A Study On Word Sense Disambiguation Of English Modal Verb MUST

Generally speaking, ambiguity widely exists in human natural languages. In order to avoid this linguistic phenomenon, the appropriate sense of an ambiguous word in a given context has to be identified by analyzing semantic, contextual information and even the common knowledge beyond the text. This way for research is called Word Sense Disambiguation (WSD), which is regarded as one difficulty but also a heating topic in the study of Natural Language Processing. In the meantime, it is a key problem in the field of Machine Translation, Information Retrieval, Text Categorization and also Linguistics. There are many ways to solve the problem of WSD. Currently, with the rise and development of corpus linguistics, the machine learning methods based on statistics are booming, such as Decision Tree, Decision List, Genetic Algorithm, Na?ve-Bayesian Classifier, Maximum Entropy Model and so on. They have eventually become the leading methods in the research of WSD. BP neural network is one of the most popular machine learning methods, and additionally, its possibility and superiority for building a WSD neural network model have been proved by the researches done by predecessors.The methods in the study of WSD have been developed rapidly, whereas the object of such research is still mainly focused on the regular nouns and verbs with obvious distinction and easily identified senses. Few of them can touch those words like English modal verbs that are more ambiguous, fuzzy and sensitive to the context. Therefore, word sense disambiguation of English modal verbs will be studied on the basis of the theory of connectionism in this thesis. It aims to first build a BP neural network model based on a large scale corpus to realize the word sense disambiguation of English modal verb MUST. And then based on the analysis of the different results worked out by different models, the influence of different semantic and syntactic features on the effect of WSD for modal verb MUST is studied and the ranking of the degree of the influence by the different linguistic features is given.A BP neural network for WSD of modal verb MUST has been established based on a corpus. The accuracy of it reaches 96%. Both the influence and the degree of influence of different linguistic features on WSD have been studied through an analysis of the results worked out by models with different linguistic features as input vectors. This research reveals the difference of contribution by different semantic and syntactic features to the WSD of English modals, and provides a valuable basis for both linguistic study on modality and selection of linguistic features for WSD of English modals. Additionally, this study not only extends the word sense disambiguation from the regular nouns and verbs to the modals, but also extends a new approach for linguistic studies to solve a linguistic problem by means of a machine learning method.In the future, the senses of every modal verb may be sense-tagged automatically. When the linguists apply a large scale corpus to study the modals, the realization of the automatic sense-tagging of these modal verbs can save much time and energy and improve the efficiency. Therefore, this research has both of theoretical and practical meaning on either linguistic studies or word sense disambiguation studies.