One of the interesting problems that was mentioned various times at NAACL was name disambiguation. This is the problem of identifying which person (or entity) a name refers to. This seems like a trivial problem, when your name is something like Horatio Louis Fizzlewart, but what if you are John Smith, or Jim Clark, or even...Ted Pedersen!
If you search on Google for "Ted Pedersen", here are the first four results: ================================================================ Ted Pedersen - Personal Home PageHome Page of Ted Pedersen. I am an Associate Professor in the Department of Computer Science at the University of Minnesota in Duluth . ... www.d.umn.edu/~tpederse/ - 5k - May 23, 2004 - Cached - Similar pages Ted Pedersen - Software... We are interested in both manually and automatically aligning parallel text. Find some interesting tools here. By: Ted Pedersen - [EMAIL PROTECTED] www.d.umn.edu/~tpederse/code.html - 3k - May 22, 2004 - Cached - Similar pages Ted Pedersen BibliographyA bibliography of Ted Pedersen's books and short stories, with book covers and links to related authors. ... Fantastic Fiction, Ted Pedersen. Search Authors. ... www.fantasticfiction.co.uk/authors/Ted_Pedersen.htm - 16k - Cached - Similar pages Ted Pedersen - TV TomeTV Tome is your guide to Ted Pedersen. Biography, roles and appearances, gossip and more. Click Here. Ted Pedersen. . Biographical ... www.tvtome.com/tvtome/servlet/ PersonDetail/personid-43742 - 16k - Cached - Similar pages ================================================================ Now, as you probably know, I'm the first two, but the second two are not me! So the problem of name disambiguation would be to group or cluster these web pages such that each cluster refers to one particular and distinct Ted Pedersen. The good news (for us) is that name disambiguation is very much like word sense disambiguation, and might even be more like word sense discrimination (which is what SenseClusters does). Remember that word sense discrimination takes instances of a word and divides them into clusters based on their meaning. So in this case we'd want to take instances of a name (100 sentences or paragraphs or web pages) that have "John Smith", and identify how many "John Smith"s are being referred to, and which instances refer to which particular person. Interestingly, SenseClusters can do this without any modification whatsover, and would be a very easy experiment to run. ( http://senseclusters.sourceforge.net ) The even better news about name disambiguation as a problem is that it's fairly easy to create test data. Suppose you have a corpus of text. Pick out two names that occur in that text (e.g., George Bush and John Kerry) and replace those names with "PersonX". Then cluster the instances of PersonX, and see how closely they correspond to the actual clusters of John Kerry and George Bush. Once again, this kind of evaluation is something that SenseClusters already supports. Now, as far as NAACL goes, there was an approach described that made me think of SenseClusters - It was called "Cross Document Coreference in a Large Scale Corpus" and it was by Chung Heong Gooi and James Allan of UMass, and it was all about different clustering algorithms that are all supported by SenseClusters that were applied to name disambiguation, or what could be called name discrimination. There was another paper at NAACL this year (in addition to the Umass one mentioned above) that presented a probabilistic model for "Robust Reading: Identification and tracing of ambiguous names". This was by Li, Morie and Roth at Illinois. Finally, some nice previous work in name discrimination was presented last year at CoNLL: Gideon S. Mann, David Yarowsky, "Unsupervised Personal Name Disambiguation", CoNLL, Edmonton, Canada, 2003 http://www.cs.jhu.edu/~gsm/publications/pd.gsm03.pdf This is a nice paper, with lots of good examples, and it points out the unsupervised nature of the problem. In any case, name disambiguation seems to be an important problem these days. I think it has to do with the fact that information extraction from the web is a fairly big thing, and name disambiguation is a very large practical problem. It can also affect the names of organizations as well. For example, UMD - do you mean U of Minn, Duluth or University of Maryland? What about AAAI - American Association for Artificial Intelligence or American Aerospace Activities, Inc. (www.aaai.org versus www.aaai.com) Anyway, almost by accident, it turns out that SenseClusters is ideally suited for name disambiguation research! But, it's not completely an accident. I know that Amruta was always interested in using SenseClusters for more than simple word sense discrimination, so I think the adaptability to this new problem is a consequence of that. Enjoy, Ted -- Ted Pedersen http://www.d.umn.edu/~tpederse ------------------------ Yahoo! Groups Sponsor ---------------------~--> Yahoo! Domains - Claim yours for only $14.70 http://us.click.yahoo.com/Z1wmxD/DREIAA/yQLSAA/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/
