A Study Of Attribute Features Of The Primary English Modal Auxiliaries And Their Contribution To Word Sense Disambiguation

Formal concept analysis is applied to analyze data and extract rules from the formalcontext constructed on the basis of multiple attributes of different objects. Being anessential element of natural language processing, word sense disambiguation can realizethe automatic identification of the senses of target words deriving from the contextualinformation. Attributes extracted from the context and imposing influences on the sensesof target words are available to construct the models for the word sense disambiguation,which will be conductive to reveal the deep semantic relations of target words. The studyof attribute features aims to analyze the relevance and uniqueness of attributes, amongwhich unique attributes and unique composite attributes of the congeneric objects providea new perspective for the word sense disambiguation.According to the theory of formal concept analysis, models for the word sensedisambiguation of the primary English modal auxiliaries are constructed, relationsbetween attribute features and senses of target words are explored, and the contributions ofcontextual features to the word sense disambiguation of target words are measured. Thestudy is based on a corpus of three million words. It tags the senses of root meaning andepistemic meaning of target words, CAN, MAY, MUST, WILL and SHALL, calculates themutual information of each target words to obtain the semantic features, extracts nineteensyntactic features for each word from eight potential dimensions, and constructs theformal context and model for the word sense disambiguation of each word. The accuraciesof the word sense disambiguation of target words, CAN, MAY, MUST, WILL and SHALL,are87%,93.34%,96.33%,92%and96.11%respectively.In addition, unique attributes and unique composite attributes of the congenericobjects of target words are extracted and the relations between these two features andsenses of target words are explained in the thesis. The result demonstrates that differentsenses of target words are closely related to the semantic features of their own andsyntactic features have varied effect on the senses of target words. Finally, contributions of different contextual features to the word sense disambiguation of target words areexpounded. The result indicates that semantic features impose more influence on the wordsense disambiguation of MAY, MUST and WILL and syntactic features have greaterinfluence on disambiguating the senses of CAN and SHALL. Among them, epistemicmeanings of the five target words are largely affected by syntactic features of stative verb,negation and aspect while other syntactic features impose varied influence on differenttarget words. Syntactic feature of passive voice has more influence on the word sensedisambiguation of MAY, CAN and SHALL, syntactic feature of subject animacy has greaterimpact on the word sense disambiguation of MAY and CAN and syntactic feature ofsubjectivity and authority have greater influence on the word sense disambiguation ofMAY and MUST.It is conductive of this study which constructs the models for word sensedisambiguation of the primary English modal auxiliaries and explores the relationsbetween different language features to the word sense disambiguation to the furtherresearches of other English auxiliaries, and provides theoretical and practical foundationsfor the natural language processing and the semantic study of polysemy.