XPost: comp.ai, comp.ai.alife, comp.ai.edu   
   XPost: comp.ai.doc-analysis.misc   
      
   The programming language itself isn't of paramount concern just yet -   
   figure out how you want to do this.  Probably you'll want to consider   
   estimating similarity between text using Bayesian methods (consider the   
   Rainbow Bayesian tools) and/or the cosine similarity metric (or a   
   related metric).  With only two months probably you want to run some   
   tests using these methods pretty soon.  With more time you could think   
   about combining the results of the two methods.  So as far as that goes   
   you don't need symbolic processing until you start generating the title   
   itself.  Now for that, you don't necessarily need a lot of NLP   
   processing per se: titles are summaries in effect - the shortest and   
   most descriptive or most eye catching summary that you can generate.   
   So check out what the summarization people have to say about this   
   problem.  If you can predict the domain, you can probably generate a   
   grammar for title generation.  Even if you can't predict the domain,   
   you can extract some rules for title generation although for that   
   you'll probably need some kind of ontology to keep your titles safe   
   from something like: Dog flies spacecraft around the world (possibly   
   extracted from a story about the Russian dog in space Laika, for   
   example).   
      
   BTW - I didn't just invent this approach.  I did this as part of a   
   project  four years ago.   
      
   Kirt Undercoffer   
      
   [ comp.ai is moderated.  To submit, just post and be patient, or if ]   
   [ that fails mail your article to , and ]   
   [ ask your news administrator to fix the problems with your system. ]   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)