[developers] Challenge

[Ann Copestake]

Glad to hear the paper went down well.  Dan and I talked about this
very briefly with Steve Clark when he visited Stanford and I heard a
practice run of the talk in Cambridge.  There are some issues about the
details of the evaluation which might adversely affect the ERG, but
yes, I agree it's interesting.  

(BTW, note that RASP isn't supposed to deal with unbounded
dependencies!)

Ann


> Hi developers,
> 
> I saw this very well-received paper today at EMNLP:
> http://aclweb.org/anthology-new/D/D09/D09-1085.pdf. It tries to
> evaluate a couple of parsers on their performance on unbounded
> dependencies. I think she said that this dataset can be public, and
> people can try to run their own evaluation on it. Basically, parsers
> like Enju and C&C performed better than the DCU, Rasp and Stanford
> parsers. But the question is: would the ERG manage to do better than
> that?
> 
> Bart.