[developers] Re: pending/674: generator broken on everything except the ERG?

Valia Kordoni kordoni at coli.uni-sb.de
Wed Oct 26 10:30:52 CEST 2005


Stephan Oepen wrote:

>>The other thing is that it's apparently a lot slower with these
>>little grammars than it used to be.  I don't have timings for the old
>>version, but it now taking up to a minute to generate from an 8 word
>>sentence with the grammar that's supposed to be the end point of our
>>practicals - there are a lot of realisations, but I'm sure it wasn't
>>that slow before.
>>    
>>
>
>that sounds troubling.  the published claim is that we generate about
>twenty times faster now than things used to be.  do you have any kind
>of comparison data (actual timings, edge counts, whatever)?  where is
>the time spent currently?  if you provide the grammar and a pointer to
>an earlier LKB version that you know worked faster, i would be tempted
>to try and investigate this further.
>
>  
>
>>could you keep me up to date by describing anything you check in over
>>the next few weeks, please?
>>    
>>
>
>yes, in fact i think it would be useful if we all made it a habit to
>email `developers' whenever we do something potentially relevant.  how
>do others feel about this?  
>
I do agree!

>i realize ann and i are currently filling
>up quite a number of mailboxes, but myself i am typically content just
>skimming potentially relevant traffic, thus getting some idea of what
>people are currently up to ... 
>  
>
I do agree again! :-) It is v. useful and helpful! I am doing the same.

Valia

-- 

PD Dr. Valia Kordoni, 
Computational Linguistics-Saarland University, 
P.O. Box 15 11 50, 66041 Saarbruecken, Germany and 
LT-Lab DFKI GmbH
Tel: +49 681 302 4682
Fax: +49 681 302 4700
E-mail: kordoni at coli.uni-sb.de
http://www.coli.uni-saarland.de/~kordoni/
http://www.coli.uni-saarland.de/msc/lct/





More information about the developers mailing list