Disambiguating Distributional Neighbors Using a Lexical Substitution Dataset
Abstract: This paper addresses the issue of polysemy in a distributional thesaurus. In such resources, distributional neighbors can relate indistinguishably to various senses. We propose a method to cluster the neighbors of a target word with respect to its senses, i.e. to attribute one sense to each neighbor. This is made possible by the use of a lexical substitution dataset, to which the distribution of the neighbors are compared.
Many NLP applications need to know whether a word A is semantically more related to B than to C. Unsupervised corpus-based approaches to similarity ...