I will be presenting a paper written by Sid and me (mostly Sid, truth be told) at the EACL 2006 Workshop "Making Sense of Sense" in just a few weeks time.
The workshop will be on April 4 in Trento. I hope to see you at the workshop or at EACL. Below is a link to a preprint of the paper, if you are interested. Using WordNet Based Context Vectors to Estimate the Semantic Relatedness of Concepts (Patwardhan and Pedersen) - Appears in the Proceedings of the EACL 2006 Workshop Making Sense of Sense - Bringing Computational Linguistics and Psycholinguistics Together, April 4, 2006, Trento, Italy. http://www.d.umn.edu/~tpederse/Pubs/eacl2006-vector.pdf This paper describes the vector measure in some detail, and is highly recommended if you are interested in using or understanding that measure in WordNet-Similarity. As a personal aside, I feel that the vector measure (and lesk for that matter) has a great unrealized potential. I think even Sid and I do not quite yet understand its power and flexibility, and we remain very interested in pursuing these ideas. So, your comments and feedback really would be valuable. I also feel that vector really does capture something that many of these other measures don't, and that is the idea of defining or categorizing meaning based on context, which is what we think humans do too to some degree, and that is where we see the connection to psycholinguistics (and hence our interest in this workshop). WordNet of course owes its basic architecture to ideas about how humans organize semantic content (George Miller's ideas, of course), and we use that as a core for vector. But, unlike any of the other measures in WordNet-Similarity, it's possible to add contextual information from any corpora you might wish, thus overlaying this cognitive model that WordNet presents with contextual content that we think humans use to further refine and categorize those basic ideas of meaning. Now, to be clear, Sid and I have not worked out a universal theory of everything, but we really do think that vector is or can do something very different and interesting, and that it might be a bit more like what humans do than let's say path or jcn or even lesk (we don't really find overlaps in definitions in our heads too often). So, comments on all of this would be really helpful to us. We are still thinking about all these issues, and are very open to ideas that either confirm or deny what we are thinking (which I doubt that I have articuled too well here, but hopefully you get the idea). Cordially, Ted -- Ted Pedersen http://www.d.umn.edu/~tpederse Yahoo! Groups Links <*> To visit your group on the web, go to: http://groups.yahoo.com/group/nlpatumd/ <*> To unsubscribe from this group, send an email to: [EMAIL PROTECTED] <*> Your use of Yahoo! Groups is subject to: http://docs.yahoo.com/info/terms/
