Feature Selection For Sense Classification Of The English Quasi-modal Verb Be Able To Based On The Filter-Wrapper Method

English modalities express the speaker’s subjective opinions or assumptions.However,a research difficulty may be caused by the complexity and indeterminacy of the meaning of modality in the field of semantic recognition.Feature selection can improve the efficiency for sense recognition.Therefore,it is of great significance for natural language processing,language teaching and machine translation to reveal the relationships between English modal verbs and feature of attributes.This thesis studies feature selection for sense classification of the English quasi-modal verb be able to based on the Filter-Wrapper method and finds the relationship between the target word and attributes based on the optimal feature set for sense classification.In this thesis,the sense of the target word be able to is divided into three categories: ability,permission and possibility,then the 300 example sentences are selected from the 3.5 million words corpus as the objects and the extracted 12 semantic features,16 syntactic features and 4 situational features are regarded as attributes to construct a formal context.The method of Filter-Wrapper is adopted to obtain the optimal feature subset for the data set.Based on the method of Filter-Wrapper,the attribute partial ordered structure diagram(APOSD)is used to extract rules of sense classification of be able to,and the accuracy of it reaches 95.7%.The study finds that something related to inherent property of the subject is the simple class exclusive attributes of be able to(ability);the high value of MI between be able to(ability)and subject,the high value of MI between be able to(ability)and predicate verb,the passive voice and the topic related to law and regulations are the simple class exclusive attributes of be able to(permission);the sentence pattern of whether...or not is the simple class exclusive attributes of be able to(possibility).The final feature set contains 21 features,including 5 semantic features(the high value and low value of MI between the predicate verb and be able to(ability),the high value and low value of MI between the subject and be able to(permission)and the high value of MI between the predicate verb and be able to(permission)),12 syntactic features(animate subject,three person subjects,stative verb,past time,present time,interrogative sentence,active and passive voice,negative sentence and the sentence pattern of whether...or not)and 4 situational features(the topic related to law and regulations,personal authority of the subject,something related to subject’s inherent property and the topic related to external circumstances).This thesis can provide a valuable reference for the studies on language teaching and machine translation and for the study of feature selection in intelligent sense recognition,word sense disambiguation and natural language processing.