A Study of Polysemy and Sense Proximity in the Senseval-2 Test Suite

We report on an empirical study of sense relations in the Senseval-2 test suite. We apply and extend the method described in (Resnik and Yarowsky, 1999), estimating proximity of sense pairs from the evidence collected from native-speaker translations of 508 contexts across 4 Indoeuropean languages representing 3 language families. A control set composed of 65 contexts has also been annotated in 12 languages (including 2 non-Indoeuropean languages) in order to estimate the correlation between parallel polysemy and language family distance. A new parameter, sense stability, is introduced to assess the homogeneity of each individual sense definition. Finally, we combine the sense proximity estimation with a classification of semantic relations between senses.