We are very happy to announce the release of version 0.13 of WordNet-Similarity. This version has several rather major changes included. You can download from any of these sites:
http://search.cpan.org/dist/WordNet-Similarity http://wn-similarity.sourceforge.net http://www.d.umn.edu/~tpederse/similarity.html First, we have reorganized the class hierarchy to avoid the situation that previously existed where information content measures required information only need for path finding measures, and vice versa. You can see fairly detailed discussion of this problem and proposed solutions in the wn-similarity mailing list. The following message is a good summary of the issue, and you can find related messages in close proximity. http://groups.yahoo.com/group/wn-similarity/message/207 Second, the vector measure has been temporarily removed from the package (until version 0.14) and is now represented simply by a "stub". If you are using the vector measure it is disabled in 0.13, but will return in 0.14 (which will follow fairly quickly). We did this to allow for the clean and orderly inclusion of a new vector measure called vector_pairs. Both vector and vector_pairs are related to the lesk measure in that they represent concepts (word senses) via their definitions. Since definitions are so short usually, we allow the user to specify (via the lesk-relation.dat and vector-relation.dat files) additional glosses that should be added to the gloss of the concept being measured. These extra glosses come from concepts that are related according to WordNet (like the hypernym of a word sense, or the hyponyms, etc.). Now, in the case of vector a single large gloss is created (we sometimes call this a supergloss) and then we replace aech word in that gloss with a co-occurrence vector we have derived from some corpus (by default the corpus used is the WordNet glosses themselves, but it can be any corpora). Now, in the case of vector_pairs, we don't build the super-gloss, but rather we take each gloss separately and expand it using the co-occurrence matrix. Then we measure the cosine between each pair of glosses (as specified in the vector-relation.dat file) and keep a running sume of the cosine values for all the pairs of concepts/glosses specified. Of course the big question might be "why?" -- Why have both vector_pairs and vectors, and then why disable vector to introduce vector_pairs. Let's answer the second question first - it's really just an accident of how we were developing things - we essentially rewrote vector to be vector_pairs, and then realized that both had a place in the system. In addition, tehse changes have been a long time coming, and we didn't want to further complicate and delay the release of 0.13 by having both vector and vector_pairs. Now, how are they different? our feeling is this - vector creates very large super-glosses, and we replace each content word in that supe-gloss with a co-occurrence vector, and then we average all those co-occurrence vectors together. It seems (to us) that this averaging operating is probably introducing a lot of noise into the process, and that maybe a more controlled situation where you have smaller sized glosses (as we have in vector_pairs) being treated this way would result in more accurate measurements. This is simply conjecture at this point, so we want to actually be able to compare vector with vector_pairs in order to better understand this particular issue (which is why we'll have both when 0.14 comes out). Third, there have been considerable internal changes to the vector measure in general - for example, we are no longer using Berkeley DB, and we have added some options to the utility that creates the co-occurrence matrix. What follows is the Changelog for version 0.13. Please check out the new version, and let us know if you have any comments, questions, or observations about 0.13 or anything else! Enjoy, Ted, Jason, and Sid ======================================================================= NAME Revision History for WordNet::Similarity. SYNOPSIS A list of changes to the WordNet::Similarity package. These are copied from the Recently Completed Items in the file todo.pod when a new version is released. DESCRIPTION Version 0.13 (Released 5/9/2005) * 04/21/05 (1) removed LCSFinder module; moved LCS methods to DepthFinder, ICFinder, and PathFinder (JM) (2) renamed vector measure vector_pairs (JM) * 03/24/05 (1) Modified the documentation to reflect the relation file format for vector and for lesk. (Sid) * 03/02/05 (1) Set up selective test cases for "make test", depending upon the default data files installed by user. (Sid) * 02/24/05 (1) Reinstated default relation files for vector and lesk. In case the default relation files (vector-relation.dat and lesk-relation.dat) are missing, both modules would default to the glosexample-glosexample relation. (Sid) (2) Modified Makefile.PL to query the user before installing default data files. (Sid) (3) Removed infocontent file generation code from Makefile.PL. Now Makefile.PL simply calls utilities from the /utils directory (wnDepths.pl, semCorFreq.pl and wordVectors.pl) to generate the all default data files. (Sid) (4) Installation process now generates a default word vectors file. The vectordb configuration variable for vector is now optional. (Sid) (5) Earlier, the WNHOME option was given to Makefile.PL as --WNHOME <path>, whereas the PREFIX option was written as PREFIX=<path>. This inconsistent (and potentially confusing) notation has now been fixed. Now, the WNHOME option is provided to Makefile.PL as WNHOME=<path>. (Sid) (6) Added some basic tests for vector in t/vector.t. * 12/11/04 (1) Created WordNet::Similarity::GlossFinder.pm, a super-class of WordNet::Similarity::vector and WordNet::Similarity::lesk. (Sid) (2) Removed default relation file for lesk. Vector and lesk both default to glosexample-glosexample. (Sid) -- Ted Pedersen http://www.d.umn.edu/~tpederse ------------------------ Yahoo! Groups Sponsor --------------------~--> Give underprivileged students the materials they need to learn. Bring education to life by funding a specific classroom project. http://us.click.yahoo.com/LZzaMD/_WnJAA/HwKMAA/x3XolB/TM --------------------------------------------------------------------~-> 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/
