Finding Instance Names and Alternative Glosses on the Web: WordNet Reloaded

摘要

This paper presents an approach to extending existing lexical resources with instance names and alternative definitions acquired from textual documents. The experiments involve WordNet and approximately 300 million Web documents, but the method is more generally applicable. We leverage formally-structured, human-validated resources, on one hand, and data-driven instance names and definitions on the other, which opens the path to new applications of the reloaded resources.