Distribution-Based Semantic Similarity of Nouns

In our previous work we have proposed two methods for evaluating semantic similarity / dissimilarity of nouns based on their modifier sets registered in Oxford Collocation Dictionary for Student of English. In this paper we provide further details on the experimental support and discussion of these methods. Given two nouns, in the first method the similarity is measured by the relative size of the intersection of the sets of modifiers applicable to both of them. In the second method, the dissimilarity is measured by the difference between the mean values of cohesion between a noun and the two sets of modifiers: its own ones and those of the other noun in question. Here, the cohesion between words is measured via Web statistics for co-occurrences of words. The two proposed measures prove to be in approximately inverse dependency. Our experiments show that Web-based weighting (the second method) gives better results.