[Corpora-List] head of a (short) sentence

[Vincenzo (Enzo) Maltese]

Actually in my case they are all noun phrases (without verbs),
as in the example below. I'm more interested in the accuracy.

Thanks
Enzo

-----Original Message-----
From: Yannick Versley [mailto:yversley at gmail.com] 
Sent: Friday, June 24, 2011 1:11 PM
To: Vincenzo (Enzo) Maltese
Cc: corpora at uib.no
Subject: Re: [Corpora-List] head of a (short) sentence

Dear Vincenzo,

The problem you have does come up in computational linguistics occasionally,
and you'll find solutions such as by Ponzetto&Strube:
http://www.cl.uni-heidelberg.de/~ponzetto/pubs/ponzetto07b.pdf
It's probably not too difficult to do something similar by running
(say) the Stanford Parser, and use the head finder from the same package to
determine the head of the whole phrase, or use a dependency parser such as
MALTParser right away (in a dependency parse, the head of the phrase or
sentence is attached to root (0) while all the modifiers are attached to
something else.

However, there are enough small-but-important details that you'd want to
care about (speed vs accuracy tradeoffs, the exact head definition, i.e. do
you want the lexical head "has" or the semantic head "driven" of "Peter has
driven his car"; does it start from raw text or from tokens), and the task
itself is easy enough that people usually build their own version of it
if/when they need one.

Best wishes,
Yannick Versley

On Fri, Jun 24, 2011 at 12:45 PM, Vincenzo (Enzo) Maltese
<maltese at disi.unitn.it> wrote:
> Dear all,
>
> I'm not an expert in NLP, but I'm looking for a java function acting 
> as a black box
>
> that by taking a short sentence in English such as "white wines in Italy"
>
> returns the head of the sentence "wines", i.e. something like:
>
>
>
> String getHead(String)
>
>
>
> Is there any?
>
>
>
> Thanks in advance,
>
> Vincenzo Maltese
>
> _______________________________________________
> UNSUBSCRIBE from this page: http://mailman.uib.no/options/corpora
> Corpora mailing list
> Corpora at uib.no
> http://mailman.uib.no/listinfo/corpora
>
>


_______________________________________________
UNSUBSCRIBE from this page: http://mailman.uib.no/options/corpora
Corpora mailing list
Corpora at uib.no
http://mailman.uib.no/listinfo/corpora