I am forwarding this from the Senseval list. Saif wrote the following summary of a discussion at the Senseval workshop in Barcelona, and I knew that many of us might find this interesting.
There is some previous discussion of these issues on the mailing list. There is an archive, but you must be a member to see it! So if you are interested, you can find out how to join here... http://www.senseval.org/contact.html Cordially, Ted What follows is by Saif Mohammad: =============== A summary of the panel discussion on the applications of word sense disambiguation at Senseval-3 workshop, Barcelona, Spain, July 2004. Dr. Phil Edmonds co-ordinated the discussion and introduced the topic with a question like is WSD useful at all and whether we should concentrate on end user applications (those for the lay man) or applications such as machine translation. Dr. Bill Dolan from Microsoft Research is of the belief that WSD is not useful. It may seem that many NLP tasks need WSD but he wondered why we have not been ab le to *show* that WSD is useful. May be we have not reached that level accuracy in WSD or we are trying to automate a task that is not useful. Dr. Dolan believe s that the applications do not benefit from knowing the intended sense but rather by the ability to identify and generate paraphrases. The ability to identify a complex string of words to carry mor or less the same meaning as another string is more beneficial than identifying individual senses of the words. Dr. Dekang Lin gave a "yes and no" answer to whether WSD is useful or not. He us ed the example "evening attire". If one wants to get info about a particular rac ing horse and types 'evening attire' on google, we get a lot of web pages pertai ning to clothes. However, if we add 'racing' to the query one gets documents per taining to the horse named evening attire. Dr. Lin thus argued that we do not ne ed explicit WSD; google seems to do it implicitly. He says that yes, in a way, WSD is needed to determine if two instances of a word refer to the same thing. On the other hand, what is really needed is to determine that the meaning represented by something is the same as the meaning represented by some other instance. He believes that a predetermined sense inventory is not adequate and not good for this purpose. An inventory implies that the sense is discrete. Further granularity of senses can make a big difference in different contexts. For example, we would like to make a difference between opera tickets and tickets to a baseball game in certain contexts but not so in other contexts (for example when consider ing traffic tickets and movie/baseball tickets). Dr. Peter Turney asked if explicit WSD is useful or not? He pointed out that the results at DARPA suggest that implicit WSD was found to be more useful. Dr. Dolan replied saying that the state-of-the-art WSD systems are not useful presently but may be next year explicit WSD will win. Dr. Franz Och pointed out that so far people have not used WordNet etc for machine translation. He said that syntax has not been helpful for machine translation and believes that semantics will not help either. However, so far integrating semantics with the statistical system has not been tried out. He suggested that that the output of a statistical system should be looked at and we should find out what is going wrong. Then try and use use semantics to help in those cases. Dr Turney commented that as per Dr. Charniak, machine translation will be the next killer-app. _______________________________________________ Senseval-discuss mailing list [EMAIL PROTECTED] http://listserv.hum.gu.se/mailman/listinfo/senseval-discuss ------------------------ Yahoo! Groups Sponsor --------------------~--> Make a clean sweep of pop-up ads. Yahoo! Companion Toolbar. Now with Pop-Up Blocker. Get it for free! http://us.click.yahoo.com/L5YrjA/eSIIAA/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/
