Research On Frame Disamniguation Based On Conditional Random Field

The frame disambiguation technology research is a sub-task in the “Frame Semantic Structure Extraction” task of the new semantic analysis evaluation task in the international semantic evaluation SemEval-2007.This research bases on the frame semantic knowledge base resource FrameNet,and aims at the given sentence and the ambiguous target word in the sentence(which can arouse the target words of multiple frames),and based on the context information of the sentence in which the target word is currently located,in the ambiguity target.Among the multiple frameworks that can be inspired by the word,one can select the framework that best represents the semantic information of the current sentence,and finally mark the specific framework information for the ambiguous target word.At present,semantic analysis is a research hotspot and difficulty in the field of natural language understanding.Frame disambiguation is an intermediate part of semantic analysis,and its research is crucial to the development of semantic analysis and even the field of natural language understanding.If we can efficiently handle the frame disambiguation in the semantic analysis,it not only lays a solid foundation for the framework's semantic role tagging system,but also can provide a strong guarantee for the construction of the corpus,and can also be used for information retrieval,question answering systems,and machines.Applications related to natural language processing such as translation and speech recognition provide useful semantic information.This paper mainly studies the problem of frame disambiguation technology.Based on the frame semantic knowledge base resource FrameNet,lexical analysis and dependency syntactic analysis are performed on selected sentences and sentences that can excite target words of multiple frames,and use these information to construct feature templates as ambiguous goals in sentences.Words choose a context-friendly framework.The experiment uses the statistical machine learning method Conditional Random Fields(CRF)algorithm to explore the frame disambiguation technology.The specific research content is as follows:(1)Build experimental corpora.From the FrameNet Knowledge Base,select10 representative vocabularies that can stimulate two or more frames,namely "can","name","kill","say","suggest",and "number"."sense","show","see" and "know".In the experiment,for each ambiguous lexical element chosen,2000 sentences containing the lexical element are extracted,and the obvious errors in the sentence are proofread.As an experimental corpus for this experiment.(2)Feature selection.For each sentence,Stanford Parser Stanford Parser was used to carry out the part-of-speech tagging and dependency relationship parsing.Based on this,the selection of basic feature templates and the selection of dependent feature templates were performed.Finally,10 feature templates were selected.(3)Training and testing.We use the conditional random field algorithm with sequence labeling to model,make full use of its sequence annotation idea,use the CRF++ toolkit for training and testing,and compare the experimental results.The best result is 82.78% accurate.rate.(4)Explore new methods.Currently,FrameNet has not been studied using the conditional random field algorithm at home and abroad.This paper uses the conditional random field algorithm for modeling for the first time,which changes the blank situation that the current conditional random field algorithm deals with the English language FrameNet in the frame disambiguation field.