Combination of Global and Local Attributional Similarities for Synonym Detection

2000 Mathematics Subject Classification: 68T50.

In this paper, we present a new methodology for synonym detection based on the combination of global and local distributional similarities of pairs of words. The methodology is evaluated on the noun space of the 50 multiple-choice synonym questions taken from the ESL and reaches 91.30% accuracy using a conditional probabilistic model associated with the cosine similarity measure.