Relieving polysemy problem for synonymy detection

Abstract

In order to automatically identify noun synonyms, we propose a new idea which opposes classical polysemous representations of words to monosemous representations based on the "one sense per discourse" hypothesis. For that purpose, we apply the attributional similarity paradigm on two levels: corpus and document. We evaluate our methodology on well-known standard multiple choice synonymy question tests and evidence that it steadily outperforms the baseline. © 2009 Springer Berlin Heidelberg.