper discourse semantic analysis.