On Tue, Apr 17, 2012 at 8:25 AM, Jörn Kottmann <[email protected]> wrote:
> On 04/17/2012 03:20 PM, Jason Baldridge wrote: > >> I haven't followed this in detail, but I do wonder why we don't have a >> single model that just predicts all the types? That is the standard thing >> to do... >> > > We can do that. Anyway you could still end up in a situation where > you want to merge the output of multiple name finders. > Maybe you have a maxent name finder and a Dictionary or Regular Expression > Name Finder. > > Then some application have a need to merge the names so there are no > overlaps. > > > That's a very good point. > > FWIW, integrating the output of multiple classifiers and incorporating >> their probabilities is something that can be done quite cleanly with >> approaches like Integer Linear Programming. >> > > Sounds interesting, do you have a paper on this? > > Pascal Denis and I used it for coreference: http://www.sepln.org/revistaSEPLN/revista/42/03Articulos-p19-87a96.pdf There is a fairly extensive body of work in NLP using integer linear programming: http://www.google.com/cse?cx=011664571474657673452%3A4w9swzkcxiy&cof=FORID%3A0&q=integer+linear+programming#gsc.tab=0&gsc.q=integer%20linear%20programming&gsc.page=1 One of the primary people to bring it to the attention of NLP research was Dan Roth. Here's a tutorial he and his students did in 2010: http://l2r.cs.uiuc.edu/%7Edanr/Talks/ILP-CCM-Tutorial-NAACL10.ppt What do you think about having a simple baseline version? > > +1 -- Jason Baldridge Associate Professor, Department of Linguistics The University of Texas at Austin http://www.jasonbaldridge.com http://twitter.com/jasonbaldridge
