Unsupervised Word Sense Disambiguation Rivaling Supervised Methods

7 Using the one-sense-per-discourse property

1. How is one sense per discourse used after convergence?
   For error correction: low-confidence tags of a word that occurs several times in a discourse are overridden by the dominant tag, when the cumulative evidence (summed log-likelihood ratios) for the majority sense exceeds the minority’s by a threshold depending on the number of occurrences.
   • Yarowsky, 1995, Unsupervised Word Sense Disambiguation Rivaling Supervised Methods (Proceedings of the 33rd Annual Meeting of ACL), §7
2. How does using one sense per discourse at every iteration differ from using it at the end?
   Where a discourse strongly disagrees about the dominant sense, all its instances are returned to the residual instead of keeping their tags, preventing mistagged collocates from gaining a foothold and keeping training data purer.
   • Yarowsky, 1995, Unsupervised Word Sense Disambiguation Rivaling Supervised Methods (Proceedings of the 33rd Annual Meeting of ACL), §7
3. What is the main limitation of the one-sense-per-discourse constraint?
   Coverage: about half the examples occur in discourses with no other instance of the word to corroborate the sense.
   • Yarowsky, 1995, Unsupervised Word Sense Disambiguation Rivaling Supervised Methods (Proceedings of the 33rd Annual Meeting of ACL), §7