Chinese Frame Disambiguation Based On Frame Representation Learning

Frame semantic analysis tasks mainly include three sub-tasks: target word identification,frame identification,and semantic role labeling.The frame disambiguation task studied in this paper is a key sub-task of frame identification.The frame disambiguation task is defined as: how to automatically identify the correct frame for a target word that can evoke several different frames in given context.Frame disambiguation is the bottleneck of the performance of the frame identification model.A breakthrough in this task will lay the foundation for the subsequent construction of a high-accuracy semantic role labeling model,and promote the realization of the computer's effective processing of shallow semantic information in sentences.For the frame disambiguation subtask,the main research content and results of this paper are as follows:(1)This paper proposes a frame representation learning algorithm which can distinguish the correct frame from the error frame in the largest degree.different from the traditional classification algorithm extracting features manually,this paper uses a neural network model to learn frame representation based on sentences in corpus.Making full use of the CFN example sentence database and being based on the hinge-loss neural network,the algorithm learns excellent frame distributed representation vectors which effectively improve the performance of the frame disambiguation model.In addition,this paper compares three other different frame representation learning approaches,including frame representation learning based on lexunit,frame representation learning based on Word2 Vec and frame representation learning based on WSABIE,verifying that the frame representation learning based on neural networks is significantly better than the other three methods.(2)This paper constructs a Chinese frame disambiguation model based on frame representation learning.It uses the WSABIE algorithm to learn the representation vector of the context of the target word.The cosine angle between the context representation vector and the frame representation vector is used to make a decision for the task.In the experiment,three sets of two-fold cross-validation(3×2 BCV)were performed on 88 ambiguous words in CFN,and the best accuracy of frame disambiguation reaches to 72.52%.The t-test results show that the performance of our proposed method is significantly higher than other frame disambiguation methods.(3)This paper designs and develops an automatic Chinese frame disambiguation system.Based on the Chinese frame disambiguation model of frame representation learning,the system realizes the automatic processing of frame disambiguation in a syntactic and semantic analysis task based on frame semantics.In addition,the system also implements the information management module of frame database,lexunit database,and sentence database involved in frame disambiguation task,which regulates the basic data of frame disambiguation and subsequent semantic role labeling tasks.The Chinese frame disambiguation model constructed in this paper can lay the foundation for subsequent Chinese frame semantic analysis tasks.In addition,the Chinese frame distributed representation learning method proposed in this paper can also be more deeply applied and improved in other Chinese frame semantic analysis tasks.