A Model for Word Sense Disambiguation

Word sense disambiguation is one of the most difficult problems in natural language processing. This paper puts forward a model for mapping a structural semantic space from a thesaurus into a multi-dimensional, real-valued vector space and gives a word sense disambiguation method based on this mapping. The model, which uses an unsupervised learning method to acquire the disambiguation knowledge, not only saves extensive manual work, but also realizes the sense tagging of a large number of content words. Firstly, a Chinese thesaurus Cilin and a very large-scale corpus are used to construct the structure of the semantic space. Then, a dynamic disambiguation model is developed to disambiguate an ambiguous word according to the vectors of monosemous words in each of its possible categories. In order to resolve the problem of data sparseness, a method is proposed to make the model more robust. Testing results show that the model has relatively good performance and can also be used for other languages.