Output details

An investigation into feature construction to assist word sense disambiguation

<24> This paper describes a novel methodology for constructing binary features from relational knowledge representations that can be processed by robust machine learning algorithms; it led to significant performance improvements. The work resulted from an international collaboration with IBM Research Labs (co-authors), following an internship at their site. A preliminary version was awarded best application paper at the Inductive Logic Programming Conference. The successful application of this methodology to disambiguating words in context led to a strand of work at IBM for other problems, such as information extraction.