Restriction Of Co-occurred Contextual Features Upon Word Sense Disambiguation Of English Modal Verb Can

Context has a restrictive and complementary effect on semantics;therefore,it can be used as an important factor in determining the sense of ambiguity words.However,the research of using context information to deal with word sense disambiguation(WSD)is still relatively few.With strong subjectivity,the WSD of English modal verbs has always been a complex and difficult problem in linguistic field.The sense of English modal verbs is determined by the interaction of basic sense and contextual sense.Therefore,it is of great significance to study the WSD of English modal verbs from the perspective of contextual features.Based on a corpus of 120 words,this thesis adopts formal concept analysis and attribute partial-ordered structure diagram(APOSD)to investigate the restriction of co-occurred contextual features upon the word sense disambiguation of English modal verb.Based on co-occurrence relation and co-occurrence probability,the contextual features are divided into two categories: intra-linguistic context and extra-linguistic context.Among them,the context includes 12 semantic features and 6 co-occurred syntactic features,and extra-linguistic contextual features which include seven situational contextual features.English modal verb can is chosen as the target word,and the meaning of can is divided into three categories,namely,ability,possibility and permission.Through the method of formal concept analysis,50 sample sentences are selected from every kind of sense as the objects and the 25 contextual features as the attributes to build the formal context of can.Then the corresponding APOSD is generated as a model for WSD and knowledge discovery,and the accuracy of this model reaches to 92%.And it is 91.33% 1.26%,when cross validated by five-fold cross validation.The major findings are as follows: firstly,semantic features,syntactic features and situational contextual features mutually work together to determine the senses of can.Secondly,every kind of sense co-occurs with different features,and iterative aspect,passive voice and speaker's authority are the unique attributes of ability,possibility and permission respectively.Thirdly,semantic features have the greatest restriction on the sense of can,while situational contextual features have the least restriction.This thesis adopts the method of formal concept analysis and APOSD to research the restriction of co-occurred contextual features upon the word sense disambiguation of English modal verb can.The discovered knowledge contributes to the sense determination of English modal verb can and feature selection in natural language processing.