Unsupervised word sense disambiguation rivaling supervised methods

Key Points

• The aim is to develop an unsupervised algorithm that effectively disambiguates word senses without the need for extensive manual annotation.
• Developed an unsupervised learning algorithm based on two constraints: one sense per discourse and one sense per collocation.
• Utilized an iterative bootstrapping procedure to enhance word sense identification.
• Tested the algorithm on unannotated English text.
• Achieved an accuracy exceeding 96% in sense disambiguation.
• Performance rivals supervised methods that require hand annotations.