Abstract
Word sense disambiguation is an important and difficult problem in the field of natural language processing. A sense tagged corpus is often used as training data in word sense disambiguation research. But constructing a sense tagged corpus is very time consuming. This paper presents an unsupervised word sense disambiguation approach using WordNet definition and bilingual dictionary. We extract collocations as features to disambiguate word sense. The experiment uses bilingual sentences in the LDOCE as training data and Brown corpus as testing data. We have tested and evaluated the English adjective “hard”. Experimental results show that the accuracy rate using the proposed method is 93%, that indicates our approach is effective for word sense disambiguation.