W3C home > Mailing lists > Public > public-swbp-wg@w3.org > October 2005

Re: [WN] editor's draft WordNet datamodel

From: Mark van Assem <mark@cs.vu.nl>
Date: Tue, 18 Oct 2005 11:33:06 +0200
Message-ID: <4354C152.6070207@cs.vu.nl>
To: Dan Brickley <danbri@w3.org>
CC: Guus Schreiber <schreiber@cs.vu.nl>, SWBPD list <public-swbp-wg@w3.org>

Hi Dan,

> I felt licensed to map wordnet hyponym into RDF classes because the 
> wordnet documentation said that hyponym is only used when two terms fit 
> into the following template: "A ____ is a kind of ___". While I did find 
> some cases (in v1.6 at least) where terms didn't seem good as classes, 
> they also seemed to fail on Wordnet's own internal documentation rules. 
> I think "Paris" is (or was) an example.

Yep, the documents can be pretty vague or not entirely correct...
It turns out there are lots of hyponyms that are instances, e.g. 
persons such as "Rembrandt" or "Wall Street".

I certainly can't rule out that there are other relationships that we 
have not converted as they were intended because of documentation or 
misunderstanding NLP terms. For this version we have tried to stay 
neutral on the interpretation of the relations and "errors" in the 
actual data, so that these issues can be sorted out using this "as-is" 
conversion. Or e.g. specific interpretations added on top of it to fit 
different needs.

BTW in WN 2.1 the hyponym relation seems to be split into "direct 
hypernym", "inherited hypernym" and "instance". There "Paris" is an 
instance of "national capital". When we move to 2.1 we might take 
advantage of this info.

> BTW could you also cite http://xmlns.com/2001/08/wordnet/  from the 
> references -- it is the same piece of work begun in the 1999 post, but 
> from an new URL that is in use and the page is even occasionally updated...

Thanks, will add that in the next Draft!


  Mark F.J. van Assem - Vrije Universiteit Amsterdam
        mark@cs.vu.nl - http://www.cs.vu.nl/~mark
Received on Tuesday, 18 October 2005 09:33:52 UTC

This archive was generated by hypermail 2.4.0 : Friday, 17 January 2020 17:31:13 UTC