Greetings all, SenseRelate (http://search.cpan.org/dist/WordNet-SenseRelate-TargetWord/) participated in the recent Senseval-4/Semeval-1 English lexical sample task, which was a coarse grained target word disambiguation evaluation. The task description can be found here: http://nlp.cs.swarthmore.edu/semeval/tasks/task17/description.shtml
There were a few changes made to SenseRelate during the course of these experiments, and those will be released eventually as version 0.10. Right now the current version is 0.09 and is quite close to what we used in the actual experiments. What we did is described in the following paper: UMND1: Unsupervised Word Sense Disambiguation Using Contextual Semantic Relatedness (Patwardhan, Banerjee, and Pedersen) - Appears in the Proceedings of SemEval-2007: 4th International Workshop on Semantic Evaluations, June 23-24, 2007, Prague, Czech Republic. http://www.d.umn.edu/~tpederse/Pubs/umnd1-sval4.pdf The most salient points are that we used the vector measure from WordNet-Similarity, and we did not take advantage of any training or tuning data, which included NOT using information about WordNet Sense 1, so the method is purely unsupervised in the sense that we did not use any training data or information derived from training data. Finally, I have said "we" here a few times, but my role in this was in fact quite limited, Sid and Bano did all of the heavy lifting, and I think arrived at a very interesting and succesful approach for this task. Enjoy, Ted -- Ted Pedersen http://www.d.umn.edu/~tpederse
