cant Lexical Relationships

Statistical NLP inevitably deals with a large number of rare events As a consequence NLP data often vio lates the assumptions implicit in traditional statistical procedures such as signi cance testing We describe a signi cance test an exact conditional test that is appropriate for NLP data and can be performed us ing freely available software We apply this test to the study of lexical relationships and demonstrate that the results obtained using this test are both theoretically more reliable and di erent from the results obtained using previously applied tests