[developers] parse forest output

Rebecca Dridan rdrid at dridan.com
Fri Oct 22 00:46:42 CEST 2010


On 21/10/10 19:40, Bernd Kiefer wrote:
> Hi Rebecca,
>
> what do you have in mind? A packed forest of all possible trees?

Basically, yes. The unsupervised parse selection and supertagging work 
we've been doing has been over "the parse forest", meaning over all of 
the derivation trees.  We think it might be better working directly over 
the forest, and Dan suggested it might be worth coming up with a 
standard forest output.

> And what should be the forest node content?

At least rule names, spans, parents/daughters as in the derivation 
trees. I'm not sure what else - perhaps bookkeeping related to packing?


Rebecca

>
> Bernd
>
> On 21.10.2010 00:16, Rebecca Dridan wrote:
>> Hi all,
>>
>> Some of the experiments that I want to run would, we think, be better
>> run over a full parse forest, rather than the individual derivation
>> trees in the standard result file. Hence, I'm looking to make (local)
>> changes to PET to allow a parse forest output and I was curious if
>> anyone had done that before and had hints or code to pass along.
>>
>> Also, if there is anyone else who would find that output useful, maybe
>> we could talk about a standard format?
>>
>> Rebecca
>




More information about the developers mailing list