Automatic association of Web directories to word senses

The aim of this research is the development and application of algorithms to combine lexical information with web directories, in order to associate Wordnet word senses with ODP (Open Directory Project) directories.  Such associations can be used as rich domain labels and to acquire sense-tagged corpora automatically, cluster topically-related senses and detect sense specializations.

Our current algorithm has been evaluated for the 29 nouns (147 senses) used in the Senseval 2 competition, obtaining 148 word sense/ Internet directory associations covering 88% of the domain-specific word senses in the test data with  86% accuracy.

The richness of Internet directories as sense characterizations is evaluated in a supervised Word Sense Disambiguation task with the Senseval 2 test suite. The results indicate that, when the directory/word sense association is correct, the training samples acquired automatically from the Internet directories are as valid for training as the original Senseval 2 training instances.

The following data is currently available:
 

• The full set of associations between noun senses in wordnet 1.7 and ODP directories, and a summary file with the coverage of the system.
• The full set of hyponyms (sense specializations) extracted for Wordnet 1.7 from ODP directories.
• The training material obtained automatically for the Senseval 2 WSD test suite, which can be compared with the corresponding hand-tagged training material provided by the Senseval 2 organization, and tested against Senseval 2 test material.